am 1c4ad5ef: Merge branch \'upstream\' into merge-2012_09_10

* commit '1c4ad5ef4fab105f0c8af7edd026e00502fb6279': (446 commits)
  Revert r163556. Missed updates to tablegen files.
  Update function names to conform to guidelines.  No functional change intended.
  test/CodeGen/X86/ms-inline-asm.ll: Relax for non-darwin x86 targets. '##InlineAsm' could not be seen in other hosts.
  [ms-inline asm] Properly emit the asm directives when the AsmPrinterVariant and InlineAsmVariant don't match.
  Update test case for Release builds.
  Remove redundant semicolons which are null statements.
  Disable stack coloring because it makes dragonegg fail bootstrapping.
  [ms-inline asm] Pass the correct AsmVariant to the PrintAsmOperand() function and update the printOperand() function accordingly.
  [ms-inline asm] Add support for .att_syntax directive.
  Enable stack coloring.
  Don't attempt to use flags from predicated instructions.
  [Object] Extract Elf_Ehdr. Patch by Hemant Kulkarni!
  Stack Coloring: Handle the case where END markers come before BEGIN markers properly.
  Enhance PR11334 fix to support extload from v2f32/v4f32
  Add "blocked" heuristic to the Hexagon MI scheduler.
  Fold multiply by 0 or 1 when in UnsafeFPMath mode in SelectionDAG::getNode().
  whitespace
  Add boolean simplification support from CMOV
  Fix an assertion failure when optimising a shufflevector incorrectly into concat_vectors, and a followup bug with SelectionDAG::getNode() creating nodes with invalid types.
  Minor cleanup. No functional change.
  ...

26 files changed, 696 insertions, 210 deletions

diff --git a/lib/Analysis/AliasAnalysis.cpp b/lib/Analysis/AliasAnalysis.cpp
index 3b6aab1..f768eec 100644
--- a/lib/Analysis/AliasAnalysis.cpp
+++ b/lib/Analysis/AliasAnalysis.cpp
@@ -36,6 +36,7 @@
 #include "llvm/LLVMContext.h"
 #include "llvm/Type.h"
 #include "llvm/Target/TargetData.h"
+#include "llvm/Target/TargetLibraryInfo.h"
 using namespace llvm;
 
 // Register the AliasAnalysis interface, providing a nice name to refer to.
@@ -452,6 +453,7 @@ AliasAnalysis::~AliasAnalysis() {}
 ///
 void AliasAnalysis::InitializeAliasAnalysis(Pass *P) {
   TD = P->getAnalysisIfAvailable<TargetData>();
+  TLI = P->getAnalysisIfAvailable<TargetLibraryInfo>();
   AA = &P->getAnalysis<AliasAnalysis>();
 }
 
diff --git a/lib/Analysis/AliasSetTracker.cpp b/lib/Analysis/AliasSetTracker.cpp
index 92e8906..e9dcb37 100644
--- a/lib/Analysis/AliasSetTracker.cpp
+++ b/lib/Analysis/AliasSetTracker.cpp
@@ -550,7 +550,7 @@ void AliasSetTracker::copyValue(Value *From, Value *To) {
 //===----------------------------------------------------------------------===//
 
 void AliasSet::print(raw_ostream &OS) const {
-  OS << "  AliasSet[" << (void*)this << ", " << RefCount << "] ";
+  OS << "  AliasSet[" << (const void*)this << ", " << RefCount << "] ";
   OS << (AliasTy == MustAlias ? "must" : "may") << " alias, ";
   switch (AccessTy) {
   case NoModRef: OS << "No access "; break;
@@ -590,8 +590,10 @@ void AliasSetTracker::print(raw_ostream &OS) const {
   OS << "\n";
 }
 
+#ifndef NDEBUG
 void AliasSet::dump() const { print(dbgs()); }
 void AliasSetTracker::dump() const { print(dbgs()); }
+#endif
 
 //===----------------------------------------------------------------------===//
 //                     ASTCallbackVH Class Implementation
diff --git a/lib/Analysis/Analysis.cpp b/lib/Analysis/Analysis.cpp
index 0ba6af9..87a75fd 100644
--- a/lib/Analysis/Analysis.cpp
+++ b/lib/Analysis/Analysis.cpp
@@ -61,6 +61,7 @@ void llvm::initializeAnalysis(PassRegistry &Registry) {
   initializePathProfileLoaderPassPass(Registry);
   initializeProfileVerifierPassPass(Registry);
   initializePathProfileVerifierPass(Registry);
+  initializeProfileMetadataLoaderPassPass(Registry);
   initializeRegionInfoPass(Registry);
   initializeRegionViewerPass(Registry);
   initializeRegionPrinterPass(Registry);
diff --git a/lib/Analysis/BasicAliasAnalysis.cpp b/lib/Analysis/BasicAliasAnalysis.cpp
index 1d028c2..a3bc06a 100644
--- a/lib/Analysis/BasicAliasAnalysis.cpp
+++ b/lib/Analysis/BasicAliasAnalysis.cpp
@@ -85,9 +85,10 @@ static bool isEscapeSource(const Value *V) {
 /// getObjectSize - Return the size of the object specified by V, or
 /// UnknownSize if unknown.
 static uint64_t getObjectSize(const Value *V, const TargetData &TD,
+                              const TargetLibraryInfo &TLI,
                               bool RoundToAlign = false) {
   uint64_t Size;
-  if (getObjectSize(V, Size, &TD, RoundToAlign))
+  if (getObjectSize(V, Size, &TD, &TLI, RoundToAlign))
     return Size;
   return AliasAnalysis::UnknownSize;
 }
@@ -95,10 +96,11 @@ static uint64_t getObjectSize(const Value *V, const TargetData &TD,
 /// isObjectSmallerThan - Return true if we can prove that the object specified
 /// by V is smaller than Size.
 static bool isObjectSmallerThan(const Value *V, uint64_t Size,
-                                const TargetData &TD) {
+                                const TargetData &TD,
+                                const TargetLibraryInfo &TLI) {
   // This function needs to use the aligned object size because we allow
   // reads a bit past the end given sufficient alignment.
-  uint64_t ObjectSize = getObjectSize(V, TD, /*RoundToAlign*/true);
+  uint64_t ObjectSize = getObjectSize(V, TD, TLI, /*RoundToAlign*/true);
   
   return ObjectSize != AliasAnalysis::UnknownSize && ObjectSize < Size;
 }
@@ -106,8 +108,8 @@ static bool isObjectSmallerThan(const Value *V, uint64_t Size,
 /// isObjectSize - Return true if we can prove that the object specified
 /// by V has size Size.
 static bool isObjectSize(const Value *V, uint64_t Size,
-                         const TargetData &TD) {
-  uint64_t ObjectSize = getObjectSize(V, TD);
+                         const TargetData &TD, const TargetLibraryInfo &TLI) {
+  uint64_t ObjectSize = getObjectSize(V, TD, TLI);
   return ObjectSize != AliasAnalysis::UnknownSize && ObjectSize == Size;
 }
 
@@ -126,6 +128,15 @@ namespace {
     const Value *V;
     ExtensionKind Extension;
     int64_t Scale;
+
+    bool operator==(const VariableGEPIndex &Other) const {
+      return V == Other.V && Extension == Other.Extension &&
+        Scale == Other.Scale;
+    }
+
+    bool operator!=(const VariableGEPIndex &Other) const {
+      return !operator==(Other);
+    }
   };
 }
 
@@ -417,13 +428,7 @@ namespace {
   /// BasicAliasAnalysis - This is the primary alias analysis implementation.
   struct BasicAliasAnalysis : public ImmutablePass, public AliasAnalysis {
     static char ID; // Class identification, replacement for typeinfo
-    BasicAliasAnalysis() : ImmutablePass(ID),
-                           // AliasCache rarely has more than 1 or 2 elements,
-                           // so start it off fairly small so that clear()
-                           // doesn't have to tromp through 64 (the default)
-                           // elements on each alias query. This really wants
-                           // something like a SmallDenseMap.
-                           AliasCache(8) {
+    BasicAliasAnalysis() : ImmutablePass(ID) {
       initializeBasicAliasAnalysisPass(*PassRegistry::getPassRegistry());
     }
 
@@ -443,7 +448,11 @@ namespace {
              "BasicAliasAnalysis doesn't support interprocedural queries.");
       AliasResult Alias = aliasCheck(LocA.Ptr, LocA.Size, LocA.TBAATag,
                                      LocB.Ptr, LocB.Size, LocB.TBAATag);
-      AliasCache.clear();
+      // AliasCache rarely has more than 1 or 2 elements, always use
+      // shrink_and_clear so it quickly returns to the inline capacity of the
+      // SmallDenseMap if it ever grows larger.
+      // FIXME: This should really be shrink_to_inline_capacity_and_clear().
+      AliasCache.shrink_and_clear();
       return Alias;
     }
 
@@ -481,7 +490,7 @@ namespace {
   private:
     // AliasCache - Track alias queries to guard against recursion.
     typedef std::pair<Location, Location> LocPair;
-    typedef DenseMap<LocPair, AliasResult> AliasCacheTy;
+    typedef SmallDenseMap<LocPair, AliasResult, 8> AliasCacheTy;
     AliasCacheTy AliasCache;
 
     // Visited - Track instructions visited by pointsToConstantMemory.
@@ -490,6 +499,7 @@ namespace {
     // aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP
     // instruction against another.
     AliasResult aliasGEP(const GEPOperator *V1, uint64_t V1Size,
+                         const MDNode *V1TBAAInfo,
                          const Value *V2, uint64_t V2Size,
                          const MDNode *V2TBAAInfo,
                          const Value *UnderlyingV1, const Value *UnderlyingV2);
@@ -807,6 +817,21 @@ BasicAliasAnalysis::getModRefInfo(ImmutableCallSite CS,
   return ModRefResult(AliasAnalysis::getModRefInfo(CS, Loc) & Min);
 }
 
+static bool areVarIndicesEqual(SmallVector<VariableGEPIndex, 4> &Indices1,
+                               SmallVector<VariableGEPIndex, 4> &Indices2) {
+  unsigned Size1 = Indices1.size();
+  unsigned Size2 = Indices2.size();
+
+  if (Size1 != Size2)
+    return false;
+
+  for (unsigned I = 0; I != Size1; ++I)
+    if (Indices1[I] != Indices2[I])
+      return false;
+
+  return true;
+}
+
 /// aliasGEP - Provide a bunch of ad-hoc rules to disambiguate a GEP instruction
 /// against another pointer.  We know that V1 is a GEP, but we don't know
 /// anything about V2.  UnderlyingV1 is GetUnderlyingObject(GEP1, TD),
@@ -814,6 +839,7 @@ BasicAliasAnalysis::getModRefInfo(ImmutableCallSite CS,
 ///
 AliasAnalysis::AliasResult
 BasicAliasAnalysis::aliasGEP(const GEPOperator *GEP1, uint64_t V1Size,
+                             const MDNode *V1TBAAInfo,
                              const Value *V2, uint64_t V2Size,
                              const MDNode *V2TBAAInfo,
                              const Value *UnderlyingV1,
@@ -821,9 +847,41 @@ BasicAliasAnalysis::aliasGEP(const GEPOperator *GEP1, uint64_t V1Size,
   int64_t GEP1BaseOffset;
   SmallVector<VariableGEPIndex, 4> GEP1VariableIndices;
 
