Import clang, at r72995.

54 files changed, 2639 insertions, 1805 deletions

diff --git a/lib/AST/ASTContext.cpp b/lib/AST/ASTContext.cpp
index b36d1f3..e6dea7c 100644
--- a/lib/AST/ASTContext.cpp
+++ b/lib/AST/ASTContext.cpp
@@ -143,7 +143,7 @@ void ASTContext::InitBuiltinTypes() {
   // C99 6.2.5p2.
   InitBuiltinType(BoolTy,              BuiltinType::Bool);
   // C99 6.2.5p3.
-  if (Target.isCharSigned())
+  if (LangOpts.CharIsSigned)
     InitBuiltinType(CharTy,            BuiltinType::Char_S);
   else
     InitBuiltinType(CharTy,            BuiltinType::Char_U);
@@ -613,6 +613,20 @@ void ASTContext::CollectObjCIvars(const ObjCInterfaceDecl *OI,
   CollectLocalObjCIvars(this, OI, Fields);
 }
 
+/// ShallowCollectObjCIvars -
+/// Collect all ivars, including those synthesized, in the current class.
+///
+void ASTContext::ShallowCollectObjCIvars(const ObjCInterfaceDecl *OI,
+                                 llvm::SmallVectorImpl<ObjCIvarDecl*> &Ivars,
+                                 bool CollectSynthesized) {
+  for (ObjCInterfaceDecl::ivar_iterator I = OI->ivar_begin(),
+         E = OI->ivar_end(); I != E; ++I) {
+     Ivars.push_back(*I);
+  }
+  if (CollectSynthesized)
+    CollectSynthesizedIvars(OI, Ivars);
+}
+
 void ASTContext::CollectProtocolSynthesizedIvars(const ObjCProtocolDecl *PD,
                                 llvm::SmallVectorImpl<ObjCIvarDecl*> &Ivars) {
   for (ObjCContainerDecl::prop_iterator I = PD->prop_begin(*this),
@@ -645,6 +659,38 @@ void ASTContext::CollectSynthesizedIvars(const ObjCInterfaceDecl *OI,
   }
 }
 
+unsigned ASTContext::CountProtocolSynthesizedIvars(const ObjCProtocolDecl *PD) {
+  unsigned count = 0;
+  for (ObjCContainerDecl::prop_iterator I = PD->prop_begin(*this),
+       E = PD->prop_end(*this); I != E; ++I)
+    if ((*I)->getPropertyIvarDecl())
+      ++count;
+
+  // Also look into nested protocols.
+  for (ObjCProtocolDecl::protocol_iterator P = PD->protocol_begin(),
+       E = PD->protocol_end(); P != E; ++P)
+    count += CountProtocolSynthesizedIvars(*P);
+  return count;
+}
+
+unsigned ASTContext::CountSynthesizedIvars(const ObjCInterfaceDecl *OI)
+{
+  unsigned count = 0;
+  for (ObjCInterfaceDecl::prop_iterator I = OI->prop_begin(*this),
+       E = OI->prop_end(*this); I != E; ++I) {
+    if ((*I)->getPropertyIvarDecl())
+      ++count;
+  }
+  // Also look into interface's protocol list for properties declared
+  // in the protocol and whose ivars are synthesized.
+  for (ObjCInterfaceDecl::protocol_iterator P = OI->protocol_begin(),
+       PE = OI->protocol_end(); P != PE; ++P) {
+    ObjCProtocolDecl *PD = (*P);
+    count += CountProtocolSynthesizedIvars(PD);
+  }
+  return count;
+}
+
 /// getInterfaceLayoutImpl - Get or compute information about the
 /// layout of the given interface.
 ///
@@ -664,14 +710,13 @@ ASTContext::getObjCLayout(const ObjCInterfaceDecl *D,
   unsigned FieldCount = D->ivar_size();
   // Add in synthesized ivar count if laying out an implementation.
   if (Impl) {
-    llvm::SmallVector<ObjCIvarDecl*, 16> Ivars;
-    CollectSynthesizedIvars(D, Ivars);
-    FieldCount += Ivars.size();
+    unsigned SynthCount = CountSynthesizedIvars(D);
+    FieldCount += SynthCount;
     // If there aren't any sythesized ivars then reuse the interface
     // entry. Note we can't cache this because we simply free all
     // entries later; however we shouldn't look up implementations
     // frequently.
-    if (FieldCount == D->ivar_size())
+    if (SynthCount == 0)
       return getObjCLayout(D, 0);
   }
 
@@ -701,20 +746,11 @@ ASTContext::getObjCLayout(const ObjCInterfaceDecl *D,
 
   // Layout each ivar sequentially.
   unsigned i = 0;
-  for (ObjCInterfaceDecl::ivar_iterator IVI = D->ivar_begin(), 
-       IVE = D->ivar_end(); IVI != IVE; ++IVI) {
-    const ObjCIvarDecl* Ivar = (*IVI);
-    NewEntry->LayoutField(Ivar, i++, false, StructPacking, *this);
-  }
-  // And synthesized ivars, if this is an implementation.
-  if (Impl) {
-    // FIXME. Do we need to colltect twice?
-    llvm::SmallVector<ObjCIvarDecl*, 16> Ivars;
-    CollectSynthesizedIvars(D, Ivars);
-    for (unsigned k = 0, e = Ivars.size(); k != e; ++k)
-      NewEntry->LayoutField(Ivars[k], i++, false, StructPacking, *this);
-  }
-  
+  llvm::SmallVector<ObjCIvarDecl*, 16> Ivars;
+  ShallowCollectObjCIvars(D, Ivars, Impl);
+  for (unsigned k = 0, e = Ivars.size(); k != e; ++k)
+       NewEntry->LayoutField(Ivars[k], i++, false, StructPacking, *this);
+ 
   // Finally, round the size of the total struct up to the alignment of the
   // struct itself.
   NewEntry->FinalizeLayout();
diff --git a/lib/AST/Decl.cpp b/lib/AST/Decl.cpp
index cb3ec1f..dfec106 100644
--- a/lib/AST/Decl.cpp
+++ b/lib/AST/Decl.cpp
@@ -489,7 +489,7 @@ void FunctionDecl::setParams(ASTContext& C, ParmVarDecl **NewParamInfo,
 unsigned FunctionDecl::getMinRequiredArguments() const {
   unsigned NumRequiredArgs = getNumParams();
   while (NumRequiredArgs > 0
-         && getParamDecl(NumRequiredArgs-1)->getDefaultArg())
+         && getParamDecl(NumRequiredArgs-1)->hasDefaultArg())
     --NumRequiredArgs;
 
   return NumRequiredArgs;
diff --git a/lib/AST/DeclCXX.cpp b/lib/AST/DeclCXX.cpp
index 19f8958..94daf48 100644
--- a/lib/AST/DeclCXX.cpp
+++ b/lib/AST/DeclCXX.cpp
@@ -326,7 +326,7 @@ CXXConstructorDecl::isCopyConstructor(ASTContext &Context,
   //   const volatile X&, and either there are no other parameters
   //   or else all other parameters have default arguments (8.3.6).
   if ((getNumParams() < 1) ||
-      (getNumParams() > 1 && getParamDecl(1)->getDefaultArg() == 0))
+      (getNumParams() > 1 && !getParamDecl(1)->hasDefaultArg()))
     return false;
 
   const ParmVarDecl *Param = getParamDecl(0);
@@ -363,7 +363,7 @@ bool CXXConstructorDecl::isConvertingConstructor() const {
   return (getNumParams() == 0 && 
           getType()->getAsFunctionProtoType()->isVariadic()) ||
          (getNumParams() == 1) ||
-         (getNumParams() > 1 && getParamDecl(1)->getDefaultArg() != 0);
+         (getNumParams() > 1 && getParamDecl(1)->hasDefaultArg());
 }
 
 CXXDestructorDecl *
diff --git a/lib/AST/DeclObjC.cpp b/lib/AST/DeclObjC.cpp
index f4bb895..21aefdd 100644
--- a/lib/AST/DeclObjC.cpp
+++ b/lib/AST/DeclObjC.cpp
@@ -42,6 +42,19 @@ void ObjCListBase::set(void *const* InList, unsigned Elts, ASTContext &Ctx) {
 // ObjCInterfaceDecl
 //===----------------------------------------------------------------------===//
 
+/// getIvarDecl - This method looks up an ivar in this ContextDecl.
+///
+ObjCIvarDecl *
+ObjCContainerDecl::getIvarDecl(ASTContext &Context, IdentifierInfo *Id) const {
+  lookup_const_iterator Ivar, IvarEnd;
+  for (llvm::tie(Ivar, IvarEnd) = lookup(Context, Id);
+       Ivar != IvarEnd; ++Ivar) {
+    if (ObjCIvarDecl *ivar = dyn_cast<ObjCIvarDecl>(*Ivar))
+      return ivar;
+  }
+  return 0;
+}
+
 // Get the local instance method declared in this interface.
 ObjCMethodDecl *
 ObjCContainerDecl::getInstanceMethod(ASTContext &Context, Selector Sel) const {
@@ -139,12 +152,9 @@ ObjCIvarDecl *ObjCInterfaceDecl::lookupInstanceVariable(
   ASTContext &Context, IdentifierInfo *ID, ObjCInterfaceDecl *&clsDeclared) {
   ObjCInterfaceDecl* ClassDecl = this;
   while (ClassDecl != NULL) {
-    for (ivar_iterator I = ClassDecl->ivar_begin(), E = ClassDecl->ivar_end();
-         I != E; ++I) {
-      if ((*I)->getIdentifier() == ID) {
-        clsDeclared = ClassDecl;
-        return *I;
-      }
+    if (ObjCIvarDecl *I = ClassDecl->getIvarDecl(Context, ID)) {
+      clsDeclared = ClassDecl;
+      return I;
     }
     // look into properties.
     for (ObjCInterfaceDecl::prop_iterator I = ClassDecl->prop_begin(Context),
diff --git a/lib/AST/DeclPrinter.cpp b/lib/AST/DeclPrinter.cpp
index f29da8b..f231abf 100644
--- a/lib/AST/DeclPrinter.cpp
+++ b/lib/AST/DeclPrinter.cpp
@@ -508,8 +508,50 @@ void DeclPrinter::VisitLinkageSpecDecl(LinkageSpecDecl *D) {
 }
 
 void DeclPrinter::VisitTemplateDecl(TemplateDecl *D) {
-  // TODO: Write template parameters.
-  Out << "template <...> ";
+  Out << "template <";
+  
+  TemplateParameterList *Params = D->getTemplateParameters();
+  for (unsigned i = 0, e = Params->size(); i != e; ++i) {
+    if (i != 0)
+      Out << ", ";
+    
+    const Decl *Param = Params->getParam(i);
+    if (const TemplateTypeParmDecl *TTP = 
+          dyn_cast<TemplateTypeParmDecl>(Param)) {
+      
+      QualType ParamType = 
+        Context.getTypeDeclType(const_cast<TemplateTypeParmDecl*>(TTP));
+
+      if (TTP->wasDeclaredWithTypename())
+        Out << "typename ";
+      else
+        Out << "class ";
+
+      Out << ParamType.getAsString(Policy);
+
+      if (TTP->hasDefaultArgument()) {
+        Out << " = ";
+        Out << TTP->getDefaultArgument().getAsString(Policy);
+      };
+    } else if (const NonTypeTemplateParmDecl *NTTP = 
+                 dyn_cast<NonTypeTemplateParmDecl>(Param)) {
+      Out << NTTP->getType().getAsString(Policy);
+
+      if (IdentifierInfo *Name = NTTP->getIdentifier()) {
+        Out << ' ';
+        Out << Name->getName();
+      }
+      
+      if (NTTP->hasDefaultArgument()) {
+        Out << " = ";
+        NTTP->getDefaultArgument()->printPretty(Out, Context, 0, Policy, 
+                                                Indentation);
+      }
+    }
+  }
+  
+  Out << "> ";
+
   Visit(D->getTemplatedDecl());
 }
 
diff --git a/lib/AST/DeclTemplate.cpp b/lib/AST/DeclTemplate.cpp
index f38ee82..a534164 100644
--- a/lib/AST/DeclTemplate.cpp
+++ b/lib/AST/DeclTemplate.cpp
@@ -236,28 +236,42 @@ TemplateArgument::TemplateArgument(Expr *E) : Kind(Expression) {
 }
 
 //===----------------------------------------------------------------------===//
+// TemplateArgumentListBuilder Implementation
+//===----------------------------------------------------------------------===//
+void TemplateArgumentListBuilder::push_back(const TemplateArgument& Arg) {
+  switch (Arg.getKind()) {
+  default: break;
+  case TemplateArgument::Type:
+    assert(Arg.getAsType()->isCanonical() && "Type must be canonical!");
+    break;
+  }
+  
+  Args.push_back(Arg);
+}
+
+//===----------------------------------------------------------------------===//
 // TemplateArgumentList Implementation
 //===----------------------------------------------------------------------===//
 TemplateArgumentList::TemplateArgumentList(ASTContext &Context,
-                                           TemplateArgument *TemplateArgs,
-                                           unsigned NumTemplateArgs,
-                                           bool CopyArgs)
-  : NumArguments(NumTemplateArgs) {
+                                           TemplateArgumentListBuilder &Builder,
+                                           bool CopyArgs, bool FlattenArgs)
+  : NumArguments(Builder.flatSize()) {
   if (!CopyArgs) {
-    Arguments.setPointer(TemplateArgs);
+    Arguments.setPointer(Builder.getFlatArgumentList());
     Arguments.setInt(1);
     return;
   }
 
-  unsigned Size = sizeof(TemplateArgument) * NumTemplateArgs;
+  
+  unsigned Size = sizeof(TemplateArgument) * Builder.flatSize();
   unsigned Align = llvm::AlignOf<TemplateArgument>::Alignment;
   void *Mem = Context.Allocate(Size, Align);
   Arguments.setPointer((TemplateArgument *)Mem);
   Arguments.setInt(0);
 
   TemplateArgument *Args = (TemplateArgument *)Mem;
-  for (unsigned I = 0; I != NumTemplateArgs; ++I)
-    new (Args + I) TemplateArgument(TemplateArgs[I]);
+  for (unsigned I = 0; I != NumArguments; ++I)
+    new (Args + I) TemplateArgument(Builder.getFlatArgumentList()[I]);
 }
 
 TemplateArgumentList::~TemplateArgumentList() {
@@ -271,8 +285,7 @@ ClassTemplateSpecializationDecl::
 ClassTemplateSpecializationDecl(ASTContext &Context, Kind DK,
                                 DeclContext *DC, SourceLocation L,
                                 ClassTemplateDecl *SpecializedTemplate,
-                                TemplateArgument *TemplateArgs,
-                                unsigned NumTemplateArgs)
+                                TemplateArgumentListBuilder &Builder)
   : CXXRecordDecl(DK, 
                   SpecializedTemplate->getTemplatedDecl()->getTagKind(), 
                   DC, L,
@@ -280,7 +293,7 @@ ClassTemplateSpecializationDecl(ASTContext &Context, Kind DK,
                   // class template specializations?
                   SpecializedTemplate->getIdentifier()),
     SpecializedTemplate(SpecializedTemplate),
-    TemplateArgs(Context, TemplateArgs, NumTemplateArgs, /*CopyArgs=*/true),
+    TemplateArgs(Context, Builder, /*CopyArgs=*/true, /*FlattenArgs=*/true),
     SpecializationKind(TSK_Undeclared) {
 }
                   
@@ -288,16 +301,14 @@ ClassTemplateSpecializationDecl *
 ClassTemplateSpecializationDecl::Create(ASTContext &Context, 
                                         DeclContext *DC, SourceLocation L,
                                         ClassTemplateDecl *SpecializedTemplate,
-                                        TemplateArgument *TemplateArgs, 
-                                        unsigned NumTemplateArgs,
+                                        TemplateArgumentListBuilder &Builder,
                                    ClassTemplateSpecializationDecl *PrevDecl) {
   ClassTemplateSpecializationDecl *Result
     = new (Context)ClassTemplateSpecializationDecl(Context, 
                                                    ClassTemplateSpecialization,
                                                    DC, L, 
                                                    SpecializedTemplate,
-                                                   TemplateArgs, 
-                                                   NumTemplateArgs);
+                                                   Builder);
   Context.getTypeDeclType(Result, PrevDecl);
   return Result;
 }
@@ -310,14 +321,13 @@ ClassTemplatePartialSpecializationDecl::
 Create(ASTContext &Context, DeclContext *DC, SourceLocation L,
        TemplateParameterList *Params,
        ClassTemplateDecl *SpecializedTemplate,
-       TemplateArgument *TemplateArgs, unsigned NumTemplateArgs,
+       TemplateArgumentListBuilder &Builder,
        ClassTemplatePartialSpecializationDecl *PrevDecl) {
   ClassTemplatePartialSpecializationDecl *Result
     = new (Context)ClassTemplatePartialSpecializationDecl(Context, 
                                                           DC, L, Params,
                                                           SpecializedTemplate,
-                                                          TemplateArgs, 
-                                                          NumTemplateArgs);
+                                                          Builder);
   Result->setSpecializationKind(TSK_ExplicitSpecialization);
   Context.getTypeDeclType(Result, PrevDecl);
   return Result;
diff --git a/lib/AST/ExprCXX.cpp b/lib/AST/ExprCXX.cpp
index 4a15245..8fd66a2 100644
--- a/lib/AST/ExprCXX.cpp
+++ b/lib/AST/ExprCXX.cpp
@@ -305,10 +305,11 @@ void CXXConstructExpr::Destroy(ASTContext &C) {
 
 CXXExprWithTemporaries::CXXExprWithTemporaries(Expr *subexpr, 
                                                CXXTemporary **temps, 
-                                               unsigned numtemps)
+                                               unsigned numtemps,
+                                               bool destroytemps)
 : Expr(CXXExprWithTemporariesClass, subexpr->getType(),
        subexpr->isTypeDependent(), subexpr->isValueDependent()), 
-  SubExpr(subexpr), Temps(0), NumTemps(numtemps) {
+  SubExpr(subexpr), Temps(0), NumTemps(numtemps), DestroyTemps(destroytemps) {
   if (NumTemps > 0) {
     Temps = new CXXTemporary*[NumTemps];
     for (unsigned i = 0; i < NumTemps; ++i)
@@ -319,8 +320,10 @@ CXXExprWithTemporaries::CXXExprWithTemporaries(Expr *subexpr,
 CXXExprWithTemporaries *CXXExprWithTemporaries::Create(ASTContext &C, 
                                                        Expr *SubExpr,
                                                        CXXTemporary **Temps, 
-                                                       unsigned NumTemps) {
-  return new (C) CXXExprWithTemporaries(SubExpr, Temps, NumTemps);
+                                                       unsigned NumTemps,
+                                                       bool DestroyTemps) {
+  return new (C) CXXExprWithTemporaries(SubExpr, Temps, NumTemps, 
+                                        DestroyTemps);
 }
 
 void CXXExprWithTemporaries::Destroy(ASTContext &C) {
diff --git a/lib/AST/ExprConstant.cpp b/lib/AST/ExprConstant.cpp
index 50fdcfd..8e3c3ce 100644
--- a/lib/AST/ExprConstant.cpp
+++ b/lib/AST/ExprConstant.cpp
@@ -946,7 +946,8 @@ bool IntExprEvaluator::VisitBinaryOperator(const BinaryOperator *E) {
         const QualType ElementType = Type->getAsPointerType()->getPointeeType();
 
         uint64_t D = LHSValue.getLValueOffset() - RHSValue.getLValueOffset();
-        D /= Info.Ctx.getTypeSize(ElementType) / 8;
+        if (!ElementType->isVoidType() && !ElementType->isFunctionType())
+          D /= Info.Ctx.getTypeSize(ElementType) / 8;
 
         return Success(D, E);
       }
diff --git a/lib/AST/Type.cpp b/lib/AST/Type.cpp
index f573744..e304f54 100644
--- a/lib/AST/Type.cpp
+++ b/lib/AST/Type.cpp
@@ -1049,6 +1049,10 @@ TemplateSpecializationType::
 anyDependentTemplateArguments(const TemplateArgument *Args, unsigned NumArgs) {
   for (unsigned Idx = 0; Idx < NumArgs; ++Idx) {
     switch (Args[Idx].getKind()) {
+    case TemplateArgument::Null:
+      assert(false && "Should not have a NULL template argument");
+      break;
+        
     case TemplateArgument::Type:
       if (Args[Idx].getAsType()->isDependentType())
         return true;
@@ -1451,6 +1455,10 @@ TemplateSpecializationType::PrintTemplateArgumentList(
     // Print the argument into a string.
     std::string ArgString;
     switch (Args[Arg].getKind()) {
+    case TemplateArgument::Null:
+      assert(false && "Null template argument");
+      break;
+
     case TemplateArgument::Type:
       Args[Arg].getAsType().getAsStringInternal(ArgString, Policy);
       break;
diff --git a/lib/Analysis/CFRefCount.cpp b/lib/Analysis/CFRefCount.cpp
index 30ff67f..532d16d 100644
--- a/lib/Analysis/CFRefCount.cpp
+++ b/lib/Analysis/CFRefCount.cpp
@@ -636,7 +636,11 @@ class VISIBILITY_HIDDEN RetainSummaryManager {
   /// ObjCAllocRetE - Default return effect for methods returning Objective-C
   ///  objects.
   RetEffect ObjCAllocRetE;
-  
+
+	/// ObjCInitRetE - Default return effect for init methods returning Objective-C
+  ///  objects.
+  RetEffect ObjCInitRetE;
+	
   RetainSummary DefaultSummary;
   RetainSummary* StopSummary;
   
@@ -776,6 +780,8 @@ public:
      GCEnabled(gcenabled), AF(BPAlloc), ScratchArgs(AF.GetEmptyMap()),
      ObjCAllocRetE(gcenabled ? RetEffect::MakeGCNotOwned()
                              : RetEffect::MakeOwned(RetEffect::ObjC, true)),
+		 ObjCInitRetE(gcenabled ? RetEffect::MakeGCNotOwned()
+														:	RetEffect::MakeOwnedWhenTrackedReceiver()),
      DefaultSummary(AF.GetEmptyMap() /* per-argument effects (none) */,
                     RetEffect::MakeNoRet() /* return effect */,
                     MayEscape, /* default argument effect */
@@ -1156,8 +1162,7 @@ RetainSummaryManager::getInitMethodSummary(QualType RetTy) {
   // 'init' methods conceptually return a newly allocated object and claim
   // the receiver.  
   if (isTrackedObjCObjectType(RetTy) || isTrackedCFObjectType(RetTy))
-    return getPersistentSummary(RetEffect::MakeOwnedWhenTrackedReceiver(),
-                                DecRefMsg);
+    return getPersistentSummary(ObjCInitRetE, DecRefMsg);
   
   return getDefaultSummary();
 }
@@ -1168,12 +1173,19 @@ RetainSummaryManager::updateSummaryFromAnnotations(RetainSummary &Summ,
   if (!FD)
     return;
 
+	QualType RetTy = FD->getResultType();
+	
   // Determine if there is a special return effect for this method.
-  if (isTrackedObjCObjectType(FD->getResultType())) {
+  if (isTrackedObjCObjectType(RetTy)) {
     if (FD->getAttr<NSReturnsRetainedAttr>()) {
       Summ.setRetEffect(ObjCAllocRetE);
     }
-    else if (FD->getAttr<CFReturnsRetainedAttr>()) {
+		else if (FD->getAttr<CFReturnsRetainedAttr>()) {
+      Summ.setRetEffect(RetEffect::MakeOwned(RetEffect::CF, true));
+		}
+	}
+	else if (RetTy->getAsPointerType()) {
+		if (FD->getAttr<CFReturnsRetainedAttr>()) {
       Summ.setRetEffect(RetEffect::MakeOwned(RetEffect::CF, true));
     }
   }
@@ -1367,8 +1379,7 @@ void RetainSummaryManager::InitializeMethodSummaries() {
   // Create the "init" selector.  It just acts as a pass-through for the
   // receiver.
   addNSObjectMethSummary(GetNullarySelector("init", Ctx),
-                 getPersistentSummary(RetEffect::MakeOwnedWhenTrackedReceiver(),
-                 DecRefMsg));
+												 getPersistentSummary(ObjCInitRetE, DecRefMsg));
   
   // The next methods are allocators.
   RetainSummary *AllocSumm = getPersistentSummary(ObjCAllocRetE);  
diff --git a/lib/Basic/TargetInfo.cpp b/lib/Basic/TargetInfo.cpp
index 1e8ca2b..a513cb1 100644
--- a/lib/Basic/TargetInfo.cpp
+++ b/lib/Basic/TargetInfo.cpp
@@ -22,7 +22,6 @@ TargetInfo::TargetInfo(const std::string &T) : Triple(T) {
   // Set defaults.  Defaults are set for a 32-bit RISC platform,
   // like PPC or SPARC.
   // These should be overridden by concrete targets as needed.
-  CharIsSigned = true;
   TLSSupported = true;
   PointerWidth = PointerAlign = 32;
   WCharWidth = WCharAlign = 32;
diff --git a/lib/Basic/Targets.cpp b/lib/Basic/Targets.cpp
index 4b94bcf..120d525 100644
--- a/lib/Basic/Targets.cpp
+++ b/lib/Basic/Targets.cpp
@@ -182,6 +182,11 @@ static void getDarwinDefines(std::vector<char> &Defs, const LangOptions &Opts) {
     Define(Defs, "__strong", "");
   else
     Define(Defs, "__strong", "__attribute__((objc_gc(strong)))");
+
+  if (Opts.Static)
+    Define(Defs, "__STATIC__");
+  else
+    Define(Defs, "__DYNAMIC__");
 }
 
 static void getDarwinOSXDefines(std::vector<char> &Defs, const char *Triple) {
@@ -252,9 +257,8 @@ class PPCTargetInfo : public TargetInfo {
   static const TargetInfo::GCCRegAlias GCCRegAliases[];
 
 public:
-  PPCTargetInfo(const std::string& triple) : TargetInfo(triple) {
-    CharIsSigned = false;
-  }
+  PPCTargetInfo(const std::string& triple) : TargetInfo(triple) {}
+
   virtual void getTargetBuiltins(const Builtin::Info *&Records,
                                  unsigned &NumRecords) const {
     Records = BuiltinInfo;
@@ -294,6 +298,10 @@ public:
       return true;
     }
   }
+  virtual void getDefaultLangOptions(LangOptions &Opts) {
+    TargetInfo::getDefaultLangOptions(Opts);
+    Opts.CharIsSigned = false;
+  }
   virtual const char *getClobbers() const {
     return "";
   }
@@ -444,6 +452,7 @@ public:
   /// various language options.  These may be overridden by command line
   /// options.
   virtual void getDefaultLangOptions(LangOptions &Opts) {
+    PPC32TargetInfo::getDefaultLangOptions(Opts);
     GetDarwinLanguageOptions(Opts, getTargetTriple());
   }
 };
@@ -464,6 +473,7 @@ public:
   /// various language options.  These may be overridden by command line
   /// options.
   virtual void getDefaultLangOptions(LangOptions &Opts) {
+    PPC64TargetInfo::getDefaultLangOptions(Opts);
     GetDarwinLanguageOptions(Opts, getTargetTriple());
   }
 };
@@ -840,6 +850,7 @@ public:
   /// various language options.  These may be overridden by command line
   /// options.
   virtual void getDefaultLangOptions(LangOptions &Opts) {
+    X86_32TargetInfo::getDefaultLangOptions(Opts);
     GetDarwinLanguageOptions(Opts, getTargetTriple());
   }
 };
diff --git a/lib/CodeGen/ABIInfo.h b/lib/CodeGen/ABIInfo.h
index 3de4612..44af0c4 100644
--- a/lib/CodeGen/ABIInfo.h
+++ b/lib/CodeGen/ABIInfo.h
@@ -10,8 +10,13 @@
 #ifndef CLANG_CODEGEN_ABIINFO_H
 #define CLANG_CODEGEN_ABIINFO_H
 
+#include "clang/AST/Type.h"
+
+#include <cassert>
+
 namespace llvm {
   class Type;
+  class Value;
 }
 
 namespace clang {
@@ -38,32 +43,32 @@ namespace clang {
       Direct,    /// Pass the argument directly using the normal
                  /// converted LLVM type. Complex and structure types
                  /// are passed using first class aggregates.
-  
+
       Indirect,  /// Pass the argument indirectly via a hidden pointer
                  /// with the specified alignment (0 indicates default
                  /// alignment).
-  
+
       Ignore,    /// Ignore the argument (treat as void). Useful for
                  /// void and empty structs.
-  
+
       Coerce,    /// Only valid for aggregate return types, the argument
                  /// should be accessed by coercion to a provided type.
-  
+
       Expand,    /// Only valid for aggregate argument types. The
                  /// structure should be expanded into consecutive
                  /// arguments for its constituent fields. Currently
                  /// expand is only allowed on structures whose fields
                  /// are all scalar types or are themselves expandable
                  /// types.
-  
+
       KindFirst=Direct, KindLast=Expand
     };
-  
+
   private:
     Kind TheKind;
     const llvm::Type *TypeData;
     unsigned UIntData;
-  
+
     ABIArgInfo(Kind K, const llvm::Type *TD=0,
                unsigned UI=0) : TheKind(K),
                                 TypeData(TD),
@@ -71,13 +76,13 @@ namespace clang {
   public:
     ABIArgInfo() : TheKind(Direct), TypeData(0), UIntData(0) {}
 
-    static ABIArgInfo getDirect() { 
-      return ABIArgInfo(Direct); 
+    static ABIArgInfo getDirect() {
+      return ABIArgInfo(Direct);
     }
     static ABIArgInfo getIgnore() {
       return ABIArgInfo(Ignore);
     }
-    static ABIArgInfo getCoerce(const llvm::Type *T) { 
+    static ABIArgInfo getCoerce(const llvm::Type *T) {
       return ABIArgInfo(Coerce, T);
     }
     static ABIArgInfo getIndirect(unsigned Alignment) {
@@ -86,20 +91,20 @@ namespace clang {
     static ABIArgInfo getExpand() {
       return ABIArgInfo(Expand);
     }
-  
+
     Kind getKind() const { return TheKind; }
     bool isDirect() const { return TheKind == Direct; }
     bool isIgnore() const { return TheKind == Ignore; }
     bool isCoerce() const { return TheKind == Coerce; }
     bool isIndirect() const { return TheKind == Indirect; }
     bool isExpand() const { return TheKind == Expand; }
-  
+
     // Coerce accessors
     const llvm::Type *getCoerceToType() const {
       assert(TheKind == Coerce && "Invalid kind!");
       return TypeData;
     }
-  
+
     // ByVal accessors
     unsigned getIndirectAlign() const {
       assert(TheKind == Indirect && "Invalid kind!");
@@ -120,7 +125,7 @@ namespace clang {
 
     /// EmitVAArg - Emit the target dependent code to load a value of
     /// \arg Ty from the va_list pointed to by \arg VAListAddr.
-    
+
     // FIXME: This is a gaping layering violation if we wanted to drop
     // the ABI information any lower than CodeGen. Of course, for
     // VAArg handling it has to be at this level; there is no way to
diff --git a/lib/CodeGen/CGBlocks.cpp b/lib/CodeGen/CGBlocks.cpp
index ead689c..d5f803b 100644
--- a/lib/CodeGen/CGBlocks.cpp
+++ b/lib/CodeGen/CGBlocks.cpp
@@ -724,6 +724,8 @@ GenerateCopyHelperFunction(bool BlockHasCopyDispose, const llvm::StructType *T,
   const CGFunctionInfo &FI =
     CGM.getTypes().getFunctionInfo(R, Args);
 
+  // FIXME: We'd like to put these into a mergable by content, with
+  // internal linkage.
   std::string Name = std::string("__copy_helper_block_");
   CodeGenTypes &Types = CGM.getTypes();
   const llvm::FunctionType *LTy = Types.GetFunctionType(FI, false);
@@ -803,6 +805,8 @@ GenerateDestroyHelperFunction(bool BlockHasCopyDispose,
   const CGFunctionInfo &FI =
     CGM.getTypes().getFunctionInfo(R, Args);
 
+  // FIXME: We'd like to put these into a mergable by content, with
+  // internal linkage.
   std::string Name = std::string("__destroy_helper_block_");
   CodeGenTypes &Types = CGM.getTypes();
   const llvm::FunctionType *LTy = Types.GetFunctionType(FI, false);
@@ -889,6 +893,8 @@ GeneratebyrefCopyHelperFunction(const llvm::Type *T, int flag) {
   CodeGenTypes &Types = CGM.getTypes();
   const llvm::FunctionType *LTy = Types.GetFunctionType(FI, false);
 
+  // FIXME: We'd like to put these into a mergable by content, with
+  // internal linkage.
   llvm::Function *Fn =
     llvm::Function::Create(LTy, llvm::GlobalValue::InternalLinkage,
                            Name,
@@ -950,6 +956,8 @@ BlockFunction::GeneratebyrefDestroyHelperFunction(const llvm::Type *T,
   CodeGenTypes &Types = CGM.getTypes();
   const llvm::FunctionType *LTy = Types.GetFunctionType(FI, false);
 
