Import LLVM r73021.

31 files changed, 613 insertions, 366 deletions

diff --git a/docs/GoldPlugin.html b/docs/GoldPlugin.html
index 17a50ac..b5148ab 100644
--- a/docs/GoldPlugin.html
+++ b/docs/GoldPlugin.html
@@ -157,7 +157,7 @@ export AR="$PREFIX/bin/ar"
 export NM="$PREFIX/bin/nm"
 export RANLIB=/bin/true #ranlib is not needed, and doesn't support .bc files in .a
 export CFLAGS="-O4"
-     </pre>
+</pre>
      </li>
      <li>Or you can just set your path:
     <pre class="doc_code">
@@ -166,7 +166,7 @@ export CC="llvm-gcc -use-gold-plugin"
 export CXX="llvm-g++ -use-gold-plugin"
 export RANLIB=/bin/true
 export CFLAGS="-O4"
-     </pre>
+</pre>
      </li>
      <li>Configure &amp; build the project as usual: <tt>./configure &amp;&amp; make &amp;&amp; make check</tt> </li>
    </ul>
diff --git a/include/llvm/Analysis/ScalarEvolution.h b/include/llvm/Analysis/ScalarEvolution.h
index 88002cb..e3d4299 100644
--- a/include/llvm/Analysis/ScalarEvolution.h
+++ b/include/llvm/Analysis/ScalarEvolution.h
@@ -227,9 +227,9 @@ namespace llvm {
     ///
     TargetData *TD;
 
-    /// UnknownValue - This SCEV is used to represent unknown trip counts and
-    /// things.
-    SCEVHandle UnknownValue;
+    /// CouldNotCompute - This SCEV is used to represent unknown trip
+    /// counts and things.
+    SCEVHandle CouldNotCompute;
 
     /// Scalars - This is a cache of the scalars we have analyzed so far.
     ///
@@ -322,23 +322,23 @@ namespace llvm {
     /// a constant number of times (the condition evolves only from constants),
     /// try to evaluate a few iterations of the loop until we get the exit
     /// condition gets a value of ExitWhen (true or false).  If we cannot
-    /// evaluate the trip count of the loop, return UnknownValue.
+    /// evaluate the trip count of the loop, return CouldNotCompute.
     SCEVHandle ComputeBackedgeTakenCountExhaustively(const Loop *L, Value *Cond,
                                                      bool ExitWhen);
 
     /// HowFarToZero - Return the number of times a backedge comparing the
     /// specified value to zero will execute.  If not computable, return
-    /// UnknownValue.
+    /// CouldNotCompute.
     SCEVHandle HowFarToZero(const SCEV *V, const Loop *L);
 
     /// HowFarToNonZero - Return the number of times a backedge checking the
     /// specified value for nonzero will execute.  If not computable, return
-    /// UnknownValue.
+    /// CouldNotCompute.
     SCEVHandle HowFarToNonZero(const SCEV *V, const Loop *L);
 
     /// HowManyLessThans - Return the number of times a backedge containing the
     /// specified less-than comparison will execute.  If not computable, return
-    /// UnknownValue. isSigned specifies whether the less-than is signed.
+    /// CouldNotCompute. isSigned specifies whether the less-than is signed.
     BackedgeTakenInfo HowManyLessThans(const SCEV *LHS, const SCEV *RHS,
                                        const Loop *L, bool isSigned);
 
diff --git a/include/llvm/DerivedTypes.h b/include/llvm/DerivedTypes.h
index dd6d908..fa951bf 100644
--- a/include/llvm/DerivedTypes.h
+++ b/include/llvm/DerivedTypes.h
@@ -159,6 +159,10 @@ public:
   /// type.
   static bool isValidReturnType(const Type *RetTy);
 
+  /// isValidArgumentType - Return true if the specified type is valid as an
+  /// argument type.
+  static bool isValidArgumentType(const Type *ArgTy);
+
   inline bool isVarArg() const { return isVarArgs; }
   inline const Type *getReturnType() const { return ContainedTys[0]; }
 
@@ -232,6 +236,10 @@ public:
   /// an empty struct, pass NULL, NULL.
   static StructType *get(const Type *type, ...) END_WITH_NULL;
 
+  /// isValidElementType - Return true if the specified type is valid as a
+  /// element type.
+  static bool isValidElementType(const Type *ElemTy);
+
   // Iterator access to the elements
   typedef Type::subtype_iterator element_iterator;
   element_iterator element_begin() const { return ContainedTys; }
@@ -331,6 +339,10 @@ public:
   ///
   static ArrayType *get(const Type *ElementType, uint64_t NumElements);
 
+  /// isValidElementType - Return true if the specified type is valid as a
+  /// element type.
+  static bool isValidElementType(const Type *ElemTy);
+
   inline uint64_t getNumElements() const { return NumElements; }
 
   // Implement the AbstractTypeUser interface.
@@ -391,6 +403,10 @@ public:
     return VectorType::get(EltTy, VTy->getNumElements());
   }
 
+  /// isValidElementType - Return true if the specified type is valid as a
+  /// element type.
+  static bool isValidElementType(const Type *ElemTy);
+
   /// @brief Return the number of elements in the Vector type.
   inline unsigned getNumElements() const { return NumElements; }
 
@@ -431,6 +447,10 @@ public:
     return PointerType::get(ElementType, 0);
   }
 
+  /// isValidElementType - Return true if the specified type is valid as a
+  /// element type.
+  static bool isValidElementType(const Type *ElemTy);
+
   /// @brief Return the address space of the Pointer type.
   inline unsigned getAddressSpace() const { return AddressSpace; }
 
diff --git a/include/llvm/Type.h b/include/llvm/Type.h
index 5ce23ef..c1732af 100644
--- a/include/llvm/Type.h
+++ b/include/llvm/Type.h
@@ -205,7 +205,7 @@ public:
   inline bool isAbstract() const { return Abstract; }
 
   /// canLosslesslyBitCastTo - Return true if this type could be converted 
-  /// with a lossless BitCast to type 'Ty'. For example, uint to int. BitCasts 
+  /// with a lossless BitCast to type 'Ty'. For example, i8* to i32*. BitCasts 
   /// are valid for types of the same size only where no re-interpretation of 
   /// the bits is done.
   /// @brief Determine if this type could be losslessly bitcast to Ty
diff --git a/lib/Analysis/ScalarEvolution.cpp b/lib/Analysis/ScalarEvolution.cpp
index 03c5005..fd97db8 100644
--- a/lib/Analysis/ScalarEvolution.cpp
+++ b/lib/Analysis/ScalarEvolution.cpp
@@ -1771,7 +1771,7 @@ const Type *ScalarEvolution::getEffectiveSCEVType(const Type *Ty) const {
 }
 
 SCEVHandle ScalarEvolution::getCouldNotCompute() {
-  return UnknownValue;
+  return CouldNotCompute;
 }
 
 /// hasSCEV - Return true if the SCEV for this value has already been
@@ -2395,7 +2395,7 @@ ScalarEvolution::getBackedgeTakenInfo(const Loop *L) {
     BackedgeTakenCounts.insert(std::make_pair(L, getCouldNotCompute()));
   if (Pair.second) {
     BackedgeTakenInfo ItCount = ComputeBackedgeTakenCount(L);
-    if (ItCount.Exact != UnknownValue) {
+    if (ItCount.Exact != CouldNotCompute) {
       assert(ItCount.Exact->isLoopInvariant(L) &&
              ItCount.Max->isLoopInvariant(L) &&
              "Computed trip count isn't loop invariant for loop!");
@@ -2464,20 +2464,20 @@ ScalarEvolution::ComputeBackedgeTakenCount(const Loop *L) {
   // If the loop has a non-one exit block count, we can't analyze it.
   BasicBlock *ExitBlock = L->getExitBlock();
   if (!ExitBlock)
-    return UnknownValue;
+    return CouldNotCompute;
 
   // Okay, there is one exit block.  Try to find the condition that causes the
   // loop to be exited.
   BasicBlock *ExitingBlock = L->getExitingBlock();
   if (!ExitingBlock)
-    return UnknownValue;   // More than one block exiting!
+    return CouldNotCompute;   // More than one block exiting!
 
   // Okay, we've computed the exiting block.  See what condition causes us to
   // exit.
   //
   // FIXME: we should be able to handle switch instructions (with a single exit)
   BranchInst *ExitBr = dyn_cast<BranchInst>(ExitingBlock->getTerminator());
-  if (ExitBr == 0) return UnknownValue;
+  if (ExitBr == 0) return CouldNotCompute;
   assert(ExitBr->isConditional() && "If unconditional, it can't be in loop!");
   
   // At this point, we know we have a conditional branch that determines whether
@@ -2493,7 +2493,7 @@ ScalarEvolution::ComputeBackedgeTakenCount(const Loop *L) {
   if (ExitBr->getSuccessor(0) != L->getHeader() &&
       ExitBr->getSuccessor(1) != L->getHeader() &&
       ExitBr->getParent() != L->getHeader())
-    return UnknownValue;
+    return CouldNotCompute;
   
   ICmpInst *ExitCond = dyn_cast<ICmpInst>(ExitBr->getCondition());
 
@@ -2647,11 +2647,11 @@ SCEVHandle ScalarEvolution::
 ComputeLoadConstantCompareBackedgeTakenCount(LoadInst *LI, Constant *RHS,
                                              const Loop *L,
                                              ICmpInst::Predicate predicate) {
-  if (LI->isVolatile()) return UnknownValue;
+  if (LI->isVolatile()) return CouldNotCompute;
 
   // Check to see if the loaded pointer is a getelementptr of a global.
   GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(LI->getOperand(0));
-  if (!GEP) return UnknownValue;
+  if (!GEP) return CouldNotCompute;
 
   // Make sure that it is really a constant global we are gepping, with an
   // initializer, and make sure the first IDX is really 0.
@@ -2659,7 +2659,7 @@ ComputeLoadConstantCompareBackedgeTakenCount(LoadInst *LI, Constant *RHS,
   if (!GV || !GV->isConstant() || !GV->hasInitializer() ||
       GEP->getNumOperands() < 3 || !isa<Constant>(GEP->getOperand(1)) ||
       !cast<Constant>(GEP->getOperand(1))->isNullValue())
-    return UnknownValue;
+    return CouldNotCompute;
 
