Updating branches/google/stable to r176857

git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/google/stable@177040 91177308-0d34-0410-b5e6-96231b3b80d8

4 files changed, 2624 insertions, 1532 deletions

diff --git a/lib/Transforms/Vectorize/BBVectorize.cpp b/lib/Transforms/Vectorize/BBVectorize.cpp
index a48229132b..17900dabbe 100644
--- a/lib/Transforms/Vectorize/BBVectorize.cpp
+++ b/lib/Transforms/Vectorize/BBVectorize.cpp
@@ -29,26 +29,25 @@
 #include "llvm/Analysis/Dominators.h"
 #include "llvm/Analysis/ScalarEvolution.h"
 #include "llvm/Analysis/ScalarEvolutionExpressions.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/Analysis/ValueTracking.h"
-#include "llvm/Constants.h"
-#include "llvm/DataLayout.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Instructions.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/Intrinsics.h"
-#include "llvm/LLVMContext.h"
-#include "llvm/Metadata.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Metadata.h"
+#include "llvm/IR/Type.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/ValueHandle.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/TargetTransformInfo.h"
 #include "llvm/Transforms/Utils/Local.h"
-#include "llvm/Type.h"
 #include <algorithm>
-#include <map>
 using namespace llvm;
 
 static cl::opt<bool>
@@ -89,6 +88,10 @@ MaxInsts("bb-vectorize-max-instr-per-group", cl::init(500), cl::Hidden,
   cl::desc("The maximum number of pairable instructions per group"));
 
 static cl::opt<unsigned>
+MaxPairs("bb-vectorize-max-pairs-per-group", cl::init(3000), cl::Hidden,
+  cl::desc("The maximum number of candidate instruction pairs per group"));
+
+static cl::opt<unsigned>
 MaxCandPairsForCycleCheck("bb-vectorize-max-cycle-check-pairs", cl::init(200),
   cl::Hidden, cl::desc("The maximum number of candidate pairs with which to use"
                        " a full cycle check"));
@@ -199,9 +202,7 @@ namespace {
       DT = &P->getAnalysis<DominatorTree>();
       SE = &P->getAnalysis<ScalarEvolution>();
       TD = P->getAnalysisIfAvailable<DataLayout>();
-      TTI = IgnoreTargetInfo ? 0 :
-        P->getAnalysisIfAvailable<TargetTransformInfo>();
-      VTTI = TTI ? TTI->getVectorTargetTransformInfo() : 0;
+      TTI = IgnoreTargetInfo ? 0 : &P->getAnalysis<TargetTransformInfo>();
     }
 
     typedef std::pair<Value *, Value *> ValuePair;
@@ -209,18 +210,12 @@ namespace {
     typedef std::pair<ValuePair, size_t> ValuePairWithDepth;
     typedef std::pair<ValuePair, ValuePair> VPPair; // A ValuePair pair
     typedef std::pair<VPPair, unsigned> VPPairWithType;
-    typedef std::pair<std::multimap<Value *, Value *>::iterator,
-              std::multimap<Value *, Value *>::iterator> VPIteratorPair;
-    typedef std::pair<std::multimap<ValuePair, ValuePair>::iterator,
-              std::multimap<ValuePair, ValuePair>::iterator>
-                VPPIteratorPair;
 
     AliasAnalysis *AA;
     DominatorTree *DT;
     ScalarEvolution *SE;
     DataLayout *TD;
-    TargetTransformInfo *TTI;
-    const VectorTargetTransformInfo *VTTI;
+    const TargetTransformInfo *TTI;
 
     // FIXME: const correct?
 
@@ -228,7 +223,7 @@ namespace {
 
     bool getCandidatePairs(BasicBlock &BB,
                        BasicBlock::iterator &Start,
-                       std::multimap<Value *, Value *> &CandidatePairs,
+                       DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
                        DenseSet<ValuePair> &FixedOrderPairs,
                        DenseMap<ValuePair, int> &CandidatePairCostSavings,
                        std::vector<Value *> &PairableInsts, bool NonPow2Len);
@@ -242,33 +237,36 @@ namespace {
       PairConnectionSplat
     };
 
-    void computeConnectedPairs(std::multimap<Value *, Value *> &CandidatePairs,
-                       std::vector<Value *> &PairableInsts,
-                       std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                       DenseMap<VPPair, unsigned> &PairConnectionTypes);
+    void computeConnectedPairs(
+             DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+             DenseSet<ValuePair> &CandidatePairsSet,
+             std::vector<Value *> &PairableInsts,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+             DenseMap<VPPair, unsigned> &PairConnectionTypes);
 
     void buildDepMap(BasicBlock &BB,
-                       std::multimap<Value *, Value *> &CandidatePairs,
-                       std::vector<Value *> &PairableInsts,
-                       DenseSet<ValuePair> &PairableInstUsers);
-
-    void choosePairs(std::multimap<Value *, Value *> &CandidatePairs,
-                        DenseMap<ValuePair, int> &CandidatePairCostSavings,
-                        std::vector<Value *> &PairableInsts,
-                        DenseSet<ValuePair> &FixedOrderPairs,
-                        DenseMap<VPPair, unsigned> &PairConnectionTypes,
-                        std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                        std::multimap<ValuePair, ValuePair> &ConnectedPairDeps,
-                        DenseSet<ValuePair> &PairableInstUsers,
-                        DenseMap<Value *, Value *>& ChosenPairs);
+             DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+             std::vector<Value *> &PairableInsts,
+             DenseSet<ValuePair> &PairableInstUsers);
+
+    void choosePairs(DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+             DenseSet<ValuePair> &CandidatePairsSet,
+             DenseMap<ValuePair, int> &CandidatePairCostSavings,
+             std::vector<Value *> &PairableInsts,
+             DenseSet<ValuePair> &FixedOrderPairs,
+             DenseMap<VPPair, unsigned> &PairConnectionTypes,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps,
+             DenseSet<ValuePair> &PairableInstUsers,
+             DenseMap<Value *, Value *>& ChosenPairs);
 
     void fuseChosenPairs(BasicBlock &BB,
-                     std::vector<Value *> &PairableInsts,
-                     DenseMap<Value *, Value *>& ChosenPairs,
-                     DenseSet<ValuePair> &FixedOrderPairs,
-                     DenseMap<VPPair, unsigned> &PairConnectionTypes,
-                     std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                     std::multimap<ValuePair, ValuePair> &ConnectedPairDeps);
+             std::vector<Value *> &PairableInsts,
+             DenseMap<Value *, Value *>& ChosenPairs,
+             DenseSet<ValuePair> &FixedOrderPairs,
+             DenseMap<VPPair, unsigned> &PairConnectionTypes,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps);
 
 
     bool isInstVectorizable(Instruction *I, bool &IsSimpleLoadStore);
@@ -280,56 +278,63 @@ namespace {
     bool trackUsesOfI(DenseSet<Value *> &Users,
                       AliasSetTracker &WriteSet, Instruction *I,
                       Instruction *J, bool UpdateUsers = true,
-                      std::multimap<Value *, Value *> *LoadMoveSet = 0);
+                      DenseSet<ValuePair> *LoadMoveSetPairs = 0);
 
-    void computePairsConnectedTo(
-                      std::multimap<Value *, Value *> &CandidatePairs,
-                      std::vector<Value *> &PairableInsts,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                      DenseMap<VPPair, unsigned> &PairConnectionTypes,
-                      ValuePair P);
+  void computePairsConnectedTo(
+             DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+             DenseSet<ValuePair> &CandidatePairsSet,
+             std::vector<Value *> &PairableInsts,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+             DenseMap<VPPair, unsigned> &PairConnectionTypes,
+             ValuePair P);
 
     bool pairsConflict(ValuePair P, ValuePair Q,
-                 DenseSet<ValuePair> &PairableInstUsers,
-                 std::multimap<ValuePair, ValuePair> *PairableInstUserMap = 0);
+             DenseSet<ValuePair> &PairableInstUsers,
+             DenseMap<ValuePair, std::vector<ValuePair> >
+               *PairableInstUserMap = 0,
+             DenseSet<VPPair> *PairableInstUserPairSet = 0);
 
     bool pairWillFormCycle(ValuePair P,
-                       std::multimap<ValuePair, ValuePair> &PairableInstUsers,
-                       DenseSet<ValuePair> &CurrentPairs);
-
-    void pruneTreeFor(
-                      std::multimap<Value *, Value *> &CandidatePairs,
-                      std::vector<Value *> &PairableInsts,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                      DenseSet<ValuePair> &PairableInstUsers,
-                      std::multimap<ValuePair, ValuePair> &PairableInstUserMap,
-                      DenseMap<Value *, Value *> &ChosenPairs,
-                      DenseMap<ValuePair, size_t> &Tree,
-                      DenseSet<ValuePair> &PrunedTree, ValuePair J,
-                      bool UseCycleCheck);
-
-    void buildInitialTreeFor(
-                      std::multimap<Value *, Value *> &CandidatePairs,
-                      std::vector<Value *> &PairableInsts,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                      DenseSet<ValuePair> &PairableInstUsers,
-                      DenseMap<Value *, Value *> &ChosenPairs,
-                      DenseMap<ValuePair, size_t> &Tree, ValuePair J);
-
-    void findBestTreeFor(
-                      std::multimap<Value *, Value *> &CandidatePairs,
-                      DenseMap<ValuePair, int> &CandidatePairCostSavings,
-                      std::vector<Value *> &PairableInsts,
-                      DenseSet<ValuePair> &FixedOrderPairs,
-                      DenseMap<VPPair, unsigned> &PairConnectionTypes,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairDeps,
-                      DenseSet<ValuePair> &PairableInstUsers,
-                      std::multimap<ValuePair, ValuePair> &PairableInstUserMap,
-                      DenseMap<Value *, Value *> &ChosenPairs,
-                      DenseSet<ValuePair> &BestTree, size_t &BestMaxDepth,
-                      int &BestEffSize, VPIteratorPair ChoiceRange,
-                      bool UseCycleCheck);
+             DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUsers,
+             DenseSet<ValuePair> &CurrentPairs);
+
+    void pruneDAGFor(
+             DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+             std::vector<Value *> &PairableInsts,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+             DenseSet<ValuePair> &PairableInstUsers,
+             DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUserMap,
+             DenseSet<VPPair> &PairableInstUserPairSet,
+             DenseMap<Value *, Value *> &ChosenPairs,
+             DenseMap<ValuePair, size_t> &DAG,
+             DenseSet<ValuePair> &PrunedDAG, ValuePair J,
+             bool UseCycleCheck);
+
+    void buildInitialDAGFor(
+             DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+             DenseSet<ValuePair> &CandidatePairsSet,
+             std::vector<Value *> &PairableInsts,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+             DenseSet<ValuePair> &PairableInstUsers,
+             DenseMap<Value *, Value *> &ChosenPairs,
+             DenseMap<ValuePair, size_t> &DAG, ValuePair J);
+
+    void findBestDAGFor(
+             DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+             DenseSet<ValuePair> &CandidatePairsSet,
+             DenseMap<ValuePair, int> &CandidatePairCostSavings,
+             std::vector<Value *> &PairableInsts,
+             DenseSet<ValuePair> &FixedOrderPairs,
+             DenseMap<VPPair, unsigned> &PairConnectionTypes,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps,
+             DenseSet<ValuePair> &PairableInstUsers,
+             DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUserMap,
+             DenseSet<VPPair> &PairableInstUserPairSet,
+             DenseMap<Value *, Value *> &ChosenPairs,
+             DenseSet<ValuePair> &BestDAG, size_t &BestMaxDepth,
+             int &BestEffSize, Value *II, std::vector<Value *>&JJ,
+             bool UseCycleCheck);
 
     Value *getReplacementPointerInput(LLVMContext& Context, Instruction *I,
                      Instruction *J, unsigned o);
@@ -361,20 +366,22 @@ namespace {
 
     void collectPairLoadMoveSet(BasicBlock &BB,
                      DenseMap<Value *, Value *> &ChosenPairs,
-                     std::multimap<Value *, Value *> &LoadMoveSet,
+                     DenseMap<Value *, std::vector<Value *> > &LoadMoveSet,
+                     DenseSet<ValuePair> &LoadMoveSetPairs,
                      Instruction *I);
 
     void collectLoadMoveSet(BasicBlock &BB,
                      std::vector<Value *> &PairableInsts,
                      DenseMap<Value *, Value *> &ChosenPairs,
-                     std::multimap<Value *, Value *> &LoadMoveSet);
+                     DenseMap<Value *, std::vector<Value *> > &LoadMoveSet,
+                     DenseSet<ValuePair> &LoadMoveSetPairs);
 
     bool canMoveUsesOfIAfterJ(BasicBlock &BB,
-                     std::multimap<Value *, Value *> &LoadMoveSet,
+                     DenseSet<ValuePair> &LoadMoveSetPairs,
                      Instruction *I, Instruction *J);
 
     void moveUsesOfIAfterJ(BasicBlock &BB,
-                     std::multimap<Value *, Value *> &LoadMoveSet,
+                     DenseSet<ValuePair> &LoadMoveSetPairs,
                      Instruction *&InsertionPt,
                      Instruction *I, Instruction *J);
 
@@ -387,7 +394,7 @@ namespace {
         return false;
       }
 
-      DEBUG(if (VTTI) dbgs() << "BBV: using target information\n");
+      DEBUG(if (TTI) dbgs() << "BBV: using target information\n");
 
       bool changed = false;
       // Iterate a sufficient number of times to merge types of size 1 bit,
@@ -395,7 +402,7 @@ namespace {
       // target vector register.
       unsigned n = 1;
       for (unsigned v = 2;
-           (VTTI || v <= Config.VectorBits) &&
+           (TTI || v <= Config.VectorBits) &&
            (!Config.MaxIter || n <= Config.MaxIter);
            v *= 2, ++n) {
         DEBUG(dbgs() << "BBV: fusing loop #" << n <<
@@ -426,9 +433,7 @@ namespace {
       DT = &getAnalysis<DominatorTree>();
       SE = &getAnalysis<ScalarEvolution>();
       TD = getAnalysisIfAvailable<DataLayout>();
-      TTI = IgnoreTargetInfo ? 0 :
-        getAnalysisIfAvailable<TargetTransformInfo>();
-      VTTI = TTI ? TTI->getVectorTargetTransformInfo() : 0;
+      TTI = IgnoreTargetInfo ? 0 : &getAnalysis<TargetTransformInfo>();
 
       return vectorizeBB(BB);
     }
@@ -438,6 +443,7 @@ namespace {
       AU.addRequired<AliasAnalysis>();
       AU.addRequired<DominatorTree>();
       AU.addRequired<ScalarEvolution>();
+      AU.addRequired<TargetTransformInfo>();
       AU.addPreserved<AliasAnalysis>();
       AU.addPreserved<DominatorTree>();
       AU.addPreserved<ScalarEvolution>();
@@ -467,18 +473,18 @@ namespace {
 
     static inline void getInstructionTypes(Instruction *I,
                                            Type *&T1, Type *&T2) {
-      if (isa<StoreInst>(I)) {
+      if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
         // For stores, it is the value type, not the pointer type that matters
         // because the value is what will come from a vector register.
   
-        Value *IVal = cast<StoreInst>(I)->getValueOperand();
+        Value *IVal = SI->getValueOperand();
         T1 = IVal->getType();
       } else {
         T1 = I->getType();
       }
   
-      if (I->isCast())
-        T2 = cast<CastInst>(I)->getSrcTy();
+      if (CastInst *CI = dyn_cast<CastInst>(I))
+        T2 = CI->getSrcTy();
       else
         T2 = T1;
 
@@ -504,7 +510,7 @@ namespace {
       // InsertElement and ExtractElement have a depth factor of zero. This is
       // for two reasons: First, they cannot be usefully fused. Second, because
       // the pass generates a lot of these, they can confuse the simple metric
-      // used to compare the trees in the next iteration. Thus, giving them a
+      // used to compare the dags in the next iteration. Thus, giving them a
       // weight of zero allows the pass to essentially ignore them in
       // subsequent iterations when looking for vectorization opportunities
       // while still tracking dependency chains that flow through those
@@ -520,7 +526,7 @@ namespace {
       return 1;
     }
 
-    // Returns the cost of the provided instruction using VTTI.
+    // Returns the cost of the provided instruction using TTI.
     // This does not handle loads and stores.
     unsigned getInstrCost(unsigned Opcode, Type *T1, Type *T2) {
       switch (Opcode) {
@@ -531,7 +537,7 @@ namespace {
         // generate vector GEPs.
         return 0;
       case Instruction::Br:
-        return VTTI->getCFInstrCost(Opcode);
+        return TTI->getCFInstrCost(Opcode);
       case Instruction::PHI:
         return 0;
       case Instruction::Add:
@@ -552,11 +558,11 @@ namespace {
       case Instruction::And:
       case Instruction::Or:
       case Instruction::Xor:
-        return VTTI->getArithmeticInstrCost(Opcode, T1);
+        return TTI->getArithmeticInstrCost(Opcode, T1);
       case Instruction::Select:
       case Instruction::ICmp:
       case Instruction::FCmp:
-        return VTTI->getCmpSelInstrCost(Opcode, T1, T2);
+        return TTI->getCmpSelInstrCost(Opcode, T1, T2);
       case Instruction::ZExt:
       case Instruction::SExt:
       case Instruction::FPToUI:
@@ -570,7 +576,7 @@ namespace {
       case Instruction::FPTrunc:
       case Instruction::BitCast:
       case Instruction::ShuffleVector:
-        return VTTI->getCastInstrCost(Opcode, T1, T2);
+        return TTI->getCastInstrCost(Opcode, T1, T2);
       }
 
       return 1;
@@ -642,7 +648,7 @@ namespace {
       Function *F = I->getCalledFunction();
       if (!F) return false;
 
-      unsigned IID = F->getIntrinsicID();
+      Intrinsic::ID IID = (Intrinsic::ID) F->getIntrinsicID();
       if (!IID) return false;
 
       switch(IID) {
@@ -660,23 +666,11 @@ namespace {
       case Intrinsic::pow:
         return Config.VectorizeMath;
       case Intrinsic::fma:
+      case Intrinsic::fmuladd:
         return Config.VectorizeFMA;
       }
     }
 
-    // Returns true if J is the second element in some pair referenced by
-    // some multimap pair iterator pair.
-    template <typename V>
-    bool isSecondInIteratorPair(V J, std::pair<
-           typename std::multimap<V, V>::iterator,
-           typename std::multimap<V, V>::iterator> PairRange) {
-      for (typename std::multimap<V, V>::iterator K = PairRange.first;
-           K != PairRange.second; ++K)
-        if (K->second == J) return true;
-
-      return false;
-    }
-
     bool isPureIEChain(InsertElementInst *IE) {
       InsertElementInst *IENext = IE;
       do {
@@ -701,11 +695,12 @@ namespace {
     DenseMap<Value *, Value *> AllChosenPairs;
     DenseSet<ValuePair> AllFixedOrderPairs;
     DenseMap<VPPair, unsigned> AllPairConnectionTypes;
-    std::multimap<ValuePair, ValuePair> AllConnectedPairs, AllConnectedPairDeps;
+    DenseMap<ValuePair, std::vector<ValuePair> > AllConnectedPairs,
+                                                 AllConnectedPairDeps;
 
     do {
       std::vector<Value *> PairableInsts;
-      std::multimap<Value *, Value *> CandidatePairs;
+      DenseMap<Value *, std::vector<Value *> > CandidatePairs;
       DenseSet<ValuePair> FixedOrderPairs;
       DenseMap<ValuePair, int> CandidatePairCostSavings;
       ShouldContinue = getCandidatePairs(BB, Start, CandidatePairs,
@@ -714,6 +709,14 @@ namespace {
                                          PairableInsts, NonPow2Len);
       if (PairableInsts.empty()) continue;
 
+      // Build the candidate pair set for faster lookups.
+      DenseSet<ValuePair> CandidatePairsSet;
+      for (DenseMap<Value *, std::vector<Value *> >::iterator I =
+           CandidatePairs.begin(), E = CandidatePairs.end(); I != E; ++I)
+        for (std::vector<Value *>::iterator J = I->second.begin(),
+             JE = I->second.end(); J != JE; ++J)
+          CandidatePairsSet.insert(ValuePair(I->first, *J));
+
       // Now we have a map of all of the pairable instructions and we need to
       // select the best possible pairing. A good pairing is one such that the
       // users of the pair are also paired. This defines a (directed) forest
@@ -723,30 +726,33 @@ namespace {
       // Note that it only matters that both members of the second pair use some
       // element of the first pair (to allow for splatting).
 
-      std::multimap<ValuePair, ValuePair> ConnectedPairs, ConnectedPairDeps;
+      DenseMap<ValuePair, std::vector<ValuePair> > ConnectedPairs,
+                                                   ConnectedPairDeps;
       DenseMap<VPPair, unsigned> PairConnectionTypes;
-      computeConnectedPairs(CandidatePairs, PairableInsts, ConnectedPairs,
-                            PairConnectionTypes);
+      computeConnectedPairs(CandidatePairs, CandidatePairsSet,
+                            PairableInsts, ConnectedPairs, PairConnectionTypes);
       if (ConnectedPairs.empty()) continue;
 
-      for (std::multimap<ValuePair, ValuePair>::iterator
+      for (DenseMap<ValuePair, std::vector<ValuePair> >::iterator
            I = ConnectedPairs.begin(), IE = ConnectedPairs.end();
-           I != IE; ++I) {
-        ConnectedPairDeps.insert(VPPair(I->second, I->first));
-      }
+           I != IE; ++I)
+        for (std::vector<ValuePair>::iterator J = I->second.begin(),
+             JE = I->second.end(); J != JE; ++J)
+          ConnectedPairDeps[*J].push_back(I->first);
 
       // Build the pairable-instruction dependency map
       DenseSet<ValuePair> PairableInstUsers;
       buildDepMap(BB, CandidatePairs, PairableInsts, PairableInstUsers);
 
       // There is now a graph of the connected pairs. For each variable, pick
-      // the pairing with the largest tree meeting the depth requirement on at
-      // least one branch. Then select all pairings that are part of that tree
+      // the pairing with the largest dag meeting the depth requirement on at
+      // least one branch. Then select all pairings that are part of that dag
       // and remove them from the list of available pairings and pairable
       // variables.
 
       DenseMap<Value *, Value *> ChosenPairs;
-      choosePairs(CandidatePairs, CandidatePairCostSavings,
+      choosePairs(CandidatePairs, CandidatePairsSet,
+        CandidatePairCostSavings,
         PairableInsts, FixedOrderPairs, PairConnectionTypes,
         ConnectedPairs, ConnectedPairDeps,
         PairableInstUsers, ChosenPairs);
@@ -780,14 +786,15 @@ namespace {
         }
       }
 
-      for (std::multimap<ValuePair, ValuePair>::iterator
+      for (DenseMap<ValuePair, std::vector<ValuePair> >::iterator
            I = ConnectedPairs.begin(), IE = ConnectedPairs.end();
-           I != IE; ++I) {
-        if (AllPairConnectionTypes.count(*I)) {
-          AllConnectedPairs.insert(*I);
-          AllConnectedPairDeps.insert(VPPair(I->second, I->first));
-        }
-      }
+           I != IE; ++I)
+        for (std::vector<ValuePair>::iterator J = I->second.begin(),
+          JE = I->second.end(); J != JE; ++J)
+          if (AllPairConnectionTypes.count(VPPair(I->first, *J))) {
+            AllConnectedPairs[I->first].push_back(*J);
+            AllConnectedPairDeps[*J].push_back(I->first);
+          }
     } while (ShouldContinue);
 
     if (AllChosenPairs.empty()) return false;
@@ -903,8 +910,8 @@ namespace {
          T2->getScalarType()->isPointerTy()))
       return false;
 
-    if (!VTTI && (T1->getPrimitiveSizeInBits() >= Config.VectorBits ||
-                  T2->getPrimitiveSizeInBits() >= Config.VectorBits))
+    if (!TTI && (T1->getPrimitiveSizeInBits() >= Config.VectorBits ||
+                 T2->getPrimitiveSizeInBits() >= Config.VectorBits))
       return false;
 
     return true;
@@ -913,7 +920,7 @@ namespace {
   // This function returns true if the two provided instructions are compatible
   // (meaning that they can be fused into a vector instruction). This assumes
   // that I has already been determined to be vectorizable and that J is not
-  // in the use tree of I.
+  // in the use dag of I.
   bool BBVectorize::areInstsCompatible(Instruction *I, Instruction *J,
                        bool IsSimpleLoadStore, bool NonPow2Len,
                        int &CostSavings, int &FixedOrder) {
@@ -935,7 +942,7 @@ namespace {
     unsigned MaxTypeBits = std::max(
       IT1->getPrimitiveSizeInBits() + JT1->getPrimitiveSizeInBits(),
       IT2->getPrimitiveSizeInBits() + JT2->getPrimitiveSizeInBits());
-    if (!VTTI && MaxTypeBits > Config.VectorBits)
+    if (!TTI && MaxTypeBits > Config.VectorBits)
       return false;
 
     // FIXME: handle addsub-type operations!
@@ -967,21 +974,26 @@ namespace {
             return false;
         }
 
-        if (VTTI) {
-          unsigned ICost = VTTI->getMemoryOpCost(I->getOpcode(), I->getType(),
-                                                 IAlignment, IAddressSpace);
-          unsigned JCost = VTTI->getMemoryOpCost(J->getOpcode(), J->getType(),
-                                                 JAlignment, JAddressSpace);
-          unsigned VCost = VTTI->getMemoryOpCost(I->getOpcode(), VType,
-                                                 BottomAlignment,
-                                                 IAddressSpace);
+        if (TTI) {
+          unsigned ICost = TTI->getMemoryOpCost(I->getOpcode(), aTypeI,
+                                                IAlignment, IAddressSpace);
+          unsigned JCost = TTI->getMemoryOpCost(J->getOpcode(), aTypeJ,
+                                                JAlignment, JAddressSpace);
+          unsigned VCost = TTI->getMemoryOpCost(I->getOpcode(), VType,
+                                                BottomAlignment,
+                                                IAddressSpace);
+
+          ICost += TTI->getAddressComputationCost(aTypeI);
+          JCost += TTI->getAddressComputationCost(aTypeJ);
+          VCost += TTI->getAddressComputationCost(VType);
+
           if (VCost > ICost + JCost)
             return false;
 
           // We don't want to fuse to a type that will be split, even
           // if the two input types will also be split and there is no other
           // associated cost.
-          unsigned VParts = VTTI->getNumberOfParts(VType);
+          unsigned VParts = TTI->getNumberOfParts(VType);
           if (VParts > 1)
             return false;
           else if (!VParts && VCost == ICost + JCost)
@@ -992,11 +1004,17 @@ namespace {
       } else {
         return false;
       }
-    } else if (VTTI) {
+    } else if (TTI) {
       unsigned ICost = getInstrCost(I->getOpcode(), IT1, IT2);
       unsigned JCost = getInstrCost(J->getOpcode(), JT1, JT2);
       Type *VT1 = getVecTypeForPair(IT1, JT1),
            *VT2 = getVecTypeForPair(IT2, JT2);
+
+      // Note that this procedure is incorrect for insert and extract element
+      // instructions (because combining these often results in a shuffle),
+      // but this cost is ignored (because insert and extract element
+      // instructions are assigned a zero depth factor and are not really
+      // fused in general).
       unsigned VCost = getInstrCost(I->getOpcode(), VT1, VT2);
 
       if (VCost > ICost + JCost)
@@ -1005,8 +1023,8 @@ namespace {
       // We don't want to fuse to a type that will be split, even
       // if the two input types will also be split and there is no other
       // associated cost.
-      unsigned VParts1 = VTTI->getNumberOfParts(VT1),
-               VParts2 = VTTI->getNumberOfParts(VT2);
+      unsigned VParts1 = TTI->getNumberOfParts(VT1),
+               VParts2 = TTI->getNumberOfParts(VT2);
       if (VParts1 > 1 || VParts2 > 1)
         return false;
       else if ((!VParts1 || !VParts2) && VCost == ICost + JCost)
@@ -1019,14 +1037,67 @@ namespace {
     // vectorized, the second arguments must be equal.
     CallInst *CI = dyn_cast<CallInst>(I);
     Function *FI;
-    if (CI && (FI = CI->getCalledFunction()) &&
-        FI->getIntrinsicID() == Intrinsic::powi) {
-
-      Value *A1I = CI->getArgOperand(1),
-            *A1J = cast<CallInst>(J)->getArgOperand(1);
-      const SCEV *A1ISCEV = SE->getSCEV(A1I),
-                 *A1JSCEV = SE->getSCEV(A1J);
-      return (A1ISCEV == A1JSCEV);
+    if (CI && (FI = CI->getCalledFunction())) {
+      Intrinsic::ID IID = (Intrinsic::ID) FI->getIntrinsicID();
+      if (IID == Intrinsic::powi) {
+        Value *A1I = CI->getArgOperand(1),
+              *A1J = cast<CallInst>(J)->getArgOperand(1);
+        const SCEV *A1ISCEV = SE->getSCEV(A1I),
+                   *A1JSCEV = SE->getSCEV(A1J);
+        return (A1ISCEV == A1JSCEV);
+      }
+
+      if (IID && TTI) {
+        SmallVector<Type*, 4> Tys;
+        for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i)
+          Tys.push_back(CI->getArgOperand(i)->getType());
+        unsigned ICost = TTI->getIntrinsicInstrCost(IID, IT1, Tys);
+
+        Tys.clear();
+        CallInst *CJ = cast<CallInst>(J);
+        for (unsigned i = 0, ie = CJ->getNumArgOperands(); i != ie; ++i)
+          Tys.push_back(CJ->getArgOperand(i)->getType());
+        unsigned JCost = TTI->getIntrinsicInstrCost(IID, JT1, Tys);
+
+        Tys.clear();
+        assert(CI->getNumArgOperands() == CJ->getNumArgOperands() &&
+               "Intrinsic argument counts differ");
+        for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) {
+          if (IID == Intrinsic::powi && i == 1)
+            Tys.push_back(CI->getArgOperand(i)->getType());
+          else
+            Tys.push_back(getVecTypeForPair(CI->getArgOperand(i)->getType(),
+                                            CJ->getArgOperand(i)->getType()));
+        }
+
+        Type *RetTy = getVecTypeForPair(IT1, JT1);
+        unsigned VCost = TTI->getIntrinsicInstrCost(IID, RetTy, Tys);
+
+        if (VCost > ICost + JCost)
+          return false;
+
+        // We don't want to fuse to a type that will be split, even
+        // if the two input types will also be split and there is no other
+        // associated cost.
+        unsigned RetParts = TTI->getNumberOfParts(RetTy);
+        if (RetParts > 1)
+          return false;
+        else if (!RetParts && VCost == ICost + JCost)
+          return false;
+
+        for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) {
+          if (!Tys[i]->isVectorTy())
+            continue;
+
+          unsigned NumParts = TTI->getNumberOfParts(Tys[i]);
+          if (NumParts > 1)
+            return false;
+          else if (!NumParts && VCost == ICost + JCost)
+            return false;
+        }
+
+        CostSavings = ICost + JCost - VCost;
+      }
     }
 
     return true;
@@ -1040,7 +1111,7 @@ namespace {
   // to contain any memory locations to which J writes. The function returns
   // true if J uses I. By default, alias analysis is used to determine
   // whether J reads from memory that overlaps with a location in WriteSet.
-  // If LoadMoveSet is not null, then it is a previously-computed multimap
+  // If LoadMoveSet is not null, then it is a previously-computed map
   // where the key is the memory-based user instruction and the value is
   // the instruction to be compared with I. So, if LoadMoveSet is provided,
   // then the alias analysis is not used. This is necessary because this
@@ -1050,7 +1121,7 @@ namespace {
   bool BBVectorize::trackUsesOfI(DenseSet<Value *> &Users,
                        AliasSetTracker &WriteSet, Instruction *I,
                        Instruction *J, bool UpdateUsers,
-                       std::multimap<Value *, Value *> *LoadMoveSet) {
+                       DenseSet<ValuePair> *LoadMoveSetPairs) {
     bool UsesI = false;
 
     // This instruction may already be marked as a user due, for example, to
@@ -1068,9 +1139,8 @@ namespace {
         }
       }
     if (!UsesI && J->mayReadFromMemory()) {
-      if (LoadMoveSet) {
-        VPIteratorPair JPairRange = LoadMoveSet->equal_range(J);
-        UsesI = isSecondInIteratorPair<Value*>(I, JPairRange);
+      if (LoadMoveSetPairs) {
+        UsesI = LoadMoveSetPairs->count(ValuePair(J, I));
       } else {
         for (AliasSetTracker::iterator W = WriteSet.begin(),
              WE = WriteSet.end(); W != WE; ++W) {
@@ -1094,10 +1164,11 @@ namespace {
   // basic block and collects all candidate pairs for vectorization.
   bool BBVectorize::getCandidatePairs(BasicBlock &BB,
                        BasicBlock::iterator &Start,
-                       std::multimap<Value *, Value *> &CandidatePairs,
+                       DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
                        DenseSet<ValuePair> &FixedOrderPairs,
                        DenseMap<ValuePair, int> &CandidatePairCostSavings,
                        std::vector<Value *> &PairableInsts, bool NonPow2Len) {
+    size_t TotalPairs = 0;
     BasicBlock::iterator E = BB.end();
     if (Start == E) return false;
 
@@ -1143,8 +1214,9 @@ namespace {
           PairableInsts.push_back(I);
         }
 
