vendor/clang/clang-trunk-r238337

321 files changed, 38669 insertions, 16497 deletions

diff --git a/lib/ARCMigrate/ARCMT.cpp b/lib/ARCMigrate/ARCMT.cpp
index dddc886a269c..f266eaf83968 100644
--- a/lib/ARCMigrate/ARCMT.cpp
+++ b/lib/ARCMigrate/ARCMT.cpp
@@ -124,7 +124,7 @@ public:
     }
   }
 
-  virtual ~CaptureDiagnosticConsumer() {
+  ~CaptureDiagnosticConsumer() override {
     assert(!HasBegunSourceFile && "FinishCapture not called!");
   }
 
@@ -432,7 +432,7 @@ public:
   ARCMTMacroTrackerPPCallbacks(std::vector<SourceLocation> &ARCMTMacroLocs)
     : ARCMTMacroLocs(ARCMTMacroLocs) { }
 
-  void MacroExpands(const Token &MacroNameTok, const MacroDirective *MD,
+  void MacroExpands(const Token &MacroNameTok, const MacroDefinition &MD,
                     SourceRange Range, const MacroArgs *Args) override {
     if (MacroNameTok.getIdentifierInfo()->getName() == getARCMTMacroName())
       ARCMTMacroLocs.push_back(MacroNameTok.getLocation());
@@ -465,7 +465,7 @@ public:
     if (Listener)
       Listener->start(ctx);
   }
-  ~RewritesApplicator() {
+  ~RewritesApplicator() override {
     if (Listener)
       Listener->finish();
   }
diff --git a/lib/ARCMigrate/ObjCMT.cpp b/lib/ARCMigrate/ObjCMT.cpp
index 52c424c000f8..a43879c23655 100644
--- a/lib/ARCMigrate/ObjCMT.cpp
+++ b/lib/ARCMigrate/ObjCMT.cpp
@@ -48,7 +48,7 @@ class ObjCMigrateASTConsumer : public ASTConsumer {
   };
   
   void migrateDecl(Decl *D);
-  void migrateObjCInterfaceDecl(ASTContext &Ctx, ObjCContainerDecl *D);
+  void migrateObjCContainerDecl(ASTContext &Ctx, ObjCContainerDecl *D);
   void migrateProtocolConformance(ASTContext &Ctx,
                                   const ObjCImplementationDecl *ImpDecl);
   void CacheObjCNSIntegerTypedefed(const TypedefDecl *TypedefDcl);
@@ -245,17 +245,16 @@ namespace {
         (Msg->getReceiverKind() != ObjCMessageExpr::Instance &&
          Msg->getReceiverKind() != ObjCMessageExpr::SuperInstance))
       return false;
+    if (const Expr *Receiver = Msg->getInstanceReceiver())
+      if (Receiver->getType()->isObjCBuiltinType())
+        return false;
+      
     const ObjCMethodDecl *Method = Msg->getMethodDecl();
     if (!Method)
       return false;
     if (!Method->isPropertyAccessor())
       return false;
     
-    const ObjCInterfaceDecl *IFace =
-      NS.getASTContext().getObjContainingInterface(Method);
-    if (!IFace)
-      return false;
-    
     const ObjCPropertyDecl *Prop = Method->findPropertyDecl();
     if (!Prop)
       return false;
@@ -305,6 +304,10 @@ namespace {
       BegLoc = PP.getLocForEndOfToken(BegLoc);
       SourceLocation EndLoc = RHS->getLocStart();
       EndLoc = EndLoc.getLocWithOffset(-1);
+      const char *colon = PP.getSourceManager().getCharacterData(EndLoc);
+      // Add a space after '=' if there is no space between RHS and '='
+      if (colon && colon[0] == ':')
+        PropertyDotString += " ";
       SourceRange Range(BegLoc, EndLoc);
       commit.replace(Range, PropertyDotString);
       // remove '[' ']'
@@ -465,7 +468,7 @@ static void rewriteToObjCProperty(const ObjCMethodDecl *Getter,
   ASTContext &Context = NS.getASTContext();
   bool LParenAdded = false;
   std::string PropertyString = "@property ";
-  if (UseNsIosOnlyMacro && Context.Idents.get("NS_NONATOMIC_IOSONLY").hasMacroDefinition()) {
+  if (UseNsIosOnlyMacro && NS.isMacroDefined("NS_NONATOMIC_IOSONLY")) {
     PropertyString += "(NS_NONATOMIC_IOSONLY";
     LParenAdded = true;
   } else if (!Atomic) {
@@ -575,7 +578,7 @@ static bool IsCategoryNameWithDeprecatedSuffix(ObjCContainerDecl *D) {
   return false;
 }
 
-void ObjCMigrateASTConsumer::migrateObjCInterfaceDecl(ASTContext &Ctx,
+void ObjCMigrateASTConsumer::migrateObjCContainerDecl(ASTContext &Ctx,
                                                       ObjCContainerDecl *D) {
   if (D->isDeprecated() || IsCategoryNameWithDeprecatedSuffix(D))
     return;
@@ -616,7 +619,7 @@ ClassImplementsAllMethodsAndProperties(ASTContext &Ctx,
       if (Property->getPropertyImplementation() == ObjCPropertyDecl::Optional)
         continue;
       HasAtleastOneRequiredProperty = true;
-      DeclContext::lookup_const_result R = IDecl->lookup(Property->getDeclName());
+      DeclContext::lookup_result R = IDecl->lookup(Property->getDeclName());
       if (R.size() == 0) {
         // Relax the rule and look into class's implementation for a synthesize
         // or dynamic declaration. Class is implementing a property coming from
@@ -647,7 +650,7 @@ ClassImplementsAllMethodsAndProperties(ASTContext &Ctx,
         continue;
       if (MD->getImplementationControl() == ObjCMethodDecl::Optional)
         continue;
-      DeclContext::lookup_const_result R = ImpDecl->lookup(MD->getDeclName());
+      DeclContext::lookup_result R = ImpDecl->lookup(MD->getDeclName());
       if (R.size() == 0)
         return false;
       bool match = false;
@@ -768,19 +771,34 @@ static void rewriteToNSMacroDecl(ASTContext &Ctx,
                                 const TypedefDecl *TypedefDcl,
                                 const NSAPI &NS, edit::Commit &commit,
                                  bool IsNSIntegerType) {
-  QualType EnumUnderlyingT = EnumDcl->getPromotionType();
-  assert(!EnumUnderlyingT.isNull()
+  QualType DesignatedEnumType = EnumDcl->getIntegerType();
+  assert(!DesignatedEnumType.isNull()
          && "rewriteToNSMacroDecl - underlying enum type is null");
   
   PrintingPolicy Policy(Ctx.getPrintingPolicy());
-  std::string TypeString = EnumUnderlyingT.getAsString(Policy);
+  std::string TypeString = DesignatedEnumType.getAsString(Policy);
   std::string ClassString = IsNSIntegerType ? "NS_ENUM(" : "NS_OPTIONS(";
   ClassString += TypeString;
   ClassString += ", ";
   
   ClassString += TypedefDcl->getIdentifier()->getName();
   ClassString += ')';
-  SourceRange R(EnumDcl->getLocStart(), EnumDcl->getLocStart());
+  SourceLocation EndLoc;
+  if (EnumDcl->getIntegerTypeSourceInfo()) {
+    TypeSourceInfo *TSourceInfo = EnumDcl->getIntegerTypeSourceInfo();
+    TypeLoc TLoc = TSourceInfo->getTypeLoc();
+    EndLoc = TLoc.getLocEnd();
+    const char *lbrace = Ctx.getSourceManager().getCharacterData(EndLoc);
+    unsigned count = 0;
+    if (lbrace)
+      while (lbrace[count] != '{')
+        ++count;
+    if (count > 0)
+      EndLoc = EndLoc.getLocWithOffset(count-1);
+  }
+  else
+    EndLoc = EnumDcl->getLocStart();
+  SourceRange R(EnumDcl->getLocStart(), EndLoc);
   commit.replace(R, ClassString);
   // This is to remove spaces between '}' and typedef name.
   SourceLocation StartTypedefLoc = EnumDcl->getLocEnd();
@@ -902,7 +920,7 @@ bool ObjCMigrateASTConsumer::migrateNSEnumDecl(ASTContext &Ctx,
                                            const EnumDecl *EnumDcl,
                                            const TypedefDecl *TypedefDcl) {
   if (!EnumDcl->isCompleteDefinition() || EnumDcl->getIdentifier() ||
-      EnumDcl->isDeprecated() || EnumDcl->getIntegerTypeSourceInfo())
+      EnumDcl->isDeprecated())
     return false;
   if (!TypedefDcl) {
     if (NSIntegerTypedefed) {
@@ -1259,7 +1277,7 @@ void ObjCMigrateASTConsumer::migrateNsReturnsInnerPointer(ASTContext &Ctx,
 
   QualType RT = OM->getReturnType();
   if (!TypeIsInnerPointer(RT) ||
-      !Ctx.Idents.get("NS_RETURNS_INNER_POINTER").hasMacroDefinition())
+      !NSAPIObj->isMacroDefined("NS_RETURNS_INNER_POINTER"))
     return;
   
   edit::Commit commit(*Editor);
@@ -1270,9 +1288,9 @@ void ObjCMigrateASTConsumer::migrateNsReturnsInnerPointer(ASTContext &Ctx,
 void ObjCMigrateASTConsumer::migratePropertyNsReturnsInnerPointer(ASTContext &Ctx,
                                                                   ObjCPropertyDecl *P) {
   QualType T = P->getType();
-  
+
   if (!TypeIsInnerPointer(T) ||
-      !Ctx.Idents.get("NS_RETURNS_INNER_POINTER").hasMacroDefinition())
+      !NSAPIObj->isMacroDefined("NS_RETURNS_INNER_POINTER"))
     return;
   edit::Commit commit(*Editor);
   commit.insertBefore(P->getLocEnd(), " NS_RETURNS_INNER_POINTER ");
@@ -1390,7 +1408,7 @@ static bool AuditedType (QualType AT) {
 void ObjCMigrateASTConsumer::AnnotateImplicitBridging(ASTContext &Ctx) {
   if (CFFunctionIBCandidates.empty())
     return;
-  if (!Ctx.Idents.get("CF_IMPLICIT_BRIDGING_ENABLED").hasMacroDefinition()) {
+  if (!NSAPIObj->isMacroDefined("CF_IMPLICIT_BRIDGING_ENABLED")) {
     CFFunctionIBCandidates.clear();
     FileId = FileID();
     return;
@@ -1465,16 +1483,14 @@ void ObjCMigrateASTConsumer::AddCFAnnotations(ASTContext &Ctx,
     RetEffect Ret = CE.getReturnValue();
     const char *AnnotationString = nullptr;
     if (Ret.getObjKind() == RetEffect::CF) {
-      if (Ret.isOwned() &&
-          Ctx.Idents.get("CF_RETURNS_RETAINED").hasMacroDefinition())
+      if (Ret.isOwned() && NSAPIObj->isMacroDefined("CF_RETURNS_RETAINED"))
         AnnotationString = " CF_RETURNS_RETAINED";
       else if (Ret.notOwned() &&
-               Ctx.Idents.get("CF_RETURNS_NOT_RETAINED").hasMacroDefinition())
+               NSAPIObj->isMacroDefined("CF_RETURNS_NOT_RETAINED"))
         AnnotationString = " CF_RETURNS_NOT_RETAINED";
     }
     else if (Ret.getObjKind() == RetEffect::ObjC) {
-      if (Ret.isOwned() &&
-          Ctx.Idents.get("NS_RETURNS_RETAINED").hasMacroDefinition())
+      if (Ret.isOwned() && NSAPIObj->isMacroDefined("NS_RETURNS_RETAINED"))
         AnnotationString = " NS_RETURNS_RETAINED";
     }
     
@@ -1491,13 +1507,13 @@ void ObjCMigrateASTConsumer::AddCFAnnotations(ASTContext &Ctx,
     const ParmVarDecl *pd = *pi;
     ArgEffect AE = AEArgs[i];
     if (AE == DecRef && !pd->hasAttr<CFConsumedAttr>() &&
-        Ctx.Idents.get("CF_CONSUMED").hasMacroDefinition()) {
+        NSAPIObj->isMacroDefined("CF_CONSUMED")) {
       edit::Commit commit(*Editor);
       commit.insertBefore(pd->getLocation(), "CF_CONSUMED ");
       Editor->commit(commit);
     }
     else if (AE == DecRefMsg && !pd->hasAttr<NSConsumedAttr>() &&
-             Ctx.Idents.get("NS_CONSUMED").hasMacroDefinition()) {
+             NSAPIObj->isMacroDefined("NS_CONSUMED")) {
       edit::Commit commit(*Editor);
       commit.insertBefore(pd->getLocation(), "NS_CONSUMED ");
       Editor->commit(commit);
@@ -1582,11 +1598,10 @@ void ObjCMigrateASTConsumer::AddCFAnnotations(ASTContext &Ctx,
     RetEffect Ret = CE.getReturnValue();
     const char *AnnotationString = nullptr;
     if (Ret.getObjKind() == RetEffect::CF) {
-      if (Ret.isOwned() &&
-          Ctx.Idents.get("CF_RETURNS_RETAINED").hasMacroDefinition())
+      if (Ret.isOwned() && NSAPIObj->isMacroDefined("CF_RETURNS_RETAINED"))
         AnnotationString = " CF_RETURNS_RETAINED";
       else if (Ret.notOwned() &&
-               Ctx.Idents.get("CF_RETURNS_NOT_RETAINED").hasMacroDefinition())
+               NSAPIObj->isMacroDefined("CF_RETURNS_NOT_RETAINED"))
         AnnotationString = " CF_RETURNS_NOT_RETAINED";
     }
     else if (Ret.getObjKind() == RetEffect::ObjC) {
@@ -1600,8 +1615,7 @@ void ObjCMigrateASTConsumer::AddCFAnnotations(ASTContext &Ctx,
           break;
           
         default:
-          if (Ret.isOwned() &&
-              Ctx.Idents.get("NS_RETURNS_RETAINED").hasMacroDefinition())
+          if (Ret.isOwned() && NSAPIObj->isMacroDefined("NS_RETURNS_RETAINED"))
             AnnotationString = " NS_RETURNS_RETAINED";
           break;
       }
@@ -1620,7 +1634,7 @@ void ObjCMigrateASTConsumer::AddCFAnnotations(ASTContext &Ctx,
     const ParmVarDecl *pd = *pi;
     ArgEffect AE = AEArgs[i];
     if (AE == DecRef && !pd->hasAttr<CFConsumedAttr>() &&
-        Ctx.Idents.get("CF_CONSUMED").hasMacroDefinition()) {
+        NSAPIObj->isMacroDefined("CF_CONSUMED")) {
       edit::Commit commit(*Editor);
       commit.insertBefore(pd->getLocation(), "CF_CONSUMED ");
       Editor->commit(commit);
@@ -1640,12 +1654,12 @@ void ObjCMigrateASTConsumer::migrateAddMethodAnnotation(
                                   MethodDecl->hasAttr<NSReturnsRetainedAttr>() ||
                                   MethodDecl->hasAttr<NSReturnsNotRetainedAttr>() ||
                                   MethodDecl->hasAttr<NSReturnsAutoreleasedAttr>());
-  
-  if (CE.getReceiver() ==  DecRefMsg &&
+
+  if (CE.getReceiver() == DecRefMsg &&
       !MethodDecl->hasAttr<NSConsumesSelfAttr>() &&
       MethodDecl->getMethodFamily() != OMF_init &&
       MethodDecl->getMethodFamily() != OMF_release &&
-      Ctx.Idents.get("NS_CONSUMES_SELF").hasMacroDefinition()) {
+      NSAPIObj->isMacroDefined("NS_CONSUMES_SELF")) {
     edit::Commit commit(*Editor);
     commit.insertBefore(MethodDecl->getLocEnd(), " NS_CONSUMES_SELF");
     Editor->commit(commit);
@@ -1711,7 +1725,7 @@ void ObjCMigrateASTConsumer::inferDesignatedInitializers(
   const ObjCInterfaceDecl *IFace = ImplD->getClassInterface();
   if (!IFace || IFace->hasDesignatedInitializers())
     return;
-  if (!Ctx.Idents.get("NS_DESIGNATED_INITIALIZER").hasMacroDefinition())
+  if (!NSAPIObj->isMacroDefined("NS_DESIGNATED_INITIALIZER"))
     return;
 
   for (const auto *MD : ImplD->instance_methods()) {
@@ -1772,9 +1786,7 @@ public:
     : SourceMgr(SM), OS(OS) {
     OS << "[\n";
   }
-  ~JSONEditWriter() {
-    OS << "]\n";
-  }
+  ~JSONEditWriter() override { OS << "]\n"; }
 
 private:
   struct EntryWriter {
@@ -1858,13 +1870,16 @@ void ObjCMigrateASTConsumer::HandleTranslationUnit(ASTContext &Ctx) {
       
       if (ObjCInterfaceDecl *CDecl = dyn_cast<ObjCInterfaceDecl>(*D))
         if (canModify(CDecl))
-          migrateObjCInterfaceDecl(Ctx, CDecl);
+          migrateObjCContainerDecl(Ctx, CDecl);
       if (ObjCCategoryDecl *CatDecl = dyn_cast<ObjCCategoryDecl>(*D)) {
         if (canModify(CatDecl))
-          migrateObjCInterfaceDecl(Ctx, CatDecl);
+          migrateObjCContainerDecl(Ctx, CatDecl);
       }
-      else if (ObjCProtocolDecl *PDecl = dyn_cast<ObjCProtocolDecl>(*D))
+      else if (ObjCProtocolDecl *PDecl = dyn_cast<ObjCProtocolDecl>(*D)) {
         ObjCProtocolDecls.insert(PDecl->getCanonicalDecl());
+        if (canModify(PDecl))
+          migrateObjCContainerDecl(Ctx, PDecl);
+      }
       else if (const ObjCImplementationDecl *ImpDecl =
                dyn_cast<ObjCImplementationDecl>(*D)) {
         if ((ASTMigrateActions & FrontendOptions::ObjCMT_ProtocolConformance) &&
diff --git a/lib/ARCMigrate/PlistReporter.cpp b/lib/ARCMigrate/PlistReporter.cpp
index 53398b27af49..9a51690c0ce7 100644
--- a/lib/ARCMigrate/PlistReporter.cpp
+++ b/lib/ARCMigrate/PlistReporter.cpp
@@ -100,16 +100,18 @@ void arcmt::writeARCDiagsToPlist(const std::string &outPath,
 
     // Output the location of the bug.
     o << "  <key>location</key>\n";
-    EmitLocation(o, SM, LangOpts, D.getLocation(), FM, 2);
+    EmitLocation(o, SM, D.getLocation(), FM, 2);
 
     // Output the ranges (if any).
-    StoredDiagnostic::range_iterator RI = D.range_begin(), RE = D.range_end();
-
-    if (RI != RE) {
+    if (!D.getRanges().empty()) {
       o << "   <key>ranges</key>\n";
       o << "   <array>\n";
-      for (; RI != RE; ++RI)
-        EmitRange(o, SM, LangOpts, *RI, FM, 4);
+      for (auto &R : D.getRanges()) {
+        CharSourceRange ExpansionRange(SM.getExpansionRange(R.getAsRange()),
+                                       R.isTokenRange());
+        EmitRange(o, SM, Lexer::getAsCharRange(ExpansionRange, SM, LangOpts),
+                  FM, 4);
+      }
       o << "   </array>\n";
     }
 
diff --git a/lib/ARCMigrate/TransAPIUses.cpp b/lib/ARCMigrate/TransAPIUses.cpp
index 544cb0addfcd..40c8a070f813 100644
--- a/lib/ARCMigrate/TransAPIUses.cpp
+++ b/lib/ARCMigrate/TransAPIUses.cpp
@@ -95,7 +95,7 @@ public:
       Pass.TA.clearDiagnostic(diag::err_unavailable,
                               diag::err_unavailable_message,
                               E->getSelectorLoc(0));
-      Pass.TA.replace(E->getSourceRange(), getNilString(Pass.Ctx));
+      Pass.TA.replace(E->getSourceRange(), getNilString(Pass));
     }
     return true;
   }
diff --git a/lib/ARCMigrate/TransRetainReleaseDealloc.cpp b/lib/ARCMigrate/TransRetainReleaseDealloc.cpp
index bcbc9e9612ba..7db1a1c378cc 100644
--- a/lib/ARCMigrate/TransRetainReleaseDealloc.cpp
+++ b/lib/ARCMigrate/TransRetainReleaseDealloc.cpp
@@ -145,7 +145,7 @@ public:
       // when an exception is thrown.
       Pass.TA.replace(RecContainer->getSourceRange(), RecRange);
       std::string str = " = ";
-      str += getNilString(Pass.Ctx);
+      str += getNilString(Pass);
       Pass.TA.insertAfterToken(RecRange.getEnd(), str);
       return true;
     }
diff --git a/lib/ARCMigrate/TransUnusedInitDelegate.cpp b/lib/ARCMigrate/TransUnusedInitDelegate.cpp
index 98571c035dd5..70370ecc4e90 100644
--- a/lib/ARCMigrate/TransUnusedInitDelegate.cpp
+++ b/lib/ARCMigrate/TransUnusedInitDelegate.cpp
@@ -58,7 +58,7 @@ public:
       SourceRange ExprRange = ME->getSourceRange();
       Pass.TA.insert(ExprRange.getBegin(), "if (!(self = ");
       std::string retStr = ")) return ";
-      retStr += getNilString(Pass.Ctx);
+      retStr += getNilString(Pass);
       Pass.TA.insertAfterToken(ExprRange.getEnd(), retStr);
     }
     return true;
diff --git a/lib/ARCMigrate/Transforms.cpp b/lib/ARCMigrate/Transforms.cpp
index 6ff7b6b9db87..56d3af7233bf 100644
--- a/lib/ARCMigrate/Transforms.cpp
+++ b/lib/ARCMigrate/Transforms.cpp
@@ -16,6 +16,7 @@
 #include "clang/Basic/SourceManager.h"
 #include "clang/Basic/TargetInfo.h"
 #include "clang/Lex/Lexer.h"
+#include "clang/Lex/Preprocessor.h"
 #include "clang/Sema/Sema.h"
 #include "clang/Sema/SemaDiagnostic.h"
 #include "llvm/ADT/DenseSet.h"
@@ -212,11 +213,8 @@ bool trans::isGlobalVar(Expr *E) {
   return false;  
 }
 
-StringRef trans::getNilString(ASTContext &Ctx) {
-  if (Ctx.Idents.get("nil").hasMacroDefinition())
-    return "nil";
-  else
-    return "0";
+StringRef trans::getNilString(MigrationPass &Pass) {
+  return Pass.SemaRef.PP.isMacroDefined("nil") ? "nil" : "0";
 }
 
 namespace {
diff --git a/lib/ARCMigrate/Transforms.h b/lib/ARCMigrate/Transforms.h
index 12551d261d36..7e3dd34e7607 100644
--- a/lib/ARCMigrate/Transforms.h
+++ b/lib/ARCMigrate/Transforms.h
@@ -180,7 +180,7 @@ SourceLocation findSemiAfterLocation(SourceLocation loc, ASTContext &Ctx,
 bool hasSideEffects(Expr *E, ASTContext &Ctx);
 bool isGlobalVar(Expr *E);
 /// \brief Returns "nil" or "0" if 'nil' macro is not actually defined.
-StringRef getNilString(ASTContext &Ctx);
+StringRef getNilString(MigrationPass &Pass);
 
 template <typename BODY_TRANS>
 class BodyTransform : public RecursiveASTVisitor<BodyTransform<BODY_TRANS> > {
diff --git a/lib/AST/ASTContext.cpp b/lib/AST/ASTContext.cpp
index c9fb80ceaae2..4a831d90e645 100644
--- a/lib/AST/ASTContext.cpp
+++ b/lib/AST/ASTContext.cpp
@@ -738,9 +738,9 @@ ASTContext::ASTContext(LangOptions &LOpts, SourceManager &SM,
       FILEDecl(nullptr), jmp_bufDecl(nullptr), sigjmp_bufDecl(nullptr),
       ucontext_tDecl(nullptr), BlockDescriptorType(nullptr),
       BlockDescriptorExtendedType(nullptr), cudaConfigureCallDecl(nullptr),
-      FirstLocalImport(), LastLocalImport(),
+      FirstLocalImport(), LastLocalImport(), ExternCContext(nullptr),
       SourceMgr(SM), LangOpts(LOpts),
-      SanitizerBL(new SanitizerBlacklist(LangOpts.SanitizerBlacklistFile, SM)),
+      SanitizerBL(new SanitizerBlacklist(LangOpts.SanitizerBlacklistFiles, SM)),
       AddrSpaceMap(nullptr), Target(nullptr), PrintingPolicy(LOpts),
       Idents(idents), Selectors(sels), BuiltinInfo(builtins),
       DeclarationNames(*this), ExternalSource(nullptr), Listener(nullptr),
@@ -866,6 +866,38 @@ void ASTContext::PrintStats() const {
   BumpAlloc.PrintStats();
 }
 
+void ASTContext::mergeDefinitionIntoModule(NamedDecl *ND, Module *M,
+                                           bool NotifyListeners) {
+  if (NotifyListeners)
+    if (auto *Listener = getASTMutationListener())
+      Listener->RedefinedHiddenDefinition(ND, M);
+
+  if (getLangOpts().ModulesLocalVisibility)
+    MergedDefModules[ND].push_back(M);
+  else
+    ND->setHidden(false);
+}
+
+void ASTContext::deduplicateMergedDefinitonsFor(NamedDecl *ND) {
+  auto It = MergedDefModules.find(ND);
+  if (It == MergedDefModules.end())
+    return;
+
+  auto &Merged = It->second;
+  llvm::DenseSet<Module*> Found;
+  for (Module *&M : Merged)
+    if (!Found.insert(M).second)
+      M = nullptr;
+  Merged.erase(std::remove(Merged.begin(), Merged.end(), nullptr), Merged.end());
+}
+
+ExternCContextDecl *ASTContext::getExternCContextDecl() const {
+  if (!ExternCContext)
+    ExternCContext = ExternCContextDecl::Create(*this, getTranslationUnitDecl());
+
+  return ExternCContext;
+}
+
 RecordDecl *ASTContext::buildImplicitRecord(StringRef Name,
                                             RecordDecl::TagKind TK) const {
   SourceLocation Loc;
@@ -877,6 +909,8 @@ RecordDecl *ASTContext::buildImplicitRecord(StringRef Name,
     NewDecl = RecordDecl::Create(*this, TK, getTranslationUnitDecl(), Loc, Loc,
                                  &Idents.get(Name));
   NewDecl->setImplicit();
+  NewDecl->addAttr(TypeVisibilityAttr::CreateImplicit(
+      const_cast<ASTContext &>(*this), TypeVisibilityAttr::Default));
   return NewDecl;
 }
 
@@ -1326,7 +1360,7 @@ CharUnits ASTContext::getDeclAlign(const Decl *D, bool ForAlignof) const {
       }
       Align = std::max(Align, getPreferredTypeAlign(T.getTypePtr()));
       if (const VarDecl *VD = dyn_cast<VarDecl>(D)) {
-        if (VD->hasGlobalStorage())
+        if (VD->hasGlobalStorage() && !ForAlignof)
           Align = std::max(Align, getTargetInfo().getMinGlobalAlign());
       }
     }
@@ -1669,13 +1703,23 @@ TypeInfo ASTContext::getTypeInfoImpl(const Type *T) const {
       break;
     }
 
-    if (const EnumType *ET = dyn_cast<EnumType>(TT))
-      return getTypeInfo(ET->getDecl()->getIntegerType());
+    if (const EnumType *ET = dyn_cast<EnumType>(TT)) {
+      const EnumDecl *ED = ET->getDecl();
+      TypeInfo Info =
+          getTypeInfo(ED->getIntegerType()->getUnqualifiedDesugaredType());
+      if (unsigned AttrAlign = ED->getMaxAlignment()) {
+        Info.Align = AttrAlign;
+        Info.AlignIsRequired = true;
+      }
+      return Info;
+    }
 
     const RecordType *RT = cast<RecordType>(TT);
-    const ASTRecordLayout &Layout = getASTRecordLayout(RT->getDecl());
+    const RecordDecl *RD = RT->getDecl();
+    const ASTRecordLayout &Layout = getASTRecordLayout(RD);
     Width = toBits(Layout.getSize());
     Align = toBits(Layout.getAlignment());
+    AlignIsRequired = RD->hasAttr<AlignedAttr>();
     break;
   }
 
@@ -1778,13 +1822,20 @@ unsigned ASTContext::getPreferredTypeAlign(const Type *T) const {
   TypeInfo TI = getTypeInfo(T);
   unsigned ABIAlign = TI.Align;
 
+  T = T->getBaseElementTypeUnsafe();
+
+  // The preferred alignment of member pointers is that of a pointer.
+  if (T->isMemberPointerType())
+    return getPreferredTypeAlign(getPointerDiffType().getTypePtr());
+
   if (Target->getTriple().getArch() == llvm::Triple::xcore)
     return ABIAlign;  // Never overalign on XCore.
 
   // Double and long long should be naturally aligned if possible.
-  T = T->getBaseElementTypeUnsafe();
   if (const ComplexType *CT = T->getAs<ComplexType>())
     T = CT->getElementType().getTypePtr();
+  if (const EnumType *ET = T->getAs<EnumType>())
+    T = ET->getDecl()->getIntegerType().getTypePtr();
   if (T->isSpecificBuiltinType(BuiltinType::Double) ||
       T->isSpecificBuiltinType(BuiltinType::LongLong) ||
       T->isSpecificBuiltinType(BuiltinType::ULongLong))
@@ -1796,6 +1847,13 @@ unsigned ASTContext::getPreferredTypeAlign(const Type *T) const {
   return ABIAlign;
 }
 
+/// getTargetDefaultAlignForAttributeAligned - Return the default alignment
+/// for __attribute__((aligned)) on this target, to be used if no alignment
+/// value is specified.
+unsigned ASTContext::getTargetDefaultAlignForAttributeAligned(void) const {
+  return getTargetInfo().getDefaultAlignForAttributeAligned();
+}
+
 /// getAlignOfGlobalVar - Return the alignment in bits that should be given
 /// to a global variable of the specified type.
 unsigned ASTContext::getAlignOfGlobalVar(QualType T) const {
@@ -3337,7 +3395,7 @@ ASTContext::getElaboratedType(ElaboratedTypeKeyword Keyword,
     (void)CheckT;
   }
 
-  T = new (*this) ElaboratedType(Keyword, NNS, NamedType, Canon);
+  T = new (*this, TypeAlignment) ElaboratedType(Keyword, NNS, NamedType, Canon);
   Types.push_back(T);
   ElaboratedTypes.InsertNode(T, InsertPos);
   return QualType(T, 0);
@@ -3361,7 +3419,7 @@ ASTContext::getParenType(QualType InnerType) const {
     (void)CheckT;
   }
 
-  T = new (*this) ParenType(InnerType, Canon);
+  T = new (*this, TypeAlignment) ParenType(InnerType, Canon);
   Types.push_back(T);
   ParenTypes.InsertNode(T, InsertPos);
   return QualType(T, 0);
@@ -3390,7 +3448,7 @@ QualType ASTContext::getDependentNameType(ElaboratedTypeKeyword Keyword,
   if (T)
     return QualType(T, 0);
 
-  T = new (*this) DependentNameType(Keyword, NNS, Name, Canon);
+  T = new (*this, TypeAlignment) DependentNameType(Keyword, NNS, Name, Canon);
   Types.push_back(T);
   DependentNameTypes.InsertNode(T, InsertPos);
   return QualType(T, 0);
@@ -3492,7 +3550,8 @@ QualType ASTContext::getPackExpansionType(QualType Pattern,
     }
   }
 
-  T = new (*this) PackExpansionType(Pattern, Canon, NumExpansions);
+  T = new (*this, TypeAlignment)
+      PackExpansionType(Pattern, Canon, NumExpansions);
   Types.push_back(T);
   PackExpansionTypes.InsertNode(T, InsertPos);
   return QualType(T, 0);
@@ -3500,9 +3559,9 @@ QualType ASTContext::getPackExpansionType(QualType Pattern,
 
 /// CmpProtocolNames - Comparison predicate for sorting protocols
 /// alphabetically.
-static bool CmpProtocolNames(const ObjCProtocolDecl *LHS,
-                            const ObjCProtocolDecl *RHS) {
-  return LHS->getDeclName() < RHS->getDeclName();
+static int CmpProtocolNames(ObjCProtocolDecl *const *LHS,
+                            ObjCProtocolDecl *const *RHS) {
+  return DeclarationName::compare((*LHS)->getDeclName(), (*RHS)->getDeclName());
 }
 
 static bool areSortedAndUniqued(ObjCProtocolDecl * const *Protocols,
@@ -3513,7 +3572,7 @@ static bool areSortedAndUniqued(ObjCProtocolDecl * const *Protocols,
     return false;
   
   for (unsigned i = 1; i != NumProtocols; ++i)
-    if (!CmpProtocolNames(Protocols[i-1], Protocols[i]) ||
+    if (CmpProtocolNames(&Protocols[i - 1], &Protocols[i]) >= 0 ||
         Protocols[i]->getCanonicalDecl() != Protocols[i])
       return false;
   return true;
@@ -3524,7 +3583,7 @@ static void SortAndUniqueProtocols(ObjCProtocolDecl **Protocols,
   ObjCProtocolDecl **ProtocolsEnd = Protocols+NumProtocols;
 
   // Sort protocols, keyed by name.
-  std::sort(Protocols, Protocols+NumProtocols, CmpProtocolNames);
+  llvm::array_pod_sort(Protocols, ProtocolsEnd, CmpProtocolNames);
 
   // Canonicalize.
   for (unsigned I = 0, N = NumProtocols; I != N; ++I)
@@ -4325,6 +4384,19 @@ QualType ASTContext::getSignatureParameterType(QualType T) const {
   return T.getUnqualifiedType();
 }
 
+QualType ASTContext::getExceptionObjectType(QualType T) const {
+  // C++ [except.throw]p3:
+  //   A throw-expression initializes a temporary object, called the exception
+  //   object, the type of which is determined by removing any top-level
+  //   cv-qualifiers from the static type of the operand of throw and adjusting
+  //   the type from "array of T" or "function returning T" to "pointer to T"
+  //   or "pointer to function returning T", [...]
+  T = getVariableArrayDecayedType(T);
+  if (T->isArrayType() || T->isFunctionType())
+    T = getDecayedType(T);
+  return T.getUnqualifiedType();
+}
+
 /// getArrayDecayedType - Return the properly qualified result of decaying the
 /// specified array type to a pointer.  This operation is non-trivial when
 /// handling typedefs etc.  The canonical type of "T" must be an array type,
@@ -4859,7 +4931,7 @@ CharUnits ASTContext::getObjCEncodingTypeSize(QualType type) const {
 bool ASTContext::isMSStaticDataMemberInlineDefinition(const VarDecl *VD) const {
   return getLangOpts().MSVCCompat && VD->isStaticDataMember() &&
          VD->getType()->isIntegralOrEnumerationType() &&
-         !VD->getFirstDecl()->isOutOfLine() && VD->getFirstDecl()->hasInit();
+         VD->isFirstDecl() && !VD->isOutOfLine() && VD->hasInit();
 }
 
 static inline 
@@ -7544,7 +7616,7 @@ static QualType DecodeTypeFromStr(const char *&Str, const ASTContext &Context,
       break;
     case 'U':
       assert(!Signed && "Can't use both 'S' and 'U' modifiers!");
-      assert(!Unsigned && "Can't use 'S' modifier multiple times!");
+      assert(!Unsigned && "Can't use 'U' modifier multiple times!");
       Unsigned = true;
       break;
     case 'L':
@@ -7579,7 +7651,7 @@ static QualType DecodeTypeFromStr(const char *&Str, const ASTContext &Context,
     break;
   case 'h':
     assert(HowLong == 0 && !Signed && !Unsigned &&
-           "Bad modifiers used with 'f'!");
+           "Bad modifiers used with 'h'!");
     Type = Context.HalfTy;
     break;
   case 'f':
@@ -7806,6 +7878,9 @@ QualType ASTContext::GetBuiltinType(unsigned Id,
     ArgTypes.push_back(Ty);
   }
 
+  if (Id == Builtin::BI__GetExceptionInfo)
+    return QualType();
+
   assert((TypeStr[0] != '.' || TypeStr[1] == 0) &&
          "'.' should only occur at end of builtin type list!");
 
@@ -7909,7 +7984,7 @@ static GVALinkage basicGVALinkageForVariable(const ASTContext &Context,
     while (LexicalContext && !isa<FunctionDecl>(LexicalContext))
       LexicalContext = LexicalContext->getLexicalParent();
 
-    // Let the static local variable inherit it's linkage from the nearest
+    // Let the static local variable inherit its linkage from the nearest
     // enclosing function.
     if (LexicalContext)
       StaticLocalLinkage =
@@ -8042,7 +8117,9 @@ CallingConv ASTContext::getDefaultCallingConvention(bool IsVariadic,
   if (IsCXXMethod)
     return ABI->getDefaultMethodCallConv(IsVariadic);
 
-  return (LangOpts.MRTD && !IsVariadic) ? CC_X86StdCall : CC_C;
+  if (LangOpts.MRTD && !IsVariadic) return CC_X86StdCall;
+
+  return Target->getDefaultCallingConv(TargetInfo::CCMT_Unknown);
 }
 
 bool ASTContext::isNearlyEmpty(const CXXRecordDecl *RD) const {
@@ -8161,6 +8238,31 @@ MangleNumberingContext *ASTContext::createMangleNumberingContext() const {
   return ABI->createMangleNumberingContext();
 }
 
+const CXXConstructorDecl *
+ASTContext::getCopyConstructorForExceptionObject(CXXRecordDecl *RD) {
+  return ABI->getCopyConstructorForExceptionObject(
+      cast<CXXRecordDecl>(RD->getFirstDecl()));
+}
+
+void ASTContext::addCopyConstructorForExceptionObject(CXXRecordDecl *RD,
+                                                      CXXConstructorDecl *CD) {
+  return ABI->addCopyConstructorForExceptionObject(
+      cast<CXXRecordDecl>(RD->getFirstDecl()),
+      cast<CXXConstructorDecl>(CD->getFirstDecl()));
+}
+
+void ASTContext::addDefaultArgExprForConstructor(const CXXConstructorDecl *CD,
+                                                 unsigned ParmIdx, Expr *DAE) {
+  ABI->addDefaultArgExprForConstructor(
+      cast<CXXConstructorDecl>(CD->getFirstDecl()), ParmIdx, DAE);
+}
+
+Expr *ASTContext::getDefaultArgExprForConstructor(const CXXConstructorDecl *CD,
+                                                  unsigned ParmIdx) {
+  return ABI->getDefaultArgExprForConstructor(
+      cast<CXXConstructorDecl>(CD->getFirstDecl()), ParmIdx);
+}
+
 void ASTContext::setParameterIndex(const ParmVarDecl *D, unsigned int index) {
   ParamIndices[D] = index;
 }
diff --git a/lib/AST/ASTDumper.cpp b/lib/AST/ASTDumper.cpp
index ebf5e651ef9a..60cbb0601138 100644
--- a/lib/AST/ASTDumper.cpp
+++ b/lib/AST/ASTDumper.cpp
@@ -508,6 +508,8 @@ namespace  {
     void VisitCXXFunctionalCastExpr(const CXXFunctionalCastExpr *Node);
     void VisitCXXConstructExpr(const CXXConstructExpr *Node);
     void VisitCXXBindTemporaryExpr(const CXXBindTemporaryExpr *Node);
+    void VisitCXXNewExpr(const CXXNewExpr *Node);
+    void VisitCXXDeleteExpr(const CXXDeleteExpr *Node);
     void VisitMaterializeTemporaryExpr(const MaterializeTemporaryExpr *Node);
     void VisitExprWithCleanups(const ExprWithCleanups *Node);
     void VisitUnresolvedLookupExpr(const UnresolvedLookupExpr *Node);
@@ -516,6 +518,7 @@ namespace  {
       VisitExpr(Node);
       dumpDecl(Node->getLambdaClass());
     }
+    void VisitSizeOfPackExpr(const SizeOfPackExpr *Node);
 
     // ObjC
     void VisitObjCAtCatchStmt(const ObjCAtCatchStmt *Node);
@@ -974,8 +977,10 @@ void ASTDumper::dumpDecl(const Decl *D) {
     dumpSourceRange(D->getSourceRange());
     OS << ' ';
     dumpLocation(D->getLocation());
-    if (Module *M = D->getOwningModule())
+    if (Module *M = D->getImportedOwningModule())
       OS << " in " << M->getFullModuleName();
+    else if (Module *M = D->getLocalOwningModule())
+      OS << " in (local) " << M->getFullModuleName();
     if (const NamedDecl *ND = dyn_cast<NamedDecl>(D))
       if (ND->isHidden())
         OS << " hidden";
@@ -1099,10 +1104,13 @@ void ASTDumper::VisitFunctionDecl(const FunctionDecl *D) {
        E = D->getDeclsInPrototypeScope().end(); I != E; ++I)
     dumpDecl(*I);
 
-  for (FunctionDecl::param_const_iterator I = D->param_begin(),
-                                          E = D->param_end();
-       I != E; ++I)
-    dumpDecl(*I);
+  if (!D->param_begin() && D->getNumParams())
+    dumpChild([=] { OS << "<<NULL params x " << D->getNumParams() << ">>"; });
+  else
+    for (FunctionDecl::param_const_iterator I = D->param_begin(),
+                                            E = D->param_end();
+         I != E; ++I)
+      dumpDecl(*I);
 
   if (const CXXConstructorDecl *C = dyn_cast<CXXConstructorDecl>(D))
     for (CXXConstructorDecl::init_const_iterator I = C->init_begin(),
@@ -1913,6 +1921,32 @@ void ASTDumper::VisitCXXBindTemporaryExpr(const CXXBindTemporaryExpr *Node) {
   dumpCXXTemporary(Node->getTemporary());
 }
 
+void ASTDumper::VisitCXXNewExpr(const CXXNewExpr *Node) {
+  VisitExpr(Node);
+  if (Node->isGlobalNew())
+    OS << " global";
+  if (Node->isArray())
+    OS << " array";
+  if (Node->getOperatorNew()) {
+    OS << ' ';
+    dumpBareDeclRef(Node->getOperatorNew());
+  }
+  // We could dump the deallocation function used in case of error, but it's
+  // usually not that interesting.
+}
+
+void ASTDumper::VisitCXXDeleteExpr(const CXXDeleteExpr *Node) {
+  VisitExpr(Node);
+  if (Node->isGlobalDelete())
+    OS << " global";
+  if (Node->isArrayForm())
+    OS << " array";
+  if (Node->getOperatorDelete()) {
+    OS << ' ';
+    dumpBareDeclRef(Node->getOperatorDelete());
+  }
+}
+
 void
 ASTDumper::VisitMaterializeTemporaryExpr(const MaterializeTemporaryExpr *Node) {
   VisitExpr(Node);
@@ -1934,6 +1968,13 @@ void ASTDumper::dumpCXXTemporary(const CXXTemporary *Temporary) {
   OS << ")";
 }
 
+void ASTDumper::VisitSizeOfPackExpr(const SizeOfPackExpr *Node) {
+  VisitExpr(Node);
+  dumpPointer(Node->getPack());
+  dumpName(Node->getPack());
+}
+
+
 //===----------------------------------------------------------------------===//
 // Obj-C Expressions
 //===----------------------------------------------------------------------===//
@@ -2255,6 +2296,11 @@ LLVM_DUMP_METHOD void Stmt::dump(raw_ostream &OS, SourceManager &SM) const {
   P.dumpStmt(this);
 }
 
+LLVM_DUMP_METHOD void Stmt::dump(raw_ostream &OS) const {
+  ASTDumper P(OS, nullptr, nullptr);
+  P.dumpStmt(this);
+}
+
 LLVM_DUMP_METHOD void Stmt::dump() const {
   ASTDumper P(llvm::errs(), nullptr, nullptr);
   P.dumpStmt(this);
diff --git a/lib/AST/ASTImporter.cpp b/lib/AST/ASTImporter.cpp
index 2442e8ec25f1..911f1681ce61 100644
--- a/lib/AST/ASTImporter.cpp
+++ b/lib/AST/ASTImporter.cpp
@@ -67,6 +67,7 @@ namespace clang {
     // FIXME: DependentDecltypeType
     QualType VisitRecordType(const RecordType *T);
     QualType VisitEnumType(const EnumType *T);
+    QualType VisitAttributedType(const AttributedType *T);
     // FIXME: TemplateTypeParmType
     // FIXME: SubstTemplateTypeParmType
     QualType VisitTemplateSpecializationType(const TemplateSpecializationType *T);
@@ -80,7 +81,7 @@ namespace clang {
     // Importing declarations                            
     bool ImportDeclParts(NamedDecl *D, DeclContext *&DC, 
                          DeclContext *&LexicalDC, DeclarationName &Name, 
-                         SourceLocation &Loc);
+                         NamedDecl *&ToD, SourceLocation &Loc);
     void ImportDefinitionIfNeeded(Decl *FromD, Decl *ToD = nullptr);
     void ImportDeclarationNameLoc(const DeclarationNameInfo &From,
                                   DeclarationNameInfo& To);
@@ -167,7 +168,44 @@ namespace clang {
     Decl *VisitVarTemplateSpecializationDecl(VarTemplateSpecializationDecl *D);
 
     // Importing statements
+    DeclGroupRef ImportDeclGroup(DeclGroupRef DG);
+
     Stmt *VisitStmt(Stmt *S);
+    Stmt *VisitDeclStmt(DeclStmt *S);
+    Stmt *VisitNullStmt(NullStmt *S);
+    Stmt *VisitCompoundStmt(CompoundStmt *S);
+    Stmt *VisitCaseStmt(CaseStmt *S);
+    Stmt *VisitDefaultStmt(DefaultStmt *S);
+    Stmt *VisitLabelStmt(LabelStmt *S);
+    Stmt *VisitAttributedStmt(AttributedStmt *S);
+    Stmt *VisitIfStmt(IfStmt *S);
+    Stmt *VisitSwitchStmt(SwitchStmt *S);
+    Stmt *VisitWhileStmt(WhileStmt *S);
+    Stmt *VisitDoStmt(DoStmt *S);
+    Stmt *VisitForStmt(ForStmt *S);
+    Stmt *VisitGotoStmt(GotoStmt *S);
+    Stmt *VisitIndirectGotoStmt(IndirectGotoStmt *S);
+    Stmt *VisitContinueStmt(ContinueStmt *S);
+    Stmt *VisitBreakStmt(BreakStmt *S);
+    Stmt *VisitReturnStmt(ReturnStmt *S);
+    // FIXME: GCCAsmStmt
+    // FIXME: MSAsmStmt
+    // FIXME: SEHExceptStmt
+    // FIXME: SEHFinallyStmt
+    // FIXME: SEHTryStmt
+    // FIXME: SEHLeaveStmt
+    // FIXME: CapturedStmt
+    Stmt *VisitCXXCatchStmt(CXXCatchStmt *S);
+    Stmt *VisitCXXTryStmt(CXXTryStmt *S);
+    Stmt *VisitCXXForRangeStmt(CXXForRangeStmt *S);
+    // FIXME: MSDependentExistsStmt
+    Stmt *VisitObjCForCollectionStmt(ObjCForCollectionStmt *S);
+    Stmt *VisitObjCAtCatchStmt(ObjCAtCatchStmt *S);
+    Stmt *VisitObjCAtFinallyStmt(ObjCAtFinallyStmt *S);
+    Stmt *VisitObjCAtTryStmt(ObjCAtTryStmt *S);
+    Stmt *VisitObjCAtSynchronizedStmt(ObjCAtSynchronizedStmt *S);
+    Stmt *VisitObjCAtThrowStmt(ObjCAtThrowStmt *S);
+    Stmt *VisitObjCAutoreleasePoolStmt(ObjCAutoreleasePoolStmt *S);
 
     // Importing expressions
     Expr *VisitExpr(Expr *E);
@@ -181,6 +219,9 @@ namespace clang {
     Expr *VisitCompoundAssignOperator(CompoundAssignOperator *E);
     Expr *VisitImplicitCastExpr(ImplicitCastExpr *E);
     Expr *VisitCStyleCastExpr(CStyleCastExpr *E);
+    Expr *VisitCXXConstructExpr(CXXConstructExpr *E);
+    Expr *VisitMemberExpr(MemberExpr *E);
+    Expr *VisitCallExpr(CallExpr *E);
   };
 }
 using namespace clang;
@@ -1724,6 +1765,27 @@ QualType ASTNodeImporter::VisitEnumType(const EnumType *T) {
   return Importer.getToContext().getTagDeclType(ToDecl);
 }
 
+QualType ASTNodeImporter::VisitAttributedType(const AttributedType *T) {
+  QualType FromModifiedType = T->getModifiedType();
+  QualType FromEquivalentType = T->getEquivalentType();
+  QualType ToModifiedType;
+  QualType ToEquivalentType;
+
+  if (!FromModifiedType.isNull()) {
+    ToModifiedType = Importer.Import(FromModifiedType);
+    if (ToModifiedType.isNull())
+      return QualType();
+  }
+  if (!FromEquivalentType.isNull()) {
+    ToEquivalentType = Importer.Import(FromEquivalentType);
+    if (ToEquivalentType.isNull())
+      return QualType();
+  }
+
+  return Importer.getToContext().getAttributedType(T->getAttrKind(),
+    ToModifiedType, ToEquivalentType);
+}
+
 QualType ASTNodeImporter::VisitTemplateSpecializationType(
                                        const TemplateSpecializationType *T) {
   TemplateName ToTemplate = Importer.Import(T->getTemplateName());
@@ -1808,6 +1870,7 @@ ASTNodeImporter::VisitObjCObjectPointerType(const ObjCObjectPointerType *T) {
 bool ASTNodeImporter::ImportDeclParts(NamedDecl *D, DeclContext *&DC, 
                                       DeclContext *&LexicalDC, 
                                       DeclarationName &Name, 
+                                      NamedDecl *&ToD,
                                       SourceLocation &Loc) {
   // Import the context of this declaration.
   DC = Importer.ImportContext(D->getDeclContext());
@@ -1828,6 +1891,7 @@ bool ASTNodeImporter::ImportDeclParts(NamedDecl *D, DeclContext *&DC,
   
   // Import the location of this declaration.
   Loc = Importer.Import(D->getLocation());
+  ToD = cast_or_null<NamedDecl>(Importer.GetAlreadyImportedOrNull(D));
   return false;
 }
 
@@ -2009,7 +2073,7 @@ bool ASTNodeImporter::ImportDefinition(RecordDecl *From, RecordDecl *To,
 
 bool ASTNodeImporter::ImportDefinition(VarDecl *From, VarDecl *To,
                                        ImportDefinitionKind Kind) {
-  if (To->getDefinition())
+  if (To->getAnyInitializer())
     return false;
 
   // FIXME: Can we really import any initializer? Alternatively, we could force
@@ -2239,8 +2303,11 @@ Decl *ASTNodeImporter::VisitNamespaceDecl(NamespaceDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   NamespaceDecl *MergeWithNamespace = nullptr;
   if (!Name) {
@@ -2307,8 +2374,11 @@ Decl *ASTNodeImporter::VisitTypedefNameDecl(TypedefNameDecl *D, bool IsAlias) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // If this typedef is not in block scope, determine whether we've
   // seen a typedef with the same name (that we can merge with) or any
@@ -2381,8 +2451,11 @@ Decl *ASTNodeImporter::VisitEnumDecl(EnumDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // Figure out what enum name we're looking for.
   unsigned IDNS = Decl::IDNS_Tag;
@@ -2466,8 +2539,11 @@ Decl *ASTNodeImporter::VisitRecordDecl(RecordDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // Figure out what structure name we're looking for.
   unsigned IDNS = Decl::IDNS_Tag;
@@ -2592,8 +2668,11 @@ Decl *ASTNodeImporter::VisitEnumConstantDecl(EnumConstantDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   QualType T = Importer.Import(D->getType());
   if (T.isNull())
@@ -2648,8 +2727,11 @@ Decl *ASTNodeImporter::VisitFunctionDecl(FunctionDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // Try to find a function in our own ("to") context with the same name, same
   // type, and in the same context as the function we're importing.
@@ -2741,10 +2823,11 @@ Decl *ASTNodeImporter::VisitFunctionDecl(FunctionDecl *D) {
   // Create the imported function.
   TypeSourceInfo *TInfo = Importer.Import(D->getTypeSourceInfo());
   FunctionDecl *ToFunction = nullptr;
+  SourceLocation InnerLocStart = Importer.Import(D->getInnerLocStart());
   if (CXXConstructorDecl *FromConstructor = dyn_cast<CXXConstructorDecl>(D)) {
     ToFunction = CXXConstructorDecl::Create(Importer.getToContext(),
                                             cast<CXXRecordDecl>(DC),
-                                            D->getInnerLocStart(),
+                                            InnerLocStart,
                                             NameInfo, T, TInfo, 
                                             FromConstructor->isExplicit(),
                                             D->isInlineSpecified(), 
@@ -2753,7 +2836,7 @@ Decl *ASTNodeImporter::VisitFunctionDecl(FunctionDecl *D) {
   } else if (isa<CXXDestructorDecl>(D)) {
     ToFunction = CXXDestructorDecl::Create(Importer.getToContext(),
                                            cast<CXXRecordDecl>(DC),
-                                           D->getInnerLocStart(),
+                                           InnerLocStart,
                                            NameInfo, T, TInfo,
                                            D->isInlineSpecified(),
                                            D->isImplicit());
@@ -2761,7 +2844,7 @@ Decl *ASTNodeImporter::VisitFunctionDecl(FunctionDecl *D) {
                                            = dyn_cast<CXXConversionDecl>(D)) {
     ToFunction = CXXConversionDecl::Create(Importer.getToContext(), 
                                            cast<CXXRecordDecl>(DC),
-                                           D->getInnerLocStart(),
+                                           InnerLocStart,
                                            NameInfo, T, TInfo,
                                            D->isInlineSpecified(),
                                            FromConversion->isExplicit(),
@@ -2770,7 +2853,7 @@ Decl *ASTNodeImporter::VisitFunctionDecl(FunctionDecl *D) {
   } else if (CXXMethodDecl *Method = dyn_cast<CXXMethodDecl>(D)) {
     ToFunction = CXXMethodDecl::Create(Importer.getToContext(), 
                                        cast<CXXRecordDecl>(DC),
-                                       D->getInnerLocStart(),
+                                       InnerLocStart,
                                        NameInfo, T, TInfo,
                                        Method->getStorageClass(),
                                        Method->isInlineSpecified(),
@@ -2778,7 +2861,7 @@ Decl *ASTNodeImporter::VisitFunctionDecl(FunctionDecl *D) {
                                        Importer.Import(D->getLocEnd()));
   } else {
     ToFunction = FunctionDecl::Create(Importer.getToContext(), DC,
-                                      D->getInnerLocStart(),
+                                      InnerLocStart,
                                       NameInfo, T, TInfo, D->getStorageClass(),
                                       D->isInlineSpecified(),
                                       D->hasWrittenPrototype(),
@@ -2809,6 +2892,13 @@ Decl *ASTNodeImporter::VisitFunctionDecl(FunctionDecl *D) {
     ToFunction->setType(T);
   }
 
+  // Import the body, if any.
+  if (Stmt *FromBody = D->getBody()) {
+    if (Stmt *ToBody = Importer.Import(FromBody)) {
+      ToFunction->setBody(ToBody);
+    }
+  }
+
   // FIXME: Other bits to merge?
 
   // Add this function to the lexical context.
@@ -2855,8 +2945,11 @@ Decl *ASTNodeImporter::VisitFieldDecl(FieldDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // Determine whether we've already imported this field. 
   SmallVector<NamedDecl *, 2> FoundDecls;
@@ -2911,8 +3004,11 @@ Decl *ASTNodeImporter::VisitIndirectFieldDecl(IndirectFieldDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // Determine whether we've already imported this field. 
   SmallVector<NamedDecl *, 2> FoundDecls;
@@ -2978,8 +3074,11 @@ Decl *ASTNodeImporter::VisitObjCIvarDecl(ObjCIvarDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // Determine whether we've already imported this ivar 
   SmallVector<NamedDecl *, 2> FoundDecls;
@@ -3028,8 +3127,11 @@ Decl *ASTNodeImporter::VisitVarDecl(VarDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // Try to find a variable in our own ("to") context with the same name and
   // in the same context as the variable we're importing.
@@ -3137,6 +3239,10 @@ Decl *ASTNodeImporter::VisitVarDecl(VarDecl *D) {
   Importer.Imported(D, ToVar);
   LexicalDC->addDeclInternal(ToVar);
 
+  if (!D->isFileVarDecl() &&
+      D->isUsed())
+    ToVar->setIsUsed();
+
   // Merge the initializer.
   if (ImportDefinition(D, ToVar))
     return nullptr;
@@ -3196,6 +3302,10 @@ Decl *ASTNodeImporter::VisitParmVarDecl(ParmVarDecl *D) {
                                             T, TInfo, D->getStorageClass(),
                                             /*FIXME: Default argument*/nullptr);
   ToParm->setHasInheritedDefaultArg(D->hasInheritedDefaultArg());
+
+  if (D->isUsed())
+    ToParm->setIsUsed();
+
   return Importer.Imported(D, ToParm);
 }
 
@@ -3204,8 +3314,11 @@ Decl *ASTNodeImporter::VisitObjCMethodDecl(ObjCMethodDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   SmallVector<NamedDecl *, 2> FoundDecls;
   DC->getRedeclContext()->localUncachedLookup(Name, FoundDecls);
@@ -3315,8 +3428,11 @@ Decl *ASTNodeImporter::VisitObjCCategoryDecl(ObjCCategoryDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   ObjCInterfaceDecl *ToInterface
     = cast_or_null<ObjCInterfaceDecl>(Importer.Import(D->getClassInterface()));
@@ -3439,8 +3555,11 @@ Decl *ASTNodeImporter::VisitObjCProtocolDecl(ObjCProtocolDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   ObjCProtocolDecl *MergeWithProtocol = nullptr;
   SmallVector<NamedDecl *, 2> FoundDecls;
@@ -3614,8 +3733,11 @@ Decl *ASTNodeImporter::VisitObjCInterfaceDecl(ObjCInterfaceDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // Look for an existing interface with the same name.
   ObjCInterfaceDecl *MergeWithIface = nullptr;
@@ -3769,8 +3891,11 @@ Decl *ASTNodeImporter::VisitObjCPropertyDecl(ObjCPropertyDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // Check whether we have already imported this property.
   SmallVector<NamedDecl *, 2> FoundDecls;
@@ -4000,8 +4125,11 @@ Decl *ASTNodeImporter::VisitClassTemplateDecl(ClassTemplateDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // We may already have a template of the same name; try to find and match it.
   if (!DC->isFunctionOrMethod()) {
@@ -4188,8 +4316,11 @@ Decl *ASTNodeImporter::VisitVarTemplateDecl(VarTemplateDecl *D) {
   DeclContext *DC, *LexicalDC;
   DeclarationName Name;
   SourceLocation Loc;
-  if (ImportDeclParts(D, DC, LexicalDC, Name, Loc))
+  NamedDecl *ToD;
+  if (ImportDeclParts(D, DC, LexicalDC, Name, ToD, Loc))
     return nullptr;
+  if (ToD)
+    return ToD;
 
   // We may already have a template of the same name; try to find and match it.
   assert(!DC->isFunctionOrMethod() &&
@@ -4371,10 +4502,457 @@ Decl *ASTNodeImporter::VisitVarTemplateSpecializationDecl(
 // Import Statements
 //----------------------------------------------------------------------------
 
-Stmt *ASTNodeImporter::VisitStmt(Stmt *S) {
-  Importer.FromDiag(S->getLocStart(), diag::err_unsupported_ast_node)
-    << S->getStmtClassName();
-  return nullptr;
+DeclGroupRef ASTNodeImporter::ImportDeclGroup(DeclGroupRef DG) {
+  if (DG.isNull())
+    return DeclGroupRef::Create(Importer.getToContext(), nullptr, 0);
+  size_t NumDecls = DG.end() - DG.begin();
+  SmallVector<Decl *, 1> ToDecls(NumDecls);
+  auto &_Importer = this->Importer;
+  std::transform(DG.begin(), DG.end(), ToDecls.begin(),
+    [&_Importer](Decl *D) -> Decl * {
+      return _Importer.Import(D);
+    });
+  return DeclGroupRef::Create(Importer.getToContext(),
+                              ToDecls.begin(),
+                              NumDecls);
+}
+
+ Stmt *ASTNodeImporter::VisitStmt(Stmt *S) {
+   Importer.FromDiag(S->getLocStart(), diag::err_unsupported_ast_node)
+     << S->getStmtClassName();
+   return nullptr;
+ }
+ 
+Stmt *ASTNodeImporter::VisitDeclStmt(DeclStmt *S) {
+  DeclGroupRef ToDG = ImportDeclGroup(S->getDeclGroup());
+  for (Decl *ToD : ToDG) {
+    if (!ToD)
+      return nullptr;
+  }
+  SourceLocation ToStartLoc = Importer.Import(S->getStartLoc());
+  SourceLocation ToEndLoc = Importer.Import(S->getEndLoc());
+  return new (Importer.getToContext()) DeclStmt(ToDG, ToStartLoc, ToEndLoc);
+}
+
+Stmt *ASTNodeImporter::VisitNullStmt(NullStmt *S) {
+  SourceLocation ToSemiLoc = Importer.Import(S->getSemiLoc());
+  return new (Importer.getToContext()) NullStmt(ToSemiLoc,
+                                                S->hasLeadingEmptyMacro());
+}
+
+Stmt *ASTNodeImporter::VisitCompoundStmt(CompoundStmt *S) {
+  SmallVector<Stmt *, 4> ToStmts(S->size());
+  auto &_Importer = this->Importer;
+  std::transform(S->body_begin(), S->body_end(), ToStmts.begin(),
+    [&_Importer](Stmt *CS) -> Stmt * {
+      return _Importer.Import(CS);
+    });
+  for (Stmt *ToS : ToStmts) {
+    if (!ToS)
+      return nullptr;
+  }
+  SourceLocation ToLBraceLoc = Importer.Import(S->getLBracLoc());
+  SourceLocation ToRBraceLoc = Importer.Import(S->getRBracLoc());
+  return new (Importer.getToContext()) CompoundStmt(Importer.getToContext(),
+                                                    ToStmts,
+                                                    ToLBraceLoc, ToRBraceLoc);
+}
+
+Stmt *ASTNodeImporter::VisitCaseStmt(CaseStmt *S) {
+  Expr *ToLHS = Importer.Import(S->getLHS());
+  if (!ToLHS)
+    return nullptr;
+  Expr *ToRHS = Importer.Import(S->getRHS());
+  if (!ToRHS && S->getRHS())
+    return nullptr;
+  SourceLocation ToCaseLoc = Importer.Import(S->getCaseLoc());
+  SourceLocation ToEllipsisLoc = Importer.Import(S->getEllipsisLoc());
+  SourceLocation ToColonLoc = Importer.Import(S->getColonLoc());
+  return new (Importer.getToContext()) CaseStmt(ToLHS, ToRHS,
+                                                ToCaseLoc, ToEllipsisLoc,
+                                                ToColonLoc);
+}
+
+Stmt *ASTNodeImporter::VisitDefaultStmt(DefaultStmt *S) {
+  SourceLocation ToDefaultLoc = Importer.Import(S->getDefaultLoc());
+  SourceLocation ToColonLoc = Importer.Import(S->getColonLoc());
+  Stmt *ToSubStmt = Importer.Import(S->getSubStmt());
+  if (!ToSubStmt && S->getSubStmt())
+    return nullptr;
+  return new (Importer.getToContext()) DefaultStmt(ToDefaultLoc, ToColonLoc,
+                                                   ToSubStmt);
+}
+
+Stmt *ASTNodeImporter::VisitLabelStmt(LabelStmt *S) {
+  SourceLocation ToIdentLoc = Importer.Import(S->getIdentLoc());
+  LabelDecl *ToLabelDecl =
+    cast_or_null<LabelDecl>(Importer.Import(S->getDecl()));
+  if (!ToLabelDecl && S->getDecl())
+    return nullptr;
+  Stmt *ToSubStmt = Importer.Import(S->getSubStmt());
+  if (!ToSubStmt && S->getSubStmt())
+    return nullptr;
+  return new (Importer.getToContext()) LabelStmt(ToIdentLoc, ToLabelDecl,
+                                                 ToSubStmt);
+}
+
+Stmt *ASTNodeImporter::VisitAttributedStmt(AttributedStmt *S) {
+  SourceLocation ToAttrLoc = Importer.Import(S->getAttrLoc());
+  ArrayRef<const Attr*> FromAttrs(S->getAttrs());
+  SmallVector<const Attr *, 1> ToAttrs(FromAttrs.size());
+  ASTContext &_ToContext = Importer.getToContext();
+  std::transform(FromAttrs.begin(), FromAttrs.end(), ToAttrs.begin(),
+    [&_ToContext](const Attr *A) -> const Attr * {
+      return A->clone(_ToContext);
+    });
+  for (const Attr *ToA : ToAttrs) {
+    if (!ToA)
+      return nullptr;
+  }
+  Stmt *ToSubStmt = Importer.Import(S->getSubStmt());
+  if (!ToSubStmt && S->getSubStmt())
+    return nullptr;
+  return AttributedStmt::Create(Importer.getToContext(), ToAttrLoc,
+                                ToAttrs, ToSubStmt);
+}
+
+Stmt *ASTNodeImporter::VisitIfStmt(IfStmt *S) {
+  SourceLocation ToIfLoc = Importer.Import(S->getIfLoc());
+  VarDecl *ToConditionVariable = nullptr;
+  if (VarDecl *FromConditionVariable = S->getConditionVariable()) {
+    ToConditionVariable =
+      dyn_cast_or_null<VarDecl>(Importer.Import(FromConditionVariable));
+    if (!ToConditionVariable)
+      return nullptr;
+  }
+  Expr *ToCondition = Importer.Import(S->getCond());
+  if (!ToCondition && S->getCond())
+    return nullptr;
+  Stmt *ToThenStmt = Importer.Import(S->getThen());
+  if (!ToThenStmt && S->getThen())
+    return nullptr;
+  SourceLocation ToElseLoc = Importer.Import(S->getElseLoc());
+  Stmt *ToElseStmt = Importer.Import(S->getElse());
+  if (!ToElseStmt && S->getElse())
+    return nullptr;
+  return new (Importer.getToContext()) IfStmt(Importer.getToContext(),
+                                              ToIfLoc, ToConditionVariable,
+                                              ToCondition, ToThenStmt,
+                                              ToElseLoc, ToElseStmt);
+}
+
+Stmt *ASTNodeImporter::VisitSwitchStmt(SwitchStmt *S) {
+  VarDecl *ToConditionVariable = nullptr;
+  if (VarDecl *FromConditionVariable = S->getConditionVariable()) {
+    ToConditionVariable =
+      dyn_cast_or_null<VarDecl>(Importer.Import(FromConditionVariable));
+    if (!ToConditionVariable)
+      return nullptr;
+  }
+  Expr *ToCondition = Importer.Import(S->getCond());
+  if (!ToCondition && S->getCond())
+    return nullptr;
+  SwitchStmt *ToStmt = new (Importer.getToContext()) SwitchStmt(
+                         Importer.getToContext(), ToConditionVariable,
+                         ToCondition);
+  Stmt *ToBody = Importer.Import(S->getBody());
+  if (!ToBody && S->getBody())
+    return nullptr;
+  ToStmt->setBody(ToBody);
+  ToStmt->setSwitchLoc(Importer.Import(S->getSwitchLoc()));
+  // Now we have to re-chain the cases.
+  SwitchCase *LastChainedSwitchCase = nullptr;
+  for (SwitchCase *SC = S->getSwitchCaseList(); SC != nullptr;
+       SC = SC->getNextSwitchCase()) {
+    SwitchCase *ToSC = dyn_cast_or_null<SwitchCase>(Importer.Import(SC));
+    if (!ToSC)
+      return nullptr;
+    if (LastChainedSwitchCase)
+      LastChainedSwitchCase->setNextSwitchCase(ToSC);
+    else
+      ToStmt->setSwitchCaseList(ToSC);
+    LastChainedSwitchCase = ToSC;
+  }
+  return ToStmt;
+}
+
+Stmt *ASTNodeImporter::VisitWhileStmt(WhileStmt *S) {
+  VarDecl *ToConditionVariable = nullptr;
+  if (VarDecl *FromConditionVariable = S->getConditionVariable()) {
+    ToConditionVariable =
+      dyn_cast_or_null<VarDecl>(Importer.Import(FromConditionVariable));
+    if (!ToConditionVariable)
+      return nullptr;
+  }
+  Expr *ToCondition = Importer.Import(S->getCond());
+  if (!ToCondition && S->getCond())
+    return nullptr;
+  Stmt *ToBody = Importer.Import(S->getBody());
+  if (!ToBody && S->getBody())
+    return nullptr;
+  SourceLocation ToWhileLoc = Importer.Import(S->getWhileLoc());
+  return new (Importer.getToContext()) WhileStmt(Importer.getToContext(),
+                                                 ToConditionVariable,
+                                                 ToCondition, ToBody,
+                                                 ToWhileLoc);
+}
+
+Stmt *ASTNodeImporter::VisitDoStmt(DoStmt *S) {
+  Stmt *ToBody = Importer.Import(S->getBody());
+  if (!ToBody && S->getBody())
+    return nullptr;
+  Expr *ToCondition = Importer.Import(S->getCond());
+  if (!ToCondition && S->getCond())
+    return nullptr;
+  SourceLocation ToDoLoc = Importer.Import(S->getDoLoc());
+  SourceLocation ToWhileLoc = Importer.Import(S->getWhileLoc());
+  SourceLocation ToRParenLoc = Importer.Import(S->getRParenLoc());
+  return new (Importer.getToContext()) DoStmt(ToBody, ToCondition,
+                                              ToDoLoc, ToWhileLoc,
+                                              ToRParenLoc);
+}
+
+Stmt *ASTNodeImporter::VisitForStmt(ForStmt *S) {
+  Stmt *ToInit = Importer.Import(S->getInit());
+  if (!ToInit && S->getInit())
+    return nullptr;
+  Expr *ToCondition = Importer.Import(S->getCond());
+  if (!ToCondition && S->getCond())
+    return nullptr;
+  VarDecl *ToConditionVariable = nullptr;
+  if (VarDecl *FromConditionVariable = S->getConditionVariable()) {
+    ToConditionVariable =
+      dyn_cast_or_null<VarDecl>(Importer.Import(FromConditionVariable));
+    if (!ToConditionVariable)
+      return nullptr;
+  }
+  Expr *ToInc = Importer.Import(S->getInc());
+  if (!ToInc && S->getInc())
+    return nullptr;
+  Stmt *ToBody = Importer.Import(S->getBody());
+  if (!ToBody && S->getBody())
+    return nullptr;
+  SourceLocation ToForLoc = Importer.Import(S->getForLoc());
+  SourceLocation ToLParenLoc = Importer.Import(S->getLParenLoc());
+  SourceLocation ToRParenLoc = Importer.Import(S->getRParenLoc());
+  return new (Importer.getToContext()) ForStmt(Importer.getToContext(),
+                                               ToInit, ToCondition,
+                                               ToConditionVariable,
+                                               ToInc, ToBody,
+                                               ToForLoc, ToLParenLoc,
+                                               ToRParenLoc);
+}
+
+Stmt *ASTNodeImporter::VisitGotoStmt(GotoStmt *S) {
+  LabelDecl *ToLabel = nullptr;
+  if (LabelDecl *FromLabel = S->getLabel()) {
+    ToLabel = dyn_cast_or_null<LabelDecl>(Importer.Import(FromLabel));
+    if (!ToLabel)
+      return nullptr;
+  }
+  SourceLocation ToGotoLoc = Importer.Import(S->getGotoLoc());
+  SourceLocation ToLabelLoc = Importer.Import(S->getLabelLoc());
+  return new (Importer.getToContext()) GotoStmt(ToLabel,
+                                                ToGotoLoc, ToLabelLoc);
+}
+
+Stmt *ASTNodeImporter::VisitIndirectGotoStmt(IndirectGotoStmt *S) {
+  SourceLocation ToGotoLoc = Importer.Import(S->getGotoLoc());
+  SourceLocation ToStarLoc = Importer.Import(S->getStarLoc());
+  Expr *ToTarget = Importer.Import(S->getTarget());
+  if (!ToTarget && S->getTarget())
+    return nullptr;
+  return new (Importer.getToContext()) IndirectGotoStmt(ToGotoLoc, ToStarLoc,
+                                                        ToTarget);
+}
+
+Stmt *ASTNodeImporter::VisitContinueStmt(ContinueStmt *S) {
+  SourceLocation ToContinueLoc = Importer.Import(S->getContinueLoc());
+  return new (Importer.getToContext()) ContinueStmt(ToContinueLoc);
+}
+
+Stmt *ASTNodeImporter::VisitBreakStmt(BreakStmt *S) {
+  SourceLocation ToBreakLoc = Importer.Import(S->getBreakLoc());
+  return new (Importer.getToContext()) BreakStmt(ToBreakLoc);
+}
+
+Stmt *ASTNodeImporter::VisitReturnStmt(ReturnStmt *S) {
+  SourceLocation ToRetLoc = Importer.Import(S->getReturnLoc());
+  Expr *ToRetExpr = Importer.Import(S->getRetValue());
+  if (!ToRetExpr && S->getRetValue())
+    return nullptr;
+  VarDecl *NRVOCandidate = const_cast<VarDecl*>(S->getNRVOCandidate());
+  VarDecl *ToNRVOCandidate = cast_or_null<VarDecl>(Importer.Import(NRVOCandidate));
+  if (!ToNRVOCandidate && NRVOCandidate)
+    return nullptr;
+  return new (Importer.getToContext()) ReturnStmt(ToRetLoc, ToRetExpr,
+                                                  ToNRVOCandidate);
+}
+
+Stmt *ASTNodeImporter::VisitCXXCatchStmt(CXXCatchStmt *S) {
+  SourceLocation ToCatchLoc = Importer.Import(S->getCatchLoc());
+  VarDecl *ToExceptionDecl = nullptr;
+  if (VarDecl *FromExceptionDecl = S->getExceptionDecl()) {
+    ToExceptionDecl =
+      dyn_cast_or_null<VarDecl>(Importer.Import(FromExceptionDecl));
+    if (!ToExceptionDecl)
+      return nullptr;
+  }
+  Stmt *ToHandlerBlock = Importer.Import(S->getHandlerBlock());
+  if (!ToHandlerBlock && S->getHandlerBlock())
+    return nullptr;
+  return new (Importer.getToContext()) CXXCatchStmt(ToCatchLoc,
+                                                    ToExceptionDecl,
+                                                    ToHandlerBlock);
+}
+
+Stmt *ASTNodeImporter::VisitCXXTryStmt(CXXTryStmt *S) {
+  SourceLocation ToTryLoc = Importer.Import(S->getTryLoc());
+  Stmt *ToTryBlock = Importer.Import(S->getTryBlock());
+  if (!ToTryBlock && S->getTryBlock())
+    return nullptr;
+  SmallVector<Stmt *, 1> ToHandlers(S->getNumHandlers());
+  for (unsigned HI = 0, HE = S->getNumHandlers(); HI != HE; ++HI) {
+    CXXCatchStmt *FromHandler = S->getHandler(HI);
+    if (Stmt *ToHandler = Importer.Import(FromHandler))
+      ToHandlers[HI] = ToHandler;
+    else
+      return nullptr;
+  }
+  return CXXTryStmt::Create(Importer.getToContext(), ToTryLoc, ToTryBlock,
+                            ToHandlers);
+}
+
+Stmt *ASTNodeImporter::VisitCXXForRangeStmt(CXXForRangeStmt *S) {
+  DeclStmt *ToRange =
+    dyn_cast_or_null<DeclStmt>(Importer.Import(S->getRangeStmt()));
+  if (!ToRange && S->getRangeStmt())
+    return nullptr;
+  DeclStmt *ToBeginEnd =
+    dyn_cast_or_null<DeclStmt>(Importer.Import(S->getBeginEndStmt()));
+  if (!ToBeginEnd && S->getBeginEndStmt())
+    return nullptr;
+  Expr *ToCond = Importer.Import(S->getCond());
+  if (!ToCond && S->getCond())
+    return nullptr;
+  Expr *ToInc = Importer.Import(S->getInc());
+  if (!ToInc && S->getInc())
+    return nullptr;
+  DeclStmt *ToLoopVar =
+    dyn_cast_or_null<DeclStmt>(Importer.Import(S->getLoopVarStmt()));
+  if (!ToLoopVar && S->getLoopVarStmt())
+    return nullptr;
+  Stmt *ToBody = Importer.Import(S->getBody());
+  if (!ToBody && S->getBody())
+    return nullptr;
+  SourceLocation ToForLoc = Importer.Import(S->getForLoc());
+  SourceLocation ToColonLoc = Importer.Import(S->getColonLoc());
+  SourceLocation ToRParenLoc = Importer.Import(S->getRParenLoc());
+  return new (Importer.getToContext()) CXXForRangeStmt(ToRange, ToBeginEnd,
+                                                       ToCond, ToInc,
+                                                       ToLoopVar, ToBody,
+                                                       ToForLoc, ToColonLoc,
+                                                       ToRParenLoc);
+}
+
+Stmt *ASTNodeImporter::VisitObjCForCollectionStmt(ObjCForCollectionStmt *S) {
+  Stmt *ToElem = Importer.Import(S->getElement());
+  if (!ToElem && S->getElement())
+    return nullptr;
+  Expr *ToCollect = Importer.Import(S->getCollection());
+  if (!ToCollect && S->getCollection())
+    return nullptr;
+  Stmt *ToBody = Importer.Import(S->getBody());
+  if (!ToBody && S->getBody())
+    return nullptr;
+  SourceLocation ToForLoc = Importer.Import(S->getForLoc());
+  SourceLocation ToRParenLoc = Importer.Import(S->getRParenLoc());
+  return new (Importer.getToContext()) ObjCForCollectionStmt(ToElem,
+                                                             ToCollect,
+                                                             ToBody, ToForLoc,
+                                                             ToRParenLoc);
+}
+
+Stmt *ASTNodeImporter::VisitObjCAtCatchStmt(ObjCAtCatchStmt *S) {
+  SourceLocation ToAtCatchLoc = Importer.Import(S->getAtCatchLoc());
+  SourceLocation ToRParenLoc = Importer.Import(S->getRParenLoc());
+  VarDecl *ToExceptionDecl = nullptr;
+  if (VarDecl *FromExceptionDecl = S->getCatchParamDecl()) {
+    ToExceptionDecl =
+      dyn_cast_or_null<VarDecl>(Importer.Import(FromExceptionDecl));
+    if (!ToExceptionDecl)
+      return nullptr;
+  }
+  Stmt *ToBody = Importer.Import(S->getCatchBody());
+  if (!ToBody && S->getCatchBody())
+    return nullptr;
+  return new (Importer.getToContext()) ObjCAtCatchStmt(ToAtCatchLoc,
+                                                       ToRParenLoc,
+                                                       ToExceptionDecl,
+                                                       ToBody);
+}
+
+Stmt *ASTNodeImporter::VisitObjCAtFinallyStmt(ObjCAtFinallyStmt *S) {
+  SourceLocation ToAtFinallyLoc = Importer.Import(S->getAtFinallyLoc());
+  Stmt *ToAtFinallyStmt = Importer.Import(S->getFinallyBody());
+  if (!ToAtFinallyStmt && S->getFinallyBody())
+    return nullptr;
+  return new (Importer.getToContext()) ObjCAtFinallyStmt(ToAtFinallyLoc,
+                                                         ToAtFinallyStmt);
+}
+
+Stmt *ASTNodeImporter::VisitObjCAtTryStmt(ObjCAtTryStmt *S) {
+  SourceLocation ToAtTryLoc = Importer.Import(S->getAtTryLoc());
+  Stmt *ToAtTryStmt = Importer.Import(S->getTryBody());
+  if (!ToAtTryStmt && S->getTryBody())
+    return nullptr;
+  SmallVector<Stmt *, 1> ToCatchStmts(S->getNumCatchStmts());
+  for (unsigned CI = 0, CE = S->getNumCatchStmts(); CI != CE; ++CI) {
+    ObjCAtCatchStmt *FromCatchStmt = S->getCatchStmt(CI);
+    if (Stmt *ToCatchStmt = Importer.Import(FromCatchStmt))
+      ToCatchStmts[CI] = ToCatchStmt;
+    else
+      return nullptr;
+  }
+  Stmt *ToAtFinallyStmt = Importer.Import(S->getFinallyStmt());
+  if (!ToAtFinallyStmt && S->getFinallyStmt())
+    return nullptr;
+  return ObjCAtTryStmt::Create(Importer.getToContext(),
+                               ToAtTryLoc, ToAtTryStmt,
+                               ToCatchStmts.begin(), ToCatchStmts.size(),
+                               ToAtFinallyStmt);
+}
+
+Stmt *ASTNodeImporter::VisitObjCAtSynchronizedStmt
+  (ObjCAtSynchronizedStmt *S) {
+  SourceLocation ToAtSynchronizedLoc =
+    Importer.Import(S->getAtSynchronizedLoc());
+  Expr *ToSynchExpr = Importer.Import(S->getSynchExpr());
+  if (!ToSynchExpr && S->getSynchExpr())
+    return nullptr;
+  Stmt *ToSynchBody = Importer.Import(S->getSynchBody());
+  if (!ToSynchBody && S->getSynchBody())
+    return nullptr;
+  return new (Importer.getToContext()) ObjCAtSynchronizedStmt(
+    ToAtSynchronizedLoc, ToSynchExpr, ToSynchBody);
+}
+
+Stmt *ASTNodeImporter::VisitObjCAtThrowStmt(ObjCAtThrowStmt *S) {
+  SourceLocation ToAtThrowLoc = Importer.Import(S->getThrowLoc());
+  Expr *ToThrow = Importer.Import(S->getThrowExpr());
+  if (!ToThrow && S->getThrowExpr())
+    return nullptr;
+  return new (Importer.getToContext()) ObjCAtThrowStmt(ToAtThrowLoc, ToThrow);
+}
+
+Stmt *ASTNodeImporter::VisitObjCAutoreleasePoolStmt
+  (ObjCAutoreleasePoolStmt *S) {
+  SourceLocation ToAtLoc = Importer.Import(S->getAtLoc());
+  Stmt *ToSubStmt = Importer.Import(S->getSubStmt());
+  if (!ToSubStmt && S->getSubStmt())
+    return nullptr;
+  return new (Importer.getToContext()) ObjCAutoreleasePoolStmt(ToAtLoc,
+                                                               ToSubStmt);
 }
 
 //----------------------------------------------------------------------------
@@ -4585,6 +5163,107 @@ Expr *ASTNodeImporter::VisitCStyleCastExpr(CStyleCastExpr *E) {
                                 Importer.Import(E->getRParenLoc()));
 }
 
+Expr *ASTNodeImporter::VisitCXXConstructExpr(CXXConstructExpr *E) {
+  QualType T = Importer.Import(E->getType());
+  if (T.isNull())
+    return nullptr;
+
+  CXXConstructorDecl *ToCCD =
+    dyn_cast<CXXConstructorDecl>(Importer.Import(E->getConstructor()));
+  if (!ToCCD && E->getConstructor())
+    return nullptr;
+
+  size_t NumArgs = E->getNumArgs();
+  SmallVector<Expr *, 1> ToArgs(NumArgs);
+  ASTImporter &_Importer = Importer;
+  std::transform(E->arg_begin(), E->arg_end(), ToArgs.begin(),
+    [&_Importer](Expr *AE) -> Expr * {
+      return _Importer.Import(AE);
+    });
+  for (Expr *ToA : ToArgs) {
+    if (!ToA)
+      return nullptr;
+  }
+
+  return CXXConstructExpr::Create(Importer.getToContext(), T,
+                                  Importer.Import(E->getLocation()),
+                                  ToCCD, E->isElidable(),
+                                  ToArgs, E->hadMultipleCandidates(),
+                                  E->isListInitialization(),
+                                  E->isStdInitListInitialization(),
+                                  E->requiresZeroInitialization(),
+                                  E->getConstructionKind(),
+                                  Importer.Import(E->getParenOrBraceRange()));
+}
+
+Expr *ASTNodeImporter::VisitMemberExpr(MemberExpr *E) {
+  QualType T = Importer.Import(E->getType());
+  if (T.isNull())
+    return nullptr;
+
+  Expr *ToBase = Importer.Import(E->getBase());
+  if (!ToBase && E->getBase())
+    return nullptr;
+
+  ValueDecl *ToMember = dyn_cast<ValueDecl>(Importer.Import(E->getMemberDecl()));
+  if (!ToMember && E->getMemberDecl())
+    return nullptr;
+
+  DeclAccessPair ToFoundDecl = DeclAccessPair::make(
+    dyn_cast<NamedDecl>(Importer.Import(E->getFoundDecl().getDecl())),
+    E->getFoundDecl().getAccess());
+
+  DeclarationNameInfo ToMemberNameInfo(
+    Importer.Import(E->getMemberNameInfo().getName()),
+    Importer.Import(E->getMemberNameInfo().getLoc()));
+
+  if (E->hasExplicitTemplateArgs()) {
+    return nullptr; // FIXME: handle template arguments
+  }
+
+  return MemberExpr::Create(Importer.getToContext(), ToBase,
+                            E->isArrow(),
+                            Importer.Import(E->getOperatorLoc()),
+                            Importer.Import(E->getQualifierLoc()),
+                            Importer.Import(E->getTemplateKeywordLoc()),
+                            ToMember, ToFoundDecl, ToMemberNameInfo,
+                            nullptr, T, E->getValueKind(),
+                            E->getObjectKind());
+}
+
+Expr *ASTNodeImporter::VisitCallExpr(CallExpr *E) {
+  QualType T = Importer.Import(E->getType());
+  if (T.isNull())
+    return nullptr;
+
+  Expr *ToCallee = Importer.Import(E->getCallee());
+  if (!ToCallee && E->getCallee())
+    return nullptr;
+
+  unsigned NumArgs = E->getNumArgs();
+
+  llvm::SmallVector<Expr *, 2> ToArgs(NumArgs);
+
+  for (unsigned ai = 0, ae = NumArgs; ai != ae; ++ai) {
+    Expr *FromArg = E->getArg(ai);
+    Expr *ToArg = Importer.Import(FromArg);
+    if (!ToArg)
+      return nullptr;
+    ToArgs[ai] = ToArg;
+  }
+
+  Expr **ToArgs_Copied = new (Importer.getToContext()) 
+    Expr*[NumArgs];
+
+  for (unsigned ai = 0, ae = NumArgs; ai != ae; ++ai)
+    ToArgs_Copied[ai] = ToArgs[ai];
+
+  return new (Importer.getToContext())
+    CallExpr(Importer.getToContext(), ToCallee, 
+             ArrayRef<Expr*>(ToArgs_Copied, NumArgs), T, E->getValueKind(),
+             Importer.Import(E->getRParenLoc()));
+}
+
 ASTImporter::ASTImporter(ASTContext &ToContext, FileManager &ToFileManager,
                          ASTContext &FromContext, FileManager &FromFileManager,
                          bool MinimalImport)
@@ -4636,6 +5315,17 @@ TypeSourceInfo *ASTImporter::Import(TypeSourceInfo *FromTSI) {
                         FromTSI->getTypeLoc().getLocStart());
 }
 
+Decl *ASTImporter::GetAlreadyImportedOrNull(Decl *FromD) {
+  llvm::DenseMap<Decl *, Decl *>::iterator Pos = ImportedDecls.find(FromD);
+  if (Pos != ImportedDecls.end()) {
+    Decl *ToD = Pos->second;
+    ASTNodeImporter(*this).ImportDefinitionIfNeeded(FromD, ToD);
+    return ToD;
+  } else {
+    return nullptr;
+  }
+}
+
 Decl *ASTImporter::Import(Decl *FromD) {
   if (!FromD)
     return nullptr;
@@ -4927,8 +5617,9 @@ SourceLocation ASTImporter::Import(SourceLocation FromLoc) {
   FileID ToFileID = Import(Decomposed.first);
   if (ToFileID.isInvalid())
     return SourceLocation();
-  return ToSM.getLocForStartOfFile(ToFileID)
-             .getLocWithOffset(Decomposed.second);
+  SourceLocation ret = ToSM.getLocForStartOfFile(ToFileID)
+                           .getLocWithOffset(Decomposed.second);
+  return ret;
 }
 
 SourceRange ASTImporter::Import(SourceRange FromRange) {
@@ -4952,7 +5643,7 @@ FileID ASTImporter::Import(FileID FromID) {
   // Map the FileID for to the "to" source manager.
   FileID ToID;
   const SrcMgr::ContentCache *Cache = FromSLoc.getFile().getContentCache();
-  if (Cache->OrigEntry) {
+  if (Cache->OrigEntry && Cache->OrigEntry->getDir()) {
     // FIXME: We probably want to use getVirtualFile(), so we don't hit the
     // disk again
     // FIXME: We definitely want to re-use the existing MemoryBuffer, rather
diff --git a/lib/AST/AttrImpl.cpp b/lib/AST/AttrImpl.cpp
index 0bf6bcd9092e..cb608700133c 100644
--- a/lib/AST/AttrImpl.cpp
+++ b/lib/AST/AttrImpl.cpp
@@ -7,7 +7,7 @@
 //
 //===----------------------------------------------------------------------===//
 //
-//  This file contains out-of-line virtual methods for Attr classes.
+//  This file contains out-of-line methods for Attr classes.
 //
 //===----------------------------------------------------------------------===//
 
@@ -18,10 +18,4 @@
 #include "llvm/ADT/StringSwitch.h"
 using namespace clang;
 
-Attr::~Attr() { }
-
-void InheritableAttr::anchor() { }
-
-void InheritableParamAttr::anchor() { }
-
 #include "clang/AST/AttrImpl.inc"
diff --git a/lib/AST/CXXABI.h b/lib/AST/CXXABI.h
index 8e9e358525e8..dad226474fa7 100644
--- a/lib/AST/CXXABI.h
+++ b/lib/AST/CXXABI.h
@@ -20,6 +20,8 @@
 namespace clang {
 
 class ASTContext;
+class CXXConstructorDecl;
+class Expr;
 class MemberPointerType;
 class MangleNumberingContext;
 
@@ -41,6 +43,20 @@ public:
 
   /// Returns a new mangling number context for this C++ ABI.
   virtual MangleNumberingContext *createMangleNumberingContext() const = 0;
+
+  /// Adds a mapping from class to copy constructor for this C++ ABI.
+  virtual void addCopyConstructorForExceptionObject(CXXRecordDecl *,
+                                                    CXXConstructorDecl *) = 0;
+
+  /// Retrieves the mapping from class to copy constructor for this C++ ABI.
+  virtual const CXXConstructorDecl *
+  getCopyConstructorForExceptionObject(CXXRecordDecl *) = 0;
+
+  virtual void addDefaultArgExprForConstructor(const CXXConstructorDecl *CD,
+                                               unsigned ParmIdx, Expr *DAE) = 0;
+
+  virtual Expr *getDefaultArgExprForConstructor(const CXXConstructorDecl *CD,
+                                                unsigned ParmIdx) = 0;
 };
 
 /// Creates an instance of a C++ ABI class.
diff --git a/lib/AST/CXXInheritance.cpp b/lib/AST/CXXInheritance.cpp
index 6e80ee7c28a1..800c8f83b880 100644
--- a/lib/AST/CXXInheritance.cpp
+++ b/lib/AST/CXXInheritance.cpp
@@ -318,48 +318,36 @@ bool CXXRecordDecl::lookupInBases(BaseMatchesCallback *BaseMatches,
   //
   // FIXME: This is an O(N^2) algorithm, but DPG doesn't see an easy
   // way to make it any faster.
-  for (CXXBasePaths::paths_iterator P = Paths.begin(), PEnd = Paths.end();
-       P != PEnd; /* increment in loop */) {
-    bool Hidden = false;
-
-    for (CXXBasePath::iterator PE = P->begin(), PEEnd = P->end();
-         PE != PEEnd && !Hidden; ++PE) {
-      if (PE->Base->isVirtual()) {
-        CXXRecordDecl *VBase = nullptr;
-        if (const RecordType *Record = PE->Base->getType()->getAs<RecordType>())
-          VBase = cast<CXXRecordDecl>(Record->getDecl());
-        if (!VBase)
+  Paths.Paths.remove_if([&Paths](const CXXBasePath &Path) {
+    for (const CXXBasePathElement &PE : Path) {
+      if (!PE.Base->isVirtual())
+        continue;
+
+      CXXRecordDecl *VBase = nullptr;
+      if (const RecordType *Record = PE.Base->getType()->getAs<RecordType>())
+        VBase = cast<CXXRecordDecl>(Record->getDecl());
+      if (!VBase)
+        break;
+
+      // The declaration(s) we found along this path were found in a
+      // subobject of a virtual base. Check whether this virtual
+      // base is a subobject of any other path; if so, then the
+      // declaration in this path are hidden by that patch.
+      for (const CXXBasePath &HidingP : Paths) {
+        CXXRecordDecl *HidingClass = nullptr;
+        if (const RecordType *Record =
+                HidingP.back().Base->getType()->getAs<RecordType>())
+          HidingClass = cast<CXXRecordDecl>(Record->getDecl());
+        if (!HidingClass)
           break;
 
-        // The declaration(s) we found along this path were found in a
-        // subobject of a virtual base. Check whether this virtual
-        // base is a subobject of any other path; if so, then the
-        // declaration in this path are hidden by that patch.
-        for (CXXBasePaths::paths_iterator HidingP = Paths.begin(),
-                                       HidingPEnd = Paths.end();
-             HidingP != HidingPEnd;
-             ++HidingP) {
-          CXXRecordDecl *HidingClass = nullptr;
-          if (const RecordType *Record
-                       = HidingP->back().Base->getType()->getAs<RecordType>())
-            HidingClass = cast<CXXRecordDecl>(Record->getDecl());
-          if (!HidingClass)
-            break;
-
-          if (HidingClass->isVirtuallyDerivedFrom(VBase)) {
-            Hidden = true;
-            break;
-          }
-        }
+        if (HidingClass->isVirtuallyDerivedFrom(VBase))
+          return true;
       }
     }
+    return false;
+  });
 
-    if (Hidden)
-      P = Paths.Paths.erase(P);
-    else
-      ++P;
-  }
-  
   return true;
 }
 
@@ -569,18 +557,14 @@ void FinalOverriderCollector::Collect(const CXXRecordDecl *RD,
     // overrider. To do so, we dig down to the original virtual
     // functions using data recursion and update all of the methods it
     // overrides.
-    typedef std::pair<CXXMethodDecl::method_iterator, 
-                      CXXMethodDecl::method_iterator> OverriddenMethods;
+    typedef llvm::iterator_range<CXXMethodDecl::method_iterator>
+        OverriddenMethods;
     SmallVector<OverriddenMethods, 4> Stack;
-    Stack.push_back(std::make_pair(CanonM->begin_overridden_methods(),
-                                   CanonM->end_overridden_methods()));
+    Stack.push_back(llvm::make_range(CanonM->begin_overridden_methods(),
+                                     CanonM->end_overridden_methods()));
     while (!Stack.empty()) {
-      OverriddenMethods OverMethods = Stack.back();
-      Stack.pop_back();
-
-      for (; OverMethods.first != OverMethods.second; ++OverMethods.first) {
-        const CXXMethodDecl *CanonOM
-          = cast<CXXMethodDecl>((*OverMethods.first)->getCanonicalDecl());
+      for (const CXXMethodDecl *OM : Stack.pop_back_val()) {
+        const CXXMethodDecl *CanonOM = OM->getCanonicalDecl();
 
         // C++ [class.virtual]p2:
         //   A virtual member function C::vf of a class object S is
@@ -601,8 +585,8 @@ void FinalOverriderCollector::Collect(const CXXRecordDecl *RD,
 
         // Continue recursion to the methods that this virtual method
         // overrides.
-        Stack.push_back(std::make_pair(CanonOM->begin_overridden_methods(),
-                                       CanonOM->end_overridden_methods()));
+        Stack.push_back(llvm::make_range(CanonOM->begin_overridden_methods(),
+                                         CanonOM->end_overridden_methods()));
       }
     }
 
@@ -630,54 +614,32 @@ CXXRecordDecl::getFinalOverriders(CXXFinalOverriderMap &FinalOverriders) const {
   // Weed out any final overriders that come from virtual base class
   // subobjects that were hidden by other subobjects along any path.
   // This is the final-overrider variant of C++ [class.member.lookup]p10.
-  for (CXXFinalOverriderMap::iterator OM = FinalOverriders.begin(), 
-                           OMEnd = FinalOverriders.end();
-       OM != OMEnd;
-       ++OM) {
-    for (OverridingMethods::iterator SO = OM->second.begin(), 
-                                  SOEnd = OM->second.end();
-         SO != SOEnd; 
-         ++SO) {
-      SmallVectorImpl<UniqueVirtualMethod> &Overriding = SO->second;
+  for (auto &OM : FinalOverriders) {
+    for (auto &SO : OM.second) {
+      SmallVectorImpl<UniqueVirtualMethod> &Overriding = SO.second;
       if (Overriding.size() < 2)
         continue;
 
-      for (SmallVectorImpl<UniqueVirtualMethod>::iterator
-             Pos = Overriding.begin(), PosEnd = Overriding.end();
-           Pos != PosEnd;
-           /* increment in loop */) {
-        if (!Pos->InVirtualSubobject) {
-          ++Pos;
-          continue;
-        }
+      auto IsHidden = [&Overriding](const UniqueVirtualMethod &M) {
+        if (!M.InVirtualSubobject)
+          return false;
 
         // We have an overriding method in a virtual base class
         // subobject (or non-virtual base class subobject thereof);
         // determine whether there exists an other overriding method
         // in a base class subobject that hides the virtual base class
         // subobject.
-        bool Hidden = false;
-        for (SmallVectorImpl<UniqueVirtualMethod>::iterator
-               OP = Overriding.begin(), OPEnd = Overriding.end();
-             OP != OPEnd && !Hidden; 
-             ++OP) {
-          if (Pos == OP)
-            continue;
-
-          if (OP->Method->getParent()->isVirtuallyDerivedFrom(
-                         const_cast<CXXRecordDecl *>(Pos->InVirtualSubobject)))
-            Hidden = true;
-        }
-
-        if (Hidden) {
-          // The current overriding function is hidden by another
-          // overriding function; remove this one.
-          Pos = Overriding.erase(Pos);
-          PosEnd = Overriding.end();
-        } else {
-          ++Pos;
-        }
-      }
+        for (const UniqueVirtualMethod &OP : Overriding)
+          if (&M != &OP &&
+              OP.Method->getParent()->isVirtuallyDerivedFrom(
+                  M.InVirtualSubobject))
+            return true;
+        return false;
+      };
+
+      Overriding.erase(
+          std::remove_if(Overriding.begin(), Overriding.end(), IsHidden),
+          Overriding.end());
     }
   }
 }
diff --git a/lib/AST/CommentLexer.cpp b/lib/AST/CommentLexer.cpp
index 06a08bdad45f..98b7e367950c 100644
--- a/lib/AST/CommentLexer.cpp
+++ b/lib/AST/CommentLexer.cpp
@@ -514,6 +514,12 @@ void Lexer::lexVerbatimBlockBody(Token &T) {
   if (CommentState == LCS_InsideCComment)
     skipLineStartingDecorations();
 
+  if (BufferPtr == CommentEnd) {
+    formTokenWithChars(T, BufferPtr, tok::verbatim_block_line);
+    T.setVerbatimBlockText("");
+    return;
+  }
+
   lexVerbatimBlockFirstLine(T);
 }
 
diff --git a/lib/AST/Decl.cpp b/lib/AST/Decl.cpp
index dc08d23a24b5..8eff4c4427b3 100644
--- a/lib/AST/Decl.cpp
+++ b/lib/AST/Decl.cpp
@@ -44,6 +44,12 @@ bool Decl::isOutOfLine() const {
   return !getLexicalDeclContext()->Equals(getDeclContext());
 }
 
+TranslationUnitDecl::TranslationUnitDecl(ASTContext &ctx)
+    : Decl(TranslationUnit, nullptr, SourceLocation()),
+      DeclContext(TranslationUnit), Ctx(ctx), AnonymousNamespace(nullptr) {
+  Hidden = Ctx.getLangOpts().ModulesLocalVisibility;
+}
+
 //===----------------------------------------------------------------------===//
 // NamedDecl Implementation
 //===----------------------------------------------------------------------===//
@@ -894,13 +900,13 @@ static LinkageInfo getLVForClassMember(const NamedDecl *D,
   if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(D)) {
     // If the type of the function uses a type with unique-external
     // linkage, it's not legally usable from outside this translation unit.
-    // But only look at the type-as-written. If this function has an auto-deduced
-    // return type, we can't compute the linkage of that type because it could
-    // require looking at the linkage of this function, and we don't need this
-    // for correctness because the type is not part of the function's
-    // signature.
-    // FIXME: This is a hack. We should be able to solve this circularity and the
-    // one in getLVForNamespaceScopeDecl for Functions some other way.
+    // But only look at the type-as-written. If this function has an
+    // auto-deduced return type, we can't compute the linkage of that type
+    // because it could require looking at the linkage of this function, and we
+    // don't need this for correctness because the type is not part of the
+    // function's signature.
+    // FIXME: This is a hack. We should be able to solve this circularity and
+    // the one in getLVForNamespaceScopeDecl for Functions some other way.
     {
       QualType TypeAsWritten = MD->getType();
       if (TypeSourceInfo *TSI = MD->getTypeSourceInfo())
@@ -1445,74 +1451,127 @@ void NamedDecl::getNameForDiagnostic(raw_ostream &OS,
     printName(OS);
 }
 
-bool NamedDecl::declarationReplaces(NamedDecl *OldD) const {
-  assert(getDeclName() == OldD->getDeclName() && "Declaration name mismatch");
+static bool isKindReplaceableBy(Decl::Kind OldK, Decl::Kind NewK) {
+  // For method declarations, we never replace.
+  if (ObjCMethodDecl::classofKind(NewK))
+    return false;
 
-  // UsingDirectiveDecl's are not really NamedDecl's, and all have same name.
-  // We want to keep it, unless it nominates same namespace.
-  if (getKind() == Decl::UsingDirective) {
-    return cast<UsingDirectiveDecl>(this)->getNominatedNamespace()
-             ->getOriginalNamespace() ==
-           cast<UsingDirectiveDecl>(OldD)->getNominatedNamespace()
-             ->getOriginalNamespace();
+  if (OldK == NewK)
+    return true;
+
+  // A compatibility alias for a class can be replaced by an interface.
+  if (ObjCCompatibleAliasDecl::classofKind(OldK) &&
+      ObjCInterfaceDecl::classofKind(NewK))
+    return true;
+
+  // A typedef-declaration, alias-declaration, or Objective-C class declaration
+  // can replace another declaration of the same type. Semantic analysis checks
+  // that we have matching types.
+  if ((TypedefNameDecl::classofKind(OldK) ||
+       ObjCInterfaceDecl::classofKind(OldK)) &&
+      (TypedefNameDecl::classofKind(NewK) ||
+       ObjCInterfaceDecl::classofKind(NewK)))
+    return true;
+
+  // Otherwise, a kind mismatch implies that the declaration is not replaced.
+  return false;
+}
+
+template<typename T> static bool isRedeclarableImpl(Redeclarable<T> *) {
+  return true;
+}
+static bool isRedeclarableImpl(...) { return false; }
+static bool isRedeclarable(Decl::Kind K) {
+  switch (K) {
+#define DECL(Type, Base) \
+  case Decl::Type: \
+    return isRedeclarableImpl((Type##Decl *)nullptr);
+#define ABSTRACT_DECL(DECL)
+#include "clang/AST/DeclNodes.inc"
   }
+  llvm_unreachable("unknown decl kind");
+}
 
-  if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(this))
-    // For function declarations, we keep track of redeclarations.
-    return FD->getPreviousDecl() == OldD;
+bool NamedDecl::declarationReplaces(NamedDecl *OldD, bool IsKnownNewer) const {
+  assert(getDeclName() == OldD->getDeclName() && "Declaration name mismatch");
 
-  // For function templates, the underlying function declarations are linked.
-  if (const FunctionTemplateDecl *FunctionTemplate
-        = dyn_cast<FunctionTemplateDecl>(this))
-    if (const FunctionTemplateDecl *OldFunctionTemplate
-          = dyn_cast<FunctionTemplateDecl>(OldD))
-      return FunctionTemplate->getTemplatedDecl()
-               ->declarationReplaces(OldFunctionTemplate->getTemplatedDecl());
-
-  // For method declarations, we keep track of redeclarations.
-  if (isa<ObjCMethodDecl>(this))
+  // Never replace one imported declaration with another; we need both results
+  // when re-exporting.
+  if (OldD->isFromASTFile() && isFromASTFile())
     return false;
 
-  // FIXME: Is this correct if one of the decls comes from an inline namespace?
-  if (isa<ObjCInterfaceDecl>(this) && isa<ObjCCompatibleAliasDecl>(OldD))
-    return true;
+  if (!isKindReplaceableBy(OldD->getKind(), getKind()))
+    return false;
 
-  if (isa<UsingShadowDecl>(this) && isa<UsingShadowDecl>(OldD))
-    return cast<UsingShadowDecl>(this)->getTargetDecl() ==
-           cast<UsingShadowDecl>(OldD)->getTargetDecl();
+  // Inline namespaces can give us two declarations with the same
+  // name and kind in the same scope but different contexts; we should
+  // keep both declarations in this case.
+  if (!this->getDeclContext()->getRedeclContext()->Equals(
+          OldD->getDeclContext()->getRedeclContext()))
+    return false;
 
-  if (isa<UsingDecl>(this) && isa<UsingDecl>(OldD)) {
+  if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(this))
+    // For function declarations, we keep track of redeclarations.
+    // FIXME: This returns false for functions that should in fact be replaced.
+    // Instead, perform some kind of type check?
+    if (FD->getPreviousDecl() != OldD)
+      return false;
+
+  // For function templates, the underlying function declarations are linked.
+  if (const FunctionTemplateDecl *FunctionTemplate =
+          dyn_cast<FunctionTemplateDecl>(this))
+    return FunctionTemplate->getTemplatedDecl()->declarationReplaces(
+        cast<FunctionTemplateDecl>(OldD)->getTemplatedDecl());
+
+  // Using shadow declarations can be overloaded on their target declarations
+  // if they introduce functions.
+  // FIXME: If our target replaces the old target, can we replace the old
+  //        shadow declaration?
+  if (auto *USD = dyn_cast<UsingShadowDecl>(this))
+    if (USD->getTargetDecl() != cast<UsingShadowDecl>(OldD)->getTargetDecl())
+      return false;
+
+  // Using declarations can be overloaded if they introduce functions.
+  if (auto *UD = dyn_cast<UsingDecl>(this)) {
     ASTContext &Context = getASTContext();
-    return Context.getCanonicalNestedNameSpecifier(
-                                     cast<UsingDecl>(this)->getQualifier()) ==
+    return Context.getCanonicalNestedNameSpecifier(UD->getQualifier()) ==
            Context.getCanonicalNestedNameSpecifier(
-                                        cast<UsingDecl>(OldD)->getQualifier());
+               cast<UsingDecl>(OldD)->getQualifier());
   }
-
-  if (isa<UnresolvedUsingValueDecl>(this) &&
-      isa<UnresolvedUsingValueDecl>(OldD)) {
+  if (auto *UUVD = dyn_cast<UnresolvedUsingValueDecl>(this)) {
     ASTContext &Context = getASTContext();
-    return Context.getCanonicalNestedNameSpecifier(
-                      cast<UnresolvedUsingValueDecl>(this)->getQualifier()) ==
+    return Context.getCanonicalNestedNameSpecifier(UUVD->getQualifier()) ==
            Context.getCanonicalNestedNameSpecifier(
                         cast<UnresolvedUsingValueDecl>(OldD)->getQualifier());
   }
 
-  // A typedef of an Objective-C class type can replace an Objective-C class
-  // declaration or definition, and vice versa.
-  // FIXME: Is this correct if one of the decls comes from an inline namespace?
-  if ((isa<TypedefNameDecl>(this) && isa<ObjCInterfaceDecl>(OldD)) ||
-      (isa<ObjCInterfaceDecl>(this) && isa<TypedefNameDecl>(OldD)))
-    return true;
+  // UsingDirectiveDecl's are not really NamedDecl's, and all have same name.
+  // We want to keep it, unless it nominates same namespace.
+  if (auto *UD = dyn_cast<UsingDirectiveDecl>(this))
+    return UD->getNominatedNamespace()->getOriginalNamespace() ==
+           cast<UsingDirectiveDecl>(OldD)->getNominatedNamespace()
+               ->getOriginalNamespace();
+
+  if (!IsKnownNewer && isRedeclarable(getKind())) {
+    // Check whether this is actually newer than OldD. We want to keep the
+    // newer declaration. This loop will usually only iterate once, because
+    // OldD is usually the previous declaration.
+    for (auto D : redecls()) {
+      if (D == OldD)
+        break;
 
-  // For non-function declarations, if the declarations are of the
-  // same kind and have the same parent then this must be a redeclaration,
-  // or semantic analysis would not have given us the new declaration.
-  // Note that inline namespaces can give us two declarations with the same
-  // name and kind in the same scope but different contexts.
-  return this->getKind() == OldD->getKind() &&
-         this->getDeclContext()->getRedeclContext()->Equals(
-             OldD->getDeclContext()->getRedeclContext());
+      // If we reach the canonical declaration, then OldD is not actually older
+      // than this one.
+      //
+      // FIXME: In this case, we should not add this decl to the lookup table.
+      if (D->isCanonicalDecl())
+        return false;
+    }
+  }
+
+  // It's a newer declaration of the same kind of declaration in the same scope,
+  // and not an overload: we want this decl instead of the existing one.
+  return true;
 }
 
 bool NamedDecl::hasLinkage() const {
@@ -1684,8 +1743,7 @@ QualifierInfo::setTemplateParameterListsInfo(ASTContext &Context,
   if (NumTPLists > 0) {
     TemplParamLists = new (Context) TemplateParameterList*[NumTPLists];
     NumTemplParamLists = NumTPLists;
-    for (unsigned i = NumTPLists; i-- > 0; )
-      TemplParamLists[i] = TPLists[i];
+    std::copy(TPLists, TPLists + NumTPLists, TemplParamLists);
   }
 }
 
@@ -1717,6 +1775,8 @@ VarDecl::VarDecl(Kind DK, ASTContext &C, DeclContext *DC,
                 "VarDeclBitfields too large!");
   static_assert(sizeof(ParmVarDeclBitfields) <= sizeof(unsigned),
                 "ParmVarDeclBitfields too large!");
+  static_assert(sizeof(NonParmVarDeclBitfields) <= sizeof(unsigned),
+                "NonParmVarDeclBitfields too large!");
   AllBits = 0;
   VarDeclBits.SClass = SC;
   // Everything else is implicitly initialized to false.
@@ -1743,9 +1803,12 @@ void VarDecl::setStorageClass(StorageClass SC) {
 VarDecl::TLSKind VarDecl::getTLSKind() const {
   switch (VarDeclBits.TSCSpec) {
   case TSCS_unspecified:
-    if (hasAttr<ThreadAttr>())
-      return TLS_Static;
-    return TLS_None;
+    if (!hasAttr<ThreadAttr>())
+      return TLS_None;
+    return getASTContext().getLangOpts().isCompatibleWithMSVC(
+               LangOptions::MSVC2015)
+               ? TLS_Dynamic
+               : TLS_Static;
   case TSCS___thread: // Fall through.
   case TSCS__Thread_local:
       return TLS_Static;
@@ -1825,9 +1888,8 @@ bool VarDecl::isInExternCXXContext() const {
 
 VarDecl *VarDecl::getCanonicalDecl() { return getFirstDecl(); }
 
-VarDecl::DefinitionKind VarDecl::isThisDeclarationADefinition(
-  ASTContext &C) const
-{
+VarDecl::DefinitionKind
+VarDecl::isThisDeclarationADefinition(ASTContext &C) const {
   // C++ [basic.def]p2:
   //   A declaration is a definition unless [...] it contains the 'extern'
   //   specifier or a linkage-specification and neither an initializer [...],
@@ -1867,6 +1929,10 @@ VarDecl::DefinitionKind VarDecl::isThisDeclarationADefinition(
   if (hasAttr<AliasAttr>())
     return Definition;
 
+  if (const auto *SAA = getAttr<SelectAnyAttr>())
+    if (!SAA->isInherited())
+      return Definition;
+
   // A variable template specialization (other than a static data member
   // template or an explicit specialization) is a declaration until we
   // instantiate its initializer.
@@ -2460,39 +2526,6 @@ bool FunctionDecl::isReplaceableGlobalAllocationFunction() const {
   return RD && isNamed(RD, "nothrow_t") && RD->isInStdNamespace();
 }
 
-FunctionDecl *
-FunctionDecl::getCorrespondingUnsizedGlobalDeallocationFunction() const {
-  ASTContext &Ctx = getASTContext();
-  if (!Ctx.getLangOpts().SizedDeallocation)
-    return nullptr;
-
-  if (getDeclName().getNameKind() != DeclarationName::CXXOperatorName)
-    return nullptr;
-  if (getDeclName().getCXXOverloadedOperator() != OO_Delete &&
-      getDeclName().getCXXOverloadedOperator() != OO_Array_Delete)
-    return nullptr;
-  if (isa<CXXRecordDecl>(getDeclContext()))
-    return nullptr;
-
-  if (!getDeclContext()->getRedeclContext()->isTranslationUnit())
-    return nullptr;
-
-  if (getNumParams() != 2 || isVariadic() ||
-      !Ctx.hasSameType(getType()->castAs<FunctionProtoType>()->getParamType(1),
-                       Ctx.getSizeType()))
-    return nullptr;
-
-  // This is a sized deallocation function. Find the corresponding unsized
-  // deallocation function.
-  lookup_const_result R = getDeclContext()->lookup(getDeclName());
-  for (lookup_const_result::iterator RI = R.begin(), RE = R.end(); RI != RE;
-       ++RI)
-    if (FunctionDecl *FD = dyn_cast<FunctionDecl>(*RI))
-      if (FD->getNumParams() == 1 && !FD->isVariadic())
-        return FD;
-  return nullptr;
-}
-
 LanguageLinkage FunctionDecl::getLanguageLinkage() const {
   return getDeclLanguageLinkage(*this);
 }
@@ -2550,10 +2583,6 @@ FunctionDecl::setPreviousDeclaration(FunctionDecl *PrevDecl) {
     IsInline = true;
 }
 
-const FunctionDecl *FunctionDecl::getCanonicalDecl() const {
-  return getFirstDecl();
-}
-
 FunctionDecl *FunctionDecl::getCanonicalDecl() { return getFirstDecl(); }
 
 /// \brief Returns a value indicating whether this function
@@ -2581,7 +2610,14 @@ unsigned FunctionDecl::getBuiltinID() const {
     // extern "C".
     // FIXME: A recognised library function may not be directly in an extern "C"
     // declaration, for instance "extern "C" { namespace std { decl } }".
-    if (!LinkageDecl || LinkageDecl->getLanguage() != LinkageSpecDecl::lang_c)
+    if (!LinkageDecl) {
+      if (BuiltinID == Builtin::BI__GetExceptionInfo &&
+          Context.getTargetInfo().getCXXABI().isMicrosoft() &&
+          isInStdNamespace())
+        return Builtin::BI__GetExceptionInfo;
+      return 0;
+    }
+    if (LinkageDecl->getLanguage() != LinkageSpecDecl::lang_c)
       return 0;
   }
 
@@ -2796,6 +2832,18 @@ SourceRange FunctionDecl::getReturnTypeSourceRange() const {
   return RTRange;
 }
 