-  // If we have two gep instructions with must-alias'ing base pointers, figure
-  // out if the indexes to the GEP tell us anything about the derived pointer.
+  // If we have two gep instructions with must-alias or not-alias'ing base
+  // pointers, figure out if the indexes to the GEP tell us anything about the
+  // derived pointer.
   if (const GEPOperator *GEP2 = dyn_cast<GEPOperator>(V2)) {
+    // Check for geps of non-aliasing underlying pointers where the offsets are
+    // identical.
+    if (V1Size == V2Size) {
+      // Do the base pointers alias assuming type and size.
+      AliasResult PreciseBaseAlias = aliasCheck(UnderlyingV1, V1Size,
+                                                V1TBAAInfo, UnderlyingV2,
+                                                V2Size, V2TBAAInfo);
+      if (PreciseBaseAlias == NoAlias) {
+        // See if the computed offset from the common pointer tells us about the
+        // relation of the resulting pointer.
+        int64_t GEP2BaseOffset;
+        SmallVector<VariableGEPIndex, 4> GEP2VariableIndices;
+        const Value *GEP2BasePtr =
+          DecomposeGEPExpression(GEP2, GEP2BaseOffset, GEP2VariableIndices, TD);
+        const Value *GEP1BasePtr =
+          DecomposeGEPExpression(GEP1, GEP1BaseOffset, GEP1VariableIndices, TD);
+        // DecomposeGEPExpression and GetUnderlyingObject should return the
+        // same result except when DecomposeGEPExpression has no TargetData.
+        if (GEP1BasePtr != UnderlyingV1 || GEP2BasePtr != UnderlyingV2) {
+          assert(TD == 0 &&
+             "DecomposeGEPExpression and GetUnderlyingObject disagree!");
+          return MayAlias;
+        }
+        // Same offsets.
+        if (GEP1BaseOffset == GEP2BaseOffset &&
+            areVarIndicesEqual(GEP1VariableIndices, GEP2VariableIndices))
+          return NoAlias;
+        GEP1VariableIndices.clear();
+      }
+    }
+
     // Do the base pointers alias?
     AliasResult BaseAlias = aliasCheck(UnderlyingV1, UnknownSize, 0,
                                        UnderlyingV2, UnknownSize, 0);
@@ -843,9 +901,8 @@ BasicAliasAnalysis::aliasGEP(const GEPOperator *GEP1, uint64_t V1Size,
     const Value *GEP2BasePtr =
       DecomposeGEPExpression(GEP2, GEP2BaseOffset, GEP2VariableIndices, TD);
     
-    // If DecomposeGEPExpression isn't able to look all the way through the
-    // addressing operation, we must not have TD and this is too complex for us
-    // to handle without it.
+    // DecomposeGEPExpression and GetUnderlyingObject should return the
+    // same result except when DecomposeGEPExpression has no TargetData.
     if (GEP1BasePtr != UnderlyingV1 || GEP2BasePtr != UnderlyingV2) {
       assert(TD == 0 &&
              "DecomposeGEPExpression and GetUnderlyingObject disagree!");
@@ -879,9 +936,8 @@ BasicAliasAnalysis::aliasGEP(const GEPOperator *GEP1, uint64_t V1Size,
     const Value *GEP1BasePtr =
       DecomposeGEPExpression(GEP1, GEP1BaseOffset, GEP1VariableIndices, TD);
     
-    // If DecomposeGEPExpression isn't able to look all the way through the
-    // addressing operation, we must not have TD and this is too complex for us
-    // to handle without it.
+    // DecomposeGEPExpression and GetUnderlyingObject should return the
+    // same result except when DecomposeGEPExpression has no TargetData.
     if (GEP1BasePtr != UnderlyingV1) {
       assert(TD == 0 &&
              "DecomposeGEPExpression and GetUnderlyingObject disagree!");
@@ -1004,12 +1060,42 @@ BasicAliasAnalysis::aliasPHI(const PHINode *PN, uint64_t PNSize,
   // on corresponding edges.
   if (const PHINode *PN2 = dyn_cast<PHINode>(V2))
     if (PN2->getParent() == PN->getParent()) {
+      LocPair Locs(Location(PN, PNSize, PNTBAAInfo),
+                   Location(V2, V2Size, V2TBAAInfo));
+      if (PN > V2)
+        std::swap(Locs.first, Locs.second);
+
       AliasResult Alias =
         aliasCheck(PN->getIncomingValue(0), PNSize, PNTBAAInfo,
                    PN2->getIncomingValueForBlock(PN->getIncomingBlock(0)),
                    V2Size, V2TBAAInfo);
       if (Alias == MayAlias)
         return MayAlias;
+
+      // If the first source of the PHI nodes NoAlias and the other inputs are
+      // the PHI node itself through some amount of recursion this does not add
+      // any new information so just return NoAlias.
+      // bb:
+      //    ptr = ptr2 + 1
+      // loop:
+      //    ptr_phi = phi [bb, ptr], [loop, ptr_plus_one]
+      //    ptr2_phi = phi [bb, ptr2], [loop, ptr2_plus_one]
+      //    ...
+      //    ptr_plus_one = gep ptr_phi, 1
+      //    ptr2_plus_one = gep ptr2_phi, 1
+      // We assume for the recursion that the the phis (ptr_phi, ptr2_phi) do
+      // not alias each other.
+      bool ArePhisAssumedNoAlias = false;
+      AliasResult OrigAliasResult;
+      if (Alias == NoAlias) {
+        // Pretend the phis do not alias.
+        assert(AliasCache.count(Locs) &&
+               "There must exist an entry for the phi node");
+        OrigAliasResult = AliasCache[Locs];
+        AliasCache[Locs] = NoAlias;
+        ArePhisAssumedNoAlias = true;
+      }
+
       for (unsigned i = 1, e = PN->getNumIncomingValues(); i != e; ++i) {
         AliasResult ThisAlias =
           aliasCheck(PN->getIncomingValue(i), PNSize, PNTBAAInfo,
@@ -1019,6 +1105,11 @@ BasicAliasAnalysis::aliasPHI(const PHINode *PN, uint64_t PNSize,
         if (Alias == MayAlias)
           break;
       }
+
+      // Reset if speculation failed.
+      if (ArePhisAssumedNoAlias && Alias != NoAlias)
+        AliasCache[Locs] = OrigAliasResult;
+
       return Alias;
     }
 
@@ -1133,8 +1224,8 @@ BasicAliasAnalysis::aliasCheck(const Value *V1, uint64_t V1Size,
   // If the size of one access is larger than the entire object on the other
   // side, then we know such behavior is undefined and can assume no alias.
   if (TD)
-    if ((V1Size != UnknownSize && isObjectSmallerThan(O2, V1Size, *TD)) ||
-        (V2Size != UnknownSize && isObjectSmallerThan(O1, V2Size, *TD)))
+    if ((V1Size != UnknownSize && isObjectSmallerThan(O2, V1Size, *TD, *TLI)) ||
+        (V2Size != UnknownSize && isObjectSmallerThan(O1, V2Size, *TD, *TLI)))
       return NoAlias;
   
   // Check the cache before climbing up use-def chains. This also terminates
@@ -1156,7 +1247,7 @@ BasicAliasAnalysis::aliasCheck(const Value *V1, uint64_t V1Size,
     std::swap(O1, O2);
   }
   if (const GEPOperator *GV1 = dyn_cast<GEPOperator>(V1)) {
-    AliasResult Result = aliasGEP(GV1, V1Size, V2, V2Size, V2TBAAInfo, O1, O2);
+    AliasResult Result = aliasGEP(GV1, V1Size, V1TBAAInfo, V2, V2Size, V2TBAAInfo, O1, O2);
     if (Result != MayAlias) return AliasCache[Locs] = Result;
   }
 
@@ -1184,8 +1275,8 @@ BasicAliasAnalysis::aliasCheck(const Value *V1, uint64_t V1Size,
   // accesses is accessing the entire object, then the accesses must
   // overlap in some way.
   if (TD && O1 == O2)
-    if ((V1Size != UnknownSize && isObjectSize(O1, V1Size, *TD)) ||
-        (V2Size != UnknownSize && isObjectSize(O2, V2Size, *TD)))
+    if ((V1Size != UnknownSize && isObjectSize(O1, V1Size, *TD, *TLI)) ||
+        (V2Size != UnknownSize && isObjectSize(O2, V2Size, *TD, *TLI)))
       return AliasCache[Locs] = PartialAlias;
 
   AliasResult Result =
diff --git a/lib/Analysis/BranchProbabilityInfo.cpp b/lib/Analysis/BranchProbabilityInfo.cpp
index b255ce6..04a6560 100644
--- a/lib/Analysis/BranchProbabilityInfo.cpp
+++ b/lib/Analysis/BranchProbabilityInfo.cpp
@@ -115,14 +115,14 @@ bool BranchProbabilityInfo::calcUnreachableHeuristics(BasicBlock *BB) {
     return false;
   }
 
-  SmallPtrSet<BasicBlock *, 4> UnreachableEdges;
-  SmallPtrSet<BasicBlock *, 4> ReachableEdges;
+  SmallVector<unsigned, 4> UnreachableEdges;
+  SmallVector<unsigned, 4> ReachableEdges;
 
   for (succ_iterator I = succ_begin(BB), E = succ_end(BB); I != E; ++I) {
     if (PostDominatedByUnreachable.count(*I))
-      UnreachableEdges.insert(*I);
+      UnreachableEdges.push_back(I.getSuccessorIndex());
     else
-      ReachableEdges.insert(*I);
+      ReachableEdges.push_back(I.getSuccessorIndex());
   }
 
   // If all successors are in the set of blocks post-dominated by unreachable,
@@ -136,18 +136,19 @@ bool BranchProbabilityInfo::calcUnreachableHeuristics(BasicBlock *BB) {
     return false;
 
   uint32_t UnreachableWeight =
-    std::max(UR_TAKEN_WEIGHT / UnreachableEdges.size(), MIN_WEIGHT);
-  for (SmallPtrSet<BasicBlock *, 4>::iterator I = UnreachableEdges.begin(),
-                                              E = UnreachableEdges.end();
+    std::max(UR_TAKEN_WEIGHT / (unsigned)UnreachableEdges.size(), MIN_WEIGHT);
+  for (SmallVector<unsigned, 4>::iterator I = UnreachableEdges.begin(),
+                                          E = UnreachableEdges.end();
        I != E; ++I)
     setEdgeWeight(BB, *I, UnreachableWeight);
 
   if (ReachableEdges.empty())
     return true;
   uint32_t ReachableWeight =
-    std::max(UR_NONTAKEN_WEIGHT / ReachableEdges.size(), NORMAL_WEIGHT);
-  for (SmallPtrSet<BasicBlock *, 4>::iterator I = ReachableEdges.begin(),
-                                              E = ReachableEdges.end();
+    std::max(UR_NONTAKEN_WEIGHT / (unsigned)ReachableEdges.size(),
+             NORMAL_WEIGHT);
+  for (SmallVector<unsigned, 4>::iterator I = ReachableEdges.begin(),
+                                          E = ReachableEdges.end();
        I != E; ++I)
     setEdgeWeight(BB, *I, ReachableWeight);
 