-        CandidatePairs.insert(ValuePair(I, J));
-        if (VTTI)
+        CandidatePairs[I].push_back(J);
+        ++TotalPairs;
+        if (TTI)
           CandidatePairCostSavings.insert(ValuePairWithCost(ValuePair(I, J),
                                                             CostSavings));
 
@@ -1167,7 +1239,8 @@ namespace {
         // If we have already found too many pairs, break here and this function
         // will be called again starting after the last instruction selected
         // during this invocation.
-        if (PairableInsts.size() >= Config.MaxInsts) {
+        if (PairableInsts.size() >= Config.MaxInsts ||
+            TotalPairs >= Config.MaxPairs) {
           ShouldContinue = true;
           break;
         }
@@ -1187,11 +1260,12 @@ namespace {
   // it looks for pairs such that both members have an input which is an
   // output of PI or PJ.
   void BBVectorize::computePairsConnectedTo(
-                      std::multimap<Value *, Value *> &CandidatePairs,
-                      std::vector<Value *> &PairableInsts,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                      DenseMap<VPPair, unsigned> &PairConnectionTypes,
-                      ValuePair P) {
+                  DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+                  DenseSet<ValuePair> &CandidatePairsSet,
+                  std::vector<Value *> &PairableInsts,
+                  DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+                  DenseMap<VPPair, unsigned> &PairConnectionTypes,
+                  ValuePair P) {
     StoreInst *SI, *SJ;
 
     // For each possible pairing for this variable, look at the uses of
@@ -1209,8 +1283,6 @@ namespace {
         continue;
       }
 
-      VPIteratorPair IPairRange = CandidatePairs.equal_range(*I);
-
       // For each use of the first variable, look for uses of the second
       // variable...
       for (Value::use_iterator J = P.second->use_begin(),
@@ -1219,19 +1291,17 @@ namespace {
             P.second == SJ->getPointerOperand())
           continue;
 
-        VPIteratorPair JPairRange = CandidatePairs.equal_range(*J);
-
         // Look for <I, J>:
-        if (isSecondInIteratorPair<Value*>(*J, IPairRange)) {
+        if (CandidatePairsSet.count(ValuePair(*I, *J))) {
           VPPair VP(P, ValuePair(*I, *J));
-          ConnectedPairs.insert(VP);
+          ConnectedPairs[VP.first].push_back(VP.second);
           PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionDirect));
         }
 
         // Look for <J, I>:
-        if (isSecondInIteratorPair<Value*>(*I, JPairRange)) {
+        if (CandidatePairsSet.count(ValuePair(*J, *I))) {
           VPPair VP(P, ValuePair(*J, *I));
-          ConnectedPairs.insert(VP);
+          ConnectedPairs[VP.first].push_back(VP.second);
           PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionSwap));
         }
       }
@@ -1244,9 +1314,9 @@ namespace {
             P.first == SJ->getPointerOperand())
           continue;
 
-        if (isSecondInIteratorPair<Value*>(*J, IPairRange)) {
+        if (CandidatePairsSet.count(ValuePair(*I, *J))) {
           VPPair VP(P, ValuePair(*I, *J));
-          ConnectedPairs.insert(VP);
+          ConnectedPairs[VP.first].push_back(VP.second);
           PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionSplat));
         }
       }
@@ -1263,16 +1333,14 @@ namespace {
                P.second == SI->getPointerOperand())
         continue;
 
-      VPIteratorPair IPairRange = CandidatePairs.equal_range(*I);
-
       for (Value::use_iterator J = P.second->use_begin(); J != E; ++J) {
         if ((SJ = dyn_cast<StoreInst>(*J)) &&
             P.second == SJ->getPointerOperand())
           continue;
 
-        if (isSecondInIteratorPair<Value*>(*J, IPairRange)) {
+        if (CandidatePairsSet.count(ValuePair(*I, *J))) {
           VPPair VP(P, ValuePair(*I, *J));
-          ConnectedPairs.insert(VP);
+          ConnectedPairs[VP.first].push_back(VP.second);
           PairConnectionTypes.insert(VPPairWithType(VP, PairConnectionSplat));
         }
       }
@@ -1283,55 +1351,73 @@ namespace {
   // connected if some output of the first pair forms an input to both members
   // of the second pair.
   void BBVectorize::computeConnectedPairs(
-                      std::multimap<Value *, Value *> &CandidatePairs,
-                      std::vector<Value *> &PairableInsts,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                      DenseMap<VPPair, unsigned> &PairConnectionTypes) {
-
+                  DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+                  DenseSet<ValuePair> &CandidatePairsSet,
+                  std::vector<Value *> &PairableInsts,
+                  DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+                  DenseMap<VPPair, unsigned> &PairConnectionTypes) {
     for (std::vector<Value *>::iterator PI = PairableInsts.begin(),
          PE = PairableInsts.end(); PI != PE; ++PI) {
-      VPIteratorPair choiceRange = CandidatePairs.equal_range(*PI);
+      DenseMap<Value *, std::vector<Value *> >::iterator PP =
+        CandidatePairs.find(*PI);
+      if (PP == CandidatePairs.end())
+        continue;
 
-      for (std::multimap<Value *, Value *>::iterator P = choiceRange.first;
-           P != choiceRange.second; ++P)
-        computePairsConnectedTo(CandidatePairs, PairableInsts,
-                                ConnectedPairs, PairConnectionTypes, *P);
+      for (std::vector<Value *>::iterator P = PP->second.begin(),
+           E = PP->second.end(); P != E; ++P)
+        computePairsConnectedTo(CandidatePairs, CandidatePairsSet,
+                                PairableInsts, ConnectedPairs,
+                                PairConnectionTypes, ValuePair(*PI, *P));
     }
 
-    DEBUG(dbgs() << "BBV: found " << ConnectedPairs.size()
+    DEBUG(size_t TotalPairs = 0;
+          for (DenseMap<ValuePair, std::vector<ValuePair> >::iterator I =
+               ConnectedPairs.begin(), IE = ConnectedPairs.end(); I != IE; ++I)
+            TotalPairs += I->second.size();
+          dbgs() << "BBV: found " << TotalPairs
                  << " pair connections.\n");
   }
 
   // This function builds a set of use tuples such that <A, B> is in the set
-  // if B is in the use tree of A. If B is in the use tree of A, then B
+  // if B is in the use dag of A. If B is in the use dag of A, then B
   // depends on the output of A.
   void BBVectorize::buildDepMap(
                       BasicBlock &BB,
-                      std::multimap<Value *, Value *> &CandidatePairs,
+                      DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
                       std::vector<Value *> &PairableInsts,
                       DenseSet<ValuePair> &PairableInstUsers) {
     DenseSet<Value *> IsInPair;
-    for (std::multimap<Value *, Value *>::iterator C = CandidatePairs.begin(),
-         E = CandidatePairs.end(); C != E; ++C) {
+    for (DenseMap<Value *, std::vector<Value *> >::iterator C =
+         CandidatePairs.begin(), E = CandidatePairs.end(); C != E; ++C) {
       IsInPair.insert(C->first);
-      IsInPair.insert(C->second);
+      IsInPair.insert(C->second.begin(), C->second.end());
     }
 
-    // Iterate through the basic block, recording all Users of each
+    // Iterate through the basic block, recording all users of each
     // pairable instruction.
 
-    BasicBlock::iterator E = BB.end();
+    BasicBlock::iterator E = BB.end(), EL =
+      BasicBlock::iterator(cast<Instruction>(PairableInsts.back()));
     for (BasicBlock::iterator I = BB.getFirstInsertionPt(); I != E; ++I) {
       if (IsInPair.find(I) == IsInPair.end()) continue;
 
       DenseSet<Value *> Users;
       AliasSetTracker WriteSet(*AA);
-      for (BasicBlock::iterator J = llvm::next(I); J != E; ++J)
+      for (BasicBlock::iterator J = llvm::next(I); J != E; ++J) {
         (void) trackUsesOfI(Users, WriteSet, I, J);
 
+        if (J == EL)
+          break;
+      }
+
       for (DenseSet<Value *>::iterator U = Users.begin(), E = Users.end();
-           U != E; ++U)
+           U != E; ++U) {
+        if (IsInPair.find(*U) == IsInPair.end()) continue;
         PairableInstUsers.insert(ValuePair(I, *U));
+      }
+
+      if (I == EL)
+        break;
     }
   }
 
@@ -1339,8 +1425,9 @@ namespace {
   // input of pair Q is an output of pair P. If this is the case, then these
   // two pairs cannot be simultaneously fused.
   bool BBVectorize::pairsConflict(ValuePair P, ValuePair Q,
-                     DenseSet<ValuePair> &PairableInstUsers,
-                     std::multimap<ValuePair, ValuePair> *PairableInstUserMap) {
+             DenseSet<ValuePair> &PairableInstUsers,
+             DenseMap<ValuePair, std::vector<ValuePair> > *PairableInstUserMap,
+             DenseSet<VPPair> *PairableInstUserPairSet) {
     // Two pairs are in conflict if they are mutual Users of eachother.
     bool QUsesP = PairableInstUsers.count(ValuePair(P.first,  Q.first))  ||
                   PairableInstUsers.count(ValuePair(P.first,  Q.second)) ||
@@ -1353,17 +1440,14 @@ namespace {
     if (PairableInstUserMap) {
       // FIXME: The expensive part of the cycle check is not so much the cycle
       // check itself but this edge insertion procedure. This needs some
-      // profiling and probably a different data structure (same is true of
-      // most uses of std::multimap).
+      // profiling and probably a different data structure.
       if (PUsesQ) {
-        VPPIteratorPair QPairRange = PairableInstUserMap->equal_range(Q);
-        if (!isSecondInIteratorPair(P, QPairRange))
-          PairableInstUserMap->insert(VPPair(Q, P));
+        if (PairableInstUserPairSet->insert(VPPair(Q, P)).second)
+          (*PairableInstUserMap)[Q].push_back(P);
       }
       if (QUsesP) {
-        VPPIteratorPair PPairRange = PairableInstUserMap->equal_range(P);
-        if (!isSecondInIteratorPair(Q, PPairRange))
-          PairableInstUserMap->insert(VPPair(P, Q));
+        if (PairableInstUserPairSet->insert(VPPair(P, Q)).second)
+          (*PairableInstUserMap)[P].push_back(Q);
       }
     }
 
@@ -1373,8 +1457,8 @@ namespace {
   // This function walks the use graph of current pairs to see if, starting
   // from P, the walk returns to P.
   bool BBVectorize::pairWillFormCycle(ValuePair P,
-                       std::multimap<ValuePair, ValuePair> &PairableInstUserMap,
-                       DenseSet<ValuePair> &CurrentPairs) {
+             DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUserMap,
+             DenseSet<ValuePair> &CurrentPairs) {
     DEBUG(if (DebugCycleCheck)
             dbgs() << "BBV: starting cycle check for : " << *P.first << " <-> "
                    << *P.second << "\n");
@@ -1391,36 +1475,41 @@ namespace {
       DEBUG(if (DebugCycleCheck)
               dbgs() << "BBV: cycle check visiting: " << *QTop.first << " <-> "
                      << *QTop.second << "\n");
-      VPPIteratorPair QPairRange = PairableInstUserMap.equal_range(QTop);
-      for (std::multimap<ValuePair, ValuePair>::iterator C = QPairRange.first;
-           C != QPairRange.second; ++C) {
-        if (C->second == P) {
+      DenseMap<ValuePair, std::vector<ValuePair> >::iterator QQ =
+        PairableInstUserMap.find(QTop);
+      if (QQ == PairableInstUserMap.end())
+        continue;
+
+      for (std::vector<ValuePair>::iterator C = QQ->second.begin(),
+           CE = QQ->second.end(); C != CE; ++C) {
+        if (*C == P) {
           DEBUG(dbgs()
                  << "BBV: rejected to prevent non-trivial cycle formation: "
-                 << *C->first.first << " <-> " << *C->first.second << "\n");
+                 << QTop.first << " <-> " << C->second << "\n");
           return true;
         }
 
-        if (CurrentPairs.count(C->second) && !Visited.count(C->second))
-          Q.push_back(C->second);
+        if (CurrentPairs.count(*C) && !Visited.count(*C))
+          Q.push_back(*C);
       }
     } while (!Q.empty());
 
     return false;
   }
 
-  // This function builds the initial tree of connected pairs with the
+  // This function builds the initial dag of connected pairs with the
   // pair J at the root.
-  void BBVectorize::buildInitialTreeFor(
-                      std::multimap<Value *, Value *> &CandidatePairs,
-                      std::vector<Value *> &PairableInsts,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                      DenseSet<ValuePair> &PairableInstUsers,
-                      DenseMap<Value *, Value *> &ChosenPairs,
-                      DenseMap<ValuePair, size_t> &Tree, ValuePair J) {
-    // Each of these pairs is viewed as the root node of a Tree. The Tree
+  void BBVectorize::buildInitialDAGFor(
+                  DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+                  DenseSet<ValuePair> &CandidatePairsSet,
+                  std::vector<Value *> &PairableInsts,
+                  DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+                  DenseSet<ValuePair> &PairableInstUsers,
+                  DenseMap<Value *, Value *> &ChosenPairs,
+                  DenseMap<ValuePair, size_t> &DAG, ValuePair J) {
+    // Each of these pairs is viewed as the root node of a DAG. The DAG
     // is then walked (depth-first). As this happens, we keep track of
-    // the pairs that compose the Tree and the maximum depth of the Tree.
+    // the pairs that compose the DAG and the maximum depth of the DAG.
     SmallVector<ValuePairWithDepth, 32> Q;
     // General depth-first post-order traversal:
     Q.push_back(ValuePairWithDepth(J, getDepthFactor(J.first)));
@@ -1430,69 +1519,65 @@ namespace {
       // Push each child onto the queue:
       bool MoreChildren = false;
       size_t MaxChildDepth = QTop.second;
-      VPPIteratorPair qtRange = ConnectedPairs.equal_range(QTop.first);
-      for (std::multimap<ValuePair, ValuePair>::iterator k = qtRange.first;
-           k != qtRange.second; ++k) {
-        // Make sure that this child pair is still a candidate:
-        bool IsStillCand = false;
-        VPIteratorPair checkRange =
-          CandidatePairs.equal_range(k->second.first);
-        for (std::multimap<Value *, Value *>::iterator m = checkRange.first;
-             m != checkRange.second; ++m) {
-          if (m->second == k->second.second) {
-            IsStillCand = true;
-            break;
-          }
-        }
-
-        if (IsStillCand) {
-          DenseMap<ValuePair, size_t>::iterator C = Tree.find(k->second);
-          if (C == Tree.end()) {
-            size_t d = getDepthFactor(k->second.first);
-            Q.push_back(ValuePairWithDepth(k->second, QTop.second+d));
-            MoreChildren = true;
-          } else {
-            MaxChildDepth = std::max(MaxChildDepth, C->second);
+      DenseMap<ValuePair, std::vector<ValuePair> >::iterator QQ =
+        ConnectedPairs.find(QTop.first);
+      if (QQ != ConnectedPairs.end())
+        for (std::vector<ValuePair>::iterator k = QQ->second.begin(),
+             ke = QQ->second.end(); k != ke; ++k) {
+          // Make sure that this child pair is still a candidate:
+          if (CandidatePairsSet.count(*k)) {
+            DenseMap<ValuePair, size_t>::iterator C = DAG.find(*k);
+            if (C == DAG.end()) {
+              size_t d = getDepthFactor(k->first);
+              Q.push_back(ValuePairWithDepth(*k, QTop.second+d));
+              MoreChildren = true;
+            } else {
+              MaxChildDepth = std::max(MaxChildDepth, C->second);
+            }
           }
         }
-      }
 
       if (!MoreChildren) {
-        // Record the current pair as part of the Tree:
-        Tree.insert(ValuePairWithDepth(QTop.first, MaxChildDepth));
+        // Record the current pair as part of the DAG:
+        DAG.insert(ValuePairWithDepth(QTop.first, MaxChildDepth));
         Q.pop_back();
       }
     } while (!Q.empty());
   }
 
-  // Given some initial tree, prune it by removing conflicting pairs (pairs
+  // Given some initial dag, prune it by removing conflicting pairs (pairs
   // that cannot be simultaneously chosen for vectorization).
-  void BBVectorize::pruneTreeFor(
-                      std::multimap<Value *, Value *> &CandidatePairs,
-                      std::vector<Value *> &PairableInsts,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                      DenseSet<ValuePair> &PairableInstUsers,
-                      std::multimap<ValuePair, ValuePair> &PairableInstUserMap,
-                      DenseMap<Value *, Value *> &ChosenPairs,
-                      DenseMap<ValuePair, size_t> &Tree,
-                      DenseSet<ValuePair> &PrunedTree, ValuePair J,
-                      bool UseCycleCheck) {
+  void BBVectorize::pruneDAGFor(
+              DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+              std::vector<Value *> &PairableInsts,
+              DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+              DenseSet<ValuePair> &PairableInstUsers,
+              DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUserMap,
+              DenseSet<VPPair> &PairableInstUserPairSet,
+              DenseMap<Value *, Value *> &ChosenPairs,
+              DenseMap<ValuePair, size_t> &DAG,
+              DenseSet<ValuePair> &PrunedDAG, ValuePair J,
+              bool UseCycleCheck) {
     SmallVector<ValuePairWithDepth, 32> Q;
     // General depth-first post-order traversal:
     Q.push_back(ValuePairWithDepth(J, getDepthFactor(J.first)));
     do {
       ValuePairWithDepth QTop = Q.pop_back_val();
-      PrunedTree.insert(QTop.first);
+      PrunedDAG.insert(QTop.first);
 
       // Visit each child, pruning as necessary...
       SmallVector<ValuePairWithDepth, 8> BestChildren;
-      VPPIteratorPair QTopRange = ConnectedPairs.equal_range(QTop.first);
-      for (std::multimap<ValuePair, ValuePair>::iterator K = QTopRange.first;
-           K != QTopRange.second; ++K) {
-        DenseMap<ValuePair, size_t>::iterator C = Tree.find(K->second);
-        if (C == Tree.end()) continue;
+      DenseMap<ValuePair, std::vector<ValuePair> >::iterator QQ =
+        ConnectedPairs.find(QTop.first);
+      if (QQ == ConnectedPairs.end())
+        continue;
+
+      for (std::vector<ValuePair>::iterator K = QQ->second.begin(),
+           KE = QQ->second.end(); K != KE; ++K) {
+        DenseMap<ValuePair, size_t>::iterator C = DAG.find(*K);
+        if (C == DAG.end()) continue;
 
-        // This child is in the Tree, now we need to make sure it is the
+        // This child is in the DAG, now we need to make sure it is the
         // best of any conflicting children. There could be multiple
         // conflicting children, so first, determine if we're keeping
         // this child, then delete conflicting children as necessary.
@@ -1506,7 +1591,7 @@ namespace {
         // fusing (a,b) we have y .. a/b .. x where y is an input
         // to a/b and x is an output to a/b: x and y can no longer
         // be legally fused. To prevent this condition, we must
-        // make sure that a child pair added to the Tree is not
+        // make sure that a child pair added to the DAG is not
         // both an input and output of an already-selected pair.
 
         // Pairing-induced dependencies can also form from more complicated
@@ -1525,7 +1610,8 @@ namespace {
               C2->first.second == C->first.first ||
               C2->first.second == C->first.second ||
               pairsConflict(C2->first, C->first, PairableInstUsers,
-                            UseCycleCheck ? &PairableInstUserMap : 0)) {
+                            UseCycleCheck ? &PairableInstUserMap : 0,
+                            UseCycleCheck ? &PairableInstUserPairSet : 0)) {
             if (C2->second >= C->second) {
               CanAdd = false;
               break;
@@ -1537,15 +1623,16 @@ namespace {
         if (!CanAdd) continue;
 
         // Even worse, this child could conflict with another node already
-        // selected for the Tree. If that is the case, ignore this child.
-        for (DenseSet<ValuePair>::iterator T = PrunedTree.begin(),
-             E2 = PrunedTree.end(); T != E2; ++T) {
+        // selected for the DAG. If that is the case, ignore this child.
+        for (DenseSet<ValuePair>::iterator T = PrunedDAG.begin(),
+             E2 = PrunedDAG.end(); T != E2; ++T) {
           if (T->first == C->first.first ||
               T->first == C->first.second ||
               T->second == C->first.first ||
               T->second == C->first.second ||
               pairsConflict(*T, C->first, PairableInstUsers,
-                            UseCycleCheck ? &PairableInstUserMap : 0)) {
+                            UseCycleCheck ? &PairableInstUserMap : 0,
+                            UseCycleCheck ? &PairableInstUserPairSet : 0)) {
             CanAdd = false;
             break;
           }
@@ -1562,7 +1649,8 @@ namespace {
               C2->first.second == C->first.first ||
               C2->first.second == C->first.second ||
               pairsConflict(C2->first, C->first, PairableInstUsers,
-                            UseCycleCheck ? &PairableInstUserMap : 0)) {
+                            UseCycleCheck ? &PairableInstUserMap : 0,
+                            UseCycleCheck ? &PairableInstUserPairSet : 0)) {
             CanAdd = false;
             break;
           }
@@ -1577,7 +1665,8 @@ namespace {
               ChosenPairs.begin(), E2 = ChosenPairs.end();
              C2 != E2; ++C2) {
           if (pairsConflict(*C2, C->first, PairableInstUsers,
-                            UseCycleCheck ? &PairableInstUserMap : 0)) {
+                            UseCycleCheck ? &PairableInstUserMap : 0,
+                            UseCycleCheck ? &PairableInstUserPairSet : 0)) {
             CanAdd = false;
             break;
           }
@@ -1589,7 +1678,7 @@ namespace {
         // To check for non-trivial cycles formed by the addition of the
         // current pair we've formed a list of all relevant pairs, now use a
         // graph walk to check for a cycle. We start from the current pair and
-        // walk the use tree to see if we again reach the current pair. If we
+        // walk the use dag to see if we again reach the current pair. If we
         // do, then the current pair is rejected.
 
         // FIXME: It may be more efficient to use a topological-ordering
@@ -1626,34 +1715,40 @@ namespace {
     } while (!Q.empty());
   }
 
-  // This function finds the best tree of mututally-compatible connected
+  // This function finds the best dag of mututally-compatible connected
   // pairs, given the choice of root pairs as an iterator range.
-  void BBVectorize::findBestTreeFor(
-                      std::multimap<Value *, Value *> &CandidatePairs,
-                      DenseMap<ValuePair, int> &CandidatePairCostSavings,
-                      std::vector<Value *> &PairableInsts,
-                      DenseSet<ValuePair> &FixedOrderPairs,
-                      DenseMap<VPPair, unsigned> &PairConnectionTypes,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairDeps,
-                      DenseSet<ValuePair> &PairableInstUsers,
-                      std::multimap<ValuePair, ValuePair> &PairableInstUserMap,
-                      DenseMap<Value *, Value *> &ChosenPairs,
-                      DenseSet<ValuePair> &BestTree, size_t &BestMaxDepth,
-                      int &BestEffSize, VPIteratorPair ChoiceRange,
-                      bool UseCycleCheck) {
-    for (std::multimap<Value *, Value *>::iterator J = ChoiceRange.first;
-         J != ChoiceRange.second; ++J) {
+  void BBVectorize::findBestDAGFor(
+              DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+              DenseSet<ValuePair> &CandidatePairsSet,
+              DenseMap<ValuePair, int> &CandidatePairCostSavings,
+              std::vector<Value *> &PairableInsts,
+              DenseSet<ValuePair> &FixedOrderPairs,
+              DenseMap<VPPair, unsigned> &PairConnectionTypes,
+              DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+              DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps,
+              DenseSet<ValuePair> &PairableInstUsers,
+              DenseMap<ValuePair, std::vector<ValuePair> > &PairableInstUserMap,
+              DenseSet<VPPair> &PairableInstUserPairSet,
+              DenseMap<Value *, Value *> &ChosenPairs,
+              DenseSet<ValuePair> &BestDAG, size_t &BestMaxDepth,
+              int &BestEffSize, Value *II, std::vector<Value *>&JJ,
+              bool UseCycleCheck) {
+    for (std::vector<Value *>::iterator J = JJ.begin(), JE = JJ.end();
+         J != JE; ++J) {
+      ValuePair IJ(II, *J);
+      if (!CandidatePairsSet.count(IJ))
+        continue;
 
       // Before going any further, make sure that this pair does not
       // conflict with any already-selected pairs (see comment below
-      // near the Tree pruning for more details).
+      // near the DAG pruning for more details).
       DenseSet<ValuePair> ChosenPairSet;
       bool DoesConflict = false;
       for (DenseMap<Value *, Value *>::iterator C = ChosenPairs.begin(),
            E = ChosenPairs.end(); C != E; ++C) {
-        if (pairsConflict(*C, *J, PairableInstUsers,
-                          UseCycleCheck ? &PairableInstUserMap : 0)) {
+        if (pairsConflict(*C, IJ, PairableInstUsers,
+                          UseCycleCheck ? &PairableInstUserMap : 0,
+                          UseCycleCheck ? &PairableInstUserPairSet : 0)) {
           DoesConflict = true;
           break;
         }
@@ -1663,40 +1758,42 @@ namespace {
       if (DoesConflict) continue;
 
       if (UseCycleCheck &&
-          pairWillFormCycle(*J, PairableInstUserMap, ChosenPairSet))
+          pairWillFormCycle(IJ, PairableInstUserMap, ChosenPairSet))
         continue;
 
-      DenseMap<ValuePair, size_t> Tree;
-      buildInitialTreeFor(CandidatePairs, PairableInsts, ConnectedPairs,
-                          PairableInstUsers, ChosenPairs, Tree, *J);
+      DenseMap<ValuePair, size_t> DAG;
+      buildInitialDAGFor(CandidatePairs, CandidatePairsSet,
+                          PairableInsts, ConnectedPairs,
+                          PairableInstUsers, ChosenPairs, DAG, IJ);
 
       // Because we'll keep the child with the largest depth, the largest
-      // depth is still the same in the unpruned Tree.
-      size_t MaxDepth = Tree.lookup(*J);
+      // depth is still the same in the unpruned DAG.
+      size_t MaxDepth = DAG.lookup(IJ);
 
-      DEBUG(if (DebugPairSelection) dbgs() << "BBV: found Tree for pair {"
-                   << *J->first << " <-> " << *J->second << "} of depth " <<
-                   MaxDepth << " and size " << Tree.size() << "\n");
+      DEBUG(if (DebugPairSelection) dbgs() << "BBV: found DAG for pair {"
+                   << *IJ.first << " <-> " << *IJ.second << "} of depth " <<
+                   MaxDepth << " and size " << DAG.size() << "\n");
 
-      // At this point the Tree has been constructed, but, may contain
+      // At this point the DAG has been constructed, but, may contain
       // contradictory children (meaning that different children of
-      // some tree node may be attempting to fuse the same instruction).
-      // So now we walk the tree again, in the case of a conflict,
+      // some dag node may be attempting to fuse the same instruction).
+      // So now we walk the dag again, in the case of a conflict,
       // keep only the child with the largest depth. To break a tie,
       // favor the first child.
 
-      DenseSet<ValuePair> PrunedTree;
-      pruneTreeFor(CandidatePairs, PairableInsts, ConnectedPairs,
-                   PairableInstUsers, PairableInstUserMap, ChosenPairs, Tree,
-                   PrunedTree, *J, UseCycleCheck);
+      DenseSet<ValuePair> PrunedDAG;
+      pruneDAGFor(CandidatePairs, PairableInsts, ConnectedPairs,
+                   PairableInstUsers, PairableInstUserMap,
+                   PairableInstUserPairSet,
+                   ChosenPairs, DAG, PrunedDAG, IJ, UseCycleCheck);
 
       int EffSize = 0;
-      if (VTTI) {
-        DenseSet<Value *> PrunedTreeInstrs;
-        for (DenseSet<ValuePair>::iterator S = PrunedTree.begin(),
-             E = PrunedTree.end(); S != E; ++S) {
-          PrunedTreeInstrs.insert(S->first);
-          PrunedTreeInstrs.insert(S->second);
+      if (TTI) {
+        DenseSet<Value *> PrunedDAGInstrs;
+        for (DenseSet<ValuePair>::iterator S = PrunedDAG.begin(),
+             E = PrunedDAG.end(); S != E; ++S) {
+          PrunedDAGInstrs.insert(S->first);
+          PrunedDAGInstrs.insert(S->second);
         }
 
         // The set of pairs that have already contributed to the total cost.
@@ -1709,8 +1806,8 @@ namespace {
 
         // The node weights represent the cost savings associated with
         // fusing the pair of instructions.
-        for (DenseSet<ValuePair>::iterator S = PrunedTree.begin(),
-             E = PrunedTree.end(); S != E; ++S) {
+        for (DenseSet<ValuePair>::iterator S = PrunedDAG.begin(),
+             E = PrunedDAG.end(); S != E; ++S) {
           if (!isa<ShuffleVectorInst>(S->first) &&
               !isa<InsertElementInst>(S->first) &&
               !isa<ExtractElementInst>(S->first))
@@ -1728,15 +1825,17 @@ namespace {
 
           // The edge weights contribute in a negative sense: they represent
           // the cost of shuffles.
-          VPPIteratorPair IP = ConnectedPairDeps.equal_range(*S);
-          if (IP.first != ConnectedPairDeps.end()) {
+          DenseMap<ValuePair, std::vector<ValuePair> >::iterator SS =
+            ConnectedPairDeps.find(*S);
+          if (SS != ConnectedPairDeps.end()) {
             unsigned NumDepsDirect = 0, NumDepsSwap = 0;
-            for (std::multimap<ValuePair, ValuePair>::iterator Q = IP.first;
-                 Q != IP.second; ++Q) {
-              if (!PrunedTree.count(Q->second))
+            for (std::vector<ValuePair>::iterator T = SS->second.begin(),
+                 TE = SS->second.end(); T != TE; ++T) {
+              VPPair Q(*S, *T);
+              if (!PrunedDAG.count(Q.second))
                 continue;
               DenseMap<VPPair, unsigned>::iterator R =
-                PairConnectionTypes.find(VPPair(Q->second, Q->first));
+                PairConnectionTypes.find(VPPair(Q.second, Q.first));
               assert(R != PairConnectionTypes.end() &&
                      "Cannot find pair connection type");
               if (R->second == PairConnectionDirect)
@@ -1752,24 +1851,35 @@ namespace {
               ((NumDepsSwap > NumDepsDirect) ||
                 FixedOrderPairs.count(ValuePair(S->second, S->first)));
 
-            for (std::multimap<ValuePair, ValuePair>::iterator Q = IP.first;
-                 Q != IP.second; ++Q) {
-              if (!PrunedTree.count(Q->second))
+            for (std::vector<ValuePair>::iterator T = SS->second.begin(),
+                 TE = SS->second.end(); T != TE; ++T) {
+              VPPair Q(*S, *T);
+              if (!PrunedDAG.count(Q.second))
                 continue;
               DenseMap<VPPair, unsigned>::iterator R =
-                PairConnectionTypes.find(VPPair(Q->second, Q->first));
+                PairConnectionTypes.find(VPPair(Q.second, Q.first));
               assert(R != PairConnectionTypes.end() &&
                      "Cannot find pair connection type");
-              Type *Ty1 = Q->second.first->getType(),
-                   *Ty2 = Q->second.second->getType();
+              Type *Ty1 = Q.second.first->getType(),
+                   *Ty2 = Q.second.second->getType();
               Type *VTy = getVecTypeForPair(Ty1, Ty2);
               if ((R->second == PairConnectionDirect && FlipOrder) ||
                   (R->second == PairConnectionSwap && !FlipOrder)  ||
                   R->second == PairConnectionSplat) {
                 int ESContrib = (int) getInstrCost(Instruction::ShuffleVector,
                                                    VTy, VTy);
+
+                if (VTy->getVectorNumElements() == 2) {
+                  if (R->second == PairConnectionSplat)
+                    ESContrib = std::min(ESContrib, (int) TTI->getShuffleCost(
+                      TargetTransformInfo::SK_Broadcast, VTy));
+                  else
+                    ESContrib = std::min(ESContrib, (int) TTI->getShuffleCost(
+                      TargetTransformInfo::SK_Reverse, VTy));
+                }
+
                 DEBUG(if (DebugPairSelection) dbgs() << "\tcost {" <<
-                  *Q->second.first << " <-> " << *Q->second.second <<
+                  *Q.second.first << " <-> " << *Q.second.second <<
                     "} -> {" <<
                   *S->first << " <-> " << *S->second << "} = " <<
                    ESContrib << "\n");
@@ -1796,7 +1906,7 @@ namespace {
               }
               if (isa<ExtractElementInst>(*I))
                 continue;
-              if (PrunedTreeInstrs.count(*I))
+              if (PrunedDAGInstrs.count(*I))
                 continue;
               NeedsExtraction = true;
               break;
@@ -1804,11 +1914,13 @@ namespace {
 
             if (NeedsExtraction) {
               int ESContrib;
-              if (Ty1->isVectorTy())
+              if (Ty1->isVectorTy()) {
                 ESContrib = (int) getInstrCost(Instruction::ShuffleVector,
                                                Ty1, VTy);
-              else
-                ESContrib = (int) VTTI->getVectorInstrCost(
+                ESContrib = std::min(ESContrib, (int) TTI->getShuffleCost(
+                  TargetTransformInfo::SK_ExtractSubvector, VTy, 0, Ty1));
+              } else
+                ESContrib = (int) TTI->getVectorInstrCost(
                                     Instruction::ExtractElement, VTy, 0);
 