+bool FunctionDecl::hasUnusedResultAttr() const {
+  QualType RetType = getReturnType();
+  if (RetType->isRecordType()) {
+    const CXXRecordDecl *Ret = RetType->getAsCXXRecordDecl();
+    const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(this);
+    if (Ret && Ret->hasAttr<WarnUnusedResultAttr>() &&
+        !(MD && MD->getCorrespondingMethodInClass(Ret, true)))
+      return true;
+  }
+  return hasAttr<WarnUnusedResultAttr>();
+}
+
 /// \brief For an inline function definition in C, or for a gnu_inline function
 /// in C++, determine whether the definition will be externally visible.
 ///
@@ -3077,6 +3125,8 @@ DependentFunctionTemplateSpecializationInfo::
 DependentFunctionTemplateSpecializationInfo(const UnresolvedSetImpl &Ts,
                                       const TemplateArgumentListInfo &TArgs)
   : AngleLocs(TArgs.getLAngleLoc(), TArgs.getRAngleLoc()) {
+  static_assert(sizeof(*this) % llvm::AlignOf<void *>::Alignment == 0,
+                "Trailing data is unaligned!");
 
   d.NumTemplates = Ts.size();
   d.NumArgs = TArgs.size();
@@ -3590,7 +3640,7 @@ void RecordDecl::completeDefinition() {
 /// This which can be turned on with an attribute, pragma, or the
 /// -mms-bitfields command-line option.
 bool RecordDecl::isMsStruct(const ASTContext &C) const {
-  return hasAttr<MsStructAttr>() || C.getLangOpts().MSBitfields == 1;
+  return hasAttr<MSStructAttr>() || C.getLangOpts().MSBitfields == 1;
 }
 
 static bool isFieldOrIndirectField(Decl::Kind K) {
@@ -3747,6 +3797,13 @@ TranslationUnitDecl *TranslationUnitDecl::Create(ASTContext &C) {
   return new (C, (DeclContext *)nullptr) TranslationUnitDecl(C);
 }
 
+void ExternCContextDecl::anchor() { }
+
+ExternCContextDecl *ExternCContextDecl::Create(const ASTContext &C,
+                                               TranslationUnitDecl *DC) {
+  return new (C, DC) ExternCContextDecl(DC);
+}
+
 void LabelDecl::anchor() { }
 
 LabelDecl *LabelDecl::Create(ASTContext &C, DeclContext *DC,
@@ -3885,6 +3942,21 @@ TypedefDecl *TypedefDecl::Create(ASTContext &C, DeclContext *DC,
 
 void TypedefNameDecl::anchor() { }
 
+TagDecl *TypedefNameDecl::getAnonDeclWithTypedefName(bool AnyRedecl) const {
+  if (auto *TT = getTypeSourceInfo()->getType()->getAs<TagType>()) {
+    auto *OwningTypedef = TT->getDecl()->getTypedefNameForAnonDecl();
+    auto *ThisTypedef = this;
+    if (AnyRedecl && OwningTypedef) {
+      OwningTypedef = OwningTypedef->getCanonicalDecl();
+      ThisTypedef = ThisTypedef->getCanonicalDecl();
+    }
+    if (OwningTypedef == ThisTypedef)
+      return TT->getDecl();
+  }
+
+  return nullptr;
+}
+
 TypedefDecl *TypedefDecl::CreateDeserialized(ASTContext &C, unsigned ID) {
   return new (C, ID) TypedefDecl(C, nullptr, SourceLocation(), SourceLocation(),
                                  nullptr, nullptr);
diff --git a/lib/AST/DeclBase.cpp b/lib/AST/DeclBase.cpp
index a46787fb0e8f..79cadcfcb167 100644
--- a/lib/AST/DeclBase.cpp
+++ b/lib/AST/DeclBase.cpp
@@ -66,6 +66,12 @@ void *Decl::operator new(std::size_t Size, const ASTContext &Context,
 void *Decl::operator new(std::size_t Size, const ASTContext &Ctx,
                          DeclContext *Parent, std::size_t Extra) {
   assert(!Parent || &Parent->getParentASTContext() == &Ctx);
+  // With local visibility enabled, we track the owning module even for local
+  // declarations.
+  if (Ctx.getLangOpts().ModulesLocalVisibility) {
+    void *Buffer = ::operator new(sizeof(Module *) + Size + Extra, Ctx);
+    return new (Buffer) Module*(nullptr) + 1;
+  }
   return ::operator new(Size + Extra, Ctx);
 }
 
@@ -336,20 +342,34 @@ bool Decl::isReferenced() const {
 static AvailabilityResult CheckAvailability(ASTContext &Context,
                                             const AvailabilityAttr *A,
                                             std::string *Message) {
-  StringRef TargetPlatform = Context.getTargetInfo().getPlatformName();
-  StringRef PrettyPlatformName
-    = AvailabilityAttr::getPrettyPlatformName(TargetPlatform);
-  if (PrettyPlatformName.empty())
-    PrettyPlatformName = TargetPlatform;
+  VersionTuple TargetMinVersion =
+    Context.getTargetInfo().getPlatformMinVersion();
 
-  VersionTuple TargetMinVersion = Context.getTargetInfo().getPlatformMinVersion();
   if (TargetMinVersion.empty())
     return AR_Available;
 
+  // Check if this is an App Extension "platform", and if so chop off
+  // the suffix for matching with the actual platform.
+  StringRef ActualPlatform = A->getPlatform()->getName();
+  StringRef RealizedPlatform = ActualPlatform;
+  if (Context.getLangOpts().AppExt) {
+    size_t suffix = RealizedPlatform.rfind("_app_extension");
+    if (suffix != StringRef::npos)
+      RealizedPlatform = RealizedPlatform.slice(0, suffix);
+  }
+
+  StringRef TargetPlatform = Context.getTargetInfo().getPlatformName();
+
   // Match the platform name.
-  if (A->getPlatform()->getName() != TargetPlatform)
+  if (RealizedPlatform != TargetPlatform)
     return AR_Available;
-  
+
+  StringRef PrettyPlatformName
+    = AvailabilityAttr::getPrettyPlatformName(ActualPlatform);
+
+  if (PrettyPlatformName.empty())
+    PrettyPlatformName = ActualPlatform;
+
   std::string HintMessage;
   if (!A->getMessage().empty()) {
     HintMessage = " - ";
@@ -583,6 +603,7 @@ unsigned Decl::getIdentifierNamespaceForKind(Kind DeclKind) {
     case Block:
     case Captured:
     case TranslationUnit:
+    case ExternCContext:
 
     case UsingDirective:
     case ClassTemplateSpecialization:
@@ -846,6 +867,10 @@ bool DeclContext::isDependentContext() const {
       return getLexicalParent()->isDependentContext();
   }
 
+  // FIXME: A variable template is a dependent context, but is not a
+  // DeclContext. A context within it (such as a lambda-expression)
+  // should be considered dependent.
+
   return getParent() && getParent()->isDependentContext();
 }
 
@@ -889,6 +914,7 @@ bool DeclContext::Encloses(const DeclContext *DC) const {
 DeclContext *DeclContext::getPrimaryContext() {
   switch (DeclKind) {
   case Decl::TranslationUnit:
+  case Decl::ExternCContext:
   case Decl::LinkageSpec:
   case Decl::Block:
   case Decl::Captured:
@@ -991,23 +1017,24 @@ DeclContext::BuildDeclChain(ArrayRef<Decl*> Decls,
 /// built a lookup map. For every name in the map, pull in the new names from
 /// the external storage.
 void DeclContext::reconcileExternalVisibleStorage() const {
-  assert(NeedToReconcileExternalVisibleStorage && LookupPtr.getPointer());
+  assert(NeedToReconcileExternalVisibleStorage && LookupPtr);
   NeedToReconcileExternalVisibleStorage = false;
 
-  for (auto &Lookup : *LookupPtr.getPointer())
+  for (auto &Lookup : *LookupPtr)
     Lookup.second.setHasExternalDecls();
 }
 
 /// \brief Load the declarations within this lexical storage from an
 /// external source.
-void
+/// \return \c true if any declarations were added.
+bool
 DeclContext::LoadLexicalDeclsFromExternalStorage() const {
   ExternalASTSource *Source = getParentASTContext().getExternalSource();
   assert(hasExternalLexicalStorage() && Source && "No external storage?");
 
   // Notify that we have a DeclContext that is initializing.
   ExternalASTSource::Deserializing ADeclContext(Source);
-  
+
   // Load the external declarations, if any.
   SmallVector<Decl*, 64> Decls;
   ExternalLexicalStorage = false;
@@ -1017,11 +1044,11 @@ DeclContext::LoadLexicalDeclsFromExternalStorage() const {
     
   case ELR_Failure:
   case ELR_AlreadyLoaded:
-    return;
+    return false;
   }
 
   if (Decls.empty())
-    return;
+    return false;
 
   // We may have already loaded just the fields of this record, in which case
   // we need to ignore them.
@@ -1038,6 +1065,7 @@ DeclContext::LoadLexicalDeclsFromExternalStorage() const {
   FirstDecl = ExternalFirst;
   if (!LastDecl)
     LastDecl = ExternalLast;
+  return true;
 }
 
 DeclContext::lookup_result
@@ -1045,7 +1073,7 @@ ExternalASTSource::SetNoExternalVisibleDeclsForName(const DeclContext *DC,
                                                     DeclarationName Name) {
   ASTContext &Context = DC->getParentASTContext();
   StoredDeclsMap *Map;
-  if (!(Map = DC->LookupPtr.getPointer()))
+  if (!(Map = DC->LookupPtr))
     Map = DC->CreateStoredDeclsMap(Context);
   if (DC->NeedToReconcileExternalVisibleStorage)
     DC->reconcileExternalVisibleStorage();
@@ -1061,7 +1089,7 @@ ExternalASTSource::SetExternalVisibleDeclsForName(const DeclContext *DC,
                                                   ArrayRef<NamedDecl*> Decls) {
   ASTContext &Context = DC->getParentASTContext();
   StoredDeclsMap *Map;
-  if (!(Map = DC->LookupPtr.getPointer()))
+  if (!(Map = DC->LookupPtr))
     Map = DC->CreateStoredDeclsMap(Context);
   if (DC->NeedToReconcileExternalVisibleStorage)
     DC->reconcileExternalVisibleStorage();
@@ -1078,7 +1106,7 @@ ExternalASTSource::SetExternalVisibleDeclsForName(const DeclContext *DC,
     // first.
     llvm::SmallVector<unsigned, 8> Skip;
     for (unsigned I = 0, N = Decls.size(); I != N; ++I)
-      if (List.HandleRedeclaration(Decls[I]))
+      if (List.HandleRedeclaration(Decls[I], /*IsKnownNewer*/false))
         Skip.push_back(I);
     Skip.push_back(Decls.size());
 
@@ -1155,7 +1183,7 @@ void DeclContext::removeDecl(Decl *D) {
     // Remove only decls that have a name
     if (!ND->getDeclName()) return;
 
-    StoredDeclsMap *Map = getPrimaryContext()->LookupPtr.getPointer();
+    StoredDeclsMap *Map = getPrimaryContext()->LookupPtr;
     if (!Map) return;
 
     StoredDeclsMap::iterator Pos = Map->find(ND->getDeclName());
@@ -1243,32 +1271,38 @@ static bool shouldBeHidden(NamedDecl *D) {
 StoredDeclsMap *DeclContext::buildLookup() {
   assert(this == getPrimaryContext() && "buildLookup called on non-primary DC");
 
-  // FIXME: Should we keep going if hasExternalVisibleStorage?
-  if (!LookupPtr.getInt())
-    return LookupPtr.getPointer();
+  if (!HasLazyLocalLexicalLookups && !HasLazyExternalLexicalLookups)
+    return LookupPtr;
 
   SmallVector<DeclContext *, 2> Contexts;
   collectAllContexts(Contexts);
-  for (unsigned I = 0, N = Contexts.size(); I != N; ++I)
-    buildLookupImpl<&DeclContext::decls_begin,
-                    &DeclContext::decls_end>(Contexts[I]);
+
+  if (HasLazyExternalLexicalLookups) {
+    HasLazyExternalLexicalLookups = false;
+    for (auto *DC : Contexts) {
+      if (DC->hasExternalLexicalStorage())
+        HasLazyLocalLexicalLookups |=
+            DC->LoadLexicalDeclsFromExternalStorage();
+    }
+
+    if (!HasLazyLocalLexicalLookups)
+      return LookupPtr;
+  }
+
+  for (auto *DC : Contexts)
+    buildLookupImpl(DC, hasExternalVisibleStorage());
 
   // We no longer have any lazy decls.
-  LookupPtr.setInt(false);
-  return LookupPtr.getPointer();
+  HasLazyLocalLexicalLookups = false;
+  return LookupPtr;
 }
 
 /// buildLookupImpl - Build part of the lookup data structure for the
 /// declarations contained within DCtx, which will either be this
 /// DeclContext, a DeclContext linked to it, or a transparent context
 /// nested within it.
-template<DeclContext::decl_iterator (DeclContext::*Begin)() const,
-         DeclContext::decl_iterator (DeclContext::*End)() const>
-void DeclContext::buildLookupImpl(DeclContext *DCtx) {
-  for (decl_iterator I = (DCtx->*Begin)(), E = (DCtx->*End)();
-       I != E; ++I) {
-    Decl *D = *I;
-
+void DeclContext::buildLookupImpl(DeclContext *DCtx, bool Internal) {
+  for (Decl *D : DCtx->noload_decls()) {
     // Insert this declaration into the lookup structure, but only if
     // it's semantically within its decl context. Any other decls which
     // should be found in this context are added eagerly.
@@ -1282,39 +1316,46 @@ void DeclContext::buildLookupImpl(DeclContext *DCtx) {
           (!ND->isFromASTFile() ||
            (isTranslationUnit() &&
             !getParentASTContext().getLangOpts().CPlusPlus)))
-        makeDeclVisibleInContextImpl(ND, false);
+        makeDeclVisibleInContextImpl(ND, Internal);
 
     // If this declaration is itself a transparent declaration context
     // or inline namespace, add the members of this declaration of that
     // context (recursively).
     if (DeclContext *InnerCtx = dyn_cast<DeclContext>(D))
       if (InnerCtx->isTransparentContext() || InnerCtx->isInlineNamespace())
-        buildLookupImpl<Begin, End>(InnerCtx);
+        buildLookupImpl(InnerCtx, Internal);
   }
 }
 
+NamedDecl *const DeclContextLookupResult::SingleElementDummyList = nullptr;
+
 DeclContext::lookup_result
-DeclContext::lookup(DeclarationName Name) {
+DeclContext::lookup(DeclarationName Name) const {
   assert(DeclKind != Decl::LinkageSpec &&
          "Should not perform lookups into linkage specs!");
 
-  DeclContext *PrimaryContext = getPrimaryContext();
+  const DeclContext *PrimaryContext = getPrimaryContext();
   if (PrimaryContext != this)
     return PrimaryContext->lookup(Name);
 
-  // If this is a namespace, ensure that any later redeclarations of it have
-  // been loaded, since they may add names to the result of this lookup.
-  if (auto *ND = dyn_cast<NamespaceDecl>(this))
-    (void)ND->getMostRecentDecl();
+  // If we have an external source, ensure that any later redeclarations of this
+  // context have been loaded, since they may add names to the result of this
+  // lookup (or add external visible storage).
+  ExternalASTSource *Source = getParentASTContext().getExternalSource();
+  if (Source)
+    (void)cast<Decl>(this)->getMostRecentDecl();
 
   if (hasExternalVisibleStorage()) {
+    assert(Source && "external visible storage but no external source?");
+
     if (NeedToReconcileExternalVisibleStorage)
       reconcileExternalVisibleStorage();
 
-    StoredDeclsMap *Map = LookupPtr.getPointer();
+    StoredDeclsMap *Map = LookupPtr;
 
-    if (LookupPtr.getInt())
-      Map = buildLookup();
+    if (HasLazyLocalLexicalLookups || HasLazyExternalLexicalLookups)
+      // FIXME: Make buildLookup const?
+      Map = const_cast<DeclContext*>(this)->buildLookup();
 
     if (!Map)
       Map = CreateStoredDeclsMap(getParentASTContext());
@@ -1325,28 +1366,27 @@ DeclContext::lookup(DeclarationName Name) {
     if (!R.second && !R.first->second.hasExternalDecls())
       return R.first->second.getLookupResult();
 
-    ExternalASTSource *Source = getParentASTContext().getExternalSource();
     if (Source->FindExternalVisibleDeclsByName(this, Name) || !R.second) {
-      if (StoredDeclsMap *Map = LookupPtr.getPointer()) {
+      if (StoredDeclsMap *Map = LookupPtr) {
         StoredDeclsMap::iterator I = Map->find(Name);
         if (I != Map->end())
           return I->second.getLookupResult();
       }
     }
 
-    return lookup_result(lookup_iterator(nullptr), lookup_iterator(nullptr));
+    return lookup_result();
   }
 
-  StoredDeclsMap *Map = LookupPtr.getPointer();
-  if (LookupPtr.getInt())
-    Map = buildLookup();
+  StoredDeclsMap *Map = LookupPtr;
+  if (HasLazyLocalLexicalLookups || HasLazyExternalLexicalLookups)
+    Map = const_cast<DeclContext*>(this)->buildLookup();
 
   if (!Map)
-    return lookup_result(lookup_iterator(nullptr), lookup_iterator(nullptr));
+    return lookup_result();
 
   StoredDeclsMap::iterator I = Map->find(Name);
   if (I == Map->end())
-    return lookup_result(lookup_iterator(nullptr), lookup_iterator(nullptr));
+    return lookup_result();
 
   return I->second.getLookupResult();
 }
@@ -1355,40 +1395,29 @@ DeclContext::lookup_result
 DeclContext::noload_lookup(DeclarationName Name) {
   assert(DeclKind != Decl::LinkageSpec &&
          "Should not perform lookups into linkage specs!");
-  if (!hasExternalVisibleStorage())
-    return lookup(Name);
 
   DeclContext *PrimaryContext = getPrimaryContext();
   if (PrimaryContext != this)
     return PrimaryContext->noload_lookup(Name);
 
-  StoredDeclsMap *Map = LookupPtr.getPointer();
-  if (LookupPtr.getInt()) {
-    // Carefully build the lookup map, without deserializing anything.
+  // If we have any lazy lexical declarations not in our lookup map, add them
+  // now. Don't import any external declarations, not even if we know we have
+  // some missing from the external visible lookups.
+  if (HasLazyLocalLexicalLookups) {
     SmallVector<DeclContext *, 2> Contexts;
     collectAllContexts(Contexts);
     for (unsigned I = 0, N = Contexts.size(); I != N; ++I)
-      buildLookupImpl<&DeclContext::noload_decls_begin,
-                      &DeclContext::noload_decls_end>(Contexts[I]);
-
-    // We no longer have any lazy decls.
-    LookupPtr.setInt(false);
-
-    // There may now be names for which we have local decls but are
-    // missing the external decls. FIXME: Just set the hasExternalDecls
-    // flag on those names that have external decls.
-    NeedToReconcileExternalVisibleStorage = true;
-
-    Map = LookupPtr.getPointer();
+      buildLookupImpl(Contexts[I], hasExternalVisibleStorage());
+    HasLazyLocalLexicalLookups = false;
   }
 
+  StoredDeclsMap *Map = LookupPtr;
   if (!Map)
-    return lookup_result(lookup_iterator(nullptr), lookup_iterator(nullptr));
+    return lookup_result();
 
   StoredDeclsMap::iterator I = Map->find(Name);
   return I != Map->end() ? I->second.getLookupResult()
-                         : lookup_result(lookup_iterator(nullptr),
-                                         lookup_iterator(nullptr));
+                         : lookup_result();
 }
 
 void DeclContext::localUncachedLookup(DeclarationName Name,
@@ -1404,8 +1433,9 @@ void DeclContext::localUncachedLookup(DeclarationName Name,
   }
 
   // If we have a lookup table, check there first. Maybe we'll get lucky.
-  if (Name && !LookupPtr.getInt()) {
-    if (StoredDeclsMap *Map = LookupPtr.getPointer()) {
+  // FIXME: Should we be checking these flags on the primary context?
+  if (Name && !HasLazyLocalLexicalLookups && !HasLazyExternalLexicalLookups) {
+    if (StoredDeclsMap *Map = LookupPtr) {
       StoredDeclsMap::iterator Pos = Map->find(Name);
       if (Pos != Map->end()) {
         Results.insert(Results.end(),
@@ -1418,6 +1448,8 @@ void DeclContext::localUncachedLookup(DeclarationName Name,
 
   // Slow case: grovel through the declarations in our chain looking for 
   // matches.
+  // FIXME: If we have lazy external declarations, this will not find them!
+  // FIXME: Should we CollectAllContexts and walk them all here?
   for (Decl *D = FirstDecl; D; D = D->getNextDeclInContext()) {
     if (NamedDecl *ND = dyn_cast<NamedDecl>(D))
       if (ND->getDeclName() == Name)
@@ -1498,7 +1530,7 @@ void DeclContext::makeDeclVisibleInContextWithFlags(NamedDecl *D, bool Internal,
   // FIXME: As a performance hack, don't add such decls into the translation
   // unit unless we're in C++, since qualified lookup into the TU is never
   // performed.
-  if (LookupPtr.getPointer() || hasExternalVisibleStorage() ||
+  if (LookupPtr || hasExternalVisibleStorage() ||
       ((!Recoverable || D->getDeclContext() != D->getLexicalDeclContext()) &&
        (getParentASTContext().getLangOpts().CPlusPlus ||
         !isTranslationUnit()))) {
@@ -1508,7 +1540,7 @@ void DeclContext::makeDeclVisibleInContextWithFlags(NamedDecl *D, bool Internal,
     buildLookup();
     makeDeclVisibleInContextImpl(D, Internal);
   } else {
-    LookupPtr.setInt(true);
+    HasLazyLocalLexicalLookups = true;
   }
 
   // If we are a transparent context or inline namespace, insert into our
@@ -1526,7 +1558,7 @@ void DeclContext::makeDeclVisibleInContextWithFlags(NamedDecl *D, bool Internal,
 
 void DeclContext::makeDeclVisibleInContextImpl(NamedDecl *D, bool Internal) {
   // Find or create the stored declaration map.
-  StoredDeclsMap *Map = LookupPtr.getPointer();
+  StoredDeclsMap *Map = LookupPtr;
   if (!Map) {
     ASTContext *C = &getParentASTContext();
     Map = CreateStoredDeclsMap(*C);
@@ -1555,12 +1587,12 @@ void DeclContext::makeDeclVisibleInContextImpl(NamedDecl *D, bool Internal) {
     return;
   }
 
-  else if (DeclNameEntries.isNull()) {
+  if (DeclNameEntries.isNull()) {
     DeclNameEntries.setOnlyValue(D);
     return;
   }
 
-  if (DeclNameEntries.HandleRedeclaration(D)) {
+  if (DeclNameEntries.HandleRedeclaration(D, /*IsKnownNewer*/!Internal)) {
     // This declaration has replaced an existing one for which
     // declarationReplaces returns true.
     return;
@@ -1570,15 +1602,17 @@ void DeclContext::makeDeclVisibleInContextImpl(NamedDecl *D, bool Internal) {
   DeclNameEntries.AddSubsequentDecl(D);
 }
 
+UsingDirectiveDecl *DeclContext::udir_iterator::operator*() const {
+  return cast<UsingDirectiveDecl>(*I);
+}
+
 /// Returns iterator range [First, Last) of UsingDirectiveDecls stored within
 /// this context.
 DeclContext::udir_range DeclContext::using_directives() const {
   // FIXME: Use something more efficient than normal lookup for using
   // directives. In C++, using directives are looked up more than anything else.
-  lookup_const_result Result = lookup(UsingDirectiveDecl::getName());
-  return udir_range(
-      reinterpret_cast<UsingDirectiveDecl *const *>(Result.begin()),
-      reinterpret_cast<UsingDirectiveDecl *const *>(Result.end()));
+  lookup_result Result = lookup(UsingDirectiveDecl::getName());
+  return udir_range(Result.begin(), Result.end());
 }
 
 //===----------------------------------------------------------------------===//
@@ -1586,7 +1620,7 @@ DeclContext::udir_range DeclContext::using_directives() const {
 //===----------------------------------------------------------------------===//
 
 StoredDeclsMap *DeclContext::CreateStoredDeclsMap(ASTContext &C) const {
-  assert(!LookupPtr.getPointer() && "context already has a decls map");
+  assert(!LookupPtr && "context already has a decls map");
   assert(getPrimaryContext() == this &&
          "creating decls map on non-primary context");
 
@@ -1598,7 +1632,7 @@ StoredDeclsMap *DeclContext::CreateStoredDeclsMap(ASTContext &C) const {
     M = new StoredDeclsMap();
   M->Previous = C.LastSDM;
   C.LastSDM = llvm::PointerIntPair<StoredDeclsMap*,1>(M, Dependent);
-  LookupPtr.setPointer(M);
+  LookupPtr = M;
   return M;
 }
 
@@ -1630,11 +1664,11 @@ DependentDiagnostic *DependentDiagnostic::Create(ASTContext &C,
   assert(Parent->isDependentContext()
          && "cannot iterate dependent diagnostics of non-dependent context");
   Parent = Parent->getPrimaryContext();
-  if (!Parent->LookupPtr.getPointer())
+  if (!Parent->LookupPtr)
     Parent->CreateStoredDeclsMap(C);
 
-  DependentStoredDeclsMap *Map
-    = static_cast<DependentStoredDeclsMap*>(Parent->LookupPtr.getPointer());
+  DependentStoredDeclsMap *Map =
+      static_cast<DependentStoredDeclsMap *>(Parent->LookupPtr);
 
   // Allocate the copy of the PartialDiagnostic via the ASTContext's
   // BumpPtrAllocator, rather than the ASTContext itself.
diff --git a/lib/AST/DeclCXX.cpp b/lib/AST/DeclCXX.cpp
index 9af8c4b242a6..8dc62dd07f2c 100644
--- a/lib/AST/DeclCXX.cpp
+++ b/lib/AST/DeclCXX.cpp
@@ -991,7 +991,7 @@ CXXMethodDecl* CXXRecordDecl::getLambdaCallOperator() const {
   if (!isLambda()) return nullptr;
   DeclarationName Name = 
     getASTContext().DeclarationNames.getCXXOperatorName(OO_Call);
-  DeclContext::lookup_const_result Calls = lookup(Name);
+  DeclContext::lookup_result Calls = lookup(Name);
 
   assert(!Calls.empty() && "Missing lambda call operator!");
   assert(Calls.size() == 1 && "More than one lambda call operator!"); 
@@ -1008,7 +1008,7 @@ CXXMethodDecl* CXXRecordDecl::getLambdaStaticInvoker() const {
   if (!isLambda()) return nullptr;
   DeclarationName Name = 
     &getASTContext().Idents.get(getLambdaStaticInvokerName());
-  DeclContext::lookup_const_result Invoker = lookup(Name);
+  DeclContext::lookup_result Invoker = lookup(Name);
   if (Invoker.empty()) return nullptr;
   assert(Invoker.size() == 1 && "More than one static invoker operator!");  
   NamedDecl *InvokerFun = Invoker.front();
@@ -1173,7 +1173,7 @@ static void CollectVisibleConversions(ASTContext &Context,
 
 /// getVisibleConversionFunctions - get all conversion functions visible
 /// in current class; including conversion function templates.
-std::pair<CXXRecordDecl::conversion_iterator,CXXRecordDecl::conversion_iterator>
+llvm::iterator_range<CXXRecordDecl::conversion_iterator>
 CXXRecordDecl::getVisibleConversionFunctions() {
   ASTContext &Ctx = getASTContext();
 
@@ -1189,7 +1189,7 @@ CXXRecordDecl::getVisibleConversionFunctions() {
       data().ComputedVisibleConversions = true;
     }
   }
-  return std::make_pair(Set->begin(), Set->end());
+  return llvm::make_range(Set->begin(), Set->end());
 }
 
 void CXXRecordDecl::removeConversion(const NamedDecl *ConvDecl) {
@@ -1307,7 +1307,7 @@ CXXDestructorDecl *CXXRecordDecl::getDestructor() const {
     = Context.DeclarationNames.getCXXDestructorName(
                                           Context.getCanonicalType(ClassType));
 
-  DeclContext::lookup_const_result R = lookup(Name);
+  DeclContext::lookup_result R = lookup(Name);
   if (R.empty())
     return nullptr;
 
@@ -1418,9 +1418,8 @@ CXXMethodDecl::getCorrespondingMethodInClass(const CXXRecordDecl *RD,
     return nullptr;
   }
 
-  lookup_const_result Candidates = RD->lookup(getDeclName());
-  for (NamedDecl * const * I = Candidates.begin(); I != Candidates.end(); ++I) {
-    CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(*I);
+  for (auto *ND : RD->lookup(getDeclName())) {
+    CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(ND);
     if (!MD)
       continue;
     if (recursivelyOverrides(MD, this))
@@ -1491,8 +1490,8 @@ bool CXXMethodDecl::isUsualDeallocationFunction() const {
                  
   // This function is a usual deallocation function if there are no 
   // single-parameter deallocation functions of the same kind.
-  DeclContext::lookup_const_result R = getDeclContext()->lookup(getDeclName());
-  for (DeclContext::lookup_const_result::iterator I = R.begin(), E = R.end();
+  DeclContext::lookup_result R = getDeclContext()->lookup(getDeclName());
+  for (DeclContext::lookup_result::iterator I = R.begin(), E = R.end();
        I != E; ++I) {
     if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(*I))
       if (FD->getNumParams() == 1)
@@ -1740,6 +1739,10 @@ CXXConstructorDecl::Create(ASTContext &C, CXXRecordDecl *RD,
                                         isImplicitlyDeclared, isConstexpr);
 }
 
+CXXConstructorDecl::init_const_iterator CXXConstructorDecl::init_begin() const {
+  return CtorInitializers.get(getASTContext().getExternalSource());
+}
+
 CXXConstructorDecl *CXXConstructorDecl::getTargetConstructor() const {
   assert(isDelegatingConstructor() && "Not a delegating constructor!");
   Expr *E = (*init_begin())->getInit()->IgnoreImplicit();
@@ -1887,6 +1890,15 @@ CXXDestructorDecl::Create(ASTContext &C, CXXRecordDecl *RD,
                                        isInline, isImplicitlyDeclared);
 }
 
+void CXXDestructorDecl::setOperatorDelete(FunctionDecl *OD) {
+  auto *First = cast<CXXDestructorDecl>(getFirstDecl());
+  if (OD && !First->OperatorDelete) {
+    First->OperatorDelete = OD;
+    if (auto *L = getASTMutationListener())
+      L->ResolvedOperatorDelete(First, OD);
+  }
+}
+
 void CXXConversionDecl::anchor() { }
 
 CXXConversionDecl *
diff --git a/lib/AST/DeclGroup.cpp b/lib/AST/DeclGroup.cpp
index 9861f2278f9a..512837fdf3f4 100644
--- a/lib/AST/DeclGroup.cpp
+++ b/lib/AST/DeclGroup.cpp
@@ -18,6 +18,8 @@
 using namespace clang;
 
 DeclGroup* DeclGroup::Create(ASTContext &C, Decl **Decls, unsigned NumDecls) {
+  static_assert(sizeof(DeclGroup) % llvm::AlignOf<void *>::Alignment == 0,
+                "Trailing data is unaligned!");
   assert(NumDecls > 1 && "Invalid DeclGroup");
   unsigned Size = sizeof(DeclGroup) + sizeof(Decl*) * NumDecls;
   void* Mem = C.Allocate(Size, llvm::AlignOf<DeclGroup>::Alignment);
diff --git a/lib/AST/DeclObjC.cpp b/lib/AST/DeclObjC.cpp
index ed5367514c30..a63ba7e698a8 100644
--- a/lib/AST/DeclObjC.cpp
+++ b/lib/AST/DeclObjC.cpp
@@ -54,8 +54,8 @@ void ObjCContainerDecl::anchor() { }
 ///
 ObjCIvarDecl *
 ObjCContainerDecl::getIvarDecl(IdentifierInfo *Id) const {
-  lookup_const_result R = lookup(Id);
-  for (lookup_const_iterator Ivar = R.begin(), IvarEnd = R.end();
+  lookup_result R = lookup(Id);
+  for (lookup_iterator Ivar = R.begin(), IvarEnd = R.end();
        Ivar != IvarEnd; ++Ivar) {
     if (ObjCIvarDecl *ivar = dyn_cast<ObjCIvarDecl>(*Ivar))
       return ivar;
@@ -83,8 +83,8 @@ ObjCContainerDecl::getMethod(Selector Sel, bool isInstance,
   // + (float) class_method;
   // @end
   //
-  lookup_const_result R = lookup(Sel);
-  for (lookup_const_iterator Meth = R.begin(), MethEnd = R.end();
+  lookup_result R = lookup(Sel);
+  for (lookup_iterator Meth = R.begin(), MethEnd = R.end();
        Meth != MethEnd; ++Meth) {
     ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(*Meth);
     if (MD && MD->isInstanceMethod() == isInstance)
@@ -101,8 +101,8 @@ ObjCContainerDecl::getMethod(Selector Sel, bool isInstance,
 bool ObjCContainerDecl::HasUserDeclaredSetterMethod(
     const ObjCPropertyDecl *Property) const {
   Selector Sel = Property->getSetterName();
-  lookup_const_result R = lookup(Sel);
-  for (lookup_const_iterator Meth = R.begin(), MethEnd = R.end();
+  lookup_result R = lookup(Sel);
+  for (lookup_iterator Meth = R.begin(), MethEnd = R.end();
        Meth != MethEnd; ++Meth) {
     ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(*Meth);
     if (MD && MD->isInstanceMethod() && !MD->isImplicit())
@@ -161,8 +161,8 @@ ObjCPropertyDecl::findPropertyDecl(const DeclContext *DC,
         return nullptr;
   }
 
-  DeclContext::lookup_const_result R = DC->lookup(propertyID);
-  for (DeclContext::lookup_const_iterator I = R.begin(), E = R.end(); I != E;
+  DeclContext::lookup_result R = DC->lookup(propertyID);
+  for (DeclContext::lookup_iterator I = R.begin(), E = R.end(); I != E;
        ++I)
     if (ObjCPropertyDecl *PD = dyn_cast<ObjCPropertyDecl>(*I))
       return PD;
@@ -334,9 +334,8 @@ void ObjCInterfaceDecl::mergeClassExtensionProtocolList(
     return;
 
   // Merge ProtocolRefs into class's protocol list;
-  for (auto *P : all_referenced_protocols()) {
-    ProtocolRefs.push_back(P);
-  }
+  ProtocolRefs.append(all_referenced_protocol_begin(),
+                      all_referenced_protocol_end());
 
   data().AllReferencedProtocols.set(ProtocolRefs.data(), ProtocolRefs.size(),C);
 }
@@ -617,8 +616,7 @@ ObjCMethodDecl *ObjCInterfaceDecl::lookupPrivateMethod(
 
   // Look through local category implementations associated with the class.
   if (!Method)
-    Method = Instance ? getCategoryInstanceMethod(Sel)
-                      : getCategoryClassMethod(Sel);
+    Method = getCategoryMethod(Sel, Instance);
 
   // Before we give up, check if the selector is an instance method.
   // But only in the root. This matches gcc's behavior and what the
@@ -1101,7 +1099,7 @@ ObjCMethodDecl::findPropertyDecl(bool CheckOverrides) const {
   if (NumArgs > 1)
     return nullptr;
 
-  if (!isInstanceMethod() || getMethodFamily() != OMF_None)
+  if (!isInstanceMethod())
     return nullptr;
 
   if (isPropertyAccessor()) {
@@ -1822,6 +1820,11 @@ void ObjCImplementationDecl::setIvarInitializers(ASTContext &C,
   }
 }
 
+ObjCImplementationDecl::init_const_iterator
+ObjCImplementationDecl::init_begin() const {
+  return IvarInitializers.get(getASTContext().getExternalSource());
+}
+
 raw_ostream &clang::operator<<(raw_ostream &OS,
                                const ObjCImplementationDecl &ID) {
   OS << ID.getName();
diff --git a/lib/AST/DeclTemplate.cpp b/lib/AST/DeclTemplate.cpp
index 0d1d2a4613dc..6374a92621a4 100644
--- a/lib/AST/DeclTemplate.cpp
+++ b/lib/AST/DeclTemplate.cpp
@@ -160,17 +160,43 @@ RedeclarableTemplateDecl::CommonBase *RedeclarableTemplateDecl::getCommonPtr() c
   return Common;
 }
 
-template <class EntryType>
-typename RedeclarableTemplateDecl::SpecEntryTraits<EntryType>::DeclType*
+template<class EntryType>
+typename RedeclarableTemplateDecl::SpecEntryTraits<EntryType>::DeclType *
 RedeclarableTemplateDecl::findSpecializationImpl(
-                                 llvm::FoldingSetVector<EntryType> &Specs,
-                                 ArrayRef<TemplateArgument> Args,
-                                 void *&InsertPos) {
+    llvm::FoldingSetVector<EntryType> &Specs, ArrayRef<TemplateArgument> Args,
+    void *&InsertPos) {
   typedef SpecEntryTraits<EntryType> SETraits;
   llvm::FoldingSetNodeID ID;
   EntryType::Profile(ID,Args, getASTContext());
   EntryType *Entry = Specs.FindNodeOrInsertPos(ID, InsertPos);
-  return Entry ? SETraits::getMostRecentDecl(Entry) : nullptr;
+  return Entry ? SETraits::getDecl(Entry)->getMostRecentDecl() : nullptr;
+}
+
+template<class Derived, class EntryType>
+void RedeclarableTemplateDecl::addSpecializationImpl(
+    llvm::FoldingSetVector<EntryType> &Specializations, EntryType *Entry,
+    void *InsertPos) {
+  typedef SpecEntryTraits<EntryType> SETraits;
+  if (InsertPos) {
+#ifndef NDEBUG
+    void *CorrectInsertPos;
+    assert(!findSpecializationImpl(Specializations,
+                                   SETraits::getTemplateArgs(Entry),
+                                   CorrectInsertPos) &&
+           InsertPos == CorrectInsertPos &&
+           "given incorrect InsertPos for specialization");
+#endif
+    Specializations.InsertNode(Entry, InsertPos);
+  } else {
+    EntryType *Existing = Specializations.GetOrInsertNode(Entry);
+    (void)Existing;
+    assert(SETraits::getDecl(Existing)->isCanonicalDecl() &&
+           "non-canonical specialization?");
+  }
+
+  if (ASTMutationListener *L = getASTMutationListener())
+    L->AddedCXXTemplateSpecialization(cast<Derived>(this),
+                                      SETraits::getDecl(Entry));
 }
 
 /// \brief Generate the injected template arguments for the given template
@@ -246,7 +272,11 @@ FunctionTemplateDecl::newCommon(ASTContext &C) const {
 }
 
 void FunctionTemplateDecl::LoadLazySpecializations() const {
-  Common *CommonPtr = getCommonPtr();
+  // Grab the most recent declaration to ensure we've loaded any lazy
+  // redeclarations of this template.
+  //
+  // FIXME: Avoid walking the entire redeclaration chain here.
+  Common *CommonPtr = getMostRecentDecl()->getCommonPtr();
   if (CommonPtr->LazySpecializations) {
     ASTContext &Context = getASTContext();
     uint32_t *Specs = CommonPtr->LazySpecializations;
@@ -270,12 +300,8 @@ FunctionTemplateDecl::findSpecialization(ArrayRef<TemplateArgument> Args,
 
 void FunctionTemplateDecl::addSpecialization(
       FunctionTemplateSpecializationInfo *Info, void *InsertPos) {
-  if (InsertPos)
-    getSpecializations().InsertNode(Info, InsertPos);
-  else
-    getSpecializations().GetOrInsertNode(Info);
-  if (ASTMutationListener *L = getASTMutationListener())
-    L->AddedCXXTemplateSpecialization(this, Info->Function);
+  addSpecializationImpl<FunctionTemplateDecl>(getSpecializations(), Info,
+                                              InsertPos);
 }
 
 ArrayRef<TemplateArgument> FunctionTemplateDecl::getInjectedTemplateArgs() {
@@ -320,7 +346,11 @@ ClassTemplateDecl *ClassTemplateDecl::CreateDeserialized(ASTContext &C,
 }
 
 void ClassTemplateDecl::LoadLazySpecializations() const {
-  Common *CommonPtr = getCommonPtr();
+  // Grab the most recent declaration to ensure we've loaded any lazy
+  // redeclarations of this template.
+  //
+  // FIXME: Avoid walking the entire redeclaration chain here.
+  Common *CommonPtr = getMostRecentDecl()->getCommonPtr();
   if (CommonPtr->LazySpecializations) {
     ASTContext &Context = getASTContext();
     uint32_t *Specs = CommonPtr->LazySpecializations;
@@ -357,16 +387,7 @@ ClassTemplateDecl::findSpecialization(ArrayRef<TemplateArgument> Args,
 
 void ClassTemplateDecl::AddSpecialization(ClassTemplateSpecializationDecl *D,
                                           void *InsertPos) {
-  if (InsertPos)
-    getSpecializations().InsertNode(D, InsertPos);
-  else {
-    ClassTemplateSpecializationDecl *Existing 
-      = getSpecializations().GetOrInsertNode(D);
-    (void)Existing;
-    assert(Existing->isCanonicalDecl() && "Non-canonical specialization?");
-  }
-  if (ASTMutationListener *L = getASTMutationListener())
-    L->AddedCXXTemplateSpecialization(this, D);
+  addSpecializationImpl<ClassTemplateDecl>(getSpecializations(), D, InsertPos);
 }
 
 ClassTemplatePartialSpecializationDecl *
@@ -953,7 +974,11 @@ VarTemplateDecl *VarTemplateDecl::CreateDeserialized(ASTContext &C,
 // TODO: Unify across class, function and variable templates?
 //       May require moving this and Common to RedeclarableTemplateDecl.
 void VarTemplateDecl::LoadLazySpecializations() const {
-  Common *CommonPtr = getCommonPtr();
+  // Grab the most recent declaration to ensure we've loaded any lazy
+  // redeclarations of this template.
+  //
+  // FIXME: Avoid walking the entire redeclaration chain here.
+  Common *CommonPtr = getMostRecentDecl()->getCommonPtr();
   if (CommonPtr->LazySpecializations) {
     ASTContext &Context = getASTContext();
     uint32_t *Specs = CommonPtr->LazySpecializations;
@@ -990,16 +1015,7 @@ VarTemplateDecl::findSpecialization(ArrayRef<TemplateArgument> Args,
 
 void VarTemplateDecl::AddSpecialization(VarTemplateSpecializationDecl *D,
                                         void *InsertPos) {
-  if (InsertPos)
-    getSpecializations().InsertNode(D, InsertPos);
-  else {
-    VarTemplateSpecializationDecl *Existing =
-        getSpecializations().GetOrInsertNode(D);
-    (void)Existing;
-    assert(Existing->isCanonicalDecl() && "Non-canonical specialization?");
-  }
-  if (ASTMutationListener *L = getASTMutationListener())
-    L->AddedCXXTemplateSpecialization(this, D);
+  addSpecializationImpl<VarTemplateDecl>(getSpecializations(), D, InsertPos);
 }
 
 VarTemplatePartialSpecializationDecl *
diff --git a/lib/AST/Expr.cpp b/lib/AST/Expr.cpp
index 712de5056e8d..76a4da2371b5 100644
--- a/lib/AST/Expr.cpp
+++ b/lib/AST/Expr.cpp
@@ -1141,40 +1141,13 @@ CallExpr::CallExpr(const ASTContext& C, StmtClass SC, Expr *fn,
   RParenLoc = rparenloc;
 }
 
-CallExpr::CallExpr(const ASTContext& C, Expr *fn, ArrayRef<Expr*> args,
+CallExpr::CallExpr(const ASTContext &C, Expr *fn, ArrayRef<Expr *> args,
                    QualType t, ExprValueKind VK, SourceLocation rparenloc)
-  : Expr(CallExprClass, t, VK, OK_Ordinary,
-         fn->isTypeDependent(),
-         fn->isValueDependent(),
-         fn->isInstantiationDependent(),
-         fn->containsUnexpandedParameterPack()),
-    NumArgs(args.size()) {
-
-  SubExprs = new (C) Stmt*[args.size()+PREARGS_START];
-  SubExprs[FN] = fn;
-  for (unsigned i = 0; i != args.size(); ++i) {
-    if (args[i]->isTypeDependent())
-      ExprBits.TypeDependent = true;
-    if (args[i]->isValueDependent())
-      ExprBits.ValueDependent = true;
-    if (args[i]->isInstantiationDependent())
-      ExprBits.InstantiationDependent = true;
-    if (args[i]->containsUnexpandedParameterPack())
-      ExprBits.ContainsUnexpandedParameterPack = true;
-
-    SubExprs[i+PREARGS_START] = args[i];
-  }
-
-  CallExprBits.NumPreArgs = 0;
-  RParenLoc = rparenloc;
+    : CallExpr(C, CallExprClass, fn, /*NumPreArgs=*/0, args, t, VK, rparenloc) {
 }
 
 CallExpr::CallExpr(const ASTContext &C, StmtClass SC, EmptyShell Empty)
-  : Expr(SC, Empty), SubExprs(nullptr), NumArgs(0) {
-  // FIXME: Why do we allocate this?
-  SubExprs = new (C) Stmt*[PREARGS_START];
-  CallExprBits.NumPreArgs = 0;
-}
+    : CallExpr(C, SC, /*NumPreArgs=*/0, Empty) {}
 
 CallExpr::CallExpr(const ASTContext &C, StmtClass SC, unsigned NumPreArgs,
                    EmptyShell Empty)
@@ -1271,16 +1244,21 @@ bool CallExpr::isUnevaluatedBuiltinCall(ASTContext &Ctx) const {
   return false;
 }
 
-QualType CallExpr::getCallReturnType() const {
-  QualType CalleeType = getCallee()->getType();
-  if (const PointerType *FnTypePtr = CalleeType->getAs<PointerType>())
+QualType CallExpr::getCallReturnType(const ASTContext &Ctx) const {
+  const Expr *Callee = getCallee();
+  QualType CalleeType = Callee->getType();
+  if (const auto *FnTypePtr = CalleeType->getAs<PointerType>()) {
     CalleeType = FnTypePtr->getPointeeType();
-  else if (const BlockPointerType *BPT = CalleeType->getAs<BlockPointerType>())
+  } else if (const auto *BPT = CalleeType->getAs<BlockPointerType>()) {
     CalleeType = BPT->getPointeeType();
-  else if (CalleeType->isSpecificPlaceholderType(BuiltinType::BoundMember))
+  } else if (CalleeType->isSpecificPlaceholderType(BuiltinType::BoundMember)) {
+    if (isa<CXXPseudoDestructorExpr>(Callee->IgnoreParens()))
+      return Ctx.VoidTy;
+
     // This should never be overloaded and so should never return null.
-    CalleeType = Expr::findBoundMemberType(getCallee());
-    
+    CalleeType = Expr::findBoundMemberType(Callee);
+  }
+
   const FunctionType *FnType = CalleeType->castAs<FunctionType>();
   return FnType->getReturnType();
 }
@@ -1360,16 +1338,50 @@ IdentifierInfo *OffsetOfExpr::OffsetOfNode::getFieldName() const {
   return reinterpret_cast<IdentifierInfo *> (Data & ~(uintptr_t)Mask);
 }
 
-MemberExpr *MemberExpr::Create(const ASTContext &C, Expr *base, bool isarrow,
-                               NestedNameSpecifierLoc QualifierLoc,
-                               SourceLocation TemplateKWLoc,
-                               ValueDecl *memberdecl,
-                               DeclAccessPair founddecl,
-                               DeclarationNameInfo nameinfo,
-                               const TemplateArgumentListInfo *targs,
-                               QualType ty,
-                               ExprValueKind vk,
-                               ExprObjectKind ok) {
+UnaryExprOrTypeTraitExpr::UnaryExprOrTypeTraitExpr(
+    UnaryExprOrTypeTrait ExprKind, Expr *E, QualType resultType,
+    SourceLocation op, SourceLocation rp)
+    : Expr(UnaryExprOrTypeTraitExprClass, resultType, VK_RValue, OK_Ordinary,
+           false, // Never type-dependent (C++ [temp.dep.expr]p3).
+           // Value-dependent if the argument is type-dependent.
+           E->isTypeDependent(), E->isInstantiationDependent(),
+           E->containsUnexpandedParameterPack()),
+      OpLoc(op), RParenLoc(rp) {
+  UnaryExprOrTypeTraitExprBits.Kind = ExprKind;
+  UnaryExprOrTypeTraitExprBits.IsType = false;
+  Argument.Ex = E;
+
+  // Check to see if we are in the situation where alignof(decl) should be
+  // dependent because decl's alignment is dependent.
+  if (ExprKind == UETT_AlignOf) {
+    if (!isValueDependent() || !isInstantiationDependent()) {
+      E = E->IgnoreParens();
+
+      const ValueDecl *D = nullptr;
+      if (const auto *DRE = dyn_cast<DeclRefExpr>(E))
+        D = DRE->getDecl();
+      else if (const auto *ME = dyn_cast<MemberExpr>(E))
+        D = ME->getMemberDecl();
+
+      if (D) {
+        for (const auto *I : D->specific_attrs<AlignedAttr>()) {
+          if (I->isAlignmentDependent()) {
+            setValueDependent(true);
+            setInstantiationDependent(true);
+            break;
+          }
+        }
+      }
+    }
+  }
+}
+
+MemberExpr *MemberExpr::Create(
+    const ASTContext &C, Expr *base, bool isarrow, SourceLocation OperatorLoc,
+    NestedNameSpecifierLoc QualifierLoc, SourceLocation TemplateKWLoc,
+    ValueDecl *memberdecl, DeclAccessPair founddecl,
+    DeclarationNameInfo nameinfo, const TemplateArgumentListInfo *targs,
+    QualType ty, ExprValueKind vk, ExprObjectKind ok) {
   std::size_t Size = sizeof(MemberExpr);
 
   bool hasQualOrFound = (QualifierLoc ||
@@ -1384,8 +1396,8 @@ MemberExpr *MemberExpr::Create(const ASTContext &C, Expr *base, bool isarrow,
     Size += ASTTemplateKWAndArgsInfo::sizeFor(0);
 
   void *Mem = C.Allocate(Size, llvm::alignOf<MemberExpr>());
-  MemberExpr *E = new (Mem) MemberExpr(base, isarrow, memberdecl, nameinfo,
-                                       ty, vk, ok);
+  MemberExpr *E = new (Mem)
+      MemberExpr(base, isarrow, OperatorLoc, memberdecl, nameinfo, ty, vk, ok);
 
   if (hasQualOrFound) {
     // FIXME: Wrong. We should be looking at the member declaration we found.
@@ -2132,8 +2144,8 @@ bool Expr::isUnusedResultAWarning(const Expr *&WarnE, SourceLocation &Loc,
     case OO_Greater:
     case OO_GreaterEqual:
     case OO_LessEqual:
-      if (Op->getCallReturnType()->isReferenceType() ||
-          Op->getCallReturnType()->isVoidType())
+      if (Op->getCallReturnType(Ctx)->isReferenceType() ||
+          Op->getCallReturnType(Ctx)->isVoidType())
         break;
       WarnE = this;
       Loc = Op->getOperatorLoc();
@@ -2149,12 +2161,16 @@ bool Expr::isUnusedResultAWarning(const Expr *&WarnE, SourceLocation &Loc,
     // If this is a direct call, get the callee.
     const CallExpr *CE = cast<CallExpr>(this);
     if (const Decl *FD = CE->getCalleeDecl()) {
+      const FunctionDecl *Func = dyn_cast<FunctionDecl>(FD);
+      bool HasWarnUnusedResultAttr = Func ? Func->hasUnusedResultAttr()
+                                          : FD->hasAttr<WarnUnusedResultAttr>();
+
       // If the callee has attribute pure, const, or warn_unused_result, warn
       // about it. void foo() { strlen("bar"); } should warn.
       //
       // Note: If new cases are added here, DiagnoseUnusedExprResult should be
       // updated to match for QoI.
-      if (FD->hasAttr<WarnUnusedResultAttr>() ||
+      if (HasWarnUnusedResultAttr ||
           FD->hasAttr<PureAttr>() || FD->hasAttr<ConstAttr>()) {
         WarnE = this;
         Loc = CE->getCallee()->getLocStart();
@@ -2200,9 +2216,7 @@ bool Expr::isUnusedResultAWarning(const Expr *&WarnE, SourceLocation &Loc,
     }
 
     if (const ObjCMethodDecl *MD = ME->getMethodDecl())
-      if (MD->hasAttr<WarnUnusedResultAttr>() ||
-          (MD->isPropertyAccessor() && !MD->getReturnType()->isVoidType() &&
-           !ME->getReceiverType()->isObjCIdType())) {
+      if (MD->hasAttr<WarnUnusedResultAttr>()) {
         WarnE = this;
         Loc = getExprLoc();
         return true;
@@ -2387,7 +2401,7 @@ QualType Expr::findBoundMemberType(const Expr *expr) {
     return type;
   }
 
-  assert(isa<UnresolvedMemberExpr>(expr));
+  assert(isa<UnresolvedMemberExpr>(expr) || isa<CXXPseudoDestructorExpr>(expr));
   return QualType();
 }
 
@@ -2932,11 +2946,19 @@ bool Expr::HasSideEffects(const ASTContext &Ctx,
   case CXXOperatorCallExprClass:
   case CXXMemberCallExprClass:
   case CUDAKernelCallExprClass:
+  case UserDefinedLiteralClass: {
+    // We don't know a call definitely has side effects, except for calls
+    // to pure/const functions that definitely don't.
+    // If the call itself is considered side-effect free, check the operands.
+    const Decl *FD = cast<CallExpr>(this)->getCalleeDecl();
+    bool IsPure = FD && (FD->hasAttr<ConstAttr>() || FD->hasAttr<PureAttr>());
+    if (IsPure || !IncludePossibleEffects)
+      break;
+    return true;
+  }
+
   case BlockExprClass:
   case CXXBindTemporaryExprClass:
-  case UserDefinedLiteralClass:
-    // We don't know a call definitely has side effects, but we can check the
-    // call's operands.
     if (!IncludePossibleEffects)
       break;
     return true;
diff --git a/lib/AST/ExprCXX.cpp b/lib/AST/ExprCXX.cpp
index 93361666183b..d6f2ce63a0a5 100644
--- a/lib/AST/ExprCXX.cpp
+++ b/lib/AST/ExprCXX.cpp
@@ -208,8 +208,9 @@ void CXXNewExpr::AllocateArgsArray(const ASTContext &C, bool isArray,
 }
 
 bool CXXNewExpr::shouldNullCheckAllocation(const ASTContext &Ctx) const {
-  return getOperatorNew()->getType()->
-    castAs<FunctionProtoType>()->isNothrow(Ctx);
+  return getOperatorNew()->getType()->castAs<FunctionProtoType>()->isNothrow(
+             Ctx) &&
+         !getOperatorNew()->isReservedGlobalPlacementOperator();
 }
 
 // CXXDeleteExpr
@@ -237,10 +238,7 @@ CXXPseudoDestructorExpr::CXXPseudoDestructorExpr(const ASTContext &Context,
                 SourceLocation ColonColonLoc, SourceLocation TildeLoc, 
                 PseudoDestructorTypeStorage DestroyedType)
   : Expr(CXXPseudoDestructorExprClass,
-         Context.getPointerType(Context.getFunctionType(
-             Context.VoidTy, None,
-             FunctionProtoType::ExtProtoInfo(
-                 Context.getDefaultCallingConvention(false, true)))),
+         Context.BoundMemberTy,
          VK_RValue, OK_Ordinary,
          /*isTypeDependent=*/(Base->isTypeDependent() ||
            (DestroyedType.getTypeSourceInfo() &&
@@ -359,8 +357,7 @@ OverloadExpr::OverloadExpr(StmtClass K, const ASTContext &C,
     Results = static_cast<DeclAccessPair *>(
                                 C.Allocate(sizeof(DeclAccessPair) * NumResults, 
                                            llvm::alignOf<DeclAccessPair>()));
-    memcpy(Results, &*Begin.getIterator(), 
-           NumResults * sizeof(DeclAccessPair));
+    memcpy(Results, Begin.I, NumResults * sizeof(DeclAccessPair));
   }
 
   // If we have explicit template arguments, check for dependent
@@ -401,8 +398,7 @@ void OverloadExpr::initializeResults(const ASTContext &C,
                                C.Allocate(sizeof(DeclAccessPair) * NumResults,
  
                                           llvm::alignOf<DeclAccessPair>()));
-     memcpy(Results, &*Begin.getIterator(), 
-            NumResults * sizeof(DeclAccessPair));
+     memcpy(Results, Begin.I, NumResults * sizeof(DeclAccessPair));
   }
 }
 
@@ -1031,6 +1027,11 @@ LambdaExpr *LambdaExpr::CreateDeserialized(const ASTContext &C,
   return new (Mem) LambdaExpr(EmptyShell(), NumCaptures, NumArrayIndexVars > 0);
 }
 
+bool LambdaExpr::isInitCapture(const LambdaCapture *C) const {
+  return (C->capturesVariable() && C->getCapturedVar()->isInitCapture() &&
+          (getCallOperator() == C->getCapturedVar()->getDeclContext()));
+}
+
 LambdaExpr::capture_iterator LambdaExpr::capture_begin() const {
   return getLambdaClass()->getLambdaData().Captures;
 }
diff --git a/lib/AST/ExprClassification.cpp b/lib/AST/ExprClassification.cpp
index 933ea97fa2ba..5b320c2694ab 100644
--- a/lib/AST/ExprClassification.cpp
+++ b/lib/AST/ExprClassification.cpp
@@ -283,7 +283,7 @@ static Cl::Kinds ClassifyInternal(ASTContext &Ctx, const Expr *E) {
   case Expr::CXXMemberCallExprClass:
   case Expr::UserDefinedLiteralClass:
   case Expr::CUDAKernelCallExprClass:
-    return ClassifyUnnamed(Ctx, cast<CallExpr>(E)->getCallReturnType());
+    return ClassifyUnnamed(Ctx, cast<CallExpr>(E)->getCallReturnType(Ctx));
 
     // __builtin_choose_expr is equivalent to the chosen expression.
   case Expr::ChooseExprClass:
@@ -418,9 +418,10 @@ static Cl::Kinds ClassifyDecl(ASTContext &Ctx, const Decl *D) {
     islvalue = NTTParm->getType()->isReferenceType();
   else
     islvalue = isa<VarDecl>(D) || isa<FieldDecl>(D) ||
-	  isa<IndirectFieldDecl>(D) ||
-      (Ctx.getLangOpts().CPlusPlus &&
-        (isa<FunctionDecl>(D) || isa<FunctionTemplateDecl>(D)));
+               isa<IndirectFieldDecl>(D) ||
+               (Ctx.getLangOpts().CPlusPlus &&
+                (isa<FunctionDecl>(D) || isa<MSPropertyDecl>(D) ||
+                 isa<FunctionTemplateDecl>(D)));
 
   return islvalue ? Cl::CL_LValue : Cl::CL_PRValue;
 }
@@ -605,7 +606,7 @@ static Cl::ModifiableType IsModifiable(ASTContext &Ctx, const Expr *E,
   if (CT.isConstQualified())
     return Cl::CM_ConstQualified;
   if (CT.getQualifiers().getAddressSpace() == LangAS::opencl_constant)
-    return Cl::CM_ConstQualified;
+    return Cl::CM_ConstAddrSpace;
 
   // Arrays are not modifiable, only their elements are.
   if (CT->isArrayType())
@@ -671,6 +672,7 @@ Expr::isModifiableLvalue(ASTContext &Ctx, SourceLocation *Loc) const {
     llvm_unreachable("CM_LValueCast and CL_LValue don't match");
   case Cl::CM_NoSetterProperty: return MLV_NoSetterProperty;
   case Cl::CM_ConstQualified: return MLV_ConstQualified;
+  case Cl::CM_ConstAddrSpace: return MLV_ConstAddrSpace;
   case Cl::CM_ArrayType: return MLV_ArrayType;
   case Cl::CM_IncompleteType: return MLV_IncompleteType;
   }
diff --git a/lib/AST/ExprConstant.cpp b/lib/AST/ExprConstant.cpp
index 3d7f2dca7a2f..d1ec7aea1d2a 100644
--- a/lib/AST/ExprConstant.cpp
+++ b/lib/AST/ExprConstant.cpp
@@ -1406,7 +1406,7 @@ static bool CheckConstantExpression(EvalInfo &Info, SourceLocation DiagLoc,
   return true;
 }
 
-const ValueDecl *GetLValueBaseDecl(const LValue &LVal) {
+static const ValueDecl *GetLValueBaseDecl(const LValue &LVal) {
   return LVal.Base.dyn_cast<const ValueDecl*>();
 }
 
@@ -2173,7 +2173,7 @@ struct CompleteObject {
     assert(Value && "missing value for complete object");
   }
 
-  LLVM_EXPLICIT operator bool() const { return Value; }
+  explicit operator bool() const { return Value; }
 };
 
 /// Find the designated sub-object of an rvalue.
@@ -2502,8 +2502,9 @@ static bool AreElementsOfSameArray(QualType ObjType,
 }
 
 /// Find the complete object to which an LValue refers.
-CompleteObject findCompleteObject(EvalInfo &Info, const Expr *E, AccessKinds AK,
-                                  const LValue &LVal, QualType LValType) {
+static CompleteObject findCompleteObject(EvalInfo &Info, const Expr *E,
+                                         AccessKinds AK, const LValue &LVal,
+                                         QualType LValType) {
   if (!LVal.Base) {
     Info.Diag(E, diag::note_constexpr_access_null) << AK;
     return CompleteObject();
@@ -3726,8 +3727,9 @@ static bool HandleFunctionCall(SourceLocation CallLoc,
   // Skip this for non-union classes with no fields; in that case, the defaulted
   // copy/move does not actually read the object.
   const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(Callee);
-  if (MD && MD->isDefaulted() && MD->isTrivial() &&
-      (MD->getParent()->isUnion() || hasFields(MD->getParent()))) {
+  if (MD && MD->isDefaulted() &&
+      (MD->getParent()->isUnion() ||
+       (MD->isTrivial() && hasFields(MD->getParent())))) {
     assert(This &&
            (MD->isCopyAssignmentOperator() || MD->isMoveAssignmentOperator()));
     LValue RHS;
@@ -3791,11 +3793,9 @@ static bool HandleConstructorCall(SourceLocation CallLoc, const LValue &This,
   // Skip this for empty non-union classes; we should not perform an
   // lvalue-to-rvalue conversion on them because their copy constructor does not
   // actually read them.
-  if (Definition->isDefaulted() &&
-      ((Definition->isCopyConstructor() && Definition->isTrivial()) ||
-       (Definition->isMoveConstructor() && Definition->isTrivial())) &&
+  if (Definition->isDefaulted() && Definition->isCopyOrMoveConstructor() &&
       (Definition->getParent()->isUnion() ||
-       hasFields(Definition->getParent()))) {
+       (Definition->isTrivial() && hasFields(Definition->getParent())))) {
     LValue RHS;
     RHS.setFrom(Info.Ctx, ArgValues[0]);
     return handleLValueToRValueConversion(Info, Args[0], Args[0]->getType(),
@@ -6834,7 +6834,7 @@ void DataRecursiveIntBinOpEvaluator::process(EvalResult &Result) {
 }
 
 bool IntExprEvaluator::VisitBinaryOperator(const BinaryOperator *E) {
-  if (E->isAssignmentOp())
+  if (!Info.keepEvaluatingAfterFailure() && E->isAssignmentOp())
     return Error(E);
 
   if (DataRecursiveIntBinOpEvaluator::shouldEnqueue(E))
@@ -6846,7 +6846,11 @@ bool IntExprEvaluator::VisitBinaryOperator(const BinaryOperator *E) {
   if (LHSTy->isAnyComplexType() || RHSTy->isAnyComplexType()) {
     ComplexValue LHS, RHS;
     bool LHSOK;
-    if (E->getLHS()->getType()->isRealFloatingType()) {
+    if (E->isAssignmentOp()) {
+      LValue LV;
+      EvaluateLValue(E->getLHS(), LV, Info);
+      LHSOK = false;
+    } else if (LHSTy->isRealFloatingType()) {
       LHSOK = EvaluateFloat(E->getLHS(), LHS.FloatReal, Info);
       if (LHSOK) {
         LHS.makeComplexFloat();
@@ -7586,10 +7590,23 @@ static bool TryEvaluateBuiltinNaN(const ASTContext &Context,
   else if (S->getString().getAsInteger(0, fill))
     return false;
 
-  if (SNaN)
-    Result = llvm::APFloat::getSNaN(Sem, false, &fill);
-  else
-    Result = llvm::APFloat::getQNaN(Sem, false, &fill);
+  if (Context.getTargetInfo().isNan2008()) {
+    if (SNaN)
+      Result = llvm::APFloat::getSNaN(Sem, false, &fill);
+    else
+      Result = llvm::APFloat::getQNaN(Sem, false, &fill);
+  } else {
+    // Prior to IEEE 754-2008, architectures were allowed to choose whether
+    // the first bit of their significand was set for qNaN or sNaN. MIPS chose
+    // a different encoding to what became a standard in 2008, and for pre-
+    // 2008 revisions, MIPS interpreted sNaN-2008 as qNan and qNaN-2008 as
+    // sNaN. This is now known as "legacy NaN" encoding.
+    if (SNaN)
+      Result = llvm::APFloat::getQNaN(Sem, false, &fill);
+    else
+      Result = llvm::APFloat::getSNaN(Sem, false, &fill);
+  }
+
   return true;
 }
 
diff --git a/lib/AST/ExternalASTSource.cpp b/lib/AST/ExternalASTSource.cpp
index 88941075dd06..730842a28f6a 100644
--- a/lib/AST/ExternalASTSource.cpp
+++ b/lib/AST/ExternalASTSource.cpp
@@ -66,6 +66,11 @@ Stmt *ExternalASTSource::GetExternalDeclStmt(uint64_t Offset) {
   return nullptr;
 }
 
+CXXCtorInitializer **
+ExternalASTSource::GetExternalCXXCtorInitializers(uint64_t Offset) {
+  return nullptr;
+}
+
 CXXBaseSpecifier *
 ExternalASTSource::GetExternalCXXBaseSpecifiers(uint64_t Offset) {
   return nullptr;
diff --git a/lib/AST/InheritViz.cpp b/lib/AST/InheritViz.cpp
index eb3020c5f518..0b82da133fa7 100644
--- a/lib/AST/InheritViz.cpp
+++ b/lib/AST/InheritViz.cpp
@@ -22,11 +22,9 @@
 #include "llvm/Support/raw_ostream.h"
 #include <map>
 #include <set>
+using namespace clang;
 
-using namespace llvm;
-
-namespace clang {
-
+namespace {
 /// InheritanceHierarchyWriter - Helper class that writes out a
 /// GraphViz file that diagrams the inheritance hierarchy starting at
 /// a given C++ class type. Note that we do not use LLVM's
@@ -44,7 +42,8 @@ public:
     : Context(Context), Out(Out) { }
 
   void WriteGraph(QualType Type) {
-    Out << "digraph \"" << DOT::EscapeString(Type.getAsString()) << "\" {\n";
+    Out << "digraph \"" << llvm::DOT::EscapeString(Type.getAsString())
+        << "\" {\n";
     WriteNode(Type, false);
     Out << "}\n";
   }
@@ -59,6 +58,7 @@ protected:
   /// (only) virtual base.
   raw_ostream& WriteNodeReference(QualType Type, bool FromVirtual);
 };
+} // namespace
 
 void InheritanceHierarchyWriter::WriteNode(QualType Type, bool FromVirtual) {
   QualType CanonType = Context.getCanonicalType(Type);
@@ -78,7 +78,7 @@ void InheritanceHierarchyWriter::WriteNode(QualType Type, bool FromVirtual) {
 
   // Give the node a label based on the name of the class.
   std::string TypeName = Type.getAsString();
-  Out << " [ shape=\"box\", label=\"" << DOT::EscapeString(TypeName);
+  Out << " [ shape=\"box\", label=\"" << llvm::DOT::EscapeString(TypeName);
 
   // If the name of the class was a typedef or something different
   // from the "real" class name, show the real class name in
@@ -139,9 +139,8 @@ void CXXRecordDecl::viewInheritance(ASTContext& Context) const {
 
   int FD;
   SmallString<128> Filename;
-  std::error_code EC =
-      sys::fs::createTemporaryFile(Self.getAsString(), "dot", FD, Filename);
-  if (EC) {
+  if (std::error_code EC = llvm::sys::fs::createTemporaryFile(
+          Self.getAsString(), "dot", FD, Filename)) {
     llvm::errs() << "Error: " << EC.message() << "\n";
     return;
   }
@@ -159,5 +158,3 @@ void CXXRecordDecl::viewInheritance(ASTContext& Context) const {
   // Display the graph
   DisplayGraph(Filename);
 }
-
-}
diff --git a/lib/AST/ItaniumCXXABI.cpp b/lib/AST/ItaniumCXXABI.cpp
index 378121c8e5b9..7503cbfc9805 100644
--- a/lib/AST/ItaniumCXXABI.cpp
+++ b/lib/AST/ItaniumCXXABI.cpp
@@ -106,7 +106,7 @@ public:
     TargetInfo::IntType PtrDiff = Target.getPtrDiffType(0);
     uint64_t Width = Target.getTypeWidth(PtrDiff);
     unsigned Align = Target.getTypeAlign(PtrDiff);
-    if (MPT->getPointeeType()->isFunctionType())
+    if (MPT->isMemberFunctionPointer())
       Width = 2 * Width;
     return std::make_pair(Width, Align);
   }
@@ -133,6 +133,22 @@ public:
     return Layout.getNonVirtualSize() == PointerSize;
   }
 
+  const CXXConstructorDecl *
+  getCopyConstructorForExceptionObject(CXXRecordDecl *RD) override {
+    return nullptr;
+  }
+
+  void addCopyConstructorForExceptionObject(CXXRecordDecl *RD,
+                                            CXXConstructorDecl *CD) override {}
+
+  void addDefaultArgExprForConstructor(const CXXConstructorDecl *CD,
+                                       unsigned ParmIdx, Expr *DAE) override {}
+
+  Expr *getDefaultArgExprForConstructor(const CXXConstructorDecl *CD,
+                                        unsigned ParmIdx) override {
+    return nullptr;
+  }
+
   MangleNumberingContext *createMangleNumberingContext() const override {
     return new ItaniumNumberingContext();
   }
diff --git a/lib/AST/ItaniumMangle.cpp b/lib/AST/ItaniumMangle.cpp
index 156ad646fa6d..d07efaee7bba 100644
--- a/lib/AST/ItaniumMangle.cpp
+++ b/lib/AST/ItaniumMangle.cpp
@@ -42,8 +42,8 @@ using namespace clang;
 
 namespace {
 
-/// \brief Retrieve the declaration context that should be used when mangling 
-/// the given declaration.
+/// Retrieve the declaration context that should be used when mangling the given
+/// declaration.
 static const DeclContext *getEffectiveDeclContext(const Decl *D) {
   // The ABI assumes that lambda closure types that occur within 
   // default arguments live in the context of the function. However, due to
@@ -69,6 +69,14 @@ static const DeclContext *getEffectiveDeclContext(const Decl *D) {
   if (const CapturedDecl *CD = dyn_cast<CapturedDecl>(DC))
     return getEffectiveDeclContext(CD);
 
+  if (const auto *VD = dyn_cast<VarDecl>(D))
+    if (VD->isExternC())
+      return VD->getASTContext().getTranslationUnitDecl();
+
+  if (const auto *FD = dyn_cast<FunctionDecl>(D))
+    if (FD->isExternC())
+      return FD->getASTContext().getTranslationUnitDecl();
+
   return DC;
 }
 
@@ -156,12 +164,18 @@ public:
   void mangleDynamicInitializer(const VarDecl *D, raw_ostream &Out) override;
   void mangleDynamicAtExitDestructor(const VarDecl *D,
                                      raw_ostream &Out) override;
+  void mangleSEHFilterExpression(const NamedDecl *EnclosingDecl,
+                                 raw_ostream &Out) override;
+  void mangleSEHFinallyBlock(const NamedDecl *EnclosingDecl,
+                             raw_ostream &Out) override;
   void mangleItaniumThreadLocalInit(const VarDecl *D, raw_ostream &) override;
   void mangleItaniumThreadLocalWrapper(const VarDecl *D,
                                        raw_ostream &) override;
 
   void mangleStringLiteral(const StringLiteral *, raw_ostream &) override;
 
+  void mangleCXXVTableBitSet(const CXXRecordDecl *RD, raw_ostream &) override;
+
   bool getNextDiscriminator(const NamedDecl *ND, unsigned &disc) {
     // Lambda closure types are already numbered.
     if (isLambda(ND))
@@ -196,7 +210,7 @@ public:
   /// @}
 };
 
-/// CXXNameMangler - Manage the mangling of a single name.
+/// Manage the mangling of a single name.
 class CXXNameMangler {
   ItaniumMangleContextImpl &Context;
   raw_ostream &Out;
@@ -207,7 +221,7 @@ class CXXNameMangler {
   const NamedDecl *Structor;
   unsigned StructorType;
 
-  /// SeqID - The next subsitution sequence number.
+  /// The next substitution sequence number.
   unsigned SeqID;
 
   class FunctionTypeDepthState {
@@ -284,7 +298,7 @@ public:
 #endif
   raw_ostream &getStream() { return Out; }
 
-  void mangle(const NamedDecl *D, StringRef Prefix = "_Z");
+  void mangle(const NamedDecl *D);
   void mangleCallOffset(int64_t NonVirtual, int64_t Virtual);
   void mangleNumber(const llvm::APSInt &I);
   void mangleNumber(int64_t Number);
@@ -317,10 +331,8 @@ private:
   void addSubstitution(uintptr_t Ptr);
 
   void mangleUnresolvedPrefix(NestedNameSpecifier *qualifier,
-                              NamedDecl *firstQualifierLookup,
                               bool recursive = false);
   void mangleUnresolvedName(NestedNameSpecifier *qualifier,
-                            NamedDecl *firstQualifierLookup,
                             DeclarationName name,
                             unsigned KnownArity = UnknownArity);
 
@@ -350,6 +362,9 @@ private:
   void manglePrefix(QualType type);
   void mangleTemplatePrefix(const TemplateDecl *ND, bool NoFunction=false);
   void mangleTemplatePrefix(TemplateName Template);
+  bool mangleUnresolvedTypeOrSimpleId(QualType DestroyedType,
+                                      StringRef Prefix = "");
+  void mangleOperatorName(DeclarationName Name, unsigned Arity);
   void mangleOperatorName(OverloadedOperatorKind OO, unsigned Arity);
   void mangleQualifiers(Qualifiers Quals);
   void mangleRefQualifier(RefQualifierKind RefQualifier);
@@ -370,12 +385,14 @@ private:
   void mangleAArch64NeonVectorType(const VectorType *T);
 
   void mangleIntegerLiteral(QualType T, const llvm::APSInt &Value);
+  void mangleMemberExprBase(const Expr *base, bool isArrow);
   void mangleMemberExpr(const Expr *base, bool isArrow,
                         NestedNameSpecifier *qualifier,
                         NamedDecl *firstQualifierLookup,
                         DeclarationName name,
                         unsigned knownArity);
   void mangleCastExpression(const Expr *E, StringRef CastEncoding);
+  void mangleInitListElements(const InitListExpr *InitList);
   void mangleExpression(const Expr *E, unsigned Arity = UnknownArity);
   void mangleCXXCtorType(CXXCtorType T);
   void mangleCXXDtorType(CXXDtorType T);
@@ -439,11 +456,11 @@ bool ItaniumMangleContextImpl::shouldMangleCXXName(const NamedDecl *D) {
   return true;
 }
 
-void CXXNameMangler::mangle(const NamedDecl *D, StringRef Prefix) {
+void CXXNameMangler::mangle(const NamedDecl *D) {
   // <mangled-name> ::= _Z <encoding>
   //            ::= <data name>
   //            ::= <special-name>
-  Out << Prefix;
+  Out << "_Z";
   if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D))
     mangleFunctionEncoding(FD);
   else if (const VarDecl *VD = dyn_cast<VarDecl>(D))
@@ -519,7 +536,7 @@ static const DeclContext *IgnoreLinkageSpecDecls(const DeclContext *DC) {
   return DC;
 }
 
-/// isStd - Return whether a given namespace is the 'std' namespace.
+/// Return whether a given namespace is the 'std' namespace.
 static bool isStd(const NamespaceDecl *NS) {
   if (!IgnoreLinkageSpecDecls(getEffectiveParentContext(NS))
                                 ->isTranslationUnit())
@@ -748,8 +765,7 @@ void CXXNameMangler::mangleCallOffset(int64_t NonVirtual, int64_t Virtual) {
 }
 
 void CXXNameMangler::manglePrefix(QualType type) {
-  if (const TemplateSpecializationType *TST =
-        type->getAs<TemplateSpecializationType>()) {
+  if (const auto *TST = type->getAs<TemplateSpecializationType>()) {
     if (!mangleSubstitution(QualType(TST, 0))) {
       mangleTemplatePrefix(TST->getTemplateName());
         
@@ -759,17 +775,19 @@ void CXXNameMangler::manglePrefix(QualType type) {
       mangleTemplateArgs(TST->getArgs(), TST->getNumArgs());
       addSubstitution(QualType(TST, 0));
     }
-  } else if (const DependentTemplateSpecializationType *DTST
-               = type->getAs<DependentTemplateSpecializationType>()) {
-    TemplateName Template
-      = getASTContext().getDependentTemplateName(DTST->getQualifier(), 
-                                                 DTST->getIdentifier());
-    mangleTemplatePrefix(Template);
+  } else if (const auto *DTST =
+                 type->getAs<DependentTemplateSpecializationType>()) {
+    if (!mangleSubstitution(QualType(DTST, 0))) {
+      TemplateName Template = getASTContext().getDependentTemplateName(
+          DTST->getQualifier(), DTST->getIdentifier());
+      mangleTemplatePrefix(Template);
 
-    // FIXME: GCC does not appear to mangle the template arguments when
-    // the template in question is a dependent template name. Should we
-    // emulate that badness?
-    mangleTemplateArgs(DTST->getArgs(), DTST->getNumArgs());
+      // FIXME: GCC does not appear to mangle the template arguments when
+      // the template in question is a dependent template name. Should we
+      // emulate that badness?
+      mangleTemplateArgs(DTST->getArgs(), DTST->getNumArgs());
+      addSubstitution(QualType(DTST, 0));
+    }
   } else {
     // We use the QualType mangle type variant here because it handles
     // substitutions.
@@ -779,12 +797,9 @@ void CXXNameMangler::manglePrefix(QualType type) {
 
 /// Mangle everything prior to the base-unresolved-name in an unresolved-name.
 ///
-/// \param firstQualifierLookup - the entity found by unqualified lookup
-///   for the first name in the qualifier, if this is for a member expression
 /// \param recursive - true if this is being called recursively,
 ///   i.e. if there is more prefix "to the right".
 void CXXNameMangler::mangleUnresolvedPrefix(NestedNameSpecifier *qualifier,
-                                            NamedDecl *firstQualifierLookup,
                                             bool recursive) {
 
   // x, ::x
@@ -817,7 +832,7 @@ void CXXNameMangler::mangleUnresolvedPrefix(NestedNameSpecifier *qualifier,
 
   case NestedNameSpecifier::Namespace:
     if (qualifier->getPrefix())
-      mangleUnresolvedPrefix(qualifier->getPrefix(), firstQualifierLookup,
+      mangleUnresolvedPrefix(qualifier->getPrefix(),
                              /*recursive*/ true);
     else
       Out << "sr";
@@ -825,7 +840,7 @@ void CXXNameMangler::mangleUnresolvedPrefix(NestedNameSpecifier *qualifier,
     break;
   case NestedNameSpecifier::NamespaceAlias:
     if (qualifier->getPrefix())
-      mangleUnresolvedPrefix(qualifier->getPrefix(), firstQualifierLookup,
+      mangleUnresolvedPrefix(qualifier->getPrefix(),
                              /*recursive*/ true);
     else
       Out << "sr";
@@ -842,193 +857,26 @@ void CXXNameMangler::mangleUnresolvedPrefix(NestedNameSpecifier *qualifier,
     //   - a template template parameter with arguments
     // In all of these cases, we should have no prefix.
     if (qualifier->getPrefix()) {
-      mangleUnresolvedPrefix(qualifier->getPrefix(), firstQualifierLookup,
+      mangleUnresolvedPrefix(qualifier->getPrefix(),
                              /*recursive*/ true);
     } else {
       // Otherwise, all the cases want this.
       Out << "sr";
     }
 
-    // Only certain other types are valid as prefixes;  enumerate them.
-    switch (type->getTypeClass()) {
-    case Type::Builtin:
-    case Type::Complex:
-    case Type::Adjusted:
-    case Type::Decayed:
-    case Type::Pointer:
-    case Type::BlockPointer:
-    case Type::LValueReference:
-    case Type::RValueReference:
-    case Type::MemberPointer:
-    case Type::ConstantArray:
-    case Type::IncompleteArray:
-    case Type::VariableArray:
-    case Type::DependentSizedArray:
-    case Type::DependentSizedExtVector:
-    case Type::Vector:
-    case Type::ExtVector:
-    case Type::FunctionProto:
-    case Type::FunctionNoProto:
-    case Type::Enum:
-    case Type::Paren:
-    case Type::Elaborated:
-    case Type::Attributed:
-    case Type::Auto:
-    case Type::PackExpansion:
-    case Type::ObjCObject:
-    case Type::ObjCInterface:
-    case Type::ObjCObjectPointer:
-    case Type::Atomic:
-      llvm_unreachable("type is illegal as a nested name specifier");
-
-    case Type::SubstTemplateTypeParmPack:
-      // FIXME: not clear how to mangle this!
-      // template <class T...> class A {
-      //   template <class U...> void foo(decltype(T::foo(U())) x...);
-      // };
-      Out << "_SUBSTPACK_";
-      break;
-
-    // <unresolved-type> ::= <template-param>
-    //                   ::= <decltype>
-    //                   ::= <template-template-param> <template-args>
-    // (this last is not official yet)
-    case Type::TypeOfExpr:
-    case Type::TypeOf:
-    case Type::Decltype:
-    case Type::TemplateTypeParm:
-    case Type::UnaryTransform:
-    case Type::SubstTemplateTypeParm:
-    unresolvedType:
-      assert(!qualifier->getPrefix());
-
-      // We only get here recursively if we're followed by identifiers.
-      if (recursive) Out << 'N';
-
-      // This seems to do everything we want.  It's not really
-      // sanctioned for a substituted template parameter, though.
-      mangleType(QualType(type, 0));
-
-      // We never want to print 'E' directly after an unresolved-type,
-      // so we return directly.
+    if (mangleUnresolvedTypeOrSimpleId(QualType(type, 0), recursive ? "N" : ""))
       return;
 
-    case Type::Typedef:
-      mangleSourceName(cast<TypedefType>(type)->getDecl()->getIdentifier());
-      break;
-
-    case Type::UnresolvedUsing:
-      mangleSourceName(cast<UnresolvedUsingType>(type)->getDecl()
-                         ->getIdentifier());
-      break;
-
-    case Type::Record:
-      mangleSourceName(cast<RecordType>(type)->getDecl()->getIdentifier());
-      break;
-
-    case Type::TemplateSpecialization: {
-      const TemplateSpecializationType *tst
-        = cast<TemplateSpecializationType>(type);
-      TemplateName name = tst->getTemplateName();
-      switch (name.getKind()) {
-      case TemplateName::Template:
-      case TemplateName::QualifiedTemplate: {
-        TemplateDecl *temp = name.getAsTemplateDecl();
-
-        // If the base is a template template parameter, this is an
-        // unresolved type.
-        assert(temp && "no template for template specialization type");
-        if (isa<TemplateTemplateParmDecl>(temp)) goto unresolvedType;
-
-        mangleSourceName(temp->getIdentifier());
-        break;
-      }
-
-      case TemplateName::OverloadedTemplate:
-      case TemplateName::DependentTemplate:
-        llvm_unreachable("invalid base for a template specialization type");
-
-      case TemplateName::SubstTemplateTemplateParm: {
-        SubstTemplateTemplateParmStorage *subst
-          = name.getAsSubstTemplateTemplateParm();
-        mangleExistingSubstitution(subst->getReplacement());
-        break;
-      }
-
-      case TemplateName::SubstTemplateTemplateParmPack: {
-        // FIXME: not clear how to mangle this!
-        // template <template <class U> class T...> class A {
-        //   template <class U...> void foo(decltype(T<U>::foo) x...);
-        // };
-        Out << "_SUBSTPACK_";
-        break;
-      }
-      }
-
-      mangleTemplateArgs(tst->getArgs(), tst->getNumArgs());
-      break;
-    }
-
-    case Type::InjectedClassName:
-      mangleSourceName(cast<InjectedClassNameType>(type)->getDecl()
-                         ->getIdentifier());
-      break;
-
-    case Type::DependentName:
-      mangleSourceName(cast<DependentNameType>(type)->getIdentifier());
-      break;
-
-    case Type::DependentTemplateSpecialization: {
-      const DependentTemplateSpecializationType *tst
-        = cast<DependentTemplateSpecializationType>(type);
-      mangleSourceName(tst->getIdentifier());
-      mangleTemplateArgs(tst->getArgs(), tst->getNumArgs());
-      break;
-    }
-    }
     break;
   }
 
   case NestedNameSpecifier::Identifier:
     // Member expressions can have these without prefixes.
-    if (qualifier->getPrefix()) {
-      mangleUnresolvedPrefix(qualifier->getPrefix(), firstQualifierLookup,
+    if (qualifier->getPrefix())
+      mangleUnresolvedPrefix(qualifier->getPrefix(),
                              /*recursive*/ true);
-    } else if (firstQualifierLookup) {
-
-      // Try to make a proper qualifier out of the lookup result, and
-      // then just recurse on that.
-      NestedNameSpecifier *newQualifier;
-      if (TypeDecl *typeDecl = dyn_cast<TypeDecl>(firstQualifierLookup)) {
-        QualType type = getASTContext().getTypeDeclType(typeDecl);
-
-        // Pretend we had a different nested name specifier.
-        newQualifier = NestedNameSpecifier::Create(getASTContext(),
-                                                   /*prefix*/ nullptr,
-                                                   /*template*/ false,
-                                                   type.getTypePtr());
-      } else if (NamespaceDecl *nspace =
-                   dyn_cast<NamespaceDecl>(firstQualifierLookup)) {
-        newQualifier = NestedNameSpecifier::Create(getASTContext(),
-                                                   /*prefix*/ nullptr,
-                                                   nspace);
-      } else if (NamespaceAliasDecl *alias =
-                   dyn_cast<NamespaceAliasDecl>(firstQualifierLookup)) {
-        newQualifier = NestedNameSpecifier::Create(getASTContext(),
-                                                   /*prefix*/ nullptr,
-                                                   alias);
-      } else {
-        // No sensible mangling to do here.
-        newQualifier = nullptr;
-      }
-
-      if (newQualifier)
-        return mangleUnresolvedPrefix(newQualifier, /*lookup*/ nullptr,
-                                      recursive);
-
-    } else {
+    else
       Out << "sr";
-    }
 
     mangleSourceName(qualifier->getAsIdentifier());
     break;
@@ -1043,16 +891,41 @@ void CXXNameMangler::mangleUnresolvedPrefix(NestedNameSpecifier *qualifier,
 /// Mangle an unresolved-name, which is generally used for names which
 /// weren't resolved to specific entities.
 void CXXNameMangler::mangleUnresolvedName(NestedNameSpecifier *qualifier,
-                                          NamedDecl *firstQualifierLookup,
                                           DeclarationName name,
                                           unsigned knownArity) {
-  if (qualifier) mangleUnresolvedPrefix(qualifier, firstQualifierLookup);
-  mangleUnqualifiedName(nullptr, name, knownArity);
+  if (qualifier) mangleUnresolvedPrefix(qualifier);
+  switch (name.getNameKind()) {
+    // <base-unresolved-name> ::= <simple-id>
+    case DeclarationName::Identifier:
+      mangleSourceName(name.getAsIdentifierInfo());
+      break;
+    // <base-unresolved-name> ::= dn <destructor-name>
+    case DeclarationName::CXXDestructorName:
+      Out << "dn";
+      mangleUnresolvedTypeOrSimpleId(name.getCXXNameType());
+      break;
+    // <base-unresolved-name> ::= on <operator-name>
+    case DeclarationName::CXXConversionFunctionName:
+    case DeclarationName::CXXLiteralOperatorName:
+    case DeclarationName::CXXOperatorName:
+      Out << "on";
+      mangleOperatorName(name, knownArity);
+      break;
+    case DeclarationName::CXXConstructorName:
+      llvm_unreachable("Can't mangle a constructor name!");
+    case DeclarationName::CXXUsingDirective:
+      llvm_unreachable("Can't mangle a using directive name!");
+    case DeclarationName::ObjCMultiArgSelector:
+    case DeclarationName::ObjCOneArgSelector:
+    case DeclarationName::ObjCZeroArgSelector:
+      llvm_unreachable("Can't mangle Objective-C selector names here!");
+  }
 }
 
 void CXXNameMangler::mangleUnqualifiedName(const NamedDecl *ND,
                                            DeclarationName Name,
                                            unsigned KnownArity) {
+  unsigned Arity = KnownArity;
   //  <unqualified-name> ::= <operator-name>
   //                     ::= <ctor-dtor-name>
   //                     ::= <source-name>
@@ -1163,7 +1036,7 @@ void CXXNameMangler::mangleUnqualifiedName(const NamedDecl *ND,
     Str += llvm::utostr(AnonStructId);
 
     Out << Str.size();
-    Out << Str.str();
+    Out << Str;
     break;
   }
 
@@ -1194,33 +1067,19 @@ void CXXNameMangler::mangleUnqualifiedName(const NamedDecl *ND,
       mangleCXXDtorType(Dtor_Complete);
     break;
 
-  case DeclarationName::CXXConversionFunctionName:
-    // <operator-name> ::= cv <type>    # (cast)
-    Out << "cv";
-    mangleType(Name.getCXXNameType());
-    break;
-
-  case DeclarationName::CXXOperatorName: {
-    unsigned Arity;
-    if (ND) {
+  case DeclarationName::CXXOperatorName:
+    if (ND && Arity == UnknownArity) {
       Arity = cast<FunctionDecl>(ND)->getNumParams();
 
-      // If we have a C++ member function, we need to include the 'this' pointer.
-      // FIXME: This does not make sense for operators that are static, but their
-      // names stay the same regardless of the arity (operator new for instance).
-      if (isa<CXXMethodDecl>(ND))
-        Arity++;
-    } else
-      Arity = KnownArity;
-
-    mangleOperatorName(Name.getCXXOverloadedOperator(), Arity);
-    break;
-  }
-
+      // If we have a member function, we need to include the 'this' pointer.
+      if (const auto *MD = dyn_cast<CXXMethodDecl>(ND))
+        if (!MD->isStatic())
+          Arity++;
+    }
+  // FALLTHROUGH
+  case DeclarationName::CXXConversionFunctionName:
   case DeclarationName::CXXLiteralOperatorName:
-    // FIXME: This mangling is not yet official.
-    Out << "li";
-    mangleSourceName(Name.getCXXLiteralIdentifier());
+    mangleOperatorName(Name, Arity);
     break;
 
   case DeclarationName::CXXUsingDirective:
@@ -1529,7 +1388,8 @@ void CXXNameMangler::mangleTemplatePrefix(TemplateName Template) {
    
   DependentTemplateName *Dependent = Template.getAsDependentTemplateName();
   assert(Dependent && "Unknown template name kind?");
-  manglePrefix(Dependent->getQualifier());
+  if (NestedNameSpecifier *Qualifier = Dependent->getQualifier())
+    manglePrefix(Qualifier);
   mangleUnscopedTemplateName(Template);
 }
 
@@ -1591,7 +1451,7 @@ void CXXNameMangler::mangleType(TemplateName TN) {
 
     // <class-enum-type> ::= <name>
     // <name> ::= <nested-name>
-    mangleUnresolvedPrefix(Dependent->getQualifier(), nullptr);
+    mangleUnresolvedPrefix(Dependent->getQualifier());
     mangleSourceName(Dependent->getIdentifier());
     break;
   }
@@ -1620,6 +1480,181 @@ void CXXNameMangler::mangleType(TemplateName TN) {
   addSubstitution(TN);
 }
 
+bool CXXNameMangler::mangleUnresolvedTypeOrSimpleId(QualType Ty,
+                                                    StringRef Prefix) {
+  // Only certain other types are valid as prefixes;  enumerate them.
+  switch (Ty->getTypeClass()) {
+  case Type::Builtin:
+  case Type::Complex:
+  case Type::Adjusted:
+  case Type::Decayed:
+  case Type::Pointer:
+  case Type::BlockPointer:
+  case Type::LValueReference:
+  case Type::RValueReference:
+  case Type::MemberPointer:
+  case Type::ConstantArray:
+  case Type::IncompleteArray:
+  case Type::VariableArray:
+  case Type::DependentSizedArray:
+  case Type::DependentSizedExtVector:
+  case Type::Vector:
+  case Type::ExtVector:
+  case Type::FunctionProto:
+  case Type::FunctionNoProto:
+  case Type::Paren:
+  case Type::Attributed:
+  case Type::Auto:
+  case Type::PackExpansion:
+  case Type::ObjCObject:
+  case Type::ObjCInterface:
+  case Type::ObjCObjectPointer:
+  case Type::Atomic:
+    llvm_unreachable("type is illegal as a nested name specifier");
+
+  case Type::SubstTemplateTypeParmPack:
+    // FIXME: not clear how to mangle this!
+    // template <class T...> class A {
+    //   template <class U...> void foo(decltype(T::foo(U())) x...);
+    // };
+    Out << "_SUBSTPACK_";
+    break;
+
+  // <unresolved-type> ::= <template-param>
+  //                   ::= <decltype>
+  //                   ::= <template-template-param> <template-args>
+  // (this last is not official yet)
+  case Type::TypeOfExpr:
+  case Type::TypeOf:
+  case Type::Decltype:
+  case Type::TemplateTypeParm:
+  case Type::UnaryTransform:
+  case Type::SubstTemplateTypeParm:
+  unresolvedType:
+    // Some callers want a prefix before the mangled type.
+    Out << Prefix;
+
+    // This seems to do everything we want.  It's not really
+    // sanctioned for a substituted template parameter, though.
+    mangleType(Ty);
+
+    // We never want to print 'E' directly after an unresolved-type,
+    // so we return directly.
+    return true;
+
+  case Type::Typedef:
+    mangleSourceName(cast<TypedefType>(Ty)->getDecl()->getIdentifier());
+    break;
+
+  case Type::UnresolvedUsing:
+    mangleSourceName(
+        cast<UnresolvedUsingType>(Ty)->getDecl()->getIdentifier());
+    break;
+
+  case Type::Enum:
+  case Type::Record:
+    mangleSourceName(cast<TagType>(Ty)->getDecl()->getIdentifier());
+    break;
+
+  case Type::TemplateSpecialization: {
+    const TemplateSpecializationType *TST =
+        cast<TemplateSpecializationType>(Ty);
+    TemplateName TN = TST->getTemplateName();
+    switch (TN.getKind()) {
+    case TemplateName::Template:
+    case TemplateName::QualifiedTemplate: {
+      TemplateDecl *TD = TN.getAsTemplateDecl();
+
+      // If the base is a template template parameter, this is an
+      // unresolved type.
+      assert(TD && "no template for template specialization type");
+      if (isa<TemplateTemplateParmDecl>(TD))
+        goto unresolvedType;
+
+      mangleSourceName(TD->getIdentifier());
+      break;
+    }
+
+    case TemplateName::OverloadedTemplate:
+    case TemplateName::DependentTemplate:
+      llvm_unreachable("invalid base for a template specialization type");
+
+    case TemplateName::SubstTemplateTemplateParm: {
+      SubstTemplateTemplateParmStorage *subst =
+          TN.getAsSubstTemplateTemplateParm();
+      mangleExistingSubstitution(subst->getReplacement());
+      break;
+    }
+
+    case TemplateName::SubstTemplateTemplateParmPack: {
+      // FIXME: not clear how to mangle this!
+      // template <template <class U> class T...> class A {
+      //   template <class U...> void foo(decltype(T<U>::foo) x...);
+      // };
+      Out << "_SUBSTPACK_";
+      break;
+    }
+    }
+
+    mangleTemplateArgs(TST->getArgs(), TST->getNumArgs());
+    break;
+  }
+
+  case Type::InjectedClassName:
+    mangleSourceName(
+        cast<InjectedClassNameType>(Ty)->getDecl()->getIdentifier());
+    break;
+
+  case Type::DependentName:
+    mangleSourceName(cast<DependentNameType>(Ty)->getIdentifier());
+    break;
+
+  case Type::DependentTemplateSpecialization: {
+    const DependentTemplateSpecializationType *DTST =
+        cast<DependentTemplateSpecializationType>(Ty);
+    mangleSourceName(DTST->getIdentifier());
+    mangleTemplateArgs(DTST->getArgs(), DTST->getNumArgs());
+    break;
+  }
+
+  case Type::Elaborated:
+    return mangleUnresolvedTypeOrSimpleId(
+        cast<ElaboratedType>(Ty)->getNamedType(), Prefix);
+  }
+
+  return false;
+}
+
+void CXXNameMangler::mangleOperatorName(DeclarationName Name, unsigned Arity) {
+  switch (Name.getNameKind()) {
+  case DeclarationName::CXXConstructorName:
+  case DeclarationName::CXXDestructorName:
+  case DeclarationName::CXXUsingDirective:
+  case DeclarationName::Identifier:
+  case DeclarationName::ObjCMultiArgSelector:
+  case DeclarationName::ObjCOneArgSelector:
+  case DeclarationName::ObjCZeroArgSelector:
+    llvm_unreachable("Not an operator name");
+
+  case DeclarationName::CXXConversionFunctionName:
+    // <operator-name> ::= cv <type>    # (cast)
+    Out << "cv";
+    mangleType(Name.getCXXNameType());
+    break;
+
+  case DeclarationName::CXXLiteralOperatorName:
+    Out << "li";
+    mangleSourceName(Name.getCXXLiteralIdentifier());
+    return;
+
+  case DeclarationName::CXXOperatorName:
+    mangleOperatorName(Name.getCXXOverloadedOperator(), Arity);
+    break;
+  }
+}
+
+
+
 void
 CXXNameMangler::mangleOperatorName(OverloadedOperatorKind OO, unsigned Arity) {
   switch (OO) {
@@ -2276,6 +2311,7 @@ void CXXNameMangler::mangleAArch64NeonVectorType(const VectorType *T) {
       EltName = "Poly16";
       break;
     case BuiltinType::ULong:
+    case BuiltinType::ULongLong:
       EltName = "Poly64";
       break;
     default:
@@ -2519,6 +2555,29 @@ void CXXNameMangler::mangleIntegerLiteral(QualType T,
 
 }
 
+void CXXNameMangler::mangleMemberExprBase(const Expr *Base, bool IsArrow) {
+  // Ignore member expressions involving anonymous unions.
+  while (const auto *RT = Base->getType()->getAs<RecordType>()) {
+    if (!RT->getDecl()->isAnonymousStructOrUnion())
+      break;
+    const auto *ME = dyn_cast<MemberExpr>(Base);
+    if (!ME)
+      break;
+    Base = ME->getBase();
+    IsArrow = ME->isArrow();
+  }
+
+  if (Base->isImplicitCXXThis()) {
+    // Note: GCC mangles member expressions to the implicit 'this' as
+    // *this., whereas we represent them as this->. The Itanium C++ ABI
+    // does not specify anything here, so we follow GCC.
+    Out << "dtdefpT";
+  } else {
+    Out << (IsArrow ? "pt" : "dt");
+    mangleExpression(Base);
+  }
+}
+
 /// Mangles a member expression.
 void CXXNameMangler::mangleMemberExpr(const Expr *base,
                                       bool isArrow,
@@ -2528,30 +2587,9 @@ void CXXNameMangler::mangleMemberExpr(const Expr *base,
                                       unsigned arity) {
   // <expression> ::= dt <expression> <unresolved-name>
   //              ::= pt <expression> <unresolved-name>
-  if (base) {
-
-    // Ignore member expressions involving anonymous unions.
-    while (const auto *RT = base->getType()->getAs<RecordType>()) {
-      if (!RT->getDecl()->isAnonymousStructOrUnion())
-        break;
-      const auto *ME = dyn_cast<MemberExpr>(base);
-      if (!ME)
-        break;
-      base = ME->getBase();
-      isArrow = ME->isArrow();
-    }
-
-    if (base->isImplicitCXXThis()) {
-      // Note: GCC mangles member expressions to the implicit 'this' as
-      // *this., whereas we represent them as this->. The Itanium C++ ABI
-      // does not specify anything here, so we follow GCC.
-      Out << "dtdefpT";
-    } else {
-      Out << (isArrow ? "pt" : "dt");
-      mangleExpression(base);
-    }
-  }
-  mangleUnresolvedName(qualifier, firstQualifierLookup, member, arity);
+  if (base)
+    mangleMemberExprBase(base, isArrow);
+  mangleUnresolvedName(qualifier, member, arity);
 }
 
 /// Look at the callee of the given call expression and determine if
@@ -2592,6 +2630,13 @@ void CXXNameMangler::mangleCastExpression(const Expr *E, StringRef CastEncoding)
   mangleExpression(ECE->getSubExpr());
 }
 
+void CXXNameMangler::mangleInitListElements(const InitListExpr *InitList) {
+  if (auto *Syntactic = InitList->getSyntacticForm())
+    InitList = Syntactic;
+  for (unsigned i = 0, e = InitList->getNumInits(); i != e; ++i)
+    mangleExpression(InitList->getInit(i));
+}
+
 void CXXNameMangler::mangleExpression(const Expr *E, unsigned Arity) {
   // <expression> ::= <unary operator-name> <expression>
   //              ::= <binary operator-name> <expression> <expression>
@@ -2631,7 +2676,6 @@ recurse:
   // These all can only appear in local or variable-initialization
   // contexts and so should never appear in a mangling.
   case Expr::AddrLabelExprClass:
-  case Expr::DesignatedInitExprClass:
   case Expr::ImplicitValueInitExprClass:
   case Expr::ParenListExprClass:
   case Expr::LambdaExprClass:
@@ -2641,9 +2685,9 @@ recurse:
 
   // FIXME: invent manglings for all these.
   case Expr::BlockExprClass:
-  case Expr::CXXPseudoDestructorExprClass:
   case Expr::ChooseExprClass:
   case Expr::CompoundLiteralExprClass:
+  case Expr::DesignatedInitExprClass:
   case Expr::ExtVectorElementExprClass:
   case Expr::GenericSelectionExprClass:
   case Expr::ObjCEncodeExprClass:
@@ -2713,9 +2757,7 @@ recurse:
 
   case Expr::InitListExprClass: {
     Out << "il";
-    const InitListExpr *InitList = cast<InitListExpr>(E);
-    for (unsigned i = 0, e = InitList->getNumInits(); i != e; ++i)
-      mangleExpression(InitList->getInit(i));
+    mangleInitListElements(cast<InitListExpr>(E));
     Out << "E";
     break;
   }
@@ -2759,9 +2801,14 @@ recurse:
       Out << "cl";
     }
 
-    mangleExpression(CE->getCallee(), CE->getNumArgs());
-    for (unsigned I = 0, N = CE->getNumArgs(); I != N; ++I)
-      mangleExpression(CE->getArg(I));
+    unsigned CallArity = CE->getNumArgs();
+    for (const Expr *Arg : CE->arguments())
+      if (isa<PackExpansionExpr>(Arg))
+        CallArity = UnknownArity;
+
+    mangleExpression(CE->getCallee(), CallArity);
+    for (const Expr *Arg : CE->arguments())
+      mangleExpression(Arg);
     Out << 'E';
     break;
   }
@@ -2793,9 +2840,7 @@ recurse:
       } else if (New->getInitializationStyle() == CXXNewExpr::ListInit &&
                  isa<InitListExpr>(Init)) {
         // Only take InitListExprs apart for list-initialization.
-        const InitListExpr *InitList = cast<InitListExpr>(Init);
-        for (unsigned i = 0, e = InitList->getNumInits(); i != e; ++i)
-          mangleExpression(InitList->getInit(i));
+        mangleInitListElements(cast<InitListExpr>(Init));
       } else
         mangleExpression(Init);
     }
@@ -2803,6 +2848,33 @@ recurse:
     break;
   }
 
+  case Expr::CXXPseudoDestructorExprClass: {
+    const auto *PDE = cast<CXXPseudoDestructorExpr>(E);
+    if (const Expr *Base = PDE->getBase())
+      mangleMemberExprBase(Base, PDE->isArrow());
+    NestedNameSpecifier *Qualifier = PDE->getQualifier();
+    QualType ScopeType;
+    if (TypeSourceInfo *ScopeInfo = PDE->getScopeTypeInfo()) {
+      if (Qualifier) {
+        mangleUnresolvedPrefix(Qualifier,
+                               /*Recursive=*/true);
+        mangleUnresolvedTypeOrSimpleId(ScopeInfo->getType());
+        Out << 'E';
+      } else {
+        Out << "sr";
+        if (!mangleUnresolvedTypeOrSimpleId(ScopeInfo->getType()))
+          Out << 'E';
+      }
+    } else if (Qualifier) {
+      mangleUnresolvedPrefix(Qualifier);
+    }
+    // <base-unresolved-name> ::= dn <destructor-name>
+    Out << "dn";
+    QualType DestroyedType = PDE->getDestroyedType();
+    mangleUnresolvedTypeOrSimpleId(DestroyedType);
+    break;
+  }
+
   case Expr::MemberExprClass: {
     const MemberExpr *ME = cast<MemberExpr>(E);
     mangleMemberExpr(ME->getBase(), ME->isArrow(),
@@ -2813,9 +2885,9 @@ recurse:
 
   case Expr::UnresolvedMemberExprClass: {
     const UnresolvedMemberExpr *ME = cast<UnresolvedMemberExpr>(E);
-    mangleMemberExpr(ME->getBase(), ME->isArrow(),
-                     ME->getQualifier(), nullptr, ME->getMemberName(),
-                     Arity);
+    mangleMemberExpr(ME->isImplicitAccess() ? nullptr : ME->getBase(),
+                     ME->isArrow(), ME->getQualifier(), nullptr,
+                     ME->getMemberName(), Arity);
     if (ME->hasExplicitTemplateArgs())
       mangleTemplateArgs(ME->getExplicitTemplateArgs());
     break;
@@ -2824,8 +2896,9 @@ recurse:
   case Expr::CXXDependentScopeMemberExprClass: {
     const CXXDependentScopeMemberExpr *ME
       = cast<CXXDependentScopeMemberExpr>(E);
-    mangleMemberExpr(ME->getBase(), ME->isArrow(),
-                     ME->getQualifier(), ME->getFirstQualifierFoundInScope(),
+    mangleMemberExpr(ME->isImplicitAccess() ? nullptr : ME->getBase(),
+                     ME->isArrow(), ME->getQualifier(),
+                     ME->getFirstQualifierFoundInScope(),
                      ME->getMember(), Arity);
     if (ME->hasExplicitTemplateArgs())
       mangleTemplateArgs(ME->getExplicitTemplateArgs());
@@ -2834,7 +2907,7 @@ recurse:
 
   case Expr::UnresolvedLookupExprClass: {
     const UnresolvedLookupExpr *ULE = cast<UnresolvedLookupExpr>(E);
-    mangleUnresolvedName(ULE->getQualifier(), nullptr, ULE->getName(), Arity);
+    mangleUnresolvedName(ULE->getQualifier(), ULE->getName(), Arity);
 
     // All the <unresolved-name> productions end in a
     // base-unresolved-name, where <template-args> are just tacked
@@ -2856,26 +2929,55 @@ recurse:
     break;
   }
 
-  case Expr::CXXTemporaryObjectExprClass:
   case Expr::CXXConstructExprClass: {
-    const CXXConstructExpr *CE = cast<CXXConstructExpr>(E);
+    const auto *CE = cast<CXXConstructExpr>(E);
+    if (!CE->isListInitialization() || CE->isStdInitListInitialization()) {
+      assert(
+          CE->getNumArgs() >= 1 &&
+          (CE->getNumArgs() == 1 || isa<CXXDefaultArgExpr>(CE->getArg(1))) &&
+          "implicit CXXConstructExpr must have one argument");
+      return mangleExpression(cast<CXXConstructExpr>(E)->getArg(0));
+    }
+    Out << "il";
+    for (auto *E : CE->arguments())
+      mangleExpression(E);
+    Out << "E";
+    break;
+  }
+
+  case Expr::CXXTemporaryObjectExprClass: {
+    const auto *CE = cast<CXXTemporaryObjectExpr>(E);
     unsigned N = CE->getNumArgs();
+    bool List = CE->isListInitialization();
 
-    if (CE->isListInitialization())
+    if (List)
       Out << "tl";
     else
       Out << "cv";
     mangleType(CE->getType());
-    if (N != 1) Out << '_';
-    for (unsigned I = 0; I != N; ++I) mangleExpression(CE->getArg(I));
-    if (N != 1) Out << 'E';
+    if (!List && N != 1)
+      Out << '_';
+    if (CE->isStdInitListInitialization()) {
+      // We implicitly created a std::initializer_list<T> for the first argument
+      // of a constructor of type U in an expression of the form U{a, b, c}.
+      // Strip all the semantic gunk off the initializer list.
+      auto *SILE =
+          cast<CXXStdInitializerListExpr>(CE->getArg(0)->IgnoreImplicit());
+      auto *ILE = cast<InitListExpr>(SILE->getSubExpr()->IgnoreImplicit());
+      mangleInitListElements(ILE);
+    } else {
+      for (auto *E : CE->arguments())
+        mangleExpression(E);
+    }
+    if (List || N != 1)
+      Out << 'E';
     break;
   }
 
   case Expr::CXXScalarValueInitExprClass:
-    Out <<"cv";
+    Out << "cv";
     mangleType(E->getType());
-    Out <<"_E";
+    Out << "_E";
     break;
 
   case Expr::CXXNoexceptExprClass:
@@ -3020,10 +3122,28 @@ recurse:
   // Fall through to mangle the cast itself.
       
   case Expr::CStyleCastExprClass:
-  case Expr::CXXFunctionalCastExprClass:
     mangleCastExpression(E, "cv");
     break;
 
+  case Expr::CXXFunctionalCastExprClass: {
+    auto *Sub = cast<ExplicitCastExpr>(E)->getSubExpr()->IgnoreImplicit();
+    // FIXME: Add isImplicit to CXXConstructExpr.
+    if (auto *CCE = dyn_cast<CXXConstructExpr>(Sub))
+      if (CCE->getParenOrBraceRange().isInvalid())
+        Sub = CCE->getArg(0)->IgnoreImplicit();
+    if (auto *StdInitList = dyn_cast<CXXStdInitializerListExpr>(Sub))
+      Sub = StdInitList->getSubExpr()->IgnoreImplicit();
+    if (auto *IL = dyn_cast<InitListExpr>(Sub)) {
+      Out << "tl";
+      mangleType(E->getType());
+      mangleInitListElements(IL);
+      Out << "E";
+    } else {
+      mangleCastExpression(E, "cv");
+    }
+    break;
+  }
+
   case Expr::CXXStaticCastExprClass:
     mangleCastExpression(E, "sc");
     break;
@@ -3058,7 +3178,7 @@ recurse:
     default:
       //  <expr-primary> ::= L <mangled-name> E # external name
       Out << 'L';
-      mangle(D, "_Z");
+      mangle(D);
       Out << 'E';
       break;
 
@@ -3101,8 +3221,7 @@ recurse:
 
   case Expr::DependentScopeDeclRefExprClass: {
     const DependentScopeDeclRefExpr *DRE = cast<DependentScopeDeclRefExpr>(E);
-    mangleUnresolvedName(DRE->getQualifier(), nullptr, DRE->getDeclName(),
-                         Arity);
+    mangleUnresolvedName(DRE->getQualifier(), DRE->getDeclName(), Arity);
 
     // All the <unresolved-name> productions end in a
     // base-unresolved-name, where <template-args> are just tacked
@@ -3327,6 +3446,9 @@ void CXXNameMangler::mangleCXXCtorType(CXXCtorType T) {
   case Ctor_Comdat:
     Out << "C5";
     break;
+  case Ctor_DefaultClosure:
+  case Ctor_CopyingClosure:
+    llvm_unreachable("closure constructors don't exist for the Itanium ABI!");
   }
 }
 
@@ -3410,8 +3532,8 @@ void CXXNameMangler::mangleTemplateArg(TemplateArgument A) {
     if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) {
       const ValueDecl *D = DRE->getDecl();
       if (isa<VarDecl>(D) || isa<FunctionDecl>(D)) {
-        Out << "L";
-        mangle(D, "_Z");
+        Out << 'L';
+        mangle(D);
         Out << 'E';
         break;
       }
@@ -3440,13 +3562,7 @@ void CXXNameMangler::mangleTemplateArg(TemplateArgument A) {
     Out << 'L';
     // References to external entities use the mangled name; if the name would
     // not normally be manged then mangle it as unqualified.
-    //
-    // FIXME: The ABI specifies that external names here should have _Z, but
-    // gcc leaves this off.
-    if (compensateMangling)
-      mangle(D, "_Z");
-    else
-      mangle(D, "Z");
+    mangle(D);
     Out << 'E';
 
     if (compensateMangling)
@@ -3524,8 +3640,8 @@ bool CXXNameMangler::mangleSubstitution(const NamedDecl *ND) {
   return mangleSubstitution(reinterpret_cast<uintptr_t>(ND));
 }
 
-/// \brief Determine whether the given type has any qualifiers that are
-/// relevant for substitutions.
+/// Determine whether the given type has any qualifiers that are relevant for
+/// substitutions.
 static bool hasMangledSubstitutionQualifiers(QualType T) {
   Qualifiers Qs = T.getQualifiers();
   return Qs.getCVRQualifiers() || Qs.hasAddressSpace();
@@ -3571,8 +3687,8 @@ static bool isCharType(QualType T) {
     T->isSpecificBuiltinType(BuiltinType::Char_U);
 }
 
-/// isCharSpecialization - Returns whether a given type is a template
-/// specialization of a given name with a single argument of type char.
+/// Returns whether a given type is a template specialization of a given name
+/// with a single argument of type char.
 static bool isCharSpecialization(QualType T, const char *Name) {
   if (T.isNull())
     return false;
@@ -3722,8 +3838,8 @@ void CXXNameMangler::addSubstitution(uintptr_t Ptr) {
 
 //
 
-/// \brief Mangles the name of the declaration D and emits that name to the
-/// given output stream.
+/// Mangles the name of the declaration D and emits that name to the given
+/// output stream.
 ///
 /// If the declaration D requires a mangled name, this routine will emit that
 /// mangled name to \p os and return true. Otherwise, \p os will be unchanged
@@ -3815,8 +3931,7 @@ void ItaniumMangleContextImpl::mangleCXXDtorThunk(
   Mangler.mangleFunctionEncoding(DD);
 }
 
-/// mangleGuardVariable - Returns the mangled name for a guard variable
-/// for the passed in VarDecl.
+/// Returns the mangled name for a guard variable for the passed in VarDecl.
 void ItaniumMangleContextImpl::mangleStaticGuardVariable(const VarDecl *D,
                                                          raw_ostream &Out) {
   //  <special-name> ::= GV <object name>       # Guard variable for one-time
@@ -3845,6 +3960,26 @@ void ItaniumMangleContextImpl::mangleDynamicAtExitDestructor(const VarDecl *D,
     Mangler.getStream() << D->getName();
 }
 
+void ItaniumMangleContextImpl::mangleSEHFilterExpression(
+    const NamedDecl *EnclosingDecl, raw_ostream &Out) {
+  CXXNameMangler Mangler(*this, Out);
+  Mangler.getStream() << "__filt_";
+  if (shouldMangleDeclName(EnclosingDecl))
+    Mangler.mangle(EnclosingDecl);
+  else
+    Mangler.getStream() << EnclosingDecl->getName();
+}
+
+void ItaniumMangleContextImpl::mangleSEHFinallyBlock(
+    const NamedDecl *EnclosingDecl, raw_ostream &Out) {
+  CXXNameMangler Mangler(*this, Out);
+  Mangler.getStream() << "__fin_";
+  if (shouldMangleDeclName(EnclosingDecl))
+    Mangler.mangle(EnclosingDecl);
+  else
+    Mangler.getStream() << EnclosingDecl->getName();
+}
+
 void ItaniumMangleContextImpl::mangleItaniumThreadLocalInit(const VarDecl *D,
                                                             raw_ostream &Out) {
   //  <special-name> ::= TH <object name>
@@ -3923,6 +4058,22 @@ void ItaniumMangleContextImpl::mangleTypeName(QualType Ty, raw_ostream &Out) {
   mangleCXXRTTIName(Ty, Out);
 }
 