   // Okay, we allow one non-constant index into the GEP instruction.
   Value *VarIdx = 0;
@@ -2669,7 +2669,7 @@ ComputeLoadConstantCompareBackedgeTakenCount(LoadInst *LI, Constant *RHS,
     if (ConstantInt *CI = dyn_cast<ConstantInt>(GEP->getOperand(i))) {
       Indexes.push_back(CI);
     } else if (!isa<ConstantInt>(GEP->getOperand(i))) {
-      if (VarIdx) return UnknownValue;  // Multiple non-constant idx's.
+      if (VarIdx) return CouldNotCompute;  // Multiple non-constant idx's.
       VarIdx = GEP->getOperand(i);
       VarIdxNum = i-2;
       Indexes.push_back(0);
@@ -2686,7 +2686,7 @@ ComputeLoadConstantCompareBackedgeTakenCount(LoadInst *LI, Constant *RHS,
   if (!IdxExpr || !IdxExpr->isAffine() || IdxExpr->isLoopInvariant(L) ||
       !isa<SCEVConstant>(IdxExpr->getOperand(0)) ||
       !isa<SCEVConstant>(IdxExpr->getOperand(1)))
-    return UnknownValue;
+    return CouldNotCompute;
 
   unsigned MaxSteps = MaxBruteForceIterations;
   for (unsigned IterationNum = 0; IterationNum != MaxSteps; ++IterationNum) {
@@ -2713,7 +2713,7 @@ ComputeLoadConstantCompareBackedgeTakenCount(LoadInst *LI, Constant *RHS,
       return getConstant(ItCst);   // Found terminating iteration!
     }
   }
-  return UnknownValue;
+  return CouldNotCompute;
 }
 
 
@@ -2852,11 +2852,11 @@ getConstantEvolutionLoopExitValue(PHINode *PN, const APInt& BEs, const Loop *L){
 /// constant number of times (the condition evolves only from constants),
 /// try to evaluate a few iterations of the loop until we get the exit
 /// condition gets a value of ExitWhen (true or false).  If we cannot
-/// evaluate the trip count of the loop, return UnknownValue.
+/// evaluate the trip count of the loop, return CouldNotCompute.
 SCEVHandle ScalarEvolution::
 ComputeBackedgeTakenCountExhaustively(const Loop *L, Value *Cond, bool ExitWhen) {
   PHINode *PN = getConstantEvolvingPHI(Cond, L);
-  if (PN == 0) return UnknownValue;
+  if (PN == 0) return CouldNotCompute;
 
   // Since the loop is canonicalized, the PHI node must have two entries.  One
   // entry must be a constant (coming in from outside of the loop), and the
@@ -2864,11 +2864,11 @@ ComputeBackedgeTakenCountExhaustively(const Loop *L, Value *Cond, bool ExitWhen)
   bool SecondIsBackedge = L->contains(PN->getIncomingBlock(1));
   Constant *StartCST =
     dyn_cast<Constant>(PN->getIncomingValue(!SecondIsBackedge));
-  if (StartCST == 0) return UnknownValue;  // Must be a constant.
+  if (StartCST == 0) return CouldNotCompute;  // Must be a constant.
 
   Value *BEValue = PN->getIncomingValue(SecondIsBackedge);
   PHINode *PN2 = getConstantEvolvingPHI(BEValue, L);
-  if (PN2 != PN) return UnknownValue;  // Not derived from same PHI.
+  if (PN2 != PN) return CouldNotCompute;  // Not derived from same PHI.
 
   // Okay, we find a PHI node that defines the trip count of this loop.  Execute
   // the loop symbolically to determine when the condition gets a value of
@@ -2881,7 +2881,7 @@ ComputeBackedgeTakenCountExhaustively(const Loop *L, Value *Cond, bool ExitWhen)
       dyn_cast_or_null<ConstantInt>(EvaluateExpression(Cond, PHIVal));
 
     // Couldn't symbolically evaluate.
-    if (!CondVal) return UnknownValue;
+    if (!CondVal) return CouldNotCompute;
 
     if (CondVal->getValue() == uint64_t(ExitWhen)) {
       ConstantEvolutionLoopExitValue[PN] = PHIVal;
@@ -2892,12 +2892,12 @@ ComputeBackedgeTakenCountExhaustively(const Loop *L, Value *Cond, bool ExitWhen)
     // Compute the value of the PHI node for the next iteration.
     Constant *NextPHI = EvaluateExpression(BEValue, PHIVal);
     if (NextPHI == 0 || NextPHI == PHIVal)
-      return UnknownValue;  // Couldn't evaluate or not making progress...
+      return CouldNotCompute;   // Couldn't evaluate or not making progress...
     PHIVal = NextPHI;
   }
 
   // Too many iterations were needed to evaluate.
-  return UnknownValue;
+  return CouldNotCompute;
 }
 
 /// getSCEVAtScope - Return a SCEV expression handle for the specified value
@@ -3052,7 +3052,7 @@ SCEVHandle ScalarEvolution::getSCEVAtScope(const SCEV *V, const Loop *L) {
       // To evaluate this recurrence, we need to know how many times the AddRec
       // loop iterates.  Compute this now.
       SCEVHandle BackedgeTakenCount = getBackedgeTakenCount(AddRec->getLoop());
-      if (BackedgeTakenCount == UnknownValue) return AddRec;
+      if (BackedgeTakenCount == CouldNotCompute) return AddRec;
 
       // Then, evaluate the AddRec.
       return AddRec->evaluateAtIteration(BackedgeTakenCount, *this);
@@ -3201,18 +3201,18 @@ SolveQuadraticEquation(const SCEVAddRecExpr *AddRec, ScalarEvolution &SE) {
 }
 
 /// HowFarToZero - Return the number of times a backedge comparing the specified
-/// value to zero will execute.  If not computable, return UnknownValue.
+/// value to zero will execute.  If not computable, return CouldNotCompute.
 SCEVHandle ScalarEvolution::HowFarToZero(const SCEV *V, const Loop *L) {
   // If the value is a constant
   if (const SCEVConstant *C = dyn_cast<SCEVConstant>(V)) {
     // If the value is already zero, the branch will execute zero times.
     if (C->getValue()->isZero()) return C;
-    return UnknownValue;  // Otherwise it will loop infinitely.
+    return CouldNotCompute;  // Otherwise it will loop infinitely.
   }
 
   const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(V);
   if (!AddRec || AddRec->getLoop() != L)
-    return UnknownValue;
+    return CouldNotCompute;
 
   if (AddRec->isAffine()) {
     // If this is an affine expression, the execution count of this branch is
@@ -3274,12 +3274,12 @@ SCEVHandle ScalarEvolution::HowFarToZero(const SCEV *V, const Loop *L) {
     }
   }
 
-  return UnknownValue;
+  return CouldNotCompute;
 }
 
 /// HowFarToNonZero - Return the number of times a backedge checking the
 /// specified value for nonzero will execute.  If not computable, return
-/// UnknownValue
+/// CouldNotCompute
 SCEVHandle ScalarEvolution::HowFarToNonZero(const SCEV *V, const Loop *L) {
   // Loops that look like: while (X == 0) are very strange indeed.  We don't
   // handle them yet except for the trivial case.  This could be expanded in the
@@ -3290,12 +3290,12 @@ SCEVHandle ScalarEvolution::HowFarToNonZero(const SCEV *V, const Loop *L) {
   if (const SCEVConstant *C = dyn_cast<SCEVConstant>(V)) {
     if (!C->getValue()->isNullValue())
       return getIntegerSCEV(0, C->getType());
-    return UnknownValue;  // Otherwise it will loop infinitely.
+    return CouldNotCompute;  // Otherwise it will loop infinitely.
   }
 
   // We could implement others, but I really doubt anyone writes loops like
   // this, and if they did, they would already be constant folded.
-  return UnknownValue;
+  return CouldNotCompute;
 }
 
 /// getLoopPredecessor - If the given loop's header has exactly one unique
@@ -3446,16 +3446,16 @@ bool ScalarEvolution::isLoopGuardedByCond(const Loop *L,
 
 /// HowManyLessThans - Return the number of times a backedge containing the
 /// specified less-than comparison will execute.  If not computable, return
-/// UnknownValue.
+/// CouldNotCompute.
 ScalarEvolution::BackedgeTakenInfo ScalarEvolution::
 HowManyLessThans(const SCEV *LHS, const SCEV *RHS,
                  const Loop *L, bool isSigned) {
   // Only handle:  "ADDREC < LoopInvariant".
-  if (!RHS->isLoopInvariant(L)) return UnknownValue;
+  if (!RHS->isLoopInvariant(L)) return CouldNotCompute;
 
   const SCEVAddRecExpr *AddRec = dyn_cast<SCEVAddRecExpr>(LHS);
   if (!AddRec || AddRec->getLoop() != L)
-    return UnknownValue;
+    return CouldNotCompute;
 
   if (AddRec->isAffine()) {
     // FORNOW: We only support unit strides.
@@ -3466,7 +3466,7 @@ HowManyLessThans(const SCEV *LHS, const SCEV *RHS,
     // TODO: handle non-constant strides.
     const SCEVConstant *CStep = dyn_cast<SCEVConstant>(Step);
     if (!CStep || CStep->isZero())
-      return UnknownValue;
+      return CouldNotCompute;
     if (CStep->isOne()) {
       // With unit stride, the iteration never steps past the limit value.
     } else if (CStep->getValue()->getValue().isStrictlyPositive()) {
@@ -3477,19 +3477,19 @@ HowManyLessThans(const SCEV *LHS, const SCEV *RHS,
           APInt Max = APInt::getSignedMaxValue(BitWidth);
           if ((Max - CStep->getValue()->getValue())
                 .slt(CLimit->getValue()->getValue()))
-            return UnknownValue;
+            return CouldNotCompute;
         } else {
           APInt Max = APInt::getMaxValue(BitWidth);
           if ((Max - CStep->getValue()->getValue())
                 .ult(CLimit->getValue()->getValue()))
-            return UnknownValue;
+            return CouldNotCompute;
         }
       } else
         // TODO: handle non-constant limit values below.
-        return UnknownValue;
+        return CouldNotCompute;
     } else
       // TODO: handle negative strides below.
-      return UnknownValue;
+      return CouldNotCompute;
 
     // We know the LHS is of the form {n,+,s} and the RHS is some loop-invariant
     // m.  So, we count the number of iterations in which {n,+,s} < m is true.
@@ -3536,7 +3536,7 @@ HowManyLessThans(const SCEV *LHS, const SCEV *RHS,
     return BackedgeTakenInfo(BECount, MaxBECount);
   }
 