               DEBUG(if (DebugPairSelection) dbgs() << "\tcost {" <<
@@ -1826,7 +1938,7 @@ namespace {
               }
               if (isa<ExtractElementInst>(*I))
                 continue;
-              if (PrunedTreeInstrs.count(*I))
+              if (PrunedDAGInstrs.count(*I))
                 continue;
               NeedsExtraction = true;
               break;
@@ -1834,11 +1946,14 @@ namespace {
 
             if (NeedsExtraction) {
               int ESContrib;
-              if (Ty2->isVectorTy())
+              if (Ty2->isVectorTy()) {
                 ESContrib = (int) getInstrCost(Instruction::ShuffleVector,
                                                Ty2, VTy);
-              else
-                ESContrib = (int) VTTI->getVectorInstrCost(
+                ESContrib = std::min(ESContrib, (int) TTI->getShuffleCost(
+                  TargetTransformInfo::SK_ExtractSubvector, VTy,
+                  Ty1->isVectorTy() ? Ty1->getVectorNumElements() : 1, Ty2));
+              } else
+                ESContrib = (int) TTI->getVectorInstrCost(
                                     Instruction::ExtractElement, VTy, 1);
               DEBUG(if (DebugPairSelection) dbgs() << "\tcost {" <<
                 *S->second << "} = " << ESContrib << "\n");
@@ -1865,7 +1980,7 @@ namespace {
               ValuePair VPR = ValuePair(O2, O1);
 
               // Internal edges are not handled here.
-              if (PrunedTree.count(VP) || PrunedTree.count(VPR))
+              if (PrunedDAG.count(VP) || PrunedDAG.count(VPR))
                 continue;
 
               Type *Ty1 = O1->getType(),
@@ -1913,22 +2028,26 @@ namespace {
               } else if (IncomingPairs.count(VPR)) {
                 ESContrib = (int) getInstrCost(Instruction::ShuffleVector,
                                                VTy, VTy);
+
+                if (VTy->getVectorNumElements() == 2)
+                  ESContrib = std::min(ESContrib, (int) TTI->getShuffleCost(
+                    TargetTransformInfo::SK_Reverse, VTy));
               } else if (!Ty1->isVectorTy() && !Ty2->isVectorTy()) {
-                ESContrib = (int) VTTI->getVectorInstrCost(
+                ESContrib = (int) TTI->getVectorInstrCost(
                                     Instruction::InsertElement, VTy, 0);
-                ESContrib += (int) VTTI->getVectorInstrCost(
+                ESContrib += (int) TTI->getVectorInstrCost(
                                      Instruction::InsertElement, VTy, 1);
               } else if (!Ty1->isVectorTy()) {
                 // O1 needs to be inserted into a vector of size O2, and then
                 // both need to be shuffled together.
-                ESContrib = (int) VTTI->getVectorInstrCost(
+                ESContrib = (int) TTI->getVectorInstrCost(
                                     Instruction::InsertElement, Ty2, 0);
                 ESContrib += (int) getInstrCost(Instruction::ShuffleVector,
                                                 VTy, Ty2);
               } else if (!Ty2->isVectorTy()) {
                 // O2 needs to be inserted into a vector of size O1, and then
                 // both need to be shuffled together.
-                ESContrib = (int) VTTI->getVectorInstrCost(
+                ESContrib = (int) TTI->getVectorInstrCost(
                                     Instruction::InsertElement, Ty1, 0);
                 ESContrib += (int) getInstrCost(Instruction::ShuffleVector,
                                                 VTy, Ty1);
@@ -1955,27 +2074,27 @@ namespace {
 
         if (!HasNontrivialInsts) {
           DEBUG(if (DebugPairSelection) dbgs() <<
-                "\tNo non-trivial instructions in tree;"
+                "\tNo non-trivial instructions in DAG;"
                 " override to zero effective size\n");
           EffSize = 0;
         }
       } else {
-        for (DenseSet<ValuePair>::iterator S = PrunedTree.begin(),
-             E = PrunedTree.end(); S != E; ++S)
+        for (DenseSet<ValuePair>::iterator S = PrunedDAG.begin(),
+             E = PrunedDAG.end(); S != E; ++S)
           EffSize += (int) getDepthFactor(S->first);
       }
 
       DEBUG(if (DebugPairSelection)
-             dbgs() << "BBV: found pruned Tree for pair {"
-             << *J->first << " <-> " << *J->second << "} of depth " <<
-             MaxDepth << " and size " << PrunedTree.size() <<
+             dbgs() << "BBV: found pruned DAG for pair {"
+             << *IJ.first << " <-> " << *IJ.second << "} of depth " <<
+             MaxDepth << " and size " << PrunedDAG.size() <<
             " (effective size: " << EffSize << ")\n");
-      if (((VTTI && !UseChainDepthWithTI) ||
+      if (((TTI && !UseChainDepthWithTI) ||
             MaxDepth >= Config.ReqChainDepth) &&
           EffSize > 0 && EffSize > BestEffSize) {
         BestMaxDepth = MaxDepth;
         BestEffSize = EffSize;
-        BestTree = PrunedTree;
+        BestDAG = PrunedDAG;
       }
     }
   }
@@ -1983,66 +2102,98 @@ namespace {
   // Given the list of candidate pairs, this function selects those
   // that will be fused into vector instructions.
   void BBVectorize::choosePairs(
-                      std::multimap<Value *, Value *> &CandidatePairs,
-                      DenseMap<ValuePair, int> &CandidatePairCostSavings,
-                      std::vector<Value *> &PairableInsts,
-                      DenseSet<ValuePair> &FixedOrderPairs,
-                      DenseMap<VPPair, unsigned> &PairConnectionTypes,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                      std::multimap<ValuePair, ValuePair> &ConnectedPairDeps,
-                      DenseSet<ValuePair> &PairableInstUsers,
-                      DenseMap<Value *, Value *>& ChosenPairs) {
+                DenseMap<Value *, std::vector<Value *> > &CandidatePairs,
+                DenseSet<ValuePair> &CandidatePairsSet,
+                DenseMap<ValuePair, int> &CandidatePairCostSavings,
+                std::vector<Value *> &PairableInsts,
+                DenseSet<ValuePair> &FixedOrderPairs,
+                DenseMap<VPPair, unsigned> &PairConnectionTypes,
+                DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+                DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps,
+                DenseSet<ValuePair> &PairableInstUsers,
+                DenseMap<Value *, Value *>& ChosenPairs) {
     bool UseCycleCheck =
-     CandidatePairs.size() <= Config.MaxCandPairsForCycleCheck;
-    std::multimap<ValuePair, ValuePair> PairableInstUserMap;
+     CandidatePairsSet.size() <= Config.MaxCandPairsForCycleCheck;
+
+    DenseMap<Value *, std::vector<Value *> > CandidatePairs2;
+    for (DenseSet<ValuePair>::iterator I = CandidatePairsSet.begin(),
+         E = CandidatePairsSet.end(); I != E; ++I) {
+      std::vector<Value *> &JJ = CandidatePairs2[I->second];
+      if (JJ.empty()) JJ.reserve(32);
+      JJ.push_back(I->first);
+    }
+
+    DenseMap<ValuePair, std::vector<ValuePair> > PairableInstUserMap;
+    DenseSet<VPPair> PairableInstUserPairSet;
     for (std::vector<Value *>::iterator I = PairableInsts.begin(),
          E = PairableInsts.end(); I != E; ++I) {
       // The number of possible pairings for this variable:
-      size_t NumChoices = CandidatePairs.count(*I);
+      size_t NumChoices = CandidatePairs.lookup(*I).size();
       if (!NumChoices) continue;
 
-      VPIteratorPair ChoiceRange = CandidatePairs.equal_range(*I);
+      std::vector<Value *> &JJ = CandidatePairs[*I];
 
-      // The best pair to choose and its tree:
+      // The best pair to choose and its dag:
       size_t BestMaxDepth = 0;
       int BestEffSize = 0;
-      DenseSet<ValuePair> BestTree;
-      findBestTreeFor(CandidatePairs, CandidatePairCostSavings,
+      DenseSet<ValuePair> BestDAG;
+      findBestDAGFor(CandidatePairs, CandidatePairsSet,
+                      CandidatePairCostSavings,
                       PairableInsts, FixedOrderPairs, PairConnectionTypes,
                       ConnectedPairs, ConnectedPairDeps,
-                      PairableInstUsers, PairableInstUserMap, ChosenPairs,
-                      BestTree, BestMaxDepth, BestEffSize, ChoiceRange,
+                      PairableInstUsers, PairableInstUserMap,
+                      PairableInstUserPairSet, ChosenPairs,
+                      BestDAG, BestMaxDepth, BestEffSize, *I, JJ,
                       UseCycleCheck);
 
-      // A tree has been chosen (or not) at this point. If no tree was
+      if (BestDAG.empty())
+        continue;
+
+      // A dag has been chosen (or not) at this point. If no dag was
       // chosen, then this instruction, I, cannot be paired (and is no longer
       // considered).
 
-      DEBUG(if (BestTree.size() > 0)
-              dbgs() << "BBV: selected pairs in the best tree for: "
-                     << *cast<Instruction>(*I) << "\n");
+      DEBUG(dbgs() << "BBV: selected pairs in the best DAG for: "
+                   << *cast<Instruction>(*I) << "\n");
 
-      for (DenseSet<ValuePair>::iterator S = BestTree.begin(),
-           SE2 = BestTree.end(); S != SE2; ++S) {
-        // Insert the members of this tree into the list of chosen pairs.
+      for (DenseSet<ValuePair>::iterator S = BestDAG.begin(),
+           SE2 = BestDAG.end(); S != SE2; ++S) {
+        // Insert the members of this dag into the list of chosen pairs.
         ChosenPairs.insert(ValuePair(S->first, S->second));
         DEBUG(dbgs() << "BBV: selected pair: " << *S->first << " <-> " <<
                *S->second << "\n");
 
-        // Remove all candidate pairs that have values in the chosen tree.
-        for (std::multimap<Value *, Value *>::iterator K =
-               CandidatePairs.begin(); K != CandidatePairs.end();) {
-          if (K->first == S->first || K->second == S->first ||
-              K->second == S->second || K->first == S->second) {
-            // Don't remove the actual pair chosen so that it can be used
-            // in subsequent tree selections.
-            if (!(K->first == S->first && K->second == S->second))
-              CandidatePairs.erase(K++);
-            else
-              ++K;
-          } else {
-            ++K;
-          }
+        // Remove all candidate pairs that have values in the chosen dag.
+        std::vector<Value *> &KK = CandidatePairs[S->first];
+        for (std::vector<Value *>::iterator K = KK.begin(), KE = KK.end();
+             K != KE; ++K) {
+          if (*K == S->second)
+            continue;
+
+          CandidatePairsSet.erase(ValuePair(S->first, *K));
+        }
+
+        std::vector<Value *> &LL = CandidatePairs2[S->second];
+        for (std::vector<Value *>::iterator L = LL.begin(), LE = LL.end();
+             L != LE; ++L) {
+          if (*L == S->first)
+            continue;
+
+          CandidatePairsSet.erase(ValuePair(*L, S->second));
+        }
+
+        std::vector<Value *> &MM = CandidatePairs[S->second];
+        for (std::vector<Value *>::iterator M = MM.begin(), ME = MM.end();
+             M != ME; ++M) {
+          assert(*M != S->first && "Flipped pair in candidate list?");
+          CandidatePairsSet.erase(ValuePair(S->second, *M));
+        }
+
+        std::vector<Value *> &NN = CandidatePairs2[S->first];
+        for (std::vector<Value *>::iterator N = NN.begin(), NE = NN.end();
+             N != NE; ++N) {
+          assert(*N != S->second && "Flipped pair in candidate list?");
+          CandidatePairsSet.erase(ValuePair(*N, S->first));
         }
       }
     }
@@ -2550,7 +2701,7 @@ namespace {
         continue;
       } else if (isa<CallInst>(I)) {
         Function *F = cast<CallInst>(I)->getCalledFunction();
-        unsigned IID = F->getIntrinsicID();
+        Intrinsic::ID IID = (Intrinsic::ID) F->getIntrinsicID();
         if (o == NumOperands-1) {
           BasicBlock &BB = *I->getParent();
 
@@ -2559,8 +2710,7 @@ namespace {
           Type *ArgTypeJ = J->getType();
           Type *VArgType = getVecTypeForPair(ArgTypeI, ArgTypeJ);
 
-          ReplacedOperands[o] = Intrinsic::getDeclaration(M,
-            (Intrinsic::ID) IID, VArgType);
+          ReplacedOperands[o] = Intrinsic::getDeclaration(M, IID, VArgType);
           continue;
         } else if (IID == Intrinsic::powi && o == 1) {
           // The second argument of powi is a single integer and we've already
@@ -2647,7 +2797,7 @@ namespace {
 
   // Move all uses of the function I (including pairing-induced uses) after J.
   bool BBVectorize::canMoveUsesOfIAfterJ(BasicBlock &BB,
-                     std::multimap<Value *, Value *> &LoadMoveSet,
+                     DenseSet<ValuePair> &LoadMoveSetPairs,
                      Instruction *I, Instruction *J) {
     // Skip to the first instruction past I.
     BasicBlock::iterator L = llvm::next(BasicBlock::iterator(I));
@@ -2655,18 +2805,18 @@ namespace {
     DenseSet<Value *> Users;
     AliasSetTracker WriteSet(*AA);
     for (; cast<Instruction>(L) != J; ++L)
-      (void) trackUsesOfI(Users, WriteSet, I, L, true, &LoadMoveSet);
+      (void) trackUsesOfI(Users, WriteSet, I, L, true, &LoadMoveSetPairs);
 
     assert(cast<Instruction>(L) == J &&
       "Tracking has not proceeded far enough to check for dependencies");
     // If J is now in the use set of I, then trackUsesOfI will return true
     // and we have a dependency cycle (and the fusing operation must abort).
-    return !trackUsesOfI(Users, WriteSet, I, J, true, &LoadMoveSet);
+    return !trackUsesOfI(Users, WriteSet, I, J, true, &LoadMoveSetPairs);
   }
 
   // Move all uses of the function I (including pairing-induced uses) after J.
   void BBVectorize::moveUsesOfIAfterJ(BasicBlock &BB,
-                     std::multimap<Value *, Value *> &LoadMoveSet,
+                     DenseSet<ValuePair> &LoadMoveSetPairs,
                      Instruction *&InsertionPt,
                      Instruction *I, Instruction *J) {
     // Skip to the first instruction past I.
@@ -2675,7 +2825,7 @@ namespace {
     DenseSet<Value *> Users;
     AliasSetTracker WriteSet(*AA);
     for (; cast<Instruction>(L) != J;) {
-      if (trackUsesOfI(Users, WriteSet, I, L, true, &LoadMoveSet)) {
+      if (trackUsesOfI(Users, WriteSet, I, L, true, &LoadMoveSetPairs)) {
         // Move this instruction
         Instruction *InstToMove = L; ++L;
 
@@ -2695,7 +2845,8 @@ namespace {
   // to be moved after J (the second instruction) when the pair is fused.
   void BBVectorize::collectPairLoadMoveSet(BasicBlock &BB,
                      DenseMap<Value *, Value *> &ChosenPairs,
-                     std::multimap<Value *, Value *> &LoadMoveSet,
+                     DenseMap<Value *, std::vector<Value *> > &LoadMoveSet,
+                     DenseSet<ValuePair> &LoadMoveSetPairs,
                      Instruction *I) {
     // Skip to the first instruction past I.
     BasicBlock::iterator L = llvm::next(BasicBlock::iterator(I));
@@ -2708,8 +2859,10 @@ namespace {
     // could be before I if this is an inverted input.
     for (BasicBlock::iterator E = BB.end(); cast<Instruction>(L) != E; ++L) {
       if (trackUsesOfI(Users, WriteSet, I, L)) {
-        if (L->mayReadFromMemory())
-          LoadMoveSet.insert(ValuePair(L, I));
+        if (L->mayReadFromMemory()) {
+          LoadMoveSet[L].push_back(I);
+          LoadMoveSetPairs.insert(ValuePair(L, I));
+        }
       }
     }
   }
@@ -2718,20 +2871,22 @@ namespace {
   // are chosen for vectorization, we can end up in a situation where the
   // aliasing analysis starts returning different query results as the
   // process of fusing instruction pairs continues. Because the algorithm
-  // relies on finding the same use trees here as were found earlier, we'll
+  // relies on finding the same use dags here as were found earlier, we'll
   // need to precompute the necessary aliasing information here and then
   // manually update it during the fusion process.
   void BBVectorize::collectLoadMoveSet(BasicBlock &BB,
                      std::vector<Value *> &PairableInsts,
                      DenseMap<Value *, Value *> &ChosenPairs,
-                     std::multimap<Value *, Value *> &LoadMoveSet) {
+                     DenseMap<Value *, std::vector<Value *> > &LoadMoveSet,
+                     DenseSet<ValuePair> &LoadMoveSetPairs) {
     for (std::vector<Value *>::iterator PI = PairableInsts.begin(),
          PIE = PairableInsts.end(); PI != PIE; ++PI) {
       DenseMap<Value *, Value *>::iterator P = ChosenPairs.find(*PI);
       if (P == ChosenPairs.end()) continue;
 
       Instruction *I = cast<Instruction>(P->first);
-      collectPairLoadMoveSet(BB, ChosenPairs, LoadMoveSet, I);
+      collectPairLoadMoveSet(BB, ChosenPairs, LoadMoveSet,
+                             LoadMoveSetPairs, I);
     }
   }
 
@@ -2767,12 +2922,12 @@ namespace {
   // because the vector instruction is inserted in the location of the pair's
   // second member).
   void BBVectorize::fuseChosenPairs(BasicBlock &BB,
-                     std::vector<Value *> &PairableInsts,
-                     DenseMap<Value *, Value *> &ChosenPairs,
-                     DenseSet<ValuePair> &FixedOrderPairs,
-                     DenseMap<VPPair, unsigned> &PairConnectionTypes,
-                     std::multimap<ValuePair, ValuePair> &ConnectedPairs,
-                     std::multimap<ValuePair, ValuePair> &ConnectedPairDeps) {
+             std::vector<Value *> &PairableInsts,
+             DenseMap<Value *, Value *> &ChosenPairs,
+             DenseSet<ValuePair> &FixedOrderPairs,
+             DenseMap<VPPair, unsigned> &PairConnectionTypes,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairs,
+             DenseMap<ValuePair, std::vector<ValuePair> > &ConnectedPairDeps) {
     LLVMContext& Context = BB.getContext();
 
     // During the vectorization process, the order of the pairs to be fused
@@ -2786,8 +2941,10 @@ namespace {
          E = FlippedPairs.end(); P != E; ++P)
       ChosenPairs.insert(*P);
 
-    std::multimap<Value *, Value *> LoadMoveSet;
-    collectLoadMoveSet(BB, PairableInsts, ChosenPairs, LoadMoveSet);
+    DenseMap<Value *, std::vector<Value *> > LoadMoveSet;
+    DenseSet<ValuePair> LoadMoveSetPairs;
+    collectLoadMoveSet(BB, PairableInsts, ChosenPairs,
+                       LoadMoveSet, LoadMoveSetPairs);
 
     DEBUG(dbgs() << "BBV: initial: \n" << BB << "\n");
 
@@ -2819,7 +2976,7 @@ namespace {
       ChosenPairs.erase(FP);
       ChosenPairs.erase(P);
 
-      if (!canMoveUsesOfIAfterJ(BB, LoadMoveSet, I, J)) {
+      if (!canMoveUsesOfIAfterJ(BB, LoadMoveSetPairs, I, J)) {
         DEBUG(dbgs() << "BBV: fusion of: " << *I <<
                " <-> " << *J <<
                " aborted because of non-trivial dependency cycle\n");
@@ -2836,18 +2993,20 @@ namespace {
         // of dependencies connected via swaps, and those directly connected,
         // and flip the order if the number of swaps is greater.
         bool OrigOrder = true;
-        VPPIteratorPair IP = ConnectedPairDeps.equal_range(ValuePair(I, J));
-        if (IP.first == ConnectedPairDeps.end()) {
-          IP = ConnectedPairDeps.equal_range(ValuePair(J, I));
+        DenseMap<ValuePair, std::vector<ValuePair> >::iterator IJ =
+          ConnectedPairDeps.find(ValuePair(I, J));
+        if (IJ == ConnectedPairDeps.end()) {
+          IJ = ConnectedPairDeps.find(ValuePair(J, I));
           OrigOrder = false;
         }
 
-        if (IP.first != ConnectedPairDeps.end()) {
+        if (IJ != ConnectedPairDeps.end()) {
           unsigned NumDepsDirect = 0, NumDepsSwap = 0;
-          for (std::multimap<ValuePair, ValuePair>::iterator Q = IP.first;
-               Q != IP.second; ++Q) {
+          for (std::vector<ValuePair>::iterator T = IJ->second.begin(),
+               TE = IJ->second.end(); T != TE; ++T) {
+            VPPair Q(IJ->first, *T);
             DenseMap<VPPair, unsigned>::iterator R =
-              PairConnectionTypes.find(VPPair(Q->second, Q->first));
+              PairConnectionTypes.find(VPPair(Q.second, Q.first));
             assert(R != PairConnectionTypes.end() &&
                    "Cannot find pair connection type");
             if (R->second == PairConnectionDirect)
@@ -2873,17 +3032,20 @@ namespace {
 
       // If the pair being fused uses the opposite order from that in the pair
       // connection map, then we need to flip the types.
-      VPPIteratorPair IP = ConnectedPairs.equal_range(ValuePair(H, L));
-      for (std::multimap<ValuePair, ValuePair>::iterator Q = IP.first;
-           Q != IP.second; ++Q) {
-        DenseMap<VPPair, unsigned>::iterator R = PairConnectionTypes.find(*Q);
-        assert(R != PairConnectionTypes.end() &&
-               "Cannot find pair connection type");
-        if (R->second == PairConnectionDirect)
-          R->second = PairConnectionSwap;
-        else if (R->second == PairConnectionSwap)
-          R->second = PairConnectionDirect;
-      }
+      DenseMap<ValuePair, std::vector<ValuePair> >::iterator HL =
+        ConnectedPairs.find(ValuePair(H, L));
+      if (HL != ConnectedPairs.end())
+        for (std::vector<ValuePair>::iterator T = HL->second.begin(),
+             TE = HL->second.end(); T != TE; ++T) {
+          VPPair Q(HL->first, *T);
+          DenseMap<VPPair, unsigned>::iterator R = PairConnectionTypes.find(Q);
+          assert(R != PairConnectionTypes.end() &&
+                 "Cannot find pair connection type");
+          if (R->second == PairConnectionDirect)
+            R->second = PairConnectionSwap;
+          else if (R->second == PairConnectionSwap)
+            R->second = PairConnectionDirect;
+        }
 
       bool LBeforeH = !FlipPairOrder;
       unsigned NumOperands = I->getNumOperands();
@@ -2915,12 +3077,12 @@ namespace {
       Instruction *K1 = 0, *K2 = 0;
       replaceOutputsOfPair(Context, L, H, K, InsertionPt, K1, K2);
 
-      // The use tree of the first original instruction must be moved to after
-      // the location of the second instruction. The entire use tree of the
-      // first instruction is disjoint from the input tree of the second
+      // The use dag of the first original instruction must be moved to after
+      // the location of the second instruction. The entire use dag of the
+      // first instruction is disjoint from the input dag of the second
       // (by definition), and so commutes with it.
 
-      moveUsesOfIAfterJ(BB, LoadMoveSet, InsertionPt, I, J);
+      moveUsesOfIAfterJ(BB, LoadMoveSetPairs, InsertionPt, I, J);
 
       if (!isa<StoreInst>(I)) {
         L->replaceAllUsesWith(K1);
@@ -2937,17 +3099,23 @@ namespace {
       // yet-to-be-fused pair. The loads in question are the keys of the map.
       if (I->mayReadFromMemory()) {
         std::vector<ValuePair> NewSetMembers;
-        VPIteratorPair IPairRange = LoadMoveSet.equal_range(I);
-        VPIteratorPair JPairRange = LoadMoveSet.equal_range(J);
-        for (std::multimap<Value *, Value *>::iterator N = IPairRange.first;
-             N != IPairRange.second; ++N)
-          NewSetMembers.push_back(ValuePair(K, N->second));
-        for (std::multimap<Value *, Value *>::iterator N = JPairRange.first;
-             N != JPairRange.second; ++N)
-          NewSetMembers.push_back(ValuePair(K, N->second));
+        DenseMap<Value *, std::vector<Value *> >::iterator II =
+          LoadMoveSet.find(I);
+        if (II != LoadMoveSet.end())
+          for (std::vector<Value *>::iterator N = II->second.begin(),
+               NE = II->second.end(); N != NE; ++N)
+            NewSetMembers.push_back(ValuePair(K, *N));
+        DenseMap<Value *, std::vector<Value *> >::iterator JJ =
+          LoadMoveSet.find(J);
+        if (JJ != LoadMoveSet.end())
+          for (std::vector<Value *>::iterator N = JJ->second.begin(),
+               NE = JJ->second.end(); N != NE; ++N)
+            NewSetMembers.push_back(ValuePair(K, *N));
         for (std::vector<ValuePair>::iterator A = NewSetMembers.begin(),
-             AE = NewSetMembers.end(); A != AE; ++A)
-          LoadMoveSet.insert(*A);
+             AE = NewSetMembers.end(); A != AE; ++A) {
+          LoadMoveSet[A->first].push_back(A->second);
+          LoadMoveSetPairs.insert(*A);
+        }
       }
 
       // Before removing I, set the iterator to the next instruction.
@@ -2972,6 +3140,7 @@ char BBVectorize::ID = 0;
 static const char bb_vectorize_name[] = "Basic-Block Vectorization";
 INITIALIZE_PASS_BEGIN(BBVectorize, BBV_NAME, bb_vectorize_name, false, false)
 INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_AG_DEPENDENCY(TargetTransformInfo)
 INITIALIZE_PASS_DEPENDENCY(DominatorTree)
 INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
 INITIALIZE_PASS_END(BBVectorize, BBV_NAME, bb_vectorize_name, false, false)
@@ -3006,6 +3175,7 @@ VectorizeConfig::VectorizeConfig() {
   MaxCandPairsForCycleCheck = ::MaxCandPairsForCycleCheck;
   SplatBreaksChain = ::SplatBreaksChain;
   MaxInsts = ::MaxInsts;
+  MaxPairs = ::MaxPairs;
   MaxIter = ::MaxIter;
   Pow2LenOnly = ::Pow2LenOnly;
   NoMemOpBoost = ::NoMemOpBoost;
diff --git a/lib/Transforms/Vectorize/LoopVectorize.cpp b/lib/Transforms/Vectorize/LoopVectorize.cpp
index feeececedb..07dd453424 100644
--- a/lib/Transforms/Vectorize/LoopVectorize.cpp
+++ b/lib/Transforms/Vectorize/LoopVectorize.cpp
@@ -6,7 +6,51 @@
 // License. See LICENSE.TXT for details.
 //
 //===----------------------------------------------------------------------===//
-#include "LoopVectorize.h"
+//
+// This is the LLVM loop vectorizer. This pass modifies 'vectorizable' loops
+// and generates target-independent LLVM-IR. Legalization of the IR is done
+// in the codegen. However, the vectorizer uses (will use) the codegen
+// interfaces to generate IR that is likely to result in an optimal binary.
+//
+// The loop vectorizer combines consecutive loop iterations into a single
+// 'wide' iteration. After this transformation the index is incremented
+// by the SIMD vector width, and not by one.
+//
+// This pass has three parts:
+// 1. The main loop pass that drives the different parts.
+// 2. LoopVectorizationLegality - A unit that checks for the legality
+//    of the vectorization.
+// 3. InnerLoopVectorizer - A unit that performs the actual
+//    widening of instructions.
+// 4. LoopVectorizationCostModel - A unit that checks for the profitability
+//    of vectorization. It decides on the optimal vector width, which
+//    can be one, if vectorization is not profitable.
+//
+//===----------------------------------------------------------------------===//
+//
+// The reduction-variable vectorization is based on the paper:
+//  D. Nuzman and R. Henderson. Multi-platform Auto-vectorization.
+//
+// Variable uniformity checks are inspired by:
+//  Karrenberg, R. and Hack, S. Whole Function Vectorization.
+//
+// Other ideas/concepts are from:
+//  A. Zaks and D. Nuzman. Autovectorization in GCC-two years later.
+//
+//  S. Maleki, Y. Gao, M. Garzaran, T. Wong and D. Padua.  An Evaluation of
+//  Vectorizing Compilers.
+//
+//===----------------------------------------------------------------------===//
+
+#define LV_NAME "loop-vectorize"
+#define DEBUG_TYPE LV_NAME
+
+#include "llvm/Transforms/Vectorize.h"
+#include "llvm/ADT/DenseMap.h"
+#include "llvm/ADT/MapVector.h"
+#include "llvm/ADT/SmallPtrSet.h"
+#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/Analysis/AliasAnalysis.h"
 #include "llvm/Analysis/AliasSetTracker.h"
@@ -14,46 +58,586 @@
 #include "llvm/Analysis/LoopInfo.h"
 #include "llvm/Analysis/LoopIterator.h"
 #include "llvm/Analysis/LoopPass.h"
-#include "llvm/Analysis/ScalarEvolutionExpander.h"
+#include "llvm/Analysis/ScalarEvolution.h"
 #include "llvm/Analysis/ScalarEvolutionExpander.h"
 #include "llvm/Analysis/ScalarEvolutionExpressions.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/Analysis/ValueTracking.h"
 #include "llvm/Analysis/Verifier.h"
-#include "llvm/Constants.h"
-#include "llvm/DataLayout.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Function.h"
-#include "llvm/Instructions.h"
-#include "llvm/IntrinsicInst.h"
-#include "llvm/LLVMContext.h"
-#include "llvm/Module.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Module.h"
+#include "llvm/IR/Type.h"
+#include "llvm/IR/Value.h"
 #include "llvm/Pass.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/TargetTransformInfo.h"
+#include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Scalar.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/Local.h"
-#include "llvm/Transforms/Vectorize.h"
-#include "llvm/Type.h"
-#include "llvm/Value.h"
+#include <algorithm>
+#include <map>
+
+using namespace llvm;
 
 static cl::opt<unsigned>
 VectorizationFactor("force-vector-width", cl::init(0), cl::Hidden,
                     cl::desc("Sets the SIMD width. Zero is autoselect."));
 
+static cl::opt<unsigned>
+VectorizationUnroll("force-vector-unroll", cl::init(0), cl::Hidden,
+                    cl::desc("Sets the vectorization unroll count. "
+                             "Zero is autoselect."));
+
 static cl::opt<bool>
 EnableIfConversion("enable-if-conversion", cl::init(true), cl::Hidden,
                    cl::desc("Enable if-conversion during vectorization."));
 
+/// We don't vectorize loops with a known constant trip count below this number.
+static cl::opt<unsigned>
+TinyTripCountVectorThreshold("vectorizer-min-trip-count", cl::init(16),
+                             cl::Hidden,
+                             cl::desc("Don't vectorize loops with a constant "
+                                      "trip count that is smaller than this "
+                                      "value."));
+
+/// We don't unroll loops with a known constant trip count below this number.
+static const unsigned TinyTripCountUnrollThreshold = 128;
+
+/// When performing a runtime memory check, do not check more than this
+/// number of pointers. Notice that the check is quadratic!
+static const unsigned RuntimeMemoryCheckThreshold = 4;
+
+/// We use a metadata with this name  to indicate that a scalar loop was
+/// vectorized and that we don't need to re-vectorize it if we run into it
+/// again.
+static const char*
+AlreadyVectorizedMDName = "llvm.vectorizer.already_vectorized";
+
 namespace {
 