@@ -187,7 +188,7 @@ bool BranchProbabilityInfo::calcMetadataWeights(BasicBlock *BB) {
   }
   assert(Weights.size() == TI->getNumSuccessors() && "Checked above");
   for (unsigned i = 0, e = TI->getNumSuccessors(); i != e; ++i)
-    setEdgeWeight(BB, TI->getSuccessor(i), Weights[i]);
+    setEdgeWeight(BB, i, Weights[i]);
 
   return true;
 }
@@ -211,19 +212,17 @@ bool BranchProbabilityInfo::calcPointerHeuristics(BasicBlock *BB) {
 
   assert(CI->getOperand(1)->getType()->isPointerTy());
 
-  BasicBlock *Taken = BI->getSuccessor(0);
-  BasicBlock *NonTaken = BI->getSuccessor(1);
-
   // p != 0   ->   isProb = true
   // p == 0   ->   isProb = false
   // p != q   ->   isProb = true
   // p == q   ->   isProb = false;
+  unsigned TakenIdx = 0, NonTakenIdx = 1;
   bool isProb = CI->getPredicate() == ICmpInst::ICMP_NE;
   if (!isProb)
-    std::swap(Taken, NonTaken);
+    std::swap(TakenIdx, NonTakenIdx);
 
-  setEdgeWeight(BB, Taken, PH_TAKEN_WEIGHT);
-  setEdgeWeight(BB, NonTaken, PH_NONTAKEN_WEIGHT);
+  setEdgeWeight(BB, TakenIdx, PH_TAKEN_WEIGHT);
+  setEdgeWeight(BB, NonTakenIdx, PH_NONTAKEN_WEIGHT);
   return true;
 }
 
@@ -234,17 +233,17 @@ bool BranchProbabilityInfo::calcLoopBranchHeuristics(BasicBlock *BB) {
   if (!L)
     return false;
 
-  SmallPtrSet<BasicBlock *, 8> BackEdges;
-  SmallPtrSet<BasicBlock *, 8> ExitingEdges;
-  SmallPtrSet<BasicBlock *, 8> InEdges; // Edges from header to the loop.
+  SmallVector<unsigned, 8> BackEdges;
+  SmallVector<unsigned, 8> ExitingEdges;
+  SmallVector<unsigned, 8> InEdges; // Edges from header to the loop.
 
   for (succ_iterator I = succ_begin(BB), E = succ_end(BB); I != E; ++I) {
     if (!L->contains(*I))
-      ExitingEdges.insert(*I);
+      ExitingEdges.push_back(I.getSuccessorIndex());
     else if (L->getHeader() == *I)
-      BackEdges.insert(*I);
+      BackEdges.push_back(I.getSuccessorIndex());
     else
-      InEdges.insert(*I);
+      InEdges.push_back(I.getSuccessorIndex());
   }
 
   if (uint32_t numBackEdges = BackEdges.size()) {
@@ -252,10 +251,9 @@ bool BranchProbabilityInfo::calcLoopBranchHeuristics(BasicBlock *BB) {
     if (backWeight < NORMAL_WEIGHT)
       backWeight = NORMAL_WEIGHT;
 
-    for (SmallPtrSet<BasicBlock *, 8>::iterator EI = BackEdges.begin(),
+    for (SmallVector<unsigned, 8>::iterator EI = BackEdges.begin(),
          EE = BackEdges.end(); EI != EE; ++EI) {
-      BasicBlock *Back = *EI;
-      setEdgeWeight(BB, Back, backWeight);
+      setEdgeWeight(BB, *EI, backWeight);
     }
   }
 
@@ -264,10 +262,9 @@ bool BranchProbabilityInfo::calcLoopBranchHeuristics(BasicBlock *BB) {
     if (inWeight < NORMAL_WEIGHT)
       inWeight = NORMAL_WEIGHT;
 
-    for (SmallPtrSet<BasicBlock *, 8>::iterator EI = InEdges.begin(),
+    for (SmallVector<unsigned, 8>::iterator EI = InEdges.begin(),
          EE = InEdges.end(); EI != EE; ++EI) {
-      BasicBlock *Back = *EI;
-      setEdgeWeight(BB, Back, inWeight);
+      setEdgeWeight(BB, *EI, inWeight);
     }
   }
 
@@ -276,10 +273,9 @@ bool BranchProbabilityInfo::calcLoopBranchHeuristics(BasicBlock *BB) {
     if (exitWeight < MIN_WEIGHT)
       exitWeight = MIN_WEIGHT;
 
-    for (SmallPtrSet<BasicBlock *, 8>::iterator EI = ExitingEdges.begin(),
+    for (SmallVector<unsigned, 8>::iterator EI = ExitingEdges.begin(),
          EE = ExitingEdges.end(); EI != EE; ++EI) {
-      BasicBlock *Exiting = *EI;
-      setEdgeWeight(BB, Exiting, exitWeight);
+      setEdgeWeight(BB, *EI, exitWeight);
     }
   }
 
@@ -335,14 +331,13 @@ bool BranchProbabilityInfo::calcZeroHeuristics(BasicBlock *BB) {
     return false;
   }
 
-  BasicBlock *Taken = BI->getSuccessor(0);
-  BasicBlock *NonTaken = BI->getSuccessor(1);
+  unsigned TakenIdx = 0, NonTakenIdx = 1;
 
   if (!isProb)
-    std::swap(Taken, NonTaken);
+    std::swap(TakenIdx, NonTakenIdx);
 
-  setEdgeWeight(BB, Taken, ZH_TAKEN_WEIGHT);
-  setEdgeWeight(BB, NonTaken, ZH_NONTAKEN_WEIGHT);
+  setEdgeWeight(BB, TakenIdx, ZH_TAKEN_WEIGHT);
+  setEdgeWeight(BB, NonTakenIdx, ZH_NONTAKEN_WEIGHT);
 
   return true;
 }
@@ -372,14 +367,13 @@ bool BranchProbabilityInfo::calcFloatingPointHeuristics(BasicBlock *BB) {
     return false;
   }
 
-  BasicBlock *Taken = BI->getSuccessor(0);
-  BasicBlock *NonTaken = BI->getSuccessor(1);
+  unsigned TakenIdx = 0, NonTakenIdx = 1;
 
   if (!isProb)
-    std::swap(Taken, NonTaken);
+    std::swap(TakenIdx, NonTakenIdx);
 
-  setEdgeWeight(BB, Taken, FPH_TAKEN_WEIGHT);
-  setEdgeWeight(BB, NonTaken, FPH_NONTAKEN_WEIGHT);
+  setEdgeWeight(BB, TakenIdx, FPH_TAKEN_WEIGHT);
+  setEdgeWeight(BB, NonTakenIdx, FPH_NONTAKEN_WEIGHT);
 
   return true;
 }
@@ -389,11 +383,8 @@ bool BranchProbabilityInfo::calcInvokeHeuristics(BasicBlock *BB) {
   if (!II)
     return false;
 
-  BasicBlock *Normal = II->getNormalDest();
-  BasicBlock *Unwind = II->getUnwindDest();
-
-  setEdgeWeight(BB, Normal, IH_TAKEN_WEIGHT);
-  setEdgeWeight(BB, Unwind, IH_NONTAKEN_WEIGHT);
+  setEdgeWeight(BB, 0/*Index for Normal*/, IH_TAKEN_WEIGHT);
+  setEdgeWeight(BB, 1/*Index for Unwind*/, IH_NONTAKEN_WEIGHT);
   return true;
 }
 
@@ -450,8 +441,7 @@ uint32_t BranchProbabilityInfo::getSumForBlock(const BasicBlock *BB) const {
   uint32_t Sum = 0;
 
   for (succ_const_iterator I = succ_begin(BB), E = succ_end(BB); I != E; ++I) {
-    const BasicBlock *Succ = *I;
-    uint32_t Weight = getEdgeWeight(BB, Succ);
+    uint32_t Weight = getEdgeWeight(BB, I.getSuccessorIndex());
     uint32_t PrevSum = Sum;
 
     Sum += Weight;
@@ -494,11 +484,13 @@ BasicBlock *BranchProbabilityInfo::getHotSucc(BasicBlock *BB) const {
   return 0;
 }
 
-// Return edge's weight. If can't find it, return DEFAULT_WEIGHT value.
+/// Get the raw edge weight for the edge. If can't find it, return
+/// DEFAULT_WEIGHT value. Here an edge is specified using PredBlock and an index
+/// to the successors.
 uint32_t BranchProbabilityInfo::
-getEdgeWeight(const BasicBlock *Src, const BasicBlock *Dst) const {
-  Edge E(Src, Dst);
-  DenseMap<Edge, uint32_t>::const_iterator I = Weights.find(E);
+getEdgeWeight(const BasicBlock *Src, unsigned IndexInSuccessors) const {
+  DenseMap<Edge, uint32_t>::const_iterator I =
+      Weights.find(std::make_pair(Src, IndexInSuccessors));
 
   if (I != Weights.end())
     return I->second;
@@ -506,15 +498,43 @@ getEdgeWeight(const BasicBlock *Src, const BasicBlock *Dst) const {
   return DEFAULT_WEIGHT;
 }
 
+/// Get the raw edge weight calculated for the block pair. This returns the sum
+/// of all raw edge weights from Src to Dst.
+uint32_t BranchProbabilityInfo::
+getEdgeWeight(const BasicBlock *Src, const BasicBlock *Dst) const {
+  uint32_t Weight = 0;
+  DenseMap<Edge, uint32_t>::const_iterator MapI;
+  for (succ_const_iterator I = succ_begin(Src), E = succ_end(Src); I != E; ++I)
+    if (*I == Dst) {
+      MapI = Weights.find(std::make_pair(Src, I.getSuccessorIndex()));
+      if (MapI != Weights.end())
+        Weight += MapI->second;
+    }
+  return (Weight == 0) ? DEFAULT_WEIGHT : Weight;
+}
+
+/// Set the edge weight for a given edge specified by PredBlock and an index
+/// to the successors.
 void BranchProbabilityInfo::
-setEdgeWeight(const BasicBlock *Src, const BasicBlock *Dst, uint32_t Weight) {
-  Weights[std::make_pair(Src, Dst)] = Weight;
+setEdgeWeight(const BasicBlock *Src, unsigned IndexInSuccessors,
+              uint32_t Weight) {
+  Weights[std::make_pair(Src, IndexInSuccessors)] = Weight;
   DEBUG(dbgs() << "set edge " << Src->getName() << " -> "
-               << Dst->getName() << " weight to " << Weight
-               << (isEdgeHot(Src, Dst) ? " [is HOT now]\n" : "\n"));
+               << IndexInSuccessors << " successor weight to "
+               << Weight << "\n");
 }
 