+void ItaniumMangleContextImpl::mangleCXXVTableBitSet(const CXXRecordDecl *RD,
+                                                     raw_ostream &Out) {
+  Linkage L = RD->getLinkageInternal();
+  if (L == InternalLinkage || L == UniqueExternalLinkage) {
+    // This part of the identifier needs to be unique across all translation
+    // units in the linked program. The scheme fails if multiple translation
+    // units are compiled using the same relative source file path, or if
+    // multiple translation units are built from the same source file.
+    SourceManager &SM = getASTContext().getSourceManager();
+    Out << "[" << SM.getFileEntryForID(SM.getMainFileID())->getName() << "]";
+  }
+
+  CXXNameMangler Mangler(*this, Out);
+  Mangler.mangleType(QualType(RD->getTypeForDecl(), 0));
+}
+
 void ItaniumMangleContextImpl::mangleStringLiteral(const StringLiteral *, raw_ostream &) {
   llvm_unreachable("Can't mangle string literals");
 }
diff --git a/lib/AST/MicrosoftCXXABI.cpp b/lib/AST/MicrosoftCXXABI.cpp
index 0603d3b7b9b5..93ff77a2e964 100644
--- a/lib/AST/MicrosoftCXXABI.cpp
+++ b/lib/AST/MicrosoftCXXABI.cpp
@@ -31,11 +31,12 @@ class MicrosoftNumberingContext : public MangleNumberingContext {
   llvm::DenseMap<const Type *, unsigned> ManglingNumbers;
   unsigned LambdaManglingNumber;
   unsigned StaticLocalNumber;
+  unsigned StaticThreadlocalNumber;
 
 public:
   MicrosoftNumberingContext()
       : MangleNumberingContext(), LambdaManglingNumber(0),
-        StaticLocalNumber(0) {}
+        StaticLocalNumber(0), StaticThreadlocalNumber(0) {}
 
   unsigned getManglingNumber(const CXXMethodDecl *CallOperator) override {
     return ++LambdaManglingNumber;
@@ -47,6 +48,8 @@ public:
   }
 
   unsigned getStaticLocalNumber(const VarDecl *VD) override {
+    if (VD->getTLSKind())
+      return ++StaticThreadlocalNumber;
     return ++StaticLocalNumber;
   }
 
@@ -63,6 +66,10 @@ public:
 
 class MicrosoftCXXABI : public CXXABI {
   ASTContext &Context;
+  llvm::SmallDenseMap<CXXRecordDecl *, CXXConstructorDecl *> RecordToCopyCtor;
+  llvm::SmallDenseMap<std::pair<const CXXConstructorDecl *, unsigned>, Expr *>
+      CtorToDefaultArgExpr;
+
 public:
   MicrosoftCXXABI(ASTContext &Ctx) : Context(Ctx) { }
 
@@ -82,13 +89,36 @@ public:
       return false;
 
     const ASTRecordLayout &Layout = Context.getASTRecordLayout(RD);
-    
+
     // In the Microsoft ABI, classes can have one or two vtable pointers.
-    CharUnits PointerSize = 
-      Context.toCharUnitsFromBits(Context.getTargetInfo().getPointerWidth(0));
+    CharUnits PointerSize =
+        Context.toCharUnitsFromBits(Context.getTargetInfo().getPointerWidth(0));
     return Layout.getNonVirtualSize() == PointerSize ||
       Layout.getNonVirtualSize() == PointerSize * 2;
-  }    
+  }
+
+  void addDefaultArgExprForConstructor(const CXXConstructorDecl *CD,
+                                       unsigned ParmIdx, Expr *DAE) override {
+    CtorToDefaultArgExpr[std::make_pair(CD, ParmIdx)] = DAE;
+  }
+
+  Expr *getDefaultArgExprForConstructor(const CXXConstructorDecl *CD,
+                                        unsigned ParmIdx) override {
+    return CtorToDefaultArgExpr[std::make_pair(CD, ParmIdx)];
+  }
+
+  const CXXConstructorDecl *
+  getCopyConstructorForExceptionObject(CXXRecordDecl *RD) override {
+    return RecordToCopyCtor[RD];
+  }
+
+  void
+  addCopyConstructorForExceptionObject(CXXRecordDecl *RD,
+                                       CXXConstructorDecl *CD) override {
+    assert(CD != nullptr);
+    assert(RecordToCopyCtor[RD] == nullptr || RecordToCopyCtor[RD] == CD);
+    RecordToCopyCtor[RD] = CD;
+  }
 
   MangleNumberingContext *createMangleNumberingContext() const override {
     return new MicrosoftNumberingContext();
@@ -186,29 +216,28 @@ getMSMemberPointerSlots(const MemberPointerType *MPT) {
 
 std::pair<uint64_t, unsigned> MicrosoftCXXABI::getMemberPointerWidthAndAlign(
     const MemberPointerType *MPT) const {
-  const TargetInfo &Target = Context.getTargetInfo();
-  assert(Target.getTriple().getArch() == llvm::Triple::x86 ||
-         Target.getTriple().getArch() == llvm::Triple::x86_64);
-  unsigned Ptrs, Ints;
-  std::tie(Ptrs, Ints) = getMSMemberPointerSlots(MPT);
   // The nominal struct is laid out with pointers followed by ints and aligned
   // to a pointer width if any are present and an int width otherwise.
+  const TargetInfo &Target = Context.getTargetInfo();
   unsigned PtrSize = Target.getPointerWidth(0);
   unsigned IntSize = Target.getIntWidth();
+
+  unsigned Ptrs, Ints;
+  std::tie(Ptrs, Ints) = getMSMemberPointerSlots(MPT);
   uint64_t Width = Ptrs * PtrSize + Ints * IntSize;
   unsigned Align;
 
   // When MSVC does x86_32 record layout, it aligns aggregate member pointers to
   // 8 bytes.  However, __alignof usually returns 4 for data memptrs and 8 for
   // function memptrs.
-  if (Ptrs + Ints > 1 && Target.getTriple().getArch() == llvm::Triple::x86)
-    Align = 8 * 8;
+  if (Ptrs + Ints > 1 && Target.getTriple().isArch32Bit())
+    Align = 64;
   else if (Ptrs)
     Align = Target.getPointerAlign(0);
   else
     Align = Target.getIntAlign();
 
-  if (Target.getTriple().getArch() == llvm::Triple::x86_64)
+  if (Target.getTriple().isArch64Bit())
     Width = llvm::RoundUpToAlignment(Width, Align);
   return std::make_pair(Width, Align);
 }
diff --git a/lib/AST/MicrosoftMangle.cpp b/lib/AST/MicrosoftMangle.cpp
index 72f90f67cbdf..77522c1f9c5d 100644
--- a/lib/AST/MicrosoftMangle.cpp
+++ b/lib/AST/MicrosoftMangle.cpp
@@ -67,11 +67,15 @@ static const DeclContext *getEffectiveParentContext(const DeclContext *DC) {
   return getEffectiveDeclContext(cast<Decl>(DC));
 }
 
-static const FunctionDecl *getStructor(const FunctionDecl *fn) {
-  if (const FunctionTemplateDecl *ftd = fn->getPrimaryTemplate())
-    return ftd->getTemplatedDecl();
+static const FunctionDecl *getStructor(const NamedDecl *ND) {
+  if (const auto *FTD = dyn_cast<FunctionTemplateDecl>(ND))
+    return FTD->getTemplatedDecl();
 
-  return fn;
+  const auto *FD = cast<FunctionDecl>(ND);
+  if (const auto *FTD = FD->getPrimaryTemplate())
+    return FTD->getTemplatedDecl();
+
+  return FD;
 }
 
 static bool isLambda(const NamedDecl *ND) {
@@ -89,6 +93,8 @@ class MicrosoftMangleContextImpl : public MicrosoftMangleContext {
   llvm::DenseMap<DiscriminatorKeyTy, unsigned> Discriminator;
   llvm::DenseMap<const NamedDecl *, unsigned> Uniquifier;
   llvm::DenseMap<const CXXRecordDecl *, unsigned> LambdaIds;
+  llvm::DenseMap<const NamedDecl *, unsigned> SEHFilterIds;
+  llvm::DenseMap<const NamedDecl *, unsigned> SEHFinallyIds;
 
 public:
   MicrosoftMangleContextImpl(ASTContext &Context, DiagnosticsEngine &Diags)
@@ -109,6 +115,16 @@ public:
   void mangleCXXVBTable(const CXXRecordDecl *Derived,
                         ArrayRef<const CXXRecordDecl *> BasePath,
                         raw_ostream &Out) override;
+  void mangleCXXThrowInfo(QualType T, bool IsConst, bool IsVolatile,
+                          uint32_t NumEntries, raw_ostream &Out) override;
+  void mangleCXXCatchableTypeArray(QualType T, uint32_t NumEntries,
+                                   raw_ostream &Out) override;
+  void mangleCXXCatchableType(QualType T, const CXXConstructorDecl *CD,
+                              CXXCtorType CT, uint32_t Size, uint32_t NVOffset,
+                              int32_t VBPtrOffset, uint32_t VBIndex,
+                              raw_ostream &Out) override;
+  void mangleCXXCatchHandlerType(QualType T, uint32_t Flags,
+                                 raw_ostream &Out) override;
   void mangleCXXRTTI(QualType T, raw_ostream &Out) override;
   void mangleCXXRTTIName(QualType T, raw_ostream &Out) override;
   void mangleCXXRTTIBaseClassDescriptor(const CXXRecordDecl *Derived,
@@ -131,10 +147,18 @@ public:
   void mangleReferenceTemporary(const VarDecl *, unsigned ManglingNumber,
                                 raw_ostream &) override;
   void mangleStaticGuardVariable(const VarDecl *D, raw_ostream &Out) override;
+  void mangleThreadSafeStaticGuardVariable(const VarDecl *D, unsigned GuardNum,
+                                           raw_ostream &Out) override;
   void mangleDynamicInitializer(const VarDecl *D, raw_ostream &Out) override;
   void mangleDynamicAtExitDestructor(const VarDecl *D,
                                      raw_ostream &Out) override;
+  void mangleSEHFilterExpression(const NamedDecl *EnclosingDecl,
+                                 raw_ostream &Out) override;
+  void mangleSEHFinallyBlock(const NamedDecl *EnclosingDecl,
+                             raw_ostream &Out) override;
   void mangleStringLiteral(const StringLiteral *SL, raw_ostream &Out) override;
+  void mangleCXXVTableBitSet(const CXXRecordDecl *RD,
+                             raw_ostream &Out) override;
   bool getNextDiscriminator(const NamedDecl *ND, unsigned &disc) {
     // Lambda closure types are already numbered.
     if (isLambda(ND))
@@ -211,6 +235,12 @@ public:
                          64) {}
 
   MicrosoftCXXNameMangler(MicrosoftMangleContextImpl &C, raw_ostream &Out_,
+                          const CXXConstructorDecl *D, CXXCtorType Type)
+      : Context(C), Out(Out_), Structor(getStructor(D)), StructorType(Type),
+        PointersAre64Bit(C.getASTContext().getTargetInfo().getPointerWidth(0) ==
+                         64) {}
+
+  MicrosoftCXXNameMangler(MicrosoftMangleContextImpl &C, raw_ostream &Out_,
                           const CXXDestructorDecl *D, CXXDtorType Type)
       : Context(C), Out(Out_), Structor(getStructor(D)), StructorType(Type),
         PointersAre64Bit(C.getASTContext().getTargetInfo().getPointerWidth(0) ==
@@ -220,7 +250,7 @@ public:
 
   void mangle(const NamedDecl *D, StringRef Prefix = "\01?");
   void mangleName(const NamedDecl *ND);
-  void mangleFunctionEncoding(const FunctionDecl *FD);
+  void mangleFunctionEncoding(const FunctionDecl *FD, bool ShouldMangle);
   void mangleVariableEncoding(const VarDecl *VD);
   void mangleMemberDataPointer(const CXXRecordDecl *RD, const ValueDecl *VD);
   void mangleMemberFunctionPointer(const CXXRecordDecl *RD,
@@ -247,7 +277,7 @@ private:
   void mangleQualifiers(Qualifiers Quals, bool IsMember);
   void mangleRefQualifier(RefQualifierKind RefQualifier);
   void manglePointerCVQualifiers(Qualifiers Quals);
-  void manglePointerExtQualifiers(Qualifiers Quals, const Type *PointeeType);
+  void manglePointerExtQualifiers(Qualifiers Quals, QualType PointeeType);
 
   void mangleUnscopedTemplateName(const TemplateDecl *ND);
   void
@@ -261,6 +291,7 @@ private:
 #define ABSTRACT_TYPE(CLASS, PARENT)
 #define NON_CANONICAL_TYPE(CLASS, PARENT)
 #define TYPE(CLASS, PARENT) void mangleType(const CLASS##Type *T, \
+                                            Qualifiers Quals, \
                                             SourceRange Range);
 #include "clang/AST/TypeNodes.def"
 #undef ABSTRACT_TYPE
@@ -271,6 +302,7 @@ private:
   void mangleDecayedArrayType(const ArrayType *T);
   void mangleArrayType(const ArrayType *T);
   void mangleFunctionClass(const FunctionDecl *FD);
+  void mangleCallingConvention(CallingConv CC);
   void mangleCallingConvention(const FunctionType *T);
   void mangleIntegerLiteral(const llvm::APSInt &Number, bool IsBoolean);
   void mangleExpression(const Expr *E);
@@ -352,7 +384,7 @@ void MicrosoftCXXNameMangler::mangle(const NamedDecl *D, StringRef Prefix) {
   Out << Prefix;
   mangleName(D);
   if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D))
-    mangleFunctionEncoding(FD);
+    mangleFunctionEncoding(FD, Context.shouldMangleDeclName(FD));
   else if (const VarDecl *VD = dyn_cast<VarDecl>(D))
     mangleVariableEncoding(VD);
   else {
@@ -365,7 +397,8 @@ void MicrosoftCXXNameMangler::mangle(const NamedDecl *D, StringRef Prefix) {
   }
 }
 
-void MicrosoftCXXNameMangler::mangleFunctionEncoding(const FunctionDecl *FD) {
+void MicrosoftCXXNameMangler::mangleFunctionEncoding(const FunctionDecl *FD,
+                                                     bool ShouldMangle) {
   // <type-encoding> ::= <function-class> <function-type>
 
   // Since MSVC operates on the type as written and not the canonical type, it
@@ -380,13 +413,20 @@ void MicrosoftCXXNameMangler::mangleFunctionEncoding(const FunctionDecl *FD) {
   // extern "C" functions can hold entities that must be mangled.
   // As it stands, these functions still need to get expressed in the full
   // external name.  They have their class and type omitted, replaced with '9'.
-  if (Context.shouldMangleDeclName(FD)) {
-    // First, the function class.
+  if (ShouldMangle) {
+    // We would like to mangle all extern "C" functions using this additional
+    // component but this would break compatibility with MSVC's behavior.
+    // Instead, do this when we know that compatibility isn't important (in
+    // other words, when it is an overloaded extern "C" funciton).
+    if (FD->isExternC() && FD->hasAttr<OverloadableAttr>())
+      Out << "$$J0";
+
     mangleFunctionClass(FD);
 
     mangleFunctionType(FT, FD);
-  } else
+  } else {
     Out << '9';
+  }
 }
 
 void MicrosoftCXXNameMangler::mangleVariableEncoding(const VarDecl *VD) {
@@ -422,7 +462,7 @@ void MicrosoftCXXNameMangler::mangleVariableEncoding(const VarDecl *VD) {
       Ty->isMemberPointerType()) {
     mangleType(Ty, SR, QMM_Drop);
     manglePointerExtQualifiers(
-        Ty.getDesugaredType(getASTContext()).getLocalQualifiers(), nullptr);
+        Ty.getDesugaredType(getASTContext()).getLocalQualifiers(), QualType());
     if (const MemberPointerType *MPT = Ty->getAs<MemberPointerType>()) {
       mangleQualifiers(MPT->getPointeeType().getQualifiers(), true);
       // Member pointers are suffixed with a back reference to the member
@@ -525,7 +565,7 @@ MicrosoftCXXNameMangler::mangleMemberFunctionPointer(const CXXRecordDecl *RD,
       }
     } else {
       mangleName(MD);
-      mangleFunctionEncoding(MD);
+      mangleFunctionEncoding(MD, /*ShouldMangle=*/true);
     }
   } else {
     // Null single inheritance member functions are encoded as a simple nullptr.
@@ -559,7 +599,7 @@ void MicrosoftCXXNameMangler::mangleVirtualMemPtrThunk(
   Out << "$B";
   mangleNumber(OffsetInVFTable);
   Out << 'A';
-  Out << (PointersAre64Bit ? 'A' : 'E');
+  mangleCallingConvention(MD->getType()->getAs<FunctionProtoType>());
 }
 
 void MicrosoftCXXNameMangler::mangleName(const NamedDecl *ND) {
@@ -757,12 +797,18 @@ void MicrosoftCXXNameMangler::mangleUnqualifiedName(const NamedDecl *ND,
       llvm_unreachable("Can't mangle Objective-C selector names here!");
 
     case DeclarationName::CXXConstructorName:
-      if (ND == Structor) {
-        assert(StructorType == Ctor_Complete &&
-               "Should never be asked to mangle a ctor other than complete");
+      if (Structor == getStructor(ND)) {
+        if (StructorType == Ctor_CopyingClosure) {
+          Out << "?_O";
+          return;
+        }
+        if (StructorType == Ctor_DefaultClosure) {
+          Out << "?_F";
+          return;
+        }
       }
       Out << "?0";
-      break;
+      return;
 
     case DeclarationName::CXXDestructorName:
       if (ND == Structor)
@@ -1134,10 +1180,13 @@ void MicrosoftCXXNameMangler::mangleTemplateArg(const TemplateDecl *TD,
           cast<ValueDecl>(ND));
     } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(ND)) {
       const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(FD);
-      if (MD && MD->isInstance())
+      if (MD && MD->isInstance()) {
         mangleMemberFunctionPointer(MD->getParent()->getMostRecentDecl(), MD);
-      else
-        mangle(FD, "$1?");
+      } else {
+        Out << "$1?";
+        mangleName(FD);
+        mangleFunctionEncoding(FD, /*ShouldMangle=*/true);
+      }
     } else {
       mangle(ND, TA.getParamTypeForDecl()->isReferenceType() ? "$E?" : "$1?");
     }
@@ -1171,7 +1220,12 @@ void MicrosoftCXXNameMangler::mangleTemplateArg(const TemplateDecl *TD,
     if (TemplateArgs.empty()) {
       if (isa<TemplateTypeParmDecl>(Parm) ||
           isa<TemplateTemplateParmDecl>(Parm))
-        Out << "$$V";
+        // MSVC 2015 changed the mangling for empty expanded template packs,
+        // use the old mangling for link compatibility for old versions.
+        Out << (Context.getASTContext().getLangOpts().isCompatibleWithMSVC(
+                    LangOptions::MSVC2015)
+                    ? "$$V"
+                    : "$$$V");
       else if (isa<NonTypeTemplateParmDecl>(Parm))
         Out << "$S";
       else
@@ -1298,11 +1352,11 @@ MicrosoftCXXNameMangler::mangleRefQualifier(RefQualifierKind RefQualifier) {
   }
 }
 
-void
-MicrosoftCXXNameMangler::manglePointerExtQualifiers(Qualifiers Quals,
-                                                    const Type *PointeeType) {
+void MicrosoftCXXNameMangler::manglePointerExtQualifiers(Qualifiers Quals,
+                                                         QualType PointeeType) {
   bool HasRestrict = Quals.hasRestrict();
-  if (PointersAre64Bit && (!PointeeType || !PointeeType->isFunctionType()))
+  if (PointersAre64Bit &&
+      (PointeeType.isNull() || !PointeeType->isFunctionType()))
     Out << 'E';
 
   if (HasRestrict)
@@ -1338,29 +1392,38 @@ void MicrosoftCXXNameMangler::mangleArgumentType(QualType T,
   // e.g.
   // void (*x)(void) will not form a backreference with void x(void)
   void *TypePtr;
-  if (const DecayedType *DT = T->getAs<DecayedType>()) {
-    TypePtr = DT->getOriginalType().getCanonicalType().getAsOpaquePtr();
+  if (const auto *DT = T->getAs<DecayedType>()) {
+    QualType OriginalType = DT->getOriginalType();
+    // All decayed ArrayTypes should be treated identically; as-if they were
+    // a decayed IncompleteArrayType.
+    if (const auto *AT = getASTContext().getAsArrayType(OriginalType))
+      OriginalType = getASTContext().getIncompleteArrayType(
+          AT->getElementType(), AT->getSizeModifier(),
+          AT->getIndexTypeCVRQualifiers());
+
+    TypePtr = OriginalType.getCanonicalType().getAsOpaquePtr();
     // If the original parameter was textually written as an array,
     // instead treat the decayed parameter like it's const.
     //
     // e.g.
     // int [] -> int * const
-    if (DT->getOriginalType()->isArrayType())
+    if (OriginalType->isArrayType())
       T = T.withConst();
-  } else
+  } else {
     TypePtr = T.getCanonicalType().getAsOpaquePtr();
+  }
 
   ArgBackRefMap::iterator Found = TypeBackReferences.find(TypePtr);
 
   if (Found == TypeBackReferences.end()) {
-    size_t OutSizeBefore = Out.GetNumBytesInBuffer();
+    size_t OutSizeBefore = Out.tell();
 
     mangleType(T, Range, QMM_Drop);
 
     // See if it's worth creating a back reference.
     // Only types longer than 1 character are considered
     // and only 10 back references slots are available:
-    bool LongerThanOneChar = (Out.GetNumBytesInBuffer() - OutSizeBefore > 1);
+    bool LongerThanOneChar = (Out.tell() - OutSizeBefore > 1);
     if (LongerThanOneChar && TypeBackReferences.size() < 10) {
       size_t Size = TypeBackReferences.size();
       TypeBackReferences[TypePtr] = Size;
@@ -1388,7 +1451,7 @@ void MicrosoftCXXNameMangler::mangleType(QualType T, SourceRange Range,
   }
 
   bool IsPointer = T->isAnyPointerType() || T->isMemberPointerType() ||
-                   T->isBlockPointerType();
+                   T->isReferenceType() || T->isBlockPointerType();
 
   switch (QMM) {
   case QMM_Drop:
@@ -1415,11 +1478,6 @@ void MicrosoftCXXNameMangler::mangleType(QualType T, SourceRange Range,
     break;
   }
 
-  // We have to mangle these now, while we still have enough information.
-  if (IsPointer) {
-    manglePointerCVQualifiers(Quals);
-    manglePointerExtQualifiers(Quals, T->getPointeeType().getTypePtr());
-  }
   const Type *ty = T.getTypePtr();
 
   switch (ty->getTypeClass()) {
@@ -1430,7 +1488,7 @@ void MicrosoftCXXNameMangler::mangleType(QualType T, SourceRange Range,
     return;
 #define TYPE(CLASS, PARENT) \
   case Type::CLASS: \
-    mangleType(cast<CLASS##Type>(ty), Range); \
+    mangleType(cast<CLASS##Type>(ty), Quals, Range); \
     break;
 #include "clang/AST/TypeNodes.def"
 #undef ABSTRACT_TYPE
@@ -1439,7 +1497,7 @@ void MicrosoftCXXNameMangler::mangleType(QualType T, SourceRange Range,
   }
 }
 
-void MicrosoftCXXNameMangler::mangleType(const BuiltinType *T,
+void MicrosoftCXXNameMangler::mangleType(const BuiltinType *T, Qualifiers,
                                          SourceRange Range) {
   //  <type>         ::= <builtin-type>
   //  <builtin-type> ::= X  # void
@@ -1525,7 +1583,7 @@ void MicrosoftCXXNameMangler::mangleType(const BuiltinType *T,
 }
 
 // <type>          ::= <function-type>
-void MicrosoftCXXNameMangler::mangleType(const FunctionProtoType *T,
+void MicrosoftCXXNameMangler::mangleType(const FunctionProtoType *T, Qualifiers,
                                          SourceRange) {
   // Structors only appear in decls, so at this point we know it's not a
   // structor type.
@@ -1539,7 +1597,7 @@ void MicrosoftCXXNameMangler::mangleType(const FunctionProtoType *T,
   }
 }
 void MicrosoftCXXNameMangler::mangleType(const FunctionNoProtoType *T,
-                                         SourceRange) {
+                                         Qualifiers, SourceRange) {
   llvm_unreachable("Can't mangle K&R function prototypes");
 }
 
@@ -1553,24 +1611,34 @@ void MicrosoftCXXNameMangler::mangleFunctionType(const FunctionType *T,
   SourceRange Range;
   if (D) Range = D->getSourceRange();
 
-  bool IsStructor = false, HasThisQuals = ForceThisQuals;
+  bool IsStructor = false, HasThisQuals = ForceThisQuals, IsCtorClosure = false;
+  CallingConv CC = T->getCallConv();
   if (const CXXMethodDecl *MD = dyn_cast_or_null<CXXMethodDecl>(D)) {
     if (MD->isInstance())
       HasThisQuals = true;
-    if (isa<CXXConstructorDecl>(MD) || isa<CXXDestructorDecl>(MD))
+    if (isa<CXXDestructorDecl>(MD)) {
+      IsStructor = true;
+    } else if (isa<CXXConstructorDecl>(MD)) {
       IsStructor = true;
+      IsCtorClosure = (StructorType == Ctor_CopyingClosure ||
+                       StructorType == Ctor_DefaultClosure) &&
+                      getStructor(MD) == Structor;
+      if (IsCtorClosure)
+        CC = getASTContext().getDefaultCallingConvention(
+            /*IsVariadic=*/false, /*IsCXXMethod=*/true);
+    }
   }
 
   // If this is a C++ instance method, mangle the CVR qualifiers for the
   // this pointer.
   if (HasThisQuals) {
     Qualifiers Quals = Qualifiers::fromCVRMask(Proto->getTypeQuals());
-    manglePointerExtQualifiers(Quals, /*PointeeType=*/nullptr);
+    manglePointerExtQualifiers(Quals, /*PointeeType=*/QualType());
     mangleRefQualifier(Proto->getRefQualifier());
     mangleQualifiers(Quals, /*IsMember=*/false);
   }
 
-  mangleCallingConvention(T);
+  mangleCallingConvention(CC);
 
   // <return-type> ::= <type>
   //               ::= @ # structors (they have no declared return type)
@@ -1584,6 +1652,29 @@ void MicrosoftCXXNameMangler::mangleFunctionType(const FunctionType *T,
       Out << (PointersAre64Bit ? "PEAXI@Z" : "PAXI@Z");
       return;
     }
+    if (IsCtorClosure) {
+      // Default constructor closure and copy constructor closure both return
+      // void.
+      Out << 'X';
+
+      if (StructorType == Ctor_DefaultClosure) {
+        // Default constructor closure always has no arguments.
+        Out << 'X';
+      } else if (StructorType == Ctor_CopyingClosure) {
+        // Copy constructor closure always takes an unqualified reference.
+        mangleArgumentType(getASTContext().getLValueReferenceType(
+                               Proto->getParamType(0)
+                                   ->getAs<LValueReferenceType>()
+                                   ->getPointeeType(),
+                               /*SpelledAsLValue=*/true),
+                           Range);
+        Out << '@';
+      } else {
+        llvm_unreachable("unexpected constructor closure!");
+      }
+      Out << 'Z';
+      return;
+    }
     Out << '@';
   } else {
     QualType ResultType = Proto->getReturnType();
@@ -1608,7 +1699,7 @@ void MicrosoftCXXNameMangler::mangleFunctionType(const FunctionType *T,
     Out << 'X';
   } else {
     // Happens for function pointer type arguments for example.
-    for (const QualType Arg : Proto->param_types())
+    for (const QualType &Arg : Proto->param_types())
       mangleArgumentType(Arg, Range);
     // <builtin-type>      ::= Z  # ellipsis
     if (Proto->isVariadic())
@@ -1673,10 +1764,11 @@ void MicrosoftCXXNameMangler::mangleFunctionClass(const FunctionDecl *FD) {
         else
           Out << 'Q';
     }
-  } else
+  } else {
     Out << 'Y';
+  }
 }
-void MicrosoftCXXNameMangler::mangleCallingConvention(const FunctionType *T) {
+void MicrosoftCXXNameMangler::mangleCallingConvention(CallingConv CC) {
   // <calling-convention> ::= A # __cdecl
   //                      ::= B # __export __cdecl
   //                      ::= C # __pascal
@@ -1693,7 +1785,7 @@ void MicrosoftCXXNameMangler::mangleCallingConvention(const FunctionType *T) {
   // that keyword. (It didn't actually export them, it just made them so
   // that they could be in a DLL and somebody from another module could call
   // them.)
-  CallingConv CC = T->getCallConv();
+
   switch (CC) {
     default:
       llvm_unreachable("Unsupported CC for mangling");
@@ -1707,6 +1799,9 @@ void MicrosoftCXXNameMangler::mangleCallingConvention(const FunctionType *T) {
     case CC_X86VectorCall: Out << 'Q'; break;
   }
 }
+void MicrosoftCXXNameMangler::mangleCallingConvention(const FunctionType *T) {
+  mangleCallingConvention(T->getCallConv());
+}
 void MicrosoftCXXNameMangler::mangleThrowSpecification(
                                                 const FunctionProtoType *FT) {
   // <throw-spec> ::= Z # throw(...) (default)
@@ -1719,7 +1814,7 @@ void MicrosoftCXXNameMangler::mangleThrowSpecification(
 }
 
 void MicrosoftCXXNameMangler::mangleType(const UnresolvedUsingType *T,
-                                         SourceRange Range) {
+                                         Qualifiers, SourceRange Range) {
   // Probably should be mangled as a template instantiation; need to see what
   // VC does first.
   DiagnosticsEngine &Diags = Context.getDiags();
@@ -1734,10 +1829,12 @@ void MicrosoftCXXNameMangler::mangleType(const UnresolvedUsingType *T,
 // <struct-type> ::= U <name>
 // <class-type>  ::= V <name>
 // <enum-type>   ::= W4 <name>
-void MicrosoftCXXNameMangler::mangleType(const EnumType *T, SourceRange) {
+void MicrosoftCXXNameMangler::mangleType(const EnumType *T, Qualifiers,
+                                         SourceRange) {
   mangleType(cast<TagType>(T)->getDecl());
 }
-void MicrosoftCXXNameMangler::mangleType(const RecordType *T, SourceRange) {
+void MicrosoftCXXNameMangler::mangleType(const RecordType *T, Qualifiers,
+                                         SourceRange) {
   mangleType(cast<TagType>(T)->getDecl());
 }
 void MicrosoftCXXNameMangler::mangleType(const TagDecl *TD) {
@@ -1772,39 +1869,41 @@ void MicrosoftCXXNameMangler::mangleDecayedArrayType(const ArrayType *T) {
   manglePointerCVQualifiers(T->getElementType().getQualifiers());
   mangleType(T->getElementType(), SourceRange());
 }
-void MicrosoftCXXNameMangler::mangleType(const ConstantArrayType *T,
+void MicrosoftCXXNameMangler::mangleType(const ConstantArrayType *T, Qualifiers,
                                          SourceRange) {
   llvm_unreachable("Should have been special cased");
 }
-void MicrosoftCXXNameMangler::mangleType(const VariableArrayType *T,
+void MicrosoftCXXNameMangler::mangleType(const VariableArrayType *T, Qualifiers,
                                          SourceRange) {
   llvm_unreachable("Should have been special cased");
 }
 void MicrosoftCXXNameMangler::mangleType(const DependentSizedArrayType *T,
-                                         SourceRange) {
+                                         Qualifiers, SourceRange) {
   llvm_unreachable("Should have been special cased");
 }
 void MicrosoftCXXNameMangler::mangleType(const IncompleteArrayType *T,
-                                         SourceRange) {
+                                         Qualifiers, SourceRange) {
   llvm_unreachable("Should have been special cased");
 }
 void MicrosoftCXXNameMangler::mangleArrayType(const ArrayType *T) {
   QualType ElementTy(T, 0);
   SmallVector<llvm::APInt, 3> Dimensions;
   for (;;) {
-    if (const ConstantArrayType *CAT =
-            getASTContext().getAsConstantArrayType(ElementTy)) {
+    if (ElementTy->isConstantArrayType()) {
+      const ConstantArrayType *CAT =
+          getASTContext().getAsConstantArrayType(ElementTy);
       Dimensions.push_back(CAT->getSize());
       ElementTy = CAT->getElementType();
+    } else if (ElementTy->isIncompleteArrayType()) {
+      const IncompleteArrayType *IAT =
+          getASTContext().getAsIncompleteArrayType(ElementTy);
+      Dimensions.push_back(llvm::APInt(32, 0));
+      ElementTy = IAT->getElementType();
     } else if (ElementTy->isVariableArrayType()) {
       const VariableArrayType *VAT =
         getASTContext().getAsVariableArrayType(ElementTy);
-      DiagnosticsEngine &Diags = Context.getDiags();
-      unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
-        "cannot mangle this variable-length array yet");
-      Diags.Report(VAT->getSizeExpr()->getExprLoc(), DiagID)
-        << VAT->getBracketsRange();
-      return;
+      Dimensions.push_back(llvm::APInt(32, 0));
+      ElementTy = VAT->getElementType();
     } else if (ElementTy->isDependentSizedArrayType()) {
       // The dependent expression has to be folded into a constant (TODO).
       const DependentSizedArrayType *DSAT =
@@ -1815,12 +1914,9 @@ void MicrosoftCXXNameMangler::mangleArrayType(const ArrayType *T) {
       Diags.Report(DSAT->getSizeExpr()->getExprLoc(), DiagID)
         << DSAT->getBracketsRange();
       return;
-    } else if (const IncompleteArrayType *IAT =
-                   getASTContext().getAsIncompleteArrayType(ElementTy)) {
-      Dimensions.push_back(llvm::APInt(32, 0));
-      ElementTy = IAT->getElementType();
+    } else {
+      break;
     }
-    else break;
   }
   Out << 'Y';
   // <dimension-count> ::= <number> # number of extra dimensions
@@ -1833,9 +1929,11 @@ void MicrosoftCXXNameMangler::mangleArrayType(const ArrayType *T) {
 // <type>                   ::= <pointer-to-member-type>
 // <pointer-to-member-type> ::= <pointer-cvr-qualifiers> <cvr-qualifiers>
 //                                                          <class name> <type>
-void MicrosoftCXXNameMangler::mangleType(const MemberPointerType *T,
+void MicrosoftCXXNameMangler::mangleType(const MemberPointerType *T, Qualifiers Quals,
                                          SourceRange Range) {
   QualType PointeeType = T->getPointeeType();
+  manglePointerCVQualifiers(Quals);
+  manglePointerExtQualifiers(Quals, PointeeType);
   if (const FunctionProtoType *FPT = PointeeType->getAs<FunctionProtoType>()) {
     Out << '8';
     mangleName(T->getClass()->castAs<RecordType>()->getDecl());
@@ -1848,7 +1946,7 @@ void MicrosoftCXXNameMangler::mangleType(const MemberPointerType *T,
 }
 
 void MicrosoftCXXNameMangler::mangleType(const TemplateTypeParmType *T,
-                                         SourceRange Range) {
+                                         Qualifiers, SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
     "cannot mangle this template type parameter type yet");
@@ -1856,9 +1954,8 @@ void MicrosoftCXXNameMangler::mangleType(const TemplateTypeParmType *T,
     << Range;
 }
 
-void MicrosoftCXXNameMangler::mangleType(
-                                       const SubstTemplateTypeParmPackType *T,
-                                       SourceRange Range) {
+void MicrosoftCXXNameMangler::mangleType(const SubstTemplateTypeParmPackType *T,
+                                         Qualifiers, SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
     "cannot mangle this substituted parameter pack yet");
@@ -1869,40 +1966,46 @@ void MicrosoftCXXNameMangler::mangleType(
 // <type> ::= <pointer-type>
 // <pointer-type> ::= E? <pointer-cvr-qualifiers> <cvr-qualifiers> <type>
 //                       # the E is required for 64-bit non-static pointers
-void MicrosoftCXXNameMangler::mangleType(const PointerType *T,
+void MicrosoftCXXNameMangler::mangleType(const PointerType *T, Qualifiers Quals,
                                          SourceRange Range) {
-  QualType PointeeTy = T->getPointeeType();
-  mangleType(PointeeTy, Range);
+  QualType PointeeType = T->getPointeeType();
+  manglePointerCVQualifiers(Quals);
+  manglePointerExtQualifiers(Quals, PointeeType);
+  mangleType(PointeeType, Range);
 }
 void MicrosoftCXXNameMangler::mangleType(const ObjCObjectPointerType *T,
-                                         SourceRange Range) {
+                                         Qualifiers Quals, SourceRange Range) {
+  QualType PointeeType = T->getPointeeType();
+  manglePointerCVQualifiers(Quals);
+  manglePointerExtQualifiers(Quals, PointeeType);
   // Object pointers never have qualifiers.
   Out << 'A';
-  manglePointerExtQualifiers(Qualifiers(), T->getPointeeType().getTypePtr());
-  mangleType(T->getPointeeType(), Range);
+  mangleType(PointeeType, Range);
 }
 
 // <type> ::= <reference-type>
 // <reference-type> ::= A E? <cvr-qualifiers> <type>
 //                 # the E is required for 64-bit non-static lvalue references
 void MicrosoftCXXNameMangler::mangleType(const LValueReferenceType *T,
-                                         SourceRange Range) {
-  Out << 'A';
-  manglePointerExtQualifiers(Qualifiers(), T->getPointeeType().getTypePtr());
-  mangleType(T->getPointeeType(), Range);
+                                         Qualifiers Quals, SourceRange Range) {
+  QualType PointeeType = T->getPointeeType();
+  Out << (Quals.hasVolatile() ? 'B' : 'A');
+  manglePointerExtQualifiers(Quals, PointeeType);
+  mangleType(PointeeType, Range);
 }
 
 // <type> ::= <r-value-reference-type>
 // <r-value-reference-type> ::= $$Q E? <cvr-qualifiers> <type>
 //                 # the E is required for 64-bit non-static rvalue references
 void MicrosoftCXXNameMangler::mangleType(const RValueReferenceType *T,
-                                         SourceRange Range) {
-  Out << "$$Q";
-  manglePointerExtQualifiers(Qualifiers(), T->getPointeeType().getTypePtr());
-  mangleType(T->getPointeeType(), Range);
+                                         Qualifiers Quals, SourceRange Range) {
+  QualType PointeeType = T->getPointeeType();
+  Out << (Quals.hasVolatile() ? "$$R" : "$$Q");
+  manglePointerExtQualifiers(Quals, PointeeType);
+  mangleType(PointeeType, Range);
 }
 
-void MicrosoftCXXNameMangler::mangleType(const ComplexType *T,
+void MicrosoftCXXNameMangler::mangleType(const ComplexType *T, Qualifiers,
                                          SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
@@ -1911,7 +2014,7 @@ void MicrosoftCXXNameMangler::mangleType(const ComplexType *T,
     << Range;
 }
 
-void MicrosoftCXXNameMangler::mangleType(const VectorType *T,
+void MicrosoftCXXNameMangler::mangleType(const VectorType *T, Qualifiers Quals,
                                          SourceRange Range) {
   const BuiltinType *ET = T->getElementType()->getAs<BuiltinType>();
   assert(ET && "vectors with non-builtin elements are unsupported");
@@ -1939,13 +2042,13 @@ void MicrosoftCXXNameMangler::mangleType(const VectorType *T,
     // our own mangling to handle uses of __vector_size__ on user-specified
     // types, and for extensions like __v4sf.
     Out << "T__clang_vec" << T->getNumElements() << '_';
-    mangleType(ET, Range);
+    mangleType(ET, Quals, Range);
   }
 
   Out << "@@";
 }
 
-void MicrosoftCXXNameMangler::mangleType(const ExtVectorType *T,
+void MicrosoftCXXNameMangler::mangleType(const ExtVectorType *T, Qualifiers,
                                          SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
@@ -1954,7 +2057,7 @@ void MicrosoftCXXNameMangler::mangleType(const ExtVectorType *T,
     << Range;
 }
 void MicrosoftCXXNameMangler::mangleType(const DependentSizedExtVectorType *T,
-                                         SourceRange Range) {
+                                         Qualifiers, SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
     "cannot mangle this dependent-sized extended vector type yet");
@@ -1962,14 +2065,14 @@ void MicrosoftCXXNameMangler::mangleType(const DependentSizedExtVectorType *T,
     << Range;
 }
 
-void MicrosoftCXXNameMangler::mangleType(const ObjCInterfaceType *T,
+void MicrosoftCXXNameMangler::mangleType(const ObjCInterfaceType *T, Qualifiers,
                                          SourceRange) {
   // ObjC interfaces have structs underlying them.
   Out << 'U';
   mangleName(T->getDecl());
 }
 
-void MicrosoftCXXNameMangler::mangleType(const ObjCObjectType *T,
+void MicrosoftCXXNameMangler::mangleType(const ObjCObjectType *T, Qualifiers,
                                          SourceRange Range) {
   // We don't allow overloading by different protocol qualification,
   // so mangling them isn't necessary.
@@ -1977,20 +2080,23 @@ void MicrosoftCXXNameMangler::mangleType(const ObjCObjectType *T,
 }
 
 void MicrosoftCXXNameMangler::mangleType(const BlockPointerType *T,
-                                         SourceRange Range) {
+                                         Qualifiers Quals, SourceRange Range) {
+  QualType PointeeType = T->getPointeeType();
+  manglePointerCVQualifiers(Quals);
+  manglePointerExtQualifiers(Quals, PointeeType);
+
   Out << "_E";
 
-  QualType pointee = T->getPointeeType();
-  mangleFunctionType(pointee->castAs<FunctionProtoType>());
+  mangleFunctionType(PointeeType->castAs<FunctionProtoType>());
 }
 
 void MicrosoftCXXNameMangler::mangleType(const InjectedClassNameType *,
-                                         SourceRange) {
+                                         Qualifiers, SourceRange) {
   llvm_unreachable("Cannot mangle injected class name type.");
 }
 
 void MicrosoftCXXNameMangler::mangleType(const TemplateSpecializationType *T,
-                                         SourceRange Range) {
+                                         Qualifiers, SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
     "cannot mangle this template specialization type yet");
@@ -1998,7 +2104,7 @@ void MicrosoftCXXNameMangler::mangleType(const TemplateSpecializationType *T,
     << Range;
 }
 
-void MicrosoftCXXNameMangler::mangleType(const DependentNameType *T,
+void MicrosoftCXXNameMangler::mangleType(const DependentNameType *T, Qualifiers,
                                          SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
@@ -2008,8 +2114,8 @@ void MicrosoftCXXNameMangler::mangleType(const DependentNameType *T,
 }
 
 void MicrosoftCXXNameMangler::mangleType(
-                                 const DependentTemplateSpecializationType *T,
-                                 SourceRange Range) {
+    const DependentTemplateSpecializationType *T, Qualifiers,
+    SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
     "cannot mangle this dependent template specialization type yet");
@@ -2017,7 +2123,7 @@ void MicrosoftCXXNameMangler::mangleType(
     << Range;
 }
 
-void MicrosoftCXXNameMangler::mangleType(const PackExpansionType *T,
+void MicrosoftCXXNameMangler::mangleType(const PackExpansionType *T, Qualifiers,
                                          SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
@@ -2026,7 +2132,7 @@ void MicrosoftCXXNameMangler::mangleType(const PackExpansionType *T,
     << Range;
 }
 
-void MicrosoftCXXNameMangler::mangleType(const TypeOfType *T,
+void MicrosoftCXXNameMangler::mangleType(const TypeOfType *T, Qualifiers,
                                          SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
@@ -2035,7 +2141,7 @@ void MicrosoftCXXNameMangler::mangleType(const TypeOfType *T,
     << Range;
 }
 
-void MicrosoftCXXNameMangler::mangleType(const TypeOfExprType *T,
+void MicrosoftCXXNameMangler::mangleType(const TypeOfExprType *T, Qualifiers,
                                          SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
@@ -2044,7 +2150,7 @@ void MicrosoftCXXNameMangler::mangleType(const TypeOfExprType *T,
     << Range;
 }
 
-void MicrosoftCXXNameMangler::mangleType(const DecltypeType *T,
+void MicrosoftCXXNameMangler::mangleType(const DecltypeType *T, Qualifiers,
                                          SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
@@ -2054,7 +2160,7 @@ void MicrosoftCXXNameMangler::mangleType(const DecltypeType *T,
 }
 
 void MicrosoftCXXNameMangler::mangleType(const UnaryTransformType *T,
-                                         SourceRange Range) {
+                                         Qualifiers, SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
     "cannot mangle this unary transform type yet");
@@ -2062,7 +2168,8 @@ void MicrosoftCXXNameMangler::mangleType(const UnaryTransformType *T,
     << Range;
 }
 
-void MicrosoftCXXNameMangler::mangleType(const AutoType *T, SourceRange Range) {
+void MicrosoftCXXNameMangler::mangleType(const AutoType *T, Qualifiers,
+                                         SourceRange Range) {
   assert(T->getDeducedType().isNull() && "expecting a dependent type!");
 
   DiagnosticsEngine &Diags = Context.getDiags();
@@ -2072,7 +2179,7 @@ void MicrosoftCXXNameMangler::mangleType(const AutoType *T, SourceRange Range) {
     << Range;
 }
 
-void MicrosoftCXXNameMangler::mangleType(const AtomicType *T,
+void MicrosoftCXXNameMangler::mangleType(const AtomicType *T, Qualifiers,
                                          SourceRange Range) {
   DiagnosticsEngine &Diags = Context.getDiags();
   unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
@@ -2273,6 +2380,74 @@ void MicrosoftMangleContextImpl::mangleCXXRTTIName(QualType T,
   Mangler.mangleType(T, SourceRange(), MicrosoftCXXNameMangler::QMM_Result);
 }
 
+void MicrosoftMangleContextImpl::mangleCXXCatchHandlerType(QualType T,
+                                                           uint32_t Flags,
+                                                           raw_ostream &Out) {
+  MicrosoftCXXNameMangler Mangler(*this, Out);
+  Mangler.getStream() << "llvm.eh.handlertype.";
+  Mangler.mangleType(T, SourceRange(), MicrosoftCXXNameMangler::QMM_Result);
+  Mangler.getStream() << '.' << Flags;
+}
+
+void MicrosoftMangleContextImpl::mangleCXXThrowInfo(QualType T,
+                                                    bool IsConst,
+                                                    bool IsVolatile,
+                                                    uint32_t NumEntries,
+                                                    raw_ostream &Out) {
+  MicrosoftCXXNameMangler Mangler(*this, Out);
+  Mangler.getStream() << "_TI";
+  if (IsConst)
+    Mangler.getStream() << 'C';
+  if (IsVolatile)
+    Mangler.getStream() << 'V';
+  Mangler.getStream() << NumEntries;
+  Mangler.mangleType(T, SourceRange(), MicrosoftCXXNameMangler::QMM_Result);
+}
+
+void MicrosoftMangleContextImpl::mangleCXXCatchableTypeArray(
+    QualType T, uint32_t NumEntries, raw_ostream &Out) {
+  MicrosoftCXXNameMangler Mangler(*this, Out);
+  Mangler.getStream() << "_CTA";
+  Mangler.getStream() << NumEntries;
+  Mangler.mangleType(T, SourceRange(), MicrosoftCXXNameMangler::QMM_Result);
+}
+
+void MicrosoftMangleContextImpl::mangleCXXCatchableType(
+    QualType T, const CXXConstructorDecl *CD, CXXCtorType CT, uint32_t Size,
+    uint32_t NVOffset, int32_t VBPtrOffset, uint32_t VBIndex,
+    raw_ostream &Out) {
+  MicrosoftCXXNameMangler Mangler(*this, Out);
+  Mangler.getStream() << "_CT";
+
+  llvm::SmallString<64> RTTIMangling;
+  {
+    llvm::raw_svector_ostream Stream(RTTIMangling);
+    mangleCXXRTTI(T, Stream);
+  }
+  Mangler.getStream() << RTTIMangling.substr(1);
+
+  // VS2015 CTP6 omits the copy-constructor in the mangled name.  This name is,
+  // in fact, superfluous but I'm not sure the change was made consciously.
+  // TODO: Revisit this when VS2015 gets released.
+  llvm::SmallString<64> CopyCtorMangling;
+  if (CD) {
+    llvm::raw_svector_ostream Stream(CopyCtorMangling);
+    mangleCXXCtor(CD, CT, Stream);
+  }
+  Mangler.getStream() << CopyCtorMangling.substr(1);
+
+  Mangler.getStream() << Size;
+  if (VBPtrOffset == -1) {
+    if (NVOffset) {
+      Mangler.getStream() << NVOffset;
+    }
+  } else {
+    Mangler.getStream() << NVOffset;
+    Mangler.getStream() << VBPtrOffset;
+    Mangler.getStream() << VBIndex;
+  }
+}
+
 void MicrosoftMangleContextImpl::mangleCXXRTTIBaseClassDescriptor(
     const CXXRecordDecl *Derived, uint32_t NVOffset, int32_t VBPtrOffset,
     uint32_t VBTableOffset, uint32_t Flags, raw_ostream &Out) {
@@ -2318,6 +2493,28 @@ void MicrosoftMangleContextImpl::mangleCXXRTTICompleteObjectLocator(
   Mangler.getStream() << '@';
 }
 
+void MicrosoftMangleContextImpl::mangleSEHFilterExpression(
+    const NamedDecl *EnclosingDecl, raw_ostream &Out) {
+  MicrosoftCXXNameMangler Mangler(*this, Out);
+  // The function body is in the same comdat as the function with the handler,
+  // so the numbering here doesn't have to be the same across TUs.
+  //
+  // <mangled-name> ::= ?filt$ <filter-number> @0
+  Mangler.getStream() << "\01?filt$" << SEHFilterIds[EnclosingDecl]++ << "@0@";
+  Mangler.mangleName(EnclosingDecl);
+}
+
+void MicrosoftMangleContextImpl::mangleSEHFinallyBlock(
+    const NamedDecl *EnclosingDecl, raw_ostream &Out) {
+  MicrosoftCXXNameMangler Mangler(*this, Out);
+  // The function body is in the same comdat as the function with the handler,
+  // so the numbering here doesn't have to be the same across TUs.
+  //
+  // <mangled-name> ::= ?fin$ <filter-number> @0
+  Mangler.getStream() << "\01?fin$" << SEHFinallyIds[EnclosingDecl]++ << "@0@";
+  Mangler.mangleName(EnclosingDecl);
+}
+
 void MicrosoftMangleContextImpl::mangleTypeName(QualType T, raw_ostream &Out) {
   // This is just a made up unique string for the purposes of tbaa.  undname
   // does *not* know how to demangle it.
@@ -2329,7 +2526,7 @@ void MicrosoftMangleContextImpl::mangleTypeName(QualType T, raw_ostream &Out) {
 void MicrosoftMangleContextImpl::mangleCXXCtor(const CXXConstructorDecl *D,
                                                CXXCtorType Type,
                                                raw_ostream &Out) {
-  MicrosoftCXXNameMangler mangler(*this, Out);
+  MicrosoftCXXNameMangler mangler(*this, Out, D, Type);
   mangler.mangle(D);
 }
 
@@ -2348,18 +2545,18 @@ void MicrosoftMangleContextImpl::mangleReferenceTemporary(const VarDecl *VD,
   getDiags().Report(VD->getLocation(), DiagID);
 }
 
+void MicrosoftMangleContextImpl::mangleThreadSafeStaticGuardVariable(
+    const VarDecl *VD, unsigned GuardNum, raw_ostream &Out) {
+  MicrosoftCXXNameMangler Mangler(*this, Out);
+
+  Mangler.getStream() << "\01?$TSS" << GuardNum << '@';
+  Mangler.mangleNestedName(VD);
+}
+
 void MicrosoftMangleContextImpl::mangleStaticGuardVariable(const VarDecl *VD,
                                                            raw_ostream &Out) {
-  // TODO: This is not correct, especially with respect to VS "14".  VS "14"
-  // utilizes thread local variables to implement thread safe, re-entrant
-  // initialization for statics.  They no longer differentiate between an
-  // externally visible and non-externally visible static with respect to
-  // mangling, they all get $TSS <number>.
-  //
-  // N.B. This means that they can get more than 32 static variable guards in a
-  // scope.  It also means that they broke compatibility with their own ABI.
-
   // <guard-name> ::= ?_B <postfix> @5 <scope-depth>
+  //              ::= ?__J <postfix> @5 <scope-depth>
   //              ::= ?$S <guard-num> @ <postfix> @4IA
 
   // The first mangling is what MSVC uses to guard static locals in inline
@@ -2371,8 +2568,11 @@ void MicrosoftMangleContextImpl::mangleStaticGuardVariable(const VarDecl *VD,
   MicrosoftCXXNameMangler Mangler(*this, Out);
 
   bool Visible = VD->isExternallyVisible();
-  // <operator-name> ::= ?_B # local static guard
-  Mangler.getStream() << (Visible ? "\01??_B" : "\01?$S1@");
+  if (Visible) {
+    Mangler.getStream() << (VD->getTLSKind() ? "\01??__J" : "\01??_B");
+  } else {
+    Mangler.getStream() << "\01?$S1@";
+  }
   unsigned ScopeDepth = 0;
   if (Visible && !getNextDiscriminator(VD, ScopeDepth))
     // If we do not have a discriminator and are emitting a guard variable for
@@ -2553,6 +2753,11 @@ void MicrosoftMangleContextImpl::mangleStringLiteral(const StringLiteral *SL,
   Mangler.getStream() << '@';
 }
 
+void MicrosoftMangleContextImpl::mangleCXXVTableBitSet(const CXXRecordDecl *RD,
+                                                       raw_ostream &Out) {
+  llvm::report_fatal_error("Cannot mangle bitsets yet");
+}
+
 MicrosoftMangleContext *
 MicrosoftMangleContext::create(ASTContext &Context, DiagnosticsEngine &Diags) {
   return new MicrosoftMangleContextImpl(Context, Diags);
diff --git a/lib/AST/NSAPI.cpp b/lib/AST/NSAPI.cpp
index 3dc750a81787..2749100e14a2 100644
--- a/lib/AST/NSAPI.cpp
+++ b/lib/AST/NSAPI.cpp
@@ -27,7 +27,10 @@ IdentifierInfo *NSAPI::getNSClassId(NSClassIdKindKind K) const {
     "NSMutableArray",
     "NSDictionary",
     "NSMutableDictionary",
-    "NSNumber"
+    "NSNumber",
+    "NSMutableSet",
+    "NSCountedSet",
+    "NSMutableOrderedSet"
   };
 
   if (!ClassIds[K])
@@ -124,6 +127,25 @@ Selector NSAPI::getNSArraySelector(NSArrayMethodKind MK) const {
       Sel = Ctx.Selectors.getSelector(2, KeyIdents);
       break;
     }
+    case NSMutableArr_addObject:
+      Sel = Ctx.Selectors.getUnarySelector(&Ctx.Idents.get("addObject"));
+      break;
+    case NSMutableArr_insertObjectAtIndex: {
+      IdentifierInfo *KeyIdents[] = {
+        &Ctx.Idents.get("insertObject"),
+        &Ctx.Idents.get("atIndex")
+      };
+      Sel = Ctx.Selectors.getSelector(2, KeyIdents);
+      break;
+    }
+    case NSMutableArr_setObjectAtIndexedSubscript: {
+      IdentifierInfo *KeyIdents[] = {
+        &Ctx.Idents.get("setObject"),
+        &Ctx.Idents.get("atIndexedSubscript")
+      };
+      Sel = Ctx.Selectors.getSelector(2, KeyIdents);
+      break;
+    }
     }
     return (NSArraySelectors[MK] = Sel);
   }
@@ -209,6 +231,22 @@ Selector NSAPI::getNSDictionarySelector(
       Sel = Ctx.Selectors.getSelector(2, KeyIdents);
       break;
     }
+    case NSMutableDict_setObjectForKeyedSubscript: {
+      IdentifierInfo *KeyIdents[] = {
+        &Ctx.Idents.get("setObject"),
+        &Ctx.Idents.get("forKeyedSubscript")
+      };
+      Sel = Ctx.Selectors.getSelector(2, KeyIdents);
+      break;
+    }
+    case NSMutableDict_setValueForKey: {
+      IdentifierInfo *KeyIdents[] = {
+        &Ctx.Idents.get("setValue"),
+        &Ctx.Idents.get("forKey")
+      };
+      Sel = Ctx.Selectors.getSelector(2, KeyIdents);
+      break;
+    }
     }
     return (NSDictionarySelectors[MK] = Sel);
   }
@@ -227,6 +265,63 @@ NSAPI::getNSDictionaryMethodKind(Selector Sel) {
   return None;
 }
 
+Selector NSAPI::getNSSetSelector(NSSetMethodKind MK) const {
+  if (NSSetSelectors[MK].isNull()) {
+    Selector Sel;
+    switch (MK) {
+    case NSMutableSet_addObject:
+      Sel = Ctx.Selectors.getUnarySelector(&Ctx.Idents.get("addObject"));
+      break;
+    case NSOrderedSet_insertObjectAtIndex: {
+      IdentifierInfo *KeyIdents[] = {
+        &Ctx.Idents.get("insertObject"),
+        &Ctx.Idents.get("atIndex")
+      };
+      Sel = Ctx.Selectors.getSelector(2, KeyIdents);
+      break;
+    }
+    case NSOrderedSet_setObjectAtIndex: {
+      IdentifierInfo *KeyIdents[] = {
+        &Ctx.Idents.get("setObject"),
+        &Ctx.Idents.get("atIndex")
+      };
+      Sel = Ctx.Selectors.getSelector(2, KeyIdents);
+      break;
+    }
+    case NSOrderedSet_setObjectAtIndexedSubscript: {
+      IdentifierInfo *KeyIdents[] = {
+        &Ctx.Idents.get("setObject"),
+        &Ctx.Idents.get("atIndexedSubscript")
+      };
+      Sel = Ctx.Selectors.getSelector(2, KeyIdents);
+      break;
+    }
+    case NSOrderedSet_replaceObjectAtIndexWithObject: {
+      IdentifierInfo *KeyIdents[] = {
+        &Ctx.Idents.get("replaceObjectAtIndex"),
+        &Ctx.Idents.get("withObject")
+      };
+      Sel = Ctx.Selectors.getSelector(2, KeyIdents);
+      break;
+    }
+    }
+    return (NSSetSelectors[MK] = Sel);
+  }
+
+  return NSSetSelectors[MK];
+}
+
+Optional<NSAPI::NSSetMethodKind>
+NSAPI::getNSSetMethodKind(Selector Sel) {
+  for (unsigned i = 0; i != NumNSSetMethods; ++i) {
+    NSSetMethodKind MK = NSSetMethodKind(i);
+    if (Sel == getNSSetSelector(MK))
+      return MK;
+  }
+
+  return None;
+}
+
 Selector NSAPI::getNSNumberLiteralSelector(NSNumberLiteralMethodKind MK,
                                            bool Instance) const {
   static const char *ClassSelectorName[NumNSNumberLiteralMethods] = {
@@ -410,6 +505,11 @@ StringRef NSAPI::GetNSIntegralKind(QualType T) const {
   return StringRef();
 }
 
+bool NSAPI::isMacroDefined(StringRef Id) const {
+  // FIXME: Check whether the relevant module macros are visible.
+  return Ctx.Idents.get(Id).hasMacroDefinition();
+}
+
 bool NSAPI::isObjCTypedef(QualType T,
                           StringRef name, IdentifierInfo *&II) const {
   if (!Ctx.getLangOpts().ObjC1)
diff --git a/lib/AST/RecordLayoutBuilder.cpp b/lib/AST/RecordLayoutBuilder.cpp
index 0d070a4bfbfc..2101a5534a6b 100644
--- a/lib/AST/RecordLayoutBuilder.cpp
+++ b/lib/AST/RecordLayoutBuilder.cpp
@@ -55,6 +55,52 @@ struct BaseSubobjectInfo {
   const BaseSubobjectInfo *Derived;
 };
 
+/// \brief Externally provided layout. Typically used when the AST source, such
+/// as DWARF, lacks all the information that was available at compile time, such
+/// as alignment attributes on fields and pragmas in effect.
+struct ExternalLayout {
+  ExternalLayout() : Size(0), Align(0) {}
+
+  /// \brief Overall record size in bits.
+  uint64_t Size;
+
+  /// \brief Overall record alignment in bits.
+  uint64_t Align;
+
+  /// \brief Record field offsets in bits.
+  llvm::DenseMap<const FieldDecl *, uint64_t> FieldOffsets;
+
+  /// \brief Direct, non-virtual base offsets.
+  llvm::DenseMap<const CXXRecordDecl *, CharUnits> BaseOffsets;
+
+  /// \brief Virtual base offsets.
+  llvm::DenseMap<const CXXRecordDecl *, CharUnits> VirtualBaseOffsets;
+
+  /// Get the offset of the given field. The external source must provide
+  /// entries for all fields in the record.
+  uint64_t getExternalFieldOffset(const FieldDecl *FD) {
+    assert(FieldOffsets.count(FD) &&
+           "Field does not have an external offset");
+    return FieldOffsets[FD];
+  }
+
+  bool getExternalNVBaseOffset(const CXXRecordDecl *RD, CharUnits &BaseOffset) {
+    auto Known = BaseOffsets.find(RD);
+    if (Known == BaseOffsets.end())
+      return false;
+    BaseOffset = Known->second;
+    return true;
+  }
+
+  bool getExternalVBaseOffset(const CXXRecordDecl *RD, CharUnits &BaseOffset) {
+    auto Known = VirtualBaseOffsets.find(RD);
+    if (Known == VirtualBaseOffsets.end())
+      return false;
+    BaseOffset = Known->second;
+    return true;
+  }
+};
+
 /// EmptySubobjectMap - Keeps track of which empty subobjects exist at different
 /// offsets while laying out a C++ class.
 class EmptySubobjectMap {
@@ -541,7 +587,7 @@ protected:
 
   /// \brief Whether the external AST source has provided a layout for this
   /// record.
-  unsigned ExternalLayout : 1;
+  unsigned UseExternalLayout : 1;
 
   /// \brief Whether we need to infer alignment, even when we have an 
   /// externally-provided layout.
@@ -607,26 +653,14 @@ protected:
   /// avoid visiting virtual bases more than once.
   llvm::SmallPtrSet<const CXXRecordDecl *, 4> VisitedVirtualBases;
 
-  /// \brief Externally-provided size.
-  uint64_t ExternalSize;
-  
-  /// \brief Externally-provided alignment.
-  uint64_t ExternalAlign;
-  
-  /// \brief Externally-provided field offsets.
-  llvm::DenseMap<const FieldDecl *, uint64_t> ExternalFieldOffsets;
-
-  /// \brief Externally-provided direct, non-virtual base offsets.
-  llvm::DenseMap<const CXXRecordDecl *, CharUnits> ExternalBaseOffsets;
-
-  /// \brief Externally-provided virtual base offsets.
-  llvm::DenseMap<const CXXRecordDecl *, CharUnits> ExternalVirtualBaseOffsets;
+  /// Valid if UseExternalLayout is true.
+  ExternalLayout External;
 
   RecordLayoutBuilder(const ASTContext &Context,
                       EmptySubobjectMap *EmptySubobjects)
     : Context(Context), EmptySubobjects(EmptySubobjects), Size(0), 
       Alignment(CharUnits::One()), UnpackedAlignment(CharUnits::One()),
-      ExternalLayout(false), InferAlignment(false), 
+      UseExternalLayout(false), InferAlignment(false),
       Packed(false), IsUnion(false), IsMac68kAlign(false), IsMsStruct(false),
       UnfilledBitsInLastUnit(0), LastBitfieldTypeSize(0),
       MaxFieldAlignment(CharUnits::Zero()), 
@@ -748,8 +782,8 @@ protected:
   void setDataSize(CharUnits NewSize) { DataSize = Context.toBits(NewSize); }
   void setDataSize(uint64_t NewSize) { DataSize = NewSize; }
 
-  RecordLayoutBuilder(const RecordLayoutBuilder &) LLVM_DELETED_FUNCTION;
-  void operator=(const RecordLayoutBuilder &) LLVM_DELETED_FUNCTION;
+  RecordLayoutBuilder(const RecordLayoutBuilder &) = delete;
+  void operator=(const RecordLayoutBuilder &) = delete;
 };
 } // end anonymous namespace
 
@@ -1134,21 +1168,12 @@ CharUnits RecordLayoutBuilder::LayoutBase(const BaseSubobjectInfo *Base) {
   
   // Query the external layout to see if it provides an offset.
   bool HasExternalLayout = false;
-  if (ExternalLayout) {
+  if (UseExternalLayout) {
     llvm::DenseMap<const CXXRecordDecl *, CharUnits>::iterator Known;
-    if (Base->IsVirtual) {
-      Known = ExternalVirtualBaseOffsets.find(Base->Class);
-      if (Known != ExternalVirtualBaseOffsets.end()) {
-        Offset = Known->second;
-        HasExternalLayout = true;
-      }
-    } else {
-      Known = ExternalBaseOffsets.find(Base->Class);
-      if (Known != ExternalBaseOffsets.end()) {
-        Offset = Known->second;
-        HasExternalLayout = true;
-      }
-    }
+    if (Base->IsVirtual)
+      HasExternalLayout = External.getExternalNVBaseOffset(Base->Class, Offset);
+    else
+      HasExternalLayout = External.getExternalVBaseOffset(Base->Class, Offset);
   }
   
   CharUnits UnpackedBaseAlign = Layout.getNonVirtualAlignment();
@@ -1235,18 +1260,15 @@ void RecordLayoutBuilder::InitializeLayout(const Decl *D) {
   
   // If there is an external AST source, ask it for the various offsets.
   if (const RecordDecl *RD = dyn_cast<RecordDecl>(D))
-    if (ExternalASTSource *External = Context.getExternalSource()) {
-      ExternalLayout = External->layoutRecordType(RD, 
-                                                  ExternalSize,
-                                                  ExternalAlign,
-                                                  ExternalFieldOffsets,
-                                                  ExternalBaseOffsets,
-                                                  ExternalVirtualBaseOffsets);
-      
+    if (ExternalASTSource *Source = Context.getExternalSource()) {
+      UseExternalLayout = Source->layoutRecordType(
+          RD, External.Size, External.Align, External.FieldOffsets,
+          External.BaseOffsets, External.VirtualBaseOffsets);
+
       // Update based on external alignment.
-      if (ExternalLayout) {
-        if (ExternalAlign > 0) {
-          Alignment = Context.toCharUnitsFromBits(ExternalAlign);
+      if (UseExternalLayout) {
+        if (External.Align > 0) {
+          Alignment = Context.toCharUnitsFromBits(External.Align);
         } else {
           // The external source didn't have alignment information; infer it.
           InferAlignment = true;
@@ -1588,7 +1610,7 @@ void RecordLayoutBuilder::LayoutBitField(const FieldDecl *D) {
 
   // If we're using external layout, give the external layout a chance
   // to override this information.
-  if (ExternalLayout)
+  if (UseExternalLayout)
     FieldOffset = updateExternalFieldOffset(D, FieldOffset);
 
   // Okay, place the bitfield at the calculated offset.
@@ -1604,7 +1626,7 @@ void RecordLayoutBuilder::LayoutBitField(const FieldDecl *D) {
     FieldAlign = UnpackedFieldAlign = 1;
 
   // Diagnose differences in layout due to padding or packing.
-  if (!ExternalLayout)
+  if (!UseExternalLayout)
     CheckFieldPadding(FieldOffset, UnpaddedFieldOffset, UnpackedFieldOffset,
                       UnpackedFieldAlign, FieldPacked, D);
 
@@ -1727,7 +1749,7 @@ void RecordLayoutBuilder::LayoutField(const FieldDecl *D,
   UnpackedFieldOffset = 
     UnpackedFieldOffset.RoundUpToAlignment(UnpackedFieldAlign);
 
-  if (ExternalLayout) {
+  if (UseExternalLayout) {
     FieldOffset = Context.toCharUnitsFromBits(
                     updateExternalFieldOffset(D, Context.toBits(FieldOffset)));
     
@@ -1750,7 +1772,7 @@ void RecordLayoutBuilder::LayoutField(const FieldDecl *D,
   // Place this field at the current location.
   FieldOffsets.push_back(Context.toBits(FieldOffset));
 
-  if (!ExternalLayout)
+  if (!UseExternalLayout)
     CheckFieldPadding(Context.toBits(FieldOffset), UnpaddedFieldOffset, 
                       Context.toBits(UnpackedFieldOffset),
                       Context.toBits(UnpackedFieldAlign), FieldPacked, D);
@@ -1802,15 +1824,15 @@ void RecordLayoutBuilder::FinishLayout(const NamedDecl *D) {
   uint64_t RoundedSize
     = llvm::RoundUpToAlignment(getSizeInBits(), Context.toBits(Alignment));
 
-  if (ExternalLayout) {
+  if (UseExternalLayout) {
     // If we're inferring alignment, and the external size is smaller than
     // our size after we've rounded up to alignment, conservatively set the
     // alignment to 1.
-    if (InferAlignment && ExternalSize < RoundedSize) {
+    if (InferAlignment && External.Size < RoundedSize) {
       Alignment = CharUnits::One();
       InferAlignment = false;
     }
-    setSize(ExternalSize);
+    setSize(External.Size);
     return;
   }
 
@@ -1846,18 +1868,18 @@ void RecordLayoutBuilder::UpdateAlignment(CharUnits NewAlignment,
                                           CharUnits UnpackedNewAlignment) {
   // The alignment is not modified when using 'mac68k' alignment or when
   // we have an externally-supplied layout that also provides overall alignment.
-  if (IsMac68kAlign || (ExternalLayout && !InferAlignment))
+  if (IsMac68kAlign || (UseExternalLayout && !InferAlignment))
     return;
 
   if (NewAlignment > Alignment) {
-    assert(llvm::isPowerOf2_32(NewAlignment.getQuantity() && 
-           "Alignment not a power of 2"));
+    assert(llvm::isPowerOf2_64(NewAlignment.getQuantity()) &&
+           "Alignment not a power of 2");
     Alignment = NewAlignment;
   }
 
   if (UnpackedNewAlignment > UnpackedAlignment) {
-    assert(llvm::isPowerOf2_32(UnpackedNewAlignment.getQuantity() &&
-           "Alignment not a power of 2"));
+    assert(llvm::isPowerOf2_64(UnpackedNewAlignment.getQuantity()) &&
+           "Alignment not a power of 2");
     UnpackedAlignment = UnpackedNewAlignment;
   }
 }
@@ -1865,11 +1887,8 @@ void RecordLayoutBuilder::UpdateAlignment(CharUnits NewAlignment,
 uint64_t
 RecordLayoutBuilder::updateExternalFieldOffset(const FieldDecl *Field, 
                                                uint64_t ComputedOffset) {
-  assert(ExternalFieldOffsets.find(Field) != ExternalFieldOffsets.end() &&
-         "Field does not have an external offset");
-  
-  uint64_t ExternalFieldOffset = ExternalFieldOffsets[Field];
-  
+  uint64_t ExternalFieldOffset = External.getExternalFieldOffset(Field);
+
   if (InferAlignment && ExternalFieldOffset < ComputedOffset) {
     // The externally-supplied field offset is before the field offset we
     // computed. Assume that the structure is packed.
@@ -2152,9 +2171,8 @@ struct MicrosoftRecordLayoutBuilder {
   typedef llvm::DenseMap<const CXXRecordDecl *, CharUnits> BaseOffsetsMapTy;
   MicrosoftRecordLayoutBuilder(const ASTContext &Context) : Context(Context) {}
 private:
-  MicrosoftRecordLayoutBuilder(const MicrosoftRecordLayoutBuilder &)
-  LLVM_DELETED_FUNCTION;
-  void operator=(const MicrosoftRecordLayoutBuilder &) LLVM_DELETED_FUNCTION;
+  MicrosoftRecordLayoutBuilder(const MicrosoftRecordLayoutBuilder &) = delete;
+  void operator=(const MicrosoftRecordLayoutBuilder &) = delete;
 public:
   void layout(const RecordDecl *RD);
   void cxxLayout(const CXXRecordDecl *RD);
@@ -2252,6 +2270,13 @@ public:
   /// \brief True if this class is zero sized or first base is zero sized or
   /// has this property.  Only used for MS-ABI.
   bool LeadsWithZeroSizedBase : 1;
+
+  /// \brief True if the external AST source provided a layout for this record.
+  bool UseExternalLayout : 1;
+
+  /// \brief The layout provided by the external AST source. Only active if
+  /// UseExternalLayout is true.
+  ExternalLayout External;
 };
 } // namespace
 
@@ -2354,8 +2379,9 @@ void MicrosoftRecordLayoutBuilder::initializeLayout(const RecordDecl *RD) {
   // In 64-bit mode we always perform an alignment step after laying out vbases.
   // In 32-bit mode we do not.  The check to see if we need to perform alignment
   // checks the RequiredAlignment field and performs alignment if it isn't 0.
-  RequiredAlignment = Context.getTargetInfo().getPointerWidth(0) == 64 ?
-                      CharUnits::One() : CharUnits::Zero();
+  RequiredAlignment = Context.getTargetInfo().getTriple().isArch64Bit()
+                          ? CharUnits::One()
+                          : CharUnits::Zero();
   // Compute the maximum field alignment.
   MaxFieldAlignment = CharUnits::Zero();
   // Honor the default struct packing maximum alignment flag.
@@ -2371,6 +2397,13 @@ void MicrosoftRecordLayoutBuilder::initializeLayout(const RecordDecl *RD) {
   // Packed attribute forces max field alignment to be 1.
   if (RD->hasAttr<PackedAttr>())
     MaxFieldAlignment = CharUnits::One();
+
+  // Try to respect the external layout if present.
+  UseExternalLayout = false;
+  if (ExternalASTSource *Source = Context.getExternalSource())
+    UseExternalLayout = Source->layoutRecordType(
+        RD, External.Size, External.Align, External.FieldOffsets,
+        External.BaseOffsets, External.VirtualBaseOffsets);
 }
 
 void
@@ -2385,7 +2418,8 @@ MicrosoftRecordLayoutBuilder::initializeCXXLayout(const CXXRecordDecl *RD) {
   // injection.
   PointerInfo.Size =
       Context.toCharUnitsFromBits(Context.getTargetInfo().getPointerWidth(0));
-  PointerInfo.Alignment = PointerInfo.Size;
+  PointerInfo.Alignment =
+      Context.toCharUnitsFromBits(Context.getTargetInfo().getPointerAlign(0));
   // Respect pragma pack.
   if (!MaxFieldAlignment.isZero())
     PointerInfo.Alignment = std::min(PointerInfo.Alignment, MaxFieldAlignment);
@@ -2475,7 +2509,18 @@ void MicrosoftRecordLayoutBuilder::layoutNonVirtualBase(
       BaseLayout.leadsWithZeroSizedBase())
     Size++;
   ElementInfo Info = getAdjustedElementInfo(BaseLayout);
-  CharUnits BaseOffset = Size.RoundUpToAlignment(Info.Alignment);
+  CharUnits BaseOffset;
+
+  // Respect the external AST source base offset, if present.
+  bool FoundBase = false;
+  if (UseExternalLayout) {
+    FoundBase = External.getExternalNVBaseOffset(BaseDecl, BaseOffset);
+    if (FoundBase)
+      assert(BaseOffset >= Size && "base offset already allocated");
+  }
+
+  if (!FoundBase)
+    BaseOffset = Size.RoundUpToAlignment(Info.Alignment);
   Bases.insert(std::make_pair(BaseDecl, BaseOffset));
   Size = BaseOffset + BaseLayout.getNonVirtualSize();
   PreviousBaseLayout = &BaseLayout;
@@ -2499,7 +2544,14 @@ void MicrosoftRecordLayoutBuilder::layoutField(const FieldDecl *FD) {
     placeFieldAtOffset(CharUnits::Zero());
     Size = std::max(Size, Info.Size);
   } else {
-    CharUnits FieldOffset = Size.RoundUpToAlignment(Info.Alignment);
+    CharUnits FieldOffset;
+    if (UseExternalLayout) {
+      FieldOffset =
+          Context.toCharUnitsFromBits(External.getExternalFieldOffset(FD));
+      assert(FieldOffset >= Size && "field offset already allocated");
+    } else {
+      FieldOffset = Size.RoundUpToAlignment(Info.Alignment);
+    }
     placeFieldAtOffset(FieldOffset);
     Size = FieldOffset + Info.Size;
   }
@@ -2573,14 +2625,16 @@ void MicrosoftRecordLayoutBuilder::injectVBPtr(const CXXRecordDecl *RD) {
   CharUnits InjectionSite = VBPtrOffset;
   // But before we do, make sure it's properly aligned.
   VBPtrOffset = VBPtrOffset.RoundUpToAlignment(PointerInfo.Alignment);
+  // Shift everything after the vbptr down, unless we're using an external
+  // layout.
+  if (UseExternalLayout)
+    return;
   // Determine where the first field should be laid out after the vbptr.
   CharUnits FieldStart = VBPtrOffset + PointerInfo.Size;
   // Make sure that the amount we push the fields back by is a multiple of the
   // alignment.
   CharUnits Offset = (FieldStart - InjectionSite).RoundUpToAlignment(
       std::max(RequiredAlignment, Alignment));
-  // Increase the size of the object and push back all fields by the offset
-  // amount.
   Size += Offset;
   for (uint64_t &FieldOffset : FieldOffsets)
     FieldOffset += Context.toBits(Offset);
@@ -2647,7 +2701,18 @@ void MicrosoftRecordLayoutBuilder::layoutVirtualBases(const CXXRecordDecl *RD) {
     }
     // Insert the virtual base.
     ElementInfo Info = getAdjustedElementInfo(BaseLayout);
-    CharUnits BaseOffset = Size.RoundUpToAlignment(Info.Alignment);
+    CharUnits BaseOffset;
+
+    // Respect the external AST source base offset, if present.
+    bool FoundBase = false;
+    if (UseExternalLayout) {
+      FoundBase = External.getExternalVBaseOffset(BaseDecl, BaseOffset);
+      if (FoundBase)
+        assert(BaseOffset >= Size && "base offset already allocated");
+    }
+    if (!FoundBase)
+      BaseOffset = Size.RoundUpToAlignment(Info.Alignment);
+
     VBases.insert(std::make_pair(BaseDecl,
         ASTRecordLayout::VBaseInfo(BaseOffset, HasVtordisp)));
     Size = BaseOffset + BaseLayout.getNonVirtualSize();
@@ -2677,6 +2742,12 @@ void MicrosoftRecordLayoutBuilder::finalizeLayout(const RecordDecl *RD) {
     else
       Size = MinEmptyStructSize;
   }
+
+  if (UseExternalLayout) {
+    Size = Context.toCharUnitsFromBits(External.Size);
+    if (External.Align)
+      Alignment = Context.toCharUnitsFromBits(External.Align);
+  }
 }
 
 // Recursively walks the non-virtual bases of a class and determines if any of
@@ -2815,7 +2886,7 @@ ASTContext::getASTRecordLayout(const RecordDecl *D) const {
 
   const ASTRecordLayout *NewEntry = nullptr;
 
-  if (isMsLayout(D) && !D->getASTContext().getExternalSource()) {
+  if (isMsLayout(D)) {
     NewEntry = BuildMicrosoftASTRecordLayout(D);
   } else if (const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(D)) {
     EmptySubobjectMap EmptySubobjects(*this, RD);
@@ -2905,11 +2976,11 @@ void ASTContext::setNonKeyFunction(const CXXMethodDecl *Method) {
   // Look up the cache entry.  Since we're working with the first
   // declaration, its parent must be the class definition, which is
   // the correct key for the KeyFunctions hash.
-  llvm::DenseMap<const CXXRecordDecl*, LazyDeclPtr>::iterator
-    I = KeyFunctions.find(Method->getParent());
+  const auto &Map = KeyFunctions;
+  auto I = Map.find(Method->getParent());
 
   // If it's not cached, there's nothing to do.
-  if (I == KeyFunctions.end()) return;
+  if (I == Map.end()) return;
 
   // If it is cached, check whether it's the target method, and if so,
   // remove it from the cache. Note, the call to 'get' might invalidate
diff --git a/lib/AST/Stmt.cpp b/lib/AST/Stmt.cpp
index 68c7e7278432..6baa99bbc5bf 100644
--- a/lib/AST/Stmt.cpp
+++ b/lib/AST/Stmt.cpp
@@ -95,10 +95,16 @@ void Stmt::EnableStatistics() {
 Stmt *Stmt::IgnoreImplicit() {
   Stmt *s = this;
 
-  if (ExprWithCleanups *ewc = dyn_cast<ExprWithCleanups>(s))
+  if (auto *ewc = dyn_cast<ExprWithCleanups>(s))
     s = ewc->getSubExpr();
 
-  while (ImplicitCastExpr *ice = dyn_cast<ImplicitCastExpr>(s))
+  if (auto *mte = dyn_cast<MaterializeTemporaryExpr>(s))
+    s = mte->GetTemporaryExpr();
+
+  if (auto *bte = dyn_cast<CXXBindTemporaryExpr>(s))
+    s = bte->getSubExpr();
+
+  while (auto *ice = dyn_cast<ImplicitCastExpr>(s))
     s = ice->getSubExpr();
 
   return s;
@@ -934,8 +940,7 @@ void ForStmt::setConditionVariable(const ASTContext &C, VarDecl *V) {
 }
 
 SwitchStmt::SwitchStmt(const ASTContext &C, VarDecl *Var, Expr *cond)
-  : Stmt(SwitchStmtClass), FirstCase(nullptr), AllEnumCasesCovered(0)
-{
+    : Stmt(SwitchStmtClass), FirstCase(nullptr, false) {
   setConditionVariable(C, Var);
   SubExprs[COND] = cond;
   SubExprs[BODY] = nullptr;
@@ -1241,17 +1246,47 @@ OMPFirstprivateClause *OMPFirstprivateClause::CreateEmpty(const ASTContext &C,
   return new (Mem) OMPFirstprivateClause(N);
 }
 
-OMPLastprivateClause *OMPLastprivateClause::Create(const ASTContext &C,
-                                                   SourceLocation StartLoc,
-                                                   SourceLocation LParenLoc,
-                                                   SourceLocation EndLoc,
-                                                   ArrayRef<Expr *> VL) {
+void OMPLastprivateClause::setPrivateCopies(ArrayRef<Expr *> PrivateCopies) {
+  assert(PrivateCopies.size() == varlist_size() &&
+         "Number of private copies is not the same as the preallocated buffer");
+  std::copy(PrivateCopies.begin(), PrivateCopies.end(), varlist_end());
+}
+
+void OMPLastprivateClause::setSourceExprs(ArrayRef<Expr *> SrcExprs) {
+  assert(SrcExprs.size() == varlist_size() && "Number of source expressions is "
+                                              "not the same as the "
+                                              "preallocated buffer");
+  std::copy(SrcExprs.begin(), SrcExprs.end(), getPrivateCopies().end());
+}
+
+void OMPLastprivateClause::setDestinationExprs(ArrayRef<Expr *> DstExprs) {
+  assert(DstExprs.size() == varlist_size() && "Number of destination "
+                                              "expressions is not the same as "
+                                              "the preallocated buffer");
+  std::copy(DstExprs.begin(), DstExprs.end(), getSourceExprs().end());
+}
+
+void OMPLastprivateClause::setAssignmentOps(ArrayRef<Expr *> AssignmentOps) {
+  assert(AssignmentOps.size() == varlist_size() &&
+         "Number of assignment expressions is not the same as the preallocated "
+         "buffer");
+  std::copy(AssignmentOps.begin(), AssignmentOps.end(),
+            getDestinationExprs().end());
+}
+
+OMPLastprivateClause *OMPLastprivateClause::Create(
+    const ASTContext &C, SourceLocation StartLoc, SourceLocation LParenLoc,
+    SourceLocation EndLoc, ArrayRef<Expr *> VL, ArrayRef<Expr *> SrcExprs,
+    ArrayRef<Expr *> DstExprs, ArrayRef<Expr *> AssignmentOps) {
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPLastprivateClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * VL.size());
+                         5 * sizeof(Expr *) * VL.size());
   OMPLastprivateClause *Clause =
       new (Mem) OMPLastprivateClause(StartLoc, LParenLoc, EndLoc, VL.size());
   Clause->setVarRefs(VL);
+  Clause->setSourceExprs(SrcExprs);
+  Clause->setDestinationExprs(DstExprs);
+  Clause->setAssignmentOps(AssignmentOps);
   return Clause;
 }
 
@@ -1259,7 +1294,7 @@ OMPLastprivateClause *OMPLastprivateClause::CreateEmpty(const ASTContext &C,
                                                         unsigned N) {
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPLastprivateClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * N);
+                         5 * sizeof(Expr *) * N);
   return new (Mem) OMPLastprivateClause(N);
 }
 
@@ -1285,27 +1320,56 @@ OMPSharedClause *OMPSharedClause::CreateEmpty(const ASTContext &C,
   return new (Mem) OMPSharedClause(N);
 }
 
-OMPLinearClause *OMPLinearClause::Create(const ASTContext &C,
-                                         SourceLocation StartLoc,
-                                         SourceLocation LParenLoc,
-                                         SourceLocation ColonLoc,
-                                         SourceLocation EndLoc,
-                                         ArrayRef<Expr *> VL, Expr *Step) {
+void OMPLinearClause::setInits(ArrayRef<Expr *> IL) {
+  assert(IL.size() == varlist_size() &&
+         "Number of inits is not the same as the preallocated buffer");
+  std::copy(IL.begin(), IL.end(), varlist_end());
+}
+
+void OMPLinearClause::setUpdates(ArrayRef<Expr *> UL) {
+  assert(UL.size() == varlist_size() &&
+         "Number of updates is not the same as the preallocated buffer");
+  std::copy(UL.begin(), UL.end(), getInits().end());
+}
+
+void OMPLinearClause::setFinals(ArrayRef<Expr *> FL) {
+  assert(FL.size() == varlist_size() &&
+         "Number of final updates is not the same as the preallocated buffer");
+  std::copy(FL.begin(), FL.end(), getUpdates().end());
+}
+
+OMPLinearClause *
+OMPLinearClause::Create(const ASTContext &C, SourceLocation StartLoc,
+                        SourceLocation LParenLoc, SourceLocation ColonLoc,
+                        SourceLocation EndLoc, ArrayRef<Expr *> VL,
+                        ArrayRef<Expr *> IL, Expr *Step, Expr *CalcStep) {
+  // Allocate space for 4 lists (Vars, Inits, Updates, Finals) and 2 expressions
+  // (Step and CalcStep).
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPLinearClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * (VL.size() + 1));
+                         (4 * VL.size() + 2) * sizeof(Expr *));
   OMPLinearClause *Clause = new (Mem)
       OMPLinearClause(StartLoc, LParenLoc, ColonLoc, EndLoc, VL.size());
   Clause->setVarRefs(VL);
+  Clause->setInits(IL);
+  // Fill update and final expressions with zeroes, they are provided later,
+  // after the directive construction.
+  std::fill(Clause->getInits().end(), Clause->getInits().end() + VL.size(),
+            nullptr);
+  std::fill(Clause->getUpdates().end(), Clause->getUpdates().end() + VL.size(),
+            nullptr);
   Clause->setStep(Step);
+  Clause->setCalcStep(CalcStep);
   return Clause;
 }
 
 OMPLinearClause *OMPLinearClause::CreateEmpty(const ASTContext &C,
                                               unsigned NumVars) {
+  // Allocate space for 4 lists (Vars, Inits, Updates, Finals) and 2 expressions
+  // (Step and CalcStep).
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPLinearClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * (NumVars + 1));
+                         (4 * NumVars + 2) * sizeof(Expr *));
   return new (Mem) OMPLinearClause(NumVars);
 }
 
@@ -1331,17 +1395,41 @@ OMPAlignedClause *OMPAlignedClause::CreateEmpty(const ASTContext &C,
   return new (Mem) OMPAlignedClause(NumVars);
 }
 
-OMPCopyinClause *OMPCopyinClause::Create(const ASTContext &C,
-                                         SourceLocation StartLoc,
-                                         SourceLocation LParenLoc,
-                                         SourceLocation EndLoc,
-                                         ArrayRef<Expr *> VL) {
+void OMPCopyinClause::setSourceExprs(ArrayRef<Expr *> SrcExprs) {
+  assert(SrcExprs.size() == varlist_size() && "Number of source expressions is "
+                                              "not the same as the "
+                                              "preallocated buffer");
+  std::copy(SrcExprs.begin(), SrcExprs.end(), varlist_end());
+}
+
+void OMPCopyinClause::setDestinationExprs(ArrayRef<Expr *> DstExprs) {
+  assert(DstExprs.size() == varlist_size() && "Number of destination "
+                                              "expressions is not the same as "
+                                              "the preallocated buffer");
+  std::copy(DstExprs.begin(), DstExprs.end(), getSourceExprs().end());
+}
+
+void OMPCopyinClause::setAssignmentOps(ArrayRef<Expr *> AssignmentOps) {
+  assert(AssignmentOps.size() == varlist_size() &&
+         "Number of assignment expressions is not the same as the preallocated "
+         "buffer");
+  std::copy(AssignmentOps.begin(), AssignmentOps.end(),
+            getDestinationExprs().end());
+}
+
+OMPCopyinClause *OMPCopyinClause::Create(
+    const ASTContext &C, SourceLocation StartLoc, SourceLocation LParenLoc,
+    SourceLocation EndLoc, ArrayRef<Expr *> VL, ArrayRef<Expr *> SrcExprs,
+    ArrayRef<Expr *> DstExprs, ArrayRef<Expr *> AssignmentOps) {
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPCopyinClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * VL.size());
+                         4 * sizeof(Expr *) * VL.size());
   OMPCopyinClause *Clause = new (Mem) OMPCopyinClause(StartLoc, LParenLoc,
                                                       EndLoc, VL.size());
   Clause->setVarRefs(VL);
+  Clause->setSourceExprs(SrcExprs);
+  Clause->setDestinationExprs(DstExprs);
+  Clause->setAssignmentOps(AssignmentOps);
   return Clause;
 }
 
@@ -1349,21 +1437,45 @@ OMPCopyinClause *OMPCopyinClause::CreateEmpty(const ASTContext &C,
                                               unsigned N) {
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPCopyinClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * N);
+                         4 * sizeof(Expr *) * N);
   return new (Mem) OMPCopyinClause(N);
 }
 
-OMPCopyprivateClause *OMPCopyprivateClause::Create(const ASTContext &C,
-                                                   SourceLocation StartLoc,
-                                                   SourceLocation LParenLoc,
-                                                   SourceLocation EndLoc,
-                                                   ArrayRef<Expr *> VL) {
+void OMPCopyprivateClause::setSourceExprs(ArrayRef<Expr *> SrcExprs) {
+  assert(SrcExprs.size() == varlist_size() && "Number of source expressions is "
+                                              "not the same as the "
+                                              "preallocated buffer");
+  std::copy(SrcExprs.begin(), SrcExprs.end(), varlist_end());
+}
+
+void OMPCopyprivateClause::setDestinationExprs(ArrayRef<Expr *> DstExprs) {
+  assert(DstExprs.size() == varlist_size() && "Number of destination "
+                                              "expressions is not the same as "
+                                              "the preallocated buffer");
+  std::copy(DstExprs.begin(), DstExprs.end(), getSourceExprs().end());
+}
+
+void OMPCopyprivateClause::setAssignmentOps(ArrayRef<Expr *> AssignmentOps) {
+  assert(AssignmentOps.size() == varlist_size() &&
+         "Number of assignment expressions is not the same as the preallocated "
+         "buffer");
+  std::copy(AssignmentOps.begin(), AssignmentOps.end(),
+            getDestinationExprs().end());
+}
+
+OMPCopyprivateClause *OMPCopyprivateClause::Create(
+    const ASTContext &C, SourceLocation StartLoc, SourceLocation LParenLoc,
+    SourceLocation EndLoc, ArrayRef<Expr *> VL, ArrayRef<Expr *> SrcExprs,
+    ArrayRef<Expr *> DstExprs, ArrayRef<Expr *> AssignmentOps) {
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPCopyprivateClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * VL.size());
+                         4 * sizeof(Expr *) * VL.size());
   OMPCopyprivateClause *Clause =
       new (Mem) OMPCopyprivateClause(StartLoc, LParenLoc, EndLoc, VL.size());
   Clause->setVarRefs(VL);
+  Clause->setSourceExprs(SrcExprs);
+  Clause->setDestinationExprs(DstExprs);
+  Clause->setAssignmentOps(AssignmentOps);
   return Clause;
 }
 
@@ -1371,7 +1483,7 @@ OMPCopyprivateClause *OMPCopyprivateClause::CreateEmpty(const ASTContext &C,
                                                         unsigned N) {
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPCopyprivateClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * N);
+                         4 * sizeof(Expr *) * N);
   return new (Mem) OMPCopyprivateClause(N);
 }
 
@@ -1399,16 +1511,42 @@ void OMPLoopDirective::setFinals(ArrayRef<Expr *> A) {
   std::copy(A.begin(), A.end(), getFinals().begin());
 }
 
+void OMPReductionClause::setLHSExprs(ArrayRef<Expr *> LHSExprs) {
+  assert(
+      LHSExprs.size() == varlist_size() &&
+      "Number of LHS expressions is not the same as the preallocated buffer");
+  std::copy(LHSExprs.begin(), LHSExprs.end(), varlist_end());
+}
+
+void OMPReductionClause::setRHSExprs(ArrayRef<Expr *> RHSExprs) {
+  assert(
+      RHSExprs.size() == varlist_size() &&
+      "Number of RHS expressions is not the same as the preallocated buffer");
+  std::copy(RHSExprs.begin(), RHSExprs.end(), getLHSExprs().end());
+}
+
+void OMPReductionClause::setReductionOps(ArrayRef<Expr *> ReductionOps) {
+  assert(ReductionOps.size() == varlist_size() && "Number of reduction "
+                                                  "expressions is not the same "
+                                                  "as the preallocated buffer");
+  std::copy(ReductionOps.begin(), ReductionOps.end(), getRHSExprs().end());
+}
+
 OMPReductionClause *OMPReductionClause::Create(
     const ASTContext &C, SourceLocation StartLoc, SourceLocation LParenLoc,
     SourceLocation EndLoc, SourceLocation ColonLoc, ArrayRef<Expr *> VL,
-    NestedNameSpecifierLoc QualifierLoc, const DeclarationNameInfo &NameInfo) {
+    NestedNameSpecifierLoc QualifierLoc, const DeclarationNameInfo &NameInfo,
+    ArrayRef<Expr *> LHSExprs, ArrayRef<Expr *> RHSExprs,
+    ArrayRef<Expr *> ReductionOps) {
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPReductionClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * VL.size());
+                         4 * sizeof(Expr *) * VL.size());
   OMPReductionClause *Clause = new (Mem) OMPReductionClause(
       StartLoc, LParenLoc, EndLoc, ColonLoc, VL.size(), QualifierLoc, NameInfo);
   Clause->setVarRefs(VL);
+  Clause->setLHSExprs(LHSExprs);
+  Clause->setRHSExprs(RHSExprs);
+  Clause->setReductionOps(ReductionOps);
   return Clause;
 }
 
@@ -1416,7 +1554,7 @@ OMPReductionClause *OMPReductionClause::CreateEmpty(const ASTContext &C,
                                                     unsigned N) {
   void *Mem = C.Allocate(llvm::RoundUpToAlignment(sizeof(OMPReductionClause),
                                                   llvm::alignOf<Expr *>()) +
-                         sizeof(Expr *) * N);
+                         4 * sizeof(Expr *) * N);
   return new (Mem) OMPReductionClause(N);
 }
 
@@ -1443,10 +1581,7 @@ OMPFlushClause *OMPFlushClause::CreateEmpty(const ASTContext &C, unsigned N) {
 
 const OMPClause *
 OMPExecutableDirective::getSingleClause(OpenMPClauseKind K) const {
-  auto ClauseFilter =
-      [=](const OMPClause *C) -> bool { return C->getClauseKind() == K; };
-  OMPExecutableDirective::filtered_clause_iterator<decltype(ClauseFilter)> I(
-      clauses(), ClauseFilter);
+  auto &&I = getClausesOfKind(K);
 
   if (I) {
     auto *Clause = *I;
@@ -1948,14 +2083,14 @@ OMPOrderedDirective *OMPOrderedDirective::CreateEmpty(const ASTContext &C,
   return new (Mem) OMPOrderedDirective();
 }
 
-OMPAtomicDirective *
-OMPAtomicDirective::Create(const ASTContext &C, SourceLocation StartLoc,
-                           SourceLocation EndLoc, ArrayRef<OMPClause *> Clauses,
-                           Stmt *AssociatedStmt, Expr *X, Expr *V, Expr *E) {
+OMPAtomicDirective *OMPAtomicDirective::Create(
+    const ASTContext &C, SourceLocation StartLoc, SourceLocation EndLoc,
+    ArrayRef<OMPClause *> Clauses, Stmt *AssociatedStmt, Expr *X, Expr *V,
+    Expr *E, Expr *UE, bool IsXLHSInRHSPart, bool IsPostfixUpdate) {
   unsigned Size = llvm::RoundUpToAlignment(sizeof(OMPAtomicDirective),
                                            llvm::alignOf<OMPClause *>());
   void *Mem = C.Allocate(Size + sizeof(OMPClause *) * Clauses.size() +
-                         4 * sizeof(Stmt *));
+                         5 * sizeof(Stmt *));
   OMPAtomicDirective *Dir =
       new (Mem) OMPAtomicDirective(StartLoc, EndLoc, Clauses.size());
   Dir->setClauses(Clauses);
@@ -1963,6 +2098,9 @@ OMPAtomicDirective::Create(const ASTContext &C, SourceLocation StartLoc,
   Dir->setX(X);
   Dir->setV(V);
   Dir->setExpr(E);
+  Dir->setUpdateExpr(UE);
+  Dir->IsXLHSInRHSPart = IsXLHSInRHSPart;
+  Dir->IsPostfixUpdate = IsPostfixUpdate;
   return Dir;
 }
 
@@ -1972,7 +2110,7 @@ OMPAtomicDirective *OMPAtomicDirective::CreateEmpty(const ASTContext &C,
   unsigned Size = llvm::RoundUpToAlignment(sizeof(OMPAtomicDirective),
                                            llvm::alignOf<OMPClause *>());
   void *Mem =
-      C.Allocate(Size + sizeof(OMPClause *) * NumClauses + 4 * sizeof(Stmt *));
+      C.Allocate(Size + sizeof(OMPClause *) * NumClauses + 5 * sizeof(Stmt *));
   return new (Mem) OMPAtomicDirective(NumClauses);
 }
 
diff --git a/lib/AST/StmtPrinter.cpp b/lib/AST/StmtPrinter.cpp
index 927a679244b5..dc4f9964c7a6 100644
--- a/lib/AST/StmtPrinter.cpp
+++ b/lib/AST/StmtPrinter.cpp
@@ -1069,7 +1069,8 @@ void StmtPrinter::VisitIntegerLiteral(IntegerLiteral *Node) {
   // Emit suffixes.  Integer literals are always a builtin integer type.
   switch (Node->getType()->getAs<BuiltinType>()->getKind()) {
   default: llvm_unreachable("Unexpected type for integer literal!");
-  case BuiltinType::SChar:     OS << "i8"; break;
+  case BuiltinType::Char_S:
+  case BuiltinType::Char_U:    OS << "i8"; break;
   case BuiltinType::UChar:     OS << "Ui8"; break;
   case BuiltinType::Short:     OS << "i16"; break;
   case BuiltinType::UShort:    OS << "Ui16"; break;
@@ -1373,24 +1374,24 @@ void StmtPrinter::VisitInitListExpr(InitListExpr* Node) {
     return;
   }
 
-  OS << "{ ";
+  OS << "{";
   for (unsigned i = 0, e = Node->getNumInits(); i != e; ++i) {
     if (i) OS << ", ";
     if (Node->getInit(i))
       PrintExpr(Node->getInit(i));
     else
-      OS << "0";
+      OS << "{}";
   }
-  OS << " }";
+  OS << "}";
 }
 
 void StmtPrinter::VisitParenListExpr(ParenListExpr* Node) {
-  OS << "( ";
+  OS << "(";
   for (unsigned i = 0, e = Node->getNumExprs(); i != e; ++i) {
     if (i) OS << ", ";
     PrintExpr(Node->getExpr(i));
   }
-  OS << " )";
+  OS << ")";
 }
 
 void StmtPrinter::VisitDesignatedInitExpr(DesignatedInitExpr *Node) {
@@ -1621,6 +1622,15 @@ void StmtPrinter::VisitUserDefinedLiteral(UserDefinedLiteral *Node) {
     const TemplateArgumentList *Args =
       cast<FunctionDecl>(DRE->getDecl())->getTemplateSpecializationArgs();
     assert(Args);
+
+    if (Args->size() != 1) {
+      OS << "operator \"\" " << Node->getUDSuffix()->getName();
+      TemplateSpecializationType::PrintTemplateArgumentList(
+          OS, Args->data(), Args->size(), Policy);
+      OS << "()";
+      return;
+    }
+
     const TemplateArgument &Pack = Args->get(0);
     for (const auto &P : Pack.pack_elements()) {
       char C = (char)P.getAsIntegral().getZExtValue();
@@ -1679,9 +1689,13 @@ void StmtPrinter::VisitCXXDefaultInitExpr(CXXDefaultInitExpr *Node) {
 
 void StmtPrinter::VisitCXXFunctionalCastExpr(CXXFunctionalCastExpr *Node) {
   Node->getType().print(OS, Policy);
-  OS << "(";
+  // If there are no parens, this is list-initialization, and the braces are
+  // part of the syntax of the inner construct.
+  if (Node->getLParenLoc().isValid())
+    OS << "(";
   PrintExpr(Node->getSubExpr());
-  OS << ")";
+  if (Node->getLParenLoc().isValid())
+    OS << ")";
 }
 
 void StmtPrinter::VisitCXXBindTemporaryExpr(CXXBindTemporaryExpr *Node) {
@@ -1690,7 +1704,12 @@ void StmtPrinter::VisitCXXBindTemporaryExpr(CXXBindTemporaryExpr *Node) {
 
 void StmtPrinter::VisitCXXTemporaryObjectExpr(CXXTemporaryObjectExpr *Node) {
   Node->getType().print(OS, Policy);
-  OS << "(";
+  if (Node->isStdInitListInitialization())
+    /* Nothing to do; braces are part of creating the std::initializer_list. */;
+  else if (Node->isListInitialization())
+    OS << "{";
+  else
+    OS << "(";
   for (CXXTemporaryObjectExpr::arg_iterator Arg = Node->arg_begin(),
                                          ArgEnd = Node->arg_end();
        Arg != ArgEnd; ++Arg) {
@@ -1700,7 +1719,12 @@ void StmtPrinter::VisitCXXTemporaryObjectExpr(CXXTemporaryObjectExpr *Node) {
       OS << ", ";
     PrintExpr(*Arg);
   }
-  OS << ")";
+  if (Node->isStdInitListInitialization())
+    /* See above. */;
+  else if (Node->isListInitialization())
+    OS << "}";
+  else
+    OS << ")";
 }
 
 void StmtPrinter::VisitLambdaExpr(LambdaExpr *Node) {
@@ -1734,7 +1758,7 @@ void StmtPrinter::VisitLambdaExpr(LambdaExpr *Node) {
       break;
 
     case LCK_ByRef:
-      if (Node->getCaptureDefault() != LCD_ByRef || C->isInitCapture())
+      if (Node->getCaptureDefault() != LCD_ByRef || Node->isInitCapture(C))
         OS << '&';
       OS << C->getCapturedVar()->getName();
       break;
@@ -1746,7 +1770,7 @@ void StmtPrinter::VisitLambdaExpr(LambdaExpr *Node) {
       llvm_unreachable("VLA type in explicit captures.");
     }
 
-    if (C->isInitCapture())
+    if (Node->isInitCapture(C))
       PrintExpr(C->getCapturedVar()->getInit());
   }
   OS << ']';
@@ -1866,8 +1890,8 @@ void StmtPrinter::VisitCXXPseudoDestructorExpr(CXXPseudoDestructorExpr *E) {
 }
 
 void StmtPrinter::VisitCXXConstructExpr(CXXConstructExpr *E) {
-  if (E->isListInitialization())
-    OS << "{ ";
+  if (E->isListInitialization() && !E->isStdInitListInitialization())
+    OS << "{";
 
   for (unsigned i = 0, e = E->getNumArgs(); i != e; ++i) {
     if (isa<CXXDefaultArgExpr>(E->getArg(i))) {
@@ -1879,8 +1903,8 @@ void StmtPrinter::VisitCXXConstructExpr(CXXConstructExpr *E) {
     PrintExpr(E->getArg(i));
   }
 
-  if (E->isListInitialization())
-    OS << " }";
+  if (E->isListInitialization() && !E->isStdInitListInitialization())
+    OS << "}";
 }
 
 void StmtPrinter::VisitCXXStdInitializerListExpr(CXXStdInitializerListExpr *E) {
diff --git a/lib/AST/StmtProfile.cpp b/lib/AST/StmtProfile.cpp
index d1f25d63b3f9..c66b153eadbb 100644
--- a/lib/AST/StmtProfile.cpp
+++ b/lib/AST/StmtProfile.cpp
@@ -298,8 +298,12 @@ void OMPClauseProfiler::VisitOMPDefaultClause(const OMPDefaultClause *C) { }
 void OMPClauseProfiler::VisitOMPProcBindClause(const OMPProcBindClause *C) { }
 
 void OMPClauseProfiler::VisitOMPScheduleClause(const OMPScheduleClause *C) {
-  if (C->getChunkSize())
+  if (C->getChunkSize()) {
     Profiler->VisitStmt(C->getChunkSize());
+    if (C->getHelperChunkSize()) {
+      Profiler->VisitStmt(C->getChunkSize());
+    }
+  }
 }
 
 void OMPClauseProfiler::VisitOMPOrderedClause(const OMPOrderedClause *) {}
@@ -346,6 +350,15 @@ OMPClauseProfiler::VisitOMPFirstprivateClause(const OMPFirstprivateClause *C) {
 void
 OMPClauseProfiler::VisitOMPLastprivateClause(const OMPLastprivateClause *C) {
   VisitOMPClauseList(C);
+  for (auto *E : C->source_exprs()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->destination_exprs()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->assignment_ops()) {
+    Profiler->VisitStmt(E);
+  }
 }
 void OMPClauseProfiler::VisitOMPSharedClause(const OMPSharedClause *C) {
   VisitOMPClauseList(C);
@@ -356,10 +369,29 @@ void OMPClauseProfiler::VisitOMPReductionClause(
       C->getQualifierLoc().getNestedNameSpecifier());
   Profiler->VisitName(C->getNameInfo().getName());
   VisitOMPClauseList(C);
+  for (auto *E : C->lhs_exprs()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->rhs_exprs()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->reduction_ops()) {
+    Profiler->VisitStmt(E);
+  }
 }
 void OMPClauseProfiler::VisitOMPLinearClause(const OMPLinearClause *C) {
   VisitOMPClauseList(C);
+  for (auto *E : C->inits()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->updates()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->finals()) {
+    Profiler->VisitStmt(E);
+  }
   Profiler->VisitStmt(C->getStep());
+  Profiler->VisitStmt(C->getCalcStep());
 }
 void OMPClauseProfiler::VisitOMPAlignedClause(const OMPAlignedClause *C) {
   VisitOMPClauseList(C);
@@ -367,10 +399,28 @@ void OMPClauseProfiler::VisitOMPAlignedClause(const OMPAlignedClause *C) {
 }
 void OMPClauseProfiler::VisitOMPCopyinClause(const OMPCopyinClause *C) {
   VisitOMPClauseList(C);
+  for (auto *E : C->source_exprs()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->destination_exprs()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->assignment_ops()) {
+    Profiler->VisitStmt(E);
+  }
 }
 void
 OMPClauseProfiler::VisitOMPCopyprivateClause(const OMPCopyprivateClause *C) {
   VisitOMPClauseList(C);
+  for (auto *E : C->source_exprs()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->destination_exprs()) {
+    Profiler->VisitStmt(E);
+  }
+  for (auto *E : C->assignment_ops()) {
+    Profiler->VisitStmt(E);
+  }
 }
 void OMPClauseProfiler::VisitOMPFlushClause(const OMPFlushClause *C) {
   VisitOMPClauseList(C);
diff --git a/lib/AST/Type.cpp b/lib/AST/Type.cpp
index e4f364d04f86..09bb7692596c 100644
--- a/lib/AST/Type.cpp
+++ b/lib/AST/Type.cpp
@@ -170,15 +170,7 @@ DependentSizedExtVectorType::Profile(llvm::FoldingSetNodeID &ID,
 
 VectorType::VectorType(QualType vecType, unsigned nElements, QualType canonType,
                        VectorKind vecKind)
-  : Type(Vector, canonType, vecType->isDependentType(),
-         vecType->isInstantiationDependentType(),
-         vecType->isVariablyModifiedType(),
-         vecType->containsUnexpandedParameterPack()),
-    ElementType(vecType) 
-{
-  VectorTypeBits.VecKind = vecKind;
-  VectorTypeBits.NumElements = nElements;
-}
+    : VectorType(Vector, vecType, nElements, canonType, vecKind) {}
 
 VectorType::VectorType(TypeClass tc, QualType vecType, unsigned nElements,
                        QualType canonType, VectorKind vecKind)
@@ -640,12 +632,13 @@ bool Type::hasIntegerRepresentation() const {
 bool Type::isIntegralType(ASTContext &Ctx) const {
   if (const BuiltinType *BT = dyn_cast<BuiltinType>(CanonicalType))
     return BT->getKind() >= BuiltinType::Bool &&
-    BT->getKind() <= BuiltinType::Int128;
-  
+           BT->getKind() <= BuiltinType::Int128;
+
+  // Complete enum types are integral in C.
   if (!Ctx.getLangOpts().CPlusPlus)
     if (const EnumType *ET = dyn_cast<EnumType>(CanonicalType))
-      return ET->getDecl()->isComplete(); // Complete enum types are integral in C.
-  
+      return ET->getDecl()->isComplete();
+
   return false;
 }
 
@@ -736,7 +729,7 @@ bool Type::isSignedIntegerType() const {
 bool Type::isSignedIntegerOrEnumerationType() const {
   if (const BuiltinType *BT = dyn_cast<BuiltinType>(CanonicalType)) {
     return BT->getKind() >= BuiltinType::Char_S &&
-    BT->getKind() <= BuiltinType::Int128;
+           BT->getKind() <= BuiltinType::Int128;
   }
   
   if (const EnumType *ET = dyn_cast<EnumType>(CanonicalType)) {
@@ -1089,7 +1082,7 @@ bool QualType::isTrivialType(ASTContext &Context) const {
 
 bool QualType::isTriviallyCopyableType(ASTContext &Context) const {
   if ((*this)->isArrayType())
-    return Context.getBaseElementType(*this).isTrivialType(Context);
+    return Context.getBaseElementType(*this).isTriviallyCopyableType(Context);
 
   if (Context.getLangOpts().ObjCAutoRefCount) {
     switch (getObjCLifetime()) {
@@ -1586,8 +1579,9 @@ StringRef FunctionType::getNameForCallConv(CallingConv CC) {
   case CC_X86_64SysV: return "sysv_abi";
   case CC_AAPCS: return "aapcs";
   case CC_AAPCS_VFP: return "aapcs-vfp";
-  case CC_PnaclCall: return "pnaclcall";
   case CC_IntelOclBicc: return "intel_ocl_bicc";
+  case CC_SpirFunction: return "spir_function";
+  case CC_SpirKernel: return "spir_kernel";
   }
 
   llvm_unreachable("Invalid calling convention.");
@@ -1720,7 +1714,7 @@ bool FunctionProtoType::isNothrow(const ASTContext &Ctx,
   if (EST == EST_DynamicNone || EST == EST_BasicNoexcept)
     return true;
 
-  if (EST == EST_Dynamic && ResultIfDependent == true) {
+  if (EST == EST_Dynamic && ResultIfDependent) {
     // A dynamic exception specification is throwing unless every exception
     // type is an (unexpanded) pack expansion type.
     for (unsigned I = 0, N = NumExceptions; I != N; ++I)
@@ -1936,7 +1930,6 @@ bool AttributedType::isCallingConv() const {
   case attr_pascal:
   case attr_ms_abi:
   case attr_sysv_abi:
-  case attr_pnaclcall:
   case attr_inteloclbicc:
     return true;
   }
@@ -2378,6 +2371,11 @@ bool Type::isObjCNSObjectType() const {
     return typedefType->getDecl()->hasAttr<ObjCNSObjectAttr>();
   return false;
 }
+bool Type::isObjCIndependentClassType() const {
+  if (const TypedefType *typedefType = dyn_cast<TypedefType>(this))
+    return typedefType->getDecl()->hasAttr<ObjCIndependentClassAttr>();
+  return false;
+}
 bool Type::isObjCRetainableType() const {
   return isObjCObjectPointerType() ||
          isBlockPointerType() ||
diff --git a/lib/AST/TypePrinter.cpp b/lib/AST/TypePrinter.cpp
index e36fc175c449..3928fe8f8942 100644
--- a/lib/AST/TypePrinter.cpp
+++ b/lib/AST/TypePrinter.cpp
@@ -110,7 +110,7 @@ namespace {
   };
 }
 
-static void AppendTypeQualList(raw_ostream &OS, unsigned TypeQuals) {
+static void AppendTypeQualList(raw_ostream &OS, unsigned TypeQuals, bool C99) {
   bool appendSpace = false;
   if (TypeQuals & Qualifiers::Const) {
     OS << "const";
@@ -123,7 +123,11 @@ static void AppendTypeQualList(raw_ostream &OS, unsigned TypeQuals) {
   }
   if (TypeQuals & Qualifiers::Restrict) {
     if (appendSpace) OS << ' ';
-    OS << "restrict";
+    if (C99) {
+      OS << "restrict";
+    } else {
+      OS << "__restrict";
+    }
   }
 }
 
@@ -432,7 +436,7 @@ void TypePrinter::printConstantArrayAfter(const ConstantArrayType *T,
                                           raw_ostream &OS) {
   OS << '[';
   if (T->getIndexTypeQualifiers().hasQualifiers()) {
-    AppendTypeQualList(OS, T->getIndexTypeCVRQualifiers());
+    AppendTypeQualList(OS, T->getIndexTypeCVRQualifiers(), Policy.LangOpts.C99);
     OS << ' ';
   }
 
@@ -465,7 +469,7 @@ void TypePrinter::printVariableArrayAfter(const VariableArrayType *T,
                                           raw_ostream &OS) {
   OS << '[';
   if (T->getIndexTypeQualifiers().hasQualifiers()) {
-    AppendTypeQualList(OS, T->getIndexTypeCVRQualifiers());
+    AppendTypeQualList(OS, T->getIndexTypeCVRQualifiers(), Policy.LangOpts.C99);
     OS << ' ';
   }
 
@@ -685,9 +689,6 @@ void TypePrinter::printFunctionProtoAfter(const FunctionProtoType *T,
     case CC_AAPCS_VFP:
       OS << " __attribute__((pcs(\"aapcs-vfp\")))";
       break;
-    case CC_PnaclCall:
-      OS << " __attribute__((pnaclcall))";
-      break;
     case CC_IntelOclBicc:
       OS << " __attribute__((intel_ocl_bicc))";
       break;
@@ -697,6 +698,10 @@ void TypePrinter::printFunctionProtoAfter(const FunctionProtoType *T,
     case CC_X86_64SysV:
       OS << " __attribute__((sysv_abi))";
       break;
+    case CC_SpirFunction:
+    case CC_SpirKernel:
+      // Do nothing. These CCs are not available as attributes.
+      break;
     }
   }
 