+// Forward declarations.
+class LoopVectorizationLegality;
+class LoopVectorizationCostModel;
+
+/// InnerLoopVectorizer vectorizes loops which contain only one basic
+/// block to a specified vectorization factor (VF).
+/// This class performs the widening of scalars into vectors, or multiple
+/// scalars. This class also implements the following features:
+/// * It inserts an epilogue loop for handling loops that don't have iteration
+///   counts that are known to be a multiple of the vectorization factor.
+/// * It handles the code generation for reduction variables.
+/// * Scalarization (implementation using scalars) of un-vectorizable
+///   instructions.
+/// InnerLoopVectorizer does not perform any vectorization-legality
+/// checks, and relies on the caller to check for the different legality
+/// aspects. The InnerLoopVectorizer relies on the
+/// LoopVectorizationLegality class to provide information about the induction
+/// and reduction variables that were found to a given vectorization factor.
+class InnerLoopVectorizer {
+public:
+  InnerLoopVectorizer(Loop *OrigLoop, ScalarEvolution *SE, LoopInfo *LI,
+                      DominatorTree *DT, DataLayout *DL,
+                      const TargetLibraryInfo *TLI, unsigned VecWidth,
+                      unsigned UnrollFactor)
+      : OrigLoop(OrigLoop), SE(SE), LI(LI), DT(DT), DL(DL), TLI(TLI),
+        VF(VecWidth), UF(UnrollFactor), Builder(SE->getContext()), Induction(0),
+        OldInduction(0), WidenMap(UnrollFactor) {}
+
+  // Perform the actual loop widening (vectorization).
+  void vectorize(LoopVectorizationLegality *Legal) {
+    // Create a new empty loop. Unlink the old loop and connect the new one.
+    createEmptyLoop(Legal);
+    // Widen each instruction in the old loop to a new one in the new loop.
+    // Use the Legality module to find the induction and reduction variables.
+    vectorizeLoop(Legal);
+    // Register the new loop and update the analysis passes.
+    updateAnalysis();
+  }
+
+private:
+  /// A small list of PHINodes.
+  typedef SmallVector<PHINode*, 4> PhiVector;
+  /// When we unroll loops we have multiple vector values for each scalar.
+  /// This data structure holds the unrolled and vectorized values that
+  /// originated from one scalar instruction.
+  typedef SmallVector<Value*, 2> VectorParts;
+
+  /// Add code that checks at runtime if the accessed arrays overlap.
+  /// Returns the comparator value or NULL if no check is needed.
+  Instruction *addRuntimeCheck(LoopVectorizationLegality *Legal,
+                               Instruction *Loc);
+  /// Create an empty loop, based on the loop ranges of the old loop.
+  void createEmptyLoop(LoopVectorizationLegality *Legal);
+  /// Copy and widen the instructions from the old loop.
+  void vectorizeLoop(LoopVectorizationLegality *Legal);
+
+  /// A helper function that computes the predicate of the block BB, assuming
+  /// that the header block of the loop is set to True. It returns the *entry*
+  /// mask for the block BB.
+  VectorParts createBlockInMask(BasicBlock *BB);
+  /// A helper function that computes the predicate of the edge between SRC
+  /// and DST.
+  VectorParts createEdgeMask(BasicBlock *Src, BasicBlock *Dst);
+
+  /// A helper function to vectorize a single BB within the innermost loop.
+  void vectorizeBlockInLoop(LoopVectorizationLegality *Legal, BasicBlock *BB,
+                            PhiVector *PV);
+
+  /// Insert the new loop to the loop hierarchy and pass manager
+  /// and update the analysis passes.
+  void updateAnalysis();
+
+  /// This instruction is un-vectorizable. Implement it as a sequence
+  /// of scalars.
+  void scalarizeInstruction(Instruction *Instr);
+
+  /// Vectorize Load and Store instructions,
+  void vectorizeMemoryInstruction(Instruction *Instr,
+                                  LoopVectorizationLegality *Legal);
+
+  /// Create a broadcast instruction. This method generates a broadcast
+  /// instruction (shuffle) for loop invariant values and for the induction
+  /// value. If this is the induction variable then we extend it to N, N+1, ...
+  /// this is needed because each iteration in the loop corresponds to a SIMD
+  /// element.
+  Value *getBroadcastInstrs(Value *V);
+
+  /// This function adds 0, 1, 2 ... to each vector element, starting at zero.
+  /// If Negate is set then negative numbers are added e.g. (0, -1, -2, ...).
+  /// The sequence starts at StartIndex.
+  Value *getConsecutiveVector(Value* Val, unsigned StartIdx, bool Negate);
+
+  /// When we go over instructions in the basic block we rely on previous
+  /// values within the current basic block or on loop invariant values.
+  /// When we widen (vectorize) values we place them in the map. If the values
+  /// are not within the map, they have to be loop invariant, so we simply
+  /// broadcast them into a vector.
+  VectorParts &getVectorValue(Value *V);
+
+  /// Generate a shuffle sequence that will reverse the vector Vec.
+  Value *reverseVector(Value *Vec);
+
+  /// This is a helper class that holds the vectorizer state. It maps scalar
+  /// instructions to vector instructions. When the code is 'unrolled' then
+  /// then a single scalar value is mapped to multiple vector parts. The parts
+  /// are stored in the VectorPart type.
+  struct ValueMap {
+    /// C'tor.  UnrollFactor controls the number of vectors ('parts') that
+    /// are mapped.
+    ValueMap(unsigned UnrollFactor) : UF(UnrollFactor) {}
+
+    /// \return True if 'Key' is saved in the Value Map.
+    bool has(Value *Key) const { return MapStorage.count(Key); }
+
+    /// Initializes a new entry in the map. Sets all of the vector parts to the
+    /// save value in 'Val'.
+    /// \return A reference to a vector with splat values.
+    VectorParts &splat(Value *Key, Value *Val) {
+      VectorParts &Entry = MapStorage[Key];
+      Entry.assign(UF, Val);
+      return Entry;
+    }
+
+    ///\return A reference to the value that is stored at 'Key'.
+    VectorParts &get(Value *Key) {
+      VectorParts &Entry = MapStorage[Key];
+      if (Entry.empty())
+        Entry.resize(UF);
+      assert(Entry.size() == UF);
+      return Entry;
+    }
+
+  private:
+    /// The unroll factor. Each entry in the map stores this number of vector
+    /// elements.
+    unsigned UF;
+
+    /// Map storage. We use std::map and not DenseMap because insertions to a
+    /// dense map invalidates its iterators.
+    std::map<Value *, VectorParts> MapStorage;
+  };
+
+  /// The original loop.
+  Loop *OrigLoop;
+  /// Scev analysis to use.
+  ScalarEvolution *SE;
+  /// Loop Info.
+  LoopInfo *LI;
+  /// Dominator Tree.
+  DominatorTree *DT;
+  /// Data Layout.
+  DataLayout *DL;
+  /// Target Library Info.
+  const TargetLibraryInfo *TLI;
+
+  /// The vectorization SIMD factor to use. Each vector will have this many
+  /// vector elements.
+  unsigned VF;
+  /// The vectorization unroll factor to use. Each scalar is vectorized to this
+  /// many different vector instructions.
+  unsigned UF;
+
+  /// The builder that we use
+  IRBuilder<> Builder;
+
+  // --- Vectorization state ---
+
+  /// The vector-loop preheader.
+  BasicBlock *LoopVectorPreHeader;
+  /// The scalar-loop preheader.
+  BasicBlock *LoopScalarPreHeader;
+  /// Middle Block between the vector and the scalar.
+  BasicBlock *LoopMiddleBlock;
+  ///The ExitBlock of the scalar loop.
+  BasicBlock *LoopExitBlock;
+  ///The vector loop body.
+  BasicBlock *LoopVectorBody;
+  ///The scalar loop body.
+  BasicBlock *LoopScalarBody;
+  /// A list of all bypass blocks. The first block is the entry of the loop.
+  SmallVector<BasicBlock *, 4> LoopBypassBlocks;
+
+  /// The new Induction variable which was added to the new block.
+  PHINode *Induction;
+  /// The induction variable of the old basic block.
+  PHINode *OldInduction;
+  /// Maps scalars to widened vectors.
+  ValueMap WidenMap;
+};
+
+/// LoopVectorizationLegality checks if it is legal to vectorize a loop, and
+/// to what vectorization factor.
+/// This class does not look at the profitability of vectorization, only the
+/// legality. This class has two main kinds of checks:
+/// * Memory checks - The code in canVectorizeMemory checks if vectorization
+///   will change the order of memory accesses in a way that will change the
+///   correctness of the program.
+/// * Scalars checks - The code in canVectorizeInstrs and canVectorizeMemory
+/// checks for a number of different conditions, such as the availability of a
+/// single induction variable, that all types are supported and vectorize-able,
+/// etc. This code reflects the capabilities of InnerLoopVectorizer.
+/// This class is also used by InnerLoopVectorizer for identifying
+/// induction variable and the different reduction variables.
+class LoopVectorizationLegality {
+public:
+  LoopVectorizationLegality(Loop *L, ScalarEvolution *SE, DataLayout *DL,
+                            DominatorTree *DT, TargetTransformInfo* TTI,
+                            AliasAnalysis *AA, TargetLibraryInfo *TLI)
+      : TheLoop(L), SE(SE), DL(DL), DT(DT), TTI(TTI), AA(AA), TLI(TLI),
+        Induction(0) {}
+
+  /// This enum represents the kinds of reductions that we support.
+  enum ReductionKind {
+    RK_NoReduction, ///< Not a reduction.
+    RK_IntegerAdd,  ///< Sum of integers.
+    RK_IntegerMult, ///< Product of integers.
+    RK_IntegerOr,   ///< Bitwise or logical OR of numbers.
+    RK_IntegerAnd,  ///< Bitwise or logical AND of numbers.
+    RK_IntegerXor,  ///< Bitwise or logical XOR of numbers.
+    RK_FloatAdd,    ///< Sum of floats.
+    RK_FloatMult    ///< Product of floats.
+  };
+
+  /// This enum represents the kinds of inductions that we support.
+  enum InductionKind {
+    IK_NoInduction,         ///< Not an induction variable.
+    IK_IntInduction,        ///< Integer induction variable. Step = 1.
+    IK_ReverseIntInduction, ///< Reverse int induction variable. Step = -1.
+    IK_PtrInduction,        ///< Pointer induction var. Step = sizeof(elem).
+    IK_ReversePtrInduction  ///< Reverse ptr indvar. Step = - sizeof(elem).
+  };
+
+  /// This POD struct holds information about reduction variables.
+  struct ReductionDescriptor {
+    ReductionDescriptor() : StartValue(0), LoopExitInstr(0),
+      Kind(RK_NoReduction) {}
+
+    ReductionDescriptor(Value *Start, Instruction *Exit, ReductionKind K)
+        : StartValue(Start), LoopExitInstr(Exit), Kind(K) {}
+
+    // The starting value of the reduction.
+    // It does not have to be zero!
+    Value *StartValue;
+    // The instruction who's value is used outside the loop.
+    Instruction *LoopExitInstr;
+    // The kind of the reduction.
+    ReductionKind Kind;
+  };
+
+  // This POD struct holds information about the memory runtime legality
+  // check that a group of pointers do not overlap.
+  struct RuntimePointerCheck {
+    RuntimePointerCheck() : Need(false) {}
+
+    /// Reset the state of the pointer runtime information.
+    void reset() {
+      Need = false;
+      Pointers.clear();
+      Starts.clear();
+      Ends.clear();
+    }
+
+    /// Insert a pointer and calculate the start and end SCEVs.
+    void insert(ScalarEvolution *SE, Loop *Lp, Value *Ptr);
+
+    /// This flag indicates if we need to add the runtime check.
+    bool Need;
+    /// Holds the pointers that we need to check.
+    SmallVector<Value*, 2> Pointers;
+    /// Holds the pointer value at the beginning of the loop.
+    SmallVector<const SCEV*, 2> Starts;
+    /// Holds the pointer value at the end of the loop.
+    SmallVector<const SCEV*, 2> Ends;
+  };
+
+  /// A POD for saving information about induction variables.
+  struct InductionInfo {
+    InductionInfo(Value *Start, InductionKind K) : StartValue(Start), IK(K) {}
+    InductionInfo() : StartValue(0), IK(IK_NoInduction) {}
+    /// Start value.
+    Value *StartValue;
+    /// Induction kind.
+    InductionKind IK;
+  };
+
+  /// ReductionList contains the reduction descriptors for all
+  /// of the reductions that were found in the loop.
+  typedef DenseMap<PHINode*, ReductionDescriptor> ReductionList;
+
+  /// InductionList saves induction variables and maps them to the
+  /// induction descriptor.
+  typedef MapVector<PHINode*, InductionInfo> InductionList;
+
+  /// Alias(Multi)Map stores the values (GEPs or underlying objects and their
+  /// respective Store/Load instruction(s) to calculate aliasing.
+  typedef MapVector<Value*, Instruction* > AliasMap;
+  typedef DenseMap<Value*, std::vector<Instruction*> > AliasMultiMap;
+
+  /// Returns true if it is legal to vectorize this loop.
+  /// This does not mean that it is profitable to vectorize this
+  /// loop, only that it is legal to do so.
+  bool canVectorize();
+
+  /// Returns the Induction variable.
+  PHINode *getInduction() { return Induction; }
+
+  /// Returns the reduction variables found in the loop.
+  ReductionList *getReductionVars() { return &Reductions; }
+
+  /// Returns the induction variables found in the loop.
+  InductionList *getInductionVars() { return &Inductions; }
+
+  /// Returns True if V is an induction variable in this loop.
+  bool isInductionVariable(const Value *V);
+
+  /// Return true if the block BB needs to be predicated in order for the loop
+  /// to be vectorized.
+  bool blockNeedsPredication(BasicBlock *BB);
+
+  /// Check if this  pointer is consecutive when vectorizing. This happens
+  /// when the last index of the GEP is the induction variable, or that the
+  /// pointer itself is an induction variable.
+  /// This check allows us to vectorize A[idx] into a wide load/store.
+  /// Returns:
+  /// 0 - Stride is unknown or non consecutive.
+  /// 1 - Address is consecutive.
+  /// -1 - Address is consecutive, and decreasing.
+  int isConsecutivePtr(Value *Ptr);
+
+  /// Returns true if the value V is uniform within the loop.
+  bool isUniform(Value *V);
+
+  /// Returns true if this instruction will remain scalar after vectorization.
+  bool isUniformAfterVectorization(Instruction* I) { return Uniforms.count(I); }
+
+  /// Returns the information that we collected about runtime memory check.
+  RuntimePointerCheck *getRuntimePointerCheck() { return &PtrRtCheck; }
+private:
+  /// Check if a single basic block loop is vectorizable.
+  /// At this point we know that this is a loop with a constant trip count
+  /// and we only need to check individual instructions.
+  bool canVectorizeInstrs();
+
+  /// When we vectorize loops we may change the order in which
+  /// we read and write from memory. This method checks if it is
+  /// legal to vectorize the code, considering only memory constrains.
+  /// Returns true if the loop is vectorizable
+  bool canVectorizeMemory();
+
+  /// Return true if we can vectorize this loop using the IF-conversion
+  /// transformation.
+  bool canVectorizeWithIfConvert();
+
+  /// Collect the variables that need to stay uniform after vectorization.
+  void collectLoopUniforms();
+
+  /// Return true if all of the instructions in the block can be speculatively
+  /// executed.
+  bool blockCanBePredicated(BasicBlock *BB);
+
+  /// Returns True, if 'Phi' is the kind of reduction variable for type
+  /// 'Kind'. If this is a reduction variable, it adds it to ReductionList.
+  bool AddReductionVar(PHINode *Phi, ReductionKind Kind);
+  /// Returns true if the instruction I can be a reduction variable of type
+  /// 'Kind'.
+  bool isReductionInstr(Instruction *I, ReductionKind Kind);
+  /// Returns the induction kind of Phi. This function may return NoInduction
+  /// if the PHI is not an induction variable.
+  InductionKind isInductionVariable(PHINode *Phi);
+  /// Return true if can compute the address bounds of Ptr within the loop.
+  bool hasComputableBounds(Value *Ptr);
+  /// Return true if there is the chance of write reorder.
+  bool hasPossibleGlobalWriteReorder(Value *Object,
+                                     Instruction *Inst,
+                                     AliasMultiMap &WriteObjects,
+                                     unsigned MaxByteWidth);
+  /// Return the AA location for a load or a store.
+  AliasAnalysis::Location getLoadStoreLocation(Instruction *Inst);
+
+
+  /// The loop that we evaluate.
+  Loop *TheLoop;
+  /// Scev analysis.
+  ScalarEvolution *SE;
+  /// DataLayout analysis.
+  DataLayout *DL;
+  /// Dominators.
+  DominatorTree *DT;
+  /// Target Info.
+  TargetTransformInfo *TTI;
+  /// Alias Analysis.
+  AliasAnalysis *AA;
+  /// Target Library Info.
+  TargetLibraryInfo *TLI;
+
+  //  ---  vectorization state --- //
+
+  /// Holds the integer induction variable. This is the counter of the
+  /// loop.
+  PHINode *Induction;
+  /// Holds the reduction variables.
+  ReductionList Reductions;
+  /// Holds all of the induction variables that we found in the loop.
+  /// Notice that inductions don't need to start at zero and that induction
+  /// variables can be pointers.
+  InductionList Inductions;
+
+  /// Allowed outside users. This holds the reduction
+  /// vars which can be accessed from outside the loop.
+  SmallPtrSet<Value*, 4> AllowedExit;
+  /// This set holds the variables which are known to be uniform after
+  /// vectorization.
+  SmallPtrSet<Instruction*, 4> Uniforms;
+  /// We need to check that all of the pointers in this list are disjoint
+  /// at runtime.
+  RuntimePointerCheck PtrRtCheck;
+};
+
+/// LoopVectorizationCostModel - estimates the expected speedups due to
+/// vectorization.
+/// In many cases vectorization is not profitable. This can happen because of
+/// a number of reasons. In this class we mainly attempt to predict the
+/// expected speedup/slowdowns due to the supported instruction set. We use the
+/// TargetTransformInfo to query the different backends for the cost of
+/// different operations.
+class LoopVectorizationCostModel {
+public:
+  LoopVectorizationCostModel(Loop *L, ScalarEvolution *SE, LoopInfo *LI,
+                             LoopVectorizationLegality *Legal,
+                             const TargetTransformInfo &TTI,
+                             DataLayout *DL, const TargetLibraryInfo *TLI)
+      : TheLoop(L), SE(SE), LI(LI), Legal(Legal), TTI(TTI), DL(DL), TLI(TLI) {}
+
+  /// Information about vectorization costs
+  struct VectorizationFactor {
+    unsigned Width; // Vector width with best cost
+    unsigned Cost; // Cost of the loop with that width
+  };
+  /// \return The most profitable vectorization factor and the cost of that VF.
+  /// This method checks every power of two up to VF. If UserVF is not ZERO
+  /// then this vectorization factor will be selected if vectorization is
+  /// possible.
+  VectorizationFactor selectVectorizationFactor(bool OptForSize,
+                                                unsigned UserVF);
+
+  /// \return The size (in bits) of the widest type in the code that
+  /// needs to be vectorized. We ignore values that remain scalar such as
+  /// 64 bit loop indices.
+  unsigned getWidestType();
+
+  /// \return The most profitable unroll factor.
+  /// If UserUF is non-zero then this method finds the best unroll-factor
+  /// based on register pressure and other parameters.
+  /// VF and LoopCost are the selected vectorization factor and the cost of the
+  /// selected VF.
+  unsigned selectUnrollFactor(bool OptForSize, unsigned UserUF, unsigned VF,
+                              unsigned LoopCost);
+
+  /// \brief A struct that represents some properties of the register usage
+  /// of a loop.
+  struct RegisterUsage {
+    /// Holds the number of loop invariant values that are used in the loop.
+    unsigned LoopInvariantRegs;
+    /// Holds the maximum number of concurrent live intervals in the loop.
+    unsigned MaxLocalUsers;
+    /// Holds the number of instructions in the loop.
+    unsigned NumInstructions;
+  };
+
+  /// \return  information about the register usage of the loop.
+  RegisterUsage calculateRegisterUsage();
+
+private:
+  /// Returns the expected execution cost. The unit of the cost does
+  /// not matter because we use the 'cost' units to compare different
+  /// vector widths. The cost that is returned is *not* normalized by
+  /// the factor width.
+  unsigned expectedCost(unsigned VF);
+
+  /// Returns the execution time cost of an instruction for a given vector
+  /// width. Vector width of one means scalar.
+  unsigned getInstructionCost(Instruction *I, unsigned VF);
+
+  /// A helper function for converting Scalar types to vector types.
+  /// If the incoming type is void, we return void. If the VF is 1, we return
+  /// the scalar type.
+  static Type* ToVectorTy(Type *Scalar, unsigned VF);
+
+  /// Returns whether the instruction is a load or store and will be a emitted
+  /// as a vector operation.
+  bool isConsecutiveLoadOrStore(Instruction *I);
+
+  /// The loop that we evaluate.
+  Loop *TheLoop;
+  /// Scev analysis.
+  ScalarEvolution *SE;
+  /// Loop Info analysis.
+  LoopInfo *LI;
+  /// Vectorization legality.
+  LoopVectorizationLegality *Legal;
+  /// Vector target information.
+  const TargetTransformInfo &TTI;
+  /// Target data layout information.
+  DataLayout *DL;
+  /// Target Library Info.
+  const TargetLibraryInfo *TLI;
+};
+
 /// The LoopVectorize Pass.
 struct LoopVectorize : public LoopPass {
-  static char ID; // Pass identification, replacement for typeid
+  /// Pass identification, replacement for typeid
+  static char ID;
 
-  LoopVectorize() : LoopPass(ID) {
+  explicit LoopVectorize() : LoopPass(ID) {
     initializeLoopVectorizePass(*PassRegistry::getPassRegistry());
   }
 
@@ -62,6 +646,8 @@ struct LoopVectorize : public LoopPass {
   LoopInfo *LI;
   TargetTransformInfo *TTI;
   DominatorTree *DT;
+  AliasAnalysis *AA;
+  TargetLibraryInfo *TLI;
 
   virtual bool runOnLoop(Loop *L, LPPassManager &LPM) {
     // We only vectorize innermost loops.
@@ -71,45 +657,57 @@ struct LoopVectorize : public LoopPass {
     SE = &getAnalysis<ScalarEvolution>();
     DL = getAnalysisIfAvailable<DataLayout>();
     LI = &getAnalysis<LoopInfo>();
-    TTI = getAnalysisIfAvailable<TargetTransformInfo>();
+    TTI = &getAnalysis<TargetTransformInfo>();
     DT = &getAnalysis<DominatorTree>();
+    AA = getAnalysisIfAvailable<AliasAnalysis>();
+    TLI = getAnalysisIfAvailable<TargetLibraryInfo>();
 
     DEBUG(dbgs() << "LV: Checking a loop in \"" <<
           L->getHeader()->getParent()->getName() << "\"\n");
 
     // Check if it is legal to vectorize the loop.
-    LoopVectorizationLegality LVL(L, SE, DL, DT);
+    LoopVectorizationLegality LVL(L, SE, DL, DT, TTI, AA, TLI);
     if (!LVL.canVectorize()) {
       DEBUG(dbgs() << "LV: Not vectorizing.\n");
       return false;
     }
 
-    // Select the preffered vectorization factor.
-    unsigned VF = 1;
-    if (VectorizationFactor == 0) {
-      const VectorTargetTransformInfo *VTTI = 0;
-      if (TTI)
-        VTTI = TTI->getVectorTargetTransformInfo();
-      // Use the cost model.
-      LoopVectorizationCostModel CM(L, SE, &LVL, VTTI);
-      VF = CM.findBestVectorizationFactor();
-
-      if (VF == 1) {
-        DEBUG(dbgs() << "LV: Vectorization is possible but not beneficial.\n");
-        return false;
-      }
+    // Use the cost model.
+    LoopVectorizationCostModel CM(L, SE, LI, &LVL, *TTI, DL, TLI);
+
+    // Check the function attributes to find out if this function should be
+    // optimized for size.
+    Function *F = L->getHeader()->getParent();
+    Attribute::AttrKind SzAttr = Attribute::OptimizeForSize;
+    Attribute::AttrKind FlAttr = Attribute::NoImplicitFloat;
+    unsigned FnIndex = AttributeSet::FunctionIndex;
+    bool OptForSize = F->getAttributes().hasAttribute(FnIndex, SzAttr);
+    bool NoFloat = F->getAttributes().hasAttribute(FnIndex, FlAttr);
+
+    if (NoFloat) {
+      DEBUG(dbgs() << "LV: Can't vectorize when the NoImplicitFloat"
+            "attribute is used.\n");
+      return false;
+    }
 
-    } else {
-      // Use the user command flag.
-      VF = VectorizationFactor;
+    // Select the optimal vectorization factor.
+    LoopVectorizationCostModel::VectorizationFactor VF;
+    VF = CM.selectVectorizationFactor(OptForSize, VectorizationFactor);
+    // Select the unroll factor.
+    unsigned UF = CM.selectUnrollFactor(OptForSize, VectorizationUnroll,
+                                        VF.Width, VF.Cost);
+
+    if (VF.Width == 1) {
+      DEBUG(dbgs() << "LV: Vectorization is possible but not beneficial.\n");
+      return false;
     }
 
-    DEBUG(dbgs() << "LV: Found a vectorizable loop ("<< VF << ") in "<<
-          L->getHeader()->getParent()->getParent()->getModuleIdentifier()<<
-          "\n");
+    DEBUG(dbgs() << "LV: Found a vectorizable loop ("<< VF.Width << ") in "<<
+          F->getParent()->getModuleIdentifier()<<"\n");
+    DEBUG(dbgs() << "LV: Unroll Factor is " << UF << "\n");
 
-    // If we decided that it is *legal* to vectorizer the loop then do it.
-    InnerLoopVectorizer LB(L, SE, LI, DT, DL, VF);
+    // If we decided that it is *legal* to vectorize the loop then do it.
+    InnerLoopVectorizer LB(L, SE, LI, DT, DL, TLI, VF.Width, UF);
     LB.vectorize(&LVL);
 
     DEBUG(verifyFunction(*L->getHeader()->getParent()));
@@ -120,16 +718,17 @@ struct LoopVectorize : public LoopPass {
     LoopPass::getAnalysisUsage(AU);
     AU.addRequiredID(LoopSimplifyID);
     AU.addRequiredID(LCSSAID);
+    AU.addRequired<DominatorTree>();
     AU.addRequired<LoopInfo>();
     AU.addRequired<ScalarEvolution>();
-    AU.addRequired<DominatorTree>();
+    AU.addRequired<TargetTransformInfo>();
     AU.addPreserved<LoopInfo>();
     AU.addPreserved<DominatorTree>();
   }
 
 };
 
-}// namespace
+} // end anonymous namespace
 
 //===----------------------------------------------------------------------===//
 // Implementation of LoopVectorizationLegality, InnerLoopVectorizer and
@@ -150,11 +749,6 @@ LoopVectorizationLegality::RuntimePointerCheck::insert(ScalarEvolution *SE,
 }
 
 Value *InnerLoopVectorizer::getBroadcastInstrs(Value *V) {
-  // Create the types.
-  LLVMContext &C = V->getContext();
-  Type *VTy = VectorType::get(V->getType(), VF);
-  Type *I32 = IntegerType::getInt32Ty(C);
-
   // Save the current insertion location.
   Instruction *Loc = Builder.GetInsertPoint();
 
@@ -167,14 +761,8 @@ Value *InnerLoopVectorizer::getBroadcastInstrs(Value *V) {
   if (Invariant)
     Builder.SetInsertPoint(LoopVectorPreHeader->getTerminator());
 
-  Constant *Zero = ConstantInt::get(I32, 0);
-  Value *Zeros = ConstantAggregateZero::get(VectorType::get(I32, VF));
-  Value *UndefVal = UndefValue::get(VTy);
-  // Insert the value into a new vector.
-  Value *SingleElem = Builder.CreateInsertElement(UndefVal, V, Zero);
   // Broadcast the scalar into all locations in the vector.
-  Value *Shuf = Builder.CreateShuffleVector(SingleElem, UndefVal, Zeros,
-                                            "broadcast");
+  Value *Shuf = Builder.CreateVectorSplat(VF, V, "broadcast");
 
   // Restore the builder insertion point.
   if (Invariant)
@@ -183,7 +771,8 @@ Value *InnerLoopVectorizer::getBroadcastInstrs(Value *V) {
   return Shuf;
 }
 
-Value *InnerLoopVectorizer::getConsecutiveVector(Value* Val, bool Negate) {
+Value *InnerLoopVectorizer::getConsecutiveVector(Value* Val, unsigned StartIdx,
+                                                 bool Negate) {
   assert(Val->getType()->isVectorTy() && "Must be a vector");
   assert(Val->getType()->getScalarType()->isIntegerTy() &&
          "Elem must be an integer");
@@ -194,8 +783,10 @@ Value *InnerLoopVectorizer::getConsecutiveVector(Value* Val, bool Negate) {
   SmallVector<Constant*, 8> Indices;
 
   // Create a vector of consecutive numbers from zero to VF.
-  for (int i = 0; i < VLen; ++i)
-    Indices.push_back(ConstantInt::get(ITy, Negate ? (-i): i ));
+  for (int i = 0; i < VLen; ++i) {
+    int Idx = Negate ? (-i): i;
+    Indices.push_back(ConstantInt::get(ITy, StartIdx + Idx));
+  }
 
   // Add the consecutive indices to the vector value.
   Constant *Cv = ConstantVector::get(Indices);
@@ -203,28 +794,56 @@ Value *InnerLoopVectorizer::getConsecutiveVector(Value* Val, bool Negate) {
   return Builder.CreateAdd(Val, Cv, "induction");
 }
 
-bool LoopVectorizationLegality::isConsecutivePtr(Value *Ptr) {
+int LoopVectorizationLegality::isConsecutivePtr(Value *Ptr) {
   assert(Ptr->getType()->isPointerTy() && "Unexpected non ptr");
+  // Make sure that the pointer does not point to structs.
+  if (cast<PointerType>(Ptr->getType())->getElementType()->isAggregateType())
+    return 0;
 
   // If this value is a pointer induction variable we know it is consecutive.
   PHINode *Phi = dyn_cast_or_null<PHINode>(Ptr);
   if (Phi && Inductions.count(Phi)) {
     InductionInfo II = Inductions[Phi];
-    if (PtrInduction == II.IK)
-      return true;
+    if (IK_PtrInduction == II.IK)
+      return 1;
+    else if (IK_ReversePtrInduction == II.IK)
+      return -1;
   }
 
   GetElementPtrInst *Gep = dyn_cast_or_null<GetElementPtrInst>(Ptr);
   if (!Gep)
-    return false;
+    return 0;
 
   unsigned NumOperands = Gep->getNumOperands();
   Value *LastIndex = Gep->getOperand(NumOperands - 1);
 
+  Value *GpPtr = Gep->getPointerOperand();
+  // If this GEP value is a consecutive pointer induction variable and all of
+  // the indices are constant then we know it is consecutive. We can
+  Phi = dyn_cast<PHINode>(GpPtr);
+  if (Phi && Inductions.count(Phi)) {
+
+    // Make sure that the pointer does not point to structs.
+    PointerType *GepPtrType = cast<PointerType>(GpPtr->getType());
+    if (GepPtrType->getElementType()->isAggregateType())
+      return 0;
+
+    // Make sure that all of the index operands are loop invariant.
+    for (unsigned i = 1; i < NumOperands; ++i)
+      if (!SE->isLoopInvariant(SE->getSCEV(Gep->getOperand(i)), TheLoop))
+        return 0;
+
+    InductionInfo II = Inductions[Phi];
+    if (IK_PtrInduction == II.IK)
+      return 1;
+    else if (IK_ReversePtrInduction == II.IK)
+      return -1;
+  }
+
   // Check that all of the gep indices are uniform except for the last.
   for (unsigned i = 0; i < NumOperands - 1; ++i)
     if (!SE->isLoopInvariant(SE->getSCEV(Gep->getOperand(i)), TheLoop))
-      return false;
+      return 0;
 
   // We can emit wide load/stores only if the last index is the induction
   // variable.
@@ -235,39 +854,153 @@ bool LoopVectorizationLegality::isConsecutivePtr(Value *Ptr) {
     // The memory is consecutive because the last index is consecutive
     // and all other indices are loop invariant.
     if (Step->isOne())
-      return true;
+      return 1;
+    if (Step->isAllOnesValue())
+      return -1;
   }
 
-  return false;
+  return 0;
 }
 
 bool LoopVectorizationLegality::isUniform(Value *V) {
   return (SE->isLoopInvariant(SE->getSCEV(V), TheLoop));
 }
 
-Value *InnerLoopVectorizer::getVectorValue(Value *V) {
+InnerLoopVectorizer::VectorParts&
+InnerLoopVectorizer::getVectorValue(Value *V) {
   assert(V != Induction && "The new induction variable should not be used.");
   assert(!V->getType()->isVectorTy() && "Can't widen a vector");
-  // If we saved a vectorized copy of V, use it.
-  Value *&MapEntry = WidenMap[V];
-  if (MapEntry)
-    return MapEntry;
 
-  // Broadcast V and save the value for future uses.
+  // If we have this scalar in the map, return it.
+  if (WidenMap.has(V))
+    return WidenMap.get(V);
+
+  // If this scalar is unknown, assume that it is a constant or that it is
+  // loop invariant. Broadcast V and save the value for future uses.
   Value *B = getBroadcastInstrs(V);
-  MapEntry = B;
-  return B;
+  return WidenMap.splat(V, B);
 }
 
