[ARM][ParallelDSP] Relax alias checks

[samparker]

Summary

When deciding the safety of generating smlad, we checked for any writes within the block that may alias with any of the loads that need to be widened. This is overly conservative because it only matters when there's a potential aliasing write to a location accessed by a pair of loads.

Now we check for aliasing writes only once, during setup. If two loads are found to have an aliasing write between them, we don't add these loads to LoadPairs. This means that later during the transform, we can safely widened a pair without worrying about aliasing.

However, to maintain correctness, we also need to change the way that wide loads are inserted because the order is now important.

The MatchSMLAD method has also been changed, absorbing MatchReductions and AddMACCandidate to hopefully improve readability.

[SjoerdMeijer]

lib/Target/ARM/ARMParallelDSP.cpp

351

Functions CheckRAWDeps and CheckWARDeps are the same, just the order of Base and Offset are different, and whether WarDeps or RawDeps are queried, but this could be passed in as an argument.

379

Can you elaborate on what this function is doing? My first reaction would be, given that we are looking at 2 loads, how come are we looking for RAW dependencies? Perhaps the function name is a bit misleading?

[samparker]

lib/Target/ARM/ARMParallelDSP.cpp

351

I will have a look into changing these.

379

Yeah, it is... even looking at it myself I'm getting confused - not a great sign! First we check is the dependency sets are the same, which means there's no write between the loads. The next checks allow the sets not to be equal as long as the dominating load is the only one with RAW dependency. This means we can safely schedule the write and then the combined base + offset pair.

[samparker]

Greatly simplified the alias checks after they didn't seem to make much sense...

[SjoerdMeijer]

lib/Target/ARM/ARMParallelDSP.cpp

369

I think I am struggling with the algorithm here. Here, and in the line below, we get the "write sets" of the 2 loads. We then start iterating over them , and if a write does not occur in both sets then we say it is safe to pair. But is this enough?

[samparker]

lib/Target/ARM/ARMParallelDSP.cpp

369

Good point! No... we also need to check that it's possible to either move the first load forward or the second load backward.

[samparker]

When we create a wide load, we pull the second load up to the first. So for aliasing worries, we only need to check whether there's a write between the two loads that would prevent us from hoisting the second load.

[SjoerdMeijer]

This looks okay to me; just some nits inline.

lib/Target/ARM/ARMParallelDSP.cpp

323

nit: instruction -> instructions

344

nit: I don't think this condition can be true. The loads/reads are Simple loads (not atomics), and the only writes that can be loads are loads with atomic ordering.

368

nit: newline

619

Not even sure if it is possible, but looking at this condition I am wondering about i128s.

676

Don't think we need to check for hasOneUse again here; we only add loads with one use to the list. Asserts for BaseSExt and OffetSext might be useful.

[samparker]

lib/Target/ARM/ARMParallelDSP.cpp

344

Makes sense, good catch!

619

The IR supports arbitrary bit widths, but we only support 32- and 64-bit macs.

676

Cheers, I'll remove the check - and we are asserting on SExt!

[samparker]

Committed in rL360567.

lib/Target/ARM/ARMParallelDSP.cpp

namespace {
   using PMACPairList    = SmallVector<PMACPair, 8>;
   using Instructions    = SmallVector<Instruction*,16>;
   using MemLocList      = SmallVector<MemoryLocation, 4>;
 
   struct OpChain {
     Instruction   *Root;
     ValueList     AllValues;
     MemInstList   VecLd;    // List of all load instructions.
-    MemLocList    MemLocs;  // All memory locations read by this tree.
+    MemInstList   Loads;
     bool          ReadOnly = true;
 
     OpChain(Instruction *I, ValueList &vl) : Root(I), AllValues(vl) { }
     virtual ~OpChain() = default;
 
-    void SetMemoryLocations() {
-      const auto Size = LocationSize::unknown();
+    void PopulateLoads() {
       for (auto *V : AllValues) {
-        if (auto *I = dyn_cast<Instruction>(V)) {
-          if (I->mayWriteToMemory())
-            ReadOnly = false;
-          if (auto *Ld = dyn_cast<LoadInst>(V))
-            MemLocs.push_back(MemoryLocation(Ld->getPointerOperand(), Size));
-        }
+          Loads.push_back(Ld);
       }
     }
 
     unsigned size() const { return AllValues.size(); }
   };
 
   // 'BinOpChain' and 'Reduction' are just some bookkeeping data structures.
   // 'Reduction' contains the phi-node and accumulator statement from where we
class ARMParallelDSP : public LoopPass {
     DominatorTree     *DT;
     LoopInfo          *LI;
     Loop              *L;
     const DataLayout  *DL;
     Module            *M;
     std::map<LoadInst*, LoadInst*> LoadPairs;
     std::map<LoadInst*, std::unique_ptr<WidenedLoad>> WideLoads;
 
-    bool RecordSequentialLoads(BasicBlock *BB);
+    bool RecordMemoryOps(BasicBlock *BB);
     bool InsertParallelMACs(Reduction &Reduction);
     bool AreSequentialLoads(LoadInst *Ld0, LoadInst *Ld1, MemInstList &VecMem);
-    LoadInst* CreateLoadIns(IRBuilder<NoFolder> &IRB,
-                            SmallVectorImpl<LoadInst*> &Loads,
+    LoadInst* CreateWideLoad(SmallVectorImpl<LoadInst*> &Loads,
                              IntegerType *LoadTy);
     void CreateParallelMACPairs(Reduction &R);
     Instruction *CreateSMLADCall(SmallVectorImpl<LoadInst*> &VecLd0,
                                  SmallVectorImpl<LoadInst*> &VecLd1,
                                  Instruction *Acc, bool Exchange,
                                  Instruction *InsertAfter);
 
     /// Try to match and generate: SMLAD, SMLADX - Signed Multiply Accumulate
     /// Dual performs two signed 16x16-bit multiplications. It adds the
     /// products to a 32-bit accumulate operand. Optionally, the instruction can
     /// exchange the halfwords of the second operand before performing the
     /// arithmetic.
     bool MatchSMLAD(Function &F);
 
   public:
     static char ID;
 
     ARMParallelDSP() : LoopPass(ID) { }
 
+    bool doInitialization(Loop *L, LPPassManager &LPM) override {
+      LoadPairs.clear();
+      WideLoads.clear();
+      return true;
+    }
+
     void getAnalysisUsage(AnalysisUsage &AU) const override {
       LoopPass::getAnalysisUsage(AU);
       AU.addRequired<AssumptionCacheTracker>();
       AU.addRequired<ScalarEvolutionWrapperPass>();
       AU.addRequired<AAResultsWrapperPass>();
       AU.addRequired<TargetLibraryInfoWrapperPass>();
       AU.addRequired<LoopInfoWrapperPass>();
       AU.addRequired<DominatorTreeWrapperPass>();
bool runOnLoop(Loop *TheLoop, LPPassManager &) override {
       if (!ST->hasDSP()) {
         LLVM_DEBUG(dbgs() << "DSP extension not enabled: not running pass "
                              "ARMParallelDSP\n");
         return false;
       }
 
       if (!ST->isLittle()) {
         LLVM_DEBUG(dbgs() << "Only supporting little endian: not running pass "
-                             "ARMParallelDSP\n");
+                          << "ARMParallelDSP\n");
         return false;
       }
 
       LoopAccessInfo LAI(L, SE, TLI, AA, DT, LI);
 
       LLVM_DEBUG(dbgs() << "\n== Parallel DSP pass ==\n");
       LLVM_DEBUG(dbgs() << " - " << F.getName() << "\n\n");
 
-      if (!RecordSequentialLoads(Header)) {
+      if (!RecordMemoryOps(Header)) {
         LLVM_DEBUG(dbgs() << " - No sequential loads found.\n");
         return false;
       }
 
       bool Changes = MatchSMLAD(F);
       return Changes;
     }
   };
bool ARMParallelDSP::AreSequentialLoads(LoadInst *Ld0, LoadInst *Ld1,
   );
 
   VecMem.clear();
   VecMem.push_back(Ld0);
   VecMem.push_back(Ld1);
   return true;
 }
 
-/// Iterate through the block and record base, offset pairs of loads as well as
-/// maximal sequences of sequential loads.
-bool ARMParallelDSP::RecordSequentialLoads(BasicBlock *BB) {
+/// Iterate through the block and record base, offset pairs of loads which can
+/// be widened into a single load.
+bool ARMParallelDSP::RecordMemoryOps(BasicBlock *BB) {
   SmallVector<LoadInst*, 8> Loads;
+  SmallVector<Instruction*, 8> Writes;
+
+  // Collect loads and instruction that may write to memory. For now we only
+  // record loads which are simple, sign-extended and have a single user.
+  // TODO: Allow zero-extended loads.
   for (auto &I : *BB) {
+    if (I.mayWriteToMemory())
+      Writes.push_back(&I);
     auto *Ld = dyn_cast<LoadInst>(&I);
     if (!Ld || !Ld->isSimple() ||
         !Ld->hasOneUse() || !isa<SExtInst>(Ld->user_back()))
       continue;
     Loads.push_back(Ld);
   }
 
-  for (auto *Ld0 : Loads) {
-    for (auto *Ld1 : Loads) {
-      if (Ld0 == Ld1)
+  using InstSet = std::set<Instruction*>;
+  using DepMap = std::map<Instruction*, InstSet>;
+  DepMap RAWDeps;
+  DepMap WARDeps;
+
+  // Record any writes that may alias a load.
+  const auto Size = LocationSize::unknown();
+  for (auto Read : Loads) {
+    for (auto Write : Writes) {
+      if (Read == Write)
         continue;
 