@@ -708,7 +713,7 @@ void TypePrinter::printFunctionProtoAfter(const FunctionProtoType *T,
 
   if (unsigned quals = T->getTypeQuals()) {
     OS << ' ';
-    AppendTypeQualList(OS, quals);
+    AppendTypeQualList(OS, quals, Policy.LangOpts.C99);
   }
 
   switch (T->getRefQualifier()) {
@@ -1253,7 +1258,6 @@ void TypePrinter::printAttributedAfter(const AttributedType *T,
    OS << ')';
    break;
   }
-  case AttributedType::attr_pnaclcall: OS << "pnaclcall"; break;
   case AttributedType::attr_inteloclbicc: OS << "inteloclbicc"; break;
   }
   OS << "))";
@@ -1473,7 +1477,7 @@ void Qualifiers::print(raw_ostream &OS, const PrintingPolicy& Policy,
 
   unsigned quals = getCVRQualifiers();
   if (quals) {
-    AppendTypeQualList(OS, quals);
+    AppendTypeQualList(OS, quals, Policy.LangOpts.C99);
     addSpace = true;
   }
   if (unsigned addrspace = getAddressSpace()) {
diff --git a/lib/AST/VTableBuilder.cpp b/lib/AST/VTableBuilder.cpp
index ddb1f057ac8e..ca5f0aad0013 100644
--- a/lib/AST/VTableBuilder.cpp
+++ b/lib/AST/VTableBuilder.cpp
@@ -13,9 +13,11 @@
 
 #include "clang/AST/VTableBuilder.h"
 #include "clang/AST/ASTContext.h"
+#include "clang/AST/ASTDiagnostic.h"
 #include "clang/AST/CXXInheritance.h"
 #include "clang/AST/RecordLayout.h"
 #include "clang/Basic/TargetInfo.h"
+#include "llvm/ADT/SetOperations.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/Support/Format.h"
 #include "llvm/Support/raw_ostream.h"
@@ -216,7 +218,7 @@ FinalOverriders::FinalOverriders(const CXXRecordDecl *MostDerivedClass,
 #endif
 }
 
-static BaseOffset ComputeBaseOffset(ASTContext &Context, 
+static BaseOffset ComputeBaseOffset(const ASTContext &Context,
                                     const CXXRecordDecl *DerivedRD,
                                     const CXXBasePath &Path) {
   CharUnits NonVirtualOffset = CharUnits::Zero();
@@ -255,7 +257,7 @@ static BaseOffset ComputeBaseOffset(ASTContext &Context,
   
 }
 
-static BaseOffset ComputeBaseOffset(ASTContext &Context, 
+static BaseOffset ComputeBaseOffset(const ASTContext &Context,
                                     const CXXRecordDecl *BaseRD,
                                     const CXXRecordDecl *DerivedRD) {
   CXXBasePaths Paths(/*FindAmbiguities=*/false,
@@ -411,7 +413,8 @@ void FinalOverriders::dump(raw_ostream &Out, BaseSubobject Base,
   for (const auto *MD : RD->methods()) {
     if (!MD->isVirtual())
       continue;
-  
+    MD = MD->getCanonicalDecl();
+
     OverriderInfo Overrider = getOverrider(MD, Base.getBaseOffset());
 
     Out << "  ";
@@ -695,6 +698,7 @@ void VCallAndVBaseOffsetBuilder::AddVCallOffsets(BaseSubobject Base,
   for (const auto *MD : RD->methods()) {
     if (!MD->isVirtual())
       continue;
+    MD = MD->getCanonicalDecl();
 
     CharUnits OffsetOffset = getCurrentOffsetOffset();
     
@@ -1514,6 +1518,7 @@ void ItaniumVTableBuilder::AddMethods(
   for (const auto *MD : RD->methods()) {
     if (!MD->isVirtual())
       continue;
+    MD = MD->getCanonicalDecl();
 
     // Get the final overrider.
     FinalOverriders::OverriderInfo Overrider = 
@@ -2196,6 +2201,7 @@ void ItaniumVTableBuilder::dumpLayout(raw_ostream &Out) {
     // We only want virtual member functions.
     if (!MD->isVirtual())
       continue;
+    MD = MD->getCanonicalDecl();
 
     std::string MethodName =
       PredefinedExpr::ComputeName(PredefinedExpr::PrettyFunctionNoVirtual,
@@ -2585,7 +2591,9 @@ public:
     // Only include the RTTI component if we know that we will provide a
     // definition of the vftable.
     HasRTTIComponent = Context.getLangOpts().RTTIData &&
-                       !MostDerivedClass->hasAttr<DLLImportAttr>();
+                       !MostDerivedClass->hasAttr<DLLImportAttr>() &&
+                       MostDerivedClass->getTemplateSpecializationKind() !=
+                           TSK_ExplicitInstantiationDeclaration;
 
     LayoutVFTable();
 
@@ -2625,8 +2633,6 @@ public:
   void dumpLayout(raw_ostream &);
 };
 
-} // end namespace
-
 /// InitialOverriddenDefinitionCollector - Finds the set of least derived bases
 /// that define the given method.
 struct InitialOverriddenDefinitionCollector {
@@ -2641,6 +2647,8 @@ struct InitialOverriddenDefinitionCollector {
   }
 };
 
+} // end namespace
+
 static bool BaseInSet(const CXXBaseSpecifier *Specifier,
                       CXXBasePath &Path, void *BasesSet) {
   BasesSetVectorTy *Bases = (BasesSetVectorTy *)BasesSet;
@@ -2730,8 +2738,9 @@ VFTableBuilder::ComputeThisOffset(FinalOverriders::OverriderInfo Overrider) {
     CharUnits ThisOffset = Overrider.Offset;
     CharUnits LastVBaseOffset;
 
-    // For each path from the overrider to the parents of the overridden methods,
-    // traverse the path, calculating the this offset in the most derived class.
+    // For each path from the overrider to the parents of the overridden
+    // methods, traverse the path, calculating the this offset in the most
+    // derived class.
     for (int J = 0, F = Path.size(); J != F; ++J) {
       const CXXBasePathElement &Element = Path[J];
       QualType CurTy = Element.Base->getType();
@@ -2930,6 +2939,7 @@ static void GroupNewVirtualOverloads(
   typedef llvm::DenseMap<DeclarationName, unsigned> VisitedGroupIndicesTy;
   VisitedGroupIndicesTy VisitedGroupIndices;
   for (const auto *MD : RD->methods()) {
+    MD = MD->getCanonicalDecl();
     VisitedGroupIndicesTy::iterator J;
     bool Inserted;
     std::tie(J, Inserted) = VisitedGroupIndices.insert(
@@ -2962,7 +2972,8 @@ void VFTableBuilder::AddMethods(BaseSubobject Base, unsigned BaseDepth,
   const ASTRecordLayout &Layout = Context.getASTRecordLayout(RD);
 
   // See if this class expands a vftable of the base we look at, which is either
-  // the one defined by the vfptr base path or the primary base of the current class.
+  // the one defined by the vfptr base path or the primary base of the current
+  // class.
   const CXXRecordDecl *NextBase = nullptr, *NextLastVBase = LastVBase;
   CharUnits NextBaseOffset;
   if (BaseDepth < WhichVFPtr.PathToBaseWithVPtr.size()) {
@@ -3020,7 +3031,8 @@ void VFTableBuilder::AddMethods(BaseSubobject Base, unsigned BaseDepth,
                                   ThisAdjustmentOffset);
 
     if (OverriddenMD) {
-      // If MD overrides anything in this vftable, we need to update the entries.
+      // If MD overrides anything in this vftable, we need to update the
+      // entries.
       MethodInfoMapTy::iterator OverriddenMDIterator =
           MethodInfoMap.find(OverriddenMD);
 
@@ -3435,55 +3447,176 @@ MicrosoftVTableContext::~MicrosoftVTableContext() {
   llvm::DeleteContainerSeconds(VBaseInfo);
 }
 
-static bool
-findPathForVPtr(ASTContext &Context, const ASTRecordLayout &MostDerivedLayout,
-                const CXXRecordDecl *RD, CharUnits Offset,
-                llvm::SmallPtrSetImpl<const CXXRecordDecl *> &VBasesSeen,
-                VPtrInfo::BasePath &FullPath, VPtrInfo *Info) {
-  if (RD == Info->BaseWithVPtr && Offset == Info->FullOffsetInMDC) {
-    Info->PathToBaseWithVPtr = FullPath;
-    return true;
+namespace {
+typedef llvm::SetVector<BaseSubobject, std::vector<BaseSubobject>,
+                        llvm::DenseSet<BaseSubobject>> FullPathTy;
+}
+
+// This recursive function finds all paths from a subobject centered at
+// (RD, Offset) to the subobject located at BaseWithVPtr.
+static void findPathsToSubobject(ASTContext &Context,
+                                 const ASTRecordLayout &MostDerivedLayout,
+                                 const CXXRecordDecl *RD, CharUnits Offset,
+                                 BaseSubobject BaseWithVPtr,
+                                 FullPathTy &FullPath,
+                                 std::list<FullPathTy> &Paths) {
+  if (BaseSubobject(RD, Offset) == BaseWithVPtr) {
+    Paths.push_back(FullPath);
+    return;
   }
 
   const ASTRecordLayout &Layout = Context.getASTRecordLayout(RD);
 
-  // Recurse with non-virtual bases first.
-  // FIXME: Does this need to be in layout order? Virtual bases will be in base
-  // specifier order, which isn't necessarily layout order.
-  SmallVector<CXXBaseSpecifier, 4> Bases(RD->bases_begin(), RD->bases_end());
-  std::stable_partition(Bases.begin(), Bases.end(),
-                        [](CXXBaseSpecifier bs) { return !bs.isVirtual(); });
-
-  for (const auto &B : Bases) {
-    const CXXRecordDecl *Base = B.getType()->getAsCXXRecordDecl();
-    CharUnits NewOffset;
-    if (!B.isVirtual())
-      NewOffset = Offset + Layout.getBaseClassOffset(Base);
-    else {
-      if (!VBasesSeen.insert(Base).second)
-        return false;
-      NewOffset = MostDerivedLayout.getVBaseClassOffset(Base);
-    }
-    FullPath.push_back(Base);
-    if (findPathForVPtr(Context, MostDerivedLayout, Base, NewOffset, VBasesSeen,
-                        FullPath, Info))
-      return true;
+  for (const CXXBaseSpecifier &BS : RD->bases()) {
+    const CXXRecordDecl *Base = BS.getType()->getAsCXXRecordDecl();
+    CharUnits NewOffset = BS.isVirtual()
+                              ? MostDerivedLayout.getVBaseClassOffset(Base)
+                              : Offset + Layout.getBaseClassOffset(Base);
+    FullPath.insert(BaseSubobject(Base, NewOffset));
+    findPathsToSubobject(Context, MostDerivedLayout, Base, NewOffset,
+                         BaseWithVPtr, FullPath, Paths);
     FullPath.pop_back();
   }
-  return false;
+}
+
+// Return the paths which are not subsets of other paths.
+static void removeRedundantPaths(std::list<FullPathTy> &FullPaths) {
+  FullPaths.remove_if([&](const FullPathTy &SpecificPath) {
+    for (const FullPathTy &OtherPath : FullPaths) {
+      if (&SpecificPath == &OtherPath)
+        continue;
+      if (std::all_of(SpecificPath.begin(), SpecificPath.end(),
+                      [&](const BaseSubobject &BSO) {
+                        return OtherPath.count(BSO) != 0;
+                      })) {
+        return true;
+      }
+    }
+    return false;
+  });
+}
+
+static CharUnits getOffsetOfFullPath(ASTContext &Context,
+                                     const CXXRecordDecl *RD,
+                                     const FullPathTy &FullPath) {
+  const ASTRecordLayout &MostDerivedLayout =
+      Context.getASTRecordLayout(RD);
+  CharUnits Offset = CharUnits::fromQuantity(-1);
+  for (const BaseSubobject &BSO : FullPath) {
+    const CXXRecordDecl *Base = BSO.getBase();
+    // The first entry in the path is always the most derived record, skip it.
+    if (Base == RD) {
+      assert(Offset.getQuantity() == -1);
+      Offset = CharUnits::Zero();
+      continue;
+    }
+    assert(Offset.getQuantity() != -1);
+    const ASTRecordLayout &Layout = Context.getASTRecordLayout(RD);
+    // While we know which base has to be traversed, we don't know if that base
+    // was a virtual base.
+    const CXXBaseSpecifier *BaseBS = std::find_if(
+        RD->bases_begin(), RD->bases_end(), [&](const CXXBaseSpecifier &BS) {
+          return BS.getType()->getAsCXXRecordDecl() == Base;
+        });
+    Offset = BaseBS->isVirtual() ? MostDerivedLayout.getVBaseClassOffset(Base)
+                                 : Offset + Layout.getBaseClassOffset(Base);
+    RD = Base;
+  }
+  return Offset;
+}
+
+// We want to select the path which introduces the most covariant overrides.  If
+// two paths introduce overrides which the other path doesn't contain, issue a
+// diagnostic.
+static const FullPathTy *selectBestPath(ASTContext &Context,
+                                        const CXXRecordDecl *RD, VPtrInfo *Info,
+                                        std::list<FullPathTy> &FullPaths) {
+  // Handle some easy cases first.
+  if (FullPaths.empty())
+    return nullptr;
+  if (FullPaths.size() == 1)
+    return &FullPaths.front();
+
+  const FullPathTy *BestPath = nullptr;
+  typedef std::set<const CXXMethodDecl *> OverriderSetTy;
+  OverriderSetTy LastOverrides;
+  for (const FullPathTy &SpecificPath : FullPaths) {
+    assert(!SpecificPath.empty());
+    OverriderSetTy CurrentOverrides;
+    const CXXRecordDecl *TopLevelRD = SpecificPath.begin()->getBase();
+    // Find the distance from the start of the path to the subobject with the
+    // VPtr.
+    CharUnits BaseOffset =
+        getOffsetOfFullPath(Context, TopLevelRD, SpecificPath);
+    FinalOverriders Overriders(TopLevelRD, CharUnits::Zero(), TopLevelRD);
+    for (const CXXMethodDecl *MD : Info->BaseWithVPtr->methods()) {
+      if (!MD->isVirtual())
+        continue;
+      FinalOverriders::OverriderInfo OI =
+          Overriders.getOverrider(MD->getCanonicalDecl(), BaseOffset);
+      const CXXMethodDecl *OverridingMethod = OI.Method;
+      // Only overriders which have a return adjustment introduce problematic
+      // thunks.
+      if (ComputeReturnAdjustmentBaseOffset(Context, OverridingMethod, MD)
+              .isEmpty())
+        continue;
+      // It's possible that the overrider isn't in this path.  If so, skip it
+      // because this path didn't introduce it.
+      const CXXRecordDecl *OverridingParent = OverridingMethod->getParent();
+      if (std::none_of(SpecificPath.begin(), SpecificPath.end(),
+                       [&](const BaseSubobject &BSO) {
+                         return BSO.getBase() == OverridingParent;
+                       }))
+        continue;
+      CurrentOverrides.insert(OverridingMethod);
+    }
+    OverriderSetTy NewOverrides =
+        llvm::set_difference(CurrentOverrides, LastOverrides);
+    if (NewOverrides.empty())
+      continue;
+    OverriderSetTy MissingOverrides =
+        llvm::set_difference(LastOverrides, CurrentOverrides);
+    if (MissingOverrides.empty()) {
+      // This path is a strict improvement over the last path, let's use it.
+      BestPath = &SpecificPath;
+      std::swap(CurrentOverrides, LastOverrides);
+    } else {
+      // This path introduces an overrider with a conflicting covariant thunk.
+      DiagnosticsEngine &Diags = Context.getDiagnostics();
+      const CXXMethodDecl *CovariantMD = *NewOverrides.begin();
+      const CXXMethodDecl *ConflictMD = *MissingOverrides.begin();
+      Diags.Report(RD->getLocation(), diag::err_vftable_ambiguous_component)
+          << RD;
+      Diags.Report(CovariantMD->getLocation(), diag::note_covariant_thunk)
+          << CovariantMD;
+      Diags.Report(ConflictMD->getLocation(), diag::note_covariant_thunk)
+          << ConflictMD;
+    }
+  }
+  // Go with the path that introduced the most covariant overrides.  If there is
+  // no such path, pick the first path.
+  return BestPath ? BestPath : &FullPaths.front();
 }
 
 static void computeFullPathsForVFTables(ASTContext &Context,
                                         const CXXRecordDecl *RD,
                                         VPtrInfoVector &Paths) {
-  llvm::SmallPtrSet<const CXXRecordDecl*, 4> VBasesSeen;
   const ASTRecordLayout &MostDerivedLayout = Context.getASTRecordLayout(RD);
-  VPtrInfo::BasePath FullPath;
+  FullPathTy FullPath;
+  std::list<FullPathTy> FullPaths;
   for (VPtrInfo *Info : Paths) {
-    findPathForVPtr(Context, MostDerivedLayout, RD, CharUnits::Zero(),
-                    VBasesSeen, FullPath, Info);
-    VBasesSeen.clear();
+    findPathsToSubobject(
+        Context, MostDerivedLayout, RD, CharUnits::Zero(),
+        BaseSubobject(Info->BaseWithVPtr, Info->FullOffsetInMDC), FullPath,
+        FullPaths);
     FullPath.clear();
+    removeRedundantPaths(FullPaths);
+    Info->PathToBaseWithVPtr.clear();
+    if (const FullPathTy *BestPath =
+            selectBestPath(Context, RD, Info, FullPaths))
+      for (const BaseSubobject &BSO : *BestPath)
+        Info->PathToBaseWithVPtr.push_back(BSO.getBase());
+    FullPaths.clear();
   }
 }
 
diff --git a/lib/ASTMatchers/ASTMatchFinder.cpp b/lib/ASTMatchers/ASTMatchFinder.cpp
index fa7968a805c4..c5f3063fb4f3 100644
--- a/lib/ASTMatchers/ASTMatchFinder.cpp
+++ b/lib/ASTMatchers/ASTMatchFinder.cpp
@@ -300,7 +300,7 @@ public:
                   const MatchFinder::MatchFinderOptions &Options)
       : Matchers(Matchers), Options(Options), ActiveASTContext(nullptr) {}
 
-  ~MatchASTVisitor() {
+  ~MatchASTVisitor() override {
     if (Options.CheckProfiling) {
       Options.CheckProfiling->Records = std::move(TimeByBucket);
     }
diff --git a/lib/ASTMatchers/CMakeLists.txt b/lib/ASTMatchers/CMakeLists.txt
index 8a8e8c36569f..b49528ede227 100644
--- a/lib/ASTMatchers/CMakeLists.txt
+++ b/lib/ASTMatchers/CMakeLists.txt
@@ -8,4 +8,5 @@ add_clang_library(clangASTMatchers
 
   LINK_LIBS
   clangAST
+  clangBasic
   )
diff --git a/lib/ASTMatchers/Dynamic/Diagnostics.cpp b/lib/ASTMatchers/Dynamic/Diagnostics.cpp
index b0abdc7cf810..f6d34494dec5 100644
--- a/lib/ASTMatchers/Dynamic/Diagnostics.cpp
+++ b/lib/ASTMatchers/Dynamic/Diagnostics.cpp
@@ -12,7 +12,6 @@
 namespace clang {
 namespace ast_matchers {
 namespace dynamic {
-
 Diagnostics::ArgStream Diagnostics::pushContextFrame(ContextType Type,
                                                      SourceRange Range) {
   ContextStack.push_back(ContextFrame());
@@ -75,7 +74,7 @@ Diagnostics::ArgStream Diagnostics::addError(const SourceRange &Range,
   return ArgStream(&Last.Messages.back().Args);
 }
 
-StringRef contextTypeToFormatString(Diagnostics::ContextType Type) {
+static StringRef contextTypeToFormatString(Diagnostics::ContextType Type) {
   switch (Type) {
     case Diagnostics::CT_MatcherConstruct:
       return "Error building matcher $0.";
@@ -85,7 +84,7 @@ StringRef contextTypeToFormatString(Diagnostics::ContextType Type) {
   llvm_unreachable("Unknown ContextType value.");
 }
 
-StringRef errorTypeToFormatString(Diagnostics::ErrorType Type) {
+static StringRef errorTypeToFormatString(Diagnostics::ErrorType Type) {
   switch (Type) {
   case Diagnostics::ET_RegistryMatcherNotFound:
     return "Matcher not found: $0";
@@ -130,8 +129,9 @@ StringRef errorTypeToFormatString(Diagnostics::ErrorType Type) {
   llvm_unreachable("Unknown ErrorType value.");
 }
 
-void formatErrorString(StringRef FormatString, ArrayRef<std::string> Args,
-                       llvm::raw_ostream &OS) {
+static void formatErrorString(StringRef FormatString,
+                              ArrayRef<std::string> Args,
+                              llvm::raw_ostream &OS) {
   while (!FormatString.empty()) {
     std::pair<StringRef, StringRef> Pieces = FormatString.split("$");
     OS << Pieces.first.str();
diff --git a/lib/ASTMatchers/Dynamic/Marshallers.h b/lib/ASTMatchers/Dynamic/Marshallers.h
index b78bc0381990..36a6415ae82c 100644
--- a/lib/ASTMatchers/Dynamic/Marshallers.h
+++ b/lib/ASTMatchers/Dynamic/Marshallers.h
@@ -181,18 +181,20 @@ public:
         ArgKinds(ArgKinds.begin(), ArgKinds.end()) {}
 
   VariantMatcher create(const SourceRange &NameRange,
-                        ArrayRef<ParserValue> Args, Diagnostics *Error) const {
+                        ArrayRef<ParserValue> Args,
+                        Diagnostics *Error) const override {
     return Marshaller(Func, MatcherName, NameRange, Args, Error);
   }
 
-  bool isVariadic() const { return false; }
-  unsigned getNumArgs() const { return ArgKinds.size(); }
+  bool isVariadic() const override { return false; }
+  unsigned getNumArgs() const override { return ArgKinds.size(); }
   void getArgKinds(ast_type_traits::ASTNodeKind ThisKind, unsigned ArgNo,
-                   std::vector<ArgKind> &Kinds) const {
+                   std::vector<ArgKind> &Kinds) const override {
     Kinds.push_back(ArgKinds[ArgNo]);
   }
-  bool isConvertibleTo(ast_type_traits::ASTNodeKind Kind, unsigned *Specificity,
-                       ast_type_traits::ASTNodeKind *LeastDerivedKind) const {
+  bool isConvertibleTo(
+      ast_type_traits::ASTNodeKind Kind, unsigned *Specificity,
+      ast_type_traits::ASTNodeKind *LeastDerivedKind) const override {
     return isRetKindConvertibleTo(RetKinds, Kind, Specificity,
                                   LeastDerivedKind);
   }
@@ -333,18 +335,20 @@ public:
   }
 
   VariantMatcher create(const SourceRange &NameRange,
-                        ArrayRef<ParserValue> Args, Diagnostics *Error) const {
+                        ArrayRef<ParserValue> Args,
+                        Diagnostics *Error) const override {
     return Func(MatcherName, NameRange, Args, Error);
   }
 
-  bool isVariadic() const { return true; }
-  unsigned getNumArgs() const { return 0; }
+  bool isVariadic() const override { return true; }
+  unsigned getNumArgs() const override { return 0; }
   void getArgKinds(ast_type_traits::ASTNodeKind ThisKind, unsigned ArgNo,
-                   std::vector<ArgKind> &Kinds) const {
+                   std::vector<ArgKind> &Kinds) const override {
     Kinds.push_back(ArgsKind);
   }
-  bool isConvertibleTo(ast_type_traits::ASTNodeKind Kind, unsigned *Specificity,
-                       ast_type_traits::ASTNodeKind *LeastDerivedKind) const {
+  bool isConvertibleTo(
+      ast_type_traits::ASTNodeKind Kind, unsigned *Specificity,
+      ast_type_traits::ASTNodeKind *LeastDerivedKind) const override {
     return isRetKindConvertibleTo(RetKinds, Kind, Specificity,
                                   LeastDerivedKind);
   }
@@ -487,11 +491,11 @@ public:
   OverloadedMatcherDescriptor(ArrayRef<MatcherDescriptor *> Callbacks)
       : Overloads(Callbacks.begin(), Callbacks.end()) {}
 
-  virtual ~OverloadedMatcherDescriptor() {}
+  ~OverloadedMatcherDescriptor() override {}
 
-  virtual VariantMatcher create(const SourceRange &NameRange,
-                                ArrayRef<ParserValue> Args,
-                                Diagnostics *Error) const {
+  VariantMatcher create(const SourceRange &NameRange,
+                        ArrayRef<ParserValue> Args,
+                        Diagnostics *Error) const override {
     std::vector<VariantMatcher> Constructed;
     Diagnostics::OverloadContext Ctx(Error);
     for (const auto &O : Overloads) {
@@ -512,7 +516,7 @@ public:
     return Constructed[0];
   }
 
-  bool isVariadic() const {
+  bool isVariadic() const override {
     bool Overload0Variadic = Overloads[0]->isVariadic();
 #ifndef NDEBUG
     for (const auto &O : Overloads) {
@@ -522,7 +526,7 @@ public:
     return Overload0Variadic;
   }
 
-  unsigned getNumArgs() const {
+  unsigned getNumArgs() const override {
     unsigned Overload0NumArgs = Overloads[0]->getNumArgs();
 #ifndef NDEBUG
     for (const auto &O : Overloads) {
@@ -533,15 +537,16 @@ public:
   }
 
   void getArgKinds(ast_type_traits::ASTNodeKind ThisKind, unsigned ArgNo,
-                   std::vector<ArgKind> &Kinds) const {
+                   std::vector<ArgKind> &Kinds) const override {
     for (const auto &O : Overloads) {
       if (O->isConvertibleTo(ThisKind))
         O->getArgKinds(ThisKind, ArgNo, Kinds);
     }
   }
 
-  bool isConvertibleTo(ast_type_traits::ASTNodeKind Kind, unsigned *Specificity,
-                       ast_type_traits::ASTNodeKind *LeastDerivedKind) const {
+  bool isConvertibleTo(
+      ast_type_traits::ASTNodeKind Kind, unsigned *Specificity,
+      ast_type_traits::ASTNodeKind *LeastDerivedKind) const override {
     for (const auto &O : Overloads) {
       if (O->isConvertibleTo(Kind, Specificity, LeastDerivedKind))
         return true;
@@ -562,9 +567,9 @@ public:
       : MinCount(MinCount), MaxCount(MaxCount), Op(Op),
         MatcherName(MatcherName) {}
 
-  virtual VariantMatcher create(const SourceRange &NameRange,
-                                ArrayRef<ParserValue> Args,
-                                Diagnostics *Error) const override {
+  VariantMatcher create(const SourceRange &NameRange,
+                        ArrayRef<ParserValue> Args,
+                        Diagnostics *Error) const override {
     if (Args.size() < MinCount || MaxCount < Args.size()) {
       const std::string MaxStr =
           (MaxCount == UINT_MAX ? "" : Twine(MaxCount)).str();
diff --git a/lib/ASTMatchers/Dynamic/Registry.cpp b/lib/ASTMatchers/Dynamic/Registry.cpp
index d550a89cad49..04d3a3256313 100644
--- a/lib/ASTMatchers/Dynamic/Registry.cpp
+++ b/lib/ASTMatchers/Dynamic/Registry.cpp
@@ -128,6 +128,7 @@ RegistryMaps::RegistryMaps() {
   REGISTER_MATCHER(constructorDecl);
   REGISTER_MATCHER(containsDeclaration);
   REGISTER_MATCHER(continueStmt);
+  REGISTER_MATCHER(conversionDecl);
   REGISTER_MATCHER(cStyleCastExpr);
   REGISTER_MATCHER(ctorInitializer);
   REGISTER_MATCHER(CUDAKernelCallExpr);
@@ -198,6 +199,7 @@ RegistryMaps::RegistryMaps() {
   REGISTER_MATCHER(hasIncrement);
   REGISTER_MATCHER(hasIndex);
   REGISTER_MATCHER(hasInitializer);
+  REGISTER_MATCHER(hasKeywordSelector);
   REGISTER_MATCHER(hasLHS);
   REGISTER_MATCHER(hasLocalQualifiers);
   REGISTER_MATCHER(hasLocalStorage);
@@ -205,6 +207,7 @@ RegistryMaps::RegistryMaps() {
   REGISTER_MATCHER(hasLoopVariable);
   REGISTER_MATCHER(hasMethod);
   REGISTER_MATCHER(hasName);
+  REGISTER_MATCHER(hasNullSelector);
   REGISTER_MATCHER(hasObjectExpression);
   REGISTER_MATCHER(hasOperatorName);
   REGISTER_MATCHER(hasOverloadedOperatorName);
@@ -212,7 +215,9 @@ RegistryMaps::RegistryMaps() {
   REGISTER_MATCHER(hasParent);
   REGISTER_MATCHER(hasQualifier);
   REGISTER_MATCHER(hasRangeInit);
+  REGISTER_MATCHER(hasReceiverType);
   REGISTER_MATCHER(hasRHS);
+  REGISTER_MATCHER(hasSelector);
   REGISTER_MATCHER(hasSingleDecl);
   REGISTER_MATCHER(hasSize);
   REGISTER_MATCHER(hasSizeExpr);
@@ -223,6 +228,7 @@ RegistryMaps::RegistryMaps() {
   REGISTER_MATCHER(hasTrueExpression);
   REGISTER_MATCHER(hasTypeLoc);
   REGISTER_MATCHER(hasUnaryOperand);
+  REGISTER_MATCHER(hasUnarySelector);
   REGISTER_MATCHER(hasValueType);
   REGISTER_MATCHER(ifStmt);
   REGISTER_MATCHER(ignoringImpCasts);
@@ -262,6 +268,7 @@ RegistryMaps::RegistryMaps() {
   REGISTER_MATCHER(lambdaExpr);
   REGISTER_MATCHER(lValueReferenceType);
   REGISTER_MATCHER(matchesName);
+  REGISTER_MATCHER(matchesSelector);
   REGISTER_MATCHER(materializeTemporaryExpr);
   REGISTER_MATCHER(member);
   REGISTER_MATCHER(memberCallExpr);
@@ -276,7 +283,9 @@ RegistryMaps::RegistryMaps() {
   REGISTER_MATCHER(newExpr);
   REGISTER_MATCHER(nullPtrLiteralExpr);
   REGISTER_MATCHER(nullStmt);
+  REGISTER_MATCHER(numSelectorArgs);
   REGISTER_MATCHER(ofClass);
+  REGISTER_MATCHER(objcMessageExpr);
   REGISTER_MATCHER(on);
   REGISTER_MATCHER(onImplicitObjectArgument);
   REGISTER_MATCHER(operatorCallExpr);
@@ -315,6 +324,7 @@ RegistryMaps::RegistryMaps() {
   REGISTER_MATCHER(throughUsingDecl);
   REGISTER_MATCHER(throwExpr);
   REGISTER_MATCHER(to);
+  REGISTER_MATCHER(translationUnitDecl);
   REGISTER_MATCHER(tryStmt);
   REGISTER_MATCHER(type);
   REGISTER_MATCHER(typedefDecl);
diff --git a/lib/Analysis/AnalysisDeclContext.cpp b/lib/Analysis/AnalysisDeclContext.cpp
index be66f32e77be..4e623c8d6c39 100644
--- a/lib/Analysis/AnalysisDeclContext.cpp
+++ b/lib/Analysis/AnalysisDeclContext.cpp
@@ -530,14 +530,14 @@ static DeclVec* LazyInitializeReferencedDecls(const BlockDecl *BD,
   return BV;
 }
 
-std::pair<AnalysisDeclContext::referenced_decls_iterator,
-          AnalysisDeclContext::referenced_decls_iterator>
+llvm::iterator_range<AnalysisDeclContext::referenced_decls_iterator>
 AnalysisDeclContext::getReferencedBlockVars(const BlockDecl *BD) {
   if (!ReferencedBlockVars)
     ReferencedBlockVars = new llvm::DenseMap<const BlockDecl*,void*>();
 
-  DeclVec *V = LazyInitializeReferencedDecls(BD, (*ReferencedBlockVars)[BD], A);
-  return std::make_pair(V->begin(), V->end());
+  const DeclVec *V =
+      LazyInitializeReferencedDecls(BD, (*ReferencedBlockVars)[BD], A);
+  return llvm::make_range(V->begin(), V->end());
 }
 
 ManagedAnalysis *&AnalysisDeclContext::getAnalysisImpl(const void *tag) {
diff --git a/lib/Analysis/CFG.cpp b/lib/Analysis/CFG.cpp
index d9073aa63b16..2744c5fbe72d 100644
--- a/lib/Analysis/CFG.cpp
+++ b/lib/Analysis/CFG.cpp
@@ -156,7 +156,7 @@ public:
       return !(*this == rhs);
     }
 
-    LLVM_EXPLICIT operator bool() const {
+    explicit operator bool() const {
       return *this != const_iterator();
     }
 
@@ -203,9 +203,9 @@ int LocalScope::const_iterator::distance(LocalScope::const_iterator L) {
   return D;
 }
 
-/// BlockScopePosPair - Structure for specifying position in CFG during its
-/// build process. It consists of CFGBlock that specifies position in CFG graph
-/// and  LocalScope::const_iterator that specifies position in LocalScope graph.
+/// Structure for specifying position in CFG during its build process. It
+/// consists of CFGBlock that specifies position in CFG and
+/// LocalScope::const_iterator that specifies position in LocalScope graph.
 struct BlockScopePosPair {
   BlockScopePosPair() : block(nullptr) {}
   BlockScopePosPair(CFGBlock *b, LocalScope::const_iterator scopePos)
@@ -841,12 +841,12 @@ private:
             // must be false.
             llvm::APSInt IntVal;
             if (Bop->getLHS()->EvaluateAsInt(IntVal, *Context)) {
-              if (IntVal.getBoolValue() == false) {
+              if (!IntVal.getBoolValue()) {
                 return TryResult(false);
               }
             }
             if (Bop->getRHS()->EvaluateAsInt(IntVal, *Context)) {
-              if (IntVal.getBoolValue() == false) {
+              if (!IntVal.getBoolValue()) {
                 return TryResult(false);
               }
             }
@@ -3950,9 +3950,8 @@ public:
       }
     }
   }
-  
 
-  virtual ~StmtPrinterHelper() {}
+  ~StmtPrinterHelper() override {}
 
   const LangOptions &getLangOpts() const { return LangOpts; }
   void setBlockID(signed i) { currentBlock = i; }
diff --git a/lib/Analysis/Consumed.cpp b/lib/Analysis/Consumed.cpp
index 2b2da2c69a41..fa985ee02e59 100644
--- a/lib/Analysis/Consumed.cpp
+++ b/lib/Analysis/Consumed.cpp
@@ -946,10 +946,9 @@ void ConsumedStmtVisitor::VisitVarDecl(const VarDecl *Var) {
 namespace clang {
 namespace consumed {
 
-void splitVarStateForIf(const IfStmt * IfNode, const VarTestResult &Test,
-                        ConsumedStateMap *ThenStates,
-                        ConsumedStateMap *ElseStates) {
-
+static void splitVarStateForIf(const IfStmt *IfNode, const VarTestResult &Test,
+                               ConsumedStateMap *ThenStates,
+                               ConsumedStateMap *ElseStates) {
   ConsumedState VarState = ThenStates->getState(Test.Var);
   
   if (VarState == CS_Unknown) {
@@ -964,9 +963,9 @@ void splitVarStateForIf(const IfStmt * IfNode, const VarTestResult &Test,
   }
 }
 
-void splitVarStateForIfBinOp(const PropagationInfo &PInfo,
-  ConsumedStateMap *ThenStates, ConsumedStateMap *ElseStates) {
-  
+static void splitVarStateForIfBinOp(const PropagationInfo &PInfo,
+                                    ConsumedStateMap *ThenStates,
+                                    ConsumedStateMap *ElseStates) {
   const VarTestResult &LTest = PInfo.getLTest(),
                       &RTest = PInfo.getRTest();
   
@@ -1443,7 +1442,7 @@ void ConsumedAnalyzer::run(AnalysisDeclContext &AC) {
                                                            CurrStates,
                                                            WarningsHandler);
             
-            if (BlockInfo.allBackEdgesVisited(*SI, CurrBlock))
+            if (BlockInfo.allBackEdgesVisited(CurrBlock, *SI))
               BlockInfo.discardInfo(*SI);
           } else {
             BlockInfo.addInfo(*SI, CurrStates, OwnershipTaken);
diff --git a/lib/Analysis/FormatString.cpp b/lib/Analysis/FormatString.cpp
index 8c663d856f6a..0948bc0b08a8 100644
--- a/lib/Analysis/FormatString.cpp
+++ b/lib/Analysis/FormatString.cpp
@@ -256,16 +256,17 @@ clang::analyze_format_string::ParseLengthModifier(FormatSpecifier &FS,
 // Methods on ArgType.
 //===----------------------------------------------------------------------===//
 
-bool ArgType::matchesType(ASTContext &C, QualType argTy) const {
+clang::analyze_format_string::ArgType::MatchKind
+ArgType::matchesType(ASTContext &C, QualType argTy) const {
   if (Ptr) {
     // It has to be a pointer.
     const PointerType *PT = argTy->getAs<PointerType>();
     if (!PT)
-      return false;
+      return NoMatch;
 
     // We cannot write through a const qualified pointer.
     if (PT->getPointeeType().isConstQualified())
-      return false;
+      return NoMatch;
 
     argTy = PT->getPointeeType();
   }
@@ -275,8 +276,8 @@ bool ArgType::matchesType(ASTContext &C, QualType argTy) const {
       llvm_unreachable("ArgType must be valid");
 
     case UnknownTy:
-      return true;
-      
+      return Match;
+
     case AnyCharTy: {
       if (const EnumType *ETy = argTy->getAs<EnumType>())
         argTy = ETy->getDecl()->getIntegerType();
@@ -289,18 +290,18 @@ bool ArgType::matchesType(ASTContext &C, QualType argTy) const {
           case BuiltinType::SChar:
           case BuiltinType::UChar:
           case BuiltinType::Char_U:
-            return true;            
+            return Match;
         }
-      return false;
+      return NoMatch;
     }
-      
+
     case SpecificTy: {
       if (const EnumType *ETy = argTy->getAs<EnumType>())
         argTy = ETy->getDecl()->getIntegerType();
       argTy = C.getCanonicalType(argTy).getUnqualifiedType();
 
       if (T == argTy)
-        return true;
+        return Match;
       // Check for "compatible types".
       if (const BuiltinType *BT = argTy->getAs<BuiltinType>())
         switch (BT->getKind()) {
@@ -309,32 +310,33 @@ bool ArgType::matchesType(ASTContext &C, QualType argTy) const {
           case BuiltinType::Char_S:
           case BuiltinType::SChar:
           case BuiltinType::Char_U:
-          case BuiltinType::UChar:                    
-            return T == C.UnsignedCharTy || T == C.SignedCharTy;
+          case BuiltinType::UChar:
+            return T == C.UnsignedCharTy || T == C.SignedCharTy ? Match
+                                                                : NoMatch;
           case BuiltinType::Short:
-            return T == C.UnsignedShortTy;
+            return T == C.UnsignedShortTy ? Match : NoMatch;
           case BuiltinType::UShort:
-            return T == C.ShortTy;
+            return T == C.ShortTy ? Match : NoMatch;
           case BuiltinType::Int:
-            return T == C.UnsignedIntTy;
+            return T == C.UnsignedIntTy ? Match : NoMatch;
           case BuiltinType::UInt:
-            return T == C.IntTy;
+            return T == C.IntTy ? Match : NoMatch;
           case BuiltinType::Long:
-            return T == C.UnsignedLongTy;
+            return T == C.UnsignedLongTy ? Match : NoMatch;
           case BuiltinType::ULong:
-            return T == C.LongTy;
+            return T == C.LongTy ? Match : NoMatch;
           case BuiltinType::LongLong:
-            return T == C.UnsignedLongLongTy;
+            return T == C.UnsignedLongLongTy ? Match : NoMatch;
           case BuiltinType::ULongLong:
-            return T == C.LongLongTy;
+            return T == C.LongLongTy ? Match : NoMatch;
         }
-      return false;
+      return NoMatch;
     }
 
     case CStrTy: {
       const PointerType *PT = argTy->getAs<PointerType>();
       if (!PT)
-        return false;
+        return NoMatch;
       QualType pointeeTy = PT->getPointeeType();
       if (const BuiltinType *BT = pointeeTy->getAs<BuiltinType>())
         switch (BT->getKind()) {
@@ -343,50 +345,56 @@ bool ArgType::matchesType(ASTContext &C, QualType argTy) const {
           case BuiltinType::UChar:
           case BuiltinType::Char_S:
           case BuiltinType::SChar:
-            return true;
+            return Match;
           default:
             break;
         }
 
-      return false;
+      return NoMatch;
     }
 
     case WCStrTy: {
       const PointerType *PT = argTy->getAs<PointerType>();
       if (!PT)
-        return false;
+        return NoMatch;
       QualType pointeeTy =
         C.getCanonicalType(PT->getPointeeType()).getUnqualifiedType();
-      return pointeeTy == C.getWideCharType();
+      return pointeeTy == C.getWideCharType() ? Match : NoMatch;
     }
-    
+
     case WIntTy: {
-      
+
       QualType PromoArg = 
         argTy->isPromotableIntegerType()
           ? C.getPromotedIntegerType(argTy) : argTy;
-      
+
       QualType WInt = C.getCanonicalType(C.getWIntType()).getUnqualifiedType();
       PromoArg = C.getCanonicalType(PromoArg).getUnqualifiedType();
-      
+
       // If the promoted argument is the corresponding signed type of the
       // wint_t type, then it should match.
       if (PromoArg->hasSignedIntegerRepresentation() &&
           C.getCorrespondingUnsignedType(PromoArg) == WInt)
-        return true;
+        return Match;
 
-      return WInt == PromoArg;
+      return WInt == PromoArg ? Match : NoMatch;
     }
 
     case CPointerTy:
-      return argTy->isPointerType() || argTy->isObjCObjectPointerType() ||
-             argTy->isBlockPointerType() || argTy->isNullPtrType();
+      if (argTy->isVoidPointerType()) {
+        return Match;
+      } if (argTy->isPointerType() || argTy->isObjCObjectPointerType() ||
+            argTy->isBlockPointerType() || argTy->isNullPtrType()) {
+        return NoMatchPedantic;
+      } else {
+        return NoMatch;
+      }
 
     case ObjCPointerTy: {
       if (argTy->getAs<ObjCObjectPointerType>() ||
           argTy->getAs<BlockPointerType>())
-        return true;
-      
+        return Match;
+
       // Handle implicit toll-free bridging.
       if (const PointerType *PT = argTy->getAs<PointerType>()) {
         // Things such as CFTypeRef are really just opaque pointers
@@ -395,9 +403,9 @@ bool ArgType::matchesType(ASTContext &C, QualType argTy) const {
         // structs can be toll-free bridged, we just accept them all.
         QualType pointee = PT->getPointeeType();
         if (pointee->getAsStructureType() || pointee->isVoidType())
-          return true;
+          return Match;
       }
-      return false;      
+      return NoMatch;
     }
   }
 
@@ -552,6 +560,12 @@ const char *ConversionSpecifier::toString() const {
   // Objective-C specific specifiers.
   case ObjCObjArg: return "@";
 
+  // FreeBSD kernel specific specifiers.
+  case FreeBSDbArg: return "b";
+  case FreeBSDDArg: return "D";
+  case FreeBSDrArg: return "r";
+  case FreeBSDyArg: return "y";
+
   // GlibC specific specifiers.
   case PrintErrno: return "m";
 
@@ -647,6 +661,9 @@ bool FormatSpecifier::hasValidLengthModifier(const TargetInfo &Target) const {
         case ConversionSpecifier::XArg:
         case ConversionSpecifier::nArg:
           return true;
+        case ConversionSpecifier::FreeBSDrArg:
+        case ConversionSpecifier::FreeBSDyArg:
+          return Target.getTriple().isOSFreeBSD();
         default:
           return false;
       }
@@ -677,6 +694,9 @@ bool FormatSpecifier::hasValidLengthModifier(const TargetInfo &Target) const {
         case ConversionSpecifier::ScanListArg:
         case ConversionSpecifier::ZArg:
           return true;
+        case ConversionSpecifier::FreeBSDrArg:
+        case ConversionSpecifier::FreeBSDyArg:
+          return Target.getTriple().isOSFreeBSD();
         default:
           return false;
       }
@@ -779,7 +799,8 @@ bool FormatSpecifier::hasStandardLengthModifier() const {
   llvm_unreachable("Invalid LengthModifier Kind!");
 }
 
-bool FormatSpecifier::hasStandardConversionSpecifier(const LangOptions &LangOpt) const {
+bool FormatSpecifier::hasStandardConversionSpecifier(
+    const LangOptions &LangOpt) const {
   switch (CS.getKind()) {
     case ConversionSpecifier::cArg:
     case ConversionSpecifier::dArg:
@@ -807,6 +828,10 @@ bool FormatSpecifier::hasStandardConversionSpecifier(const LangOptions &LangOpt)
     case ConversionSpecifier::SArg:
       return LangOpt.ObjC1 || LangOpt.ObjC2;
     case ConversionSpecifier::InvalidSpecifier:
+    case ConversionSpecifier::FreeBSDbArg:
+    case ConversionSpecifier::FreeBSDDArg:
+    case ConversionSpecifier::FreeBSDrArg:
+    case ConversionSpecifier::FreeBSDyArg:
     case ConversionSpecifier::PrintErrno:
     case ConversionSpecifier::DArg:
     case ConversionSpecifier::OArg:
diff --git a/lib/Analysis/LiveVariables.cpp b/lib/Analysis/LiveVariables.cpp
index 86b679cb155b..0ab158036d84 100644
--- a/lib/Analysis/LiveVariables.cpp
+++ b/lib/Analysis/LiveVariables.cpp
@@ -356,11 +356,8 @@ void TransferFunctions::VisitBinaryOperator(BinaryOperator *B) {
 }
 
 void TransferFunctions::VisitBlockExpr(BlockExpr *BE) {
-  AnalysisDeclContext::referenced_decls_iterator I, E;
-  std::tie(I, E) =
-    LV.analysisContext.getReferencedBlockVars(BE->getBlockDecl());
-  for ( ; I != E ; ++I) {
-    const VarDecl *VD = *I;
+  for (const VarDecl *VD :
+       LV.analysisContext.getReferencedBlockVars(BE->getBlockDecl())) {
     if (isAlwaysAlive(VD))
       continue;
     val.liveDecls = LV.DSetFact.add(val.liveDecls, VD);
diff --git a/lib/Analysis/PrintfFormatString.cpp b/lib/Analysis/PrintfFormatString.cpp
index 146635b88702..b8d3ec18016a 100644
--- a/lib/Analysis/PrintfFormatString.cpp
+++ b/lib/Analysis/PrintfFormatString.cpp
@@ -55,7 +55,8 @@ static PrintfSpecifierResult ParsePrintfSpecifier(FormatStringHandler &H,
                                                   unsigned &argIndex,
                                                   const LangOptions &LO,
                                                   const TargetInfo &Target,
-                                                  bool Warn) {
+                                                  bool Warn,
+                                                  bool isFreeBSDKPrintf) {
 
   using namespace clang::analyze_format_string;
   using namespace clang::analyze_printf;
@@ -206,9 +207,24 @@ static PrintfSpecifierResult ParsePrintfSpecifier(FormatStringHandler &H,
     case '@': k = ConversionSpecifier::ObjCObjArg; break;
     // Glibc specific.
     case 'm': k = ConversionSpecifier::PrintErrno; break;
+    // FreeBSD kernel specific.
+    case 'b':
+      if (isFreeBSDKPrintf)
+        k = ConversionSpecifier::FreeBSDbArg; // int followed by char *
+      break;
+    case 'r':
+      if (isFreeBSDKPrintf)
+        k = ConversionSpecifier::FreeBSDrArg; // int
+      break;
+    case 'y':
+      if (isFreeBSDKPrintf)
+        k = ConversionSpecifier::FreeBSDyArg; // int
+      break;
     // Apple-specific.
     case 'D':
-      if (Target.getTriple().isOSDarwin())
+      if (isFreeBSDKPrintf)
+        k = ConversionSpecifier::FreeBSDDArg; // void * followed by char *
+      else if (Target.getTriple().isOSDarwin())
         k = ConversionSpecifier::DArg;
       break;
     case 'O':
@@ -228,6 +244,10 @@ static PrintfSpecifierResult ParsePrintfSpecifier(FormatStringHandler &H,
   FS.setConversionSpecifier(CS);
   if (CS.consumesDataArgument() && !FS.usesPositionalArg())
     FS.setArgIndex(argIndex++);
+  // FreeBSD kernel specific.
+  if (k == ConversionSpecifier::FreeBSDbArg ||
+      k == ConversionSpecifier::FreeBSDDArg)
+    argIndex++;
 
   if (k == ConversionSpecifier::InvalidSpecifier) {
     // Assume the conversion takes one argument.
@@ -240,14 +260,16 @@ bool clang::analyze_format_string::ParsePrintfString(FormatStringHandler &H,
                                                      const char *I,
                                                      const char *E,
                                                      const LangOptions &LO,
-                                                     const TargetInfo &Target) {
+                                                     const TargetInfo &Target,
+                                                     bool isFreeBSDKPrintf) {
 
   unsigned argIndex = 0;
 
   // Keep looking for a format specifier until we have exhausted the string.
   while (I != E) {
     const PrintfSpecifierResult &FSR = ParsePrintfSpecifier(H, I, E, argIndex,
-                                                            LO, Target, true);
+                                                            LO, Target, true,
+                                                            isFreeBSDKPrintf);
     // Did a fail-stop error of any kind occur when parsing the specifier?
     // If so, don't do any more processing.
     if (FSR.shouldStop())
@@ -276,7 +298,8 @@ bool clang::analyze_format_string::ParseFormatStringHasSArg(const char *I,
   FormatStringHandler H;
   while (I != E) {
     const PrintfSpecifierResult &FSR = ParsePrintfSpecifier(H, I, E, argIndex,
-                                                            LO, Target, false);
+                                                            LO, Target, false,
+                                                            false);
     // Did a fail-stop error of any kind occur when parsing the specifier?
     // If so, don't do any more processing.
     if (FSR.shouldStop())
@@ -674,6 +697,8 @@ bool PrintfSpecifier::hasValidPlusPrefix() const {
   case ConversionSpecifier::GArg:
   case ConversionSpecifier::aArg:
   case ConversionSpecifier::AArg:
+  case ConversionSpecifier::FreeBSDrArg:
+  case ConversionSpecifier::FreeBSDyArg:
     return true;
 
   default:
@@ -699,6 +724,8 @@ bool PrintfSpecifier::hasValidAlternativeForm() const {
   case ConversionSpecifier::FArg:
   case ConversionSpecifier::gArg:
   case ConversionSpecifier::GArg:
+  case ConversionSpecifier::FreeBSDrArg:
+  case ConversionSpecifier::FreeBSDyArg:
     return true;
 
   default:
@@ -729,6 +756,8 @@ bool PrintfSpecifier::hasValidLeadingZeros() const {
   case ConversionSpecifier::FArg:
   case ConversionSpecifier::gArg:
   case ConversionSpecifier::GArg:
+  case ConversionSpecifier::FreeBSDrArg:
+  case ConversionSpecifier::FreeBSDyArg:
     return true;
 
   default:
@@ -753,6 +782,8 @@ bool PrintfSpecifier::hasValidSpacePrefix() const {
   case ConversionSpecifier::GArg:
   case ConversionSpecifier::aArg:
   case ConversionSpecifier::AArg:
+  case ConversionSpecifier::FreeBSDrArg:
+  case ConversionSpecifier::FreeBSDyArg:
     return true;
 
   default:
@@ -818,6 +849,8 @@ bool PrintfSpecifier::hasValidPrecision() const {
   case ConversionSpecifier::gArg:
   case ConversionSpecifier::GArg:
   case ConversionSpecifier::sArg:
+  case ConversionSpecifier::FreeBSDrArg:
+  case ConversionSpecifier::FreeBSDyArg:
     return true;
 
   default:
diff --git a/lib/Analysis/ThreadSafety.cpp b/lib/Analysis/ThreadSafety.cpp
index f45d6e7ef139..e2c6ab5d9485 100644
--- a/lib/Analysis/ThreadSafety.cpp
+++ b/lib/Analysis/ThreadSafety.cpp
@@ -44,14 +44,13 @@
 #include <sstream>
 #include <utility>
 #include <vector>
-
-
-namespace clang {
-namespace threadSafety {
+using namespace clang;
+using namespace threadSafety;
 
 // Key method definition
 ThreadSafetyHandler::~ThreadSafetyHandler() {}
 
+namespace {
 class TILPrinter :
   public til::PrettyPrinter<TILPrinter, llvm::raw_ostream> {};
 
@@ -69,7 +68,6 @@ static void warnInvalidLock(ThreadSafetyHandler &Handler,
     Handler.handleInvalidLockExp(Kind, Loc);
 }
 
-
 /// \brief A set of CapabilityInfo objects, which are compiled from the
 /// requires attributes on a function.
 class CapExprSet : public SmallVector<CapabilityExpr, 4> {
@@ -101,17 +99,22 @@ private:
   LockKind          LKind;            ///<  exclusive or shared
   SourceLocation    AcquireLoc;       ///<  where it was acquired.
   bool              Asserted;         ///<  true if the lock was asserted
+  bool              Declared;         ///<  true if the lock was declared
 
 public:
   FactEntry(const CapabilityExpr &CE, LockKind LK, SourceLocation Loc,
-            bool Asrt)
-      : CapabilityExpr(CE), LKind(LK), AcquireLoc(Loc), Asserted(Asrt) {}
+            bool Asrt, bool Declrd = false)
+      : CapabilityExpr(CE), LKind(LK), AcquireLoc(Loc), Asserted(Asrt),
+        Declared(Declrd) {}
 
   virtual ~FactEntry() {}
 
-  LockKind          kind()       const { return LKind;    }
+  LockKind          kind()       const { return LKind;      }
   SourceLocation    loc()        const { return AcquireLoc; }
   bool              asserted()   const { return Asserted; }
+  bool              declared()   const { return Declared; }
+
+  void setDeclared(bool D) { Declared = D; }
 
   virtual void
   handleRemovalFromIntersection(const FactSet &FSet, FactManager &FactMan,
@@ -231,14 +234,61 @@ public:
 
   FactEntry *findPartialMatch(FactManager &FM,
                               const CapabilityExpr &CapE) const {
-    auto I = std::find_if(begin(), end(), [&](FactID ID) {
+    auto I = std::find_if(begin(), end(), [&](FactID ID) -> bool {
       return FM[ID].partiallyMatches(CapE);
     });
     return I != end() ? &FM[*I] : nullptr;
   }
+
+  bool containsMutexDecl(FactManager &FM, const ValueDecl* Vd) const {
+    auto I = std::find_if(begin(), end(), [&](FactID ID) -> bool {
+      return FM[ID].valueDecl() == Vd;
+    });
+    return I != end();
+  }
 };
 
+class ThreadSafetyAnalyzer;
+} // namespace
+
+namespace clang {
+namespace threadSafety {
+class BeforeSet {
+private:
+  typedef SmallVector<const ValueDecl*, 4>  BeforeVect;
+
+  struct BeforeInfo {
+    BeforeInfo() : Vect(nullptr), Visited(false) { }
+    BeforeInfo(BeforeInfo &&O)
+        : Vect(std::move(O.Vect)), Visited(O.Visited)
+    {}
+
+    std::unique_ptr<BeforeVect> Vect;
+    int                         Visited;
+  };
+
+  typedef llvm::DenseMap<const ValueDecl*, BeforeInfo>  BeforeMap;
+  typedef llvm::DenseMap<const ValueDecl*, bool>        CycleMap;
+
+public:
+  BeforeSet() { }
+
+  BeforeInfo* insertAttrExprs(const ValueDecl* Vd,
+                              ThreadSafetyAnalyzer& Analyzer);
+
+  void checkBeforeAfter(const ValueDecl* Vd,
+                        const FactSet& FSet,
+                        ThreadSafetyAnalyzer& Analyzer,
+                        SourceLocation Loc, StringRef CapKind);
+
+private:
+  BeforeMap BMap;
+  CycleMap  CycMap;
+};
+} // end namespace threadSafety
+} // end namespace clang
 
+namespace {
 typedef llvm::ImmutableMap<const NamedDecl*, unsigned> LocalVarContext;
 class LocalVariableMap;
 
@@ -853,6 +903,7 @@ public:
 /// \brief Class which implements the core thread safety analysis routines.
 class ThreadSafetyAnalyzer {
   friend class BuildLockset;
+  friend class threadSafety::BeforeSet;
 
   llvm::BumpPtrAllocator Bpa;
   threadSafety::til::MemRegionRef Arena;
@@ -864,9 +915,11 @@ class ThreadSafetyAnalyzer {
   FactManager               FactMan;
   std::vector<CFGBlockInfo> BlockInfo;
 
+  BeforeSet* GlobalBeforeSet;
+
 public:
-  ThreadSafetyAnalyzer(ThreadSafetyHandler &H)
-     : Arena(&Bpa), SxBuilder(Arena), Handler(H) {}
+  ThreadSafetyAnalyzer(ThreadSafetyHandler &H, BeforeSet* Bset)
+     : Arena(&Bpa), SxBuilder(Arena), Handler(H), GlobalBeforeSet(Bset) {}
 
   bool inCurrentScope(const CapabilityExpr &CapE);
 
@@ -906,6 +959,134 @@ public:
 
   void runAnalysis(AnalysisDeclContext &AC);
 };
+} // namespace
+
+/// Process acquired_before and acquired_after attributes on Vd.
+BeforeSet::BeforeInfo* BeforeSet::insertAttrExprs(const ValueDecl* Vd,
+    ThreadSafetyAnalyzer& Analyzer) {
+  // Create a new entry for Vd.
+  auto& Entry = BMap.FindAndConstruct(Vd);
+  BeforeInfo* Info = &Entry.second;
+  BeforeVect* Bv = nullptr;
+
+  for (Attr* At : Vd->attrs()) {
+    switch (At->getKind()) {
+      case attr::AcquiredBefore: {
+        auto *A = cast<AcquiredBeforeAttr>(At);
+
+        // Create a new BeforeVect for Vd if necessary.
+        if (!Bv) {
+          Bv = new BeforeVect;
+          Info->Vect.reset(Bv);
+        }
+        // Read exprs from the attribute, and add them to BeforeVect.
+        for (const auto *Arg : A->args()) {
+          CapabilityExpr Cp =
+            Analyzer.SxBuilder.translateAttrExpr(Arg, nullptr);
+          if (const ValueDecl *Cpvd = Cp.valueDecl()) {
+            Bv->push_back(Cpvd);
+            auto It = BMap.find(Cpvd);
+            if (It == BMap.end())
+              insertAttrExprs(Cpvd, Analyzer);
+          }
+        }
+        break;
+      }
+      case attr::AcquiredAfter: {
+        auto *A = cast<AcquiredAfterAttr>(At);
+
+        // Read exprs from the attribute, and add them to BeforeVect.
+        for (const auto *Arg : A->args()) {
+          CapabilityExpr Cp =
+            Analyzer.SxBuilder.translateAttrExpr(Arg, nullptr);
+          if (const ValueDecl *ArgVd = Cp.valueDecl()) {
+            // Get entry for mutex listed in attribute
+            BeforeInfo* ArgInfo;
+            auto It = BMap.find(ArgVd);
+            if (It == BMap.end())
+              ArgInfo = insertAttrExprs(ArgVd, Analyzer);
+            else
+              ArgInfo = &It->second;
+
+            // Create a new BeforeVect if necessary.
+            BeforeVect* ArgBv = ArgInfo->Vect.get();
+            if (!ArgBv) {
+              ArgBv = new BeforeVect;
+              ArgInfo->Vect.reset(ArgBv);
+            }
+            ArgBv->push_back(Vd);
+          }
+        }
+        break;
+      }
+      default:
+        break;
+    }
+  }
+
+  return Info;
+}
+
+
+/// Return true if any mutexes in FSet are in the acquired_before set of Vd.
+void BeforeSet::checkBeforeAfter(const ValueDecl* StartVd,
+                                 const FactSet& FSet,
+                                 ThreadSafetyAnalyzer& Analyzer,
+                                 SourceLocation Loc, StringRef CapKind) {
+  SmallVector<BeforeInfo*, 8> InfoVect;
+
+  // Do a depth-first traversal of Vd.
+  // Return true if there are cycles.
+  std::function<bool (const ValueDecl*)> traverse = [&](const ValueDecl* Vd) {
+    if (!Vd)
+      return false;
+
+    BeforeSet::BeforeInfo* Info;
+    auto It = BMap.find(Vd);
+    if (It == BMap.end())
+      Info = insertAttrExprs(Vd, Analyzer);
+    else
+      Info = &It->second;
+
+    if (Info->Visited == 1)
+      return true;
+
+    if (Info->Visited == 2)
+      return false;
+
+    BeforeVect* Bv = Info->Vect.get();
+    if (!Bv)
+      return false;
+
+    InfoVect.push_back(Info);
+    Info->Visited = 1;
+    for (auto *Vdb : *Bv) {
+      // Exclude mutexes in our immediate before set.
+      if (FSet.containsMutexDecl(Analyzer.FactMan, Vdb)) {
+        StringRef L1 = StartVd->getName();
+        StringRef L2 = Vdb->getName();
+        Analyzer.Handler.handleLockAcquiredBefore(CapKind, L1, L2, Loc);
+      }
+      // Transitively search other before sets, and warn on cycles.
+      if (traverse(Vdb)) {
+        if (CycMap.find(Vd) == CycMap.end()) {
+          CycMap.insert(std::make_pair(Vd, true));
+          StringRef L1 = Vd->getName();
+          Analyzer.Handler.handleBeforeAfterCycle(L1, Vd->getLocation());
+        }
+      }
+    }
+    Info->Visited = 2;
+    return false;
+  };
+
+  traverse(StartVd);
+
+  for (auto* Info : InfoVect)
+    Info->Visited = 0;
+}
+
+
 
 /// \brief Gets the value decl pointer from DeclRefExprs or MemberExprs.
 static const ValueDecl *getValueDecl(const Expr *Exp) {
@@ -921,6 +1102,7 @@ static const ValueDecl *getValueDecl(const Expr *Exp) {
   return nullptr;
 }
 
+namespace {
 template <typename Ty>
 class has_arg_iterator_range {
   typedef char yes[1];
@@ -935,6 +1117,7 @@ class has_arg_iterator_range {
 public:
   static const bool value = sizeof(test<Ty>(nullptr)) == sizeof(yes);
 };
+} // namespace
 
 static StringRef ClassifyDiagnostic(const CapabilityAttr *A) {
   return A->getName();
@@ -1020,7 +1203,13 @@ void ThreadSafetyAnalyzer::addLock(FactSet &FSet,
     }
   }
 
-  // FIXME: deal with acquired before/after annotations.
+  // Check before/after constraints
+  if (Handler.issueBetaWarnings() &&
+      !Entry->asserted() && !Entry->declared()) {
+    GlobalBeforeSet->checkBeforeAfter(Entry->valueDecl(), FSet, *this,
+                                      Entry->loc(), DiagKind);
+  }
+
   // FIXME: Don't always warn when we have support for reentrant locks.
   if (FSet.findLock(FactMan, *Entry)) {
     if (!Entry->asserted())
@@ -1119,8 +1308,7 @@ void ThreadSafetyAnalyzer::getMutexIDs(CapExprSet &Mtxs, AttrType *Attr,
   }
 }
 
-
-bool getStaticBooleanValue(Expr* E, bool& TCond) {
+static bool getStaticBooleanValue(Expr *E, bool &TCond) {
   if (isa<CXXNullPtrLiteralExpr>(E) || isa<GNUNullExpr>(E)) {
     TCond = false;
     return true;
@@ -1230,7 +1418,7 @@ void ThreadSafetyAnalyzer::getEdgeLockset(FactSet& Result,
   CapExprSet SharedLocksToAdd;
 
   // If the condition is a call to a Trylock function, then grab the attributes
-  for (auto *Attr : FunDecl->getAttrs()) {
+  for (auto *Attr : FunDecl->attrs()) {
     switch (Attr->getKind()) {
       case attr::ExclusiveTrylockFunction: {
         ExclusiveTrylockFunctionAttr *A =
@@ -1265,6 +1453,7 @@ void ThreadSafetyAnalyzer::getEdgeLockset(FactSet& Result,
             CapDiagKind);
 }
 
+namespace {
 /// \brief We use this class to visit different types of expressions in
 /// CFGBlocks, and build up the lockset.
 /// An expression may cause us to add or remove locks from the lockset, or else
@@ -1308,7 +1497,7 @@ public:
   void VisitCXXConstructExpr(CXXConstructExpr *Exp);
   void VisitDeclStmt(DeclStmt *S);
 };
-
+} // namespace
 
 /// \brief Warn if the LSet does not contain a lock sufficient to protect access
 /// of at least the passed in AccessKind.
@@ -1500,13 +1689,23 @@ void BuildLockset::checkPtAccess(const Expr *Exp, AccessKind AK,
 ///
 void BuildLockset::handleCall(Expr *Exp, const NamedDecl *D, VarDecl *VD) {
   SourceLocation Loc = Exp->getExprLoc();
-  const AttrVec &ArgAttrs = D->getAttrs();
   CapExprSet ExclusiveLocksToAdd, SharedLocksToAdd;
   CapExprSet ExclusiveLocksToRemove, SharedLocksToRemove, GenericLocksToRemove;
+  CapExprSet ScopedExclusiveReqs, ScopedSharedReqs;
   StringRef CapDiagKind = "mutex";
 
-  for(unsigned i = 0; i < ArgAttrs.size(); ++i) {
-    Attr *At = const_cast<Attr*>(ArgAttrs[i]);
+  // Figure out if we're calling the constructor of scoped lockable class
+  bool isScopedVar = false;
+  if (VD) {
+    if (const CXXConstructorDecl *CD = dyn_cast<const CXXConstructorDecl>(D)) {
+      const CXXRecordDecl* PD = CD->getParent();
+      if (PD && PD->hasAttr<ScopedLockableAttr>())
+        isScopedVar = true;
+    }
+  }
+
+  for(Attr *Atconst : D->attrs()) {
+    Attr* At = const_cast<Attr*>(Atconst);
     switch (At->getKind()) {
       // When we encounter a lock function, we need to add the lock to our
       // lockset.
@@ -1564,10 +1763,17 @@ void BuildLockset::handleCall(Expr *Exp, const NamedDecl *D, VarDecl *VD) {
 
       case attr::RequiresCapability: {
         RequiresCapabilityAttr *A = cast<RequiresCapabilityAttr>(At);
-        for (auto *Arg : A->args())
+        for (auto *Arg : A->args()) {
           warnIfMutexNotHeld(D, Exp, A->isShared() ? AK_Read : AK_Written, Arg,
                              POK_FunctionCall, ClassifyDiagnostic(A),
                              Exp->getExprLoc());
+          // use for adopting a lock
+          if (isScopedVar) {
+            Analyzer->getMutexIDs(A->isShared() ? ScopedSharedReqs
+                                                : ScopedExclusiveReqs,
+                                  A, Exp, D, VD);
+          }
+        }
         break;
       }
 
@@ -1584,16 +1790,6 @@ void BuildLockset::handleCall(Expr *Exp, const NamedDecl *D, VarDecl *VD) {
     }
   }
 
-  // Figure out if we're calling the constructor of scoped lockable class
-  bool isScopedVar = false;
-  if (VD) {
-    if (const CXXConstructorDecl *CD = dyn_cast<const CXXConstructorDecl>(D)) {
-      const CXXRecordDecl* PD = CD->getParent();
-      if (PD && PD->hasAttr<ScopedLockableAttr>())
-        isScopedVar = true;
-    }
-  }
-
   // Add locks.
   for (const auto &M : ExclusiveLocksToAdd)
     Analyzer->addLock(FSet, llvm::make_unique<LockableFactEntry>(
@@ -1611,9 +1807,10 @@ void BuildLockset::handleCall(Expr *Exp, const NamedDecl *D, VarDecl *VD) {
     // FIXME: does this store a pointer to DRE?
     CapabilityExpr Scp = Analyzer->SxBuilder.translateAttrExpr(&DRE, nullptr);
 
-    CapExprSet UnderlyingMutexes(ExclusiveLocksToAdd);
-    std::copy(SharedLocksToAdd.begin(), SharedLocksToAdd.end(),
-              std::back_inserter(UnderlyingMutexes));
+    std::copy(ScopedExclusiveReqs.begin(), ScopedExclusiveReqs.end(),
+              std::back_inserter(ExclusiveLocksToAdd));
+    std::copy(ScopedSharedReqs.begin(), ScopedSharedReqs.end(),
+              std::back_inserter(SharedLocksToAdd));
     Analyzer->addLock(FSet,
                       llvm::make_unique<ScopedLockableFactEntry>(
                           Scp, MLoc, ExclusiveLocksToAdd, SharedLocksToAdd),
@@ -1863,7 +2060,7 @@ void ThreadSafetyAnalyzer::intersectAndWarn(FactSet &FSet1,
 
 
 // Return true if block B never continues to its successors.
-inline bool neverReturns(const CFGBlock* B) {
+static bool neverReturns(const CFGBlock *B) {
   if (B->hasNoReturnElement())
     return true;
   if (B->empty())
@@ -1940,14 +2137,13 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) {
   if (!SortedGraph->empty() && D->hasAttrs()) {
     const CFGBlock *FirstBlock = *SortedGraph->begin();
     FactSet &InitialLockset = BlockInfo[FirstBlock->getBlockID()].EntrySet;
-    const AttrVec &ArgAttrs = D->getAttrs();
 
     CapExprSet ExclusiveLocksToAdd;
     CapExprSet SharedLocksToAdd;
     StringRef CapDiagKind = "mutex";
 
     SourceLocation Loc = D->getLocation();
-    for (const auto *Attr : ArgAttrs) {
+    for (const auto *Attr : D->attrs()) {
       Loc = Attr->getLocation();
       if (const auto *A = dyn_cast<RequiresCapabilityAttr>(Attr)) {
         getMutexIDs(A->isShared() ? SharedLocksToAdd : ExclusiveLocksToAdd, A,
@@ -1979,14 +2175,16 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) {
     }
 
     // FIXME -- Loc can be wrong here.
-    for (const auto &Mu : ExclusiveLocksToAdd)
-      addLock(InitialLockset,
-              llvm::make_unique<LockableFactEntry>(Mu, LK_Exclusive, Loc),
-              CapDiagKind, true);
-    for (const auto &Mu : SharedLocksToAdd)
-      addLock(InitialLockset,
-              llvm::make_unique<LockableFactEntry>(Mu, LK_Shared, Loc),
-              CapDiagKind, true);
+    for (const auto &Mu : ExclusiveLocksToAdd) {
+      auto Entry = llvm::make_unique<LockableFactEntry>(Mu, LK_Exclusive, Loc);
+      Entry->setDeclared(true);
+      addLock(InitialLockset, std::move(Entry), CapDiagKind, true);
+    }
+    for (const auto &Mu : SharedLocksToAdd) {
+      auto Entry = llvm::make_unique<LockableFactEntry>(Mu, LK_Shared, Loc);
+      Entry->setDeclared(true);
+      addLock(InitialLockset, std::move(Entry), CapDiagKind, true);
+    }
   }
 
   for (const auto *CurrBlock : *SortedGraph) {
@@ -2179,15 +2377,20 @@ void ThreadSafetyAnalyzer::runAnalysis(AnalysisDeclContext &AC) {
 /// We traverse the blocks in the CFG, compute the set of mutexes that are held
 /// at the end of each block, and issue warnings for thread safety violations.
 /// Each block in the CFG is traversed exactly once.
-void runThreadSafetyAnalysis(AnalysisDeclContext &AC,
-                             ThreadSafetyHandler &Handler) {
-  ThreadSafetyAnalyzer Analyzer(Handler);
+void threadSafety::runThreadSafetyAnalysis(AnalysisDeclContext &AC,
+                                           ThreadSafetyHandler &Handler,
+                                           BeforeSet **BSet) {
+  if (!*BSet)
+    *BSet = new BeforeSet;
+  ThreadSafetyAnalyzer Analyzer(Handler, *BSet);
   Analyzer.runAnalysis(AC);
 }
 
+void threadSafety::threadSafetyCleanup(BeforeSet *Cache) { delete Cache; }
+
 /// \brief Helper function that returns a LockKind required for the given level
 /// of access.
-LockKind getLockKindFromAccessKind(AccessKind AK) {
+LockKind threadSafety::getLockKindFromAccessKind(AccessKind AK) {
   switch (AK) {
     case AK_Read :
       return LK_Shared;
@@ -2196,5 +2399,3 @@ LockKind getLockKindFromAccessKind(AccessKind AK) {
   }
   llvm_unreachable("Unknown AccessKind");
 }
-
-}} // end namespace clang::threadSafety
diff --git a/lib/Analysis/ThreadSafetyCommon.cpp b/lib/Analysis/ThreadSafetyCommon.cpp
index 563e0590aacb..d4b1ce26d4b3 100644
--- a/lib/Analysis/ThreadSafetyCommon.cpp
+++ b/lib/Analysis/ThreadSafetyCommon.cpp
@@ -31,13 +31,11 @@
 #include <algorithm>
 #include <climits>
 #include <vector>
-
-
-namespace clang {
-namespace threadSafety {
+using namespace clang;
+using namespace threadSafety;
 
 // From ThreadSafetyUtil.h
-std::string getSourceLiteralString(const clang::Expr *CE) {
+std::string threadSafety::getSourceLiteralString(const clang::Expr *CE) {
   switch (CE->getStmtClass()) {
     case Stmt::IntegerLiteralClass:
       return cast<IntegerLiteral>(CE)->getValue().toString(10, true);
@@ -59,18 +57,13 @@ std::string getSourceLiteralString(const clang::Expr *CE) {
   }
 }
 
-namespace til {
-
 // Return true if E is a variable that points to an incomplete Phi node.
-static bool isIncompletePhi(const SExpr *E) {
-  if (const auto *Ph = dyn_cast<Phi>(E))
-    return Ph->status() == Phi::PH_Incomplete;
+static bool isIncompletePhi(const til::SExpr *E) {
+  if (const auto *Ph = dyn_cast<til::Phi>(E))
+    return Ph->status() == til::Phi::PH_Incomplete;
   return false;
 }
 
-}  // end namespace til
-
-
 typedef SExprBuilder::CallingContext CallingContext;
 
 
@@ -87,9 +80,7 @@ til::SCFG *SExprBuilder::buildCFG(CFGWalker &Walker) {
   return Scfg;
 }
 
-
-
-inline bool isCalleeArrow(const Expr *E) {
+static bool isCalleeArrow(const Expr *E) {
   const MemberExpr *ME = dyn_cast<MemberExpr>(E->IgnoreParenCasts());
   return ME ? ME->isArrow() : false;
 }
@@ -313,8 +304,7 @@ til::SExpr *SExprBuilder::translateCXXThisExpr(const CXXThisExpr *TE,
   return SelfVar;
 }
 
-
-const ValueDecl *getValueDeclFromSExpr(const til::SExpr *E) {
+static const ValueDecl *getValueDeclFromSExpr(const til::SExpr *E) {
   if (auto *V = dyn_cast<til::Variable>(E))
     return V->clangDecl();
   if (auto *Ph = dyn_cast<til::Phi>(E))
@@ -326,7 +316,7 @@ const ValueDecl *getValueDeclFromSExpr(const til::SExpr *E) {
   return 0;
 }
 
-bool hasCppPointerType(const til::SExpr *E) {
+static bool hasCppPointerType(const til::SExpr *E) {
   auto *VD = getValueDeclFromSExpr(E);
   if (VD && VD->getType()->isPointerType())
     return true;
@@ -336,9 +326,8 @@ bool hasCppPointerType(const til::SExpr *E) {
   return false;
 }
 
-
 // Grab the very first declaration of virtual method D
-const CXXMethodDecl* getFirstVirtualDecl(const CXXMethodDecl *D) {
+static const CXXMethodDecl *getFirstVirtualDecl(const CXXMethodDecl *D) {
   while (true) {
     D = D->getCanonicalDecl();
     CXXMethodDecl::method_iterator I = D->begin_overridden_methods(),
@@ -663,7 +652,7 @@ til::SExpr *SExprBuilder::lookupVarDecl(const ValueDecl *VD) {
 
 
 // if E is a til::Variable, update its clangDecl.
-inline void maybeUpdateVD(til::SExpr *E, const ValueDecl *VD) {
+static void maybeUpdateVD(til::SExpr *E, const ValueDecl *VD) {
   if (!E)
     return;
   if (til::Variable *V = dyn_cast<til::Variable>(E)) {
@@ -986,8 +975,3 @@ void printSCFG(CFGWalker &Walker) {
   TILPrinter::print(Scfg, llvm::errs());
 }
 */
-
-
-} // end namespace threadSafety
-
-} // end namespace clang
diff --git a/lib/Analysis/ThreadSafetyTIL.cpp b/lib/Analysis/ThreadSafetyTIL.cpp
index ebe374ea609d..2923f7e66a61 100644
--- a/lib/Analysis/ThreadSafetyTIL.cpp
+++ b/lib/Analysis/ThreadSafetyTIL.cpp
@@ -9,13 +9,11 @@
 
 #include "clang/Analysis/Analyses/ThreadSafetyTIL.h"
 #include "clang/Analysis/Analyses/ThreadSafetyTraverse.h"
+using namespace clang;
+using namespace threadSafety;
+using namespace til;
 
-namespace clang {
-namespace threadSafety {
-namespace til {
-
-
-StringRef getUnaryOpcodeString(TIL_UnaryOpcode Op) {
+StringRef til::getUnaryOpcodeString(TIL_UnaryOpcode Op) {
   switch (Op) {
     case UOP_Minus:    return "-";
     case UOP_BitNot:   return "~";
@@ -24,8 +22,7 @@ StringRef getUnaryOpcodeString(TIL_UnaryOpcode Op) {
   return "";
 }
 
-
-StringRef getBinaryOpcodeString(TIL_BinaryOpcode Op) {
+StringRef til::getBinaryOpcodeString(TIL_BinaryOpcode Op) {
   switch (Op) {
     case BOP_Mul:      return "*";
     case BOP_Div:      return "/";
@@ -82,7 +79,7 @@ void BasicBlock::reservePredecessors(unsigned NumPreds) {
 
 // If E is a variable, then trace back through any aliases or redundant
 // Phi nodes to find the canonical definition.
-const SExpr *getCanonicalVal(const SExpr *E) {
+const SExpr *til::getCanonicalVal(const SExpr *E) {
   while (true) {
     if (auto *V = dyn_cast<Variable>(E)) {
       if (V->kind() == Variable::VK_Let) {
@@ -105,7 +102,7 @@ const SExpr *getCanonicalVal(const SExpr *E) {
 // If E is a variable, then trace back through any aliases or redundant
 // Phi nodes to find the canonical definition.
 // The non-const version will simplify incomplete Phi nodes.
-SExpr *simplifyToCanonicalVal(SExpr *E) {
+SExpr *til::simplifyToCanonicalVal(SExpr *E) {
   while (true) {
     if (auto *V = dyn_cast<Variable>(E)) {
       if (V->kind() != Variable::VK_Let)
@@ -135,7 +132,7 @@ SExpr *simplifyToCanonicalVal(SExpr *E) {
 // Trace the arguments of an incomplete Phi node to see if they have the same
 // canonical definition.  If so, mark the Phi node as redundant.
 // getCanonicalVal() will recursively call simplifyIncompletePhi().
-void simplifyIncompleteArg(til::Phi *Ph) {
+void til::simplifyIncompleteArg(til::Phi *Ph) {
   assert(Ph && Ph->status() == Phi::PH_Incomplete);
 
   // eliminate infinite recursion -- assume that this node is not redundant.
@@ -337,7 +334,3 @@ void SCFG::computeNormalForm() {
     computeNodeID(Block, &BasicBlock::PostDominatorNode);
   }
 }
-
-}  // end namespace til
-}  // end namespace threadSafety
-}  // end namespace clang
diff --git a/lib/Analysis/UninitializedValues.cpp b/lib/Analysis/UninitializedValues.cpp
index 61a259217d7d..f2f791957aa3 100644
--- a/lib/Analysis/UninitializedValues.cpp
+++ b/lib/Analysis/UninitializedValues.cpp
@@ -14,6 +14,7 @@
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/Attr.h"
 #include "clang/AST/Decl.h"
+#include "clang/AST/DeclCXX.h"
 #include "clang/AST/StmtVisitor.h"
 #include "clang/Analysis/Analyses/PostOrderCFGView.h"
 #include "clang/Analysis/Analyses/UninitializedValues.h"
@@ -35,9 +36,9 @@ using namespace clang;
 static bool isTrackedVar(const VarDecl *vd, const DeclContext *dc) {
   if (vd->isLocalVarDecl() && !vd->hasGlobalStorage() &&
       !vd->isExceptionVariable() && !vd->isInitCapture() &&
-      vd->getDeclContext() == dc) {
+      !vd->isImplicit() && vd->getDeclContext() == dc) {
     QualType ty = vd->getType();
-    return ty->isScalarType() || ty->isVectorType();
+    return ty->isScalarType() || ty->isVectorType() || ty->isRecordType();
   }
   return false;
 }
@@ -347,6 +348,7 @@ public:
 }
 
 static const DeclRefExpr *getSelfInitExpr(VarDecl *VD) {
+  if (VD->getType()->isRecordType()) return nullptr;
   if (Expr *Init = VD->getInit()) {
     const DeclRefExpr *DRE
       = dyn_cast<DeclRefExpr>(stripCasts(VD->getASTContext(), Init));
@@ -376,10 +378,26 @@ void ClassifyRefs::classify(const Expr *E, Class C) {
     return;
   }
 
+  if (const MemberExpr *ME = dyn_cast<MemberExpr>(E)) {
+    if (VarDecl *VD = dyn_cast<VarDecl>(ME->getMemberDecl())) {
+      if (!VD->isStaticDataMember())
+        classify(ME->getBase(), C);
+    }
+    return;
+  }
+
   if (const BinaryOperator *BO = dyn_cast<BinaryOperator>(E)) {
-    if (BO->getOpcode() == BO_Comma)
+    switch (BO->getOpcode()) {
+    case BO_PtrMemD:
+    case BO_PtrMemI:
+      classify(BO->getLHS(), C);
+      return;
+    case BO_Comma:
       classify(BO->getRHS(), C);
-    return;
+      return;
+    default:
+      return;
+    }
   }
 
   FindVarResult Var = findVar(E, DC);
@@ -404,7 +422,7 @@ void ClassifyRefs::VisitBinaryOperator(BinaryOperator *BO) {
   // use.
   if (BO->isCompoundAssignmentOp())
     classify(BO->getLHS(), Use);
-  else if (BO->getOpcode() == BO_Assign)
+  else if (BO->getOpcode() == BO_Assign || BO->getOpcode() == BO_Comma)
     classify(BO->getLHS(), Ignore);
 }
 
@@ -415,25 +433,40 @@ void ClassifyRefs::VisitUnaryOperator(UnaryOperator *UO) {
     classify(UO->getSubExpr(), Use);
 }
 
+static bool isPointerToConst(const QualType &QT) {
+  return QT->isAnyPointerType() && QT->getPointeeType().isConstQualified();
+}
+
 void ClassifyRefs::VisitCallExpr(CallExpr *CE) {
   // Classify arguments to std::move as used.
   if (CE->getNumArgs() == 1) {
     if (FunctionDecl *FD = CE->getDirectCallee()) {
       if (FD->isInStdNamespace() && FD->getIdentifier() &&
           FD->getIdentifier()->isStr("move")) {
-        classify(CE->getArg(0), Use);
+        // RecordTypes are handled in SemaDeclCXX.cpp.
+        if (!CE->getArg(0)->getType()->isRecordType())
+          classify(CE->getArg(0), Use);
         return;
       }
     }
   }
 
-  // If a value is passed by const reference to a function, we should not assume
-  // that it is initialized by the call, and we conservatively do not assume
-  // that it is used.
+  // If a value is passed by const pointer or by const reference to a function,
+  // we should not assume that it is initialized by the call, and we
+  // conservatively do not assume that it is used.
   for (CallExpr::arg_iterator I = CE->arg_begin(), E = CE->arg_end();
-       I != E; ++I)
-    if ((*I)->getType().isConstQualified() && (*I)->isGLValue())
-      classify(*I, Ignore);
+       I != E; ++I) {
+    if ((*I)->isGLValue()) {
+      if ((*I)->getType().isConstQualified())
+        classify((*I), Ignore);
+    } else if (isPointerToConst((*I)->getType())) {
+      const Expr *Ex = stripCasts(DC->getParentASTContext(), *I);
+      const UnaryOperator *UO = dyn_cast<UnaryOperator>(Ex);
+      if (UO && UO->getOpcode() == UO_AddrOf)
+        Ex = UO->getSubExpr();
+      classify(Ex, Ignore);
+    }
+  }
 }
 
 void ClassifyRefs::VisitCastExpr(CastExpr *CE) {
@@ -804,7 +837,7 @@ struct PruneBlocksHandler : public UninitVariablesHandler {
     : hadUse(numBlocks, false), hadAnyUse(false),
       currentBlock(0) {}
 
-  virtual ~PruneBlocksHandler() {}
+  ~PruneBlocksHandler() override {}
 
   /// Records if a CFGBlock had a potential use of an uninitialized variable.
   llvm::BitVector hadUse;
diff --git a/lib/Basic/Diagnostic.cpp b/lib/Basic/Diagnostic.cpp
index 83228ad3c55d..631b97899bd5 100644
--- a/lib/Basic/Diagnostic.cpp
+++ b/lib/Basic/Diagnostic.cpp
@@ -633,7 +633,8 @@ FormatDiagnostic(const char *DiagStr, const char *DiagEnd,
   // When the diagnostic string is only "%0", the entire string is being given
   // by an outside source.  Remove unprintable characters from this string
   // and skip all the other string processing.
-  if (DiagEnd - DiagStr == 2 && DiagStr[0] == '%' && DiagStr[1] == '0' &&
+  if (DiagEnd - DiagStr == 2 &&
+      StringRef(DiagStr, DiagEnd - DiagStr).equals("%0") &&
       getArgKind(0) == DiagnosticsEngine::ak_std_string) {
     const std::string &S = getArgStdStr(0);
     for (char c : S) {
@@ -948,14 +949,8 @@ StoredDiagnostic::StoredDiagnostic(DiagnosticsEngine::Level Level,
   SmallString<64> Message;
   Info.FormatDiagnostic(Message);
   this->Message.assign(Message.begin(), Message.end());
-
-  Ranges.reserve(Info.getNumRanges());
-  for (unsigned I = 0, N = Info.getNumRanges(); I != N; ++I)
-    Ranges.push_back(Info.getRange(I));
-
-  FixIts.reserve(Info.getNumFixItHints());
-  for (unsigned I = 0, N = Info.getNumFixItHints(); I != N; ++I)
-    FixIts.push_back(Info.getFixItHint(I));
+  this->Ranges.assign(Info.getRanges().begin(), Info.getRanges().end());
+  this->FixIts.assign(Info.getFixItHints().begin(), Info.getFixItHints().end());
 }
 
 StoredDiagnostic::StoredDiagnostic(DiagnosticsEngine::Level Level, unsigned ID,
diff --git a/lib/Basic/DiagnosticIDs.cpp b/lib/Basic/DiagnosticIDs.cpp
index 1c68375f3cdd..643503b00b91 100644
--- a/lib/Basic/DiagnosticIDs.cpp
+++ b/lib/Basic/DiagnosticIDs.cpp
@@ -528,7 +528,7 @@ static bool getDiagnosticsInGroup(diag::Flavor Flavor,
   // An empty group is considered to be a warning group: we have empty groups
   // for GCC compatibility, and GCC does not have remarks.
   if (!Group->Members && !Group->SubGroups)
-    return Flavor == diag::Flavor::Remark ? true : false;
+    return Flavor == diag::Flavor::Remark;
 
   bool NotFound = true;
 
diff --git a/lib/Basic/FileManager.cpp b/lib/Basic/FileManager.cpp
index 214e0f35a52e..c46e2c7db380 100644
--- a/lib/Basic/FileManager.cpp
+++ b/lib/Basic/FileManager.cpp
@@ -430,7 +430,7 @@ FileManager::getBufferForFile(const FileEntry *Entry, bool isVolatile,
 
   SmallString<128> FilePath(Entry->getName());
   FixupRelativePath(FilePath);
-  return FS->getBufferForFile(FilePath.str(), FileSize,
+  return FS->getBufferForFile(FilePath, FileSize,
                               /*RequiresNullTerminator=*/true, isVolatile);
 }
 
diff --git a/lib/Basic/IdentifierTable.cpp b/lib/Basic/IdentifierTable.cpp
index 613b43fce95f..e830be9603e6 100644
--- a/lib/Basic/IdentifierTable.cpp
+++ b/lib/Basic/IdentifierTable.cpp
@@ -35,7 +35,7 @@ IdentifierInfo::IdentifierInfo() {
   HasMacro = false;
   HadMacro = false;
   IsExtension = false;
-  IsCXX11CompatKeyword = false;
+  IsFutureCompatKeyword = false;
   IsPoisoned = false;
   IsCPPOperatorKeyword = false;
   NeedsHandleIdentifier = false;
@@ -105,10 +105,12 @@ namespace {
     KEYOPENCL = 0x200,
     KEYC11 = 0x400,
     KEYARC = 0x800,
-    KEYNOMS = 0x01000,
-    WCHARSUPPORT = 0x02000,
-    HALFSUPPORT = 0x04000,
-    KEYALL = (0xffff & ~KEYNOMS) // Because KEYNOMS is used to exclude.
+    KEYNOMS18 = 0x01000,
+    KEYNOOPENCL = 0x02000,
+    WCHARSUPPORT = 0x04000,
+    HALFSUPPORT = 0x08000,
+    KEYALL = (0xffff & ~KEYNOMS18 &
+              ~KEYNOOPENCL) // KEYNOMS18 and KEYNOOPENCL are used to exclude.
   };
 
   /// \brief How a keyword is treated in the selected standard.
@@ -154,15 +156,21 @@ static void AddKeyword(StringRef Keyword,
   KeywordStatus AddResult = getKeywordStatus(LangOpts, Flags);
 
   // Don't add this keyword under MSVCCompat.
-  if (LangOpts.MSVCCompat && (Flags & KEYNOMS))
-     return;
+  if (LangOpts.MSVCCompat && (Flags & KEYNOMS18) &&
+      !LangOpts.isCompatibleWithMSVC(LangOptions::MSVC2015))
+    return;
+
+  // Don't add this keyword under OpenCL.
+  if (LangOpts.OpenCL && (Flags & KEYNOOPENCL))
+    return;
+
   // Don't add this keyword if disabled in this language.
   if (AddResult == KS_Disabled) return;
 
   IdentifierInfo &Info =
       Table.get(Keyword, AddResult == KS_Future ? tok::identifier : TokenCode);
   Info.setIsExtensionToken(AddResult == KS_Extension);
-  Info.setIsCXX11CompatKeyword(AddResult == KS_Future);
+  Info.setIsFutureCompatKeyword(AddResult == KS_Future);
 }
 
 /// AddCXXOperatorKeyword - Register a C++ operator keyword alternative
@@ -207,6 +215,12 @@ void IdentifierTable::AddKeywords(const LangOptions &LangOpts) {
   if (LangOpts.ParseUnknownAnytype)
     AddKeyword("__unknown_anytype", tok::kw___unknown_anytype, KEYALL,
                LangOpts, *this);
+
+  // FIXME: __declspec isn't really a CUDA extension, however it is required for
+  // supporting cuda_builtin_vars.h, which uses __declspec(property). Once that
+  // has been rewritten in terms of something more generic, remove this code.
+  if (LangOpts.CUDA)
+    AddKeyword("__declspec", tok::kw___declspec, KEYALL, LangOpts, *this);
 }
 
 /// \brief Checks if the specified token kind represents a keyword in the
diff --git a/lib/Basic/LangOptions.cpp b/lib/Basic/LangOptions.cpp
index dcbd22817111..2c87845b7b0e 100644
--- a/lib/Basic/LangOptions.cpp
+++ b/lib/Basic/LangOptions.cpp
@@ -30,7 +30,7 @@ void LangOptions::resetNonModularOptions() {
   // FIXME: This should not be reset; modules can be different with different
   // sanitizer options (this affects __has_feature(address_sanitizer) etc).
   Sanitize.clear();
-  SanitizerBlacklistFile.clear();
+  SanitizerBlacklistFiles.clear();
 
   CurrentModule.clear();
   ImplementationOfModule.clear();
diff --git a/lib/Basic/Module.cpp b/lib/Basic/Module.cpp
index 03f9bd3f329c..7308665bd2d8 100644
--- a/lib/Basic/Module.cpp
+++ b/lib/Basic/Module.cpp
@@ -25,14 +25,14 @@
 using namespace clang;
 
 Module::Module(StringRef Name, SourceLocation DefinitionLoc, Module *Parent,
-               bool IsFramework, bool IsExplicit)
+               bool IsFramework, bool IsExplicit, unsigned VisibilityID)
     : Name(Name), DefinitionLoc(DefinitionLoc), Parent(Parent), Directory(),
-      Umbrella(), ASTFile(nullptr), IsMissingRequirement(false),
-      IsAvailable(true), IsFromModuleFile(false), IsFramework(IsFramework),
-      IsExplicit(IsExplicit), IsSystem(false), IsExternC(false),
-      IsInferred(false), InferSubmodules(false), InferExplicitSubmodules(false),
-      InferExportWildcard(false), ConfigMacrosExhaustive(false),
-      NameVisibility(Hidden) {
+      Umbrella(), ASTFile(nullptr), VisibilityID(VisibilityID),
+      IsMissingRequirement(false), IsAvailable(true), IsFromModuleFile(false),
+      IsFramework(IsFramework), IsExplicit(IsExplicit), IsSystem(false),
+      IsExternC(false), IsInferred(false), InferSubmodules(false),
+      InferExplicitSubmodules(false), InferExportWildcard(false),
+      ConfigMacrosExhaustive(false), NameVisibility(Hidden) {
   if (Parent) {
     if (!Parent->isAvailable())
       IsAvailable = false;
@@ -58,16 +58,21 @@ Module::~Module() {
 /// language options has the given feature.
 static bool hasFeature(StringRef Feature, const LangOptions &LangOpts,
                        const TargetInfo &Target) {
-  return llvm::StringSwitch<bool>(Feature)
-           .Case("altivec", LangOpts.AltiVec)
-           .Case("blocks", LangOpts.Blocks)
-           .Case("cplusplus", LangOpts.CPlusPlus)
-           .Case("cplusplus11", LangOpts.CPlusPlus11)
-           .Case("objc", LangOpts.ObjC1)
-           .Case("objc_arc", LangOpts.ObjCAutoRefCount)
-           .Case("opencl", LangOpts.OpenCL)
-           .Case("tls", Target.isTLSSupported())
-           .Default(Target.hasFeature(Feature));
+  bool HasFeature = llvm::StringSwitch<bool>(Feature)
+                        .Case("altivec", LangOpts.AltiVec)
+                        .Case("blocks", LangOpts.Blocks)
+                        .Case("cplusplus", LangOpts.CPlusPlus)
+                        .Case("cplusplus11", LangOpts.CPlusPlus11)
+                        .Case("objc", LangOpts.ObjC1)
+                        .Case("objc_arc", LangOpts.ObjCAutoRefCount)
+                        .Case("opencl", LangOpts.OpenCL)
+                        .Case("tls", Target.isTLSSupported())
+                        .Default(Target.hasFeature(Feature));
+  if (!HasFeature)
+    HasFeature = std::find(LangOpts.ModuleFeatures.begin(),
+                           LangOpts.ModuleFeatures.end(),
+                           Feature) != LangOpts.ModuleFeatures.end();
+  return HasFeature;
 }
 
 bool Module::isAvailable(const LangOptions &LangOpts, const TargetInfo &Target,
@@ -133,11 +138,11 @@ std::string Module::getFullModuleName() const {
   return Result;
 }
 
-const DirectoryEntry *Module::getUmbrellaDir() const {
-  if (const FileEntry *Header = getUmbrellaHeader())
-    return Header->getDir();
+Module::DirectoryName Module::getUmbrellaDir() const {
+  if (Header U = getUmbrellaHeader())
+    return {"", U.Entry->getDir()};
   
-  return Umbrella.dyn_cast<const DirectoryEntry *>();
+  return {UmbrellaAsWritten, Umbrella.dyn_cast<const DirectoryEntry *>()};
 }
 
 ArrayRef<const FileEntry *> Module::getTopHeaders(FileManager &FileMgr) {
@@ -153,6 +158,19 @@ ArrayRef<const FileEntry *> Module::getTopHeaders(FileManager &FileMgr) {
   return llvm::makeArrayRef(TopHeaders.begin(), TopHeaders.end());
 }
 
+bool Module::directlyUses(const Module *Requested) const {
+  auto *Top = getTopLevelModule();
+
+  // A top-level module implicitly uses itself.
+  if (Requested->isSubModuleOf(Top))
+    return true;
+
+  for (auto *Use : Top->DirectUses)
+    if (Requested->isSubModuleOf(Use))
+      return true;
+  return false;
+}
+
 void Module::addRequirement(StringRef Feature, bool RequiredState,
                             const LangOptions &LangOpts,
                             const TargetInfo &Target) {
@@ -316,15 +334,15 @@ void Module::print(raw_ostream &OS, unsigned Indent) const {
     OS << "\n";
   }
   
-  if (const FileEntry *UmbrellaHeader = getUmbrellaHeader()) {
+  if (Header H = getUmbrellaHeader()) {
     OS.indent(Indent + 2);
     OS << "umbrella header \"";
-    OS.write_escaped(UmbrellaHeader->getName());
+    OS.write_escaped(H.NameAsWritten);
     OS << "\"\n";
-  } else if (const DirectoryEntry *UmbrellaDir = getUmbrellaDir()) {
+  } else if (DirectoryName D = getUmbrellaDir()) {
     OS.indent(Indent + 2);
     OS << "umbrella \"";
-    OS.write_escaped(UmbrellaDir->getName());
+    OS.write_escaped(D.NameAsWritten);
     OS << "\"\n";    
   }
 
@@ -457,4 +475,47 @@ void Module::dump() const {
   print(llvm::errs());
 }
 
+void VisibleModuleSet::setVisible(Module *M, SourceLocation Loc,
+                                  VisibleCallback Vis, ConflictCallback Cb) {
+  if (isVisible(M))
+    return;
 
+  ++Generation;
+
+  struct Visiting {
+    Module *M;
+    Visiting *ExportedBy;
+  };
+
+  std::function<void(Visiting)> VisitModule = [&](Visiting V) {
+    // Modules that aren't available cannot be made visible.
+    if (!V.M->isAvailable())
+      return;
+
+    // Nothing to do for a module that's already visible.
+    unsigned ID = V.M->getVisibilityID();
+    if (ImportLocs.size() <= ID)
+      ImportLocs.resize(ID + 1);
+    else if (ImportLocs[ID].isValid())
+      return;
+
+    ImportLocs[ID] = Loc;
+    Vis(M);
+
+    // Make any exported modules visible.
+    SmallVector<Module *, 16> Exports;
+    V.M->getExportedModules(Exports);
+    for (Module *E : Exports)
+      VisitModule({E, &V});
+
+    for (auto &C : V.M->Conflicts) {
+      if (isVisible(C.Other)) {
+        llvm::SmallVector<Module*, 8> Path;
+        for (Visiting *I = &V; I; I = I->ExportedBy)
+          Path.push_back(I->M);
+        Cb(Path, C.Other, C.Message);
+      }
+    }
+  };
+  VisitModule({M, nullptr});
+}
diff --git a/lib/Basic/OpenMPKinds.cpp b/lib/Basic/OpenMPKinds.cpp
index 6e98d48c27d3..b83a0692c132 100644
--- a/lib/Basic/OpenMPKinds.cpp
+++ b/lib/Basic/OpenMPKinds.cpp
@@ -374,7 +374,6 @@ bool clang::isOpenMPPrivate(OpenMPClauseKind Kind) {
 }
 
 bool clang::isOpenMPThreadPrivate(OpenMPClauseKind Kind) {
-  return Kind == OMPC_threadprivate ||
-         Kind == OMPC_copyin; // TODO add next clauses like 'copyprivate'.
+  return Kind == OMPC_threadprivate || Kind == OMPC_copyin;
 }
 
diff --git a/lib/Basic/SanitizerBlacklist.cpp b/lib/Basic/SanitizerBlacklist.cpp
index ea5b8d0da845..095fcd6ccaeb 100644
--- a/lib/Basic/SanitizerBlacklist.cpp
+++ b/lib/Basic/SanitizerBlacklist.cpp
@@ -15,9 +15,9 @@
 
 using namespace clang;
 
-SanitizerBlacklist::SanitizerBlacklist(StringRef BlacklistPath,
-                                       SourceManager &SM)
-    : SCL(llvm::SpecialCaseList::createOrDie(BlacklistPath)), SM(SM) {}
+SanitizerBlacklist::SanitizerBlacklist(
+    const std::vector<std::string> &BlacklistPaths, SourceManager &SM)
+    : SCL(llvm::SpecialCaseList::createOrDie(BlacklistPaths)), SM(SM) {}
 
 bool SanitizerBlacklist::isBlacklistedGlobal(StringRef GlobalName,
                                              StringRef Category) const {
diff --git a/lib/Basic/Sanitizers.cpp b/lib/Basic/Sanitizers.cpp
index e9aaa366e32b..8c4884b8ec36 100644
--- a/lib/Basic/Sanitizers.cpp
+++ b/lib/Basic/Sanitizers.cpp
@@ -11,25 +11,48 @@
 //
 //===----------------------------------------------------------------------===//
 #include "clang/Basic/Sanitizers.h"
+#include "clang/Basic/LLVM.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/ADT/StringSwitch.h"
+#include "llvm/Support/MathExtras.h"
 
 using namespace clang;
 
-SanitizerSet::SanitizerSet() : Kinds(0) {}
+SanitizerSet::SanitizerSet() : Mask(0) {}
 
-bool SanitizerSet::has(SanitizerKind K) const {
-  unsigned Bit = static_cast<unsigned>(K);
-  return Kinds & (1 << Bit);
+bool SanitizerSet::has(SanitizerMask K) const {
+  assert(llvm::countPopulation(K) == 1);
+  return Mask & K;
 }
 
-void SanitizerSet::set(SanitizerKind K, bool Value) {
-  unsigned Bit = static_cast<unsigned>(K);
-  Kinds = Value ? (Kinds | (1 << Bit)) : (Kinds & ~(1 << Bit));
+void SanitizerSet::set(SanitizerMask K, bool Value) {
+  assert(llvm::countPopulation(K) == 1);
+  Mask = Value ? (Mask | K) : (Mask & ~K);
 }
 
 void SanitizerSet::clear() {
-  Kinds = 0;
+  Mask = 0;
 }
 
 bool SanitizerSet::empty() const {
-  return Kinds == 0;
+  return Mask == 0;
+}
+
+SanitizerMask clang::parseSanitizerValue(StringRef Value, bool AllowGroups) {
+  SanitizerMask ParsedKind = llvm::StringSwitch<SanitizerMask>(Value)
+#define SANITIZER(NAME, ID) .Case(NAME, SanitizerKind::ID)
+#define SANITIZER_GROUP(NAME, ID, ALIAS)                                       \
+  .Case(NAME, AllowGroups ? SanitizerKind::ID##Group : 0)
+#include "clang/Basic/Sanitizers.def"
+    .Default(0);
+  return ParsedKind;
+}
+
+SanitizerMask clang::expandSanitizerGroups(SanitizerMask Kinds) {
+#define SANITIZER(NAME, ID)
+#define SANITIZER_GROUP(NAME, ID, ALIAS)                                       \
+  if (Kinds & SanitizerKind::ID##Group)                                        \
+    Kinds |= SanitizerKind::ID;
+#include "clang/Basic/Sanitizers.def"
+  return Kinds;
 }
diff --git a/lib/Basic/SourceManager.cpp b/lib/Basic/SourceManager.cpp
index 118e3f32008b..c0b045331dd6 100644
--- a/lib/Basic/SourceManager.cpp
+++ b/lib/Basic/SourceManager.cpp
@@ -110,8 +110,8 @@ llvm::MemoryBuffer *ContentCache::getBuffer(DiagnosticsEngine &Diag,
   // possible.
   if (!BufferOrError) {
     StringRef FillStr("<<<MISSING SOURCE FILE>>>\n");
-    Buffer.setPointer(MemoryBuffer::getNewMemBuffer(ContentsEntry->getSize(),
-                                                    "<invalid>").release());
+    Buffer.setPointer(MemoryBuffer::getNewUninitMemBuffer(
+                          ContentsEntry->getSize(), "<invalid>").release());
     char *Ptr = const_cast<char*>(Buffer.getPointer()->getBufferStart());
     for (unsigned i = 0, e = ContentsEntry->getSize(); i != e; ++i)
       Ptr[i] = FillStr[i % FillStr.size()];
diff --git a/lib/Basic/TargetInfo.cpp b/lib/Basic/TargetInfo.cpp
index 871bbd5ef43f..330258b025b5 100644
--- a/lib/Basic/TargetInfo.cpp
+++ b/lib/Basic/TargetInfo.cpp
@@ -36,6 +36,7 @@ TargetInfo::TargetInfo(const llvm::Triple &T) : TargetOpts(), Triple(T) {
   LongWidth = LongAlign = 32;
   LongLongWidth = LongLongAlign = 64;
   SuitableAlign = 64;
+  DefaultAlignForAttributeAligned = 128;
   MinGlobalAlign = 0;
   HalfWidth = 16;
   HalfAlign = 16;
diff --git a/lib/Basic/Targets.cpp b/lib/Basic/Targets.cpp
index a7a00570d127..a768081e70e0 100644
--- a/lib/Basic/Targets.cpp
+++ b/lib/Basic/Targets.cpp
@@ -25,9 +25,9 @@
 #include "llvm/ADT/StringRef.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/ADT/Triple.h"
-#include "llvm/IR/Type.h"
 #include "llvm/MC/MCSectionMachO.h"
 #include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/TargetParser.h"
 #include <algorithm>
 #include <memory>
 using namespace clang;
@@ -185,6 +185,28 @@ static void getDarwinDefines(MacroBuilder &Builder, const LangOptions &Opts,
 }
 
 namespace {
+// CloudABI Target
+template <typename Target>
+class CloudABITargetInfo : public OSTargetInfo<Target> {
+protected:
+  void getOSDefines(const LangOptions &Opts, const llvm::Triple &Triple,
+                    MacroBuilder &Builder) const override {
+    Builder.defineMacro("__CloudABI__");
+    Builder.defineMacro("__ELF__");
+
+    // CloudABI uses ISO/IEC 10646:2012 for wchar_t, char16_t and char32_t.
+    Builder.defineMacro("__STDC_ISO_10646__", "201206L");
+    Builder.defineMacro("__STDC_UTF_16__");
+    Builder.defineMacro("__STDC_UTF_32__");
+  }
+
+public:
+  CloudABITargetInfo(const llvm::Triple &Triple)
+      : OSTargetInfo<Target>(Triple) {
+    this->UserLabelPrefix = "";
+  }
+};
+
 template<typename Target>
 class DarwinTargetInfo : public OSTargetInfo<Target> {
 protected:
@@ -364,8 +386,13 @@ protected:
     DefineStd(Builder, "linux", Opts);
     Builder.defineMacro("__gnu_linux__");
     Builder.defineMacro("__ELF__");
-    if (Triple.getEnvironment() == llvm::Triple::Android)
+    if (Triple.getEnvironment() == llvm::Triple::Android) {
       Builder.defineMacro("__ANDROID__", "1");
+      unsigned Maj, Min, Rev;
+      Triple.getOSVersion(Maj, Min, Rev);
+      this->PlatformName = "android";
+      this->PlatformMinVersion = VersionTuple(Maj, Min, Rev);
+    }
     if (Opts.POSIXThreads)
       Builder.defineMacro("_REENTRANT");
     if (Opts.CPlusPlus)
@@ -473,6 +500,17 @@ protected:
     Builder.defineMacro("__ELF__");
     if (Opts.POSIXThreads)
       Builder.defineMacro("_REENTRANT");
+
+    switch (Triple.getArch()) {
+    default:
+      break;
+    case llvm::Triple::arm:
+    case llvm::Triple::armeb:
+    case llvm::Triple::thumb:
+    case llvm::Triple::thumbeb:
+      Builder.defineMacro("__ARM_DWARF_EH__");
+      break;
+    }
   }
 public:
   BitrigTargetInfo(const llvm::Triple &Triple) : OSTargetInfo<Target>(Triple) {
@@ -526,6 +564,33 @@ public:
   }
 };
 
+template <typename Target>
+class PS4OSTargetInfo : public OSTargetInfo<Target> {
+protected:
+  void getOSDefines(const LangOptions &Opts, const llvm::Triple &Triple,
+                    MacroBuilder &Builder) const override {
+    Builder.defineMacro("__FreeBSD__", "9");
+    Builder.defineMacro("__FreeBSD_cc_version", "900001");
+    Builder.defineMacro("__KPRINTF_ATTRIBUTE__");
+    DefineStd(Builder, "unix", Opts);
+    Builder.defineMacro("__ELF__");
+    Builder.defineMacro("__PS4__");
+  }
+public:
+  PS4OSTargetInfo(const llvm::Triple &Triple) : OSTargetInfo<Target>(Triple) {
+    this->WCharType = this->UnsignedShort;
+
+    this->UserLabelPrefix = "";
+
+    switch (Triple.getArch()) {
+    default:
+    case llvm::Triple::x86_64:
+      this->MCountName = ".mcount";
+      break;
+    }
+  }
+};
+
 // Solaris target
 template<typename Target>
 class SolarisTargetInfo : public OSTargetInfo<Target> {
@@ -574,7 +639,7 @@ protected:
       if (Opts.RTTIData)
         Builder.defineMacro("_CPPRTTI");
 
-      if (Opts.Exceptions)
+      if (Opts.CXXExceptions)
         Builder.defineMacro("_CPPUNWIND");
     }
 
@@ -592,6 +657,9 @@ protected:
       Builder.defineMacro("_MSC_FULL_VER", Twine(Opts.MSCompatibilityVersion));
       // FIXME We cannot encode the revision information into 32-bits
       Builder.defineMacro("_MSC_BUILD", Twine(1));
+
+      if (Opts.CPlusPlus11 && Opts.isCompatibleWithMSVC(LangOptions::MSVC2015))
+        Builder.defineMacro("_HAS_CHAR16_T_LANGUAGE_SUPPORT", Twine(1));
     }
 
     if (Opts.MicrosoftExt) {
@@ -647,8 +715,7 @@ public:
     // RegParmMax is inherited from the underlying architecture
     this->LongDoubleFormat = &llvm::APFloat::IEEEdouble;
     if (Triple.getArch() == llvm::Triple::arm) {
-      this->DescriptionString =
-          "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S128";
+      // Handled in ARM's setABI().
     } else if (Triple.getArch() == llvm::Triple::x86) {
       this->DescriptionString = "e-m:e-p:32:32-i64:64-n8:16:32-S128";
     } else if (Triple.getArch() == llvm::Triple::x86_64) {
@@ -660,19 +727,12 @@ public:
       this->DescriptionString = "e-p:32:32-i64:64";
     }
   }
-  typename Target::CallingConvCheckResult checkCallingConvention(
-      CallingConv CC) const override {
-    return CC == CC_PnaclCall ? Target::CCCR_OK :
-        Target::checkCallingConvention(CC);
-  }
 };
-} // end anonymous namespace.
 
 //===----------------------------------------------------------------------===//
 // Specific target implementations.
 //===----------------------------------------------------------------------===//
 
-namespace {
 // PPC abstract base class
 class PPCTargetInfo : public TargetInfo {
   static const Builtin::Info BuiltinInfo[];
@@ -683,13 +743,21 @@ class PPCTargetInfo : public TargetInfo {
   // Target cpu features.
   bool HasVSX;
   bool HasP8Vector;
+  bool HasP8Crypto;
+  bool HasDirectMove;
+  bool HasQPX;
+  bool HasHTM;
+  bool HasBPERMD;
+  bool HasExtDiv;
 
 protected:
   std::string ABI;
 
 public:
   PPCTargetInfo(const llvm::Triple &Triple)
-    : TargetInfo(Triple), HasVSX(false), HasP8Vector(false) {
+    : TargetInfo(Triple), HasVSX(false), HasP8Vector(false),
+      HasP8Crypto(false), HasDirectMove(false), HasQPX(false), HasHTM(false),
+      HasBPERMD(false), HasExtDiv(false) {
     BigEndian = (Triple.getArch() != llvm::Triple::ppc64le);
     LongDoubleWidth = LongDoubleAlign = 128;
     LongDoubleFormat = &llvm::APFloat::PPCDoubleDouble;
@@ -949,8 +1017,40 @@ bool PPCTargetInfo::handleTargetFeatures(std::vector<std::string> &Features,
       continue;
     }
 
+    if (Feature == "bpermd") {
+      HasBPERMD = true;
+      continue;
+    }
+
+    if (Feature == "extdiv") {
+      HasExtDiv = true;
+      continue;
+    }
+
     if (Feature == "power8-vector") {
       HasP8Vector = true;
+      HasVSX = true;
+      continue;
+    }
+
+    if (Feature == "crypto") {
+      HasP8Crypto = true;
+      continue;
+    }
+
+    if (Feature == "direct-move") {
+      HasDirectMove = true;
+      HasVSX = true;
+      continue;
+    }
+
+    if (Feature == "qpx") {
+      HasQPX = true;
+      continue;
+    }
+
+    if (Feature == "htm") {
+      HasHTM = true;
       continue;
     }
 
@@ -988,7 +1088,7 @@ void PPCTargetInfo::getTargetDefines(const LangOptions &Opts,
   }
 
   // ABI options.
-  if (ABI == "elfv1")
+  if (ABI == "elfv1" || ABI == "elfv1-qpx")
     Builder.defineMacro("_CALL_ELF", "1");
   if (ABI == "elfv2")
     Builder.defineMacro("_CALL_ELF", "2");
@@ -1106,6 +1206,18 @@ void PPCTargetInfo::getTargetDefines(const LangOptions &Opts,
     Builder.defineMacro("__VSX__");
   if (HasP8Vector)
     Builder.defineMacro("__POWER8_VECTOR__");
+  if (HasP8Crypto)
+    Builder.defineMacro("__CRYPTO__");
+  if (HasHTM)
+    Builder.defineMacro("__HTM__");
+  if (getTriple().getArch() == llvm::Triple::ppc64le ||
+      (defs & ArchDefinePwr8) || (CPU == "pwr8")) {
+    Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_1");
+    Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_2");
+    Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4");
+    if (PointerWidth == 64)
+      Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_8");
+  }
 
   // FIXME: The following are not yet generated here by Clang, but are
   //        generated by GCC:
@@ -1144,9 +1256,28 @@ void PPCTargetInfo::getDefaultFeatures(llvm::StringMap<bool> &Features) const {
     .Default(false);
 
   Features["qpx"] = (CPU == "a2q");
-
-  if (!ABI.empty())
-    Features[ABI] = true;
+  Features["crypto"] = llvm::StringSwitch<bool>(CPU)
+    .Case("ppc64le", true)
+    .Case("pwr8", true)
+    .Default(false);
+  Features["power8-vector"] = llvm::StringSwitch<bool>(CPU)
+    .Case("ppc64le", true)
+    .Case("pwr8", true)
+    .Default(false);
+  Features["bpermd"] = llvm::StringSwitch<bool>(CPU)
+    .Case("ppc64le", true)
+    .Case("pwr8", true)
+    .Case("pwr7", true)
+    .Default(false);
+  Features["extdiv"] = llvm::StringSwitch<bool>(CPU)
+    .Case("ppc64le", true)
+    .Case("pwr8", true)
+    .Case("pwr7", true)
+    .Default(false);
+  Features["direct-move"] = llvm::StringSwitch<bool>(CPU)
+    .Case("ppc64le", true)
+    .Case("pwr8", true)
+    .Default(false);
 }
 
 bool PPCTargetInfo::hasFeature(StringRef Feature) const {
@@ -1154,6 +1285,12 @@ bool PPCTargetInfo::hasFeature(StringRef Feature) const {
     .Case("powerpc", true)
     .Case("vsx", HasVSX)
     .Case("power8-vector", HasP8Vector)
+    .Case("crypto", HasP8Crypto)
+    .Case("direct-move", HasDirectMove)
+    .Case("qpx", HasQPX)
+    .Case("htm", HasHTM)
+    .Case("bpermd", HasBPERMD)
+    .Case("extdiv", HasExtDiv)
     .Default(false);
 }
 
@@ -1259,9 +1396,7 @@ void PPCTargetInfo::getGCCRegAliases(const GCCRegAlias *&Aliases,
   Aliases = GCCRegAliases;
   NumAliases = llvm::array_lengthof(GCCRegAliases);
 }
-} // end anonymous namespace.
 