+/// Get an edge's probability, relative to other out-edges from Src.
+BranchProbability BranchProbabilityInfo::
+getEdgeProbability(const BasicBlock *Src, unsigned IndexInSuccessors) const {
+  uint32_t N = getEdgeWeight(Src, IndexInSuccessors);
+  uint32_t D = getSumForBlock(Src);
+
+  return BranchProbability(N, D);
+}
 
+/// Get the probability of going from Src to Dst. It returns the sum of all
+/// probabilities for edges from Src to Dst.
 BranchProbability BranchProbabilityInfo::
 getEdgeProbability(const BasicBlock *Src, const BasicBlock *Dst) const {
 
diff --git a/lib/Analysis/CMakeLists.txt b/lib/Analysis/CMakeLists.txt
index 96e68b4..e461848 100644
--- a/lib/Analysis/CMakeLists.txt
+++ b/lib/Analysis/CMakeLists.txt
@@ -44,6 +44,8 @@ add_llvm_library(LLVMAnalysis
   ProfileInfoLoader.cpp
   ProfileInfoLoaderPass.cpp
   ProfileVerifierPass.cpp
+  ProfileDataLoader.cpp
+  ProfileDataLoaderPass.cpp
   RegionInfo.cpp
   RegionPass.cpp
   RegionPrinter.cpp
diff --git a/lib/Analysis/ConstantFolding.cpp b/lib/Analysis/ConstantFolding.cpp
index f5e619c..4ad613c 100644
--- a/lib/Analysis/ConstantFolding.cpp
+++ b/lib/Analysis/ConstantFolding.cpp
@@ -659,7 +659,8 @@ static Constant *SymbolicallyEvaluateGEP(ArrayRef<Constant *> Ops,
   unsigned BitWidth = TD->getTypeSizeInBits(IntPtrTy);
   APInt Offset =
     APInt(BitWidth, TD->getIndexedOffset(Ptr->getType(),
-                                         makeArrayRef((Value **)Ops.data() + 1,
+                                         makeArrayRef((Value *const*)
+                                                        Ops.data() + 1,
                                                       Ops.size() - 1)));
   Ptr = StripPtrCastKeepAS(Ptr);
 
diff --git a/lib/Analysis/DominanceFrontier.cpp b/lib/Analysis/DominanceFrontier.cpp
index 1604576..5536a9b 100644
--- a/lib/Analysis/DominanceFrontier.cpp
+++ b/lib/Analysis/DominanceFrontier.cpp
@@ -133,7 +133,9 @@ void DominanceFrontierBase::print(raw_ostream &OS, const Module* ) const {
   }
 }
 
+#ifndef NDEBUG
 void DominanceFrontierBase::dump() const {
   print(dbgs());
 }
+#endif
 
diff --git a/lib/Analysis/IPA/CallGraph.cpp b/lib/Analysis/IPA/CallGraph.cpp
index 0df3e8a..947ad51 100644
--- a/lib/Analysis/IPA/CallGraph.cpp
+++ b/lib/Analysis/IPA/CallGraph.cpp
@@ -198,9 +198,11 @@ void CallGraph::print(raw_ostream &OS, Module*) const {
   for (CallGraph::const_iterator I = begin(), E = end(); I != E; ++I)
     I->second->print(OS);
 }
+#ifndef NDEBUG
 void CallGraph::dump() const {
   print(dbgs(), 0);
 }
+#endif
 
 //===----------------------------------------------------------------------===//
 // Implementations of public modification methods
@@ -267,7 +269,9 @@ void CallGraphNode::print(raw_ostream &OS) const {
   OS << '\n';
 }
 
+#ifndef NDEBUG
 void CallGraphNode::dump() const { print(dbgs()); }
+#endif
 
 /// removeCallEdgeFor - This method removes the edge in the node for the
 /// specified call site.  Note that this method takes linear time, so it
diff --git a/lib/Analysis/IPA/GlobalsModRef.cpp b/lib/Analysis/IPA/GlobalsModRef.cpp
index 22f6e96..990caa8 100644
--- a/lib/Analysis/IPA/GlobalsModRef.cpp
+++ b/lib/Analysis/IPA/GlobalsModRef.cpp
@@ -263,7 +263,7 @@ bool GlobalsModRef::AnalyzeUsesOfPointer(Value *V,
     } else if (BitCastInst *BCI = dyn_cast<BitCastInst>(U)) {
       if (AnalyzeUsesOfPointer(BCI, Readers, Writers, OkayStoreDest))
         return true;
-    } else if (isFreeCall(U)) {
+    } else if (isFreeCall(U, TLI)) {
       Writers.push_back(cast<Instruction>(U)->getParent()->getParent());
     } else if (CallInst *CI = dyn_cast<CallInst>(U)) {
       // Make sure that this is just the function being called, not that it is
@@ -329,7 +329,7 @@ bool GlobalsModRef::AnalyzeIndirectGlobalMemory(GlobalValue *GV) {
       // Check the value being stored.
       Value *Ptr = GetUnderlyingObject(SI->getOperand(0));
 
-      if (!isAllocLikeFn(Ptr))
+      if (!isAllocLikeFn(Ptr, TLI))
         return false;  // Too hard to analyze.
 
       // Analyze all uses of the allocation.  If any of them are used in a
@@ -458,7 +458,7 @@ void GlobalsModRef::AnalyzeCallGraph(CallGraph &CG, Module &M) {
           if (SI->isVolatile())
             // Treat volatile stores as reading memory somewhere.
             FunctionEffect |= Ref;
-        } else if (isAllocationFn(&*II) || isFreeCall(&*II)) {
+        } else if (isAllocationFn(&*II, TLI) || isFreeCall(&*II, TLI)) {
           FunctionEffect |= ModRef;
         } else if (IntrinsicInst *Intrinsic = dyn_cast<IntrinsicInst>(&*II)) {
           // The callgraph doesn't include intrinsic calls.
diff --git a/lib/Analysis/IVUsers.cpp b/lib/Analysis/IVUsers.cpp
index 0a6682a..f705181 100644
--- a/lib/Analysis/IVUsers.cpp
+++ b/lib/Analysis/IVUsers.cpp
@@ -273,9 +273,11 @@ void IVUsers::print(raw_ostream &OS, const Module *M) const {
   }
 }
 
+#ifndef NDEBUG
 void IVUsers::dump() const {
   print(dbgs());
 }
+#endif
 
 void IVUsers::releaseMemory() {
   Processed.clear();
diff --git a/lib/Analysis/InlineCost.cpp b/lib/Analysis/InlineCost.cpp
index bc1ecd2..12be7fd 100644
--- a/lib/Analysis/InlineCost.cpp
+++ b/lib/Analysis/InlineCost.cpp
@@ -974,6 +974,7 @@ bool CallAnalyzer::analyzeCall(CallSite CS) {
   return AlwaysInline || Cost < Threshold;
 }
 
+#ifndef NDEBUG
 /// \brief Dump stats about this call's analysis.
 void CallAnalyzer::dump() {
 #define DEBUG_PRINT_STAT(x) llvm::dbgs() << "      " #x ": " << x << "\n"
@@ -987,6 +988,7 @@ void CallAnalyzer::dump() {
   DEBUG_PRINT_STAT(SROACostSavingsLost);
 #undef DEBUG_PRINT_STAT
 }
+#endif
 
 InlineCost InlineCostAnalyzer::getInlineCost(CallSite CS, int Threshold) {
   return getInlineCost(CS, CS.getCalledFunction(), Threshold);
diff --git a/lib/Analysis/LazyValueInfo.cpp b/lib/Analysis/LazyValueInfo.cpp
index 9140786..ec618fa 100644
--- a/lib/Analysis/LazyValueInfo.cpp
+++ b/lib/Analysis/LazyValueInfo.cpp
@@ -470,8 +470,10 @@ bool LazyValueInfoCache::hasBlockValue(Value *Val, BasicBlock *BB) {
     return true;
 
   LVIValueHandle ValHandle(Val, this);
-  if (!ValueCache.count(ValHandle)) return false;
-  return ValueCache[ValHandle].count(BB);
+  std::map<LVIValueHandle, ValueCacheEntryTy>::iterator I =
+    ValueCache.find(ValHandle);
+  if (I == ValueCache.end()) return false;
+  return I->second.count(BB);
 }
 
 LVILatticeVal LazyValueInfoCache::getBlockValue(Value *Val, BasicBlock *BB) {
@@ -845,9 +847,12 @@ static bool getEdgeValueLocal(Value *Val, BasicBlock *BBFrom,
     for (SwitchInst::CaseIt i = SI->case_begin(), e = SI->case_end();
          i != e; ++i) {
       ConstantRange EdgeVal(i.getCaseValue()->getValue());
-      if (DefaultCase)
-        EdgesVals = EdgesVals.difference(EdgeVal);
-      else if (i.getCaseSuccessor() == BBTo)
+      if (DefaultCase) {
+        // It is possible that the default destination is the destination of
+        // some cases. There is no need to perform difference for those cases.
+        if (i.getCaseSuccessor() != BBTo)
+          EdgesVals = EdgesVals.difference(EdgeVal);
+      } else if (i.getCaseSuccessor() == BBTo)
         EdgesVals = EdgesVals.unionWith(EdgeVal);
     }
     Result = LVILatticeVal::getRange(EdgesVals);
diff --git a/lib/Analysis/LoopInfo.cpp b/lib/Analysis/LoopInfo.cpp
index 20c33a3..4a18104 100644
--- a/lib/Analysis/LoopInfo.cpp
+++ b/lib/Analysis/LoopInfo.cpp
@@ -306,9 +306,11 @@ BasicBlock *Loop::getUniqueExitBlock() const {
   return 0;
 }
 
+#ifndef NDEBUG
 void Loop::dump() const {
   print(dbgs());
 }
+#endif
 
 //===----------------------------------------------------------------------===//
 // UnloopUpdater implementation
@@ -429,8 +431,8 @@ void UnloopUpdater::updateSubloopParents() {
     Unloop->removeChildLoop(llvm::prior(Unloop->end()));
 
     assert(SubloopParents.count(Subloop) && "DFS failed to visit subloop");
-    if (SubloopParents[Subloop])
-      SubloopParents[Subloop]->addChildLoop(Subloop);
+    if (Loop *Parent = SubloopParents[Subloop])
+      Parent->addChildLoop(Subloop);
     else
       LI->addTopLevelLoop(Subloop);
   }
@@ -456,9 +458,8 @@ Loop *UnloopUpdater::getNearestLoop(BasicBlock *BB, Loop *BBLoop) {
       assert(Subloop && "subloop is not an ancestor of the original loop");
     }
     // Get the current nearest parent of the Subloop exits, initially Unloop.
-    if (!SubloopParents.count(Subloop))
-      SubloopParents[Subloop] = Unloop;
-    NearLoop = SubloopParents[Subloop];
+    NearLoop =
+      SubloopParents.insert(std::make_pair(Subloop, Unloop)).first->second;
   }
 
   succ_iterator I = succ_begin(BB), E = succ_end(BB);
diff --git a/lib/Analysis/MemoryBuiltins.cpp b/lib/Analysis/MemoryBuiltins.cpp
index b986b32..5b2313e 100644
--- a/lib/Analysis/MemoryBuiltins.cpp
+++ b/lib/Analysis/MemoryBuiltins.cpp
@@ -26,6 +26,7 @@
 #include "llvm/Support/MathExtras.h"
 #include "llvm/Support/raw_ostream.h"
 #include "llvm/Target/TargetData.h"
+#include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Transforms/Utils/Local.h"
 using namespace llvm;
 