-      if (AreSequentialAccesses<LoadInst>(Ld0, Ld1, *DL, *SE)) {
-        LoadPairs[Ld0] = Ld1;
+      MemoryLocation ReadLoc =
+        MemoryLocation(Read->getPointerOperand(), Size);
+
+      if (!isModOrRefSet(intersectModRef(AA->getModRefInfo(Write, ReadLoc),
+          ModRefInfo::ModRef)))
+        continue;
+      if (DT->dominates(Write, Read))
+        RAWDeps[Read].insert(Write);
+      else
+        WARDeps[Read].insert(Write);
+    }
+  }
+
+  // Check there's not a write inbetween the two loads that may alias both of
+  // them. If there's a write after the dominating load and before the dominated
+  // load, then we can't merge the loads.
+  auto SafeToPair = [&](LoadInst *Base, LoadInst *Offset) {
+    LoadInst *Dominator = DT->dominates(Base, Offset) ? Base : Offset;
+    LoadInst *Dominated = DT->dominates(Base, Offset) ? Offset : Base;
+
+    if (WARDeps.count(Dominator) && RAWDeps.count(Dominated)) {
+      InstSet &WARs = WARDeps[Dominator];
+      InstSet &RAWs = RAWDeps[Dominated];
+      for (auto WAR : WARs)
+        for (auto RAW : RAWs)
+          if (WAR == RAW)
+            return false;
+    }
+    return true;
+  };
+
+  // Record base, offset load pairs.
+  for (auto *Base : Loads) {
+    for (auto *Offset : Loads) {
+      if (Base == Offset)
+        continue;
+
+      if (AreSequentialAccesses<LoadInst>(Base, Offset, *DL, *SE) &&
+          SafeToPair(Base, Offset)) {
+        LoadPairs[Base] = Offset;
         break;
       }
     }
   }
 
   LLVM_DEBUG(if (!LoadPairs.empty()) {
                dbgs() << "Consecutive load pairs:\n";
                for (auto &MapIt : LoadPairs) {
bool ARMParallelDSP::InsertParallelMACs(Reduction &Reduction) {
   if (Acc != Reduction.Phi) {
     LLVM_DEBUG(dbgs() << "Replace Accumulate: "; Acc->dump());
     Reduction.AccIntAdd->replaceAllUsesWith(Acc);
     return true;
   }
   return false;
 }
 
-static void MatchReductions(Function &F, Loop *TheLoop, BasicBlock *Header,
-                            ReductionList &Reductions) {
-  RecurrenceDescriptor RecDesc;
-  const bool HasFnNoNaNAttr =
-    F.getFnAttribute("no-nans-fp-math").getValueAsString() == "true";
-  const BasicBlock *Latch = TheLoop->getLoopLatch();
-
-  for (PHINode &Phi : Header->phis()) {
-    const auto *Ty = Phi.getType();
-    if (!Ty->isIntegerTy(32) && !Ty->isIntegerTy(64))
-      continue;
-
-    const bool IsReduction =
-      RecurrenceDescriptor::AddReductionVar(&Phi,
-                                            RecurrenceDescriptor::RK_IntegerAdd,
-                                            TheLoop, HasFnNoNaNAttr, RecDesc);
-    if (!IsReduction)
-      continue;
-
-    Instruction *Acc = dyn_cast<Instruction>(Phi.getIncomingValueForBlock(Latch));
-    if (!Acc)
-      continue;
-
-    Reductions.push_back(Reduction(&Phi, Acc));
-  }
-
-  LLVM_DEBUG(
-    dbgs() << "\nAccumulating integer additions (reductions) found:\n";
-    for (auto &R : Reductions) {
-      dbgs() << "-  "; R.Phi->dump();
-      dbgs() << "-> "; R.AccIntAdd->dump();
-    }
-  );
-}
-
-static void AddMACCandidate(OpChainList &Candidates,
-                            Instruction *Mul,
-                            Value *MulOp0, Value *MulOp1) {
-  assert(Mul->getOpcode() == Instruction::Mul &&
-         "expected mul instruction");
-  ValueList LHS;
-  ValueList RHS;
-  if (IsNarrowSequence<16>(MulOp0, LHS) &&
-      IsNarrowSequence<16>(MulOp1, RHS)) {
-    Candidates.push_back(make_unique<BinOpChain>(Mul, LHS, RHS));
-  }
-}
-
 static void MatchParallelMACSequences(Reduction &R,
                                       OpChainList &Candidates) {
   Instruction *Acc = R.AccIntAdd;
   LLVM_DEBUG(dbgs() << "\n- Analysing:\t" << *Acc << "\n");
 
   // Returns false to signal the search should be stopped.
   std::function<bool(Value*)> Match =
     [&Candidates, &Match](Value *V) -> bool {
 
     auto *I = dyn_cast<Instruction>(V);
     if (!I)
       return false;
 
     switch (I->getOpcode()) {
     case Instruction::Add:
       if (Match(I->getOperand(0)) || (Match(I->getOperand(1))))
         return true;
       break;
     case Instruction::Mul: {
       Value *MulOp0 = I->getOperand(0);
       Value *MulOp1 = I->getOperand(1);
-      if (isa<SExtInst>(MulOp0) && isa<SExtInst>(MulOp1))
-        AddMACCandidate(Candidates, I, MulOp0, MulOp1);
+      if (isa<SExtInst>(MulOp0) && isa<SExtInst>(MulOp1)) {
+        ValueList LHS;
+        ValueList RHS;
+        if (IsNarrowSequence<16>(MulOp0, LHS) &&
+            IsNarrowSequence<16>(MulOp1, RHS)) {
+          Candidates.push_back(make_unique<BinOpChain>(I, LHS, RHS));
+        }
+      }
       return false;
     }
     case Instruction::SExt:
       return Match(I->getOperand(0));
     }
     return false;
   };
 
   while (Match (Acc));
   LLVM_DEBUG(dbgs() << "Finished matching MAC sequences, found "
              << Candidates.size() << " candidates.\n");
 }
 
-// Collects all instructions that are not part of the MAC chains, which is the
-// set of instructions that can potentially alias with the MAC operands.
-static void AliasCandidates(BasicBlock *Header, Instructions &Reads,
-                            Instructions &Writes) {
-  for (auto &I : *Header) {
-    if (I.mayReadFromMemory())
-      Reads.push_back(&I);
-    if (I.mayWriteToMemory())
-      Writes.push_back(&I);
-  }
-}
-
-// Check whether statements in the basic block that write to memory alias with
-// the memory locations accessed by the MAC-chains.
-// TODO: we need the read statements when we accept more complicated chains.
-static bool AreAliased(AliasAnalysis *AA, Instructions &Reads,
-                       Instructions &Writes, OpChainList &MACCandidates) {
-  LLVM_DEBUG(dbgs() << "Alias checks:\n");
-  for (auto &MAC : MACCandidates) {
-    LLVM_DEBUG(dbgs() << "mul: "; MAC->Root->dump());
-
-    // At the moment, we allow only simple chains that only consist of reads,
-    // accumulate their result with an integer add, and thus that don't write
-    // memory, and simply bail if they do.
-    if (!MAC->ReadOnly)
-      return true;
-
-    // Now for all writes in the basic block, check that they don't alias with
-    // the memory locations accessed by our MAC-chain:
-    for (auto *I : Writes) {
-      LLVM_DEBUG(dbgs() << "- "; I->dump());
-      assert(MAC->MemLocs.size() >= 2 && "expecting at least 2 memlocs");
-      for (auto &MemLoc : MAC->MemLocs) {
-        if (isModOrRefSet(intersectModRef(AA->getModRefInfo(I, MemLoc),
-                                          ModRefInfo::ModRef))) {
-          LLVM_DEBUG(dbgs() << "Yes, aliases found\n");
-          return true;
-        }
-      }
-    }
-  }
-
-  LLVM_DEBUG(dbgs() << "OK: no aliases found!\n");
-  return false;
-}
-
 static bool CheckMACMemory(OpChainList &Candidates) {
   for (auto &C : Candidates) {
     // A mul has 2 operands, and a narrow op consist of sext and a load; thus
     // we expect at least 4 items in this operand value list.
     if (C->size() < 4) {
       LLVM_DEBUG(dbgs() << "Operand list too short.\n");
       return false;
     }
-    C->SetMemoryLocations();
+    C->PopulateLoads();
     ValueList &LHS = static_cast<BinOpChain*>(C.get())->LHS;
     ValueList &RHS = static_cast<BinOpChain*>(C.get())->RHS;
 
     // Use +=2 to skip over the expected extend instructions.
     for (unsigned i = 0, e = LHS.size(); i < e; i += 2) {
       if (!isa<LoadInst>(LHS[i]) || !isa<LoadInst>(RHS[i]))
         return false;
     }

 //
 // If constants are used instead of loads, these will need to be hoisted
 // out and into a register.
 //
 // If loop invariants are used instead of loads, these need to be packed
 // before the loop begins.
 //
 bool ARMParallelDSP::MatchSMLAD(Function &F) {
-  BasicBlock *Header = L->getHeader();
-  LLVM_DEBUG(dbgs() << "= Matching SMLAD =\n";
-             dbgs() << "Header block:\n"; Header->dump();
-             dbgs() << "Loop info:\n\n"; L->dump());
 
-  bool Changed = false;
+  auto FindReductions = [&](ReductionList &Reductions) {
+    RecurrenceDescriptor RecDesc;
+    const bool HasFnNoNaNAttr =
+      F.getFnAttribute("no-nans-fp-math").getValueAsString() == "true";
+    BasicBlock *Latch = L->getLoopLatch();
+
+    for (PHINode &Phi : Latch->phis()) {
+      const auto *Ty = Phi.getType();
+      if (!Ty->isIntegerTy(32) && !Ty->isIntegerTy(64))
+        continue;
+
+      const bool IsReduction = RecurrenceDescriptor::AddReductionVar(
+        &Phi, RecurrenceDescriptor::RK_IntegerAdd, L, HasFnNoNaNAttr, RecDesc);
+
+      if (!IsReduction)
+        continue;
+
+      Instruction *Acc = dyn_cast<Instruction>(Phi.getIncomingValueForBlock(Latch));
+      if (!Acc)
+        continue;
+
+      Reductions.push_back(Reduction(&Phi, Acc));
+    }
+    return !Reductions.empty();
+  };
+
   ReductionList Reductions;
-  MatchReductions(F, L, Header, Reductions);
+  if (!FindReductions(Reductions))
+    return false;
 
   for (auto &R : Reductions) {
     OpChainList MACCandidates;
     MatchParallelMACSequences(R, MACCandidates);
     if (!CheckMACMemory(MACCandidates))
       continue;
 