-Constant*
-InnerLoopVectorizer::getUniformVector(unsigned Val, Type* ScalarTy) {
-  return ConstantVector::getSplat(VF, ConstantInt::get(ScalarTy, Val, true));
+Value *InnerLoopVectorizer::reverseVector(Value *Vec) {
+  assert(Vec->getType()->isVectorTy() && "Invalid type");
+  SmallVector<Constant*, 8> ShuffleMask;
+  for (unsigned i = 0; i < VF; ++i)
+    ShuffleMask.push_back(Builder.getInt32(VF - i - 1));
+
+  return Builder.CreateShuffleVector(Vec, UndefValue::get(Vec->getType()),
+                                     ConstantVector::get(ShuffleMask),
+                                     "reverse");
+}
+
+
+void InnerLoopVectorizer::vectorizeMemoryInstruction(Instruction *Instr,
+                                             LoopVectorizationLegality *Legal) {
+  // Attempt to issue a wide load.
+  LoadInst *LI = dyn_cast<LoadInst>(Instr);
+  StoreInst *SI = dyn_cast<StoreInst>(Instr);
+
+  assert((LI || SI) && "Invalid Load/Store instruction");
+
+  Type *ScalarDataTy = LI ? LI->getType() : SI->getValueOperand()->getType();
+  Type *DataTy = VectorType::get(ScalarDataTy, VF);
+  Value *Ptr = LI ? LI->getPointerOperand() : SI->getPointerOperand();
+  unsigned Alignment = LI ? LI->getAlignment() : SI->getAlignment();
+
+  // If the pointer is loop invariant or if it is non consecutive,
+  // scalarize the load.
+  int Stride = Legal->isConsecutivePtr(Ptr);
+  bool Reverse = Stride < 0;
+  bool UniformLoad = LI && Legal->isUniform(Ptr);
+  if (Stride == 0 || UniformLoad)
+    return scalarizeInstruction(Instr);
+
+  Constant *Zero = Builder.getInt32(0);
+  VectorParts &Entry = WidenMap.get(Instr);
+
+  // Handle consecutive loads/stores.
+  GetElementPtrInst *Gep = dyn_cast<GetElementPtrInst>(Ptr);
+  if (Gep && Legal->isInductionVariable(Gep->getPointerOperand())) {
+    Value *PtrOperand = Gep->getPointerOperand();
+    Value *FirstBasePtr = getVectorValue(PtrOperand)[0];
+    FirstBasePtr = Builder.CreateExtractElement(FirstBasePtr, Zero);
+
+    // Create the new GEP with the new induction variable.
+    GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone());
+    Gep2->setOperand(0, FirstBasePtr);
+    Gep2->setName("gep.indvar.base");
+    Ptr = Builder.Insert(Gep2);
+  } else if (Gep) {
+    assert(SE->isLoopInvariant(SE->getSCEV(Gep->getPointerOperand()),
+                               OrigLoop) && "Base ptr must be invariant");
+
+    // The last index does not have to be the induction. It can be
+    // consecutive and be a function of the index. For example A[I+1];
+    unsigned NumOperands = Gep->getNumOperands();
+
+    Value *LastGepOperand = Gep->getOperand(NumOperands - 1);
+    VectorParts &GEPParts = getVectorValue(LastGepOperand);
+    Value *LastIndex = GEPParts[0];
+    LastIndex = Builder.CreateExtractElement(LastIndex, Zero);
+
+    // Create the new GEP with the new induction variable.
+    GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone());
+    Gep2->setOperand(NumOperands - 1, LastIndex);
+    Gep2->setName("gep.indvar.idx");
+    Ptr = Builder.Insert(Gep2);
+  } else {
+    // Use the induction element ptr.
+    assert(isa<PHINode>(Ptr) && "Invalid induction ptr");
+    VectorParts &PtrVal = getVectorValue(Ptr);
+    Ptr = Builder.CreateExtractElement(PtrVal[0], Zero);
+  }
+
+  // Handle Stores:
+  if (SI) {
+    assert(!Legal->isUniform(SI->getPointerOperand()) &&
+           "We do not allow storing to uniform addresses");
+
+    VectorParts &StoredVal = getVectorValue(SI->getValueOperand());
+    for (unsigned Part = 0; Part < UF; ++Part) {
+      // Calculate the pointer for the specific unroll-part.
+      Value *PartPtr = Builder.CreateGEP(Ptr, Builder.getInt32(Part * VF));
+
+      if (Reverse) {
+        // If we store to reverse consecutive memory locations then we need
+        // to reverse the order of elements in the stored value.
+        StoredVal[Part] = reverseVector(StoredVal[Part]);
+        // If the address is consecutive but reversed, then the
+        // wide store needs to start at the last vector element.
+        PartPtr = Builder.CreateGEP(Ptr, Builder.getInt32(-Part * VF));
+        PartPtr = Builder.CreateGEP(PartPtr, Builder.getInt32(1 - VF));
+      }
+
+      Value *VecPtr = Builder.CreateBitCast(PartPtr, DataTy->getPointerTo());
+      Builder.CreateStore(StoredVal[Part], VecPtr)->setAlignment(Alignment);
+    }
+  }
+
+  for (unsigned Part = 0; Part < UF; ++Part) {
+    // Calculate the pointer for the specific unroll-part.
+    Value *PartPtr = Builder.CreateGEP(Ptr, Builder.getInt32(Part * VF));
+
+    if (Reverse) {
+      // If the address is consecutive but reversed, then the
+      // wide store needs to start at the last vector element.
+      PartPtr = Builder.CreateGEP(Ptr, Builder.getInt32(-Part * VF));
+      PartPtr = Builder.CreateGEP(PartPtr, Builder.getInt32(1 - VF));
+    }
+
+    Value *VecPtr = Builder.CreateBitCast(PartPtr, DataTy->getPointerTo());
+    Value *LI = Builder.CreateLoad(VecPtr, "wide.load");
+    cast<LoadInst>(LI)->setAlignment(Alignment);
+    Entry[Part] = Reverse ? reverseVector(LI) :  LI;
+  }
 }
 
 void InnerLoopVectorizer::scalarizeInstruction(Instruction *Instr) {
   assert(!Instr->getType()->isAggregateType() && "Can't handle vectors");
   // Holds vector parameters or scalars, in case of uniform vals.
-  SmallVector<Value*, 8> Params;
+  SmallVector<VectorParts, 4> Params;
 
   // Find all of the vectorized parameters.
   for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
@@ -284,13 +1017,15 @@ void InnerLoopVectorizer::scalarizeInstruction(Instruction *Instr) {
 
     // If the src is an instruction that appeared earlier in the basic block
     // then it should already be vectorized.
-    if (SrcInst && SrcInst->getParent() == Instr->getParent()) {
-      assert(WidenMap.count(SrcInst) && "Source operand is unavailable");
+    if (SrcInst && OrigLoop->contains(SrcInst)) {
+      assert(WidenMap.has(SrcInst) && "Source operand is unavailable");
       // The parameter is a vector value from earlier.
-      Params.push_back(WidenMap[SrcInst]);
+      Params.push_back(WidenMap.get(SrcInst));
     } else {
       // The parameter is a scalar from outside the loop. Maybe even a constant.
-      Params.push_back(SrcOp);
+      VectorParts Scalars;
+      Scalars.append(UF, SrcOp);
+      Params.push_back(Scalars);
     }
   }
 
@@ -299,42 +1034,41 @@ void InnerLoopVectorizer::scalarizeInstruction(Instruction *Instr) {
 
   // Does this instruction return a value ?
   bool IsVoidRetTy = Instr->getType()->isVoidTy();
-  Value *VecResults = 0;
 
-  // If we have a return value, create an empty vector. We place the scalarized
-  // instructions in this vector.
-  if (!IsVoidRetTy)
-    VecResults = UndefValue::get(VectorType::get(Instr->getType(), VF));
+  Value *UndefVec = IsVoidRetTy ? 0 :
+    UndefValue::get(VectorType::get(Instr->getType(), VF));
+  // Create a new entry in the WidenMap and initialize it to Undef or Null.
+  VectorParts &VecResults = WidenMap.splat(Instr, UndefVec);
 
   // For each scalar that we create:
-  for (unsigned i = 0; i < VF; ++i) {
-    Instruction *Cloned = Instr->clone();
-    if (!IsVoidRetTy)
-      Cloned->setName(Instr->getName() + ".cloned");
-    // Replace the operands of the cloned instrucions with extracted scalars.
-    for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
-      Value *Op = Params[op];
-      // Param is a vector. Need to extract the right lane.
-      if (Op->getType()->isVectorTy())
-        Op = Builder.CreateExtractElement(Op, Builder.getInt32(i));
-      Cloned->setOperand(op, Op);
-    }
+  for (unsigned Width = 0; Width < VF; ++Width) {
+    // For each vector unroll 'part':
+    for (unsigned Part = 0; Part < UF; ++Part) {
+      Instruction *Cloned = Instr->clone();
+      if (!IsVoidRetTy)
+        Cloned->setName(Instr->getName() + ".cloned");
+      // Replace the operands of the cloned instrucions with extracted scalars.
+      for (unsigned op = 0, e = Instr->getNumOperands(); op != e; ++op) {
+        Value *Op = Params[op][Part];
+        // Param is a vector. Need to extract the right lane.
+        if (Op->getType()->isVectorTy())
+          Op = Builder.CreateExtractElement(Op, Builder.getInt32(Width));
+        Cloned->setOperand(op, Op);
+      }
 
-    // Place the cloned scalar in the new loop.
-    Builder.Insert(Cloned);
+      // Place the cloned scalar in the new loop.
+      Builder.Insert(Cloned);
 
-    // If the original scalar returns a value we need to place it in a vector
-    // so that future users will be able to use it.
-    if (!IsVoidRetTy)
-      VecResults = Builder.CreateInsertElement(VecResults, Cloned,
-                                               Builder.getInt32(i));
+      // If the original scalar returns a value we need to place it in a vector
+      // so that future users will be able to use it.
+      if (!IsVoidRetTy)
+        VecResults[Part] = Builder.CreateInsertElement(VecResults[Part], Cloned,
+                                                       Builder.getInt32(Width));
+    }
   }
-
-  if (!IsVoidRetTy)
-    WidenMap[Instr] = VecResults;
 }
 
-Value*
+Instruction *
 InnerLoopVectorizer::addRuntimeCheck(LoopVectorizationLegality *Legal,
                                      Instruction *Loc) {
   LoopVectorizationLegality::RuntimePointerCheck *PtrRtCheck =
@@ -343,7 +1077,7 @@ InnerLoopVectorizer::addRuntimeCheck(LoopVectorizationLegality *Legal,
   if (!PtrRtCheck->Need)
     return NULL;
 
-  Value *MemoryRuntimeCheck = 0;
+  Instruction *MemoryRuntimeCheck = 0;
   unsigned NumPointers = PtrRtCheck->Pointers.size();
   SmallVector<Value* , 2> Starts;
   SmallVector<Value* , 2> Ends;
@@ -372,28 +1106,23 @@ InnerLoopVectorizer::addRuntimeCheck(LoopVectorizationLegality *Legal,
     }
   }
 
+  IRBuilder<> ChkBuilder(Loc);
+
   for (unsigned i = 0; i < NumPointers; ++i) {
     for (unsigned j = i+1; j < NumPointers; ++j) {
-      Instruction::CastOps Op = Instruction::BitCast;
-      Value *Start0 = CastInst::Create(Op, Starts[i], PtrArithTy, "bc", Loc);
-      Value *Start1 = CastInst::Create(Op, Starts[j], PtrArithTy, "bc", Loc);
-      Value *End0 =   CastInst::Create(Op, Ends[i],   PtrArithTy, "bc", Loc);
-      Value *End1 =   CastInst::Create(Op, Ends[j],   PtrArithTy, "bc", Loc);
-
-      Value *Cmp0 = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_ULE,
-                                    Start0, End1, "bound0", Loc);
-      Value *Cmp1 = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_ULE,
-                                    Start1, End0, "bound1", Loc);
-      Value *IsConflict = BinaryOperator::Create(Instruction::And, Cmp0, Cmp1,
-                                                 "found.conflict", Loc);
+      Value *Start0 = ChkBuilder.CreateBitCast(Starts[i], PtrArithTy, "bc");
+      Value *Start1 = ChkBuilder.CreateBitCast(Starts[j], PtrArithTy, "bc");
+      Value *End0 =   ChkBuilder.CreateBitCast(Ends[i],   PtrArithTy, "bc");
+      Value *End1 =   ChkBuilder.CreateBitCast(Ends[j],   PtrArithTy, "bc");
+
+      Value *Cmp0 = ChkBuilder.CreateICmpULE(Start0, End1, "bound0");
+      Value *Cmp1 = ChkBuilder.CreateICmpULE(Start1, End0, "bound1");
+      Value *IsConflict = ChkBuilder.CreateAnd(Cmp0, Cmp1, "found.conflict");
       if (MemoryRuntimeCheck)
-        MemoryRuntimeCheck = BinaryOperator::Create(Instruction::Or,
-                                                    MemoryRuntimeCheck,
-                                                    IsConflict,
-                                                    "conflict.rdx", Loc);
-      else
-        MemoryRuntimeCheck = IsConflict;
+        IsConflict = ChkBuilder.CreateOr(MemoryRuntimeCheck, IsConflict,
+                                         "conflict.rdx");
 
+      MemoryRuntimeCheck = cast<Instruction>(IsConflict);
     }
   }
 
@@ -407,27 +1136,27 @@ InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
    the vectorized instructions while the old loop will continue to run the
    scalar remainder.
 
-   [ ] <-- vector loop bypass.
-   /  |
-   /   v
+       [ ] <-- vector loop bypass (may consist of multiple blocks).
+     /  |
+    /   v
    |   [ ]     <-- vector pre header.
    |    |
    |    v
    |   [  ] \
    |   [  ]_|   <-- vector loop.
    |    |
-   \   v
-   >[ ]   <--- middle-block.
-   /  |
-   /   v
+    \   v
+      >[ ]   <--- middle-block.
+     /  |
+    /   v
    |   [ ]     <--- new preheader.
    |    |
    |    v
    |   [ ] \
    |   [ ]_|   <-- old scalar loop to handle remainder.
-   \   |
-   \  v
-   >[ ]     <-- exit block.
+    \   |
+     \  v
+      >[ ]     <-- exit block.
    ...
    */
 
@@ -436,6 +1165,11 @@ InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
   BasicBlock *ExitBlock = OrigLoop->getExitBlock();
   assert(ExitBlock && "Must have an exit block");
 
+  // Mark the old scalar loop with metadata that tells us not to vectorize this
+  // loop again if we run into it.
+  MDNode *MD = MDNode::get(OldBasicBlock->getContext(), ArrayRef<Value*>());
+  OldBasicBlock->getTerminator()->setMetadata(AlreadyVectorizedMDName, MD);
+
   // Some loops have a single integer induction variable, while other loops
   // don't. One example is c++ iterators that often have multiple pointer
   // induction variables. In the code below we also support a case where we
@@ -468,10 +1202,7 @@ InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
   ConstantInt::get(IdxTy, 0);
 
   assert(BypassBlock && "Invalid loop structure");
-
-  // Generate the code that checks in runtime if arrays overlap.
-  Value *MemoryRuntimeCheck = addRuntimeCheck(Legal,
-                                              BypassBlock->getTerminator());
+  LoopBypassBlocks.push_back(BypassBlock);
 
   // Split the single block loop into the two loop structure described above.
   BasicBlock *VectorPH =
@@ -483,17 +1214,19 @@ InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
   BasicBlock *ScalarPH =
   MiddleBlock->splitBasicBlock(MiddleBlock->getTerminator(), "scalar.ph");
 
-  // This is the location in which we add all of the logic for bypassing
-  // the new vector loop.
-  Instruction *Loc = BypassBlock->getTerminator();
-
   // Use this IR builder to create the loop instructions (Phi, Br, Cmp)
   // inside the loop.
   Builder.SetInsertPoint(VecBody->getFirstInsertionPt());
 
   // Generate the induction variable.
   Induction = Builder.CreatePHI(IdxTy, 2, "index");
-  Constant *Step = ConstantInt::get(IdxTy, VF);
+  // The loop step is equal to the vectorization factor (num of SIMD elements)
+  // times the unroll factor (num of SIMD instructions).
+  Constant *Step = ConstantInt::get(IdxTy, VF * UF);
+
+  // This is the IR builder that we use to add all of the logic for bypassing
+  // the new vector loop.
+  IRBuilder<> BypassBuilder(BypassBlock->getTerminator());
 
   // We may need to extend the index in case there is a type mismatch.
   // We know that the count starts at zero and does not overflow.
@@ -501,37 +1234,52 @@ InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
     // The exit count can be of pointer type. Convert it to the correct
     // integer type.
     if (ExitCount->getType()->isPointerTy())
-      Count = CastInst::CreatePointerCast(Count, IdxTy, "ptrcnt.to.int", Loc);
+      Count = BypassBuilder.CreatePointerCast(Count, IdxTy, "ptrcnt.to.int");
     else
-      Count = CastInst::CreateZExtOrBitCast(Count, IdxTy, "zext.cnt", Loc);
+      Count = BypassBuilder.CreateZExtOrTrunc(Count, IdxTy, "cnt.cast");
   }
 
   // Add the start index to the loop count to get the new end index.
-  Value *IdxEnd = BinaryOperator::CreateAdd(Count, StartIdx, "end.idx", Loc);
+  Value *IdxEnd = BypassBuilder.CreateAdd(Count, StartIdx, "end.idx");
 
   // Now we need to generate the expression for N - (N % VF), which is
   // the part that the vectorized body will execute.
-  Constant *CIVF = ConstantInt::get(IdxTy, VF);
-  Value *R = BinaryOperator::CreateURem(Count, CIVF, "n.mod.vf", Loc);
-  Value *CountRoundDown = BinaryOperator::CreateSub(Count, R, "n.vec", Loc);
-  Value *IdxEndRoundDown = BinaryOperator::CreateAdd(CountRoundDown, StartIdx,
-                                                     "end.idx.rnd.down", Loc);
+  Value *R = BypassBuilder.CreateURem(Count, Step, "n.mod.vf");
+  Value *CountRoundDown = BypassBuilder.CreateSub(Count, R, "n.vec");
+  Value *IdxEndRoundDown = BypassBuilder.CreateAdd(CountRoundDown, StartIdx,
+                                                     "end.idx.rnd.down");
 
   // Now, compare the new count to zero. If it is zero skip the vector loop and
   // jump to the scalar loop.
-  Value *Cmp = CmpInst::Create(Instruction::ICmp, CmpInst::ICMP_EQ,
-                               IdxEndRoundDown,
-                               StartIdx,
-                               "cmp.zero", Loc);
-
-  // If we are using memory runtime checks, include them in.
-  if (MemoryRuntimeCheck)
-    Cmp = BinaryOperator::Create(Instruction::Or, Cmp, MemoryRuntimeCheck,
-                                 "CntOrMem", Loc);
+  Value *Cmp = BypassBuilder.CreateICmpEQ(IdxEndRoundDown, StartIdx,
+                                          "cmp.zero");
+
+  BasicBlock *LastBypassBlock = BypassBlock;
+
+  // Generate the code that checks in runtime if arrays overlap. We put the
+  // checks into a separate block to make the more common case of few elements
+  // faster.
+  Instruction *MemRuntimeCheck = addRuntimeCheck(Legal,
+                                                 BypassBlock->getTerminator());
+  if (MemRuntimeCheck) {
+    // Create a new block containing the memory check.
+    BasicBlock *CheckBlock = BypassBlock->splitBasicBlock(MemRuntimeCheck,
+                                                          "vector.memcheck");
+    LoopBypassBlocks.push_back(CheckBlock);
+
+    // Replace the branch into the memory check block with a conditional branch
+    // for the "few elements case".
+    Instruction *OldTerm = BypassBlock->getTerminator();
+    BranchInst::Create(MiddleBlock, CheckBlock, Cmp, OldTerm);
+    OldTerm->eraseFromParent();
+
+    Cmp = MemRuntimeCheck;
+    LastBypassBlock = CheckBlock;
+  }
 
-  BranchInst::Create(MiddleBlock, VectorPH, Cmp, Loc);
-  // Remove the old terminator.
-  Loc->eraseFromParent();
+  LastBypassBlock->getTerminator()->eraseFromParent();
+  BranchInst::Create(MiddleBlock, VectorPH, Cmp,
+                     LastBypassBlock);
 
   // We are going to resume the execution of the scalar loop.
   // Go over all of the induction variables that we found and fix the
@@ -552,9 +1300,9 @@ InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
                                          MiddleBlock->getTerminator());
     Value *EndValue = 0;
     switch (II.IK) {
-    case LoopVectorizationLegality::NoInduction:
+    case LoopVectorizationLegality::IK_NoInduction:
       llvm_unreachable("Unknown induction");
-    case LoopVectorizationLegality::IntInduction: {
+    case LoopVectorizationLegality::IK_IntInduction: {
       // Handle the integer induction counter:
       assert(OrigPhi->getType()->isIntegerTy() && "Invalid type");
       assert(OrigPhi == OldInduction && "Unknown integer PHI");
@@ -564,37 +1312,52 @@ InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
       ResumeIndex = ResumeVal;
       break;
     }
-    case LoopVectorizationLegality::ReverseIntInduction: {
+    case LoopVectorizationLegality::IK_ReverseIntInduction: {
       // Convert the CountRoundDown variable to the PHI size.
       unsigned CRDSize = CountRoundDown->getType()->getScalarSizeInBits();
       unsigned IISize = II.StartValue->getType()->getScalarSizeInBits();
       Value *CRD = CountRoundDown;
       if (CRDSize > IISize)
         CRD = CastInst::Create(Instruction::Trunc, CountRoundDown,
-                               II.StartValue->getType(),
-                               "tr.crd", BypassBlock->getTerminator());
+                               II.StartValue->getType(), "tr.crd",
+                               LoopBypassBlocks.back()->getTerminator());
       else if (CRDSize < IISize)
         CRD = CastInst::Create(Instruction::SExt, CountRoundDown,
                                II.StartValue->getType(),
-                               "sext.crd", BypassBlock->getTerminator());
+                               "sext.crd",
+                               LoopBypassBlocks.back()->getTerminator());
       // Handle reverse integer induction counter:
-      EndValue = BinaryOperator::CreateSub(II.StartValue, CRD, "rev.ind.end",
-                                           BypassBlock->getTerminator());
+      EndValue =
+        BinaryOperator::CreateSub(II.StartValue, CRD, "rev.ind.end",
+                                  LoopBypassBlocks.back()->getTerminator());
       break;
     }
-    case LoopVectorizationLegality::PtrInduction: {
+    case LoopVectorizationLegality::IK_PtrInduction: {
       // For pointer induction variables, calculate the offset using
       // the end index.
-      EndValue = GetElementPtrInst::Create(II.StartValue, CountRoundDown,
-                                           "ptr.ind.end",
-                                           BypassBlock->getTerminator());
+      EndValue =
+        GetElementPtrInst::Create(II.StartValue, CountRoundDown, "ptr.ind.end",
+                                  LoopBypassBlocks.back()->getTerminator());
+      break;
+    }
+    case LoopVectorizationLegality::IK_ReversePtrInduction: {
+      // The value at the end of the loop for the reverse pointer is calculated
+      // by creating a GEP with a negative index starting from the start value.
+      Value *Zero = ConstantInt::get(CountRoundDown->getType(), 0);
+      Value *NegIdx = BinaryOperator::CreateSub(Zero, CountRoundDown,
+                                  "rev.ind.end",
+                                  LoopBypassBlocks.back()->getTerminator());
+      EndValue = GetElementPtrInst::Create(II.StartValue, NegIdx,
+                                  "rev.ptr.ind.end",
+                                  LoopBypassBlocks.back()->getTerminator());
       break;
     }
     }// end of case
 
     // The new PHI merges the original incoming value, in case of a bypass,
     // or the value at the end of the vectorized loop.
-    ResumeVal->addIncoming(II.StartValue, BypassBlock);
+    for (unsigned I = 0, E = LoopBypassBlocks.size(); I != E; ++I)
+      ResumeVal->addIncoming(II.StartValue, LoopBypassBlocks[I]);
     ResumeVal->addIncoming(EndValue, VecBody);
 
     // Fix the scalar body counter (PHI node).
@@ -610,7 +1373,8 @@ InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
     assert(!ResumeIndex && "Unexpected resume value found");
     ResumeIndex = PHINode::Create(IdxTy, 2, "new.indc.resume.val",
                                   MiddleBlock->getTerminator());
-    ResumeIndex->addIncoming(StartIdx, BypassBlock);
+    for (unsigned I = 0, E = LoopBypassBlocks.size(); I != E; ++I)
+      ResumeIndex->addIncoming(StartIdx, LoopBypassBlocks[I]);
     ResumeIndex->addIncoming(IdxEndRoundDown, VecBody);
   }
 
@@ -650,6 +1414,8 @@ InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
   // Insert the new loop into the loop nest and register the new basic blocks.
   if (ParentLoop) {
     ParentLoop->addChildLoop(Lp);
+    for (unsigned I = 1, E = LoopBypassBlocks.size(); I != E; ++I)
+      ParentLoop->addBasicBlockToLoop(LoopBypassBlocks[I], LI->getBase());
     ParentLoop->addBasicBlockToLoop(ScalarPH, LI->getBase());
     ParentLoop->addBasicBlockToLoop(VectorPH, LI->getBase());
     ParentLoop->addBasicBlockToLoop(MiddleBlock, LI->getBase());
@@ -666,57 +1432,160 @@ InnerLoopVectorizer::createEmptyLoop(LoopVectorizationLegality *Legal) {
   LoopExitBlock = ExitBlock;
   LoopVectorBody = VecBody;
   LoopScalarBody = OldBasicBlock;
-  LoopBypassBlock = BypassBlock;
 }
 
 /// This function returns the identity element (or neutral element) for
 /// the operation K.
-static unsigned
-getReductionIdentity(LoopVectorizationLegality::ReductionKind K) {
+static Constant*
+getReductionIdentity(LoopVectorizationLegality::ReductionKind K, Type *Tp) {
   switch (K) {
-  case LoopVectorizationLegality::IntegerXor:
-  case LoopVectorizationLegality::IntegerAdd:
-  case LoopVectorizationLegality::IntegerOr:
+  case LoopVectorizationLegality:: RK_IntegerXor:
+  case LoopVectorizationLegality:: RK_IntegerAdd:
+  case LoopVectorizationLegality:: RK_IntegerOr:
     // Adding, Xoring, Oring zero to a number does not change it.
-    return 0;
-  case LoopVectorizationLegality::IntegerMult:
+    return ConstantInt::get(Tp, 0);
+  case LoopVectorizationLegality:: RK_IntegerMult:
     // Multiplying a number by 1 does not change it.
-    return 1;
-  case LoopVectorizationLegality::IntegerAnd:
+    return ConstantInt::get(Tp, 1);
+  case LoopVectorizationLegality:: RK_IntegerAnd:
     // AND-ing a number with an all-1 value does not change it.
-    return -1;
+    return ConstantInt::get(Tp, -1, true);
+  case LoopVectorizationLegality:: RK_FloatMult:
+    // Multiplying a number by 1 does not change it.
+    return ConstantFP::get(Tp, 1.0L);
+  case LoopVectorizationLegality:: RK_FloatAdd:
+    // Adding zero to a number does not change it.
+    return ConstantFP::get(Tp, 0.0L);
   default:
     llvm_unreachable("Unknown reduction kind");
   }
 }
 
-static bool
-isTriviallyVectorizableIntrinsic(Instruction *Inst) {
-  IntrinsicInst *II = dyn_cast<IntrinsicInst>(Inst);
-  if (!II)
-    return false;
-  switch (II->getIntrinsicID()) {
-  case Intrinsic::sqrt:
-  case Intrinsic::sin:
-  case Intrinsic::cos:
-  case Intrinsic::exp:
-  case Intrinsic::exp2:
-  case Intrinsic::log:
-  case Intrinsic::log10:
-  case Intrinsic::log2:
-  case Intrinsic::fabs:
-  case Intrinsic::floor:
-  case Intrinsic::ceil:
-  case Intrinsic::trunc:
-  case Intrinsic::rint:
-  case Intrinsic::nearbyint:
-  case Intrinsic::pow:
-  case Intrinsic::fma:
-    return true;
+static Intrinsic::ID
+getIntrinsicIDForCall(CallInst *CI, const TargetLibraryInfo *TLI) {
+  // If we have an intrinsic call, check if it is trivially vectorizable.
+  if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI)) {
+    switch (II->getIntrinsicID()) {
+    case Intrinsic::sqrt:
+    case Intrinsic::sin:
+    case Intrinsic::cos:
+    case Intrinsic::exp:
+    case Intrinsic::exp2:
+    case Intrinsic::log:
+    case Intrinsic::log10:
+    case Intrinsic::log2:
+    case Intrinsic::fabs:
+    case Intrinsic::floor:
+    case Intrinsic::ceil:
+    case Intrinsic::trunc:
+    case Intrinsic::rint:
+    case Intrinsic::nearbyint:
+    case Intrinsic::pow:
+    case Intrinsic::fma:
+    case Intrinsic::fmuladd:
+      return II->getIntrinsicID();
+    default:
+      return Intrinsic::not_intrinsic;
+    }
+  }
+
+  if (!TLI)
+    return Intrinsic::not_intrinsic;
+
+  LibFunc::Func Func;
+  Function *F = CI->getCalledFunction();
+  // We're going to make assumptions on the semantics of the functions, check
+  // that the target knows that it's available in this environment.
+  if (!F || !TLI->getLibFunc(F->getName(), Func))
+    return Intrinsic::not_intrinsic;
+
+  // Otherwise check if we have a call to a function that can be turned into a
+  // vector intrinsic.
+  switch (Func) {
   default:
-    return false;
+    break;
+  case LibFunc::sin:
+  case LibFunc::sinf:
+  case LibFunc::sinl:
+    return Intrinsic::sin;
+  case LibFunc::cos:
+  case LibFunc::cosf:
+  case LibFunc::cosl:
+    return Intrinsic::cos;
+  case LibFunc::exp:
+  case LibFunc::expf:
+  case LibFunc::expl:
+    return Intrinsic::exp;
+  case LibFunc::exp2:
+  case LibFunc::exp2f:
+  case LibFunc::exp2l:
+    return Intrinsic::exp2;
+  case LibFunc::log:
+  case LibFunc::logf:
+  case LibFunc::logl:
+    return Intrinsic::log;
+  case LibFunc::log10:
+  case LibFunc::log10f:
+  case LibFunc::log10l:
+    return Intrinsic::log10;
+  case LibFunc::log2:
+  case LibFunc::log2f:
+  case LibFunc::log2l:
+    return Intrinsic::log2;
+  case LibFunc::fabs:
+  case LibFunc::fabsf:
+  case LibFunc::fabsl:
+    return Intrinsic::fabs;
+  case LibFunc::floor:
+  case LibFunc::floorf:
+  case LibFunc::floorl:
+    return Intrinsic::floor;
+  case LibFunc::ceil:
+  case LibFunc::ceilf:
+  case LibFunc::ceill:
+    return Intrinsic::ceil;
+  case LibFunc::trunc:
+  case LibFunc::truncf:
+  case LibFunc::truncl:
+    return Intrinsic::trunc;
+  case LibFunc::rint:
+  case LibFunc::rintf:
+  case LibFunc::rintl:
+    return Intrinsic::rint;
+  case LibFunc::nearbyint:
+  case LibFunc::nearbyintf:
+  case LibFunc::nearbyintl:
+    return Intrinsic::nearbyint;
+  case LibFunc::pow:
+  case LibFunc::powf:
+  case LibFunc::powl:
+    return Intrinsic::pow;
+  }
+
+  return Intrinsic::not_intrinsic;
+}
+
+/// This function translates the reduction kind to an LLVM binary operator.
+static Instruction::BinaryOps
+getReductionBinOp(LoopVectorizationLegality::ReductionKind Kind) {
+  switch (Kind) {
+    case LoopVectorizationLegality::RK_IntegerAdd:
+      return Instruction::Add;
+    case LoopVectorizationLegality::RK_IntegerMult:
+      return Instruction::Mul;
+    case LoopVectorizationLegality::RK_IntegerOr:
+      return Instruction::Or;
+    case LoopVectorizationLegality::RK_IntegerAnd:
+      return Instruction::And;
+    case LoopVectorizationLegality::RK_IntegerXor:
+      return Instruction::Xor;
+    case LoopVectorizationLegality::RK_FloatMult:
+      return Instruction::FMul;
+    case LoopVectorizationLegality::RK_FloatAdd:
+      return Instruction::FAdd;
+    default:
+      llvm_unreachable("Unknown reduction operation");
   }
-  return false;
 }
 
 void
@@ -728,9 +1597,7 @@ InnerLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
   // the cost-model.
   //
   //===------------------------------------------------===//
-  BasicBlock &BB = *OrigLoop->getHeader();
-  Constant *Zero =
-  ConstantInt::get(IntegerType::getInt32Ty(BB.getContext()), 0);
+  Constant *Zero = Builder.getInt32(0);
 
   // In order to support reduction variables we need to be able to vectorize
   // Phi nodes. Phi nodes have cycles, so we need to vectorize them in two
@@ -764,7 +1631,6 @@ InnerLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
   for (PhiVector::iterator it = RdxPHIsToFix.begin(), e = RdxPHIsToFix.end();
        it != e; ++it) {
     PHINode *RdxPhi = *it;
-    PHINode *VecRdxPhi = dyn_cast<PHINode>(WidenMap[RdxPhi]);
     assert(RdxPhi && "Unable to recover vectorized PHI");
 
     // Find the reduction variable descriptor.
@@ -777,16 +1643,16 @@ InnerLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
     // To do so, we need to generate the 'identity' vector and overide
     // one of the elements with the incoming scalar reduction. We need
     // to do it in the vector-loop preheader.
-    Builder.SetInsertPoint(LoopBypassBlock->getTerminator());
+    Builder.SetInsertPoint(LoopBypassBlocks.front()->getTerminator());
 
     // This is the vector-clone of the value that leaves the loop.
-    Value *VectorExit = getVectorValue(RdxDesc.LoopExitInstr);
-    Type *VecTy = VectorExit->getType();
+    VectorParts &VectorExit = getVectorValue(RdxDesc.LoopExitInstr);
+    Type *VecTy = VectorExit[0]->getType();
 
     // Find the reduction identity variable. Zero for addition, or, xor,
     // one for multiplication, -1 for And.
-    Constant *Identity = getUniformVector(getReductionIdentity(RdxDesc.Kind),
-                                          VecTy->getScalarType());
+    Constant *Iden = getReductionIdentity(RdxDesc.Kind, VecTy->getScalarType());
+    Constant *Identity = ConstantVector::getSplat(VF, Iden);
 
     // This vector is the Identity vector where the first element is the
     // incoming scalar reduction.
@@ -800,10 +1666,17 @@ InnerLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
 