-  return UnknownValue;
+  return CouldNotCompute;
 }
 
 /// getNumIterationsInRange - Return the number of iterations of this loop that
@@ -3724,7 +3724,7 @@ ScalarEvolution::SCEVCallbackVH::SCEVCallbackVH(Value *V, ScalarEvolution *se)
 //===----------------------------------------------------------------------===//
 
 ScalarEvolution::ScalarEvolution()
-  : FunctionPass(&ID), UnknownValue(new SCEVCouldNotCompute()) {
+  : FunctionPass(&ID), CouldNotCompute(new SCEVCouldNotCompute()) {
 }
 
 bool ScalarEvolution::runOnFunction(Function &F) {
diff --git a/lib/AsmParser/LLParser.cpp b/lib/AsmParser/LLParser.cpp
index 5c44502..4863f3c 100644
--- a/lib/AsmParser/LLParser.cpp
+++ b/lib/AsmParser/LLParser.cpp
@@ -1043,6 +1043,8 @@ bool LLParser::ParseTypeRec(PATypeHolder &Result) {
         return TokError("basic block pointers are invalid");
       if (Result.get() == Type::VoidTy)
         return TokError("pointers to void are invalid; use i8* instead");
+      if (!PointerType::isValidElementType(Result.get()))
+        return TokError("pointer to this type is invalid");
       Result = HandleUpRefs(PointerType::getUnqual(Result.get()));
       Lex.Lex();
       break;
@@ -1053,6 +1055,8 @@ bool LLParser::ParseTypeRec(PATypeHolder &Result) {
         return TokError("basic block pointers are invalid");
       if (Result.get() == Type::VoidTy)
         return TokError("pointers to void are invalid; use i8* instead");
+      if (!PointerType::isValidElementType(Result.get()))
+        return TokError("pointer to this type is invalid");
       unsigned AddrSpace;
       if (ParseOptionalAddrSpace(AddrSpace) ||
           ParseToken(lltok::star, "expected '*' in address space"))
@@ -1149,7 +1153,7 @@ bool LLParser::ParseArgumentList(std::vector<ArgInfo> &ArgList,
       Lex.Lex();
     }
 
-    if (!ArgTy->isFirstClassType() && !isa<OpaqueType>(ArgTy))
+    if (!FunctionType::isValidArgumentType(ArgTy))
       return Error(TypeLoc, "invalid type for function argument");
     
     ArgList.push_back(ArgInfo(TypeLoc, ArgTy, Attrs, Name));
@@ -1245,6 +1249,8 @@ bool LLParser::ParseStructType(PATypeHolder &Result, bool Packed) {
   
   if (Result == Type::VoidTy)
     return Error(EltTyLoc, "struct element can not have void type");
+  if (!StructType::isValidElementType(Result))
+    return Error(EltTyLoc, "invalid element type for struct");
   
   while (EatIfPresent(lltok::comma)) {
     EltTyLoc = Lex.getLoc();
@@ -1252,6 +1258,8 @@ bool LLParser::ParseStructType(PATypeHolder &Result, bool Packed) {
     
     if (Result == Type::VoidTy)
       return Error(EltTyLoc, "struct element can not have void type");
+    if (!StructType::isValidElementType(Result))
+      return Error(EltTyLoc, "invalid element type for struct");
     
     ParamsList.push_back(Result);
   }
@@ -1299,11 +1307,11 @@ bool LLParser::ParseArrayVectorType(PATypeHolder &Result, bool isVector) {
       return Error(SizeLoc, "zero element vector is illegal");
     if ((unsigned)Size != Size)
       return Error(SizeLoc, "size too large for vector");
-    if (!EltTy->isFloatingPoint() && !EltTy->isInteger())
+    if (!VectorType::isValidElementType(EltTy))
       return Error(TypeLoc, "vector element type must be fp or integer");
     Result = VectorType::get(EltTy, unsigned(Size));
   } else {
-    if (!EltTy->isFirstClassType() && !isa<OpaqueType>(EltTy))
+    if (!ArrayType::isValidElementType(EltTy))
       return Error(TypeLoc, "invalid array element type");
     Result = HandleUpRefs(ArrayType::get(EltTy, Size));
   }
diff --git a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index 5ae183e..f3c2833 100644
--- a/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -1785,48 +1785,41 @@ SDValue SelectionDAGLegalize::ExpandSCALAR_TO_VECTOR(SDNode *Node) {
 /// support the operation, but do support the resultant vector type.
 SDValue SelectionDAGLegalize::ExpandBUILD_VECTOR(SDNode *Node) {
   unsigned NumElems = Node->getNumOperands();
-  SDValue SplatValue = Node->getOperand(0);
+  SDValue Value1, Value2;
   DebugLoc dl = Node->getDebugLoc();
   MVT VT = Node->getValueType(0);
-  MVT OpVT = SplatValue.getValueType();
+  MVT OpVT = Node->getOperand(0).getValueType();
   MVT EltVT = VT.getVectorElementType();
 
   // If the only non-undef value is the low element, turn this into a
   // SCALAR_TO_VECTOR node.  If this is { X, X, X, X }, determine X.
   bool isOnlyLowElement = true;
-
-  // FIXME: it would be far nicer to change this into map<SDValue,uint64_t>
-  // and use a bitmask instead of a list of elements.
-  // FIXME: this doesn't treat <0, u, 0, u> for example, as a splat.
-  std::map<SDValue, std::vector<unsigned> > Values;
-  Values[SplatValue].push_back(0);
+  bool MoreThanTwoValues = false;
   bool isConstant = true;
-  if (!isa<ConstantFPSDNode>(SplatValue) && !isa<ConstantSDNode>(SplatValue) &&
-      SplatValue.getOpcode() != ISD::UNDEF)
-    isConstant = false;
-
-  for (unsigned i = 1; i < NumElems; ++i) {
+  for (unsigned i = 0; i < NumElems; ++i) {
     SDValue V = Node->getOperand(i);
-    Values[V].push_back(i);
-    if (V.getOpcode() != ISD::UNDEF)
+    if (V.getOpcode() == ISD::UNDEF)
+      continue;
+    if (i > 0)
       isOnlyLowElement = false;
-    if (SplatValue != V)
-      SplatValue = SDValue(0, 0);
-
-    // If this isn't a constant element or an undef, we can't use a constant
-    // pool load.
-    if (!isa<ConstantFPSDNode>(V) && !isa<ConstantSDNode>(V) &&
-        V.getOpcode() != ISD::UNDEF)
+    if (!isa<ConstantFPSDNode>(V) && !isa<ConstantSDNode>(V))
       isConstant = false;
+
+    if (!Value1.getNode()) {
+      Value1 = V;
+    } else if (!Value2.getNode()) {
+      if (V != Value1)
+        Value2 = V;
+    } else if (V != Value1 && V != Value2) {
+      MoreThanTwoValues = true;
+    }
   }
 
-  if (isOnlyLowElement) {
-    // If the low element is an undef too, then this whole things is an undef.
-    if (Node->getOperand(0).getOpcode() == ISD::UNDEF)
-      return DAG.getUNDEF(VT);
-    // Otherwise, turn this into a scalar_to_vector node.
+  if (!Value1.getNode())
+    return DAG.getUNDEF(VT);
+
+  if (isOnlyLowElement)
     return DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Node->getOperand(0));
-  }
 
   // If all elements are constants, create a load from the constant pool.
   if (isConstant) {
@@ -1852,59 +1845,25 @@ SDValue SelectionDAGLegalize::ExpandBUILD_VECTOR(SDNode *Node) {
                        false, Alignment);
   }
 
-  if (SplatValue.getNode()) {   // Splat of one value?
-    // Build the shuffle constant vector: <0, 0, 0, 0>
-    SmallVector<int, 8> ZeroVec(NumElems, 0);
-
-    // If the target supports VECTOR_SHUFFLE and this shuffle mask, use it.
-    if (TLI.isShuffleMaskLegal(ZeroVec, Node->getValueType(0))) {
+  if (!MoreThanTwoValues) {
+    SmallVector<int, 8> ShuffleVec(NumElems, -1);
+    for (unsigned i = 0; i < NumElems; ++i) {
+      SDValue V = Node->getOperand(i);
+      if (V.getOpcode() == ISD::UNDEF)
+        continue;
+      ShuffleVec[i] = V == Value1 ? 0 : NumElems;
+    }
+    if (TLI.isShuffleMaskLegal(ShuffleVec, Node->getValueType(0))) {
       // Get the splatted value into the low element of a vector register.
-      SDValue LowValVec =
-        DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, SplatValue);
+      SDValue Vec1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value1);
+      SDValue Vec2;
+      if (Value2.getNode())
+        Vec2 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Value2);
+      else
+        Vec2 = DAG.getUNDEF(VT);
 
       // Return shuffle(LowValVec, undef, <0,0,0,0>)
-      return DAG.getVectorShuffle(VT, dl, LowValVec, DAG.getUNDEF(VT),
-                                  &ZeroVec[0]);
-    }
-  }
-
-  // If there are only two unique elements, we may be able to turn this into a
-  // vector shuffle.
-  if (Values.size() == 2) {
-    // Get the two values in deterministic order.
-    SDValue Val1 = Node->getOperand(1);
-    SDValue Val2;
-    std::map<SDValue, std::vector<unsigned> >::iterator MI = Values.begin();
-    if (MI->first != Val1)
-      Val2 = MI->first;
-    else
-      Val2 = (++MI)->first;
-
-    // If Val1 is an undef, make sure it ends up as Val2, to ensure that our
-    // vector shuffle has the undef vector on the RHS.
-    if (Val1.getOpcode() == ISD::UNDEF)
-      std::swap(Val1, Val2);
-
-    // Build the shuffle constant vector: e.g. <0, 4, 0, 4>
-    SmallVector<int, 8> ShuffleMask(NumElems, -1);
-
-    // Set elements of the shuffle mask for Val1.
-    std::vector<unsigned> &Val1Elts = Values[Val1];
-    for (unsigned i = 0, e = Val1Elts.size(); i != e; ++i)
-      ShuffleMask[Val1Elts[i]] = 0;
-
-    // Set elements of the shuffle mask for Val2.
-    std::vector<unsigned> &Val2Elts = Values[Val2];
-    for (unsigned i = 0, e = Val2Elts.size(); i != e; ++i)
-      if (Val2.getOpcode() != ISD::UNDEF)
-        ShuffleMask[Val2Elts[i]] = NumElems;
-
-    // If the target supports SCALAR_TO_VECTOR and this shuffle mask, use it.
-    if (TLI.isOperationLegalOrCustom(ISD::SCALAR_TO_VECTOR, VT) &&
-        TLI.isShuffleMaskLegal(ShuffleMask, VT)) {
-      Val1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Val1);
-      Val2 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, Val2);
-      return DAG.getVectorShuffle(VT, dl, Val1, Val2, &ShuffleMask[0]);
+      return DAG.getVectorShuffle(VT, dl, Vec1, Vec2, ShuffleVec.data());
     }
   }
 