     R.MACCandidates = std::move(MACCandidates);
 
     LLVM_DEBUG(dbgs() << "MAC candidates:\n";
       for (auto &M : R.MACCandidates)
         M->Root->dump();
       dbgs() << "\n";);
   }
 
-  // Collect all instructions that may read or write memory. Our alias
-  // analysis checks bail out if any of these instructions aliases with an
-  // instruction from the MAC-chain.
-  Instructions Reads, Writes;
-  AliasCandidates(Header, Reads, Writes);
-
+  bool Changed = false;
+  // Check whether statements in the basic block that write to memory alias
+  // with the memory locations accessed by the MAC-chains.
   for (auto &R : Reductions) {
-    if (AreAliased(AA, Reads, Writes, R.MACCandidates))
-      return false;
     CreateParallelMACPairs(R);
     Changed |= InsertParallelMACs(R);
   }
 
-  LLVM_DEBUG(if (Changed) dbgs() << "Header block:\n"; Header->dump(););
   return Changed;
 }
 
-LoadInst* ARMParallelDSP::CreateLoadIns(IRBuilder<NoFolder> &IRB,
-                                        SmallVectorImpl<LoadInst*> &Loads,
+LoadInst* ARMParallelDSP::CreateWideLoad(SmallVectorImpl<LoadInst*> &Loads,
                                          IntegerType *LoadTy) {
   assert(Loads.size() == 2 && "currently only support widening two loads");
- 
-  const unsigned AddrSpace = Loads[0]->getPointerAddressSpace();
-  Value *VecPtr = IRB.CreateBitCast(Loads[0]->getPointerOperand(),
-                                    LoadTy->getPointerTo(AddrSpace));
-  LoadInst *WideLoad = IRB.CreateAlignedLoad(LoadTy, VecPtr,
-                                             Loads[0]->getAlignment());
-  // Fix up users, Loads[0] needs trunc while Loads[1] needs a lshr and trunc.
-  Instruction *SExt0 = dyn_cast<SExtInst>(Loads[0]->user_back());
-  Instruction *SExt1 = dyn_cast<SExtInst>(Loads[1]->user_back());
+  LoadInst *Base = Loads[0];
+  LoadInst *Offset = Loads[1];
+
+  Instruction *BaseSExt = dyn_cast<SExtInst>(Base->user_back());
+  Instruction *OffsetSExt = dyn_cast<SExtInst>(Offset->user_back());
+
+  assert((Base->hasOneUse() && Offset->hasOneUse() && BaseSExt && OffsetSExt)
+         && "Loads should have a single, extending, user");
 
-  assert((Loads[0]->hasOneUse() && Loads[1]->hasOneUse() && SExt0 && SExt1) &&
-         "Loads should have a single, extending, user");
+  std::function<void(Value*, Value*)> MoveBefore =
+    [&](Value *A, Value *B) -> void {
+      if (!isa<Instruction>(A) || !isa<Instruction>(B))
+        return;
+
+      auto *Source = cast<Instruction>(A);
+      auto *Sink = cast<Instruction>(B);
 
-  std::function<void(Instruction*, Instruction*)> MoveAfter =
-    [&](Instruction* Source, Instruction* Sink) -> void {
       if (DT->dominates(Source, Sink) ||
           Source->getParent() != Sink->getParent() ||
           isa<PHINode>(Source) || isa<PHINode>(Sink))
         return;
 
-    Sink->moveAfter(Source);
-    for (auto &U : Sink->uses())
-      MoveAfter(Sink, cast<Instruction>(U.getUser()));
+      Source->moveBefore(Sink);
+      for (auto &U : Source->uses())
+        MoveBefore(Source, U.getUser());
     };
 
+  // Insert the load at the point of the original dominating load.
+  LoadInst *DomLoad = DT->dominates(Base, Offset) ? Base : Offset;
+  IRBuilder<NoFolder> IRB(DomLoad->getParent(),
+                          ++BasicBlock::iterator(DomLoad));
+
+  // Bitcast the pointer to a wider type and create the wide load, while making
+  // sure to maintain the original alignment as this prevents ldrd from being
+  // generated when it could be illegal due to memory alignment.
+  const unsigned AddrSpace = DomLoad->getPointerAddressSpace();
+  Value *VecPtr = IRB.CreateBitCast(Base->getPointerOperand(),
+                                    LoadTy->getPointerTo(AddrSpace));
+  LoadInst *WideLoad = IRB.CreateAlignedLoad(LoadTy, VecPtr,
+                                             Base->getAlignment());
+
+  // Make sure everything is in the correct order in the basic block.
+  MoveBefore(Base->getPointerOperand(), VecPtr);
+  MoveBefore(VecPtr, WideLoad);
+
   // From the wide load, create two values that equal the original two loads.
-  Value *Bottom = IRB.CreateTrunc(WideLoad, Loads[0]->getType());
-  SExt0->setOperand(0, Bottom);
-  if (auto *I = dyn_cast<Instruction>(Bottom)) {
-    I->moveAfter(WideLoad);
-    MoveAfter(I, SExt0);
-  }
+  // Loads[0] needs trunc while Loads[1] needs a lshr and trunc.
+  // TODO: Support big-endian as well.
+  Value *Bottom = IRB.CreateTrunc(WideLoad, Base->getType());
+  BaseSExt->setOperand(0, Bottom);
 
-  IntegerType *Ld1Ty = cast<IntegerType>(Loads[1]->getType());
-  Value *ShiftVal = ConstantInt::get(LoadTy, Ld1Ty->getBitWidth());
+  IntegerType *OffsetTy = cast<IntegerType>(Offset->getType());
+  Value *ShiftVal = ConstantInt::get(LoadTy, OffsetTy->getBitWidth());
   Value *Top = IRB.CreateLShr(WideLoad, ShiftVal);
-  if (auto *I = dyn_cast<Instruction>(Top))
-    MoveAfter(WideLoad, I);
-
-  Value *Trunc = IRB.CreateTrunc(Top, Ld1Ty);
-  SExt1->setOperand(0, Trunc);
-  if (auto *I = dyn_cast<Instruction>(Trunc))
-    MoveAfter(I, SExt1);
+  Value *Trunc = IRB.CreateTrunc(Top, OffsetTy);
+  OffsetSExt->setOperand(0, Trunc);
 
-  WideLoads.emplace(std::make_pair(Loads[0],
+  WideLoads.emplace(std::make_pair(Base,
                                    make_unique<WidenedLoad>(Loads, WideLoad)));
   return WideLoad;
 }
 
 Instruction *ARMParallelDSP::CreateSMLADCall(SmallVectorImpl<LoadInst*> &VecLd0,
                                              SmallVectorImpl<LoadInst*> &VecLd1,
                                              Instruction *Acc, bool Exchange,
                                              Instruction *InsertAfter) {
   LLVM_DEBUG(dbgs() << "Create SMLAD intrinsic using:\n"
              << "- " << *VecLd0[0] << "\n"
              << "- " << *VecLd0[1] << "\n"
              << "- " << *VecLd1[0] << "\n"
              << "- " << *VecLd1[1] << "\n"
              << "- " << *Acc << "\n"
              << "- Exchange: " << Exchange << "\n");
 
-  IRBuilder<NoFolder> Builder(InsertAfter->getParent(),
-                              ++BasicBlock::iterator(InsertAfter));
-
   // Replace the reduction chain with an intrinsic call
   IntegerType *Ty = IntegerType::get(M->getContext(), 32);
   LoadInst *WideLd0 = WideLoads.count(VecLd0[0]) ?
-    WideLoads[VecLd0[0]]->getLoad() : CreateLoadIns(Builder, VecLd0, Ty);
+    WideLoads[VecLd0[0]]->getLoad() : CreateWideLoad(VecLd0, Ty);
   LoadInst *WideLd1 = WideLoads.count(VecLd1[0]) ?
-    WideLoads[VecLd1[0]]->getLoad() : CreateLoadIns(Builder, VecLd1, Ty);
+    WideLoads[VecLd1[0]]->getLoad() : CreateWideLoad(VecLd1, Ty);
+
   Value* Args[] = { WideLd0, WideLd1, Acc };
   Function *SMLAD = nullptr;
   if (Exchange)
     SMLAD = Acc->getType()->isIntegerTy(32) ?
       Intrinsic::getDeclaration(M, Intrinsic::arm_smladx) :
       Intrinsic::getDeclaration(M, Intrinsic::arm_smlaldx);
   else
     SMLAD = Acc->getType()->isIntegerTy(32) ?
       Intrinsic::getDeclaration(M, Intrinsic::arm_smlad) :
       Intrinsic::getDeclaration(M, Intrinsic::arm_smlald);
+
+  IRBuilder<NoFolder> Builder(InsertAfter->getParent(),
+                              ++BasicBlock::iterator(InsertAfter));
   CallInst *Call = Builder.CreateCall(SMLAD, Args);
   NumSMLAD++;
   return Call;
 }
 