+  // FIXME: We'd like to put these into a mergable by content, with
+  // internal linkage.
   llvm::Function *Fn =
     llvm::Function::Create(LTy, llvm::GlobalValue::InternalLinkage,
                            Name,
@@ -980,13 +988,36 @@ BlockFunction::GeneratebyrefDestroyHelperFunction(const llvm::Type *T,
 }
 
 llvm::Constant *BlockFunction::BuildbyrefCopyHelper(const llvm::Type *T,
-                                                    int flag) {
-  return CodeGenFunction(CGM).GeneratebyrefCopyHelperFunction(T, flag);
+                                                    int flag, unsigned Align) {
+  // All alignments below that of pointer alignment collpase down to just
+  // pointer alignment, as we always have at least that much alignment to begin
+  // with.
+  Align /= unsigned(CGF.Target.getPointerAlign(0)/8);
+  // As an optimization, we only generate a single function of each kind we
+  // might need.  We need a different one for each alignment and for each
+  // setting of flags.  We mix Align and flag to get the kind.
+  uint64_t kind = (uint64_t)Align*BLOCK_BYREF_CURRENT_MAX + flag;
+  llvm::Constant *& Entry = CGM.AssignCache[kind];
+  if (Entry)
+    return Entry;
+  return Entry=CodeGenFunction(CGM).GeneratebyrefCopyHelperFunction(T, flag);
 }
 
 llvm::Constant *BlockFunction::BuildbyrefDestroyHelper(const llvm::Type *T,
-                                                       int flag) {
-  return CodeGenFunction(CGM).GeneratebyrefDestroyHelperFunction(T, flag);
+                                                       int flag,
+                                                       unsigned Align) {
+  // All alignments below that of pointer alignment collpase down to just
+  // pointer alignment, as we always have at least that much alignment to begin
+  // with.
+  Align /= unsigned(CGF.Target.getPointerAlign(0)/8);
+  // As an optimization, we only generate a single function of each kind we
+  // might need.  We need a different one for each alignment and for each
+  // setting of flags.  We mix Align and flag to get the kind.
+  uint64_t kind = (uint64_t)Align*BLOCK_BYREF_CURRENT_MAX + flag;
+  llvm::Constant *& Entry = CGM.DestroyCache[kind];
+  if (Entry)
+    return Entry;
+  return Entry=CodeGenFunction(CGM).GeneratebyrefDestroyHelperFunction(T, flag);
 }
 
 llvm::Value *BlockFunction::getBlockObjectDispose() {
diff --git a/lib/CodeGen/CGBlocks.h b/lib/CodeGen/CGBlocks.h
index 56d3a2d..5d46ac7 100644
--- a/lib/CodeGen/CGBlocks.h
+++ b/lib/CodeGen/CGBlocks.h
@@ -98,6 +98,9 @@ public:
   llvm::Value *BlockObjectDispose;
   const llvm::Type *PtrToInt8Ty;
 
+  std::map<uint64_t, llvm::Constant *> AssignCache;
+  std::map<uint64_t, llvm::Constant *> DestroyCache;
+
   BlockModule(ASTContext &C, llvm::Module &M, const llvm::TargetData &TD,
               CodeGenTypes &T, CodeGenModule &CodeGen)
     : Context(C), TheModule(M), TheTargetData(TD), Types(T),
@@ -131,8 +134,9 @@ public:
                                       variable */
     BLOCK_FIELD_IS_WEAK     = 16,  /* declared __weak, only used in byref copy
                                       helpers */
-    BLOCK_BYREF_CALLER      = 128  /* called from __block (byref) copy/dispose
+    BLOCK_BYREF_CALLER      = 128,  /* called from __block (byref) copy/dispose
                                       support routines */
+    BLOCK_BYREF_CURRENT_MAX = 256
   };
 
   /// BlockInfo - Information to generate a block literal.
@@ -199,8 +203,10 @@ public:
   llvm::Constant *GeneratebyrefCopyHelperFunction(const llvm::Type *, int flag);
   llvm::Constant *GeneratebyrefDestroyHelperFunction(const llvm::Type *T, int);
 
-  llvm::Constant *BuildbyrefCopyHelper(const llvm::Type *T, int flag);
-  llvm::Constant *BuildbyrefDestroyHelper(const llvm::Type *T, int flag);
+  llvm::Constant *BuildbyrefCopyHelper(const llvm::Type *T, int flag,
+                                       unsigned Align);
+  llvm::Constant *BuildbyrefDestroyHelper(const llvm::Type *T, int flag,
+                                          unsigned Align);
 
   llvm::Value *getBlockObjectAssign();
   llvm::Value *getBlockObjectDispose();
diff --git a/lib/CodeGen/CGCXXTemp.cpp b/lib/CodeGen/CGCXXTemp.cpp
index d53a56f..141726a 100644
--- a/lib/CodeGen/CGCXXTemp.cpp
+++ b/lib/CodeGen/CGCXXTemp.cpp
@@ -18,8 +18,29 @@ using namespace CodeGen;
 void CodeGenFunction::PushCXXTemporary(const CXXTemporary *Temporary, 
                                        llvm::Value *Ptr) {
   llvm::BasicBlock *DtorBlock = createBasicBlock("temp.dtor");
-    
-  LiveTemporaries.push_back(CXXLiveTemporaryInfo(Temporary, Ptr, DtorBlock, 0));
+  
+  llvm::Value *CondPtr = 0;
+  
+  // Check if temporaries need to be conditional. If so, we'll create a 
+  // condition boolean, initialize it to 0 and 
+  if (!ConditionalTempDestructionStack.empty()) {
+    CondPtr = CreateTempAlloca(llvm::Type::Int1Ty, "cond");
+  
+    // Initialize it to false. This initialization takes place right after
+    // the alloca insert point.
+    llvm::StoreInst *SI = 
+      new llvm::StoreInst(llvm::ConstantInt::getFalse(), CondPtr);
+    llvm::BasicBlock *Block = AllocaInsertPt->getParent();
+    Block->getInstList().insertAfter((llvm::Instruction *)AllocaInsertPt, SI);
+
+    // Now set it to true.
+    Builder.CreateStore(llvm::ConstantInt::getTrue(), CondPtr);
+  }
+  
+  LiveTemporaries.push_back(CXXLiveTemporaryInfo(Temporary, Ptr, DtorBlock, 
+                                                 CondPtr));
+
+  PushCleanupBlock(DtorBlock);
 }
 
 void CodeGenFunction::PopCXXTemporary() {
@@ -35,9 +56,28 @@ void CodeGenFunction::PopCXXTemporary() {
   
   EmitBlock(Info.DtorBlock);
 
+  llvm::BasicBlock *CondEnd = 0;
+
+  // If this is a conditional temporary, we need to check the condition
+  // boolean and only call the destructor if it's true.
+  if (Info.CondPtr) {
+    llvm::BasicBlock *CondBlock = createBasicBlock("cond.dtor.call");
+    CondEnd = createBasicBlock("cond.dtor.end");
+      
+    llvm::Value *Cond = Builder.CreateLoad(Info.CondPtr);
+    Builder.CreateCondBr(Cond, CondBlock, CondEnd);
+    EmitBlock(CondBlock);
+  }
+  
   EmitCXXDestructorCall(Info.Temporary->getDestructor(),
                         Dtor_Complete, Info.ThisPtr);
 
+  if (CondEnd) {
+    // Reset the condition. to false.
+    Builder.CreateStore(llvm::ConstantInt::getFalse(), Info.CondPtr);
+    EmitBlock(CondEnd);
+  }
+  
   LiveTemporaries.pop_back();
 }
 
@@ -47,21 +87,38 @@ CodeGenFunction::EmitCXXExprWithTemporaries(const CXXExprWithTemporaries *E,
                                             bool isAggLocVolatile) {
   // Keep track of the current cleanup stack depth.
   size_t CleanupStackDepth = CleanupEntries.size();
+  (void) CleanupStackDepth;
 
   unsigned OldNumLiveTemporaries = LiveTemporaries.size();
   
   RValue RV = EmitAnyExpr(E->getSubExpr(), AggLoc, isAggLocVolatile);
   
-  // Go through the temporaries backwards.
-  for (unsigned i = E->getNumTemporaries(); i != 0; --i) {
-    assert(LiveTemporaries.back().Temporary == E->getTemporary(i - 1));
-    LiveTemporaries.pop_back();
-  }
+  // Pop temporaries.
+  while (LiveTemporaries.size() > OldNumLiveTemporaries)
+    PopCXXTemporary();
+  
+  assert(CleanupEntries.size() == CleanupStackDepth &&
+         "Cleanup size mismatch!");
+  
+  return RV;
+}
+
+void 
+CodeGenFunction::PushConditionalTempDestruction() {
+  // Store the current number of live temporaries.
+  ConditionalTempDestructionStack.push_back(LiveTemporaries.size());
+}
 
-  assert(OldNumLiveTemporaries == LiveTemporaries.size() &&
-         "Live temporary stack mismatch!");
+void CodeGenFunction::PopConditionalTempDestruction() {
+ size_t NumLiveTemporaries = ConditionalTempDestructionStack.back();
+ ConditionalTempDestructionStack.pop_back();
   
-  EmitCleanupBlocks(CleanupStackDepth);
+  // Pop temporaries.
+  while (LiveTemporaries.size() > NumLiveTemporaries) {
+    assert(LiveTemporaries.back().CondPtr && 
+           "Conditional temporary must have a cond ptr!");
 
-  return RV;
+    PopCXXTemporary();
+  }  
 }
+  
diff --git a/lib/CodeGen/CGCall.cpp b/lib/CodeGen/CGCall.cpp
index ea0b887..b46e860 100644
--- a/lib/CodeGen/CGCall.cpp
+++ b/lib/CodeGen/CGCall.cpp
@@ -16,15 +16,12 @@
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
 #include "clang/Basic/TargetInfo.h"
-#include "clang/AST/ASTContext.h"
 #include "clang/AST/Decl.h"
 #include "clang/AST/DeclCXX.h"
 #include "clang/AST/DeclObjC.h"
-#include "clang/AST/RecordLayout.h"
-#include "llvm/ADT/StringExtras.h"
+#include "clang/Frontend/CompileOptions.h"
 #include "llvm/Attributes.h"
 #include "llvm/Support/CallSite.h"
-#include "llvm/Support/MathExtras.h"
 #include "llvm/Target/TargetData.h"
 
 #include "ABIInfo.h"
@@ -126,1370 +123,6 @@ const CGFunctionInfo &CodeGenTypes::getFunctionInfo(QualType ResTy,
   return *FI;
 }
 