diff --git a/lib/Target/PowerPC/PPCInstrAltivec.td b/lib/Target/PowerPC/PPCInstrAltivec.td
index 9a5be79..3f4d329 100644
--- a/lib/Target/PowerPC/PPCInstrAltivec.td
+++ b/lib/Target/PowerPC/PPCInstrAltivec.td
@@ -666,3 +666,25 @@ def : Pat<(int_ppc_altivec_vnmsubfp VRRC:$A, VRRC:$B, VRRC:$C),
 
 def : Pat<(PPCvperm (v16i8 VRRC:$vA), VRRC:$vB, VRRC:$vC),
           (VPERM VRRC:$vA, VRRC:$vB, VRRC:$vC)>;
+
+// Vector shifts
+def : Pat<(v16i8 (shl (v16i8 VRRC:$vA), (v16i8 VRRC:$vB))),
+          (v16i8 (VSLB VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v8i16 (shl (v8i16 VRRC:$vA), (v8i16 VRRC:$vB))),
+          (v8i16 (VSLH VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v4i32 (shl (v4i32 VRRC:$vA), (v4i32 VRRC:$vB))),
+          (v4i32 (VSLW VRRC:$vA, VRRC:$vB))>;
+
+def : Pat<(v16i8 (srl (v16i8 VRRC:$vA), (v16i8 VRRC:$vB))),
+          (v16i8 (VSRB VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v8i16 (srl (v8i16 VRRC:$vA), (v8i16 VRRC:$vB))),
+          (v8i16 (VSRH VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v4i32 (srl (v4i32 VRRC:$vA), (v4i32 VRRC:$vB))),
+          (v4i32 (VSRW VRRC:$vA, VRRC:$vB))>;
+
+def : Pat<(v16i8 (sra (v16i8 VRRC:$vA), (v16i8 VRRC:$vB))),
+          (v16i8 (VSRAB VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v8i16 (sra (v8i16 VRRC:$vA), (v8i16 VRRC:$vB))),
+          (v8i16 (VSRAH VRRC:$vA, VRRC:$vB))>;
+def : Pat<(v4i32 (sra (v4i32 VRRC:$vA), (v4i32 VRRC:$vB))),
+          (v4i32 (VSRAW VRRC:$vA, VRRC:$vB))>;
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index ef60ff5..9e15a54 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -7691,13 +7691,15 @@ static bool isBaseAlignmentOfN(unsigned N, SDNode *Base,
 }
 
 static bool EltsFromConsecutiveLoads(ShuffleVectorSDNode *N, unsigned NumElems,
-                                     MVT EVT, SDNode *&Base,
+                                     MVT EVT, LoadSDNode *&LDBase,
+                                     unsigned &LastLoadedElt,
                                      SelectionDAG &DAG, MachineFrameInfo *MFI,
                                      const TargetLowering &TLI) {
-  Base = NULL;
+  LDBase = NULL;
+  LastLoadedElt = -1;
   for (unsigned i = 0; i < NumElems; ++i) {
     if (N->getMaskElt(i) < 0) {
-      if (!Base)
+      if (!LDBase)
         return false;
       continue;
     }
@@ -7706,19 +7708,20 @@ static bool EltsFromConsecutiveLoads(ShuffleVectorSDNode *N, unsigned NumElems,
     if (!Elt.getNode() ||
         (Elt.getOpcode() != ISD::UNDEF && !ISD::isNON_EXTLoad(Elt.getNode())))
       return false;
-    if (!Base) {
-      Base = Elt.getNode();
-      if (Base->getOpcode() == ISD::UNDEF)
+    if (!LDBase) {
+      if (Elt.getNode()->getOpcode() == ISD::UNDEF)
         return false;
+      LDBase = cast<LoadSDNode>(Elt.getNode());
+      LastLoadedElt = i;
       continue;
     }
     if (Elt.getOpcode() == ISD::UNDEF)
       continue;
 
     LoadSDNode *LD = cast<LoadSDNode>(Elt);
-    LoadSDNode *LDBase = cast<LoadSDNode>(Base);
     if (!TLI.isConsecutiveLoad(LD, LDBase, EVT.getSizeInBits()/8, i, MFI))
       return false;
+    LastLoadedElt = i;
   }
   return true;
 }
@@ -7737,134 +7740,29 @@ static SDValue PerformShuffleCombine(SDNode *N, SelectionDAG &DAG,
   ShuffleVectorSDNode *SVN = cast<ShuffleVectorSDNode>(N);
   unsigned NumElems = VT.getVectorNumElements();
 
-  // For x86-32 machines, if we see an insert and then a shuffle in a v2i64
-  // where the upper half is 0, it is advantageous to rewrite it as a build
-  // vector of (0, val) so it can use movq.
-  if (VT == MVT::v2i64) {
-    SDValue In[2];
-    In[0] = N->getOperand(0);
-    In[1] = N->getOperand(1);
-    int Idx0 = SVN->getMaskElt(0);
-    int Idx1 = SVN->getMaskElt(1);
-    // FIXME: can we take advantage of undef index?
-    if (Idx0 >= 0 && Idx1 >= 0 &&
-        In[Idx0/2].getOpcode() == ISD::INSERT_VECTOR_ELT &&
-        In[Idx1/2].getOpcode() == ISD::BUILD_VECTOR) {
-      ConstantSDNode* InsertVecIdx =
-                             dyn_cast<ConstantSDNode>(In[Idx0/2].getOperand(2));
-      if (InsertVecIdx &&
-          InsertVecIdx->getZExtValue() == (unsigned)(Idx0 % 2) &&
-          isZeroNode(In[Idx1/2].getOperand(Idx1 % 2))) {
-        return DAG.getNode(ISD::BUILD_VECTOR, dl, VT,
-                           In[Idx0/2].getOperand(1),
-                           In[Idx1/2].getOperand(Idx1 % 2));
-      }
-    }
-  }
+  if (VT.getSizeInBits() != 128)
+    return SDValue();
 
   // Try to combine a vector_shuffle into a 128-bit load.
   MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
-  SDNode *Base = NULL;
-  if (!EltsFromConsecutiveLoads(SVN, NumElems, EVT, Base, DAG, MFI, TLI))
+  LoadSDNode *LD = NULL;
+  unsigned LastLoadedElt;
+  if (!EltsFromConsecutiveLoads(SVN, NumElems, EVT, LD, LastLoadedElt, DAG,
+                                MFI, TLI))
     return SDValue();
 
-  LoadSDNode *LD = cast<LoadSDNode>(Base);
-  if (isBaseAlignmentOfN(16, Base->getOperand(1).getNode(), TLI))
+  if (LastLoadedElt == NumElems - 1) {
+    if (isBaseAlignmentOfN(16, LD->getBasePtr().getNode(), TLI))
+      return DAG.getLoad(VT, dl, LD->getChain(), LD->getBasePtr(),
+                         LD->getSrcValue(), LD->getSrcValueOffset(),
+                         LD->isVolatile());
     return DAG.getLoad(VT, dl, LD->getChain(), LD->getBasePtr(),
                        LD->getSrcValue(), LD->getSrcValueOffset(),
-                       LD->isVolatile());
-  return DAG.getLoad(VT, dl, LD->getChain(), LD->getBasePtr(),
-                     LD->getSrcValue(), LD->getSrcValueOffset(),
-                     LD->isVolatile(), LD->getAlignment());
-}
-
-/// PerformBuildVectorCombine - build_vector 0,(load i64 / f64) -> movq / movsd.
-static SDValue PerformBuildVectorCombine(SDNode *N, SelectionDAG &DAG,
-                                         TargetLowering::DAGCombinerInfo &DCI,
-                                         const X86Subtarget *Subtarget,
-                                         const TargetLowering &TLI) {
-  unsigned NumOps = N->getNumOperands();
-  DebugLoc dl = N->getDebugLoc();
-
-  // Ignore single operand BUILD_VECTOR.
-  if (NumOps == 1)
-    return SDValue();
-
-  MVT VT = N->getValueType(0);
-  MVT EVT = VT.getVectorElementType();
-  
-  // Before or during type legalization, we want to try and convert a
-  // build_vector of an i64 load and a zero value into vzext_movl before the 
-  // legalizer can break it up.  
-  // FIXME: does the case below remove the need to do this?
-  if (DCI.isBeforeLegalize() || DCI.isCalledByLegalizer()) {
-    if ((EVT != MVT::i64 && EVT != MVT::f64) || Subtarget->is64Bit())
-      return SDValue();
-    
-    // This must be an insertion into a zero vector.
-    SDValue HighElt = N->getOperand(1);
-    if (!isZeroNode(HighElt))
-      return SDValue();
-    
-    // Value must be a load.
-    SDNode *Base = N->getOperand(0).getNode();
-    if (!isa<LoadSDNode>(Base)) {
-      if (Base->getOpcode() != ISD::BIT_CONVERT)
-        return SDValue();
-      Base = Base->getOperand(0).getNode();
-      if (!isa<LoadSDNode>(Base))
-        return SDValue();
-    }
-    
-    // Transform it into VZEXT_LOAD addr.
-    LoadSDNode *LD = cast<LoadSDNode>(Base);
-    
-    // Load must not be an extload.
-    if (LD->getExtensionType() != ISD::NON_EXTLOAD)
-      return SDValue();
-    
-    // Load type should legal type so we don't have to legalize it.
-    if (!TLI.isTypeLegal(VT))
-      return SDValue();
-    
-    SDVTList Tys = DAG.getVTList(VT, MVT::Other);
-    SDValue Ops[] = { LD->getChain(), LD->getBasePtr() };
-    SDValue ResNode = DAG.getNode(X86ISD::VZEXT_LOAD, dl, Tys, Ops, 2);
-    TargetLowering::TargetLoweringOpt TLO(DAG);
-    TLO.CombineTo(SDValue(Base, 1), ResNode.getValue(1));
-    DCI.CommitTargetLoweringOpt(TLO);
-    return ResNode;
-  }
-
-  // The type legalizer will have broken apart v2i64 build_vector created during
-  // widening before the code which handles that case is run.  Look for build
-  // vector (load, load + 4, 0/undef, 0/undef)
-  if (VT == MVT::v4i32 || VT == MVT::v4f32) {
-    LoadSDNode *LD0 = dyn_cast<LoadSDNode>(N->getOperand(0));
-    LoadSDNode *LD1 = dyn_cast<LoadSDNode>(N->getOperand(1));
-    if (!LD0 || !LD1)
-      return SDValue();
-    if (LD0->getExtensionType() != ISD::NON_EXTLOAD ||
-        LD1->getExtensionType() != ISD::NON_EXTLOAD)
-      return SDValue();
-    // Make sure the second elt is a consecutive load.
-    if (!TLI.isConsecutiveLoad(LD1, LD0, EVT.getSizeInBits()/8, 1,
-                               DAG.getMachineFunction().getFrameInfo()))
-      return SDValue();
-
-    SDValue N2 = N->getOperand(2);
-    SDValue N3 = N->getOperand(3);
-    if (!isZeroNode(N2) && N2.getOpcode() != ISD::UNDEF)
-      return SDValue();
-    if (!isZeroNode(N3) && N3.getOpcode() != ISD::UNDEF)
-      return SDValue();
-    
+                       LD->isVolatile(), LD->getAlignment());
+  } else if (NumElems == 4 && LastLoadedElt == 1) {
     SDVTList Tys = DAG.getVTList(MVT::v2i64, MVT::Other);
-    SDValue Ops[] = { LD0->getChain(), LD0->getBasePtr() };
+    SDValue Ops[] = { LD->getChain(), LD->getBasePtr() };
     SDValue ResNode = DAG.getNode(X86ISD::VZEXT_LOAD, dl, Tys, Ops, 2);
-    TargetLowering::TargetLoweringOpt TLO(DAG);
-    TLO.CombineTo(SDValue(LD0, 1), ResNode.getValue(1));
-    DCI.CommitTargetLoweringOpt(TLO);
     return DAG.getNode(ISD::BIT_CONVERT, dl, VT, ResNode);
   }
   return SDValue();
@@ -8466,14 +8364,25 @@ static SDValue PerformBTCombine(SDNode *N,
   return SDValue();
 }
 