 // Compare the value lists in Other to this chain.
 bool BinOpChain::AreSymmetrical(BinOpChain *Other) {
   // Element-by-element comparison of Value lists returning true if they are

test/CodeGen/ARM/ParallelDSP/aliasing.ll

+; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s
+;
+; Alias check: check that the rewrite isn't triggered when there's a store
+; instruction possibly aliasing any mul load operands; arguments are passed
+; without 'restrict' enabled.
+;
+; CHECK-NOT:  call i32 @llvm.arm.smlad
+;
+define dso_local i32 @no_restrict(i32 %arg, i32* nocapture %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
+entry:
+  %cmp24 = icmp sgt i32 %arg, 0
+  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %.pre = load i16, i16* %arg3, align 2
+  %.pre27 = load i16, i16* %arg2, align 2
+  br label %for.body
+
+for.cond.cleanup:
+  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
+  ret i32 %mac1.0.lcssa
+
+for.body:
+  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
+  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
+  %0 = load i16, i16* %arrayidx, align 2
+
+; Store inserted here, aliasing with arrayidx, arrayidx1, arrayidx3
+  store i16 42, i16* %arrayidx, align 2
+
+  %add = add nuw nsw i32 %i.025, 1
+  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
+  %1 = load i16, i16* %arrayidx1, align 2
+  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
+  %2 = load i16, i16* %arrayidx3, align 2
+  %conv = sext i16 %2 to i32
+  %conv4 = sext i16 %0 to i32
+  %mul = mul nsw i32 %conv, %conv4
+  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
+  %3 = load i16, i16* %arrayidx6, align 2
+  %conv7 = sext i16 %3 to i32
+  %conv8 = sext i16 %1 to i32
+  %mul9 = mul nsw i32 %conv7, %conv8
+  %add10 = add i32 %mul, %mac1.026
+  %add11 = add i32 %mul9, %add10
+  %exitcond = icmp ne i32 %add, %arg
+  br i1 %exitcond, label %for.body, label %for.cond.cleanup
+}
+
+; Alias check: check that the rewrite isn't triggered when there's a store
+; aliasing one of the mul load operands. Arguments are now annotated with
+; 'noalias'.
+;
+; CHECK-NOT:  call i32 @llvm.arm.smlad
+;
+define dso_local i32 @restrict(i32 %arg, i32* noalias %arg1, i16* noalias readonly %arg2, i16* noalias readonly %arg3) {
+entry:
+  %cmp24 = icmp sgt i32 %arg, 0
+  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %.pre = load i16, i16* %arg3, align 2
+  %.pre27 = load i16, i16* %arg2, align 2
+  br label %for.body
+
+for.cond.cleanup:
+  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
+  ret i32 %mac1.0.lcssa
+
+for.body:
+  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
+  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
+  %0 = load i16, i16* %arrayidx, align 2
+
+; Store inserted here, aliasing only with loads from 'arrayidx'.
+  store i16 42, i16* %arrayidx, align 2
+
+  %add = add nuw nsw i32 %i.025, 1
+  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
+  %1 = load i16, i16* %arrayidx1, align 2
+  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
+  %2 = load i16, i16* %arrayidx3, align 2
+  %conv = sext i16 %2 to i32
+  %conv4 = sext i16 %0 to i32
+  %mul = mul nsw i32 %conv, %conv4
+  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
+  %3 = load i16, i16* %arrayidx6, align 2
+  %conv7 = sext i16 %3 to i32
+  %conv8 = sext i16 %1 to i32
+  %mul9 = mul nsw i32 %conv7, %conv8
+  %add10 = add i32 %mul, %mac1.026
+
+; Here the Mul is the LHS, and the Add the RHS.
+  %add11 = add i32 %mul9, %add10
+
+  %exitcond = icmp ne i32 %add, %arg
+  br i1 %exitcond, label %for.body, label %for.cond.cleanup
+}
+
+; CHECK-LABEL: store_dominates_all
+; CHECK: store
+; CHECK: load
+; CHECK: load
+; CHECK: load
+; CHECK: load
+; CHECK: smlad
+define dso_local i32 @store_dominates_all(i32 %arg, i32* nocapture %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
+entry:
+  %cmp24 = icmp sgt i32 %arg, 0
+  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %.pre = load i16, i16* %arg3, align 2
+  %.pre27 = load i16, i16* %arg2, align 2
+  br label %for.body
+
+for.cond.cleanup:
+  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
+  ret i32 %mac1.0.lcssa
+
+for.body:
+  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
+  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
+  store i16 42, i16* %arrayidx, align 2
+  %0 = load i16, i16* %arrayidx, align 2
+  %add = add nuw nsw i32 %i.025, 1
+  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
+  %1 = load i16, i16* %arrayidx1, align 2
+  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
+  %2 = load i16, i16* %arrayidx3, align 2
+  %conv = sext i16 %2 to i32
+  %conv4 = sext i16 %0 to i32
+  %mul = mul nsw i32 %conv, %conv4
+  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
+  %3 = load i16, i16* %arrayidx6, align 2
+  %conv7 = sext i16 %3 to i32
+  %conv8 = sext i16 %1 to i32
+  %mul9 = mul nsw i32 %conv7, %conv8
+  %add10 = add i32 %mul, %mac1.026
+  %add11 = add i32 %mul9, %add10
+  %exitcond = icmp ne i32 %add, %arg
+  br i1 %exitcond, label %for.body, label %for.cond.cleanup
+}
+
+; CHECK-LABEL: loads_dominate
+; CHECK-NOT: store 
+; CHECK: load i32
+; CHECK-NOT: store
+; CHECK: load i32
+; CHECK: store
+define dso_local i32 @loads_dominate(i32 %arg, i32* nocapture %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
+entry:
+  %cmp24 = icmp sgt i32 %arg, 0
+  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %.pre = load i16, i16* %arg3, align 2
+  %.pre27 = load i16, i16* %arg2, align 2
+  br label %for.body
+
+for.cond.cleanup:
+  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
+  ret i32 %mac1.0.lcssa
+
+for.body:
+  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
+  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
+  %0 = load i16, i16* %arrayidx, align 2
+  %add = add nuw nsw i32 %i.025, 1
+  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
+  %1 = load i16, i16* %arrayidx1, align 2
+  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
+  %2 = load i16, i16* %arrayidx3, align 2
+  %conv = sext i16 %2 to i32
+  %conv4 = sext i16 %0 to i32
+  %mul = mul nsw i32 %conv, %conv4
+  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
+  %3 = load i16, i16* %arrayidx6, align 2
+  %conv7 = sext i16 %3 to i32
+  %conv8 = sext i16 %1 to i32
+  %mul9 = mul nsw i32 %conv7, %conv8
+  %add10 = add i32 %mul, %mac1.026
+  %add11 = add i32 %mul9, %add10
+  store i16 42, i16* %arrayidx, align 2
+  %exitcond = icmp ne i32 %add, %arg
+  br i1 %exitcond, label %for.body, label %for.cond.cleanup
+}
+
+; CHECK-LABEL: store_alias_arg3_legal_1
+; CHECK-NOT: store
+; CHECK: phi i32
+; CHECK: [[IV:%[^ ]+]] = phi i32 [ %add
+; CHECK: [[ARG3_GEP:%[^ ]+]] = getelementptr inbounds i16, i16* %arg3, i32 [[IV]]
+; CHECK: [[ARG3:%[^ ]+]] = bitcast i16* [[ARG3_GEP]] to i32*
+; CHECK: load i32, i32* [[ARG3]]
+; CHECK: [[ARG2_GEP:%[^ ]+]] = getelementptr inbounds i16, i16* %arg2, i32 [[IV]]
+; CHECK: [[ARG2:%[^ ]+]] = bitcast i16* [[ARG2_GEP]] to i32*
+; CHECK: load i32, i32* [[ARG2]]
+; CHECK: store
+define dso_local i32 @store_alias_arg3_legal_1(i32 %arg, i32* nocapture %arg1, i16* noalias nocapture readonly %arg2, i16* nocapture readonly %arg3) {
+entry:
+  %cmp24 = icmp sgt i32 %arg, 0
+  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %.pre = load i16, i16* %arg3, align 2
+  %.pre27 = load i16, i16* %arg2, align 2
+  br label %for.body
+
+for.cond.cleanup:
+  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
+  ret i32 %mac1.0.lcssa
+
+for.body:
+  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
+  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
+  %0 = load i16, i16* %arrayidx, align 2
+  %add = add nuw nsw i32 %i.025, 1
+  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
+  %1 = load i16, i16* %arrayidx1, align 2
+  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
+  %2 = load i16, i16* %arrayidx3, align 2
+  %conv = sext i16 %2 to i32
+  %conv4 = sext i16 %0 to i32
+  %mul = mul nsw i32 %conv, %conv4
+  store i16 42, i16* %arrayidx, align 2
+  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
+  %3 = load i16, i16* %arrayidx6, align 2
+  %conv7 = sext i16 %3 to i32
+  %conv8 = sext i16 %1 to i32
+  %mul9 = mul nsw i32 %conv7, %conv8
+  %add10 = add i32 %mul, %mac1.026
+  %add11 = add i32 %mul9, %add10
+  %exitcond = icmp ne i32 %add, %arg
+  br i1 %exitcond, label %for.body, label %for.cond.cleanup
+}
+
+; CHECK-LABEL: store_alias_arg3_legal_2
+; CHECK-NOT: store
+; CHECK: [[BITCAST:[^ ]+]] = bitcast i16* %arrayidx to i32*
+; CHECK: load i32, i32* [[BITCAST]]
+; CHECK: store i16 42, i16* %arrayidx
+; CHECK: [[BITCAST3:[^ ]+]] = bitcast i16* %arrayidx3 to i32*
+; CHECK: load i32, i32* [[BITCAST3]]
+; CHECK: smlad
+define dso_local i32 @store_alias_arg3_legal_2(i32 %arg, i32* nocapture %arg1, i16* noalias nocapture readonly %arg2, i16* nocapture readonly %arg3) {
+entry:
+  %cmp24 = icmp sgt i32 %arg, 0
+  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %.pre = load i16, i16* %arg3, align 2
+  %.pre27 = load i16, i16* %arg2, align 2
+  br label %for.body
+
+for.cond.cleanup:
+  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
+  ret i32 %mac1.0.lcssa
+
+for.body:
+  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
+  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
+  %0 = load i16, i16* %arrayidx, align 2
+  %add = add nuw nsw i32 %i.025, 1
+  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
+  %1 = load i16, i16* %arrayidx1, align 2
+  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
+  store i16 42, i16* %arrayidx, align 2
+  %2 = load i16, i16* %arrayidx3, align 2
+  %conv = sext i16 %2 to i32
+  %conv4 = sext i16 %0 to i32
+  %mul = mul nsw i32 %conv, %conv4
+  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
+  %3 = load i16, i16* %arrayidx6, align 2
+  %conv7 = sext i16 %3 to i32
+  %conv8 = sext i16 %1 to i32
+  %mul9 = mul nsw i32 %conv7, %conv8
+  %add10 = add i32 %mul, %mac1.026
+  %add11 = add i32 %mul9, %add10
+  %exitcond = icmp ne i32 %add, %arg
+  br i1 %exitcond, label %for.body, label %for.cond.cleanup
+}
+
+; CHECK-LABEL: store_alias_arg3_illegal_1
+; CHECK-NOT: load i32
+define dso_local i32 @store_alias_arg3_illegal_1(i32 %arg, i32* nocapture %arg1, i16* noalias nocapture readonly %arg2, i16* noalias nocapture readonly %arg3) {
+entry:
+  %cmp24 = icmp sgt i32 %arg, 0
+  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %.pre = load i16, i16* %arg3, align 2
+  %.pre27 = load i16, i16* %arg2, align 2
+  br label %for.body
+
+for.cond.cleanup:
+  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
+  ret i32 %mac1.0.lcssa
+
+for.body:
+  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
+  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
+  %0 = load i16, i16* %arrayidx, align 2
+  %add = add nuw nsw i32 %i.025, 1
+  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
+  store i16 42, i16* %arrayidx1, align 2
+  %1 = load i16, i16* %arrayidx1, align 2
+  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
+  %2 = load i16, i16* %arrayidx3, align 2
+  %conv = sext i16 %2 to i32
+  %conv4 = sext i16 %0 to i32
+  %mul = mul nsw i32 %conv, %conv4
+  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
+  %3 = load i16, i16* %arrayidx6, align 2
+  %conv7 = sext i16 %3 to i32
+  %conv8 = sext i16 %1 to i32
+  %mul9 = mul nsw i32 %conv7, %conv8
+  %add10 = add i32 %mul, %mac1.026
+  %add11 = add i32 %mul9, %add10
+  %exitcond = icmp ne i32 %add, %arg
+  br i1 %exitcond, label %for.body, label %for.cond.cleanup
+}
+
+; CHECK-LABEL: store_alias_arg3_illegal_2
+; CHECK-NOT: load i32
+define dso_local i32 @store_alias_arg3_illegal_2(i32 %arg, i32* nocapture %arg1, i16* noalias nocapture readonly %arg2, i16* noalias nocapture readonly %arg3) {
+entry:
+  %cmp24 = icmp sgt i32 %arg, 0
+  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %.pre = load i16, i16* %arg3, align 2
+  %.pre27 = load i16, i16* %arg2, align 2
+  br label %for.body
+
+for.cond.cleanup:
+  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
+  ret i32 %mac1.0.lcssa
+
+for.body:
+  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
+  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
+  %0 = load i16, i16* %arrayidx, align 2
+  %add = add nuw nsw i32 %i.025, 1
+  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
+  store i16 42, i16* %arrayidx, align 2
+  %1 = load i16, i16* %arrayidx1, align 2
+  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
+  %2 = load i16, i16* %arrayidx3, align 2
+  %conv = sext i16 %2 to i32
+  %conv4 = sext i16 %0 to i32
+  %mul = mul nsw i32 %conv, %conv4
+  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
+  %3 = load i16, i16* %arrayidx6, align 2
+  %conv7 = sext i16 %3 to i32
+  %conv8 = sext i16 %1 to i32
+  %mul9 = mul nsw i32 %conv7, %conv8
+  %add10 = add i32 %mul, %mac1.026
+  %add11 = add i32 %mul9, %add10
+  %exitcond = icmp ne i32 %add, %arg
+  br i1 %exitcond, label %for.body, label %for.cond.cleanup
+}
+
+; CHECK-LABEL: store_alias_arg2_illegal_1
+; CHECK-NOT: load i32
+define dso_local i32 @store_alias_arg2_illegal_1(i32 %arg, i32* nocapture %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
+entry:
+  %cmp24 = icmp sgt i32 %arg, 0
+  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %.pre = load i16, i16* %arg3, align 2
+  %.pre27 = load i16, i16* %arg2, align 2
+  br label %for.body
+
+for.cond.cleanup:
+  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
+  ret i32 %mac1.0.lcssa
+
+for.body:
+  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
+  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
+  %0 = load i16, i16* %arrayidx, align 2
+  %add = add nuw nsw i32 %i.025, 1
+  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
+  %1 = load i16, i16* %arrayidx1, align 2
+  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
+  %2 = load i16, i16* %arrayidx3, align 2
+  %conv = sext i16 %2 to i32
+  %conv4 = sext i16 %0 to i32
+  %mul = mul nsw i32 %conv, %conv4
+  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
+  store i16 42, i16* %arrayidx6, align 2
+  %3 = load i16, i16* %arrayidx6, align 2
+  %conv7 = sext i16 %3 to i32
+  %conv8 = sext i16 %1 to i32
+  %mul9 = mul nsw i32 %conv7, %conv8
+  %add10 = add i32 %mul, %mac1.026
+  %add11 = add i32 %mul9, %add10
+  %exitcond = icmp ne i32 %add, %arg
+  br i1 %exitcond, label %for.body, label %for.cond.cleanup
+}
+
+; CHECK-LABEL: store_alias_arg2_illegal_2
+; CHECK-NOT: load i32
+define dso_local i32 @store_alias_arg2_illegal_2(i32 %arg, i32* nocapture %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
+entry:
+  %cmp24 = icmp sgt i32 %arg, 0
+  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
+
+for.body.preheader:
+  %.pre = load i16, i16* %arg3, align 2
+  %.pre27 = load i16, i16* %arg2, align 2
+  br label %for.body
+
+for.cond.cleanup:
+  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
+  ret i32 %mac1.0.lcssa
+
+for.body:
+  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
+  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
+  %0 = load i16, i16* %arrayidx, align 2
+  %add = add nuw nsw i32 %i.025, 1
+  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
+  %1 = load i16, i16* %arrayidx1, align 2
+  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
+  %2 = load i16, i16* %arrayidx3, align 2
+  %conv = sext i16 %2 to i32
+  %conv4 = sext i16 %0 to i32
+  %mul = mul nsw i32 %conv, %conv4
+  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
+  store i16 42, i16* %arrayidx3, align 2
+  %3 = load i16, i16* %arrayidx6, align 2
+  %conv7 = sext i16 %3 to i32
+  %conv8 = sext i16 %1 to i32
+  %mul9 = mul nsw i32 %conv7, %conv8
+  %add10 = add i32 %mul, %mac1.026
+  %add11 = add i32 %mul9, %add10
+  %exitcond = icmp ne i32 %add, %arg
+  br i1 %exitcond, label %for.body, label %for.cond.cleanup
+}