-/***/
-
-ABIInfo::~ABIInfo() {}
-
-void ABIArgInfo::dump() const {
-  fprintf(stderr, "(ABIArgInfo Kind=");
-  switch (TheKind) {
-  case Direct: 
-    fprintf(stderr, "Direct");
-    break;
-  case Ignore: 
-    fprintf(stderr, "Ignore");
-    break;
-  case Coerce: 
-    fprintf(stderr, "Coerce Type=");
-    getCoerceToType()->print(llvm::errs());
-    break;
-  case Indirect: 
-    fprintf(stderr, "Indirect Align=%d", getIndirectAlign());
-    break;
-  case Expand: 
-    fprintf(stderr, "Expand");
-    break;
-  }
-  fprintf(stderr, ")\n");
-}
-
-/***/
-
-static bool isEmptyRecord(ASTContext &Context, QualType T);
-
-/// isEmptyField - Return true iff a the field is "empty", that is it
-/// is an unnamed bit-field or an (array of) empty record(s).
-static bool isEmptyField(ASTContext &Context, const FieldDecl *FD) {
-  if (FD->isUnnamedBitfield())
-    return true;
-
-  QualType FT = FD->getType();
-  // Constant arrays of empty records count as empty, strip them off.
-  while (const ConstantArrayType *AT = Context.getAsConstantArrayType(FT))
-    FT = AT->getElementType();
-  
-  return isEmptyRecord(Context, FT);
-}
-
-/// isEmptyRecord - Return true iff a structure contains only empty
-/// fields. Note that a structure with a flexible array member is not
-/// considered empty.
-static bool isEmptyRecord(ASTContext &Context, QualType T) {
-  const RecordType *RT = T->getAsRecordType();
-  if (!RT)
-    return 0;
-  const RecordDecl *RD = RT->getDecl();
-  if (RD->hasFlexibleArrayMember())
-    return false;
-  for (RecordDecl::field_iterator i = RD->field_begin(Context), 
-         e = RD->field_end(Context); i != e; ++i)
-    if (!isEmptyField(Context, *i))
-      return false;
-  return true;
-}
-
-/// isSingleElementStruct - Determine if a structure is a "single
-/// element struct", i.e. it has exactly one non-empty field or
-/// exactly one field which is itself a single element
-/// struct. Structures with flexible array members are never
-/// considered single element structs.
-///
-/// \return The field declaration for the single non-empty field, if
-/// it exists.
-static const Type *isSingleElementStruct(QualType T, ASTContext &Context) {
-  const RecordType *RT = T->getAsStructureType();
-  if (!RT)
-    return 0;
-
-  const RecordDecl *RD = RT->getDecl();
-  if (RD->hasFlexibleArrayMember())
-    return 0;
-
-  const Type *Found = 0;
-  for (RecordDecl::field_iterator i = RD->field_begin(Context), 
-         e = RD->field_end(Context); i != e; ++i) {
-    const FieldDecl *FD = *i;
-    QualType FT = FD->getType();
-
-    // Ignore empty fields.
-    if (isEmptyField(Context, FD))
-      continue;
-
-    // If we already found an element then this isn't a single-element
-    // struct.
-    if (Found)
-      return 0;
-
-    // Treat single element arrays as the element.
-    while (const ConstantArrayType *AT = Context.getAsConstantArrayType(FT)) {
-      if (AT->getSize().getZExtValue() != 1)
-        break;
-      FT = AT->getElementType();
-    }
-
-    if (!CodeGenFunction::hasAggregateLLVMType(FT)) {
-      Found = FT.getTypePtr();
-    } else {
-      Found = isSingleElementStruct(FT, Context);
-      if (!Found)
-        return 0;
-    }
-  }
-
-  return Found;
-}
-
-static bool is32Or64BitBasicType(QualType Ty, ASTContext &Context) {
-  if (!Ty->getAsBuiltinType() && !Ty->isPointerType())
-    return false;
-
-  uint64_t Size = Context.getTypeSize(Ty);
-  return Size == 32 || Size == 64;
-}
-
-static bool areAllFields32Or64BitBasicType(const RecordDecl *RD,
-                                           ASTContext &Context) {
-  for (RecordDecl::field_iterator i = RD->field_begin(Context), 
-         e = RD->field_end(Context); i != e; ++i) {
-    const FieldDecl *FD = *i;
-
-    if (!is32Or64BitBasicType(FD->getType(), Context))
-      return false;
-    
-    // FIXME: Reject bit-fields wholesale; there are two problems, we don't know
-    // how to expand them yet, and the predicate for telling if a bitfield still
-    // counts as "basic" is more complicated than what we were doing previously.
-    if (FD->isBitField())
-      return false;
-  }
-
-  return true;
-}
-
-namespace {
-/// DefaultABIInfo - The default implementation for ABI specific
-/// details. This implementation provides information which results in
-/// self-consistent and sensible LLVM IR generation, but does not
-/// conform to any particular ABI.
-class DefaultABIInfo : public ABIInfo {
-  ABIArgInfo classifyReturnType(QualType RetTy, 
-                                ASTContext &Context) const;
-  
-  ABIArgInfo classifyArgumentType(QualType RetTy,
-                                  ASTContext &Context) const;
-
-  virtual void computeInfo(CGFunctionInfo &FI, ASTContext &Context) const {
-    FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), Context);
-    for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
-         it != ie; ++it)
-      it->info = classifyArgumentType(it->type, Context);
-  }
-
-  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                 CodeGenFunction &CGF) const;
-};
-
-/// X86_32ABIInfo - The X86-32 ABI information.
-class X86_32ABIInfo : public ABIInfo {
-  ASTContext &Context;
-  bool IsDarwin;
-
-  static bool isRegisterSize(unsigned Size) {
-    return (Size == 8 || Size == 16 || Size == 32 || Size == 64);
-  }
-
-  static bool shouldReturnTypeInRegister(QualType Ty, ASTContext &Context);
-
-public:
-  ABIArgInfo classifyReturnType(QualType RetTy, 
-                                ASTContext &Context) const;
-
-  ABIArgInfo classifyArgumentType(QualType RetTy,
-                                  ASTContext &Context) const;
-
-  virtual void computeInfo(CGFunctionInfo &FI, ASTContext &Context) const {
-    FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), Context);
-    for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
-         it != ie; ++it)
-      it->info = classifyArgumentType(it->type, Context);
-  }
-
-  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                 CodeGenFunction &CGF) const;
-
-  X86_32ABIInfo(ASTContext &Context, bool d) 
-    : ABIInfo(), Context(Context), IsDarwin(d) {}
-};
-}
-
-
-/// shouldReturnTypeInRegister - Determine if the given type should be
-/// passed in a register (for the Darwin ABI).
-bool X86_32ABIInfo::shouldReturnTypeInRegister(QualType Ty,
-                                               ASTContext &Context) {
-  uint64_t Size = Context.getTypeSize(Ty);
-
-  // Type must be register sized.
-  if (!isRegisterSize(Size))
-    return false;
-
-  if (Ty->isVectorType()) {
-    // 64- and 128- bit vectors inside structures are not returned in
-    // registers.
-    if (Size == 64 || Size == 128)
-      return false;
-
-    return true;
-  }
-
-  // If this is a builtin, pointer, or complex type, it is ok.
-  if (Ty->getAsBuiltinType() || Ty->isPointerType() || Ty->isAnyComplexType())
-    return true;
-
-  // Arrays are treated like records.
-  if (const ConstantArrayType *AT = Context.getAsConstantArrayType(Ty))
-    return shouldReturnTypeInRegister(AT->getElementType(), Context);
-
-  // Otherwise, it must be a record type.
-  const RecordType *RT = Ty->getAsRecordType();
-  if (!RT) return false;
-
-  // Structure types are passed in register if all fields would be
-  // passed in a register.
-  for (RecordDecl::field_iterator i = RT->getDecl()->field_begin(Context), 
-         e = RT->getDecl()->field_end(Context); i != e; ++i) {
-    const FieldDecl *FD = *i;
-    
-    // Empty fields are ignored.
-    if (isEmptyField(Context, FD))
-      continue;
-
-    // Check fields recursively.
-    if (!shouldReturnTypeInRegister(FD->getType(), Context))
-      return false;
-  }
-
-  return true;
-}
-
-ABIArgInfo X86_32ABIInfo::classifyReturnType(QualType RetTy,
-                                            ASTContext &Context) const {
-  if (RetTy->isVoidType()) {
-    return ABIArgInfo::getIgnore();
-  } else if (const VectorType *VT = RetTy->getAsVectorType()) {
-    // On Darwin, some vectors are returned in registers.
-    if (IsDarwin) {
-      uint64_t Size = Context.getTypeSize(RetTy);
-
-      // 128-bit vectors are a special case; they are returned in
-      // registers and we need to make sure to pick a type the LLVM
-      // backend will like.
-      if (Size == 128)
-        return ABIArgInfo::getCoerce(llvm::VectorType::get(llvm::Type::Int64Ty, 
-                                                           2));
-
-      // Always return in register if it fits in a general purpose
-      // register, or if it is 64 bits and has a single element.
-      if ((Size == 8 || Size == 16 || Size == 32) ||
-          (Size == 64 && VT->getNumElements() == 1))
-        return ABIArgInfo::getCoerce(llvm::IntegerType::get(Size));
-        
-      return ABIArgInfo::getIndirect(0);
-    }
-
-    return ABIArgInfo::getDirect();
-  } else if (CodeGenFunction::hasAggregateLLVMType(RetTy)) {
-    // Structures with flexible arrays are always indirect.
-    if (const RecordType *RT = RetTy->getAsStructureType())
-      if (RT->getDecl()->hasFlexibleArrayMember())
-        return ABIArgInfo::getIndirect(0);
-
-    // Outside of Darwin, structs and unions are always indirect.
-    if (!IsDarwin && !RetTy->isAnyComplexType())
-      return ABIArgInfo::getIndirect(0);
-
-    // Classify "single element" structs as their element type.
-    if (const Type *SeltTy = isSingleElementStruct(RetTy, Context)) {
-      if (const BuiltinType *BT = SeltTy->getAsBuiltinType()) {
-        if (BT->isIntegerType()) {
-          // We need to use the size of the structure, padding
-          // bit-fields can adjust that to be larger than the single
-          // element type.
-          uint64_t Size = Context.getTypeSize(RetTy);
-          return ABIArgInfo::getCoerce(llvm::IntegerType::get((unsigned) Size));
-        } else if (BT->getKind() == BuiltinType::Float) {
-          assert(Context.getTypeSize(RetTy) == Context.getTypeSize(SeltTy) &&
-                 "Unexpect single element structure size!");
-          return ABIArgInfo::getCoerce(llvm::Type::FloatTy);
-        } else if (BT->getKind() == BuiltinType::Double) {
-          assert(Context.getTypeSize(RetTy) == Context.getTypeSize(SeltTy) &&
-                 "Unexpect single element structure size!");
-          return ABIArgInfo::getCoerce(llvm::Type::DoubleTy);
-        }
-      } else if (SeltTy->isPointerType()) {
-        // FIXME: It would be really nice if this could come out as the proper
-        // pointer type.
-        llvm::Type *PtrTy = 
-          llvm::PointerType::getUnqual(llvm::Type::Int8Ty);
-        return ABIArgInfo::getCoerce(PtrTy);
-      } else if (SeltTy->isVectorType()) {
-        // 64- and 128-bit vectors are never returned in a
-        // register when inside a structure.
-        uint64_t Size = Context.getTypeSize(RetTy);
-        if (Size == 64 || Size == 128)
-          return ABIArgInfo::getIndirect(0);
-
-        return classifyReturnType(QualType(SeltTy, 0), Context);
-      }
-    }
-
-    // Small structures which are register sized are generally returned
-    // in a register.
-    if (X86_32ABIInfo::shouldReturnTypeInRegister(RetTy, Context)) {
-      uint64_t Size = Context.getTypeSize(RetTy);
-      return ABIArgInfo::getCoerce(llvm::IntegerType::get(Size));
-    }
-
-    return ABIArgInfo::getIndirect(0);
-  } else {
-    return ABIArgInfo::getDirect();
-  }
-}
-
-ABIArgInfo X86_32ABIInfo::classifyArgumentType(QualType Ty,
-                                               ASTContext &Context) const {
-  // FIXME: Set alignment on indirect arguments.
-  if (CodeGenFunction::hasAggregateLLVMType(Ty)) {
-    // Structures with flexible arrays are always indirect.
-    if (const RecordType *RT = Ty->getAsStructureType())
-      if (RT->getDecl()->hasFlexibleArrayMember())
-        return ABIArgInfo::getIndirect(0);
-
-    // Ignore empty structs.
-    uint64_t Size = Context.getTypeSize(Ty);
-    if (Ty->isStructureType() && Size == 0)
-      return ABIArgInfo::getIgnore();
-
-    // Expand structs with size <= 128-bits which consist only of
-    // basic types (int, long long, float, double, xxx*). This is
-    // non-recursive and does not ignore empty fields.
-    if (const RecordType *RT = Ty->getAsStructureType()) {
-      if (Context.getTypeSize(Ty) <= 4*32 &&
-          areAllFields32Or64BitBasicType(RT->getDecl(), Context))
-        return ABIArgInfo::getExpand();
-    }
-
-    return ABIArgInfo::getIndirect(0);
-  } else {
-    return ABIArgInfo::getDirect();
-  }
-}
-
-llvm::Value *X86_32ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                      CodeGenFunction &CGF) const {
-  const llvm::Type *BP = llvm::PointerType::getUnqual(llvm::Type::Int8Ty);
-  const llvm::Type *BPP = llvm::PointerType::getUnqual(BP);
-
-  CGBuilderTy &Builder = CGF.Builder;
-  llvm::Value *VAListAddrAsBPP = Builder.CreateBitCast(VAListAddr, BPP, 
-                                                       "ap");
-  llvm::Value *Addr = Builder.CreateLoad(VAListAddrAsBPP, "ap.cur");
-  llvm::Type *PTy = 
-    llvm::PointerType::getUnqual(CGF.ConvertType(Ty));
-  llvm::Value *AddrTyped = Builder.CreateBitCast(Addr, PTy);
-  
-  uint64_t Offset = 
-    llvm::RoundUpToAlignment(CGF.getContext().getTypeSize(Ty) / 8, 4);
-  llvm::Value *NextAddr = 
-    Builder.CreateGEP(Addr, 
-                      llvm::ConstantInt::get(llvm::Type::Int32Ty, Offset),
-                      "ap.next");
-  Builder.CreateStore(NextAddr, VAListAddrAsBPP);
-
-  return AddrTyped;
-}
-
-namespace {
-/// X86_64ABIInfo - The X86_64 ABI information.
-class X86_64ABIInfo : public ABIInfo {
-  enum Class {
-    Integer = 0,
-    SSE,
-    SSEUp,
-    X87,
-    X87Up,
-    ComplexX87,
-    NoClass,
-    Memory
-  };
-
-  /// merge - Implement the X86_64 ABI merging algorithm.
-  ///
-  /// Merge an accumulating classification \arg Accum with a field
-  /// classification \arg Field.
-  ///
-  /// \param Accum - The accumulating classification. This should
-  /// always be either NoClass or the result of a previous merge
-  /// call. In addition, this should never be Memory (the caller
-  /// should just return Memory for the aggregate).
-  Class merge(Class Accum, Class Field) const;
-
-  /// classify - Determine the x86_64 register classes in which the
-  /// given type T should be passed.
-  ///
-  /// \param Lo - The classification for the parts of the type
-  /// residing in the low word of the containing object.
-  ///
-  /// \param Hi - The classification for the parts of the type
-  /// residing in the high word of the containing object.
-  ///
-  /// \param OffsetBase - The bit offset of this type in the
-  /// containing object.  Some parameters are classified different
-  /// depending on whether they straddle an eightbyte boundary.
-  ///
-  /// If a word is unused its result will be NoClass; if a type should
-  /// be passed in Memory then at least the classification of \arg Lo
-  /// will be Memory.
-  ///
-  /// The \arg Lo class will be NoClass iff the argument is ignored.
-  ///
-  /// If the \arg Lo class is ComplexX87, then the \arg Hi class will
-  /// also be ComplexX87.
-  void classify(QualType T, ASTContext &Context, uint64_t OffsetBase,
-                Class &Lo, Class &Hi) const;
-  
-  /// getCoerceResult - Given a source type \arg Ty and an LLVM type
-  /// to coerce to, chose the best way to pass Ty in the same place
-  /// that \arg CoerceTo would be passed, but while keeping the
-  /// emitted code as simple as possible.
-  ///
-  /// FIXME: Note, this should be cleaned up to just take an enumeration of all
-  /// the ways we might want to pass things, instead of constructing an LLVM
-  /// type. This makes this code more explicit, and it makes it clearer that we
-  /// are also doing this for correctness in the case of passing scalar types.
-  ABIArgInfo getCoerceResult(QualType Ty,
-                             const llvm::Type *CoerceTo,
-                             ASTContext &Context) const;
-
-  /// getIndirectResult - Give a source type \arg Ty, return a suitable result
-  /// such that the argument will be passed in memory.
-  ABIArgInfo getIndirectResult(QualType Ty,
-                               ASTContext &Context) const;
-
-  ABIArgInfo classifyReturnType(QualType RetTy, 
-                                ASTContext &Context) const;  
-
-  ABIArgInfo classifyArgumentType(QualType Ty,
-                                  ASTContext &Context,
-                                  unsigned &neededInt,
-                                  unsigned &neededSSE) const;
-
-public:
-  virtual void computeInfo(CGFunctionInfo &FI, ASTContext &Context) const;
-
-  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                 CodeGenFunction &CGF) const;
-};
-}
-
-X86_64ABIInfo::Class X86_64ABIInfo::merge(Class Accum, 
-                                          Class Field) const {
-  // AMD64-ABI 3.2.3p2: Rule 4. Each field of an object is
-  // classified recursively so that always two fields are
-  // considered. The resulting class is calculated according to
-  // the classes of the fields in the eightbyte:
-  //
-  // (a) If both classes are equal, this is the resulting class.
-  //
-  // (b) If one of the classes is NO_CLASS, the resulting class is
-  // the other class.
-  //
-  // (c) If one of the classes is MEMORY, the result is the MEMORY
-  // class.
-  //
-  // (d) If one of the classes is INTEGER, the result is the
-  // INTEGER.
-  //
-  // (e) If one of the classes is X87, X87UP, COMPLEX_X87 class,
-  // MEMORY is used as class.
-  //
-  // (f) Otherwise class SSE is used.
-
-  // Accum should never be memory (we should have returned) or
-  // ComplexX87 (because this cannot be passed in a structure).
-  assert((Accum != Memory && Accum != ComplexX87) &&
-         "Invalid accumulated classification during merge.");
-  if (Accum == Field || Field == NoClass)
-    return Accum;
-  else if (Field == Memory)
-    return Memory;
-  else if (Accum == NoClass)
-    return Field;
-  else if (Accum == Integer || Field == Integer) 
-    return Integer;
-  else if (Field == X87 || Field == X87Up || Field == ComplexX87 ||
-           Accum == X87 || Accum == X87Up)
-    return Memory;
-  else
-    return SSE;
-}
-
-void X86_64ABIInfo::classify(QualType Ty,
-                             ASTContext &Context,
-                             uint64_t OffsetBase,
-                             Class &Lo, Class &Hi) const {
-  // FIXME: This code can be simplified by introducing a simple value class for
-  // Class pairs with appropriate constructor methods for the various
-  // situations.
-
-  // FIXME: Some of the split computations are wrong; unaligned vectors
-  // shouldn't be passed in registers for example, so there is no chance they
-  // can straddle an eightbyte. Verify & simplify.
-
-  Lo = Hi = NoClass;
-
-  Class &Current = OffsetBase < 64 ? Lo : Hi;
-  Current = Memory;
-
-  if (const BuiltinType *BT = Ty->getAsBuiltinType()) {
-    BuiltinType::Kind k = BT->getKind();
-
-    if (k == BuiltinType::Void) {
-      Current = NoClass; 
-    } else if (k == BuiltinType::Int128 || k == BuiltinType::UInt128) {
-      Lo = Integer;
-      Hi = Integer;
-    } else if (k >= BuiltinType::Bool && k <= BuiltinType::LongLong) {
-      Current = Integer;
-    } else if (k == BuiltinType::Float || k == BuiltinType::Double) {
-      Current = SSE;
-    } else if (k == BuiltinType::LongDouble) {
-      Lo = X87;
-      Hi = X87Up;
-    }
-    // FIXME: _Decimal32 and _Decimal64 are SSE.
-    // FIXME: _float128 and _Decimal128 are (SSE, SSEUp).
-  } else if (const EnumType *ET = Ty->getAsEnumType()) {
-    // Classify the underlying integer type.
-    classify(ET->getDecl()->getIntegerType(), Context, OffsetBase, Lo, Hi);
-  } else if (Ty->hasPointerRepresentation()) {
-    Current = Integer;
-  } else if (const VectorType *VT = Ty->getAsVectorType()) {
-    uint64_t Size = Context.getTypeSize(VT);
-    if (Size == 32) {
-      // gcc passes all <4 x char>, <2 x short>, <1 x int>, <1 x
-      // float> as integer.      
-      Current = Integer;
-
-      // If this type crosses an eightbyte boundary, it should be
-      // split.
-      uint64_t EB_Real = (OffsetBase) / 64;
-      uint64_t EB_Imag = (OffsetBase + Size - 1) / 64;
-      if (EB_Real != EB_Imag)
-        Hi = Lo;      
-    } else if (Size == 64) {
-      // gcc passes <1 x double> in memory. :(
-      if (VT->getElementType()->isSpecificBuiltinType(BuiltinType::Double))
-        return;
-      
-      // gcc passes <1 x long long> as INTEGER.
-      if (VT->getElementType()->isSpecificBuiltinType(BuiltinType::LongLong))
-        Current = Integer;
-      else
-        Current = SSE;
-
-      // If this type crosses an eightbyte boundary, it should be
-      // split.
-      if (OffsetBase && OffsetBase != 64)
-        Hi = Lo;
-    } else if (Size == 128) {
-      Lo = SSE;
-      Hi = SSEUp;
-    }
-  } else if (const ComplexType *CT = Ty->getAsComplexType()) {
-    QualType ET = Context.getCanonicalType(CT->getElementType());
-    
-    uint64_t Size = Context.getTypeSize(Ty);
-    if (ET->isIntegralType()) {
-      if (Size <= 64)
-        Current = Integer;
-      else if (Size <= 128)
-        Lo = Hi = Integer;
-    } else if (ET == Context.FloatTy) 
-      Current = SSE;
-    else if (ET == Context.DoubleTy)
-      Lo = Hi = SSE;
-    else if (ET == Context.LongDoubleTy)
-      Current = ComplexX87;
-
-    // If this complex type crosses an eightbyte boundary then it
-    // should be split.
-    uint64_t EB_Real = (OffsetBase) / 64;
-    uint64_t EB_Imag = (OffsetBase + Context.getTypeSize(ET)) / 64;
-    if (Hi == NoClass && EB_Real != EB_Imag)
-      Hi = Lo;
-  } else if (const ConstantArrayType *AT = Context.getAsConstantArrayType(Ty)) {
-    // Arrays are treated like structures.
-
-    uint64_t Size = Context.getTypeSize(Ty);
-    
-    // AMD64-ABI 3.2.3p2: Rule 1. If the size of an object is larger
-    // than two eightbytes, ..., it has class MEMORY.
-    if (Size > 128)
-      return;
-    
-    // AMD64-ABI 3.2.3p2: Rule 1. If ..., or it contains unaligned
-    // fields, it has class MEMORY.
-    //
-    // Only need to check alignment of array base.
-    if (OffsetBase % Context.getTypeAlign(AT->getElementType()))
-      return;
-
-    // Otherwise implement simplified merge. We could be smarter about
-    // this, but it isn't worth it and would be harder to verify.
-    Current = NoClass;
-    uint64_t EltSize = Context.getTypeSize(AT->getElementType());
-    uint64_t ArraySize = AT->getSize().getZExtValue();
-    for (uint64_t i=0, Offset=OffsetBase; i<ArraySize; ++i, Offset += EltSize) {
-      Class FieldLo, FieldHi;
-      classify(AT->getElementType(), Context, Offset, FieldLo, FieldHi);
-      Lo = merge(Lo, FieldLo);
-      Hi = merge(Hi, FieldHi);
-      if (Lo == Memory || Hi == Memory)
-        break;
-    }
-    
-    // Do post merger cleanup (see below). Only case we worry about is Memory.
-    if (Hi == Memory)
-      Lo = Memory;
-    assert((Hi != SSEUp || Lo == SSE) && "Invalid SSEUp array classification.");
-  } else if (const RecordType *RT = Ty->getAsRecordType()) {
-    uint64_t Size = Context.getTypeSize(Ty);
-    
-    // AMD64-ABI 3.2.3p2: Rule 1. If the size of an object is larger
-    // than two eightbytes, ..., it has class MEMORY.
-    if (Size > 128)
-      return;
-
-    const RecordDecl *RD = RT->getDecl();
-
-    // Assume variable sized types are passed in memory.
-    if (RD->hasFlexibleArrayMember())
-      return;
-
-    const ASTRecordLayout &Layout = Context.getASTRecordLayout(RD);
-    
-    // Reset Lo class, this will be recomputed.
-    Current = NoClass;
-    unsigned idx = 0;
-    for (RecordDecl::field_iterator i = RD->field_begin(Context), 
-           e = RD->field_end(Context); i != e; ++i, ++idx) {
-      uint64_t Offset = OffsetBase + Layout.getFieldOffset(idx);
-      bool BitField = i->isBitField();
-
-      // AMD64-ABI 3.2.3p2: Rule 1. If ..., or it contains unaligned
-      // fields, it has class MEMORY.
-      //
-      // Note, skip this test for bit-fields, see below.
-      if (!BitField && Offset % Context.getTypeAlign(i->getType())) {
-        Lo = Memory;
-        return;
-      }
-
-      // Classify this field.
-      //
-      // AMD64-ABI 3.2.3p2: Rule 3. If the size of the aggregate
-      // exceeds a single eightbyte, each is classified
-      // separately. Each eightbyte gets initialized to class
-      // NO_CLASS.
-      Class FieldLo, FieldHi;
-      
-      // Bit-fields require special handling, they do not force the
-      // structure to be passed in memory even if unaligned, and
-      // therefore they can straddle an eightbyte.
-      if (BitField) {
-        // Ignore padding bit-fields.
-        if (i->isUnnamedBitfield())
-          continue;
-
-        uint64_t Offset = OffsetBase + Layout.getFieldOffset(idx);
-        uint64_t Size = i->getBitWidth()->EvaluateAsInt(Context).getZExtValue();
-
-        uint64_t EB_Lo = Offset / 64;
-        uint64_t EB_Hi = (Offset + Size - 1) / 64;
-        FieldLo = FieldHi = NoClass;
-        if (EB_Lo) {
-          assert(EB_Hi == EB_Lo && "Invalid classification, type > 16 bytes.");
-          FieldLo = NoClass;
-          FieldHi = Integer;
-        } else { 
-          FieldLo = Integer;
-          FieldHi = EB_Hi ? Integer : NoClass;
-        }
-      } else
-        classify(i->getType(), Context, Offset, FieldLo, FieldHi);
-      Lo = merge(Lo, FieldLo);
-      Hi = merge(Hi, FieldHi);
-      if (Lo == Memory || Hi == Memory)
-        break;
-    }
-
-    // AMD64-ABI 3.2.3p2: Rule 5. Then a post merger cleanup is done:
-    //
-    // (a) If one of the classes is MEMORY, the whole argument is
-    // passed in memory.
-    //
-    // (b) If SSEUP is not preceeded by SSE, it is converted to SSE.
-
-    // The first of these conditions is guaranteed by how we implement
-    // the merge (just bail). 
-    //
-    // The second condition occurs in the case of unions; for example
-    // union { _Complex double; unsigned; }.
-    if (Hi == Memory)
-      Lo = Memory;
-    if (Hi == SSEUp && Lo != SSE)
-      Hi = SSE;
-  }
-}
-
-ABIArgInfo X86_64ABIInfo::getCoerceResult(QualType Ty,
-                                          const llvm::Type *CoerceTo,
-                                          ASTContext &Context) const {
-  if (CoerceTo == llvm::Type::Int64Ty) {
-    // Integer and pointer types will end up in a general purpose
-    // register.
-    if (Ty->isIntegralType() || Ty->isPointerType())
-      return ABIArgInfo::getDirect();
-
-  } else if (CoerceTo == llvm::Type::DoubleTy) {
-    // FIXME: It would probably be better to make CGFunctionInfo only map using
-    // canonical types than to canonize here.
-    QualType CTy = Context.getCanonicalType(Ty);
-  
-    // Float and double end up in a single SSE reg.
-    if (CTy == Context.FloatTy || CTy == Context.DoubleTy)
-      return ABIArgInfo::getDirect();
-
-  }
-
-  return ABIArgInfo::getCoerce(CoerceTo);
-}
-
-ABIArgInfo X86_64ABIInfo::getIndirectResult(QualType Ty,
-                                            ASTContext &Context) const {
-  // If this is a scalar LLVM value then assume LLVM will pass it in the right
-  // place naturally.
-  if (!CodeGenFunction::hasAggregateLLVMType(Ty))
-    return ABIArgInfo::getDirect();
-
-  // FIXME: Set alignment correctly.
-  return ABIArgInfo::getIndirect(0);
-}
-
-ABIArgInfo X86_64ABIInfo::classifyReturnType(QualType RetTy,
-                                            ASTContext &Context) const {
-  // AMD64-ABI 3.2.3p4: Rule 1. Classify the return type with the
-  // classification algorithm.
-  X86_64ABIInfo::Class Lo, Hi;
-  classify(RetTy, Context, 0, Lo, Hi);
-
-  // Check some invariants.
-  assert((Hi != Memory || Lo == Memory) && "Invalid memory classification.");
-  assert((Lo != NoClass || Hi == NoClass) && "Invalid null classification.");
-  assert((Hi != SSEUp || Lo == SSE) && "Invalid SSEUp classification.");
-
-  const llvm::Type *ResType = 0;
-  switch (Lo) {
-  case NoClass:
-    return ABIArgInfo::getIgnore();
-
-  case SSEUp:
-  case X87Up:
-    assert(0 && "Invalid classification for lo word.");
-
-    // AMD64-ABI 3.2.3p4: Rule 2. Types of class memory are returned via
-    // hidden argument.
-  case Memory:
-    return getIndirectResult(RetTy, Context);
-
-    // AMD64-ABI 3.2.3p4: Rule 3. If the class is INTEGER, the next
-    // available register of the sequence %rax, %rdx is used.
-  case Integer:
-    ResType = llvm::Type::Int64Ty; break;
-
-    // AMD64-ABI 3.2.3p4: Rule 4. If the class is SSE, the next
-    // available SSE register of the sequence %xmm0, %xmm1 is used.
-  case SSE:
-    ResType = llvm::Type::DoubleTy; break;
-
-    // AMD64-ABI 3.2.3p4: Rule 6. If the class is X87, the value is
-    // returned on the X87 stack in %st0 as 80-bit x87 number.
-  case X87:
-    ResType = llvm::Type::X86_FP80Ty; break;
-
-    // AMD64-ABI 3.2.3p4: Rule 8. If the class is COMPLEX_X87, the real
-    // part of the value is returned in %st0 and the imaginary part in
-    // %st1.
-  case ComplexX87:
-    assert(Hi == ComplexX87 && "Unexpected ComplexX87 classification.");
-    ResType = llvm::StructType::get(llvm::Type::X86_FP80Ty,
-                                    llvm::Type::X86_FP80Ty,
-                                    NULL);
-    break;    
-  }
-
-  switch (Hi) {
-    // Memory was handled previously and X87 should
-    // never occur as a hi class.
-  case Memory:
-  case X87:
-    assert(0 && "Invalid classification for hi word.");
-
-  case ComplexX87: // Previously handled.
-  case NoClass: break;
-
-  case Integer:
-    ResType = llvm::StructType::get(ResType, llvm::Type::Int64Ty, NULL);
-    break;
-  case SSE:    
-    ResType = llvm::StructType::get(ResType, llvm::Type::DoubleTy, NULL);
-    break;
-
-    // AMD64-ABI 3.2.3p4: Rule 5. If the class is SSEUP, the eightbyte
-    // is passed in the upper half of the last used SSE register.
-    //
-    // SSEUP should always be preceeded by SSE, just widen.
-  case SSEUp:
-    assert(Lo == SSE && "Unexpected SSEUp classification.");
-    ResType = llvm::VectorType::get(llvm::Type::DoubleTy, 2);
-    break;
-
-    // AMD64-ABI 3.2.3p4: Rule 7. If the class is X87UP, the value is
-    // returned together with the previous X87 value in %st0.
-  case X87Up:
-    // If X87Up is preceeded by X87, we don't need to do
-    // anything. However, in some cases with unions it may not be
-    // preceeded by X87. In such situations we follow gcc and pass the
-    // extra bits in an SSE reg.
-    if (Lo != X87) 
-      ResType = llvm::StructType::get(ResType, llvm::Type::DoubleTy, NULL);
-    break;
-  }
-
-  return getCoerceResult(RetTy, ResType, Context);
-}
-
-ABIArgInfo X86_64ABIInfo::classifyArgumentType(QualType Ty, ASTContext &Context,
-                                               unsigned &neededInt,
-                                               unsigned &neededSSE) const {
-  X86_64ABIInfo::Class Lo, Hi;
-  classify(Ty, Context, 0, Lo, Hi);
-  
-  // Check some invariants.
-  // FIXME: Enforce these by construction.
-  assert((Hi != Memory || Lo == Memory) && "Invalid memory classification.");
-  assert((Lo != NoClass || Hi == NoClass) && "Invalid null classification.");
-  assert((Hi != SSEUp || Lo == SSE) && "Invalid SSEUp classification.");
-
-  neededInt = 0;
-  neededSSE = 0;
-  const llvm::Type *ResType = 0;
-  switch (Lo) {
-  case NoClass:
-    return ABIArgInfo::getIgnore();
-
-    // AMD64-ABI 3.2.3p3: Rule 1. If the class is MEMORY, pass the argument
-    // on the stack.
-  case Memory:
-    
-    // AMD64-ABI 3.2.3p3: Rule 5. If the class is X87, X87UP or
-    // COMPLEX_X87, it is passed in memory.
-  case X87:
-  case ComplexX87:
-    return getIndirectResult(Ty, Context);
-
-  case SSEUp:
-  case X87Up:
-    assert(0 && "Invalid classification for lo word.");
-
-    // AMD64-ABI 3.2.3p3: Rule 2. If the class is INTEGER, the next
-    // available register of the sequence %rdi, %rsi, %rdx, %rcx, %r8
-    // and %r9 is used.
-  case Integer:
-    ++neededInt; 
-    ResType = llvm::Type::Int64Ty;
-    break;
-
-    // AMD64-ABI 3.2.3p3: Rule 3. If the class is SSE, the next
-    // available SSE register is used, the registers are taken in the
-    // order from %xmm0 to %xmm7.
-  case SSE:
-    ++neededSSE; 
-    ResType = llvm::Type::DoubleTy;
-    break;
-  }
-
-  switch (Hi) {
-    // Memory was handled previously, ComplexX87 and X87 should
-    // never occur as hi classes, and X87Up must be preceed by X87,
-    // which is passed in memory.
-  case Memory:
-  case X87:
-  case ComplexX87:
-    assert(0 && "Invalid classification for hi word.");
-    break;
-
-  case NoClass: break;
-  case Integer:
-    ResType = llvm::StructType::get(ResType, llvm::Type::Int64Ty, NULL);
-    ++neededInt;
-    break;
-
-    // X87Up generally doesn't occur here (long double is passed in
-    // memory), except in situations involving unions.
-  case X87Up:
-  case SSE:
-    ResType = llvm::StructType::get(ResType, llvm::Type::DoubleTy, NULL);
-    ++neededSSE;
-    break;
-
-    // AMD64-ABI 3.2.3p3: Rule 4. If the class is SSEUP, the
-    // eightbyte is passed in the upper half of the last used SSE
-    // register.
-  case SSEUp:
-    assert(Lo == SSE && "Unexpected SSEUp classification.");
-    ResType = llvm::VectorType::get(llvm::Type::DoubleTy, 2);
-    break;
-  }
-
-  return getCoerceResult(Ty, ResType, Context);
-}
-
-void X86_64ABIInfo::computeInfo(CGFunctionInfo &FI, ASTContext &Context) const {
-  FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), Context);
-
-  // Keep track of the number of assigned registers.
-  unsigned freeIntRegs = 6, freeSSERegs = 8;
-  
-  // If the return value is indirect, then the hidden argument is consuming one
-  // integer register.
-  if (FI.getReturnInfo().isIndirect())
-    --freeIntRegs;
-
-  // AMD64-ABI 3.2.3p3: Once arguments are classified, the registers
-  // get assigned (in left-to-right order) for passing as follows...
-  for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
-       it != ie; ++it) {
-    unsigned neededInt, neededSSE;
-    it->info = classifyArgumentType(it->type, Context, neededInt, neededSSE);
-
-    // AMD64-ABI 3.2.3p3: If there are no registers available for any
-    // eightbyte of an argument, the whole argument is passed on the
-    // stack. If registers have already been assigned for some
-    // eightbytes of such an argument, the assignments get reverted.
-    if (freeIntRegs >= neededInt && freeSSERegs >= neededSSE) {
-      freeIntRegs -= neededInt;
-      freeSSERegs -= neededSSE;
-    } else {
-      it->info = getIndirectResult(it->type, Context);
-    }
-  }
-}
-
-static llvm::Value *EmitVAArgFromMemory(llvm::Value *VAListAddr, 
-                                        QualType Ty,
-                                        CodeGenFunction &CGF) {
-  llvm::Value *overflow_arg_area_p = 
-    CGF.Builder.CreateStructGEP(VAListAddr, 2, "overflow_arg_area_p");
-  llvm::Value *overflow_arg_area = 
-    CGF.Builder.CreateLoad(overflow_arg_area_p, "overflow_arg_area");
-
-  // AMD64-ABI 3.5.7p5: Step 7. Align l->overflow_arg_area upwards to a 16
-  // byte boundary if alignment needed by type exceeds 8 byte boundary.
-  uint64_t Align = CGF.getContext().getTypeAlign(Ty) / 8;
-  if (Align > 8) {
-    // Note that we follow the ABI & gcc here, even though the type
-    // could in theory have an alignment greater than 16. This case
-    // shouldn't ever matter in practice.
-
-    // overflow_arg_area = (overflow_arg_area + 15) & ~15;
-    llvm::Value *Offset = llvm::ConstantInt::get(llvm::Type::Int32Ty, 15);
-    overflow_arg_area = CGF.Builder.CreateGEP(overflow_arg_area, Offset);
-    llvm::Value *AsInt = CGF.Builder.CreatePtrToInt(overflow_arg_area,
-                                                    llvm::Type::Int64Ty);
-    llvm::Value *Mask = llvm::ConstantInt::get(llvm::Type::Int64Ty, ~15LL);
-    overflow_arg_area = 
-      CGF.Builder.CreateIntToPtr(CGF.Builder.CreateAnd(AsInt, Mask),
-                                 overflow_arg_area->getType(),
-                                 "overflow_arg_area.align");
-  }
-
-  // AMD64-ABI 3.5.7p5: Step 8. Fetch type from l->overflow_arg_area.
-  const llvm::Type *LTy = CGF.ConvertTypeForMem(Ty);
-  llvm::Value *Res = 
-    CGF.Builder.CreateBitCast(overflow_arg_area, 
-                              llvm::PointerType::getUnqual(LTy));
-
-  // AMD64-ABI 3.5.7p5: Step 9. Set l->overflow_arg_area to:
-  // l->overflow_arg_area + sizeof(type).
-  // AMD64-ABI 3.5.7p5: Step 10. Align l->overflow_arg_area upwards to
-  // an 8 byte boundary.
-
-  uint64_t SizeInBytes = (CGF.getContext().getTypeSize(Ty) + 7) / 8;
-  llvm::Value *Offset = llvm::ConstantInt::get(llvm::Type::Int32Ty,
-                                               (SizeInBytes + 7)  & ~7);
-  overflow_arg_area = CGF.Builder.CreateGEP(overflow_arg_area, Offset,
-                                            "overflow_arg_area.next");
-  CGF.Builder.CreateStore(overflow_arg_area, overflow_arg_area_p);
-
-  // AMD64-ABI 3.5.7p5: Step 11. Return the fetched type.  
-  return Res;
-}
-
-llvm::Value *X86_64ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                      CodeGenFunction &CGF) const {
-  // Assume that va_list type is correct; should be pointer to LLVM type:
-  // struct {
-  //   i32 gp_offset;
-  //   i32 fp_offset;
-  //   i8* overflow_arg_area;
-  //   i8* reg_save_area;
-  // }; 
-  unsigned neededInt, neededSSE;
-  ABIArgInfo AI = classifyArgumentType(Ty, CGF.getContext(), 
-                                       neededInt, neededSSE);
-
-  // AMD64-ABI 3.5.7p5: Step 1. Determine whether type may be passed
-  // in the registers. If not go to step 7.
-  if (!neededInt && !neededSSE)
-    return EmitVAArgFromMemory(VAListAddr, Ty, CGF);
-
-  // AMD64-ABI 3.5.7p5: Step 2. Compute num_gp to hold the number of
-  // general purpose registers needed to pass type and num_fp to hold
-  // the number of floating point registers needed.
-
-  // AMD64-ABI 3.5.7p5: Step 3. Verify whether arguments fit into
-  // registers. In the case: l->gp_offset > 48 - num_gp * 8 or
-  // l->fp_offset > 304 - num_fp * 16 go to step 7.
-  // 
-  // NOTE: 304 is a typo, there are (6 * 8 + 8 * 16) = 176 bytes of
-  // register save space).
-
-  llvm::Value *InRegs = 0;
-  llvm::Value *gp_offset_p = 0, *gp_offset = 0;
-  llvm::Value *fp_offset_p = 0, *fp_offset = 0;
-  if (neededInt) {
-    gp_offset_p = CGF.Builder.CreateStructGEP(VAListAddr, 0, "gp_offset_p");
-    gp_offset = CGF.Builder.CreateLoad(gp_offset_p, "gp_offset");
-    InRegs = 
-      CGF.Builder.CreateICmpULE(gp_offset,
-                                llvm::ConstantInt::get(llvm::Type::Int32Ty,
-                                                       48 - neededInt * 8),
-                                "fits_in_gp");
-  }
-
-  if (neededSSE) {
-    fp_offset_p = CGF.Builder.CreateStructGEP(VAListAddr, 1, "fp_offset_p");
-    fp_offset = CGF.Builder.CreateLoad(fp_offset_p, "fp_offset");
-    llvm::Value *FitsInFP = 
-      CGF.Builder.CreateICmpULE(fp_offset,
-                                llvm::ConstantInt::get(llvm::Type::Int32Ty,
-                                                       176 - neededSSE * 16),
-                                "fits_in_fp");
-    InRegs = InRegs ? CGF.Builder.CreateAnd(InRegs, FitsInFP) : FitsInFP;
-  }
-
-  llvm::BasicBlock *InRegBlock = CGF.createBasicBlock("vaarg.in_reg");
-  llvm::BasicBlock *InMemBlock = CGF.createBasicBlock("vaarg.in_mem");
-  llvm::BasicBlock *ContBlock = CGF.createBasicBlock("vaarg.end");
-  CGF.Builder.CreateCondBr(InRegs, InRegBlock, InMemBlock);
-  
-  // Emit code to load the value if it was passed in registers.
-  
-  CGF.EmitBlock(InRegBlock);
-
-  // AMD64-ABI 3.5.7p5: Step 4. Fetch type from l->reg_save_area with
-  // an offset of l->gp_offset and/or l->fp_offset. This may require
-  // copying to a temporary location in case the parameter is passed
-  // in different register classes or requires an alignment greater
-  // than 8 for general purpose registers and 16 for XMM registers.
-  //
-  // FIXME: This really results in shameful code when we end up needing to
-  // collect arguments from different places; often what should result in a
-  // simple assembling of a structure from scattered addresses has many more
-  // loads than necessary. Can we clean this up?
-  const llvm::Type *LTy = CGF.ConvertTypeForMem(Ty);
-  llvm::Value *RegAddr = 
-    CGF.Builder.CreateLoad(CGF.Builder.CreateStructGEP(VAListAddr, 3), 
-                           "reg_save_area");
-  if (neededInt && neededSSE) {
-    // FIXME: Cleanup.
-    assert(AI.isCoerce() && "Unexpected ABI info for mixed regs");
-    const llvm::StructType *ST = cast<llvm::StructType>(AI.getCoerceToType());
-    llvm::Value *Tmp = CGF.CreateTempAlloca(ST);
-    assert(ST->getNumElements() == 2 && "Unexpected ABI info for mixed regs");
-    const llvm::Type *TyLo = ST->getElementType(0);
-    const llvm::Type *TyHi = ST->getElementType(1);
-    assert((TyLo->isFloatingPoint() ^ TyHi->isFloatingPoint()) &&
-           "Unexpected ABI info for mixed regs");
-    const llvm::Type *PTyLo = llvm::PointerType::getUnqual(TyLo);
-    const llvm::Type *PTyHi = llvm::PointerType::getUnqual(TyHi);
-    llvm::Value *GPAddr = CGF.Builder.CreateGEP(RegAddr, gp_offset);
-    llvm::Value *FPAddr = CGF.Builder.CreateGEP(RegAddr, fp_offset);
-    llvm::Value *RegLoAddr = TyLo->isFloatingPoint() ? FPAddr : GPAddr;
-    llvm::Value *RegHiAddr = TyLo->isFloatingPoint() ? GPAddr : FPAddr;
-    llvm::Value *V = 
-      CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegLoAddr, PTyLo));
-    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 0));
-    V = CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegHiAddr, PTyHi));
-    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 1));
-    
-    RegAddr = CGF.Builder.CreateBitCast(Tmp, llvm::PointerType::getUnqual(LTy));
-  } else if (neededInt) {
-    RegAddr = CGF.Builder.CreateGEP(RegAddr, gp_offset);
-    RegAddr = CGF.Builder.CreateBitCast(RegAddr, 
-                                        llvm::PointerType::getUnqual(LTy));
-  } else {
-    if (neededSSE == 1) {
-      RegAddr = CGF.Builder.CreateGEP(RegAddr, fp_offset);
-      RegAddr = CGF.Builder.CreateBitCast(RegAddr, 
-                                          llvm::PointerType::getUnqual(LTy));
-    } else {
-      assert(neededSSE == 2 && "Invalid number of needed registers!");
-      // SSE registers are spaced 16 bytes apart in the register save
-      // area, we need to collect the two eightbytes together.
-      llvm::Value *RegAddrLo = CGF.Builder.CreateGEP(RegAddr, fp_offset);
-      llvm::Value *RegAddrHi = 
-        CGF.Builder.CreateGEP(RegAddrLo, 
-                              llvm::ConstantInt::get(llvm::Type::Int32Ty, 16));
-      const llvm::Type *DblPtrTy = 
-        llvm::PointerType::getUnqual(llvm::Type::DoubleTy);
-      const llvm::StructType *ST = llvm::StructType::get(llvm::Type::DoubleTy,
-                                                         llvm::Type::DoubleTy,
-                                                         NULL);
-      llvm::Value *V, *Tmp = CGF.CreateTempAlloca(ST);
-      V = CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegAddrLo, 
-                                                           DblPtrTy));
-      CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 0));
-      V = CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegAddrHi, 
-                                                           DblPtrTy));
-      CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 1));
-      RegAddr = CGF.Builder.CreateBitCast(Tmp, 
-                                          llvm::PointerType::getUnqual(LTy));
-    }
-  }
-
-  // AMD64-ABI 3.5.7p5: Step 5. Set: 
-  // l->gp_offset = l->gp_offset + num_gp * 8 
-  // l->fp_offset = l->fp_offset + num_fp * 16.
-  if (neededInt) {
-    llvm::Value *Offset = llvm::ConstantInt::get(llvm::Type::Int32Ty,
-                                                 neededInt * 8);
-    CGF.Builder.CreateStore(CGF.Builder.CreateAdd(gp_offset, Offset),
-                            gp_offset_p);
-  }
-  if (neededSSE) {
-    llvm::Value *Offset = llvm::ConstantInt::get(llvm::Type::Int32Ty,
-                                                 neededSSE * 16);
-    CGF.Builder.CreateStore(CGF.Builder.CreateAdd(fp_offset, Offset),
-                            fp_offset_p);
-  }
-  CGF.EmitBranch(ContBlock);
-
-  // Emit code to load the value if it was passed in memory.
-  
-  CGF.EmitBlock(InMemBlock);
-  llvm::Value *MemAddr = EmitVAArgFromMemory(VAListAddr, Ty, CGF);
-
-  // Return the appropriate result.
-
-  CGF.EmitBlock(ContBlock);  
-  llvm::PHINode *ResAddr = CGF.Builder.CreatePHI(RegAddr->getType(), 
-                                                 "vaarg.addr");
-  ResAddr->reserveOperandSpace(2);
-  ResAddr->addIncoming(RegAddr, InRegBlock);
-  ResAddr->addIncoming(MemAddr, InMemBlock);
-  
-  return ResAddr;
-}
-
-// ABI Info for PIC16 
-class PIC16ABIInfo : public ABIInfo {
-  ABIArgInfo classifyReturnType(QualType RetTy, 
-                                ASTContext &Context) const;
-  
-  ABIArgInfo classifyArgumentType(QualType RetTy,
-                                  ASTContext &Context) const;
-
-  virtual void computeInfo(CGFunctionInfo &FI, ASTContext &Context) const {
-    FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), Context);
-    for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
-         it != ie; ++it)
-      it->info = classifyArgumentType(it->type, Context);
-  }
-
-  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                 CodeGenFunction &CGF) const;
-
-};
-
-ABIArgInfo PIC16ABIInfo::classifyReturnType(QualType RetTy,
-                                              ASTContext &Context) const {
-  if (RetTy->isVoidType()) {
-    return ABIArgInfo::getIgnore();
-  } else {
-    return ABIArgInfo::getDirect();
-  }
-}
-
-ABIArgInfo PIC16ABIInfo::classifyArgumentType(QualType Ty,
-                                                ASTContext &Context) const {
-  return ABIArgInfo::getDirect();
-}
-
-llvm::Value *PIC16ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                       CodeGenFunction &CGF) const {
-  return 0;
-}
-
-class ARMABIInfo : public ABIInfo {
-  ABIArgInfo classifyReturnType(QualType RetTy, 
-                                ASTContext &Context) const;
-  
-  ABIArgInfo classifyArgumentType(QualType RetTy,
-                                  ASTContext &Context) const;
-
-  virtual void computeInfo(CGFunctionInfo &FI, ASTContext &Context) const;
-
-  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                 CodeGenFunction &CGF) const;
-};
-
-void ARMABIInfo::computeInfo(CGFunctionInfo &FI, ASTContext &Context) const {
-  FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), Context);
-  for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
-       it != ie; ++it) {
-    it->info = classifyArgumentType(it->type, Context);
-  }
-}
-
-ABIArgInfo ARMABIInfo::classifyArgumentType(QualType Ty,
-                                            ASTContext &Context) const {
-  if (!CodeGenFunction::hasAggregateLLVMType(Ty)) {
-    return ABIArgInfo::getDirect();
-  }
-  // FIXME: This is kind of nasty... but there isn't much choice because the ARM
-  // backend doesn't support byval.
-  // FIXME: This doesn't handle alignment > 64 bits.
-  const llvm::Type* ElemTy;
-  unsigned SizeRegs;
-  if (Context.getTypeAlign(Ty) > 32) {
-    ElemTy = llvm::Type::Int64Ty;
-    SizeRegs = (Context.getTypeSize(Ty) + 63) / 64;
-  } else {
-    ElemTy = llvm::Type::Int32Ty;
-    SizeRegs = (Context.getTypeSize(Ty) + 31) / 32;
-  }
-  std::vector<const llvm::Type*> LLVMFields;
-  LLVMFields.push_back(llvm::ArrayType::get(ElemTy, SizeRegs));
-  const llvm::Type* STy = llvm::StructType::get(LLVMFields, true);
-  return ABIArgInfo::getCoerce(STy);
-}
-
-ABIArgInfo ARMABIInfo::classifyReturnType(QualType RetTy,
-                                          ASTContext &Context) const {
-  if (RetTy->isVoidType()) {
-    return ABIArgInfo::getIgnore();
-  } else if (CodeGenFunction::hasAggregateLLVMType(RetTy)) {
-    // Aggregates <= 4 bytes are returned in r0; other aggregates
-    // are returned indirectly.
-    uint64_t Size = Context.getTypeSize(RetTy);
-    if (Size <= 32)
-      return ABIArgInfo::getCoerce(llvm::Type::Int32Ty);
-    return ABIArgInfo::getIndirect(0);
-  } else {
-    return ABIArgInfo::getDirect();
-  }
-}
-
-llvm::Value *ARMABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                      CodeGenFunction &CGF) const {
-  // FIXME: Need to handle alignment
-  const llvm::Type *BP = llvm::PointerType::getUnqual(llvm::Type::Int8Ty);
-  const llvm::Type *BPP = llvm::PointerType::getUnqual(BP);
-
-  CGBuilderTy &Builder = CGF.Builder;
-  llvm::Value *VAListAddrAsBPP = Builder.CreateBitCast(VAListAddr, BPP, 
-                                                       "ap");
-  llvm::Value *Addr = Builder.CreateLoad(VAListAddrAsBPP, "ap.cur");
-  llvm::Type *PTy = 
-    llvm::PointerType::getUnqual(CGF.ConvertType(Ty));
-  llvm::Value *AddrTyped = Builder.CreateBitCast(Addr, PTy);
-  
-  uint64_t Offset = 
-    llvm::RoundUpToAlignment(CGF.getContext().getTypeSize(Ty) / 8, 4);
-  llvm::Value *NextAddr = 
-    Builder.CreateGEP(Addr, 
-                      llvm::ConstantInt::get(llvm::Type::Int32Ty, Offset),
-                      "ap.next");
-  Builder.CreateStore(NextAddr, VAListAddrAsBPP);
-
-  return AddrTyped;
-}
-
-ABIArgInfo DefaultABIInfo::classifyReturnType(QualType RetTy,
-                                              ASTContext &Context) const {
-  if (RetTy->isVoidType()) {
-    return ABIArgInfo::getIgnore();
-  } else if (CodeGenFunction::hasAggregateLLVMType(RetTy)) {
-    return ABIArgInfo::getIndirect(0);
-  } else {
-    return ABIArgInfo::getDirect();
-  }
-}
-
-ABIArgInfo DefaultABIInfo::classifyArgumentType(QualType Ty,
-                                                ASTContext &Context) const {
-  if (CodeGenFunction::hasAggregateLLVMType(Ty)) {
-    return ABIArgInfo::getIndirect(0);
-  } else {
-    return ABIArgInfo::getDirect();
-  }
-}
-
-llvm::Value *DefaultABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                       CodeGenFunction &CGF) const {
-  return 0;
-}
-
-const ABIInfo &CodeGenTypes::getABIInfo() const {
-  if (TheABIInfo)
-    return *TheABIInfo;
-
-  // For now we just cache this in the CodeGenTypes and don't bother
-  // to free it.
-  const char *TargetPrefix = getContext().Target.getTargetPrefix();
-  if (strcmp(TargetPrefix, "x86") == 0) {
-    bool IsDarwin = strstr(getContext().Target.getTargetTriple(), "darwin");
-    switch (getContext().Target.getPointerWidth(0)) {
-    case 32:
-      return *(TheABIInfo = new X86_32ABIInfo(Context, IsDarwin));
-    case 64:
-      return *(TheABIInfo = new X86_64ABIInfo());
-    }
-  } else if (strcmp(TargetPrefix, "arm") == 0) {
-    // FIXME: Support for OABI?
-    return *(TheABIInfo = new ARMABIInfo());
-  } else if (strcmp(TargetPrefix, "pic16") == 0) {
-    return *(TheABIInfo = new PIC16ABIInfo());
-  }
-
-  return *(TheABIInfo = new DefaultABIInfo);
-}
-
-/***/
-
 CGFunctionInfo::CGFunctionInfo(QualType ResTy, 
                                const llvm::SmallVector<QualType, 16> &ArgTys) {
   NumArgs = ArgTys.size();
@@ -1637,13 +270,7 @@ static void CreateCoercedStore(llvm::Value *Src,
   uint64_t DstSize = CGF.CGM.getTargetData().getTypeAllocSize(DstTy);
 
   // If store is legal, just bitcast the src pointer.
-  if (SrcSize >= DstSize) {
-    // Generally SrcSize is never greater than DstSize, since this means we are
-    // losing bits. However, this can happen in cases where the structure has
-    // additional padding, for example due to a user specified alignment.
-    //
-    // FIXME: Assert that we aren't truncating non-padding bits when have access
-    // to that information.
+  if (SrcSize <= DstSize) {
     llvm::Value *Casted =
       CGF.Builder.CreateBitCast(DstPtr, llvm::PointerType::getUnqual(SrcTy));
     // FIXME: Use better alignment / avoid requiring aligned store.
@@ -1651,6 +278,13 @@ static void CreateCoercedStore(llvm::Value *Src,
   } else {
     // Otherwise do coercion through memory. This is stupid, but
     // simple.
+
+    // Generally SrcSize is never greater than DstSize, since this means we are
+    // losing bits. However, this can happen in cases where the structure has
+    // additional padding, for example due to a user specified alignment.
+    //
+    // FIXME: Assert that we aren't truncating non-padding bits when have access
+    // to that information.
     llvm::Value *Tmp = CGF.CreateTempAlloca(SrcTy);
     CGF.Builder.CreateStore(Src, Tmp);
     llvm::Value *Casted = 
@@ -1751,6 +385,11 @@ void CodeGenModule::ConstructAttributeList(const CGFunctionInfo &FI,
       FuncAttrs |= llvm::Attribute::ReadOnly;
   }
 