-namespace {
 class PPC32TargetInfo : public PPCTargetInfo {
 public:
   PPC32TargetInfo(const llvm::Triple &Triple) : PPCTargetInfo(Triple) {
@@ -1293,11 +1428,9 @@ public:
     return TargetInfo::PowerABIBuiltinVaList;
   }
 };
-} // end anonymous namespace.
 
 // Note: ABI differences may eventually require us to have a separate
 // TargetInfo for little endian.
-namespace {
 class PPC64TargetInfo : public PPCTargetInfo {
 public:
   PPC64TargetInfo(const llvm::Triple &Triple) : PPCTargetInfo(Triple) {
@@ -1334,17 +1467,14 @@ public:
   }
   // PPC64 Linux-specifc ABI options.
   bool setABI(const std::string &Name) override {
-    if (Name == "elfv1" || Name == "elfv2") {
+    if (Name == "elfv1" || Name == "elfv1-qpx" || Name == "elfv2") {
       ABI = Name;
       return true;
     }
     return false;
   }
 };
-} // end anonymous namespace.
-
 
-namespace {
 class DarwinPPC32TargetInfo :
   public DarwinTargetInfo<PPC32TargetInfo> {
 public:
@@ -1372,9 +1502,7 @@ public:
     DescriptionString = "E-m:o-i64:64-n32:64";
   }
 };
-} // end anonymous namespace.
 
-namespace {
   static const unsigned NVPTXAddrSpaceMap[] = {
     1,    // opencl_global
     3,    // opencl_local
@@ -1396,6 +1524,7 @@ namespace {
     GK_SM21,
     GK_SM30,
     GK_SM35,
+    GK_SM37,
   } GPU;
 
   public:
@@ -1431,6 +1560,9 @@ namespace {
         case GK_SM35:
           CUDAArchCode = "350";
           break;
+        case GK_SM37:
+          CUDAArchCode = "370";
+          break;
         default:
           llvm_unreachable("Unhandled target CPU");
         }
@@ -1483,6 +1615,7 @@ namespace {
                 .Case("sm_21", GK_SM21)
                 .Case("sm_30", GK_SM30)
                 .Case("sm_35", GK_SM35)
+                .Case("sm_37", GK_SM37)
                 .Default(GK_NONE);
 
       return GPU != GK_NONE;
@@ -1510,24 +1643,23 @@ namespace {
   public:
     NVPTX32TargetInfo(const llvm::Triple &Triple) : NVPTXTargetInfo(Triple) {
       PointerWidth = PointerAlign = 32;
-      SizeType     = PtrDiffType = TargetInfo::UnsignedInt;
+      SizeType = TargetInfo::UnsignedInt;
+      PtrDiffType = TargetInfo::SignedInt;
       IntPtrType = TargetInfo::SignedInt;
       DescriptionString = "e-p:32:32-i64:64-v16:16-v32:32-n16:32:64";
-  }
+    }
   };
 
   class NVPTX64TargetInfo : public NVPTXTargetInfo {
   public:
     NVPTX64TargetInfo(const llvm::Triple &Triple) : NVPTXTargetInfo(Triple) {
       PointerWidth = PointerAlign = 64;
-      SizeType     = PtrDiffType = TargetInfo::UnsignedLongLong;
-      IntPtrType = TargetInfo::SignedLongLong;
+      SizeType = TargetInfo::UnsignedLong;
+      PtrDiffType = TargetInfo::SignedLong;
+      IntPtrType = TargetInfo::SignedLong;
       DescriptionString = "e-i64:64-v16:16-v32:32-n16:32:64";
-  }
+    }
   };
-}
-
-namespace {
 
 static const unsigned R600AddrSpaceMap[] = {
   1,    // opencl_global
@@ -1557,6 +1689,7 @@ static const char *DescriptionStringSI =
 
 class R600TargetInfo : public TargetInfo {
   static const Builtin::Info BuiltinInfo[];
+  static const char * const GCCRegNames[];
 
   /// \brief The GPU profiles supported by the R600 target.
   enum GPUKind {
@@ -1573,10 +1706,27 @@ class R600TargetInfo : public TargetInfo {
     GK_SEA_ISLANDS
   } GPU;
 
+  bool hasFP64:1;
+  bool hasFMAF:1;
+  bool hasLDEXPF:1;
+
 public:
   R600TargetInfo(const llvm::Triple &Triple)
-      : TargetInfo(Triple), GPU(GK_R600) {
-    DescriptionString = DescriptionStringR600;
+      : TargetInfo(Triple) {
+
+    if (Triple.getArch() == llvm::Triple::amdgcn) {
+      DescriptionString = DescriptionStringSI;
+      GPU = GK_SOUTHERN_ISLANDS;
+      hasFP64 = true;
+      hasFMAF = true;
+      hasLDEXPF = true;
+    } else {
+      DescriptionString = DescriptionStringR600;
+      GPU = GK_R600;
+      hasFP64 = false;
+      hasFMAF = false;
+      hasLDEXPF = false;
+    }
     AddrSpaceMap = &R600AddrSpaceMap;
     UseAddrSpaceMapMangling = true;
   }
@@ -1600,10 +1750,7 @@ public:
   }
 
   void getGCCRegNames(const char * const *&Names,
-                      unsigned &numNames) const override {
-    Names = nullptr;
-    numNames = 0;
-  }
+                      unsigned &NumNames) const override;
 
   void getGCCRegAliases(const GCCRegAlias *&Aliases,
                         unsigned &NumAliases) const override {
@@ -1625,6 +1772,13 @@ public:
   void getTargetDefines(const LangOptions &Opts,
                         MacroBuilder &Builder) const override {
     Builder.defineMacro("__R600__");
+    if (hasFMAF)
+      Builder.defineMacro("__HAS_FMAF__");
+    if (hasLDEXPF)
+      Builder.defineMacro("__HAS_LDEXPF__");
+    if (hasFP64 && Opts.OpenCL) {
+      Builder.defineMacro("cl_khr_fp64");
+    }
   }
 
   BuiltinVaListKind getBuiltinVaListKind() const override {
@@ -1682,16 +1836,25 @@ public:
     case GK_EVERGREEN:
     case GK_NORTHERN_ISLANDS:
       DescriptionString = DescriptionStringR600;
+      hasFP64 = false;
+      hasFMAF = false;
+      hasLDEXPF = false;
       break;
     case GK_R600_DOUBLE_OPS:
     case GK_R700_DOUBLE_OPS:
     case GK_EVERGREEN_DOUBLE_OPS:
     case GK_CAYMAN:
       DescriptionString = DescriptionStringR600DoubleOps;
+      hasFP64 = true;
+      hasFMAF = true;
+      hasLDEXPF = false;
       break;
     case GK_SOUTHERN_ISLANDS:
     case GK_SEA_ISLANDS:
       DescriptionString = DescriptionStringSI;
+      hasFP64 = true;
+      hasFMAF = true;
+      hasLDEXPF = true;
       break;
     }
 
@@ -1704,10 +1867,65 @@ const Builtin::Info R600TargetInfo::BuiltinInfo[] = {
   { #ID, TYPE, ATTRS, 0, ALL_LANGUAGES },
 #include "clang/Basic/BuiltinsR600.def"
 };
+const char * const R600TargetInfo::GCCRegNames[] = {
+  "v0", "v1", "v2", "v3", "v4", "v5", "v6", "v7",
+  "v8", "v9", "v10", "v11", "v12", "v13", "v14", "v15",
+  "v16", "v17", "v18", "v19", "v20", "v21", "v22", "v23",
+  "v24", "v25", "v26", "v27", "v28", "v29", "v30", "v31",
+  "v32", "v33", "v34", "v35", "v36", "v37", "v38", "v39",
+  "v40", "v41", "v42", "v43", "v44", "v45", "v46", "v47",
+  "v48", "v49", "v50", "v51", "v52", "v53", "v54", "v55",
+  "v56", "v57", "v58", "v59", "v60", "v61", "v62", "v63",
+  "v64", "v65", "v66", "v67", "v68", "v69", "v70", "v71",
+  "v72", "v73", "v74", "v75", "v76", "v77", "v78", "v79",
+  "v80", "v81", "v82", "v83", "v84", "v85", "v86", "v87",
+  "v88", "v89", "v90", "v91", "v92", "v93", "v94", "v95",
+  "v96", "v97", "v98", "v99", "v100", "v101", "v102", "v103",
+  "v104", "v105", "v106", "v107", "v108", "v109", "v110", "v111",
+  "v112", "v113", "v114", "v115", "v116", "v117", "v118", "v119",
+  "v120", "v121", "v122", "v123", "v124", "v125", "v126", "v127",
+  "v128", "v129", "v130", "v131", "v132", "v133", "v134", "v135",
+  "v136", "v137", "v138", "v139", "v140", "v141", "v142", "v143",
+  "v144", "v145", "v146", "v147", "v148", "v149", "v150", "v151",
+  "v152", "v153", "v154", "v155", "v156", "v157", "v158", "v159",
+  "v160", "v161", "v162", "v163", "v164", "v165", "v166", "v167",
+  "v168", "v169", "v170", "v171", "v172", "v173", "v174", "v175",
+  "v176", "v177", "v178", "v179", "v180", "v181", "v182", "v183",
+  "v184", "v185", "v186", "v187", "v188", "v189", "v190", "v191",
+  "v192", "v193", "v194", "v195", "v196", "v197", "v198", "v199",
+  "v200", "v201", "v202", "v203", "v204", "v205", "v206", "v207",
+  "v208", "v209", "v210", "v211", "v212", "v213", "v214", "v215",
+  "v216", "v217", "v218", "v219", "v220", "v221", "v222", "v223",
+  "v224", "v225", "v226", "v227", "v228", "v229", "v230", "v231",
+  "v232", "v233", "v234", "v235", "v236", "v237", "v238", "v239",
+  "v240", "v241", "v242", "v243", "v244", "v245", "v246", "v247",
+  "v248", "v249", "v250", "v251", "v252", "v253", "v254", "v255",
+  "s0", "s1", "s2", "s3", "s4", "s5", "s6", "s7",
+  "s8", "s9", "s10", "s11", "s12", "s13", "s14", "s15",
+  "s16", "s17", "s18", "s19", "s20", "s21", "s22", "s23",
+  "s24", "s25", "s26", "s27", "s28", "s29", "s30", "s31",
+  "s32", "s33", "s34", "s35", "s36", "s37", "s38", "s39",
+  "s40", "s41", "s42", "s43", "s44", "s45", "s46", "s47",
+  "s48", "s49", "s50", "s51", "s52", "s53", "s54", "s55",
+  "s56", "s57", "s58", "s59", "s60", "s61", "s62", "s63",
+  "s64", "s65", "s66", "s67", "s68", "s69", "s70", "s71",
+  "s72", "s73", "s74", "s75", "s76", "s77", "s78", "s79",
+  "s80", "s81", "s82", "s83", "s84", "s85", "s86", "s87",
+  "s88", "s89", "s90", "s91", "s92", "s93", "s94", "s95",
+  "s96", "s97", "s98", "s99", "s100", "s101", "s102", "s103",
+  "s104", "s105", "s106", "s107", "s108", "s109", "s110", "s111",
+  "s112", "s113", "s114", "s115", "s116", "s117", "s118", "s119",
+  "s120", "s121", "s122", "s123", "s124", "s125", "s126", "s127"
+  "exec", "vcc", "scc", "m0", "flat_scr", "exec_lo", "exec_hi",
+  "vcc_lo", "vcc_hi", "flat_scr_lo", "flat_scr_hi"
+};
 
-} // end anonymous namespace
+void R600TargetInfo::getGCCRegNames(const char * const *&Names,
+                                    unsigned &NumNames) const {
+  Names = GCCRegNames;
+  NumNames = llvm::array_lengthof(GCCRegNames);
+}
 
-namespace {
 // Namespace for x86 abstract base class
 const Builtin::Info BuiltinInfo[] = {
 #define BUILTIN(ID, TYPE, ATTRS) { #ID, TYPE, ATTRS, 0, ALL_LANGUAGES },
@@ -2496,11 +2714,6 @@ void X86TargetInfo::setXOPLevel(llvm::StringMap<bool> &Features, XOPEnum Level,
 
 void X86TargetInfo::setFeatureEnabledImpl(llvm::StringMap<bool> &Features,
                                           StringRef Name, bool Enabled) {
-  // FIXME: This *really* should not be here.  We need some way of translating
-  // options into llvm subtarget features.
-  if (Name == "sse4")
-    Name = "sse4.2";
-
   Features[Name] = Enabled;
 
   if (Name == "mmx") {
@@ -3236,9 +3449,7 @@ X86TargetInfo::convertConstraint(const char *&Constraint) const {
     return std::string(1, *Constraint);
   }
 }
-} // end anonymous namespace
 
-namespace {
 // X86-32 generic target
 class X86_32TargetInfo : public X86TargetInfo {
 public:
@@ -3293,9 +3504,7 @@ public:
     return X86TargetInfo::validateOperandSize(Constraint, Size);
   }
 };
-} // end anonymous namespace
 
-namespace {
 class NetBSDI386TargetInfo : public NetBSDTargetInfo<X86_32TargetInfo> {
 public:
   NetBSDI386TargetInfo(const llvm::Triple &Triple)
@@ -3311,9 +3520,7 @@ public:
     return 1;
   }
 };
-} // end anonymous namespace
 
-namespace {
 class OpenBSDI386TargetInfo : public OpenBSDTargetInfo<X86_32TargetInfo> {
 public:
   OpenBSDI386TargetInfo(const llvm::Triple &Triple)
@@ -3323,9 +3530,7 @@ public:
     PtrDiffType = SignedLong;
   }
 };
-} // end anonymous namespace
 
-namespace {
 class BitrigI386TargetInfo : public BitrigTargetInfo<X86_32TargetInfo> {
 public:
   BitrigI386TargetInfo(const llvm::Triple &Triple)
@@ -3335,9 +3540,7 @@ public:
     PtrDiffType = SignedLong;
   }
 };
-} // end anonymous namespace
 
-namespace {
 class DarwinI386TargetInfo : public DarwinTargetInfo<X86_32TargetInfo> {
 public:
   DarwinI386TargetInfo(const llvm::Triple &Triple)
@@ -3353,9 +3556,7 @@ public:
   }
 
 };
-} // end anonymous namespace
 
-namespace {
 // x86-32 Windows target
 class WindowsX86_32TargetInfo : public WindowsTargetInfo<X86_32TargetInfo> {
 public:
@@ -3365,8 +3566,9 @@ public:
     DoubleAlign = LongLongAlign = 64;
     bool IsWinCOFF =
         getTriple().isOSWindows() && getTriple().isOSBinFormatCOFF();
-    DescriptionString = IsWinCOFF ? "e-m:w-p:32:32-i64:64-f80:32-n8:16:32-S32"
-                                  : "e-m:e-p:32:32-i64:64-f80:32-n8:16:32-S32";
+    DescriptionString = IsWinCOFF
+                            ? "e-m:x-p:32:32-i64:64-f80:32-n8:16:32-a:0:32-S32"
+                            : "e-m:e-p:32:32-i64:64-f80:32-n8:16:32-a:0:32-S32";
   }
   void getTargetDefines(const LangOptions &Opts,
                         MacroBuilder &Builder) const override {
@@ -3436,9 +3638,7 @@ public:
     addMinGWDefines(Opts, Builder);
   }
 };
-} // end anonymous namespace
 
-namespace {
 // x86-32 Cygwin target
 class CygwinX86_32TargetInfo : public X86_32TargetInfo {
 public:
@@ -3447,7 +3647,7 @@ public:
     TLSSupported = false;
     WCharType = UnsignedShort;
     DoubleAlign = LongLongAlign = 64;
-    DescriptionString = "e-m:w-p:32:32-i64:64-f80:32-n8:16:32-S32";
+    DescriptionString = "e-m:x-p:32:32-i64:64-f80:32-n8:16:32-a:0:32-S32";
   }
   void getTargetDefines(const LangOptions &Opts,
                         MacroBuilder &Builder) const override {
@@ -3460,9 +3660,7 @@ public:
       Builder.defineMacro("_GNU_SOURCE");
   }
 };
-} // end anonymous namespace
 
-namespace {
 // x86-32 Haiku target
 class HaikuX86_32TargetInfo : public X86_32TargetInfo {
 public:
@@ -3481,7 +3679,6 @@ public:
     Builder.defineMacro("__HAIKU__");
   }
 };
-} // end anonymous namespace
 
 // RTEMS Target
 template<typename Target>
@@ -3518,7 +3715,6 @@ public:
   }
 };
 
-namespace {
 // x86-32 RTEMS target
 class RTEMSX86_32TargetInfo : public X86_32TargetInfo {
 public:
@@ -3535,14 +3731,14 @@ public:
     Builder.defineMacro("__rtems__");
   }
 };
-} // end anonymous namespace
 
-namespace {
 // x86-64 generic target
 class X86_64TargetInfo : public X86TargetInfo {
 public:
   X86_64TargetInfo(const llvm::Triple &Triple) : X86TargetInfo(Triple) {
     const bool IsX32 = getTriple().getEnvironment() == llvm::Triple::GNUX32;
+    bool IsWinCOFF =
+        getTriple().isOSWindows() && getTriple().isOSBinFormatCOFF();
     LongWidth = LongAlign = PointerWidth = PointerAlign = IsX32 ? 32 : 64;
     LongDoubleWidth = 128;
     LongDoubleAlign = 128;
@@ -3557,9 +3753,10 @@ public:
     RegParmMax = 6;
 
     // Pointers are 32-bit in x32.
-    DescriptionString = (IsX32)
-                            ? "e-m:e-p:32:32-i64:64-f80:128-n8:16:32:64-S128"
-                            : "e-m:e-i64:64-f80:128-n8:16:32:64-S128";
+    DescriptionString = IsX32 ? "e-m:e-p:32:32-i64:64-f80:128-n8:16:32:64-S128"
+                              : IsWinCOFF
+                                    ? "e-m:w-i64:64-f80:128-n8:16:32:64-S128"
+                                    : "e-m:e-i64:64-f80:128-n8:16:32:64-S128";
 
     // Use fpret only for long double.
     RealTypeUsesObjCFPRet = (1 << TargetInfo::LongDouble);
@@ -3595,9 +3792,7 @@ public:
   // for x32 we need it here explicitly
   bool hasInt128Type() const override { return true; }
 };
-} // end anonymous namespace
 
-namespace {
 // x86-64 Windows target
 class WindowsX86_64TargetInfo : public WindowsTargetInfo<X86_64TargetInfo> {
 public:
@@ -3613,24 +3808,34 @@ public:
     IntPtrType = SignedLongLong;
     this->UserLabelPrefix = "";
   }
+
   void getTargetDefines(const LangOptions &Opts,
                                 MacroBuilder &Builder) const override {
     WindowsTargetInfo<X86_64TargetInfo>::getTargetDefines(Opts, Builder);
     Builder.defineMacro("_WIN64");
   }
+
   BuiltinVaListKind getBuiltinVaListKind() const override {
     return TargetInfo::CharPtrBuiltinVaList;
   }
+
   CallingConvCheckResult checkCallingConvention(CallingConv CC) const override {
-    return (CC == CC_C ||
-            CC == CC_X86VectorCall ||
-            CC == CC_IntelOclBicc ||
-            CC == CC_X86_64SysV) ? CCCR_OK : CCCR_Warning;
+    switch (CC) {
+    case CC_X86StdCall:
+    case CC_X86ThisCall:
+    case CC_X86FastCall:
+      return CCCR_Ignore;
+    case CC_C:
+    case CC_X86VectorCall:
+    case CC_IntelOclBicc:
+    case CC_X86_64SysV:
+      return CCCR_OK;
+    default:
+      return CCCR_Warning;
+    }
   }
 };
-} // end anonymous namespace
 
-namespace {
 // x86-64 Windows Visual Studio target
 class MicrosoftX86_64TargetInfo : public WindowsX86_64TargetInfo {
 public:
@@ -3647,9 +3852,7 @@ public:
     Builder.defineMacro("_M_AMD64");
   }
 };
-} // end anonymous namespace
 
-namespace {
 // x86-64 MinGW target
 class MinGWX86_64TargetInfo : public WindowsX86_64TargetInfo {
 public:
@@ -3667,9 +3870,7 @@ public:
       Builder.defineMacro("__SEH__");
   }
 };
-} // end anonymous namespace
 
-namespace {
 class DarwinX86_64TargetInfo : public DarwinTargetInfo<X86_64TargetInfo> {
 public:
   DarwinX86_64TargetInfo(const llvm::Triple &Triple)
@@ -3683,9 +3884,7 @@ public:
     DescriptionString = "e-m:o-i64:64-f80:128-n8:16:32:64-S128";
   }
 };
-} // end anonymous namespace
 
-namespace {
 class OpenBSDX86_64TargetInfo : public OpenBSDTargetInfo<X86_64TargetInfo> {
 public:
   OpenBSDX86_64TargetInfo(const llvm::Triple &Triple)
@@ -3694,9 +3893,7 @@ public:
     Int64Type = SignedLongLong;
   }
 };
-} // end anonymous namespace
 
-namespace {
 class BitrigX86_64TargetInfo : public BitrigTargetInfo<X86_64TargetInfo> {
 public:
   BitrigX86_64TargetInfo(const llvm::Triple &Triple)
@@ -3705,10 +3902,7 @@ public:
     Int64Type = SignedLongLong;
   }
 };
-}
 
-
-namespace {
 class ARMTargetInfo : public TargetInfo {
   // Possible FPU choices.
   enum FPUMode {
@@ -3800,8 +3994,9 @@ class ARMTargetInfo : public TargetInfo {
     DoubleAlign = LongLongAlign = LongDoubleAlign = SuitableAlign = 64;
     const llvm::Triple &T = getTriple();
 
-    // size_t is unsigned long on MachO-derived environments and NetBSD.
-    if (T.isOSBinFormatMachO() || T.getOS() == llvm::Triple::NetBSD)
+    // size_t is unsigned long on MachO-derived environments, NetBSD and Bitrig.
+    if (T.isOSBinFormatMachO() || T.getOS() == llvm::Triple::NetBSD ||
+        T.getOS() == llvm::Triple::Bitrig)
       SizeType = UnsignedLong;
     else
       SizeType = UnsignedInt;
@@ -3831,16 +4026,18 @@ class ARMTargetInfo : public TargetInfo {
           BigEndian ? "E-m:o-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
                     : "e-m:o-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64";
     } else if (T.isOSWindows()) {
-      // FIXME: this is invalid for WindowsCE
       assert(!BigEndian && "Windows on ARM does not support big endian");
       DescriptionString = "e"
-                          "-m:e"
+                          "-m:w"
                           "-p:32:32"
                           "-i64:64"
                           "-v128:64:128"
                           "-a:0:32"
                           "-n32"
                           "-S64";
+    } else if (T.isOSNaCl()) {
+      assert(!BigEndian && "NaCl on ARM does not support big endian");
+      DescriptionString = "e-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S128";
     } else {
       DescriptionString =
           BigEndian ? "E-m:e-p:32:32-i64:64-v128:64:128-a:0:32-n32-S64"
@@ -3984,8 +4181,15 @@ public:
     return false;
   }
 
+  // FIXME: This should be based on Arch attributes, not CPU names.
   void getDefaultFeatures(llvm::StringMap<bool> &Features) const override {
     StringRef ArchName = getTriple().getArchName();
+    unsigned ArchKind =
+                llvm::ARMTargetParser::parseArch(
+                llvm::ARMTargetParser::getCanonicalArchName(ArchName));
+    bool IsV8 = (ArchKind == llvm::ARM::AK_ARMV8A ||
+                 ArchKind == llvm::ARM::AK_ARMV8_1A);
+
     if (CPU == "arm1136jf-s" || CPU == "arm1176jzf-s" || CPU == "mpcore")
       Features["vfp2"] = true;
     else if (CPU == "cortex-a8" || CPU == "cortex-a9") {
@@ -4002,28 +4206,19 @@ public:
       Features["neon"] = true;
       Features["hwdiv"] = true;
       Features["hwdiv-arm"] = true;
-    } else if (CPU == "cyclone") {
-      Features["v8fp"] = true;
-      Features["neon"] = true;
-      Features["hwdiv"] = true;
-      Features["hwdiv-arm"] = true;
-    } else if (CPU == "cortex-a53" || CPU == "cortex-a57") {
+    } else if (CPU == "cyclone" || CPU == "cortex-a53" || CPU == "cortex-a57" ||
+               CPU == "cortex-a72") {
       Features["fp-armv8"] = true;
       Features["neon"] = true;
       Features["hwdiv"] = true;
       Features["hwdiv-arm"] = true;
       Features["crc"] = true;
       Features["crypto"] = true;
-    } else if (CPU == "cortex-r5" ||
-               // Enable the hwdiv extension for all v8a AArch32 cores by
-               // default.
-               ArchName == "armv8a" || ArchName == "armv8" ||
-               ArchName == "armebv8a" || ArchName == "armebv8" ||
-               ArchName == "thumbv8a" || ArchName == "thumbv8" ||
-               ArchName == "thumbebv8a" || ArchName == "thumbebv8") {
+    } else if (CPU == "cortex-r5" || CPU == "cortex-r7" || IsV8) {
       Features["hwdiv"] = true;
       Features["hwdiv-arm"] = true;
-    } else if (CPU == "cortex-m3" || CPU == "cortex-m4" || CPU == "cortex-m7") {
+    } else if (CPU == "cortex-m3" || CPU == "cortex-m4" || CPU == "cortex-m7" ||
+               CPU == "sc300" || CPU == "cortex-r4" || CPU == "cortex-r4f") {
       Features["hwdiv"] = true;
     }
   }
@@ -4080,12 +4275,10 @@ public:
       Features.push_back("-neonfp");
 
     // Remove front-end specific options which the backend handles differently.
-    const StringRef FrontEndFeatures[] = { "+soft-float", "+soft-float-abi" };
-    for (const auto &FEFeature : FrontEndFeatures) {
-      auto Feature = std::find(Features.begin(), Features.end(), FEFeature);
-      if (Feature != Features.end())
-        Features.erase(Feature);
-    }
+    auto Feature =
+        std::find(Features.begin(), Features.end(), "+soft-float-abi");
+    if (Feature != Features.end())
+      Features.erase(Feature);
 
     return true;
   }
@@ -4101,7 +4294,18 @@ public:
         .Default(false);
   }
   // FIXME: Should we actually have some table instead of these switches?
-  static const char *getCPUDefineSuffix(StringRef Name) {
+  const char *getCPUDefineSuffix(StringRef Name) const {
+    // FIXME: Use ARMTargetParser
+    if(Name == "generic") {
+      auto subarch = getTriple().getSubArch();
+      switch (subarch) {
+        case llvm::Triple::SubArchType::ARMSubArch_v8_1a: 
+          return "8_1A";
+        default:
+          break;
+      }
+    }
+
     return llvm::StringSwitch<const char *>(Name)
         .Cases("arm8", "arm810", "4")
         .Cases("strongarm", "strongarm110", "strongarm1100", "strongarm1110",
@@ -4115,30 +4319,48 @@ public:
         .Cases("arm10e", "arm1020e", "arm1022e", "5TE")
         .Cases("xscale", "iwmmxt", "5TE")
         .Case("arm1136j-s", "6J")
-        .Cases("arm1176jz-s", "arm1176jzf-s", "6ZK")
-        .Cases("arm1136jf-s", "mpcorenovfp", "mpcore", "6K")
+        .Case("arm1136jf-s", "6")
+        .Cases("mpcorenovfp", "mpcore", "6K")
+        .Cases("arm1176jz-s", "arm1176jzf-s", "6K")
         .Cases("arm1156t2-s", "arm1156t2f-s", "6T2")
         .Cases("cortex-a5", "cortex-a7", "cortex-a8", "7A")
         .Cases("cortex-a9", "cortex-a12", "cortex-a15", "cortex-a17", "krait",
                "7A")
-        .Cases("cortex-r4", "cortex-r5", "7R")
+        .Cases("cortex-r4", "cortex-r4f", "cortex-r5", "cortex-r7", "7R")
         .Case("swift", "7S")
         .Case("cyclone", "8A")
-        .Case("cortex-m3", "7M")
+        .Cases("sc300", "cortex-m3", "7M")
         .Cases("cortex-m4", "cortex-m7", "7EM")
-        .Case("cortex-m0", "6M")
-        .Cases("cortex-a53", "cortex-a57", "8A")
+        .Cases("sc000", "cortex-m0", "cortex-m0plus", "cortex-m1", "6M")
+        .Cases("cortex-a53", "cortex-a57", "cortex-a72", "8A")
         .Default(nullptr);
   }
-  static const char *getCPUProfile(StringRef Name) {
-    return llvm::StringSwitch<const char *>(Name)
-        .Cases("cortex-a5", "cortex-a7", "cortex-a8", "A")
-        .Cases("cortex-a9", "cortex-a12", "cortex-a15", "cortex-a17", "krait",
-               "A")
-        .Cases("cortex-a53", "cortex-a57", "A")
-        .Cases("cortex-m3", "cortex-m4", "cortex-m0", "cortex-m7", "M")
-        .Cases("cortex-r4", "cortex-r5", "R")
-        .Default("");
+  const char *getCPUProfile(StringRef Name) const {
+    if(Name == "generic") {
+      auto subarch = getTriple().getSubArch();
+      switch (subarch) {
+        case llvm::Triple::SubArchType::ARMSubArch_v8_1a: 
+          return "A";
+        default:
+          break;
+      }
+    }
+
+    unsigned CPUArch = llvm::ARMTargetParser::parseCPUArch(Name);
+    if (CPUArch == llvm::ARM::AK_INVALID)
+      return "";
+
+    StringRef ArchName = llvm::ARMTargetParser::getArchName(CPUArch);
+    switch(llvm::ARMTargetParser::parseArchProfile(ArchName)) {
+      case llvm::ARM::PK_A:
+        return "A";
+      case llvm::ARM::PK_R:
+        return "R";
+      case llvm::ARM::PK_M:
+        return "M";
+      default:
+        return "";
+    }
   }
   bool setCPU(const std::string &Name) override {
     if (!getCPUDefineSuffix(Name))
@@ -4165,6 +4387,7 @@ public:
     // We check both CPUArchVer and ArchName because when only triple is
     // specified, the default CPU is arm1136j-s.
     return ArchName.endswith("v6t2") || ArchName.endswith("v7") ||
+           ArchName.endswith("v8.1a") ||
            ArchName.endswith("v8") || CPUArch == "6T2" || CPUArchVer >= 7;
   }
   void getTargetDefines(const LangOptions &Opts,
@@ -4499,8 +4722,8 @@ class ARMleTargetInfo : public ARMTargetInfo {
 public:
   ARMleTargetInfo(const llvm::Triple &Triple)
     : ARMTargetInfo(Triple, false) { }
-  virtual void getTargetDefines(const LangOptions &Opts,
-                                MacroBuilder &Builder) const {
+  void getTargetDefines(const LangOptions &Opts,
+                        MacroBuilder &Builder) const override {
     Builder.defineMacro("__ARMEL__");
     ARMTargetInfo::getTargetDefines(Opts, Builder);
   }
@@ -4510,16 +4733,14 @@ class ARMbeTargetInfo : public ARMTargetInfo {
 public:
   ARMbeTargetInfo(const llvm::Triple &Triple)
     : ARMTargetInfo(Triple, true) { }
-  virtual void getTargetDefines(const LangOptions &Opts,
-                                MacroBuilder &Builder) const {
+  void getTargetDefines(const LangOptions &Opts,
+                        MacroBuilder &Builder) const override {
     Builder.defineMacro("__ARMEB__");
     Builder.defineMacro("__ARM_BIG_ENDIAN");
     ARMTargetInfo::getTargetDefines(Opts, Builder);
   }
 };
-} // end anonymous namespace.
 
-namespace {
 class WindowsARMTargetInfo : public WindowsTargetInfo<ARMleTargetInfo> {
   const llvm::Triple Triple;
 public:
@@ -4585,10 +4806,7 @@ public:
     WindowsARMTargetInfo::getVisualStudioDefines(Opts, Builder);
   }
 };
-}
-
 
-namespace {
 class DarwinARMTargetInfo :
   public DarwinTargetInfo<ARMleTargetInfo> {
 protected:
@@ -4610,10 +4828,7 @@ public:
     TheCXXABI.set(TargetCXXABI::iOS);
   }
 };
-} // end anonymous namespace.
-
 
-namespace {
 class AArch64TargetInfo : public TargetInfo {
   virtual void setDescriptionString() = 0;
   static const TargetInfo::GCCRegAlias GCCRegAliases[];
@@ -4655,13 +4870,20 @@ public:
     MaxAtomicInlineWidth = 128;
     MaxAtomicPromoteWidth = 128;
 
-    LongDoubleWidth = LongDoubleAlign = 128;
+    LongDoubleWidth = LongDoubleAlign = SuitableAlign = 128;
     LongDoubleFormat = &llvm::APFloat::IEEEquad;
 
     // {} in inline assembly are neon specifiers, not assembly variant
     // specifiers.
     NoAsmVariants = true;
 
+    // AAPCS gives rules for bitfields. 7.1.7 says: "The container type
+    // contributes to the alignment of the containing aggregate in the same way
+    // a plain (non bit-field) member of that type would, without exception for
+    // zero-sized or anonymous bit-fields."
+    UseBitFieldTypeAlignment = true;
+    UseZeroLengthBitfieldAlignment = true;
+
     // AArch64 targets default to using the ARM C++ ABI.
     TheCXXABI.set(TargetCXXABI::GenericAArch64);
   }
@@ -4678,14 +4900,14 @@ public:
   bool setCPU(const std::string &Name) override {
     bool CPUKnown = llvm::StringSwitch<bool>(Name)
                         .Case("generic", true)
-                        .Cases("cortex-a53", "cortex-a57", true)
+                        .Cases("cortex-a53", "cortex-a57", "cortex-a72", true)
                         .Case("cyclone", true)
                         .Default(false);
     return CPUKnown;
   }
 
-  virtual void getTargetDefines(const LangOptions &Opts,
-                                MacroBuilder &Builder) const  override {
+  void getTargetDefines(const LangOptions &Opts,
+                        MacroBuilder &Builder) const override {
     // Target identification.
     Builder.defineMacro("__aarch64__");
 
@@ -4742,10 +4964,16 @@ public:
 
     if (Crypto)
       Builder.defineMacro("__ARM_FEATURE_CRYPTO");
+
+    // All of the __sync_(bool|val)_compare_and_swap_(1|2|4|8) builtins work.
+    Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_1");
+    Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_2");
+    Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4");
+    Builder.defineMacro("__GCC_HAVE_SYNC_COMPARE_AND_SWAP_8");
   }
 
-  virtual void getTargetBuiltins(const Builtin::Info *&Records,
-                                 unsigned &NumRecords) const override {
+  void getTargetBuiltins(const Builtin::Info *&Records,
+                         unsigned &NumRecords) const override {
     Records = BuiltinInfo;
     NumRecords = clang::AArch64::LastTSBuiltin - Builtin::FirstTSBuiltin;
   }
@@ -4781,14 +5009,13 @@ public:
     return TargetInfo::AArch64ABIBuiltinVaList;
   }
 
-  virtual void getGCCRegNames(const char *const *&Names,
-                              unsigned &NumNames) const override;
-  virtual void getGCCRegAliases(const GCCRegAlias *&Aliases,
-                                unsigned &NumAliases) const override;
+  void getGCCRegNames(const char *const *&Names,
+                      unsigned &NumNames) const override;
+  void getGCCRegAliases(const GCCRegAlias *&Aliases,
+                        unsigned &NumAliases) const override;
 
-  virtual bool
-  validateAsmConstraint(const char *&Name,
-                        TargetInfo::ConstraintInfo &Info) const override {
+  bool validateAsmConstraint(const char *&Name,
+                             TargetInfo::ConstraintInfo &Info) const override {
     switch (*Name) {
     default:
       return false;
@@ -4963,9 +5190,7 @@ public:
     AArch64TargetInfo::getTargetDefines(Opts, Builder);
   }
 };
-} // end anonymous namespace.
 
-namespace {
 class DarwinAArch64TargetInfo : public DarwinTargetInfo<AArch64leTargetInfo> {
 protected:
   void getOSDefines(const LangOptions &Opts, const llvm::Triple &Triple,
@@ -4988,7 +5213,7 @@ public:
     WCharType = SignedInt;
     UseSignedCharForObjCBool = false;
 
-    LongDoubleWidth = LongDoubleAlign = 64;
+    LongDoubleWidth = LongDoubleAlign = SuitableAlign = 64;
     LongDoubleFormat = &llvm::APFloat::IEEEdouble;
 
     TheCXXABI.set(TargetCXXABI::iOS64);
@@ -4998,9 +5223,7 @@ public:
     return TargetInfo::CharPtrBuiltinVaList;
   }
 };
-} // end anonymous namespace
 
-namespace {
 // Hexagon abstract base class
 class HexagonTargetInfo : public TargetInfo {
   static const Builtin::Info BuiltinInfo[];
@@ -5149,24 +5372,25 @@ const Builtin::Info HexagonTargetInfo::BuiltinInfo[] = {
                                               ALL_LANGUAGES },
 #include "clang/Basic/BuiltinsHexagon.def"
 };
-}
-
 
-namespace {
 // Shared base class for SPARC v8 (32-bit) and SPARC v9 (64-bit).
 class SparcTargetInfo : public TargetInfo {
   static const TargetInfo::GCCRegAlias GCCRegAliases[];
   static const char * const GCCRegNames[];
   bool SoftFloat;
 public:
-  SparcTargetInfo(const llvm::Triple &Triple) : TargetInfo(Triple) {}
+  SparcTargetInfo(const llvm::Triple &Triple)
+      : TargetInfo(Triple), SoftFloat(false) {}
 
   bool handleTargetFeatures(std::vector<std::string> &Features,
                             DiagnosticsEngine &Diags) override {
-    SoftFloat = false;
-    for (unsigned i = 0, e = Features.size(); i != e; ++i)
-      if (Features[i] == "+soft-float")
-        SoftFloat = true;
+    // The backend doesn't actually handle soft float yet, but in case someone
+    // is using the support for the front end continue to support it.
+    auto Feature = std::find(Features.begin(), Features.end(), "+soft-float");
+    if (Feature != Features.end()) {
+      SoftFloat = true;
+      Features.erase(Feature);
+    }
     return true;
   }
   void getTargetDefines(const LangOptions &Opts,
@@ -5285,6 +5509,15 @@ public:
   }
 };
 
+// SPARCV8el is the 32-bit little-endian mode selected by Triple::sparcel.
+class SparcV8elTargetInfo : public SparcV8TargetInfo {
+ public:
+  SparcV8elTargetInfo(const llvm::Triple &Triple) : SparcV8TargetInfo(Triple) {
+    DescriptionString = "e-m:e-p:32:32-i64:64-f128:64-n32-S64";
+    BigEndian = false;
+  }
+};
+
 // SPARC v9 is the 64-bit mode selected by Triple::sparcv9.
 class SparcV9TargetInfo : public SparcTargetInfo {
 public:
@@ -5339,9 +5572,6 @@ public:
   }
 };
 
-} // end anonymous namespace.
-
-namespace {
 class SolarisSparcV8TargetInfo : public SolarisTargetInfo<SparcV8TargetInfo> {
 public:
   SolarisSparcV8TargetInfo(const llvm::Triple &Triple)
@@ -5350,14 +5580,19 @@ public:
     PtrDiffType = SignedInt;
   }
 };
-} // end anonymous namespace.
 
-namespace {
 class SystemZTargetInfo : public TargetInfo {
+  static const Builtin::Info BuiltinInfo[];
   static const char *const GCCRegNames[];
+  std::string CPU;
+  bool HasTransactionalExecution;
+  bool HasVector;
 
 public:
-  SystemZTargetInfo(const llvm::Triple &Triple) : TargetInfo(Triple) {
+  SystemZTargetInfo(const llvm::Triple &Triple)
+    : TargetInfo(Triple), CPU("z10"), HasTransactionalExecution(false), HasVector(false) {
+    IntMaxType = SignedLong;
+    Int64Type = SignedLong;
     TLSSupported = true;
     IntWidth = IntAlign = 32;
     LongWidth = LongLongWidth = LongAlign = LongLongAlign = 64;
@@ -5365,6 +5600,7 @@ public:
     LongDoubleWidth = 128;
     LongDoubleAlign = 64;
     LongDoubleFormat = &llvm::APFloat::IEEEquad;
+    DefaultAlignForAttributeAligned = 64;
     MinGlobalAlign = 16;
     DescriptionString = "E-m:e-i1:8:16-i8:8:16-i64:64-f128:64-a:8:16-n32:64";
     MaxAtomicPromoteWidth = MaxAtomicInlineWidth = 64;
@@ -5375,12 +5611,13 @@ public:
     Builder.defineMacro("__s390x__");
     Builder.defineMacro("__zarch__");
     Builder.defineMacro("__LONG_DOUBLE_128__");
+    if (HasTransactionalExecution)
+      Builder.defineMacro("__HTM__");
   }
   void getTargetBuiltins(const Builtin::Info *&Records,
                          unsigned &NumRecords) const override {
-    // FIXME: Implement.
-    Records = nullptr;
-    NumRecords = 0;
+    Records = BuiltinInfo;
+    NumRecords = clang::SystemZ::LastTSBuiltin-Builtin::FirstTSBuiltin;
   }
 
   void getGCCRegNames(const char *const *&Names,
@@ -5401,16 +5638,62 @@ public:
     return TargetInfo::SystemZBuiltinVaList;
   }
   bool setCPU(const std::string &Name) override {
+    CPU = Name;
     bool CPUKnown = llvm::StringSwitch<bool>(Name)
       .Case("z10", true)
       .Case("z196", true)
       .Case("zEC12", true)
+      .Case("z13", true)
       .Default(false);
 
-    // No need to store the CPU yet.  There aren't any CPU-specific
-    // macros to define.
     return CPUKnown;
   }
+  void getDefaultFeatures(llvm::StringMap<bool> &Features) const override {
+    if (CPU == "zEC12")
+      Features["transactional-execution"] = true;
+    if (CPU == "z13") {
+      Features["transactional-execution"] = true;
+      Features["vector"] = true;
+    }
+  }
+
+  bool handleTargetFeatures(std::vector<std::string> &Features,
+                            DiagnosticsEngine &Diags) override {
+    HasTransactionalExecution = false;
+    for (unsigned i = 0, e = Features.size(); i != e; ++i) {
+      if (Features[i] == "+transactional-execution")
+        HasTransactionalExecution = true;
+      if (Features[i] == "+vector")
+        HasVector = true;
+    }
+    // If we use the vector ABI, vector types are 64-bit aligned.
+    if (HasVector) {
+      MaxVectorAlign = 64;
+      DescriptionString = "E-m:e-i1:8:16-i8:8:16-i64:64-f128:64"
+                          "-v128:64-a:8:16-n32:64";
+    }
+    return true;
+  }
+
+  bool hasFeature(StringRef Feature) const override {
+    return llvm::StringSwitch<bool>(Feature)
+        .Case("systemz", true)
+        .Case("htm", HasTransactionalExecution)
+        .Case("vx", HasVector)
+        .Default(false);
+  }
+
+  StringRef getABI() const override {
+    if (HasVector)
+      return "vector";
+    return "";
+  }
+};
+
+const Builtin::Info SystemZTargetInfo::BuiltinInfo[] = {
+#define BUILTIN(ID, TYPE, ATTRS)                                               \
+  { #ID, TYPE, ATTRS, 0, ALL_LANGUAGES },
+#include "clang/Basic/BuiltinsSystemZ.def"
 };
 
 const char *const SystemZTargetInfo::GCCRegNames[] = {
@@ -5454,9 +5737,7 @@ validateAsmConstraint(const char *&Name,
     return true;
   }
 }
-}
 
-namespace {
   class MSP430TargetInfo : public TargetInfo {
     static const char * const GCCRegNames[];
   public:
@@ -5531,9 +5812,6 @@ namespace {
     Names = GCCRegNames;
     NumNames = llvm::array_lengthof(GCCRegNames);
   }
-}
-
-namespace {
 
   // LLVM and Clang cannot be used directly to output native binaries for
   // target, but is used to compile C code to llvm bitcode with correct
@@ -5611,9 +5889,7 @@ namespace {
     void getGCCRegAliases(const GCCRegAlias *&Aliases,
                           unsigned &NumAliases) const override {}
   };
-}
 
-namespace {
 class MipsTargetInfoBase : public TargetInfo {
   virtual void setDescriptionString() = 0;
 
@@ -5652,6 +5928,10 @@ public:
     return CPU == "mips32r6" || ABI == "n32" || ABI == "n64" || ABI == "64";
   }
 
+  bool isNan2008() const override {
+    return IsNan2008;
+  }
+
   StringRef getABI() const override { return ABI; }
   bool setCPU(const std::string &Name) override {
     bool IsMips32 = getTriple().getArch() == llvm::Triple::mips ||
@@ -5665,23 +5945,19 @@ public:
         .Case("mips5", true)
         .Case("mips32", IsMips32)
         .Case("mips32r2", IsMips32)
+        .Case("mips32r3", IsMips32)
+        .Case("mips32r5", IsMips32)
         .Case("mips32r6", IsMips32)
         .Case("mips64", true)
         .Case("mips64r2", true)
+        .Case("mips64r3", true)
+        .Case("mips64r5", true)
         .Case("mips64r6", true)
         .Case("octeon", true)
         .Default(false);
   }
   const std::string& getCPU() const { return CPU; }
   void getDefaultFeatures(llvm::StringMap<bool> &Features) const override {
-    // The backend enables certain ABI's by default according to the
-    // architecture.
-    // Disable both possible defaults so that we don't end up with multiple
-    // ABI's selected and trigger an assertion.
-    Features["o32"] = false;
-    Features["n64"] = false;
-
-    Features[ABI] = true;
     if (CPU == "octeon")
       Features["mips64r2"] = Features["cnmips"] = true;
     else
@@ -5818,7 +6094,28 @@ public:
     case 'R': // An address that can be used in a non-macro load or store
       Info.setAllowsMemory();
       return true;
+    case 'Z':
+      if (Name[1] == 'C') { // An address usable by ll, and sc.
+        Info.setAllowsMemory();
+        Name++; // Skip over 'Z'.
+        return true;
+      }
+      return false;
+    }
+  }
+
+  std::string convertConstraint(const char *&Constraint) const override {
+    std::string R;
+    switch (*Constraint) {
+    case 'Z': // Two-character constraint; add "^" hint for later parsing.
+      if (Constraint[1] == 'C') {
+        R = std::string("^") + std::string(Constraint, 2);
+        Constraint++;
+        return R;
+      }
+      break;
     }
+    return TargetInfo::convertConstraint(Constraint);
   }
 
   const char *getClobbers() const override {
@@ -5882,12 +6179,6 @@ public:
         IsNan2008 = false;
     }
 
-    // Remove front-end specific options.
-    std::vector<std::string>::iterator it =
-      std::find(Features.begin(), Features.end(), "+soft-float");
-    if (it != Features.end())
-      Features.erase(it);
-
     setDescriptionString();
 
     return true;
@@ -5938,6 +6229,10 @@ public:
       Builder.defineMacro("__mips_isa_rev", "1");
     else if (CPUStr == "mips32r2")
       Builder.defineMacro("__mips_isa_rev", "2");
+    else if (CPUStr == "mips32r3")
+      Builder.defineMacro("__mips_isa_rev", "3");
+    else if (CPUStr == "mips32r5")
+      Builder.defineMacro("__mips_isa_rev", "5");
     else if (CPUStr == "mips32r6")
       Builder.defineMacro("__mips_isa_rev", "6");
 
@@ -6087,6 +6382,10 @@ public:
       Builder.defineMacro("__mips_isa_rev", "1");
     else if (CPUStr == "mips64r2")
       Builder.defineMacro("__mips_isa_rev", "2");
+    else if (CPUStr == "mips64r3")
+      Builder.defineMacro("__mips_isa_rev", "3");
+    else if (CPUStr == "mips64r5")
+      Builder.defineMacro("__mips_isa_rev", "5");
     else if (CPUStr == "mips64r6")
       Builder.defineMacro("__mips_isa_rev", "6");
 
@@ -6185,9 +6484,7 @@ public:
     Mips64TargetInfoBase::getTargetDefines(Opts, Builder);
   }
 };
-} // end anonymous namespace.
 
-namespace {
 class PNaClTargetInfo : public TargetInfo {
 public:
   PNaClTargetInfo(const llvm::Triple &Triple) : TargetInfo(Triple) {
@@ -6252,9 +6549,7 @@ void PNaClTargetInfo::getGCCRegAliases(const GCCRegAlias *&Aliases,
   Aliases = nullptr;
   NumAliases = 0;
 }
-} // end anonymous namespace.
 
-namespace {
 class Le64TargetInfo : public TargetInfo {
   static const Builtin::Info BuiltinInfo[];
 
@@ -6359,6 +6654,15 @@ namespace {
     BuiltinVaListKind getBuiltinVaListKind() const override {
       return TargetInfo::VoidPtrBuiltinVaList;
     }
+
+    CallingConvCheckResult checkCallingConvention(CallingConv CC) const override {
+      return (CC == CC_SpirFunction ||
+              CC == CC_SpirKernel) ? CCCR_OK : CCCR_Warning;
+    }
+
+    CallingConv getDefaultCallingConv(CallingConvMethodType MT) const override {
+      return CC_SpirFunction;
+    }
   };
 
 
@@ -6392,9 +6696,7 @@ namespace {
       DefineStd(Builder, "SPIR64", Opts);
     }
   };
-}
 
-namespace {
 class XCoreTargetInfo : public TargetInfo {
   static const Builtin::Info BuiltinInfo[];
 public:
@@ -6460,6 +6762,30 @@ const Builtin::Info XCoreTargetInfo::BuiltinInfo[] = {
 };
 } // end anonymous namespace.
 
+namespace {
+// x86_32 Android target
+class AndroidX86_32TargetInfo : public LinuxTargetInfo<X86_32TargetInfo> {
+public:
+  AndroidX86_32TargetInfo(const llvm::Triple &Triple)
+      : LinuxTargetInfo<X86_32TargetInfo>(Triple) {
+    SuitableAlign = 32;
+    LongDoubleWidth = 64;
+    LongDoubleFormat = &llvm::APFloat::IEEEdouble;
+  }
+};
+} // end anonymous namespace
+
+namespace {
+// x86_64 Android target
+class AndroidX86_64TargetInfo : public LinuxTargetInfo<X86_64TargetInfo> {
+public:
+  AndroidX86_64TargetInfo(const llvm::Triple &Triple)
+      : LinuxTargetInfo<X86_64TargetInfo>(Triple) {
+    LongDoubleFormat = &llvm::APFloat::IEEEquad;
+  }
+};
+} // end anonymous namespace
+
 
 //===----------------------------------------------------------------------===//
 // Driver code
@@ -6629,10 +6955,10 @@ static TargetInfo *AllocateTarget(const llvm::Triple &Triple) {
 
   case llvm::Triple::le32:
     switch (os) {
-      case llvm::Triple::NaCl:
-        return new NaClTargetInfo<PNaClTargetInfo>(Triple);
-      default:
-        return nullptr;
+    case llvm::Triple::NaCl:
+      return new NaClTargetInfo<PNaClTargetInfo>(Triple);
+    default:
+      return nullptr;
     }
 
   case llvm::Triple::le64:
@@ -6707,6 +7033,21 @@ static TargetInfo *AllocateTarget(const llvm::Triple &Triple) {
       return new SparcV8TargetInfo(Triple);
     }
 
+  // The 'sparcel' architecture copies all the above cases except for Solaris.
+  case llvm::Triple::sparcel:
+    switch (os) {
+    case llvm::Triple::Linux:
+      return new LinuxTargetInfo<SparcV8elTargetInfo>(Triple);
+    case llvm::Triple::NetBSD:
+      return new NetBSDTargetInfo<SparcV8elTargetInfo>(Triple);
+    case llvm::Triple::OpenBSD:
+      return new OpenBSDTargetInfo<SparcV8elTargetInfo>(Triple);
+    case llvm::Triple::RTEMS:
+      return new RTEMSTargetInfo<SparcV8elTargetInfo>(Triple);
+    default:
+      return new SparcV8elTargetInfo(Triple);
+    }
+
   case llvm::Triple::sparcv9:
     switch (os) {
     case llvm::Triple::Linux:
@@ -6739,8 +7080,14 @@ static TargetInfo *AllocateTarget(const llvm::Triple &Triple) {
       return new DarwinI386TargetInfo(Triple);
 
     switch (os) {
-    case llvm::Triple::Linux:
-      return new LinuxTargetInfo<X86_32TargetInfo>(Triple);
+    case llvm::Triple::Linux: {
+      switch (Triple.getEnvironment()) {
+      default:
+        return new LinuxTargetInfo<X86_32TargetInfo>(Triple);
+      case llvm::Triple::Android:
+        return new AndroidX86_32TargetInfo(Triple);
+      }
+    }
     case llvm::Triple::DragonFly:
       return new DragonFlyBSDTargetInfo<X86_32TargetInfo>(Triple);
     case llvm::Triple::NetBSD:
@@ -6785,8 +7132,16 @@ static TargetInfo *AllocateTarget(const llvm::Triple &Triple) {
       return new DarwinX86_64TargetInfo(Triple);
 
     switch (os) {
-    case llvm::Triple::Linux:
-      return new LinuxTargetInfo<X86_64TargetInfo>(Triple);
+    case llvm::Triple::CloudABI:
+      return new CloudABITargetInfo<X86_64TargetInfo>(Triple);
+    case llvm::Triple::Linux: {
+      switch (Triple.getEnvironment()) {
+      default:
+        return new LinuxTargetInfo<X86_64TargetInfo>(Triple);
+      case llvm::Triple::Android:
+        return new AndroidX86_64TargetInfo(Triple);
+      }
+    }
     case llvm::Triple::DragonFly:
       return new DragonFlyBSDTargetInfo<X86_64TargetInfo>(Triple);
     case llvm::Triple::NetBSD:
@@ -6813,22 +7168,24 @@ static TargetInfo *AllocateTarget(const llvm::Triple &Triple) {
     }
     case llvm::Triple::NaCl:
       return new NaClTargetInfo<X86_64TargetInfo>(Triple);
+    case llvm::Triple::PS4:
+      return new PS4OSTargetInfo<X86_64TargetInfo>(Triple);
     default:
       return new X86_64TargetInfo(Triple);
     }
 
-    case llvm::Triple::spir: {
-      if (Triple.getOS() != llvm::Triple::UnknownOS ||
-          Triple.getEnvironment() != llvm::Triple::UnknownEnvironment)
-        return nullptr;
-      return new SPIR32TargetInfo(Triple);
-    }
-    case llvm::Triple::spir64: {
-      if (Triple.getOS() != llvm::Triple::UnknownOS ||
-          Triple.getEnvironment() != llvm::Triple::UnknownEnvironment)
-        return nullptr;
-      return new SPIR64TargetInfo(Triple);
-    }
+  case llvm::Triple::spir: {
+    if (Triple.getOS() != llvm::Triple::UnknownOS ||
+        Triple.getEnvironment() != llvm::Triple::UnknownEnvironment)
+      return nullptr;
+    return new SPIR32TargetInfo(Triple);
+  }
+  case llvm::Triple::spir64: {
+    if (Triple.getOS() != llvm::Triple::UnknownOS ||
+        Triple.getEnvironment() != llvm::Triple::UnknownEnvironment)
+      return nullptr;
+    return new SPIR64TargetInfo(Triple);
+  }
   }
 }
 
diff --git a/lib/Basic/Version.cpp b/lib/Basic/Version.cpp
index b631b3e8a763..a1a67c2bc144 100644
--- a/lib/Basic/Version.cpp
+++ b/lib/Basic/Version.cpp
@@ -36,7 +36,7 @@ std::string getClangRepositoryPath() {
 
   // If the SVN_REPOSITORY is empty, try to use the SVN keyword. This helps us
   // pick up a tag in an SVN export, for example.
-  StringRef SVNRepository("$URL: https://llvm.org/svn/llvm-project/cfe/tags/RELEASE_360/final/lib/Basic/Version.cpp $");
+  StringRef SVNRepository("$URL: https://llvm.org/svn/llvm-project/cfe/trunk/lib/Basic/Version.cpp $");
   if (URL.empty()) {
     URL = SVNRepository.slice(SVNRepository.find(':'),
                               SVNRepository.find("/lib/Basic"));
diff --git a/lib/Basic/VersionTuple.cpp b/lib/Basic/VersionTuple.cpp
index aa43ae298e23..9c73fd98a174 100644
--- a/lib/Basic/VersionTuple.cpp
+++ b/lib/Basic/VersionTuple.cpp
@@ -32,6 +32,8 @@ raw_ostream& clang::operator<<(raw_ostream &Out,
     Out << (V.usesUnderscores() ? '_' : '.') << *Minor;
   if (Optional<unsigned> Subminor = V.getSubminor())
     Out << (V.usesUnderscores() ? '_' : '.') << *Subminor;
+  if (Optional<unsigned> Build = V.getBuild())
+    Out << (V.usesUnderscores() ? '_' : '.') << *Build;
   return Out;
 }
 
@@ -55,7 +57,7 @@ static bool parseInt(StringRef &input, unsigned &value) {
 }
 
 bool VersionTuple::tryParse(StringRef input) {
-  unsigned major = 0, minor = 0, micro = 0;
+  unsigned major = 0, minor = 0, micro = 0, build = 0;
 
   // Parse the major version, [0-9]+
   if (parseInt(input, major)) return true;
@@ -80,9 +82,19 @@ bool VersionTuple::tryParse(StringRef input) {
   input = input.substr(1);
   if (parseInt(input, micro)) return true;
 
+  if (input.empty()) {
+    *this = VersionTuple(major, minor, micro);
+    return false;
+  }
+
+  // If we're not done, parse the micro version, \.[0-9]+
+  if (input[0] != '.') return true;
+  input = input.substr(1);
+  if (parseInt(input, build)) return true;
+
   // If we have characters left over, it's an error.
   if (!input.empty()) return true;
 
-  *this = VersionTuple(major, minor, micro);
+  *this = VersionTuple(major, minor, micro, build);
   return false;
 }
diff --git a/lib/Basic/VirtualFileSystem.cpp b/lib/Basic/VirtualFileSystem.cpp
index c89195e9326c..8a882e13f7a5 100644
--- a/lib/Basic/VirtualFileSystem.cpp
+++ b/lib/Basic/VirtualFileSystem.cpp
@@ -92,7 +92,7 @@ class RealFile : public File {
   }
 
 public:
-  ~RealFile();
+  ~RealFile() override;
   ErrorOr<Status> status() override;
   ErrorOr<std::unique_ptr<MemoryBuffer>>
   getBuffer(const Twine &Name, int64_t FileSize = -1,
@@ -362,7 +362,7 @@ class DirectoryEntry : public Entry {
   Status S;
 
 public:
-  virtual ~DirectoryEntry();
+  ~DirectoryEntry() override;
   DirectoryEntry(StringRef Name, std::vector<Entry *> Contents, Status S)
       : Entry(EK_Directory, Name), Contents(std::move(Contents)),
         S(std::move(S)) {}
@@ -498,7 +498,7 @@ private:
   ErrorOr<Status> status(const Twine &Path, Entry *E);
 
 public:
-  ~VFSFromYAML();
+  ~VFSFromYAML() override;
 
   /// \brief Parses \p Buffer, which is expected to be in YAML format and
   /// returns a virtual file system representing its contents.
@@ -1134,7 +1134,7 @@ VFSFromYamlDirIterImpl::VFSFromYamlDirIterImpl(const Twine &_Path,
   if (Current != End) {
     SmallString<128> PathStr(Dir);
     llvm::sys::path::append(PathStr, (*Current)->getName());
-    llvm::ErrorOr<vfs::Status> S = FS.status(PathStr.str());
+    llvm::ErrorOr<vfs::Status> S = FS.status(PathStr);
     if (S)
       CurrentEntry = *S;
     else
@@ -1147,7 +1147,7 @@ std::error_code VFSFromYamlDirIterImpl::increment() {
   if (++Current != End) {
     SmallString<128> PathStr(Dir);
     llvm::sys::path::append(PathStr, (*Current)->getName());
-    llvm::ErrorOr<vfs::Status> S = FS.status(PathStr.str());
+    llvm::ErrorOr<vfs::Status> S = FS.status(PathStr);
     if (!S)
       return S.getError();
     CurrentEntry = *S;
diff --git a/lib/CodeGen/ABIInfo.h b/lib/CodeGen/ABIInfo.h
index 7e7f7fa20679..cc8652e169d8 100644
--- a/lib/CodeGen/ABIInfo.h
+++ b/lib/CodeGen/ABIInfo.h
@@ -87,6 +87,8 @@ namespace clang {
     virtual bool isHomogeneousAggregateSmallEnough(const Type *Base,
                                                    uint64_t Members) const;
 
+    virtual bool shouldSignExtUnsignedType(QualType Ty) const;
+
     bool isHomogeneousAggregate(QualType Ty, const Type *&Base,
                                 uint64_t &Members) const;
 
diff --git a/lib/CodeGen/BackendUtil.cpp b/lib/CodeGen/BackendUtil.cpp
index 25ecec586244..7f0c7bafc046 100644
--- a/lib/CodeGen/BackendUtil.cpp
+++ b/lib/CodeGen/BackendUtil.cpp
@@ -15,22 +15,22 @@
 #include "clang/Frontend/FrontendDiagnostic.h"
 #include "clang/Frontend/Utils.h"
 #include "llvm/ADT/StringSwitch.h"
+#include "llvm/Analysis/TargetLibraryInfo.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/Bitcode/BitcodeWriterPass.h"
 #include "llvm/CodeGen/RegAllocRegistry.h"
 #include "llvm/CodeGen/SchedulerRegistry.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/IRPrintingPasses.h"
+#include "llvm/IR/LegacyPassManager.h"
 #include "llvm/IR/Module.h"
 #include "llvm/IR/Verifier.h"
 #include "llvm/MC/SubtargetFeature.h"
-#include "llvm/PassManager.h"
 #include "llvm/Support/CommandLine.h"
-#include "llvm/Support/FormattedStream.h"
 #include "llvm/Support/PrettyStackTrace.h"
 #include "llvm/Support/TargetRegistry.h"
 #include "llvm/Support/Timer.h"
 #include "llvm/Support/raw_ostream.h"
-#include "llvm/Target/TargetLibraryInfo.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetOptions.h"
 #include "llvm/Target/TargetSubtargetInfo.h"
@@ -55,44 +55,48 @@ class EmitAssemblyHelper {
 
   Timer CodeGenerationTime;
 
-  mutable PassManager *CodeGenPasses;
-  mutable PassManager *PerModulePasses;
-  mutable FunctionPassManager *PerFunctionPasses;
+  mutable legacy::PassManager *CodeGenPasses;
+  mutable legacy::PassManager *PerModulePasses;
+  mutable legacy::FunctionPassManager *PerFunctionPasses;
 
 private:
-  PassManager *getCodeGenPasses() const {
+  TargetIRAnalysis getTargetIRAnalysis() const {
+    if (TM)
+      return TM->getTargetIRAnalysis();
+
+    return TargetIRAnalysis();
+  }
+
+  legacy::PassManager *getCodeGenPasses() const {
     if (!CodeGenPasses) {
-      CodeGenPasses = new PassManager();
-      CodeGenPasses->add(new DataLayoutPass());
-      if (TM)
-        TM->addAnalysisPasses(*CodeGenPasses);
+      CodeGenPasses = new legacy::PassManager();
+      CodeGenPasses->add(
+          createTargetTransformInfoWrapperPass(getTargetIRAnalysis()));
     }
     return CodeGenPasses;
   }
 
-  PassManager *getPerModulePasses() const {
+  legacy::PassManager *getPerModulePasses() const {
     if (!PerModulePasses) {
-      PerModulePasses = new PassManager();
-      PerModulePasses->add(new DataLayoutPass());
-      if (TM)
-        TM->addAnalysisPasses(*PerModulePasses);
+      PerModulePasses = new legacy::PassManager();
+      PerModulePasses->add(
+          createTargetTransformInfoWrapperPass(getTargetIRAnalysis()));
     }
     return PerModulePasses;
   }
 
-  FunctionPassManager *getPerFunctionPasses() const {
+  legacy::FunctionPassManager *getPerFunctionPasses() const {
     if (!PerFunctionPasses) {
-      PerFunctionPasses = new FunctionPassManager(TheModule);
-      PerFunctionPasses->add(new DataLayoutPass());
-      if (TM)
-        TM->addAnalysisPasses(*PerFunctionPasses);
+      PerFunctionPasses = new legacy::FunctionPassManager(TheModule);
+      PerFunctionPasses->add(
+          createTargetTransformInfoWrapperPass(getTargetIRAnalysis()));
     }
     return PerFunctionPasses;
   }
 
   void CreatePasses();
 
-  /// CreateTargetMachine - Generates the TargetMachine.
+  /// Generates the TargetMachine.
   /// Returns Null if it is unable to create the target machine.
   /// Some of our clang tests specify triples which are not built
   /// into clang. This is okay because these tests check the generated
@@ -102,10 +106,10 @@ private:
   /// the requested target.
   TargetMachine *CreateTargetMachine(bool MustCreateTM);
 
-  /// AddEmitPasses - Add passes necessary to emit assembly or LLVM IR.
+  /// Add passes necessary to emit assembly or LLVM IR.
   ///
   /// \return True on success.
-  bool AddEmitPasses(BackendAction Action, formatted_raw_ostream &OS);
+  bool AddEmitPasses(BackendAction Action, raw_pwrite_stream &OS);
 
 public:
   EmitAssemblyHelper(DiagnosticsEngine &_Diags,
@@ -128,7 +132,7 @@ public:
 
   std::unique_ptr<TargetMachine> TM;
 
-  void EmitAssembly(BackendAction Action, raw_ostream *OS);
+  void EmitAssembly(BackendAction Action, raw_pwrite_stream *OS);
 };
 
 // We need this wrapper to access LangOpts and CGOpts from extension functions
@@ -163,7 +167,7 @@ static void addObjCARCOptPass(const PassManagerBuilder &Builder, PassManagerBase
 }
 
 static void addSampleProfileLoaderPass(const PassManagerBuilder &Builder,
-                                       PassManagerBase &PM) {
+                                       legacy::PassManagerBase &PM) {
   const PassManagerBuilderWrapper &BuilderWrapper =
       static_cast<const PassManagerBuilderWrapper &>(Builder);
   const CodeGenOptions &CGOpts = BuilderWrapper.getCGOpts();
@@ -171,31 +175,38 @@ static void addSampleProfileLoaderPass(const PassManagerBuilder &Builder,
 }
 
 static void addAddDiscriminatorsPass(const PassManagerBuilder &Builder,
-                                     PassManagerBase &PM) {
+                                     legacy::PassManagerBase &PM) {
   PM.add(createAddDiscriminatorsPass());
 }
 
 static void addBoundsCheckingPass(const PassManagerBuilder &Builder,
-                                    PassManagerBase &PM) {
+                                    legacy::PassManagerBase &PM) {
   PM.add(createBoundsCheckingPass());
 }
 
 static void addSanitizerCoveragePass(const PassManagerBuilder &Builder,
-                                     PassManagerBase &PM) {
+                                     legacy::PassManagerBase &PM) {
   const PassManagerBuilderWrapper &BuilderWrapper =
       static_cast<const PassManagerBuilderWrapper&>(Builder);
   const CodeGenOptions &CGOpts = BuilderWrapper.getCGOpts();
-  PM.add(createSanitizerCoverageModulePass(CGOpts.SanitizeCoverage));
+  SanitizerCoverageOptions Opts;
+  Opts.CoverageType =
+      static_cast<SanitizerCoverageOptions::Type>(CGOpts.SanitizeCoverageType);
+  Opts.IndirectCalls = CGOpts.SanitizeCoverageIndirectCalls;
+  Opts.TraceBB = CGOpts.SanitizeCoverageTraceBB;
+  Opts.TraceCmp = CGOpts.SanitizeCoverageTraceCmp;
+  Opts.Use8bitCounters = CGOpts.SanitizeCoverage8bitCounters;
+  PM.add(createSanitizerCoverageModulePass(Opts));
 }
 
 static void addAddressSanitizerPasses(const PassManagerBuilder &Builder,
-                                      PassManagerBase &PM) {
+                                      legacy::PassManagerBase &PM) {
   PM.add(createAddressSanitizerFunctionPass());
   PM.add(createAddressSanitizerModulePass());
 }
 
 static void addMemorySanitizerPass(const PassManagerBuilder &Builder,
-                                   PassManagerBase &PM) {
+                                   legacy::PassManagerBase &PM) {
   const PassManagerBuilderWrapper &BuilderWrapper =
       static_cast<const PassManagerBuilderWrapper&>(Builder);
   const CodeGenOptions &CGOpts = BuilderWrapper.getCGOpts();
@@ -215,28 +226,36 @@ static void addMemorySanitizerPass(const PassManagerBuilder &Builder,
 }
 
 static void addThreadSanitizerPass(const PassManagerBuilder &Builder,
-                                   PassManagerBase &PM) {
+                                   legacy::PassManagerBase &PM) {
   PM.add(createThreadSanitizerPass());
 }
 
 static void addDataFlowSanitizerPass(const PassManagerBuilder &Builder,
-                                     PassManagerBase &PM) {
+                                     legacy::PassManagerBase &PM) {
   const PassManagerBuilderWrapper &BuilderWrapper =
       static_cast<const PassManagerBuilderWrapper&>(Builder);
   const LangOptions &LangOpts = BuilderWrapper.getLangOpts();
-  PM.add(createDataFlowSanitizerPass(LangOpts.SanitizerBlacklistFile));
+  PM.add(createDataFlowSanitizerPass(LangOpts.SanitizerBlacklistFiles));
 }
 
-static TargetLibraryInfo *createTLI(llvm::Triple &TargetTriple,
-                                    const CodeGenOptions &CodeGenOpts) {
-  TargetLibraryInfo *TLI = new TargetLibraryInfo(TargetTriple);
+static TargetLibraryInfoImpl *createTLII(llvm::Triple &TargetTriple,
+                                         const CodeGenOptions &CodeGenOpts) {
+  TargetLibraryInfoImpl *TLII = new TargetLibraryInfoImpl(TargetTriple);
   if (!CodeGenOpts.SimplifyLibCalls)
-    TLI->disableAllFunctions();
-  return TLI;
+    TLII->disableAllFunctions();
+
+  switch (CodeGenOpts.getVecLib()) {
+  case CodeGenOptions::Accelerate:
+    TLII->addVectorizableFunctionsFromVecLib(TargetLibraryInfoImpl::Accelerate);
+    break;
+  default:
+    break;
+  }
+  return TLII;
 }
 
 static void addSymbolRewriterPass(const CodeGenOptions &Opts,
-                                  PassManager *MPM) {
+                                  legacy::PassManager *MPM) {
   llvm::SymbolRewriter::RewriteDescriptorList DL;
 
   llvm::SymbolRewriter::RewriteMapParser MapParser;
@@ -294,7 +313,9 @@ void EmitAssemblyHelper::CreatePasses() {
                            addBoundsCheckingPass);
   }
 
-  if (CodeGenOpts.SanitizeCoverage) {
+  if (CodeGenOpts.SanitizeCoverageType ||
+      CodeGenOpts.SanitizeCoverageIndirectCalls ||
+      CodeGenOpts.SanitizeCoverageTraceCmp) {
     PMBuilder.addExtension(PassManagerBuilder::EP_OptimizerLast,
                            addSanitizerCoveragePass);
     PMBuilder.addExtension(PassManagerBuilder::EP_EnabledOnOptLevel0,
@@ -331,7 +352,7 @@ void EmitAssemblyHelper::CreatePasses() {
 
   // Figure out TargetLibraryInfo.
   Triple TargetTriple(TheModule->getTargetTriple());
-  PMBuilder.LibraryInfo = createTLI(TargetTriple, CodeGenOpts);
+  PMBuilder.LibraryInfo = createTLII(TargetTriple, CodeGenOpts);
 
   switch (Inlining) {
   case CodeGenOptions::NoInlining: break;
@@ -351,17 +372,15 @@ void EmitAssemblyHelper::CreatePasses() {
   }
 
   // Set up the per-function pass manager.
-  FunctionPassManager *FPM = getPerFunctionPasses();
+  legacy::FunctionPassManager *FPM = getPerFunctionPasses();
   if (CodeGenOpts.VerifyModule)
     FPM->add(createVerifierPass());
   PMBuilder.populateFunctionPassManager(*FPM);
 
   // Set up the per-module pass manager.
-  PassManager *MPM = getPerModulePasses();
+  legacy::PassManager *MPM = getPerModulePasses();
   if (!CodeGenOpts.RewriteMapFiles.empty())
     addSymbolRewriterPass(CodeGenOpts, MPM);
-  if (CodeGenOpts.VerifyModule)
-    MPM->add(createDebugInfoVerifierPass());
 
   if (!CodeGenOpts.DisableGCov &&
       (CodeGenOpts.EmitGcovArcs || CodeGenOpts.EmitGcovNotes)) {
@@ -375,6 +394,7 @@ void EmitAssemblyHelper::CreatePasses() {
     Options.NoRedZone = CodeGenOpts.DisableRedZone;
     Options.FunctionNamesInData =
         !CodeGenOpts.CoverageNoFunctionNamesInData;
+    Options.ExitBlockBeforeBody = CodeGenOpts.CoverageExitBlockBeforeBody;
     MPM->add(createGCOVProfilerPass(Options));
     if (CodeGenOpts.getDebugInfo() == CodeGenOptions::NoDebugInfo)
       MPM->add(createStripSymbolsPass(true));
@@ -383,6 +403,7 @@ void EmitAssemblyHelper::CreatePasses() {
   if (CodeGenOpts.ProfileInstrGenerate) {
     InstrProfOptions Options;
     Options.NoRedZone = CodeGenOpts.DisableRedZone;
+    Options.InstrProfileOutput = CodeGenOpts.InstrProfileOutput;
     MPM->add(createInstrProfilingPass(Options));
   }
 
@@ -425,14 +446,12 @@ TargetMachine *EmitAssemblyHelper::CreateTargetMachine(bool MustCreateTM) {
     BackendArgs.push_back("-time-passes");
   for (unsigned i = 0, e = CodeGenOpts.BackendOptions.size(); i != e; ++i)
     BackendArgs.push_back(CodeGenOpts.BackendOptions[i].c_str());
-  if (CodeGenOpts.NoGlobalMerge)
-    BackendArgs.push_back("-enable-global-merge=false");
   BackendArgs.push_back(nullptr);
   llvm::cl::ParseCommandLineOptions(BackendArgs.size() - 1,
                                     BackendArgs.data());
 
   std::string FeaturesStr;
-  if (TargetOpts.Features.size()) {
+  if (!TargetOpts.Features.empty()) {
     SubtargetFeatures Features;
     for (std::vector<std::string>::const_iterator
            it = TargetOpts.Features.begin(),
@@ -472,15 +491,6 @@ TargetMachine *EmitAssemblyHelper::CreateTargetMachine(bool MustCreateTM) {
   if (CodeGenOpts.CompressDebugSections)
     Options.CompressDebugSections = true;
 
-  // Set frame pointer elimination mode.
-  if (!CodeGenOpts.DisableFPElim) {
-    Options.NoFramePointerElim = false;
-  } else if (CodeGenOpts.OmitLeafFramePointer) {
-    Options.NoFramePointerElim = false;
-  } else {
-    Options.NoFramePointerElim = true;
-  }
-
   if (CodeGenOpts.UseInitArray)
     Options.UseInitArray = true;
 
@@ -512,13 +522,13 @@ TargetMachine *EmitAssemblyHelper::CreateTargetMachine(bool MustCreateTM) {
   Options.NoNaNsFPMath = CodeGenOpts.NoNaNsFPMath;
   Options.NoZerosInBSS = CodeGenOpts.NoZeroInitializedInBSS;
   Options.UnsafeFPMath = CodeGenOpts.UnsafeFPMath;
-  Options.UseSoftFloat = CodeGenOpts.SoftFloat;
   Options.StackAlignmentOverride = CodeGenOpts.StackAlignment;
   Options.DisableTailCalls = CodeGenOpts.DisableTailCalls;
   Options.TrapFuncName = CodeGenOpts.TrapFuncName;
   Options.PositionIndependentExecutable = LangOpts.PIELevel != 0;
   Options.FunctionSections = CodeGenOpts.FunctionSections;
   Options.DataSections = CodeGenOpts.DataSections;
+  Options.UniqueSectionNames = CodeGenOpts.UniqueSectionNames;
 
   Options.MCOptions.MCRelaxAll = CodeGenOpts.RelaxAll;
   Options.MCOptions.MCSaveTempLabels = CodeGenOpts.SaveTempLabels;
@@ -536,17 +546,16 @@ TargetMachine *EmitAssemblyHelper::CreateTargetMachine(bool MustCreateTM) {
 }
 
 bool EmitAssemblyHelper::AddEmitPasses(BackendAction Action,
-                                       formatted_raw_ostream &OS) {
+                                       raw_pwrite_stream &OS) {
 
   // Create the code generator passes.
-  PassManager *PM = getCodeGenPasses();
+  legacy::PassManager *PM = getCodeGenPasses();
 
   // Add LibraryInfo.
   llvm::Triple TargetTriple(TheModule->getTargetTriple());
-  PM->add(createTLI(TargetTriple, CodeGenOpts));
-
-  // Add Target specific analysis passes.
-  TM->addAnalysisPasses(*PM);
+  std::unique_ptr<TargetLibraryInfoImpl> TLII(
+      createTLII(TargetTriple, CodeGenOpts));
+  PM->add(new TargetLibraryInfoWrapperPass(*TLII));
 
   // Normal mode, emit a .s or .o file by running the code generator. Note,
   // this also adds codegenerator level optimization passes.
@@ -561,8 +570,7 @@ bool EmitAssemblyHelper::AddEmitPasses(BackendAction Action,
   // Add ObjC ARC final-cleanup optimizations. This is done as part of the
   // "codegen" passes so that it isn't run multiple times when there is
   // inlining happening.
-  if (LangOpts.ObjCAutoRefCount &&
-      CodeGenOpts.OptimizationLevel > 0)
+  if (CodeGenOpts.OptimizationLevel > 0)
     PM->add(createObjCARCContractPass());
 
   if (TM->addPassesToEmitFile(*PM, OS, CGFT,
@@ -574,9 +582,9 @@ bool EmitAssemblyHelper::AddEmitPasses(BackendAction Action,
   return true;
 }
 
-void EmitAssemblyHelper::EmitAssembly(BackendAction Action, raw_ostream *OS) {
+void EmitAssemblyHelper::EmitAssembly(BackendAction Action,
+                                      raw_pwrite_stream *OS) {
   TimeRegion Region(llvm::TimePassesIsEnabled ? &CodeGenerationTime : nullptr);
-  llvm::formatted_raw_ostream FormattedOS;
 
   bool UsesCodeGen = (Action != Backend_EmitNothing &&
                       Action != Backend_EmitBC &&
@@ -592,17 +600,17 @@ void EmitAssemblyHelper::EmitAssembly(BackendAction Action, raw_ostream *OS) {
     break;
 
   case Backend_EmitBC:
-    getPerModulePasses()->add(createBitcodeWriterPass(*OS));
+    getPerModulePasses()->add(
+        createBitcodeWriterPass(*OS, CodeGenOpts.EmitLLVMUseLists));
     break;
 
   case Backend_EmitLL:
-    FormattedOS.setStream(*OS, formatted_raw_ostream::PRESERVE_STREAM);
-    getPerModulePasses()->add(createPrintModulePass(FormattedOS));
+    getPerModulePasses()->add(
+        createPrintModulePass(*OS, "", CodeGenOpts.EmitLLVMUseLists));
     break;
 
   default:
-    FormattedOS.setStream(*OS, formatted_raw_ostream::PRESERVE_STREAM);
-    if (!AddEmitPasses(Action, FormattedOS))
+    if (!AddEmitPasses(Action, *OS))
       return;
   }
 
@@ -639,7 +647,7 @@ void clang::EmitBackendOutput(DiagnosticsEngine &Diags,
                               const clang::TargetOptions &TOpts,
                               const LangOptions &LOpts, StringRef TDesc,
                               Module *M, BackendAction Action,
-                              raw_ostream *OS) {
+                              raw_pwrite_stream *OS) {
   EmitAssemblyHelper AsmHelper(Diags, CGOpts, TOpts, LOpts, M);
 
   AsmHelper.EmitAssembly(Action, OS);
@@ -647,9 +655,8 @@ void clang::EmitBackendOutput(DiagnosticsEngine &Diags,
   // If an optional clang TargetInfo description string was passed in, use it to
   // verify the LLVM TargetMachine's DataLayout.
   if (AsmHelper.TM && !TDesc.empty()) {
-    std::string DLDesc = AsmHelper.TM->getSubtargetImpl()
-                             ->getDataLayout()
-                             ->getStringRepresentation();
+    std::string DLDesc =
+        AsmHelper.TM->getDataLayout()->getStringRepresentation();
     if (DLDesc != TDesc) {
       unsigned DiagID = Diags.getCustomDiagID(
           DiagnosticsEngine::Error, "backend data layout '%0' does not match "
diff --git a/lib/CodeGen/CGAtomic.cpp b/lib/CodeGen/CGAtomic.cpp
index daac174c8e0c..da82249fe114 100644
--- a/lib/CodeGen/CGAtomic.cpp
+++ b/lib/CodeGen/CGAtomic.cpp
@@ -13,6 +13,7 @@
 
 #include "CodeGenFunction.h"
 #include "CGCall.h"
+#include "CGRecordLayout.h"
 #include "CodeGenModule.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/CodeGen/CGFunctionInfo.h"
@@ -36,34 +37,94 @@ namespace {
     CharUnits LValueAlign;
     TypeEvaluationKind EvaluationKind;
     bool UseLibcall;
+    LValue LVal;
+    CGBitFieldInfo BFI;
   public:
-    AtomicInfo(CodeGenFunction &CGF, LValue &lvalue) : CGF(CGF) {
-      assert(lvalue.isSimple());
-
-      AtomicTy = lvalue.getType();
-      ValueTy = AtomicTy->castAs<AtomicType>()->getValueType();
-      EvaluationKind = CGF.getEvaluationKind(ValueTy);
-
+    AtomicInfo(CodeGenFunction &CGF, LValue &lvalue)
+        : CGF(CGF), AtomicSizeInBits(0), ValueSizeInBits(0),
+          EvaluationKind(TEK_Scalar), UseLibcall(true) {
+      assert(!lvalue.isGlobalReg());
       ASTContext &C = CGF.getContext();
-
-      uint64_t ValueAlignInBits;
-      uint64_t AtomicAlignInBits;
-      TypeInfo ValueTI = C.getTypeInfo(ValueTy);
-      ValueSizeInBits = ValueTI.Width;
-      ValueAlignInBits = ValueTI.Align;
-
-      TypeInfo AtomicTI = C.getTypeInfo(AtomicTy);
-      AtomicSizeInBits = AtomicTI.Width;
-      AtomicAlignInBits = AtomicTI.Align;
-
-      assert(ValueSizeInBits <= AtomicSizeInBits);
-      assert(ValueAlignInBits <= AtomicAlignInBits);
-
-      AtomicAlign = C.toCharUnitsFromBits(AtomicAlignInBits);
-      ValueAlign = C.toCharUnitsFromBits(ValueAlignInBits);
-      if (lvalue.getAlignment().isZero())
-        lvalue.setAlignment(AtomicAlign);
-
+      if (lvalue.isSimple()) {
+        AtomicTy = lvalue.getType();
+        if (auto *ATy = AtomicTy->getAs<AtomicType>())
+          ValueTy = ATy->getValueType();
+        else
+          ValueTy = AtomicTy;
+        EvaluationKind = CGF.getEvaluationKind(ValueTy);
+
+        uint64_t ValueAlignInBits;
+        uint64_t AtomicAlignInBits;
+        TypeInfo ValueTI = C.getTypeInfo(ValueTy);
+        ValueSizeInBits = ValueTI.Width;
+        ValueAlignInBits = ValueTI.Align;
+
+        TypeInfo AtomicTI = C.getTypeInfo(AtomicTy);
+        AtomicSizeInBits = AtomicTI.Width;
+        AtomicAlignInBits = AtomicTI.Align;
+
+        assert(ValueSizeInBits <= AtomicSizeInBits);
+        assert(ValueAlignInBits <= AtomicAlignInBits);
+
+        AtomicAlign = C.toCharUnitsFromBits(AtomicAlignInBits);
+        ValueAlign = C.toCharUnitsFromBits(ValueAlignInBits);
+        if (lvalue.getAlignment().isZero())
+          lvalue.setAlignment(AtomicAlign);
+
+        LVal = lvalue;
+      } else if (lvalue.isBitField()) {
+        ValueTy = lvalue.getType();
+        ValueSizeInBits = C.getTypeSize(ValueTy);
+        auto &OrigBFI = lvalue.getBitFieldInfo();
+        auto Offset = OrigBFI.Offset % C.toBits(lvalue.getAlignment());
+        AtomicSizeInBits = C.toBits(
+            C.toCharUnitsFromBits(Offset + OrigBFI.Size + C.getCharWidth() - 1)
+                .RoundUpToAlignment(lvalue.getAlignment()));
+        auto VoidPtrAddr = CGF.EmitCastToVoidPtr(lvalue.getBitFieldAddr());
+        auto OffsetInChars =
+            (C.toCharUnitsFromBits(OrigBFI.Offset) / lvalue.getAlignment()) *
+            lvalue.getAlignment();
+        VoidPtrAddr = CGF.Builder.CreateConstGEP1_64(
+            VoidPtrAddr, OffsetInChars.getQuantity());
+        auto Addr = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+            VoidPtrAddr,
+            CGF.Builder.getIntNTy(AtomicSizeInBits)->getPointerTo(),
+            "atomic_bitfield_base");
+        BFI = OrigBFI;
+        BFI.Offset = Offset;
+        BFI.StorageSize = AtomicSizeInBits;
+        LVal = LValue::MakeBitfield(Addr, BFI, lvalue.getType(),
+                                    lvalue.getAlignment());
+        LVal.setTBAAInfo(lvalue.getTBAAInfo());
+        AtomicTy = C.getIntTypeForBitwidth(AtomicSizeInBits, OrigBFI.IsSigned);
+        if (AtomicTy.isNull()) {
+          llvm::APInt Size(
+              /*numBits=*/32,
+              C.toCharUnitsFromBits(AtomicSizeInBits).getQuantity());
+          AtomicTy = C.getConstantArrayType(C.CharTy, Size, ArrayType::Normal,
+                                            /*IndexTypeQuals=*/0);
+        }
+        AtomicAlign = ValueAlign = lvalue.getAlignment();
+      } else if (lvalue.isVectorElt()) {
+        ValueTy = lvalue.getType()->getAs<VectorType>()->getElementType();
+        ValueSizeInBits = C.getTypeSize(ValueTy);
+        AtomicTy = lvalue.getType();
+        AtomicSizeInBits = C.getTypeSize(AtomicTy);
+        AtomicAlign = ValueAlign = lvalue.getAlignment();
+        LVal = lvalue;
+      } else {
+        assert(lvalue.isExtVectorElt());
+        ValueTy = lvalue.getType();
+        ValueSizeInBits = C.getTypeSize(ValueTy);
+        AtomicTy = ValueTy = CGF.getContext().getExtVectorType(
+            lvalue.getType(), lvalue.getExtVectorAddr()
+                                  ->getType()
+                                  ->getPointerElementType()
+                                  ->getVectorNumElements());
+        AtomicSizeInBits = C.getTypeSize(AtomicTy);
+        AtomicAlign = ValueAlign = lvalue.getAlignment();
+        LVal = lvalue;
+      }
       UseLibcall = !C.getTargetInfo().hasBuiltinAtomic(
           AtomicSizeInBits, C.toBits(lvalue.getAlignment()));
     }
@@ -76,6 +137,17 @@ namespace {
     uint64_t getValueSizeInBits() const { return ValueSizeInBits; }
     TypeEvaluationKind getEvaluationKind() const { return EvaluationKind; }
     bool shouldUseLibcall() const { return UseLibcall; }
+    const LValue &getAtomicLValue() const { return LVal; }
+    llvm::Value *getAtomicAddress() const {
+      if (LVal.isSimple())
+        return LVal.getAddress();
+      else if (LVal.isBitField())
+        return LVal.getBitFieldAddr();
+      else if (LVal.isVectorElt())
+        return LVal.getVectorAddr();
+      assert(LVal.isExtVectorElt());
+      return LVal.getExtVectorAddr();
+    }
 
     /// Is the atomic size larger than the underlying value type?
     ///
@@ -87,7 +159,7 @@ namespace {
       return (ValueSizeInBits != AtomicSizeInBits);
     }
 
-    bool emitMemSetZeroIfNecessary(LValue dest) const;
+    bool emitMemSetZeroIfNecessary() const;
 
     llvm::Value *getAtomicSizeValue() const {
       CharUnits size = CGF.getContext().toCharUnitsFromBits(AtomicSizeInBits);
@@ -99,37 +171,141 @@ namespace {
     llvm::Value *emitCastToAtomicIntPointer(llvm::Value *addr) const;
 
     /// Turn an atomic-layout object into an r-value.
-    RValue convertTempToRValue(llvm::Value *addr,
-                               AggValueSlot resultSlot,
-                               SourceLocation loc) const;
+    RValue convertTempToRValue(llvm::Value *addr, AggValueSlot resultSlot,
+                               SourceLocation loc, bool AsValue) const;
 
     /// \brief Converts a rvalue to integer value.
     llvm::Value *convertRValueToInt(RValue RVal) const;
 
-    RValue convertIntToValue(llvm::Value *IntVal, AggValueSlot ResultSlot,
-                             SourceLocation Loc) const;
+    RValue ConvertIntToValueOrAtomic(llvm::Value *IntVal,
+                                     AggValueSlot ResultSlot,
+                                     SourceLocation Loc, bool AsValue) const;
 
     /// Copy an atomic r-value into atomic-layout memory.
-    void emitCopyIntoMemory(RValue rvalue, LValue lvalue) const;
+    void emitCopyIntoMemory(RValue rvalue) const;
 
     /// Project an l-value down to the value field.
-    LValue projectValue(LValue lvalue) const {
-      llvm::Value *addr = lvalue.getAddress();
+    LValue projectValue() const {
+      assert(LVal.isSimple());
+      llvm::Value *addr = getAtomicAddress();
       if (hasPadding())
-        addr = CGF.Builder.CreateStructGEP(addr, 0);
+        addr = CGF.Builder.CreateStructGEP(nullptr, addr, 0);
 
-      return LValue::MakeAddr(addr, getValueType(), lvalue.getAlignment(),
-                              CGF.getContext(), lvalue.getTBAAInfo());
+      return LValue::MakeAddr(addr, getValueType(), LVal.getAlignment(),
+                              CGF.getContext(), LVal.getTBAAInfo());
     }
 
+    /// \brief Emits atomic load.
+    /// \returns Loaded value.
+    RValue EmitAtomicLoad(AggValueSlot ResultSlot, SourceLocation Loc,
+                          bool AsValue, llvm::AtomicOrdering AO,
+                          bool IsVolatile);
+
+    /// \brief Emits atomic compare-and-exchange sequence.
+    /// \param Expected Expected value.
+    /// \param Desired Desired value.
+    /// \param Success Atomic ordering for success operation.
+    /// \param Failure Atomic ordering for failed operation.
+    /// \param IsWeak true if atomic operation is weak, false otherwise.
+    /// \returns Pair of values: previous value from storage (value type) and
+    /// boolean flag (i1 type) with true if success and false otherwise.
+    std::pair<RValue, llvm::Value *> EmitAtomicCompareExchange(
+        RValue Expected, RValue Desired,
+        llvm::AtomicOrdering Success = llvm::SequentiallyConsistent,
+        llvm::AtomicOrdering Failure = llvm::SequentiallyConsistent,
+        bool IsWeak = false);
+
+    /// \brief Emits atomic update.
+    /// \param AO Atomic ordering.
+    /// \param UpdateOp Update operation for the current lvalue.
+    void EmitAtomicUpdate(llvm::AtomicOrdering AO,
+                          const llvm::function_ref<RValue(RValue)> &UpdateOp,
+                          bool IsVolatile);
+    /// \brief Emits atomic update.
+    /// \param AO Atomic ordering.
+    void EmitAtomicUpdate(llvm::AtomicOrdering AO, RValue UpdateRVal,
+                          bool IsVolatile);
+
     /// Materialize an atomic r-value in atomic-layout memory.
     llvm::Value *materializeRValue(RValue rvalue) const;
 
+    /// \brief Translates LLVM atomic ordering to GNU atomic ordering for
+    /// libcalls.
+    static AtomicExpr::AtomicOrderingKind
+    translateAtomicOrdering(const llvm::AtomicOrdering AO);
+
   private:
     bool requiresMemSetZero(llvm::Type *type) const;
+
+    /// \brief Creates temp alloca for intermediate operations on atomic value.
+    llvm::Value *CreateTempAlloca() const;
+
+    /// \brief Emits atomic load as a libcall.
+    void EmitAtomicLoadLibcall(llvm::Value *AddForLoaded,
+                               llvm::AtomicOrdering AO, bool IsVolatile);
+    /// \brief Emits atomic load as LLVM instruction.
+    llvm::Value *EmitAtomicLoadOp(llvm::AtomicOrdering AO, bool IsVolatile);
+    /// \brief Emits atomic compare-and-exchange op as a libcall.
+    llvm::Value *EmitAtomicCompareExchangeLibcall(
+        llvm::Value *ExpectedAddr, llvm::Value *DesiredAddr,
+        llvm::AtomicOrdering Success = llvm::SequentiallyConsistent,
+        llvm::AtomicOrdering Failure = llvm::SequentiallyConsistent);
+    /// \brief Emits atomic compare-and-exchange op as LLVM instruction.
+    std::pair<llvm::Value *, llvm::Value *> EmitAtomicCompareExchangeOp(
+        llvm::Value *ExpectedVal, llvm::Value *DesiredVal,
+        llvm::AtomicOrdering Success = llvm::SequentiallyConsistent,
+        llvm::AtomicOrdering Failure = llvm::SequentiallyConsistent,
+        bool IsWeak = false);
+    /// \brief Emit atomic update as libcalls.
+    void
+    EmitAtomicUpdateLibcall(llvm::AtomicOrdering AO,
+                            const llvm::function_ref<RValue(RValue)> &UpdateOp,
+                            bool IsVolatile);
+    /// \brief Emit atomic update as LLVM instructions.
+    void EmitAtomicUpdateOp(llvm::AtomicOrdering AO,
+                            const llvm::function_ref<RValue(RValue)> &UpdateOp,
+                            bool IsVolatile);
+    /// \brief Emit atomic update as libcalls.
+    void EmitAtomicUpdateLibcall(llvm::AtomicOrdering AO, RValue UpdateRVal,
+                                 bool IsVolatile);
+    /// \brief Emit atomic update as LLVM instructions.
+    void EmitAtomicUpdateOp(llvm::AtomicOrdering AO, RValue UpdateRal,
+                            bool IsVolatile);
   };
 }
 
+AtomicExpr::AtomicOrderingKind
+AtomicInfo::translateAtomicOrdering(const llvm::AtomicOrdering AO) {
+  switch (AO) {
+  case llvm::Unordered:
+  case llvm::NotAtomic:
+  case llvm::Monotonic:
+    return AtomicExpr::AO_ABI_memory_order_relaxed;
+  case llvm::Acquire:
+    return AtomicExpr::AO_ABI_memory_order_acquire;
+  case llvm::Release:
+    return AtomicExpr::AO_ABI_memory_order_release;
+  case llvm::AcquireRelease:
+    return AtomicExpr::AO_ABI_memory_order_acq_rel;
+  case llvm::SequentiallyConsistent:
+    return AtomicExpr::AO_ABI_memory_order_seq_cst;
+  }
+  llvm_unreachable("Unhandled AtomicOrdering");
+}
+
+llvm::Value *AtomicInfo::CreateTempAlloca() const {
+  auto *TempAlloca = CGF.CreateMemTemp(
+      (LVal.isBitField() && ValueSizeInBits > AtomicSizeInBits) ? ValueTy
+                                                                : AtomicTy,
+      "atomic-temp");
+  TempAlloca->setAlignment(getAtomicAlignment().getQuantity());
+  // Cast to pointer to value type for bitfields.
+  if (LVal.isBitField())
+    return CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+        TempAlloca, getAtomicAddress()->getType());
+  return TempAlloca;
+}
+
 static RValue emitAtomicLibcall(CodeGenFunction &CGF,
                                 StringRef fnName,
                                 QualType resultType,
@@ -172,14 +348,16 @@ bool AtomicInfo::requiresMemSetZero(llvm::Type *type) const {
   llvm_unreachable("bad evaluation kind");
 }
 
-bool AtomicInfo::emitMemSetZeroIfNecessary(LValue dest) const {
-  llvm::Value *addr = dest.getAddress();
+bool AtomicInfo::emitMemSetZeroIfNecessary() const {
+  assert(LVal.isSimple());
+  llvm::Value *addr = LVal.getAddress();
   if (!requiresMemSetZero(addr->getType()->getPointerElementType()))
     return false;
 
-  CGF.Builder.CreateMemSet(addr, llvm::ConstantInt::get(CGF.Int8Ty, 0),
-                           AtomicSizeInBits / 8,
-                           dest.getAlignment().getQuantity());
+  CGF.Builder.CreateMemSet(
+      addr, llvm::ConstantInt::get(CGF.Int8Ty, 0),
+      CGF.getContext().toCharUnitsFromBits(AtomicSizeInBits).getQuantity(),
+      LVal.getAlignment().getQuantity());
   return true;
 }
 
@@ -901,29 +1079,53 @@ llvm::Value *AtomicInfo::emitCastToAtomicIntPointer(llvm::Value *addr) const {
 
 RValue AtomicInfo::convertTempToRValue(llvm::Value *addr,
                                        AggValueSlot resultSlot,
-                                       SourceLocation loc) const {
-  if (EvaluationKind == TEK_Aggregate)
-    return resultSlot.asRValue();
-
-  // Drill into the padding structure if we have one.
-  if (hasPadding())
-    addr = CGF.Builder.CreateStructGEP(addr, 0);
-
-  // Otherwise, just convert the temporary to an r-value using the
-  // normal conversion routine.
-  return CGF.convertTempToRValue(addr, getValueType(), loc);
+                                       SourceLocation loc, bool AsValue) const {
+  if (LVal.isSimple()) {
+    if (EvaluationKind == TEK_Aggregate)
+      return resultSlot.asRValue();
+
+    // Drill into the padding structure if we have one.
+    if (hasPadding())
+      addr = CGF.Builder.CreateStructGEP(nullptr, addr, 0);
+
+    // Otherwise, just convert the temporary to an r-value using the
+    // normal conversion routine.
+    return CGF.convertTempToRValue(addr, getValueType(), loc);
+  }
+  if (!AsValue)
+    // Get RValue from temp memory as atomic for non-simple lvalues
+    return RValue::get(
+        CGF.Builder.CreateAlignedLoad(addr, AtomicAlign.getQuantity()));
+  if (LVal.isBitField())
+    return CGF.EmitLoadOfBitfieldLValue(LValue::MakeBitfield(
+        addr, LVal.getBitFieldInfo(), LVal.getType(), LVal.getAlignment()));
+  if (LVal.isVectorElt())
+    return CGF.EmitLoadOfLValue(LValue::MakeVectorElt(addr, LVal.getVectorIdx(),
+                                                      LVal.getType(),
+                                                      LVal.getAlignment()),
+                                loc);
+  assert(LVal.isExtVectorElt());
+  return CGF.EmitLoadOfExtVectorElementLValue(LValue::MakeExtVectorElt(
+      addr, LVal.getExtVectorElts(), LVal.getType(), LVal.getAlignment()));
 }
 
-RValue AtomicInfo::convertIntToValue(llvm::Value *IntVal,
-                                     AggValueSlot ResultSlot,
-                                     SourceLocation Loc) const {
+RValue AtomicInfo::ConvertIntToValueOrAtomic(llvm::Value *IntVal,
+                                             AggValueSlot ResultSlot,
+                                             SourceLocation Loc,
+                                             bool AsValue) const {
   // Try not to in some easy cases.
   assert(IntVal->getType()->isIntegerTy() && "Expected integer value");
-  if (getEvaluationKind() == TEK_Scalar && !hasPadding()) {
-    auto *ValTy = CGF.ConvertTypeForMem(ValueTy);
+  if (getEvaluationKind() == TEK_Scalar &&
+      (((!LVal.isBitField() ||
+         LVal.getBitFieldInfo().Size == ValueSizeInBits) &&
+        !hasPadding()) ||
+       !AsValue)) {
+    auto *ValTy = AsValue
+                      ? CGF.ConvertTypeForMem(ValueTy)
+                      : getAtomicAddress()->getType()->getPointerElementType();
     if (ValTy->isIntegerTy()) {
       assert(IntVal->getType() == ValTy && "Different integer types.");
-      return RValue::get(IntVal);
+      return RValue::get(CGF.EmitFromMemory(IntVal, ValueTy));
     } else if (ValTy->isPointerTy())
       return RValue::get(CGF.Builder.CreateIntToPtr(IntVal, ValTy));
     else if (llvm::CastInst::isBitCastable(IntVal->getType(), ValTy))
@@ -935,13 +1137,13 @@ RValue AtomicInfo::convertIntToValue(llvm::Value *IntVal,
   llvm::Value *Temp;
   bool TempIsVolatile = false;
   CharUnits TempAlignment;
-  if (getEvaluationKind() == TEK_Aggregate) {
+  if (AsValue && getEvaluationKind() == TEK_Aggregate) {
     assert(!ResultSlot.isIgnored());
     Temp = ResultSlot.getAddr();
     TempAlignment = getValueAlignment();
     TempIsVolatile = ResultSlot.isVolatile();
   } else {
-    Temp = CGF.CreateMemTemp(getAtomicType(), "atomic-temp");
+    Temp = CreateTempAlloca();
     TempAlignment = getAtomicAlignment();
   }
 
@@ -950,93 +1152,146 @@ RValue AtomicInfo::convertIntToValue(llvm::Value *IntVal,
   CGF.Builder.CreateAlignedStore(IntVal, CastTemp, TempAlignment.getQuantity())
       ->setVolatile(TempIsVolatile);
 
-  return convertTempToRValue(Temp, ResultSlot, Loc);
+  return convertTempToRValue(Temp, ResultSlot, Loc, AsValue);
 }
 
-/// Emit a load from an l-value of atomic type.  Note that the r-value
-/// we produce is an r-value of the atomic *value* type.
-RValue CodeGenFunction::EmitAtomicLoad(LValue src, SourceLocation loc,
-                                       AggValueSlot resultSlot) {
-  AtomicInfo atomics(*this, src);
+void AtomicInfo::EmitAtomicLoadLibcall(llvm::Value *AddForLoaded,
+                                       llvm::AtomicOrdering AO, bool) {
+  // void __atomic_load(size_t size, void *mem, void *return, int order);
+  CallArgList Args;
+  Args.add(RValue::get(getAtomicSizeValue()), CGF.getContext().getSizeType());
+  Args.add(RValue::get(CGF.EmitCastToVoidPtr(getAtomicAddress())),
+           CGF.getContext().VoidPtrTy);
+  Args.add(RValue::get(CGF.EmitCastToVoidPtr(AddForLoaded)),
+           CGF.getContext().VoidPtrTy);
+  Args.add(RValue::get(
+               llvm::ConstantInt::get(CGF.IntTy, translateAtomicOrdering(AO))),
+           CGF.getContext().IntTy);
+  emitAtomicLibcall(CGF, "__atomic_load", CGF.getContext().VoidTy, Args);
+}
 
-  // Check whether we should use a library call.
-  if (atomics.shouldUseLibcall()) {
-    llvm::Value *tempAddr;
-    if (!resultSlot.isIgnored()) {
-      assert(atomics.getEvaluationKind() == TEK_Aggregate);
-      tempAddr = resultSlot.getAddr();
-    } else {
-      tempAddr = CreateMemTemp(atomics.getAtomicType(), "atomic-load-temp");
-    }
+llvm::Value *AtomicInfo::EmitAtomicLoadOp(llvm::AtomicOrdering AO,
+                                          bool IsVolatile) {
+  // Okay, we're doing this natively.
+  llvm::Value *Addr = emitCastToAtomicIntPointer(getAtomicAddress());
+  llvm::LoadInst *Load = CGF.Builder.CreateLoad(Addr, "atomic-load");
+  Load->setAtomic(AO);
 
-    // void __atomic_load(size_t size, void *mem, void *return, int order);
-    CallArgList args;
-    args.add(RValue::get(atomics.getAtomicSizeValue()),
-             getContext().getSizeType());
-    args.add(RValue::get(EmitCastToVoidPtr(src.getAddress())),
-             getContext().VoidPtrTy);
-    args.add(RValue::get(EmitCastToVoidPtr(tempAddr)),
-             getContext().VoidPtrTy);
-    args.add(RValue::get(llvm::ConstantInt::get(
-                 IntTy, AtomicExpr::AO_ABI_memory_order_seq_cst)),
-             getContext().IntTy);
-    emitAtomicLibcall(*this, "__atomic_load", getContext().VoidTy, args);
+  // Other decoration.
+  Load->setAlignment(getAtomicAlignment().getQuantity());
+  if (IsVolatile)
+    Load->setVolatile(true);
+  if (LVal.getTBAAInfo())
+    CGF.CGM.DecorateInstruction(Load, LVal.getTBAAInfo());
+  return Load;
+}
+
+/// An LValue is a candidate for having its loads and stores be made atomic if
+/// we are operating under /volatile:ms *and* the LValue itself is volatile and
+/// performing such an operation can be performed without a libcall.
+bool CodeGenFunction::LValueIsSuitableForInlineAtomic(LValue LV) {
+  AtomicInfo AI(*this, LV);
+  bool IsVolatile = LV.isVolatile() || hasVolatileMember(LV.getType());
+  // An atomic is inline if we don't need to use a libcall.
+  bool AtomicIsInline = !AI.shouldUseLibcall();
+  return CGM.getCodeGenOpts().MSVolatile && IsVolatile && AtomicIsInline;
+}
 
-    // Produce the r-value.
-    return atomics.convertTempToRValue(tempAddr, resultSlot, loc);
+/// An type is a candidate for having its loads and stores be made atomic if
+/// we are operating under /volatile:ms *and* we know the access is volatile and
+/// performing such an operation can be performed without a libcall.
+bool CodeGenFunction::typeIsSuitableForInlineAtomic(QualType Ty,
+                                                    bool IsVolatile) const {
+  // An atomic is inline if we don't need to use a libcall (e.g. it is builtin).
+  bool AtomicIsInline = getContext().getTargetInfo().hasBuiltinAtomic(
+      getContext().getTypeSize(Ty), getContext().getTypeAlign(Ty));
+  return CGM.getCodeGenOpts().MSVolatile && IsVolatile && AtomicIsInline;
+}
+
+RValue CodeGenFunction::EmitAtomicLoad(LValue LV, SourceLocation SL,
+                                       AggValueSlot Slot) {
+  llvm::AtomicOrdering AO;
+  bool IsVolatile = LV.isVolatileQualified();
+  if (LV.getType()->isAtomicType()) {
+    AO = llvm::SequentiallyConsistent;
+  } else {
+    AO = llvm::Acquire;
+    IsVolatile = true;
   }
+  return EmitAtomicLoad(LV, SL, AO, IsVolatile, Slot);
+}
 
-  // Okay, we're doing this natively.
-  llvm::Value *addr = atomics.emitCastToAtomicIntPointer(src.getAddress());
-  llvm::LoadInst *load = Builder.CreateLoad(addr, "atomic-load");
-  load->setAtomic(llvm::SequentiallyConsistent);
+RValue AtomicInfo::EmitAtomicLoad(AggValueSlot ResultSlot, SourceLocation Loc,
+                                  bool AsValue, llvm::AtomicOrdering AO,
+                                  bool IsVolatile) {
+  // Check whether we should use a library call.
+  if (shouldUseLibcall()) {
+    llvm::Value *TempAddr;
+    if (LVal.isSimple() && !ResultSlot.isIgnored()) {
+      assert(getEvaluationKind() == TEK_Aggregate);
+      TempAddr = ResultSlot.getAddr();
+    } else
+      TempAddr = CreateTempAlloca();
+
+    EmitAtomicLoadLibcall(TempAddr, AO, IsVolatile);
+
+    // Okay, turn that back into the original value or whole atomic (for
+    // non-simple lvalues) type.
+    return convertTempToRValue(TempAddr, ResultSlot, Loc, AsValue);
+  }
 
-  // Other decoration.
-  load->setAlignment(src.getAlignment().getQuantity());
-  if (src.isVolatileQualified())
-    load->setVolatile(true);
-  if (src.getTBAAInfo())
-    CGM.DecorateInstruction(load, src.getTBAAInfo());
+  // Okay, we're doing this natively.
+  auto *Load = EmitAtomicLoadOp(AO, IsVolatile);
 
   // If we're ignoring an aggregate return, don't do anything.
-  if (atomics.getEvaluationKind() == TEK_Aggregate && resultSlot.isIgnored())
+  if (getEvaluationKind() == TEK_Aggregate && ResultSlot.isIgnored())
     return RValue::getAggregate(nullptr, false);
 
-  // Okay, turn that back into the original value type.
-  return atomics.convertIntToValue(load, resultSlot, loc);
+  // Okay, turn that back into the original value or atomic (for non-simple
+  // lvalues) type.
+  return ConvertIntToValueOrAtomic(Load, ResultSlot, Loc, AsValue);
 }
 
-
+/// Emit a load from an l-value of atomic type.  Note that the r-value
+/// we produce is an r-value of the atomic *value* type.
+RValue CodeGenFunction::EmitAtomicLoad(LValue src, SourceLocation loc,
+                                       llvm::AtomicOrdering AO, bool IsVolatile,
+                                       AggValueSlot resultSlot) {
+  AtomicInfo Atomics(*this, src);
+  return Atomics.EmitAtomicLoad(resultSlot, loc, /*AsValue=*/true, AO,
+                                IsVolatile);
+}
 
 /// Copy an r-value into memory as part of storing to an atomic type.
 /// This needs to create a bit-pattern suitable for atomic operations.
-void AtomicInfo::emitCopyIntoMemory(RValue rvalue, LValue dest) const {
+void AtomicInfo::emitCopyIntoMemory(RValue rvalue) const {
+  assert(LVal.isSimple());
   // If we have an r-value, the rvalue should be of the atomic type,
   // which means that the caller is responsible for having zeroed
   // any padding.  Just do an aggregate copy of that type.
   if (rvalue.isAggregate()) {
-    CGF.EmitAggregateCopy(dest.getAddress(),
+    CGF.EmitAggregateCopy(getAtomicAddress(),
                           rvalue.getAggregateAddr(),
                           getAtomicType(),
                           (rvalue.isVolatileQualified()
-                           || dest.isVolatileQualified()),
-                          dest.getAlignment());
+                           || LVal.isVolatileQualified()),
+                          LVal.getAlignment());
     return;
   }
 
   // Okay, otherwise we're copying stuff.
 