test/CodeGen/ARM/ParallelDSP/smlad0.ll

 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s
 ; RUN: opt -mtriple=armeb-arm-eabi -mcpu=cortex-m0 < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 ;
 ; The Cortex-M0 does not support unaligned accesses:
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m0 < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 ;
 ; Check DSP extension:
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 -mattr=-dsp < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 
 define dso_local i32 @OneReduction(i32 %arg, i32* nocapture readnone %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
 ;
 ; CHECK-LABEL: @OneReduction
 ; CHECK:  %mac1{{\.}}026 = phi i32 [ [[V8:%[0-9]+]], %for.body ], [ 0, %for.body.preheader ]
-; CHECK:  [[V4:%[0-9]+]] = bitcast i16* %arrayidx3 to i32*
+; CHECK:  [[V4:%[0-9]+]] = bitcast i16* %arrayidx to i32*
 ; CHECK:  [[V5:%[0-9]+]] = load i32, i32* [[V4]], align 2
-; CHECK:  [[V6:%[0-9]+]] = bitcast i16* %arrayidx to i32*
+; CHECK:  [[V6:%[0-9]+]] = bitcast i16* %arrayidx3 to i32*
 ; CHECK:  [[V7:%[0-9]+]] = load i32, i32* [[V6]], align 2
-; CHECK:  [[V8]] = call i32 @llvm.arm.smlad(i32 [[V5]], i32 [[V7]], i32 %mac1{{\.}}026)
+; CHECK:  [[V8]] = call i32 @llvm.arm.smlad(i32 [[V7]], i32 [[V5]], i32 %mac1{{\.}}026)
 ; CHECK-NOT: call i32 @llvm.arm.smlad
 ;
 ; CHECK-UNSUPPORTED-NOT:  call i32 @llvm.arm.smlad
 ;
 entry:
   %cmp24 = icmp sgt i32 %arg, 0
   br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
 

test/CodeGen/ARM/ParallelDSP/smlad1.ll

 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s
 
 ; CHECK-LABEL: @test1
 ; CHECK:  %mac1{{\.}}026 = phi i32 [ [[V8:%[0-9]+]], %for.body ], [ 0, %for.body.preheader ]
-; CHECK:  [[V4:%[0-9]+]] = bitcast i16* %arrayidx3 to i32*
+; CHECK:  [[V4:%[0-9]+]] = bitcast i16* %arrayidx to i32*
 ; CHECK:  [[V5:%[0-9]+]] = load i32, i32* [[V4]], align 2
-; CHECK:  [[V6:%[0-9]+]] = bitcast i16* %arrayidx to i32*
+; CHECK:  [[V6:%[0-9]+]] = bitcast i16* %arrayidx3 to i32*
 ; CHECK:  [[V7:%[0-9]+]] = load i32, i32* [[V6]], align 2
-; CHECK:  [[V8]] = call i32 @llvm.arm.smlad(i32 [[V5]], i32 [[V7]], i32 %mac1{{\.}}026)
+; CHECK:  [[V8]] = call i32 @llvm.arm.smlad(i32 [[V7]], i32 [[V5]], i32 %mac1{{\.}}026)
 
 define dso_local i32 @test1(i32 %arg, i32* nocapture readnone %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
 entry:
   %cmp24 = icmp sgt i32 %arg, 0
   br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
 