+  if (CompileOpts.DisableRedZone)
+    FuncAttrs |= llvm::Attribute::NoRedZone;
+  if (CompileOpts.NoImplicitFloat)
+    FuncAttrs |= llvm::Attribute::NoImplicitFloat;
+
   QualType RetTy = FI.getReturnType();
   unsigned Index = 1;
   const ABIArgInfo &RetAI = FI.getReturnInfo();
diff --git a/lib/CodeGen/CGDecl.cpp b/lib/CodeGen/CGDecl.cpp
index bcad77b..29eaaad 100644
--- a/lib/CodeGen/CGDecl.cpp
+++ b/lib/CodeGen/CGDecl.cpp
@@ -232,7 +232,9 @@ const llvm::Type *CodeGenFunction::BuildByRefType(QualType Ty,
     Types[4] = PtrToInt8Ty;
     Types[5] = PtrToInt8Ty;
   }
-  // FIXME: Align this on at least an Align boundary.
+  // FIXME: Align this on at least an Align boundary, assert if we can't.
+  assert((Align <= unsigned(Target.getPointerAlign(0))/8)
+         && "Can't align more thqn pointer yet");
   Types[needsCopyDispose*2 + 4] = LTy;
   return llvm::StructType::get(Types, false);
 }
@@ -244,22 +246,22 @@ void CodeGenFunction::EmitLocalBlockVarDecl(const VarDecl &D) {
   QualType Ty = D.getType();
   bool isByRef = D.hasAttr<BlocksAttr>();
   bool needsDispose = false;
+  unsigned Align = 0;
 
   llvm::Value *DeclPtr;
   if (Ty->isConstantSizeType()) {
     if (!Target.useGlobalsForAutomaticVariables()) {
       // A normal fixed sized variable becomes an alloca in the entry block.
       const llvm::Type *LTy = ConvertTypeForMem(Ty);
+      Align = getContext().getDeclAlignInBytes(&D);
       if (isByRef)
-        LTy = BuildByRefType(Ty, getContext().getDeclAlignInBytes(&D));
+        LTy = BuildByRefType(Ty, Align);
       llvm::AllocaInst *Alloc = CreateTempAlloca(LTy);
       Alloc->setName(D.getNameAsString().c_str());
       
       if (isByRef)
-        Alloc->setAlignment(std::max(getContext().getDeclAlignInBytes(&D),
-                                     unsigned(Target.getPointerAlign(0) / 8)));
-      else
-        Alloc->setAlignment(getContext().getDeclAlignInBytes(&D));
+        Align = std::max(Align, unsigned(Target.getPointerAlign(0) / 8));
+      Alloc->setAlignment(Align);
       DeclPtr = Alloc;
     } else {
       // Targets that don't support recursion emit locals as globals.
@@ -401,11 +403,12 @@ void CodeGenFunction::EmitLocalBlockVarDecl(const VarDecl &D) {
     if (flags & BLOCK_HAS_COPY_DISPOSE) {
       BlockHasCopyDispose = true;
       llvm::Value *copy_helper = Builder.CreateStructGEP(DeclPtr, 4);
-      Builder.CreateStore(BuildbyrefCopyHelper(DeclPtr->getType(), flag),
+      Builder.CreateStore(BuildbyrefCopyHelper(DeclPtr->getType(), flag, Align),
                           copy_helper);
 
       llvm::Value *destroy_helper = Builder.CreateStructGEP(DeclPtr, 5);
-      Builder.CreateStore(BuildbyrefDestroyHelper(DeclPtr->getType(), flag),
+      Builder.CreateStore(BuildbyrefDestroyHelper(DeclPtr->getType(), flag,
+                                                  Align),
                           destroy_helper);
     }
   }
diff --git a/lib/CodeGen/CGExprAgg.cpp b/lib/CodeGen/CGExprAgg.cpp
index 8d903d9..4268ae3 100644
--- a/lib/CodeGen/CGExprAgg.cpp
+++ b/lib/CodeGen/CGExprAgg.cpp
@@ -263,17 +263,21 @@ void AggExprEmitter::VisitConditionalOperator(const ConditionalOperator *E) {
   llvm::Value *Cond = CGF.EvaluateExprAsBool(E->getCond());
   Builder.CreateCondBr(Cond, LHSBlock, RHSBlock);
   
+  CGF.PushConditionalTempDestruction();
   CGF.EmitBlock(LHSBlock);
   
   // Handle the GNU extension for missing LHS.
   assert(E->getLHS() && "Must have LHS for aggregate value");
 
   Visit(E->getLHS());
+  CGF.PopConditionalTempDestruction();
   CGF.EmitBranch(ContBlock);
   
+  CGF.PushConditionalTempDestruction();
   CGF.EmitBlock(RHSBlock);
   
   Visit(E->getRHS());
+  CGF.PopConditionalTempDestruction();
   CGF.EmitBranch(ContBlock);
   
   CGF.EmitBlock(ContBlock);
diff --git a/lib/CodeGen/CGExprScalar.cpp b/lib/CodeGen/CGExprScalar.cpp
index 950e9e5..ed18d32 100644
--- a/lib/CodeGen/CGExprScalar.cpp
+++ b/lib/CodeGen/CGExprScalar.cpp
@@ -1290,8 +1290,10 @@ Value *ScalarExprEmitter::VisitBinLAnd(const BinaryOperator *E) {
        PI != PE; ++PI)
     PN->addIncoming(llvm::ConstantInt::getFalse(), *PI);
   
+  CGF.PushConditionalTempDestruction();
   CGF.EmitBlock(RHSBlock);
   Value *RHSCond = CGF.EvaluateExprAsBool(E->getRHS());
+  CGF.PopConditionalTempDestruction();
   
   // Reaquire the RHS block, as there may be subblocks inserted.
   RHSBlock = Builder.GetInsertBlock();
@@ -1335,10 +1337,14 @@ Value *ScalarExprEmitter::VisitBinLOr(const BinaryOperator *E) {
        PI != PE; ++PI)
     PN->addIncoming(llvm::ConstantInt::getTrue(), *PI);
 
+  CGF.PushConditionalTempDestruction();
+
   // Emit the RHS condition as a bool value.
   CGF.EmitBlock(RHSBlock);
   Value *RHSCond = CGF.EvaluateExprAsBool(E->getRHS());
   
+  CGF.PopConditionalTempDestruction();
+  
   // Reaquire the RHS block, as there may be subblocks inserted.
   RHSBlock = Builder.GetInsertBlock();
   
@@ -1446,7 +1452,8 @@ VisitConditionalOperator(const ConditionalOperator *E) {
                                CGF.getContext().BoolTy);
     Builder.CreateCondBr(CondBoolVal, LHSBlock, RHSBlock);
   }
-  
+
+  CGF.PushConditionalTempDestruction();
   CGF.EmitBlock(LHSBlock);
   
   // Handle the GNU extension for missing LHS.
@@ -1456,12 +1463,15 @@ VisitConditionalOperator(const ConditionalOperator *E) {
   else    // Perform promotions, to handle cases like "short ?: int"
     LHS = EmitScalarConversion(CondVal, E->getCond()->getType(), E->getType());
   
+  CGF.PopConditionalTempDestruction();
   LHSBlock = Builder.GetInsertBlock();
   CGF.EmitBranch(ContBlock);
   
+  CGF.PushConditionalTempDestruction();
   CGF.EmitBlock(RHSBlock);
   
   Value *RHS = Visit(E->getRHS());
+  CGF.PopConditionalTempDestruction();
   RHSBlock = Builder.GetInsertBlock();
   CGF.EmitBranch(ContBlock);
   
diff --git a/lib/CodeGen/CGObjCGNU.cpp b/lib/CodeGen/CGObjCGNU.cpp
index 5e7eec9..a70f718 100644
--- a/lib/CodeGen/CGObjCGNU.cpp
+++ b/lib/CodeGen/CGObjCGNU.cpp
@@ -1544,14 +1544,8 @@ LValue CGObjCGNU::EmitObjCValueForIvar(CodeGen::CodeGenFunction &CGF,
 static const ObjCInterfaceDecl *FindIvarInterface(ASTContext &Context,
                                                   const ObjCInterfaceDecl *OID,
                                                   const ObjCIvarDecl *OIVD) {
-  for (ObjCInterfaceDecl::ivar_iterator IVI = OID->ivar_begin(), 
-         IVE = OID->ivar_end(); IVI != IVE; ++IVI)
-    if (OIVD == *IVI)
-      return OID;
-  
-  // Also look in synthesized ivars.
   llvm::SmallVector<ObjCIvarDecl*, 16> Ivars;
-  Context.CollectSynthesizedIvars(OID, Ivars);
+  Context.ShallowCollectObjCIvars(OID, Ivars);
   for (unsigned k = 0, e = Ivars.size(); k != e; ++k) {
     if (OIVD == Ivars[k])
       return OID;
diff --git a/lib/CodeGen/CGObjCMac.cpp b/lib/CodeGen/CGObjCMac.cpp
index 8f1404d..75755ec 100644
--- a/lib/CodeGen/CGObjCMac.cpp
+++ b/lib/CodeGen/CGObjCMac.cpp
@@ -47,14 +47,8 @@ static const ObjCInterfaceDecl *FindIvarInterface(ASTContext &Context,
   // ASTContext::getObjCLayout is implemented. This should be fixed to
   // get the information from the layout directly.
   Index = 0;
-  for (ObjCInterfaceDecl::ivar_iterator IVI = OID->ivar_begin(), 
-         IVE = OID->ivar_end(); IVI != IVE; ++IVI, ++Index)
-    if (OIVD == *IVI)
-      return OID;
-  
-  // Also look in synthesized ivars.
   llvm::SmallVector<ObjCIvarDecl*, 16> Ivars;
-  Context.CollectSynthesizedIvars(OID, Ivars);
+  Context.ShallowCollectObjCIvars(OID, Ivars);
   for (unsigned k = 0, e = Ivars.size(); k != e; ++k) {
     if (OIVD == Ivars[k])
       return OID;
@@ -908,15 +902,6 @@ protected:
                                           unsigned Align,
                                           bool AddToUsed);
 
-  /// GetNamedIvarList - Return the list of ivars in the interface
-  /// itself (not including super classes and not including unnamed
-  /// bitfields).
-  /// 
-  /// For the non-fragile ABI, this also includes synthesized property
-  /// ivars.
-  void GetNamedIvarList(const ObjCInterfaceDecl *OID,
-                        llvm::SmallVector<ObjCIvarDecl*, 16> &Res) const;
-  
   CodeGen::RValue EmitLegacyMessageSend(CodeGen::CodeGenFunction &CGF,
                                         QualType ResultType,
                                         llvm::Value *Sel,
@@ -2198,10 +2183,13 @@ llvm::Constant *CGObjCMac::EmitIvarList(const ObjCImplementationDecl *ID,
     const_cast<ObjCInterfaceDecl*>(ID->getClassInterface());
   
   llvm::SmallVector<ObjCIvarDecl*, 16> OIvars;
-  GetNamedIvarList(OID, OIvars);
+  CGM.getContext().ShallowCollectObjCIvars(OID, OIvars);
     
   for (unsigned i = 0, e = OIvars.size(); i != e; ++i) {
     ObjCIvarDecl *IVD = OIvars[i];
+    // Ignore unnamed bit-fields.
+    if (!IVD->getDeclName())
+      continue;    
     Ivar[0] = GetMethodVarName(IVD->getIdentifier());
     Ivar[1] = GetMethodVarType(IVD);
     Ivar[2] = llvm::ConstantInt::get(ObjCTypes.IntTy, 
@@ -4696,25 +4684,6 @@ llvm::Constant * CGObjCNonFragileABIMac::EmitIvarOffsetVar(
 /// }
 ///
 
-void CGObjCCommonMac::GetNamedIvarList(const ObjCInterfaceDecl *OID,
-                              llvm::SmallVector<ObjCIvarDecl*, 16> &Res) const {
-  for (ObjCInterfaceDecl::ivar_iterator I = OID->ivar_begin(),
-         E = OID->ivar_end(); I != E; ++I) {
-    // Ignore unnamed bit-fields.
-    if (!(*I)->getDeclName())
-      continue;
-
-     Res.push_back(*I);
-  }
-  
-  // Also save synthesize ivars.
-  // FIXME. Why can't we just use passed in Res small vector?
-  llvm::SmallVector<ObjCIvarDecl*, 16> Ivars;
-  CGM.getContext().CollectSynthesizedIvars(OID, Ivars);
-  for (unsigned k = 0, e = Ivars.size(); k != e; ++k)
-    Res.push_back(Ivars[k]);
-}
-
 llvm::Constant *CGObjCNonFragileABIMac::EmitIvarList(
                                             const ObjCImplementationDecl *ID) {
   
@@ -4727,10 +4696,13 @@ llvm::Constant *CGObjCNonFragileABIMac::EmitIvarList(
   
   // Collect declared and synthesized ivars in a small vector.
   llvm::SmallVector<ObjCIvarDecl*, 16> OIvars;
-  GetNamedIvarList(OID, OIvars);
+  CGM.getContext().ShallowCollectObjCIvars(OID, OIvars);
     
   for (unsigned i = 0, e = OIvars.size(); i != e; ++i) {
     ObjCIvarDecl *IVD = OIvars[i];
+    // Ignore unnamed bit-fields.
+    if (!IVD->getDeclName())
+      continue;
     Ivar[0] = EmitIvarOffsetVar(ID->getClassInterface(), IVD, 
                                 ComputeIvarBaseOffset(CGM, ID, IVD));
     Ivar[1] = GetMethodVarName(IVD->getIdentifier());
diff --git a/lib/CodeGen/CMakeLists.txt b/lib/CodeGen/CMakeLists.txt
index d6c46a8..c206a3b 100644
--- a/lib/CodeGen/CMakeLists.txt
+++ b/lib/CodeGen/CMakeLists.txt
@@ -5,6 +5,7 @@ add_clang_library(clangCodeGen
   CGBlocks.cpp
   CGCall.cpp
   CGCXX.cpp
+  CGCXXTemp.cpp
   CGDebugInfo.cpp
   CGDecl.cpp
   CGExprAgg.cpp
@@ -21,4 +22,5 @@ add_clang_library(clangCodeGen
   CodeGenTypes.cpp
   Mangle.cpp
   ModuleBuilder.cpp
+  TargetABIInfo.cpp
   )
diff --git a/lib/CodeGen/CodeGenFunction.h b/lib/CodeGen/CodeGenFunction.h
index c91a052..72c4aa4 100644
--- a/lib/CodeGen/CodeGenFunction.h
+++ b/lib/CodeGen/CodeGenFunction.h
@@ -160,6 +160,20 @@ public:
   /// this behavior for branches?
   void EmitBranchThroughCleanup(llvm::BasicBlock *Dest);
 
+  /// PushConditionalTempDestruction - Should be called before a conditional 
+  /// part of an expression is emitted. For example, before the RHS of the
+  /// expression below is emitted:
+  /// 
+  /// b && f(T());
+  ///
+  /// This is used to make sure that any temporaryes created in the conditional
+  /// branch are only destroyed if the branch is taken.
+  void PushConditionalTempDestruction();
+  
+  /// PopConditionalTempDestruction - Should be called after a conditional 
+  /// part of an expression has been emitted.
+  void PopConditionalTempDestruction();
+  
 private:
   CGDebugInfo* DebugInfo;
 
@@ -263,6 +277,11 @@ private:
   };
   
   llvm::SmallVector<CXXLiveTemporaryInfo, 4> LiveTemporaries;
+
+  /// ConditionalTempDestructionStack - Contains the number of live temporaries 
+  /// when PushConditionalTempDestruction was called. This is used so that
+  /// we know how many temporaries were created by a certain expression.
+  llvm::SmallVector<size_t, 4> ConditionalTempDestructionStack;
   
 public:
   CodeGenFunction(CodeGenModule &cgm);
diff --git a/lib/CodeGen/CodeGenModule.cpp b/lib/CodeGen/CodeGenModule.cpp
index b69301e..5c12c81 100644
--- a/lib/CodeGen/CodeGenModule.cpp
+++ b/lib/CodeGen/CodeGenModule.cpp
@@ -298,7 +298,7 @@ void CodeGenModule::SetFunctionDefinitionAttributes(const FunctionDecl *D,
     GV->setLinkage(llvm::Function::InternalLinkage);
   } else if (D->hasAttr<DLLExportAttr>()) {
     GV->setLinkage(llvm::Function::DLLExportLinkage);
-  } else if (D->hasAttr<WeakAttr>() || D->hasAttr<WeakImportAttr>()) {
+  } else if (D->hasAttr<WeakAttr>()) {
     GV->setLinkage(llvm::Function::WeakAnyLinkage);
   } else if (Linkage == GVA_C99Inline) {
     // In C99 mode, 'inline' functions are guaranteed to have a strong
@@ -853,7 +853,7 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D) {
     GV->setLinkage(llvm::Function::DLLImportLinkage);
   else if (D->hasAttr<DLLExportAttr>())
     GV->setLinkage(llvm::Function::DLLExportLinkage);
-  else if (D->hasAttr<WeakAttr>() || D->hasAttr<WeakImportAttr>())
+  else if (D->hasAttr<WeakAttr>())
     GV->setLinkage(llvm::GlobalVariable::WeakAnyLinkage);
   else if (!CompileOpts.NoCommon &&
            (!D->hasExternalStorage() && !D->getInit()))
@@ -891,8 +891,9 @@ static void ReplaceUsesOfNonProtoTypeWithRealFunction(llvm::GlobalValue *Old,
   for (llvm::Value::use_iterator UI = OldFn->use_begin(), E = OldFn->use_end();
        UI != E; ) {
     // TODO: Do invokes ever occur in C code?  If so, we should handle them too.
+    unsigned OpNo = UI.getOperandNo();
     llvm::CallInst *CI = dyn_cast<llvm::CallInst>(*UI++);
-    if (!CI) continue;
+    if (!CI || OpNo != 0) continue;
     