   // Zero out the buffer if necessary.
-  emitMemSetZeroIfNecessary(dest);
+  emitMemSetZeroIfNecessary();
 
   // Drill past the padding if present.
-  dest = projectValue(dest);
+  LValue TempLVal = projectValue();
 
   // Okay, store the rvalue in.
   if (rvalue.isScalar()) {
-    CGF.EmitStoreOfScalar(rvalue.getScalarVal(), dest, /*init*/ true);
+    CGF.EmitStoreOfScalar(rvalue.getScalarVal(), TempLVal, /*init*/ true);
   } else {
-    CGF.EmitStoreOfComplex(rvalue.getComplexVal(), dest, /*init*/ true);
+    CGF.EmitStoreOfComplex(rvalue.getComplexVal(), TempLVal, /*init*/ true);
   }
 }
 
@@ -1050,22 +1305,24 @@ llvm::Value *AtomicInfo::materializeRValue(RValue rvalue) const {
     return rvalue.getAggregateAddr();
 
   // Otherwise, make a temporary and materialize into it.
-  llvm::Value *temp = CGF.CreateMemTemp(getAtomicType(), "atomic-store-temp");
-  LValue tempLV = CGF.MakeAddrLValue(temp, getAtomicType(), getAtomicAlignment());
-  emitCopyIntoMemory(rvalue, tempLV);
-  return temp;
+  LValue TempLV = CGF.MakeAddrLValue(CreateTempAlloca(), getAtomicType(),
+                                     getAtomicAlignment());
+  AtomicInfo Atomics(CGF, TempLV);
+  Atomics.emitCopyIntoMemory(rvalue);
+  return TempLV.getAddress();
 }
 
 llvm::Value *AtomicInfo::convertRValueToInt(RValue RVal) const {
   // If we've got a scalar value of the right size, try to avoid going
   // through memory.
-  if (RVal.isScalar() && !hasPadding()) {
+  if (RVal.isScalar() && (!hasPadding() || !LVal.isSimple())) {
     llvm::Value *Value = RVal.getScalarVal();
     if (isa<llvm::IntegerType>(Value->getType()))
-      return Value;
+      return CGF.EmitToMemory(Value, ValueTy);
     else {
-      llvm::IntegerType *InputIntTy =
-          llvm::IntegerType::get(CGF.getLLVMContext(), getValueSizeInBits());
+      llvm::IntegerType *InputIntTy = llvm::IntegerType::get(
+          CGF.getLLVMContext(),
+          LVal.isSimple() ? getValueSizeInBits() : getAtomicSizeInBits());
       if (isa<llvm::PointerType>(Value->getType()))
         return CGF.Builder.CreatePtrToInt(Value, InputIntTy);
       else if (llvm::BitCastInst::isBitCastable(Value->getType(), InputIntTy))
@@ -1082,12 +1339,324 @@ llvm::Value *AtomicInfo::convertRValueToInt(RValue RVal) const {
                                        getAtomicAlignment().getQuantity());
 }
 
+std::pair<llvm::Value *, llvm::Value *> AtomicInfo::EmitAtomicCompareExchangeOp(
+    llvm::Value *ExpectedVal, llvm::Value *DesiredVal,
+    llvm::AtomicOrdering Success, llvm::AtomicOrdering Failure, bool IsWeak) {
+  // Do the atomic store.
+  auto *Addr = emitCastToAtomicIntPointer(getAtomicAddress());
+  auto *Inst = CGF.Builder.CreateAtomicCmpXchg(Addr, ExpectedVal, DesiredVal,
+                                               Success, Failure);
+  // Other decoration.
+  Inst->setVolatile(LVal.isVolatileQualified());
+  Inst->setWeak(IsWeak);
+
+  // Okay, turn that back into the original value type.
+  auto *PreviousVal = CGF.Builder.CreateExtractValue(Inst, /*Idxs=*/0);
+  auto *SuccessFailureVal = CGF.Builder.CreateExtractValue(Inst, /*Idxs=*/1);
+  return std::make_pair(PreviousVal, SuccessFailureVal);
+}
+
+llvm::Value *
+AtomicInfo::EmitAtomicCompareExchangeLibcall(llvm::Value *ExpectedAddr,
+                                             llvm::Value *DesiredAddr,
+                                             llvm::AtomicOrdering Success,
+                                             llvm::AtomicOrdering Failure) {
+  // bool __atomic_compare_exchange(size_t size, void *obj, void *expected,
+  // void *desired, int success, int failure);
+  CallArgList Args;
+  Args.add(RValue::get(getAtomicSizeValue()), CGF.getContext().getSizeType());
+  Args.add(RValue::get(CGF.EmitCastToVoidPtr(getAtomicAddress())),
+           CGF.getContext().VoidPtrTy);
+  Args.add(RValue::get(CGF.EmitCastToVoidPtr(ExpectedAddr)),
+           CGF.getContext().VoidPtrTy);
+  Args.add(RValue::get(CGF.EmitCastToVoidPtr(DesiredAddr)),
+           CGF.getContext().VoidPtrTy);
+  Args.add(RValue::get(llvm::ConstantInt::get(
+               CGF.IntTy, translateAtomicOrdering(Success))),
+           CGF.getContext().IntTy);
+  Args.add(RValue::get(llvm::ConstantInt::get(
+               CGF.IntTy, translateAtomicOrdering(Failure))),
+           CGF.getContext().IntTy);
+  auto SuccessFailureRVal = emitAtomicLibcall(CGF, "__atomic_compare_exchange",
+                                              CGF.getContext().BoolTy, Args);
+
+  return SuccessFailureRVal.getScalarVal();
+}
+
+std::pair<RValue, llvm::Value *> AtomicInfo::EmitAtomicCompareExchange(
+    RValue Expected, RValue Desired, llvm::AtomicOrdering Success,
+    llvm::AtomicOrdering Failure, bool IsWeak) {
+  if (Failure >= Success)
+    // Don't assert on undefined behavior.
+    Failure = llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(Success);
+
+  // Check whether we should use a library call.
+  if (shouldUseLibcall()) {
+    // Produce a source address.
+    auto *ExpectedAddr = materializeRValue(Expected);
+    auto *DesiredAddr = materializeRValue(Desired);
+    auto *Res = EmitAtomicCompareExchangeLibcall(ExpectedAddr, DesiredAddr,
+                                                 Success, Failure);
+    return std::make_pair(
+        convertTempToRValue(ExpectedAddr, AggValueSlot::ignored(),
+                            SourceLocation(), /*AsValue=*/false),
+        Res);
+  }
+
+  // If we've got a scalar value of the right size, try to avoid going
+  // through memory.
+  auto *ExpectedVal = convertRValueToInt(Expected);
+  auto *DesiredVal = convertRValueToInt(Desired);
+  auto Res = EmitAtomicCompareExchangeOp(ExpectedVal, DesiredVal, Success,
+                                         Failure, IsWeak);
+  return std::make_pair(
+      ConvertIntToValueOrAtomic(Res.first, AggValueSlot::ignored(),
+                                SourceLocation(), /*AsValue=*/false),
+      Res.second);
+}
+
+static void
+EmitAtomicUpdateValue(CodeGenFunction &CGF, AtomicInfo &Atomics, RValue OldRVal,
+                      const llvm::function_ref<RValue(RValue)> &UpdateOp,
+                      llvm::Value *DesiredAddr) {
+  llvm::Value *Ptr = nullptr;
+  LValue UpdateLVal;
+  RValue UpRVal;
+  LValue AtomicLVal = Atomics.getAtomicLValue();
+  LValue DesiredLVal;
+  if (AtomicLVal.isSimple()) {
+    UpRVal = OldRVal;
+    DesiredLVal =
+        LValue::MakeAddr(DesiredAddr, AtomicLVal.getType(),
+                         AtomicLVal.getAlignment(), CGF.CGM.getContext());
+  } else {
+    // Build new lvalue for temp address
+    Ptr = Atomics.materializeRValue(OldRVal);
+    if (AtomicLVal.isBitField()) {
+      UpdateLVal =
+          LValue::MakeBitfield(Ptr, AtomicLVal.getBitFieldInfo(),
+                               AtomicLVal.getType(), AtomicLVal.getAlignment());
+      DesiredLVal =
+          LValue::MakeBitfield(DesiredAddr, AtomicLVal.getBitFieldInfo(),
+                               AtomicLVal.getType(), AtomicLVal.getAlignment());
+    } else if (AtomicLVal.isVectorElt()) {
+      UpdateLVal = LValue::MakeVectorElt(Ptr, AtomicLVal.getVectorIdx(),
+                                         AtomicLVal.getType(),
+                                         AtomicLVal.getAlignment());
+      DesiredLVal = LValue::MakeVectorElt(
+          DesiredAddr, AtomicLVal.getVectorIdx(), AtomicLVal.getType(),
+          AtomicLVal.getAlignment());
+    } else {
+      assert(AtomicLVal.isExtVectorElt());
+      UpdateLVal = LValue::MakeExtVectorElt(Ptr, AtomicLVal.getExtVectorElts(),
+                                            AtomicLVal.getType(),
+                                            AtomicLVal.getAlignment());
+      DesiredLVal = LValue::MakeExtVectorElt(
+          DesiredAddr, AtomicLVal.getExtVectorElts(), AtomicLVal.getType(),
+          AtomicLVal.getAlignment());
+    }
+    UpdateLVal.setTBAAInfo(AtomicLVal.getTBAAInfo());
+    DesiredLVal.setTBAAInfo(AtomicLVal.getTBAAInfo());
+    UpRVal = CGF.EmitLoadOfLValue(UpdateLVal, SourceLocation());
+  }
+  // Store new value in the corresponding memory area
+  RValue NewRVal = UpdateOp(UpRVal);
+  if (NewRVal.isScalar()) {
+    CGF.EmitStoreThroughLValue(NewRVal, DesiredLVal);
+  } else {
+    assert(NewRVal.isComplex());
+    CGF.EmitStoreOfComplex(NewRVal.getComplexVal(), DesiredLVal,
+                           /*isInit=*/false);
+  }
+}
+
+void AtomicInfo::EmitAtomicUpdateLibcall(
+    llvm::AtomicOrdering AO, const llvm::function_ref<RValue(RValue)> &UpdateOp,
+    bool IsVolatile) {
+  auto Failure = llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(AO);
+
+  llvm::Value *ExpectedAddr = CreateTempAlloca();
+
+  EmitAtomicLoadLibcall(ExpectedAddr, AO, IsVolatile);
+  auto *ContBB = CGF.createBasicBlock("atomic_cont");
+  auto *ExitBB = CGF.createBasicBlock("atomic_exit");
+  CGF.EmitBlock(ContBB);
+  auto *DesiredAddr = CreateTempAlloca();
+  if ((LVal.isBitField() && BFI.Size != ValueSizeInBits) ||
+      requiresMemSetZero(
+          getAtomicAddress()->getType()->getPointerElementType())) {
+    auto *OldVal = CGF.Builder.CreateAlignedLoad(
+        ExpectedAddr, getAtomicAlignment().getQuantity());
+    CGF.Builder.CreateAlignedStore(OldVal, DesiredAddr,
+                                   getAtomicAlignment().getQuantity());
+  }
+  auto OldRVal = convertTempToRValue(ExpectedAddr, AggValueSlot::ignored(),
+                                    SourceLocation(), /*AsValue=*/false);
+  EmitAtomicUpdateValue(CGF, *this, OldRVal, UpdateOp, DesiredAddr);
+  auto *Res =
+      EmitAtomicCompareExchangeLibcall(ExpectedAddr, DesiredAddr, AO, Failure);
+  CGF.Builder.CreateCondBr(Res, ExitBB, ContBB);
+  CGF.EmitBlock(ExitBB, /*IsFinished=*/true);
+}
+
+void AtomicInfo::EmitAtomicUpdateOp(
+    llvm::AtomicOrdering AO, const llvm::function_ref<RValue(RValue)> &UpdateOp,
+    bool IsVolatile) {
+  auto Failure = llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(AO);
+
+  // Do the atomic load.
+  auto *OldVal = EmitAtomicLoadOp(AO, IsVolatile);
+  // For non-simple lvalues perform compare-and-swap procedure.
+  auto *ContBB = CGF.createBasicBlock("atomic_cont");
+  auto *ExitBB = CGF.createBasicBlock("atomic_exit");
+  auto *CurBB = CGF.Builder.GetInsertBlock();
+  CGF.EmitBlock(ContBB);
+  llvm::PHINode *PHI = CGF.Builder.CreatePHI(OldVal->getType(),
+                                             /*NumReservedValues=*/2);
+  PHI->addIncoming(OldVal, CurBB);
+  auto *NewAtomicAddr = CreateTempAlloca();
+  auto *NewAtomicIntAddr = emitCastToAtomicIntPointer(NewAtomicAddr);
+  if ((LVal.isBitField() && BFI.Size != ValueSizeInBits) ||
+      requiresMemSetZero(
+          getAtomicAddress()->getType()->getPointerElementType())) {
+    CGF.Builder.CreateAlignedStore(PHI, NewAtomicIntAddr,
+                                   getAtomicAlignment().getQuantity());
+  }
+  auto OldRVal = ConvertIntToValueOrAtomic(PHI, AggValueSlot::ignored(),
+                                           SourceLocation(), /*AsValue=*/false);
+  EmitAtomicUpdateValue(CGF, *this, OldRVal, UpdateOp, NewAtomicAddr);
+  auto *DesiredVal = CGF.Builder.CreateAlignedLoad(
+      NewAtomicIntAddr, getAtomicAlignment().getQuantity());
+  // Try to write new value using cmpxchg operation
+  auto Res = EmitAtomicCompareExchangeOp(PHI, DesiredVal, AO, Failure);
+  PHI->addIncoming(Res.first, CGF.Builder.GetInsertBlock());
+  CGF.Builder.CreateCondBr(Res.second, ExitBB, ContBB);
+  CGF.EmitBlock(ExitBB, /*IsFinished=*/true);
+}
+
+static void EmitAtomicUpdateValue(CodeGenFunction &CGF, AtomicInfo &Atomics,
+                                  RValue UpdateRVal, llvm::Value *DesiredAddr) {
+  LValue AtomicLVal = Atomics.getAtomicLValue();
+  LValue DesiredLVal;
+  // Build new lvalue for temp address
+  if (AtomicLVal.isBitField()) {
+    DesiredLVal =
+        LValue::MakeBitfield(DesiredAddr, AtomicLVal.getBitFieldInfo(),
+                             AtomicLVal.getType(), AtomicLVal.getAlignment());
+  } else if (AtomicLVal.isVectorElt()) {
+    DesiredLVal =
+        LValue::MakeVectorElt(DesiredAddr, AtomicLVal.getVectorIdx(),
+                              AtomicLVal.getType(), AtomicLVal.getAlignment());
+  } else {
+    assert(AtomicLVal.isExtVectorElt());
+    DesiredLVal = LValue::MakeExtVectorElt(
+        DesiredAddr, AtomicLVal.getExtVectorElts(), AtomicLVal.getType(),
+        AtomicLVal.getAlignment());
+  }
+  DesiredLVal.setTBAAInfo(AtomicLVal.getTBAAInfo());
+  // Store new value in the corresponding memory area
+  assert(UpdateRVal.isScalar());
+  CGF.EmitStoreThroughLValue(UpdateRVal, DesiredLVal);
+}
+
+void AtomicInfo::EmitAtomicUpdateLibcall(llvm::AtomicOrdering AO,
+                                         RValue UpdateRVal, bool IsVolatile) {
+  auto Failure = llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(AO);
+
+  llvm::Value *ExpectedAddr = CreateTempAlloca();
+
+  EmitAtomicLoadLibcall(ExpectedAddr, AO, IsVolatile);
+  auto *ContBB = CGF.createBasicBlock("atomic_cont");
+  auto *ExitBB = CGF.createBasicBlock("atomic_exit");
+  CGF.EmitBlock(ContBB);
+  auto *DesiredAddr = CreateTempAlloca();
+  if ((LVal.isBitField() && BFI.Size != ValueSizeInBits) ||
+      requiresMemSetZero(
+          getAtomicAddress()->getType()->getPointerElementType())) {
+    auto *OldVal = CGF.Builder.CreateAlignedLoad(
+        ExpectedAddr, getAtomicAlignment().getQuantity());
+    CGF.Builder.CreateAlignedStore(OldVal, DesiredAddr,
+                                   getAtomicAlignment().getQuantity());
+  }
+  EmitAtomicUpdateValue(CGF, *this, UpdateRVal, DesiredAddr);
+  auto *Res =
+      EmitAtomicCompareExchangeLibcall(ExpectedAddr, DesiredAddr, AO, Failure);
+  CGF.Builder.CreateCondBr(Res, ExitBB, ContBB);
+  CGF.EmitBlock(ExitBB, /*IsFinished=*/true);
+}
+
+void AtomicInfo::EmitAtomicUpdateOp(llvm::AtomicOrdering AO, RValue UpdateRVal,
+                                    bool IsVolatile) {
+  auto Failure = llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(AO);
+
+  // Do the atomic load.
+  auto *OldVal = EmitAtomicLoadOp(AO, IsVolatile);
+  // For non-simple lvalues perform compare-and-swap procedure.
+  auto *ContBB = CGF.createBasicBlock("atomic_cont");
+  auto *ExitBB = CGF.createBasicBlock("atomic_exit");
+  auto *CurBB = CGF.Builder.GetInsertBlock();
+  CGF.EmitBlock(ContBB);
+  llvm::PHINode *PHI = CGF.Builder.CreatePHI(OldVal->getType(),
+                                             /*NumReservedValues=*/2);
+  PHI->addIncoming(OldVal, CurBB);
+  auto *NewAtomicAddr = CreateTempAlloca();
+  auto *NewAtomicIntAddr = emitCastToAtomicIntPointer(NewAtomicAddr);
+  if ((LVal.isBitField() && BFI.Size != ValueSizeInBits) ||
+      requiresMemSetZero(
+          getAtomicAddress()->getType()->getPointerElementType())) {
+    CGF.Builder.CreateAlignedStore(PHI, NewAtomicIntAddr,
+                                   getAtomicAlignment().getQuantity());
+  }
+  EmitAtomicUpdateValue(CGF, *this, UpdateRVal, NewAtomicAddr);
+  auto *DesiredVal = CGF.Builder.CreateAlignedLoad(
+      NewAtomicIntAddr, getAtomicAlignment().getQuantity());
+  // Try to write new value using cmpxchg operation
+  auto Res = EmitAtomicCompareExchangeOp(PHI, DesiredVal, AO, Failure);
+  PHI->addIncoming(Res.first, CGF.Builder.GetInsertBlock());
+  CGF.Builder.CreateCondBr(Res.second, ExitBB, ContBB);
+  CGF.EmitBlock(ExitBB, /*IsFinished=*/true);
+}
+
+void AtomicInfo::EmitAtomicUpdate(
+    llvm::AtomicOrdering AO, const llvm::function_ref<RValue(RValue)> &UpdateOp,
+    bool IsVolatile) {
+  if (shouldUseLibcall()) {
+    EmitAtomicUpdateLibcall(AO, UpdateOp, IsVolatile);
+  } else {
+    EmitAtomicUpdateOp(AO, UpdateOp, IsVolatile);
+  }
+}
+
+void AtomicInfo::EmitAtomicUpdate(llvm::AtomicOrdering AO, RValue UpdateRVal,
+                                  bool IsVolatile) {
+  if (shouldUseLibcall()) {
+    EmitAtomicUpdateLibcall(AO, UpdateRVal, IsVolatile);
+  } else {
+    EmitAtomicUpdateOp(AO, UpdateRVal, IsVolatile);
+  }
+}
+
+void CodeGenFunction::EmitAtomicStore(RValue rvalue, LValue lvalue,
+                                      bool isInit) {
+  bool IsVolatile = lvalue.isVolatileQualified();
+  llvm::AtomicOrdering AO;
+  if (lvalue.getType()->isAtomicType()) {
+    AO = llvm::SequentiallyConsistent;
+  } else {
+    AO = llvm::Release;
+    IsVolatile = true;
+  }
+  return EmitAtomicStore(rvalue, lvalue, AO, IsVolatile, isInit);
+}
+
 /// Emit a store to an l-value of atomic type.
 ///
 /// Note that the r-value is expected to be an r-value *of the atomic
 /// type*; this means that for aggregate r-values, it should include
 /// storage for any padding that was necessary.
-void CodeGenFunction::EmitAtomicStore(RValue rvalue, LValue dest, bool isInit) {
+void CodeGenFunction::EmitAtomicStore(RValue rvalue, LValue dest,
+                                      llvm::AtomicOrdering AO, bool IsVolatile,
+                                      bool isInit) {
   // If this is an aggregate r-value, it should agree in type except
   // maybe for address-space qualification.
   assert(!rvalue.isAggregate() ||
@@ -1095,54 +1664,64 @@ void CodeGenFunction::EmitAtomicStore(RValue rvalue, LValue dest, bool isInit) {
            == dest.getAddress()->getType()->getPointerElementType());
 
   AtomicInfo atomics(*this, dest);
+  LValue LVal = atomics.getAtomicLValue();
 
   // If this is an initialization, just put the value there normally.
-  if (isInit) {
-    atomics.emitCopyIntoMemory(rvalue, dest);
-    return;
-  }
+  if (LVal.isSimple()) {
+    if (isInit) {
+      atomics.emitCopyIntoMemory(rvalue);
+      return;
+    }
 
-  // Check whether we should use a library call.
-  if (atomics.shouldUseLibcall()) {
-    // Produce a source address.
-    llvm::Value *srcAddr = atomics.materializeRValue(rvalue);
+    // Check whether we should use a library call.
+    if (atomics.shouldUseLibcall()) {
+      // Produce a source address.
+      llvm::Value *srcAddr = atomics.materializeRValue(rvalue);
 
-    // void __atomic_store(size_t size, void *mem, void *val, int order)
-    CallArgList args;
-    args.add(RValue::get(atomics.getAtomicSizeValue()),
-             getContext().getSizeType());
-    args.add(RValue::get(EmitCastToVoidPtr(dest.getAddress())),
-             getContext().VoidPtrTy);
-    args.add(RValue::get(EmitCastToVoidPtr(srcAddr)),
-             getContext().VoidPtrTy);
-    args.add(RValue::get(llvm::ConstantInt::get(
-                 IntTy, AtomicExpr::AO_ABI_memory_order_seq_cst)),
-             getContext().IntTy);
-    emitAtomicLibcall(*this, "__atomic_store", getContext().VoidTy, args);
+      // void __atomic_store(size_t size, void *mem, void *val, int order)
+      CallArgList args;
+      args.add(RValue::get(atomics.getAtomicSizeValue()),
+               getContext().getSizeType());
+      args.add(RValue::get(EmitCastToVoidPtr(atomics.getAtomicAddress())),
+               getContext().VoidPtrTy);
+      args.add(RValue::get(EmitCastToVoidPtr(srcAddr)), getContext().VoidPtrTy);
+      args.add(RValue::get(llvm::ConstantInt::get(
+                   IntTy, AtomicInfo::translateAtomicOrdering(AO))),
+               getContext().IntTy);
+      emitAtomicLibcall(*this, "__atomic_store", getContext().VoidTy, args);
+      return;
+    }
+
+    // Okay, we're doing this natively.
+    llvm::Value *intValue = atomics.convertRValueToInt(rvalue);
+
+    // Do the atomic store.
+    llvm::Value *addr =
+        atomics.emitCastToAtomicIntPointer(atomics.getAtomicAddress());
+    intValue = Builder.CreateIntCast(
+        intValue, addr->getType()->getPointerElementType(), /*isSigned=*/false);
+    llvm::StoreInst *store = Builder.CreateStore(intValue, addr);
+
+    // Initializations don't need to be atomic.
+    if (!isInit)
+      store->setAtomic(AO);
+
+    // Other decoration.
+    store->setAlignment(dest.getAlignment().getQuantity());
+    if (IsVolatile)
+      store->setVolatile(true);
+    if (dest.getTBAAInfo())
+      CGM.DecorateInstruction(store, dest.getTBAAInfo());
     return;
   }
 
-  // Okay, we're doing this natively.
-  llvm::Value *intValue = atomics.convertRValueToInt(rvalue);
-
-  // Do the atomic store.
-  llvm::Value *addr = atomics.emitCastToAtomicIntPointer(dest.getAddress());
-  llvm::StoreInst *store = Builder.CreateStore(intValue, addr);
-
-  // Initializations don't need to be atomic.
-  if (!isInit) store->setAtomic(llvm::SequentiallyConsistent);
-
-  // Other decoration.
-  store->setAlignment(dest.getAlignment().getQuantity());
-  if (dest.isVolatileQualified())
-    store->setVolatile(true);
-  if (dest.getTBAAInfo())
-    CGM.DecorateInstruction(store, dest.getTBAAInfo());
+  // Emit simple atomic update operation.
+  atomics.EmitAtomicUpdate(AO, rvalue, IsVolatile);
 }
 
 /// Emit a compare-and-exchange op for atomic type.
 ///
-std::pair<RValue, RValue> CodeGenFunction::EmitAtomicCompareExchange(
+std::pair<RValue, llvm::Value *> CodeGenFunction::EmitAtomicCompareExchange(
     LValue Obj, RValue Expected, RValue Desired, SourceLocation Loc,
     llvm::AtomicOrdering Success, llvm::AtomicOrdering Failure, bool IsWeak,
     AggValueSlot Slot) {
@@ -1156,56 +1735,15 @@ std::pair<RValue, RValue> CodeGenFunction::EmitAtomicCompareExchange(
              Obj.getAddress()->getType()->getPointerElementType());
   AtomicInfo Atomics(*this, Obj);
 
-  if (Failure >= Success)
-    // Don't assert on undefined behavior.
-    Failure = llvm::AtomicCmpXchgInst::getStrongestFailureOrdering(Success);
-
-  auto Alignment = Atomics.getValueAlignment();
-  // Check whether we should use a library call.
-  if (Atomics.shouldUseLibcall()) {
-    auto *ExpectedAddr = Atomics.materializeRValue(Expected);
-    // Produce a source address.
-    auto *DesiredAddr = Atomics.materializeRValue(Desired);
-    // bool __atomic_compare_exchange(size_t size, void *obj, void *expected,
-    // void *desired, int success, int failure);
-    CallArgList Args;
-    Args.add(RValue::get(Atomics.getAtomicSizeValue()),
-             getContext().getSizeType());
-    Args.add(RValue::get(EmitCastToVoidPtr(Obj.getAddress())),
-             getContext().VoidPtrTy);
-    Args.add(RValue::get(EmitCastToVoidPtr(ExpectedAddr)),
-             getContext().VoidPtrTy);
-    Args.add(RValue::get(EmitCastToVoidPtr(DesiredAddr)),
-             getContext().VoidPtrTy);
-    Args.add(RValue::get(llvm::ConstantInt::get(IntTy, Success)),
-             getContext().IntTy);
-    Args.add(RValue::get(llvm::ConstantInt::get(IntTy, Failure)),
-             getContext().IntTy);
-    auto SuccessFailureRVal = emitAtomicLibcall(
-        *this, "__atomic_compare_exchange", getContext().BoolTy, Args);
-    auto *PreviousVal =
-        Builder.CreateAlignedLoad(ExpectedAddr, Alignment.getQuantity());
-    return std::make_pair(RValue::get(PreviousVal), SuccessFailureRVal);
-  }
-
-  // If we've got a scalar value of the right size, try to avoid going
-  // through memory.
-  auto *ExpectedIntVal = Atomics.convertRValueToInt(Expected);
-  auto *DesiredIntVal = Atomics.convertRValueToInt(Desired);
-
-  // Do the atomic store.
-  auto *Addr = Atomics.emitCastToAtomicIntPointer(Obj.getAddress());
-  auto *Inst = Builder.CreateAtomicCmpXchg(Addr, ExpectedIntVal, DesiredIntVal,
-                                          Success, Failure);
-  // Other decoration.
-  Inst->setVolatile(Obj.isVolatileQualified());
-  Inst->setWeak(IsWeak);
+  return Atomics.EmitAtomicCompareExchange(Expected, Desired, Success, Failure,
+                                           IsWeak);
+}
 
-  // Okay, turn that back into the original value type.
-  auto *PreviousVal = Builder.CreateExtractValue(Inst, /*Idxs=*/0);
-  auto *SuccessFailureVal = Builder.CreateExtractValue(Inst, /*Idxs=*/1);
-  return std::make_pair(Atomics.convertIntToValue(PreviousVal, Slot, Loc),
-                        RValue::get(SuccessFailureVal));
+void CodeGenFunction::EmitAtomicUpdate(
+    LValue LVal, llvm::AtomicOrdering AO,
+    const llvm::function_ref<RValue(RValue)> &UpdateOp, bool IsVolatile) {
+  AtomicInfo Atomics(*this, LVal);
+  Atomics.EmitAtomicUpdate(AO, UpdateOp, IsVolatile);
 }
 
 void CodeGenFunction::EmitAtomicInit(Expr *init, LValue dest) {
@@ -1214,13 +1752,13 @@ void CodeGenFunction::EmitAtomicInit(Expr *init, LValue dest) {
   switch (atomics.getEvaluationKind()) {
   case TEK_Scalar: {
     llvm::Value *value = EmitScalarExpr(init);
-    atomics.emitCopyIntoMemory(RValue::get(value), dest);
+    atomics.emitCopyIntoMemory(RValue::get(value));
     return;
   }
 
   case TEK_Complex: {
     ComplexPairTy value = EmitComplexExpr(init);
-    atomics.emitCopyIntoMemory(RValue::getComplex(value), dest);
+    atomics.emitCopyIntoMemory(RValue::getComplex(value));
     return;
   }
 
@@ -1229,8 +1767,8 @@ void CodeGenFunction::EmitAtomicInit(Expr *init, LValue dest) {
     // of atomic type.
     bool Zeroed = false;
     if (!init->getType()->isAtomicType()) {
-      Zeroed = atomics.emitMemSetZeroIfNecessary(dest);
-      dest = atomics.projectValue(dest);
+      Zeroed = atomics.emitMemSetZeroIfNecessary();
+      dest = atomics.projectValue();
     }
 
     // Evaluate the expression directly into the destination.
diff --git a/lib/CodeGen/CGBlocks.cpp b/lib/CodeGen/CGBlocks.cpp
index b98460a9ddd8..3fd344c389a5 100644
--- a/lib/CodeGen/CGBlocks.cpp
+++ b/lib/CodeGen/CGBlocks.cpp
@@ -621,8 +621,8 @@ static void enterBlockScope(CodeGenFunction &CGF, BlockDecl *block) {
     }
 
     // GEP down to the address.
-    llvm::Value *addr = CGF.Builder.CreateStructGEP(blockInfo.Address,
-                                                    capture.getIndex());
+    llvm::Value *addr = CGF.Builder.CreateStructGEP(
+        blockInfo.StructureType, blockInfo.Address, capture.getIndex());
 
     // We can use that GEP as the dominating IP.
     if (!blockInfo.DominatingIP)
@@ -721,6 +721,7 @@ llvm::Value *CodeGenFunction::EmitBlockLiteral(const CGBlockInfo &blockInfo) {
   // Build the block descriptor.
   llvm::Constant *descriptor = buildBlockDescriptor(CGM, blockInfo);
 
+  llvm::Type *blockTy = blockInfo.StructureType;
   llvm::AllocaInst *blockAddr = blockInfo.Address;
   assert(blockAddr && "block has no address!");
 
@@ -732,14 +733,17 @@ llvm::Value *CodeGenFunction::EmitBlockLiteral(const CGBlockInfo &blockInfo) {
   if (blockInfo.UsesStret) flags |= BLOCK_USE_STRET;
 
   // Initialize the block literal.
-  Builder.CreateStore(isa, Builder.CreateStructGEP(blockAddr, 0, "block.isa"));
-  Builder.CreateStore(llvm::ConstantInt::get(IntTy, flags.getBitMask()),
-                      Builder.CreateStructGEP(blockAddr, 1, "block.flags"));
-  Builder.CreateStore(llvm::ConstantInt::get(IntTy, 0),
-                      Builder.CreateStructGEP(blockAddr, 2, "block.reserved"));
-  Builder.CreateStore(blockFn, Builder.CreateStructGEP(blockAddr, 3,
-                                                       "block.invoke"));
-  Builder.CreateStore(descriptor, Builder.CreateStructGEP(blockAddr, 4,
+  Builder.CreateStore(
+      isa, Builder.CreateStructGEP(blockTy, blockAddr, 0, "block.isa"));
+  Builder.CreateStore(
+      llvm::ConstantInt::get(IntTy, flags.getBitMask()),
+      Builder.CreateStructGEP(blockTy, blockAddr, 1, "block.flags"));
+  Builder.CreateStore(
+      llvm::ConstantInt::get(IntTy, 0),
+      Builder.CreateStructGEP(blockTy, blockAddr, 2, "block.reserved"));
+  Builder.CreateStore(
+      blockFn, Builder.CreateStructGEP(blockTy, blockAddr, 3, "block.invoke"));
+  Builder.CreateStore(descriptor, Builder.CreateStructGEP(blockTy, blockAddr, 4,
                                                           "block.descriptor"));
 
   // Finally, capture all the values into the block.
@@ -747,9 +751,8 @@ llvm::Value *CodeGenFunction::EmitBlockLiteral(const CGBlockInfo &blockInfo) {
 
   // First, 'this'.
   if (blockDecl->capturesCXXThis()) {
-    llvm::Value *addr = Builder.CreateStructGEP(blockAddr,
-                                                blockInfo.CXXThisIndex,
-                                                "block.captured-this.addr");
+    llvm::Value *addr = Builder.CreateStructGEP(
+        blockTy, blockAddr, blockInfo.CXXThisIndex, "block.captured-this.addr");
     Builder.CreateStore(LoadCXXThis(), addr);
   }
 
@@ -766,9 +769,8 @@ llvm::Value *CodeGenFunction::EmitBlockLiteral(const CGBlockInfo &blockInfo) {
 
     // This will be a [[type]]*, except that a byref entry will just be
     // an i8**.
-    llvm::Value *blockField =
-      Builder.CreateStructGEP(blockAddr, capture.getIndex(),
-                              "block.captured");
+    llvm::Value *blockField = Builder.CreateStructGEP(
+        blockTy, blockAddr, capture.getIndex(), "block.captured");
 
     // Compute the address of the thing we're going to move into the
     // block literal.
@@ -779,7 +781,7 @@ llvm::Value *CodeGenFunction::EmitBlockLiteral(const CGBlockInfo &blockInfo) {
         BlockInfo->getCapture(variable);
 
       // This is a [[type]]*, except that a byref entry wil just be an i8**.
-      src = Builder.CreateStructGEP(LoadBlockStruct(),
+      src = Builder.CreateStructGEP(BlockInfo->StructureType, LoadBlockStruct(),
                                     enclosingCapture.getIndex(),
                                     "block.capture.addr");
     } else if (blockDecl->isConversionFromLambda()) {
@@ -964,7 +966,8 @@ RValue CodeGenFunction::EmitBlockCallExpr(const CallExpr *E,
     Builder.CreateBitCast(Callee, BlockLiteralTy, "block.literal");
 
   // Get the function pointer from the literal.
-  llvm::Value *FuncPtr = Builder.CreateStructGEP(BlockLiteral, 3);
+  llvm::Value *FuncPtr = Builder.CreateStructGEP(
+      CGM.getGenericBlockLiteralType(), BlockLiteral, 3);
 
   BlockLiteral = Builder.CreateBitCast(BlockLiteral, VoidPtrTy);
 
@@ -1004,26 +1007,27 @@ llvm::Value *CodeGenFunction::GetAddrOfBlockDecl(const VarDecl *variable,
   if (capture.isConstant()) return LocalDeclMap[variable];
 
   llvm::Value *addr =
-    Builder.CreateStructGEP(LoadBlockStruct(), capture.getIndex(),
-                            "block.capture.addr");
+      Builder.CreateStructGEP(BlockInfo->StructureType, LoadBlockStruct(),
+                              capture.getIndex(), "block.capture.addr");
 
   if (isByRef) {
     // addr should be a void** right now.  Load, then cast the result
     // to byref*.
 
     addr = Builder.CreateLoad(addr);
-    llvm::PointerType *byrefPointerType
-      = llvm::PointerType::get(BuildByRefType(variable), 0);
+    auto *byrefType = BuildByRefType(variable);
+    llvm::PointerType *byrefPointerType = llvm::PointerType::get(byrefType, 0);
     addr = Builder.CreateBitCast(addr, byrefPointerType,
                                  "byref.addr");
 
     // Follow the forwarding pointer.
-    addr = Builder.CreateStructGEP(addr, 1, "byref.forwarding");
+    addr = Builder.CreateStructGEP(byrefType, addr, 1, "byref.forwarding");
     addr = Builder.CreateLoad(addr, "byref.addr.forwarded");
 
     // Cast back to byref* and GEP over to the actual object.
     addr = Builder.CreateBitCast(addr, byrefPointerType);
-    addr = Builder.CreateStructGEP(addr, getByRefValueLLVMField(variable), 
+    addr = Builder.CreateStructGEP(byrefType, addr,
+                                   getByRefValueLLVMField(variable).second,
                                    variable->getNameAsString());
   }
 
@@ -1136,8 +1140,7 @@ CodeGenFunction::GenerateBlockFunction(GlobalDecl GD,
   args.push_back(&selfDecl);
 
   // Now add the rest of the parameters.
-  for (auto i : blockDecl->params())
-    args.push_back(i);
+  args.append(blockDecl->param_begin(), blockDecl->param_end());
 
   // Create the function declaration.
   const FunctionProtoType *fnType = blockInfo.getBlockExpr()->getFunctionType();
@@ -1178,7 +1181,7 @@ CodeGenFunction::GenerateBlockFunction(GlobalDecl GD,
     Alloca->setAlignment(Align);
     // Set the DebugLocation to empty, so the store is recognized as a
     // frame setup instruction by llvm::DwarfDebug::beginFunction().
-    ApplyDebugLocation NL(*this);
+    auto NL = ApplyDebugLocation::CreateEmpty(*this);
     Builder.CreateAlignedStore(BlockPointer, Alloca, Align);
     BlockPointerDbgLoc = Alloca;
   }
@@ -1186,9 +1189,9 @@ CodeGenFunction::GenerateBlockFunction(GlobalDecl GD,
   // If we have a C++ 'this' reference, go ahead and force it into
   // existence now.
   if (blockDecl->capturesCXXThis()) {
-    llvm::Value *addr = Builder.CreateStructGEP(BlockPointer,
-                                                blockInfo.CXXThisIndex,
-                                                "block.captured-this");
+    llvm::Value *addr =
+        Builder.CreateStructGEP(blockInfo.StructureType, BlockPointer,
+                                blockInfo.CXXThisIndex, "block.captured-this");
     CXXThisValue = Builder.CreateLoad(addr, "this");
   }
 
@@ -1218,8 +1221,7 @@ CodeGenFunction::GenerateBlockFunction(GlobalDecl GD,
     EmitLambdaBlockInvokeBody();
   else {
     PGO.assignRegionCounters(blockDecl, fn);
-    RegionCounter Cnt = getPGORegionCounter(blockDecl->getBody());
-    Cnt.beginRegion(Builder);
+    incrementProfileCounter(blockDecl->getBody());
     EmitStmt(blockDecl->getBody());
   }
 
@@ -1328,11 +1330,10 @@ CodeGenFunction::GenerateCopyHelperFunction(const CGBlockInfo &blockInfo) {
                                           nullptr, SC_Static,
                                           false,
                                           false);
-  // Create a scope with an artificial location for the body of this function.
-  ApplyDebugLocation NL(*this);
+  auto NL = ApplyDebugLocation::CreateEmpty(*this);
   StartFunction(FD, C.VoidTy, Fn, FI, args);
-  ArtificialLocation AL(*this);
-
+  // Create a scope with an artificial location for the body of this function.
+  auto AL = ApplyDebugLocation::CreateArtificial(*this);
   llvm::Type *structPtrTy = blockInfo.StructureType->getPointerTo();
 
   llvm::Value *src = GetAddrOfLocalVar(&srcDecl);
@@ -1404,8 +1405,10 @@ CodeGenFunction::GenerateCopyHelperFunction(const CGBlockInfo &blockInfo) {
     }
 
     unsigned index = capture.getIndex();
-    llvm::Value *srcField = Builder.CreateStructGEP(src, index);
-    llvm::Value *dstField = Builder.CreateStructGEP(dst, index);
+    llvm::Value *srcField =
+        Builder.CreateStructGEP(blockInfo.StructureType, src, index);
+    llvm::Value *dstField =
+        Builder.CreateStructGEP(blockInfo.StructureType, dst, index);
 
     // If there's an explicit copy expression, we do that.
     if (copyExpr) {
@@ -1500,9 +1503,9 @@ CodeGenFunction::GenerateDestroyHelperFunction(const CGBlockInfo &blockInfo) {
                                           nullptr, SC_Static,
                                           false, false);
   // Create a scope with an artificial location for the body of this function.
-  ApplyDebugLocation NL(*this);
+  auto NL = ApplyDebugLocation::CreateEmpty(*this);
   StartFunction(FD, C.VoidTy, Fn, FI, args);
-  ArtificialLocation AL(*this);
+  auto AL = ApplyDebugLocation::CreateArtificial(*this);
 
   llvm::Type *structPtrTy = blockInfo.StructureType->getPointerTo();
 
@@ -1562,7 +1565,8 @@ CodeGenFunction::GenerateDestroyHelperFunction(const CGBlockInfo &blockInfo) {
     }
 
     unsigned index = capture.getIndex();
-    llvm::Value *srcField = Builder.CreateStructGEP(src, index);
+    llvm::Value *srcField =
+        Builder.CreateStructGEP(blockInfo.StructureType, src, index);
 
     // If there's an explicit copy expression, we do that.
     if (dtor) {
@@ -1801,13 +1805,15 @@ generateByrefCopyHelper(CodeGenFunction &CGF,
     llvm::Value *destField = CGF.GetAddrOfLocalVar(&dst);
     destField = CGF.Builder.CreateLoad(destField);
     destField = CGF.Builder.CreateBitCast(destField, byrefPtrType);
-    destField = CGF.Builder.CreateStructGEP(destField, valueFieldIndex, "x");
+    destField = CGF.Builder.CreateStructGEP(&byrefType, destField,
+                                            valueFieldIndex, "x");
 
     // src->x
     llvm::Value *srcField = CGF.GetAddrOfLocalVar(&src);
     srcField = CGF.Builder.CreateLoad(srcField);
     srcField = CGF.Builder.CreateBitCast(srcField, byrefPtrType);
-    srcField = CGF.Builder.CreateStructGEP(srcField, valueFieldIndex, "x");
+    srcField =
+        CGF.Builder.CreateStructGEP(&byrefType, srcField, valueFieldIndex, "x");
 
     byrefInfo.emitCopy(CGF, destField, srcField);
   }  
@@ -1868,7 +1874,7 @@ generateByrefDisposeHelper(CodeGenFunction &CGF,
     llvm::Value *V = CGF.GetAddrOfLocalVar(&src);
     V = CGF.Builder.CreateLoad(V);
     V = CGF.Builder.CreateBitCast(V, byrefType.getPointerTo(0));
-    V = CGF.Builder.CreateStructGEP(V, byrefValueIndex, "x");
+    V = CGF.Builder.CreateStructGEP(&byrefType, V, byrefValueIndex, "x");
 
     byrefInfo.emitDispose(CGF, V);
   }
@@ -1925,7 +1931,7 @@ CodeGenFunction::buildByrefHelpers(llvm::StructType &byrefType,
   const VarDecl &var = *emission.Variable;
   QualType type = var.getType();
 
-  unsigned byrefValueIndex = getByRefValueLLVMField(&var);
+  unsigned byrefValueIndex = getByRefValueLLVMField(&var).second;
 
   if (const CXXRecordDecl *record = type->getAsCXXRecordDecl()) {
     const Expr *copyExpr = CGM.getContext().getBlockVarCopyInits(&var);
@@ -1995,18 +2001,20 @@ CodeGenFunction::buildByrefHelpers(llvm::StructType &byrefType,
   return ::buildByrefHelpers(CGM, byrefType, byrefValueIndex, byrefInfo);
 }
 
-unsigned CodeGenFunction::getByRefValueLLVMField(const ValueDecl *VD) const {
+std::pair<llvm::Type *, unsigned>
+CodeGenFunction::getByRefValueLLVMField(const ValueDecl *VD) const {
   assert(ByRefValueInfo.count(VD) && "Did not find value!");
-  
-  return ByRefValueInfo.find(VD)->second.second;
+
+  return ByRefValueInfo.find(VD)->second;
 }
 
 llvm::Value *CodeGenFunction::BuildBlockByrefAddress(llvm::Value *BaseAddr,
                                                      const VarDecl *V) {
-  llvm::Value *Loc = Builder.CreateStructGEP(BaseAddr, 1, "forwarding");
+  auto P = getByRefValueLLVMField(V);
+  llvm::Value *Loc =
+      Builder.CreateStructGEP(P.first, BaseAddr, 1, "forwarding");
   Loc = Builder.CreateLoad(Loc);
-  Loc = Builder.CreateStructGEP(Loc, getByRefValueLLVMField(V),
-                                V->getNameAsString());
+  Loc = Builder.CreateStructGEP(P.first, Loc, P.second, V->getNameAsString());
   return Loc;
 }
 
@@ -2143,11 +2151,12 @@ void CodeGenFunction::emitByrefStructureInit(const AutoVarEmission &emission) {
   if (type.isObjCGCWeak())
     isa = 1;
   V = Builder.CreateIntToPtr(Builder.getInt32(isa), Int8PtrTy, "isa");
-  Builder.CreateStore(V, Builder.CreateStructGEP(addr, 0, "byref.isa"));
+  Builder.CreateStore(V,
+                      Builder.CreateStructGEP(nullptr, addr, 0, "byref.isa"));
 
   // Store the address of the variable into its own forwarding pointer.
-  Builder.CreateStore(addr,
-                      Builder.CreateStructGEP(addr, 1, "byref.forwarding"));
+  Builder.CreateStore(
+      addr, Builder.CreateStructGEP(nullptr, addr, 1, "byref.forwarding"));
 
   // Blocks ABI:
   //   c) the flags field is set to either 0 if no helper functions are
@@ -2193,25 +2202,26 @@ void CodeGenFunction::emitByrefStructureInit(const AutoVarEmission &emission) {
       printf("\n");
     }
   }
-  
+
   Builder.CreateStore(llvm::ConstantInt::get(IntTy, flags.getBitMask()),
-                      Builder.CreateStructGEP(addr, 2, "byref.flags"));
+                      Builder.CreateStructGEP(nullptr, addr, 2, "byref.flags"));
 
   CharUnits byrefSize = CGM.GetTargetTypeStoreSize(byrefType);
   V = llvm::ConstantInt::get(IntTy, byrefSize.getQuantity());
-  Builder.CreateStore(V, Builder.CreateStructGEP(addr, 3, "byref.size"));
+  Builder.CreateStore(V,
+                      Builder.CreateStructGEP(nullptr, addr, 3, "byref.size"));
 
   if (helpers) {
-    llvm::Value *copy_helper = Builder.CreateStructGEP(addr, 4);
+    llvm::Value *copy_helper = Builder.CreateStructGEP(nullptr, addr, 4);
     Builder.CreateStore(helpers->CopyHelper, copy_helper);
 
-    llvm::Value *destroy_helper = Builder.CreateStructGEP(addr, 5);
+    llvm::Value *destroy_helper = Builder.CreateStructGEP(nullptr, addr, 5);
     Builder.CreateStore(helpers->DisposeHelper, destroy_helper);
   }
   if (ByRefHasLifetime && HasByrefExtendedLayout) {
     llvm::Constant* ByrefLayoutInfo = CGM.getObjCRuntime().BuildByrefLayout(CGM, type);
-    llvm::Value *ByrefInfoAddr = Builder.CreateStructGEP(addr, helpers ? 6 : 4,
-                                                         "byref.layout");
+    llvm::Value *ByrefInfoAddr =
+        Builder.CreateStructGEP(nullptr, addr, helpers ? 6 : 4, "byref.layout");
     // cast destination to pointer to source type.
     llvm::Type *DesTy = ByrefLayoutInfo->getType();
     DesTy = DesTy->getPointerTo();
diff --git a/lib/CodeGen/CGBuilder.h b/lib/CodeGen/CGBuilder.h
index 72ba4faa3c7c..6610659131f7 100644
--- a/lib/CodeGen/CGBuilder.h
+++ b/lib/CodeGen/CGBuilder.h
@@ -33,7 +33,7 @@ protected:
                     llvm::BasicBlock *BB,
                     llvm::BasicBlock::iterator InsertPt) const;
 private:
-  void operator=(const CGBuilderInserter &) LLVM_DELETED_FUNCTION;
+  void operator=(const CGBuilderInserter &) = delete;
 
   CodeGenFunction *CGF;
 };
diff --git a/lib/CodeGen/CGBuiltin.cpp b/lib/CodeGen/CGBuiltin.cpp
index 635e34207de7..272baac80897 100644
--- a/lib/CodeGen/CGBuiltin.cpp
+++ b/lib/CodeGen/CGBuiltin.cpp
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "CodeGenFunction.h"
+#include "CGCXXABI.h"
 #include "CGObjCRuntime.h"
 #include "CodeGenModule.h"
 #include "TargetInfo.h"
@@ -21,9 +22,11 @@
 #include "clang/Basic/TargetInfo.h"
 #include "clang/CodeGen/CGFunctionInfo.h"
 #include "llvm/ADT/StringExtras.h"
+#include "llvm/IR/CallSite.h"
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/InlineAsm.h"
 #include "llvm/IR/Intrinsics.h"
+#include <sstream>
 
 using namespace clang;
 using namespace CodeGen;
@@ -156,6 +159,27 @@ static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) {
   return Call;
 }
 
+/// Emit the computation of the sign bit for a floating point value. Returns
+/// the i1 sign bit value.
+static Value *EmitSignBit(CodeGenFunction &CGF, Value *V) {
+  LLVMContext &C = CGF.CGM.getLLVMContext();
+
+  llvm::Type *Ty = V->getType();
+  int Width = Ty->getPrimitiveSizeInBits();
+  llvm::Type *IntTy = llvm::IntegerType::get(C, Width);
+  V = CGF.Builder.CreateBitCast(V, IntTy);
+  if (Ty->isPPC_FP128Ty()) {
+    // The higher-order double comes first, and so we need to truncate the
+    // pair to extract the overall sign. The order of the pair is the same
+    // in both little- and big-Endian modes.
+    Width >>= 1;
+    IntTy = llvm::IntegerType::get(C, Width);
+    V = CGF.Builder.CreateTrunc(V, IntTy);
+  }
+  Value *Zero = llvm::Constant::getNullValue(IntTy);
+  return CGF.Builder.CreateICmpSLT(V, Zero);
+}
+
 static RValue emitLibraryCall(CodeGenFunction &CGF, const FunctionDecl *Fn,
                               const CallExpr *E, llvm::Value *calleeValue) {
   return CGF.EmitCall(E->getCallee()->getType(), calleeValue, E,
@@ -181,7 +205,7 @@ static llvm::Value *EmitOverflowIntrinsic(CodeGenFunction &CGF,
          "arguments have the same integer width?)");
 
   llvm::Value *Callee = CGF.CGM.getIntrinsic(IntrinsicID, X->getType());
-  llvm::Value *Tmp = CGF.Builder.CreateCall2(Callee, X, Y);
+  llvm::Value *Tmp = CGF.Builder.CreateCall(Callee, {X, Y});
   Carry = CGF.Builder.CreateExtractValue(Tmp, 1);
   return CGF.Builder.CreateExtractValue(Tmp, 0);
 }
@@ -230,8 +254,8 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
 
     DstPtr = Builder.CreateBitCast(DstPtr, Type);
     SrcPtr = Builder.CreateBitCast(SrcPtr, Type);
-    return RValue::get(Builder.CreateCall2(CGM.getIntrinsic(Intrinsic::vacopy),
-                                           DstPtr, SrcPtr));
+    return RValue::get(Builder.CreateCall(CGM.getIntrinsic(Intrinsic::vacopy),
+                                          {DstPtr, SrcPtr}));
   }
   case Builtin::BI__builtin_abs:
   case Builtin::BI__builtin_labs:
@@ -309,7 +333,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
 
     llvm::Type *ResultType = ConvertType(E->getType());
     Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef());
-    Value *Result = Builder.CreateCall2(F, ArgValue, ZeroUndef);
+    Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef});
     if (Result->getType() != ResultType)
       Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
                                      "cast");
@@ -326,7 +350,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
 
     llvm::Type *ResultType = ConvertType(E->getType());
     Value *ZeroUndef = Builder.getInt1(getTarget().isCLZForZeroUndef());
-    Value *Result = Builder.CreateCall2(F, ArgValue, ZeroUndef);
+    Value *Result = Builder.CreateCall(F, {ArgValue, ZeroUndef});
     if (Result->getType() != ResultType)
       Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
                                      "cast");
@@ -342,9 +366,9 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     Value *F = CGM.getIntrinsic(Intrinsic::cttz, ArgType);
 
     llvm::Type *ResultType = ConvertType(E->getType());
-    Value *Tmp = Builder.CreateAdd(Builder.CreateCall2(F, ArgValue,
-                                                       Builder.getTrue()),
-                                   llvm::ConstantInt::get(ArgType, 1));
+    Value *Tmp =
+        Builder.CreateAdd(Builder.CreateCall(F, {ArgValue, Builder.getTrue()}),
+                          llvm::ConstantInt::get(ArgType, 1));
     Value *Zero = llvm::Constant::getNullValue(ArgType);
     Value *IsZero = Builder.CreateICmpEQ(ArgValue, Zero, "iszero");
     Value *Result = Builder.CreateSelect(IsZero, Zero, Tmp, "ffs");
@@ -389,11 +413,16 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     Value *ArgValue = EmitScalarExpr(E->getArg(0));
     llvm::Type *ArgType = ArgValue->getType();
 
-    Value *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType);
     Value *ExpectedValue = EmitScalarExpr(E->getArg(1));
+    // Don't generate llvm.expect on -O0 as the backend won't use it for
+    // anything.
+    // Note, we still IRGen ExpectedValue because it could have side-effects.
+    if (CGM.getCodeGenOpts().OptimizationLevel == 0)
+      return RValue::get(ArgValue);
 
-    Value *Result = Builder.CreateCall2(FnExpect, ArgValue, ExpectedValue,
-                                        "expval");
+    Value *FnExpect = CGM.getIntrinsic(Intrinsic::expect, ArgType);
+    Value *Result =
+        Builder.CreateCall(FnExpect, {ArgValue, ExpectedValue}, "expval");
     return RValue::get(Result);
   }
   case Builtin::BI__builtin_assume_aligned: {
@@ -444,7 +473,8 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     // FIXME: Get right address space.
     llvm::Type *Tys[] = { ResType, Builder.getInt8PtrTy(0) };
     Value *F = CGM.getIntrinsic(Intrinsic::objectsize, Tys);
-    return RValue::get(Builder.CreateCall2(F, EmitScalarExpr(E->getArg(0)),CI));
+    return RValue::get(
+        Builder.CreateCall(F, {EmitScalarExpr(E->getArg(0)), CI}));
   }
   case Builtin::BI__builtin_prefetch: {
     Value *Locality, *RW, *Address = EmitScalarExpr(E->getArg(0));
@@ -455,25 +485,25 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
       llvm::ConstantInt::get(Int32Ty, 3);
     Value *Data = llvm::ConstantInt::get(Int32Ty, 1);
     Value *F = CGM.getIntrinsic(Intrinsic::prefetch);
-    return RValue::get(Builder.CreateCall4(F, Address, RW, Locality, Data));
+    return RValue::get(Builder.CreateCall(F, {Address, RW, Locality, Data}));
   }
   case Builtin::BI__builtin_readcyclecounter: {
     Value *F = CGM.getIntrinsic(Intrinsic::readcyclecounter);
-    return RValue::get(Builder.CreateCall(F));
+    return RValue::get(Builder.CreateCall(F, {}));
   }
   case Builtin::BI__builtin___clear_cache: {
     Value *Begin = EmitScalarExpr(E->getArg(0));
     Value *End = EmitScalarExpr(E->getArg(1));
     Value *F = CGM.getIntrinsic(Intrinsic::clear_cache);
-    return RValue::get(Builder.CreateCall2(F, Begin, End));
+    return RValue::get(Builder.CreateCall(F, {Begin, End}));
   }
   case Builtin::BI__builtin_trap: {
     Value *F = CGM.getIntrinsic(Intrinsic::trap);
-    return RValue::get(Builder.CreateCall(F));
+    return RValue::get(Builder.CreateCall(F, {}));
   }
   case Builtin::BI__debugbreak: {
     Value *F = CGM.getIntrinsic(Intrinsic::debugtrap);
-    return RValue::get(Builder.CreateCall(F));
+    return RValue::get(Builder.CreateCall(F, {}));
   }
   case Builtin::BI__builtin_unreachable: {
     if (SanOpts.has(SanitizerKind::Unreachable)) {
@@ -498,7 +528,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     Value *Exponent = EmitScalarExpr(E->getArg(1));
     llvm::Type *ArgType = Base->getType();
     Value *F = CGM.getIntrinsic(Intrinsic::powi, ArgType);
-    return RValue::get(Builder.CreateCall2(F, Base, Exponent));
+    return RValue::get(Builder.CreateCall(F, {Base, Exponent}));
   }
 
   case Builtin::BI__builtin_isgreater:
@@ -551,8 +581,22 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     return RValue::get(Builder.CreateZExt(V, ConvertType(E->getType())));
   }
 
-  // TODO: BI__builtin_isinf_sign
-  //   isinf_sign(x) -> isinf(x) ? (signbit(x) ? -1 : 1) : 0
+  case Builtin::BI__builtin_isinf_sign: {
+    // isinf_sign(x) -> fabs(x) == infinity ? (signbit(x) ? -1 : 1) : 0
+    Value *Arg = EmitScalarExpr(E->getArg(0));
+    Value *AbsArg = EmitFAbs(*this, Arg);
+    Value *IsInf = Builder.CreateFCmpOEQ(
+        AbsArg, ConstantFP::getInfinity(Arg->getType()), "isinf");
+    Value *IsNeg = EmitSignBit(*this, Arg);
+
+    llvm::Type *IntTy = ConvertType(E->getType());
+    Value *Zero = Constant::getNullValue(IntTy);
+    Value *One = ConstantInt::get(IntTy, 1);
+    Value *NegativeOne = ConstantInt::get(IntTy, -1);
+    Value *SignResult = Builder.CreateSelect(IsNeg, NegativeOne, One);
+    Value *Result = Builder.CreateSelect(IsInf, SignResult, Zero);
+    return RValue::get(Result);
+  }
 
   case Builtin::BI__builtin_isnormal: {
     // isnormal(x) --> x == x && fabsf(x) < infinity && fabsf(x) >= float_min
@@ -815,7 +859,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     Value *F = CGM.getIntrinsic(IntTy->getBitWidth() == 32
                                   ? Intrinsic::eh_return_i32
                                   : Intrinsic::eh_return_i64);
-    Builder.CreateCall2(F, Int, Ptr);
+    Builder.CreateCall(F, {Int, Ptr});
     Builder.CreateUnreachable();
 
     // We do need to preserve an insertion point.
@@ -825,7 +869,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
   }
   case Builtin::BI__builtin_unwind_init: {
     Value *F = CGM.getIntrinsic(Intrinsic::eh_unwind_init);
-    return RValue::get(Builder.CreateCall(F));
+    return RValue::get(Builder.CreateCall(F, {}));
   }
   case Builtin::BI__builtin_extend_pointer: {
     // Extends a pointer to the size of an _Unwind_Word, which is
@@ -864,7 +908,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
 
     // Store the stack pointer to the setjmp buffer.
     Value *StackAddr =
-      Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave));
+        Builder.CreateCall(CGM.getIntrinsic(Intrinsic::stacksave), {});
     Value *StackSaveSlot =
       Builder.CreateGEP(Buf, ConstantInt::get(Int32Ty, 2));
     Builder.CreateStore(StackAddr, StackSaveSlot);
@@ -1357,6 +1401,9 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     return RValue::get(Builder.CreateCall(F, Arg0));
   }
 
+  case Builtin::BI__builtin_pow:
+  case Builtin::BI__builtin_powf:
+  case Builtin::BI__builtin_powl:
   case Builtin::BIpow:
   case Builtin::BIpowf:
   case Builtin::BIpowl: {
@@ -1367,7 +1414,7 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     Value *Exponent = EmitScalarExpr(E->getArg(1));
     llvm::Type *ArgType = Base->getType();
     Value *F = CGM.getIntrinsic(Intrinsic::pow, ArgType);
-    return RValue::get(Builder.CreateCall2(F, Base, Exponent));
+    return RValue::get(Builder.CreateCall(F, {Base, Exponent}));
   }
 
   case Builtin::BIfma:
@@ -1380,32 +1427,17 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     Value *FirstArg = EmitScalarExpr(E->getArg(0));
     llvm::Type *ArgType = FirstArg->getType();
     Value *F = CGM.getIntrinsic(Intrinsic::fma, ArgType);
-    return RValue::get(Builder.CreateCall3(F, FirstArg,
-                                              EmitScalarExpr(E->getArg(1)),
-                                              EmitScalarExpr(E->getArg(2))));
+    return RValue::get(
+        Builder.CreateCall(F, {FirstArg, EmitScalarExpr(E->getArg(1)),
+                               EmitScalarExpr(E->getArg(2))}));
   }
 
   case Builtin::BI__builtin_signbit:
   case Builtin::BI__builtin_signbitf:
   case Builtin::BI__builtin_signbitl: {
-    LLVMContext &C = CGM.getLLVMContext();
-
-    Value *Arg = EmitScalarExpr(E->getArg(0));
-    llvm::Type *ArgTy = Arg->getType();
-    int ArgWidth = ArgTy->getPrimitiveSizeInBits();
-    llvm::Type *ArgIntTy = llvm::IntegerType::get(C, ArgWidth);
-    Value *BCArg = Builder.CreateBitCast(Arg, ArgIntTy);
-    if (ArgTy->isPPC_FP128Ty()) {
-      // The higher-order double comes first, and so we need to truncate the
-      // pair to extract the overall sign. The order of the pair is the same
-      // in both little- and big-Endian modes.
-      ArgWidth >>= 1;
-      ArgIntTy = llvm::IntegerType::get(C, ArgWidth);
-      BCArg = Builder.CreateTrunc(BCArg, ArgIntTy);
-    }
-    Value *ZeroCmp = llvm::Constant::getNullValue(ArgIntTy);
-    Value *Result = Builder.CreateICmpSLT(BCArg, ZeroCmp);
-    return RValue::get(Builder.CreateZExt(Result, ConvertType(E->getType())));
+    return RValue::get(
+        Builder.CreateZExt(EmitSignBit(*this, EmitScalarExpr(E->getArg(0))),
+                           ConvertType(E->getType())));
   }
   case Builtin::BI__builtin_annotation: {
     llvm::Value *AnnVal = EmitScalarExpr(E->getArg(0));
@@ -1650,6 +1682,76 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
         Builder.CreateAlignedLoad(IntToPtr, /*Align=*/4, /*isVolatile=*/true);
     return RValue::get(Load);
   }
+
+  case Builtin::BI__exception_code:
+  case Builtin::BI_exception_code:
+    return RValue::get(EmitSEHExceptionCode());
+  case Builtin::BI__exception_info:
+  case Builtin::BI_exception_info:
+    return RValue::get(EmitSEHExceptionInfo());
+  case Builtin::BI__abnormal_termination:
+  case Builtin::BI_abnormal_termination:
+    return RValue::get(EmitSEHAbnormalTermination());
+  case Builtin::BI_setjmpex: {
+    if (getTarget().getTriple().isOSMSVCRT()) {
+      llvm::Type *ArgTypes[] = {Int8PtrTy, Int8PtrTy};
+      llvm::AttributeSet ReturnsTwiceAttr =
+          AttributeSet::get(getLLVMContext(), llvm::AttributeSet::FunctionIndex,
+                            llvm::Attribute::ReturnsTwice);
+      llvm::Constant *SetJmpEx = CGM.CreateRuntimeFunction(
+          llvm::FunctionType::get(IntTy, ArgTypes, /*isVarArg=*/false),
+          "_setjmpex", ReturnsTwiceAttr);
+      llvm::Value *Buf = Builder.CreateBitOrPointerCast(
+          EmitScalarExpr(E->getArg(0)), Int8PtrTy);
+      llvm::Value *FrameAddr =
+          Builder.CreateCall(CGM.getIntrinsic(Intrinsic::frameaddress),
+                             ConstantInt::get(Int32Ty, 0));
+      llvm::Value *Args[] = {Buf, FrameAddr};
+      llvm::CallSite CS = EmitRuntimeCallOrInvoke(SetJmpEx, Args);
+      CS.setAttributes(ReturnsTwiceAttr);
+      return RValue::get(CS.getInstruction());
+    }
+    break;
+  }
+  case Builtin::BI_setjmp: {
+    if (getTarget().getTriple().isOSMSVCRT()) {
+      llvm::AttributeSet ReturnsTwiceAttr =
+          AttributeSet::get(getLLVMContext(), llvm::AttributeSet::FunctionIndex,
+                            llvm::Attribute::ReturnsTwice);
+      llvm::Value *Buf = Builder.CreateBitOrPointerCast(
+          EmitScalarExpr(E->getArg(0)), Int8PtrTy);
+      llvm::CallSite CS;
+      if (getTarget().getTriple().getArch() == llvm::Triple::x86) {
+        llvm::Type *ArgTypes[] = {Int8PtrTy, IntTy};
+        llvm::Constant *SetJmp3 = CGM.CreateRuntimeFunction(
+            llvm::FunctionType::get(IntTy, ArgTypes, /*isVarArg=*/true),
+            "_setjmp3", ReturnsTwiceAttr);
+        llvm::Value *Count = ConstantInt::get(IntTy, 0);
+        llvm::Value *Args[] = {Buf, Count};
+        CS = EmitRuntimeCallOrInvoke(SetJmp3, Args);
+      } else {
+        llvm::Type *ArgTypes[] = {Int8PtrTy, Int8PtrTy};
+        llvm::Constant *SetJmp = CGM.CreateRuntimeFunction(
+            llvm::FunctionType::get(IntTy, ArgTypes, /*isVarArg=*/false),
+            "_setjmp", ReturnsTwiceAttr);
+        llvm::Value *FrameAddr =
+            Builder.CreateCall(CGM.getIntrinsic(Intrinsic::frameaddress),
+                               ConstantInt::get(Int32Ty, 0));
+        llvm::Value *Args[] = {Buf, FrameAddr};
+        CS = EmitRuntimeCallOrInvoke(SetJmp, Args);
+      }
+      CS.setAttributes(ReturnsTwiceAttr);
+      return RValue::get(CS.getInstruction());
+    }
+    break;
+  }
+
+  case Builtin::BI__GetExceptionInfo: {
+    if (llvm::GlobalVariable *GV =
+            CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType()))
+      return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy));
+    break;
+  }
   }
 
   // If this is an alias for a lib function (e.g. __builtin_sin), emit
@@ -1764,6 +1866,8 @@ Value *CodeGenFunction::EmitTargetBuiltinExpr(unsigned BuiltinID,
   case llvm::Triple::r600:
   case llvm::Triple::amdgcn:
     return EmitR600BuiltinExpr(BuiltinID, E);
+  case llvm::Triple::systemz:
+    return EmitSystemZBuiltinExpr(BuiltinID, E);
   default:
     return nullptr;
   }
@@ -2534,7 +2638,7 @@ Function *CodeGenFunction::LookupNeonLLVMIntrinsic(unsigned IntrinsicID,
   // Return type.
   SmallVector<llvm::Type *, 3> Tys;
   if (Modifier & AddRetType) {
-    llvm::Type *Ty = ConvertType(E->getCallReturnType());
+    llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext()));
     if (Modifier & VectorizeRetType)
       Ty = llvm::VectorType::get(
           Ty, VectorSize ? VectorSize / Ty->getPrimitiveSizeInBits() : 1);
@@ -2812,7 +2916,7 @@ Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
     Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
 
     // NEON intrinsic puts accumulator first, unlike the LLVM fma.
-    return Builder.CreateCall3(F, Ops[1], Ops[2], Ops[0]);
+    return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]});
   }
   case NEON::BI__builtin_neon_vld1_v:
   case NEON::BI__builtin_neon_vld1q_v:
@@ -2825,7 +2929,7 @@ Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
   case NEON::BI__builtin_neon_vld4_v:
   case NEON::BI__builtin_neon_vld4q_v: {
     Function *F = CGM.getIntrinsic(LLVMIntrinsic, Ty);
-    Ops[1] = Builder.CreateCall2(F, Ops[1], Align, NameHint);
+    Ops[1] = Builder.CreateCall(F, {Ops[1], Align}, NameHint);
     Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
     Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
     return Builder.CreateStore(Ops[1], Ops[0]);
@@ -3004,7 +3108,7 @@ Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
         Indices.push_back(Builder.getInt32(i+vi));
         Indices.push_back(Builder.getInt32(i+e+vi));
       }
-      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ops[0], vi);
+      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
       SV = llvm::ConstantVector::get(Indices);
       SV = Builder.CreateShuffleVector(Ops[1], Ops[2], SV, "vtrn");
       SV = Builder.CreateStore(SV, Addr);
@@ -3032,7 +3136,7 @@ Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
       for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
         Indices.push_back(ConstantInt::get(Int32Ty, 2*i+vi));
 
-      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ops[0], vi);
+      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
       SV = llvm::ConstantVector::get(Indices);
       SV = Builder.CreateShuffleVector(Ops[1], Ops[2], SV, "vuzp");
       SV = Builder.CreateStore(SV, Addr);
@@ -3052,7 +3156,7 @@ Value *CodeGenFunction::EmitCommonNeonBuiltinExpr(
         Indices.push_back(ConstantInt::get(Int32Ty, (i + vi*e) >> 1));
         Indices.push_back(ConstantInt::get(Int32Ty, ((i + vi*e) >> 1)+e));
       }
-      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ops[0], vi);
+      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
       SV = llvm::ConstantVector::get(Indices);
       SV = Builder.CreateShuffleVector(Ops[1], Ops[2], SV, "vzip");
       SV = Builder.CreateStore(SV, Addr);
@@ -3185,7 +3289,7 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
                 : InlineAsm::get(FTy, ".inst 0x" + utohexstr(ZExtValue), "",
                                  /*SideEffects=*/true);
 
-    return Builder.CreateCall(Emit);
+    return Builder.CreateCall(Emit, {});
   }
 
   if (BuiltinID == ARM::BI__builtin_arm_dbg) {
@@ -3202,7 +3306,7 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
     Value *Locality = llvm::ConstantInt::get(Int32Ty, 3);
 
     Value *F = CGM.getIntrinsic(Intrinsic::prefetch);
-    return Builder.CreateCall4(F, Address, RW, Locality, IsData);
+    return Builder.CreateCall(F, {Address, RW, Locality, IsData});
   }
 
   if (BuiltinID == ARM::BI__builtin_arm_rbit) {
@@ -3300,7 +3404,7 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
     Value *Arg0 = Builder.CreateExtractValue(Val, 0);
     Value *Arg1 = Builder.CreateExtractValue(Val, 1);
     Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)), Int8PtrTy);
-    return Builder.CreateCall3(F, Arg0, Arg1, StPtr, "strexd");
+    return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "strexd");
   }
 
   if (BuiltinID == ARM::BI__builtin_arm_strex ||
@@ -3324,12 +3428,12 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
                                        ? Intrinsic::arm_stlex
                                        : Intrinsic::arm_strex,
                                    StoreAddr->getType());
-    return Builder.CreateCall2(F, StoreVal, StoreAddr, "strex");
+    return Builder.CreateCall(F, {StoreVal, StoreAddr}, "strex");
   }
 
   if (BuiltinID == ARM::BI__builtin_arm_clrex) {
     Function *F = CGM.getIntrinsic(Intrinsic::arm_clrex);
-    return Builder.CreateCall(F);
+    return Builder.CreateCall(F, {});
   }
 
   // CRC32
@@ -3365,13 +3469,13 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
       Arg1b = Builder.CreateTruncOrBitCast(Arg1b, Int32Ty);
 
       Function *F = CGM.getIntrinsic(CRCIntrinsicID);
-      Value *Res = Builder.CreateCall2(F, Arg0, Arg1a);
-      return Builder.CreateCall2(F, Res, Arg1b);
+      Value *Res = Builder.CreateCall(F, {Arg0, Arg1a});
+      return Builder.CreateCall(F, {Res, Arg1b});
     } else {
       Arg1 = Builder.CreateZExtOrBitCast(Arg1, Int32Ty);
 
       Function *F = CGM.getIntrinsic(CRCIntrinsicID);
-      return Builder.CreateCall2(F, Arg0, Arg1);
+      return Builder.CreateCall(F, {Arg0, Arg1});
     }
   }
 
@@ -3547,7 +3651,7 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
       // Load the value as a one-element vector.
       Ty = llvm::VectorType::get(VTy->getElementType(), 1);
       Function *F = CGM.getIntrinsic(Intrinsic::arm_neon_vld1, Ty);
-      Value *Ld = Builder.CreateCall2(F, Ops[0], Align);
+      Value *Ld = Builder.CreateCall(F, {Ops[0], Align});
       // Combine them.
       SmallVector<Constant*, 2> Indices;
       Indices.push_back(ConstantInt::get(Int32Ty, 1-Lane));
@@ -3582,7 +3686,7 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
       default: llvm_unreachable("unknown vld_dup intrinsic?");
       }
       Function *F = CGM.getIntrinsic(Int, Ty);
-      Ops[1] = Builder.CreateCall2(F, Ops[1], Align, "vld_dup");
+      Ops[1] = Builder.CreateCall(F, {Ops[1], Align}, "vld_dup");
       Ty = llvm::PointerType::getUnqual(Ops[1]->getType());
       Ops[0] = Builder.CreateBitCast(Ops[0], Ty);
       return Builder.CreateStore(Ops[1], Ops[0]);
@@ -3651,7 +3755,7 @@ Value *CodeGenFunction::EmitARMBuiltinExpr(unsigned BuiltinID,
     Ops[1] = Builder.CreateBitCast(Ops[1], Ty);
     Ops[2] = EmitNeonShiftVector(Ops[2], Ty, true);
     Int = usgn ? Intrinsic::arm_neon_vrshiftu : Intrinsic::arm_neon_vrshifts;
-    Ops[1] = Builder.CreateCall2(CGM.getIntrinsic(Int, Ty), Ops[1], Ops[2]);
+    Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Ty), {Ops[1], Ops[2]});
     return Builder.CreateAdd(Ops[0], Ops[1], "vrsra_n");
   case NEON::BI__builtin_neon_vsri_n_v:
   case NEON::BI__builtin_neon_vsriq_n_v:
@@ -3979,7 +4083,7 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
     // FIXME: We need AArch64 specific LLVM intrinsic if we want to specify
     // PLDL3STRM or PLDL2STRM.
     Value *F = CGM.getIntrinsic(Intrinsic::prefetch);
-    return Builder.CreateCall4(F, Address, RW, Locality, IsData);
+    return Builder.CreateCall(F, {Address, RW, Locality, IsData});
   }
 
   if (BuiltinID == AArch64::BI__builtin_arm_rbit) {
@@ -4074,9 +4178,11 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
     Value *Arg1 = Builder.CreateExtractValue(Val, 1);
     Value *StPtr = Builder.CreateBitCast(EmitScalarExpr(E->getArg(1)),
                                          Int8PtrTy);
-    return Builder.CreateCall3(F, Arg0, Arg1, StPtr, "stxp");
-  } else if (BuiltinID == AArch64::BI__builtin_arm_strex ||
-             BuiltinID == AArch64::BI__builtin_arm_stlex) {
+    return Builder.CreateCall(F, {Arg0, Arg1, StPtr}, "stxp");
+  }
+
+  if (BuiltinID == AArch64::BI__builtin_arm_strex ||
+      BuiltinID == AArch64::BI__builtin_arm_stlex) {
     Value *StoreVal = EmitScalarExpr(E->getArg(0));
     Value *StoreAddr = EmitScalarExpr(E->getArg(1));
 
@@ -4096,12 +4202,12 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
                                        ? Intrinsic::aarch64_stlxr
                                        : Intrinsic::aarch64_stxr,
                                    StoreAddr->getType());
-    return Builder.CreateCall2(F, StoreVal, StoreAddr, "stxr");
+    return Builder.CreateCall(F, {StoreVal, StoreAddr}, "stxr");
   }
 
   if (BuiltinID == AArch64::BI__builtin_arm_clrex) {
     Function *F = CGM.getIntrinsic(Intrinsic::aarch64_clrex);
-    return Builder.CreateCall(F);
+    return Builder.CreateCall(F, {});
   }
 
   // CRC32
@@ -4133,7 +4239,7 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
     llvm::Type *DataTy = F->getFunctionType()->getParamType(1);
     Arg1 = Builder.CreateZExtOrBitCast(Arg1, DataTy);
 
-    return Builder.CreateCall2(F, Arg0, Arg1);
+    return Builder.CreateCall(F, {Arg0, Arg1});
   }
 
   llvm::SmallVector<Value*, 4> Ops;
@@ -4248,36 +4354,36 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
   case NEON::BI__builtin_neon_vceqzs_f32:
     Ops.push_back(EmitScalarExpr(E->getArg(0)));
     return EmitAArch64CompareBuiltinExpr(
-        Ops[0], ConvertType(E->getCallReturnType()), ICmpInst::FCMP_OEQ,
-        ICmpInst::ICMP_EQ, "vceqz");
+        Ops[0], ConvertType(E->getCallReturnType(getContext())),
+        ICmpInst::FCMP_OEQ, ICmpInst::ICMP_EQ, "vceqz");
   case NEON::BI__builtin_neon_vcgezd_s64:
   case NEON::BI__builtin_neon_vcgezd_f64:
   case NEON::BI__builtin_neon_vcgezs_f32:
     Ops.push_back(EmitScalarExpr(E->getArg(0)));
     return EmitAArch64CompareBuiltinExpr(
-        Ops[0], ConvertType(E->getCallReturnType()), ICmpInst::FCMP_OGE,
-        ICmpInst::ICMP_SGE, "vcgez");
+        Ops[0], ConvertType(E->getCallReturnType(getContext())),
+        ICmpInst::FCMP_OGE, ICmpInst::ICMP_SGE, "vcgez");
   case NEON::BI__builtin_neon_vclezd_s64:
   case NEON::BI__builtin_neon_vclezd_f64:
   case NEON::BI__builtin_neon_vclezs_f32:
     Ops.push_back(EmitScalarExpr(E->getArg(0)));
     return EmitAArch64CompareBuiltinExpr(
-        Ops[0], ConvertType(E->getCallReturnType()), ICmpInst::FCMP_OLE,
-        ICmpInst::ICMP_SLE, "vclez");
+        Ops[0], ConvertType(E->getCallReturnType(getContext())),
+        ICmpInst::FCMP_OLE, ICmpInst::ICMP_SLE, "vclez");
   case NEON::BI__builtin_neon_vcgtzd_s64:
   case NEON::BI__builtin_neon_vcgtzd_f64:
   case NEON::BI__builtin_neon_vcgtzs_f32:
     Ops.push_back(EmitScalarExpr(E->getArg(0)));
     return EmitAArch64CompareBuiltinExpr(
-        Ops[0], ConvertType(E->getCallReturnType()), ICmpInst::FCMP_OGT,
-        ICmpInst::ICMP_SGT, "vcgtz");
+        Ops[0], ConvertType(E->getCallReturnType(getContext())),
+        ICmpInst::FCMP_OGT, ICmpInst::ICMP_SGT, "vcgtz");
   case NEON::BI__builtin_neon_vcltzd_s64:
   case NEON::BI__builtin_neon_vcltzd_f64:
   case NEON::BI__builtin_neon_vcltzs_f32:
     Ops.push_back(EmitScalarExpr(E->getArg(0)));
     return EmitAArch64CompareBuiltinExpr(
-        Ops[0], ConvertType(E->getCallReturnType()), ICmpInst::FCMP_OLT,
-        ICmpInst::ICMP_SLT, "vcltz");
+        Ops[0], ConvertType(E->getCallReturnType(getContext())),
+        ICmpInst::FCMP_OLT, ICmpInst::ICMP_SLT, "vcltz");
 
   case NEON::BI__builtin_neon_vceqzd_u64: {
     llvm::Type *Ty = llvm::Type::getInt64Ty(getLLVMContext());
@@ -4528,8 +4634,8 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
                                    : Intrinsic::aarch64_neon_srshl;
     Ops[1] = Builder.CreateBitCast(Ops[1], Int64Ty);
     Ops.push_back(Builder.CreateNeg(EmitScalarExpr(E->getArg(2))));
-    Ops[1] = Builder.CreateCall2(CGM.getIntrinsic(Int, Int64Ty), Ops[1],
-                                 Builder.CreateSExt(Ops[2], Int64Ty));
+    Ops[1] = Builder.CreateCall(CGM.getIntrinsic(Int, Int64Ty),
+                                {Ops[1], Builder.CreateSExt(Ops[2], Int64Ty)});
     return Builder.CreateAdd(Ops[0], Builder.CreateBitCast(Ops[1], Int64Ty));
   }
   case NEON::BI__builtin_neon_vshld_n_s64:
@@ -4699,7 +4805,7 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
       Ops[2] = Builder.CreateBitCast(Ops[2], VTy);
       Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract");
       Value *F = CGM.getIntrinsic(Intrinsic::fma, DoubleTy);
-      Value *Result = Builder.CreateCall3(F, Ops[1], Ops[2], Ops[0]);
+      Value *Result = Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]});
       return Builder.CreateBitCast(Result, Ty);
     }
     Value *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
@@ -4713,7 +4819,7 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
                                                cast<ConstantInt>(Ops[3]));
     Ops[2] = Builder.CreateShuffleVector(Ops[2], Ops[2], SV, "lane");
 
-    return Builder.CreateCall3(F, Ops[2], Ops[1], Ops[0]);
+    return Builder.CreateCall(F, {Ops[2], Ops[1], Ops[0]});
   }
   case NEON::BI__builtin_neon_vfmaq_laneq_v: {
     Value *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
@@ -4722,17 +4828,17 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
 
     Ops[2] = Builder.CreateBitCast(Ops[2], Ty);
     Ops[2] = EmitNeonSplat(Ops[2], cast<ConstantInt>(Ops[3]));
-    return Builder.CreateCall3(F, Ops[2], Ops[1], Ops[0]);
+    return Builder.CreateCall(F, {Ops[2], Ops[1], Ops[0]});
   }
   case NEON::BI__builtin_neon_vfmas_lane_f32:
   case NEON::BI__builtin_neon_vfmas_laneq_f32:
   case NEON::BI__builtin_neon_vfmad_lane_f64:
   case NEON::BI__builtin_neon_vfmad_laneq_f64: {
     Ops.push_back(EmitScalarExpr(E->getArg(3)));
-    llvm::Type *Ty = ConvertType(E->getCallReturnType());
+    llvm::Type *Ty = ConvertType(E->getCallReturnType(getContext()));
     Value *F = CGM.getIntrinsic(Intrinsic::fma, Ty);
     Ops[2] = Builder.CreateExtractElement(Ops[2], Ops[3], "extract");
-    return Builder.CreateCall3(F, Ops[1], Ops[2], Ops[0]);
+    return Builder.CreateCall(F, {Ops[1], Ops[2], Ops[0]});
   }
   case NEON::BI__builtin_neon_vfms_v:
   case NEON::BI__builtin_neon_vfmsq_v: {  // Only used for FP types
@@ -5667,7 +5773,7 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
         Indices.push_back(ConstantInt::get(Int32Ty, i+vi));
         Indices.push_back(ConstantInt::get(Int32Ty, i+e+vi));
       }
-      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ops[0], vi);
+      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
       SV = llvm::ConstantVector::get(Indices);
       SV = Builder.CreateShuffleVector(Ops[1], Ops[2], SV, "vtrn");
       SV = Builder.CreateStore(SV, Addr);
@@ -5686,7 +5792,7 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
       for (unsigned i = 0, e = VTy->getNumElements(); i != e; ++i)
         Indices.push_back(ConstantInt::get(Int32Ty, 2*i+vi));
 
-      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ops[0], vi);
+      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
       SV = llvm::ConstantVector::get(Indices);
       SV = Builder.CreateShuffleVector(Ops[1], Ops[2], SV, "vuzp");
       SV = Builder.CreateStore(SV, Addr);
@@ -5706,7 +5812,7 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
         Indices.push_back(ConstantInt::get(Int32Ty, (i + vi*e) >> 1));
         Indices.push_back(ConstantInt::get(Int32Ty, ((i + vi*e) >> 1)+e));
       }
-      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ops[0], vi);
+      Value *Addr = Builder.CreateConstInBoundsGEP1_32(Ty, Ops[0], vi);
       SV = llvm::ConstantVector::get(Indices);
       SV = Builder.CreateShuffleVector(Ops[1], Ops[2], SV, "vzip");
       SV = Builder.CreateStore(SV, Addr);
@@ -5817,7 +5923,7 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
     Value *Locality = EmitScalarExpr(E->getArg(1));
     Value *Data = ConstantInt::get(Int32Ty, 1);
     Value *F = CGM.getIntrinsic(Intrinsic::prefetch);
-    return Builder.CreateCall4(F, Address, RW, Locality, Data);
+    return Builder.CreateCall(F, {Address, RW, Locality, Data});
   }
   case X86::BI__builtin_ia32_vec_init_v8qi:
   case X86::BI__builtin_ia32_vec_init_v4hi:
@@ -5856,104 +5962,95 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
     Ops[0] = Builder.CreateBitCast(Ops[0], PtrTy);
     return Builder.CreateStore(Ops[1], Ops[0]);
   }
-  case X86::BI__builtin_ia32_palignr: {
-    unsigned shiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
-
-    // If palignr is shifting the pair of input vectors less than 9 bytes,
-    // emit a shuffle instruction.
-    if (shiftVal <= 8) {
-      SmallVector<llvm::Constant*, 8> Indices;
-      for (unsigned i = 0; i != 8; ++i)
-        Indices.push_back(llvm::ConstantInt::get(Int32Ty, shiftVal + i));
-
-      Value* SV = llvm::ConstantVector::get(Indices);
-      return Builder.CreateShuffleVector(Ops[1], Ops[0], SV, "palignr");
+  case X86::BI__builtin_ia32_palignr128:
+  case X86::BI__builtin_ia32_palignr256: {
+    unsigned ShiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
+
+    unsigned NumElts =
+      cast<llvm::VectorType>(Ops[0]->getType())->getNumElements();
+    assert(NumElts % 16 == 0);
+    unsigned NumLanes = NumElts / 16;
+    unsigned NumLaneElts = NumElts / NumLanes;
+
+    // If palignr is shifting the pair of vectors more than the size of two
+    // lanes, emit zero.
+    if (ShiftVal >= (2 * NumLaneElts))
+      return llvm::Constant::getNullValue(ConvertType(E->getType()));
+
+    // If palignr is shifting the pair of input vectors more than one lane,
+    // but less than two lanes, convert to shifting in zeroes.
+    if (ShiftVal > NumLaneElts) {
+      ShiftVal -= NumLaneElts;
+      Ops[0] = llvm::Constant::getNullValue(Ops[0]->getType());
     }
 
-    // If palignr is shifting the pair of input vectors more than 8 but less
-    // than 16 bytes, emit a logical right shift of the destination.
-    if (shiftVal < 16) {
-      // MMX has these as 1 x i64 vectors for some odd optimization reasons.
-      llvm::Type *VecTy = llvm::VectorType::get(Int64Ty, 1);
-
-      Ops[0] = Builder.CreateBitCast(Ops[0], VecTy, "cast");
-      Ops[1] = llvm::ConstantInt::get(VecTy, (shiftVal-8) * 8);
-
-      // create i32 constant
-      llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_mmx_psrl_q);
-      return Builder.CreateCall(F, makeArrayRef(Ops.data(), 2), "palignr");
+    SmallVector<llvm::Constant*, 32> Indices;
+    // 256-bit palignr operates on 128-bit lanes so we need to handle that
+    for (unsigned l = 0; l != NumElts; l += NumLaneElts) {
+      for (unsigned i = 0; i != NumLaneElts; ++i) {
+        unsigned Idx = ShiftVal + i;
+        if (Idx >= NumLaneElts)
+          Idx += NumElts - NumLaneElts; // End of lane, switch operand.
+        Indices.push_back(llvm::ConstantInt::get(Int32Ty, Idx + l));
+      }
     }
 
-    // If palignr is shifting the pair of vectors more than 16 bytes, emit zero.
-    return llvm::Constant::getNullValue(ConvertType(E->getType()));
+    Value* SV = llvm::ConstantVector::get(Indices);
+    return Builder.CreateShuffleVector(Ops[1], Ops[0], SV, "palignr");
   }
-  case X86::BI__builtin_ia32_palignr128: {
-    unsigned shiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
+  case X86::BI__builtin_ia32_pslldqi256: {
+    // Shift value is in bits so divide by 8.
+    unsigned shiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() >> 3;
 
-    // If palignr is shifting the pair of input vectors less than 17 bytes,
-    // emit a shuffle instruction.
-    if (shiftVal <= 16) {
-      SmallVector<llvm::Constant*, 16> Indices;
-      for (unsigned i = 0; i != 16; ++i)
-        Indices.push_back(llvm::ConstantInt::get(Int32Ty, shiftVal + i));
+    // If pslldq is shifting the vector more than 15 bytes, emit zero.
+    if (shiftVal >= 16)
+      return llvm::Constant::getNullValue(ConvertType(E->getType()));
 
-      Value* SV = llvm::ConstantVector::get(Indices);
-      return Builder.CreateShuffleVector(Ops[1], Ops[0], SV, "palignr");
+    SmallVector<llvm::Constant*, 32> Indices;
+    // 256-bit pslldq operates on 128-bit lanes so we need to handle that
+    for (unsigned l = 0; l != 32; l += 16) {
+      for (unsigned i = 0; i != 16; ++i) {
+        unsigned Idx = 32 + i - shiftVal;
+        if (Idx < 32) Idx -= 16; // end of lane, switch operand.
+        Indices.push_back(llvm::ConstantInt::get(Int32Ty, Idx + l));
+      }
     }
 
-    // If palignr is shifting the pair of input vectors more than 16 but less
-    // than 32 bytes, emit a logical right shift of the destination.
-    if (shiftVal < 32) {
-      llvm::Type *VecTy = llvm::VectorType::get(Int64Ty, 2);
-
-      Ops[0] = Builder.CreateBitCast(Ops[0], VecTy, "cast");
-      Ops[1] = llvm::ConstantInt::get(Int32Ty, (shiftVal-16) * 8);
-
-      // create i32 constant
-      llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_sse2_psrl_dq);
-      return Builder.CreateCall(F, makeArrayRef(Ops.data(), 2), "palignr");
-    }
+    llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, 32);
+    Ops[0] = Builder.CreateBitCast(Ops[0], VecTy, "cast");
+    Value *Zero = llvm::Constant::getNullValue(VecTy);
 
-    // If palignr is shifting the pair of vectors more than 32 bytes, emit zero.
-    return llvm::Constant::getNullValue(ConvertType(E->getType()));
-  }
-  case X86::BI__builtin_ia32_palignr256: {
-    unsigned shiftVal = cast<llvm::ConstantInt>(Ops[2])->getZExtValue();
-
-    // If palignr is shifting the pair of input vectors less than 17 bytes,
-    // emit a shuffle instruction.
-    if (shiftVal <= 16) {
-      SmallVector<llvm::Constant*, 32> Indices;
-      // 256-bit palignr operates on 128-bit lanes so we need to handle that
-      for (unsigned l = 0; l != 2; ++l) {
-        unsigned LaneStart = l * 16;
-        unsigned LaneEnd = (l+1) * 16;
-        for (unsigned i = 0; i != 16; ++i) {
-          unsigned Idx = shiftVal + i + LaneStart;
-          if (Idx >= LaneEnd) Idx += 16; // end of lane, switch operand
-          Indices.push_back(llvm::ConstantInt::get(Int32Ty, Idx));
-        }
+    Value *SV = llvm::ConstantVector::get(Indices);
+    SV = Builder.CreateShuffleVector(Zero, Ops[0], SV, "pslldq");
+    llvm::Type *ResultType = ConvertType(E->getType());
+    return Builder.CreateBitCast(SV, ResultType, "cast");
+  }
+  case X86::BI__builtin_ia32_psrldqi256: {
+    // Shift value is in bits so divide by 8.
+    unsigned shiftVal = cast<llvm::ConstantInt>(Ops[1])->getZExtValue() >> 3;
+
+    // If psrldq is shifting the vector more than 15 bytes, emit zero.
+    if (shiftVal >= 16)
+      return llvm::Constant::getNullValue(ConvertType(E->getType()));
+
+    SmallVector<llvm::Constant*, 32> Indices;
+    // 256-bit psrldq operates on 128-bit lanes so we need to handle that
+    for (unsigned l = 0; l != 32; l += 16) {
+      for (unsigned i = 0; i != 16; ++i) {
+        unsigned Idx = i + shiftVal;
+        if (Idx >= 16) Idx += 16; // end of lane, switch operand.
+        Indices.push_back(llvm::ConstantInt::get(Int32Ty, Idx + l));
       }
-
-      Value* SV = llvm::ConstantVector::get(Indices);
-      return Builder.CreateShuffleVector(Ops[1], Ops[0], SV, "palignr");
     }
 
-    // If palignr is shifting the pair of input vectors more than 16 but less
-    // than 32 bytes, emit a logical right shift of the destination.
-    if (shiftVal < 32) {
-      llvm::Type *VecTy = llvm::VectorType::get(Int64Ty, 4);
-
-      Ops[0] = Builder.CreateBitCast(Ops[0], VecTy, "cast");
-      Ops[1] = llvm::ConstantInt::get(Int32Ty, (shiftVal-16) * 8);
-
-      // create i32 constant
-      llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_avx2_psrl_dq);
-      return Builder.CreateCall(F, makeArrayRef(Ops.data(), 2), "palignr");
-    }
+    llvm::Type *VecTy = llvm::VectorType::get(Int8Ty, 32);
+    Ops[0] = Builder.CreateBitCast(Ops[0], VecTy, "cast");
+    Value *Zero = llvm::Constant::getNullValue(VecTy);
 
-    // If palignr is shifting the pair of vectors more than 32 bytes, emit zero.
-    return llvm::Constant::getNullValue(ConvertType(E->getType()));
+    Value *SV = llvm::ConstantVector::get(Indices);
+    SV = Builder.CreateShuffleVector(Ops[0], Zero, SV, "psrldq");
+    llvm::Type *ResultType = ConvertType(E->getType());
+    return Builder.CreateBitCast(SV, ResultType, "cast");
   }
   case X86::BI__builtin_ia32_movntps:
   case X86::BI__builtin_ia32_movntps256:
@@ -5987,20 +6084,10 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
   // 3DNow!
   case X86::BI__builtin_ia32_pswapdsf:
   case X86::BI__builtin_ia32_pswapdsi: {
-    const char *name;
-    Intrinsic::ID ID;
-    switch(BuiltinID) {
-    default: llvm_unreachable("Unsupported intrinsic!");
-    case X86::BI__builtin_ia32_pswapdsf:
-    case X86::BI__builtin_ia32_pswapdsi:
-      name = "pswapd";
-      ID = Intrinsic::x86_3dnowa_pswapd;
-      break;
-    }
     llvm::Type *MMXTy = llvm::Type::getX86_MMXTy(getLLVMContext());
     Ops[0] = Builder.CreateBitCast(Ops[0], MMXTy, "cast");
-    llvm::Function *F = CGM.getIntrinsic(ID);
-    return Builder.CreateCall(F, Ops, name);
+    llvm::Function *F = CGM.getIntrinsic(Intrinsic::x86_3dnowa_pswapd);
+    return Builder.CreateCall(F, Ops, "pswapd");
   }
   case X86::BI__builtin_ia32_rdrand16_step:
   case X86::BI__builtin_ia32_rdrand32_step:
@@ -6031,17 +6118,10 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
       break;
     }
 
-    Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID));
+    Value *Call = Builder.CreateCall(CGM.getIntrinsic(ID), {});
     Builder.CreateStore(Builder.CreateExtractValue(Call, 0), Ops[0]);
     return Builder.CreateExtractValue(Call, 1);
   }
-  // AVX2 broadcast
-  case X86::BI__builtin_ia32_vbroadcastsi256: {
-    Value *VecTmp = CreateMemTemp(E->getArg(0)->getType());
-    Builder.CreateStore(Ops[0], VecTmp);
-    Value *F = CGM.getIntrinsic(Intrinsic::x86_avx2_vbroadcasti128);
-    return Builder.CreateCall(F, Builder.CreateBitCast(VecTmp, Int8PtrTy));
-  }
   // SSE comparison intrisics
   case X86::BI__builtin_ia32_cmpeqps:
   case X86::BI__builtin_ia32_cmpltps:
@@ -6318,7 +6398,7 @@ static Value *emitTernaryFPBuiltin(CodeGenFunction &CGF,
   llvm::Value *Src2 = CGF.EmitScalarExpr(E->getArg(2));
 
   Value *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
-  return CGF.Builder.CreateCall3(F, Src0, Src1, Src2);
+  return CGF.Builder.CreateCall(F, {Src0, Src1, Src2});
 }
 
 // Emit an intrinsic that has 1 float or double operand, and 1 integer.
@@ -6329,7 +6409,7 @@ static Value *emitFPIntBuiltin(CodeGenFunction &CGF,
   llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
 
   Value *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType());
-  return CGF.Builder.CreateCall2(F, Src0, Src1);
+  return CGF.Builder.CreateCall(F, {Src0, Src1});
 }
 
 Value *CodeGenFunction::EmitR600BuiltinExpr(unsigned BuiltinID,
@@ -6350,7 +6430,7 @@ Value *CodeGenFunction::EmitR600BuiltinExpr(unsigned BuiltinID,
     llvm::Value *Callee = CGM.getIntrinsic(Intrinsic::AMDGPU_div_scale,
                                            X->getType());
 
-    llvm::Value *Tmp = Builder.CreateCall3(Callee, X, Y, Z);
+    llvm::Value *Tmp = Builder.CreateCall(Callee, {X, Y, Z});
 
     llvm::Value *Result = Builder.CreateExtractValue(Tmp, 0);
     llvm::Value *Flag = Builder.CreateExtractValue(Tmp, 1);
@@ -6373,7 +6453,7 @@ Value *CodeGenFunction::EmitR600BuiltinExpr(unsigned BuiltinID,
     llvm::Value *F = CGM.getIntrinsic(Intrinsic::AMDGPU_div_fmas,
                                       Src0->getType());
     llvm::Value *Src3ToBool = Builder.CreateIsNotNull(Src3);
-    return Builder.CreateCall4(F, Src0, Src1, Src2, Src3ToBool);
+    return Builder.CreateCall(F, {Src0, Src1, Src2, Src3ToBool});
   }
   case R600::BI__builtin_amdgpu_div_fixup:
   case R600::BI__builtin_amdgpu_div_fixupf:
@@ -6400,3 +6480,246 @@ Value *CodeGenFunction::EmitR600BuiltinExpr(unsigned BuiltinID,
     return nullptr;
   }
 }
+
+/// Handle a SystemZ function in which the final argument is a pointer
+/// to an int that receives the post-instruction CC value.  At the LLVM level
+/// this is represented as a function that returns a {result, cc} pair.
+static Value *EmitSystemZIntrinsicWithCC(CodeGenFunction &CGF,
+                                         unsigned IntrinsicID,
+                                         const CallExpr *E) {
+  unsigned NumArgs = E->getNumArgs() - 1;
+  SmallVector<Value *, 8> Args(NumArgs);
+  for (unsigned I = 0; I < NumArgs; ++I)
+    Args[I] = CGF.EmitScalarExpr(E->getArg(I));
+  Value *CCPtr = CGF.EmitScalarExpr(E->getArg(NumArgs));
+  Value *F = CGF.CGM.getIntrinsic(IntrinsicID);
+  Value *Call = CGF.Builder.CreateCall(F, Args);
+  Value *CC = CGF.Builder.CreateExtractValue(Call, 1);
+  CGF.Builder.CreateStore(CC, CCPtr);
+  return CGF.Builder.CreateExtractValue(Call, 0);
+}
+
+Value *CodeGenFunction::EmitSystemZBuiltinExpr(unsigned BuiltinID,
+                                               const CallExpr *E) {
+  switch (BuiltinID) {
+  case SystemZ::BI__builtin_tbegin: {
+    Value *TDB = EmitScalarExpr(E->getArg(0));
+    Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c);
+    Value *F = CGM.getIntrinsic(Intrinsic::s390_tbegin);
+    return Builder.CreateCall(F, {TDB, Control});
+  }
+  case SystemZ::BI__builtin_tbegin_nofloat: {
+    Value *TDB = EmitScalarExpr(E->getArg(0));
+    Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff0c);
+    Value *F = CGM.getIntrinsic(Intrinsic::s390_tbegin_nofloat);
+    return Builder.CreateCall(F, {TDB, Control});
+  }
+  case SystemZ::BI__builtin_tbeginc: {
+    Value *TDB = llvm::ConstantPointerNull::get(Int8PtrTy);
+    Value *Control = llvm::ConstantInt::get(Int32Ty, 0xff08);
+    Value *F = CGM.getIntrinsic(Intrinsic::s390_tbeginc);
+    return Builder.CreateCall(F, {TDB, Control});
+  }
+  case SystemZ::BI__builtin_tabort: {
+    Value *Data = EmitScalarExpr(E->getArg(0));
+    Value *F = CGM.getIntrinsic(Intrinsic::s390_tabort);
+    return Builder.CreateCall(F, Builder.CreateSExt(Data, Int64Ty, "tabort"));
+  }
+  case SystemZ::BI__builtin_non_tx_store: {
+    Value *Address = EmitScalarExpr(E->getArg(0));
+    Value *Data = EmitScalarExpr(E->getArg(1));
+    Value *F = CGM.getIntrinsic(Intrinsic::s390_ntstg);
+    return Builder.CreateCall(F, {Data, Address});
+  }
+
+  // Vector builtins.  Note that most vector builtins are mapped automatically
+  // to target-specific LLVM intrinsics.  The ones handled specially here can
+  // be represented via standard LLVM IR, which is preferable to enable common
+  // LLVM optimizations.
+
+  case SystemZ::BI__builtin_s390_vpopctb:
+  case SystemZ::BI__builtin_s390_vpopcth:
+  case SystemZ::BI__builtin_s390_vpopctf:
+  case SystemZ::BI__builtin_s390_vpopctg: {
+    llvm::Type *ResultType = ConvertType(E->getType());
+    Value *X = EmitScalarExpr(E->getArg(0));
+    Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ResultType);
+    return Builder.CreateCall(F, X);
+  }
+
+  case SystemZ::BI__builtin_s390_vclzb:
+  case SystemZ::BI__builtin_s390_vclzh:
+  case SystemZ::BI__builtin_s390_vclzf:
+  case SystemZ::BI__builtin_s390_vclzg: {
+    llvm::Type *ResultType = ConvertType(E->getType());
+    Value *X = EmitScalarExpr(E->getArg(0));
+    Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
+    Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ResultType);
+    return Builder.CreateCall(F, {X, Undef});
+  }
+
+  case SystemZ::BI__builtin_s390_vctzb:
+  case SystemZ::BI__builtin_s390_vctzh:
+  case SystemZ::BI__builtin_s390_vctzf:
+  case SystemZ::BI__builtin_s390_vctzg: {
+    llvm::Type *ResultType = ConvertType(E->getType());
+    Value *X = EmitScalarExpr(E->getArg(0));
+    Value *Undef = ConstantInt::get(Builder.getInt1Ty(), false);
+    Function *F = CGM.getIntrinsic(Intrinsic::cttz, ResultType);
+    return Builder.CreateCall(F, {X, Undef});
+  }
+
+  case SystemZ::BI__builtin_s390_vfsqdb: {
+    llvm::Type *ResultType = ConvertType(E->getType());
+    Value *X = EmitScalarExpr(E->getArg(0));
+    Function *F = CGM.getIntrinsic(Intrinsic::sqrt, ResultType);
+    return Builder.CreateCall(F, X);
+  }
+  case SystemZ::BI__builtin_s390_vfmadb: {
+    llvm::Type *ResultType = ConvertType(E->getType());
+    Value *X = EmitScalarExpr(E->getArg(0));
+    Value *Y = EmitScalarExpr(E->getArg(1));
+    Value *Z = EmitScalarExpr(E->getArg(2));
+    Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
+    return Builder.CreateCall(F, {X, Y, Z});
+  }
+  case SystemZ::BI__builtin_s390_vfmsdb: {
+    llvm::Type *ResultType = ConvertType(E->getType());
+    Value *X = EmitScalarExpr(E->getArg(0));
+    Value *Y = EmitScalarExpr(E->getArg(1));
+    Value *Z = EmitScalarExpr(E->getArg(2));
+    Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
+    Function *F = CGM.getIntrinsic(Intrinsic::fma, ResultType);
+    return Builder.CreateCall(F, {X, Y, Builder.CreateFSub(Zero, Z, "sub")});
+  }
+  case SystemZ::BI__builtin_s390_vflpdb: {
+    llvm::Type *ResultType = ConvertType(E->getType());
+    Value *X = EmitScalarExpr(E->getArg(0));
+    Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
+    return Builder.CreateCall(F, X);
+  }
+  case SystemZ::BI__builtin_s390_vflndb: {
+    llvm::Type *ResultType = ConvertType(E->getType());
+    Value *X = EmitScalarExpr(E->getArg(0));
+    Value *Zero = llvm::ConstantFP::getZeroValueForNegation(ResultType);
+    Function *F = CGM.getIntrinsic(Intrinsic::fabs, ResultType);
+    return Builder.CreateFSub(Zero, Builder.CreateCall(F, X), "sub");
+  }
+  case SystemZ::BI__builtin_s390_vfidb: {
+    llvm::Type *ResultType = ConvertType(E->getType());
+    Value *X = EmitScalarExpr(E->getArg(0));
+    // Constant-fold the M4 and M5 mask arguments.
+    llvm::APSInt M4, M5;
+    bool IsConstM4 = E->getArg(1)->isIntegerConstantExpr(M4, getContext());
+    bool IsConstM5 = E->getArg(2)->isIntegerConstantExpr(M5, getContext());
+    assert(IsConstM4 && IsConstM5 && "Constant arg isn't actually constant?");
+    (void)IsConstM4; (void)IsConstM5;
+    // Check whether this instance of vfidb can be represented via a LLVM
+    // standard intrinsic.  We only support some combinations of M4 and M5.
+    Intrinsic::ID ID = Intrinsic::not_intrinsic;
+    switch (M4.getZExtValue()) {
+    default: break;
+    case 0:  // IEEE-inexact exception allowed
+      switch (M5.getZExtValue()) {
+      default: break;
+      case 0: ID = Intrinsic::rint; break;
+      }
+      break;
+    case 4:  // IEEE-inexact exception suppressed
+      switch (M5.getZExtValue()) {
+      default: break;
+      case 0: ID = Intrinsic::nearbyint; break;
+      case 1: ID = Intrinsic::round; break;
+      case 5: ID = Intrinsic::trunc; break;
+      case 6: ID = Intrinsic::ceil; break;
+      case 7: ID = Intrinsic::floor; break;
+      }
+      break;
+    }
+    if (ID != Intrinsic::not_intrinsic) {
+      Function *F = CGM.getIntrinsic(ID, ResultType);
+      return Builder.CreateCall(F, X);
+    }
+    Function *F = CGM.getIntrinsic(Intrinsic::s390_vfidb);
+    Value *M4Value = llvm::ConstantInt::get(getLLVMContext(), M4);
+    Value *M5Value = llvm::ConstantInt::get(getLLVMContext(), M5);
+    return Builder.CreateCall(F, {X, M4Value, M5Value});
+  }
+
+  // Vector intrisincs that output the post-instruction CC value.
+
+#define INTRINSIC_WITH_CC(NAME) \
+    case SystemZ::BI__builtin_##NAME: \
+      return EmitSystemZIntrinsicWithCC(*this, Intrinsic::NAME, E)
+
+  INTRINSIC_WITH_CC(s390_vpkshs);
+  INTRINSIC_WITH_CC(s390_vpksfs);
+  INTRINSIC_WITH_CC(s390_vpksgs);
+
+  INTRINSIC_WITH_CC(s390_vpklshs);
+  INTRINSIC_WITH_CC(s390_vpklsfs);
+  INTRINSIC_WITH_CC(s390_vpklsgs);
+
+  INTRINSIC_WITH_CC(s390_vceqbs);
+  INTRINSIC_WITH_CC(s390_vceqhs);
+  INTRINSIC_WITH_CC(s390_vceqfs);
+  INTRINSIC_WITH_CC(s390_vceqgs);
+
+  INTRINSIC_WITH_CC(s390_vchbs);
+  INTRINSIC_WITH_CC(s390_vchhs);
+  INTRINSIC_WITH_CC(s390_vchfs);
+  INTRINSIC_WITH_CC(s390_vchgs);
+
+  INTRINSIC_WITH_CC(s390_vchlbs);
+  INTRINSIC_WITH_CC(s390_vchlhs);
+  INTRINSIC_WITH_CC(s390_vchlfs);
+  INTRINSIC_WITH_CC(s390_vchlgs);
+
+  INTRINSIC_WITH_CC(s390_vfaebs);
+  INTRINSIC_WITH_CC(s390_vfaehs);
+  INTRINSIC_WITH_CC(s390_vfaefs);
+
+  INTRINSIC_WITH_CC(s390_vfaezbs);
+  INTRINSIC_WITH_CC(s390_vfaezhs);
+  INTRINSIC_WITH_CC(s390_vfaezfs);
+
+  INTRINSIC_WITH_CC(s390_vfeebs);
+  INTRINSIC_WITH_CC(s390_vfeehs);
+  INTRINSIC_WITH_CC(s390_vfeefs);
+
+  INTRINSIC_WITH_CC(s390_vfeezbs);
+  INTRINSIC_WITH_CC(s390_vfeezhs);
+  INTRINSIC_WITH_CC(s390_vfeezfs);
+
+  INTRINSIC_WITH_CC(s390_vfenebs);
+  INTRINSIC_WITH_CC(s390_vfenehs);
+  INTRINSIC_WITH_CC(s390_vfenefs);
+
+  INTRINSIC_WITH_CC(s390_vfenezbs);
+  INTRINSIC_WITH_CC(s390_vfenezhs);
+  INTRINSIC_WITH_CC(s390_vfenezfs);
+
+  INTRINSIC_WITH_CC(s390_vistrbs);
+  INTRINSIC_WITH_CC(s390_vistrhs);
+  INTRINSIC_WITH_CC(s390_vistrfs);
+
+  INTRINSIC_WITH_CC(s390_vstrcbs);
+  INTRINSIC_WITH_CC(s390_vstrchs);
+  INTRINSIC_WITH_CC(s390_vstrcfs);
+
+  INTRINSIC_WITH_CC(s390_vstrczbs);
+  INTRINSIC_WITH_CC(s390_vstrczhs);
+  INTRINSIC_WITH_CC(s390_vstrczfs);
+
+  INTRINSIC_WITH_CC(s390_vfcedbs);
+  INTRINSIC_WITH_CC(s390_vfchdbs);
+  INTRINSIC_WITH_CC(s390_vfchedbs);
+
+  INTRINSIC_WITH_CC(s390_vftcidb);
+
+#undef INTRINSIC_WITH_CC
+
+  default:
+    return nullptr;
+  }
+}
diff --git a/lib/CodeGen/CGCUDANV.cpp b/lib/CodeGen/CGCUDANV.cpp
index fb11751fc91e..67d0ab7a82f7 100644
--- a/lib/CodeGen/CGCUDANV.cpp
+++ b/lib/CodeGen/CGCUDANV.cpp
@@ -20,7 +20,6 @@
 #include "llvm/IR/CallSite.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/DerivedTypes.h"
-#include <vector>
 
 using namespace clang;
 using namespace CodeGen;
@@ -30,29 +29,66 @@ namespace {
 class CGNVCUDARuntime : public CGCUDARuntime {
 
 private:
-  llvm::Type *IntTy, *SizeTy;
-  llvm::PointerType *CharPtrTy, *VoidPtrTy;
+  llvm::Type *IntTy, *SizeTy, *VoidTy;
+  llvm::PointerType *CharPtrTy, *VoidPtrTy, *VoidPtrPtrTy;
+
+  /// Convenience reference to LLVM Context
+  llvm::LLVMContext &Context;
+  /// Convenience reference to the current module
+  llvm::Module &TheModule;
+  /// Keeps track of kernel launch stubs emitted in this module
+  llvm::SmallVector<llvm::Function *, 16> EmittedKernels;
+  /// Keeps track of variables containing handles of GPU binaries. Populated by
+  /// ModuleCtorFunction() and used to create corresponding cleanup calls in
+  /// ModuleDtorFunction()
+  llvm::SmallVector<llvm::GlobalVariable *, 16> GpuBinaryHandles;
 
   llvm::Constant *getSetupArgumentFn() const;
   llvm::Constant *getLaunchFn() const;
 
+  /// Creates a function to register all kernel stubs generated in this module.
+  llvm::Function *makeRegisterKernelsFn();
+
+  /// Helper function that generates a constant string and returns a pointer to
+  /// the start of the string.  The result of this function can be used anywhere
+  /// where the C code specifies const char*.
+  llvm::Constant *makeConstantString(const std::string &Str,
+                                     const std::string &Name = "",
+                                     unsigned Alignment = 0) {
+    llvm::Constant *Zeros[] = {llvm::ConstantInt::get(SizeTy, 0),
+                               llvm::ConstantInt::get(SizeTy, 0)};
+    auto *ConstStr = CGM.GetAddrOfConstantCString(Str, Name.c_str());
+    return llvm::ConstantExpr::getGetElementPtr(ConstStr->getValueType(),
+                                                ConstStr, Zeros);
+ }
+
+  void emitDeviceStubBody(CodeGenFunction &CGF, FunctionArgList &Args);
+
 public:
   CGNVCUDARuntime(CodeGenModule &CGM);
 
-  void EmitDeviceStubBody(CodeGenFunction &CGF, FunctionArgList &Args) override;
+  void emitDeviceStub(CodeGenFunction &CGF, FunctionArgList &Args) override;
+  /// Creates module constructor function
+  llvm::Function *makeModuleCtorFunction() override;
+  /// Creates module destructor function
+  llvm::Function *makeModuleDtorFunction() override;
 };
 
 }
 
-CGNVCUDARuntime::CGNVCUDARuntime(CodeGenModule &CGM) : CGCUDARuntime(CGM) {
+CGNVCUDARuntime::CGNVCUDARuntime(CodeGenModule &CGM)
+    : CGCUDARuntime(CGM), Context(CGM.getLLVMContext()),
+      TheModule(CGM.getModule()) {
   CodeGen::CodeGenTypes &Types = CGM.getTypes();
   ASTContext &Ctx = CGM.getContext();
 
   IntTy = Types.ConvertType(Ctx.IntTy);
   SizeTy = Types.ConvertType(Ctx.getSizeType());
+  VoidTy = llvm::Type::getVoidTy(Context);
 
   CharPtrTy = llvm::PointerType::getUnqual(Types.ConvertType(Ctx.CharTy));
   VoidPtrTy = cast<llvm::PointerType>(Types.ConvertType(Ctx.VoidPtrTy));
+  VoidPtrPtrTy = VoidPtrTy->getPointerTo();
 }
 
 llvm::Constant *CGNVCUDARuntime::getSetupArgumentFn() const {
@@ -68,14 +104,17 @@ llvm::Constant *CGNVCUDARuntime::getSetupArgumentFn() const {
 
 llvm::Constant *CGNVCUDARuntime::getLaunchFn() const {
   // cudaError_t cudaLaunch(char *)
-  std::vector<llvm::Type*> Params;
-  Params.push_back(CharPtrTy);
-  return CGM.CreateRuntimeFunction(llvm::FunctionType::get(IntTy,
-                                                           Params, false),
-                                   "cudaLaunch");
+  return CGM.CreateRuntimeFunction(
+      llvm::FunctionType::get(IntTy, CharPtrTy, false), "cudaLaunch");
+}
+
+void CGNVCUDARuntime::emitDeviceStub(CodeGenFunction &CGF,
+                                     FunctionArgList &Args) {
+  EmittedKernels.push_back(CGF.CurFn);
+  emitDeviceStubBody(CGF, Args);
 }
 
-void CGNVCUDARuntime::EmitDeviceStubBody(CodeGenFunction &CGF,
+void CGNVCUDARuntime::emitDeviceStubBody(CodeGenFunction &CGF,
                                          FunctionArgList &Args) {
   // Build the argument value list and the argument stack struct type.
   SmallVector<llvm::Value *, 16> ArgValues;
@@ -87,8 +126,7 @@ void CGNVCUDARuntime::EmitDeviceStubBody(CodeGenFunction &CGF,
     assert(isa<llvm::PointerType>(V->getType()) && "Arg type not PointerType");
     ArgTypes.push_back(cast<llvm::PointerType>(V->getType())->getElementType());
   }
-  llvm::StructType *ArgStackTy = llvm::StructType::get(
-      CGF.getLLVMContext(), ArgTypes);
+  llvm::StructType *ArgStackTy = llvm::StructType::get(Context, ArgTypes);
 
   llvm::BasicBlock *EndBlock = CGF.createBasicBlock("setup.end");
 
@@ -120,6 +158,160 @@ void CGNVCUDARuntime::EmitDeviceStubBody(CodeGenFunction &CGF,
   CGF.EmitBlock(EndBlock);
 }
 