@@ -39,7 +40,7 @@ enum AllocType {
 };
 
 struct AllocFnsTy {
-  const char *Name;
+  LibFunc::Func Func;
   AllocType AllocTy;
   unsigned char NumParams;
   // First and Second size parameters (or -1 if unused)
@@ -49,22 +50,22 @@ struct AllocFnsTy {
 // FIXME: certain users need more information. E.g., SimplifyLibCalls needs to
 // know which functions are nounwind, noalias, nocapture parameters, etc.
 static const AllocFnsTy AllocationFnData[] = {
-  {"malloc",              MallocLike,  1, 0,  -1},
-  {"valloc",              MallocLike,  1, 0,  -1},
-  {"_Znwj",               MallocLike,  1, 0,  -1}, // new(unsigned int)
-  {"_ZnwjRKSt9nothrow_t", MallocLike,  2, 0,  -1}, // new(unsigned int, nothrow)
-  {"_Znwm",               MallocLike,  1, 0,  -1}, // new(unsigned long)
-  {"_ZnwmRKSt9nothrow_t", MallocLike,  2, 0,  -1}, // new(unsigned long, nothrow)
-  {"_Znaj",               MallocLike,  1, 0,  -1}, // new[](unsigned int)
-  {"_ZnajRKSt9nothrow_t", MallocLike,  2, 0,  -1}, // new[](unsigned int, nothrow)
-  {"_Znam",               MallocLike,  1, 0,  -1}, // new[](unsigned long)
-  {"_ZnamRKSt9nothrow_t", MallocLike,  2, 0,  -1}, // new[](unsigned long, nothrow)
-  {"posix_memalign",      MallocLike,  3, 2,  -1},
-  {"calloc",              CallocLike,  2, 0,  1},
-  {"realloc",             ReallocLike, 2, 1,  -1},
-  {"reallocf",            ReallocLike, 2, 1,  -1},
-  {"strdup",              StrDupLike,  1, -1, -1},
-  {"strndup",             StrDupLike,  2, 1,  -1}
+  {LibFunc::malloc,              MallocLike,  1, 0,  -1},
+  {LibFunc::valloc,              MallocLike,  1, 0,  -1},
+  {LibFunc::Znwj,                MallocLike,  1, 0,  -1}, // new(unsigned int)
+  {LibFunc::ZnwjRKSt9nothrow_t,  MallocLike,  2, 0,  -1}, // new(unsigned int, nothrow)
+  {LibFunc::Znwm,                MallocLike,  1, 0,  -1}, // new(unsigned long)
+  {LibFunc::ZnwmRKSt9nothrow_t,  MallocLike,  2, 0,  -1}, // new(unsigned long, nothrow)
+  {LibFunc::Znaj,                MallocLike,  1, 0,  -1}, // new[](unsigned int)
+  {LibFunc::ZnajRKSt9nothrow_t,  MallocLike,  2, 0,  -1}, // new[](unsigned int, nothrow)
+  {LibFunc::Znam,                MallocLike,  1, 0,  -1}, // new[](unsigned long)
+  {LibFunc::ZnamRKSt9nothrow_t,  MallocLike,  2, 0,  -1}, // new[](unsigned long, nothrow)
+  {LibFunc::posix_memalign,      MallocLike,  3, 2,  -1},
+  {LibFunc::calloc,              CallocLike,  2, 0,   1},
+  {LibFunc::realloc,             ReallocLike, 2, 1,  -1},
+  {LibFunc::reallocf,            ReallocLike, 2, 1,  -1},
+  {LibFunc::strdup,              StrDupLike,  1, -1, -1},
+  {LibFunc::strndup,             StrDupLike,  2, 1,  -1}
 };
 
 
@@ -85,15 +86,22 @@ static Function *getCalledFunction(const Value *V, bool LookThroughBitCast) {
 /// \brief Returns the allocation data for the given value if it is a call to a
 /// known allocation function, and NULL otherwise.
 static const AllocFnsTy *getAllocationData(const Value *V, AllocType AllocTy,
+                                           const TargetLibraryInfo *TLI,
                                            bool LookThroughBitCast = false) {
   Function *Callee = getCalledFunction(V, LookThroughBitCast);
   if (!Callee)
     return 0;
 
+  // Make sure that the function is available.
+  StringRef FnName = Callee->getName();
+  LibFunc::Func TLIFn;
+  if (!TLI || !TLI->getLibFunc(FnName, TLIFn) || !TLI->has(TLIFn))
+    return 0;
+
   unsigned i = 0;
   bool found = false;
   for ( ; i < array_lengthof(AllocationFnData); ++i) {
-    if (Callee->getName() == AllocationFnData[i].Name) {
+    if (AllocationFnData[i].Func == TLIFn) {
       found = true;
       break;
     }
@@ -106,7 +114,6 @@ static const AllocFnsTy *getAllocationData(const Value *V, AllocType AllocTy,
     return 0;
 
   // Check function prototype.
-  // FIXME: Check the nobuiltin metadata?? (PR5130)
   int FstParam = FnData->FstParam;
   int SndParam = FnData->SndParam;
   FunctionType *FTy = Callee->getFunctionType();
@@ -132,57 +139,65 @@ static bool hasNoAliasAttr(const Value *V, bool LookThroughBitCast) {
 /// \brief Tests if a value is a call or invoke to a library function that
 /// allocates or reallocates memory (either malloc, calloc, realloc, or strdup
 /// like).
-bool llvm::isAllocationFn(const Value *V, bool LookThroughBitCast) {
-  return getAllocationData(V, AnyAlloc, LookThroughBitCast);
+bool llvm::isAllocationFn(const Value *V, const TargetLibraryInfo *TLI,
+                          bool LookThroughBitCast) {
+  return getAllocationData(V, AnyAlloc, TLI, LookThroughBitCast);
 }
 
 /// \brief Tests if a value is a call or invoke to a function that returns a
 /// NoAlias pointer (including malloc/calloc/realloc/strdup-like functions).
-bool llvm::isNoAliasFn(const Value *V, bool LookThroughBitCast) {
+bool llvm::isNoAliasFn(const Value *V, const TargetLibraryInfo *TLI,
+                       bool LookThroughBitCast) {
   // it's safe to consider realloc as noalias since accessing the original
   // pointer is undefined behavior
-  return isAllocationFn(V, LookThroughBitCast) ||
+  return isAllocationFn(V, TLI, LookThroughBitCast) ||
          hasNoAliasAttr(V, LookThroughBitCast);
 }
 
 /// \brief Tests if a value is a call or invoke to a library function that
 /// allocates uninitialized memory (such as malloc).
-bool llvm::isMallocLikeFn(const Value *V, bool LookThroughBitCast) {
-  return getAllocationData(V, MallocLike, LookThroughBitCast);
+bool llvm::isMallocLikeFn(const Value *V, const TargetLibraryInfo *TLI,
+                          bool LookThroughBitCast) {
+  return getAllocationData(V, MallocLike, TLI, LookThroughBitCast);
 }
 
 /// \brief Tests if a value is a call or invoke to a library function that
 /// allocates zero-filled memory (such as calloc).
-bool llvm::isCallocLikeFn(const Value *V, bool LookThroughBitCast) {
-  return getAllocationData(V, CallocLike, LookThroughBitCast);
+bool llvm::isCallocLikeFn(const Value *V, const TargetLibraryInfo *TLI,
+                          bool LookThroughBitCast) {
+  return getAllocationData(V, CallocLike, TLI, LookThroughBitCast);
 }
 
 /// \brief Tests if a value is a call or invoke to a library function that
 /// allocates memory (either malloc, calloc, or strdup like).
-bool llvm::isAllocLikeFn(const Value *V, bool LookThroughBitCast) {
-  return getAllocationData(V, AllocLike, LookThroughBitCast);
+bool llvm::isAllocLikeFn(const Value *V, const TargetLibraryInfo *TLI,
+                         bool LookThroughBitCast) {
+  return getAllocationData(V, AllocLike, TLI, LookThroughBitCast);
 }
 
 /// \brief Tests if a value is a call or invoke to a library function that
 /// reallocates memory (such as realloc).
-bool llvm::isReallocLikeFn(const Value *V, bool LookThroughBitCast) {
-  return getAllocationData(V, ReallocLike, LookThroughBitCast);
+bool llvm::isReallocLikeFn(const Value *V, const TargetLibraryInfo *TLI,
+                           bool LookThroughBitCast) {
+  return getAllocationData(V, ReallocLike, TLI, LookThroughBitCast);
 }
 
 /// extractMallocCall - Returns the corresponding CallInst if the instruction
 /// is a malloc call.  Since CallInst::CreateMalloc() only creates calls, we
 /// ignore InvokeInst here.
-const CallInst *llvm::extractMallocCall(const Value *I) {
-  return isMallocLikeFn(I) ? dyn_cast<CallInst>(I) : 0;
+const CallInst *llvm::extractMallocCall(const Value *I,
+                                        const TargetLibraryInfo *TLI) {
+  return isMallocLikeFn(I, TLI) ? dyn_cast<CallInst>(I) : 0;
 }
 
 static Value *computeArraySize(const CallInst *CI, const TargetData *TD,
+                               const TargetLibraryInfo *TLI,
                                bool LookThroughSExt = false) {
   if (!CI)
     return NULL;
 
   // The size of the malloc's result type must be known to determine array size.
-  Type *T = getMallocAllocatedType(CI);
+  Type *T = getMallocAllocatedType(CI, TLI);
   if (!T || !T->isSized() || !TD)
     return NULL;
 