+static SDValue PerformVZEXT_MOVLCombine(SDNode *N, SelectionDAG &DAG) {
+  SDValue Op = N->getOperand(0);
+  if (Op.getOpcode() == ISD::BIT_CONVERT)
+    Op = Op.getOperand(0);
+  MVT VT = N->getValueType(0), OpVT = Op.getValueType();
+  if (Op.getOpcode() == X86ISD::VZEXT_LOAD &&
+      VT.getVectorElementType().getSizeInBits() == 
+      OpVT.getVectorElementType().getSizeInBits()) {
+    return DAG.getNode(ISD::BIT_CONVERT, N->getDebugLoc(), VT, Op);
+  }
+  return SDValue();
+}
+
 SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
                                              DAGCombinerInfo &DCI) const {
   SelectionDAG &DAG = DCI.DAG;
   switch (N->getOpcode()) {
   default: break;
   case ISD::VECTOR_SHUFFLE: return PerformShuffleCombine(N, DAG, *this);
-  case ISD::BUILD_VECTOR:
-    return PerformBuildVectorCombine(N, DAG, DCI, Subtarget, *this);
   case ISD::SELECT:         return PerformSELECTCombine(N, DAG, Subtarget);
   case X86ISD::CMOV:        return PerformCMOVCombine(N, DAG, DCI);
   case ISD::MUL:            return PerformMulCombine(N, DAG, DCI);
@@ -8485,6 +8394,7 @@ SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
   case X86ISD::FOR:         return PerformFORCombine(N, DAG);
   case X86ISD::FAND:        return PerformFANDCombine(N, DAG);
   case X86ISD::BT:          return PerformBTCombine(N, DAG, DCI);
+  case X86ISD::VZEXT_MOVL:  return PerformVZEXT_MOVLCombine(N, DAG);
   }
 
   return SDValue();
diff --git a/lib/Transforms/IPO/GlobalOpt.cpp b/lib/Transforms/IPO/GlobalOpt.cpp
index 2c01cc3..5f12825 100644
--- a/lib/Transforms/IPO/GlobalOpt.cpp
+++ b/lib/Transforms/IPO/GlobalOpt.cpp
@@ -932,8 +932,7 @@ static bool ValueIsOnlyUsedLocallyOrStoredToOneGlobal(Instruction *V,
                                                       GlobalVariable *GV,
                                               SmallPtrSet<PHINode*, 8> &PHIs) {
   for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){
-    Instruction *Inst = dyn_cast<Instruction>(*UI);
-    if (Inst == 0) return false;
+    Instruction *Inst = cast<Instruction>(*UI);
     
     if (isa<LoadInst>(Inst) || isa<CmpInst>(Inst)) {
       continue; // Fine, ignore.
diff --git a/lib/Transforms/IPO/IPConstantPropagation.cpp b/lib/Transforms/IPO/IPConstantPropagation.cpp
index 2dc8558..e4a9dea 100644
--- a/lib/Transforms/IPO/IPConstantPropagation.cpp
+++ b/lib/Transforms/IPO/IPConstantPropagation.cpp
@@ -241,15 +241,11 @@ bool IPCP::PropagateConstantReturn(Function &F) {
    
     for (Value::use_iterator I = Call->use_begin(), E = Call->use_end();
          I != E;) {
-      Instruction *Ins = dyn_cast<Instruction>(*I);
+      Instruction *Ins = cast<Instruction>(*I);
 
       // Increment now, so we can remove the use
       ++I;
 
-      // Not an instruction? Ignore
-      if (!Ins)
-        continue;
-
       // Find the index of the retval to replace with
       int index = -1;
       if (ExtractValueInst *EV = dyn_cast<ExtractValueInst>(Ins))
diff --git a/lib/Transforms/Scalar/InstructionCombining.cpp b/lib/Transforms/Scalar/InstructionCombining.cpp
index 97bd34c..6d2ff0e 100644
--- a/lib/Transforms/Scalar/InstructionCombining.cpp
+++ b/lib/Transforms/Scalar/InstructionCombining.cpp
@@ -12579,6 +12579,12 @@ Instruction *InstCombiner::visitInsertElementInst(InsertElementInst &IE) {
     }
   }
 
+  unsigned VWidth = cast<VectorType>(VecOp->getType())->getNumElements();
+  APInt UndefElts(VWidth, 0);
+  APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
+  if (SimplifyDemandedVectorElts(&IE, AllOnesEltMask, UndefElts))
+    return &IE;
+
   return 0;
 }
 
diff --git a/lib/Transforms/Scalar/PredicateSimplifier.cpp b/lib/Transforms/Scalar/PredicateSimplifier.cpp
index a7e4d6e..b9b5688 100644
--- a/lib/Transforms/Scalar/PredicateSimplifier.cpp
+++ b/lib/Transforms/Scalar/PredicateSimplifier.cpp
@@ -1525,12 +1525,12 @@ namespace {
         Instruction *I2 = dyn_cast<Instruction>(R);
         if (I2 && below(I2)) {
           std::vector<Instruction *> ToNotify;
-          for (Value::use_iterator UI = R->use_begin(), UE = R->use_end();
+          for (Value::use_iterator UI = I2->use_begin(), UE = I2->use_end();
                UI != UE;) {
             Use &TheUse = UI.getUse();
             ++UI;
-            if (Instruction *I = dyn_cast<Instruction>(TheUse.getUser()))
-              ToNotify.push_back(I);
+            Instruction *I = cast<Instruction>(TheUse.getUser());
+            ToNotify.push_back(I);
           }
 
           DOUT << "Simply removing " << *I2
@@ -1658,10 +1658,9 @@ namespace {
           ++UI;
           Value *V = TheUse.getUser();
           if (!V->use_empty()) {
-            if (Instruction *Inst = dyn_cast<Instruction>(V)) {
-              if (aboveOrBelow(Inst))
-                opsToDef(Inst);
-            }
+            Instruction *Inst = cast<Instruction>(V);
+            if (aboveOrBelow(Inst))
+              opsToDef(Inst);
           }
         }
       }
@@ -2262,10 +2261,9 @@ namespace {
                    UE = O.LHS->use_end(); UI != UE;) {
                 Use &TheUse = UI.getUse();
                 ++UI;
-                if (Instruction *I = dyn_cast<Instruction>(TheUse.getUser())) {
-                  if (aboveOrBelow(I))
-                    opsToDef(I);
-                }
+                Instruction *I = cast<Instruction>(TheUse.getUser());
+                if (aboveOrBelow(I))
+                  opsToDef(I);
               }
             }
             if (Instruction *I2 = dyn_cast<Instruction>(O.RHS)) {
@@ -2277,10 +2275,9 @@ namespace {
                    UE = O.RHS->use_end(); UI != UE;) {
                 Use &TheUse = UI.getUse();
                 ++UI;
-                if (Instruction *I = dyn_cast<Instruction>(TheUse.getUser())) {
-                  if (aboveOrBelow(I))
-                    opsToDef(I);
-                }
+                Instruction *I = cast<Instruction>(TheUse.getUser());
+                if (aboveOrBelow(I))
+                  opsToDef(I);
               }
             }
           }
diff --git a/lib/Transforms/Scalar/SCCP.cpp b/lib/Transforms/Scalar/SCCP.cpp
index d73519c..3deee54 100644
--- a/lib/Transforms/Scalar/SCCP.cpp
+++ b/lib/Transforms/Scalar/SCCP.cpp
@@ -639,8 +639,8 @@ void SCCPSolver::visitReturnInst(ReturnInst &I) {
       DenseMap<std::pair<Function*, unsigned>, LatticeVal>::iterator
         It = TrackedMultipleRetVals.find(std::make_pair(F, i));
       if (It == TrackedMultipleRetVals.end()) break;
-      Value *Val = FindInsertedValue(I.getOperand(0), i);
-      mergeInValue(It->second, F, getValueState(Val));
+      if (Value *Val = FindInsertedValue(I.getOperand(0), i))
+        mergeInValue(It->second, F, getValueState(Val));
     }
   }
 }
diff --git a/lib/Transforms/Scalar/ScalarReplAggregates.cpp b/lib/Transforms/Scalar/ScalarReplAggregates.cpp
index 9935f12..7143c7b 100644
--- a/lib/Transforms/Scalar/ScalarReplAggregates.cpp
+++ b/lib/Transforms/Scalar/ScalarReplAggregates.cpp
@@ -1228,7 +1228,8 @@ void SROA::CleanupAllocaUsers(AllocationInst *AI) {
     User *U = *UI++;
     if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(U))
       CleanupGEP(GEPI);
-    else if (Instruction *I = dyn_cast<Instruction>(U)) {
+    else {
+      Instruction *I = cast<Instruction>(U);
       SmallVector<DbgInfoIntrinsic *, 2> DbgInUses;
       if (!isa<StoreInst>(I) && OnlyUsedByDbgInfoIntrinsics(I, &DbgInUses)) {
         // Safe to remove debug info uses.
diff --git a/lib/VMCore/Type.cpp b/lib/VMCore/Type.cpp
index 11b93ff..a1e6c42 100644
--- a/lib/VMCore/Type.cpp
+++ b/lib/VMCore/Type.cpp
@@ -135,8 +135,8 @@ bool Type::isFPOrFPVector() const {
   return cast<VectorType>(this)->getElementType()->isFloatingPoint();
 }
 