+/// Creates internal function to register all kernel stubs generated in this
+/// module with the CUDA runtime.
+/// \code
+/// void __cuda_register_kernels(void** GpuBinaryHandle) {
+///    __cudaRegisterFunction(GpuBinaryHandle,Kernel0,...);
+///    ...
+///    __cudaRegisterFunction(GpuBinaryHandle,KernelM,...);
+/// }
+/// \endcode
+llvm::Function *CGNVCUDARuntime::makeRegisterKernelsFn() {
+  llvm::Function *RegisterKernelsFunc = llvm::Function::Create(
+      llvm::FunctionType::get(VoidTy, VoidPtrPtrTy, false),
+      llvm::GlobalValue::InternalLinkage, "__cuda_register_kernels", &TheModule);
+  llvm::BasicBlock *EntryBB =
+      llvm::BasicBlock::Create(Context, "entry", RegisterKernelsFunc);
+  CGBuilderTy Builder(Context);
+  Builder.SetInsertPoint(EntryBB);
+
+  // void __cudaRegisterFunction(void **, const char *, char *, const char *,
+  //                             int, uint3*, uint3*, dim3*, dim3*, int*)
+  std::vector<llvm::Type *> RegisterFuncParams = {
+      VoidPtrPtrTy, CharPtrTy, CharPtrTy, CharPtrTy, IntTy,
+      VoidPtrTy,    VoidPtrTy, VoidPtrTy, VoidPtrTy, IntTy->getPointerTo()};
+  llvm::Constant *RegisterFunc = CGM.CreateRuntimeFunction(
+      llvm::FunctionType::get(IntTy, RegisterFuncParams, false),
+      "__cudaRegisterFunction");
+
+  // Extract GpuBinaryHandle passed as the first argument passed to
+  // __cuda_register_kernels() and generate __cudaRegisterFunction() call for
+  // each emitted kernel.
+  llvm::Argument &GpuBinaryHandlePtr = *RegisterKernelsFunc->arg_begin();
+  for (llvm::Function *Kernel : EmittedKernels) {
+    llvm::Constant *KernelName = makeConstantString(Kernel->getName());
+    llvm::Constant *NullPtr = llvm::ConstantPointerNull::get(VoidPtrTy);
+    llvm::Value *args[] = {
+        &GpuBinaryHandlePtr, Builder.CreateBitCast(Kernel, VoidPtrTy),
+        KernelName, KernelName, llvm::ConstantInt::get(IntTy, -1), NullPtr,
+        NullPtr, NullPtr, NullPtr,
+        llvm::ConstantPointerNull::get(IntTy->getPointerTo())};
+    Builder.CreateCall(RegisterFunc, args);
+  }
+
+  Builder.CreateRetVoid();
+  return RegisterKernelsFunc;
+}
+
+/// Creates a global constructor function for the module:
+/// \code
+/// void __cuda_module_ctor(void*) {
+///     Handle0 = __cudaRegisterFatBinary(GpuBinaryBlob0);
+///     __cuda_register_kernels(Handle0);
+///     ...
+///     HandleN = __cudaRegisterFatBinary(GpuBinaryBlobN);
+///     __cuda_register_kernels(HandleN);
+/// }
+/// \endcode
+llvm::Function *CGNVCUDARuntime::makeModuleCtorFunction() {
+  // void __cuda_register_kernels(void* handle);
+  llvm::Function *RegisterKernelsFunc = makeRegisterKernelsFn();
+  // void ** __cudaRegisterFatBinary(void *);
+  llvm::Constant *RegisterFatbinFunc = CGM.CreateRuntimeFunction(
+      llvm::FunctionType::get(VoidPtrPtrTy, VoidPtrTy, false),
+      "__cudaRegisterFatBinary");
+  // struct { int magic, int version, void * gpu_binary, void * dont_care };
+  llvm::StructType *FatbinWrapperTy =
+      llvm::StructType::get(IntTy, IntTy, VoidPtrTy, VoidPtrTy, nullptr);
+
+  llvm::Function *ModuleCtorFunc = llvm::Function::Create(
+      llvm::FunctionType::get(VoidTy, VoidPtrTy, false),
+      llvm::GlobalValue::InternalLinkage, "__cuda_module_ctor", &TheModule);
+  llvm::BasicBlock *CtorEntryBB =
+      llvm::BasicBlock::Create(Context, "entry", ModuleCtorFunc);
+  CGBuilderTy CtorBuilder(Context);
+
+  CtorBuilder.SetInsertPoint(CtorEntryBB);
+
+  // For each GPU binary, register it with the CUDA runtime and store returned
+  // handle in a global variable and save the handle in GpuBinaryHandles vector
+  // to be cleaned up in destructor on exit. Then associate all known kernels
+  // with the GPU binary handle so CUDA runtime can figure out what to call on
+  // the GPU side.
+  for (const std::string &GpuBinaryFileName :
+       CGM.getCodeGenOpts().CudaGpuBinaryFileNames) {
+    llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> GpuBinaryOrErr =
+        llvm::MemoryBuffer::getFileOrSTDIN(GpuBinaryFileName);
+    if (std::error_code EC = GpuBinaryOrErr.getError()) {
+      CGM.getDiags().Report(diag::err_cannot_open_file) << GpuBinaryFileName
+                                                        << EC.message();
+      continue;
+    }
+
+    // Create initialized wrapper structure that points to the loaded GPU binary
+    llvm::Constant *Values[] = {
+        llvm::ConstantInt::get(IntTy, 0x466243b1), // Fatbin wrapper magic.
+        llvm::ConstantInt::get(IntTy, 1),          // Fatbin version.
+        makeConstantString(GpuBinaryOrErr.get()->getBuffer(), "", 16), // Data.
+        llvm::ConstantPointerNull::get(VoidPtrTy)}; // Unused in fatbin v1.
+    llvm::GlobalVariable *FatbinWrapper = new llvm::GlobalVariable(
+        TheModule, FatbinWrapperTy, true, llvm::GlobalValue::InternalLinkage,
+        llvm::ConstantStruct::get(FatbinWrapperTy, Values),
+        "__cuda_fatbin_wrapper");
+
+    // GpuBinaryHandle = __cudaRegisterFatBinary(&FatbinWrapper);
+    llvm::CallInst *RegisterFatbinCall = CtorBuilder.CreateCall(
+        RegisterFatbinFunc,
+        CtorBuilder.CreateBitCast(FatbinWrapper, VoidPtrTy));
+    llvm::GlobalVariable *GpuBinaryHandle = new llvm::GlobalVariable(
+        TheModule, VoidPtrPtrTy, false, llvm::GlobalValue::InternalLinkage,
+        llvm::ConstantPointerNull::get(VoidPtrPtrTy), "__cuda_gpubin_handle");
+    CtorBuilder.CreateStore(RegisterFatbinCall, GpuBinaryHandle, false);
+
+    // Call __cuda_register_kernels(GpuBinaryHandle);
+    CtorBuilder.CreateCall(RegisterKernelsFunc, RegisterFatbinCall);
+
+    // Save GpuBinaryHandle so we can unregister it in destructor.
+    GpuBinaryHandles.push_back(GpuBinaryHandle);
+  }
+
+  CtorBuilder.CreateRetVoid();
+  return ModuleCtorFunc;
+}
+
+/// Creates a global destructor function that unregisters all GPU code blobs
+/// registered by constructor.
+/// \code
+/// void __cuda_module_dtor(void*) {
+///     __cudaUnregisterFatBinary(Handle0);
+///     ...
+///     __cudaUnregisterFatBinary(HandleN);
+/// }
+/// \endcode
+llvm::Function *CGNVCUDARuntime::makeModuleDtorFunction() {
+  // void __cudaUnregisterFatBinary(void ** handle);
+  llvm::Constant *UnregisterFatbinFunc = CGM.CreateRuntimeFunction(
+      llvm::FunctionType::get(VoidTy, VoidPtrPtrTy, false),
+      "__cudaUnregisterFatBinary");
+
+  llvm::Function *ModuleDtorFunc = llvm::Function::Create(
+      llvm::FunctionType::get(VoidTy, VoidPtrTy, false),
+      llvm::GlobalValue::InternalLinkage, "__cuda_module_dtor", &TheModule);
+  llvm::BasicBlock *DtorEntryBB =
+      llvm::BasicBlock::Create(Context, "entry", ModuleDtorFunc);
+  CGBuilderTy DtorBuilder(Context);
+  DtorBuilder.SetInsertPoint(DtorEntryBB);
+
+  for (llvm::GlobalVariable *GpuBinaryHandle : GpuBinaryHandles) {
+    DtorBuilder.CreateCall(UnregisterFatbinFunc,
+                           DtorBuilder.CreateLoad(GpuBinaryHandle, false));
+  }
+
+  DtorBuilder.CreateRetVoid();
+  return ModuleDtorFunc;
+}
+
 CGCUDARuntime *CodeGen::CreateNVCUDARuntime(CodeGenModule &CGM) {
   return new CGNVCUDARuntime(CGM);
 }
diff --git a/lib/CodeGen/CGCUDARuntime.h b/lib/CodeGen/CGCUDARuntime.h
index 8c162fb05ab9..dcacf9703277 100644
--- a/lib/CodeGen/CGCUDARuntime.h
+++ b/lib/CodeGen/CGCUDARuntime.h
@@ -16,6 +16,10 @@
 #ifndef LLVM_CLANG_LIB_CODEGEN_CGCUDARUNTIME_H
 #define LLVM_CLANG_LIB_CODEGEN_CGCUDARUNTIME_H
 
+namespace llvm {
+class Function;
+}
+
 namespace clang {
 
 class CUDAKernelCallExpr;
@@ -39,10 +43,17 @@ public:
   virtual RValue EmitCUDAKernelCallExpr(CodeGenFunction &CGF,
                                         const CUDAKernelCallExpr *E,
                                         ReturnValueSlot ReturnValue);
-  
-  virtual void EmitDeviceStubBody(CodeGenFunction &CGF,
-                                  FunctionArgList &Args) = 0;
 
+  /// Emits a kernel launch stub.
+  virtual void emitDeviceStub(CodeGenFunction &CGF, FunctionArgList &Args) = 0;
+
+  /// Constructs and returns a module initialization function or nullptr if it's
+  /// not needed. Must be called after all kernels have been emitted.
+  virtual llvm::Function *makeModuleCtorFunction() = 0;
+
+  /// Returns a module cleanup function or nullptr if it's not needed.
+  /// Must be called after ModuleCtorFunction
+  virtual llvm::Function *makeModuleDtorFunction() = 0;
 };
 
 /// Creates an instance of a CUDA runtime class.
diff --git a/lib/CodeGen/CGCXX.cpp b/lib/CodeGen/CGCXX.cpp
index 9f0e67e42176..29a199d6973a 100644
--- a/lib/CodeGen/CGCXX.cpp
+++ b/lib/CodeGen/CGCXX.cpp
@@ -182,8 +182,8 @@ bool CodeGenModule::TryEmitDefinitionAsAlias(GlobalDecl AliasDecl,
     return true;
 
   // Create the alias with no name.
-  auto *Alias = llvm::GlobalAlias::create(AliasType->getElementType(), 0,
-                                          Linkage, "", Aliasee, &getModule());
+  auto *Alias =
+      llvm::GlobalAlias::create(AliasType, Linkage, "", Aliasee, &getModule());
 
   // Switch any previous uses to the alias.
   if (Entry) {
@@ -231,8 +231,7 @@ llvm::GlobalValue *CodeGenModule::getAddrOfCXXStructor(
   if (auto *CD = dyn_cast<CXXConstructorDecl>(MD)) {
     GD = GlobalDecl(CD, toCXXCtorType(Type));
   } else {
-    auto *DD = dyn_cast<CXXDestructorDecl>(MD);
-    GD = GlobalDecl(DD, toCXXDtorType(Type));
+    GD = GlobalDecl(cast<CXXDestructorDecl>(MD), toCXXDtorType(Type));
   }
 
   StringRef Name = getMangledName(GD);
diff --git a/lib/CodeGen/CGCXXABI.cpp b/lib/CodeGen/CGCXXABI.cpp
index d31331de6868..cb7e6dfca1c5 100644
--- a/lib/CodeGen/CGCXXABI.cpp
+++ b/lib/CodeGen/CGCXXABI.cpp
@@ -302,3 +302,10 @@ CGCXXABI::EmitCtorCompleteObjectHandler(CodeGenFunction &CGF,
 bool CGCXXABI::NeedsVTTParameter(GlobalDecl GD) {
   return false;
 }
+
+llvm::CallInst *
+CGCXXABI::emitTerminateForUnexpectedException(CodeGenFunction &CGF,
+                                              llvm::Value *Exn) {
+  // Just call std::terminate and ignore the violating exception.
+  return CGF.EmitNounwindRuntimeCall(CGF.CGM.getTerminateFn());
+}
diff --git a/lib/CodeGen/CGCXXABI.h b/lib/CodeGen/CGCXXABI.h
index cc5c1b2e0ae6..2c7392188f78 100644
--- a/lib/CodeGen/CGCXXABI.h
+++ b/lib/CodeGen/CGCXXABI.h
@@ -22,6 +22,7 @@ namespace llvm {
 class Constant;
 class Type;
 class Value;
+class CallInst;
 }
 
 namespace clang {
@@ -214,8 +215,18 @@ public:
                                        llvm::Value *Ptr, QualType ElementType,
                                        const CXXDestructorDecl *Dtor) = 0;
   virtual void emitRethrow(CodeGenFunction &CGF, bool isNoReturn) = 0;
+  virtual void emitThrow(CodeGenFunction &CGF, const CXXThrowExpr *E) = 0;
+  virtual llvm::GlobalVariable *getThrowInfo(QualType T) { return nullptr; }
+
+  virtual void emitBeginCatch(CodeGenFunction &CGF, const CXXCatchStmt *C) = 0;
+
+  virtual llvm::CallInst *
+  emitTerminateForUnexpectedException(CodeGenFunction &CGF,
+                                      llvm::Value *Exn);
 
   virtual llvm::Constant *getAddrOfRTTIDescriptor(QualType Ty) = 0;
+  virtual llvm::Constant *
+  getAddrOfCXXCatchHandlerType(QualType Ty, QualType CatchHandlerType) = 0;
 
   virtual bool shouldTypeidBeNullChecked(bool IsDeref,
                                          QualType SrcRecordTy) = 0;
diff --git a/lib/CodeGen/CGCall.cpp b/lib/CodeGen/CGCall.cpp
index 6403fa99aa7b..f2340530d66d 100644
--- a/lib/CodeGen/CGCall.cpp
+++ b/lib/CodeGen/CGCall.cpp
@@ -30,7 +30,9 @@
 #include "llvm/IR/DataLayout.h"
 #include "llvm/IR/InlineAsm.h"
 #include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicInst.h"
 #include "llvm/Transforms/Utils/Local.h"
+#include <sstream>
 using namespace clang;
 using namespace CodeGen;
 
@@ -51,6 +53,8 @@ static unsigned ClangCallConvToLLVMCallConv(CallingConv CC) {
   case CC_X86Pascal: return llvm::CallingConv::C;
   // TODO: Add support for __vectorcall to LLVM.
   case CC_X86VectorCall: return llvm::CallingConv::X86_VectorCall;
+  case CC_SpirFunction: return llvm::CallingConv::SPIR_FUNC;
+  case CC_SpirKernel: return llvm::CallingConv::SPIR_KERNEL;
   }
 }
 
@@ -96,8 +100,7 @@ arrangeLLVMFunctionInfo(CodeGenTypes &CGT, bool instanceMethod,
                         CanQual<FunctionProtoType> FTP) {
   RequiredArgs required = RequiredArgs::forPrototypePlus(FTP, prefix.size());
   // FIXME: Kill copy.
-  for (unsigned i = 0, e = FTP->getNumParams(); i != e; ++i)
-    prefix.push_back(FTP->getParamType(i));
+  prefix.append(FTP->param_type_begin(), FTP->param_type_end());
   CanQualType resultType = FTP->getReturnType().getUnqualifiedType();
   return CGT.arrangeLLVMFunctionInfo(resultType, instanceMethod,
                                      /*chainCall=*/false, prefix,
@@ -133,9 +136,6 @@ static CallingConv getCallingConventionForDecl(const Decl *D, bool IsWindows) {
   if (PcsAttr *PCS = D->getAttr<PcsAttr>())
     return (PCS->getPCS() == PcsAttr::AAPCS ? CC_AAPCS : CC_AAPCS_VFP);
 
-  if (D->hasAttr<PnaclCallAttr>())
-    return CC_PnaclCall;
-
   if (D->hasAttr<IntelOclBiccAttr>())
     return CC_IntelOclBicc;
 
@@ -208,8 +208,7 @@ CodeGenTypes::arrangeCXXStructorDeclaration(const CXXMethodDecl *MD,
   CanQual<FunctionProtoType> FTP = GetFormalType(MD);
 
   // Add the formal parameters.
-  for (unsigned i = 0, e = FTP->getNumParams(); i != e; ++i)
-    argTypes.push_back(FTP->getParamType(i));
+  argTypes.append(FTP->param_type_begin(), FTP->param_type_end());
 
   TheCXXABI.buildStructorSignature(MD, Type, argTypes);
 
@@ -349,6 +348,26 @@ CodeGenTypes::arrangeMSMemberPointerThunk(const CXXMethodDecl *MD) {
                                  FTP->getExtInfo(), RequiredArgs(1));
 }
 
+const CGFunctionInfo &
+CodeGenTypes::arrangeMSCtorClosure(const CXXConstructorDecl *CD,
+                                   CXXCtorType CT) {
+  assert(CT == Ctor_CopyingClosure || CT == Ctor_DefaultClosure);
+
+  CanQual<FunctionProtoType> FTP = GetFormalType(CD);
+  SmallVector<CanQualType, 2> ArgTys;
+  const CXXRecordDecl *RD = CD->getParent();
+  ArgTys.push_back(GetThisType(Context, RD));
+  if (CT == Ctor_CopyingClosure)
+    ArgTys.push_back(*FTP->param_type_begin());
+  if (RD->getNumVBases() > 0)
+    ArgTys.push_back(Context.IntTy);
+  CallingConv CC = Context.getDefaultCallingConvention(
+      /*IsVariadic=*/false, /*IsCXXMethod=*/true);
+  return arrangeLLVMFunctionInfo(Context.VoidTy, /*instanceMethod=*/true,
+                                 /*chainCall=*/false, ArgTys,
+                                 FunctionType::ExtInfo(CC), RequiredArgs::All);
+}
+
 /// Arrange a call as unto a free function, except possibly with an
 /// additional number of formal parameters considered required.
 static const CGFunctionInfo &
@@ -716,7 +735,8 @@ void CodeGenFunction::ExpandTypeFromArgs(
   auto Exp = getTypeExpansion(Ty, getContext());
   if (auto CAExp = dyn_cast<ConstantArrayExpansion>(Exp.get())) {
     for (int i = 0, n = CAExp->NumElts; i < n; i++) {
-      llvm::Value *EltAddr = Builder.CreateConstGEP2_32(LV.getAddress(), 0, i);
+      llvm::Value *EltAddr =
+          Builder.CreateConstGEP2_32(nullptr, LV.getAddress(), 0, i);
       LValue LV = MakeAddrLValue(EltAddr, CAExp->EltTy);
       ExpandTypeFromArgs(CAExp->EltTy, LV, AI);
     }
@@ -738,10 +758,12 @@ void CodeGenFunction::ExpandTypeFromArgs(
       ExpandTypeFromArgs(FD->getType(), SubLV, AI);
     }
   } else if (auto CExp = dyn_cast<ComplexExpansion>(Exp.get())) {
-    llvm::Value *RealAddr = Builder.CreateStructGEP(LV.getAddress(), 0, "real");
+    llvm::Value *RealAddr =
+        Builder.CreateStructGEP(nullptr, LV.getAddress(), 0, "real");
     EmitStoreThroughLValue(RValue::get(*AI++),
                            MakeAddrLValue(RealAddr, CExp->EltTy));
-    llvm::Value *ImagAddr = Builder.CreateStructGEP(LV.getAddress(), 1, "imag");
+    llvm::Value *ImagAddr =
+        Builder.CreateStructGEP(nullptr, LV.getAddress(), 1, "imag");
     EmitStoreThroughLValue(RValue::get(*AI++),
                            MakeAddrLValue(ImagAddr, CExp->EltTy));
   } else {
@@ -757,7 +779,7 @@ void CodeGenFunction::ExpandTypeToArgs(
   if (auto CAExp = dyn_cast<ConstantArrayExpansion>(Exp.get())) {
     llvm::Value *Addr = RV.getAggregateAddr();
     for (int i = 0, n = CAExp->NumElts; i < n; i++) {
-      llvm::Value *EltAddr = Builder.CreateConstGEP2_32(Addr, 0, i);
+      llvm::Value *EltAddr = Builder.CreateConstGEP2_32(nullptr, Addr, 0, i);
       RValue EltRV =
           convertTempToRValue(EltAddr, CAExp->EltTy, SourceLocation());
       ExpandTypeToArgs(CAExp->EltTy, EltRV, IRFuncTy, IRCallArgs, IRCallArgPos);
@@ -825,7 +847,7 @@ EnterStructPointerForCoercedAccess(llvm::Value *SrcPtr,
     return SrcPtr;
 
   // GEP into the first element.
-  SrcPtr = CGF.Builder.CreateConstGEP2_32(SrcPtr, 0, 0, "coerce.dive");
+  SrcPtr = CGF.Builder.CreateConstGEP2_32(SrcSTy, SrcPtr, 0, 0, "coerce.dive");
 
   // If the first element is a struct, recurse.
   llvm::Type *SrcTy =
@@ -963,7 +985,7 @@ static void BuildAggStore(CodeGenFunction &CGF, llvm::Value *Val,
   if (llvm::StructType *STy =
         dyn_cast<llvm::StructType>(Val->getType())) {
     for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
-      llvm::Value *EltPtr = CGF.Builder.CreateConstGEP2_32(DestPtr, 0, i);
+      llvm::Value *EltPtr = CGF.Builder.CreateConstGEP2_32(STy, DestPtr, 0, i);
       llvm::Value *Elt = CGF.Builder.CreateExtractValue(Val, i);
       llvm::StoreInst *SI = CGF.Builder.CreateStore(Elt, EltPtr,
                                                     DestIsVolatile);
@@ -1403,7 +1425,7 @@ void CodeGenModule::ConstructAttributeList(const CGFunctionInfo &FI,
       FuncAttrs.addAttribute(llvm::Attribute::ReadOnly);
       FuncAttrs.addAttribute(llvm::Attribute::NoUnwind);
     }
-    if (TargetDecl->hasAttr<MallocAttr>())
+    if (TargetDecl->hasAttr<RestrictAttr>())
       RetAttrs.addAttribute(llvm::Attribute::NoAlias);
     if (TargetDecl->hasAttr<ReturnsNonNullAttr>())
       RetAttrs.addAttribute(llvm::Attribute::NonNull);
@@ -1458,6 +1480,26 @@ void CodeGenModule::ConstructAttributeList(const CGFunctionInfo &FI,
 
     if (!CodeGenOpts.StackRealignment)
       FuncAttrs.addAttribute("no-realign-stack");
+
+    // Add target-cpu and target-features work if they differ from the defaults.
+    std::string &CPU = getTarget().getTargetOpts().CPU;
+    if (CPU != "")
+      FuncAttrs.addAttribute("target-cpu", CPU);
+
+    // TODO: Features gets us the features on the command line including
+    // feature dependencies. For canonicalization purposes we might want to
+    // avoid putting features in the target-features set if we know it'll be one
+    // of the default features in the backend, e.g. corei7-avx and +avx or figure
+    // out non-explicit dependencies.
+    std::vector<std::string> &Features = getTarget().getTargetOpts().Features;
+    if (!Features.empty()) {
+      std::stringstream S;
+      std::copy(Features.begin(), Features.end(),
+                std::ostream_iterator<std::string>(S, ","));
+      // The drop_back gets rid of the trailing space.
+      FuncAttrs.addAttribute("target-features",
+                             StringRef(S.str()).drop_back(1));
+    }
   }
 
   ClangToLLVMArgMapping IRFunctionArgs(getContext(), FI);
@@ -1546,8 +1588,12 @@ void CodeGenModule::ConstructAttributeList(const CGFunctionInfo &FI,
     case ABIArgInfo::Extend:
       if (ParamType->isSignedIntegerOrEnumerationType())
         Attrs.addAttribute(llvm::Attribute::SExt);
-      else if (ParamType->isUnsignedIntegerOrEnumerationType())
-        Attrs.addAttribute(llvm::Attribute::ZExt);
+      else if (ParamType->isUnsignedIntegerOrEnumerationType()) {
+        if (getTypes().getABIInfo().shouldSignExtUnsignedType(ParamType))
+          Attrs.addAttribute(llvm::Attribute::SExt);
+        else
+          Attrs.addAttribute(llvm::Attribute::ZExt);
+      }
       // FALL THROUGH
     case ABIArgInfo::Direct:
       if (ArgNo == 0 && FI.isChainCall())
@@ -1734,8 +1780,9 @@ void CodeGenFunction::EmitFunctionProlog(const CGFunctionInfo &FI,
     switch (ArgI.getKind()) {
     case ABIArgInfo::InAlloca: {
       assert(NumIRArgs == 0);
-      llvm::Value *V = Builder.CreateStructGEP(
-          ArgStruct, ArgI.getInAllocaFieldIndex(), Arg->getName());
+      llvm::Value *V =
+          Builder.CreateStructGEP(FI.getArgStruct(), ArgStruct,
+                                  ArgI.getInAllocaFieldIndex(), Arg->getName());
       ArgVals.push_back(ValueAndIsPtr(V, HavePointer));
       break;
     }
@@ -1770,8 +1817,7 @@ void CodeGenFunction::EmitFunctionProlog(const CGFunctionInfo &FI,
         ArgVals.push_back(ValueAndIsPtr(V, HavePointer));
       } else {
         // Load scalar value from indirect argument.
-        CharUnits Alignment = getContext().getTypeAlignInChars(Ty);
-        V = EmitLoadOfScalar(V, false, Alignment.getQuantity(), Ty,
+        V = EmitLoadOfScalar(V, false, ArgI.getIndirectAlign(), Ty,
                              Arg->getLocStart());
 
         if (isPromoted)
@@ -1901,7 +1947,7 @@ void CodeGenFunction::EmitFunctionProlog(const CGFunctionInfo &FI,
       // If the value is offset in memory, apply the offset now.
       if (unsigned Offs = ArgI.getDirectOffset()) {
         Ptr = Builder.CreateBitCast(Ptr, Builder.getInt8PtrTy());
-        Ptr = Builder.CreateConstGEP1_32(Ptr, Offs);
+        Ptr = Builder.CreateConstGEP1_32(Builder.getInt8Ty(), Ptr, Offs);
         Ptr = Builder.CreateBitCast(Ptr,
                           llvm::PointerType::getUnqual(ArgI.getCoerceToType()));
       }
@@ -1923,7 +1969,7 @@ void CodeGenFunction::EmitFunctionProlog(const CGFunctionInfo &FI,
           for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
             auto AI = FnArgs[FirstIRArg + i];
             AI->setName(Arg->getName() + ".coerce" + Twine(i));
-            llvm::Value *EltPtr = Builder.CreateConstGEP2_32(Ptr, 0, i);
+            llvm::Value *EltPtr = Builder.CreateConstGEP2_32(STy, Ptr, 0, i);
             Builder.CreateStore(AI, EltPtr);
           }
         } else {
@@ -1936,7 +1982,8 @@ void CodeGenFunction::EmitFunctionProlog(const CGFunctionInfo &FI,
           for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
             auto AI = FnArgs[FirstIRArg + i];
             AI->setName(Arg->getName() + ".coerce" + Twine(i));
-            llvm::Value *EltPtr = Builder.CreateConstGEP2_32(TempV, 0, i);
+            llvm::Value *EltPtr =
+                Builder.CreateConstGEP2_32(ArgI.getCoerceToType(), TempV, 0, i);
             Builder.CreateStore(AI, EltPtr);
           }
 
@@ -2173,7 +2220,29 @@ static llvm::StoreInst *findDominatingStoreToReturnValue(CodeGenFunction &CGF) {
   if (!CGF.ReturnValue->hasOneUse()) {
     llvm::BasicBlock *IP = CGF.Builder.GetInsertBlock();
     if (IP->empty()) return nullptr;
-    llvm::StoreInst *store = dyn_cast<llvm::StoreInst>(&IP->back());
+    llvm::Instruction *I = &IP->back();
+
+    // Skip lifetime markers
+    for (llvm::BasicBlock::reverse_iterator II = IP->rbegin(),
+                                            IE = IP->rend();
+         II != IE; ++II) {
+      if (llvm::IntrinsicInst *Intrinsic =
+              dyn_cast<llvm::IntrinsicInst>(&*II)) {
+        if (Intrinsic->getIntrinsicID() == llvm::Intrinsic::lifetime_end) {
+          const llvm::Value *CastAddr = Intrinsic->getArgOperand(1);
+          ++II;
+          if (isa<llvm::BitCastInst>(&*II)) {
+            if (CastAddr == &*II) {
+              continue;
+            }
+          }
+        }
+      }
+      I = &*II;
+      break;
+    }
+
+    llvm::StoreInst *store = dyn_cast<llvm::StoreInst>(I);
     if (!store) return nullptr;
     if (store->getPointerOperand() != CGF.ReturnValue) return nullptr;
     assert(!store->isAtomic() && !store->isVolatile()); // see below
@@ -2231,8 +2300,8 @@ void CodeGenFunction::EmitFunctionEpilog(const CGFunctionInfo &FI,
       llvm::Function::arg_iterator EI = CurFn->arg_end();
       --EI;
       llvm::Value *ArgStruct = EI;
-      llvm::Value *SRet =
-          Builder.CreateStructGEP(ArgStruct, RetAI.getInAllocaFieldIndex());
+      llvm::Value *SRet = Builder.CreateStructGEP(
+          nullptr, ArgStruct, RetAI.getInAllocaFieldIndex());
       RV = Builder.CreateLoad(SRet, "sret");
     }
     break;
@@ -2271,7 +2340,8 @@ void CodeGenFunction::EmitFunctionEpilog(const CGFunctionInfo &FI,
 
       // If there is a dominating store to ReturnValue, we can elide
       // the load, zap the store, and usually zap the alloca.
-      if (llvm::StoreInst *SI = findDominatingStoreToReturnValue(*this)) {
+      if (llvm::StoreInst *SI =
+              findDominatingStoreToReturnValue(*this)) {
         // Reuse the debug location from the store unless there is
         // cleanup code to be emitted between the store and return
         // instruction.
@@ -2296,7 +2366,7 @@ void CodeGenFunction::EmitFunctionEpilog(const CGFunctionInfo &FI,
       // If the value is offset in memory, apply the offset now.
       if (unsigned Offs = RetAI.getDirectOffset()) {
         V = Builder.CreateBitCast(V, Builder.getInt8PtrTy());
-        V = Builder.CreateConstGEP1_32(V, Offs);
+        V = Builder.CreateConstGEP1_32(Builder.getInt8Ty(), V, Offs);
         V = Builder.CreateBitCast(V,
                          llvm::PointerType::getUnqual(RetAI.getCoerceToType()));
       }
@@ -2342,8 +2412,8 @@ void CodeGenFunction::EmitFunctionEpilog(const CGFunctionInfo &FI,
     Ret = Builder.CreateRetVoid();
   }
 
-  if (!RetDbgLoc.isUnknown())
-    Ret->setDebugLoc(RetDbgLoc);
+  if (RetDbgLoc)
+    Ret->setDebugLoc(std::move(RetDbgLoc));
 }
 
 static bool isInAllocaArgument(CGCXXABI &ABI, QualType type) {
@@ -2626,7 +2696,7 @@ void CallArgList::allocateArgumentMemory(CodeGenFunction &CGF) {
 
   // Save the stack.
   llvm::Function *F = CGF.CGM.getIntrinsic(llvm::Intrinsic::stacksave);
-  StackBase = CGF.Builder.CreateCall(F, "inalloca.save");
+  StackBase = CGF.Builder.CreateCall(F, {}, "inalloca.save");
 
   // Control gets really tied up in landing pads, so we have to spill the
   // stacksave to an alloca to avoid violating SSA form.
@@ -2678,12 +2748,7 @@ void CodeGenFunction::EmitCallArgs(CallArgList &Args,
                                    CallExpr::const_arg_iterator ArgBeg,
                                    CallExpr::const_arg_iterator ArgEnd,
                                    const FunctionDecl *CalleeDecl,
-                                   unsigned ParamsToSkip,
-                                   bool ForceColumnInfo) {
-  CGDebugInfo *DI = getDebugInfo();
-  SourceLocation CallLoc;
-  if (DI) CallLoc = DI->getLocation();
-
+                                   unsigned ParamsToSkip) {
   // We *have* to evaluate arguments from right to left in the MS C++ ABI,
   // because arguments are destroyed left to right in the callee.
   if (CGM.getTarget().getCXXABI().areArgsDestroyedLeftToRightInCallee()) {
@@ -2704,8 +2769,6 @@ void CodeGenFunction::EmitCallArgs(CallArgList &Args,
       EmitCallArg(Args, *Arg, ArgTypes[I]);
       emitNonNullArgCheck(*this, Args.back().RV, ArgTypes[I], Arg->getExprLoc(),
                           CalleeDecl, ParamsToSkip + I);
-      // Restore the debug location.
-      if (DI) DI->EmitLocation(Builder, CallLoc, ForceColumnInfo);
     }
 
     // Un-reverse the arguments we just evaluated so they match up with the LLVM
@@ -2720,8 +2783,6 @@ void CodeGenFunction::EmitCallArgs(CallArgList &Args,
     EmitCallArg(Args, *Arg, ArgTypes[I]);
     emitNonNullArgCheck(*this, Args.back().RV, ArgTypes[I], Arg->getExprLoc(),
                         CalleeDecl, ParamsToSkip + I);
-    // Restore the debug location.
-    if (DI) DI->EmitLocation(Builder, CallLoc, ForceColumnInfo);
   }
 }
 
@@ -2744,8 +2805,22 @@ struct DestroyUnpassedArg : EHScopeStack::Cleanup {
 
 }
 
+struct DisableDebugLocationUpdates {
+  CodeGenFunction &CGF;
+  bool disabledDebugInfo;
+  DisableDebugLocationUpdates(CodeGenFunction &CGF, const Expr *E) : CGF(CGF) {
+    if ((disabledDebugInfo = isa<CXXDefaultArgExpr>(E) && CGF.getDebugInfo()))
+      CGF.disableDebugInfo();
+  }
+  ~DisableDebugLocationUpdates() {
+    if (disabledDebugInfo)
+      CGF.enableDebugInfo();
+  }
+};
+
 void CodeGenFunction::EmitCallArg(CallArgList &args, const Expr *E,
                                   QualType type) {
+  DisableDebugLocationUpdates Dis(*this, E);
   if (const ObjCIndirectCopyRestoreExpr *CRE
         = dyn_cast<ObjCIndirectCopyRestoreExpr>(E)) {
     assert(getLangOpts().ObjCAutoRefCount);
@@ -2900,7 +2975,6 @@ void CodeGenFunction::EmitNoreturnRuntimeCallOrInvoke(llvm::Value *callee,
     call->setCallingConv(getRuntimeCC());
     Builder.CreateUnreachable();
   }
-  PGO.setCurrentRegionUnreachable();
 }
 
 /// Emits a call or invoke instruction to the given nullary runtime
@@ -2949,7 +3023,7 @@ CodeGenFunction::EmitCallOrInvoke(llvm::Value *Callee,
   if (CGM.getLangOpts().ObjCAutoRefCount)
     AddObjCARCExceptionMetadata(Inst);
 
-  return Inst;
+  return llvm::CallSite(Inst);
 }
 
 /// \brief Store a non-aggregate value to an address to initialize it.  For
@@ -2986,7 +3060,7 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
 
   // If we're using inalloca, insert the allocation after the stack save.
   // FIXME: Do this earlier rather than hacking it in here!
-  llvm::Value *ArgMemory = nullptr;
+  llvm::AllocaInst *ArgMemory = nullptr;
   if (llvm::StructType *ArgStruct = CallInfo.getArgStruct()) {
     llvm::Instruction *IP = CallArgs.getStackBase();
     llvm::AllocaInst *AI;
@@ -3015,7 +3089,8 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
       IRCallArgs[IRFunctionArgs.getSRetArgNo()] = SRetPtr;
     } else {
       llvm::Value *Addr =
-          Builder.CreateStructGEP(ArgMemory, RetAI.getInAllocaFieldIndex());
+          Builder.CreateStructGEP(ArgMemory->getAllocatedType(), ArgMemory,
+                                  RetAI.getInAllocaFieldIndex());
       Builder.CreateStore(SRetPtr, Addr);
     }
   }
@@ -3049,14 +3124,16 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
             cast<llvm::Instruction>(RV.getAggregateAddr());
         CGBuilderTy::InsertPoint IP = Builder.saveIP();
         Builder.SetInsertPoint(Placeholder);
-        llvm::Value *Addr = Builder.CreateStructGEP(
-            ArgMemory, ArgInfo.getInAllocaFieldIndex());
+        llvm::Value *Addr =
+            Builder.CreateStructGEP(ArgMemory->getAllocatedType(), ArgMemory,
+                                    ArgInfo.getInAllocaFieldIndex());
         Builder.restoreIP(IP);
         deferPlaceholderReplacement(Placeholder, Addr);
       } else {
         // Store the RValue into the argument struct.
         llvm::Value *Addr =
-            Builder.CreateStructGEP(ArgMemory, ArgInfo.getInAllocaFieldIndex());
+            Builder.CreateStructGEP(ArgMemory->getAllocatedType(), ArgMemory,
+                                    ArgInfo.getInAllocaFieldIndex());
         unsigned AS = Addr->getType()->getPointerAddressSpace();
         llvm::Type *MemType = ConvertTypeForMem(I->Ty)->getPointerTo(AS);
         // There are some cases where a trivial bitcast is not avoidable.  The
@@ -3100,8 +3177,8 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
                  : 0);
         if ((!ArgInfo.getIndirectByVal() && I->NeedsCopy) ||
             (ArgInfo.getIndirectByVal() && TypeAlign.getQuantity() < Align &&
-             llvm::getOrEnforceKnownAlignment(Addr, Align, TD) < Align) ||
-             (ArgInfo.getIndirectByVal() && (RVAddrSpace != ArgAddrSpace))) {
+             llvm::getOrEnforceKnownAlignment(Addr, Align, *TD) < Align) ||
+            (ArgInfo.getIndirectByVal() && (RVAddrSpace != ArgAddrSpace))) {
           // Create an aligned temporary, and copy to it.
           llvm::AllocaInst *AI = CreateMemTemp(I->Ty);
           if (Align > AI->getAlignment())
@@ -3158,7 +3235,7 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
       // If the value is offset in memory, apply the offset now.
       if (unsigned Offs = ArgInfo.getDirectOffset()) {
         SrcPtr = Builder.CreateBitCast(SrcPtr, Builder.getInt8PtrTy());
-        SrcPtr = Builder.CreateConstGEP1_32(SrcPtr, Offs);
+        SrcPtr = Builder.CreateConstGEP1_32(Builder.getInt8Ty(), SrcPtr, Offs);
         SrcPtr = Builder.CreateBitCast(SrcPtr,
                        llvm::PointerType::getUnqual(ArgInfo.getCoerceToType()));
 
@@ -3190,7 +3267,7 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
 
         assert(NumIRArgs == STy->getNumElements());
         for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
-          llvm::Value *EltPtr = Builder.CreateConstGEP2_32(SrcPtr, 0, i);
+          llvm::Value *EltPtr = Builder.CreateConstGEP2_32(STy, SrcPtr, 0, i);
           llvm::LoadInst *LI = Builder.CreateLoad(EltPtr);
           // We don't know what we're loading from.
           LI->setAlignment(1);
@@ -3300,7 +3377,8 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
 
   llvm::BasicBlock *InvokeDest = nullptr;
   if (!Attrs.hasAttribute(llvm::AttributeSet::FunctionIndex,
-                          llvm::Attribute::NoUnwind))
+                          llvm::Attribute::NoUnwind) ||
+      currentFunctionUsesSEHTry())
     InvokeDest = getInvokeDest();
 
   llvm::CallSite CS;
@@ -3320,6 +3398,12 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
         Attrs.addAttribute(getLLVMContext(), llvm::AttributeSet::FunctionIndex,
                            llvm::Attribute::AlwaysInline);
 
+  // Disable inlining inside SEH __try blocks.
+  if (isSEHTryScope())
+    Attrs =
+        Attrs.addAttribute(getLLVMContext(), llvm::AttributeSet::FunctionIndex,
+                           llvm::Attribute::NoInline);
+
   CS.setAttributes(Attrs);
   CS.setCallingConv(static_cast<llvm::CallingConv::ID>(CallingConv));
 
@@ -3413,7 +3497,8 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
       llvm::Value *StorePtr = DestPtr;
       if (unsigned Offs = RetAI.getDirectOffset()) {
         StorePtr = Builder.CreateBitCast(StorePtr, Builder.getInt8PtrTy());
-        StorePtr = Builder.CreateConstGEP1_32(StorePtr, Offs);
+        StorePtr =
+            Builder.CreateConstGEP1_32(Builder.getInt8Ty(), StorePtr, Offs);
         StorePtr = Builder.CreateBitCast(StorePtr,
                            llvm::PointerType::getUnqual(RetAI.getCoerceToType()));
       }
diff --git a/lib/CodeGen/CGClass.cpp b/lib/CodeGen/CGClass.cpp
index d72eda95d7cf..cd75da276287 100644
--- a/lib/CodeGen/CGClass.cpp
+++ b/lib/CodeGen/CGClass.cpp
@@ -24,35 +24,36 @@
 #include "clang/Basic/TargetBuiltins.h"
 #include "clang/CodeGen/CGFunctionInfo.h"
 #include "clang/Frontend/CodeGenOptions.h"
+#include "llvm/IR/Intrinsics.h"
 
 using namespace clang;
 using namespace CodeGen;
 
-static CharUnits 
-ComputeNonVirtualBaseClassOffset(ASTContext &Context, 
+static CharUnits
+ComputeNonVirtualBaseClassOffset(ASTContext &Context,
                                  const CXXRecordDecl *DerivedClass,
                                  CastExpr::path_const_iterator Start,
                                  CastExpr::path_const_iterator End) {
   CharUnits Offset = CharUnits::Zero();
-  
+
   const CXXRecordDecl *RD = DerivedClass;
-  
+
   for (CastExpr::path_const_iterator I = Start; I != End; ++I) {
     const CXXBaseSpecifier *Base = *I;
     assert(!Base->isVirtual() && "Should not see virtual bases here!");
 
     // Get the layout.
     const ASTRecordLayout &Layout = Context.getASTRecordLayout(RD);
-    
-    const CXXRecordDecl *BaseDecl = 
+
+    const CXXRecordDecl *BaseDecl =
       cast<CXXRecordDecl>(Base->getType()->getAs<RecordType>()->getDecl());
-    
+
     // Add the offset.
     Offset += Layout.getBaseClassOffset(BaseDecl);
-    
+
     RD = BaseDecl;
   }
-  
+
   return Offset;
 }
 
@@ -62,15 +63,15 @@ CodeGenModule::GetNonVirtualBaseClassOffset(const CXXRecordDecl *ClassDecl,
                                    CastExpr::path_const_iterator PathEnd) {
   assert(PathBegin != PathEnd && "Base path should not be empty!");
 
-  CharUnits Offset = 
+  CharUnits Offset =
     ComputeNonVirtualBaseClassOffset(getContext(), ClassDecl,
                                      PathBegin, PathEnd);
   if (Offset.isZero())
     return nullptr;
 
-  llvm::Type *PtrDiffTy = 
+  llvm::Type *PtrDiffTy =
   Types.ConvertType(getContext().getPointerDiffType());
-  
+
   return llvm::ConstantInt::get(PtrDiffTy, Offset.getQuantity());
 }
 
@@ -127,7 +128,7 @@ ApplyNonVirtualAndVirtualOffset(CodeGenFunction &CGF, llvm::Value *ptr,
   } else {
     baseOffset = virtualOffset;
   }
-  
+
   // Apply the base offset.
   ptr = CGF.Builder.CreateBitCast(ptr, CGF.Int8PtrTy);
   ptr = CGF.Builder.CreateInBoundsGEP(ptr, baseOffset, "add.ptr");
@@ -149,7 +150,7 @@ llvm::Value *CodeGenFunction::GetAddressOfBaseClass(
   // *start* with a step down to the correct virtual base subobject,
   // and hence will not require any further steps.
   if ((*Start)->isVirtual()) {
-    VBase = 
+    VBase =
       cast<CXXRecordDecl>((*Start)->getType()->getAs<RecordType>()->getDecl());
     ++Start;
   }
@@ -157,7 +158,7 @@ llvm::Value *CodeGenFunction::GetAddressOfBaseClass(
   // Compute the static offset of the ultimate destination within its
   // allocating subobject (the virtual base, if there is one, or else
   // the "complete" object that we see).
-  CharUnits NonVirtualOffset = 
+  CharUnits NonVirtualOffset =
     ComputeNonVirtualBaseClassOffset(getContext(), VBase ? VBase : Derived,
                                      Start, PathEnd);
 
@@ -172,7 +173,7 @@ llvm::Value *CodeGenFunction::GetAddressOfBaseClass(
   }
 
   // Get the base pointer type.
-  llvm::Type *BasePtrTy = 
+  llvm::Type *BasePtrTy =
     ConvertType((PathEnd[-1])->getType())->getPointerTo();
 
   QualType DerivedTy = getContext().getRecordType(Derived);
@@ -197,7 +198,7 @@ llvm::Value *CodeGenFunction::GetAddressOfBaseClass(
     origBB = Builder.GetInsertBlock();
     llvm::BasicBlock *notNullBB = createBasicBlock("cast.notnull");
     endBB = createBasicBlock("cast.end");
-    
+
     llvm::Value *isNull = Builder.CreateIsNull(Value);
     Builder.CreateCondBr(isNull, endBB, notNullBB);
     EmitBlock(notNullBB);
@@ -216,10 +217,10 @@ llvm::Value *CodeGenFunction::GetAddressOfBaseClass(
   }
 
   // Apply both offsets.
-  Value = ApplyNonVirtualAndVirtualOffset(*this, Value, 
+  Value = ApplyNonVirtualAndVirtualOffset(*this, Value,
                                           NonVirtualOffset,
                                           VirtualOffset);
-  
+
   // Cast to the destination type.
   Value = Builder.CreateBitCast(Value, BasePtrTy);
 
@@ -228,13 +229,13 @@ llvm::Value *CodeGenFunction::GetAddressOfBaseClass(
     llvm::BasicBlock *notNullBB = Builder.GetInsertBlock();
     Builder.CreateBr(endBB);
     EmitBlock(endBB);
-    
+
     llvm::PHINode *PHI = Builder.CreatePHI(BasePtrTy, 2, "cast.result");
     PHI->addIncoming(Value, notNullBB);
     PHI->addIncoming(llvm::Constant::getNullValue(BasePtrTy), origBB);
     Value = PHI;
   }
-  
+
   return Value;
 }
 
@@ -252,7 +253,7 @@ CodeGenFunction::GetAddressOfDerivedClass(llvm::Value *Value,
 
   llvm::Value *NonVirtualOffset =
     CGM.GetNonVirtualBaseClassOffset(Derived, PathBegin, PathEnd);
-  
+
   if (!NonVirtualOffset) {
     // No offset, we can just cast back.
     return Builder.CreateBitCast(Value, DerivedPtrTy);
@@ -266,12 +267,12 @@ CodeGenFunction::GetAddressOfDerivedClass(llvm::Value *Value,
     CastNull = createBasicBlock("cast.null");
     CastNotNull = createBasicBlock("cast.notnull");
     CastEnd = createBasicBlock("cast.end");
-    
+
     llvm::Value *IsNull = Builder.CreateIsNull(Value);
     Builder.CreateCondBr(IsNull, CastNull, CastNotNull);
     EmitBlock(CastNotNull);
   }
-  
+
   // Apply the offset.
   Value = Builder.CreateBitCast(Value, Int8PtrTy);
   Value = Builder.CreateGEP(Value, Builder.CreateNeg(NonVirtualOffset),
@@ -285,14 +286,14 @@ CodeGenFunction::GetAddressOfDerivedClass(llvm::Value *Value,
     EmitBlock(CastNull);
     Builder.CreateBr(CastEnd);
     EmitBlock(CastEnd);
-    
+
     llvm::PHINode *PHI = Builder.CreatePHI(Value->getType(), 2);
     PHI->addIncoming(Value, CastNotNull);
-    PHI->addIncoming(llvm::Constant::getNullValue(Value->getType()), 
+    PHI->addIncoming(llvm::Constant::getNullValue(Value->getType()),
                      CastNull);
     Value = PHI;
   }
-  
+
   return Value;
 }
 
@@ -303,7 +304,7 @@ llvm::Value *CodeGenFunction::GetVTTParameter(GlobalDecl GD,
     // This constructor/destructor does not need a VTT parameter.
     return nullptr;
   }
-  
+
   const CXXRecordDecl *RD = cast<CXXMethodDecl>(CurCodeDecl)->getParent();
   const CXXRecordDecl *Base = cast<CXXMethodDecl>(GD.getDecl())->getParent();
 
@@ -323,15 +324,15 @@ llvm::Value *CodeGenFunction::GetVTTParameter(GlobalDecl GD,
     SubVTTIndex = 0;
   } else {
     const ASTRecordLayout &Layout = getContext().getASTRecordLayout(RD);
-    CharUnits BaseOffset = ForVirtualBase ? 
-      Layout.getVBaseClassOffset(Base) : 
+    CharUnits BaseOffset = ForVirtualBase ?
+      Layout.getVBaseClassOffset(Base) :
       Layout.getBaseClassOffset(Base);
 
-    SubVTTIndex = 
+    SubVTTIndex =
       CGM.getVTables().getSubVTTIndex(RD, BaseSubobject(Base, BaseOffset));
     assert(SubVTTIndex != 0 && "Sub-VTT index must be greater than zero!");
   }
-  
+
   if (CGM.getCXXABI().NeedsVTTParameter(CurGD)) {
     // A VTT parameter was passed to the constructor, use it.
     VTT = LoadCXXVTT();
@@ -358,7 +359,7 @@ namespace {
         cast<CXXMethodDecl>(CGF.CurCodeDecl)->getParent();
 
       const CXXDestructorDecl *D = BaseClass->getDestructor();
-      llvm::Value *Addr = 
+      llvm::Value *Addr =
         CGF.GetAddressOfDirectBaseInCompleteClass(CGF.LoadCXXThis(),
                                                   DerivedClass, BaseClass,
                                                   BaseIsVirtual);
@@ -391,7 +392,7 @@ static bool BaseInitializerUsesThis(ASTContext &C, const Expr *Init) {
   return Checker.UsesThis;
 }
 
-static void EmitBaseInitializer(CodeGenFunction &CGF, 
+static void EmitBaseInitializer(CodeGenFunction &CGF,
                                 const CXXRecordDecl *ClassDecl,
                                 CXXCtorInitializer *BaseInit,
                                 CXXCtorType CtorType) {
@@ -399,7 +400,7 @@ static void EmitBaseInitializer(CodeGenFunction &CGF,
          "Must have base initializer!");
 
   llvm::Value *ThisPtr = CGF.LoadCXXThis();
-  
+
   const Type *BaseType = BaseInit->getBaseClass();
   CXXRecordDecl *BaseClassDecl =
     cast<CXXRecordDecl>(BaseType->getAs<RecordType>()->getDecl());
@@ -418,7 +419,7 @@ static void EmitBaseInitializer(CodeGenFunction &CGF,
 
   // We can pretend to be a complete class because it only matters for
   // virtual bases, and we only do virtual bases for complete ctors.
-  llvm::Value *V = 
+  llvm::Value *V =
     CGF.GetAddressOfDirectBaseInCompleteClass(ThisPtr, ClassDecl,
                                               BaseClassDecl,
                                               isBaseVirtual);
@@ -430,8 +431,8 @@ static void EmitBaseInitializer(CodeGenFunction &CGF,
                           AggValueSlot::IsNotAliased);
 
   CGF.EmitAggExpr(BaseInit->getInit(), AggSlot);
-  
-  if (CGF.CGM.getLangOpts().Exceptions && 
+
+  if (CGF.CGM.getLangOpts().Exceptions &&
       !BaseClassDecl->hasTrivialDestructor())
     CGF.EHStack.pushCleanup<CallBaseDtor>(EHCleanup, BaseClassDecl,
                                           isBaseVirtual);
@@ -490,17 +491,17 @@ static void EmitAggMemberInitializer(CodeGenFunction &CGF,
   llvm::Value *IndexVar
     = CGF.GetAddrOfLocalVar(ArrayIndexes[Index]);
   assert(IndexVar && "Array index variable not loaded");
-  
+
   // Initialize this index variable to zero.
   llvm::Value* Zero
     = llvm::Constant::getNullValue(
                               CGF.ConvertType(CGF.getContext().getSizeType()));
   CGF.Builder.CreateStore(Zero, IndexVar);
-                                   
+
   // Start the loop with a block that tests the condition.
   llvm::BasicBlock *CondBlock = CGF.createBasicBlock("for.cond");
   llvm::BasicBlock *AfterFor = CGF.createBasicBlock("for.end");
-  
+
   CGF.EmitBlock(CondBlock);
 
   llvm::BasicBlock *ForBody = CGF.createBasicBlock("for.body");
@@ -512,7 +513,7 @@ static void EmitAggMemberInitializer(CodeGenFunction &CGF,
     llvm::ConstantInt::get(Counter->getType(), NumElements);
   llvm::Value *IsLess = CGF.Builder.CreateICmpULT(Counter, NumElementsPtr,
                                                   "isless");
-                                   
+
   // If the condition is true, execute the body.
   CGF.Builder.CreateCondBr(IsLess, ForBody, AfterFor);
 
@@ -539,6 +540,23 @@ static void EmitAggMemberInitializer(CodeGenFunction &CGF,
   CGF.EmitBlock(AfterFor, true);
 }
 
+static bool isMemcpyEquivalentSpecialMember(const CXXMethodDecl *D) {
+  auto *CD = dyn_cast<CXXConstructorDecl>(D);
+  if (!(CD && CD->isCopyOrMoveConstructor()) &&
+      !D->isCopyAssignmentOperator() && !D->isMoveAssignmentOperator())
+    return false;
+
+  // We can emit a memcpy for a trivial copy or move constructor/assignment.
+  if (D->isTrivial() && !D->getParent()->mayInsertExtraPadding())
+    return true;
+
+  // We *must* emit a memcpy for a defaulted union copy or move op.
+  if (D->getParent()->isUnion() && D->isDefaulted())
+    return true;
+
+  return false;
+}
+
 static void EmitMemberInitializer(CodeGenFunction &CGF,
                                   const CXXRecordDecl *ClassDecl,
                                   CXXCtorInitializer *MemberInit,
@@ -548,7 +566,7 @@ static void EmitMemberInitializer(CodeGenFunction &CGF,
   assert(MemberInit->isAnyMemberInitializer() &&
          "Must have member initializer!");
   assert(MemberInit->getInit() && "Must have initializer!");
-  
+
   // non-static data member initializers.
   FieldDecl *Field = MemberInit->getAnyMember();
   QualType FieldType = Field->getType();
@@ -580,14 +598,14 @@ static void EmitMemberInitializer(CodeGenFunction &CGF,
     QualType BaseElementTy = CGF.getContext().getBaseElementType(Array);
     CXXConstructExpr *CE = dyn_cast<CXXConstructExpr>(MemberInit->getInit());
     if (BaseElementTy.isPODType(CGF.getContext()) ||
-        (CE && CE->getConstructor()->isTrivial())) {
+        (CE && isMemcpyEquivalentSpecialMember(CE->getConstructor()))) {
       unsigned SrcArgIndex =
           CGF.CGM.getCXXABI().getSrcArgforCopyCtor(Constructor, Args);
       llvm::Value *SrcPtr
         = CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(Args[SrcArgIndex]));
       LValue ThisRHSLV = CGF.MakeNaturalAlignAddrLValue(SrcPtr, RecordTy);
       LValue Src = CGF.EmitLValueForFieldInitialization(ThisRHSLV, Field);
-      
+
       // Copy the aggregate.
       CGF.EmitAggregateCopy(LHS.getAddress(), Src.getAddress(), FieldType,
                             LHS.isVolatileQualified());
@@ -621,28 +639,28 @@ void CodeGenFunction::EmitInitializerForField(
     llvm::Value *ArrayIndexVar = nullptr;
     if (ArrayIndexes.size()) {
       llvm::Type *SizeTy = ConvertType(getContext().getSizeType());
-      
+
       // The LHS is a pointer to the first object we'll be constructing, as
       // a flat array.
       QualType BaseElementTy = getContext().getBaseElementType(FieldType);
       llvm::Type *BasePtr = ConvertType(BaseElementTy);
       BasePtr = llvm::PointerType::getUnqual(BasePtr);
-      llvm::Value *BaseAddrPtr = Builder.CreateBitCast(LHS.getAddress(), 
+      llvm::Value *BaseAddrPtr = Builder.CreateBitCast(LHS.getAddress(),
                                                        BasePtr);
       LHS = MakeAddrLValue(BaseAddrPtr, BaseElementTy);
-      
+
       // Create an array index that will be used to walk over all of the
       // objects we're constructing.
       ArrayIndexVar = CreateTempAlloca(SizeTy, "object.index");
       llvm::Value *Zero = llvm::Constant::getNullValue(SizeTy);
       Builder.CreateStore(Zero, ArrayIndexVar);
-      
-      
+
+
       // Emit the block variables for the array indices, if any.
       for (unsigned I = 0, N = ArrayIndexes.size(); I != N; ++I)
         EmitAutoVarDecl(*ArrayIndexes[I]);
     }
-    
+
     EmitAggMemberInitializer(*this, LHS, Init, ArrayIndexVar, FieldType,
                              ArrayIndexes, 0);
   }
@@ -762,9 +780,9 @@ void CodeGenFunction::EmitAsanPrologueOrEpilogue(bool Prologue) {
     if (PoisonSize < AsanAlignment || !SSV[i].Size ||
         (NextField % AsanAlignment) != 0)
       continue;
-    Builder.CreateCall2(
-        F, Builder.CreateAdd(ThisPtr, Builder.getIntN(PtrSize, EndOffset)),
-        Builder.getIntN(PtrSize, PoisonSize));
+    Builder.CreateCall(
+        F, {Builder.CreateAdd(ThisPtr, Builder.getIntN(PtrSize, EndOffset)),
+            Builder.getIntN(PtrSize, PoisonSize)});
   }
 }
 
@@ -796,8 +814,7 @@ void CodeGenFunction::EmitConstructorBody(FunctionArgList &Args) {
   if (IsTryBody)
     EnterCXXTryStmt(*cast<CXXTryStmt>(Body), true);
 
-  RegionCounter Cnt = getPGORegionCounter(Body);
-  Cnt.beginRegion(Builder);
+  incrementProfileCounter(Body);
 
   RunCleanupsScope RunCleanups(*this);
 
@@ -850,7 +867,7 @@ namespace {
   public:
     FieldMemcpyizer(CodeGenFunction &CGF, const CXXRecordDecl *ClassDecl,
                     const VarDecl *SrcRec)
-      : CGF(CGF), ClassDecl(ClassDecl), SrcRec(SrcRec), 
+      : CGF(CGF), ClassDecl(ClassDecl), SrcRec(SrcRec),
         RecLayout(CGF.getContext().getASTRecordLayout(ClassDecl)),
         FirstField(nullptr), LastField(nullptr), FirstFieldOffset(0),
         LastFieldOffset(0), LastAddedFieldIndex(0) {}
@@ -876,7 +893,7 @@ namespace {
       unsigned LastFieldSize =
         LastField->isBitField() ?
           LastField->getBitWidthValue(CGF.getContext()) :
-          CGF.getContext().getTypeSize(LastField->getType()); 
+          CGF.getContext().getTypeSize(LastField->getType());
       uint64_t MemcpySizeBits =
         LastFieldOffset + LastFieldSize - FirstByteOffset +
         CGF.getContext().getCharWidth() - 1;
@@ -1021,8 +1038,8 @@ namespace {
       QualType FieldType = Field->getType();
       CXXConstructExpr *CE = dyn_cast<CXXConstructExpr>(MemberInit->getInit());
 
-      // Bail out on non-POD, not-trivially-constructable members.
-      if (!(CE && CE->getConstructor()->isTrivial()) &&
+      // Bail out on non-memcpyable, not-trivially-copyable members.
+      if (!(CE && isMemcpyEquivalentSpecialMember(CE->getConstructor())) &&
           !(FieldType.isTriviallyCopyableType(CGF.getContext()) ||
             FieldType->isReferenceType()))
         return false;
@@ -1127,9 +1144,7 @@ namespace {
         return Field;
       } else if (CXXMemberCallExpr *MCE = dyn_cast<CXXMemberCallExpr>(S)) {
         CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(MCE->getCalleeDecl());
-        if (!(MD && (MD->isCopyAssignmentOperator() ||
-                       MD->isMoveAssignmentOperator()) &&
-              MD->isTrivial()))
+        if (!(MD && isMemcpyEquivalentSpecialMember(MD)))
           return nullptr;
         MemberExpr *IOA = dyn_cast<MemberExpr>(MCE->getImplicitObjectArgument());
         if (!IOA)
@@ -1189,7 +1204,7 @@ namespace {
       if (F) {
         addMemcpyableField(F);
         AggregatedStmts.push_back(S);
-      } else {  
+      } else {
         emitAggregatedStmts();
         CGF.EmitStmt(S);
       }
@@ -1274,7 +1289,7 @@ static bool
 FieldHasTrivialDestructorBody(ASTContext &Context, const FieldDecl *Field);
 
 static bool
-HasTrivialDestructorBody(ASTContext &Context, 
+HasTrivialDestructorBody(ASTContext &Context,
                          const CXXRecordDecl *BaseClassDecl,
                          const CXXRecordDecl *MostDerivedClassDecl)
 {
@@ -1309,7 +1324,7 @@ HasTrivialDestructorBody(ASTContext &Context,
         cast<CXXRecordDecl>(I.getType()->castAs<RecordType>()->getDecl());
       if (!HasTrivialDestructorBody(Context, VirtualBase,
                                     MostDerivedClassDecl))
-        return false;      
+        return false;
     }
   }
 
@@ -1325,7 +1340,7 @@ FieldHasTrivialDestructorBody(ASTContext &Context,
   const RecordType *RT = FieldBaseElementType->getAs<RecordType>();
   if (!RT)
     return true;
-  
+
   CXXRecordDecl *FieldClassDecl = cast<CXXRecordDecl>(RT->getDecl());
   return HasTrivialDestructorBody(Context, FieldClassDecl, FieldClassDecl);
 }
@@ -1351,6 +1366,10 @@ void CodeGenFunction::EmitDestructorBody(FunctionArgList &Args) {
   const CXXDestructorDecl *Dtor = cast<CXXDestructorDecl>(CurGD.getDecl());
   CXXDtorType DtorType = CurGD.getDtorType();
 
+  Stmt *Body = Dtor->getBody();
+  if (Body)
+    incrementProfileCounter(Body);
+
   // The call to operator delete in a deleting destructor happens
   // outside of the function-try-block, which means it's always
   // possible to delegate the destructor body to the complete
@@ -1363,8 +1382,6 @@ void CodeGenFunction::EmitDestructorBody(FunctionArgList &Args) {
     return;
   }
 
-  Stmt *Body = Dtor->getBody();
-
   // If the body is a function-try-block, enter the try before
   // anything else.
   bool isTryBody = (Body && isa<CXXTryStmt>(Body));
@@ -1374,11 +1391,11 @@ void CodeGenFunction::EmitDestructorBody(FunctionArgList &Args) {
 
   // Enter the epilogue cleanups.
   RunCleanupsScope DtorEpilogue(*this);
-  
+
   // If this is the complete variant, just invoke the base variant;
   // the epilogue will destruct the virtual bases.  But we can't do
   // this optimization if the body is a function-try-block, because
-  // we'd introduce *two* handler blocks.  In the Microsoft ABI, we 
+  // we'd introduce *two* handler blocks.  In the Microsoft ABI, we
   // always delegate because we might not have a definition in this TU.
   switch (DtorType) {
   case Dtor_Comdat:
@@ -1399,13 +1416,10 @@ void CodeGenFunction::EmitDestructorBody(FunctionArgList &Args) {
       break;
     }
     // Fallthrough: act like we're in the base variant.
-      
+
   case Dtor_Base:
     assert(Body);
 
-    RegionCounter Cnt = getPGORegionCounter(Body);
-    Cnt.beginRegion(Builder);
-
     // Enter the cleanup scopes for fields and non-virtual bases.
     EnterDtorCleanups(Dtor, Dtor_Base);
 
@@ -1447,7 +1461,7 @@ void CodeGenFunction::emitImplicitAssignmentOperatorBody(FunctionArgList &Args)
 
   AssignmentMemcpyizer AM(*this, AssignOp, Args);
   for (auto *I : RootCS->body())
-    AM.emitAssignment(I);  
+    AM.emitAssignment(I);
   AM.finish();
 }
 
@@ -1508,7 +1522,7 @@ namespace {
       LValue ThisLV = CGF.MakeAddrLValue(thisValue, RecordTy);
       LValue LV = CGF.EmitLValueForField(ThisLV, field);
       assert(LV.isSimple());
-      
+
       CGF.emitDestroy(LV.getAddress(), field->getType(), destroyer,
                       flags.isForNormalCleanup() && useEHCleanupForArray);
     }
@@ -1526,7 +1540,7 @@ void CodeGenFunction::EnterDtorCleanups(const CXXDestructorDecl *DD,
   // The deleting-destructor phase just needs to call the appropriate
   // operator delete that Sema picked up.
   if (DtorType == Dtor_Deleting) {
-    assert(DD->getOperatorDelete() && 
+    assert(DD->getOperatorDelete() &&
            "operator delete missing - EnterDtorCleanups");
     if (CXXStructorImplicitParamValue) {
       // If there is an implicit param to the deleting dtor, it's a boolean
@@ -1553,7 +1567,7 @@ void CodeGenFunction::EnterDtorCleanups(const CXXDestructorDecl *DD,
     for (const auto &Base : ClassDecl->vbases()) {
       CXXRecordDecl *BaseClassDecl
         = cast<CXXRecordDecl>(Base.getType()->getAs<RecordType>()->getDecl());
-    
+
       // Ignore trivial destructors.
       if (BaseClassDecl->hasTrivialDestructor())
         continue;
@@ -1567,15 +1581,15 @@ void CodeGenFunction::EnterDtorCleanups(const CXXDestructorDecl *DD,
   }
 
   assert(DtorType == Dtor_Base);
-  
+
   // Destroy non-virtual bases.
   for (const auto &Base : ClassDecl->bases()) {
     // Ignore virtual bases.
     if (Base.isVirtual())
       continue;
-    
+
     CXXRecordDecl *BaseClassDecl = Base.getType()->getAsCXXRecordDecl();
-    
+
     // Ignore trivial destructors.
     if (BaseClassDecl->hasTrivialDestructor())
       continue;
@@ -1656,7 +1670,7 @@ void CodeGenFunction::EmitCXXAggrConstructorCall(const CXXConstructorDecl *ctor,
     zeroCheckBranch = Builder.CreateCondBr(iszero, loopBB, loopBB);
     EmitBlock(loopBB);
   }
-      
+
   // Find the end of the array.
   llvm::Value *arrayEnd = Builder.CreateInBoundsGEP(arrayBegin, numElements,
                                                     "arrayctor.end");
@@ -1676,15 +1690,15 @@ void CodeGenFunction::EmitCXXAggrConstructorCall(const CXXConstructorDecl *ctor,
   // Zero initialize the storage, if requested.
   if (zeroInitialize)
     EmitNullInitialization(cur, type);
-  
-  // C++ [class.temporary]p4: 
+
+  // C++ [class.temporary]p4:
   // There are two contexts in which temporaries are destroyed at a different
   // point than the end of the full-expression. The first context is when a
-  // default constructor is called to initialize an element of an array. 
-  // If the constructor has one or more default arguments, the destruction of 
-  // every temporary created in a default argument expression is sequenced 
+  // default constructor is called to initialize an element of an array.
+  // If the constructor has one or more default arguments, the destruction of
+  // every temporary created in a default argument expression is sequenced
   // before the construction of the next array element, if any.
-  
+
   {
     RunCleanupsScope Scope(*this);
 
@@ -1733,33 +1747,32 @@ void CodeGenFunction::EmitCXXConstructorCall(const CXXConstructorDecl *D,
                                              bool ForVirtualBase,
                                              bool Delegating, llvm::Value *This,
                                              const CXXConstructExpr *E) {
-  // If this is a trivial constructor, just emit what's needed.
-  if (D->isTrivial() && !D->getParent()->mayInsertExtraPadding()) {
-    if (E->getNumArgs() == 0) {
-      // Trivial default constructor, no codegen required.
-      assert(D->isDefaultConstructor() &&
-             "trivial 0-arg ctor not a default ctor");
-      return;
-    }
+  // C++11 [class.mfct.non-static]p2:
+  //   If a non-static member function of a class X is called for an object that
+  //   is not of type X, or of a type derived from X, the behavior is undefined.
+  // FIXME: Provide a source location here.
+  EmitTypeCheck(CodeGenFunction::TCK_ConstructorCall, SourceLocation(), This,
+                getContext().getRecordType(D->getParent()));
+
+  if (D->isTrivial() && D->isDefaultConstructor()) {
+    assert(E->getNumArgs() == 0 && "trivial default ctor with args");
+    return;
+  }
 
+  // If this is a trivial constructor, just emit what's needed. If this is a
+  // union copy constructor, we must emit a memcpy, because the AST does not
+  // model that copy.
+  if (isMemcpyEquivalentSpecialMember(D)) {
     assert(E->getNumArgs() == 1 && "unexpected argcount for trivial ctor");
-    assert(D->isCopyOrMoveConstructor() &&
-           "trivial 1-arg ctor not a copy/move ctor");
 
     const Expr *Arg = E->getArg(0);
-    QualType Ty = Arg->getType();
+    QualType SrcTy = Arg->getType();
     llvm::Value *Src = EmitLValue(Arg).getAddress();
-    EmitAggregateCopy(This, Src, Ty);
+    QualType DestTy = getContext().getTypeDeclType(D->getParent());
+    EmitAggregateCopyCtor(This, Src, DestTy, SrcTy);
     return;
   }
 
-  // C++11 [class.mfct.non-static]p2:
-  //   If a non-static member function of a class X is called for an object that
-  //   is not of type X, or of a type derived from X, the behavior is undefined.
-  // FIXME: Provide a source location here.
-  EmitTypeCheck(CodeGenFunction::TCK_ConstructorCall, SourceLocation(), This,
-                getContext().getRecordType(D->getParent()));
-
   CallArgList Args;
 
   // Push the this ptr.
@@ -1784,25 +1797,26 @@ void
 CodeGenFunction::EmitSynthesizedCXXCopyCtorCall(const CXXConstructorDecl *D,
                                         llvm::Value *This, llvm::Value *Src,
                                         const CXXConstructExpr *E) {
-  if (D->isTrivial() &&
-      !D->getParent()->mayInsertExtraPadding()) {
+  if (isMemcpyEquivalentSpecialMember(D)) {
     assert(E->getNumArgs() == 1 && "unexpected argcount for trivial ctor");
     assert(D->isCopyOrMoveConstructor() &&
            "trivial 1-arg ctor not a copy/move ctor");
-    EmitAggregateCopy(This, Src, E->arg_begin()->getType());
+    EmitAggregateCopyCtor(This, Src,
+                          getContext().getTypeDeclType(D->getParent()),
+                          E->arg_begin()->getType());
     return;
   }
   llvm::Value *Callee = CGM.getAddrOfCXXStructor(D, StructorType::Complete);
   assert(D->isInstance() &&
          "Trying to emit a member call expr on a static method!");
-  
+
   const FunctionProtoType *FPT = D->getType()->castAs<FunctionProtoType>();
-  
+
   CallArgList Args;
-  
+
   // Push the this ptr.
   Args.add(RValue::get(This), D->getThisType(getContext()));
-  
+
   // Push the src ptr.
   QualType QT = *(FPT->param_type_begin());
   llvm::Type *t = CGM.getTypes().ConvertType(QT);
@@ -1945,10 +1959,18 @@ void CodeGenFunction::PushDestructorCleanup(QualType T, llvm::Value *Addr) {
 }
 
 void
-CodeGenFunction::InitializeVTablePointer(BaseSubobject Base, 
+CodeGenFunction::InitializeVTablePointer(BaseSubobject Base,
                                          const CXXRecordDecl *NearestVBase,
                                          CharUnits OffsetFromNearestVBase,
                                          const CXXRecordDecl *VTableClass) {
+  const CXXRecordDecl *RD = Base.getBase();
+
+  // Don't initialize the vtable pointer if the class is marked with the
+  // 'novtable' attribute.
+  if ((RD == VTableClass || RD == NearestVBase) &&
+      VTableClass->hasAttr<MSNoVTableAttr>())
+    return;
+
   // Compute the address point.
   bool NeedsVirtualOffset;
   llvm::Value *VTableAddressPoint =
@@ -1960,7 +1982,7 @@ CodeGenFunction::InitializeVTablePointer(BaseSubobject Base,
   // Compute where to store the address point.
   llvm::Value *VirtualOffset = nullptr;
   CharUnits NonVirtualOffset = CharUnits::Zero();
-  
+
   if (NeedsVirtualOffset) {
     // We need to use the virtual base offset offset because the virtual base
     // might have a different offset in the most derived class.
@@ -1973,12 +1995,12 @@ CodeGenFunction::InitializeVTablePointer(BaseSubobject Base,
     // We can just use the base offset in the complete class.
     NonVirtualOffset = Base.getBaseOffset();
   }
-  
+
   // Apply the offsets.
   llvm::Value *VTableField = LoadCXXThis();
-  
+
   if (!NonVirtualOffset.isZero() || VirtualOffset)
-    VTableField = ApplyNonVirtualAndVirtualOffset(*this, VTableField, 
+    VTableField = ApplyNonVirtualAndVirtualOffset(*this, VTableField,
                                                   NonVirtualOffset,
                                                   VirtualOffset);
 
@@ -1995,7 +2017,7 @@ CodeGenFunction::InitializeVTablePointer(BaseSubobject Base,
 }
 
 void
-CodeGenFunction::InitializeVTablePointers(BaseSubobject Base, 
+CodeGenFunction::InitializeVTablePointers(BaseSubobject Base,
                                           const CXXRecordDecl *NearestVBase,
                                           CharUnits OffsetFromNearestVBase,
                                           bool BaseIsNonVirtualPrimaryBase,
@@ -2008,7 +2030,7 @@ CodeGenFunction::InitializeVTablePointers(BaseSubobject Base,
     InitializeVTablePointer(Base, NearestVBase, OffsetFromNearestVBase,
                             VTableClass);
   }
-  
+
   const CXXRecordDecl *RD = Base.getBase();
 
   // Traverse bases.
@@ -2029,7 +2051,7 @@ CodeGenFunction::InitializeVTablePointers(BaseSubobject Base,
       if (!VBases.insert(BaseDecl).second)
         continue;
 
-      const ASTRecordLayout &Layout = 
+      const ASTRecordLayout &Layout =
         getContext().getASTRecordLayout(VTableClass);
 
       BaseOffset = Layout.getVBaseClassOffset(BaseDecl);
@@ -2039,15 +2061,15 @@ CodeGenFunction::InitializeVTablePointers(BaseSubobject Base,
       const ASTRecordLayout &Layout = getContext().getASTRecordLayout(RD);
 
       BaseOffset = Base.getBaseOffset() + Layout.getBaseClassOffset(BaseDecl);
-      BaseOffsetFromNearestVBase = 
+      BaseOffsetFromNearestVBase =
         OffsetFromNearestVBase + Layout.getBaseClassOffset(BaseDecl);
       BaseDeclIsNonVirtualPrimaryBase = Layout.getPrimaryBase() == BaseDecl;
     }
-    
-    InitializeVTablePointers(BaseSubobject(BaseDecl, BaseOffset), 
+
+    InitializeVTablePointers(BaseSubobject(BaseDecl, BaseOffset),
                              I.isVirtual() ? BaseDecl : NearestVBase,
                              BaseOffsetFromNearestVBase,
-                             BaseDeclIsNonVirtualPrimaryBase, 
+                             BaseDeclIsNonVirtualPrimaryBase,
                              VTableClass, VBases);
   }
 }
@@ -2059,7 +2081,7 @@ void CodeGenFunction::InitializeVTablePointers(const CXXRecordDecl *RD) {
 
   // Initialize the vtable pointers for this class and all of its bases.
   VisitedVirtualBasesSetTy VBases;
-  InitializeVTablePointers(BaseSubobject(RD, CharUnits::Zero()), 
+  InitializeVTablePointers(BaseSubobject(RD, CharUnits::Zero()),
                            /*NearestVBase=*/nullptr,
                            /*OffsetFromNearestVBase=*/CharUnits::Zero(),
                            /*BaseIsNonVirtualPrimaryBase=*/false, RD, VBases);
@@ -2076,6 +2098,128 @@ llvm::Value *CodeGenFunction::GetVTablePtr(llvm::Value *This,
   return VTable;
 }
 
+// If a class has a single non-virtual base and does not introduce or override
+// virtual member functions or fields, it will have the same layout as its base.
+// This function returns the least derived such class.
+//
+// Casting an instance of a base class to such a derived class is technically
+// undefined behavior, but it is a relatively common hack for introducing member
+// functions on class instances with specific properties (e.g. llvm::Operator)
+// that works under most compilers and should not have security implications, so
+// we allow it by default. It can be disabled with -fsanitize=cfi-cast-strict.
+static const CXXRecordDecl *
+LeastDerivedClassWithSameLayout(const CXXRecordDecl *RD) {
+  if (!RD->field_empty())
+    return RD;
+
+  if (RD->getNumVBases() != 0)
+    return RD;
+
+  if (RD->getNumBases() != 1)
+    return RD;
+
+  for (const CXXMethodDecl *MD : RD->methods()) {
+    if (MD->isVirtual()) {
+      // Virtual member functions are only ok if they are implicit destructors
+      // because the implicit destructor will have the same semantics as the
+      // base class's destructor if no fields are added.
+      if (isa<CXXDestructorDecl>(MD) && MD->isImplicit())
+        continue;
+      return RD;
+    }
+  }
+
+  return LeastDerivedClassWithSameLayout(
+      RD->bases_begin()->getType()->getAsCXXRecordDecl());
+}
+
+void CodeGenFunction::EmitVTablePtrCheckForCall(const CXXMethodDecl *MD,
+                                                llvm::Value *VTable) {
+  const CXXRecordDecl *ClassDecl = MD->getParent();
+  if (!SanOpts.has(SanitizerKind::CFICastStrict))
+    ClassDecl = LeastDerivedClassWithSameLayout(ClassDecl);
+
+  EmitVTablePtrCheck(ClassDecl, VTable);
+}
+
+void CodeGenFunction::EmitVTablePtrCheckForCast(QualType T,
+                                                llvm::Value *Derived,
+                                                bool MayBeNull) {
+  if (!getLangOpts().CPlusPlus)
+    return;
+
+  auto *ClassTy = T->getAs<RecordType>();
+  if (!ClassTy)
+    return;
+
+  const CXXRecordDecl *ClassDecl = cast<CXXRecordDecl>(ClassTy->getDecl());
+
+  if (!ClassDecl->isCompleteDefinition() || !ClassDecl->isDynamicClass())
+    return;
+
+  SmallString<64> MangledName;
+  llvm::raw_svector_ostream Out(MangledName);
+  CGM.getCXXABI().getMangleContext().mangleCXXRTTI(T.getUnqualifiedType(),
+                                                   Out);
+
+  // Blacklist based on the mangled type.
+  if (CGM.getContext().getSanitizerBlacklist().isBlacklistedType(Out.str()))
+    return;
+
+  if (!SanOpts.has(SanitizerKind::CFICastStrict))
+    ClassDecl = LeastDerivedClassWithSameLayout(ClassDecl);
+
+  llvm::BasicBlock *ContBlock = 0;
+
+  if (MayBeNull) {
+    llvm::Value *DerivedNotNull =
+        Builder.CreateIsNotNull(Derived, "cast.nonnull");
+
+    llvm::BasicBlock *CheckBlock = createBasicBlock("cast.check");
+    ContBlock = createBasicBlock("cast.cont");
+
+    Builder.CreateCondBr(DerivedNotNull, CheckBlock, ContBlock);
+
+    EmitBlock(CheckBlock);
+  }
+
+  llvm::Value *VTable = GetVTablePtr(Derived, Int8PtrTy);
+  EmitVTablePtrCheck(ClassDecl, VTable);
+
+  if (MayBeNull) {
+    Builder.CreateBr(ContBlock);
+    EmitBlock(ContBlock);
+  }
+}
+
+void CodeGenFunction::EmitVTablePtrCheck(const CXXRecordDecl *RD,
+                                         llvm::Value *VTable) {
+  // FIXME: Add blacklisting scheme.
+  if (RD->isInStdNamespace())
+    return;
+
+  std::string OutName;
+  llvm::raw_string_ostream Out(OutName);
+  CGM.getCXXABI().getMangleContext().mangleCXXVTableBitSet(RD, Out);
+
+  llvm::Value *BitSetName = llvm::MetadataAsValue::get(
+      getLLVMContext(), llvm::MDString::get(getLLVMContext(), Out.str()));
+
+  llvm::Value *BitSetTest = Builder.CreateCall(
+      CGM.getIntrinsic(llvm::Intrinsic::bitset_test),
+      {Builder.CreateBitCast(VTable, CGM.Int8PtrTy), BitSetName});
+
+  llvm::BasicBlock *ContBlock = createBasicBlock("vtable.check.cont");
+  llvm::BasicBlock *TrapBlock = createBasicBlock("vtable.check.trap");
+
+  Builder.CreateCondBr(BitSetTest, ContBlock, TrapBlock);
+
+  EmitBlock(TrapBlock);
+  Builder.CreateCall(CGM.getIntrinsic(llvm::Intrinsic::trap), {});
+  Builder.CreateUnreachable();
+
+  EmitBlock(ContBlock);
+}
 
 // FIXME: Ideally Expr::IgnoreParenNoopCasts should do this, but it doesn't do
 // quite what we want.
@@ -2140,7 +2284,7 @@ CodeGenFunction::CanDevirtualizeMemberFunctionCall(const Expr *Base,
       // This is a record decl. We know the type and can devirtualize it.
       return VD->getType()->isRecordType();
     }
-    
+
     return false;
   }
 
@@ -2154,14 +2298,14 @@ CodeGenFunction::CanDevirtualizeMemberFunctionCall(const Expr *Base,
   // We can always devirtualize calls on temporary object expressions.
   if (isa<CXXConstructExpr>(Base))
     return true;
-  
+
   // And calls on bound temporaries.
   if (isa<CXXBindTemporaryExpr>(Base))
     return true;
-  
+
   // Check if this is a call expr that returns a record type.
   if (const CallExpr *CE = dyn_cast<CallExpr>(Base))
-    return CE->getCallReturnType()->isRecordType();
+    return CE->getCallReturnType(getContext())->isRecordType();
 
   // We can't devirtualize the call.
   return false;
@@ -2190,7 +2334,7 @@ void CodeGenFunction::EmitForwardingCallToLambda(
   // We don't need to separately arrange the call arguments because
   // the call can't be variadic anyway --- it's impossible to forward
   // variadic arguments.
-  
+
   // Now emit our call.
   RValue RV = EmitCall(calleeFnInfo, callee, returnSlot,
                        callArgs, callOperator);
@@ -2218,7 +2362,7 @@ void CodeGenFunction::EmitLambdaBlockInvokeBody() {
   for (auto param : BD->params())
     EmitDelegateCallArg(CallArgs, param, param->getLocStart());
 
-  assert(!Lambda->isGenericLambda() && 
+  assert(!Lambda->isGenericLambda() &&
             "generic lambda interconversion to block not implemented");
   EmitForwardingCallToLambda(Lambda->getLambdaCallOperator(), CallArgs);
 }
@@ -2256,7 +2400,7 @@ void CodeGenFunction::EmitLambdaDelegatingInvokeBody(const CXXMethodDecl *MD) {
     const TemplateArgumentList *TAL = MD->getTemplateSpecializationArgs();
     FunctionTemplateDecl *CallOpTemplate = CallOp->getDescribedFunctionTemplate();
     void *InsertPos = nullptr;
-    FunctionDecl *CorrespondingCallOpSpecialization = 
+    FunctionDecl *CorrespondingCallOpSpecialization =
         CallOpTemplate->findSpecialization(TAL->asArray(), InsertPos);
     assert(CorrespondingCallOpSpecialization);
     CallOp = cast<CXXMethodDecl>(CorrespondingCallOpSpecialization);
diff --git a/lib/CodeGen/CGCleanup.cpp b/lib/CodeGen/CGCleanup.cpp
index 18ed3e543d20..d97e40554ef2 100644
--- a/lib/CodeGen/CGCleanup.cpp
+++ b/lib/CodeGen/CGCleanup.cpp
@@ -52,8 +52,10 @@ DominatingValue<RValue>::saved_type::save(CodeGenFunction &CGF, RValue rv) {
       llvm::StructType::get(V.first->getType(), V.second->getType(),
                             (void*) nullptr);
     llvm::Value *addr = CGF.CreateTempAlloca(ComplexTy, "saved-complex");
-    CGF.Builder.CreateStore(V.first, CGF.Builder.CreateStructGEP(addr, 0));
-    CGF.Builder.CreateStore(V.second, CGF.Builder.CreateStructGEP(addr, 1));
+    CGF.Builder.CreateStore(V.first,
+                            CGF.Builder.CreateStructGEP(ComplexTy, addr, 0));
+    CGF.Builder.CreateStore(V.second,
+                            CGF.Builder.CreateStructGEP(ComplexTy, addr, 1));
     return saved_type(addr, ComplexAddress);
   }
 
@@ -82,9 +84,9 @@ RValue DominatingValue<RValue>::saved_type::restore(CodeGenFunction &CGF) {
     return RValue::getAggregate(CGF.Builder.CreateLoad(Value));
   case ComplexAddress: {
     llvm::Value *real =
-      CGF.Builder.CreateLoad(CGF.Builder.CreateStructGEP(Value, 0));
+        CGF.Builder.CreateLoad(CGF.Builder.CreateStructGEP(nullptr, Value, 0));
     llvm::Value *imag =
-      CGF.Builder.CreateLoad(CGF.Builder.CreateStructGEP(Value, 1));
+        CGF.Builder.CreateLoad(CGF.Builder.CreateStructGEP(nullptr, Value, 1));
     return RValue::getComplex(real, imag);
   }
   }
@@ -123,6 +125,17 @@ char *EHScopeStack::allocate(size_t Size) {
   return StartOfData;
 }
 
+bool EHScopeStack::containsOnlyLifetimeMarkers(
+    EHScopeStack::stable_iterator Old) const {
+  for (EHScopeStack::iterator it = begin(); stabilize(it) != Old; it++) {
+    EHCleanupScope *cleanup = dyn_cast<EHCleanupScope>(&*it);
+    if (!cleanup || !cleanup->isLifetimeMarker())
+      return false;
+  }
+
+  return true;
+}
+
 EHScopeStack::stable_iterator
 EHScopeStack::getInnermostActiveNormalCleanup() const {
   for (stable_iterator si = getInnermostNormalCleanup(), se = stable_end();
@@ -469,8 +482,14 @@ static void EmitCleanup(CodeGenFunction &CGF,
                         EHScopeStack::Cleanup *Fn,
                         EHScopeStack::Cleanup::Flags flags,
                         llvm::Value *ActiveFlag) {
-  // EH cleanups always occur within a terminate scope.
-  if (flags.isForEHCleanup()) CGF.EHStack.pushTerminate();
+  // Itanium EH cleanups occur within a terminate scope. Microsoft SEH doesn't
+  // have this behavior, and the Microsoft C++ runtime will call terminate for
+  // us if the cleanup throws.
+  bool PushedTerminate = false;
+  if (flags.isForEHCleanup() && !CGF.getTarget().getCXXABI().isMicrosoft()) {
+    CGF.EHStack.pushTerminate();
+    PushedTerminate = true;
+  }
 
   // If there's an active flag, load it and skip the cleanup if it's
   // false.
@@ -493,7 +512,8 @@ static void EmitCleanup(CodeGenFunction &CGF,
     CGF.EmitBlock(ContBB);
 
   // Leave the terminate scope.
-  if (flags.isForEHCleanup()) CGF.EHStack.popTerminate();
+  if (PushedTerminate)
+    CGF.EHStack.popTerminate();
 }
 
 static void ForwardPrebranchedFallthrough(llvm::BasicBlock *Exit,
@@ -739,7 +759,15 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
           Scope.getNumBranchAfters() == 1) {
         assert(!BranchThroughDest || !IsActive);
 
-        // TODO: clean up the possibly dead stores to the cleanup dest slot.
+        // Clean up the possibly dead store to the cleanup dest slot.
+        llvm::Instruction *NormalCleanupDestSlot =
+            cast<llvm::Instruction>(getNormalCleanupDestSlot());
+        if (NormalCleanupDestSlot->hasOneUse()) {
+          NormalCleanupDestSlot->user_back()->eraseFromParent();
+          NormalCleanupDestSlot->eraseFromParent();
+          NormalCleanupDest = nullptr;
+        }
+
         llvm::BasicBlock *BranchAfter = Scope.getBranchAfterBlock(0);
         InstsToAppend.push_back(llvm::BranchInst::Create(BranchAfter));
 
@@ -861,8 +889,6 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
 
   // Emit the EH cleanup if required.
   if (RequiresEHCleanup) {
-    ApplyDebugLocation AutoRestoreLocation(*this, CurEHLocation);
-
     CGBuilderTy::InsertPoint SavedIP = Builder.saveAndClearIP();
 
     EmitBlock(EHEntry);
diff --git a/lib/CodeGen/CGCleanup.h b/lib/CodeGen/CGCleanup.h
index dd156c696ad3..81c64123dfdb 100644
--- a/lib/CodeGen/CGCleanup.h
+++ b/lib/CodeGen/CGCleanup.h
@@ -62,6 +62,9 @@ protected:
     /// Whether this cleanup is currently active.
     unsigned IsActive : 1;
 
+    /// Whether this cleanup is a lifetime marker
+    unsigned IsLifetimeMarker : 1;
+
     /// Whether the normal cleanup should test the activation flag.
     unsigned TestFlagInNormalCleanup : 1;
 
@@ -75,7 +78,7 @@ protected:
     /// The number of fixups required by enclosing scopes (not including
     /// this one).  If this is the top cleanup scope, all the fixups
     /// from this index onwards belong to this scope.
-    unsigned FixupDepth : 32 - 17 - NumCommonBits; // currently 13    
+    unsigned FixupDepth : 32 - 18 - NumCommonBits; // currently 13
   };
 
   class FilterBitFields {
@@ -272,6 +275,7 @@ public:
     CleanupBits.IsNormalCleanup = isNormal;
     CleanupBits.IsEHCleanup = isEH;
     CleanupBits.IsActive = isActive;
+    CleanupBits.IsLifetimeMarker = false;
     CleanupBits.TestFlagInNormalCleanup = false;
     CleanupBits.TestFlagInEHCleanup = false;
     CleanupBits.CleanupSize = cleanupSize;
@@ -284,19 +288,20 @@ public:
     delete ExtInfo;
   }
   // Objects of EHCleanupScope are not destructed. Use Destroy().
-  ~EHCleanupScope() LLVM_DELETED_FUNCTION;
+  ~EHCleanupScope() = delete;
 
   bool isNormalCleanup() const { return CleanupBits.IsNormalCleanup; }
   llvm::BasicBlock *getNormalBlock() const { return NormalBlock; }
   void setNormalBlock(llvm::BasicBlock *BB) { NormalBlock = BB; }
 
   bool isEHCleanup() const { return CleanupBits.IsEHCleanup; }
-  llvm::BasicBlock *getEHBlock() const { return getCachedEHDispatchBlock(); }
-  void setEHBlock(llvm::BasicBlock *BB) { setCachedEHDispatchBlock(BB); }
 
   bool isActive() const { return CleanupBits.IsActive; }
   void setActive(bool A) { CleanupBits.IsActive = A; }
 
+  bool isLifetimeMarker() const { return CleanupBits.IsLifetimeMarker; }
+  void setLifetimeMarker() { CleanupBits.IsLifetimeMarker = true; }
+
   llvm::AllocaInst *getActiveFlag() const { return ActiveFlag; }
   void setActiveFlag(llvm::AllocaInst *Var) { ActiveFlag = Var; }
 
diff --git a/lib/CodeGen/CGDebugInfo.cpp b/lib/CodeGen/CGDebugInfo.cpp
index 978e1bb5b81f..48458dbd6003 100644
--- a/lib/CodeGen/CGDebugInfo.cpp
+++ b/lib/CodeGen/CGDebugInfo.cpp
@@ -52,35 +52,48 @@ CGDebugInfo::~CGDebugInfo() {
          "Region stack mismatch, stack not empty!");
 }
 
-ArtificialLocation::ArtificialLocation(CodeGenFunction &CGF)
-    : ApplyDebugLocation(CGF) {
-  if (auto *DI = CGF.getDebugInfo()) {
-    // Construct a location that has a valid scope, but no line info.
-    assert(!DI->LexicalBlockStack.empty());
-    llvm::DIDescriptor Scope(DI->LexicalBlockStack.back());
-    CGF.Builder.SetCurrentDebugLocation(llvm::DebugLoc::get(0, 0, Scope));
-  }
+ApplyDebugLocation::ApplyDebugLocation(CodeGenFunction &CGF,
+                                       SourceLocation TemporaryLocation)
+    : CGF(CGF) {
+  init(TemporaryLocation);
 }
 
 ApplyDebugLocation::ApplyDebugLocation(CodeGenFunction &CGF,
-                                       SourceLocation TemporaryLocation,
-                                       bool ForceColumnInfo)
+                                       bool DefaultToEmpty,
+                                       SourceLocation TemporaryLocation)
     : CGF(CGF) {
+  init(TemporaryLocation, DefaultToEmpty);
+}
+
+void ApplyDebugLocation::init(SourceLocation TemporaryLocation,
+                              bool DefaultToEmpty) {
   if (auto *DI = CGF.getDebugInfo()) {
     OriginalLocation = CGF.Builder.getCurrentDebugLocation();
-    if (TemporaryLocation.isInvalid())
-      CGF.Builder.SetCurrentDebugLocation(llvm::DebugLoc());
-    else
-      DI->EmitLocation(CGF.Builder, TemporaryLocation, ForceColumnInfo);
+    if (TemporaryLocation.isInvalid()) {
+      if (DefaultToEmpty)
+        CGF.Builder.SetCurrentDebugLocation(llvm::DebugLoc());
+      else {
+        // Construct a location that has a valid scope, but no line info.
+        assert(!DI->LexicalBlockStack.empty());
+        CGF.Builder.SetCurrentDebugLocation(
+            llvm::DebugLoc::get(0, 0, DI->LexicalBlockStack.back()));
+      }
+    } else
+      DI->EmitLocation(CGF.Builder, TemporaryLocation);
   }
 }
 
+ApplyDebugLocation::ApplyDebugLocation(CodeGenFunction &CGF, const Expr *E)
+    : CGF(CGF) {
+  init(E->getExprLoc());
+}
+
 ApplyDebugLocation::ApplyDebugLocation(CodeGenFunction &CGF, llvm::DebugLoc Loc)
     : CGF(CGF) {
   if (CGF.getDebugInfo()) {
     OriginalLocation = CGF.Builder.getCurrentDebugLocation();
-    if (!Loc.isUnknown())
-      CGF.Builder.SetCurrentDebugLocation(Loc);
+    if (Loc)
+      CGF.Builder.SetCurrentDebugLocation(std::move(Loc));
   }
 }
 
@@ -88,7 +101,7 @@ ApplyDebugLocation::~ApplyDebugLocation() {
   // Query CGF so the location isn't overwritten when location updates are
   // temporarily disabled (for C++ default function arguments)
   if (CGF.getDebugInfo())
-    CGF.Builder.SetCurrentDebugLocation(OriginalLocation);
+    CGF.Builder.SetCurrentDebugLocation(std::move(OriginalLocation));
 }
 
 /// ArtificialLocation - An RAII object that temporarily switches to
@@ -107,37 +120,33 @@ void CGDebugInfo::setLocation(SourceLocation Loc) {
     return;
 
   SourceManager &SM = CGM.getContext().getSourceManager();
-  llvm::DIScope Scope(LexicalBlockStack.back());
+  auto *Scope = cast<llvm::DIScope>(LexicalBlockStack.back());
   PresumedLoc PCLoc = SM.getPresumedLoc(CurLoc);
 
-  if (PCLoc.isInvalid() || Scope.getFilename() == PCLoc.getFilename())
+  if (PCLoc.isInvalid() || Scope->getFilename() == PCLoc.getFilename())
     return;
 
-  if (Scope.isLexicalBlockFile()) {
-    llvm::DILexicalBlockFile LBF = llvm::DILexicalBlockFile(Scope);
-    llvm::DIDescriptor D = DBuilder.createLexicalBlockFile(
-        LBF.getScope(), getOrCreateFile(CurLoc));
-    llvm::MDNode *N = D;
+  if (auto *LBF = dyn_cast<llvm::DILexicalBlockFile>(Scope)) {
     LexicalBlockStack.pop_back();
-    LexicalBlockStack.emplace_back(N);
-  } else if (Scope.isLexicalBlock() || Scope.isSubprogram()) {
-    llvm::DIDescriptor D =
-        DBuilder.createLexicalBlockFile(Scope, getOrCreateFile(CurLoc));
-    llvm::MDNode *N = D;
+    LexicalBlockStack.emplace_back(DBuilder.createLexicalBlockFile(
+        LBF->getScope(), getOrCreateFile(CurLoc)));
+  } else if (isa<llvm::DILexicalBlock>(Scope) ||
+             isa<llvm::DISubprogram>(Scope)) {
     LexicalBlockStack.pop_back();
-    LexicalBlockStack.emplace_back(N);
+    LexicalBlockStack.emplace_back(
+        DBuilder.createLexicalBlockFile(Scope, getOrCreateFile(CurLoc)));
   }
 }
 
 /// getContextDescriptor - Get context info for the decl.
-llvm::DIScope CGDebugInfo::getContextDescriptor(const Decl *Context) {
+llvm::DIScope *CGDebugInfo::getContextDescriptor(const Decl *Context) {
   if (!Context)
     return TheCU;
 
   auto I = RegionMap.find(Context);
   if (I != RegionMap.end()) {
     llvm::Metadata *V = I->second;
-    return llvm::DIScope(dyn_cast_or_null<llvm::MDNode>(V));
+    return dyn_cast_or_null<llvm::DIScope>(V);
   }
 
   // Check namespace.
@@ -234,17 +243,17 @@ StringRef CGDebugInfo::getClassName(const RecordDecl *RD) {
 }
 
 /// getOrCreateFile - Get the file debug info descriptor for the input location.
-llvm::DIFile CGDebugInfo::getOrCreateFile(SourceLocation Loc) {
+llvm::DIFile *CGDebugInfo::getOrCreateFile(SourceLocation Loc) {
   if (!Loc.isValid())
     // If Location is not valid then use main input file.
-    return DBuilder.createFile(TheCU.getFilename(), TheCU.getDirectory());
+    return DBuilder.createFile(TheCU->getFilename(), TheCU->getDirectory());
 
   SourceManager &SM = CGM.getContext().getSourceManager();
   PresumedLoc PLoc = SM.getPresumedLoc(Loc);
 
   if (PLoc.isInvalid() || StringRef(PLoc.getFilename()).empty())
     // If the location is not valid then use main input file.
-    return DBuilder.createFile(TheCU.getFilename(), TheCU.getDirectory());
+    return DBuilder.createFile(TheCU->getFilename(), TheCU->getDirectory());
 
   // Cache the results.
   const char *fname = PLoc.getFilename();
@@ -253,18 +262,19 @@ llvm::DIFile CGDebugInfo::getOrCreateFile(SourceLocation Loc) {
   if (it != DIFileCache.end()) {
     // Verify that the information still exists.
     if (llvm::Metadata *V = it->second)
-      return llvm::DIFile(cast<llvm::MDNode>(V));
+      return cast<llvm::DIFile>(V);
   }
 
-  llvm::DIFile F = DBuilder.createFile(PLoc.getFilename(), getCurrentDirname());
+  llvm::DIFile *F =
+      DBuilder.createFile(PLoc.getFilename(), getCurrentDirname());
 
   DIFileCache[fname].reset(F);
   return F;
 }
 
 /// getOrCreateMainFile - Get the file info for main compile unit.
-llvm::DIFile CGDebugInfo::getOrCreateMainFile() {
-  return DBuilder.createFile(TheCU.getFilename(), TheCU.getDirectory());
+llvm::DIFile *CGDebugInfo::getOrCreateMainFile() {
+  return DBuilder.createFile(TheCU->getFilename(), TheCU->getDirectory());
 }
 
 /// getLineNumber - Get line number for the location. If location is invalid
@@ -369,12 +379,13 @@ void CGDebugInfo::CreateCompileUnit() {
       DebugKind <= CodeGenOptions::DebugLineTablesOnly
           ? llvm::DIBuilder::LineTablesOnly
           : llvm::DIBuilder::FullDebug,
+      0 /* DWOid */,
       DebugKind != CodeGenOptions::LocTrackingOnly);
 }
 
 /// CreateType - Get the Basic type from the cache or create a new
 /// one if necessary.
-llvm::DIType CGDebugInfo::CreateType(const BuiltinType *BT) {
+llvm::DIType *CGDebugInfo::CreateType(const BuiltinType *BT) {
   llvm::dwarf::TypeKind Encoding;
   StringRef BTName;
   switch (BT->getKind()) {
@@ -386,7 +397,7 @@ llvm::DIType CGDebugInfo::CreateType(const BuiltinType *BT) {
   case BuiltinType::NullPtr:
     return DBuilder.createNullPtrType();
   case BuiltinType::Void:
-    return llvm::DIType();
+    return nullptr;
   case BuiltinType::ObjCClass:
     if (!ClassTy)
       ClassTy = DBuilder.createForwardDecl(llvm::dwarf::DW_TAG_structure_type,
@@ -409,11 +420,11 @@ llvm::DIType CGDebugInfo::CreateType(const BuiltinType *BT) {
 
     unsigned Size = CGM.getContext().getTypeSize(CGM.getContext().VoidPtrTy);
 
-    llvm::DIType ISATy = DBuilder.createPointerType(ClassTy, Size);
+    auto *ISATy = DBuilder.createPointerType(ClassTy, Size);
 
     ObjTy =
         DBuilder.createStructType(TheCU, "objc_object", getOrCreateMainFile(),
-                                  0, 0, 0, 0, llvm::DIType(), llvm::DIArray());
+                                  0, 0, 0, 0, nullptr, llvm::DINodeArray());
 
     DBuilder.replaceArrays(
         ObjTy,
@@ -510,11 +521,10 @@ llvm::DIType CGDebugInfo::CreateType(const BuiltinType *BT) {
   // Bit size, align and offset of the type.
   uint64_t Size = CGM.getContext().getTypeSize(BT);
   uint64_t Align = CGM.getContext().getTypeAlign(BT);
-  llvm::DIType DbgTy = DBuilder.createBasicType(BTName, Size, Align, Encoding);
-  return DbgTy;
+  return DBuilder.createBasicType(BTName, Size, Align, Encoding);
 }
 
-llvm::DIType CGDebugInfo::CreateType(const ComplexType *Ty) {
+llvm::DIType *CGDebugInfo::CreateType(const ComplexType *Ty) {
   // Bit size, align and offset of the type.
   llvm::dwarf::TypeKind Encoding = llvm::dwarf::DW_ATE_complex_float;
   if (Ty->isComplexIntegerType())
@@ -522,15 +532,13 @@ llvm::DIType CGDebugInfo::CreateType(const ComplexType *Ty) {
 
   uint64_t Size = CGM.getContext().getTypeSize(Ty);
   uint64_t Align = CGM.getContext().getTypeAlign(Ty);
-  llvm::DIType DbgTy =
-      DBuilder.createBasicType("complex", Size, Align, Encoding);
-
-  return DbgTy;
+  return DBuilder.createBasicType("complex", Size, Align, Encoding);
 }
 
 /// CreateCVRType - Get the qualified type from the cache or create
 /// a new one if necessary.
-llvm::DIType CGDebugInfo::CreateQualifiedType(QualType Ty, llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateQualifiedType(QualType Ty,
+                                               llvm::DIFile *Unit) {
   QualifierCollector Qc;
   const Type *T = Qc.strip(Ty);
 
@@ -556,17 +564,15 @@ llvm::DIType CGDebugInfo::CreateQualifiedType(QualType Ty, llvm::DIFile Unit) {
     return getOrCreateType(QualType(T, 0), Unit);
   }
 
-  llvm::DIType FromTy = getOrCreateType(Qc.apply(CGM.getContext(), T), Unit);
+  auto *FromTy = getOrCreateType(Qc.apply(CGM.getContext(), T), Unit);
 
   // No need to fill in the Name, Line, Size, Alignment, Offset in case of
   // CVR derived types.
-  llvm::DIType DbgTy = DBuilder.createQualifiedType(Tag, FromTy);
-
-  return DbgTy;
+  return DBuilder.createQualifiedType(Tag, FromTy);
 }
 
-llvm::DIType CGDebugInfo::CreateType(const ObjCObjectPointerType *Ty,
-                                     llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateType(const ObjCObjectPointerType *Ty,
+                                      llvm::DIFile *Unit) {
 
   // The frontend treats 'id' as a typedef to an ObjCObjectType,
   // whereas 'id<protocol>' is treated as an ObjCPointerType. For the
@@ -574,12 +580,12 @@ llvm::DIType CGDebugInfo::CreateType(const ObjCObjectPointerType *Ty,
   if (Ty->isObjCQualifiedIdType())
     return getOrCreateType(CGM.getContext().getObjCIdType(), Unit);
 
-  llvm::DIType DbgTy = CreatePointerLikeType(llvm::dwarf::DW_TAG_pointer_type,
-                                             Ty, Ty->getPointeeType(), Unit);
-  return DbgTy;
+  return CreatePointerLikeType(llvm::dwarf::DW_TAG_pointer_type, Ty,
+                               Ty->getPointeeType(), Unit);
 }
 
-llvm::DIType CGDebugInfo::CreateType(const PointerType *Ty, llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateType(const PointerType *Ty,
+                                      llvm::DIFile *Unit) {
   return CreatePointerLikeType(llvm::dwarf::DW_TAG_pointer_type, Ty,
                                Ty->getPointeeType(), Unit);
 }
@@ -588,12 +594,12 @@ llvm::DIType CGDebugInfo::CreateType(const PointerType *Ty, llvm::DIFile Unit) {
 /// on their mangled names, if they're external.
 static SmallString<256> getUniqueTagTypeName(const TagType *Ty,
                                              CodeGenModule &CGM,
-                                             llvm::DICompileUnit TheCU) {
+                                             llvm::DICompileUnit *TheCU) {
   SmallString<256> FullName;
   // FIXME: ODR should apply to ObjC++ exactly the same wasy it does to C++.
   // For now, only apply ODR with C++.
   const TagDecl *TD = Ty->getDecl();
-  if (TheCU.getLanguage() != llvm::dwarf::DW_LANG_C_plus_plus ||
+  if (TheCU->getSourceLanguage() != llvm::dwarf::DW_LANG_C_plus_plus ||
       !TD->isExternallyVisible())
     return FullName;
   // Microsoft Mangler does not have support for mangleCXXRTTIName yet.
@@ -608,41 +614,56 @@ static SmallString<256> getUniqueTagTypeName(const TagType *Ty,
   return FullName;
 }
 
-// Creates a forward declaration for a RecordDecl in the given context.
-llvm::DICompositeType
-CGDebugInfo::getOrCreateRecordFwdDecl(const RecordType *Ty,
-                                      llvm::DIDescriptor Ctx) {
-  const RecordDecl *RD = Ty->getDecl();
-  if (llvm::DIType T = getTypeOrNull(CGM.getContext().getRecordType(RD)))
-    return llvm::DICompositeType(T);
-  llvm::DIFile DefUnit = getOrCreateFile(RD->getLocation());
-  unsigned Line = getLineNumber(RD->getLocation());
-  StringRef RDName = getClassName(RD);
-
-  llvm::dwarf::Tag Tag;
+static llvm::dwarf::Tag getTagForRecord(const RecordDecl *RD) {
+   llvm::dwarf::Tag Tag;
   if (RD->isStruct() || RD->isInterface())
     Tag = llvm::dwarf::DW_TAG_structure_type;
   else if (RD->isUnion())
     Tag = llvm::dwarf::DW_TAG_union_type;
   else {
+    // FIXME: This could be a struct type giving a default visibility different
+    // than C++ class type, but needs llvm metadata changes first.
     assert(RD->isClass());
     Tag = llvm::dwarf::DW_TAG_class_type;
   }
+  return Tag;
+}
+
+// Creates a forward declaration for a RecordDecl in the given context.
+llvm::DICompositeType *
+CGDebugInfo::getOrCreateRecordFwdDecl(const RecordType *Ty,
+                                      llvm::DIScope *Ctx) {
+  const RecordDecl *RD = Ty->getDecl();
+  if (llvm::DIType *T = getTypeOrNull(CGM.getContext().getRecordType(RD)))
+    return cast<llvm::DICompositeType>(T);
+  llvm::DIFile *DefUnit = getOrCreateFile(RD->getLocation());
+  unsigned Line = getLineNumber(RD->getLocation());
+  StringRef RDName = getClassName(RD);
+
+  uint64_t Size = 0;
+  uint64_t Align = 0;
+
+  const RecordDecl *D = RD->getDefinition();
+  if (D && D->isCompleteDefinition()) {
+    Size = CGM.getContext().getTypeSize(Ty);
+    Align = CGM.getContext().getTypeAlign(Ty);
+  }
 
   // Create the type.
   SmallString<256> FullName = getUniqueTagTypeName(Ty, CGM, TheCU);
-  llvm::DICompositeType RetTy = DBuilder.createReplaceableForwardDecl(
-      Tag, RDName, Ctx, DefUnit, Line, 0, 0, 0, FullName);
+  llvm::DICompositeType *RetTy = DBuilder.createReplaceableCompositeType(
+      getTagForRecord(RD), RDName, Ctx, DefUnit, Line, 0, Size, Align,
+      llvm::DINode::FlagFwdDecl, FullName);
   ReplaceMap.emplace_back(
       std::piecewise_construct, std::make_tuple(Ty),
       std::make_tuple(static_cast<llvm::Metadata *>(RetTy)));
   return RetTy;
 }
 
-llvm::DIType CGDebugInfo::CreatePointerLikeType(llvm::dwarf::Tag Tag,
-                                                const Type *Ty,
-                                                QualType PointeeTy,
-                                                llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreatePointerLikeType(llvm::dwarf::Tag Tag,
+                                                 const Type *Ty,
+                                                 QualType PointeeTy,
+                                                 llvm::DIFile *Unit) {
   if (Tag == llvm::dwarf::DW_TAG_reference_type ||
       Tag == llvm::dwarf::DW_TAG_rvalue_reference_type)
     return DBuilder.createReferenceType(Tag, getOrCreateType(PointeeTy, Unit));
@@ -658,8 +679,8 @@ llvm::DIType CGDebugInfo::CreatePointerLikeType(llvm::dwarf::Tag Tag,
                                     Align);
 }
 
-llvm::DIType CGDebugInfo::getOrCreateStructPtrType(StringRef Name,
-                                                   llvm::DIType &Cache) {
+llvm::DIType *CGDebugInfo::getOrCreateStructPtrType(StringRef Name,
+                                                    llvm::DIType *&Cache) {
   if (Cache)
     return Cache;
   Cache = DBuilder.createForwardDecl(llvm::dwarf::DW_TAG_structure_type, Name,
@@ -669,18 +690,16 @@ llvm::DIType CGDebugInfo::getOrCreateStructPtrType(StringRef Name,
   return Cache;
 }
 
-llvm::DIType CGDebugInfo::CreateType(const BlockPointerType *Ty,
-                                     llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateType(const BlockPointerType *Ty,
+                                      llvm::DIFile *Unit) {
   if (BlockLiteralGeneric)
     return BlockLiteralGeneric;
 
   SmallVector<llvm::Metadata *, 8> EltTys;
-  llvm::DIType FieldTy;
   QualType FType;
   uint64_t FieldSize, FieldOffset;
   unsigned FieldAlign;
-  llvm::DIArray Elements;
-  llvm::DIType EltTy, DescTy;
+  llvm::DINodeArray Elements;
 
   FieldOffset = 0;
   FType = CGM.getContext().UnsignedLongTy;
@@ -690,17 +709,17 @@ llvm::DIType CGDebugInfo::CreateType(const BlockPointerType *Ty,
   Elements = DBuilder.getOrCreateArray(EltTys);
   EltTys.clear();
 
-  unsigned Flags = llvm::DIDescriptor::FlagAppleBlock;
+  unsigned Flags = llvm::DINode::FlagAppleBlock;
   unsigned LineNo = getLineNumber(CurLoc);
 
-  EltTy = DBuilder.createStructType(Unit, "__block_descriptor", Unit, LineNo,
-                                    FieldOffset, 0, Flags, llvm::DIType(),
-                                    Elements);
+  auto *EltTy =
+      DBuilder.createStructType(Unit, "__block_descriptor", Unit, LineNo,
+                                FieldOffset, 0, Flags, nullptr, Elements);
 
   // Bit size, align and offset of the type.
   uint64_t Size = CGM.getContext().getTypeSize(Ty);
 
-  DescTy = DBuilder.createPointerType(EltTy, Size);
+  auto *DescTy = DBuilder.createPointerType(EltTy, Size);
 
   FieldOffset = 0;
   FType = CGM.getContext().getPointerType(CGM.getContext().VoidTy);
@@ -712,29 +731,27 @@ llvm::DIType CGDebugInfo::CreateType(const BlockPointerType *Ty,
   EltTys.push_back(CreateMemberType(Unit, FType, "__FuncPtr", &FieldOffset));
 
   FType = CGM.getContext().getPointerType(CGM.getContext().VoidTy);
-  FieldTy = DescTy;
   FieldSize = CGM.getContext().getTypeSize(Ty);
   FieldAlign = CGM.getContext().getTypeAlign(Ty);
-  FieldTy =
-      DBuilder.createMemberType(Unit, "__descriptor", Unit, LineNo, FieldSize,
-                                FieldAlign, FieldOffset, 0, FieldTy);
-  EltTys.push_back(FieldTy);
+  EltTys.push_back(DBuilder.createMemberType(Unit, "__descriptor", Unit, LineNo,
+                                             FieldSize, FieldAlign, FieldOffset,
+                                             0, DescTy));
 
   FieldOffset += FieldSize;
   Elements = DBuilder.getOrCreateArray(EltTys);
 
-  EltTy = DBuilder.createStructType(Unit, "__block_literal_generic", Unit,
-                                    LineNo, FieldOffset, 0, Flags,
-                                    llvm::DIType(), Elements);
+  EltTy =
+      DBuilder.createStructType(Unit, "__block_literal_generic", Unit, LineNo,
+                                FieldOffset, 0, Flags, nullptr, Elements);
 
   BlockLiteralGeneric = DBuilder.createPointerType(EltTy, Size);
   return BlockLiteralGeneric;
 }
 
-llvm::DIType CGDebugInfo::CreateType(const TemplateSpecializationType *Ty,
-                                     llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateType(const TemplateSpecializationType *Ty,
+                                      llvm::DIFile *Unit) {
   assert(Ty->isTypeAlias());
-  llvm::DIType Src = getOrCreateType(Ty->getAliasedType(), Unit);
+  llvm::DIType *Src = getOrCreateType(Ty->getAliasedType(), Unit);
 
   SmallString<128> NS;
   llvm::raw_svector_ostream OS(NS);
@@ -749,35 +766,26 @@ llvm::DIType CGDebugInfo::CreateType(const TemplateSpecializationType *Ty,
       Ty->getTemplateName().getAsTemplateDecl())->getTemplatedDecl();
 
   SourceLocation Loc = AliasDecl->getLocation();
-  llvm::DIFile File = getOrCreateFile(Loc);
-  unsigned Line = getLineNumber(Loc);
-
-  llvm::DIDescriptor Ctxt =
-      getContextDescriptor(cast<Decl>(AliasDecl->getDeclContext()));
-
-  return DBuilder.createTypedef(Src, internString(OS.str()), File, Line, Ctxt);
+  return DBuilder.createTypedef(
+      Src, internString(OS.str()), getOrCreateFile(Loc), getLineNumber(Loc),
+      getContextDescriptor(cast<Decl>(AliasDecl->getDeclContext())));
 }
 
-llvm::DIType CGDebugInfo::CreateType(const TypedefType *Ty, llvm::DIFile Unit) {
-  // Typedefs are derived from some other type.  If we have a typedef of a
-  // typedef, make sure to emit the whole chain.
-  llvm::DIType Src = getOrCreateType(Ty->getDecl()->getUnderlyingType(), Unit);
+llvm::DIType *CGDebugInfo::CreateType(const TypedefType *Ty,
+                                      llvm::DIFile *Unit) {
   // We don't set size information, but do specify where the typedef was
   // declared.
   SourceLocation Loc = Ty->getDecl()->getLocation();
-  llvm::DIFile File = getOrCreateFile(Loc);
-  unsigned Line = getLineNumber(Loc);
-  const TypedefNameDecl *TyDecl = Ty->getDecl();
 
-  llvm::DIDescriptor TypedefContext =
-      getContextDescriptor(cast<Decl>(Ty->getDecl()->getDeclContext()));
-
-  return DBuilder.createTypedef(Src, TyDecl->getName(), File, Line,
-                                TypedefContext);
+  // Typedefs are derived from some other type.
+  return DBuilder.createTypedef(
+      getOrCreateType(Ty->getDecl()->getUnderlyingType(), Unit),
+      Ty->getDecl()->getName(), getOrCreateFile(Loc), getLineNumber(Loc),
+      getContextDescriptor(cast<Decl>(Ty->getDecl()->getDeclContext())));
 }
 
-llvm::DIType CGDebugInfo::CreateType(const FunctionType *Ty,
-                                     llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateType(const FunctionType *Ty,
+                                      llvm::DIFile *Unit) {
   SmallVector<llvm::Metadata *, 16> EltTys;
 
   // Add the result type at least.
@@ -794,11 +802,11 @@ llvm::DIType CGDebugInfo::CreateType(const FunctionType *Ty,
       EltTys.push_back(DBuilder.createUnspecifiedParameter());
   }
 
-  llvm::DITypeArray EltTypeArray = DBuilder.getOrCreateTypeArray(EltTys);
+  llvm::DITypeRefArray EltTypeArray = DBuilder.getOrCreateTypeArray(EltTys);
   return DBuilder.createSubroutineType(Unit, EltTypeArray);
 }
 
-/// Convert an AccessSpecifier into the corresponding DIDescriptor flag.
+/// Convert an AccessSpecifier into the corresponding DINode flag.
 /// As an optimization, return 0 if the access specifier equals the
 /// default for the containing type.
 static unsigned getAccessFlag(AccessSpecifier Access, const RecordDecl *RD) {
@@ -813,25 +821,25 @@ static unsigned getAccessFlag(AccessSpecifier Access, const RecordDecl *RD) {
 
   switch (Access) {
   case clang::AS_private:
-    return llvm::DIDescriptor::FlagPrivate;
+    return llvm::DINode::FlagPrivate;
   case clang::AS_protected:
-    return llvm::DIDescriptor::FlagProtected;
+    return llvm::DINode::FlagProtected;
   case clang::AS_public:
-    return llvm::DIDescriptor::FlagPublic;
+    return llvm::DINode::FlagPublic;
   case clang::AS_none:
     return 0;
   }
   llvm_unreachable("unexpected access enumerator");
 }
 