@@ -204,9 +219,11 @@ static Value *computeArraySize(const CallInst *CI, const TargetData *TD,
 /// isArrayMalloc - Returns the corresponding CallInst if the instruction 
 /// is a call to malloc whose array size can be determined and the array size
 /// is not constant 1.  Otherwise, return NULL.
-const CallInst *llvm::isArrayMalloc(const Value *I, const TargetData *TD) {
-  const CallInst *CI = extractMallocCall(I);
-  Value *ArraySize = computeArraySize(CI, TD);
+const CallInst *llvm::isArrayMalloc(const Value *I,
+                                    const TargetData *TD,
+                                    const TargetLibraryInfo *TLI) {
+  const CallInst *CI = extractMallocCall(I, TLI);
+  Value *ArraySize = computeArraySize(CI, TD, TLI);
 
   if (ArraySize &&
       ArraySize != ConstantInt::get(CI->getArgOperand(0)->getType(), 1))
@@ -221,8 +238,9 @@ const CallInst *llvm::isArrayMalloc(const Value *I, const TargetData *TD) {
 ///   0: PointerType is the calls' return type.
 ///   1: PointerType is the bitcast's result type.
 ///  >1: Unique PointerType cannot be determined, return NULL.
-PointerType *llvm::getMallocType(const CallInst *CI) {
-  assert(isMallocLikeFn(CI) && "getMallocType and not malloc call");
+PointerType *llvm::getMallocType(const CallInst *CI,
+                                 const TargetLibraryInfo *TLI) {
+  assert(isMallocLikeFn(CI, TLI) && "getMallocType and not malloc call");
   
   PointerType *MallocType = NULL;
   unsigned NumOfBitCastUses = 0;
@@ -252,8 +270,9 @@ PointerType *llvm::getMallocType(const CallInst *CI) {
 ///   0: PointerType is the malloc calls' return type.
 ///   1: PointerType is the bitcast's result type.
 ///  >1: Unique PointerType cannot be determined, return NULL.
-Type *llvm::getMallocAllocatedType(const CallInst *CI) {
-  PointerType *PT = getMallocType(CI);
+Type *llvm::getMallocAllocatedType(const CallInst *CI,
+                                   const TargetLibraryInfo *TLI) {
+  PointerType *PT = getMallocType(CI, TLI);
   return PT ? PT->getElementType() : NULL;
 }
 
@@ -263,21 +282,23 @@ Type *llvm::getMallocAllocatedType(const CallInst *CI) {
 /// constant 1.  Otherwise, return NULL for mallocs whose array size cannot be
 /// determined.
 Value *llvm::getMallocArraySize(CallInst *CI, const TargetData *TD,
+                                const TargetLibraryInfo *TLI,
                                 bool LookThroughSExt) {
-  assert(isMallocLikeFn(CI) && "getMallocArraySize and not malloc call");
-  return computeArraySize(CI, TD, LookThroughSExt);
+  assert(isMallocLikeFn(CI, TLI) && "getMallocArraySize and not malloc call");
+  return computeArraySize(CI, TD, TLI, LookThroughSExt);
 }
 
 
 /// extractCallocCall - Returns the corresponding CallInst if the instruction
 /// is a calloc call.
-const CallInst *llvm::extractCallocCall(const Value *I) {
-  return isCallocLikeFn(I) ? cast<CallInst>(I) : 0;
+const CallInst *llvm::extractCallocCall(const Value *I,
+                                        const TargetLibraryInfo *TLI) {
+  return isCallocLikeFn(I, TLI) ? cast<CallInst>(I) : 0;
 }
 
 
 /// isFreeCall - Returns non-null if the value is a call to the builtin free()
-const CallInst *llvm::isFreeCall(const Value *I) {
+const CallInst *llvm::isFreeCall(const Value *I, const TargetLibraryInfo *TLI) {
   const CallInst *CI = dyn_cast<CallInst>(I);
   if (!CI)
     return 0;
@@ -285,9 +306,14 @@ const CallInst *llvm::isFreeCall(const Value *I) {
   if (Callee == 0 || !Callee->isDeclaration())
     return 0;
 
-  if (Callee->getName() != "free" &&
-      Callee->getName() != "_ZdlPv" && // operator delete(void*)
-      Callee->getName() != "_ZdaPv")   // operator delete[](void*)
+  StringRef FnName = Callee->getName();
+  LibFunc::Func TLIFn;
+  if (!TLI || !TLI->getLibFunc(FnName, TLIFn) || !TLI->has(TLIFn))
+    return 0;
+
+  if (TLIFn != LibFunc::free &&
+      TLIFn != LibFunc::ZdlPv && // operator delete(void*)
+      TLIFn != LibFunc::ZdaPv)   // operator delete[](void*)
     return 0;
 
   // Check free prototype.
@@ -316,11 +342,11 @@ const CallInst *llvm::isFreeCall(const Value *I) {
 /// If RoundToAlign is true, then Size is rounded up to the aligment of allocas,
 /// byval arguments, and global variables.
 bool llvm::getObjectSize(const Value *Ptr, uint64_t &Size, const TargetData *TD,
-                         bool RoundToAlign) {
+                         const TargetLibraryInfo *TLI, bool RoundToAlign) {
   if (!TD)
     return false;
 
-  ObjectSizeOffsetVisitor Visitor(TD, Ptr->getContext(), RoundToAlign);
+  ObjectSizeOffsetVisitor Visitor(TD, TLI, Ptr->getContext(), RoundToAlign);
   SizeOffsetType Data = Visitor.compute(const_cast<Value*>(Ptr));
   if (!Visitor.bothKnown(Data))
     return false;
@@ -348,9 +374,10 @@ APInt ObjectSizeOffsetVisitor::align(APInt Size, uint64_t Align) {
 }
 
 ObjectSizeOffsetVisitor::ObjectSizeOffsetVisitor(const TargetData *TD,
+                                                 const TargetLibraryInfo *TLI,
                                                  LLVMContext &Context,
                                                  bool RoundToAlign)
-: TD(TD), RoundToAlign(RoundToAlign) {
+: TD(TD), TLI(TLI), RoundToAlign(RoundToAlign) {
   IntegerType *IntTy = TD->getIntPtrType(Context);
   IntTyBits = IntTy->getBitWidth();
   Zero = APInt::getNullValue(IntTyBits);
@@ -416,7 +443,8 @@ SizeOffsetType ObjectSizeOffsetVisitor::visitArgument(Argument &A) {
 }
 
 SizeOffsetType ObjectSizeOffsetVisitor::visitCallSite(CallSite CS) {
-  const AllocFnsTy *FnData = getAllocationData(CS.getInstruction(), AnyAlloc);
+  const AllocFnsTy *FnData = getAllocationData(CS.getInstruction(), AnyAlloc,
+                                               TLI);
   if (!FnData)
     return unknown();
 
@@ -532,8 +560,9 @@ SizeOffsetType ObjectSizeOffsetVisitor::visitInstruction(Instruction &I) {
 
 
 ObjectSizeOffsetEvaluator::ObjectSizeOffsetEvaluator(const TargetData *TD,
+                                                   const TargetLibraryInfo *TLI,
                                                      LLVMContext &Context)
-: TD(TD), Context(Context), Builder(Context, TargetFolder(TD)) {
+: TD(TD), TLI(TLI), Context(Context), Builder(Context, TargetFolder(TD)) {
   IntTy = TD->getIntPtrType(Context);
   Zero = ConstantInt::get(IntTy, 0);
 }
@@ -558,7 +587,7 @@ SizeOffsetEvalType ObjectSizeOffsetEvaluator::compute(Value *V) {
 }
 
 SizeOffsetEvalType ObjectSizeOffsetEvaluator::compute_(Value *V) {
-  ObjectSizeOffsetVisitor Visitor(TD, Context);
+  ObjectSizeOffsetVisitor Visitor(TD, TLI, Context);
   SizeOffsetType Const = Visitor.compute(V);
   if (Visitor.bothKnown(Const))
     return std::make_pair(ConstantInt::get(Context, Const.first),
@@ -621,7 +650,8 @@ SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitAllocaInst(AllocaInst &I) {
 }
 
 SizeOffsetEvalType ObjectSizeOffsetEvaluator::visitCallSite(CallSite CS) {
-  const AllocFnsTy *FnData = getAllocationData(CS.getInstruction(), AnyAlloc);
+  const AllocFnsTy *FnData = getAllocationData(CS.getInstruction(), AnyAlloc,
+                                               TLI);
   if (!FnData)
     return unknown();
 
diff --git a/lib/Analysis/MemoryDependenceAnalysis.cpp b/lib/Analysis/MemoryDependenceAnalysis.cpp
index 059e574..5736c35 100644
--- a/lib/Analysis/MemoryDependenceAnalysis.cpp
+++ b/lib/Analysis/MemoryDependenceAnalysis.cpp
@@ -148,7 +148,7 @@ AliasAnalysis::ModRefResult GetLocation(const Instruction *Inst,
     return AliasAnalysis::ModRef;
   }
 
-  if (const CallInst *CI = isFreeCall(Inst)) {
+  if (const CallInst *CI = isFreeCall(Inst, AA->getTargetLibraryInfo())) {
     // calls to free() deallocate the entire structure
     Loc = AliasAnalysis::Location(CI->getArgOperand(0));
     return AliasAnalysis::Mod;
@@ -479,12 +479,20 @@ getPointerDependencyFrom(const AliasAnalysis::Location &MemLoc, bool isLoad,
     // a subsequent bitcast of the malloc call result.  There can be stores to
     // the malloced memory between the malloc call and its bitcast uses, and we
     // need to continue scanning until the malloc call.
-    if (isa<AllocaInst>(Inst) || isNoAliasFn(Inst)) {
+    const TargetLibraryInfo *TLI = AA->getTargetLibraryInfo();
+    if (isa<AllocaInst>(Inst) || isNoAliasFn(Inst, TLI)) {
       const Value *AccessPtr = GetUnderlyingObject(MemLoc.Ptr, TD);
       
       if (AccessPtr == Inst || AA->isMustAlias(Inst, AccessPtr))
         return MemDepResult::getDef(Inst);
-      continue;
+      // Be conservative if the accessed pointer may alias the allocation.
+      if (AA->alias(Inst, AccessPtr) != AliasAnalysis::NoAlias)
+        return MemDepResult::getClobber(Inst);
+      // If the allocation is not aliased and does not read memory (like
+      // strdup), it is safe to ignore.
+      if (isa<AllocaInst>(Inst) ||
+          isMallocLikeFn(Inst, TLI) || isCallocLikeFn(Inst, TLI))
+        continue;
     }
 
     // See if this instruction (e.g. a call or vaarg) mod/ref's the pointer.
diff --git a/lib/Analysis/PHITransAddr.cpp b/lib/Analysis/PHITransAddr.cpp
index 38cb1c9..d6a17ca 100644
--- a/lib/Analysis/PHITransAddr.cpp
+++ b/lib/Analysis/PHITransAddr.cpp
@@ -41,6 +41,7 @@ static bool CanPHITrans(Instruction *Inst) {
   return false;
 }
 