     // If the return types don't match exactly, and if the call isn't dead, then
     // we can't transform this call.
diff --git a/lib/CodeGen/TargetABIInfo.cpp b/lib/CodeGen/TargetABIInfo.cpp
new file mode 100644
index 0000000..573ffed
--- /dev/null
+++ b/lib/CodeGen/TargetABIInfo.cpp
@@ -0,0 +1,1379 @@
+//===---- TargetABIInfo.cpp - Encapsulate target ABI details ----*- C++ -*-===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// These classes wrap the information about a call or function
+// definition used to handle ABI compliancy.
+//
+//===----------------------------------------------------------------------===//
+
+#include "ABIInfo.h"
+#include "CodeGenFunction.h"
+#include "clang/AST/RecordLayout.h"
+#include "llvm/Type.h"
+
+using namespace clang;
+using namespace CodeGen;
+
+ABIInfo::~ABIInfo() {}
+
+void ABIArgInfo::dump() const {
+  fprintf(stderr, "(ABIArgInfo Kind=");
+  switch (TheKind) {
+  case Direct:
+    fprintf(stderr, "Direct");
+    break;
+  case Ignore:
+    fprintf(stderr, "Ignore");
+    break;
+  case Coerce:
+    fprintf(stderr, "Coerce Type=");
+    getCoerceToType()->print(llvm::errs());
+    break;
+  case Indirect:
+    fprintf(stderr, "Indirect Align=%d", getIndirectAlign());
+    break;
+  case Expand:
+    fprintf(stderr, "Expand");
+    break;
+  }
+  fprintf(stderr, ")\n");
+}
+
+static bool isEmptyRecord(ASTContext &Context, QualType T);
+
+/// isEmptyField - Return true iff a the field is "empty", that is it
+/// is an unnamed bit-field or an (array of) empty record(s).
+static bool isEmptyField(ASTContext &Context, const FieldDecl *FD) {
+  if (FD->isUnnamedBitfield())
+    return true;
+
+  QualType FT = FD->getType();
+  // Constant arrays of empty records count as empty, strip them off.
+  while (const ConstantArrayType *AT = Context.getAsConstantArrayType(FT))
+    FT = AT->getElementType();
+
+  return isEmptyRecord(Context, FT);
+}
+
+/// isEmptyRecord - Return true iff a structure contains only empty
+/// fields. Note that a structure with a flexible array member is not
+/// considered empty.
+static bool isEmptyRecord(ASTContext &Context, QualType T) {
+  const RecordType *RT = T->getAsRecordType();
+  if (!RT)
+    return 0;
+  const RecordDecl *RD = RT->getDecl();
+  if (RD->hasFlexibleArrayMember())
+    return false;
+  for (RecordDecl::field_iterator i = RD->field_begin(Context),
+         e = RD->field_end(Context); i != e; ++i)
+    if (!isEmptyField(Context, *i))
+      return false;
+  return true;
+}
+
+/// isSingleElementStruct - Determine if a structure is a "single
+/// element struct", i.e. it has exactly one non-empty field or
+/// exactly one field which is itself a single element
+/// struct. Structures with flexible array members are never
+/// considered single element structs.
+///
+/// \return The field declaration for the single non-empty field, if
+/// it exists.
+static const Type *isSingleElementStruct(QualType T, ASTContext &Context) {
+  const RecordType *RT = T->getAsStructureType();
+  if (!RT)
+    return 0;
+
+  const RecordDecl *RD = RT->getDecl();
+  if (RD->hasFlexibleArrayMember())
+    return 0;
+
+  const Type *Found = 0;
+  for (RecordDecl::field_iterator i = RD->field_begin(Context),
+         e = RD->field_end(Context); i != e; ++i) {
+    const FieldDecl *FD = *i;
+    QualType FT = FD->getType();
+
+    // Ignore empty fields.
+    if (isEmptyField(Context, FD))
+      continue;
+
+    // If we already found an element then this isn't a single-element
+    // struct.
+    if (Found)
+      return 0;
+
+    // Treat single element arrays as the element.
+    while (const ConstantArrayType *AT = Context.getAsConstantArrayType(FT)) {
+      if (AT->getSize().getZExtValue() != 1)
+        break;
+      FT = AT->getElementType();
+    }
+
+    if (!CodeGenFunction::hasAggregateLLVMType(FT)) {
+      Found = FT.getTypePtr();
+    } else {
+      Found = isSingleElementStruct(FT, Context);
+      if (!Found)
+        return 0;
+    }
+  }
+
+  return Found;
+}
+
+static bool is32Or64BitBasicType(QualType Ty, ASTContext &Context) {
+  if (!Ty->getAsBuiltinType() && !Ty->isPointerType())
+    return false;
+
+  uint64_t Size = Context.getTypeSize(Ty);
+  return Size == 32 || Size == 64;
+}
+
+static bool areAllFields32Or64BitBasicType(const RecordDecl *RD,
+                                           ASTContext &Context) {
+  for (RecordDecl::field_iterator i = RD->field_begin(Context),
+         e = RD->field_end(Context); i != e; ++i) {
+    const FieldDecl *FD = *i;
+
+    if (!is32Or64BitBasicType(FD->getType(), Context))
+      return false;
+
+    // FIXME: Reject bit-fields wholesale; there are two problems, we don't know
+    // how to expand them yet, and the predicate for telling if a bitfield still
+    // counts as "basic" is more complicated than what we were doing previously.
+    if (FD->isBitField())
+      return false;
+  }
+
+  return true;
+}
+
+namespace {
+/// DefaultABIInfo - The default implementation for ABI specific
+/// details. This implementation provides information which results in
+/// self-consistent and sensible LLVM IR generation, but does not
+/// conform to any particular ABI.
+class DefaultABIInfo : public ABIInfo {
+  ABIArgInfo classifyReturnType(QualType RetTy,
+                                ASTContext &Context) const;
+
+  ABIArgInfo classifyArgumentType(QualType RetTy,
+                                  ASTContext &Context) const;
+
+  virtual void computeInfo(CGFunctionInfo &FI, ASTContext &Context) const {
+    FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), Context);
+    for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
+         it != ie; ++it)
+      it->info = classifyArgumentType(it->type, Context);
+  }
+
+  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                                 CodeGenFunction &CGF) const;
+};
+
+/// X86_32ABIInfo - The X86-32 ABI information.
+class X86_32ABIInfo : public ABIInfo {
+  ASTContext &Context;
+  bool IsDarwin;
+
+  static bool isRegisterSize(unsigned Size) {
+    return (Size == 8 || Size == 16 || Size == 32 || Size == 64);
+  }
+
+  static bool shouldReturnTypeInRegister(QualType Ty, ASTContext &Context);
+
+public:
+  ABIArgInfo classifyReturnType(QualType RetTy,
+                                ASTContext &Context) const;
+
+  ABIArgInfo classifyArgumentType(QualType RetTy,
+                                  ASTContext &Context) const;
+
+  virtual void computeInfo(CGFunctionInfo &FI, ASTContext &Context) const {
+    FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), Context);
+    for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
+         it != ie; ++it)
+      it->info = classifyArgumentType(it->type, Context);
+  }
+
+  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                                 CodeGenFunction &CGF) const;
+
+  X86_32ABIInfo(ASTContext &Context, bool d)
+    : ABIInfo(), Context(Context), IsDarwin(d) {}
+};
+}
+
+
+/// shouldReturnTypeInRegister - Determine if the given type should be
+/// passed in a register (for the Darwin ABI).
+bool X86_32ABIInfo::shouldReturnTypeInRegister(QualType Ty,
+                                               ASTContext &Context) {
+  uint64_t Size = Context.getTypeSize(Ty);
+
+  // Type must be register sized.
+  if (!isRegisterSize(Size))
+    return false;
+
+  if (Ty->isVectorType()) {
+    // 64- and 128- bit vectors inside structures are not returned in
+    // registers.
+    if (Size == 64 || Size == 128)
+      return false;
+
+    return true;
+  }
+
+  // If this is a builtin, pointer, or complex type, it is ok.
+  if (Ty->getAsBuiltinType() || Ty->isPointerType() || Ty->isAnyComplexType())
+    return true;
+
+  // Arrays are treated like records.
+  if (const ConstantArrayType *AT = Context.getAsConstantArrayType(Ty))
+    return shouldReturnTypeInRegister(AT->getElementType(), Context);
+
+  // Otherwise, it must be a record type.
+  const RecordType *RT = Ty->getAsRecordType();
+  if (!RT) return false;
+
+  // Structure types are passed in register if all fields would be
+  // passed in a register.
+  for (RecordDecl::field_iterator i = RT->getDecl()->field_begin(Context),
+         e = RT->getDecl()->field_end(Context); i != e; ++i) {
+    const FieldDecl *FD = *i;
+
+    // Empty fields are ignored.
+    if (isEmptyField(Context, FD))
+      continue;
+
+    // Check fields recursively.
+    if (!shouldReturnTypeInRegister(FD->getType(), Context))
+      return false;
+  }
+
+  return true;
+}
+
+ABIArgInfo X86_32ABIInfo::classifyReturnType(QualType RetTy,
+                                            ASTContext &Context) const {
+  if (RetTy->isVoidType()) {
+    return ABIArgInfo::getIgnore();
+  } else if (const VectorType *VT = RetTy->getAsVectorType()) {
+    // On Darwin, some vectors are returned in registers.
+    if (IsDarwin) {
+      uint64_t Size = Context.getTypeSize(RetTy);
+
+      // 128-bit vectors are a special case; they are returned in
+      // registers and we need to make sure to pick a type the LLVM
+      // backend will like.
+      if (Size == 128)
+        return ABIArgInfo::getCoerce(llvm::VectorType::get(llvm::Type::Int64Ty,
+                                                           2));
+
+      // Always return in register if it fits in a general purpose
+      // register, or if it is 64 bits and has a single element.
+      if ((Size == 8 || Size == 16 || Size == 32) ||
+          (Size == 64 && VT->getNumElements() == 1))
+        return ABIArgInfo::getCoerce(llvm::IntegerType::get(Size));
+
+      return ABIArgInfo::getIndirect(0);
+    }
+
+    return ABIArgInfo::getDirect();
+  } else if (CodeGenFunction::hasAggregateLLVMType(RetTy)) {
+    // Structures with flexible arrays are always indirect.
+    if (const RecordType *RT = RetTy->getAsStructureType())
+      if (RT->getDecl()->hasFlexibleArrayMember())
+        return ABIArgInfo::getIndirect(0);
+
+    // Outside of Darwin, structs and unions are always indirect.
+    if (!IsDarwin && !RetTy->isAnyComplexType())
+      return ABIArgInfo::getIndirect(0);
+
+    // Classify "single element" structs as their element type.
+    if (const Type *SeltTy = isSingleElementStruct(RetTy, Context)) {
+      if (const BuiltinType *BT = SeltTy->getAsBuiltinType()) {
+        if (BT->isIntegerType()) {
+          // We need to use the size of the structure, padding
+          // bit-fields can adjust that to be larger than the single
+          // element type.
+          uint64_t Size = Context.getTypeSize(RetTy);
+          return ABIArgInfo::getCoerce(llvm::IntegerType::get((unsigned) Size));
+        } else if (BT->getKind() == BuiltinType::Float) {
+          assert(Context.getTypeSize(RetTy) == Context.getTypeSize(SeltTy) &&
+                 "Unexpect single element structure size!");
+          return ABIArgInfo::getCoerce(llvm::Type::FloatTy);
+        } else if (BT->getKind() == BuiltinType::Double) {
+          assert(Context.getTypeSize(RetTy) == Context.getTypeSize(SeltTy) &&
+                 "Unexpect single element structure size!");
+          return ABIArgInfo::getCoerce(llvm::Type::DoubleTy);
+        }
+      } else if (SeltTy->isPointerType()) {
+        // FIXME: It would be really nice if this could come out as the proper
+        // pointer type.
+        llvm::Type *PtrTy =
+          llvm::PointerType::getUnqual(llvm::Type::Int8Ty);
+        return ABIArgInfo::getCoerce(PtrTy);
+      } else if (SeltTy->isVectorType()) {
+        // 64- and 128-bit vectors are never returned in a
+        // register when inside a structure.
+        uint64_t Size = Context.getTypeSize(RetTy);
+        if (Size == 64 || Size == 128)
+          return ABIArgInfo::getIndirect(0);
+
+        return classifyReturnType(QualType(SeltTy, 0), Context);
+      }
+    }
+
+    // Small structures which are register sized are generally returned
+    // in a register.
+    if (X86_32ABIInfo::shouldReturnTypeInRegister(RetTy, Context)) {
+      uint64_t Size = Context.getTypeSize(RetTy);
+      return ABIArgInfo::getCoerce(llvm::IntegerType::get(Size));
+    }
+
+    return ABIArgInfo::getIndirect(0);
+  } else {
+    return ABIArgInfo::getDirect();
+  }
+}
+
+ABIArgInfo X86_32ABIInfo::classifyArgumentType(QualType Ty,
+                                               ASTContext &Context) const {
+  // FIXME: Set alignment on indirect arguments.
+  if (CodeGenFunction::hasAggregateLLVMType(Ty)) {
+    // Structures with flexible arrays are always indirect.
+    if (const RecordType *RT = Ty->getAsStructureType())
+      if (RT->getDecl()->hasFlexibleArrayMember())
+        return ABIArgInfo::getIndirect(0);
+
+    // Ignore empty structs.
+    uint64_t Size = Context.getTypeSize(Ty);
+    if (Ty->isStructureType() && Size == 0)
+      return ABIArgInfo::getIgnore();
+
+    // Expand structs with size <= 128-bits which consist only of
+    // basic types (int, long long, float, double, xxx*). This is
+    // non-recursive and does not ignore empty fields.
+    if (const RecordType *RT = Ty->getAsStructureType()) {
+      if (Context.getTypeSize(Ty) <= 4*32 &&
+          areAllFields32Or64BitBasicType(RT->getDecl(), Context))
+        return ABIArgInfo::getExpand();
+    }
+
+    return ABIArgInfo::getIndirect(0);
+  } else {
+    return ABIArgInfo::getDirect();
+  }
+}
+
+llvm::Value *X86_32ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                                      CodeGenFunction &CGF) const {
+  const llvm::Type *BP = llvm::PointerType::getUnqual(llvm::Type::Int8Ty);
+  const llvm::Type *BPP = llvm::PointerType::getUnqual(BP);
+
+  CGBuilderTy &Builder = CGF.Builder;
+  llvm::Value *VAListAddrAsBPP = Builder.CreateBitCast(VAListAddr, BPP,
+                                                       "ap");
+  llvm::Value *Addr = Builder.CreateLoad(VAListAddrAsBPP, "ap.cur");
+  llvm::Type *PTy =
+    llvm::PointerType::getUnqual(CGF.ConvertType(Ty));
+  llvm::Value *AddrTyped = Builder.CreateBitCast(Addr, PTy);
+
+  uint64_t Offset =
+    llvm::RoundUpToAlignment(CGF.getContext().getTypeSize(Ty) / 8, 4);
+  llvm::Value *NextAddr =
+    Builder.CreateGEP(Addr,
+                      llvm::ConstantInt::get(llvm::Type::Int32Ty, Offset),
+                      "ap.next");
+  Builder.CreateStore(NextAddr, VAListAddrAsBPP);
+
+  return AddrTyped;
+}
+
+namespace {
+/// X86_64ABIInfo - The X86_64 ABI information.
+class X86_64ABIInfo : public ABIInfo {
+  enum Class {
+    Integer = 0,
+    SSE,
+    SSEUp,
+    X87,
+    X87Up,
+    ComplexX87,
+    NoClass,
+    Memory
+  };
+
+  /// merge - Implement the X86_64 ABI merging algorithm.
+  ///
+  /// Merge an accumulating classification \arg Accum with a field
+  /// classification \arg Field.
+  ///
+  /// \param Accum - The accumulating classification. This should
+  /// always be either NoClass or the result of a previous merge
+  /// call. In addition, this should never be Memory (the caller
+  /// should just return Memory for the aggregate).
+  Class merge(Class Accum, Class Field) const;
+
+  /// classify - Determine the x86_64 register classes in which the
+  /// given type T should be passed.
+  ///
+  /// \param Lo - The classification for the parts of the type
+  /// residing in the low word of the containing object.
+  ///
+  /// \param Hi - The classification for the parts of the type
+  /// residing in the high word of the containing object.
+  ///
+  /// \param OffsetBase - The bit offset of this type in the
+  /// containing object.  Some parameters are classified different
+  /// depending on whether they straddle an eightbyte boundary.
+  ///
+  /// If a word is unused its result will be NoClass; if a type should
+  /// be passed in Memory then at least the classification of \arg Lo
+  /// will be Memory.
+  ///
+  /// The \arg Lo class will be NoClass iff the argument is ignored.
+  ///
+  /// If the \arg Lo class is ComplexX87, then the \arg Hi class will
+  /// also be ComplexX87.
+  void classify(QualType T, ASTContext &Context, uint64_t OffsetBase,
+                Class &Lo, Class &Hi) const;
+
+  /// getCoerceResult - Given a source type \arg Ty and an LLVM type
+  /// to coerce to, chose the best way to pass Ty in the same place
+  /// that \arg CoerceTo would be passed, but while keeping the
+  /// emitted code as simple as possible.
+  ///
+  /// FIXME: Note, this should be cleaned up to just take an enumeration of all
+  /// the ways we might want to pass things, instead of constructing an LLVM
+  /// type. This makes this code more explicit, and it makes it clearer that we
+  /// are also doing this for correctness in the case of passing scalar types.
+  ABIArgInfo getCoerceResult(QualType Ty,
+                             const llvm::Type *CoerceTo,
+                             ASTContext &Context) const;
+
+  /// getIndirectResult - Give a source type \arg Ty, return a suitable result
+  /// such that the argument will be passed in memory.
+  ABIArgInfo getIndirectResult(QualType Ty,
+                               ASTContext &Context) const;
+
+  ABIArgInfo classifyReturnType(QualType RetTy,
+                                ASTContext &Context) const;
+
+  ABIArgInfo classifyArgumentType(QualType Ty,
+                                  ASTContext &Context,
+                                  unsigned &neededInt,
+                                  unsigned &neededSSE) const;
+
+public:
+  virtual void computeInfo(CGFunctionInfo &FI, ASTContext &Context) const;
+
+  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                                 CodeGenFunction &CGF) const;
+};
+}
+
+X86_64ABIInfo::Class X86_64ABIInfo::merge(Class Accum,
+                                          Class Field) const {
+  // AMD64-ABI 3.2.3p2: Rule 4. Each field of an object is
+  // classified recursively so that always two fields are
+  // considered. The resulting class is calculated according to
+  // the classes of the fields in the eightbyte:
+  //
+  // (a) If both classes are equal, this is the resulting class.
+  //
+  // (b) If one of the classes is NO_CLASS, the resulting class is
+  // the other class.
+  //
+  // (c) If one of the classes is MEMORY, the result is the MEMORY
+  // class.
+  //
+  // (d) If one of the classes is INTEGER, the result is the
+  // INTEGER.
+  //
+  // (e) If one of the classes is X87, X87UP, COMPLEX_X87 class,
+  // MEMORY is used as class.
+  //
+  // (f) Otherwise class SSE is used.
+
+  // Accum should never be memory (we should have returned) or
+  // ComplexX87 (because this cannot be passed in a structure).
+  assert((Accum != Memory && Accum != ComplexX87) &&
+         "Invalid accumulated classification during merge.");
+  if (Accum == Field || Field == NoClass)
+    return Accum;
+  else if (Field == Memory)
+    return Memory;
+  else if (Accum == NoClass)
+    return Field;
+  else if (Accum == Integer || Field == Integer)
+    return Integer;
+  else if (Field == X87 || Field == X87Up || Field == ComplexX87 ||
+           Accum == X87 || Accum == X87Up)
+    return Memory;
+  else
+    return SSE;
+}
+
+void X86_64ABIInfo::classify(QualType Ty,
+                             ASTContext &Context,
+                             uint64_t OffsetBase,
+                             Class &Lo, Class &Hi) const {
+  // FIXME: This code can be simplified by introducing a simple value class for
+  // Class pairs with appropriate constructor methods for the various
+  // situations.
+
+  // FIXME: Some of the split computations are wrong; unaligned vectors
+  // shouldn't be passed in registers for example, so there is no chance they
+  // can straddle an eightbyte. Verify & simplify.
+
+  Lo = Hi = NoClass;
+
+  Class &Current = OffsetBase < 64 ? Lo : Hi;
+  Current = Memory;
+
+  if (const BuiltinType *BT = Ty->getAsBuiltinType()) {
+    BuiltinType::Kind k = BT->getKind();
+
+    if (k == BuiltinType::Void) {
+      Current = NoClass;
+    } else if (k == BuiltinType::Int128 || k == BuiltinType::UInt128) {
+      Lo = Integer;
+      Hi = Integer;
+    } else if (k >= BuiltinType::Bool && k <= BuiltinType::LongLong) {
+      Current = Integer;
+    } else if (k == BuiltinType::Float || k == BuiltinType::Double) {
+      Current = SSE;
+    } else if (k == BuiltinType::LongDouble) {
+      Lo = X87;
+      Hi = X87Up;
+    }
+    // FIXME: _Decimal32 and _Decimal64 are SSE.
+    // FIXME: _float128 and _Decimal128 are (SSE, SSEUp).
+  } else if (const EnumType *ET = Ty->getAsEnumType()) {
+    // Classify the underlying integer type.
+    classify(ET->getDecl()->getIntegerType(), Context, OffsetBase, Lo, Hi);
+  } else if (Ty->hasPointerRepresentation()) {
+    Current = Integer;
+  } else if (const VectorType *VT = Ty->getAsVectorType()) {
+    uint64_t Size = Context.getTypeSize(VT);
+    if (Size == 32) {
+      // gcc passes all <4 x char>, <2 x short>, <1 x int>, <1 x
+      // float> as integer.
+      Current = Integer;
+
+      // If this type crosses an eightbyte boundary, it should be
+      // split.
+      uint64_t EB_Real = (OffsetBase) / 64;
+      uint64_t EB_Imag = (OffsetBase + Size - 1) / 64;
+      if (EB_Real != EB_Imag)
+        Hi = Lo;
+    } else if (Size == 64) {
+      // gcc passes <1 x double> in memory. :(
+      if (VT->getElementType()->isSpecificBuiltinType(BuiltinType::Double))
+        return;
+
+      // gcc passes <1 x long long> as INTEGER.
+      if (VT->getElementType()->isSpecificBuiltinType(BuiltinType::LongLong))
+        Current = Integer;
+      else
+        Current = SSE;
+
+      // If this type crosses an eightbyte boundary, it should be
+      // split.
+      if (OffsetBase && OffsetBase != 64)
+        Hi = Lo;
+    } else if (Size == 128) {
+      Lo = SSE;
+      Hi = SSEUp;
+    }
+  } else if (const ComplexType *CT = Ty->getAsComplexType()) {
+    QualType ET = Context.getCanonicalType(CT->getElementType());
+
+    uint64_t Size = Context.getTypeSize(Ty);
+    if (ET->isIntegralType()) {
+      if (Size <= 64)
+        Current = Integer;
+      else if (Size <= 128)
+        Lo = Hi = Integer;
+    } else if (ET == Context.FloatTy)
+      Current = SSE;
+    else if (ET == Context.DoubleTy)
+      Lo = Hi = SSE;
+    else if (ET == Context.LongDoubleTy)
+      Current = ComplexX87;
+
+    // If this complex type crosses an eightbyte boundary then it
+    // should be split.
+    uint64_t EB_Real = (OffsetBase) / 64;
+    uint64_t EB_Imag = (OffsetBase + Context.getTypeSize(ET)) / 64;
+    if (Hi == NoClass && EB_Real != EB_Imag)
+      Hi = Lo;
+  } else if (const ConstantArrayType *AT = Context.getAsConstantArrayType(Ty)) {
+    // Arrays are treated like structures.
+
+    uint64_t Size = Context.getTypeSize(Ty);
+
+    // AMD64-ABI 3.2.3p2: Rule 1. If the size of an object is larger
+    // than two eightbytes, ..., it has class MEMORY.
+    if (Size > 128)
+      return;
+
+    // AMD64-ABI 3.2.3p2: Rule 1. If ..., or it contains unaligned
+    // fields, it has class MEMORY.
+    //
+    // Only need to check alignment of array base.
+    if (OffsetBase % Context.getTypeAlign(AT->getElementType()))
+      return;
+
+    // Otherwise implement simplified merge. We could be smarter about
+    // this, but it isn't worth it and would be harder to verify.
+    Current = NoClass;
+    uint64_t EltSize = Context.getTypeSize(AT->getElementType());
+    uint64_t ArraySize = AT->getSize().getZExtValue();
+    for (uint64_t i=0, Offset=OffsetBase; i<ArraySize; ++i, Offset += EltSize) {
+      Class FieldLo, FieldHi;
+      classify(AT->getElementType(), Context, Offset, FieldLo, FieldHi);
+      Lo = merge(Lo, FieldLo);
+      Hi = merge(Hi, FieldHi);
+      if (Lo == Memory || Hi == Memory)
+        break;
+    }
+
+    // Do post merger cleanup (see below). Only case we worry about is Memory.
+    if (Hi == Memory)
+      Lo = Memory;
+    assert((Hi != SSEUp || Lo == SSE) && "Invalid SSEUp array classification.");
+  } else if (const RecordType *RT = Ty->getAsRecordType()) {
+    uint64_t Size = Context.getTypeSize(Ty);
+
+    // AMD64-ABI 3.2.3p2: Rule 1. If the size of an object is larger
+    // than two eightbytes, ..., it has class MEMORY.
+    if (Size > 128)
+      return;
+
+    const RecordDecl *RD = RT->getDecl();
+
+    // Assume variable sized types are passed in memory.
+    if (RD->hasFlexibleArrayMember())
+      return;
+
+    const ASTRecordLayout &Layout = Context.getASTRecordLayout(RD);
+
+    // Reset Lo class, this will be recomputed.
+    Current = NoClass;
+    unsigned idx = 0;
+    for (RecordDecl::field_iterator i = RD->field_begin(Context),
+           e = RD->field_end(Context); i != e; ++i, ++idx) {
+      uint64_t Offset = OffsetBase + Layout.getFieldOffset(idx);
+      bool BitField = i->isBitField();
+
+      // AMD64-ABI 3.2.3p2: Rule 1. If ..., or it contains unaligned
+      // fields, it has class MEMORY.
+      //
+      // Note, skip this test for bit-fields, see below.
+      if (!BitField && Offset % Context.getTypeAlign(i->getType())) {
+        Lo = Memory;
+        return;
+      }
+
+      // Classify this field.
+      //
+      // AMD64-ABI 3.2.3p2: Rule 3. If the size of the aggregate
+      // exceeds a single eightbyte, each is classified
+      // separately. Each eightbyte gets initialized to class
+      // NO_CLASS.
+      Class FieldLo, FieldHi;
+
+      // Bit-fields require special handling, they do not force the
+      // structure to be passed in memory even if unaligned, and
+      // therefore they can straddle an eightbyte.
+      if (BitField) {
+        // Ignore padding bit-fields.
+        if (i->isUnnamedBitfield())
+          continue;
+
+        uint64_t Offset = OffsetBase + Layout.getFieldOffset(idx);
+        uint64_t Size = i->getBitWidth()->EvaluateAsInt(Context).getZExtValue();
+
+        uint64_t EB_Lo = Offset / 64;
+        uint64_t EB_Hi = (Offset + Size - 1) / 64;
+        FieldLo = FieldHi = NoClass;
+        if (EB_Lo) {
+          assert(EB_Hi == EB_Lo && "Invalid classification, type > 16 bytes.");
+          FieldLo = NoClass;
+          FieldHi = Integer;
+        } else {
+          FieldLo = Integer;
+          FieldHi = EB_Hi ? Integer : NoClass;
+        }
+      } else
+        classify(i->getType(), Context, Offset, FieldLo, FieldHi);
+      Lo = merge(Lo, FieldLo);
+      Hi = merge(Hi, FieldHi);
+      if (Lo == Memory || Hi == Memory)
+        break;
+    }
+
+    // AMD64-ABI 3.2.3p2: Rule 5. Then a post merger cleanup is done:
+    //
+    // (a) If one of the classes is MEMORY, the whole argument is
+    // passed in memory.
+    //
+    // (b) If SSEUP is not preceeded by SSE, it is converted to SSE.
+
+    // The first of these conditions is guaranteed by how we implement
+    // the merge (just bail).
+    //
+    // The second condition occurs in the case of unions; for example
+    // union { _Complex double; unsigned; }.
+    if (Hi == Memory)
+      Lo = Memory;
+    if (Hi == SSEUp && Lo != SSE)
+      Hi = SSE;
+  }
+}
+
+ABIArgInfo X86_64ABIInfo::getCoerceResult(QualType Ty,
+                                          const llvm::Type *CoerceTo,
+                                          ASTContext &Context) const {
+  if (CoerceTo == llvm::Type::Int64Ty) {
+    // Integer and pointer types will end up in a general purpose
+    // register.
+    if (Ty->isIntegralType() || Ty->isPointerType())
+      return ABIArgInfo::getDirect();
+
+  } else if (CoerceTo == llvm::Type::DoubleTy) {
+    // FIXME: It would probably be better to make CGFunctionInfo only map using
+    // canonical types than to canonize here.
+    QualType CTy = Context.getCanonicalType(Ty);
+
+    // Float and double end up in a single SSE reg.
+    if (CTy == Context.FloatTy || CTy == Context.DoubleTy)
+      return ABIArgInfo::getDirect();
+
+  }
+
+  return ABIArgInfo::getCoerce(CoerceTo);
+}
+
+ABIArgInfo X86_64ABIInfo::getIndirectResult(QualType Ty,
+                                            ASTContext &Context) const {
+  // If this is a scalar LLVM value then assume LLVM will pass it in the right
+  // place naturally.
+  if (!CodeGenFunction::hasAggregateLLVMType(Ty))
+    return ABIArgInfo::getDirect();
+
+  // FIXME: Set alignment correctly.
+  return ABIArgInfo::getIndirect(0);
+}
+
+ABIArgInfo X86_64ABIInfo::classifyReturnType(QualType RetTy,
+                                            ASTContext &Context) const {
+  // AMD64-ABI 3.2.3p4: Rule 1. Classify the return type with the
+  // classification algorithm.
+  X86_64ABIInfo::Class Lo, Hi;
+  classify(RetTy, Context, 0, Lo, Hi);
+
+  // Check some invariants.
+  assert((Hi != Memory || Lo == Memory) && "Invalid memory classification.");
+  assert((Lo != NoClass || Hi == NoClass) && "Invalid null classification.");
+  assert((Hi != SSEUp || Lo == SSE) && "Invalid SSEUp classification.");
+
+  const llvm::Type *ResType = 0;
+  switch (Lo) {
+  case NoClass:
+    return ABIArgInfo::getIgnore();
+
+  case SSEUp:
+  case X87Up:
+    assert(0 && "Invalid classification for lo word.");
+
+    // AMD64-ABI 3.2.3p4: Rule 2. Types of class memory are returned via
+    // hidden argument.
+  case Memory:
+    return getIndirectResult(RetTy, Context);
+
+    // AMD64-ABI 3.2.3p4: Rule 3. If the class is INTEGER, the next
+    // available register of the sequence %rax, %rdx is used.
+  case Integer:
+    ResType = llvm::Type::Int64Ty; break;
+
+    // AMD64-ABI 3.2.3p4: Rule 4. If the class is SSE, the next
+    // available SSE register of the sequence %xmm0, %xmm1 is used.
+  case SSE:
+    ResType = llvm::Type::DoubleTy; break;
+
+    // AMD64-ABI 3.2.3p4: Rule 6. If the class is X87, the value is
+    // returned on the X87 stack in %st0 as 80-bit x87 number.
+  case X87:
+    ResType = llvm::Type::X86_FP80Ty; break;
+
+    // AMD64-ABI 3.2.3p4: Rule 8. If the class is COMPLEX_X87, the real
+    // part of the value is returned in %st0 and the imaginary part in
+    // %st1.
+  case ComplexX87:
+    assert(Hi == ComplexX87 && "Unexpected ComplexX87 classification.");
+    ResType = llvm::StructType::get(llvm::Type::X86_FP80Ty,
+                                    llvm::Type::X86_FP80Ty,
+                                    NULL);
+    break;
+  }
+
+  switch (Hi) {
+    // Memory was handled previously and X87 should
+    // never occur as a hi class.
+  case Memory:
+  case X87:
+    assert(0 && "Invalid classification for hi word.");
+
+  case ComplexX87: // Previously handled.
+  case NoClass: break;
+
+  case Integer:
+    ResType = llvm::StructType::get(ResType, llvm::Type::Int64Ty, NULL);
+    break;
+  case SSE:
+    ResType = llvm::StructType::get(ResType, llvm::Type::DoubleTy, NULL);
+    break;
+
+    // AMD64-ABI 3.2.3p4: Rule 5. If the class is SSEUP, the eightbyte
+    // is passed in the upper half of the last used SSE register.
+    //
+    // SSEUP should always be preceeded by SSE, just widen.
+  case SSEUp:
+    assert(Lo == SSE && "Unexpected SSEUp classification.");
+    ResType = llvm::VectorType::get(llvm::Type::DoubleTy, 2);
+    break;
+
+    // AMD64-ABI 3.2.3p4: Rule 7. If the class is X87UP, the value is
+    // returned together with the previous X87 value in %st0.
+  case X87Up:
+    // If X87Up is preceeded by X87, we don't need to do
+    // anything. However, in some cases with unions it may not be
+    // preceeded by X87. In such situations we follow gcc and pass the
+    // extra bits in an SSE reg.
+    if (Lo != X87)
+      ResType = llvm::StructType::get(ResType, llvm::Type::DoubleTy, NULL);
+    break;
+  }
+
+  return getCoerceResult(RetTy, ResType, Context);
+}
+
+ABIArgInfo X86_64ABIInfo::classifyArgumentType(QualType Ty, ASTContext &Context,
+                                               unsigned &neededInt,
+                                               unsigned &neededSSE) const {
+  X86_64ABIInfo::Class Lo, Hi;
+  classify(Ty, Context, 0, Lo, Hi);
+
+  // Check some invariants.
+  // FIXME: Enforce these by construction.
+  assert((Hi != Memory || Lo == Memory) && "Invalid memory classification.");
+  assert((Lo != NoClass || Hi == NoClass) && "Invalid null classification.");
+  assert((Hi != SSEUp || Lo == SSE) && "Invalid SSEUp classification.");
+
+  neededInt = 0;
+  neededSSE = 0;
+  const llvm::Type *ResType = 0;
+  switch (Lo) {
+  case NoClass:
+    return ABIArgInfo::getIgnore();
+
+    // AMD64-ABI 3.2.3p3: Rule 1. If the class is MEMORY, pass the argument
+    // on the stack.
+  case Memory:
+
+    // AMD64-ABI 3.2.3p3: Rule 5. If the class is X87, X87UP or
+    // COMPLEX_X87, it is passed in memory.
+  case X87:
+  case ComplexX87:
+    return getIndirectResult(Ty, Context);
+
+  case SSEUp:
+  case X87Up:
+    assert(0 && "Invalid classification for lo word.");
+
+    // AMD64-ABI 3.2.3p3: Rule 2. If the class is INTEGER, the next
+    // available register of the sequence %rdi, %rsi, %rdx, %rcx, %r8
+    // and %r9 is used.
+  case Integer:
+    ++neededInt;
+    ResType = llvm::Type::Int64Ty;
+    break;
+
+    // AMD64-ABI 3.2.3p3: Rule 3. If the class is SSE, the next
+    // available SSE register is used, the registers are taken in the
+    // order from %xmm0 to %xmm7.
+  case SSE:
+    ++neededSSE;
+    ResType = llvm::Type::DoubleTy;
+    break;
+  }
+
+  switch (Hi) {
+    // Memory was handled previously, ComplexX87 and X87 should
+    // never occur as hi classes, and X87Up must be preceed by X87,
+    // which is passed in memory.
+  case Memory:
+  case X87:
+  case ComplexX87:
+    assert(0 && "Invalid classification for hi word.");
+    break;
+
+  case NoClass: break;
+  case Integer:
+    ResType = llvm::StructType::get(ResType, llvm::Type::Int64Ty, NULL);
+    ++neededInt;
+    break;
+
+    // X87Up generally doesn't occur here (long double is passed in
+    // memory), except in situations involving unions.
+  case X87Up:
+  case SSE:
+    ResType = llvm::StructType::get(ResType, llvm::Type::DoubleTy, NULL);
+    ++neededSSE;
+    break;
+
+    // AMD64-ABI 3.2.3p3: Rule 4. If the class is SSEUP, the
+    // eightbyte is passed in the upper half of the last used SSE
+    // register.
+  case SSEUp:
+    assert(Lo == SSE && "Unexpected SSEUp classification.");
+    ResType = llvm::VectorType::get(llvm::Type::DoubleTy, 2);
+    break;
+  }
+
+  return getCoerceResult(Ty, ResType, Context);
+}
+
+void X86_64ABIInfo::computeInfo(CGFunctionInfo &FI, ASTContext &Context) const {
+  FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), Context);
+
+  // Keep track of the number of assigned registers.
+  unsigned freeIntRegs = 6, freeSSERegs = 8;
+
+  // If the return value is indirect, then the hidden argument is consuming one
+  // integer register.
+  if (FI.getReturnInfo().isIndirect())
+    --freeIntRegs;
+
+  // AMD64-ABI 3.2.3p3: Once arguments are classified, the registers
+  // get assigned (in left-to-right order) for passing as follows...
+  for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
+       it != ie; ++it) {
+    unsigned neededInt, neededSSE;
+    it->info = classifyArgumentType(it->type, Context, neededInt, neededSSE);
+
+    // AMD64-ABI 3.2.3p3: If there are no registers available for any
+    // eightbyte of an argument, the whole argument is passed on the
+    // stack. If registers have already been assigned for some
+    // eightbytes of such an argument, the assignments get reverted.
+    if (freeIntRegs >= neededInt && freeSSERegs >= neededSSE) {
+      freeIntRegs -= neededInt;
+      freeSSERegs -= neededSSE;
+    } else {
+      it->info = getIndirectResult(it->type, Context);
+    }
+  }
+}
+
+static llvm::Value *EmitVAArgFromMemory(llvm::Value *VAListAddr,
+                                        QualType Ty,
+                                        CodeGenFunction &CGF) {
+  llvm::Value *overflow_arg_area_p =
+    CGF.Builder.CreateStructGEP(VAListAddr, 2, "overflow_arg_area_p");
+  llvm::Value *overflow_arg_area =
+    CGF.Builder.CreateLoad(overflow_arg_area_p, "overflow_arg_area");
+
+  // AMD64-ABI 3.5.7p5: Step 7. Align l->overflow_arg_area upwards to a 16
+  // byte boundary if alignment needed by type exceeds 8 byte boundary.
+  uint64_t Align = CGF.getContext().getTypeAlign(Ty) / 8;
+  if (Align > 8) {
+    // Note that we follow the ABI & gcc here, even though the type
+    // could in theory have an alignment greater than 16. This case
+    // shouldn't ever matter in practice.
+
+    // overflow_arg_area = (overflow_arg_area + 15) & ~15;
+    llvm::Value *Offset = llvm::ConstantInt::get(llvm::Type::Int32Ty, 15);
+    overflow_arg_area = CGF.Builder.CreateGEP(overflow_arg_area, Offset);
+    llvm::Value *AsInt = CGF.Builder.CreatePtrToInt(overflow_arg_area,
+                                                    llvm::Type::Int64Ty);
+    llvm::Value *Mask = llvm::ConstantInt::get(llvm::Type::Int64Ty, ~15LL);
+    overflow_arg_area =
+      CGF.Builder.CreateIntToPtr(CGF.Builder.CreateAnd(AsInt, Mask),
+                                 overflow_arg_area->getType(),
+                                 "overflow_arg_area.align");
+  }
+
+  // AMD64-ABI 3.5.7p5: Step 8. Fetch type from l->overflow_arg_area.
+  const llvm::Type *LTy = CGF.ConvertTypeForMem(Ty);
+  llvm::Value *Res =
+    CGF.Builder.CreateBitCast(overflow_arg_area,
+                              llvm::PointerType::getUnqual(LTy));
+
+  // AMD64-ABI 3.5.7p5: Step 9. Set l->overflow_arg_area to:
+  // l->overflow_arg_area + sizeof(type).
+  // AMD64-ABI 3.5.7p5: Step 10. Align l->overflow_arg_area upwards to
+  // an 8 byte boundary.
+
+  uint64_t SizeInBytes = (CGF.getContext().getTypeSize(Ty) + 7) / 8;
+  llvm::Value *Offset = llvm::ConstantInt::get(llvm::Type::Int32Ty,
+                                               (SizeInBytes + 7)  & ~7);
+  overflow_arg_area = CGF.Builder.CreateGEP(overflow_arg_area, Offset,
+                                            "overflow_arg_area.next");
+  CGF.Builder.CreateStore(overflow_arg_area, overflow_arg_area_p);
+
+  // AMD64-ABI 3.5.7p5: Step 11. Return the fetched type.
+  return Res;
+}
+
+llvm::Value *X86_64ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                                      CodeGenFunction &CGF) const {
+  // Assume that va_list type is correct; should be pointer to LLVM type:
+  // struct {
+  //   i32 gp_offset;
+  //   i32 fp_offset;
+  //   i8* overflow_arg_area;
+  //   i8* reg_save_area;
+  // };
+  unsigned neededInt, neededSSE;
+  ABIArgInfo AI = classifyArgumentType(Ty, CGF.getContext(),
+                                       neededInt, neededSSE);
+
+  // AMD64-ABI 3.5.7p5: Step 1. Determine whether type may be passed
+  // in the registers. If not go to step 7.
+  if (!neededInt && !neededSSE)
+    return EmitVAArgFromMemory(VAListAddr, Ty, CGF);
+
+  // AMD64-ABI 3.5.7p5: Step 2. Compute num_gp to hold the number of
+  // general purpose registers needed to pass type and num_fp to hold
+  // the number of floating point registers needed.
+
+  // AMD64-ABI 3.5.7p5: Step 3. Verify whether arguments fit into
+  // registers. In the case: l->gp_offset > 48 - num_gp * 8 or
+  // l->fp_offset > 304 - num_fp * 16 go to step 7.
+  //
+  // NOTE: 304 is a typo, there are (6 * 8 + 8 * 16) = 176 bytes of
+  // register save space).
+
+  llvm::Value *InRegs = 0;
+  llvm::Value *gp_offset_p = 0, *gp_offset = 0;
+  llvm::Value *fp_offset_p = 0, *fp_offset = 0;
+  if (neededInt) {
+    gp_offset_p = CGF.Builder.CreateStructGEP(VAListAddr, 0, "gp_offset_p");
+    gp_offset = CGF.Builder.CreateLoad(gp_offset_p, "gp_offset");
+    InRegs =
+      CGF.Builder.CreateICmpULE(gp_offset,
+                                llvm::ConstantInt::get(llvm::Type::Int32Ty,
+                                                       48 - neededInt * 8),
+                                "fits_in_gp");
+  }
+
+  if (neededSSE) {
+    fp_offset_p = CGF.Builder.CreateStructGEP(VAListAddr, 1, "fp_offset_p");
+    fp_offset = CGF.Builder.CreateLoad(fp_offset_p, "fp_offset");
+    llvm::Value *FitsInFP =
+      CGF.Builder.CreateICmpULE(fp_offset,
+                                llvm::ConstantInt::get(llvm::Type::Int32Ty,
+                                                       176 - neededSSE * 16),
+                                "fits_in_fp");
+    InRegs = InRegs ? CGF.Builder.CreateAnd(InRegs, FitsInFP) : FitsInFP;
+  }
+
+  llvm::BasicBlock *InRegBlock = CGF.createBasicBlock("vaarg.in_reg");
+  llvm::BasicBlock *InMemBlock = CGF.createBasicBlock("vaarg.in_mem");
+  llvm::BasicBlock *ContBlock = CGF.createBasicBlock("vaarg.end");
+  CGF.Builder.CreateCondBr(InRegs, InRegBlock, InMemBlock);
+
+  // Emit code to load the value if it was passed in registers.
+
+  CGF.EmitBlock(InRegBlock);
+
+  // AMD64-ABI 3.5.7p5: Step 4. Fetch type from l->reg_save_area with
+  // an offset of l->gp_offset and/or l->fp_offset. This may require
+  // copying to a temporary location in case the parameter is passed
+  // in different register classes or requires an alignment greater
+  // than 8 for general purpose registers and 16 for XMM registers.
+  //
+  // FIXME: This really results in shameful code when we end up needing to
+  // collect arguments from different places; often what should result in a
+  // simple assembling of a structure from scattered addresses has many more
+  // loads than necessary. Can we clean this up?
+  const llvm::Type *LTy = CGF.ConvertTypeForMem(Ty);
+  llvm::Value *RegAddr =
+    CGF.Builder.CreateLoad(CGF.Builder.CreateStructGEP(VAListAddr, 3),
+                           "reg_save_area");
+  if (neededInt && neededSSE) {
+    // FIXME: Cleanup.
+    assert(AI.isCoerce() && "Unexpected ABI info for mixed regs");
+    const llvm::StructType *ST = cast<llvm::StructType>(AI.getCoerceToType());
+    llvm::Value *Tmp = CGF.CreateTempAlloca(ST);
+    assert(ST->getNumElements() == 2 && "Unexpected ABI info for mixed regs");
+    const llvm::Type *TyLo = ST->getElementType(0);
+    const llvm::Type *TyHi = ST->getElementType(1);
+    assert((TyLo->isFloatingPoint() ^ TyHi->isFloatingPoint()) &&
+           "Unexpected ABI info for mixed regs");
+    const llvm::Type *PTyLo = llvm::PointerType::getUnqual(TyLo);
+    const llvm::Type *PTyHi = llvm::PointerType::getUnqual(TyHi);
+    llvm::Value *GPAddr = CGF.Builder.CreateGEP(RegAddr, gp_offset);
+    llvm::Value *FPAddr = CGF.Builder.CreateGEP(RegAddr, fp_offset);
+    llvm::Value *RegLoAddr = TyLo->isFloatingPoint() ? FPAddr : GPAddr;
+    llvm::Value *RegHiAddr = TyLo->isFloatingPoint() ? GPAddr : FPAddr;
+    llvm::Value *V =
+      CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegLoAddr, PTyLo));
+    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 0));
+    V = CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegHiAddr, PTyHi));
+    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 1));
+
+    RegAddr = CGF.Builder.CreateBitCast(Tmp, llvm::PointerType::getUnqual(LTy));
+  } else if (neededInt) {
+    RegAddr = CGF.Builder.CreateGEP(RegAddr, gp_offset);
+    RegAddr = CGF.Builder.CreateBitCast(RegAddr,
+                                        llvm::PointerType::getUnqual(LTy));
+  } else {
+    if (neededSSE == 1) {
+      RegAddr = CGF.Builder.CreateGEP(RegAddr, fp_offset);
+      RegAddr = CGF.Builder.CreateBitCast(RegAddr,
+                                          llvm::PointerType::getUnqual(LTy));
+    } else {
+      assert(neededSSE == 2 && "Invalid number of needed registers!");
+      // SSE registers are spaced 16 bytes apart in the register save
+      // area, we need to collect the two eightbytes together.
+      llvm::Value *RegAddrLo = CGF.Builder.CreateGEP(RegAddr, fp_offset);
+      llvm::Value *RegAddrHi =
+        CGF.Builder.CreateGEP(RegAddrLo,
+                              llvm::ConstantInt::get(llvm::Type::Int32Ty, 16));
+      const llvm::Type *DblPtrTy =
+        llvm::PointerType::getUnqual(llvm::Type::DoubleTy);
+      const llvm::StructType *ST = llvm::StructType::get(llvm::Type::DoubleTy,
+                                                         llvm::Type::DoubleTy,
+                                                         NULL);
+      llvm::Value *V, *Tmp = CGF.CreateTempAlloca(ST);
+      V = CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegAddrLo,
+                                                           DblPtrTy));
+      CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 0));
+      V = CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegAddrHi,
+                                                           DblPtrTy));
+      CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 1));
+      RegAddr = CGF.Builder.CreateBitCast(Tmp,
+                                          llvm::PointerType::getUnqual(LTy));
+    }
+  }
+
+  // AMD64-ABI 3.5.7p5: Step 5. Set:
+  // l->gp_offset = l->gp_offset + num_gp * 8
+  // l->fp_offset = l->fp_offset + num_fp * 16.
+  if (neededInt) {
+    llvm::Value *Offset = llvm::ConstantInt::get(llvm::Type::Int32Ty,
+                                                 neededInt * 8);
+    CGF.Builder.CreateStore(CGF.Builder.CreateAdd(gp_offset, Offset),
+                            gp_offset_p);
+  }
+  if (neededSSE) {
+    llvm::Value *Offset = llvm::ConstantInt::get(llvm::Type::Int32Ty,
+                                                 neededSSE * 16);
+    CGF.Builder.CreateStore(CGF.Builder.CreateAdd(fp_offset, Offset),
+                            fp_offset_p);
+  }
+  CGF.EmitBranch(ContBlock);
+
+  // Emit code to load the value if it was passed in memory.
+
+  CGF.EmitBlock(InMemBlock);
+  llvm::Value *MemAddr = EmitVAArgFromMemory(VAListAddr, Ty, CGF);
+
+  // Return the appropriate result.
+
+  CGF.EmitBlock(ContBlock);
+  llvm::PHINode *ResAddr = CGF.Builder.CreatePHI(RegAddr->getType(),
+                                                 "vaarg.addr");
+  ResAddr->reserveOperandSpace(2);
+  ResAddr->addIncoming(RegAddr, InRegBlock);
+  ResAddr->addIncoming(MemAddr, InMemBlock);
+
+  return ResAddr;
+}
+
+// ABI Info for PIC16
+class PIC16ABIInfo : public ABIInfo {
+  ABIArgInfo classifyReturnType(QualType RetTy,
+                                ASTContext &Context) const;
+
+  ABIArgInfo classifyArgumentType(QualType RetTy,
+                                  ASTContext &Context) const;
+
+  virtual void computeInfo(CGFunctionInfo &FI, ASTContext &Context) const {
+    FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), Context);
+    for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
+         it != ie; ++it)
+      it->info = classifyArgumentType(it->type, Context);
+  }
+
+  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                                 CodeGenFunction &CGF) const;
+
+};
+
+ABIArgInfo PIC16ABIInfo::classifyReturnType(QualType RetTy,
+                                              ASTContext &Context) const {
+  if (RetTy->isVoidType()) {
+    return ABIArgInfo::getIgnore();
+  } else {
+    return ABIArgInfo::getDirect();
+  }
+}
+
+ABIArgInfo PIC16ABIInfo::classifyArgumentType(QualType Ty,
+                                                ASTContext &Context) const {
+  return ABIArgInfo::getDirect();
+}
+
+llvm::Value *PIC16ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                                       CodeGenFunction &CGF) const {
+  return 0;
+}
+
+class ARMABIInfo : public ABIInfo {
+  ABIArgInfo classifyReturnType(QualType RetTy,
+                                ASTContext &Context) const;
+
+  ABIArgInfo classifyArgumentType(QualType RetTy,
+                                  ASTContext &Context) const;
+
+  virtual void computeInfo(CGFunctionInfo &FI, ASTContext &Context) const;
+
+  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                                 CodeGenFunction &CGF) const;
+};
+
+void ARMABIInfo::computeInfo(CGFunctionInfo &FI, ASTContext &Context) const {
+  FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), Context);
+  for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
+       it != ie; ++it) {
+    it->info = classifyArgumentType(it->type, Context);
+  }
+}
+
+ABIArgInfo ARMABIInfo::classifyArgumentType(QualType Ty,
+                                            ASTContext &Context) const {
+  if (!CodeGenFunction::hasAggregateLLVMType(Ty)) {
+    return ABIArgInfo::getDirect();
+  }
+  // FIXME: This is kind of nasty... but there isn't much choice because the ARM
+  // backend doesn't support byval.
+  // FIXME: This doesn't handle alignment > 64 bits.
+  const llvm::Type* ElemTy;
+  unsigned SizeRegs;
+  if (Context.getTypeAlign(Ty) > 32) {
+    ElemTy = llvm::Type::Int64Ty;
+    SizeRegs = (Context.getTypeSize(Ty) + 63) / 64;
+  } else {
+    ElemTy = llvm::Type::Int32Ty;
+    SizeRegs = (Context.getTypeSize(Ty) + 31) / 32;
+  }
+  std::vector<const llvm::Type*> LLVMFields;
+  LLVMFields.push_back(llvm::ArrayType::get(ElemTy, SizeRegs));
+  const llvm::Type* STy = llvm::StructType::get(LLVMFields, true);
+  return ABIArgInfo::getCoerce(STy);
+}
+
+ABIArgInfo ARMABIInfo::classifyReturnType(QualType RetTy,
+                                          ASTContext &Context) const {
+  if (RetTy->isVoidType()) {
+    return ABIArgInfo::getIgnore();
+  } else if (CodeGenFunction::hasAggregateLLVMType(RetTy)) {
+    // Aggregates <= 4 bytes are returned in r0; other aggregates
+    // are returned indirectly.
+    uint64_t Size = Context.getTypeSize(RetTy);
+    if (Size <= 32)
+      return ABIArgInfo::getCoerce(llvm::Type::Int32Ty);
+    return ABIArgInfo::getIndirect(0);
+  } else {
+    return ABIArgInfo::getDirect();
+  }
+}
+
+llvm::Value *ARMABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                                      CodeGenFunction &CGF) const {
+  // FIXME: Need to handle alignment
+  const llvm::Type *BP = llvm::PointerType::getUnqual(llvm::Type::Int8Ty);
+  const llvm::Type *BPP = llvm::PointerType::getUnqual(BP);
+
+  CGBuilderTy &Builder = CGF.Builder;
+  llvm::Value *VAListAddrAsBPP = Builder.CreateBitCast(VAListAddr, BPP,
+                                                       "ap");
+  llvm::Value *Addr = Builder.CreateLoad(VAListAddrAsBPP, "ap.cur");
+  llvm::Type *PTy =
+    llvm::PointerType::getUnqual(CGF.ConvertType(Ty));
+  llvm::Value *AddrTyped = Builder.CreateBitCast(Addr, PTy);
+
+  uint64_t Offset =
+    llvm::RoundUpToAlignment(CGF.getContext().getTypeSize(Ty) / 8, 4);
+  llvm::Value *NextAddr =
+    Builder.CreateGEP(Addr,
+                      llvm::ConstantInt::get(llvm::Type::Int32Ty, Offset),
+                      "ap.next");
+  Builder.CreateStore(NextAddr, VAListAddrAsBPP);
+
+  return AddrTyped;
+}
+
+ABIArgInfo DefaultABIInfo::classifyReturnType(QualType RetTy,
+                                              ASTContext &Context) const {
+  if (RetTy->isVoidType()) {
+    return ABIArgInfo::getIgnore();
+  } else if (CodeGenFunction::hasAggregateLLVMType(RetTy)) {
+    return ABIArgInfo::getIndirect(0);
+  } else {
+    return ABIArgInfo::getDirect();
+  }
+}
+
+ABIArgInfo DefaultABIInfo::classifyArgumentType(QualType Ty,
+                                                ASTContext &Context) const {
+  if (CodeGenFunction::hasAggregateLLVMType(Ty)) {
+    return ABIArgInfo::getIndirect(0);
+  } else {
+    return ABIArgInfo::getDirect();
+  }
+}
+
+llvm::Value *DefaultABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                                       CodeGenFunction &CGF) const {
+  return 0;
+}
+
+const ABIInfo &CodeGenTypes::getABIInfo() const {
+  if (TheABIInfo)
+    return *TheABIInfo;
+
+  // For now we just cache this in the CodeGenTypes and don't bother
+  // to free it.
+  const char *TargetPrefix = getContext().Target.getTargetPrefix();
+  if (strcmp(TargetPrefix, "x86") == 0) {
+    bool IsDarwin = strstr(getContext().Target.getTargetTriple(), "darwin");
+    switch (getContext().Target.getPointerWidth(0)) {
+    case 32:
+      return *(TheABIInfo = new X86_32ABIInfo(Context, IsDarwin));
+    case 64:
+      return *(TheABIInfo = new X86_64ABIInfo());
+    }
+  } else if (strcmp(TargetPrefix, "arm") == 0) {
+    // FIXME: Support for OABI?
+    return *(TheABIInfo = new ARMABIInfo());
+  } else if (strcmp(TargetPrefix, "pic16") == 0) {
+    return *(TheABIInfo = new PIC16ABIInfo());
+  }
+
+  return *(TheABIInfo = new DefaultABIInfo);
+}
diff --git a/lib/Driver/Tools.cpp b/lib/Driver/Tools.cpp
index eca6413..6b082c8 100644
--- a/lib/Driver/Tools.cpp
+++ b/lib/Driver/Tools.cpp
@@ -329,7 +329,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   if (Args.hasFlag(options::OPT_msoft_float,
                    options::OPT_mno_soft_float,
                    false))
-    CmdArgs.push_back("--soft-float");
+    CmdArgs.push_back("--no-implicit-float");
 