 for.body.preheader:
   %.pre = load i16, i16* %arg3, align 2

test/CodeGen/ARM/ParallelDSP/smlad11.ll

 ; REQUIRES: asserts
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S -stats 2>&1 | FileCheck %s
 ;
 ; A more complicated chain: 4 mul operations, so we expect 2 smlad calls.
 ;
 ; CHECK:  %mac1{{\.}}054 = phi i32 [ [[V17:%[0-9]+]], %for.body ], [ 0, %for.body.preheader ]
-; CHECK:  [[V8:%[0-9]+]] = bitcast i16* %arrayidx8 to i32*
-; CHECK:  [[V9:%[0-9]+]] = load i32, i32* [[V8]], align 2
 ; CHECK:  [[V10:%[0-9]+]] = bitcast i16* %arrayidx to i32*
 ; CHECK:  [[V11:%[0-9]+]] = load i32, i32* [[V10]], align 2
-; CHECK:  [[V12:%[0-9]+]] = call i32 @llvm.arm.smlad(i32 [[V9]], i32 [[V11]], i32 %mac1{{\.}}054)
-; CHECK:  [[V13:%[0-9]+]] = bitcast i16* %arrayidx17 to i32*
-; CHECK:  [[V14:%[0-9]+]] = load i32, i32* [[V13]], align 2
 ; CHECK:  [[V15:%[0-9]+]] = bitcast i16* %arrayidx4 to i32*
 ; CHECK:  [[V16:%[0-9]+]] = load i32, i32* [[V15]], align 2
+; CHECK:  [[V8:%[0-9]+]] = bitcast i16* %arrayidx8 to i32*
+; CHECK:  [[V9:%[0-9]+]] = load i32, i32* [[V8]], align 2
+; CHECK:  [[V13:%[0-9]+]] = bitcast i16* %arrayidx17 to i32*
+; CHECK:  [[V14:%[0-9]+]] = load i32, i32* [[V13]], align 2
+; CHECK:  [[V12:%[0-9]+]] = call i32 @llvm.arm.smlad(i32 [[V9]], i32 [[V11]], i32 %mac1{{\.}}054)
 ; CHECK:  [[V17:%[0-9]+]] = call i32 @llvm.arm.smlad(i32 [[V14]], i32 [[V16]], i32 [[V12]])
 ;
 ; And we don't want to see a 3rd smlad:
 ; CHECK-NOT: call i32 @llvm.arm.smlad
 ;
 ; CHECK:  2 arm-parallel-dsp - Number of smlad instructions generated
 ;
 define dso_local i32 @test(i32 %arg, i32* nocapture readnone %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {

test/CodeGen/ARM/ParallelDSP/smlad6.ll

-; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s
-;
-; Alias check: check that the rewrite isn't triggered when there's a store
-; instruction possibly aliasing any mul load operands; arguments are passed
-; without 'restrict' enabled.
-;
-; CHECK-NOT:  call i32 @llvm.arm.smlad
-;
-define dso_local i32 @test(i32 %arg, i32* nocapture %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
-entry:
-  %cmp24 = icmp sgt i32 %arg, 0
-  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
-
-for.body.preheader:
-  %.pre = load i16, i16* %arg3, align 2
-  %.pre27 = load i16, i16* %arg2, align 2
-  br label %for.body
-
-for.cond.cleanup:
-  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
-  ret i32 %mac1.0.lcssa
-
-for.body:
-  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
-  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
-  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
-  %0 = load i16, i16* %arrayidx, align 2
-
-; Store inserted here, aliasing with arrayidx, arrayidx1, arrayidx3
-  store i16 42, i16* %arrayidx, align 2
-
-  %add = add nuw nsw i32 %i.025, 1
-  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
-  %1 = load i16, i16* %arrayidx1, align 2
-  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
-  %2 = load i16, i16* %arrayidx3, align 2
-  %conv = sext i16 %2 to i32
-  %conv4 = sext i16 %0 to i32
-  %mul = mul nsw i32 %conv, %conv4
-  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
-  %3 = load i16, i16* %arrayidx6, align 2
-  %conv7 = sext i16 %3 to i32
-  %conv8 = sext i16 %1 to i32
-  %mul9 = mul nsw i32 %conv7, %conv8
-  %add10 = add i32 %mul, %mac1.026
-  %add11 = add i32 %mul9, %add10
-  %exitcond = icmp ne i32 %add, %arg
-  br i1 %exitcond, label %for.body, label %for.cond.cleanup
-}
-

test/CodeGen/ARM/ParallelDSP/smlad7.ll

-; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s
-;
-; Alias check: check that the rewrite isn't triggered when there's a store
-; aliasing one of the mul load operands. Arguments are now annotated with
-; 'noalias'.
-;
-; CHECK-NOT:  call i32 @llvm.arm.smlad
-;
-define dso_local i32 @test(i32 %arg, i32* noalias %arg1, i16* noalias readonly %arg2, i16* noalias readonly %arg3) {
-entry:
-  %cmp24 = icmp sgt i32 %arg, 0
-  br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
-
-for.body.preheader:
-  %.pre = load i16, i16* %arg3, align 2
-  %.pre27 = load i16, i16* %arg2, align 2
-  br label %for.body
-
-for.cond.cleanup:
-  %mac1.0.lcssa = phi i32 [ 0, %entry ], [ %add11, %for.body ]
-  ret i32 %mac1.0.lcssa
-
-for.body:
-  %mac1.026 = phi i32 [ %add11, %for.body ], [ 0, %for.body.preheader ]
-  %i.025 = phi i32 [ %add, %for.body ], [ 0, %for.body.preheader ]
-  %arrayidx = getelementptr inbounds i16, i16* %arg3, i32 %i.025
-  %0 = load i16, i16* %arrayidx, align 2
-
-; Store inserted here, aliasing only with loads from 'arrayidx'.
-  store i16 42, i16* %arrayidx, align 2
-
-  %add = add nuw nsw i32 %i.025, 1
-  %arrayidx1 = getelementptr inbounds i16, i16* %arg3, i32 %add
-  %1 = load i16, i16* %arrayidx1, align 2
-  %arrayidx3 = getelementptr inbounds i16, i16* %arg2, i32 %i.025
-  %2 = load i16, i16* %arrayidx3, align 2
-  %conv = sext i16 %2 to i32
-  %conv4 = sext i16 %0 to i32
-  %mul = mul nsw i32 %conv, %conv4
-  %arrayidx6 = getelementptr inbounds i16, i16* %arg2, i32 %add
-  %3 = load i16, i16* %arrayidx6, align 2
-  %conv7 = sext i16 %3 to i32
-  %conv8 = sext i16 %1 to i32
-  %mul9 = mul nsw i32 %conv7, %conv8
-  %add10 = add i32 %mul, %mac1.026
-
-; Here the Mul is the LHS, and the Add the RHS.
-  %add11 = add i32 %mul9, %add10
-
-  %exitcond = icmp ne i32 %add, %arg
-  br i1 %exitcond, label %for.body, label %for.cond.cleanup
-}
-

test/CodeGen/ARM/ParallelDSP/smladx-1.ll

 ; RUN: opt -mtriple=thumbv8m.main -mcpu=cortex-m33 -arm-parallel-dsp %s -S -o - | FileCheck %s
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m0 < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 -mattr=-dsp < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 ; RUN: opt -mtriple=armeb-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 
 define i32 @smladx(i16* nocapture readonly %pIn1, i16* nocapture readonly %pIn2, i32 %j, i32 %limit) {
 
 ; CHECK-LABEL: smladx
 ; CHECK: = phi i32 [ 0, %for.body.preheader.new ],
 ; CHECK: [[ACC0:%[^ ]+]] = phi i32 [ 0, %for.body.preheader.new ], [ [[ACC2:%[^ ]+]], %for.body ]
+; CHECK: [[PIN21:%[^ ]+]] = bitcast i16* %pIn2.1 to i32*
+; CHECK: [[IN21:%[^ ]+]] = load i32, i32* [[PIN21]], align 2
+; CHECK: [[PIN10:%[^ ]+]] = bitcast i16* %pIn1.0 to i32*
+; CHECK: [[IN10:%[^ ]+]] = load i32, i32* [[PIN10]], align 2
 ; CHECK: [[PIN23:%[^ ]+]] = bitcast i16* %pIn2.3 to i32*
 ; CHECK: [[IN23:%[^ ]+]] = load i32, i32* [[PIN23]], align 2
 ; CHECK: [[PIN12:%[^ ]+]] = bitcast i16* %pIn1.2 to i32*
 ; CHECK: [[IN12:%[^ ]+]] = load i32, i32* [[PIN12]], align 2
 ; CHECK: [[ACC1:%[^ ]+]] = call i32 @llvm.arm.smladx(i32 [[IN23]], i32 [[IN12]], i32 [[ACC0]])
-; CHECK: [[PIN21:%[^ ]+]] = bitcast i16* %pIn2.1 to i32*
-; CHECK: [[IN21:%[^ ]+]] = load i32, i32* [[PIN21]], align 2
-; CHECK: [[PIN10:%[^ ]+]] = bitcast i16* %pIn1.0 to i32*
-; CHECK: [[IN10:%[^ ]+]] = load i32, i32* [[PIN10]], align 2
 ; CHECK: [[ACC2]] = call i32 @llvm.arm.smladx(i32 [[IN21]], i32 [[IN10]], i32 [[ACC1]])
 ; CHECK-NOT: call i32 @llvm.arm.smlad
 ; CHECK-UNSUPPORTED-NOT:  call i32 @llvm.arm.smlad
 
 entry:
   %cmp9 = icmp eq i32 %limit, 0
   br i1 %cmp9, label %for.cond.cleanup, label %for.body.preheader
 

 ; CHECK: [[PIN2Base:[^ ]+]] = getelementptr i16, i16* %pIn2
 
 ; CHECK: for.body:
 ; CHECK: [[PIN2:%[^ ]+]] = phi i16* [ [[PIN2_NEXT:%[^ ]+]], %for.body ], [ [[PIN2Base]], %for.body.preheader.new ]
 ; CHECK: [[PIN1:%[^ ]+]] = phi i16* [ [[PIN1_NEXT:%[^ ]+]], %for.body ], [ [[PIN1Base]], %for.body.preheader.new ]
 ; CHECK: [[IV:%[^ ]+]] = phi i32
 ; CHECK: [[ACC0:%[^ ]+]] = phi i32 [ 0, %for.body.preheader.new ], [ [[ACC2:%[^ ]+]], %for.body ]
 