+#ifndef NDEBUG
 void PHITransAddr::dump() const {
   if (Addr == 0) {
     dbgs() << "PHITransAddr: null\n";
@@ -50,6 +51,7 @@ void PHITransAddr::dump() const {
   for (unsigned i = 0, e = InstInputs.size(); i != e; ++i)
     dbgs() << "  Input #" << i << " is " << *InstInputs[i] << "\n";
 }
+#endif
 
 
 static bool VerifySubExpr(Value *Expr,
diff --git a/lib/Analysis/ProfileDataLoader.cpp b/lib/Analysis/ProfileDataLoader.cpp
new file mode 100644
index 0000000..69286ef
--- /dev/null
+++ b/lib/Analysis/ProfileDataLoader.cpp
@@ -0,0 +1,162 @@
+//===- ProfileDataLoader.cpp - Load profile information from disk ---------===//
+//
+//                      The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// The ProfileDataLoader class is used to load raw profiling data from the dump
+// file.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/OwningPtr.h"
+#include "llvm/Module.h"
+#include "llvm/InstrTypes.h"
+#include "llvm/Analysis/ProfileDataLoader.h"
+#include "llvm/Analysis/ProfileDataTypes.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/system_error.h"
+#include <cstdio>
+#include <cstdlib>
+using namespace llvm;
+
+raw_ostream &llvm::operator<<(raw_ostream &O, std::pair<const BasicBlock *,
+                                                        const BasicBlock *> E) {
+  O << "(";
+
+  if (E.first)
+    O << E.first->getName();
+  else
+    O << "0";
+
+  O << ",";
+
+  if (E.second)
+    O << E.second->getName();
+  else
+    O << "0";
+
+  return O << ")";
+}
+
+/// AddCounts - Add 'A' and 'B', accounting for the fact that the value of one
+/// (or both) may not be defined.
+static unsigned AddCounts(unsigned A, unsigned B) {
+  // If either value is undefined, use the other.
+  // Undefined + undefined = undefined.
+  if (A == ProfileDataLoader::Uncounted) return B;
+  if (B == ProfileDataLoader::Uncounted) return A;
+
+  // Saturate to the maximum storable value.  This could change taken/nottaken
+  // ratios, but is presumably better than wrapping and thus potentially
+  // inverting ratios.
+  uint64_t tmp = (uint64_t)A + (uint64_t)B;
+  if (tmp > (uint64_t)ProfileDataLoader::MaxCount)
+    tmp = ProfileDataLoader::MaxCount;
+  return (unsigned)tmp;
+}
+
+/// ReadProfilingData - Load 'NumEntries' items of type 'T' from file 'F'
+template <typename T>
+static void ReadProfilingData(const char *ToolName, FILE *F,
+                              T *Data, size_t NumEntries) {
+  // Read in the block of data...
+  if (fread(Data, sizeof(T), NumEntries, F) != NumEntries)
+    report_fatal_error(Twine(ToolName) + ": Profiling data truncated");
+}
+
+/// ReadProfilingNumEntries - Read how many entries are in this profiling data
+/// packet.
+static unsigned ReadProfilingNumEntries(const char *ToolName, FILE *F,
+                                        bool ShouldByteSwap) {
+  unsigned Entry;
+  ReadProfilingData<unsigned>(ToolName, F, &Entry, 1);
+  return ShouldByteSwap ? ByteSwap_32(Entry) : Entry;
+}
+
+/// ReadProfilingBlock - Read the number of entries in the next profiling data
+/// packet and then accumulate the entries into 'Data'.
+static void ReadProfilingBlock(const char *ToolName, FILE *F,
+                               bool ShouldByteSwap,
+                               SmallVector<unsigned, 32> &Data) {
+  // Read the number of entries...
+  unsigned NumEntries = ReadProfilingNumEntries(ToolName, F, ShouldByteSwap);
+
+  // Read in the data.
+  SmallVector<unsigned, 8> TempSpace(NumEntries);
+  ReadProfilingData<unsigned>(ToolName, F, TempSpace.data(), NumEntries);
+
+  // Make sure we have enough space ...
+  if (Data.size() < NumEntries)
+    Data.resize(NumEntries, ProfileDataLoader::Uncounted);
+
+  // Accumulate the data we just read into the existing data.
+  for (unsigned i = 0; i < NumEntries; ++i) {
+    unsigned Entry = ShouldByteSwap ? ByteSwap_32(TempSpace[i]) : TempSpace[i];
+    Data[i] = AddCounts(Entry, Data[i]);
+  }
+}
+
+/// ReadProfilingArgBlock - Read the command line arguments that the progam was
+/// run with when the current profiling data packet(s) were generated.
+static void ReadProfilingArgBlock(const char *ToolName, FILE *F,
+                                  bool ShouldByteSwap,
+                                  SmallVector<std::string, 1> &CommandLines) {
+  // Read the number of bytes ...
+  unsigned ArgLength = ReadProfilingNumEntries(ToolName, F, ShouldByteSwap);
+
+  // Read in the arguments (if there are any to read).  Round up the length to
+  // the nearest 4-byte multiple.
+  SmallVector<char, 8> Args(ArgLength+4);
+  if (ArgLength)
+    ReadProfilingData<char>(ToolName, F, Args.data(), (ArgLength+3) & ~3);
+
+  // Store the arguments.
+  CommandLines.push_back(std::string(&Args[0], &Args[ArgLength]));
+}
+
+const unsigned ProfileDataLoader::Uncounted = ~0U;
+const unsigned ProfileDataLoader::MaxCount = ~0U - 1U;
+
+/// ProfileDataLoader ctor - Read the specified profiling data file, reporting
+/// a fatal error if the file is invalid or broken.
+ProfileDataLoader::ProfileDataLoader(const char *ToolName,
+                                     const std::string &Filename)
+  : Filename(Filename) {
+  FILE *F = fopen(Filename.c_str(), "rb");
+  if (F == 0)
+    report_fatal_error(Twine(ToolName) + ": Error opening '" +
+                       Filename + "': ");
+
+  // Keep reading packets until we run out of them.
+  unsigned PacketType;
+  while (fread(&PacketType, sizeof(unsigned), 1, F) == 1) {
+    // If the low eight bits of the packet are zero, we must be dealing with an
+    // endianness mismatch.  Byteswap all words read from the profiling
+    // information.  This can happen when the compiler host and target have
+    // different endianness.
+    bool ShouldByteSwap = (char)PacketType == 0;
+    PacketType = ShouldByteSwap ? ByteSwap_32(PacketType) : PacketType;
+
+    switch (PacketType) {
+      case ArgumentInfo:
+        ReadProfilingArgBlock(ToolName, F, ShouldByteSwap, CommandLines);
+        break;
+
+      case EdgeInfo:
+        ReadProfilingBlock(ToolName, F, ShouldByteSwap, EdgeCounts);
+        break;
+
+      default:
+        report_fatal_error(std::string(ToolName)
+                           + ": Unknown profiling packet type");
+        break;
+    }
+  }
+
+  fclose(F);
+}
diff --git a/lib/Analysis/ProfileDataLoaderPass.cpp b/lib/Analysis/ProfileDataLoaderPass.cpp
new file mode 100644
index 0000000..c43cff0
--- /dev/null
+++ b/lib/Analysis/ProfileDataLoaderPass.cpp
@@ -0,0 +1,188 @@
+//===- ProfileDataLoaderPass.cpp - Set branch weight metadata from prof ---===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This pass loads profiling data from a dump file and sets branch weight
+// metadata.
+//
+// TODO: Replace all "profile-metadata-loader" strings with "profile-loader"
+// once ProfileInfo etc. has been removed.
+//
+//===----------------------------------------------------------------------===//
+#define DEBUG_TYPE "profile-metadata-loader"
+#include "llvm/ADT/ArrayRef.h"
+#include "llvm/BasicBlock.h"
+#include "llvm/InstrTypes.h"
+#include "llvm/Module.h"
+#include "llvm/LLVMContext.h"
+#include "llvm/MDBuilder.h"
+#include "llvm/Metadata.h"
+#include "llvm/Pass.h"
+#include "llvm/Analysis/Passes.h"
+#include "llvm/Analysis/ProfileDataLoader.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/CFG.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Support/Format.h"
+#include "llvm/ADT/Statistic.h"
+using namespace llvm;
+
+STATISTIC(NumEdgesRead, "The # of edges read.");
+STATISTIC(NumTermsAnnotated, "The # of terminator instructions annotated.");
+
+static cl::opt<std::string>
+ProfileMetadataFilename("profile-file", cl::init("llvmprof.out"),
+                  cl::value_desc("filename"),
+                  cl::desc("Profile file loaded by -profile-metadata-loader"));
+
+namespace {
+  /// This pass loads profiling data from a dump file and sets branch weight
+  /// metadata.
+  class ProfileMetadataLoaderPass : public ModulePass {
+    std::string Filename;
+  public:
+    static char ID; // Class identification, replacement for typeinfo
+    explicit ProfileMetadataLoaderPass(const std::string &filename = "")
+        : ModulePass(ID), Filename(filename) {
+      initializeProfileMetadataLoaderPassPass(*PassRegistry::getPassRegistry());
+      if (filename.empty()) Filename = ProfileMetadataFilename;
+    }
+
+    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+      AU.setPreservesAll();
+    }
+
+    virtual const char *getPassName() const {
+      return "Profile loader";
+    }
+
+    virtual void readEdge(unsigned, ProfileData&, ProfileData::Edge,
+                          ArrayRef<unsigned>);
+    virtual unsigned matchEdges(Module&, ProfileData&, ArrayRef<unsigned>);
+    virtual void setBranchWeightMetadata(Module&, ProfileData&);
+
+    virtual bool runOnModule(Module &M);
+  };
+}  // End of anonymous namespace
+
+char ProfileMetadataLoaderPass::ID = 0;
+INITIALIZE_PASS_BEGIN(ProfileMetadataLoaderPass, "profile-metadata-loader",
+              "Load profile information from llvmprof.out", false, true)
+INITIALIZE_PASS_END(ProfileMetadataLoaderPass, "profile-metadata-loader",
+              "Load profile information from llvmprof.out", false, true)
+
+char &llvm::ProfileMetadataLoaderPassID = ProfileMetadataLoaderPass::ID;
+
+/// createProfileMetadataLoaderPass - This function returns a Pass that loads
+/// the profiling information for the module from the specified filename,
+/// making it available to the optimizers.
+ModulePass *llvm::createProfileMetadataLoaderPass() { 
+    return new ProfileMetadataLoaderPass();
+}
+ModulePass *llvm::createProfileMetadataLoaderPass(const std::string &Filename) {
+  return new ProfileMetadataLoaderPass(Filename);
+}
+
+/// readEdge - Take the value from a profile counter and assign it to an edge.
+void ProfileMetadataLoaderPass::readEdge(unsigned ReadCount,
+                                         ProfileData &PB, ProfileData::Edge e,
+                                         ArrayRef<unsigned> Counters) {
+  if (ReadCount >= Counters.size()) return;
+
+  unsigned weight = Counters[ReadCount];
+  assert(weight != ProfileDataLoader::Uncounted);
+  PB.addEdgeWeight(e, weight);
+
+  DEBUG(dbgs() << "-- Read Edge Counter for " << e
+               << " (# "<< (ReadCount) << "): "
+               << PB.getEdgeWeight(e) << "\n");
+}
+
+/// matchEdges - Link every profile counter with an edge.
+unsigned ProfileMetadataLoaderPass::matchEdges(Module &M, ProfileData &PB,
+                                               ArrayRef<unsigned> Counters) {
+  if (Counters.size() == 0) return 0;
+
+  unsigned ReadCount = 0;
+
+  for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
+    if (F->isDeclaration()) continue;
+    DEBUG(dbgs() << "Loading edges in '" << F->getName() << "'\n");
+    readEdge(ReadCount++, PB, PB.getEdge(0, &F->getEntryBlock()), Counters);
+    for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
+      TerminatorInst *TI = BB->getTerminator();
+      for (unsigned s = 0, e = TI->getNumSuccessors(); s != e; ++s) {
+        readEdge(ReadCount++, PB, PB.getEdge(BB,TI->getSuccessor(s)),
+                 Counters);
+      }
+    }
+  }
+
+  return ReadCount;
+}
+
+/// setBranchWeightMetadata - Translate the counter values associated with each
+/// edge into branch weights for each conditional branch (a branch with 2 or
+/// more desinations).
+void ProfileMetadataLoaderPass::setBranchWeightMetadata(Module &M,
+                                                        ProfileData &PB) {
+  for (Module::iterator F = M.begin(), E = M.end(); F != E; ++F) {
+    if (F->isDeclaration()) continue;
+    DEBUG(dbgs() << "Setting branch metadata in '" << F->getName() << "'\n");
+
+    for (Function::iterator BB = F->begin(), E = F->end(); BB != E; ++BB) {
+      TerminatorInst *TI = BB->getTerminator();
+      unsigned NumSuccessors = TI->getNumSuccessors();
+
+      // If there is only one successor then we can not set a branch
+      // probability as the target is certain.
+      if (NumSuccessors < 2) continue;
+
+      // Load the weights of all edges leading from this terminator.
+      DEBUG(dbgs() << "-- Terminator with " << NumSuccessors
+                   << " successors:\n");
+      SmallVector<uint32_t, 4> Weights(NumSuccessors);
+      for (unsigned s = 0 ; s < NumSuccessors ; ++s) {
+          ProfileData::Edge edge = PB.getEdge(BB, TI->getSuccessor(s));
+          Weights[s] = (uint32_t)PB.getEdgeWeight(edge);
+          DEBUG(dbgs() << "---- Edge '" << edge << "' has weight "
+                       << Weights[s] << "\n");
+      }
+
+      // Set branch weight metadata.  This will set branch probabilities of
+      // 100%/0% if that is true of the dynamic execution.
+      // BranchProbabilityInfo can account for this when it loads this metadata
+      // (it gives the unexectuted branch a weight of 1 for the purposes of
+      // probability calculations).
+      MDBuilder MDB(TI->getContext());
+      MDNode *Node = MDB.createBranchWeights(Weights);
+      TI->setMetadata(LLVMContext::MD_prof, Node);
+      NumTermsAnnotated++;
+    }
+  }
+}
+
+bool ProfileMetadataLoaderPass::runOnModule(Module &M) {
+  ProfileDataLoader PDL("profile-data-loader", Filename);
+  ProfileData PB;
+
+  ArrayRef<unsigned> Counters = PDL.getRawEdgeCounts();
+
+  unsigned ReadCount = matchEdges(M, PB, Counters);
+
+  if (ReadCount != Counters.size()) {
+    errs() << "WARNING: profile information is inconsistent with "
+           << "the current program!\n";
+  }
+  NumEdgesRead = ReadCount;
+
+  setBranchWeightMetadata(M, PB);
+
+  return ReadCount > 0;
+}
diff --git a/lib/Analysis/ProfileEstimatorPass.cpp b/lib/Analysis/ProfileEstimatorPass.cpp
index 63468f8..12b59e0 100644
--- a/lib/Analysis/ProfileEstimatorPass.cpp
+++ b/lib/Analysis/ProfileEstimatorPass.cpp
@@ -286,7 +286,7 @@ void ProfileEstimatorPass::recurseBasicBlock(BasicBlock *BB) {
     }
   }
 