     // Reductions do not have to start at zero. They can start with
     // any loop invariant values.
-    VecRdxPhi->addIncoming(VectorStart, VecPreheader);
-    Value *Val =
-    getVectorValue(RdxPhi->getIncomingValueForBlock(OrigLoop->getLoopLatch()));
-    VecRdxPhi->addIncoming(Val, LoopVectorBody);
+    VectorParts &VecRdxPhi = WidenMap.get(RdxPhi);
+    BasicBlock *Latch = OrigLoop->getLoopLatch();
+    Value *LoopVal = RdxPhi->getIncomingValueForBlock(Latch);
+    VectorParts &Val = getVectorValue(LoopVal);
+    for (unsigned part = 0; part < UF; ++part) {
+      // Make sure to add the reduction stat value only to the 
+      // first unroll part.
+      Value *StartVal = (part == 0) ? VectorStart : Identity;
+      cast<PHINode>(VecRdxPhi[part])->addIncoming(StartVal, VecPreheader);
+      cast<PHINode>(VecRdxPhi[part])->addIncoming(Val[part], LoopVectorBody);
+    }
 
     // Before each round, move the insertion point right between
     // the PHIs and the values we are going to write.
@@ -811,40 +1684,56 @@ InnerLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
     // instructions.
     Builder.SetInsertPoint(LoopMiddleBlock->getFirstInsertionPt());
 
-    // This PHINode contains the vectorized reduction variable, or
-    // the initial value vector, if we bypass the vector loop.
-    PHINode *NewPhi = Builder.CreatePHI(VecTy, 2, "rdx.vec.exit.phi");
-    NewPhi->addIncoming(VectorStart, LoopBypassBlock);
-    NewPhi->addIncoming(getVectorValue(RdxDesc.LoopExitInstr), LoopVectorBody);
-
-    // Extract the first scalar.
-    Value *Scalar0 =
-    Builder.CreateExtractElement(NewPhi, Builder.getInt32(0));
-    // Extract and reduce the remaining vector elements.
-    for (unsigned i=1; i < VF; ++i) {
-      Value *Scalar1 =
-      Builder.CreateExtractElement(NewPhi, Builder.getInt32(i));
-      switch (RdxDesc.Kind) {
-      case LoopVectorizationLegality::IntegerAdd:
-        Scalar0 = Builder.CreateAdd(Scalar0, Scalar1, "add.rdx");
-        break;
-      case LoopVectorizationLegality::IntegerMult:
-        Scalar0 = Builder.CreateMul(Scalar0, Scalar1, "mul.rdx");
-        break;
-      case LoopVectorizationLegality::IntegerOr:
-        Scalar0 = Builder.CreateOr(Scalar0, Scalar1, "or.rdx");
-        break;
-      case LoopVectorizationLegality::IntegerAnd:
-        Scalar0 = Builder.CreateAnd(Scalar0, Scalar1, "and.rdx");
-        break;
-      case LoopVectorizationLegality::IntegerXor:
-        Scalar0 = Builder.CreateXor(Scalar0, Scalar1, "xor.rdx");
-        break;
-      default:
-        llvm_unreachable("Unknown reduction operation");
-      }
+    VectorParts RdxParts;
+    for (unsigned part = 0; part < UF; ++part) {
+      // This PHINode contains the vectorized reduction variable, or
+      // the initial value vector, if we bypass the vector loop.
+      VectorParts &RdxExitVal = getVectorValue(RdxDesc.LoopExitInstr);
+      PHINode *NewPhi = Builder.CreatePHI(VecTy, 2, "rdx.vec.exit.phi");
+      Value *StartVal = (part == 0) ? VectorStart : Identity;
+      for (unsigned I = 0, E = LoopBypassBlocks.size(); I != E; ++I)
+        NewPhi->addIncoming(StartVal, LoopBypassBlocks[I]);
+      NewPhi->addIncoming(RdxExitVal[part], LoopVectorBody);
+      RdxParts.push_back(NewPhi);
+    }
+
+    // Reduce all of the unrolled parts into a single vector.
+    Value *ReducedPartRdx = RdxParts[0];
+    for (unsigned part = 1; part < UF; ++part) {
+      Instruction::BinaryOps Op = getReductionBinOp(RdxDesc.Kind);
+      ReducedPartRdx = Builder.CreateBinOp(Op, RdxParts[part], ReducedPartRdx,
+                                           "bin.rdx");
     }
 
+    // VF is a power of 2 so we can emit the reduction using log2(VF) shuffles
+    // and vector ops, reducing the set of values being computed by half each
+    // round.
+    assert(isPowerOf2_32(VF) &&
+           "Reduction emission only supported for pow2 vectors!");
+    Value *TmpVec = ReducedPartRdx;
+    SmallVector<Constant*, 32> ShuffleMask(VF, 0);
+    for (unsigned i = VF; i != 1; i >>= 1) {
+      // Move the upper half of the vector to the lower half.
+      for (unsigned j = 0; j != i/2; ++j)
+        ShuffleMask[j] = Builder.getInt32(i/2 + j);
+
+      // Fill the rest of the mask with undef.
+      std::fill(&ShuffleMask[i/2], ShuffleMask.end(),
+                UndefValue::get(Builder.getInt32Ty()));
+
+      Value *Shuf =
+        Builder.CreateShuffleVector(TmpVec,
+                                    UndefValue::get(TmpVec->getType()),
+                                    ConstantVector::get(ShuffleMask),
+                                    "rdx.shuf");
+
+      Instruction::BinaryOps Op = getReductionBinOp(RdxDesc.Kind);
+      TmpVec = Builder.CreateBinOp(Op, TmpVec, Shuf, "bin.rdx");
+    }
+
+    // The result is in the first element of the vector.
+    Value *Scalar0 = Builder.CreateExtractElement(TmpVec, Builder.getInt32(0));
+
     // Now, we need to fix the users of the reduction variable
     // inside and outside of the scalar remainder loop.
     // We know that the loop is in LCSSA form. We need to update the
@@ -877,29 +1766,49 @@ InnerLoopVectorizer::vectorizeLoop(LoopVectorizationLegality *Legal) {
     (RdxPhi)->setIncomingValue(SelfEdgeBlockIdx, Scalar0);
     (RdxPhi)->setIncomingValue(IncomingEdgeBlockIdx, RdxDesc.LoopExitInstr);
   }// end of for each redux variable.
+
+  // The Loop exit block may have single value PHI nodes where the incoming
+  // value is 'undef'. While vectorizing we only handled real values that
+  // were defined inside the loop. Here we handle the 'undef case'.
+  // See PR14725.
+  for (BasicBlock::iterator LEI = LoopExitBlock->begin(),
+       LEE = LoopExitBlock->end(); LEI != LEE; ++LEI) {
+    PHINode *LCSSAPhi = dyn_cast<PHINode>(LEI);
+    if (!LCSSAPhi) continue;
+    if (LCSSAPhi->getNumIncomingValues() == 1)
+      LCSSAPhi->addIncoming(UndefValue::get(LCSSAPhi->getType()),
+                            LoopMiddleBlock);
+  }
 }
 
-Value *InnerLoopVectorizer::createEdgeMask(BasicBlock *Src, BasicBlock *Dst) {
+InnerLoopVectorizer::VectorParts
+InnerLoopVectorizer::createEdgeMask(BasicBlock *Src, BasicBlock *Dst) {
   assert(std::find(pred_begin(Dst), pred_end(Dst), Src) != pred_end(Dst) &&
          "Invalid edge");
 
-  Value *SrcMask = createBlockInMask(Src);
+  VectorParts SrcMask = createBlockInMask(Src);
 
   // The terminator has to be a branch inst!
   BranchInst *BI = dyn_cast<BranchInst>(Src->getTerminator());
   assert(BI && "Unexpected terminator found");
 
-  Value *EdgeMask = SrcMask;
   if (BI->isConditional()) {
-    EdgeMask = getVectorValue(BI->getCondition());
+    VectorParts EdgeMask = getVectorValue(BI->getCondition());
+
     if (BI->getSuccessor(0) != Dst)
-      EdgeMask = Builder.CreateNot(EdgeMask);
+      for (unsigned part = 0; part < UF; ++part)
+        EdgeMask[part] = Builder.CreateNot(EdgeMask[part]);
+
+    for (unsigned part = 0; part < UF; ++part)
+      EdgeMask[part] = Builder.CreateAnd(EdgeMask[part], SrcMask[part]);
+    return EdgeMask;
   }
 
-  return Builder.CreateAnd(EdgeMask, SrcMask);
+  return SrcMask;
 }
 
-Value *InnerLoopVectorizer::createBlockInMask(BasicBlock *BB) {
+InnerLoopVectorizer::VectorParts
+InnerLoopVectorizer::createBlockInMask(BasicBlock *BB) {
   assert(OrigLoop->contains(BB) && "Block is not a part of a loop");
 
   // Loop incoming mask is all-one.
@@ -910,11 +1819,14 @@ Value *InnerLoopVectorizer::createBlockInMask(BasicBlock *BB) {
 
   // This is the block mask. We OR all incoming edges, and with zero.
   Value *Zero = ConstantInt::get(IntegerType::getInt1Ty(BB->getContext()), 0);
-  Value *BlockMask = getVectorValue(Zero);
+  VectorParts BlockMask = getVectorValue(Zero);
 
   // For each pred:
-  for (pred_iterator it = pred_begin(BB), e = pred_end(BB); it != e; ++it)
-    BlockMask = Builder.CreateOr(BlockMask, createEdgeMask(*it, BB));
+  for (pred_iterator it = pred_begin(BB), e = pred_end(BB); it != e; ++it) {
+    VectorParts EM = createEdgeMask(*it, BB);
+    for (unsigned part = 0; part < UF; ++part)
+      BlockMask[part] = Builder.CreateOr(BlockMask[part], EM[part]);
+  }
 
   return BlockMask;
 }
@@ -922,11 +1834,9 @@ Value *InnerLoopVectorizer::createBlockInMask(BasicBlock *BB) {
 void
 InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
                                           BasicBlock *BB, PhiVector *PV) {
-  Constant *Zero =
-  ConstantInt::get(IntegerType::getInt32Ty(BB->getContext()), 0);
-
   // For each instruction in the old loop.
   for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
+    VectorParts &Entry = WidenMap.get(it);
     switch (it->getOpcode()) {
     case Instruction::Br:
       // Nothing to do for PHIs and BR, since we already took care of the
@@ -936,11 +1846,12 @@ InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
       PHINode* P = cast<PHINode>(it);
       // Handle reduction variables:
       if (Legal->getReductionVars()->count(P)) {
-        // This is phase one of vectorizing PHIs.
-        Type *VecTy = VectorType::get(it->getType(), VF);
-        WidenMap[it] =
-          PHINode::Create(VecTy, 2, "vec.phi",
-                          LoopVectorBody->getFirstInsertionPt());
+        for (unsigned part = 0; part < UF; ++part) {
+          // This is phase one of vectorizing PHIs.
+          Type *VecTy = VectorType::get(it->getType(), VF);
+          Entry[part] = PHINode::Create(VecTy, 2, "vec.phi",
+                                        LoopVectorBody-> getFirstInsertionPt());
+        }
         PV->push_back(P);
         continue;
       }
@@ -954,12 +1865,15 @@ InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
         // At this point we generate the predication tree. There may be
         // duplications since this is a simple recursive scan, but future
         // optimizations will clean it up.
-        Value *Cond = createBlockInMask(P->getIncomingBlock(0));
-        WidenMap[P] =
-          Builder.CreateSelect(Cond,
-                               getVectorValue(P->getIncomingValue(0)),
-                               getVectorValue(P->getIncomingValue(1)),
-                               "predphi");
+        VectorParts Cond = createEdgeMask(P->getIncomingBlock(0),
+                                               P->getParent());
+
+        for (unsigned part = 0; part < UF; ++part) {
+        VectorParts &In0 = getVectorValue(P->getIncomingValue(0));
+        VectorParts &In1 = getVectorValue(P->getIncomingValue(1));
+          Entry[part] = Builder.CreateSelect(Cond[part], In0[part], In1[part],
+                                             "predphi");
+        }
         continue;
       }
 
@@ -972,19 +1886,20 @@ InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
         Legal->getInductionVars()->lookup(P);
 
       switch (II.IK) {
-      case LoopVectorizationLegality::NoInduction:
+      case LoopVectorizationLegality::IK_NoInduction:
         llvm_unreachable("Unknown induction");
-      case LoopVectorizationLegality::IntInduction: {
+      case LoopVectorizationLegality::IK_IntInduction: {
         assert(P == OldInduction && "Unexpected PHI");
         Value *Broadcasted = getBroadcastInstrs(Induction);
         // After broadcasting the induction variable we need to make the
         // vector consecutive by adding 0, 1, 2 ...
-        Value *ConsecutiveInduction = getConsecutiveVector(Broadcasted);
-        WidenMap[OldInduction] = ConsecutiveInduction;
+        for (unsigned part = 0; part < UF; ++part)
+          Entry[part] = getConsecutiveVector(Broadcasted, VF * part, false);
         continue;
       }
-      case LoopVectorizationLegality::ReverseIntInduction:
-      case LoopVectorizationLegality::PtrInduction:
+      case LoopVectorizationLegality::IK_ReverseIntInduction:
+      case LoopVectorizationLegality::IK_PtrInduction:
+      case LoopVectorizationLegality::IK_ReversePtrInduction:
         // Handle reverse integer and pointer inductions.
         Value *StartIdx = 0;
         // If we have a single integer induction variable then use it.
@@ -1001,7 +1916,7 @@ InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
                                                  "normalized.idx");
 
         // Handle the reverse integer induction variable case.
-        if (LoopVectorizationLegality::ReverseIntInduction == II.IK) {
+        if (LoopVectorizationLegality::IK_ReverseIntInduction == II.IK) {
           IntegerType *DstTy = cast<IntegerType>(II.StartValue->getType());
           Value *CNI = Builder.CreateSExtOrTrunc(NormalizedIdx, DstTy,
                                                  "resize.norm.idx");
@@ -1012,30 +1927,39 @@ InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
           Value *Broadcasted = getBroadcastInstrs(ReverseInd);
           // After broadcasting the induction variable we need to make the
           // vector consecutive by adding  ... -3, -2, -1, 0.
-          Value *ConsecutiveInduction = getConsecutiveVector(Broadcasted,
-                                                             true);
-          WidenMap[it] = ConsecutiveInduction;
+          for (unsigned part = 0; part < UF; ++part)
+            Entry[part] = getConsecutiveVector(Broadcasted, -VF * part, true);
           continue;
         }
 
         // Handle the pointer induction variable case.
         assert(P->getType()->isPointerTy() && "Unexpected type.");
 
+        // Is this a reverse induction ptr or a consecutive induction ptr.
+        bool Reverse = (LoopVectorizationLegality::IK_ReversePtrInduction ==
+                        II.IK);
+
         // This is the vector of results. Notice that we don't generate
         // vector geps because scalar geps result in better code.
-        Value *VecVal = UndefValue::get(VectorType::get(P->getType(), VF));
-        for (unsigned int i = 0; i < VF; ++i) {
-          Constant *Idx = ConstantInt::get(Induction->getType(), i);
-          Value *GlobalIdx = Builder.CreateAdd(NormalizedIdx, Idx,
-                                               "gep.idx");
-          Value *SclrGep = Builder.CreateGEP(II.StartValue, GlobalIdx,
-                                             "next.gep");
-          VecVal = Builder.CreateInsertElement(VecVal, SclrGep,
-                                               Builder.getInt32(i),
-                                               "insert.gep");
+        for (unsigned part = 0; part < UF; ++part) {
+          Value *VecVal = UndefValue::get(VectorType::get(P->getType(), VF));
+          for (unsigned int i = 0; i < VF; ++i) {
+            int EltIndex = (i + part * VF) * (Reverse ? -1 : 1);
+            Constant *Idx = ConstantInt::get(Induction->getType(), EltIndex);
+            Value *GlobalIdx;
+            if (!Reverse)
+              GlobalIdx = Builder.CreateAdd(NormalizedIdx, Idx, "gep.idx");
+            else
+              GlobalIdx = Builder.CreateSub(Idx, NormalizedIdx, "gep.ridx");
+
+            Value *SclrGep = Builder.CreateGEP(II.StartValue, GlobalIdx,
+                                               "next.gep");
+            VecVal = Builder.CreateInsertElement(VecVal, SclrGep,
+                                                 Builder.getInt32(i),
+                                                 "insert.gep");
+          }
+          Entry[part] = VecVal;
         }
-
-        WidenMap[it] = VecVal;
         continue;
       }
 
@@ -1061,41 +1985,48 @@ InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
     case Instruction::Xor: {
       // Just widen binops.
       BinaryOperator *BinOp = dyn_cast<BinaryOperator>(it);
-      Value *A = getVectorValue(it->getOperand(0));
-      Value *B = getVectorValue(it->getOperand(1));
+      VectorParts &A = getVectorValue(it->getOperand(0));
+      VectorParts &B = getVectorValue(it->getOperand(1));
 
       // Use this vector value for all users of the original instruction.
-      Value *V = Builder.CreateBinOp(BinOp->getOpcode(), A, B);
-      WidenMap[it] = V;
-
-      // Update the NSW, NUW and Exact flags.
-      BinaryOperator *VecOp = cast<BinaryOperator>(V);
-      if (isa<OverflowingBinaryOperator>(BinOp)) {
-        VecOp->setHasNoSignedWrap(BinOp->hasNoSignedWrap());
-        VecOp->setHasNoUnsignedWrap(BinOp->hasNoUnsignedWrap());
+      for (unsigned Part = 0; Part < UF; ++Part) {
+        Value *V = Builder.CreateBinOp(BinOp->getOpcode(), A[Part], B[Part]);
+
+        // Update the NSW, NUW and Exact flags. Notice: V can be an Undef.
+        BinaryOperator *VecOp = dyn_cast<BinaryOperator>(V);
+        if (VecOp && isa<OverflowingBinaryOperator>(BinOp)) {
+          VecOp->setHasNoSignedWrap(BinOp->hasNoSignedWrap());
+          VecOp->setHasNoUnsignedWrap(BinOp->hasNoUnsignedWrap());
+        }
+        if (VecOp && isa<PossiblyExactOperator>(VecOp))
+          VecOp->setIsExact(BinOp->isExact());
+
+        Entry[Part] = V;
       }
-      if (isa<PossiblyExactOperator>(VecOp))
-        VecOp->setIsExact(BinOp->isExact());
       break;
     }
     case Instruction::Select: {
       // Widen selects.
       // If the selector is loop invariant we can create a select
       // instruction with a scalar condition. Otherwise, use vector-select.
-      Value *Cond = it->getOperand(0);
-      bool InvariantCond = SE->isLoopInvariant(SE->getSCEV(Cond), OrigLoop);
+      bool InvariantCond = SE->isLoopInvariant(SE->getSCEV(it->getOperand(0)),
+                                               OrigLoop);
 
       // The condition can be loop invariant  but still defined inside the
       // loop. This means that we can't just use the original 'cond' value.
       // We have to take the 'vectorized' value and pick the first lane.
       // Instcombine will make this a no-op.
-      Cond = getVectorValue(Cond);
-      if (InvariantCond)
-        Cond = Builder.CreateExtractElement(Cond, Builder.getInt32(0));
-
-      Value *Op0 = getVectorValue(it->getOperand(1));
-      Value *Op1 = getVectorValue(it->getOperand(2));
-      WidenMap[it] = Builder.CreateSelect(Cond, Op0, Op1);
+      VectorParts &Cond = getVectorValue(it->getOperand(0));
+      VectorParts &Op0  = getVectorValue(it->getOperand(1));
+      VectorParts &Op1  = getVectorValue(it->getOperand(2));
+      Value *ScalarCond = Builder.CreateExtractElement(Cond[0],
+                                                       Builder.getInt32(0));
+      for (unsigned Part = 0; Part < UF; ++Part) {
+        Entry[Part] = Builder.CreateSelect(
+          InvariantCond ? ScalarCond : Cond[Part],
+          Op0[Part],
+          Op1[Part]);
+      }
       break;
     }
 
@@ -1104,94 +2035,23 @@ InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
       // Widen compares. Generate vector compares.
       bool FCmp = (it->getOpcode() == Instruction::FCmp);
       CmpInst *Cmp = dyn_cast<CmpInst>(it);
-      Value *A = getVectorValue(it->getOperand(0));
-      Value *B = getVectorValue(it->getOperand(1));
-      if (FCmp)
-        WidenMap[it] = Builder.CreateFCmp(Cmp->getPredicate(), A, B);
-      else
-        WidenMap[it] = Builder.CreateICmp(Cmp->getPredicate(), A, B);
-      break;
-    }
-
-    case Instruction::Store: {
-      // Attempt to issue a wide store.
-      StoreInst *SI = dyn_cast<StoreInst>(it);
-      Type *StTy = VectorType::get(SI->getValueOperand()->getType(), VF);
-      Value *Ptr = SI->getPointerOperand();
-      unsigned Alignment = SI->getAlignment();
-
-      assert(!Legal->isUniform(Ptr) &&
-             "We do not allow storing to uniform addresses");
-
-      GetElementPtrInst *Gep = dyn_cast<GetElementPtrInst>(Ptr);
-
-      // This store does not use GEPs.
-      if (!Legal->isConsecutivePtr(Ptr)) {
-        scalarizeInstruction(it);
-        break;
+      VectorParts &A = getVectorValue(it->getOperand(0));
+      VectorParts &B = getVectorValue(it->getOperand(1));
+      for (unsigned Part = 0; Part < UF; ++Part) {
+        Value *C = 0;
+        if (FCmp)
+          C = Builder.CreateFCmp(Cmp->getPredicate(), A[Part], B[Part]);
+        else
+          C = Builder.CreateICmp(Cmp->getPredicate(), A[Part], B[Part]);
+        Entry[Part] = C;
       }
-
-      if (Gep) {
-        // The last index does not have to be the induction. It can be
-        // consecutive and be a function of the index. For example A[I+1];
-        unsigned NumOperands = Gep->getNumOperands();
-        Value *LastIndex = getVectorValue(Gep->getOperand(NumOperands - 1));
-        LastIndex = Builder.CreateExtractElement(LastIndex, Zero);
-
-        // Create the new GEP with the new induction variable.
-        GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone());
-        Gep2->setOperand(NumOperands - 1, LastIndex);
-        Ptr = Builder.Insert(Gep2);
-      } else {
-        // Use the induction element ptr.
-        assert(isa<PHINode>(Ptr) && "Invalid induction ptr");
-        Ptr = Builder.CreateExtractElement(getVectorValue(Ptr), Zero);
-      }
-      Ptr = Builder.CreateBitCast(Ptr, StTy->getPointerTo());
-      Value *Val = getVectorValue(SI->getValueOperand());
-      Builder.CreateStore(Val, Ptr)->setAlignment(Alignment);
       break;
     }
-    case Instruction::Load: {
-      // Attempt to issue a wide load.
-      LoadInst *LI = dyn_cast<LoadInst>(it);
-      Type *RetTy = VectorType::get(LI->getType(), VF);
-      Value *Ptr = LI->getPointerOperand();
-      unsigned Alignment = LI->getAlignment();
-      GetElementPtrInst *Gep = dyn_cast<GetElementPtrInst>(Ptr);
-
-      // If the pointer is loop invariant or if it is non consecutive,
-      // scalarize the load.
-      bool Con = Legal->isConsecutivePtr(Ptr);
-      if (Legal->isUniform(Ptr) || !Con) {
-        scalarizeInstruction(it);
-        break;
-      }
-
-      if (Gep) {
-        // The last index does not have to be the induction. It can be
-        // consecutive and be a function of the index. For example A[I+1];
-        unsigned NumOperands = Gep->getNumOperands();
-        Value *LastIndex = getVectorValue(Gep->getOperand(NumOperands -1));
-        LastIndex = Builder.CreateExtractElement(LastIndex, Zero);
-
-        // Create the new GEP with the new induction variable.
-        GetElementPtrInst *Gep2 = cast<GetElementPtrInst>(Gep->clone());
-        Gep2->setOperand(NumOperands - 1, LastIndex);
-        Ptr = Builder.Insert(Gep2);
-      } else {
-        // Use the induction element ptr.
-        assert(isa<PHINode>(Ptr) && "Invalid induction ptr");
-        Ptr = Builder.CreateExtractElement(getVectorValue(Ptr), Zero);
-      }
 
-      Ptr = Builder.CreateBitCast(Ptr, RetTy->getPointerTo());
-      LI = Builder.CreateLoad(Ptr);
-      LI->setAlignment(Alignment);
-      // Use this vector value for all users of the load.
-      WidenMap[it] = LI;
-      break;
-    }
+    case Instruction::Store:
+    case Instruction::Load:
+        vectorizeMemoryInstruction(it, Legal);
+        break;
     case Instruction::ZExt:
     case Instruction::SExt:
     case Instruction::FPToUI:
@@ -1204,25 +2064,48 @@ InnerLoopVectorizer::vectorizeBlockInLoop(LoopVectorizationLegality *Legal,
     case Instruction::Trunc:
     case Instruction::FPTrunc:
     case Instruction::BitCast: {
-      /// Vectorize bitcasts.
       CastInst *CI = dyn_cast<CastInst>(it);
-      Value *A = getVectorValue(it->getOperand(0));
+      /// Optimize the special case where the source is the induction
+      /// variable. Notice that we can only optimize the 'trunc' case
+      /// because: a. FP conversions lose precision, b. sext/zext may wrap,
+      /// c. other casts depend on pointer size.
+      if (CI->getOperand(0) == OldInduction &&
+          it->getOpcode() == Instruction::Trunc) {
+        Value *ScalarCast = Builder.CreateCast(CI->getOpcode(), Induction,
+                                               CI->getType());
+        Value *Broadcasted = getBroadcastInstrs(ScalarCast);
+        for (unsigned Part = 0; Part < UF; ++Part)
+          Entry[Part] = getConsecutiveVector(Broadcasted, VF * Part, false);
+        break;
+      }
+      /// Vectorize casts.
       Type *DestTy = VectorType::get(CI->getType()->getScalarType(), VF);
-      WidenMap[it] = Builder.CreateCast(CI->getOpcode(), A, DestTy);
+
+      VectorParts &A = getVectorValue(it->getOperand(0));
+      for (unsigned Part = 0; Part < UF; ++Part)
+        Entry[Part] = Builder.CreateCast(CI->getOpcode(), A[Part], DestTy);
       break;
     }
 
     case Instruction::Call: {
-      assert(isTriviallyVectorizableIntrinsic(it));
+      // Ignore dbg intrinsics.
+      if (isa<DbgInfoIntrinsic>(it))
+        break;
+
       Module *M = BB->getParent()->getParent();
-      IntrinsicInst *II = cast<IntrinsicInst>(it);
-      Intrinsic::ID ID = II->getIntrinsicID();
-      SmallVector<Value*, 4> Args;
-      for (unsigned i = 0, ie = II->getNumArgOperands(); i != ie; ++i)
-        Args.push_back(getVectorValue(II->getArgOperand(i)));
-      Type *Tys[] = { VectorType::get(II->getType()->getScalarType(), VF) };
-      Function *F = Intrinsic::getDeclaration(M, ID, Tys);
-      WidenMap[it] = Builder.CreateCall(F, Args);
+      CallInst *CI = cast<CallInst>(it);
+      Intrinsic::ID ID = getIntrinsicIDForCall(CI, TLI);
+      assert(ID && "Not an intrinsic call!");
+      for (unsigned Part = 0; Part < UF; ++Part) {
+        SmallVector<Value*, 4> Args;
+        for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i) {
+          VectorParts &Arg = getVectorValue(CI->getArgOperand(i));
+          Args.push_back(Arg[Part]);
+        }
+        Type *Tys[] = { VectorType::get(CI->getType()->getScalarType(), VF) };
+        Function *F = Intrinsic::getDeclaration(M, ID, Tys);
+        Entry[Part] = Builder.CreateCall(F, Args);
+      }
       break;
     }
 
@@ -1239,12 +2122,14 @@ void InnerLoopVectorizer::updateAnalysis() {
   SE->forgetLoop(OrigLoop);
 
   // Update the dominator tree information.
-  assert(DT->properlyDominates(LoopBypassBlock, LoopExitBlock) &&
+  assert(DT->properlyDominates(LoopBypassBlocks.front(), LoopExitBlock) &&
          "Entry does not dominate exit.");
 
-  DT->addNewBlock(LoopVectorPreHeader, LoopBypassBlock);
+  for (unsigned I = 1, E = LoopBypassBlocks.size(); I != E; ++I)
+    DT->addNewBlock(LoopBypassBlocks[I], LoopBypassBlocks[I-1]);
+  DT->addNewBlock(LoopVectorPreHeader, LoopBypassBlocks.back());
   DT->addNewBlock(LoopVectorBody, LoopVectorPreHeader);
-  DT->addNewBlock(LoopMiddleBlock, LoopBypassBlock);
+  DT->addNewBlock(LoopMiddleBlock, LoopBypassBlocks.front());
   DT->addNewBlock(LoopScalarPreHeader, LoopMiddleBlock);
   DT->changeImmediateDominator(LoopScalarBody, LoopScalarPreHeader);
   DT->changeImmediateDominator(LoopExitBlock, LoopMiddleBlock);
@@ -1263,6 +2148,10 @@ bool LoopVectorizationLegality::canVectorizeWithIfConvert() {
   for (unsigned i = 0, e = LoopBlocks.size(); i < e; ++i) {
     BasicBlock *BB = LoopBlocks[i];
 
+    // We don't support switch statements inside loops.
+    if (!isa<BranchInst>(BB->getTerminator()))
+      return false;
+
     // We must have at most two predecessors because we need to convert
     // all PHIs to selects.
     unsigned Preds = std::distance(pred_begin(BB), pred_end(BB));
@@ -1315,7 +2204,7 @@ bool LoopVectorizationLegality::canVectorize() {
 
   // Do not loop-vectorize loops with a tiny trip count.
   unsigned TC = SE->getSmallConstantTripCount(TheLoop, Latch);
-  if (TC > 0u && TC < TinyTripCountThreshold) {
+  if (TC > 0u && TC < TinyTripCountVectorThreshold) {
     DEBUG(dbgs() << "LV: Found a loop with a very small trip count. " <<
           "This loop is not worth vectorizing.\n");
     return false;
@@ -1350,6 +2239,13 @@ bool LoopVectorizationLegality::canVectorizeInstrs() {
   BasicBlock *PreHeader = TheLoop->getLoopPreheader();
   BasicBlock *Header = TheLoop->getHeader();
 
+  // If we marked the scalar loop as "already vectorized" then no need
+  // to vectorize it again.
+  if (Header->getTerminator()->getMetadata(AlreadyVectorizedMDName)) {
+    DEBUG(dbgs() << "LV: This loop was vectorized before\n");
+    return false;
+  }
+
   // For each block in the loop.
   for (Loop::block_iterator bb = TheLoop->block_begin(),
        be = TheLoop->block_end(); bb != be; ++bb) {
@@ -1367,6 +2263,7 @@ bool LoopVectorizationLegality::canVectorizeInstrs() {
 
         // Check that this PHI type is allowed.
         if (!Phi->getType()->isIntegerTy() &&
+            !Phi->getType()->isFloatingPointTy() &&
             !Phi->getType()->isPointerTy()) {
           DEBUG(dbgs() << "LV: Found an non-int non-pointer PHI.\n");
           return false;
@@ -1383,9 +2280,9 @@ bool LoopVectorizationLegality::canVectorizeInstrs() {
         // Check if this is an induction variable.
         InductionKind IK = isInductionVariable(Phi);
 
-        if (NoInduction != IK) {
+        if (IK_NoInduction != IK) {
           // Int inductions are special because we only allow one IV.
-          if (IK == IntInduction) {
+          if (IK == IK_IntInduction) {
             if (Induction) {
               DEBUG(dbgs() << "LV: Found too many inductions."<< *Phi <<"\n");
               return false;
@@ -1398,45 +2295,61 @@ bool LoopVectorizationLegality::canVectorizeInstrs() {
           continue;
         }
 
-        if (AddReductionVar(Phi, IntegerAdd)) {
+        if (AddReductionVar(Phi, RK_IntegerAdd)) {
           DEBUG(dbgs() << "LV: Found an ADD reduction PHI."<< *Phi <<"\n");
           continue;
         }
-        if (AddReductionVar(Phi, IntegerMult)) {
+        if (AddReductionVar(Phi, RK_IntegerMult)) {
           DEBUG(dbgs() << "LV: Found a MUL reduction PHI."<< *Phi <<"\n");
           continue;
         }
-        if (AddReductionVar(Phi, IntegerOr)) {
+        if (AddReductionVar(Phi, RK_IntegerOr)) {
           DEBUG(dbgs() << "LV: Found an OR reduction PHI."<< *Phi <<"\n");
           continue;
         }
-        if (AddReductionVar(Phi, IntegerAnd)) {
+        if (AddReductionVar(Phi, RK_IntegerAnd)) {
           DEBUG(dbgs() << "LV: Found an AND reduction PHI."<< *Phi <<"\n");
           continue;
         }
-        if (AddReductionVar(Phi, IntegerXor)) {
+        if (AddReductionVar(Phi, RK_IntegerXor)) {
           DEBUG(dbgs() << "LV: Found a XOR reduction PHI."<< *Phi <<"\n");
           continue;
         }
+        if (AddReductionVar(Phi, RK_FloatMult)) {
+          DEBUG(dbgs() << "LV: Found an FMult reduction PHI."<< *Phi <<"\n");
+          continue;
+        }
+        if (AddReductionVar(Phi, RK_FloatAdd)) {
+          DEBUG(dbgs() << "LV: Found an FAdd reduction PHI."<< *Phi <<"\n");
+          continue;
+        }
 