-; CHECK: [[PIN1_2:%[^ ]+]] = getelementptr i16, i16* [[PIN1]], i32 -2
-; CHECK: [[PIN2_2:%[^ ]+]] = getelementptr i16, i16* [[PIN2]], i32 -2
+; CHECK: [[PIN2_CAST:%[^ ]+]] = bitcast i16* [[PIN2]] to i32*
+; CHECK: [[IN2:%[^ ]+]] = load i32, i32* [[PIN2_CAST]], align 2
 
+; CHECK: [[PIN1_2:%[^ ]+]] = getelementptr i16, i16* [[PIN1]], i32 -2
+; CHECK: [[PIN1_2_CAST:%[^ ]+]] = bitcast i16* [[PIN1_2]] to i32*
+; CHECK: [[IN1_2:%[^ ]+]] = load i32, i32* [[PIN1_2_CAST]], align 2
 
+; CHECK: [[PIN2_2:%[^ ]+]] = getelementptr i16, i16* [[PIN2]], i32 -2
 ; CHECK: [[PIN2_2_CAST:%[^ ]+]] = bitcast i16* [[PIN2_2]] to i32*
 ; CHECK: [[IN2_2:%[^ ]+]] = load i32, i32* [[PIN2_2_CAST]], align 2
+
 ; CHECK: [[PIN1_CAST:%[^ ]+]] = bitcast i16* [[PIN1]] to i32*
 ; CHECK: [[IN1:%[^ ]+]] = load i32, i32* [[PIN1_CAST]], align 2
-; CHECK: [[ACC1:%[^ ]+]] = call i32 @llvm.arm.smladx(i32 [[IN2_2]], i32 [[IN1]], i32 [[ACC0]])
 
-; CHECK: [[PIN2_CAST:%[^ ]+]] = bitcast i16* [[PIN2]] to i32*
-; CHECK: [[IN2:%[^ ]+]] = load i32, i32* [[PIN2_CAST]], align 2
-; CHECK: [[PIN1_2_CAST:%[^ ]+]] = bitcast i16* [[PIN1_2]] to i32*
-; CHECK: [[IN1_2:%[^ ]+]] = load i32, i32* [[PIN1_2_CAST]], align 2
+; CHECK: [[ACC1:%[^ ]+]] = call i32 @llvm.arm.smladx(i32 [[IN2_2]], i32 [[IN1]], i32 [[ACC0]])
 ; CHECK: [[ACC2]] = call i32 @llvm.arm.smladx(i32 [[IN2]], i32 [[IN1_2]], i32 [[ACC1]])
 
 ; CHECK: [[PIN1_NEXT]] = getelementptr i16, i16* [[PIN1]], i32 4
 ; CHECK: [[PIN2_NEXT]] = getelementptr i16, i16* [[PIN2]], i32 -4
 
 ; CHECK-NOT: call i32 @llvm.arm.smlad
 ; CHECK-UNSUPPORTED-NOT:  call i32 @llvm.arm.smlad
 

test/CodeGen/ARM/ParallelDSP/smlald0.ll

 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s
 ; RUN: opt -mtriple=armeb-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 ;
 ; The Cortex-M0 does not support unaligned accesses:
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m0 < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 ;
 ; Check DSP extension:
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 -mattr=-dsp < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 
 define dso_local i64 @OneReduction(i32 %arg, i32* nocapture readnone %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
 ;
 ; CHECK-LABEL: @OneReduction
 ; CHECK:  %mac1{{\.}}026 = phi i64 [ [[V8:%[0-9]+]], %for.body ], [ 0, %for.body.preheader ]
-; CHECK:  [[V4:%[0-9]+]] = bitcast i16* %arrayidx3 to i32*
-; CHECK:  [[V5:%[0-9]+]] = load i32, i32* [[V4]], align 2
 ; CHECK:  [[V6:%[0-9]+]] = bitcast i16* %arrayidx to i32*
 ; CHECK:  [[V7:%[0-9]+]] = load i32, i32* [[V6]], align 2
+; CHECK:  [[V4:%[0-9]+]] = bitcast i16* %arrayidx3 to i32*
+; CHECK:  [[V5:%[0-9]+]] = load i32, i32* [[V4]], align 2
 ; CHECK:  [[V8]] = call i64 @llvm.arm.smlald(i32 [[V5]], i32 [[V7]], i64 %mac1{{\.}}026)
 ; CHECK-NOT: call i64 @llvm.arm.smlald
 ;
 ; CHECK-UNSUPPORTED-NOT:  call i64 @llvm.arm.smlald
 ;
 entry:
   %cmp24 = icmp sgt i32 %arg, 0
   br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup

test/CodeGen/ARM/ParallelDSP/smlald1.ll

 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s
 
 ; CHECK-LABEL: @test1
 ; CHECK:  %mac1{{\.}}026 = phi i64 [ [[V8:%[0-9]+]], %for.body ], [ 0, %for.body.preheader ]
-; CHECK:  [[V4:%[0-9]+]] = bitcast i16* %arrayidx3 to i32*
-; CHECK:  [[V5:%[0-9]+]] = load i32, i32* [[V4]], align 2
 ; CHECK:  [[V6:%[0-9]+]] = bitcast i16* %arrayidx to i32*
 ; CHECK:  [[V7:%[0-9]+]] = load i32, i32* [[V6]], align 2
+; CHECK:  [[V4:%[0-9]+]] = bitcast i16* %arrayidx3 to i32*
+; CHECK:  [[V5:%[0-9]+]] = load i32, i32* [[V4]], align 2
 ; CHECK:  [[V8]] = call i64 @llvm.arm.smlald(i32 [[V5]], i32 [[V7]], i64 %mac1{{\.}}026)
 
 define dso_local i64 @test1(i32 %arg, i32* nocapture readnone %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
 entry:
   %cmp24 = icmp sgt i32 %arg, 0
   br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup
 
 for.body.preheader:

test/CodeGen/ARM/ParallelDSP/smlald2.ll

 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 < %s -arm-parallel-dsp -S | FileCheck %s
 ;
 ; The Cortex-M0 does not support unaligned accesses:
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m0 < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 ;
 ; Check DSP extension:
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 -mattr=-dsp < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 
 define dso_local i64 @OneReduction(i32 %arg, i32* nocapture readnone %arg1, i16* nocapture readonly %arg2, i16* nocapture readonly %arg3) {
 ;
 ; CHECK-LABEL: @OneReduction
 ; CHECK:  %mac1{{\.}}026 = phi i64 [ [[V8:%[0-9]+]], %for.body ], [ 0, %for.body.preheader ]
-; CHECK:  [[V4:%[0-9]+]] = bitcast i16* %arrayidx3 to i32*
-; CHECK:  [[V5:%[0-9]+]] = load i32, i32* [[V4]], align 2
 ; CHECK:  [[V6:%[0-9]+]] = bitcast i16* %arrayidx to i32*
 ; CHECK:  [[V7:%[0-9]+]] = load i32, i32* [[V6]], align 2
+; CHECK:  [[V4:%[0-9]+]] = bitcast i16* %arrayidx3 to i32*
+; CHECK:  [[V5:%[0-9]+]] = load i32, i32* [[V4]], align 2
 ; CHECK:  [[V8]] = call i64 @llvm.arm.smlald(i32 [[V5]], i32 [[V7]], i64 %mac1{{\.}}026)
 ; CHECK-NOT: call i64 @llvm.arm.smlald
 ;
 ; CHECK-UNSUPPORTED-NOT:  call i64 @llvm.arm.smlald
 ;
 entry:
   %cmp24 = icmp sgt i32 %arg, 0
   br i1 %cmp24, label %for.body.preheader, label %for.cond.cleanup

test/CodeGen/ARM/ParallelDSP/smlaldx-1.ll

 ; RUN: opt -mtriple=thumbv8m.main -mcpu=cortex-m33 -arm-parallel-dsp %s -S -o - | FileCheck %s
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m0 < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 -mattr=-dsp < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 
 define i64 @smlaldx(i16* nocapture readonly %pIn1, i16* nocapture readonly %pIn2, i32 %j, i32 %limit) {
 
 ; CHECK-LABEL: smlaldx
 ; CHECK: = phi i32 [ 0, %for.body.preheader.new ],
 ; CHECK: [[ACC0:%[^ ]+]] = phi i64 [ 0, %for.body.preheader.new ], [ [[ACC2:%[^ ]+]], %for.body ]
+; CHECK: [[PIN21:%[^ ]+]] = bitcast i16* %pIn2.1 to i32*
+; CHECK: [[IN21:%[^ ]+]] = load i32, i32* [[PIN21]], align 2
+; CHECK: [[PIN10:%[^ ]+]] = bitcast i16* %pIn1.0 to i32*
+; CHECK: [[IN10:%[^ ]+]] = load i32, i32* [[PIN10]], align 2
 ; CHECK: [[PIN23:%[^ ]+]] = bitcast i16* %pIn2.3 to i32*
 ; CHECK: [[IN23:%[^ ]+]] = load i32, i32* [[PIN23]], align 2
 ; CHECK: [[PIN12:%[^ ]+]] = bitcast i16* %pIn1.2 to i32*
 ; CHECK: [[IN12:%[^ ]+]] = load i32, i32* [[PIN12]], align 2
 ; CHECK: [[ACC1:%[^ ]+]] = call i64 @llvm.arm.smlaldx(i32 [[IN23]], i32 [[IN12]], i64 [[ACC0]])
-; CHECK: [[PIN21:%[^ ]+]] = bitcast i16* %pIn2.1 to i32*
-; CHECK: [[IN21:%[^ ]+]] = load i32, i32* [[PIN21]], align 2
-; CHECK: [[PIN10:%[^ ]+]] = bitcast i16* %pIn1.0 to i32*
-; CHECK: [[IN10:%[^ ]+]] = load i32, i32* [[PIN10]], align 2
 ; CHECK: [[ACC2]] = call i64 @llvm.arm.smlaldx(i32 [[IN21]], i32 [[IN10]], i64 [[ACC1]])
 ; CHECK-NOT: call i64 @llvm.arm.smlad
 ; CHECK-UNSUPPORTED-NOT:  call i64 @llvm.arm.smlad
 
 entry:
   %cmp9 = icmp eq i32 %limit, 0
   br i1 %cmp9, label %for.cond.cleanup, label %for.body.preheader
 