-// canLosslesllyBitCastTo - Return true if this type can be converted to
-// 'Ty' without any reinterpretation of bits.  For example, uint to int.
+// canLosslesslyBitCastTo - Return true if this type can be converted to
+// 'Ty' without any reinterpretation of bits.  For example, i8* to i32*.
 //
 bool Type::canLosslesslyBitCastTo(const Type *Ty) const {
   // Identity cast means no change so return true
@@ -318,6 +318,17 @@ bool FunctionType::isValidReturnType(const Type *RetTy) {
   return true;
 }
 
+/// isValidArgumentType - Return true if the specified type is valid as an
+/// argument type.
+bool FunctionType::isValidArgumentType(const Type *ArgTy) {
+  if ((!ArgTy->isFirstClassType() && !isa<OpaqueType>(ArgTy)) ||
+      (isa<PointerType>(ArgTy) &&
+       cast<PointerType>(ArgTy)->getElementType() == Type::MetadataTy))
+    return false;
+
+  return true;
+}
+
 FunctionType::FunctionType(const Type *Result,
                            const std::vector<const Type*> &Params,
                            bool IsVarArgs)
@@ -331,11 +342,8 @@ FunctionType::FunctionType(const Type *Result,
   new (&ContainedTys[0]) PATypeHandle(Result, this);
 
   for (unsigned i = 0; i != Params.size(); ++i) {
-    assert((Params[i]->isFirstClassType() || isa<OpaqueType>(Params[i])) &&
-           "Function arguments must be value types!");
-    assert((!isa<PointerType>(Params[i]) ||
-            cast<PointerType>(Params[i])->getElementType() != Type::MetadataTy)
-           && "Attempt to use metadata* as function argument type!");
+    assert(isValidArgumentType(Params[i]) &&
+           "Not a valid type for function argument!");
     new (&ContainedTys[i+1]) PATypeHandle(Params[i], this);
     isAbstract |= Params[i]->isAbstract();
   }
@@ -352,12 +360,8 @@ StructType::StructType(const std::vector<const Type*> &Types, bool isPacked)
   bool isAbstract = false;
   for (unsigned i = 0; i < Types.size(); ++i) {
     assert(Types[i] && "<null> type for structure field!");
-    assert(Types[i] != Type::VoidTy && "Void type for structure field!");
-    assert(Types[i] != Type::LabelTy && "Label type for structure field!");
-    assert(Types[i] != Type::MetadataTy && "Metadata type for structure field");
-    assert((!isa<PointerType>(Types[i]) ||
-            cast<PointerType>(Types[i])->getElementType() != Type::MetadataTy)
-           && "Type 'metadata*' is invalid for structure field.");
+    assert(isValidElementType(Types[i]) &&
+           "Invalid type for structure element!");
     new (&ContainedTys[i]) PATypeHandle(Types[i], this);
     isAbstract |= Types[i]->isAbstract();
   }
@@ -379,8 +383,7 @@ VectorType::VectorType(const Type *ElType, unsigned NumEl)
   NumElements = NumEl;
   setAbstract(ElType->isAbstract());
   assert(NumEl > 0 && "NumEl of a VectorType must be greater than 0");
-  assert((ElType->isInteger() || ElType->isFloatingPoint() || 
-          isa<OpaqueType>(ElType)) && 
+  assert(isValidElementType(ElType) &&
          "Elements of a VectorType must be a primitive type");
 
 }
@@ -1051,12 +1054,7 @@ static ManagedStatic<TypeMap<ArrayValType, ArrayType> > ArrayTypes;
 
 ArrayType *ArrayType::get(const Type *ElementType, uint64_t NumElements) {
   assert(ElementType && "Can't get array of <null> types!");
-  assert(ElementType != Type::VoidTy && "Array of void is not valid!");
-  assert(ElementType != Type::LabelTy && "Array of labels is not valid!");
-  assert(ElementType != Type::MetadataTy && "Array of metadata is not valid!");
-  assert((!isa<PointerType>(ElementType) ||
-          cast<PointerType>(ElementType)->getElementType() != Type::MetadataTy)
-         && "Array of metadata* is not valid!");
+  assert(isValidElementType(ElementType) && "Invalid type for array element!");
 
   ArrayValType AVT(ElementType, NumElements);
   ArrayType *AT = ArrayTypes->get(AVT);
@@ -1071,6 +1069,18 @@ ArrayType *ArrayType::get(const Type *ElementType, uint64_t NumElements) {
   return AT;
 }
 
+bool ArrayType::isValidElementType(const Type *ElemTy) {
+  if (ElemTy == Type::VoidTy || ElemTy == Type::LabelTy ||
+      ElemTy == Type::MetadataTy)
+    return false;
+
+  if (const PointerType *PTy = dyn_cast<PointerType>(ElemTy))
+    if (PTy->getElementType() == Type::MetadataTy)
+      return false;
+
+  return true;
+}
+
 
 //===----------------------------------------------------------------------===//
 // Vector Type Factory...
@@ -1115,6 +1125,14 @@ VectorType *VectorType::get(const Type *ElementType, unsigned NumElements) {
   return PT;
 }
 
+bool VectorType::isValidElementType(const Type *ElemTy) {
+  if (ElemTy->isInteger() || ElemTy->isFloatingPoint() ||
+      isa<OpaqueType>(ElemTy))
+    return true;
+
+  return false;
+}
+
 //===----------------------------------------------------------------------===//
 // Struct Type Factory...
 //
@@ -1181,6 +1199,17 @@ StructType *StructType::get(const Type *type, ...) {
   return llvm::StructType::get(StructFields);
 }
 
+bool StructType::isValidElementType(const Type *ElemTy) {
+  if (ElemTy == Type::VoidTy || ElemTy == Type::LabelTy ||
+      ElemTy == Type::MetadataTy)
+    return false;
+
+  if (const PointerType *PTy = dyn_cast<PointerType>(ElemTy))
+    if (PTy->getElementType() == Type::MetadataTy)
+      return false;
+
+  return true;
+}
 
 
 //===----------------------------------------------------------------------===//
@@ -1217,10 +1246,7 @@ PointerType *PointerType::get(const Type *ValueType, unsigned AddressSpace) {
   assert(ValueType && "Can't get a pointer to <null> type!");
   assert(ValueType != Type::VoidTy &&
          "Pointer to void is not valid, use i8* instead!");
-  assert(ValueType != Type::LabelTy && "Pointer to label is not valid!");
-  assert((!isa<PointerType>(ValueType) ||
-          cast<PointerType>(ValueType)->getElementType() != Type::MetadataTy)
-         && "Pointer to metadata* is not valid!");
+  assert(isValidElementType(ValueType) && "Invalid type for pointer element!");
   PointerValType PVT(ValueType, AddressSpace);
 
   PointerType *PT = PointerTypes->get(PVT);
@@ -1239,6 +1265,18 @@ PointerType *Type::getPointerTo(unsigned addrs) const {
   return PointerType::get(this, addrs);
 }
 
+bool PointerType::isValidElementType(const Type *ElemTy) {
+  if (ElemTy == Type::VoidTy || ElemTy == Type::LabelTy)
+    return false;
+
+  if (const PointerType *PTy = dyn_cast<PointerType>(ElemTy))
+    if (PTy->getElementType() == Type::MetadataTy)
+      return false;
+
+  return true;
+}
+
+
 //===----------------------------------------------------------------------===//
 //                     Derived Type Refinement Functions
 //===----------------------------------------------------------------------===//
diff --git a/lib/VMCore/Verifier.cpp b/lib/VMCore/Verifier.cpp
index b047d0c..b1297ff 100644
--- a/lib/VMCore/Verifier.cpp
+++ b/lib/VMCore/Verifier.cpp
@@ -280,7 +280,6 @@ namespace {
                      bool isReturnValue, const Value *V);
     void VerifyFunctionAttrs(const FunctionType *FT, const AttrListPtr &Attrs,
                              const Value *V);
-    bool VerifyMDNode(const MDNode *N);
 
     void WriteValue(const Value *V) {
       if (!V) return;
@@ -380,24 +379,22 @@ void Verifier::visitGlobalVariable(GlobalVariable &GV) {
     // Verify that any metadata used in a global initializer points only to
     // other globals.
     if (MDNode *FirstNode = dyn_cast<MDNode>(GV.getInitializer())) {
-      if (VerifyMDNode(FirstNode)) {
-        SmallVector<const MDNode *, 4> NodesToAnalyze;
-        NodesToAnalyze.push_back(FirstNode);
-        while (!NodesToAnalyze.empty()) {
-          const MDNode *N = NodesToAnalyze.back();
-          NodesToAnalyze.pop_back();
-
-          for (MDNode::const_elem_iterator I = N->elem_begin(),
-                 E = N->elem_end(); I != E; ++I)
-            if (const Value *V = *I) {
-              if (const MDNode *Next = dyn_cast<MDNode>(V))
-                NodesToAnalyze.push_back(Next);
-              else
-                Assert3(isa<Constant>(V),
-                        "reference to instruction from global metadata node",
-                        &GV, N, V);
-            }
-        }
+      SmallVector<const MDNode *, 4> NodesToAnalyze;
+      NodesToAnalyze.push_back(FirstNode);
+      while (!NodesToAnalyze.empty()) {
+        const MDNode *N = NodesToAnalyze.back();
+        NodesToAnalyze.pop_back();
+
+        for (MDNode::const_elem_iterator I = N->elem_begin(),
+               E = N->elem_end(); I != E; ++I)
+          if (const Value *V = *I) {
+            if (const MDNode *Next = dyn_cast<MDNode>(V))
+              NodesToAnalyze.push_back(Next);
+            else
+              Assert3(isa<Constant>(V),
+                      "reference to instruction from global metadata node",
+                      &GV, N, V);
+          }
       }
     }
   } else {
@@ -1708,44 +1705,6 @@ void Verifier::VerifyIntrinsicPrototype(Intrinsic::ID ID, Function *F,
           "Intrinsic has wrong parameter attributes!", F);
 }
 