         DEBUG(dbgs() << "LV: Found an unidentified PHI."<< *Phi <<"\n");
         return false;
       }// end of PHI handling
 
-      // We still don't handle functions.
+      // We still don't handle functions. However, we can ignore dbg intrinsic
+      // calls and we do handle certain intrinsic and libm functions.
       CallInst *CI = dyn_cast<CallInst>(it);
-      if (CI && !isTriviallyVectorizableIntrinsic(it)) {
+      if (CI && !getIntrinsicIDForCall(CI, TLI) && !isa<DbgInfoIntrinsic>(CI)) {
         DEBUG(dbgs() << "LV: Found a call site.\n");
         return false;
       }
 
-      // We do not re-vectorize vectors.
+      // Check that the instruction return type is vectorizable.
       if (!VectorType::isValidElementType(it->getType()) &&
           !it->getType()->isVoidTy()) {
         DEBUG(dbgs() << "LV: Found unvectorizable type." << "\n");
         return false;
       }
 
+      // Check that the stored type is vectorizable.
+      if (StoreInst *ST = dyn_cast<StoreInst>(it)) {
+        Type *T = ST->getValueOperand()->getType();
+        if (!VectorType::isValidElementType(T))
+          return false;
+      }
+
       // Reduction instructions are allowed to have exit users.
       // All other instructions must not have external users.
       if (!AllowedExit.count(it))
@@ -1491,7 +2404,51 @@ void LoopVectorizationLegality::collectLoopUniforms() {
   }
 }
 
+AliasAnalysis::Location
+LoopVectorizationLegality::getLoadStoreLocation(Instruction *Inst) {
+  if (StoreInst *Store = dyn_cast<StoreInst>(Inst))
+    return AA->getLocation(Store);
+  else if (LoadInst *Load = dyn_cast<LoadInst>(Inst))
+    return AA->getLocation(Load);
+
+  llvm_unreachable("Should be either load or store instruction");
+}
+
+bool
+LoopVectorizationLegality::hasPossibleGlobalWriteReorder(
+                                                Value *Object,
+                                                Instruction *Inst,
+                                                AliasMultiMap& WriteObjects,
+                                                unsigned MaxByteWidth) {
+
+  AliasAnalysis::Location ThisLoc = getLoadStoreLocation(Inst);
+
+  std::vector<Instruction*>::iterator
+              it = WriteObjects[Object].begin(),
+              end = WriteObjects[Object].end();
+
+  for (; it != end; ++it) {
+    Instruction* I = *it;
+    if (I == Inst)
+      continue;
+
+    AliasAnalysis::Location ThatLoc = getLoadStoreLocation(I);
+    if (AA->alias(ThisLoc.getWithNewSize(MaxByteWidth),
+                  ThatLoc.getWithNewSize(MaxByteWidth)))
+      return true;
+  }
+  return false;
+}
+
 bool LoopVectorizationLegality::canVectorizeMemory() {
+
+  if (TheLoop->isAnnotatedParallel()) {
+    DEBUG(dbgs()
+          << "LV: A loop annotated parallel, ignore memory dependency "
+          << "checks.\n");
+    return true;
+  }
+
   typedef SmallVector<Value*, 16> ValueVector;
   typedef SmallPtrSet<Value*, 16> ValueSet;
   // Holds the Load and Store *instructions*.
@@ -1545,9 +2502,10 @@ bool LoopVectorizationLegality::canVectorizeMemory() {
     return true;
   }
 
-  // Holds the read and read-write *pointers* that we find.
-  ValueVector Reads;
-  ValueVector ReadWrites;
+  // Holds the read and read-write *pointers* that we find. These maps hold
+  // unique values for pointers (so no need for multi-map).
+  AliasMap Reads;
+  AliasMap ReadWrites;
 
   // Holds the analyzed pointers. We don't want to call GetUnderlyingObjects
   // multiple times on the same object. If the ptr is accessed twice, once
@@ -1558,8 +2516,7 @@ bool LoopVectorizationLegality::canVectorizeMemory() {
 
   ValueVector::iterator I, IE;
   for (I = Stores.begin(), IE = Stores.end(); I != IE; ++I) {
-    StoreInst *ST = dyn_cast<StoreInst>(*I);
-    assert(ST && "Bad StoreInst");
+    StoreInst *ST = cast<StoreInst>(*I);
     Value* Ptr = ST->getPointerOperand();
 
     if (isUniform(Ptr)) {
@@ -1570,12 +2527,11 @@ bool LoopVectorizationLegality::canVectorizeMemory() {
     // If we did *not* see this pointer before, insert it to
     // the read-write list. At this phase it is only a 'write' list.
     if (Seen.insert(Ptr))
-      ReadWrites.push_back(Ptr);
+      ReadWrites.insert(std::make_pair(Ptr, ST));
   }
 
   for (I = Loads.begin(), IE = Loads.end(); I != IE; ++I) {
-    LoadInst *LD = dyn_cast<LoadInst>(*I);
-    assert(LD && "Bad LoadInst");
+    LoadInst *LD = cast<LoadInst>(*I);
     Value* Ptr = LD->getPointerOperand();
     // If we did *not* see this pointer before, insert it to the
     // read list. If we *did* see it before, then it is already in
@@ -1585,8 +2541,8 @@ bool LoopVectorizationLegality::canVectorizeMemory() {
     // If the address of i is unknown (for example A[B[i]]) then we may
     // read a few words, modify, and write a few words, and some of the
     // words may be written to the same address.
-    if (Seen.insert(Ptr) || !isConsecutivePtr(Ptr))
-      Reads.push_back(Ptr);
+    if (Seen.insert(Ptr) || 0 == isConsecutivePtr(Ptr))
+      Reads.insert(std::make_pair(Ptr, LD));
   }
 
   // If we write (or read-write) to a single destination and there are no
@@ -1598,83 +2554,156 @@ bool LoopVectorizationLegality::canVectorizeMemory() {
 
   // Find pointers with computable bounds. We are going to use this information
   // to place a runtime bound check.
-  bool RT = true;
-  for (I = ReadWrites.begin(), IE = ReadWrites.end(); I != IE; ++I)
-    if (hasComputableBounds(*I)) {
-      PtrRtCheck.insert(SE, TheLoop, *I);
-      DEBUG(dbgs() << "LV: Found a runtime check ptr:" << **I <<"\n");
+  bool CanDoRT = true;
+  AliasMap::iterator MI, ME;
+  for (MI = ReadWrites.begin(), ME = ReadWrites.end(); MI != ME; ++MI) {
+    Value *V = (*MI).first;
+    if (hasComputableBounds(V)) {
+      PtrRtCheck.insert(SE, TheLoop, V);
+      DEBUG(dbgs() << "LV: Found a runtime check ptr:" << *V <<"\n");
     } else {
-      RT = false;
+      CanDoRT = false;
       break;
     }
-  for (I = Reads.begin(), IE = Reads.end(); I != IE; ++I)
-    if (hasComputableBounds(*I)) {
-      PtrRtCheck.insert(SE, TheLoop, *I);
-      DEBUG(dbgs() << "LV: Found a runtime check ptr:" << **I <<"\n");
+  }
+  for (MI = Reads.begin(), ME = Reads.end(); MI != ME; ++MI) {
+    Value *V = (*MI).first;
+    if (hasComputableBounds(V)) {
+      PtrRtCheck.insert(SE, TheLoop, V);
+      DEBUG(dbgs() << "LV: Found a runtime check ptr:" << *V <<"\n");
     } else {
-      RT = false;
+      CanDoRT = false;
       break;
     }
+  }
 
   // Check that we did not collect too many pointers or found a
   // unsizeable pointer.
-  if (!RT || PtrRtCheck.Pointers.size() > RuntimeMemoryCheckThreshold) {
+  if (!CanDoRT || PtrRtCheck.Pointers.size() > RuntimeMemoryCheckThreshold) {
     PtrRtCheck.reset();
-    RT = false;
+    CanDoRT = false;
   }
 
-  PtrRtCheck.Need = RT;
-
-  if (RT) {
+  if (CanDoRT) {
     DEBUG(dbgs() << "LV: We can perform a memory runtime check if needed.\n");
   }
 
+  bool NeedRTCheck = false;
+
+  // Biggest vectorized access possible, vector width * unroll factor.
+  // TODO: We're being very pessimistic here, find a way to know the
+  // real access width before getting here.
+  unsigned MaxByteWidth = (TTI->getRegisterBitWidth(true) / 8) *
+                           TTI->getMaximumUnrollFactor();
   // Now that the pointers are in two lists (Reads and ReadWrites), we
   // can check that there are no conflicts between each of the writes and
   // between the writes to the reads.
-  ValueSet WriteObjects;
+  // Note that WriteObjects duplicates the stores (indexed now by underlying
+  // objects) to avoid pointing to elements inside ReadWrites.
+  // TODO: Maybe create a new type where they can interact without duplication.
+  AliasMultiMap WriteObjects;
   ValueVector TempObjects;
 
   // Check that the read-writes do not conflict with other read-write
   // pointers.
-  for (I = ReadWrites.begin(), IE = ReadWrites.end(); I != IE; ++I) {
-    GetUnderlyingObjects(*I, TempObjects, DL);
-    for (ValueVector::iterator it=TempObjects.begin(), e=TempObjects.end();
-         it != e; ++it) {
-      if (!isIdentifiedObject(*it)) {
-        DEBUG(dbgs() << "LV: Found an unidentified write ptr:"<< **it <<"\n");
-        return RT;
+  bool AllWritesIdentified = true;
+  for (MI = ReadWrites.begin(), ME = ReadWrites.end(); MI != ME; ++MI) {
+    Value *Val = (*MI).first;
+    Instruction *Inst = (*MI).second;
+
+    GetUnderlyingObjects(Val, TempObjects, DL);
+    for (ValueVector::iterator UI=TempObjects.begin(), UE=TempObjects.end();
+         UI != UE; ++UI) {
+      if (!isIdentifiedObject(*UI)) {
+        DEBUG(dbgs() << "LV: Found an unidentified write ptr:"<< **UI <<"\n");
+        NeedRTCheck = true;
+        AllWritesIdentified = false;
       }
-      if (!WriteObjects.insert(*it)) {
+
+      // Never seen it before, can't alias.
+      if (WriteObjects[*UI].empty()) {
+        DEBUG(dbgs() << "LV: Adding Underlying value:" << **UI <<"\n");
+        WriteObjects[*UI].push_back(Inst);
+        continue;
+      }
+      // Direct alias found.
+      if (!AA || dyn_cast<GlobalValue>(*UI) == NULL) {
         DEBUG(dbgs() << "LV: Found a possible write-write reorder:"
-              << **it <<"\n");
-        return RT;
+              << **UI <<"\n");
+        return false;
       }
+      DEBUG(dbgs() << "LV: Found a conflicting global value:"
+            << **UI <<"\n");
+      DEBUG(dbgs() << "LV: While examining store:" << *Inst <<"\n");
+      DEBUG(dbgs() << "LV: On value:" << *Val <<"\n");
+
+      // If global alias, make sure they do alias.
+      if (hasPossibleGlobalWriteReorder(*UI,
+                                        Inst,
+                                        WriteObjects,
+                                        MaxByteWidth)) {
+        DEBUG(dbgs() << "LV: Found a possible write-write reorder:"
+              << *UI <<"\n");
+        return false;
+      }
+
+      // Didn't alias, insert into map for further reference.
+      WriteObjects[*UI].push_back(Inst);
     }
     TempObjects.clear();
   }
 
   /// Check that the reads don't conflict with the read-writes.
-  for (I = Reads.begin(), IE = Reads.end(); I != IE; ++I) {
-    GetUnderlyingObjects(*I, TempObjects, DL);
-    for (ValueVector::iterator it=TempObjects.begin(), e=TempObjects.end();
-         it != e; ++it) {
-      if (!isIdentifiedObject(*it)) {
-        DEBUG(dbgs() << "LV: Found an unidentified read ptr:"<< **it <<"\n");
-        return RT;
+  for (MI = Reads.begin(), ME = Reads.end(); MI != ME; ++MI) {
+    Value *Val = (*MI).first;
+    GetUnderlyingObjects(Val, TempObjects, DL);
+    for (ValueVector::iterator UI=TempObjects.begin(), UE=TempObjects.end();
+         UI != UE; ++UI) {
+      // If all of the writes are identified then we don't care if the read
+      // pointer is identified or not.
+      if (!AllWritesIdentified && !isIdentifiedObject(*UI)) {
+        DEBUG(dbgs() << "LV: Found an unidentified read ptr:"<< **UI <<"\n");
+        NeedRTCheck = true;
+      }
+
+      // Never seen it before, can't alias.
+      if (WriteObjects[*UI].empty())
+        continue;
+      // Direct alias found.
+      if (!AA || dyn_cast<GlobalValue>(*UI) == NULL) {
+        DEBUG(dbgs() << "LV: Found a possible write-write reorder:"
+              << **UI <<"\n");
+        return false;
       }
-      if (WriteObjects.count(*it)) {
-        DEBUG(dbgs() << "LV: Found a possible read/write reorder:"
-              << **it <<"\n");
-        return RT;
+      DEBUG(dbgs() << "LV: Found a global value:  "
+            << **UI <<"\n");
+      Instruction *Inst = (*MI).second;
+      DEBUG(dbgs() << "LV: While examining load:" << *Inst <<"\n");
+      DEBUG(dbgs() << "LV: On value:" << *Val <<"\n");
+
+      // If global alias, make sure they do alias.
+      if (hasPossibleGlobalWriteReorder(*UI,
+                                        Inst,
+                                        WriteObjects,
+                                        MaxByteWidth)) {
+        DEBUG(dbgs() << "LV: Found a possible read-write reorder:"
+              << *UI <<"\n");
+        return false;
       }
     }
     TempObjects.clear();
   }
 
-  // It is safe to vectorize and we don't need any runtime checks.
-  DEBUG(dbgs() << "LV: We don't need a runtime memory check.\n");
-  PtrRtCheck.reset();
+  PtrRtCheck.Need = NeedRTCheck;
+  if (NeedRTCheck && !CanDoRT) {
+    DEBUG(dbgs() << "LV: We can't vectorize because we can't find " <<
+          "the array bounds.\n");
+    PtrRtCheck.reset();
+    return false;
+  }
+
+  DEBUG(dbgs() << "LV: We "<< (NeedRTCheck ? "" : "don't") <<
+        " need a runtime memory check.\n");
   return true;
 }
 
@@ -1696,12 +2725,13 @@ bool LoopVectorizationLegality::AddReductionVar(PHINode *Phi,
   // This includes users of the reduction, variables (which form a cycle
   // which ends in the phi node).
   Instruction *ExitInstruction = 0;
+  // Indicates that we found a binary operation in our scan.
+  bool FoundBinOp = false;
 
   // Iter is our iterator. We start with the PHI node and scan for all of the
-  // users of this instruction. All users must be instructions which can be
+  // users of this instruction. All users must be instructions that can be
   // used as reduction variables (such as ADD). We may have a single
-  // out-of-block user. They cycle must end with the original PHI.
-  // Also, we can't have multiple block-local users.
+  // out-of-block user. The cycle must end with the original PHI.
   Instruction *Iter = Phi;
   while (true) {
     // If the instruction has no users then this is a broken
@@ -1709,15 +2739,17 @@ bool LoopVectorizationLegality::AddReductionVar(PHINode *Phi,
     if (Iter->use_empty())
       return false;
 
-    // Any reduction instr must be of one of the allowed kinds.
-    if (!isReductionInstr(Iter, Kind))
-      return false;
-
-    // Did we find a user inside this block ?
+    // Did we find a user inside this loop already ?
     bool FoundInBlockUser = false;
-    // Did we reach the initial PHI node ?
+    // Did we reach the initial PHI node already ?
     bool FoundStartPHI = false;
 
+    // Is this a bin op ?
+    FoundBinOp |= !isa<PHINode>(Iter);
+
+    // Remember the current instruction.
+    Instruction *OldIter = Iter;
+
     // For each of the *users* of iter.
     for (Value::use_iterator it = Iter->use_begin(), e = Iter->use_end();
          it != e; ++it) {
@@ -1740,58 +2772,82 @@ bool LoopVectorizationLegality::AddReductionVar(PHINode *Phi,
       // We allow in-loop PHINodes which are not the original reduction PHI
       // node. If this PHI is the only user of Iter (happens in IF w/ no ELSE
       // structure) then don't skip this PHI.
-      if (isa<PHINode>(U) && U->getParent() != TheLoop->getHeader() &&
-          TheLoop->contains(U) && Iter->getNumUses() > 1)
+      if (isa<PHINode>(Iter) && isa<PHINode>(U) &&
+          U->getParent() != TheLoop->getHeader() &&
+          TheLoop->contains(U) &&
+          Iter->hasNUsesOrMore(2))
         continue;
 
       // We can't have multiple inside users.
       if (FoundInBlockUser)
         return false;
       FoundInBlockUser = true;
+
+      // Any reduction instr must be of one of the allowed kinds.
+      if (!isReductionInstr(U, Kind))
+        return false;
+
+      // Reductions of instructions such as Div, and Sub is only
+      // possible if the LHS is the reduction variable.
+      if (!U->isCommutative() && !isa<PHINode>(U) && U->getOperand(0) != Iter)
+        return false;
+
       Iter = U;
     }
 
+    // If all uses were skipped this can't be a reduction variable.
+    if (Iter == OldIter)
+      return false;
+
     // We found a reduction var if we have reached the original
     // phi node and we only have a single instruction with out-of-loop
     // users.
-    if (FoundStartPHI && ExitInstruction) {
+    if (FoundStartPHI) {
       // This instruction is allowed to have out-of-loop users.
       AllowedExit.insert(ExitInstruction);
 
       // Save the description of this reduction variable.
       ReductionDescriptor RD(RdxStart, ExitInstruction, Kind);
       Reductions[Phi] = RD;
-      return true;
+      // We've ended the cycle. This is a reduction variable if we have an
+      // outside user and it has a binary op.
+      return FoundBinOp && ExitInstruction;
     }
-
-    // If we've reached the start PHI but did not find an outside user then
-    // this is dead code. Abort.
-    if (FoundStartPHI)
-      return false;
   }
 }
 
 bool
 LoopVectorizationLegality::isReductionInstr(Instruction *I,
                                             ReductionKind Kind) {
+  bool FP = I->getType()->isFloatingPointTy();
+  bool FastMath = (FP && I->isCommutative() && I->isAssociative());
+
   switch (I->getOpcode()) {
   default:
     return false;
   case Instruction::PHI:
+      if (FP && (Kind != RK_FloatMult && Kind != RK_FloatAdd))
+        return false;
     // possibly.
     return true;
-  case Instruction::Add:
   case Instruction::Sub:
-    return Kind == IntegerAdd;
+  case Instruction::Add:
+    return Kind == RK_IntegerAdd;
+  case Instruction::SDiv:
+  case Instruction::UDiv:
   case Instruction::Mul:
-    return Kind == IntegerMult;
+    return Kind == RK_IntegerMult;
   case Instruction::And:
-    return Kind == IntegerAnd;
+    return Kind == RK_IntegerAnd;
   case Instruction::Or:
-    return Kind == IntegerOr;
+    return Kind == RK_IntegerOr;
   case Instruction::Xor:
-    return Kind == IntegerXor;
-  }
+    return Kind == RK_IntegerXor;
+  case Instruction::FMul:
+    return Kind == RK_FloatMult && FastMath;
+  case Instruction::FAdd:
+    return Kind == RK_FloatAdd && FastMath;
+   }
 }
 
 LoopVectorizationLegality::InductionKind
@@ -1799,37 +2855,48 @@ LoopVectorizationLegality::isInductionVariable(PHINode *Phi) {
   Type *PhiTy = Phi->getType();
   // We only handle integer and pointer inductions variables.
   if (!PhiTy->isIntegerTy() && !PhiTy->isPointerTy())
-    return NoInduction;
+    return IK_NoInduction;
 
-  // Check that the PHI is consecutive and starts at zero.
+  // Check that the PHI is consecutive.
   const SCEV *PhiScev = SE->getSCEV(Phi);
   const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(PhiScev);
   if (!AR) {
     DEBUG(dbgs() << "LV: PHI is not a poly recurrence.\n");
-    return NoInduction;
+    return IK_NoInduction;
   }
   const SCEV *Step = AR->getStepRecurrence(*SE);
 
   // Integer inductions need to have a stride of one.
   if (PhiTy->isIntegerTy()) {
     if (Step->isOne())
-      return IntInduction;
+      return IK_IntInduction;
     if (Step->isAllOnesValue())
-      return ReverseIntInduction;
-    return NoInduction;
+      return IK_ReverseIntInduction;
+    return IK_NoInduction;
   }
 
   // Calculate the pointer stride and check if it is consecutive.
   const SCEVConstant *C = dyn_cast<SCEVConstant>(Step);
   if (!C)
-    return NoInduction;
+    return IK_NoInduction;
 
   assert(PhiTy->isPointerTy() && "The PHI must be a pointer");
   uint64_t Size = DL->getTypeAllocSize(PhiTy->getPointerElementType());
   if (C->getValue()->equalsInt(Size))
-    return PtrInduction;
+    return IK_PtrInduction;
+  else if (C->getValue()->equalsInt(0 - Size))
+    return IK_ReversePtrInduction;
+
+  return IK_NoInduction;
+}
+
+bool LoopVectorizationLegality::isInductionVariable(const Value *V) {
+  Value *In0 = const_cast<Value*>(V);
+  PHINode *PN = dyn_cast_or_null<PHINode>(In0);
+  if (!PN)
+    return false;
 
-  return NoInduction;
+  return Inductions.count(PN);
 }
 
 bool LoopVectorizationLegality::blockNeedsPredication(BasicBlock *BB)  {
@@ -1846,7 +2913,7 @@ bool LoopVectorizationLegality::blockCanBePredicated(BasicBlock *BB) {
     if (it->mayReadFromMemory() || it->mayWriteToMemory() || it->mayThrow())
       return false;
 
-    // The isntructions below can trap.
+    // The instructions below can trap.
     switch (it->getOpcode()) {
     default: continue;
     case Instruction::UDiv:
@@ -1869,11 +2936,64 @@ bool LoopVectorizationLegality::hasComputableBounds(Value *Ptr) {
   return AR->isAffine();
 }
 
-unsigned
-LoopVectorizationCostModel::findBestVectorizationFactor(unsigned VF) {
-  if (!VTTI) {
-    DEBUG(dbgs() << "LV: No vector target information. Not vectorizing. \n");
-    return 1;
+LoopVectorizationCostModel::VectorizationFactor
+LoopVectorizationCostModel::selectVectorizationFactor(bool OptForSize,
+                                                      unsigned UserVF) {
+  // Width 1 means no vectorize
+  VectorizationFactor Factor = { 1U, 0U };
+  if (OptForSize && Legal->getRuntimePointerCheck()->Need) {
+    DEBUG(dbgs() << "LV: Aborting. Runtime ptr check is required in Os.\n");
+    return Factor;
+  }
+
+  // Find the trip count.
+  unsigned TC = SE->getSmallConstantTripCount(TheLoop, TheLoop->getLoopLatch());
+  DEBUG(dbgs() << "LV: Found trip count:"<<TC<<"\n");
+
+  unsigned WidestType = getWidestType();
+  unsigned WidestRegister = TTI.getRegisterBitWidth(true);
+  unsigned MaxVectorSize = WidestRegister / WidestType;
+  DEBUG(dbgs() << "LV: The Widest type: " << WidestType << " bits.\n");
+  DEBUG(dbgs() << "LV: The Widest register is:" << WidestRegister << "bits.\n");
+
+  if (MaxVectorSize == 0) {
+    DEBUG(dbgs() << "LV: The target has no vector registers.\n");
+    MaxVectorSize = 1;
+  }
+
+  assert(MaxVectorSize <= 32 && "Did not expect to pack so many elements"
+         " into one vector!");
+
+  unsigned VF = MaxVectorSize;
+
+  // If we optimize the program for size, avoid creating the tail loop.
+  if (OptForSize) {
+    // If we are unable to calculate the trip count then don't try to vectorize.
+    if (TC < 2) {
+      DEBUG(dbgs() << "LV: Aborting. A tail loop is required in Os.\n");
+      return Factor;
+    }
+
+    // Find the maximum SIMD width that can fit within the trip count.
+    VF = TC % MaxVectorSize;
+
+    if (VF == 0)
+      VF = MaxVectorSize;
+
+    // If the trip count that we found modulo the vectorization factor is not
+    // zero then we require a tail.
+    if (VF < 2) {
+      DEBUG(dbgs() << "LV: Aborting. A tail loop is required in Os.\n");
+      return Factor;
+    }
+  }
+
+  if (UserVF != 0) {
+    assert(isPowerOf2_32(UserVF) && "VF needs to be a power of two");
+    DEBUG(dbgs() << "LV: Using user VF "<<UserVF<<".\n");
+
+    Factor.Width = UserVF;
+    return Factor;
   }
 
   float Cost = expectedCost(1);
@@ -1893,7 +3013,248 @@ LoopVectorizationCostModel::findBestVectorizationFactor(unsigned VF) {
   }
 
   DEBUG(dbgs() << "LV: Selecting VF = : "<< Width << ".\n");
-  return Width;
+  Factor.Width = Width;
+  Factor.Cost = Width * Cost;
+  return Factor;
+}
+
+unsigned LoopVectorizationCostModel::getWidestType() {
+  unsigned MaxWidth = 8;
+
+  // For each block.
+  for (Loop::block_iterator bb = TheLoop->block_begin(),
+       be = TheLoop->block_end(); bb != be; ++bb) {
+    BasicBlock *BB = *bb;
+
+    // For each instruction in the loop.
+    for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
+      Type *T = it->getType();
+
+      // Only examine Loads, Stores and PHINodes.
+      if (!isa<LoadInst>(it) && !isa<StoreInst>(it) && !isa<PHINode>(it))
+        continue;
+
+      // Examine PHI nodes that are reduction variables.
+      if (PHINode *PN = dyn_cast<PHINode>(it))
+        if (!Legal->getReductionVars()->count(PN))
+          continue;
+
+      // Examine the stored values.
+      if (StoreInst *ST = dyn_cast<StoreInst>(it))
+        T = ST->getValueOperand()->getType();
+
+      // Ignore loaded pointer types and stored pointer types that are not
+      // consecutive. However, we do want to take consecutive stores/loads of
+      // pointer vectors into account.
+      if (T->isPointerTy() && !isConsecutiveLoadOrStore(it))
+        continue;
+
+      MaxWidth = std::max(MaxWidth,
+                          (unsigned)DL->getTypeSizeInBits(T->getScalarType()));
+    }
+  }
+
+  return MaxWidth;
+}
+
+unsigned
+LoopVectorizationCostModel::selectUnrollFactor(bool OptForSize,
+                                               unsigned UserUF,
+                                               unsigned VF,
+                                               unsigned LoopCost) {
+
+  // -- The unroll heuristics --
+  // We unroll the loop in order to expose ILP and reduce the loop overhead.
+  // There are many micro-architectural considerations that we can't predict
+  // at this level. For example frontend pressure (on decode or fetch) due to
+  // code size, or the number and capabilities of the execution ports.
+  //
+  // We use the following heuristics to select the unroll factor:
+  // 1. If the code has reductions the we unroll in order to break the cross
+  // iteration dependency.
+  // 2. If the loop is really small then we unroll in order to reduce the loop
+  // overhead.
+  // 3. We don't unroll if we think that we will spill registers to memory due
+  // to the increased register pressure.
+
+  // Use the user preference, unless 'auto' is selected.
+  if (UserUF != 0)
+    return UserUF;
+
+  // When we optimize for size we don't unroll.
+  if (OptForSize)
+    return 1;
+
+  // Do not unroll loops with a relatively small trip count.
+  unsigned TC = SE->getSmallConstantTripCount(TheLoop,
+                                              TheLoop->getLoopLatch());
+  if (TC > 1 && TC < TinyTripCountUnrollThreshold)
+    return 1;
+
+  unsigned TargetVectorRegisters = TTI.getNumberOfRegisters(true);
+  DEBUG(dbgs() << "LV: The target has " << TargetVectorRegisters <<
+        " vector registers\n");
+
+  LoopVectorizationCostModel::RegisterUsage R = calculateRegisterUsage();
+  // We divide by these constants so assume that we have at least one
+  // instruction that uses at least one register.
+  R.MaxLocalUsers = std::max(R.MaxLocalUsers, 1U);
+  R.NumInstructions = std::max(R.NumInstructions, 1U);
+
+  // We calculate the unroll factor using the following formula.
+  // Subtract the number of loop invariants from the number of available
+  // registers. These registers are used by all of the unrolled instances.
+  // Next, divide the remaining registers by the number of registers that is
+  // required by the loop, in order to estimate how many parallel instances
+  // fit without causing spills.
+  unsigned UF = (TargetVectorRegisters - R.LoopInvariantRegs) / R.MaxLocalUsers;
+
+  // Clamp the unroll factor ranges to reasonable factors.
+  unsigned MaxUnrollSize = TTI.getMaximumUnrollFactor();
+
+  // If we did not calculate the cost for VF (because the user selected the VF)
+  // then we calculate the cost of VF here.
+  if (LoopCost == 0)
+    LoopCost = expectedCost(VF);
+
+  // Clamp the calculated UF to be between the 1 and the max unroll factor
+  // that the target allows.
+  if (UF > MaxUnrollSize)
+    UF = MaxUnrollSize;
+  else if (UF < 1)
+    UF = 1;
+
+  if (Legal->getReductionVars()->size()) {
+    DEBUG(dbgs() << "LV: Unrolling because of reductions. \n");
+    return UF;
+  }
+
+  // We want to unroll tiny loops in order to reduce the loop overhead.
+  // We assume that the cost overhead is 1 and we use the cost model
+  // to estimate the cost of the loop and unroll until the cost of the
+  // loop overhead is about 5% of the cost of the loop.
+  DEBUG(dbgs() << "LV: Loop cost is "<< LoopCost <<" \n");
+  if (LoopCost < 20) {
+    DEBUG(dbgs() << "LV: Unrolling to reduce branch cost. \n");
+    unsigned NewUF = 20/LoopCost + 1;
+    return std::min(NewUF, UF);
+  }
+
+  DEBUG(dbgs() << "LV: Not Unrolling. \n");
+  return 1;
+}
+
+LoopVectorizationCostModel::RegisterUsage
+LoopVectorizationCostModel::calculateRegisterUsage() {
+  // This function calculates the register usage by measuring the highest number
+  // of values that are alive at a single location. Obviously, this is a very
+  // rough estimation. We scan the loop in a topological order in order and
+  // assign a number to each instruction. We use RPO to ensure that defs are
+  // met before their users. We assume that each instruction that has in-loop
+  // users starts an interval. We record every time that an in-loop value is
+  // used, so we have a list of the first and last occurrences of each
+  // instruction. Next, we transpose this data structure into a multi map that
+  // holds the list of intervals that *end* at a specific location. This multi
+  // map allows us to perform a linear search. We scan the instructions linearly
+  // and record each time that a new interval starts, by placing it in a set.
+  // If we find this value in the multi-map then we remove it from the set.
+  // The max register usage is the maximum size of the set.
+  // We also search for instructions that are defined outside the loop, but are
+  // used inside the loop. We need this number separately from the max-interval
+  // usage number because when we unroll, loop-invariant values do not take
+  // more register.
+  LoopBlocksDFS DFS(TheLoop);
+  DFS.perform(LI);
+
+  RegisterUsage R;
+  R.NumInstructions = 0;
+
+  // Each 'key' in the map opens a new interval. The values
+  // of the map are the index of the 'last seen' usage of the
+  // instruction that is the key.
+  typedef DenseMap<Instruction*, unsigned> IntervalMap;
+  // Maps instruction to its index.
+  DenseMap<unsigned, Instruction*> IdxToInstr;
+  // Marks the end of each interval.
+  IntervalMap EndPoint;
+  // Saves the list of instruction indices that are used in the loop.
+  SmallSet<Instruction*, 8> Ends;
+  // Saves the list of values that are used in the loop but are
+  // defined outside the loop, such as arguments and constants.
+  SmallPtrSet<Value*, 8> LoopInvariants;
+
+  unsigned Index = 0;
+  for (LoopBlocksDFS::RPOIterator bb = DFS.beginRPO(),
+       be = DFS.endRPO(); bb != be; ++bb) {
+    R.NumInstructions += (*bb)->size();
+    for (BasicBlock::iterator it = (*bb)->begin(), e = (*bb)->end(); it != e;
+         ++it) {
+      Instruction *I = it;
+      IdxToInstr[Index++] = I;
+
+      // Save the end location of each USE.
+      for (unsigned i = 0; i < I->getNumOperands(); ++i) {
+        Value *U = I->getOperand(i);
+        Instruction *Instr = dyn_cast<Instruction>(U);
+
+        // Ignore non-instruction values such as arguments, constants, etc.
+        if (!Instr) continue;
+
+        // If this instruction is outside the loop then record it and continue.
+        if (!TheLoop->contains(Instr)) {
+          LoopInvariants.insert(Instr);
+          continue;
+        }
+
+        // Overwrite previous end points.
+        EndPoint[Instr] = Index;
+        Ends.insert(Instr);
+      }
+    }
+  }
+
+  // Saves the list of intervals that end with the index in 'key'.
+  typedef SmallVector<Instruction*, 2> InstrList;
+  DenseMap<unsigned, InstrList> TransposeEnds;
+
+  // Transpose the EndPoints to a list of values that end at each index.
+  for (IntervalMap::iterator it = EndPoint.begin(), e = EndPoint.end();
+       it != e; ++it)
+    TransposeEnds[it->second].push_back(it->first);
+
+  SmallSet<Instruction*, 8> OpenIntervals;
+  unsigned MaxUsage = 0;
+
+
+  DEBUG(dbgs() << "LV(REG): Calculating max register usage:\n");
+  for (unsigned int i = 0; i < Index; ++i) {
+    Instruction *I = IdxToInstr[i];
+    // Ignore instructions that are never used within the loop.
+    if (!Ends.count(I)) continue;
+
+    // Remove all of the instructions that end at this location.
+    InstrList &List = TransposeEnds[i];
+    for (unsigned int j=0, e = List.size(); j < e; ++j)
+      OpenIntervals.erase(List[j]);
+
+    // Count the number of live interals.
+    MaxUsage = std::max(MaxUsage, OpenIntervals.size());
+
+    DEBUG(dbgs() << "LV(REG): At #" << i << " Interval # " <<
+          OpenIntervals.size() <<"\n");
+
+    // Add the current instruction to the list of open intervals.
+    OpenIntervals.insert(I);
+  }
+
+  unsigned Invariant = LoopInvariants.size();
+  DEBUG(dbgs() << "LV(REG): Found max usage: " << MaxUsage << " \n");
+  DEBUG(dbgs() << "LV(REG): Found invariant usage: " << Invariant << " \n");
+  DEBUG(dbgs() << "LV(REG): LoopSize: " << R.NumInstructions << " \n");
+
+  R.LoopInvariantRegs = Invariant;
+  R.MaxLocalUsers = MaxUsage;
+  return R;
 }
 
 unsigned LoopVectorizationCostModel::expectedCost(unsigned VF) {
@@ -1907,6 +3268,10 @@ unsigned LoopVectorizationCostModel::expectedCost(unsigned VF) {
 