   // FIXME: Handle -mtune=.
   (void) Args.hasArg(options::OPT_mtune_EQ);
@@ -505,6 +505,16 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       CmdArgs.push_back("-fblocks=0");
   }
 
+  // -fsigned-char/-funsigned-char default varies depending on platform; only
+  // pass if specified.
+  if (Arg *A = Args.getLastArg(options::OPT_fsigned_char,
+                               options::OPT_funsigned_char)) {
+    if (A->getOption().matches(options::OPT_fsigned_char))
+      CmdArgs.push_back("-fsigned-char");
+    else
+      CmdArgs.push_back("-fsigned-char=0");
+  }
+
   // -fno-pascal-strings is default, only pass non-default. If the
   // -tool chain happened to translate to -mpascal-strings, we want to
   // -back translate here.
@@ -539,6 +549,9 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   if (Args.hasFlag(options::OPT_fdiagnostics_show_option, 
                    options::OPT_fno_diagnostics_show_option))
     CmdArgs.push_back("-fdiagnostics-show-option");
+  if (!Args.hasFlag(options::OPT_fcolor_diagnostics,
+                   options::OPT_fno_color_diagnostics))
+    CmdArgs.push_back("-fno-color-diagnostics");
 
   // -fdollars-in-identifiers default varies depending on platform and
   // language; only pass if specified.
diff --git a/lib/Frontend/InitPreprocessor.cpp b/lib/Frontend/InitPreprocessor.cpp
index e3a45d4..01729fa 100644
--- a/lib/Frontend/InitPreprocessor.cpp
+++ b/lib/Frontend/InitPreprocessor.cpp
@@ -23,8 +23,8 @@ namespace clang {
 // Append a #define line to Buf for Macro.  Macro should be of the form XXX,
 // in which case we emit "#define XXX 1" or "XXX=Y z W" in which case we emit
 // "#define XXX Y z W".  To get a #define with no value, use "XXX=".
-static void DefineBuiltinMacro(std::vector<char> &Buf, const char *Macro,
-                               const char *Command = "#define ") {
+static void DefineBuiltinMacro(std::vector<char> &Buf, const char *Macro) {
+  const char *Command = "#define ";
   Buf.insert(Buf.end(), Command, Command+strlen(Command));
   if (const char *Equal = strchr(Macro, '=')) {
     // Turn the = into ' '.
@@ -367,7 +367,7 @@ static void InitializePredefinedMacros(const TargetInfo &TI,
   sprintf(MacroBuf, "__POINTER_WIDTH__=%d", (int)TI.getPointerWidth(0));
   DefineBuiltinMacro(Buf, MacroBuf);
   
-  if (!TI.isCharSigned())
+  if (!LangOpts.CharIsSigned)
     DefineBuiltinMacro(Buf, "__CHAR_UNSIGNED__");  
 
   // Define fixed-sized integer types for stdint.h
@@ -403,11 +403,6 @@ static void InitializePredefinedMacros(const TargetInfo &TI,
   // command line options or something.
   DefineBuiltinMacro(Buf, "__FINITE_MATH_ONLY__=0");
 
-  if (LangOpts.Static)
-    DefineBuiltinMacro(Buf, "__STATIC__=1");
-  else
-    DefineBuiltinMacro(Buf, "__DYNAMIC__=1");
-
   if (LangOpts.GNUInline)
     DefineBuiltinMacro(Buf, "__GNUC_GNU_INLINE__=1");
   else
diff --git a/lib/Frontend/PCHReader.cpp b/lib/Frontend/PCHReader.cpp
index 63e4337..87fc839 100644
--- a/lib/Frontend/PCHReader.cpp
+++ b/lib/Frontend/PCHReader.cpp
@@ -1521,6 +1521,7 @@ bool PCHReader::ParseLanguageOptions(
   PARSE_LANGOPT_IMPORTANT(GNUInline, diag::warn_pch_gnu_inline);
   PARSE_LANGOPT_IMPORTANT(NoInline, diag::warn_pch_no_inline);
   PARSE_LANGOPT_IMPORTANT(AccessControl, diag::warn_pch_access_control);
+  PARSE_LANGOPT_IMPORTANT(CharIsSigned, diag::warn_pch_char_signed);
   if ((LangOpts.getGCMode() != 0) != (Record[Idx] != 0)) {
     Diag(diag::warn_pch_gc_mode) 
       << (unsigned)Record[Idx] << LangOpts.getGCMode();
diff --git a/lib/Frontend/PCHWriter.cpp b/lib/Frontend/PCHWriter.cpp
index 80e863b..765fecb 100644
--- a/lib/Frontend/PCHWriter.cpp
+++ b/lib/Frontend/PCHWriter.cpp
@@ -556,6 +556,8 @@ void PCHWriter::WriteLanguageOptions(const LangOptions &LangOpts) {
   Record.push_back(LangOpts.NoInline); // Should __NO_INLINE__ be defined.
   Record.push_back(LangOpts.AccessControl); // Whether C++ access control should
                                             // be enabled.
+  Record.push_back(LangOpts.CharIsSigned); // Whether char is a signed or
+                                           // unsigned type
   Record.push_back(LangOpts.getGCMode());
   Record.push_back(LangOpts.getVisibilityMode());
   Record.push_back(LangOpts.InstantiationDepth);
diff --git a/lib/Frontend/PrintParserCallbacks.cpp b/lib/Frontend/PrintParserCallbacks.cpp
index f02d5d4..b9fe068 100644
--- a/lib/Frontend/PrintParserCallbacks.cpp
+++ b/lib/Frontend/PrintParserCallbacks.cpp
@@ -220,6 +220,7 @@ namespace {
     }
   
     virtual DeclPtrTy ActOnIvar(Scope *S, SourceLocation DeclStart,
+                                DeclPtrTy IntfDecl,
                                 Declarator &D, ExprTy *BitfieldWidth,
                                 tok::ObjCKeywordKind visibility) {
       Out << __FUNCTION__ << "\n";
diff --git a/lib/Frontend/TextDiagnosticPrinter.cpp b/lib/Frontend/TextDiagnosticPrinter.cpp
index b1c0533..6699c65 100644
--- a/lib/Frontend/TextDiagnosticPrinter.cpp
+++ b/lib/Frontend/TextDiagnosticPrinter.cpp
@@ -20,6 +20,20 @@
 #include <algorithm>
 using namespace clang;
 
+static const enum llvm::raw_ostream::Colors noteColor =
+  llvm::raw_ostream::BLACK;
+static const enum llvm::raw_ostream::Colors fixitColor =
+  llvm::raw_ostream::GREEN;
+static const enum llvm::raw_ostream::Colors caretColor =
+  llvm::raw_ostream::GREEN;
+static const enum llvm::raw_ostream::Colors warningColor =
+  llvm::raw_ostream::MAGENTA;
+static const enum llvm::raw_ostream::Colors errorColor = llvm::raw_ostream::RED;
+static const enum llvm::raw_ostream::Colors fatalColor = llvm::raw_ostream::RED;
+// used for changing only the bold attribute
+static const enum llvm::raw_ostream::Colors savedColor =
+  llvm::raw_ostream::SAVEDCOLOR;
+
 /// \brief Number of spaces to indent when word-wrapping.
 const unsigned WordWrapIndentation = 6;
 
@@ -396,12 +410,22 @@ void TextDiagnosticPrinter::EmitCaretDiagnostic(SourceLocation Loc,
   
   // Emit what we have computed.
   OS << SourceLine << '\n';
+
+  if (UseColors)
+    OS.changeColor(caretColor, true);
   OS << CaretLine << '\n';
+  if (UseColors)
+    OS.resetColor();
 
   if (!FixItInsertionLine.empty()) {
+    if (UseColors)
+      // Print fixit line in color
+      OS.changeColor(fixitColor, false);
     if (PrintRangeInfo) 
       OS << ' ';
     OS << FixItInsertionLine << '\n';
+    if (UseColors)
+      OS.resetColor();
   }
 }
 
@@ -598,6 +622,8 @@ void TextDiagnosticPrinter::HandleDiagnostic(Diagnostic::Level Level,
     
     // Compute the column number.
     if (ShowLocation) {
+      if (UseColors)
+        OS.changeColor(savedColor, true);
       OS << PLoc.getFilename() << ':' << LineNo << ':';
       if (ShowColumn)
         if (unsigned ColNo = PLoc.getColumn())
@@ -638,6 +664,19 @@ void TextDiagnosticPrinter::HandleDiagnostic(Diagnostic::Level Level,
           OS << ':';
       }
       OS << ' ';
+      if (UseColors)
+        OS.resetColor();
+    }
+  }
+
+  if (UseColors) {
+    // Print diagnostic category in bold and color
+    switch (Level) {
+    case Diagnostic::Ignored: assert(0 && "Invalid diagnostic type");
+    case Diagnostic::Note:    OS.changeColor(noteColor, true); break;
+    case Diagnostic::Warning: OS.changeColor(warningColor, true); break;
+    case Diagnostic::Error:   OS.changeColor(errorColor, true); break;
+    case Diagnostic::Fatal:   OS.changeColor(fatalColor, true); break;
     }
   }
   
@@ -648,7 +687,10 @@ void TextDiagnosticPrinter::HandleDiagnostic(Diagnostic::Level Level,
   case Diagnostic::Error:   OS << "error: "; break;
   case Diagnostic::Fatal:   OS << "fatal error: "; break;
   }
-  
+
+  if (UseColors)
+    OS.resetColor();
+
   llvm::SmallString<100> OutStr;
   Info.FormatDiagnostic(OutStr);
   
@@ -659,6 +701,16 @@ void TextDiagnosticPrinter::HandleDiagnostic(Diagnostic::Level Level,
       OutStr += ']';
     }
   
+  if (UseColors) {
+    // Print warnings, errors and fatal errors in bold, no color
+    switch (Level) {
+    case Diagnostic::Warning: OS.changeColor(savedColor, true); break;
+    case Diagnostic::Error:   OS.changeColor(savedColor, true); break;
+    case Diagnostic::Fatal:   OS.changeColor(savedColor, true); break;
+    default: break; //don't bold notes
+    }
+  }
+
   if (MessageLength) {
     // We will be word-wrapping the error message, so compute the
     // column number where we currently are (after printing the
@@ -669,6 +721,8 @@ void TextDiagnosticPrinter::HandleDiagnostic(Diagnostic::Level Level,
     OS.write(OutStr.begin(), OutStr.size());
   }
   OS << '\n';
+  if (UseColors)
+    OS.resetColor();
   
   // If caret diagnostics are enabled and we have location, we want to
   // emit the caret.  However, we only do this if the location moved
diff --git a/lib/Headers/emmintrin.h b/lib/Headers/emmintrin.h
index c96000a..23a61a0 100644
--- a/lib/Headers/emmintrin.h
+++ b/lib/Headers/emmintrin.h
@@ -40,7 +40,8 @@ typedef char __v16qi __attribute__((__vector_size__(16)));
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_add_sd(__m128d a, __m128d b)
 {
-  return __builtin_ia32_addsd(a, b);
+  a[0] += b[0];
+  return a;
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
@@ -52,7 +53,8 @@ _mm_add_pd(__m128d a, __m128d b)
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_sub_sd(__m128d a, __m128d b)
 {
-  return __builtin_ia32_subsd(a, b);
+  a[0] -= b[0];
+  return a;
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
@@ -64,7 +66,8 @@ _mm_sub_pd(__m128d a, __m128d b)
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_mul_sd(__m128d a, __m128d b)
 {
-  return __builtin_ia32_mulsd(a, b);
+  a[0] *= b[0];
+  return a;
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
@@ -76,7 +79,8 @@ _mm_mul_pd(__m128d a, __m128d b)
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_div_sd(__m128d a, __m128d b)
 {
-  return __builtin_ia32_divsd(a, b);
+  a[0] /= b[0];
+  return a;
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
@@ -125,25 +129,25 @@ _mm_max_pd(__m128d a, __m128d b)
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_and_pd(__m128d a, __m128d b)
 {
-  return __builtin_ia32_andpd(a, b);
+  return (__m128d)((__v4si)a & (__v4si)b);
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_andnot_pd(__m128d a, __m128d b)
 {
-  return __builtin_ia32_andnpd(a, b);
+  return (__m128d)(~(__v4si)a & (__v4si)b);
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_or_pd(__m128d a, __m128d b)
 {
-  return __builtin_ia32_orpd(a, b);
+  return (__m128d)((__v4si)a | (__v4si)b);
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_xor_pd(__m128d a, __m128d b)
 {
-  return __builtin_ia32_xorpd(a, b);
+  return (__m128d)((__v4si)a ^ (__v4si)b);
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
@@ -383,19 +387,22 @@ _mm_cvtsd_si32(__m128d a)
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_cvtsd_ss(__m128 a, __m128d b)
 {
-  return __builtin_ia32_cvtsd2ss(a, b);
+  a[0] = b[0];
+  return a;
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_cvtsi32_sd(__m128d a, int b)
 {
-  return __builtin_ia32_cvtsi2sd(a, b);
+  a[0] = b;
+  return a;
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_cvtss_sd(__m128d a, __m128 b)
 {
-  return __builtin_ia32_cvtss2sd(a, b);
+  a[0] = b[0];
+  return a;
 }
 
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
@@ -407,7 +414,7 @@ _mm_cvttpd_epi32(__m128d a)
 static inline int __attribute__((__always_inline__, __nodebug__))
 _mm_cvttsd_si32(__m128d a)
 {
-  return __builtin_ia32_cvttsd2si(a);
+  return a[0];
 }
 
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
@@ -669,7 +676,7 @@ _mm_mulhi_epu16(__m128i a, __m128i b)
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_mullo_epi16(__m128i a, __m128i b)
 {
-  return (__m128i)__builtin_ia32_pmullw128((__v8hi)a, (__v8hi)b);
+  return (__m128i)((__v8hi)a * (__v8hi)b);
 }
 
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
@@ -747,25 +754,25 @@ _mm_subs_epu16(__m128i a, __m128i b)
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_and_si128(__m128i a, __m128i b)
 {
-  return __builtin_ia32_pand128(a, b);
+  return a & b;
 }
 
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_andnot_si128(__m128i a, __m128i b)
 {
-  return __builtin_ia32_pandn128(a, b);
+  return ~a & b;
 }
 
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_or_si128(__m128i a, __m128i b)
 {
-  return __builtin_ia32_por128(a, b);
+  return a | b;
 }
 
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_xor_si128(__m128i a, __m128i b)
 {
-  return __builtin_ia32_pxor128(a, b);
+  return a ^ b;
 }
 
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
@@ -934,7 +941,8 @@ _mm_cmplt_epi32(__m128i a, __m128i b)
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
 _mm_cvtsi64_sd(__m128d a, long long b)
 {
-  return __builtin_ia32_cvtsi642sd(a, b);
+  a[0] = b;
+  return a;
 }
 
 static inline long long __attribute__((__always_inline__, __nodebug__))
@@ -946,7 +954,7 @@ _mm_cvtsd_si64(__m128d a)
 static inline long long __attribute__((__always_inline__, __nodebug__))
 _mm_cvttsd_si64(__m128d a)
 {
-  return __builtin_ia32_cvttsd2si64(a);
+  return a[0];
 }
 #endif
 
@@ -1181,7 +1189,9 @@ _mm_extract_epi16(__m128i a, int imm)
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_insert_epi16(__m128i a, int b, int imm)
 {
-  return (__m128i)__builtin_ia32_vec_set_v8hi((__v8hi)a, b, imm);
+  __v8hi c = (__v8hi)a;
+  c[imm & 7] = b;
+  return (__m128i)c;
 }
 
 static inline int __attribute__((__always_inline__, __nodebug__))
@@ -1190,9 +1200,20 @@ _mm_movemask_epi8(__m128i a)
   return __builtin_ia32_pmovmskb128((__v16qi)a);
 }
 
-#define _mm_shuffle_epi32(a, imm) ((__m128i)__builtin_ia32_pshufd((__v4si)(a), (imm)))
-#define _mm_shufflehi_epi16(a, imm) ((__m128i)__builtin_ia32_pshufhw((__v8hi)(a), (imm)))
-#define _mm_shufflelo_epi16(a, imm) ((__m128i)__builtin_ia32_pshuflw((__v8hi)(a), (imm)))
+#define _mm_shuffle_epi32(a, imm) \
+  ((__m128i)__builtin_shufflevector((__v4si)(a), (__v4si) {0}, \
+                                    (imm) & 0x3, ((imm) & 0xc) >> 2, \
+                                    ((imm) & 0x30) >> 4, ((imm) & 0xc0) >> 6))
+#define _mm_shufflelo_epi16(a, imm) \
+  ((__m128i)__builtin_shufflevector((__v8hi)(a), (__v8hi) {0}, \
+                                    (imm) & 0x3, ((imm) & 0xc) >> 2, \
+                                    ((imm) & 0x30) >> 4, ((imm) & 0xc0) >> 6, \
+                                    4, 5, 6, 7))
+#define _mm_shufflehi_epi16(a, imm) \
+  ((__m128i)__builtin_shufflevector((__v8hi)(a), (__v8hi) {0}, 0, 1, 2, 3, \
+                                    4 + ((imm) & 0x3), 4 + ((imm) & 0xc) >> 2, \
+                                    4 + ((imm) & 0x30) >> 4, \
+                                    4 + ((imm) & 0xc0) >> 6))
 
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_unpackhi_epi8(__m128i a, __m128i b)
@@ -1257,7 +1278,7 @@ _mm_movpi64_pi64(__m64 a)
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_move_epi64(__m128i a)
 {
-  return (__m128i){ a[0], 0 };
+  return __builtin_shufflevector(a, (__m128i){ 0 }, 0, 2);
 }
 
 static inline __m128d __attribute__((__always_inline__, __nodebug__))
@@ -1278,7 +1299,8 @@ _mm_movemask_pd(__m128d a)
   return __builtin_ia32_movmskpd(a);
 }
 
-#define _mm_shuffle_pd(a, b, i) (__builtin_ia32_shufpd((a), (b), (i)))
+#define _mm_shuffle_pd(a, b, i) (__builtin_shufflevector((a), (b), (i) & 1, \
+                                                         (((i) & 2) >> 1) + 2))
 
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_castpd_ps(__m128d in)
diff --git a/lib/Headers/mmintrin.h b/lib/Headers/mmintrin.h
index 339d212..8ea3c47 100644
--- a/lib/Headers/mmintrin.h
+++ b/lib/Headers/mmintrin.h
@@ -415,13 +415,13 @@ _mm_set1_pi32(int __i)
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_set1_pi16(short __s)
 {
-    return (__m64)(__v4hi){ __s };
+    return (__m64)(__v4hi){ __s, __s, __s, __s };
 }
 
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_set1_pi8(char __b)
 {
-    return (__m64)(__v8qi){ __b };
+    return (__m64)(__v8qi){ __b, __b, __b, __b, __b, __b, __b, __b };
 }
 
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
diff --git a/lib/Headers/tmmintrin.h b/lib/Headers/tmmintrin.h
index e9715f1..374a27e 100644
--- a/lib/Headers/tmmintrin.h
+++ b/lib/Headers/tmmintrin.h
@@ -114,19 +114,19 @@ _mm_hsub_epi16(__m128i a, __m128i b)
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_hsub_epi32(__m128i a, __m128i b)
 {
-    return (__m128i)__builtin_ia32_psubd128((__v4si)a, (__v4si)b);
+    return (__m128i)__builtin_ia32_phsubd128((__v4si)a, (__v4si)b);
 }
 
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_hsub_pi16(__m64 a, __m64 b)
 {
-    return (__m64)__builtin_ia32_psubw((__v4hi)a, (__v4hi)b);
+    return (__m64)__builtin_ia32_phsubw((__v4hi)a, (__v4hi)b);
 }
 
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_hsub_pi32(__m64 a, __m64 b)
 {
-    return (__m64)__builtin_ia32_psubd((__v2si)a, (__v2si)b);
+    return (__m64)__builtin_ia32_phsubd((__v2si)a, (__v2si)b);
 }
 
 static inline __m128i __attribute__((__always_inline__, __nodebug__))
diff --git a/lib/Headers/xmmintrin.h b/lib/Headers/xmmintrin.h
index c104f63..7291f88 100644
--- a/lib/Headers/xmmintrin.h
+++ b/lib/Headers/xmmintrin.h
@@ -38,7 +38,8 @@ typedef float __m128 __attribute__((__vector_size__(16)));
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_add_ss(__m128 a, __m128 b)
 {
-  return __builtin_ia32_addss(a, b);
+  a[0] += b[0];
+  return a;
 }
 
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
@@ -50,7 +51,8 @@ _mm_add_ps(__m128 a, __m128 b)
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_sub_ss(__m128 a, __m128 b)
 {
-  return __builtin_ia32_subss(a, b);
+  a[0] -= b[0];
+  return a;
 }
 
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
@@ -62,7 +64,8 @@ _mm_sub_ps(__m128 a, __m128 b)
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_mul_ss(__m128 a, __m128 b)
 {
-  return __builtin_ia32_mulss(a, b);
+  a[0] *= b[0];
+  return a;
 }
 
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
@@ -74,7 +77,8 @@ _mm_mul_ps(__m128 a, __m128 b)
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_div_ss(__m128 a, __m128 b)
 {
-  return __builtin_ia32_divss(a, b);
+  a[0] /= b[0];
+  return a;
 }
 
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
@@ -146,25 +150,29 @@ _mm_max_ps(__m128 a, __m128 b)
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_and_ps(__m128 a, __m128 b)
 {
-  return __builtin_ia32_andps(a, b);
+  typedef int __v4si __attribute__((__vector_size__(16)));
+  return (__m128)((__v4si)a & (__v4si)b);
 }
 
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_andnot_ps(__m128 a, __m128 b)
 {
-  return __builtin_ia32_andnps(a, b);
+  typedef int __v4si __attribute__((__vector_size__(16)));
+  return (__m128)(~(__v4si)a & (__v4si)b);
 }
 
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_or_ps(__m128 a, __m128 b)
 {
-  return __builtin_ia32_orps(a, b);
+  typedef int __v4si __attribute__((__vector_size__(16)));
+  return (__m128)((__v4si)a | (__v4si)b);
 }
 
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_xor_ps(__m128 a, __m128 b)
 {
-  return __builtin_ia32_xorps(a, b);
+  typedef int __v4si __attribute__((__vector_size__(16)));
+  return (__m128)((__v4si)a ^ ~(__v4si)b);
 }
 
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
@@ -389,12 +397,16 @@ _mm_cvtss_si32(__m128 a)
   return __builtin_ia32_cvtss2si(a);
 }
 
+#ifdef __x86_64__
+
 static inline long long __attribute__((__always_inline__, __nodebug__))
 _mm_cvtss_si64(__m128 a)
 {
   return __builtin_ia32_cvtss2si64(a);
 }
 
+#endif
+
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_cvtps_pi32(__m128 a)
 {
@@ -404,13 +416,13 @@ _mm_cvtps_pi32(__m128 a)
 static inline int __attribute__((__always_inline__, __nodebug__))
 _mm_cvttss_si32(__m128 a)
 {
-  return __builtin_ia32_cvttss2si(a);
+  return a[0];
 }
 
 static inline long long __attribute__((__always_inline__, __nodebug__))
 _mm_cvttss_si64(__m128 a)
 {
-  return __builtin_ia32_cvttss2si64(a);
+  return a[0];
 }
 
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
@@ -422,7 +434,8 @@ _mm_cvttps_pi32(__m128 a)
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_cvtsi32_ss(__m128 a, int b)
 {
-  return __builtin_ia32_cvtsi2ss(a, b);
+  a[0] = b;
+  return a;
 }
 
 #ifdef __x86_64__
@@ -430,7 +443,8 @@ _mm_cvtsi32_ss(__m128 a, int b)
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_cvtsi64_ss(__m128 a, long long b)
 {
-  return __builtin_ia32_cvtsi642ss(a, b);
+  a[0] = b;
+  return a;
 }
 
 #endif
@@ -456,6 +470,13 @@ _mm_loadh_pi(__m128 a, __m64 const *p)
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_loadl_pi(__m128 a, __m64 const *p)
 {
+#if 0
+  // FIXME: This should work, but gives really crappy code at the moment
+  __m128 b;
+  b[0] = *(float*)p;
+  b[1] = *((float*)p+1);
+  return __builtin_shufflevector(a, b, 0, 1, 4, 5);
+#endif
   return __builtin_ia32_loadlps(a, (__v2si *)p);
 }
 
@@ -604,26 +625,17 @@ _mm_sfence(void)
 static inline int __attribute__((__always_inline__, __nodebug__))
 _mm_extract_pi16(__m64 a, int n)
 {
-  /* FIXME: 
-   * This should force n to be an immediate.
-   * This does not use the PEXTRW instruction. From looking at the LLVM source, the
-     instruction doesn't seem to be hooked up. 
-   * The code could probably be made better :)
-   */
   __v4hi b = (__v4hi)a;
-  return b[(n == 0) ? 0 : (n == 1 ? 1 : (n == 2 ? 2 : 3))];
+  return (unsigned short)b[n & 3];
 }
 
-/* FIXME: Implement this. We could add a __builtin_insertelement function that's similar to
-   the already existing __builtin_shufflevector.
-*/
-/*
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_insert_pi16(__m64 a, int d, int n)
 {
-   return (__m64){ 0LL };
+   __v4hi b = (__v4hi)a;
+   b[n & 3] = d;
+   return (__m64)b;
 }
-*/
 
 static inline __m64 __attribute__((__always_inline__, __nodebug__))
 _mm_max_pi16(__m64 a, __m64 b)
@@ -661,7 +673,10 @@ _mm_mulhi_pu16(__m64 a, __m64 b)
   return (__m64)__builtin_ia32_pmulhuw((__v4hi)a, (__v4hi)b);  
 }
 
-#define _mm_shuffle_pi16(a, n) ((__m64)__builtin_ia32_pshufw((__v4hi)a, n))
+#define _mm_shuffle_pi16(a, n) \
+  ((__m64)__builtin_shufflevector((__v4hi)(a), (__v4hi) {0}, \
+                                  (n) & 0x3, ((n) & 0xc) >> 2, \
+                                  ((n) & 0x30) >> 4, ((n) & 0xc0) >> 6))
 
 static inline void __attribute__((__always_inline__, __nodebug__))
 _mm_maskmove_si64(__m64 d, __m64 n, char *p)
@@ -699,7 +714,10 @@ _mm_setcsr(unsigned int i)
   __builtin_ia32_ldmxcsr(i);
 }
 