-/// Verify that an MDNode is not cyclic.
-bool Verifier::VerifyMDNode(const MDNode *N) {
-  if (N->elem_empty()) return true;
-
-  // The current DFS path through the nodes. Node and element number.
-  typedef std::pair<const MDNode *, MDNode::const_elem_iterator> Edge;
-  SmallVector<Edge, 8> Path;
-
-  Path.push_back(std::make_pair(N, N->elem_begin()));
-  while (!Path.empty()) {
-    Edge &e = Path.back();
-    const MDNode *&e_N = e.first;
-    MDNode::const_elem_iterator &e_I = e.second;
-
-    if (e_N->elem_end() == e_I) {
-      Path.pop_back();
-      continue;
-    }
-
-    for (MDNode::const_elem_iterator e_E = e_N->elem_end(); e_I != e_E; ++e_I) {
-      if (const MDNode *C = dyn_cast_or_null<MDNode>(e_I->operator Value*())) {
-        // Is child MDNode C already in the Path?
-        for (SmallVectorImpl<Edge>::iterator I = Path.begin(), E = Path.end();
-             I != E; ++I) {
-          if (I->first != C) {
-            CheckFailed("MDNode is cyclic.", C);
-            return false;
-          }
-        }
-
-        Path.push_back(std::make_pair(C, C->elem_begin()));
-        break;
-      }
-    }
-  }
-  return true;
-}
-
 
 //===----------------------------------------------------------------------===//
 //  Implement the public interfaces to this file...
diff --git a/test/CodeGen/PowerPC/vec_shift.ll b/test/CodeGen/PowerPC/vec_shift.ll
new file mode 100644
index 0000000..0cc699c
--- /dev/null
+++ b/test/CodeGen/PowerPC/vec_shift.ll
@@ -0,0 +1,10 @@
+; RUN: llvm-as < %s | llc  -march=ppc32 -mcpu=g5
+; PR3628
+
+define void @update(<4 x i32> %val, <4 x i32>* %dst) nounwind {
+entry:
+	%shl = shl <4 x i32> %val, < i32 4, i32 3, i32 2, i32 1 >
+	%shr = ashr <4 x i32> %shl, < i32 1, i32 2, i32 3, i32 4 >
+	store <4 x i32> %shr, <4 x i32>* %dst
+	ret void
+}
diff --git a/test/CodeGen/X86/vec_loadsingles.ll b/test/CodeGen/X86/vec_loadsingles.ll
new file mode 100644
index 0000000..6712276
--- /dev/null
+++ b/test/CodeGen/X86/vec_loadsingles.ll
@@ -0,0 +1,12 @@
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 | grep movq
+
+define <4 x float> @a(<4 x float> %a, float* nocapture %p) nounwind readonly {
+entry:
+	%tmp1 = load float* %p
+	%vecins = insertelement <4 x float> undef, float %tmp1, i32 0
+	%add.ptr = getelementptr float* %p, i32 1
+	%tmp5 = load float* %add.ptr
+	%vecins7 = insertelement <4 x float> %vecins, float %tmp5, i32 1
+	ret <4 x float> %vecins7
+}
+
diff --git a/test/CodeGen/X86/vec_set-5.ll b/test/CodeGen/X86/vec_set-5.ll
index 4fc652c..d332970 100644
--- a/test/CodeGen/X86/vec_set-5.ll
+++ b/test/CodeGen/X86/vec_set-5.ll
@@ -1,7 +1,6 @@
 ; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 -o %t -f
 ; RUN: grep movlhps   %t | count 1
-; RUN: grep movq      %t | count 1
-; RUN: grep movsd     %t | count 1
+; RUN: grep movq      %t | count 2
 
 define <4 x float> @test1(float %a, float %b) nounwind {
 	%tmp = insertelement <4 x float> zeroinitializer, float %a, i32 0		; <<4 x float>> [#uses=1]
diff --git a/test/CodeGen/X86/vec_set-6.ll b/test/CodeGen/X86/vec_set-6.ll
index 02df526..c7b6747 100644
--- a/test/CodeGen/X86/vec_set-6.ll
+++ b/test/CodeGen/X86/vec_set-6.ll
@@ -1,6 +1,6 @@
 ; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 -o %t -f
 ; RUN: grep movss    %t | count 1
-; RUN: grep movups   %t | count 1
+; RUN: grep movq     %t | count 1
 ; RUN: grep shufps   %t | count 1
 
 define <4 x float> @test(float %a, float %b, float %c) nounwind {
diff --git a/test/Transforms/InstCombine/vec_demanded_elts-3.ll b/test/Transforms/InstCombine/vec_demanded_elts-3.ll
new file mode 100644
index 0000000..eba3629
--- /dev/null
+++ b/test/Transforms/InstCombine/vec_demanded_elts-3.ll
@@ -0,0 +1,14 @@
+; RUN: llvm-as < %s | opt -instcombine | llvm-dis | not grep load
+; PR4340
+
+define void @vac(<4 x float>* nocapture %a) nounwind {
+entry:
+	%tmp1 = load <4 x float>* %a		; <<4 x float>> [#uses=1]
+	%vecins = insertelement <4 x float> %tmp1, float 0.000000e+00, i32 0	; <<4 x float>> [#uses=1]
+	%vecins4 = insertelement <4 x float> %vecins, float 0.000000e+00, i32 1; <<4 x float>> [#uses=1]
+	%vecins6 = insertelement <4 x float> %vecins4, float 0.000000e+00, i32 2; <<4 x float>> [#uses=1]
+	%vecins8 = insertelement <4 x float> %vecins6, float 0.000000e+00, i32 3; <<4 x float>> [#uses=1]
+	store <4 x float> %vecins8, <4 x float>* %a
+	ret void
+}
+
diff --git a/tools/gold/gold-plugin.cpp b/tools/gold/gold-plugin.cpp
index 86d3fd3..46b1717 100644
--- a/tools/gold/gold-plugin.cpp
+++ b/tools/gold/gold-plugin.cpp
@@ -45,6 +45,7 @@ namespace {
 
   bool generate_api_file = false;
   const char *gcc_path = NULL;
+  const char *as_path = NULL;
 
   struct claimed_file {
     lto_module_t M;
@@ -109,6 +110,13 @@ ld_plugin_status onload(ld_plugin_tv *tv) {
           } else {
             gcc_path = strdup(tv->tv_u.tv_string + 4);
           }
+        } else if (strncmp("as=", tv->tv_u.tv_string, 3) == 0) {
+          if (as_path) {
+            (*message)(LDPL_WARNING, "Path to as specified twice. "
+                       "Discarding %s", tv->tv_u.tv_string);
+          } else {
+            as_path = strdup(tv->tv_u.tv_string + 3);
+          }
         } else {
           (*message)(LDPL_WARNING, "Ignoring flag %s", tv->tv_u.tv_string);
         }
@@ -346,6 +354,8 @@ ld_plugin_status all_symbols_read_hook(void) {
   lto_codegen_set_debug_model(cg, LTO_DEBUG_MODEL_DWARF);
   if (gcc_path)
     lto_codegen_set_gcc_path(cg, gcc_path);
+  if (as_path)
+    lto_codegen_set_assembler_path(cg, as_path);
 