 ; CHECK: [[PIN2Base:[^ ]+]] = getelementptr i16, i16* %pIn2
 
 ; CHECK: for.body:
 ; CHECK: [[PIN2:%[^ ]+]] = phi i16* [ [[PIN2_NEXT:%[^ ]+]], %for.body ], [ [[PIN2Base]], %for.body.preheader.new ]
 ; CHECK: [[PIN1:%[^ ]+]] = phi i16* [ [[PIN1_NEXT:%[^ ]+]], %for.body ], [ [[PIN1Base]], %for.body.preheader.new ]
 ; CHECK: [[IV:%[^ ]+]] = phi i32
 ; CHECK: [[ACC0:%[^ ]+]] = phi i64 [ 0, %for.body.preheader.new ], [ [[ACC2:%[^ ]+]], %for.body ]
 
+; CHECK: [[PIN2_CAST:%[^ ]+]] = bitcast i16* [[PIN2]] to i32*
+; CHECK: [[IN2:%[^ ]+]] = load i32, i32* [[PIN2_CAST]], align 2
+
 ; CHECK: [[PIN1_2:%[^ ]+]] = getelementptr i16, i16* [[PIN1]], i32 -2
-; CHECK: [[PIN2_2:%[^ ]+]] = getelementptr i16, i16* [[PIN2]], i32 -2
+; CHECK: [[PIN1_2_CAST:%[^ ]+]] = bitcast i16* [[PIN1_2]] to i32*
+; CHECK: [[IN1_2:%[^ ]+]] = load i32, i32* [[PIN1_2_CAST]], align 2
 
+; CHECK: [[PIN2_2:%[^ ]+]] = getelementptr i16, i16* [[PIN2]], i32 -2
 ; CHECK: [[PIN2_2_CAST:%[^ ]+]] = bitcast i16* [[PIN2_2]] to i32*
 ; CHECK: [[IN2_2:%[^ ]+]] = load i32, i32* [[PIN2_2_CAST]], align 2
+
 ; CHECK: [[PIN1_CAST:%[^ ]+]] = bitcast i16* [[PIN1]] to i32*
 ; CHECK: [[IN1:%[^ ]+]] = load i32, i32* [[PIN1_CAST]], align 2
-; CHECK: [[ACC1:%[^ ]+]] = call i64 @llvm.arm.smlaldx(i32 [[IN2_2]], i32 [[IN1]], i64 [[ACC0]])
 
-; CHECK: [[PIN2_CAST:%[^ ]+]] = bitcast i16* [[PIN2]] to i32*
-; CHECK: [[IN2:%[^ ]+]] = load i32, i32* [[PIN2_CAST]], align 2
-; CHECK: [[PIN1_2_CAST:%[^ ]+]] = bitcast i16* [[PIN1_2]] to i32*
-; CHECK: [[IN1_2:%[^ ]+]] = load i32, i32* [[PIN1_2_CAST]], align 2
+; CHECK: [[ACC1:%[^ ]+]] = call i64 @llvm.arm.smlaldx(i32 [[IN2_2]], i32 [[IN1]], i64 [[ACC0]])
 ; CHECK: [[ACC2]] = call i64 @llvm.arm.smlaldx(i32 [[IN2]], i32 [[IN1_2]], i64 [[ACC1]])
 
 ; CHECK: [[PIN1_NEXT]] = getelementptr i16, i16* [[PIN1]], i32 4
 ; CHECK: [[PIN2_NEXT]] = getelementptr i16, i16* [[PIN2]], i32 -4
 
 ; CHECK-NOT: call i64 @llvm.arm.smlad
 ; CHECK-UNSUPPORTED-NOT:  call i64 @llvm.arm.smlad
 

test/CodeGen/ARM/ParallelDSP/smlaldx-2.ll

 ; RUN: opt -mtriple=thumbv8m.main -mcpu=cortex-m33 -arm-parallel-dsp %s -S -o - | FileCheck %s
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m0 < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 ; RUN: opt -mtriple=arm-arm-eabi -mcpu=cortex-m33 -mattr=-dsp < %s -arm-parallel-dsp -S | FileCheck %s --check-prefix=CHECK-UNSUPPORTED
 
 define i64 @smlaldx(i16* nocapture readonly %pIn1, i16* nocapture readonly %pIn2, i32 %j, i32 %limit) {
 
 ; CHECK-LABEL: smlaldx
 ; CHECK: = phi i32 [ 0, %for.body.preheader.new ],
 ; CHECK: [[ACC0:%[^ ]+]] = phi i64 [ 0, %for.body.preheader.new ], [ [[ACC2:%[^ ]+]], %for.body ]
+; CHECK: [[PIN21:%[^ ]+]] = bitcast i16* %pIn2.1 to i32*
+; CHECK: [[IN21:%[^ ]+]] = load i32, i32* [[PIN21]], align 2
+; CHECK: [[PIN10:%[^ ]+]] = bitcast i16* %pIn1.0 to i32*
+; CHECK: [[IN10:%[^ ]+]] = load i32, i32* [[PIN10]], align 2
 ; CHECK: [[PIN23:%[^ ]+]] = bitcast i16* %pIn2.3 to i32*
 ; CHECK: [[IN23:%[^ ]+]] = load i32, i32* [[PIN23]], align 2
 ; CHECK: [[PIN12:%[^ ]+]] = bitcast i16* %pIn1.2 to i32*
 ; CHECK: [[IN12:%[^ ]+]] = load i32, i32* [[PIN12]], align 2
 ; CHECK: [[ACC1:%[^ ]+]] = call i64 @llvm.arm.smlaldx(i32 [[IN23]], i32 [[IN12]], i64 [[ACC0]])
-; CHECK: [[PIN21:%[^ ]+]] = bitcast i16* %pIn2.1 to i32*
-; CHECK: [[IN21:%[^ ]+]] = load i32, i32* [[PIN21]], align 2
-; CHECK: [[PIN10:%[^ ]+]] = bitcast i16* %pIn1.0 to i32*
-; CHECK: [[IN10:%[^ ]+]] = load i32, i32* [[PIN10]], align 2
 ; CHECK: [[ACC2]] = call i64 @llvm.arm.smlaldx(i32 [[IN21]], i32 [[IN10]], i64 [[ACC1]])
 ; CHECK-NOT: call i64 @llvm.arm.smlad
 ; CHECK-UNSUPPORTED-NOT:  call i64 @llvm.arm.smlad
 
 entry:
   %cmp9 = icmp eq i32 %limit, 0
   br i1 %cmp9, label %for.cond.cleanup, label %for.body.preheader
 

 ; CHECK: [[PIN1Base:[^ ]+]] = getelementptr i16, i16* %pIn1
 ; CHECK: [[PIN2Base:[^ ]+]] = getelementptr i16, i16* %pIn2
 
 ; CHECK: for.body:
 ; CHECK: [[PIN2:%[^ ]+]] = phi i16* [ [[PIN2_NEXT:%[^ ]+]], %for.body ], [ [[PIN2Base]], %for.body.preheader.new ]
 ; CHECK: [[PIN1:%[^ ]+]] = phi i16* [ [[PIN1_NEXT:%[^ ]+]], %for.body ], [ [[PIN1Base]], %for.body.preheader.new ]
 ; CHECK: [[IV:%[^ ]+]] = phi i32
 ; CHECK: [[ACC0:%[^ ]+]] = phi i64 [ 0, %for.body.preheader.new ], [ [[ACC2:%[^ ]+]], %for.body ]
-; CHECK: [[PIN1_2:%[^ ]+]] = getelementptr i16, i16* [[PIN1]], i32 -2
-; CHECK: [[PIN2_2:%[^ ]+]] = getelementptr i16, i16* [[PIN2]], i32 -2
 
 ; CHECK: [[PIN2_CAST:%[^ ]+]] = bitcast i16* [[PIN2]] to i32*
 ; CHECK: [[IN2:%[^ ]+]] = load i32, i32* [[PIN2_CAST]], align 2
+
+; CHECK: [[PIN1_2:%[^ ]+]] = getelementptr i16, i16* [[PIN1]], i32 -2
 ; CHECK: [[PIN1_2_CAST:%[^ ]+]] = bitcast i16* [[PIN1_2]] to i32*
 ; CHECK: [[IN1_2:%[^ ]+]] = load i32, i32* [[PIN1_2_CAST]], align 2
-; CHECK: [[ACC1:%[^ ]+]] = call i64 @llvm.arm.smlaldx(i32 [[IN2]], i32 [[IN1_2]], i64 [[ACC0]])
 
-; CHECK: [[PIN1_CAST:%[^ ]+]] = bitcast i16* [[PIN1]] to i32*
-; CHECK: [[IN1:%[^ ]+]] = load i32, i32* [[PIN1_CAST]], align 2
+; CHECK: [[PIN2_2:%[^ ]+]] = getelementptr i16, i16* [[PIN2]], i32 -2
 ; CHECK: [[PIN2_2_CAST:%[^ ]+]] = bitcast i16* [[PIN2_2]] to i32*
 ; CHECK: [[IN2_2:%[^ ]+]] = load i32, i32* [[PIN2_2_CAST]], align 2
+
+; CHECK: [[PIN1_CAST:%[^ ]+]] = bitcast i16* [[PIN1]] to i32*
+; CHECK: [[IN1:%[^ ]+]] = load i32, i32* [[PIN1_CAST]], align 2
+
+; CHECK: [[ACC1:%[^ ]+]] = call i64 @llvm.arm.smlaldx(i32 [[IN2]], i32 [[IN1_2]], i64 [[ACC0]])
 ; CHECK: [[ACC2]] = call i64 @llvm.arm.smlaldx(i32 [[IN1]], i32 [[IN2_2]], i64 [[ACC1]])
 
 ; CHECK: [[PIN1_NEXT]] = getelementptr i16, i16* [[PIN1]], i32 4
 ; CHECK: [[PIN2_NEXT]] = getelementptr i16, i16* [[PIN2]], i32 -4
 
 ; CHECK-NOT: call i64 @llvm.arm.smlad
 ; CHECK-UNSUPPORTED-NOT:  call i64 @llvm.arm.smlad