-#define _mm_shuffle_ps(a, b, mask) (__builtin_ia32_shufps(a, b, mask))
+#define _mm_shuffle_ps(a, b, mask) \
+        (__builtin_shufflevector(a, b, (mask) & 0x3, ((mask) & 0xc) >> 2, \
+                                 (((mask) & 0x30) >> 4) + 4, \
+                                 (((mask) & 0xc0) >> 6) + 4))
 
 static inline __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_unpackhi_ps(__m128 a, __m128 b)
diff --git a/lib/Lex/LiteralSupport.cpp b/lib/Lex/LiteralSupport.cpp
index 0324c0b..4d10974 100644
--- a/lib/Lex/LiteralSupport.cpp
+++ b/lib/Lex/LiteralSupport.cpp
@@ -691,7 +691,7 @@ CharLiteralParser::CharLiteralParser(const char *begin, const char *end,
   // character constants are not sign extended in the this implementation:
   // '\xFF\xFF' = 65536 and '\x0\xFF' = 255, which matches GCC.
   if (!IsWide && NumCharsSoFar == 1 && (Value & 128) &&
-      PP.getTargetInfo().isCharSigned())
+      PP.getLangOptions().CharIsSigned)
     Value = (signed char)Value;
 }
 
diff --git a/lib/Lex/PPExpressions.cpp b/lib/Lex/PPExpressions.cpp
index 709e316..c98acc4 100644
--- a/lib/Lex/PPExpressions.cpp
+++ b/lib/Lex/PPExpressions.cpp
@@ -232,7 +232,7 @@ static bool EvaluateValue(PPValue &Result, Token &PeekTok, DefinedTracker &DT,
     // Set the value.
     Val = Literal.getValue();
     // Set the signedness.
-    Val.setIsUnsigned(!TI.isCharSigned());
+    Val.setIsUnsigned(!PP.getLangOptions().CharIsSigned);
     
     if (Result.Val.getBitWidth() > Val.getBitWidth()) {
       Result.Val = Val.extend(Result.Val.getBitWidth());
diff --git a/lib/Parse/ParseObjc.cpp b/lib/Parse/ParseObjc.cpp
index 3014f95..cb7fe58 100644
--- a/lib/Parse/ParseObjc.cpp
+++ b/lib/Parse/ParseObjc.cpp
@@ -891,6 +891,7 @@ void Parser::ParseObjCClassInstanceVariables(DeclPtrTy interfaceDecl,
       // Install the declarator into interfaceDecl.
       DeclPtrTy Field = Actions.ActOnIvar(CurScope,
                                           DS.getSourceRange().getBegin(),
+                                          interfaceDecl,
                                           FD.D, FD.BitfieldSize, visibility);
       AllIvarDecls.push_back(Field);
     }
diff --git a/lib/Parse/ParsePragma.cpp b/lib/Parse/ParsePragma.cpp
index 94695e4..58c729a 100644
--- a/lib/Parse/ParsePragma.cpp
+++ b/lib/Parse/ParsePragma.cpp
@@ -100,6 +100,12 @@ void PragmaPackHandler::HandlePragma(Preprocessor &PP, Token &PackTok) {
     return;
   }
 
+  PP.Lex(Tok);
+  if (Tok.isNot(tok::eom)) {
+    PP.Diag(Tok.getLocation(), diag::warn_pragma_extra_tokens_at_eol) << "pack";
+    return;
+  }
+
   SourceLocation RParenLoc = Tok.getLocation();
   Actions.ActOnPragmaPack(Kind, Name, Alignment.release(), PackLoc,
                           LParenLoc, RParenLoc);
@@ -172,7 +178,14 @@ void PragmaUnusedHandler::HandlePragma(Preprocessor &PP, Token &UnusedTok) {
     PP.Diag(Tok.getLocation(), diag::warn_pragma_unused_expected_punc);
     return;
   }
-  
+
+  PP.Lex(Tok);
+  if (Tok.isNot(tok::eom)) {
+    PP.Diag(Tok.getLocation(), diag::warn_pragma_extra_tokens_at_eol) <<
+        "unused";
+    return;
+  }
+
   // Verify that we have a location for the right parenthesis.
   assert(RParenLoc.isValid() && "Valid '#pragma unused' must have ')'");
   assert(!Ex.empty() && "Valid '#pragma unused' must have arguments");
@@ -180,3 +193,45 @@ void PragmaUnusedHandler::HandlePragma(Preprocessor &PP, Token &UnusedTok) {
   // Perform the action to handle the pragma.    
   Actions.ActOnPragmaUnused(&Ex[0], Ex.size(), UnusedLoc, LParenLoc, RParenLoc);
 }
+
+// #pragma weak identifier
+// #pragma weak identifier '=' identifier
+void PragmaWeakHandler::HandlePragma(Preprocessor &PP, Token &WeakTok) {
+  // FIXME: Should we be expanding macros here? My guess is no.
+  SourceLocation WeakLoc = WeakTok.getLocation();
+
+  Token Tok;
+  PP.Lex(Tok);
+  if (Tok.isNot(tok::identifier)) {
+    PP.Diag(Tok.getLocation(), diag::warn_pragma_expected_identifier) << "weak";
+    return;
+  }
+
+  IdentifierInfo *WeakName = Tok.getIdentifierInfo(), *AliasName = 0;
+  SourceLocation WeakNameLoc = Tok.getLocation(), AliasNameLoc;
+
+  PP.Lex(Tok);
+  if (Tok.is(tok::equal)) {
+    PP.Lex(Tok);
+    if (Tok.isNot(tok::identifier)) {
+      PP.Diag(Tok.getLocation(), diag::warn_pragma_expected_identifier) 
+          << "weak";
+      return;
+    }
+    AliasName = Tok.getIdentifierInfo();
+    AliasNameLoc = Tok.getLocation();
+    PP.Lex(Tok);
+  }
+
+  if (Tok.isNot(tok::eom)) {
+    PP.Diag(Tok.getLocation(), diag::warn_pragma_extra_tokens_at_eol) << "weak";
+    return;
+  }
+
+  if (AliasName) {
+    Actions.ActOnPragmaWeakAlias(WeakName, AliasName, WeakLoc, WeakNameLoc,
+                                 AliasNameLoc);
+  } else {
+    Actions.ActOnPragmaWeakID(WeakName, WeakLoc, WeakNameLoc);
+  }
+}
diff --git a/lib/Parse/ParsePragma.h b/lib/Parse/ParsePragma.h
index 31b2a5f..39c86ee 100644
--- a/lib/Parse/ParsePragma.h
+++ b/lib/Parse/ParsePragma.h
@@ -39,6 +39,15 @@ public:
   virtual void HandlePragma(Preprocessor &PP, Token &FirstToken);  
 };  
 
+class PragmaWeakHandler : public PragmaHandler {
+  Action &Actions;
+public:
+  PragmaWeakHandler(const IdentifierInfo *N, Action &A)
+    : PragmaHandler(N), Actions(A) {}
+  
+  virtual void HandlePragma(Preprocessor &PP, Token &FirstToken);  
+};
+
 }  // end namespace clang
 
 #endif
diff --git a/lib/Parse/ParseStmt.cpp b/lib/Parse/ParseStmt.cpp
index 758b662..f041d7d 100644
--- a/lib/Parse/ParseStmt.cpp
+++ b/lib/Parse/ParseStmt.cpp
@@ -550,6 +550,8 @@ Parser::OwningStmtResult Parser::ParseIfStatement() {
   if (ParseParenExprOrCondition(CondExp))
     return StmtError();
 
+  FullExprArg FullCondExp(Actions.FullExpr(CondExp));
+  
   // C99 6.8.4p3 - In C99, the body of the if statement is a scope, even if
   // there is no compound stmt.  C90 does not have this clause.  We only do this
   // if the body isn't a compound statement to avoid push/pop in common cases.
@@ -631,7 +633,7 @@ Parser::OwningStmtResult Parser::ParseIfStatement() {
   if (ElseStmt.isInvalid())
     ElseStmt = Actions.ActOnNullStmt(ElseStmtLoc);
 
-  return Actions.ActOnIfStmt(IfLoc, Actions.FullExpr(CondExp), move(ThenStmt), 
+  return Actions.ActOnIfStmt(IfLoc, FullCondExp, move(ThenStmt), 
                              ElseLoc, move(ElseStmt));
 }
 
@@ -752,6 +754,8 @@ Parser::OwningStmtResult Parser::ParseWhileStatement() {
   if (ParseParenExprOrCondition(Cond))
     return StmtError();
 
+  FullExprArg FullCond(Actions.FullExpr(Cond));
+  
   // C99 6.8.5p5 - In C99, the body of the if statement is a scope, even if
   // there is no compound stmt.  C90 does not have this clause.  We only do this
   // if the body isn't a compound statement to avoid push/pop in common cases.
@@ -776,7 +780,7 @@ Parser::OwningStmtResult Parser::ParseWhileStatement() {
   if (Cond.isInvalid() || Body.isInvalid())
     return StmtError();
 
-  return Actions.ActOnWhileStmt(WhileLoc, Actions.FullExpr(Cond), move(Body));
+  return Actions.ActOnWhileStmt(WhileLoc, FullCond, move(Body));
 }
 
 /// ParseDoStatement
diff --git a/lib/Parse/Parser.cpp b/lib/Parse/Parser.cpp
index 1c2e8a6..a2a66f9 100644
--- a/lib/Parse/Parser.cpp
+++ b/lib/Parse/Parser.cpp
@@ -39,6 +39,10 @@ Parser::Parser(Preprocessor &pp, Action &actions)
           PragmaUnusedHandler(&PP.getIdentifierTable().get("unused"), actions,
                               *this));
   PP.AddPragmaHandler(0, UnusedHandler.get());
+
+  WeakHandler.reset(new
+          PragmaWeakHandler(&PP.getIdentifierTable().get("weak"), actions));
+  PP.AddPragmaHandler(0, WeakHandler.get());
 }
 
 /// If a crash happens while the parser is active, print out a line indicating
@@ -288,6 +292,8 @@ Parser::~Parser() {
   PackHandler.reset();
   PP.RemovePragmaHandler(0, UnusedHandler.get());
   UnusedHandler.reset();
+  PP.RemovePragmaHandler(0, WeakHandler.get());
+  WeakHandler.reset();
 }
 
 /// Initialize - Warm up the parser.
diff --git a/lib/Sema/CMakeLists.txt b/lib/Sema/CMakeLists.txt
index 321dac1..85c67df 100644
--- a/lib/Sema/CMakeLists.txt
+++ b/lib/Sema/CMakeLists.txt
@@ -23,6 +23,7 @@ add_clang_library(clangSema
   SemaOverload.cpp
   SemaStmt.cpp
   SemaTemplate.cpp
+  SemaTemplateDeduction.cpp
   SemaTemplateInstantiate.cpp
   SemaTemplateInstantiateDecl.cpp
   SemaTemplateInstantiateExpr.cpp
diff --git a/lib/Sema/Sema.h b/lib/Sema/Sema.h
index c428d29..d3bfef6 100644
--- a/lib/Sema/Sema.h
+++ b/lib/Sema/Sema.h
@@ -20,6 +20,7 @@
 #include "SemaOverload.h"
 #include "clang/AST/DeclBase.h"
 #include "clang/AST/DeclObjC.h"
+#include "clang/AST/DeclTemplate.h"
 #include "clang/Parse/Action.h"
 #include "clang/Sema/SemaDiagnostic.h"
 #include "llvm/ADT/SmallVector.h"
@@ -471,6 +472,7 @@ public:
                             Declarator *D = 0);
   
   virtual DeclPtrTy ActOnIvar(Scope *S, SourceLocation DeclStart,
+                              DeclPtrTy IntfDecl,
                               Declarator &D, ExprTy *BitfieldWidth,
                               tok::ObjCKeywordKind visibility);
 
@@ -1045,7 +1047,7 @@ public:
   LookupResult LookupName(Scope *S, DeclarationName Name, 
                           LookupNameKind NameKind, 
                           bool RedeclarationOnly = false,
-                          bool AllowBuiltinCreation = true,
+                          bool AllowBuiltinCreation = false,
                           SourceLocation Loc = SourceLocation());
   LookupResult LookupQualifiedName(DeclContext *LookupCtx, DeclarationName Name,
                                    LookupNameKind NameKind, 
@@ -1054,7 +1056,7 @@ public:
                                 DeclarationName Name,
                                 LookupNameKind NameKind, 
                                 bool RedeclarationOnly = false,
-                                bool AllowBuiltinCreation = true,
+                                bool AllowBuiltinCreation = false,
                                 SourceLocation Loc = SourceLocation());
 
   ObjCProtocolDecl *LookupProtocol(IdentifierInfo *II);
@@ -1629,6 +1631,12 @@ public:
                                                TypeTy *Ty,
                                                SourceLocation RParen);
 
+  /// MaybeCreateCXXExprWithTemporaries - If the list of temporaries is 
+  /// non-empty, will create a new CXXExprWithTemporaries expression.
+  /// Otherwise, just returs the passed in expression.
+  Expr *MaybeCreateCXXExprWithTemporaries(Expr *SubExpr, 
+                                          bool DestroyTemps = true);
+  
   virtual OwningExprResult ActOnFinishFullExpr(ExprArg Expr);
 
   bool RequireCompleteDeclContext(const CXXScopeSpec &SS);
@@ -1971,7 +1979,7 @@ public:
                                  const TemplateArgument *TemplateArgs,
                                  unsigned NumTemplateArgs,
                                  SourceLocation RAngleLoc,
-                       llvm::SmallVectorImpl<TemplateArgument> &Converted);
+                                 TemplateArgumentListBuilder &Converted);
 
   bool CheckTemplateArgument(TemplateTypeParmDecl *Param, QualType Arg,
                              SourceLocation ArgLoc);
@@ -1980,7 +1988,7 @@ public:
   bool CheckTemplateArgumentPointerToMember(Expr *Arg, NamedDecl *&Member);
   bool CheckTemplateArgument(NonTypeTemplateParmDecl *Param, 
                              QualType InstantiatedParamType, Expr *&Arg,
-                       llvm::SmallVectorImpl<TemplateArgument> *Converted = 0);
+                             TemplateArgumentListBuilder *Converted = 0);
   bool CheckTemplateArgument(TemplateTemplateParmDecl *Param, DeclRefExpr *Arg);
   bool TemplateParameterListsAreEqual(TemplateParameterList *New,
                                       TemplateParameterList *Old,
@@ -2019,16 +2027,9 @@ public:
                              const IdentifierInfo &II,
                              SourceRange Range);
 
-  bool DeduceTemplateArguments(QualType Param, QualType Arg,
-                               llvm::SmallVectorImpl<TemplateArgument> &Deduced);
-  bool DeduceTemplateArguments(const TemplateArgument &Param,
-                               const TemplateArgument &Arg,
-                               llvm::SmallVectorImpl<TemplateArgument> &Deduced);
-  bool DeduceTemplateArguments(const TemplateArgumentList &ParamList,
-                               const TemplateArgumentList &ArgList,
-                               llvm::SmallVectorImpl<TemplateArgument> &Deduced);
-  bool DeduceTemplateArguments(ClassTemplatePartialSpecializationDecl *Partial,
-                               const TemplateArgumentList &TemplateArgs);
+  TemplateArgumentList *
+  DeduceTemplateArguments(ClassTemplatePartialSpecializationDecl *Partial,
+                          const TemplateArgumentList &TemplateArgs);
                              
   //===--------------------------------------------------------------------===//
   // C++ Template Instantiation
@@ -2235,7 +2236,7 @@ public:
 
   QualType InstantiateType(QualType T, const TemplateArgumentList &TemplateArgs,
                            SourceLocation Loc, DeclarationName Entity);
-
+  
   OwningExprResult InstantiateExpr(Expr *E, 
                                    const TemplateArgumentList &TemplateArgs);
 
@@ -2456,7 +2457,19 @@ public:
                                  SourceLocation PragmaLoc, 
                                  SourceLocation LParenLoc,
                                  SourceLocation RParenLoc);
-  
+
+  /// ActOnPragmaWeakID - Called on well formed #pragma weak ident.
+  virtual void ActOnPragmaWeakID(IdentifierInfo* WeakName,
+                                 SourceLocation PragmaLoc,
+                                 SourceLocation WeakNameLoc);
+
+  /// ActOnPragmaWeakAlias - Called on well formed #pragma weak ident = ident.
+  virtual void ActOnPragmaWeakAlias(IdentifierInfo* WeakName,
+                                    IdentifierInfo* AliasName,
+                                    SourceLocation PragmaLoc,
+                                    SourceLocation WeakNameLoc,
+                                    SourceLocation AliasNameLoc);
+
   /// getPragmaPackAlignment() - Return the current alignment as specified by
   /// the current #pragma pack directive, or 0 if none is currently active.
   unsigned getPragmaPackAlignment() const;
diff --git a/lib/Sema/SemaDecl.cpp b/lib/Sema/SemaDecl.cpp
index 959154c..c67af29 100644
--- a/lib/Sema/SemaDecl.cpp
+++ b/lib/Sema/SemaDecl.cpp
@@ -575,10 +575,11 @@ void Sema::MergeTypeDefDecl(TypedefDecl *New, Decl *OldD) {
 
   // If we have a redefinition of a typedef in C, emit a warning.  This warning
   // is normally mapped to an error, but can be controlled with
-  // -Wtypedef-redefinition.  If either the original was in a system header,
-  // don't emit this for compatibility with GCC.
+  // -Wtypedef-redefinition.  If either the original or the redefinition is
+  // in a system header, don't emit this for compatibility with GCC.
   if (PP.getDiagnostics().getSuppressSystemWarnings() &&
-      Context.getSourceManager().isInSystemHeader(Old->getLocation()))
+      (Context.getSourceManager().isInSystemHeader(Old->getLocation()) ||
+       Context.getSourceManager().isInSystemHeader(New->getLocation())))
     return;
   
   Diag(New->getLocation(), diag::warn_redefinition_of_typedef)
@@ -3913,6 +3914,7 @@ TranslateIvarVisibility(tok::ObjCKeywordKind ivarVisibility) {
 /// in order to create an IvarDecl object for it.
 Sema::DeclPtrTy Sema::ActOnIvar(Scope *S,
                                 SourceLocation DeclStart, 
+                                DeclPtrTy IntfDecl,
                                 Declarator &D, ExprTy *BitfieldWidth,
                                 tok::ObjCKeywordKind Visibility) {
   
@@ -3951,14 +3953,28 @@ Sema::DeclPtrTy Sema::ActOnIvar(Scope *S,
   ObjCIvarDecl::AccessControl ac = 
     Visibility != tok::objc_not_keyword ? TranslateIvarVisibility(Visibility)
                                         : ObjCIvarDecl::None;
-
+  // Must set ivar's DeclContext to its enclosing interface.
+  Decl *EnclosingDecl = IntfDecl.getAs<Decl>();
+  DeclContext *EnclosingContext;
+  if (ObjCImplementationDecl *IMPDecl = 
+      dyn_cast<ObjCImplementationDecl>(EnclosingDecl)) {
+    // Case of ivar declared in an implementation. Context is that of its class.
+    ObjCInterfaceDecl* IDecl = IMPDecl->getClassInterface();
+    assert(IDecl && "No class- ActOnIvar");
+    EnclosingContext = cast_or_null<DeclContext>(IDecl);
+  }
+  else
+    EnclosingContext = dyn_cast<DeclContext>(EnclosingDecl);
+  assert(EnclosingContext && "null DeclContext for ivar - ActOnIvar");
+  
   // Construct the decl.
-  ObjCIvarDecl *NewID = ObjCIvarDecl::Create(Context, CurContext, Loc, II, T,ac,
+  ObjCIvarDecl *NewID = ObjCIvarDecl::Create(Context, 
+                                             EnclosingContext, Loc, II, T,ac,
                                              (Expr *)BitfieldWidth);
   
   if (II) {
     NamedDecl *PrevDecl = LookupName(S, II, LookupMemberName, true);
-    if (PrevDecl && isDeclInScope(PrevDecl, CurContext, S)
+    if (PrevDecl && isDeclInScope(PrevDecl, EnclosingContext, S)
         && !isa<TagDecl>(PrevDecl)) {
       Diag(Loc, diag::err_duplicate_member) << II;
       Diag(PrevDecl->getLocation(), diag::note_previous_declaration);
@@ -4099,7 +4115,11 @@ void Sema::ActOnFields(Scope* S,
     if (ObjCInterfaceDecl *ID = dyn_cast<ObjCInterfaceDecl>(EnclosingDecl)) {
       ID->setIVarList(ClsFields, RecFields.size(), Context);
       ID->setLocEnd(RBrac);
-      
+      // Add ivar's to class's DeclContext.
+      for (unsigned i = 0, e = RecFields.size(); i != e; ++i) {
+        ClsFields[i]->setLexicalDeclContext(ID);
+        ID->addDecl(Context, ClsFields[i]);
+      }
       // Must enforce the rule that ivars in the base classes may not be
       // duplicates.
       if (ID->getSuperClass()) {
@@ -4120,12 +4140,10 @@ void Sema::ActOnFields(Scope* S,
     } else if (ObjCImplementationDecl *IMPDecl = 
                   dyn_cast<ObjCImplementationDecl>(EnclosingDecl)) {
       assert(IMPDecl && "ActOnFields - missing ObjCImplementationDecl");
-      for (unsigned I = 0, N = RecFields.size(); I != N; ++I) {
-        // FIXME: Set the DeclContext correctly when we build the
-        // declarations.
+      for (unsigned I = 0, N = RecFields.size(); I != N; ++I)
+        // Ivar declared in @implementation never belongs to the implementation.
+        // Only it is in implementation's lexical context.
         ClsFields[I]->setLexicalDeclContext(IMPDecl);
-        IMPDecl->addDecl(Context, ClsFields[I]);
-      }
       CheckImplementationIvars(IMPDecl, ClsFields, RecFields.size(), RBrac);
     }
   }
@@ -4413,3 +4431,34 @@ Sema::DeclPtrTy Sema::ActOnFileScopeAsmDecl(SourceLocation Loc,
   CurContext->addDecl(Context, New);
   return DeclPtrTy::make(New);
 }
+
+void Sema::ActOnPragmaWeakID(IdentifierInfo* Name,
+                             SourceLocation PragmaLoc,
+                             SourceLocation NameLoc) {
+  Decl *PrevDecl = LookupName(TUScope, Name, LookupOrdinaryName);
+
+  // FIXME: This implementation is an ugly hack!
+  if (PrevDecl) {
+    PrevDecl->addAttr(::new (Context) WeakAttr());
+    return;
+  }
+  Diag(PragmaLoc, diag::err_unsupported_pragma_weak);
+  return;
+}
+
+void Sema::ActOnPragmaWeakAlias(IdentifierInfo* Name,
+                                IdentifierInfo* AliasName,
+                                SourceLocation PragmaLoc,
+                                SourceLocation NameLoc,
+                                SourceLocation AliasNameLoc) {
+  Decl *PrevDecl = LookupName(TUScope, Name, LookupOrdinaryName);
+
+  // FIXME: This implementation is an ugly hack!
+  if (PrevDecl) {
+    PrevDecl->addAttr(::new (Context) AliasAttr(AliasName->getName()));
+    PrevDecl->addAttr(::new (Context) WeakAttr());
+    return;
+  }
+  Diag(PragmaLoc, diag::err_unsupported_pragma_weak);
+  return;
+}
diff --git a/lib/Sema/SemaDeclCXX.cpp b/lib/Sema/SemaDeclCXX.cpp
index f13179f..b59ac87 100644
--- a/lib/Sema/SemaDeclCXX.cpp
+++ b/lib/Sema/SemaDeclCXX.cpp
@@ -1357,7 +1357,7 @@ void Sema::CheckConstructor(CXXConstructorDecl *Constructor) {
   if (!Constructor->isInvalidDecl() &&
       ((Constructor->getNumParams() == 1) || 
        (Constructor->getNumParams() > 1 && 
-        Constructor->getParamDecl(1)->getDefaultArg() != 0))) {
+        Constructor->getParamDecl(1)->hasDefaultArg()))) {
     QualType ParamType = Constructor->getParamDecl(0)->getType();
     QualType ClassTy = Context.getTagDeclType(ClassDecl);
     if (Context.getCanonicalType(ParamType).getUnqualifiedType() == ClassTy) {
diff --git a/lib/Sema/SemaDeclObjC.cpp b/lib/Sema/SemaDeclObjC.cpp
index 8f58034..2500a8f 100644
--- a/lib/Sema/SemaDeclObjC.cpp
+++ b/lib/Sema/SemaDeclObjC.cpp
@@ -2113,27 +2113,6 @@ bool Sema::CheckObjCDeclScope(Decl *D) {
   return true;
 }
 
-/// Collect the instance variables declared in an Objective-C object.  Used in
-/// the creation of structures from objects using the @defs directive.
-/// FIXME: This should be consolidated with CollectObjCIvars as it is also
-/// part of the AST generation logic of @defs.
-static void CollectIvars(ObjCInterfaceDecl *Class, RecordDecl *Record,
-                         ASTContext& Ctx,
-                         llvm::SmallVectorImpl<Sema::DeclPtrTy> &ivars) {
-  if (Class->getSuperClass())
-    CollectIvars(Class->getSuperClass(), Record, Ctx, ivars);
-  
-  // For each ivar, create a fresh ObjCAtDefsFieldDecl.
-  for (ObjCInterfaceDecl::ivar_iterator I = Class->ivar_begin(),
-       E = Class->ivar_end(); I != E; ++I) {
-    ObjCIvarDecl* ID = *I;
-    Decl *FD = ObjCAtDefsFieldDecl::Create(Ctx, Record, ID->getLocation(),
-                                           ID->getIdentifier(), ID->getType(),
-                                           ID->getBitWidth());
-    ivars.push_back(Sema::DeclPtrTy::make(FD));
-  }
-}
-
 /// Called whenever @defs(ClassName) is encountered in the source.  Inserts the
 /// instance variables of ClassName into Decls.
 void Sema::ActOnDefs(Scope *S, DeclPtrTy TagD, SourceLocation DeclStart, 
@@ -2151,7 +2130,17 @@ void Sema::ActOnDefs(Scope *S, DeclPtrTy TagD, SourceLocation DeclStart,
   }
   
   // Collect the instance variables
-  CollectIvars(Class, dyn_cast<RecordDecl>(TagD.getAs<Decl>()), Context, Decls);
+  llvm::SmallVector<FieldDecl*, 32> RecFields;
+  Context.CollectObjCIvars(Class, RecFields);
+  // For each ivar, create a fresh ObjCAtDefsFieldDecl.
+  for (unsigned i = 0; i < RecFields.size(); i++) {
+    FieldDecl* ID = RecFields[i];
+    RecordDecl *Record = dyn_cast<RecordDecl>(TagD.getAs<Decl>());
+    Decl *FD = ObjCAtDefsFieldDecl::Create(Context, Record, ID->getLocation(),
+                                           ID->getIdentifier(), ID->getType(),
+                                           ID->getBitWidth());
+    Decls.push_back(Sema::DeclPtrTy::make(FD));
+  }
   
   // Introduce all of these fields into the appropriate scope.
   for (llvm::SmallVectorImpl<DeclPtrTy>::iterator D = Decls.begin();
diff --git a/lib/Sema/SemaExprCXX.cpp b/lib/Sema/SemaExprCXX.cpp
index 65018da..ed4ac55 100644
--- a/lib/Sema/SemaExprCXX.cpp
+++ b/lib/Sema/SemaExprCXX.cpp
@@ -1588,16 +1588,26 @@ Expr *Sema::RemoveOutermostTemporaryBinding(Expr *E) {
   return E;
 }
 
+Expr *Sema::MaybeCreateCXXExprWithTemporaries(Expr *SubExpr, 
+                                              bool DestroyTemps) {
+  assert(SubExpr && "sub expression can't be null!");
+  
+  if (ExprTemporaries.empty())
+    return SubExpr;
+  
+  Expr *E = CXXExprWithTemporaries::Create(Context, SubExpr,
+                                           &ExprTemporaries[0], 
+                                           ExprTemporaries.size(),
+                                           DestroyTemps);
+  ExprTemporaries.clear();
+  
+  return E;
+}
+
 Sema::OwningExprResult Sema::ActOnFinishFullExpr(ExprArg Arg) {
   Expr *FullExpr = Arg.takeAs<Expr>();
-
-  if (FullExpr && !ExprTemporaries.empty()) {
-    // Create a cleanup expr.
-    FullExpr = CXXExprWithTemporaries::Create(Context, FullExpr,
-                                              &ExprTemporaries[0], 
-                                              ExprTemporaries.size());
-    ExprTemporaries.clear();
-  }
+  if (FullExpr)
+    FullExpr = MaybeCreateCXXExprWithTemporaries(FullExpr);
 
   return Owned(FullExpr);
 }
diff --git a/lib/Sema/SemaTemplate.cpp b/lib/Sema/SemaTemplate.cpp
index 782a0d8..f9176ca 100644
--- a/lib/Sema/SemaTemplate.cpp
+++ b/lib/Sema/SemaTemplate.cpp
@@ -750,6 +750,10 @@ static void CanonicalizeTemplateArguments(const TemplateArgument *TemplateArgs,
   Canonical.reserve(NumTemplateArgs);
   for (unsigned Idx = 0; Idx < NumTemplateArgs; ++Idx) {
     switch (TemplateArgs[Idx].getKind()) {
+    case TemplateArgument::Null:
+      assert(false && "Should never see a NULL template argument here");
+      break;
+        
     case TemplateArgument::Expression:
       // FIXME: Build canonical expression (!)
       Canonical.push_back(TemplateArgs[Idx]);
@@ -765,11 +769,13 @@ static void CanonicalizeTemplateArguments(const TemplateArgument *TemplateArgs,
       Canonical.push_back(TemplateArgument(SourceLocation(),
                                            *TemplateArgs[Idx].getAsIntegral(),
                                         TemplateArgs[Idx].getIntegralType()));
+      break;
 
     case TemplateArgument::Type: {
       QualType CanonType 
         = Context.getCanonicalType(TemplateArgs[Idx].getAsType());
       Canonical.push_back(TemplateArgument(SourceLocation(), CanonType));
+      break;
     }
     }
   }
@@ -805,7 +811,7 @@ QualType Sema::CheckTemplateIdType(TemplateName Name,
 
   // Check that the template argument list is well-formed for this
   // template.
-  llvm::SmallVector<TemplateArgument, 16> ConvertedTemplateArgs;
+  TemplateArgumentListBuilder ConvertedTemplateArgs(Context);
   if (CheckTemplateArgumentList(Template, TemplateLoc, LAngleLoc, 
                                 TemplateArgs, NumTemplateArgs, RAngleLoc,
                                 ConvertedTemplateArgs))
@@ -829,15 +835,16 @@ QualType Sema::CheckTemplateIdType(TemplateName Name,
     //   template<typename T, typename U = T> struct A;
     TemplateName CanonName = Context.getCanonicalTemplateName(Name);
     CanonType = Context.getTemplateSpecializationType(CanonName, 
-                                                    &ConvertedTemplateArgs[0],
-                                                ConvertedTemplateArgs.size());
+                                    ConvertedTemplateArgs.getFlatArgumentList(),
+                                    ConvertedTemplateArgs.flatSize());
   } else if (ClassTemplateDecl *ClassTemplate 
                = dyn_cast<ClassTemplateDecl>(Template)) {
     // Find the class template specialization declaration that
     // corresponds to these arguments.
     llvm::FoldingSetNodeID ID;
-    ClassTemplateSpecializationDecl::Profile(ID, &ConvertedTemplateArgs[0],
-                                             ConvertedTemplateArgs.size());
+    ClassTemplateSpecializationDecl::Profile(ID, 
+                                    ConvertedTemplateArgs.getFlatArgumentList(),
+                                    ConvertedTemplateArgs.flatSize());
     void *InsertPos = 0;
     ClassTemplateSpecializationDecl *Decl
       = ClassTemplate->getSpecializations().FindNodeOrInsertPos(ID, InsertPos);
@@ -846,12 +853,10 @@ QualType Sema::CheckTemplateIdType(TemplateName Name,
       // specialization. Create the canonical declaration and add it to
       // the set of specializations.
       Decl = ClassTemplateSpecializationDecl::Create(Context, 
-                                           ClassTemplate->getDeclContext(),
-                                                     TemplateLoc,
-                                                     ClassTemplate,
-                                                     &ConvertedTemplateArgs[0],
-                                                  ConvertedTemplateArgs.size(),
-                                                     0);
+                                    ClassTemplate->getDeclContext(),
+                                    TemplateLoc,
+                                    ClassTemplate,
+                                    ConvertedTemplateArgs, 0);
       ClassTemplate->getSpecializations().InsertNode(Decl, InsertPos);
       Decl->setLexicalDeclContext(CurContext);
     }
@@ -949,7 +954,7 @@ bool Sema::CheckTemplateArgumentList(TemplateDecl *Template,
                                      const TemplateArgument *TemplateArgs,
                                      unsigned NumTemplateArgs,
                                      SourceLocation RAngleLoc,
-                          llvm::SmallVectorImpl<TemplateArgument> &Converted) {
+                                     TemplateArgumentListBuilder &Converted) {
   TemplateParameterList *Params = Template->getTemplateParameters();
   unsigned NumParams = Params->size();
   unsigned NumArgs = NumTemplateArgs;
@@ -998,13 +1003,13 @@ bool Sema::CheckTemplateArgumentList(TemplateDecl *Template,
         // on the previously-computed template arguments.
         if (ArgType->isDependentType()) {
           InstantiatingTemplate Inst(*this, TemplateLoc, 
-                                     Template, &Converted[0], 
-                                     Converted.size(),
+                                     Template, Converted.getFlatArgumentList(),
+                                     Converted.flatSize(),
                                      SourceRange(TemplateLoc, RAngleLoc));
 
-          TemplateArgumentList TemplateArgs(Context, &Converted[0],
-                                            Converted.size(), 
-                                            /*CopyArgs=*/false);
+          TemplateArgumentList TemplateArgs(Context, Converted,
+                                            /*CopyArgs=*/false,
+                                            /*FlattenArgs=*/false);
           ArgType = InstantiateType(ArgType, TemplateArgs,
                                     TTP->getDefaultArgumentLoc(),
                                     TTP->getDeclName());
@@ -1069,13 +1074,13 @@ bool Sema::CheckTemplateArgumentList(TemplateDecl *Template,
       if (NTTPType->isDependentType()) {
         // Instantiate the type of the non-type template parameter.
         InstantiatingTemplate Inst(*this, TemplateLoc, 
-                                   Template, &Converted[0], 
-                                   Converted.size(),
+                                   Template, Converted.getFlatArgumentList(),
+                                   Converted.flatSize(),
                                    SourceRange(TemplateLoc, RAngleLoc));
 