-  double fraction = floor(BBWeight/Edges.size());
+  double fraction = Edges.size() ? floor(BBWeight/Edges.size()) : 0.0;
   // Finally we know what flow is still not leaving the block, distribute this
   // flow onto the empty edges.
   for (SmallVector<Edge, 8>::iterator ei = Edges.begin(), ee = Edges.end();
diff --git a/lib/Analysis/ProfileInfo.cpp b/lib/Analysis/ProfileInfo.cpp
index 173de2c..b5b7ac1 100644
--- a/lib/Analysis/ProfileInfo.cpp
+++ b/lib/Analysis/ProfileInfo.cpp
@@ -1016,40 +1016,14 @@ void ProfileInfoT<Function,BasicBlock>::repair(const Function *F) {
   }
 }
 
-raw_ostream& operator<<(raw_ostream &O, const Function *F) {
-  return O << F->getName();
-}
-
 raw_ostream& operator<<(raw_ostream &O, const MachineFunction *MF) {
   return O << MF->getFunction()->getName() << "(MF)";
 }
 
-raw_ostream& operator<<(raw_ostream &O, const BasicBlock *BB) {
-  return O << BB->getName();
-}
-
 raw_ostream& operator<<(raw_ostream &O, const MachineBasicBlock *MBB) {
   return O << MBB->getBasicBlock()->getName() << "(MB)";
 }
 
-raw_ostream& operator<<(raw_ostream &O, std::pair<const BasicBlock *, const BasicBlock *> E) {
-  O << "(";
-
-  if (E.first)
-    O << E.first;
-  else
-    O << "0";
-
-  O << ",";
-
-  if (E.second)
-    O << E.second;
-  else
-    O << "0";
-
-  return O << ")";
-}
-
 raw_ostream& operator<<(raw_ostream &O, std::pair<const MachineBasicBlock *, const MachineBasicBlock *> E) {
   O << "(";
 
diff --git a/lib/Analysis/RegionInfo.cpp b/lib/Analysis/RegionInfo.cpp
index 868f483..0f9a8b3 100644
--- a/lib/Analysis/RegionInfo.cpp
+++ b/lib/Analysis/RegionInfo.cpp
@@ -47,7 +47,7 @@ static cl::opt<enum Region::PrintStyle> printStyle("print-region-style",
   cl::values(
     clEnumValN(Region::PrintNone, "none",  "print no details"),
     clEnumValN(Region::PrintBB, "bb",
-               "print regions in detail with block_node_iterator"),
+               "print regions in detail with block_iterator"),
     clEnumValN(Region::PrintRN, "rn",
                "print regions in detail with element_iterator"),
     clEnumValEnd));
@@ -246,22 +246,6 @@ void Region::verifyRegionNest() const {
   verifyRegion();
 }
 
-Region::block_node_iterator Region::block_node_begin() {
-  return GraphTraits<FlatIt<Region*> >::nodes_begin(this);
-}
-
-Region::block_node_iterator Region::block_node_end() {
-  return GraphTraits<FlatIt<Region*> >::nodes_end(this);
-}
-
-Region::const_block_node_iterator Region::block_node_begin() const {
-  return GraphTraits<FlatIt<const Region*> >::nodes_begin(this);
-}
-
-Region::const_block_node_iterator Region::block_node_end() const {
-  return GraphTraits<FlatIt<const Region*> >::nodes_end(this);
-}
-
 Region::element_iterator Region::element_begin() {
   return GraphTraits<Region*>::nodes_begin(this);
 }
@@ -425,10 +409,8 @@ void Region::print(raw_ostream &OS, bool print_tree, unsigned level,
     OS.indent(level*2 + 2);
 
     if (Style == PrintBB) {
-      for (const_block_node_iterator I = block_node_begin(),
-                                     E = block_node_end();
-           I != E; ++I)
-        OS << **I << ", "; // TODO: remove the last ","
+      for (const_block_iterator I = block_begin(), E = block_end(); I != E; ++I)
+        OS << (*I)->getName() << ", "; // TODO: remove the last ","
     } else if (Style == PrintRN) {
       for (const_element_iterator I = element_begin(), E = element_end(); I!=E; ++I)
         OS << **I << ", "; // TODO: remove the last ",
@@ -445,9 +427,11 @@ void Region::print(raw_ostream &OS, bool print_tree, unsigned level,
     OS.indent(level*2) << "} \n";
 }
 
+#ifndef NDEBUG
 void Region::dump() const {
   print(dbgs(), true, getDepth(), printStyle.getValue());
 }
+#endif
 
 void Region::clearNodeCache() {
   // Free the cached nodes.
diff --git a/lib/Analysis/RegionPass.cpp b/lib/Analysis/RegionPass.cpp
index c97b5eb..9208fa2 100644
--- a/lib/Analysis/RegionPass.cpp
+++ b/lib/Analysis/RegionPass.cpp
@@ -195,10 +195,9 @@ public:
 
   virtual bool runOnRegion(Region *R, RGPassManager &RGM) {
     Out << Banner;
-    for (Region::block_node_iterator I = R->block_node_begin(),
-                                     E = R->block_node_end();
+    for (Region::block_iterator I = R->block_begin(), E = R->block_end();
          I != E; ++I)
-      (*I)->getEntry()->print(Out);
+      (*I)->print(Out);
 
     return false;
   }
diff --git a/lib/Analysis/ScalarEvolution.cpp b/lib/Analysis/ScalarEvolution.cpp
index a654648..84e147b 100644
--- a/lib/Analysis/ScalarEvolution.cpp
+++ b/lib/Analysis/ScalarEvolution.cpp
@@ -122,10 +122,12 @@ char ScalarEvolution::ID = 0;
 // Implementation of the SCEV class.
 //
 
+#ifndef NDEBUG
 void SCEV::dump() const {
   print(dbgs());
   dbgs() << '\n';
 }
+#endif
 
 void SCEV::print(raw_ostream &OS) const {
   switch (getSCEVType()) {
diff --git a/lib/Analysis/Trace.cpp b/lib/Analysis/Trace.cpp
index ff5010b..dbb9535 100644
--- a/lib/Analysis/Trace.cpp
+++ b/lib/Analysis/Trace.cpp
@@ -43,9 +43,11 @@ void Trace::print(raw_ostream &O) const {
   O << "; Trace parent function: \n" << *F;
 }
 
+#ifndef NDEBUG
 /// dump - Debugger convenience method; writes trace to standard error
 /// output stream.
 ///
 void Trace::dump() const {
   print(dbgs());
 }
+#endif
diff --git a/lib/Analysis/ValueTracking.cpp b/lib/Analysis/ValueTracking.cpp
index cea34e1..491224a 100644
--- a/lib/Analysis/ValueTracking.cpp
+++ b/lib/Analysis/ValueTracking.cpp
@@ -1614,7 +1614,7 @@ Value *llvm::GetPointerBaseWithConstantOffset(Value *Ptr, int64_t &Offset,
   // right.
   unsigned PtrSize = TD.getPointerSizeInBits();
   if (PtrSize < 64)
-    Offset = (Offset << (64-PtrSize)) >> (64-PtrSize);
+    Offset = SignExtend64(Offset, PtrSize);
   
   return GetPointerBaseWithConstantOffset(GEP->getPointerOperand(), Offset, TD);
 }