   size_t bufsize = 0;
   const char *buffer = static_cast<const char *>(lto_codegen_compile(cg,
diff --git a/tools/llvmc/example/mcc16/Makefile b/tools/llvmc/example/mcc16/Makefile
new file mode 100644
index 0000000..efc9d2d
--- /dev/null
+++ b/tools/llvmc/example/mcc16/Makefile
@@ -0,0 +1,24 @@
+##===- llvmc/example/mcc16/Makefile ------------------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open
+# Source License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+
+# Change this so that $(BASE_LEVEL)/Makefile.common refers to
+# $LLVM_DIR/Makefile.common.
+export LLVMC_BASE_LEVEL = ../../../..
+
+# Change this to the name of your LLVMC-based driver.
+export LLVMC_BASED_DRIVER_NAME = mcc16
+
+# List your plugin names here
+export LLVMC_BUILTIN_PLUGINS = PIC16Base
+
+LEVEL = $(LLVMC_BASE_LEVEL)
+
+DIRS = plugins driver
+
+include $(LEVEL)/Makefile.common
diff --git a/tools/llvmc/example/mcc16/README b/tools/llvmc/example/mcc16/README
new file mode 100644
index 0000000..eeef6a4
--- /dev/null
+++ b/tools/llvmc/example/mcc16/README
@@ -0,0 +1,75 @@
+This is a basic compiler driver for the PIC16 toolchain that shows how to create
+your own llvmc-based drivers. It is based on the example/Skeleton template.
+
+The PIC16 toolchain looks like this:
+
+clang-cc (FE) -> llvm-ld (optimizer) -> llc (codegen) -> native-as -> native-ld
+
+Following features were requested by Sanjiv:
+
+From: Sanjiv Gupta <sanjiv.gupta <at> microchip.com>
+Subject: Re: llvmc for PIC16
+Newsgroups: gmane.comp.compilers.llvm.devel
+Date: 2009-06-05 06:51:14 GMT
+
+The salient features that we want to have in the driver are:
+1. llvm-ld will be used as "The Optimizer".
+2. If the user has specified to generate the final executable, then
+llvm-ld should run on all the .bc files generated by clang and create a
+single optimized .bc file for further tools.
+3. -Wo <options> - pass optimizations to the llvm-ld
+4. mcc16 -Wl <options> - pass options to native linker.
+5. mcc16 -Wa <options> - pass options to native assembler.
+
+Here are some example command lines and sample command invocations as to
+what should be done.
+
+$ mcc16 -S foo.c
+// [clang-cc foo.c] -> foo.bc
+// [llvm-ld foo.bc] -> foo.opt.bc
+// [llc foo.opt.bc] -> foo.s
+
+$ mcc16 -S foo.c bar.c
+// [clang-cc foo.c] -> foo.bc
+// [llvm-ld foo.bc] -> foo.opt.bc
+// [llc foo.opt.bc] -> foo.s
+// [clang-cc bar.c] -> bar.bc
+// [llvm-ld bar.bc] -> bar.opt.bc
+// [llc bar.opt.bc] -> bar.s
+
+** Use of -g causes llvm-ld to run with -disable-opt
+$ mcc16 -S -g foo.c
+// [clang-cc foo.c] -> foo.bc
+// [llvm-ld -disable-opt foo.bc] -> foo.opt.bc
+// [llc foo.opt.bc] -> foo.s
+
+** -I is passed to clang-cc, -pre-RA-sched=list-burr to llc.
+$ mcc16 -S -g -I ../include -pre-RA-sched=list-burr foo.c
+// [clang-cc -I ../include foo.c] -> foo.bc
+// [llvm-ld -disable-opt foo.bc] -> foo.opt.bc
+// [llc -pre-RA-sched=list-burr foo.opt.bc] -> foo.s
+
+** -Wo passes options to llvm-ld
+$ mcc16 -Wo=opt1,opt2 -S -I ../include -pre-RA-sched=list-burr foo.c
+// [clang-cc -I ../include foo.c] -> foo.bc
+// [llvm-ld -opt1 -opt2 foo.bc] -> foo.opt.bc
+// [llc -pre-RA-sched=list-burr foo.opt.bc] -> foo.s
+
+** -Wa passes options to native as.
+$ mcc16 -c foo.c -Wa=opt1
+// [clang-cc foo.c] -> foo.bc
+// [llvm-ld foo.bc] -> foo.opt.bc
+// [llc foo.opt.bc] -> foo.s
+// [native-as -opt1 foo.s] -> foo.o
+
+$ mcc16 -Wo=opt1 -Wl=opt2 -Wa=opt3 foo.c bar.c
+// [clang-cc foo.c] -> foo.bc
+// [clang-cc bar.c] -> bar.bc
+// [llvm-ld -opt1 foo.bc bar.bc] -> a.out.bc
+// [llc a.out.bc] -> a.out.s
+// [native-as -opt3 a.out.s] -> a.out.o
+// [native-ld -opt2 a.out.o] -> a.out
+
+Is this achievable by a tablegen based driver ?
+
+- Sanjiv
diff --git a/tools/llvmc/example/mcc16/driver/Main.cpp b/tools/llvmc/example/mcc16/driver/Main.cpp
new file mode 100644
index 0000000..b1f5b67
--- /dev/null
+++ b/tools/llvmc/example/mcc16/driver/Main.cpp
@@ -0,0 +1,14 @@
+//===--- Main.cpp - The LLVM Compiler Driver -------------------*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open
+// Source License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+//  Just include CompilerDriver/Main.inc.
+//
+//===----------------------------------------------------------------------===//
+
+#include "llvm/CompilerDriver/Main.inc"
diff --git a/tools/llvmc/example/mcc16/driver/Makefile b/tools/llvmc/example/mcc16/driver/Makefile
new file mode 100644
index 0000000..ed9ebfd
--- /dev/null
+++ b/tools/llvmc/example/mcc16/driver/Makefile
@@ -0,0 +1,22 @@
+##===- llvmc/example/mcc16/driver/Makefile -----------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open
+# Source License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+
+LEVEL = $(LLVMC_BASE_LEVEL)/..
+
+TOOLNAME = $(LLVMC_BASED_DRIVER_NAME)
+LLVMLIBS = CompilerDriver
+
+ifneq ($(LLVMC_BUILTIN_PLUGINS),)
+USEDLIBS += $(patsubst %,plugin_llvmc_%,$(LLVMC_BUILTIN_PLUGINS))
+endif
+
+LINK_COMPONENTS = support system
+REQUIRES_EH := 1
+
+include $(LEVEL)/Makefile.common
diff --git a/tools/llvmc/example/mcc16/plugins/Makefile b/tools/llvmc/example/mcc16/plugins/Makefile
new file mode 100644
index 0000000..fb07f23
--- /dev/null
+++ b/tools/llvmc/example/mcc16/plugins/Makefile
@@ -0,0 +1,18 @@
+##===- llvmc/example/Skeleton/plugins/Makefile -------------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open
+# Source License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+
+LEVEL = $(LLVMC_BASE_LEVEL)/..
+
+ifneq ($(LLVMC_BUILTIN_PLUGINS),)
+DIRS = $(LLVMC_BUILTIN_PLUGINS)
+endif
+
+export LLVMC_BUILTIN_PLUGIN=1
+
+include $(LEVEL)/Makefile.common
diff --git a/tools/llvmc/example/mcc16/plugins/PIC16Base/Makefile b/tools/llvmc/example/mcc16/plugins/PIC16Base/Makefile
new file mode 100644
index 0000000..5d785fd
--- /dev/null
+++ b/tools/llvmc/example/mcc16/plugins/PIC16Base/Makefile
@@ -0,0 +1,17 @@
+##===- llvmc/example/Skeleton/plugins/Plugin/Makefile ------*- Makefile -*-===##
+#
+#                     The LLVM Compiler Infrastructure
+#
+# This file is distributed under the University of Illinois Open Source
+# License. See LICENSE.TXT for details.
+#
+##===----------------------------------------------------------------------===##
+
+LEVEL = $(LLVMC_BASE_LEVEL)/../..
+
+# Change this to the name of your plugin.
+LLVMC_PLUGIN = PIC16Base
+
+BUILT_SOURCES = AutoGenerated.inc
+
+include $(LEVEL)/Makefile.common
diff --git a/tools/llvmc/example/mcc16/plugins/PIC16Base/PIC16Base.td b/tools/llvmc/example/mcc16/plugins/PIC16Base/PIC16Base.td
new file mode 100644
index 0000000..0b8a22b
--- /dev/null
+++ b/tools/llvmc/example/mcc16/plugins/PIC16Base/PIC16Base.td
@@ -0,0 +1,116 @@
+//===- PIC16Base.td - PIC16 toolchain driver ---------------*- tablegen -*-===//
+//
+// A basic driver for the PIC16 toolchain.
+//
+//===----------------------------------------------------------------------===//
+
+include "llvm/CompilerDriver/Common.td"
+
+// Options
+
+def OptionList : OptionList<[
+ (switch_option "g",
+    (help "Disable optimizations")),
+ (switch_option "S",
+    (help "Stop after compilation, do not assemble")),
+ (parameter_option "I",
+    (help "Add a directory to include path")),
+ (parameter_option "pre-RA-sched",
+    (help "Example of an option that is passed to llc")),
+ (prefix_list_option "Wa,",
+    (help "Pass options to native assembler")),
+ (prefix_list_option "Wl,",
+    (help "Pass options to native linker")),
+ (prefix_list_option "Wllc,",
+    (help "Pass options to llc")),
+ (prefix_list_option "Wo,",
+    (help "Pass options to llvm-ld"))
+]>;
+
+// Tools
+
+def clang_cc : Tool<[
+ (in_language "c"),
+ (out_language "llvm-bitcode"),
+ (output_suffix "bc"),
+ (cmd_line "clang-cc $INFILE -o $OUTFILE"),
+ (actions (case
+          (not_empty "I"), (forward "I"))),
+ (sink)
+]>;
+
+def llvm_ld : Tool<[
+ (in_language "llvm-bitcode"),
+ (out_language "llvm-bitcode"),
+ (output_suffix "bc"),
+ (cmd_line "llvm-ld $INFILE -o $OUTFILE"),
+ (actions (case
+          (switch_on "g"), (append_cmd "-disable-opt"),
+          (not_empty "Wo,"), (unpack_values "Wo,")))
+]>;
+
+def llvm_ld_lto : Tool<[
+ (in_language "llvm-bitcode"),
+ (out_language "llvm-bitcode"),
+ (output_suffix "bc"),
+ (cmd_line "llvm-ld $INFILE -o $OUTFILE"),
+ (actions (case
+          (switch_on "g"), (append_cmd "-disable-opt"),
+          (not_empty "Wo,"), (unpack_values "Wo,"))),
+ (join)
+]>;
+
+def llc : Tool<[
+ (in_language "llvm-bitcode"),
+ (out_language "assembler"),
+ (output_suffix "s"),
+ (cmd_line "llc -f $INFILE -o $OUTFILE"),
+ (actions (case
+          (switch_on "S"), (stop_compilation),
+          (not_empty "Wllc,"), (unpack_values "Wllc,"),
+          (not_empty "pre-RA-sched"), (forward "pre-RA-sched")))
+]>;
+
+def native_as : Tool<[
+ (in_language "assembler"),
+ (out_language "object-code"),
+ (output_suffix "o"),
+ (cmd_line "native-as $INFILE -o $OUTFILE"),
+ (actions (case
+          (not_empty "Wa,"), (unpack_values "Wa,")))
+]>;
+
+def native_ld : Tool<[
+ (in_language "object-code"),
+ (out_language "executable"),
+ (output_suffix "out"),
+ (cmd_line "native-ld $INFILE -o $OUTFILE"),
+ (actions (case
+          (not_empty "Wl,"), (unpack_values "Wl,"))),
+ (join)
+]>;
+
+// Language map
+
+def LanguageMap : LanguageMap<[
+    LangToSuffixes<"c", ["c"]>,
+    LangToSuffixes<"c-cpp-output", ["i"]>,
+    LangToSuffixes<"assembler", ["s"]>,
+    LangToSuffixes<"assembler-with-cpp", ["S"]>,
+    LangToSuffixes<"llvm-assembler", ["ll"]>,
+    LangToSuffixes<"llvm-bitcode", ["bc"]>,
+    LangToSuffixes<"object-code", ["o"]>,
+    LangToSuffixes<"executable", ["out"]>
+]>;
+
+// Compilation graph
+
+def CompilationGraph : CompilationGraph<[
+    Edge<"root", "clang_cc">,
+    Edge<"clang_cc", "llvm_ld_lto">,
+    Edge<"llvm_ld_lto", "llc">,
+    OptionalEdge<"clang_cc", "llvm_ld", (case (switch_on "S"), (inc_weight))>,
+    Edge<"llvm_ld", "llc">,
+    Edge<"llc", "native_as">,
+    Edge<"native_as", "native_ld">
+]>;
diff --git a/tools/llvmc/example/mcc16/plugins/PIC16Base/PluginMain.cpp b/tools/llvmc/example/mcc16/plugins/PIC16Base/PluginMain.cpp
new file mode 100644
index 0000000..add8acb
--- /dev/null
+++ b/tools/llvmc/example/mcc16/plugins/PIC16Base/PluginMain.cpp
@@ -0,0 +1 @@
+#include "AutoGenerated.inc"