     // For each instruction in the old loop.
     for (BasicBlock::iterator it = BB->begin(), e = BB->end(); it != e; ++it) {
+      // Skip dbg intrinsics.
+      if (isa<DbgInfoIntrinsic>(it))
+        continue;
+
       unsigned C = getInstructionCost(it, VF);
       Cost += C;
       DEBUG(dbgs() << "LV: Found an estimated cost of "<< C <<" for VF " <<
@@ -1927,8 +3292,6 @@ unsigned LoopVectorizationCostModel::expectedCost(unsigned VF) {
 
 unsigned
 LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) {
-  assert(VTTI && "Invalid vector target transformation info");
-
   // If we know that this instruction will remain uniform, check the cost of
   // the scalar version.
   if (Legal->isUniformAfterVectorization(I))
@@ -1940,12 +3303,13 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) {
   // TODO: We need to estimate the cost of intrinsic calls.
   switch (I->getOpcode()) {
   case Instruction::GetElementPtr:
-    // We mark this instruction as zero-cost because scalar GEPs are usually
-    // lowered to the intruction addressing mode. At the moment we don't
-    // generate vector geps.
+    // We mark this instruction as zero-cost because the cost of GEPs in
+    // vectorized code depends on whether the corresponding memory instruction
+    // is scalarized or not. Therefore, we handle GEPs with the memory
+    // instruction cost.
     return 0;
   case Instruction::Br: {
-    return VTTI->getCFInstrCost(I->getOpcode());
+    return TTI.getCFInstrCost(I->getOpcode());
   }
   case Instruction::PHI:
     //TODO: IF-converted IFs become selects.
@@ -1968,7 +3332,7 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) {
   case Instruction::And:
   case Instruction::Or:
   case Instruction::Xor:
-    return VTTI->getArithmeticInstrCost(I->getOpcode(), VectorTy);
+    return TTI.getArithmeticInstrCost(I->getOpcode(), VectorTy);
   case Instruction::Select: {
     SelectInst *SI = cast<SelectInst>(I);
     const SCEV *CondSCEV = SE->getSCEV(SI->getCondition());
@@ -1977,68 +3341,66 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) {
     if (ScalarCond)
       CondTy = VectorType::get(CondTy, VF);
 
-    return VTTI->getCmpSelInstrCost(I->getOpcode(), VectorTy, CondTy);
+    return TTI.getCmpSelInstrCost(I->getOpcode(), VectorTy, CondTy);
   }
   case Instruction::ICmp:
   case Instruction::FCmp: {
     Type *ValTy = I->getOperand(0)->getType();
     VectorTy = ToVectorTy(ValTy, VF);
-    return VTTI->getCmpSelInstrCost(I->getOpcode(), VectorTy);
+    return TTI.getCmpSelInstrCost(I->getOpcode(), VectorTy);
   }
-  case Instruction::Store: {
-    StoreInst *SI = cast<StoreInst>(I);
-    Type *ValTy = SI->getValueOperand()->getType();
+  case Instruction::Store:
+  case Instruction::Load: {
+    StoreInst *SI = dyn_cast<StoreInst>(I);
+    LoadInst *LI = dyn_cast<LoadInst>(I);
+    Type *ValTy = (SI ? SI->getValueOperand()->getType() :
+                   LI->getType());
     VectorTy = ToVectorTy(ValTy, VF);
 
+    unsigned Alignment = SI ? SI->getAlignment() : LI->getAlignment();
+    unsigned AS = SI ? SI->getPointerAddressSpace() :
+      LI->getPointerAddressSpace();
+    Value *Ptr = SI ? SI->getPointerOperand() : LI->getPointerOperand();
+    // We add the cost of address computation here instead of with the gep
+    // instruction because only here we know whether the operation is
+    // scalarized.
     if (VF == 1)
-      return VTTI->getMemoryOpCost(I->getOpcode(), ValTy,
-                                   SI->getAlignment(),
-                                   SI->getPointerAddressSpace());
+      return TTI.getAddressComputationCost(VectorTy) +
+        TTI.getMemoryOpCost(I->getOpcode(), VectorTy, Alignment, AS);
 
-    // Scalarized stores.
-    if (!Legal->isConsecutivePtr(SI->getPointerOperand())) {
+    // Scalarized loads/stores.
+    int Stride = Legal->isConsecutivePtr(Ptr);
+    bool Reverse = Stride < 0;
+    if (0 == Stride) {
       unsigned Cost = 0;
-      unsigned ExtCost = VTTI->getInstrCost(Instruction::ExtractElement,
-                                            ValTy);
-      // The cost of extracting from the value vector.
-      Cost += VF * (ExtCost);
-      // The cost of the scalar stores.
-      Cost += VF * VTTI->getMemoryOpCost(I->getOpcode(),
-                                         ValTy->getScalarType(),
-                                         SI->getAlignment(),
-                                         SI->getPointerAddressSpace());
-      return Cost;
-    }
-
-    // Wide stores.
-    return VTTI->getMemoryOpCost(I->getOpcode(), VectorTy, SI->getAlignment(),
-                                 SI->getPointerAddressSpace());
-  }
-  case Instruction::Load: {
-    LoadInst *LI = cast<LoadInst>(I);
-
-    if (VF == 1)
-      return VTTI->getMemoryOpCost(I->getOpcode(), RetTy,
-                                   LI->getAlignment(),
-                                   LI->getPointerAddressSpace());
+      // The cost of extracting from the value vector and pointer vector.
+      Type *PtrTy = ToVectorTy(Ptr->getType(), VF);
+      for (unsigned i = 0; i < VF; ++i) {
+        //  The cost of extracting the pointer operand.
+        Cost += TTI.getVectorInstrCost(Instruction::ExtractElement, PtrTy, i);
+        // In case of STORE, the cost of ExtractElement from the vector.
+        // In case of LOAD, the cost of InsertElement into the returned
+        // vector.
+        Cost += TTI.getVectorInstrCost(SI ? Instruction::ExtractElement :
+                                            Instruction::InsertElement,
+                                            VectorTy, i);
+      }
 
-    // Scalarized loads.
-    if (!Legal->isConsecutivePtr(LI->getPointerOperand())) {
-      unsigned Cost = 0;
-      unsigned InCost = VTTI->getInstrCost(Instruction::InsertElement, RetTy);
-      // The cost of inserting the loaded value into the result vector.
-      Cost += VF * (InCost);
-      // The cost of the scalar stores.
-      Cost += VF * VTTI->getMemoryOpCost(I->getOpcode(),
-                                         RetTy->getScalarType(),
-                                         LI->getAlignment(),
-                                         LI->getPointerAddressSpace());
+      // The cost of the scalar loads/stores.
+      Cost += VF * TTI.getAddressComputationCost(ValTy->getScalarType());
+      Cost += VF * TTI.getMemoryOpCost(I->getOpcode(), ValTy->getScalarType(),
+                                       Alignment, AS);
       return Cost;
     }
 
-    // Wide loads.
-    return VTTI->getMemoryOpCost(I->getOpcode(), VectorTy, LI->getAlignment(),
-                                 LI->getPointerAddressSpace());
+    // Wide load/stores.
+    unsigned Cost = TTI.getAddressComputationCost(VectorTy);
+    Cost += TTI.getMemoryOpCost(I->getOpcode(), VectorTy, Alignment, AS);
+
+    if (Reverse)
+      Cost += TTI.getShuffleCost(TargetTransformInfo::SK_Reverse,
+                                  VectorTy, 0);
+    return Cost;
   }
   case Instruction::ZExt:
   case Instruction::SExt:
@@ -2052,17 +3414,25 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) {
   case Instruction::Trunc:
   case Instruction::FPTrunc:
   case Instruction::BitCast: {
+    // We optimize the truncation of induction variable.
+    // The cost of these is the same as the scalar operation.
+    if (I->getOpcode() == Instruction::Trunc &&
+        Legal->isInductionVariable(I->getOperand(0)))
+      return TTI.getCastInstrCost(I->getOpcode(), I->getType(),
+                                  I->getOperand(0)->getType());
+
     Type *SrcVecTy = ToVectorTy(I->getOperand(0)->getType(), VF);
-    return VTTI->getCastInstrCost(I->getOpcode(), VectorTy, SrcVecTy);
+    return TTI.getCastInstrCost(I->getOpcode(), VectorTy, SrcVecTy);
   }
   case Instruction::Call: {
-    assert(isTriviallyVectorizableIntrinsic(I));
-    IntrinsicInst *II = cast<IntrinsicInst>(I);
-    Type *RetTy = ToVectorTy(II->getType(), VF);
+    CallInst *CI = cast<CallInst>(I);
+    Intrinsic::ID ID = getIntrinsicIDForCall(CI, TLI);
+    assert(ID && "Not an intrinsic call!");
+    Type *RetTy = ToVectorTy(CI->getType(), VF);
     SmallVector<Type*, 4> Tys;
-    for (unsigned i = 0, ie = II->getNumArgOperands(); i != ie; ++i)
-      Tys.push_back(ToVectorTy(II->getArgOperand(i)->getType(), VF));
-    return VTTI->getIntrinsicInstrCost(II->getIntrinsicID(), RetTy, Tys);
+    for (unsigned i = 0, ie = CI->getNumArgOperands(); i != ie; ++i)
+      Tys.push_back(ToVectorTy(CI->getArgOperand(i)->getType(), VF));
+    return TTI.getIntrinsicInstrCost(ID, RetTy, Tys);
   }
   default: {
     // We are scalarizing the instruction. Return the cost of the scalar
@@ -2070,21 +3440,20 @@ LoopVectorizationCostModel::getInstructionCost(Instruction *I, unsigned VF) {
     // elements, times the vector width.
     unsigned Cost = 0;
 
-    bool IsVoid = RetTy->isVoidTy();
-
-    unsigned InsCost = (IsVoid ? 0 :
-                        VTTI->getInstrCost(Instruction::InsertElement,
-                                           VectorTy));
+    if (!RetTy->isVoidTy() && VF != 1) {
+      unsigned InsCost = TTI.getVectorInstrCost(Instruction::InsertElement,
+                                                VectorTy);
+      unsigned ExtCost = TTI.getVectorInstrCost(Instruction::ExtractElement,
+                                                VectorTy);
 
-    unsigned ExtCost = VTTI->getInstrCost(Instruction::ExtractElement,
-                                          VectorTy);
-
-    // The cost of inserting the results plus extracting each one of the
-    // operands.
-    Cost += VF * (InsCost + ExtCost * I->getNumOperands());
+      // The cost of inserting the results plus extracting each one of the
+      // operands.
+      Cost += VF * (InsCost + ExtCost * I->getNumOperands());
+    }
 
-    // The cost of executing VF copies of the scalar instruction.
-    Cost += VF * VTTI->getInstrCost(I->getOpcode(), RetTy);
+    // The cost of executing VF copies of the scalar instruction. This opcode
+    // is unknown. Assume that it is the same as 'mul'.
+    Cost += VF * TTI.getArithmeticInstrCost(Instruction::Mul, VectorTy);
     return Cost;
   }
   }// end of switch.
@@ -2100,6 +3469,7 @@ char LoopVectorize::ID = 0;
 static const char lv_name[] = "Loop Vectorization";
 INITIALIZE_PASS_BEGIN(LoopVectorize, LV_NAME, lv_name, false, false)
 INITIALIZE_AG_DEPENDENCY(AliasAnalysis)
+INITIALIZE_AG_DEPENDENCY(TargetTransformInfo)
 INITIALIZE_PASS_DEPENDENCY(ScalarEvolution)
 INITIALIZE_PASS_DEPENDENCY(LoopSimplify)
 INITIALIZE_PASS_END(LoopVectorize, LV_NAME, lv_name, false, false)
@@ -2110,4 +3480,14 @@ namespace llvm {
   }
 }
 
+bool LoopVectorizationCostModel::isConsecutiveLoadOrStore(Instruction *Inst) {
+  // Check for a store.
+  if (StoreInst *ST = dyn_cast<StoreInst>(Inst))
+    return Legal->isConsecutivePtr(ST->getPointerOperand()) != 0;
+
+  // Check for a load.
+  if (LoadInst *LI = dyn_cast<LoadInst>(Inst))
+    return Legal->isConsecutivePtr(LI->getPointerOperand()) != 0;
 
+  return false;
+}
diff --git a/lib/Transforms/Vectorize/LoopVectorize.h b/lib/Transforms/Vectorize/LoopVectorize.h
deleted file mode 100644
index 9d6d80e22b..0000000000
--- a/lib/Transforms/Vectorize/LoopVectorize.h
+++ /dev/null
@@ -1,458 +0,0 @@
-//===- LoopVectorize.h --- A Loop Vectorizer ------------------------------===//
-//
-//                     The LLVM Compiler Infrastructure
-//
-// This file is distributed under the University of Illinois Open Source
-// License. See LICENSE.TXT for details.
-//
-//===----------------------------------------------------------------------===//
-//
-// This is the LLVM loop vectorizer. This pass modifies 'vectorizable' loops
-// and generates target-independent LLVM-IR. Legalization of the IR is done
-// in the codegen. However, the vectorizes uses (will use) the codegen
-// interfaces to generate IR that is likely to result in an optimal binary.
-//
-// The loop vectorizer combines consecutive loop iteration into a single
-// 'wide' iteration. After this transformation the index is incremented
-// by the SIMD vector width, and not by one.
-//
-// This pass has three parts:
-// 1. The main loop pass that drives the different parts.
-// 2. LoopVectorizationLegality - A unit that checks for the legality
-//    of the vectorization.
-// 3. InnerLoopVectorizer - A unit that performs the actual
-//    widening of instructions.
-// 4. LoopVectorizationCostModel - A unit that checks for the profitability
-//    of vectorization. It decides on the optimal vector width, which
-//    can be one, if vectorization is not profitable.
-//
-//===----------------------------------------------------------------------===//
-//
-// The reduction-variable vectorization is based on the paper:
-//  D. Nuzman and R. Henderson. Multi-platform Auto-vectorization.
-//
-// Variable uniformity checks are inspired by:
-// Karrenberg, R. and Hack, S. Whole Function Vectorization.
-//
-// Other ideas/concepts are from:
-//  A. Zaks and D. Nuzman. Autovectorization in GCC-two years later.
-//
-//  S. Maleki, Y. Gao, M. Garzaran, T. Wong and D. Padua.  An Evaluation of
-//  Vectorizing Compilers.
-//
-//===----------------------------------------------------------------------===//
-#ifndef LLVM_TRANSFORM_VECTORIZE_LOOP_VECTORIZE_H
-#define LLVM_TRANSFORM_VECTORIZE_LOOP_VECTORIZE_H
-
-#define LV_NAME "loop-vectorize"
-#define DEBUG_TYPE LV_NAME
-
-#include "llvm/Analysis/ScalarEvolution.h"
-#include "llvm/ADT/SmallVector.h"
-#include "llvm/ADT/DenseMap.h"
-#include "llvm/ADT/SmallPtrSet.h"
-#include "llvm/IRBuilder.h" 
-
-#include <algorithm>
-using namespace llvm;
-
-/// We don't vectorize loops with a known constant trip count below this number.
-const unsigned TinyTripCountThreshold = 16;
-
-/// When performing a runtime memory check, do not check more than this
-/// number of pointers. Notice that the check is quadratic!
-const unsigned RuntimeMemoryCheckThreshold = 4;
-
-/// This is the highest vector width that we try to generate.
-const unsigned MaxVectorSize = 8;
-
-namespace llvm {
-
-// Forward declarations.
-class LoopVectorizationLegality;
-class LoopVectorizationCostModel;
-class VectorTargetTransformInfo;
-
-/// InnerLoopVectorizer vectorizes loops which contain only one basic
-/// block to a specified vectorization factor (VF).
-/// This class performs the widening of scalars into vectors, or multiple
-/// scalars. This class also implements the following features:
-/// * It inserts an epilogue loop for handling loops that don't have iteration
-///   counts that are known to be a multiple of the vectorization factor.
-/// * It handles the code generation for reduction variables.
-/// * Scalarization (implementation using scalars) of un-vectorizable
-///   instructions.
-/// InnerLoopVectorizer does not perform any vectorization-legality
-/// checks, and relies on the caller to check for the different legality
-/// aspects. The InnerLoopVectorizer relies on the
-/// LoopVectorizationLegality class to provide information about the induction
-/// and reduction variables that were found to a given vectorization factor.
-class InnerLoopVectorizer {
-public:
-  /// Ctor.
-  InnerLoopVectorizer(Loop *Orig, ScalarEvolution *Se, LoopInfo *Li,
-                      DominatorTree *Dt, DataLayout *Dl, unsigned VecWidth):
-  OrigLoop(Orig), SE(Se), LI(Li), DT(Dt), DL(Dl), VF(VecWidth),
-  Builder(Se->getContext()), Induction(0), OldInduction(0) { }
-
-  // Perform the actual loop widening (vectorization).
-  void vectorize(LoopVectorizationLegality *Legal) {
-    // Create a new empty loop. Unlink the old loop and connect the new one.
-    createEmptyLoop(Legal);
-    // Widen each instruction in the old loop to a new one in the new loop.
-    // Use the Legality module to find the induction and reduction variables.
-    vectorizeLoop(Legal);
-    // Register the new loop and update the analysis passes.
-    updateAnalysis();
-  }
-
-private:
-  /// A small list of PHINodes.
-  typedef SmallVector<PHINode*, 4> PhiVector;
-
-  /// Add code that checks at runtime if the accessed arrays overlap.
-  /// Returns the comparator value or NULL if no check is needed.
-  Value *addRuntimeCheck(LoopVectorizationLegality *Legal,
-                         Instruction *Loc);
-  /// Create an empty loop, based on the loop ranges of the old loop.
-  void createEmptyLoop(LoopVectorizationLegality *Legal);
-  /// Copy and widen the instructions from the old loop.
-  void vectorizeLoop(LoopVectorizationLegality *Legal);
-
-  /// A helper function that computes the predicate of the block BB, assuming
-  /// that the header block of the loop is set to True. It returns the *entry*
-  /// mask for the block BB.
-  Value *createBlockInMask(BasicBlock *BB);
-  /// A helper function that computes the predicate of the edge between SRC
-  /// and DST.
-  Value *createEdgeMask(BasicBlock *Src, BasicBlock *Dst);
-
-  /// A helper function to vectorize a single BB within the innermost loop.
-  void vectorizeBlockInLoop(LoopVectorizationLegality *Legal, BasicBlock *BB,
-                            PhiVector *PV);
-
-  /// Insert the new loop to the loop hierarchy and pass manager
-  /// and update the analysis passes.
-  void updateAnalysis();
-
-  /// This instruction is un-vectorizable. Implement it as a sequence
-  /// of scalars.
-  void scalarizeInstruction(Instruction *Instr);
-
-  /// Create a broadcast instruction. This method generates a broadcast
-  /// instruction (shuffle) for loop invariant values and for the induction
-  /// value. If this is the induction variable then we extend it to N, N+1, ...
-  /// this is needed because each iteration in the loop corresponds to a SIMD
-  /// element.
-  Value *getBroadcastInstrs(Value *V);
-
-  /// This function adds 0, 1, 2 ... to each vector element, starting at zero.
-  /// If Negate is set then negative numbers are added e.g. (0, -1, -2, ...).
-  Value *getConsecutiveVector(Value* Val, bool Negate = false);
-
-  /// When we go over instructions in the basic block we rely on previous
-  /// values within the current basic block or on loop invariant values.
-  /// When we widen (vectorize) values we place them in the map. If the values
-  /// are not within the map, they have to be loop invariant, so we simply
-  /// broadcast them into a vector.
-  Value *getVectorValue(Value *V);
-
-  /// Get a uniform vector of constant integers. We use this to get
-  /// vectors of ones and zeros for the reduction code.
-  Constant* getUniformVector(unsigned Val, Type* ScalarTy);
-
-  typedef DenseMap<Value*, Value*> ValueMap;
-
-  /// The original loop.
-  Loop *OrigLoop;
-  // Scev analysis to use.
-  ScalarEvolution *SE;
-  // Loop Info.
-  LoopInfo *LI;
-  // Dominator Tree.
-  DominatorTree *DT;
-  // Data Layout.
-  DataLayout *DL;
-  // The vectorization factor to use.
-  unsigned VF;
-
-  // The builder that we use
-  IRBuilder<> Builder;
-
-  // --- Vectorization state ---
-
-  /// The vector-loop preheader.
-  BasicBlock *LoopVectorPreHeader;
-  /// The scalar-loop preheader.
-  BasicBlock *LoopScalarPreHeader;
-  /// Middle Block between the vector and the scalar.
-  BasicBlock *LoopMiddleBlock;
-  ///The ExitBlock of the scalar loop.
-  BasicBlock *LoopExitBlock;
-  ///The vector loop body.
-  BasicBlock *LoopVectorBody;
-  ///The scalar loop body.
-  BasicBlock *LoopScalarBody;
-  ///The first bypass block.
-  BasicBlock *LoopBypassBlock;
-
-  /// The new Induction variable which was added to the new block.
-  PHINode *Induction;
-  /// The induction variable of the old basic block.
-  PHINode *OldInduction;
-  // Maps scalars to widened vectors.
-  ValueMap WidenMap;
-};
-
-/// LoopVectorizationLegality checks if it is legal to vectorize a loop, and
-/// to what vectorization factor.
-/// This class does not look at the profitability of vectorization, only the
-/// legality. This class has two main kinds of checks:
-/// * Memory checks - The code in canVectorizeMemory checks if vectorization
-///   will change the order of memory accesses in a way that will change the
-///   correctness of the program.
-/// * Scalars checks - The code in canVectorizeInstrs and canVectorizeMemory
-/// checks for a number of different conditions, such as the availability of a
-/// single induction variable, that all types are supported and vectorize-able,
-/// etc. This code reflects the capabilities of InnerLoopVectorizer.
-/// This class is also used by InnerLoopVectorizer for identifying
-/// induction variable and the different reduction variables.
-class LoopVectorizationLegality {
-public:
-  LoopVectorizationLegality(Loop *Lp, ScalarEvolution *Se, DataLayout *Dl,
-                            DominatorTree *Dt):
-  TheLoop(Lp), SE(Se), DL(Dl), DT(Dt), Induction(0) { }
-
-  /// This enum represents the kinds of reductions that we support.
-  enum ReductionKind {
-    NoReduction, /// Not a reduction.
-    IntegerAdd,  /// Sum of numbers.
-    IntegerMult, /// Product of numbers.
-    IntegerOr,   /// Bitwise or logical OR of numbers.
-    IntegerAnd,  /// Bitwise or logical AND of numbers.
-    IntegerXor   /// Bitwise or logical XOR of numbers.
-  };
-
-  /// This enum represents the kinds of inductions that we support.
-  enum InductionKind {
-    NoInduction,         /// Not an induction variable.
-    IntInduction,        /// Integer induction variable. Step = 1.
-    ReverseIntInduction, /// Reverse int induction variable. Step = -1.
-    PtrInduction         /// Pointer induction variable. Step = sizeof(elem).
-  };
-
-  /// This POD struct holds information about reduction variables.
-  struct ReductionDescriptor {
-    // Default C'tor
-    ReductionDescriptor():
-    StartValue(0), LoopExitInstr(0), Kind(NoReduction) {}
-
-    // C'tor.
-    ReductionDescriptor(Value *Start, Instruction *Exit, ReductionKind K):
-    StartValue(Start), LoopExitInstr(Exit), Kind(K) {}
-
-    // The starting value of the reduction.
-    // It does not have to be zero!
-    Value *StartValue;
-    // The instruction who's value is used outside the loop.
-    Instruction *LoopExitInstr;
-    // The kind of the reduction.
-    ReductionKind Kind;
-  };
-
-  // This POD struct holds information about the memory runtime legality
-  // check that a group of pointers do not overlap.
-  struct RuntimePointerCheck {
-    RuntimePointerCheck(): Need(false) {}
-
-    /// Reset the state of the pointer runtime information.
-    void reset() {
-      Need = false;
-      Pointers.clear();
-      Starts.clear();
-      Ends.clear();
-    }
-
-    /// Insert a pointer and calculate the start and end SCEVs.
-    void insert(ScalarEvolution *SE, Loop *Lp, Value *Ptr);
-
-    /// This flag indicates if we need to add the runtime check.
-    bool Need;
-    /// Holds the pointers that we need to check.
-    SmallVector<Value*, 2> Pointers;
-    /// Holds the pointer value at the beginning of the loop.
-    SmallVector<const SCEV*, 2> Starts;
-    /// Holds the pointer value at the end of the loop.
-    SmallVector<const SCEV*, 2> Ends;
-  };
-
-  /// A POD for saving information about induction variables.
-  struct InductionInfo {
-    /// Ctors.
-    InductionInfo(Value *Start, InductionKind K):
-    StartValue(Start), IK(K) {};
-    InductionInfo(): StartValue(0), IK(NoInduction) {};
-    /// Start value.
-    Value *StartValue;
-    /// Induction kind.
-    InductionKind IK;
-  };
-
-  /// ReductionList contains the reduction descriptors for all
-  /// of the reductions that were found in the loop.
-  typedef DenseMap<PHINode*, ReductionDescriptor> ReductionList;
-
-  /// InductionList saves induction variables and maps them to the
-  /// induction descriptor.
-  typedef DenseMap<PHINode*, InductionInfo> InductionList;
-
-  /// Returns true if it is legal to vectorize this loop.
-  /// This does not mean that it is profitable to vectorize this
-  /// loop, only that it is legal to do so.
-  bool canVectorize();
-
-  /// Returns the Induction variable.
-  PHINode *getInduction() {return Induction;}
-
-  /// Returns the reduction variables found in the loop.
-  ReductionList *getReductionVars() { return &Reductions; }
-
-  /// Returns the induction variables found in the loop.
-  InductionList *getInductionVars() { return &Inductions; }
-
-  /// Return true if the block BB needs to be predicated in order for the loop
-  /// to be vectorized.
-  bool blockNeedsPredication(BasicBlock *BB);
-
-  /// Check if this  pointer is consecutive when vectorizing. This happens
-  /// when the last index of the GEP is the induction variable, or that the
-  /// pointer itself is an induction variable.
-  /// This check allows us to vectorize A[idx] into a wide load/store.
-  bool isConsecutivePtr(Value *Ptr);
-
-  /// Returns true if the value V is uniform within the loop.
-  bool isUniform(Value *V);
-
-  /// Returns true if this instruction will remain scalar after vectorization.
-  bool isUniformAfterVectorization(Instruction* I) {return Uniforms.count(I);}
-
-  /// Returns the information that we collected about runtime memory check.
-  RuntimePointerCheck *getRuntimePointerCheck() {return &PtrRtCheck; }
-private:
-  /// Check if a single basic block loop is vectorizable.
-  /// At this point we know that this is a loop with a constant trip count
-  /// and we only need to check individual instructions.
-  bool canVectorizeInstrs();
-
-  /// When we vectorize loops we may change the order in which
-  /// we read and write from memory. This method checks if it is
-  /// legal to vectorize the code, considering only memory constrains.
-  /// Returns true if the loop is vectorizable
-  bool canVectorizeMemory();
-
-  /// Return true if we can vectorize this loop using the IF-conversion
-  /// transformation.
-  bool canVectorizeWithIfConvert();
-
-  /// Collect the variables that need to stay uniform after vectorization.
-  void collectLoopUniforms();
-
-  /// Return true if all of the instructions in the block can be speculatively
-  /// executed.
-  bool blockCanBePredicated(BasicBlock *BB);
-
-  /// Returns True, if 'Phi' is the kind of reduction variable for type
-  /// 'Kind'. If this is a reduction variable, it adds it to ReductionList.
-  bool AddReductionVar(PHINode *Phi, ReductionKind Kind);
-  /// Returns true if the instruction I can be a reduction variable of type
-  /// 'Kind'.
-  bool isReductionInstr(Instruction *I, ReductionKind Kind);
-  /// Returns the induction kind of Phi. This function may return NoInduction
-  /// if the PHI is not an induction variable.
-  InductionKind isInductionVariable(PHINode *Phi);
-  /// Return true if can compute the address bounds of Ptr within the loop.
-  bool hasComputableBounds(Value *Ptr);
-
-  /// The loop that we evaluate.
-  Loop *TheLoop;
-  /// Scev analysis.
-  ScalarEvolution *SE;
-  /// DataLayout analysis.
-  DataLayout *DL;
-  // Dominators.
-  DominatorTree *DT;
-
-  //  ---  vectorization state --- //
-
-  /// Holds the integer induction variable. This is the counter of the
-  /// loop.
-  PHINode *Induction;
-  /// Holds the reduction variables.
-  ReductionList Reductions;
-  /// Holds all of the induction variables that we found in the loop.
-  /// Notice that inductions don't need to start at zero and that induction
-  /// variables can be pointers.
-  InductionList Inductions;
-
-  /// Allowed outside users. This holds the reduction
-  /// vars which can be accessed from outside the loop.
-  SmallPtrSet<Value*, 4> AllowedExit;
-  /// This set holds the variables which are known to be uniform after
-  /// vectorization.
-  SmallPtrSet<Instruction*, 4> Uniforms;
-  /// We need to check that all of the pointers in this list are disjoint
-  /// at runtime.
-  RuntimePointerCheck PtrRtCheck;
-};
-
-/// LoopVectorizationCostModel - estimates the expected speedups due to
-/// vectorization.
-/// In many cases vectorization is not profitable. This can happen because
-/// of a number of reasons. In this class we mainly attempt to predict
-/// the expected speedup/slowdowns due to the supported instruction set.
-/// We use the VectorTargetTransformInfo to query the different backends
-/// for the cost of different operations.
-class LoopVectorizationCostModel {
-public:
-  /// C'tor.
-  LoopVectorizationCostModel(Loop *Lp, ScalarEvolution *Se,
-                             LoopVectorizationLegality *Leg,
-                             const VectorTargetTransformInfo *Vtti):
-  TheLoop(Lp), SE(Se), Legal(Leg), VTTI(Vtti) { }
-
-  /// Returns the most profitable vectorization factor for the loop that is
-  /// smaller or equal to the VF argument. This method checks every power
-  /// of two up to VF.
-  unsigned findBestVectorizationFactor(unsigned VF = MaxVectorSize);
-
-private:
-  /// Returns the expected execution cost. The unit of the cost does
-  /// not matter because we use the 'cost' units to compare different
-  /// vector widths. The cost that is returned is *not* normalized by
-  /// the factor width.
-  unsigned expectedCost(unsigned VF);
-
-  /// Returns the execution time cost of an instruction for a given vector
-  /// width. Vector width of one means scalar.
-  unsigned getInstructionCost(Instruction *I, unsigned VF);
-
-  /// A helper function for converting Scalar types to vector types.
-  /// If the incoming type is void, we return void. If the VF is 1, we return
-  /// the scalar type.
-  static Type* ToVectorTy(Type *Scalar, unsigned VF);
-
-  /// The loop that we evaluate.
-  Loop *TheLoop;
-  /// Scev analysis.
-  ScalarEvolution *SE;
-
-  /// Vectorization legality.
-  LoopVectorizationLegality *Legal;
-  /// Vector target information.
-  const VectorTargetTransformInfo *VTTI;
-};
-
-}// namespace llvm
-
-#endif //LLVM_TRANSFORM_VECTORIZE_LOOP_VECTORIZE_H
-
diff --git a/lib/Transforms/Vectorize/Vectorize.cpp b/lib/Transforms/Vectorize/Vectorize.cpp
index 3fb36cadea..19eefd2f87 100644
--- a/lib/Transforms/Vectorize/Vectorize.cpp
+++ b/lib/Transforms/Vectorize/Vectorize.cpp
@@ -1,4 +1,4 @@
-//===-- Vectorize.cpp -----------------------------------------------------===//
+   //===-- Vectorize.cpp -----------------------------------------------------===//
 //
 //                     The LLVM Compiler Infrastructure
 //