-        TemplateArgumentList TemplateArgs(Context, &Converted[0],
-                                          Converted.size(), 
-                                          /*CopyArgs=*/false);
+        TemplateArgumentList TemplateArgs(Context, Converted,
+                                          /*CopyArgs=*/false,
+                                          /*FlattenArgs=*/false);
         NTTPType = InstantiateType(NTTPType, TemplateArgs,
                                    NTTP->getLocation(),
                                    NTTP->getDeclName());
@@ -1092,6 +1097,10 @@ bool Sema::CheckTemplateArgumentList(TemplateDecl *Template,
       }
 
       switch (Arg.getKind()) {
+      case TemplateArgument::Null:
+        assert(false && "Should never see a NULL template argument here");
+        break;
+          
       case TemplateArgument::Expression: {
         Expr *E = Arg.getAsExpr();
         if (CheckTemplateArgument(NTTP, NTTPType, E, &Converted))
@@ -1131,6 +1140,10 @@ bool Sema::CheckTemplateArgumentList(TemplateDecl *Template,
         = cast<TemplateTemplateParmDecl>(*Param);
      
       switch (Arg.getKind()) {
+      case TemplateArgument::Null:
+        assert(false && "Should never see a NULL template argument here");
+        break;
+          
       case TemplateArgument::Expression: {
         Expr *ArgExpr = Arg.getAsExpr();
         if (ArgExpr && isa<DeclRefExpr>(ArgExpr) &&
@@ -1379,7 +1392,7 @@ Sema::CheckTemplateArgumentPointerToMember(Expr *Arg, NamedDecl *&Member) {
 /// of this argument will be added to the end of the Converted vector.
 bool Sema::CheckTemplateArgument(NonTypeTemplateParmDecl *Param,
                                  QualType InstantiatedParamType, Expr *&Arg, 
-                         llvm::SmallVectorImpl<TemplateArgument> *Converted) {
+                                 TemplateArgumentListBuilder *Converted) {
   SourceLocation StartLoc = Arg->getSourceRange().getBegin();
 
   // If either the parameter has a dependent type or the argument is
@@ -2051,7 +2064,7 @@ Sema::ActOnClassTemplateSpecialization(Scope *S, unsigned TagSpec, TagKind TK,
 
   // Check that the template argument list is well-formed for this
   // template.
-  llvm::SmallVector<TemplateArgument, 16> ConvertedTemplateArgs;
+  TemplateArgumentListBuilder ConvertedTemplateArgs(Context);
   if (CheckTemplateArgumentList(ClassTemplate, TemplateNameLoc, LAngleLoc, 
                                 &TemplateArgs[0], TemplateArgs.size(),
                                 RAngleLoc, ConvertedTemplateArgs))
@@ -2066,11 +2079,13 @@ Sema::ActOnClassTemplateSpecialization(Scope *S, unsigned TagSpec, TagKind TK,
   llvm::FoldingSetNodeID ID;
   if (isPartialSpecialization)
     // FIXME: Template parameter list matters, too
-    ClassTemplatePartialSpecializationDecl::Profile(ID, &ConvertedTemplateArgs[0],
-                                                    ConvertedTemplateArgs.size());
+    ClassTemplatePartialSpecializationDecl::Profile(ID, 
+                                    ConvertedTemplateArgs.getFlatArgumentList(),
+                                              ConvertedTemplateArgs.flatSize());
   else
-    ClassTemplateSpecializationDecl::Profile(ID, &ConvertedTemplateArgs[0],
-                                             ConvertedTemplateArgs.size());
+    ClassTemplateSpecializationDecl::Profile(ID,
+                                    ConvertedTemplateArgs.getFlatArgumentList(),
+                                             ConvertedTemplateArgs.flatSize());
   void *InsertPos = 0;
   ClassTemplateSpecializationDecl *PrevDecl = 0;
 
@@ -2111,12 +2126,11 @@ Sema::ActOnClassTemplateSpecialization(Scope *S, unsigned TagSpec, TagKind TK,
     ClassTemplatePartialSpecializationDecl *Partial 
       = ClassTemplatePartialSpecializationDecl::Create(Context, 
                                              ClassTemplate->getDeclContext(),
-                                                TemplateNameLoc,
-                                                TemplateParams,
-                                                ClassTemplate,
-                                                &ConvertedTemplateArgs[0],
-                                                ConvertedTemplateArgs.size(),
-                                                PrevPartial);
+                                                       TemplateNameLoc,
+                                                       TemplateParams,
+                                                       ClassTemplate,
+                                                       ConvertedTemplateArgs,
+                                                       PrevPartial);
 
     if (PrevPartial) {
       ClassTemplate->getPartialSpecializations().RemoveNode(PrevPartial);
@@ -2132,9 +2146,8 @@ Sema::ActOnClassTemplateSpecialization(Scope *S, unsigned TagSpec, TagKind TK,
       = ClassTemplateSpecializationDecl::Create(Context, 
                                              ClassTemplate->getDeclContext(),
                                                 TemplateNameLoc,
-                                                ClassTemplate,
-                                                &ConvertedTemplateArgs[0],
-                                                ConvertedTemplateArgs.size(),
+                                                ClassTemplate, 
+                                                ConvertedTemplateArgs,
                                                 PrevDecl);
 
     if (PrevDecl) {
@@ -2255,9 +2268,9 @@ Sema::ActOnExplicitInstantiation(Scope *S, SourceLocation TemplateLoc,
 
   // Check that the template argument list is well-formed for this
   // template.
-  llvm::SmallVector<TemplateArgument, 16> ConvertedTemplateArgs;
+  TemplateArgumentListBuilder ConvertedTemplateArgs(Context);
   if (CheckTemplateArgumentList(ClassTemplate, TemplateNameLoc, LAngleLoc, 
-                                &TemplateArgs[0], TemplateArgs.size(),
+                                TemplateArgs.data(), TemplateArgs.size(),
                                 RAngleLoc, ConvertedTemplateArgs))
     return true;
 
@@ -2268,8 +2281,9 @@ Sema::ActOnExplicitInstantiation(Scope *S, SourceLocation TemplateLoc,
   // Find the class template specialization declaration that
   // corresponds to these arguments.
   llvm::FoldingSetNodeID ID;
-  ClassTemplateSpecializationDecl::Profile(ID, &ConvertedTemplateArgs[0],
-                                           ConvertedTemplateArgs.size());
+  ClassTemplateSpecializationDecl::Profile(ID, 
+                                    ConvertedTemplateArgs.getFlatArgumentList(),
+                                           ConvertedTemplateArgs.flatSize());
   void *InsertPos = 0;
   ClassTemplateSpecializationDecl *PrevDecl
     = ClassTemplate->getSpecializations().FindNodeOrInsertPos(ID, InsertPos);
@@ -2312,9 +2326,7 @@ Sema::ActOnExplicitInstantiation(Scope *S, SourceLocation TemplateLoc,
                                              ClassTemplate->getDeclContext(),
                                                   TemplateNameLoc,
                                                   ClassTemplate,
-                                                  &ConvertedTemplateArgs[0],
-                                                  ConvertedTemplateArgs.size(),
-                                                  0);
+                                                  ConvertedTemplateArgs, 0);
       Specialization->setLexicalDeclContext(CurContext);
       CurContext->addDecl(Context, Specialization);
       return DeclPtrTy::make(Specialization);
@@ -2340,9 +2352,7 @@ Sema::ActOnExplicitInstantiation(Scope *S, SourceLocation TemplateLoc,
                                              ClassTemplate->getDeclContext(),
                                                 TemplateNameLoc,
                                                 ClassTemplate,
-                                                &ConvertedTemplateArgs[0],
-                                                ConvertedTemplateArgs.size(),
-                                                0);
+                                                ConvertedTemplateArgs, 0);
 
     ClassTemplate->getSpecializations().InsertNode(Specialization, 
                                                    InsertPos);
@@ -2357,7 +2367,7 @@ Sema::ActOnExplicitInstantiation(Scope *S, SourceLocation TemplateLoc,
   // arguments in the specialization.
   QualType WrittenTy 
     = Context.getTemplateSpecializationType(Name, 
-                                            &TemplateArgs[0],
+                                            TemplateArgs.data(),
                                             TemplateArgs.size(),
                                   Context.getTypeDeclType(Specialization));
   Specialization->setTypeAsWritten(WrittenTy);
@@ -2563,89 +2573,3 @@ Sema::CheckTypenameType(NestedNameSpecifier *NNS, const IdentifierInfo &II,
       << Name;
   return QualType();
 }
-
-// FIXME: Move to SemaTemplateDeduction.cpp
-bool 
-Sema::DeduceTemplateArguments(QualType Param, QualType Arg,
-                             llvm::SmallVectorImpl<TemplateArgument> &Deduced) {
-  // We only want to look at the canonical types, since typedefs and
-  // sugar are not part of template argument deduction.
-  Param = Context.getCanonicalType(Param);
-  Arg = Context.getCanonicalType(Arg);
-
-  // If the parameter type is not dependent, just compare the types
-  // directly.
-  if (!Param->isDependentType())
-    return Param == Arg;
-
-  // FIXME: Use a visitor or switch to handle all of the kinds of
-  // types that the parameter may be.
-  if (const TemplateTypeParmType *TemplateTypeParm 
-        = Param->getAsTemplateTypeParmType()) {
-    (void)TemplateTypeParm; // FIXME: use this
-    // The argument type can not be less qualified than the parameter
-    // type.
-    if (Param.isMoreQualifiedThan(Arg))
-      return false;
-
-    unsigned Quals = Arg.getCVRQualifiers() & ~Param.getCVRQualifiers();
-    QualType DeducedType = Arg.getQualifiedType(Quals);
-    // FIXME: actually save the deduced type, and check that this
-    // deduction is consistent.
-    return true;
-  }
-
-  if (Param.getCVRQualifiers() != Arg.getCVRQualifiers())
-    return false;
-
-  if (const PointerType *PointerParam = Param->getAsPointerType()) {
-    const PointerType *PointerArg = Arg->getAsPointerType();
-    if (!PointerArg)
-      return false;
-
-    return DeduceTemplateArguments(PointerParam->getPointeeType(),
-                                   PointerArg->getPointeeType(),
-                                   Deduced);
-  }
-
-  // FIXME: Many more cases to go (to go).
-  return false;
-}
-
-bool
-Sema::DeduceTemplateArguments(const TemplateArgument &Param,
-                              const TemplateArgument &Arg,
-                             llvm::SmallVectorImpl<TemplateArgument> &Deduced) {
-  assert(Param.getKind() == Arg.getKind() &&
-         "Template argument kind mismatch during deduction");
-  switch (Param.getKind()) {
-  case TemplateArgument::Type: 
-    return DeduceTemplateArguments(Param.getAsType(), Arg.getAsType(),
-                                   Deduced);
-
-  default:
-    return false;
-  }
-}
-
-bool 
-Sema::DeduceTemplateArguments(const TemplateArgumentList &ParamList,
-                              const TemplateArgumentList &ArgList,
-                              llvm::SmallVectorImpl<TemplateArgument> &Deduced) {
-  assert(ParamList.size() == ArgList.size());
-  for (unsigned I = 0, N = ParamList.size(); I != N; ++I) {
-    if (!DeduceTemplateArguments(ParamList[I], ArgList[I], Deduced))
-      return false;
-  }
-  return true;
-}
-
-
-bool 
-Sema::DeduceTemplateArguments(ClassTemplatePartialSpecializationDecl *Partial,
-                              const TemplateArgumentList &TemplateArgs) {
-  llvm::SmallVector<TemplateArgument, 4> Deduced;
-  Deduced.resize(Partial->getTemplateParameters()->size());
-  return DeduceTemplateArguments(Partial->getTemplateArgs(), TemplateArgs, 
-                                 Deduced);
-}
diff --git a/lib/Sema/SemaTemplateDeduction.cpp b/lib/Sema/SemaTemplateDeduction.cpp
new file mode 100644
index 0000000..812b319
--- /dev/null
+++ b/lib/Sema/SemaTemplateDeduction.cpp
@@ -0,0 +1,395 @@
+//===------- SemaTemplateDeduction.cpp - Template Argument Deduction ------===/
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//===----------------------------------------------------------------------===/
+//
+//  This file implements C++ template argument deduction.
+//
+//===----------------------------------------------------------------------===/
+
+#include "Sema.h"
+#include "clang/AST/ASTContext.h"
+#include "clang/AST/DeclTemplate.h"
+#include "clang/AST/StmtVisitor.h"
+#include "clang/AST/Expr.h"
+#include "clang/AST/ExprCXX.h"
+#include "clang/Parse/DeclSpec.h"
+#include "llvm/Support/Compiler.h"
+using namespace clang;
+
+/// \brief If the given expression is of a form that permits the deduction
+/// of a non-type template parameter, return the declaration of that
+/// non-type template parameter.
+static NonTypeTemplateParmDecl *getDeducedParameterFromExpr(Expr *E) {
+  if (ImplicitCastExpr *IC = dyn_cast<ImplicitCastExpr>(E))
+    E = IC->getSubExpr();
+  
+  if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E))
+    return dyn_cast<NonTypeTemplateParmDecl>(DRE->getDecl());
+  
+  return 0;
+}
+
+/// \brief Deduce the value of the given non-type template parameter 
+/// from the given constant.
+///
+/// \returns true if deduction succeeded, false otherwise.
+static bool DeduceNonTypeTemplateArgument(ASTContext &Context, 
+                                          NonTypeTemplateParmDecl *NTTP, 
+                                          llvm::APInt Value,
+                             llvm::SmallVectorImpl<TemplateArgument> &Deduced) {
+  assert(NTTP->getDepth() == 0 && 
+         "Cannot deduce non-type template argument with depth > 0");
+  
+  if (Deduced[NTTP->getIndex()].isNull()) {
+    Deduced[NTTP->getIndex()] = TemplateArgument(SourceLocation(), 
+                                                 llvm::APSInt(Value),
+                                                 NTTP->getType());
+    return true;
+  }
+  
+  if (Deduced[NTTP->getIndex()].getKind() != TemplateArgument::Integral)
+    return false;
+  
+  // If the template argument was previously deduced to a negative value, 
+  // then our deduction fails.
+  const llvm::APSInt *PrevValuePtr = Deduced[NTTP->getIndex()].getAsIntegral();
+  assert(PrevValuePtr && "Not an integral template argument?");
+  if (PrevValuePtr->isSigned() && PrevValuePtr->isNegative())
+    return false;
+  
+  llvm::APInt PrevValue = *PrevValuePtr;
+  if (Value.getBitWidth() > PrevValue.getBitWidth())
+    PrevValue.zext(Value.getBitWidth());
+  else if (Value.getBitWidth() < PrevValue.getBitWidth())
+    Value.zext(PrevValue.getBitWidth());
+  return Value == PrevValue;
+}
+
+/// \brief Deduce the value of the given non-type template parameter 
+/// from the given type- or value-dependent expression.
+///
+/// \returns true if deduction succeeded, false otherwise.
+
+static bool DeduceNonTypeTemplateArgument(ASTContext &Context, 
+                                          NonTypeTemplateParmDecl *NTTP,
+                                          Expr *Value,
+                            llvm::SmallVectorImpl<TemplateArgument> &Deduced) {
+  assert(NTTP->getDepth() == 0 && 
+         "Cannot deduce non-type template argument with depth > 0");
+  assert((Value->isTypeDependent() || Value->isValueDependent()) &&
+         "Expression template argument must be type- or value-dependent.");
+  
+  if (Deduced[NTTP->getIndex()].isNull()) {
+    // FIXME: Clone the Value?
+    Deduced[NTTP->getIndex()] = TemplateArgument(Value);
+    return true;
+  }
+  
+  if (Deduced[NTTP->getIndex()].getKind() == TemplateArgument::Integral) {
+    // Okay, we deduced a constant in one case and a dependent expression 
+    // in another case. FIXME: Later, we will check that instantiating the 
+    // dependent expression gives us the constant value.
+    return true;
+  }
+  
+  // FIXME: Compare the expressions for equality!
+  return true;
+}
+
+static bool DeduceTemplateArguments(ASTContext &Context, QualType Param, 
+                                    QualType Arg,
+                             llvm::SmallVectorImpl<TemplateArgument> &Deduced) {
+  // We only want to look at the canonical types, since typedefs and
+  // sugar are not part of template argument deduction.
+  Param = Context.getCanonicalType(Param);
+  Arg = Context.getCanonicalType(Arg);
+
+  // If the parameter type is not dependent, just compare the types
+  // directly.
+  if (!Param->isDependentType())
+    return Param == Arg;
+
+  // C++ [temp.deduct.type]p9:
+  //
+  //   A template type argument T, a template template argument TT or a 
+  //   template non-type argument i can be deduced if P and A have one of 
+  //   the following forms:
+  //
+  //     T
+  //     cv-list T
+  if (const TemplateTypeParmType *TemplateTypeParm 
+        = Param->getAsTemplateTypeParmType()) {
+    // The argument type can not be less qualified than the parameter
+    // type.
+    if (Param.isMoreQualifiedThan(Arg))
+      return false;
+
+    assert(TemplateTypeParm->getDepth() == 0 && "Can't deduce with depth > 0");
+	  
+    unsigned Quals = Arg.getCVRQualifiers() & ~Param.getCVRQualifiers();
+    QualType DeducedType = Arg.getQualifiedType(Quals);
+	  unsigned Index = TemplateTypeParm->getIndex();
+
+    if (Deduced[Index].isNull())
+      Deduced[Index] = TemplateArgument(SourceLocation(), DeducedType);
+    else {
+      // C++ [temp.deduct.type]p2: 
+      //   [...] If type deduction cannot be done for any P/A pair, or if for
+      //   any pair the deduction leads to more than one possible set of 
+      //   deduced values, or if different pairs yield different deduced 
+      //   values, or if any template argument remains neither deduced nor 
+      //   explicitly specified, template argument deduction fails.
+      if (Deduced[Index].getAsType() != DeducedType)
+        return false;
+    }
+    return true;
+  }
+
+  if (Param.getCVRQualifiers() != Arg.getCVRQualifiers())
+    return false;
+
+  switch (Param->getTypeClass()) {
+    // No deduction possible for these types
+    case Type::Builtin:
+      return false;
+      
+      
+    //     T *
+    case Type::Pointer: {
+      const PointerType *PointerArg = Arg->getAsPointerType();
+      if (!PointerArg)
+        return false;
+      
+      return DeduceTemplateArguments(Context,
+                                   cast<PointerType>(Param)->getPointeeType(),
+                                     PointerArg->getPointeeType(),
+                                     Deduced);
+    }
+      
+    //     T &
+    case Type::LValueReference: {
+      const LValueReferenceType *ReferenceArg = Arg->getAsLValueReferenceType();
+      if (!ReferenceArg)
+        return false;
+      
+      return DeduceTemplateArguments(Context,
+                           cast<LValueReferenceType>(Param)->getPointeeType(),
+                                     ReferenceArg->getPointeeType(),
+                                     Deduced);
+    }
+
+    //     T && [C++0x]
+    case Type::RValueReference: {
+      const RValueReferenceType *ReferenceArg = Arg->getAsRValueReferenceType();
+      if (!ReferenceArg)
+        return false;
+      
+      return DeduceTemplateArguments(Context,
+                           cast<RValueReferenceType>(Param)->getPointeeType(),
+                                     ReferenceArg->getPointeeType(),
+                                     Deduced);
+    }
+      
+    //     T [] (implied, but not stated explicitly)
+    case Type::IncompleteArray: {
+      const IncompleteArrayType *IncompleteArrayArg = 
+        Context.getAsIncompleteArrayType(Arg);
+      if (!IncompleteArrayArg)
+        return false;
+      
+      return DeduceTemplateArguments(Context,
+                     Context.getAsIncompleteArrayType(Param)->getElementType(),
+                                     IncompleteArrayArg->getElementType(),
+                                     Deduced);
+    }
+
+    //     T [integer-constant]
+    case Type::ConstantArray: {
+      const ConstantArrayType *ConstantArrayArg = 
+        Context.getAsConstantArrayType(Arg);
+      if (!ConstantArrayArg)
+        return false;
+      
+      const ConstantArrayType *ConstantArrayParm = 
+        Context.getAsConstantArrayType(Param);
+      if (ConstantArrayArg->getSize() != ConstantArrayParm->getSize())
+        return false;
+      
+      return DeduceTemplateArguments(Context,
+                                     ConstantArrayParm->getElementType(),
+                                     ConstantArrayArg->getElementType(),
+                                     Deduced);
+    }
+
+    //     type [i]
+    case Type::DependentSizedArray: {
+      const ArrayType *ArrayArg = dyn_cast<ArrayType>(Arg);
+      if (!ArrayArg)
+        return false;
+      
+      // Check the element type of the arrays
+      const DependentSizedArrayType *DependentArrayParm
+        = cast<DependentSizedArrayType>(Param);
+      if (!DeduceTemplateArguments(Context,
+                                   DependentArrayParm->getElementType(),
+                                   ArrayArg->getElementType(),
+                                   Deduced))
+        return false;
+          
+      // Determine the array bound is something we can deduce.
+      NonTypeTemplateParmDecl *NTTP 
+        = getDeducedParameterFromExpr(DependentArrayParm->getSizeExpr());
+      if (!NTTP)
+        return true;
+      
+      // We can perform template argument deduction for the given non-type 
+      // template parameter.
+      assert(NTTP->getDepth() == 0 && 
+             "Cannot deduce non-type template argument at depth > 0");
+      if (const ConstantArrayType *ConstantArrayArg 
+            = dyn_cast<ConstantArrayType>(ArrayArg))
+        return DeduceNonTypeTemplateArgument(Context, NTTP, 
+                                             ConstantArrayArg->getSize(),
+                                             Deduced);
+      if (const DependentSizedArrayType *DependentArrayArg
+            = dyn_cast<DependentSizedArrayType>(ArrayArg))
+        return DeduceNonTypeTemplateArgument(Context, NTTP,
+                                             DependentArrayArg->getSizeExpr(),
+                                             Deduced);
+      
+      // Incomplete type does not match a dependently-sized array type
+      return false;
+    }
+      
+    default:
+      break;
+  }
+
+  // FIXME: Many more cases to go (to go).
+  return false;
+}
+
+static bool
+DeduceTemplateArguments(ASTContext &Context, const TemplateArgument &Param,
+                        const TemplateArgument &Arg,
+                        llvm::SmallVectorImpl<TemplateArgument> &Deduced) {
+  switch (Param.getKind()) {
+  case TemplateArgument::Null:
+    assert(false && "Null template argument in parameter list");
+    break;
+      
+  case TemplateArgument::Type: 
+    assert(Arg.getKind() == TemplateArgument::Type && "Type/value mismatch");
+    return DeduceTemplateArguments(Context, Param.getAsType(), 
+                                   Arg.getAsType(), Deduced);
+
+  case TemplateArgument::Declaration:
+    // FIXME: Implement this check
+    assert(false && "Unimplemented template argument deduction case");
+    return false;
+      
+  case TemplateArgument::Integral:
+    if (Arg.getKind() == TemplateArgument::Integral) {
+      // FIXME: Zero extension + sign checking here?
+      return *Param.getAsIntegral() == *Arg.getAsIntegral();
+    }
+    if (Arg.getKind() == TemplateArgument::Expression)
+      return false;
+
+    assert(false && "Type/value mismatch");
+    return false;
+      
+  case TemplateArgument::Expression: {
+    if (NonTypeTemplateParmDecl *NTTP 
+          = getDeducedParameterFromExpr(Param.getAsExpr())) {
+      if (Arg.getKind() == TemplateArgument::Integral)
+        // FIXME: Sign problems here
+        return DeduceNonTypeTemplateArgument(Context, NTTP, 
+                                             *Arg.getAsIntegral(), Deduced);
+      if (Arg.getKind() == TemplateArgument::Expression)
+        return DeduceNonTypeTemplateArgument(Context, NTTP, Arg.getAsExpr(),
+                                             Deduced);
+      
+      assert(false && "Type/value mismatch");
+      return false;
+    }
+    
+    // Can't deduce anything, but that's okay.
+    return true;
+  }
+  }
+      
+  return true;
+}
+
+static bool 
+DeduceTemplateArguments(ASTContext &Context,
+                        const TemplateArgumentList &ParamList,
+                        const TemplateArgumentList &ArgList,
+                        llvm::SmallVectorImpl<TemplateArgument> &Deduced) {
+  assert(ParamList.size() == ArgList.size());
+  for (unsigned I = 0, N = ParamList.size(); I != N; ++I) {
+    if (!DeduceTemplateArguments(Context, ParamList[I], ArgList[I], Deduced))
+      return false;
+  }
+  return true;
+}
+
+
+TemplateArgumentList * 
+Sema::DeduceTemplateArguments(ClassTemplatePartialSpecializationDecl *Partial,
+                              const TemplateArgumentList &TemplateArgs) {
+  // Deduce the template arguments for the partial specialization
+  llvm::SmallVector<TemplateArgument, 4> Deduced;
+  Deduced.resize(Partial->getTemplateParameters()->size());
+  if (! ::DeduceTemplateArguments(Context, Partial->getTemplateArgs(), 
+                                  TemplateArgs, Deduced))
+    return 0;
+  
+  // FIXME: Substitute the deduced template arguments into the template
+  // arguments of the class template partial specialization; the resulting
+  // template arguments should match TemplateArgs exactly.
+  
+  for (unsigned I = 0, N = Deduced.size(); I != N; ++I) {
+    TemplateArgument &Arg = Deduced[I];
+
+    // FIXME: If this template argument was not deduced, but the corresponding
+    // template parameter has a default argument, instantiate the default
+    // argument.
+    if (Arg.isNull()) // FIXME: Result->Destroy(Context);
+      return 0;
+    
+    if (Arg.getKind() == TemplateArgument::Integral) {
+      // FIXME: Instantiate the type, but we need some context!
+      const NonTypeTemplateParmDecl *Parm 
+        = cast<NonTypeTemplateParmDecl>(Partial->getTemplateParameters()
+                                          ->getParam(I));
+      //      QualType T = InstantiateType(Parm->getType(), *Result,
+      //                                   Parm->getLocation(), Parm->getDeclName());
+      //      if (T.isNull()) // FIXME: Result->Destroy(Context);
+      //        return 0;
+      QualType T = Parm->getType();
+      
+      // FIXME: Make sure we didn't overflow our data type!
+      llvm::APSInt &Value = *Arg.getAsIntegral();
+      unsigned AllowedBits = Context.getTypeSize(T);
+      if (Value.getBitWidth() != AllowedBits)
+        Value.extOrTrunc(AllowedBits);
+      Value.setIsSigned(T->isSignedIntegerType());
+      Arg.setIntegralType(T);
+    }
+  }
+  
+  // FIXME: This is terrible. DeduceTemplateArguments should use a 
+  // TemplateArgumentListBuilder directly.
+  TemplateArgumentListBuilder Builder(Context);
+  for (unsigned I = 0, N = Deduced.size(); I != N; ++I)
+    Builder.push_back(Deduced[I]);
+  
+  return new (Context) TemplateArgumentList(Context, Builder, /*CopyArgs=*/true,
+                                            /*FlattenArgs=*/true);
+}
diff --git a/lib/Sema/SemaTemplateInstantiate.cpp b/lib/Sema/SemaTemplateInstantiate.cpp
index d3d771b..562749e 100644
--- a/lib/Sema/SemaTemplateInstantiate.cpp
+++ b/lib/Sema/SemaTemplateInstantiate.cpp
@@ -503,6 +503,10 @@ InstantiateTemplateSpecializationType(
   for (TemplateSpecializationType::iterator Arg = T->begin(), ArgEnd = T->end();
        Arg != ArgEnd; ++Arg) {
     switch (Arg->getKind()) {
+    case TemplateArgument::Null:
+      assert(false && "Should never have a NULL template argument");
+      break;
+        
     case TemplateArgument::Type: {
       QualType T = SemaRef.InstantiateType(Arg->getAsType(), 
                                            TemplateArgs, 
@@ -829,21 +833,23 @@ Sema::InstantiateClassTemplateSpecialization(
 
   // Determine whether any class template partial specializations
   // match the given template arguments.
-  llvm::SmallVector<ClassTemplatePartialSpecializationDecl *, 4> Matched;
+  typedef std::pair<ClassTemplatePartialSpecializationDecl *,
+                    TemplateArgumentList *> MatchResult;
+  llvm::SmallVector<MatchResult, 4> Matched;
   for (llvm::FoldingSet<ClassTemplatePartialSpecializationDecl>::iterator 
          Partial = Template->getPartialSpecializations().begin(),
          PartialEnd = Template->getPartialSpecializations().end();
        Partial != PartialEnd;
        ++Partial) {
-    if (DeduceTemplateArguments(&*Partial, ClassTemplateSpec->getTemplateArgs()))
-      Matched.push_back(&*Partial);
+    if (TemplateArgumentList *Deduced 
+          = DeduceTemplateArguments(&*Partial, 
+                                    ClassTemplateSpec->getTemplateArgs()))
+      Matched.push_back(std::make_pair(&*Partial, Deduced));
   }
 
   if (Matched.size() == 1) {
-    Pattern = Matched[0];
-    // FIXME: set TemplateArgs to the template arguments of the
-    // partial specialization, instantiated with the deduced template
-    // arguments.
+    Pattern = Matched[0].first;
+    TemplateArgs = Matched[0].second;
   } else if (Matched.size() > 1) {
     // FIXME: Implement partial ordering of class template partial
     // specializations.
@@ -856,9 +862,17 @@ Sema::InstantiateClassTemplateSpecialization(
                         ExplicitInstantiation? TSK_ExplicitInstantiation 
                                              : TSK_ImplicitInstantiation);
 
-  return InstantiateClass(ClassTemplateSpec->getLocation(),
-                          ClassTemplateSpec, Pattern, *TemplateArgs,
-                          ExplicitInstantiation);
+  bool Result = InstantiateClass(ClassTemplateSpec->getLocation(),
+                                 ClassTemplateSpec, Pattern, *TemplateArgs,
+                                 ExplicitInstantiation);
+  
+  for (unsigned I = 0, N = Matched.size(); I != N; ++I) {
+    // FIXME: Implement TemplateArgumentList::Destroy!
+    //    if (Matched[I].first != Pattern)
+    //      Matched[I].second->Destroy(Context);
+  }
+  
+  return Result;
 }
 
 /// \brief Instantiate the definitions of all of the member of the
diff --git a/lib/Sema/SemaTemplateInstantiateExpr.cpp b/lib/Sema/SemaTemplateInstantiateExpr.cpp
index a6b9703..5ba42f2 100644
--- a/lib/Sema/SemaTemplateInstantiateExpr.cpp
+++ b/lib/Sema/SemaTemplateInstantiateExpr.cpp
@@ -119,12 +119,13 @@ TemplateExprInstantiator::VisitDeclRefExpr(DeclRefExpr *E) {
                                           T->isWideCharType(),
                                           T, 
                                        E->getSourceRange().getBegin()));
-    else if (T->isBooleanType())
+    if (T->isBooleanType())
       return SemaRef.Owned(new (SemaRef.Context) CXXBoolLiteralExpr(
                                           Arg.getAsIntegral()->getBoolValue(),
                                                  T, 
                                        E->getSourceRange().getBegin()));
 
+    assert(Arg.getAsIntegral()->getBitWidth() == SemaRef.Context.getIntWidth(T));
     return SemaRef.Owned(new (SemaRef.Context) IntegerLiteral(
                                                  *Arg.getAsIntegral(),
                                                  T, 
diff --git a/lib/Sema/SemaTemplateInstantiateStmt.cpp b/lib/Sema/SemaTemplateInstantiateStmt.cpp
index 1f69479..fd349df 100644
--- a/lib/Sema/SemaTemplateInstantiateStmt.cpp
+++ b/lib/Sema/SemaTemplateInstantiateStmt.cpp
@@ -194,6 +194,8 @@ Sema::OwningStmtResult TemplateStmtInstantiator::VisitIfStmt(IfStmt *S) {
   if (Cond.isInvalid())
     return SemaRef.StmtError();
 
+  Sema::FullExprArg FullCond(FullExpr(Cond));
+  
   // Instantiate the "then" branch.
   OwningStmtResult Then = SemaRef.InstantiateStmt(S->getThen(), TemplateArgs);
   if (Then.isInvalid())
@@ -204,7 +206,7 @@ Sema::OwningStmtResult TemplateStmtInstantiator::VisitIfStmt(IfStmt *S) {
   if (Else.isInvalid())
     return SemaRef.StmtError();
 
-  return SemaRef.ActOnIfStmt(S->getIfLoc(), FullExpr(Cond), move(Then),
+  return SemaRef.ActOnIfStmt(S->getIfLoc(), FullCond, move(Then),
                              S->getElseLoc(), move(Else));
 }
 
@@ -236,12 +238,14 @@ Sema::OwningStmtResult TemplateStmtInstantiator::VisitWhileStmt(WhileStmt *S) {
   if (Cond.isInvalid())
     return SemaRef.StmtError();
 
+  Sema::FullExprArg FullCond(FullExpr(Cond));
+  
   // Instantiate the body
   OwningStmtResult Body = SemaRef.InstantiateStmt(S->getBody(), TemplateArgs);
   if (Body.isInvalid())
     return SemaRef.StmtError();
 
-  return SemaRef.ActOnWhileStmt(S->getWhileLoc(), FullExpr(Cond), move(Body));
+  return SemaRef.ActOnWhileStmt(S->getWhileLoc(), FullCond, move(Body));
 }
 
 Sema::OwningStmtResult TemplateStmtInstantiator::VisitDoStmt(DoStmt *S) {