-llvm::DIType CGDebugInfo::createFieldType(
+llvm::DIType *CGDebugInfo::createFieldType(
     StringRef name, QualType type, uint64_t sizeInBitsOverride,
     SourceLocation loc, AccessSpecifier AS, uint64_t offsetInBits,
-    llvm::DIFile tunit, llvm::DIScope scope, const RecordDecl *RD) {
-  llvm::DIType debugType = getOrCreateType(type, tunit);
+    llvm::DIFile *tunit, llvm::DIScope *scope, const RecordDecl *RD) {
+  llvm::DIType *debugType = getOrCreateType(type, tunit);
 
   // Get the location for the field.
-  llvm::DIFile file = getOrCreateFile(loc);
+  llvm::DIFile *file = getOrCreateFile(loc);
   unsigned line = getLineNumber(loc);
 
   uint64_t SizeInBits = 0;
@@ -853,7 +861,7 @@ llvm::DIType CGDebugInfo::createFieldType(
 /// CollectRecordLambdaFields - Helper for CollectRecordFields.
 void CGDebugInfo::CollectRecordLambdaFields(
     const CXXRecordDecl *CXXDecl, SmallVectorImpl<llvm::Metadata *> &elements,
-    llvm::DIType RecordTy) {
+    llvm::DIType *RecordTy) {
   // For C++11 Lambdas a Field will be the same as a Capture, but the Capture
   // has the name and the location of the variable so we should iterate over
   // both concurrently.
@@ -866,14 +874,14 @@ void CGDebugInfo::CollectRecordLambdaFields(
     const LambdaCapture &C = *I;
     if (C.capturesVariable()) {
       VarDecl *V = C.getCapturedVar();
-      llvm::DIFile VUnit = getOrCreateFile(C.getLocation());
+      llvm::DIFile *VUnit = getOrCreateFile(C.getLocation());
       StringRef VName = V->getName();
       uint64_t SizeInBitsOverride = 0;
       if (Field->isBitField()) {
         SizeInBitsOverride = Field->getBitWidthValue(CGM.getContext());
         assert(SizeInBitsOverride && "found named 0-width bitfield");
       }
-      llvm::DIType fieldType = createFieldType(
+      llvm::DIType *fieldType = createFieldType(
           VName, Field->getType(), SizeInBitsOverride, C.getLocation(),
           Field->getAccess(), layout.getFieldOffset(fieldno), VUnit, RecordTy,
           CXXDecl);
@@ -884,9 +892,9 @@ void CGDebugInfo::CollectRecordLambdaFields(
       // by using AT_object_pointer for the function and having that be
       // used as 'this' for semantic references.
       FieldDecl *f = *Field;
-      llvm::DIFile VUnit = getOrCreateFile(f->getLocation());
+      llvm::DIFile *VUnit = getOrCreateFile(f->getLocation());
       QualType type = f->getType();
-      llvm::DIType fieldType = createFieldType(
+      llvm::DIType *fieldType = createFieldType(
           "this", type, 0, f->getLocation(), f->getAccess(),
           layout.getFieldOffset(fieldno), VUnit, RecordTy, CXXDecl);
 
@@ -896,14 +904,14 @@ void CGDebugInfo::CollectRecordLambdaFields(
 }
 
 /// Helper for CollectRecordFields.
-llvm::DIDerivedType CGDebugInfo::CreateRecordStaticField(const VarDecl *Var,
-                                                         llvm::DIType RecordTy,
-                                                         const RecordDecl *RD) {
+llvm::DIDerivedType *
+CGDebugInfo::CreateRecordStaticField(const VarDecl *Var, llvm::DIType *RecordTy,
+                                     const RecordDecl *RD) {
   // Create the descriptor for the static variable, with or without
   // constant initializers.
   Var = Var->getCanonicalDecl();
-  llvm::DIFile VUnit = getOrCreateFile(Var->getLocation());
-  llvm::DIType VTy = getOrCreateType(Var->getType(), VUnit);
+  llvm::DIFile *VUnit = getOrCreateFile(Var->getLocation());
+  llvm::DIType *VTy = getOrCreateType(Var->getType(), VUnit);
 
   unsigned LineNumber = getLineNumber(Var->getLocation());
   StringRef VName = Var->getName();
@@ -919,7 +927,7 @@ llvm::DIDerivedType CGDebugInfo::CreateRecordStaticField(const VarDecl *Var,
   }
 
   unsigned Flags = getAccessFlag(Var->getAccess(), RD);
-  llvm::DIDerivedType GV = DBuilder.createStaticMemberType(
+  llvm::DIDerivedType *GV = DBuilder.createStaticMemberType(
       RecordTy, VName, VUnit, LineNumber, VTy, Flags, C);
   StaticDataMemberCache[Var->getCanonicalDecl()].reset(GV);
   return GV;
@@ -927,8 +935,8 @@ llvm::DIDerivedType CGDebugInfo::CreateRecordStaticField(const VarDecl *Var,
 
 /// CollectRecordNormalField - Helper for CollectRecordFields.
 void CGDebugInfo::CollectRecordNormalField(
-    const FieldDecl *field, uint64_t OffsetInBits, llvm::DIFile tunit,
-    SmallVectorImpl<llvm::Metadata *> &elements, llvm::DIType RecordTy,
+    const FieldDecl *field, uint64_t OffsetInBits, llvm::DIFile *tunit,
+    SmallVectorImpl<llvm::Metadata *> &elements, llvm::DIType *RecordTy,
     const RecordDecl *RD) {
   StringRef name = field->getName();
   QualType type = field->getType();
@@ -943,7 +951,7 @@ void CGDebugInfo::CollectRecordNormalField(
     assert(SizeInBitsOverride && "found named 0-width bitfield");
   }
 
-  llvm::DIType fieldType =
+  llvm::DIType *fieldType =
       createFieldType(name, type, SizeInBitsOverride, field->getLocation(),
                       field->getAccess(), OffsetInBits, tunit, RecordTy, RD);
 
@@ -953,9 +961,9 @@ void CGDebugInfo::CollectRecordNormalField(
 /// CollectRecordFields - A helper function to collect debug info for
 /// record fields. This is used while creating debug info entry for a Record.
 void CGDebugInfo::CollectRecordFields(
-    const RecordDecl *record, llvm::DIFile tunit,
+    const RecordDecl *record, llvm::DIFile *tunit,
     SmallVectorImpl<llvm::Metadata *> &elements,
-    llvm::DICompositeType RecordTy) {
+    llvm::DICompositeType *RecordTy) {
   const CXXRecordDecl *CXXDecl = dyn_cast<CXXRecordDecl>(record);
 
   if (CXXDecl && CXXDecl->isLambda())
@@ -975,8 +983,7 @@ void CGDebugInfo::CollectRecordFields(
         if (MI != StaticDataMemberCache.end()) {
           assert(MI->second &&
                  "Static data member declaration should still exist");
-          elements.push_back(
-              llvm::DIDerivedType(cast<llvm::MDNode>(MI->second)));
+          elements.push_back(cast<llvm::DIDerivedTypeBase>(MI->second));
         } else {
           auto Field = CreateRecordStaticField(V, RecordTy, record);
           elements.push_back(Field);
@@ -994,27 +1001,29 @@ void CGDebugInfo::CollectRecordFields(
 /// getOrCreateMethodType - CXXMethodDecl's type is a FunctionType. This
 /// function type is not updated to include implicit "this" pointer. Use this
 /// routine to get a method type which includes "this" pointer.
-llvm::DICompositeType
+llvm::DISubroutineType *
 CGDebugInfo::getOrCreateMethodType(const CXXMethodDecl *Method,
-                                   llvm::DIFile Unit) {
+                                   llvm::DIFile *Unit) {
   const FunctionProtoType *Func = Method->getType()->getAs<FunctionProtoType>();
   if (Method->isStatic())
-    return llvm::DICompositeType(getOrCreateType(QualType(Func, 0), Unit));
+    return cast_or_null<llvm::DISubroutineType>(
+        getOrCreateType(QualType(Func, 0), Unit));
   return getOrCreateInstanceMethodType(Method->getThisType(CGM.getContext()),
                                        Func, Unit);
 }
 
-llvm::DICompositeType CGDebugInfo::getOrCreateInstanceMethodType(
-    QualType ThisPtr, const FunctionProtoType *Func, llvm::DIFile Unit) {
+llvm::DISubroutineType *CGDebugInfo::getOrCreateInstanceMethodType(
+    QualType ThisPtr, const FunctionProtoType *Func, llvm::DIFile *Unit) {
   // Add "this" pointer.
-  llvm::DITypeArray Args = llvm::DISubroutineType(
-      getOrCreateType(QualType(Func, 0), Unit)).getTypeArray();
-  assert(Args.getNumElements() && "Invalid number of arguments!");
+  llvm::DITypeRefArray Args(
+      cast<llvm::DISubroutineType>(getOrCreateType(QualType(Func, 0), Unit))
+          ->getTypeArray());
+  assert(Args.size() && "Invalid number of arguments!");
 
   SmallVector<llvm::Metadata *, 16> Elts;
 
   // First element is always return type. For 'void' functions it is NULL.
-  Elts.push_back(Args.getElement(0));
+  Elts.push_back(Args[0]);
 
   // "this" pointer is always first argument.
   const CXXRecordDecl *RD = ThisPtr->getPointeeCXXRecordDecl();
@@ -1025,8 +1034,8 @@ llvm::DICompositeType CGDebugInfo::getOrCreateInstanceMethodType(
     unsigned AS = CGM.getContext().getTargetAddressSpace(PointeeTy);
     uint64_t Size = CGM.getTarget().getPointerWidth(AS);
     uint64_t Align = CGM.getContext().getTypeAlign(ThisPtrTy);
-    llvm::DIType PointeeType = getOrCreateType(PointeeTy, Unit);
-    llvm::DIType ThisPtrType =
+    llvm::DIType *PointeeType = getOrCreateType(PointeeTy, Unit);
+    llvm::DIType *ThisPtrType =
         DBuilder.createPointerType(PointeeType, Size, Align);
     TypeCache[ThisPtr.getAsOpaquePtr()].reset(ThisPtrType);
     // TODO: This and the artificial type below are misleading, the
@@ -1035,23 +1044,23 @@ llvm::DICompositeType CGDebugInfo::getOrCreateInstanceMethodType(
     ThisPtrType = DBuilder.createObjectPointerType(ThisPtrType);
     Elts.push_back(ThisPtrType);
   } else {
-    llvm::DIType ThisPtrType = getOrCreateType(ThisPtr, Unit);
+    llvm::DIType *ThisPtrType = getOrCreateType(ThisPtr, Unit);
     TypeCache[ThisPtr.getAsOpaquePtr()].reset(ThisPtrType);
     ThisPtrType = DBuilder.createObjectPointerType(ThisPtrType);
     Elts.push_back(ThisPtrType);
   }
 
   // Copy rest of the arguments.
-  for (unsigned i = 1, e = Args.getNumElements(); i != e; ++i)
-    Elts.push_back(Args.getElement(i));
+  for (unsigned i = 1, e = Args.size(); i != e; ++i)
+    Elts.push_back(Args[i]);
 
-  llvm::DITypeArray EltTypeArray = DBuilder.getOrCreateTypeArray(Elts);
+  llvm::DITypeRefArray EltTypeArray = DBuilder.getOrCreateTypeArray(Elts);
 
   unsigned Flags = 0;
   if (Func->getExtProtoInfo().RefQualifier == RQ_LValue)
-    Flags |= llvm::DIDescriptor::FlagLValueReference;
+    Flags |= llvm::DINode::FlagLValueReference;
   if (Func->getExtProtoInfo().RefQualifier == RQ_RValue)
-    Flags |= llvm::DIDescriptor::FlagRValueReference;
+    Flags |= llvm::DINode::FlagRValueReference;
 
   return DBuilder.createSubroutineType(Unit, EltTypeArray, Flags);
 }
@@ -1066,16 +1075,15 @@ static bool isFunctionLocalClass(const CXXRecordDecl *RD) {
   return false;
 }
 
-/// CreateCXXMemberFunction - A helper function to create a DISubprogram for
+/// CreateCXXMemberFunction - A helper function to create a subprogram for
 /// a single member function GlobalDecl.
-llvm::DISubprogram
-CGDebugInfo::CreateCXXMemberFunction(const CXXMethodDecl *Method,
-                                     llvm::DIFile Unit, llvm::DIType RecordTy) {
+llvm::DISubprogram *CGDebugInfo::CreateCXXMemberFunction(
+    const CXXMethodDecl *Method, llvm::DIFile *Unit, llvm::DIType *RecordTy) {
   bool IsCtorOrDtor =
       isa<CXXConstructorDecl>(Method) || isa<CXXDestructorDecl>(Method);
 
   StringRef MethodName = getFunctionName(Method);
-  llvm::DICompositeType MethodTy = getOrCreateMethodType(Method, Unit);
+  llvm::DISubroutineType *MethodTy = getOrCreateMethodType(Method, Unit);
 
   // Since a single ctor/dtor corresponds to multiple functions, it doesn't
   // make sense to give a single ctor/dtor a linkage name.
@@ -1084,7 +1092,7 @@ CGDebugInfo::CreateCXXMemberFunction(const CXXMethodDecl *Method,
     MethodLinkageName = CGM.getMangledName(Method);
 
   // Get the location for the method.
-  llvm::DIFile MethodDefUnit;
+  llvm::DIFile *MethodDefUnit = nullptr;
   unsigned MethodLine = 0;
   if (!Method->isImplicit()) {
     MethodDefUnit = getOrCreateFile(Method->getLocation());
@@ -1092,7 +1100,7 @@ CGDebugInfo::CreateCXXMemberFunction(const CXXMethodDecl *Method,
   }
 
   // Collect virtual method info.
-  llvm::DIType ContainingType;
+  llvm::DIType *ContainingType = nullptr;
   unsigned Virtuality = 0;
   unsigned VIndex = 0;
 
@@ -1115,29 +1123,29 @@ CGDebugInfo::CreateCXXMemberFunction(const CXXMethodDecl *Method,
 
   unsigned Flags = 0;
   if (Method->isImplicit())
-    Flags |= llvm::DIDescriptor::FlagArtificial;
+    Flags |= llvm::DINode::FlagArtificial;
   Flags |= getAccessFlag(Method->getAccess(), Method->getParent());
   if (const CXXConstructorDecl *CXXC = dyn_cast<CXXConstructorDecl>(Method)) {
     if (CXXC->isExplicit())
-      Flags |= llvm::DIDescriptor::FlagExplicit;
+      Flags |= llvm::DINode::FlagExplicit;
   } else if (const CXXConversionDecl *CXXC =
                  dyn_cast<CXXConversionDecl>(Method)) {
     if (CXXC->isExplicit())
-      Flags |= llvm::DIDescriptor::FlagExplicit;
+      Flags |= llvm::DINode::FlagExplicit;
   }
   if (Method->hasPrototype())
-    Flags |= llvm::DIDescriptor::FlagPrototyped;
+    Flags |= llvm::DINode::FlagPrototyped;
   if (Method->getRefQualifier() == RQ_LValue)
-    Flags |= llvm::DIDescriptor::FlagLValueReference;
+    Flags |= llvm::DINode::FlagLValueReference;
   if (Method->getRefQualifier() == RQ_RValue)
-    Flags |= llvm::DIDescriptor::FlagRValueReference;
+    Flags |= llvm::DINode::FlagRValueReference;
 
-  llvm::DIArray TParamsArray = CollectFunctionTemplateParams(Method, Unit);
-  llvm::DISubprogram SP = DBuilder.createMethod(
+  llvm::DINodeArray TParamsArray = CollectFunctionTemplateParams(Method, Unit);
+  llvm::DISubprogram *SP = DBuilder.createMethod(
       RecordTy, MethodName, MethodLinkageName, MethodDefUnit, MethodLine,
       MethodTy, /*isLocalToUnit=*/false,
       /* isDefinition=*/false, Virtuality, VIndex, ContainingType, Flags,
-      CGM.getLangOpts().Optimize, nullptr, TParamsArray);
+      CGM.getLangOpts().Optimize, nullptr, TParamsArray.get());
 
   SPCache[Method->getCanonicalDecl()].reset(SP);
 
@@ -1148,8 +1156,8 @@ CGDebugInfo::CreateCXXMemberFunction(const CXXMethodDecl *Method,
 /// C++ member functions. This is used while creating debug info entry for
 /// a Record.
 void CGDebugInfo::CollectCXXMemberFunctions(
-    const CXXRecordDecl *RD, llvm::DIFile Unit,
-    SmallVectorImpl<llvm::Metadata *> &EltTys, llvm::DIType RecordTy) {
+    const CXXRecordDecl *RD, llvm::DIFile *Unit,
+    SmallVectorImpl<llvm::Metadata *> &EltTys, llvm::DIType *RecordTy) {
 
   // Since we want more than just the individual member decls if we
   // have templated functions iterate over every declaration to gather
@@ -1187,10 +1195,9 @@ void CGDebugInfo::CollectCXXMemberFunctions(
 /// CollectCXXBases - A helper function to collect debug info for
 /// C++ base classes. This is used while creating debug info entry for
 /// a Record.
-void CGDebugInfo::CollectCXXBases(const CXXRecordDecl *RD, llvm::DIFile Unit,
+void CGDebugInfo::CollectCXXBases(const CXXRecordDecl *RD, llvm::DIFile *Unit,
                                   SmallVectorImpl<llvm::Metadata *> &EltTys,
-                                  llvm::DIType RecordTy) {
-
+                                  llvm::DIType *RecordTy) {
   const ASTRecordLayout &RL = CGM.getContext().getASTRecordLayout(RD);
   for (const auto &BI : RD->bases()) {
     unsigned BFlags = 0;
@@ -1212,24 +1219,24 @@ void CGDebugInfo::CollectCXXBases(const CXXRecordDecl *RD, llvm::DIFile Unit,
         BaseOffset =
             4 * CGM.getMicrosoftVTableContext().getVBTableIndex(RD, Base);
       }
-      BFlags = llvm::DIDescriptor::FlagVirtual;
+      BFlags = llvm::DINode::FlagVirtual;
     } else
       BaseOffset = CGM.getContext().toBits(RL.getBaseClassOffset(Base));
     // FIXME: Inconsistent units for BaseOffset. It is in bytes when
     // BI->isVirtual() and bits when not.
 
     BFlags |= getAccessFlag(BI.getAccessSpecifier(), RD);
-    llvm::DIType DTy = DBuilder.createInheritance(
+    llvm::DIType *DTy = DBuilder.createInheritance(
         RecordTy, getOrCreateType(BI.getType(), Unit), BaseOffset, BFlags);
     EltTys.push_back(DTy);
   }
 }
 
 /// CollectTemplateParams - A helper function to collect template parameters.
-llvm::DIArray
+llvm::DINodeArray
 CGDebugInfo::CollectTemplateParams(const TemplateParameterList *TPList,
                                    ArrayRef<TemplateArgument> TAList,
-                                   llvm::DIFile Unit) {
+                                   llvm::DIFile *Unit) {
   SmallVector<llvm::Metadata *, 16> TemplateParams;
   for (unsigned i = 0, e = TAList.size(); i != e; ++i) {
     const TemplateArgument &TA = TAList[i];
@@ -1238,23 +1245,20 @@ CGDebugInfo::CollectTemplateParams(const TemplateParameterList *TPList,
       Name = TPList->getParam(i)->getName();
     switch (TA.getKind()) {
     case TemplateArgument::Type: {
-      llvm::DIType TTy = getOrCreateType(TA.getAsType(), Unit);
-      llvm::DITemplateTypeParameter TTP =
-          DBuilder.createTemplateTypeParameter(TheCU, Name, TTy);
-      TemplateParams.push_back(TTP);
+      llvm::DIType *TTy = getOrCreateType(TA.getAsType(), Unit);
+      TemplateParams.push_back(
+          DBuilder.createTemplateTypeParameter(TheCU, Name, TTy));
     } break;
     case TemplateArgument::Integral: {
-      llvm::DIType TTy = getOrCreateType(TA.getIntegralType(), Unit);
-      llvm::DITemplateValueParameter TVP =
-          DBuilder.createTemplateValueParameter(
-              TheCU, Name, TTy,
-              llvm::ConstantInt::get(CGM.getLLVMContext(), TA.getAsIntegral()));
-      TemplateParams.push_back(TVP);
+      llvm::DIType *TTy = getOrCreateType(TA.getIntegralType(), Unit);
+      TemplateParams.push_back(DBuilder.createTemplateValueParameter(
+          TheCU, Name, TTy,
+          llvm::ConstantInt::get(CGM.getLLVMContext(), TA.getAsIntegral())));
     } break;
     case TemplateArgument::Declaration: {
       const ValueDecl *D = TA.getAsDecl();
       QualType T = TA.getParamTypeForDecl().getDesugaredType(CGM.getContext());
-      llvm::DIType TTy = getOrCreateType(T, Unit);
+      llvm::DIType *TTy = getOrCreateType(T, Unit);
       llvm::Constant *V = nullptr;
       const CXXMethodDecl *MD;
       // Variable pointer template parameters have a value that is the address
@@ -1278,15 +1282,13 @@ CGDebugInfo::CollectTemplateParams(const TemplateParameterList *TPList,
             CGM.getContext().toCharUnitsFromBits((int64_t)fieldOffset);
         V = CGM.getCXXABI().EmitMemberDataPointer(MPT, chars);
       }
-      llvm::DITemplateValueParameter TVP =
-          DBuilder.createTemplateValueParameter(
-              TheCU, Name, TTy,
-              cast_or_null<llvm::Constant>(V->stripPointerCasts()));
-      TemplateParams.push_back(TVP);
+      TemplateParams.push_back(DBuilder.createTemplateValueParameter(
+          TheCU, Name, TTy,
+          cast_or_null<llvm::Constant>(V->stripPointerCasts())));
     } break;
     case TemplateArgument::NullPtr: {
       QualType T = TA.getNullPtrType();
-      llvm::DIType TTy = getOrCreateType(T, Unit);
+      llvm::DIType *TTy = getOrCreateType(T, Unit);
       llvm::Constant *V = nullptr;
       // Special case member data pointer null values since they're actually -1
       // instead of zero.
@@ -1301,24 +1303,19 @@ CGDebugInfo::CollectTemplateParams(const TemplateParameterList *TPList,
           V = CGM.getCXXABI().EmitNullMemberPointer(MPT);
       if (!V)
         V = llvm::ConstantInt::get(CGM.Int8Ty, 0);
-      llvm::DITemplateValueParameter TVP =
-          DBuilder.createTemplateValueParameter(TheCU, Name, TTy,
-                                                cast<llvm::Constant>(V));
-      TemplateParams.push_back(TVP);
-    } break;
-    case TemplateArgument::Template: {
-      llvm::DITemplateValueParameter
-      TVP = DBuilder.createTemplateTemplateParameter(
-          TheCU, Name, llvm::DIType(),
-          TA.getAsTemplate().getAsTemplateDecl()->getQualifiedNameAsString());
-      TemplateParams.push_back(TVP);
-    } break;
-    case TemplateArgument::Pack: {
-      llvm::DITemplateValueParameter TVP = DBuilder.createTemplateParameterPack(
-          TheCU, Name, llvm::DIType(),
-          CollectTemplateParams(nullptr, TA.getPackAsArray(), Unit));
-      TemplateParams.push_back(TVP);
+      TemplateParams.push_back(DBuilder.createTemplateValueParameter(
+          TheCU, Name, TTy, cast<llvm::Constant>(V)));
     } break;
+    case TemplateArgument::Template:
+      TemplateParams.push_back(DBuilder.createTemplateTemplateParameter(
+          TheCU, Name, nullptr,
+          TA.getAsTemplate().getAsTemplateDecl()->getQualifiedNameAsString()));
+      break;
+    case TemplateArgument::Pack:
+      TemplateParams.push_back(DBuilder.createTemplateParameterPack(
+          TheCU, Name, nullptr,
+          CollectTemplateParams(nullptr, TA.getPackAsArray(), Unit)));
+      break;
     case TemplateArgument::Expression: {
       const Expr *E = TA.getAsExpr();
       QualType T = E->getType();
@@ -1326,11 +1323,9 @@ CGDebugInfo::CollectTemplateParams(const TemplateParameterList *TPList,
         T = CGM.getContext().getLValueReferenceType(T);
       llvm::Constant *V = CGM.EmitConstantExpr(E, T);
       assert(V && "Expression in template argument isn't constant");
-      llvm::DIType TTy = getOrCreateType(T, Unit);
-      llvm::DITemplateValueParameter TVP =
-          DBuilder.createTemplateValueParameter(
-              TheCU, Name, TTy, cast<llvm::Constant>(V->stripPointerCasts()));
-      TemplateParams.push_back(TVP);
+      llvm::DIType *TTy = getOrCreateType(T, Unit);
+      TemplateParams.push_back(DBuilder.createTemplateValueParameter(
+          TheCU, Name, TTy, cast<llvm::Constant>(V->stripPointerCasts())));
     } break;
     // And the following should never occur:
     case TemplateArgument::TemplateExpansion:
@@ -1344,8 +1339,9 @@ CGDebugInfo::CollectTemplateParams(const TemplateParameterList *TPList,
 
 /// CollectFunctionTemplateParams - A helper function to collect debug
 /// info for function template parameters.
-llvm::DIArray CGDebugInfo::CollectFunctionTemplateParams(const FunctionDecl *FD,
-                                                         llvm::DIFile Unit) {
+llvm::DINodeArray
+CGDebugInfo::CollectFunctionTemplateParams(const FunctionDecl *FD,
+                                           llvm::DIFile *Unit) {
   if (FD->getTemplatedKind() ==
       FunctionDecl::TK_FunctionTemplateSpecialization) {
     const TemplateParameterList *TList = FD->getTemplateSpecializationInfo()
@@ -1354,13 +1350,13 @@ llvm::DIArray CGDebugInfo::CollectFunctionTemplateParams(const FunctionDecl *FD,
     return CollectTemplateParams(
         TList, FD->getTemplateSpecializationArgs()->asArray(), Unit);
   }
-  return llvm::DIArray();
+  return llvm::DINodeArray();
 }
 
 /// CollectCXXTemplateParams - A helper function to collect debug info for
 /// template parameters.
-llvm::DIArray CGDebugInfo::CollectCXXTemplateParams(
-    const ClassTemplateSpecializationDecl *TSpecial, llvm::DIFile Unit) {
+llvm::DINodeArray CGDebugInfo::CollectCXXTemplateParams(
+    const ClassTemplateSpecializationDecl *TSpecial, llvm::DIFile *Unit) {
   // Always get the full list of parameters, not just the ones from
   // the specialization.
   TemplateParameterList *TPList =
@@ -1370,18 +1366,18 @@ llvm::DIArray CGDebugInfo::CollectCXXTemplateParams(
 }
 
 /// getOrCreateVTablePtrType - Return debug info descriptor for vtable.
-llvm::DIType CGDebugInfo::getOrCreateVTablePtrType(llvm::DIFile Unit) {
-  if (VTablePtrType.isValid())
+llvm::DIType *CGDebugInfo::getOrCreateVTablePtrType(llvm::DIFile *Unit) {
+  if (VTablePtrType)
     return VTablePtrType;
 
   ASTContext &Context = CGM.getContext();
 
   /* Function type */
   llvm::Metadata *STy = getOrCreateType(Context.IntTy, Unit);
-  llvm::DITypeArray SElements = DBuilder.getOrCreateTypeArray(STy);
-  llvm::DIType SubTy = DBuilder.createSubroutineType(Unit, SElements);
+  llvm::DITypeRefArray SElements = DBuilder.getOrCreateTypeArray(STy);
+  llvm::DIType *SubTy = DBuilder.createSubroutineType(Unit, SElements);
   unsigned Size = Context.getTypeSize(Context.VoidPtrTy);
-  llvm::DIType vtbl_ptr_type =
+  llvm::DIType *vtbl_ptr_type =
       DBuilder.createPointerType(SubTy, Size, 0, "__vtbl_ptr_type");
   VTablePtrType = DBuilder.createPointerType(vtbl_ptr_type, Size);
   return VTablePtrType;
@@ -1395,7 +1391,7 @@ StringRef CGDebugInfo::getVTableName(const CXXRecordDecl *RD) {
 
 /// CollectVTableInfo - If the C++ class has vtable info then insert appropriate
 /// debug info entry in EltTys vector.
-void CGDebugInfo::CollectVTableInfo(const CXXRecordDecl *RD, llvm::DIFile Unit,
+void CGDebugInfo::CollectVTableInfo(const CXXRecordDecl *RD, llvm::DIFile *Unit,
                                     SmallVectorImpl<llvm::Metadata *> &EltTys) {
   const ASTRecordLayout &RL = CGM.getContext().getASTRecordLayout(RD);
 
@@ -1408,26 +1404,26 @@ void CGDebugInfo::CollectVTableInfo(const CXXRecordDecl *RD, llvm::DIFile Unit,
     return;
 
   unsigned Size = CGM.getContext().getTypeSize(CGM.getContext().VoidPtrTy);
-  llvm::DIType VPTR = DBuilder.createMemberType(
+  llvm::DIType *VPTR = DBuilder.createMemberType(
       Unit, getVTableName(RD), Unit, 0, Size, 0, 0,
-      llvm::DIDescriptor::FlagArtificial, getOrCreateVTablePtrType(Unit));
+      llvm::DINode::FlagArtificial, getOrCreateVTablePtrType(Unit));
   EltTys.push_back(VPTR);
 }
 
 /// getOrCreateRecordType - Emit record type's standalone debug info.
-llvm::DIType CGDebugInfo::getOrCreateRecordType(QualType RTy,
-                                                SourceLocation Loc) {
+llvm::DIType *CGDebugInfo::getOrCreateRecordType(QualType RTy,
+                                                 SourceLocation Loc) {
   assert(DebugKind >= CodeGenOptions::LimitedDebugInfo);
-  llvm::DIType T = getOrCreateType(RTy, getOrCreateFile(Loc));
+  llvm::DIType *T = getOrCreateType(RTy, getOrCreateFile(Loc));
   return T;
 }
 
 /// getOrCreateInterfaceType - Emit an objective c interface type standalone
 /// debug info.
-llvm::DIType CGDebugInfo::getOrCreateInterfaceType(QualType D,
-                                                   SourceLocation Loc) {
+llvm::DIType *CGDebugInfo::getOrCreateInterfaceType(QualType D,
+                                                    SourceLocation Loc) {
   assert(DebugKind >= CodeGenOptions::LimitedDebugInfo);
-  llvm::DIType T = getOrCreateType(D, getOrCreateFile(Loc));
+  llvm::DIType *T = getOrCreateType(D, getOrCreateFile(Loc));
   RetainedTypes.push_back(D.getAsOpaquePtr());
   return T;
 }
@@ -1438,11 +1434,10 @@ void CGDebugInfo::completeType(const EnumDecl *ED) {
   QualType Ty = CGM.getContext().getEnumType(ED);
   void *TyPtr = Ty.getAsOpaquePtr();
   auto I = TypeCache.find(TyPtr);
-  if (I == TypeCache.end() ||
-      !llvm::DIType(cast<llvm::MDNode>(I->second)).isForwardDecl())
+  if (I == TypeCache.end() || !cast<llvm::DIType>(I->second)->isForwardDecl())
     return;
-  llvm::DIType Res = CreateTypeDefinition(Ty->castAs<EnumType>());
-  assert(!Res.isForwardDecl());
+  llvm::DIType *Res = CreateTypeDefinition(Ty->castAs<EnumType>());
+  assert(!Res->isForwardDecl());
   TypeCache[TyPtr].reset(Res);
 }
 
@@ -1461,8 +1456,8 @@ void CGDebugInfo::completeRequiredType(const RecordDecl *RD) {
       return;
 
   QualType Ty = CGM.getContext().getRecordType(RD);
-  llvm::DIType T = getTypeOrNull(Ty);
-  if (T && T.isForwardDecl())
+  llvm::DIType *T = getTypeOrNull(Ty);
+  if (T && T->isForwardDecl())
     completeClassData(RD);
 }
 
@@ -1472,11 +1467,10 @@ void CGDebugInfo::completeClassData(const RecordDecl *RD) {
   QualType Ty = CGM.getContext().getRecordType(RD);
   void *TyPtr = Ty.getAsOpaquePtr();
   auto I = TypeCache.find(TyPtr);
-  if (I != TypeCache.end() &&
-      !llvm::DIType(cast<llvm::MDNode>(I->second)).isForwardDecl())
+  if (I != TypeCache.end() && !cast<llvm::DIType>(I->second)->isForwardDecl())
     return;
-  llvm::DIType Res = CreateTypeDefinition(Ty->castAs<RecordType>());
-  assert(!Res.isForwardDecl());
+  llvm::DIType *Res = CreateTypeDefinition(Ty->castAs<RecordType>());
+  assert(!Res->isForwardDecl());
   TypeCache[TyPtr].reset(Res);
 }
 
@@ -1524,9 +1518,9 @@ static bool shouldOmitDefinition(CodeGenOptions::DebugInfoKind DebugKind,
 }
 
 /// CreateType - get structure or union type.
-llvm::DIType CGDebugInfo::CreateType(const RecordType *Ty) {
+llvm::DIType *CGDebugInfo::CreateType(const RecordType *Ty) {
   RecordDecl *RD = Ty->getDecl();
-  llvm::DICompositeType T(getTypeOrNull(QualType(Ty, 0)));
+  llvm::DIType *T = cast_or_null<llvm::DIType>(getTypeOrNull(QualType(Ty, 0)));
   if (T || shouldOmitDefinition(DebugKind, RD, CGM.getLangOpts())) {
     if (!T)
       T = getOrCreateRecordFwdDecl(
@@ -1537,11 +1531,11 @@ llvm::DIType CGDebugInfo::CreateType(const RecordType *Ty) {
   return CreateTypeDefinition(Ty);
 }
 
-llvm::DIType CGDebugInfo::CreateTypeDefinition(const RecordType *Ty) {
+llvm::DIType *CGDebugInfo::CreateTypeDefinition(const RecordType *Ty) {
   RecordDecl *RD = Ty->getDecl();
 
   // Get overall information about the record type for the debug info.
-  llvm::DIFile DefUnit = getOrCreateFile(RD->getLocation());
+  llvm::DIFile *DefUnit = getOrCreateFile(RD->getLocation());
 
   // Records and classes and unions can all be recursive.  To handle them, we
   // first generate a debug descriptor for the struct as a forward declaration.
@@ -1550,11 +1544,11 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const RecordType *Ty) {
   // may refer to the forward decl if the struct is recursive) and replace all
   // uses of the forward declaration with the final definition.
 
-  llvm::DICompositeType FwdDecl(getOrCreateLimitedType(Ty, DefUnit));
-  assert(FwdDecl.isCompositeType() &&
-         "The debug type of a RecordType should be a llvm::DICompositeType");
+  auto *FwdDecl =
+      cast<llvm::DICompositeType>(getOrCreateLimitedType(Ty, DefUnit));
 
-  if (FwdDecl.isForwardDecl())
+  const RecordDecl *D = RD->getDefinition();
+  if (!D || !D->isCompleteDefinition())
     return FwdDecl;
 
   if (const CXXRecordDecl *CXXDecl = dyn_cast<CXXRecordDecl>(RD))
@@ -1586,16 +1580,20 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const RecordType *Ty) {
   LexicalBlockStack.pop_back();
   RegionMap.erase(Ty->getDecl());
 
-  llvm::DIArray Elements = DBuilder.getOrCreateArray(EltTys);
+  llvm::DINodeArray Elements = DBuilder.getOrCreateArray(EltTys);
   DBuilder.replaceArrays(FwdDecl, Elements);
 
+  if (FwdDecl->isTemporary())
+    FwdDecl =
+        llvm::MDNode::replaceWithPermanent(llvm::TempDICompositeType(FwdDecl));
+
   RegionMap[Ty->getDecl()].reset(FwdDecl);
   return FwdDecl;
 }
 
 /// CreateType - get objective-c object type.
-llvm::DIType CGDebugInfo::CreateType(const ObjCObjectType *Ty,
-                                     llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateType(const ObjCObjectType *Ty,
+                                      llvm::DIFile *Unit) {
   // Ignore protocols.
   return getOrCreateType(Ty->getBaseType(), Unit);
 }
@@ -1625,22 +1623,23 @@ static bool hasDefaultSetterName(const ObjCPropertyDecl *PD,
 }
 
 /// CreateType - get objective-c interface type.
-llvm::DIType CGDebugInfo::CreateType(const ObjCInterfaceType *Ty,
-                                     llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateType(const ObjCInterfaceType *Ty,
+                                      llvm::DIFile *Unit) {
   ObjCInterfaceDecl *ID = Ty->getDecl();
   if (!ID)
-    return llvm::DIType();
+    return nullptr;
 
   // Get overall information about the record type for the debug info.
-  llvm::DIFile DefUnit = getOrCreateFile(ID->getLocation());
+  llvm::DIFile *DefUnit = getOrCreateFile(ID->getLocation());
   unsigned Line = getLineNumber(ID->getLocation());
-  llvm::dwarf::SourceLanguage RuntimeLang = TheCU.getLanguage();
+  auto RuntimeLang =
+      static_cast<llvm::dwarf::SourceLanguage>(TheCU->getSourceLanguage());
 
   // If this is just a forward declaration return a special forward-declaration
   // debug type since we won't be able to lay out the entire type.
   ObjCInterfaceDecl *Def = ID->getDefinition();
   if (!Def || !Def->getImplementation()) {
-    llvm::DIType FwdDecl = DBuilder.createReplaceableForwardDecl(
+    llvm::DIType *FwdDecl = DBuilder.createReplaceableCompositeType(
         llvm::dwarf::DW_TAG_structure_type, ID->getName(), TheCU, DefUnit, Line,
         RuntimeLang);
     ObjCInterfaceCache.push_back(ObjCInterfaceCacheEntry(Ty, FwdDecl, Unit));
@@ -1650,12 +1649,12 @@ llvm::DIType CGDebugInfo::CreateType(const ObjCInterfaceType *Ty,
   return CreateTypeDefinition(Ty, Unit);
 }
 
-llvm::DIType CGDebugInfo::CreateTypeDefinition(const ObjCInterfaceType *Ty,
-                                               llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateTypeDefinition(const ObjCInterfaceType *Ty,
+                                                llvm::DIFile *Unit) {
   ObjCInterfaceDecl *ID = Ty->getDecl();
-  llvm::DIFile DefUnit = getOrCreateFile(ID->getLocation());
+  llvm::DIFile *DefUnit = getOrCreateFile(ID->getLocation());
   unsigned Line = getLineNumber(ID->getLocation());
-  unsigned RuntimeLang = TheCU.getLanguage();
+  unsigned RuntimeLang = TheCU->getSourceLanguage();
 
   // Bit size, align and offset of the type.
   uint64_t Size = CGM.getContext().getTypeSize(Ty);
@@ -1663,17 +1662,17 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const ObjCInterfaceType *Ty,
 
   unsigned Flags = 0;
   if (ID->getImplementation())
-    Flags |= llvm::DIDescriptor::FlagObjcClassComplete;
+    Flags |= llvm::DINode::FlagObjcClassComplete;
 
-  llvm::DICompositeType RealDecl = DBuilder.createStructType(
-      Unit, ID->getName(), DefUnit, Line, Size, Align, Flags, llvm::DIType(),
-      llvm::DIArray(), RuntimeLang);
+  llvm::DICompositeType *RealDecl = DBuilder.createStructType(
+      Unit, ID->getName(), DefUnit, Line, Size, Align, Flags, nullptr,
+      llvm::DINodeArray(), RuntimeLang);
 
   QualType QTy(Ty, 0);
   TypeCache[QTy.getAsOpaquePtr()].reset(RealDecl);
 
   // Push the struct on region stack.
-  LexicalBlockStack.emplace_back(static_cast<llvm::MDNode *>(RealDecl));
+  LexicalBlockStack.emplace_back(RealDecl);
   RegionMap[Ty->getDecl()].reset(RealDecl);
 
   // Convert all the elements.
@@ -1681,19 +1680,19 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const ObjCInterfaceType *Ty,
 
   ObjCInterfaceDecl *SClass = ID->getSuperClass();
   if (SClass) {
-    llvm::DIType SClassTy =
+    llvm::DIType *SClassTy =
         getOrCreateType(CGM.getContext().getObjCInterfaceType(SClass), Unit);
-    if (!SClassTy.isValid())
-      return llvm::DIType();
+    if (!SClassTy)
+      return nullptr;
 
-    llvm::DIType InhTag = DBuilder.createInheritance(RealDecl, SClassTy, 0, 0);
+    llvm::DIType *InhTag = DBuilder.createInheritance(RealDecl, SClassTy, 0, 0);
     EltTys.push_back(InhTag);
   }
 
   // Create entries for all of the properties.
   for (const auto *PD : ID->properties()) {
     SourceLocation Loc = PD->getLocation();
-    llvm::DIFile PUnit = getOrCreateFile(Loc);
+    llvm::DIFile *PUnit = getOrCreateFile(Loc);
     unsigned PLine = getLineNumber(Loc);
     ObjCMethodDecl *Getter = PD->getGetterMethodDecl();
     ObjCMethodDecl *Setter = PD->getSetterMethodDecl();
@@ -1711,9 +1710,9 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const ObjCInterfaceType *Ty,
   unsigned FieldNo = 0;
   for (ObjCIvarDecl *Field = ID->all_declared_ivar_begin(); Field;
        Field = Field->getNextIvar(), ++FieldNo) {
-    llvm::DIType FieldTy = getOrCreateType(Field->getType(), Unit);
-    if (!FieldTy.isValid())
-      return llvm::DIType();
+    llvm::DIType *FieldTy = getOrCreateType(Field->getType(), Unit);
+    if (!FieldTy)
+      return nullptr;
 
     StringRef FieldName = Field->getName();
 
@@ -1722,7 +1721,7 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const ObjCInterfaceType *Ty,
       continue;
 
     // Get the location for the field.
-    llvm::DIFile FieldDefUnit = getOrCreateFile(Field->getLocation());
+    llvm::DIFile *FieldDefUnit = getOrCreateFile(Field->getLocation());
     unsigned FieldLine = getLineNumber(Field->getLocation());
     QualType FType = Field->getType();
     uint64_t FieldSize = 0;
@@ -1755,11 +1754,11 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const ObjCInterfaceType *Ty,
 
     unsigned Flags = 0;
     if (Field->getAccessControl() == ObjCIvarDecl::Protected)
-      Flags = llvm::DIDescriptor::FlagProtected;
+      Flags = llvm::DINode::FlagProtected;
     else if (Field->getAccessControl() == ObjCIvarDecl::Private)
-      Flags = llvm::DIDescriptor::FlagPrivate;
+      Flags = llvm::DINode::FlagPrivate;
     else if (Field->getAccessControl() == ObjCIvarDecl::Public)
-      Flags = llvm::DIDescriptor::FlagPublic;
+      Flags = llvm::DINode::FlagPublic;
 
     llvm::MDNode *PropertyNode = nullptr;
     if (ObjCImplementationDecl *ImpD = ID->getImplementation()) {
@@ -1767,7 +1766,7 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const ObjCInterfaceType *Ty,
               ImpD->FindPropertyImplIvarDecl(Field->getIdentifier())) {
         if (ObjCPropertyDecl *PD = PImpD->getPropertyDecl()) {
           SourceLocation Loc = PD->getLocation();
-          llvm::DIFile PUnit = getOrCreateFile(Loc);
+          llvm::DIFile *PUnit = getOrCreateFile(Loc);
           unsigned PLine = getLineNumber(Loc);
           ObjCMethodDecl *Getter = PD->getGetterMethodDecl();
           ObjCMethodDecl *Setter = PD->getSetterMethodDecl();
@@ -1788,15 +1787,16 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const ObjCInterfaceType *Ty,
     EltTys.push_back(FieldTy);
   }
 
-  llvm::DIArray Elements = DBuilder.getOrCreateArray(EltTys);
+  llvm::DINodeArray Elements = DBuilder.getOrCreateArray(EltTys);
   DBuilder.replaceArrays(RealDecl, Elements);
 
   LexicalBlockStack.pop_back();
   return RealDecl;
 }
 
-llvm::DIType CGDebugInfo::CreateType(const VectorType *Ty, llvm::DIFile Unit) {
-  llvm::DIType ElementTy = getOrCreateType(Ty->getElementType(), Unit);
+llvm::DIType *CGDebugInfo::CreateType(const VectorType *Ty,
+                                      llvm::DIFile *Unit) {
+  llvm::DIType *ElementTy = getOrCreateType(Ty->getElementType(), Unit);
   int64_t Count = Ty->getNumElements();
   if (Count == 0)
     // If number of elements are not known then this is an unbounded array.
@@ -1804,7 +1804,7 @@ llvm::DIType CGDebugInfo::CreateType(const VectorType *Ty, llvm::DIFile Unit) {
     Count = -1;
 
   llvm::Metadata *Subscript = DBuilder.getOrCreateSubrange(0, Count);
-  llvm::DIArray SubscriptArray = DBuilder.getOrCreateArray(Subscript);
+  llvm::DINodeArray SubscriptArray = DBuilder.getOrCreateArray(Subscript);
 
   uint64_t Size = CGM.getContext().getTypeSize(Ty);
   uint64_t Align = CGM.getContext().getTypeAlign(Ty);
@@ -1812,7 +1812,7 @@ llvm::DIType CGDebugInfo::CreateType(const VectorType *Ty, llvm::DIFile Unit) {
   return DBuilder.createVectorType(Size, Align, ElementTy, SubscriptArray);
 }
 
-llvm::DIType CGDebugInfo::CreateType(const ArrayType *Ty, llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateType(const ArrayType *Ty, llvm::DIFile *Unit) {
   uint64_t Size;
   uint64_t Align;
 
@@ -1858,32 +1858,33 @@ llvm::DIType CGDebugInfo::CreateType(const ArrayType *Ty, llvm::DIFile Unit) {
     EltTy = Ty->getElementType();
   }
 
-  llvm::DIArray SubscriptArray = DBuilder.getOrCreateArray(Subscripts);
+  llvm::DINodeArray SubscriptArray = DBuilder.getOrCreateArray(Subscripts);
 
-  llvm::DIType DbgTy = DBuilder.createArrayType(
-      Size, Align, getOrCreateType(EltTy, Unit), SubscriptArray);
-  return DbgTy;
+  return DBuilder.createArrayType(Size, Align, getOrCreateType(EltTy, Unit),
+                                  SubscriptArray);
 }
 
-llvm::DIType CGDebugInfo::CreateType(const LValueReferenceType *Ty,
-                                     llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateType(const LValueReferenceType *Ty,
+                                      llvm::DIFile *Unit) {
   return CreatePointerLikeType(llvm::dwarf::DW_TAG_reference_type, Ty,
                                Ty->getPointeeType(), Unit);
 }
 
-llvm::DIType CGDebugInfo::CreateType(const RValueReferenceType *Ty,
-                                     llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateType(const RValueReferenceType *Ty,
+                                      llvm::DIFile *Unit) {
   return CreatePointerLikeType(llvm::dwarf::DW_TAG_rvalue_reference_type, Ty,
                                Ty->getPointeeType(), Unit);
 }
 
-llvm::DIType CGDebugInfo::CreateType(const MemberPointerType *Ty,
-                                     llvm::DIFile U) {
-  llvm::DIType ClassType = getOrCreateType(QualType(Ty->getClass(), 0), U);
-  if (!Ty->getPointeeType()->isFunctionType())
+llvm::DIType *CGDebugInfo::CreateType(const MemberPointerType *Ty,
+                                      llvm::DIFile *U) {
+  uint64_t Size = CGM.getCXXABI().isTypeInfoCalculable(QualType(Ty, 0))
+                      ? CGM.getContext().getTypeSize(Ty)
+                      : 0;
+  llvm::DIType *ClassType = getOrCreateType(QualType(Ty->getClass(), 0), U);
+  if (Ty->isMemberDataPointerType())
     return DBuilder.createMemberPointerType(
-      getOrCreateType(Ty->getPointeeType(), U), ClassType,
-      CGM.getContext().getTypeSize(Ty));
+        getOrCreateType(Ty->getPointeeType(), U), ClassType, Size);
 
   const FunctionProtoType *FPT =
       Ty->getPointeeType()->getAs<FunctionProtoType>();
@@ -1891,17 +1892,17 @@ llvm::DIType CGDebugInfo::CreateType(const MemberPointerType *Ty,
       getOrCreateInstanceMethodType(CGM.getContext().getPointerType(QualType(
                                         Ty->getClass(), FPT->getTypeQuals())),
                                     FPT, U),
-      ClassType, CGM.getContext().getTypeSize(Ty));
+      ClassType, Size);
 }
 
-llvm::DIType CGDebugInfo::CreateType(const AtomicType *Ty, llvm::DIFile U) {
+llvm::DIType *CGDebugInfo::CreateType(const AtomicType *Ty, llvm::DIFile *U) {
   // Ignore the atomic wrapping
   // FIXME: What is the correct representation?
   return getOrCreateType(Ty->getValueType(), U);
 }
 
 /// CreateEnumType - get enumeration type.
-llvm::DIType CGDebugInfo::CreateEnumType(const EnumType *Ty) {
+llvm::DIType *CGDebugInfo::CreateEnumType(const EnumType *Ty) {
   const EnumDecl *ED = Ty->getDecl();
   uint64_t Size = 0;
   uint64_t Align = 0;
@@ -1915,14 +1916,14 @@ llvm::DIType CGDebugInfo::CreateEnumType(const EnumType *Ty) {
   // If this is just a forward declaration, construct an appropriately
   // marked node and just return it.
   if (!ED->getDefinition()) {
-    llvm::DIDescriptor EDContext;
-    EDContext = getContextDescriptor(cast<Decl>(ED->getDeclContext()));
-    llvm::DIFile DefUnit = getOrCreateFile(ED->getLocation());
+    llvm::DIScope *EDContext =
+        getContextDescriptor(cast<Decl>(ED->getDeclContext()));
+    llvm::DIFile *DefUnit = getOrCreateFile(ED->getLocation());
     unsigned Line = getLineNumber(ED->getLocation());
     StringRef EDName = ED->getName();
-    llvm::DIType RetTy = DBuilder.createReplaceableForwardDecl(
+    llvm::DIType *RetTy = DBuilder.createReplaceableCompositeType(
         llvm::dwarf::DW_TAG_enumeration_type, EDName, EDContext, DefUnit, Line,
-        0, Size, Align, FullName);
+        0, Size, Align, llvm::DINode::FlagFwdDecl, FullName);
     ReplaceMap.emplace_back(
         std::piecewise_construct, std::make_tuple(Ty),
         std::make_tuple(static_cast<llvm::Metadata *>(RetTy)));
@@ -1932,7 +1933,7 @@ llvm::DIType CGDebugInfo::CreateEnumType(const EnumType *Ty) {
   return CreateTypeDefinition(Ty);
 }
 
-llvm::DIType CGDebugInfo::CreateTypeDefinition(const EnumType *Ty) {
+llvm::DIType *CGDebugInfo::CreateTypeDefinition(const EnumType *Ty) {
   const EnumDecl *ED = Ty->getDecl();
   uint64_t Size = 0;
   uint64_t Align = 0;
@@ -1943,7 +1944,7 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const EnumType *Ty) {
 
   SmallString<256> FullName = getUniqueTagTypeName(Ty, CGM, TheCU);
 
-  // Create DIEnumerator elements for each enumerator.
+  // Create elements for each enumerator.
   SmallVector<llvm::Metadata *, 16> Enumerators;
   ED = ED->getDefinition();
   for (const auto *Enum : ED->enumerators()) {
@@ -1952,19 +1953,17 @@ llvm::DIType CGDebugInfo::CreateTypeDefinition(const EnumType *Ty) {
   }
 
   // Return a CompositeType for the enum itself.
-  llvm::DIArray EltArray = DBuilder.getOrCreateArray(Enumerators);
+  llvm::DINodeArray EltArray = DBuilder.getOrCreateArray(Enumerators);
 
-  llvm::DIFile DefUnit = getOrCreateFile(ED->getLocation());
+  llvm::DIFile *DefUnit = getOrCreateFile(ED->getLocation());
   unsigned Line = getLineNumber(ED->getLocation());
-  llvm::DIDescriptor EnumContext =
+  llvm::DIScope *EnumContext =
       getContextDescriptor(cast<Decl>(ED->getDeclContext()));
-  llvm::DIType ClassTy = ED->isFixed()
-                             ? getOrCreateType(ED->getIntegerType(), DefUnit)
-                             : llvm::DIType();
-  llvm::DIType DbgTy =
-      DBuilder.createEnumerationType(EnumContext, ED->getName(), DefUnit, Line,
-                                     Size, Align, EltArray, ClassTy, FullName);
-  return DbgTy;
+  llvm::DIType *ClassTy =
+      ED->isFixed() ? getOrCreateType(ED->getIntegerType(), DefUnit) : nullptr;
+  return DBuilder.createEnumerationType(EnumContext, ED->getName(), DefUnit,
+                                        Line, Size, Align, EltArray, ClassTy,
+                                        FullName);
 }
 
 static QualType UnwrapTypeForDebugInfo(QualType T, const ASTContext &C) {
@@ -2024,7 +2023,7 @@ static QualType UnwrapTypeForDebugInfo(QualType T, const ASTContext &C) {
 
 /// getType - Get the type from the cache or return null type if it doesn't
 /// exist.
-llvm::DIType CGDebugInfo::getTypeOrNull(QualType Ty) {
+llvm::DIType *CGDebugInfo::getTypeOrNull(QualType Ty) {
 
   // Unwrap the type as needed for debug information.
   Ty = UnwrapTypeForDebugInfo(Ty, CGM.getContext());
@@ -2033,10 +2032,10 @@ llvm::DIType CGDebugInfo::getTypeOrNull(QualType Ty) {
   if (it != TypeCache.end()) {
     // Verify that the debug info still exists.
     if (llvm::Metadata *V = it->second)
-      return llvm::DIType(cast<llvm::MDNode>(V));
+      return cast<llvm::DIType>(V);
   }
 
-  return llvm::DIType();
+  return nullptr;
 }
 
 void CGDebugInfo::completeTemplateDefinition(
@@ -2052,18 +2051,18 @@ void CGDebugInfo::completeTemplateDefinition(
 
 /// getOrCreateType - Get the type from the cache or create a new
 /// one if necessary.
-llvm::DIType CGDebugInfo::getOrCreateType(QualType Ty, llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::getOrCreateType(QualType Ty, llvm::DIFile *Unit) {
   if (Ty.isNull())
-    return llvm::DIType();
+    return nullptr;
 
   // Unwrap the type as needed for debug information.
   Ty = UnwrapTypeForDebugInfo(Ty, CGM.getContext());
 
-  if (llvm::DIType T = getTypeOrNull(Ty))
+  if (auto *T = getTypeOrNull(Ty))
     return T;
 
   // Otherwise create the type.
-  llvm::DIType Res = CreateTypeNode(Ty, Unit);
+  llvm::DIType *Res = CreateTypeNode(Ty, Unit);
   void *TyPtr = Ty.getAsOpaquePtr();
 
   // And update the type cache.
@@ -2099,7 +2098,7 @@ ObjCInterfaceDecl *CGDebugInfo::getObjCInterfaceDecl(QualType Ty) {
 }
 
 /// CreateTypeNode - Create a new debug type node.
-llvm::DIType CGDebugInfo::CreateTypeNode(QualType Ty, llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::CreateTypeNode(QualType Ty, llvm::DIFile *Unit) {
   // Handle qualifiers, which recursively handles what they refer to.
   if (Ty.hasLocalQualifiers())
     return CreateQualifiedType(Ty, Unit);
@@ -2181,25 +2180,25 @@ llvm::DIType CGDebugInfo::CreateTypeNode(QualType Ty, llvm::DIFile Unit) {
 
 /// getOrCreateLimitedType - Get the type from the cache or create a new
 /// limited type if necessary.
-llvm::DIType CGDebugInfo::getOrCreateLimitedType(const RecordType *Ty,
-                                                 llvm::DIFile Unit) {
+llvm::DIType *CGDebugInfo::getOrCreateLimitedType(const RecordType *Ty,
+                                                  llvm::DIFile *Unit) {
   QualType QTy(Ty, 0);
 
-  llvm::DICompositeType T(getTypeOrNull(QTy));
+  auto *T = cast_or_null<llvm::DICompositeTypeBase>(getTypeOrNull(QTy));
 
   // We may have cached a forward decl when we could have created
   // a non-forward decl. Go ahead and create a non-forward decl
   // now.
-  if (T && !T.isForwardDecl())
+  if (T && !T->isForwardDecl())
     return T;
 
   // Otherwise create the type.
-  llvm::DICompositeType Res = CreateLimitedType(Ty);
+  llvm::DICompositeType *Res = CreateLimitedType(Ty);
 
   // Propagate members from the declaration to the definition
   // CreateType(const RecordType*) will overwrite this with the members in the
   // correct order if the full type is needed.
-  DBuilder.replaceArrays(Res, T.getElements());
+  DBuilder.replaceArrays(Res, T ? T->getElements() : llvm::DINodeArray());
 
   // And update the type cache.
   TypeCache[QTy.getAsOpaquePtr()].reset(Res);
@@ -2207,21 +2206,22 @@ llvm::DIType CGDebugInfo::getOrCreateLimitedType(const RecordType *Ty,
 }
 
 // TODO: Currently used for context chains when limiting debug info.
-llvm::DICompositeType CGDebugInfo::CreateLimitedType(const RecordType *Ty) {
+llvm::DICompositeType *CGDebugInfo::CreateLimitedType(const RecordType *Ty) {
   RecordDecl *RD = Ty->getDecl();
 
   // Get overall information about the record type for the debug info.
-  llvm::DIFile DefUnit = getOrCreateFile(RD->getLocation());
+  llvm::DIFile *DefUnit = getOrCreateFile(RD->getLocation());
   unsigned Line = getLineNumber(RD->getLocation());
   StringRef RDName = getClassName(RD);
 
-  llvm::DIDescriptor RDContext =
+  llvm::DIScope *RDContext =
       getContextDescriptor(cast<Decl>(RD->getDeclContext()));
 
   // If we ended up creating the type during the context chain construction,
   // just return that.
-  llvm::DICompositeType T(getTypeOrNull(CGM.getContext().getRecordType(RD)));
-  if (T && (!T.isForwardDecl() || !RD->getDefinition()))
+  auto *T = cast_or_null<llvm::DICompositeType>(
+      getTypeOrNull(CGM.getContext().getRecordType(RD)));
+  if (T && (!T->isForwardDecl() || !RD->getDefinition()))
     return T;
 
   // If this is just a forward or incomplete declaration, construct an
@@ -2232,38 +2232,27 @@ llvm::DICompositeType CGDebugInfo::CreateLimitedType(const RecordType *Ty) {
 
   uint64_t Size = CGM.getContext().getTypeSize(Ty);
   uint64_t Align = CGM.getContext().getTypeAlign(Ty);
-  llvm::DICompositeType RealDecl;
 
   SmallString<256> FullName = getUniqueTagTypeName(Ty, CGM, TheCU);
 
-  if (RD->isUnion())
-    RealDecl = DBuilder.createUnionType(RDContext, RDName, DefUnit, Line, Size,
-                                        Align, 0, llvm::DIArray(), 0, FullName);
-  else if (RD->isClass()) {
-    // FIXME: This could be a struct type giving a default visibility different
-    // than C++ class type, but needs llvm metadata changes first.
-    RealDecl = DBuilder.createClassType(
-        RDContext, RDName, DefUnit, Line, Size, Align, 0, 0, llvm::DIType(),
-        llvm::DIArray(), llvm::DIType(), llvm::DIArray(), FullName);
-  } else
-    RealDecl = DBuilder.createStructType(
-        RDContext, RDName, DefUnit, Line, Size, Align, 0, llvm::DIType(),
-        llvm::DIArray(), 0, llvm::DIType(), FullName);
+  llvm::DICompositeType *RealDecl = DBuilder.createReplaceableCompositeType(
+      getTagForRecord(RD), RDName, RDContext, DefUnit, Line, 0, Size, Align, 0,
+      FullName);
 
   RegionMap[Ty->getDecl()].reset(RealDecl);
   TypeCache[QualType(Ty, 0).getAsOpaquePtr()].reset(RealDecl);
 
   if (const ClassTemplateSpecializationDecl *TSpecial =
           dyn_cast<ClassTemplateSpecializationDecl>(RD))
-    DBuilder.replaceArrays(RealDecl, llvm::DIArray(),
+    DBuilder.replaceArrays(RealDecl, llvm::DINodeArray(),
                            CollectCXXTemplateParams(TSpecial, DefUnit));
   return RealDecl;
 }
 
 void CGDebugInfo::CollectContainingType(const CXXRecordDecl *RD,
-                                        llvm::DICompositeType RealDecl) {
+                                        llvm::DICompositeType *RealDecl) {
   // A class's primary base or the class itself contains the vtable.
-  llvm::DICompositeType ContainingType;
+  llvm::DICompositeType *ContainingType = nullptr;
   const ASTRecordLayout &RL = CGM.getContext().getASTRecordLayout(RD);
   if (const CXXRecordDecl *PBase = RL.getPrimaryBase()) {
     // Seek non-virtual primary base root.
@@ -2275,7 +2264,7 @@ void CGDebugInfo::CollectContainingType(const CXXRecordDecl *RD,
       else
         break;
     }
-    ContainingType = llvm::DICompositeType(
+    ContainingType = cast<llvm::DICompositeType>(
         getOrCreateType(QualType(PBase->getTypeForDecl(), 0),
                         getOrCreateFile(RD->getLocation())));
   } else if (RD->isDynamicClass())
@@ -2285,29 +2274,29 @@ void CGDebugInfo::CollectContainingType(const CXXRecordDecl *RD,
 }
 
 /// CreateMemberType - Create new member and increase Offset by FType's size.
-llvm::DIType CGDebugInfo::CreateMemberType(llvm::DIFile Unit, QualType FType,
-                                           StringRef Name, uint64_t *Offset) {
-  llvm::DIType FieldTy = CGDebugInfo::getOrCreateType(FType, Unit);
+llvm::DIType *CGDebugInfo::CreateMemberType(llvm::DIFile *Unit, QualType FType,
+                                            StringRef Name, uint64_t *Offset) {
+  llvm::DIType *FieldTy = CGDebugInfo::getOrCreateType(FType, Unit);
   uint64_t FieldSize = CGM.getContext().getTypeSize(FType);
   unsigned FieldAlign = CGM.getContext().getTypeAlign(FType);
-  llvm::DIType Ty = DBuilder.createMemberType(Unit, Name, Unit, 0, FieldSize,
-                                              FieldAlign, *Offset, 0, FieldTy);
+  llvm::DIType *Ty = DBuilder.createMemberType(Unit, Name, Unit, 0, FieldSize,
+                                               FieldAlign, *Offset, 0, FieldTy);
   *Offset += FieldSize;
   return Ty;
 }
 
-void CGDebugInfo::collectFunctionDeclProps(GlobalDecl GD,
-                                           llvm::DIFile Unit,
-                                           StringRef &Name, StringRef &LinkageName,
-                                           llvm::DIDescriptor &FDContext,
-                                           llvm::DIArray &TParamsArray,
+void CGDebugInfo::collectFunctionDeclProps(GlobalDecl GD, llvm::DIFile *Unit,
+                                           StringRef &Name,
+                                           StringRef &LinkageName,
+                                           llvm::DIScope *&FDContext,
+                                           llvm::DINodeArray &TParamsArray,
                                            unsigned &Flags) {
   const FunctionDecl *FD = cast<FunctionDecl>(GD.getDecl());
   Name = getFunctionName(FD);
   // Use mangled name as linkage name for C/C++ functions.
   if (FD->hasPrototype()) {
     LinkageName = CGM.getMangledName(GD);
-    Flags |= llvm::DIDescriptor::FlagPrototyped;
+    Flags |= llvm::DINode::FlagPrototyped;
   }
   // No need to replicate the linkage name if it isn't different from the
   // subprogram name, no need to have it at all unless coverage is enabled or
@@ -2330,10 +2319,10 @@ void CGDebugInfo::collectFunctionDeclProps(GlobalDecl GD,
   }
 }
 
-void CGDebugInfo::collectVarDeclProps(const VarDecl *VD, llvm::DIFile &Unit,
+void CGDebugInfo::collectVarDeclProps(const VarDecl *VD, llvm::DIFile *&Unit,
                                       unsigned &LineNo, QualType &T,
                                       StringRef &Name, StringRef &LinkageName,
-                                      llvm::DIDescriptor &VDContext) {
+                                      llvm::DIScope *&VDContext) {
   Unit = getOrCreateFile(VD->getLocation());
   LineNo = getLineNumber(VD->getLocation());
 
@@ -2362,19 +2351,27 @@ void CGDebugInfo::collectVarDeclProps(const VarDecl *VD, llvm::DIFile &Unit,
   // FIXME: Generalize this for even non-member global variables where the
   // declaration and definition may have different lexical decl contexts, once
   // we have support for emitting declarations of (non-member) global variables.
-  VDContext = getContextDescriptor(
-      dyn_cast<Decl>(VD->isStaticDataMember() ? VD->getLexicalDeclContext()
-                                              : VD->getDeclContext()));
-}
-
-llvm::DISubprogram
+  const DeclContext *DC = VD->isStaticDataMember() ? VD->getLexicalDeclContext()
+                                                   : VD->getDeclContext();
+  // When a record type contains an in-line initialization of a static data
+  // member, and the record type is marked as __declspec(dllexport), an implicit
+  // definition of the member will be created in the record context.  DWARF
+  // doesn't seem to have a nice way to describe this in a form that consumers
+  // are likely to understand, so fake the "normal" situation of a definition
+  // outside the class by putting it in the global scope.
+  if (DC->isRecord())
+    DC = CGM.getContext().getTranslationUnitDecl();
+  VDContext = getContextDescriptor(dyn_cast<Decl>(DC));
+}
+
+llvm::DISubprogram *
 CGDebugInfo::getFunctionForwardDeclaration(const FunctionDecl *FD) {
-  llvm::DIArray TParamsArray;
+  llvm::DINodeArray TParamsArray;
   StringRef Name, LinkageName;
   unsigned Flags = 0;
   SourceLocation Loc = FD->getLocation();
-  llvm::DIFile Unit = getOrCreateFile(Loc);
-  llvm::DIDescriptor DContext(Unit);
+  llvm::DIFile *Unit = getOrCreateFile(Loc);
+  llvm::DIScope *DContext = Unit;
   unsigned Line = getLineNumber(Loc);
 
   collectFunctionDeclProps(FD, Unit, Name, LinkageName, DContext,
@@ -2386,35 +2383,31 @@ CGDebugInfo::getFunctionForwardDeclaration(const FunctionDecl *FD) {
   QualType FnType =
     CGM.getContext().getFunctionType(FD->getReturnType(), ArgTypes,
                                      FunctionProtoType::ExtProtoInfo());
-  llvm::DISubprogram SP =
-    DBuilder.createTempFunctionFwdDecl(DContext, Name, LinkageName, Unit, Line,
-                                       getOrCreateFunctionType(FD, FnType, Unit),
-                                       !FD->isExternallyVisible(),
-                                       false /*declaration*/, 0, Flags,
-                                       CGM.getLangOpts().Optimize, nullptr,
-                                       TParamsArray, getFunctionDeclaration(FD));
+  llvm::DISubprogram *SP = DBuilder.createTempFunctionFwdDecl(
+      DContext, Name, LinkageName, Unit, Line,
+      getOrCreateFunctionType(FD, FnType, Unit), !FD->isExternallyVisible(),
+      false /*declaration*/, 0, Flags, CGM.getLangOpts().Optimize, nullptr,
+      TParamsArray.get(), getFunctionDeclaration(FD));
   const FunctionDecl *CanonDecl = cast<FunctionDecl>(FD->getCanonicalDecl());
-  FwdDeclReplaceMap.emplace_back(
-      std::piecewise_construct, std::make_tuple(CanonDecl),
-      std::make_tuple(static_cast<llvm::Metadata *>(SP)));
+  FwdDeclReplaceMap.emplace_back(std::piecewise_construct,
+                                 std::make_tuple(CanonDecl),
+                                 std::make_tuple(SP));
   return SP;
 }
 
-llvm::DIGlobalVariable
+llvm::DIGlobalVariable *
 CGDebugInfo::getGlobalVariableForwardDeclaration(const VarDecl *VD) {
   QualType T;
   StringRef Name, LinkageName;
   SourceLocation Loc = VD->getLocation();
-  llvm::DIFile Unit = getOrCreateFile(Loc);
-  llvm::DIDescriptor DContext(Unit);
+  llvm::DIFile *Unit = getOrCreateFile(Loc);
+  llvm::DIScope *DContext = Unit;
   unsigned Line = getLineNumber(Loc);
 
   collectVarDeclProps(VD, Unit, Line, T, Name, LinkageName, DContext);
-  llvm::DIGlobalVariable GV =
-    DBuilder.createTempGlobalVariableFwdDecl(DContext, Name, LinkageName, Unit,
-                                             Line, getOrCreateType(T, Unit),
-                                             !VD->isExternallyVisible(),
-                                             nullptr, nullptr);
+  auto *GV = DBuilder.createTempGlobalVariableFwdDecl(
+      DContext, Name, LinkageName, Unit, Line, getOrCreateType(T, Unit),
+      !VD->isExternallyVisible(), nullptr, nullptr);
   FwdDeclReplaceMap.emplace_back(
       std::piecewise_construct,
       std::make_tuple(cast<VarDecl>(VD->getCanonicalDecl())),
@@ -2422,7 +2415,7 @@ CGDebugInfo::getGlobalVariableForwardDeclaration(const VarDecl *VD) {
   return GV;
 }
 
-llvm::DIDescriptor CGDebugInfo::getDeclarationOrDefinition(const Decl *D) {
+llvm::DINode *CGDebugInfo::getDeclarationOrDefinition(const Decl *D) {
   // We only need a declaration (not a definition) of the type - so use whatever
   // we would otherwise do to get a type for a pointee. (forward declarations in
   // limited debug info, full definitions (if the type definition is available)
@@ -2433,7 +2426,7 @@ llvm::DIDescriptor CGDebugInfo::getDeclarationOrDefinition(const Decl *D) {
   auto I = DeclCache.find(D->getCanonicalDecl());
 
   if (I != DeclCache.end())
-    return llvm::DIDescriptor(dyn_cast_or_null<llvm::MDNode>(I->second));
+    return dyn_cast_or_null<llvm::DINode>(I->second);
 
   // No definition for now. Emit a forward definition that might be
   // merged with a potential upcoming definition.
@@ -2442,59 +2435,55 @@ llvm::DIDescriptor CGDebugInfo::getDeclarationOrDefinition(const Decl *D) {
   else if (const auto *VD = dyn_cast<VarDecl>(D))
     return getGlobalVariableForwardDeclaration(VD);
 
-  return llvm::DIDescriptor();
+  return nullptr;
 }
 
 /// getFunctionDeclaration - Return debug info descriptor to describe method
 /// declaration for the given method definition.
-llvm::DISubprogram CGDebugInfo::getFunctionDeclaration(const Decl *D) {
+llvm::DISubprogram *CGDebugInfo::getFunctionDeclaration(const Decl *D) {
   if (!D || DebugKind <= CodeGenOptions::DebugLineTablesOnly)
-    return llvm::DISubprogram();
+    return nullptr;
 
   const FunctionDecl *FD = dyn_cast<FunctionDecl>(D);
   if (!FD)
-    return llvm::DISubprogram();
+    return nullptr;
 
   // Setup context.
-  llvm::DIScope S = getContextDescriptor(cast<Decl>(D->getDeclContext()));
+  auto *S = getContextDescriptor(cast<Decl>(D->getDeclContext()));
 
   auto MI = SPCache.find(FD->getCanonicalDecl());
   if (MI == SPCache.end()) {
     if (const CXXMethodDecl *MD =
             dyn_cast<CXXMethodDecl>(FD->getCanonicalDecl())) {
-      llvm::DICompositeType T(S);
-      llvm::DISubprogram SP =
-          CreateCXXMemberFunction(MD, getOrCreateFile(MD->getLocation()), T);
-      return SP;
+      return CreateCXXMemberFunction(MD, getOrCreateFile(MD->getLocation()),
+                                     cast<llvm::DICompositeType>(S));
     }
   }
   if (MI != SPCache.end()) {
-    llvm::DISubprogram SP(dyn_cast_or_null<llvm::MDNode>(MI->second));
-    if (SP.isSubprogram() && !SP.isDefinition())
+    auto *SP = dyn_cast_or_null<llvm::DISubprogram>(MI->second);
+    if (SP && !SP->isDefinition())
       return SP;
   }
 
   for (auto NextFD : FD->redecls()) {
     auto MI = SPCache.find(NextFD->getCanonicalDecl());
     if (MI != SPCache.end()) {
-      llvm::DISubprogram SP(dyn_cast_or_null<llvm::MDNode>(MI->second));
-      if (SP.isSubprogram() && !SP.isDefinition())
+      auto *SP = dyn_cast_or_null<llvm::DISubprogram>(MI->second);
+      if (SP && !SP->isDefinition())
         return SP;
     }
   }
-  return llvm::DISubprogram();
+  return nullptr;
 }
 
-// getOrCreateFunctionType - Construct DIType. If it is a c++ method, include
+// getOrCreateFunctionType - Construct type. If it is a c++ method, include
 // implicit parameter "this".
-llvm::DICompositeType CGDebugInfo::getOrCreateFunctionType(const Decl *D,
-                                                           QualType FnType,
-                                                           llvm::DIFile F) {
+llvm::DISubroutineType *CGDebugInfo::getOrCreateFunctionType(const Decl *D,
+                                                             QualType FnType,
+                                                             llvm::DIFile *F) {
   if (!D || DebugKind <= CodeGenOptions::DebugLineTablesOnly)
-    // Create fake but valid subroutine type. Otherwise
-    // llvm::DISubprogram::Verify() would return false, and
-    // subprogram DIE will miss DW_AT_decl_file and
-    // DW_AT_decl_line fields.
+    // Create fake but valid subroutine type. Otherwise -verify would fail, and
+    // subprogram DIE will miss DW_AT_decl_file and DW_AT_decl_line fields.
     return DBuilder.createSubroutineType(F,
                                          DBuilder.getOrCreateTypeArray(None));
 
@@ -2515,11 +2504,10 @@ llvm::DICompositeType CGDebugInfo::getOrCreateFunctionType(const Decl *D,
     Elts.push_back(getOrCreateType(ResultTy, F));
     // "self" pointer is always first argument.
     QualType SelfDeclTy = OMethod->getSelfDecl()->getType();
-    llvm::DIType SelfTy = getOrCreateType(SelfDeclTy, F);
-    Elts.push_back(CreateSelfType(SelfDeclTy, SelfTy));
+    Elts.push_back(CreateSelfType(SelfDeclTy, getOrCreateType(SelfDeclTy, F)));
     // "_cmd" pointer is always second argument.
-    llvm::DIType CmdTy = getOrCreateType(OMethod->getCmdDecl()->getType(), F);
-    Elts.push_back(DBuilder.createArtificialType(CmdTy));
+    Elts.push_back(DBuilder.createArtificialType(
+        getOrCreateType(OMethod->getCmdDecl()->getType(), F)));
     // Get rest of the arguments.
     for (const auto *PI : OMethod->params())
       Elts.push_back(getOrCreateType(PI->getType(), F));
@@ -2527,7 +2515,7 @@ llvm::DICompositeType CGDebugInfo::getOrCreateFunctionType(const Decl *D,
     if (OMethod->isVariadic())
       Elts.push_back(DBuilder.createUnspecifiedParameter());
 
-    llvm::DITypeArray EltTypeArray = DBuilder.getOrCreateTypeArray(Elts);
+    llvm::DITypeRefArray EltTypeArray = DBuilder.getOrCreateTypeArray(Elts);
     return DBuilder.createSubroutineType(F, EltTypeArray);
   }
 
@@ -2541,11 +2529,11 @@ llvm::DICompositeType CGDebugInfo::getOrCreateFunctionType(const Decl *D,
         for (unsigned i = 0, e = FPT->getNumParams(); i != e; ++i)
           EltTys.push_back(getOrCreateType(FPT->getParamType(i), F));
       EltTys.push_back(DBuilder.createUnspecifiedParameter());
-      llvm::DITypeArray EltTypeArray = DBuilder.getOrCreateTypeArray(EltTys);
+      llvm::DITypeRefArray EltTypeArray = DBuilder.getOrCreateTypeArray(EltTys);
       return DBuilder.createSubroutineType(F, EltTypeArray);
     }
 
-  return llvm::DICompositeType(getOrCreateType(FnType, F));
+  return cast<llvm::DISubroutineType>(getOrCreateType(FnType, F));
 }
 
 /// EmitFunctionStart - Constructs the debug code for entering a function.
@@ -2562,20 +2550,19 @@ void CGDebugInfo::EmitFunctionStart(GlobalDecl GD, SourceLocation Loc,
   bool HasDecl = (D != nullptr);
 
   unsigned Flags = 0;
-  llvm::DIFile Unit = getOrCreateFile(Loc);
-  llvm::DIDescriptor FDContext(Unit);
-  llvm::DIArray TParamsArray;
+  llvm::DIFile *Unit = getOrCreateFile(Loc);
+  llvm::DIScope *FDContext = Unit;
+  llvm::DINodeArray TParamsArray;
   if (!HasDecl) {
     // Use llvm function name.
     LinkageName = Fn->getName();
   } else if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) {
-    // If there is a DISubprogram for this function available then use it.
+    // If there is a subprogram for this function available then use it.
     auto FI = SPCache.find(FD->getCanonicalDecl());
     if (FI != SPCache.end()) {
-      llvm::DIDescriptor SP(dyn_cast_or_null<llvm::MDNode>(FI->second));
-      if (SP.isSubprogram() && llvm::DISubprogram(SP).isDefinition()) {
-        llvm::MDNode *SPN = SP;
-        LexicalBlockStack.emplace_back(SPN);
+      auto *SP = dyn_cast_or_null<llvm::DISubprogram>(FI->second);
+      if (SP && SP->isDefinition()) {
+        LexicalBlockStack.emplace_back(SP);
         RegionMap[D].reset(SP);
         return;
       }
@@ -2584,17 +2571,17 @@ void CGDebugInfo::EmitFunctionStart(GlobalDecl GD, SourceLocation Loc,
                              TParamsArray, Flags);
   } else if (const ObjCMethodDecl *OMD = dyn_cast<ObjCMethodDecl>(D)) {
     Name = getObjCMethodName(OMD);
-    Flags |= llvm::DIDescriptor::FlagPrototyped;
+    Flags |= llvm::DINode::FlagPrototyped;
   } else {
     // Use llvm function name.
     Name = Fn->getName();
-    Flags |= llvm::DIDescriptor::FlagPrototyped;
+    Flags |= llvm::DINode::FlagPrototyped;
   }
   if (!Name.empty() && Name[0] == '\01')
     Name = Name.substr(1);
 
   if (!HasDecl || D->isImplicit()) {
-    Flags |= llvm::DIDescriptor::FlagArtificial;
+    Flags |= llvm::DINode::FlagArtificial;
     // Artificial functions without a location should not silently reuse CurLoc.
     if (Loc.isInvalid())
       CurLoc = SourceLocation();
@@ -2607,11 +2594,11 @@ void CGDebugInfo::EmitFunctionStart(GlobalDecl GD, SourceLocation Loc,
   // FunctionDecls. When/if we fix this we can have FDContext be TheCU/null for
   // all subprograms instead of the actual context since subprogram definitions
   // are emitted as CU level entities by the backend.
-  llvm::DISubprogram SP = DBuilder.createFunction(
+  llvm::DISubprogram *SP = DBuilder.createFunction(
       FDContext, Name, LinkageName, Unit, LineNo,
       getOrCreateFunctionType(D, FnType, Unit), Fn->hasInternalLinkage(),
       true /*definition*/, ScopeLine, Flags, CGM.getLangOpts().Optimize, Fn,
-      TParamsArray, getFunctionDeclaration(D));
+      TParamsArray.get(), getFunctionDeclaration(D));
   // We might get here with a VarDecl in the case we're generating
   // code for the initialization of globals. Do not record these decls
   // as they will overwrite the actual VarDecl Decl in the cache.
@@ -2619,8 +2606,7 @@ void CGDebugInfo::EmitFunctionStart(GlobalDecl GD, SourceLocation Loc,
     DeclCache[D->getCanonicalDecl()].reset(static_cast<llvm::Metadata *>(SP));
 
   // Push the function onto the lexical block stack.
-  llvm::MDNode *SPN = SP;
-  LexicalBlockStack.emplace_back(SPN);
+  LexicalBlockStack.emplace_back(SP);
 
   if (HasDecl)
     RegionMap[D].reset(SP);
@@ -2629,8 +2615,7 @@ void CGDebugInfo::EmitFunctionStart(GlobalDecl GD, SourceLocation Loc,
 /// EmitLocation - Emit metadata to indicate a change in line/column
 /// information in the source file. If the location is invalid, the
 /// previous location will be reused.
-void CGDebugInfo::EmitLocation(CGBuilderTy &Builder, SourceLocation Loc,
-                               bool ForceColumnInfo) {
+void CGDebugInfo::EmitLocation(CGBuilderTy &Builder, SourceLocation Loc) {
   // Update our current location
   setLocation(Loc);
 
@@ -2639,7 +2624,7 @@ void CGDebugInfo::EmitLocation(CGBuilderTy &Builder, SourceLocation Loc,
 
   llvm::MDNode *Scope = LexicalBlockStack.back();
   Builder.SetCurrentDebugLocation(llvm::DebugLoc::get(
-      getLineNumber(CurLoc), getColumnNumber(CurLoc, ForceColumnInfo), Scope));
+      getLineNumber(CurLoc), getColumnNumber(CurLoc), Scope));
 }
 
 /// CreateLexicalBlock - Creates a new lexical block node and pushes it on
@@ -2648,11 +2633,9 @@ void CGDebugInfo::CreateLexicalBlock(SourceLocation Loc) {
   llvm::MDNode *Back = nullptr;
   if (!LexicalBlockStack.empty())
     Back = LexicalBlockStack.back().get();
-  llvm::DIDescriptor D = DBuilder.createLexicalBlock(
-      llvm::DIDescriptor(Back), getOrCreateFile(CurLoc), getLineNumber(CurLoc),
-      getColumnNumber(CurLoc));
-  llvm::MDNode *DN = D;
-  LexicalBlockStack.emplace_back(DN);
+  LexicalBlockStack.emplace_back(DBuilder.createLexicalBlock(
+      cast<llvm::DIScope>(Back), getOrCreateFile(CurLoc), getLineNumber(CurLoc),
+      getColumnNumber(CurLoc)));
 }
 
 /// EmitLexicalBlockStart - Constructs the debug code for entering a declarative
@@ -2705,15 +2688,15 @@ void CGDebugInfo::EmitFunctionEnd(CGBuilderTy &Builder) {
 
 // EmitTypeForVarWithBlocksAttr - Build up structure info for the byref.
 // See BuildByRefType.
-llvm::DIType CGDebugInfo::EmitTypeForVarWithBlocksAttr(const VarDecl *VD,
-                                                       uint64_t *XOffset) {
+llvm::DIType *CGDebugInfo::EmitTypeForVarWithBlocksAttr(const VarDecl *VD,
+                                                        uint64_t *XOffset) {
 
   SmallVector<llvm::Metadata *, 5> EltTys;
   QualType FType;
   uint64_t FieldSize, FieldOffset;
   unsigned FieldAlign;
 
-  llvm::DIFile Unit = getOrCreateFile(VD->getLocation());
+  llvm::DIFile *Unit = getOrCreateFile(VD->getLocation());
   QualType Type = VD->getType();
 
   FieldOffset = 0;
@@ -2760,7 +2743,7 @@ llvm::DIType CGDebugInfo::EmitTypeForVarWithBlocksAttr(const VarDecl *VD,
   }
 
   FType = Type;
-  llvm::DIType FieldTy = getOrCreateType(FType, Unit);
+  llvm::DIType *FieldTy = getOrCreateType(FType, Unit);
   FieldSize = CGM.getContext().getTypeSize(FType);
   FieldAlign = CGM.getContext().toBits(Align);
 
@@ -2770,16 +2753,16 @@ llvm::DIType CGDebugInfo::EmitTypeForVarWithBlocksAttr(const VarDecl *VD,
   EltTys.push_back(FieldTy);
   FieldOffset += FieldSize;
 
-  llvm::DIArray Elements = DBuilder.getOrCreateArray(EltTys);
+  llvm::DINodeArray Elements = DBuilder.getOrCreateArray(EltTys);
 
-  unsigned Flags = llvm::DIDescriptor::FlagBlockByrefStruct;
+  unsigned Flags = llvm::DINode::FlagBlockByrefStruct;
 
   return DBuilder.createStructType(Unit, "", Unit, 0, FieldOffset, 0, Flags,
-                                   llvm::DIType(), Elements);
+                                   nullptr, Elements);
 }
 
 /// EmitDeclare - Emit local variable declaration debug info.
-void CGDebugInfo::EmitDeclare(const VarDecl *VD, llvm::dwarf::LLVMConstants Tag,
+void CGDebugInfo::EmitDeclare(const VarDecl *VD, llvm::dwarf::Tag Tag,
                               llvm::Value *Storage, unsigned ArgNo,
                               CGBuilderTy &Builder) {
   assert(DebugKind >= CodeGenOptions::LimitedDebugInfo);
@@ -2788,10 +2771,10 @@ void CGDebugInfo::EmitDeclare(const VarDecl *VD, llvm::dwarf::LLVMConstants Tag,
   bool Unwritten =
       VD->isImplicit() || (isa<Decl>(VD->getDeclContext()) &&
                            cast<Decl>(VD->getDeclContext())->isImplicit());
-  llvm::DIFile Unit;
+  llvm::DIFile *Unit = nullptr;
   if (!Unwritten)
     Unit = getOrCreateFile(VD->getLocation());
-  llvm::DIType Ty;
+  llvm::DIType *Ty;
   uint64_t XOffset = 0;
   if (VD->hasAttr<BlocksAttr>())
     Ty = EmitTypeForVarWithBlocksAttr(VD, &XOffset);
@@ -2810,58 +2793,64 @@ void CGDebugInfo::EmitDeclare(const VarDecl *VD, llvm::dwarf::LLVMConstants Tag,
     Line = getLineNumber(VD->getLocation());
     Column = getColumnNumber(VD->getLocation());
   }
+  SmallVector<int64_t, 9> Expr;
   unsigned Flags = 0;
   if (VD->isImplicit())
-    Flags |= llvm::DIDescriptor::FlagArtificial;
+    Flags |= llvm::DINode::FlagArtificial;
   // If this is the first argument and it is implicit then
   // give it an object pointer flag.
   // FIXME: There has to be a better way to do this, but for static
   // functions there won't be an implicit param at arg1 and
   // otherwise it is 'self' or 'this'.
   if (isa<ImplicitParamDecl>(VD) && ArgNo == 1)
-    Flags |= llvm::DIDescriptor::FlagObjectPointer;
+    Flags |= llvm::DINode::FlagObjectPointer;
   if (llvm::Argument *Arg = dyn_cast<llvm::Argument>(Storage))
     if (Arg->getType()->isPointerTy() && !Arg->hasByValAttr() &&
         !VD->getType()->isPointerType())
-      Flags |= llvm::DIDescriptor::FlagIndirectVariable;
+      Expr.push_back(llvm::dwarf::DW_OP_deref);
 
-  llvm::MDNode *Scope = LexicalBlockStack.back();
+  auto *Scope = cast<llvm::DIScope>(LexicalBlockStack.back());
 
   StringRef Name = VD->getName();
   if (!Name.empty()) {
     if (VD->hasAttr<BlocksAttr>()) {
       CharUnits offset = CharUnits::fromQuantity(32);
-      SmallVector<int64_t, 9> addr;
-      addr.push_back(llvm::dwarf::DW_OP_plus);
+      Expr.push_back(llvm::dwarf::DW_OP_plus);
       // offset of __forwarding field
       offset = CGM.getContext().toCharUnitsFromBits(
           CGM.getTarget().getPointerWidth(0));
-      addr.push_back(offset.getQuantity());
-      addr.push_back(llvm::dwarf::DW_OP_deref);
-      addr.push_back(llvm::dwarf::DW_OP_plus);
+      Expr.push_back(offset.getQuantity());
+      Expr.push_back(llvm::dwarf::DW_OP_deref);
+      Expr.push_back(llvm::dwarf::DW_OP_plus);
       // offset of x field
       offset = CGM.getContext().toCharUnitsFromBits(XOffset);
-      addr.push_back(offset.getQuantity());
+      Expr.push_back(offset.getQuantity());
 
       // Create the descriptor for the variable.
-      llvm::DIVariable D = DBuilder.createLocalVariable(
-          Tag, llvm::DIDescriptor(Scope), VD->getName(), Unit, Line, Ty, ArgNo);
+      auto *D = DBuilder.createLocalVariable(Tag, Scope, VD->getName(), Unit,
+                                             Line, Ty, ArgNo);
 
       // Insert an llvm.dbg.declare into the current block.
-      llvm::Instruction *Call =
-          DBuilder.insertDeclare(Storage, D, DBuilder.createExpression(addr),
-                                 Builder.GetInsertBlock());
-      Call->setDebugLoc(llvm::DebugLoc::get(Line, Column, Scope));
+      DBuilder.insertDeclare(Storage, D, DBuilder.createExpression(Expr),
+                             llvm::DebugLoc::get(Line, Column, Scope),
+                             Builder.GetInsertBlock());
       return;
     } else if (isa<VariableArrayType>(VD->getType()))
-      Flags |= llvm::DIDescriptor::FlagIndirectVariable;
+      Expr.push_back(llvm::dwarf::DW_OP_deref);
   } else if (const RecordType *RT = dyn_cast<RecordType>(VD->getType())) {
     // If VD is an anonymous union then Storage represents value for
     // all union fields.
     const RecordDecl *RD = cast<RecordDecl>(RT->getDecl());
     if (RD->isUnion() && RD->isAnonymousStructOrUnion()) {
+      // GDB has trouble finding local variables in anonymous unions, so we emit
+      // artifical local variables for each of the members.
+      //
+      // FIXME: Remove this code as soon as GDB supports this.
+      // The debug info verifier in LLVM operates based on the assumption that a
+      // variable has the same size as its storage and we had to disable the check
+      // for artificial variables.
       for (const auto *Field : RD->fields()) {
-        llvm::DIType FieldTy = getOrCreateType(Field->getType(), Unit);
+        llvm::DIType *FieldTy = getOrCreateType(Field->getType(), Unit);
         StringRef FieldName = Field->getName();
 
         // Ignore unnamed fields. Do not ignore unnamed records.
@@ -2869,28 +2858,28 @@ void CGDebugInfo::EmitDeclare(const VarDecl *VD, llvm::dwarf::LLVMConstants Tag,
           continue;
 
         // Use VarDecl's Tag, Scope and Line number.
-        llvm::DIVariable D = DBuilder.createLocalVariable(
-            Tag, llvm::DIDescriptor(Scope), FieldName, Unit, Line, FieldTy,
-            CGM.getLangOpts().Optimize, Flags, ArgNo);
+        auto *D = DBuilder.createLocalVariable(
+            Tag, Scope, FieldName, Unit, Line, FieldTy,
+            CGM.getLangOpts().Optimize, Flags | llvm::DINode::FlagArtificial,
+            ArgNo);
 
         // Insert an llvm.dbg.declare into the current block.
-        llvm::Instruction *Call = DBuilder.insertDeclare(
-            Storage, D, DBuilder.createExpression(), Builder.GetInsertBlock());
-        Call->setDebugLoc(llvm::DebugLoc::get(Line, Column, Scope));
+        DBuilder.insertDeclare(Storage, D, DBuilder.createExpression(Expr),
+                               llvm::DebugLoc::get(Line, Column, Scope),
+                               Builder.GetInsertBlock());
       }
-      return;
     }
   }
 
   // Create the descriptor for the variable.
-  llvm::DIVariable D = DBuilder.createLocalVariable(
-      Tag, llvm::DIDescriptor(Scope), Name, Unit, Line, Ty,
-      CGM.getLangOpts().Optimize, Flags, ArgNo);
+  auto *D =
+      DBuilder.createLocalVariable(Tag, Scope, Name, Unit, Line, Ty,
+                                   CGM.getLangOpts().Optimize, Flags, ArgNo);
 
   // Insert an llvm.dbg.declare into the current block.
-  llvm::Instruction *Call = DBuilder.insertDeclare(
-      Storage, D, DBuilder.createExpression(), Builder.GetInsertBlock());
-  Call->setDebugLoc(llvm::DebugLoc::get(Line, Column, Scope));
+  DBuilder.insertDeclare(Storage, D, DBuilder.createExpression(Expr),
+                         llvm::DebugLoc::get(Line, Column, Scope),
+                         Builder.GetInsertBlock());
 }
 
 void CGDebugInfo::EmitDeclareOfAutoVariable(const VarDecl *VD,
@@ -2906,9 +2895,9 @@ void CGDebugInfo::EmitDeclareOfAutoVariable(const VarDecl *VD,
 /// never happen though, since creating a type for the implicit self
 /// argument implies that we already parsed the interface definition
 /// and the ivar declarations in the implementation.
-llvm::DIType CGDebugInfo::CreateSelfType(const QualType &QualTy,
-                                         llvm::DIType Ty) {
-  llvm::DIType CachedTy = getTypeOrNull(QualTy);
+llvm::DIType *CGDebugInfo::CreateSelfType(const QualType &QualTy,
+                                          llvm::DIType *Ty) {
+  llvm::DIType *CachedTy = getTypeOrNull(QualTy);
   if (CachedTy)
     Ty = CachedTy;
   return DBuilder.createObjectPointerType(Ty);
@@ -2926,8 +2915,8 @@ void CGDebugInfo::EmitDeclareOfBlockDeclRefVariable(
   bool isByRef = VD->hasAttr<BlocksAttr>();
 
   uint64_t XOffset = 0;
-  llvm::DIFile Unit = getOrCreateFile(VD->getLocation());
-  llvm::DIType Ty;
+  llvm::DIFile *Unit = getOrCreateFile(VD->getLocation());
+  llvm::DIType *Ty;
   if (isByRef)
     Ty = EmitTypeForVarWithBlocksAttr(VD, &XOffset);
   else
@@ -2968,19 +2957,19 @@ void CGDebugInfo::EmitDeclareOfBlockDeclRefVariable(
   }
 
   // Create the descriptor for the variable.
-  llvm::DIVariable D =
-      DBuilder.createLocalVariable(llvm::dwarf::DW_TAG_auto_variable,
-                                   llvm::DIDescriptor(LexicalBlockStack.back()),
-                                   VD->getName(), Unit, Line, Ty);
+  auto *D = DBuilder.createLocalVariable(
+      llvm::dwarf::DW_TAG_auto_variable,
+      cast<llvm::DILocalScope>(LexicalBlockStack.back()), VD->getName(), Unit,
+      Line, Ty);
 
   // Insert an llvm.dbg.declare into the current block.
-  llvm::Instruction *Call = InsertPoint ?
-      DBuilder.insertDeclare(Storage, D, DBuilder.createExpression(addr),
-                             InsertPoint)
-    : DBuilder.insertDeclare(Storage, D, DBuilder.createExpression(addr),
-                             Builder.GetInsertBlock());
-  Call->setDebugLoc(
-      llvm::DebugLoc::get(Line, Column, LexicalBlockStack.back()));
+  auto DL = llvm::DebugLoc::get(Line, Column, LexicalBlockStack.back());
+  if (InsertPoint)
+    DBuilder.insertDeclare(Storage, D, DBuilder.createExpression(addr), DL,
+                           InsertPoint);
+  else
+    DBuilder.insertDeclare(Storage, D, DBuilder.createExpression(addr), DL,
+                           Builder.GetInsertBlock());
 }
 
 /// EmitDeclareOfArgVariable - Emit call to llvm.dbg.declare for an argument
@@ -3013,7 +3002,7 @@ void CGDebugInfo::EmitDeclareOfBlockLiteralArgVariable(const CGBlockInfo &block,
 
   // Collect some general information about the block's location.
   SourceLocation loc = blockDecl->getCaretLocation();
-  llvm::DIFile tunit = getOrCreateFile(loc);
+  llvm::DIFile *tunit = getOrCreateFile(loc);
   unsigned line = getLineNumber(loc);
   unsigned column = getColumnNumber(loc);
 
@@ -3096,7 +3085,7 @@ void CGDebugInfo::EmitDeclareOfBlockLiteralArgVariable(const CGBlockInfo &block,
     const VarDecl *variable = capture->getVariable();
     StringRef name = variable->getName();
 
-    llvm::DIType fieldType;
+    llvm::DIType *fieldType;
     if (capture->isByRef()) {
       TypeInfo PtrInfo = C.getTypeInfo(C.VoidPtrTy);
 
@@ -3118,70 +3107,67 @@ void CGDebugInfo::EmitDeclareOfBlockLiteralArgVariable(const CGBlockInfo &block,
   llvm::raw_svector_ostream(typeName) << "__block_literal_"
                                       << CGM.getUniqueBlockCount();
 
-  llvm::DIArray fieldsArray = DBuilder.getOrCreateArray(fields);
+  llvm::DINodeArray fieldsArray = DBuilder.getOrCreateArray(fields);
 
-  llvm::DIType type =
-      DBuilder.createStructType(tunit, typeName.str(), tunit, line,
-                                CGM.getContext().toBits(block.BlockSize),
-                                CGM.getContext().toBits(block.BlockAlign), 0,
-                                llvm::DIType(), fieldsArray);
+  llvm::DIType *type = DBuilder.createStructType(
+      tunit, typeName.str(), tunit, line,
+      CGM.getContext().toBits(block.BlockSize),
+      CGM.getContext().toBits(block.BlockAlign), 0, nullptr, fieldsArray);
   type = DBuilder.createPointerType(type, CGM.PointerWidthInBits);
 
   // Get overall information about the block.
-  unsigned flags = llvm::DIDescriptor::FlagArtificial;
-  llvm::MDNode *scope = LexicalBlockStack.back();
+  unsigned flags = llvm::DINode::FlagArtificial;
+  auto *scope = cast<llvm::DILocalScope>(LexicalBlockStack.back());
 
   // Create the descriptor for the parameter.
-  llvm::DIVariable debugVar = DBuilder.createLocalVariable(
-      llvm::dwarf::DW_TAG_arg_variable, llvm::DIDescriptor(scope),
-      Arg->getName(), tunit, line, type, CGM.getLangOpts().Optimize, flags,
-      ArgNo);
+  auto *debugVar = DBuilder.createLocalVariable(
+      llvm::dwarf::DW_TAG_arg_variable, scope, Arg->getName(), tunit, line,
+      type, CGM.getLangOpts().Optimize, flags, ArgNo);
 
   if (LocalAddr) {
     // Insert an llvm.dbg.value into the current block.
-    llvm::Instruction *DbgVal = DBuilder.insertDbgValueIntrinsic(
+    DBuilder.insertDbgValueIntrinsic(
         LocalAddr, 0, debugVar, DBuilder.createExpression(),
-        Builder.GetInsertBlock());
-    DbgVal->setDebugLoc(llvm::DebugLoc::get(line, column, scope));
+        llvm::DebugLoc::get(line, column, scope), Builder.GetInsertBlock());
   }
 
   // Insert an llvm.dbg.declare into the current block.
-  llvm::Instruction *DbgDecl = DBuilder.insertDeclare(
-      Arg, debugVar, DBuilder.createExpression(), Builder.GetInsertBlock());
-  DbgDecl->setDebugLoc(llvm::DebugLoc::get(line, column, scope));
+  DBuilder.insertDeclare(Arg, debugVar, DBuilder.createExpression(),
+                         llvm::DebugLoc::get(line, column, scope),
+                         Builder.GetInsertBlock());
 }
 
 /// If D is an out-of-class definition of a static data member of a class, find
 /// its corresponding in-class declaration.
-llvm::DIDerivedType
+llvm::DIDerivedType *
 CGDebugInfo::getOrCreateStaticDataMemberDeclarationOrNull(const VarDecl *D) {
   if (!D->isStaticDataMember())
-    return llvm::DIDerivedType();
+    return nullptr;
+
   auto MI = StaticDataMemberCache.find(D->getCanonicalDecl());
   if (MI != StaticDataMemberCache.end()) {
     assert(MI->second && "Static data member declaration should still exist");
-    return llvm::DIDerivedType(cast<llvm::MDNode>(MI->second));
+    return cast<llvm::DIDerivedType>(MI->second);
   }
 
   // If the member wasn't found in the cache, lazily construct and add it to the
   // type (used when a limited form of the type is emitted).
   auto DC = D->getDeclContext();
-  llvm::DICompositeType Ctxt(getContextDescriptor(cast<Decl>(DC)));
+  auto *Ctxt =
+      cast<llvm::DICompositeType>(getContextDescriptor(cast<Decl>(DC)));
   return CreateRecordStaticField(D, Ctxt, cast<RecordDecl>(DC));
 }
 
 /// Recursively collect all of the member fields of a global anonymous decl and
 /// create static variables for them. The first time this is called it needs
 /// to be on a union and then from there we can have additional unnamed fields.
-llvm::DIGlobalVariable
-CGDebugInfo::CollectAnonRecordDecls(const RecordDecl *RD, llvm::DIFile Unit,
-                                    unsigned LineNo, StringRef LinkageName,
-                                    llvm::GlobalVariable *Var,
-                                    llvm::DIDescriptor DContext) {
-  llvm::DIGlobalVariable GV;
+llvm::DIGlobalVariable *CGDebugInfo::CollectAnonRecordDecls(
+    const RecordDecl *RD, llvm::DIFile *Unit, unsigned LineNo,
+    StringRef LinkageName, llvm::GlobalVariable *Var, llvm::DIScope *DContext) {
+  llvm::DIGlobalVariable *GV = nullptr;
 
   for (const auto *Field : RD->fields()) {
-    llvm::DIType FieldTy = getOrCreateType(Field->getType(), Unit);
+    llvm::DIType *FieldTy = getOrCreateType(Field->getType(), Unit);
     StringRef FieldName = Field->getName();
 
     // Ignore unnamed fields, but recurse into anonymous records.
@@ -3193,9 +3179,9 @@ CGDebugInfo::CollectAnonRecordDecls(const RecordDecl *RD, llvm::DIFile Unit,
       continue;
     }
     // Use VarDecl's Tag, Scope and Line number.
-    GV = DBuilder.createGlobalVariable(
-        DContext, FieldName, LinkageName, Unit, LineNo, FieldTy,
-        Var->hasInternalLinkage(), Var, llvm::DIDerivedType());
+    GV = DBuilder.createGlobalVariable(DContext, FieldName, LinkageName, Unit,
+                                       LineNo, FieldTy,
+                                       Var->hasInternalLinkage(), Var, nullptr);
   }
   return GV;
 }
@@ -3205,8 +3191,8 @@ void CGDebugInfo::EmitGlobalVariable(llvm::GlobalVariable *Var,
                                      const VarDecl *D) {
   assert(DebugKind >= CodeGenOptions::LimitedDebugInfo);
   // Create global variable debug descriptor.
-  llvm::DIFile Unit;
-  llvm::DIDescriptor DContext;
+  llvm::DIFile *Unit = nullptr;
+  llvm::DIScope *DContext = nullptr;
   unsigned LineNo;
   StringRef DeclName, LinkageName;
   QualType T;
@@ -3214,7 +3200,7 @@ void CGDebugInfo::EmitGlobalVariable(llvm::GlobalVariable *Var,
 
   // Attempt to store one global variable for the declaration - even if we
   // emit a lot of fields.
-  llvm::DIGlobalVariable GV;
+  llvm::DIGlobalVariable *GV = nullptr;
 
   // If this is an anonymous union then we'll want to emit a global
   // variable for each member of the anonymous union so that it's possible
@@ -3238,16 +3224,18 @@ void CGDebugInfo::EmitGlobalVariable(const ValueDecl *VD,
                                      llvm::Constant *Init) {
   assert(DebugKind >= CodeGenOptions::LimitedDebugInfo);
   // Create the descriptor for the variable.
-  llvm::DIFile Unit = getOrCreateFile(VD->getLocation());
+  llvm::DIFile *Unit = getOrCreateFile(VD->getLocation());
   StringRef Name = VD->getName();
-  llvm::DIType Ty = getOrCreateType(VD->getType(), Unit);
+  llvm::DIType *Ty = getOrCreateType(VD->getType(), Unit);
   if (const EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(VD)) {
     const EnumDecl *ED = cast<EnumDecl>(ECD->getDeclContext());
     assert(isa<EnumType>(ED->getTypeForDecl()) && "Enum without EnumType?");
     Ty = getOrCreateType(QualType(ED->getTypeForDecl(), 0), Unit);
   }
-  // Do not use DIGlobalVariable for enums.
-  if (Ty.getTag() == llvm::dwarf::DW_TAG_enumeration_type)
+  // Do not use global variables for enums.
+  //
+  // FIXME: why not?
+  if (Ty->getTag() == llvm::dwarf::DW_TAG_enumeration_type)
     return;
   // Do not emit separate definitions for function local const/statics.
   if (isa<FunctionDecl>(VD->getDeclContext()))
@@ -3263,7 +3251,7 @@ void CGDebugInfo::EmitGlobalVariable(const ValueDecl *VD,
     return;
   }
 
-  llvm::DIDescriptor DContext =
+  llvm::DIScope *DContext =
       getContextDescriptor(dyn_cast<Decl>(VD->getDeclContext()));
 
   auto &GV = DeclCache[VD];
@@ -3274,9 +3262,9 @@ void CGDebugInfo::EmitGlobalVariable(const ValueDecl *VD,
       true, Init, getOrCreateStaticDataMemberDeclarationOrNull(VarD)));
 }
 
-llvm::DIScope CGDebugInfo::getCurrentContextDescriptor(const Decl *D) {
+llvm::DIScope *CGDebugInfo::getCurrentContextDescriptor(const Decl *D) {
   if (!LexicalBlockStack.empty())
-    return llvm::DIScope(LexicalBlockStack.back());
+    return LexicalBlockStack.back();
   return getContextDescriptor(D);
 }
 
@@ -3297,21 +3285,21 @@ void CGDebugInfo::EmitUsingDecl(const UsingDecl &UD) {
   // Emitting one decl is sufficient - debuggers can detect that this is an
   // overloaded name & provide lookup for all the overloads.
   const UsingShadowDecl &USD = **UD.shadow_begin();
-  if (llvm::DIDescriptor Target =
+  if (llvm::DINode *Target =
           getDeclarationOrDefinition(USD.getUnderlyingDecl()))
     DBuilder.createImportedDeclaration(
         getCurrentContextDescriptor(cast<Decl>(USD.getDeclContext())), Target,
         getLineNumber(USD.getLocation()));
 }
 
-llvm::DIImportedEntity
+llvm::DIImportedEntity *
 CGDebugInfo::EmitNamespaceAlias(const NamespaceAliasDecl &NA) {
   if (CGM.getCodeGenOpts().getDebugInfo() < CodeGenOptions::LimitedDebugInfo)
-    return llvm::DIImportedEntity(nullptr);
+    return nullptr;
   auto &VH = NamespaceAliasCache[&NA];
   if (VH)
-    return llvm::DIImportedEntity(cast<llvm::MDNode>(VH));
-  llvm::DIImportedEntity R(nullptr);
+    return cast<llvm::DIImportedEntity>(VH);
+  llvm::DIImportedEntity *R;
   if (const NamespaceAliasDecl *Underlying =
           dyn_cast<NamespaceAliasDecl>(NA.getAliasedNamespace()))
     // This could cache & dedup here rather than relying on metadata deduping.
@@ -3330,19 +3318,19 @@ CGDebugInfo::EmitNamespaceAlias(const NamespaceAliasDecl &NA) {
 
 /// getOrCreateNamesSpace - Return namespace descriptor for the given
 /// namespace decl.
-llvm::DINameSpace
+llvm::DINamespace *
 CGDebugInfo::getOrCreateNameSpace(const NamespaceDecl *NSDecl) {
   NSDecl = NSDecl->getCanonicalDecl();
   auto I = NameSpaceCache.find(NSDecl);
   if (I != NameSpaceCache.end())
-    return llvm::DINameSpace(cast<llvm::MDNode>(I->second));
+    return cast<llvm::DINamespace>(I->second);
 
   unsigned LineNo = getLineNumber(NSDecl->getLocation());
-  llvm::DIFile FileD = getOrCreateFile(NSDecl->getLocation());
-  llvm::DIDescriptor Context =
-    getContextDescriptor(dyn_cast<Decl>(NSDecl->getDeclContext()));
-  llvm::DINameSpace NS =
-    DBuilder.createNameSpace(Context, NSDecl->getName(), FileD, LineNo);
+  llvm::DIFile *FileD = getOrCreateFile(NSDecl->getLocation());
+  llvm::DIScope *Context =
+      getContextDescriptor(dyn_cast<Decl>(NSDecl->getDeclContext()));
+  llvm::DINamespace *NS =
+      DBuilder.createNameSpace(Context, NSDecl->getName(), FileD, LineNo);
   NameSpaceCache[NSDecl].reset(NS);
   return NS;
 }
@@ -3352,28 +3340,28 @@ void CGDebugInfo::finalize() {
   // element and the size(), so don't cache/reference them.
   for (size_t i = 0; i != ObjCInterfaceCache.size(); ++i) {
     ObjCInterfaceCacheEntry E = ObjCInterfaceCache[i];
-    E.Decl.replaceAllUsesWith(CGM.getLLVMContext(),
-                              E.Type->getDecl()->getDefinition()
-                                  ? CreateTypeDefinition(E.Type, E.Unit)
-                                  : E.Decl);
+    llvm::DIType *Ty = E.Type->getDecl()->getDefinition()
+                           ? CreateTypeDefinition(E.Type, E.Unit)
+                           : E.Decl;
+    DBuilder.replaceTemporary(llvm::TempDIType(E.Decl), Ty);
   }
 
   for (auto p : ReplaceMap) {
     assert(p.second);
-    llvm::DIType Ty(cast<llvm::MDNode>(p.second));
-    assert(Ty.isForwardDecl());
+    auto *Ty = cast<llvm::DIType>(p.second);
+    assert(Ty->isForwardDecl());
 
     auto it = TypeCache.find(p.first);
     assert(it != TypeCache.end());
     assert(it->second);
 
-    llvm::DIType RepTy(cast<llvm::MDNode>(it->second));
-    Ty.replaceAllUsesWith(CGM.getLLVMContext(), RepTy);
+    DBuilder.replaceTemporary(llvm::TempDIType(Ty),
+                              cast<llvm::DIType>(it->second));
   }
 
   for (const auto &p : FwdDeclReplaceMap) {
     assert(p.second);
-    llvm::DIDescriptor FwdDecl(cast<llvm::MDNode>(p.second));
+    llvm::TempMDNode FwdDecl(cast<llvm::MDNode>(p.second));
     llvm::Metadata *Repl;
 
     auto it = DeclCache.find(p.first);
@@ -3385,15 +3373,14 @@ void CGDebugInfo::finalize() {
     else
       Repl = it->second;
 
-    FwdDecl.replaceAllUsesWith(CGM.getLLVMContext(),
-                               llvm::DIDescriptor(cast<llvm::MDNode>(Repl)));
+    DBuilder.replaceTemporary(std::move(FwdDecl), cast<llvm::MDNode>(Repl));
   }
 
   // We keep our own list of retained types, because we need to look
   // up the final type in the type cache.
   for (std::vector<void *>::const_iterator RI = RetainedTypes.begin(),
          RE = RetainedTypes.end(); RI != RE; ++RI)
-    DBuilder.retainType(llvm::DIType(cast<llvm::MDNode>(TypeCache[*RI])));
+    DBuilder.retainType(cast<llvm::DIType>(TypeCache[*RI]));
 
   DBuilder.finalize();
 }
@@ -3401,7 +3388,8 @@ void CGDebugInfo::finalize() {
 void CGDebugInfo::EmitExplicitCastType(QualType Ty) {
   if (CGM.getCodeGenOpts().getDebugInfo() < CodeGenOptions::LimitedDebugInfo)
     return;
-  llvm::DIType DieTy = getOrCreateType(Ty, getOrCreateMainFile());
-  // Don't ignore in case of explicit cast where it is referenced indirectly.
-  DBuilder.retainType(DieTy);
+
+  if (auto *DieTy = getOrCreateType(Ty, getOrCreateMainFile()))
+    // Don't ignore in case of explicit cast where it is referenced indirectly.
+    DBuilder.retainType(DieTy);
 }
diff --git a/lib/CodeGen/CGDebugInfo.h b/lib/CodeGen/CGDebugInfo.h
index 0be032c1d790..8509e0770db6 100644
--- a/lib/CodeGen/CGDebugInfo.h
+++ b/lib/CodeGen/CGDebugInfo.h
@@ -43,47 +43,50 @@ namespace CodeGen {
   class CodeGenFunction;
   class CGBlockInfo;
 
-/// CGDebugInfo - This class gathers all debug information during compilation
+/// \brief This class gathers all debug information during compilation
 /// and is responsible for emitting to llvm globals or pass directly to
 /// the backend.
 class CGDebugInfo {
-  friend class ArtificialLocation;
+  friend class ApplyDebugLocation;
   friend class SaveAndRestoreLocation;
   CodeGenModule &CGM;
   const CodeGenOptions::DebugInfoKind DebugKind;
   llvm::DIBuilder DBuilder;
-  llvm::DICompileUnit TheCU;
+  llvm::DICompileUnit *TheCU = nullptr;
   SourceLocation CurLoc;
-  llvm::DIType VTablePtrType;
-  llvm::DIType ClassTy;
-  llvm::DICompositeType ObjTy;
-  llvm::DIType SelTy;
-  llvm::DIType OCLImage1dDITy, OCLImage1dArrayDITy, OCLImage1dBufferDITy;
-  llvm::DIType OCLImage2dDITy, OCLImage2dArrayDITy;
-  llvm::DIType OCLImage3dDITy;
-  llvm::DIType OCLEventDITy;
-  llvm::DIType BlockLiteralGeneric;
-
-  /// TypeCache - Cache of previously constructed Types.
+  llvm::DIType *VTablePtrType = nullptr;
+  llvm::DIType *ClassTy = nullptr;
+  llvm::DICompositeType *ObjTy = nullptr;
+  llvm::DIType *SelTy = nullptr;
+  llvm::DIType *OCLImage1dDITy = nullptr;
+  llvm::DIType *OCLImage1dArrayDITy = nullptr;
+  llvm::DIType *OCLImage1dBufferDITy = nullptr;
+  llvm::DIType *OCLImage2dDITy = nullptr;
+  llvm::DIType *OCLImage2dArrayDITy = nullptr;
+  llvm::DIType *OCLImage3dDITy = nullptr;
+  llvm::DIType *OCLEventDITy = nullptr;
+  llvm::DIType *BlockLiteralGeneric = nullptr;
+
+  /// \brief Cache of previously constructed Types.
   llvm::DenseMap<const void *, llvm::TrackingMDRef> TypeCache;
 
   struct ObjCInterfaceCacheEntry {
     const ObjCInterfaceType *Type;
-    llvm::DIType Decl;
-    llvm::DIFile Unit;
-    ObjCInterfaceCacheEntry(const ObjCInterfaceType *Type, llvm::DIType Decl,
-                            llvm::DIFile Unit)
+    llvm::DIType *Decl;
+    llvm::DIFile *Unit;
+    ObjCInterfaceCacheEntry(const ObjCInterfaceType *Type, llvm::DIType *Decl,
+                            llvm::DIFile *Unit)
         : Type(Type), Decl(Decl), Unit(Unit) {}
   };
 
-  /// ObjCInterfaceCache - Cache of previously constructed interfaces
+  /// \brief Cache of previously constructed interfaces
   /// which may change.
   llvm::SmallVector<ObjCInterfaceCacheEntry, 32> ObjCInterfaceCache;
 
-  /// RetainedTypes - list of interfaces we want to keep even if orphaned.
+  /// \brief list of interfaces we want to keep even if orphaned.
   std::vector<void *> RetainedTypes;
 
-  /// ReplaceMap - Cache of forward declared types to RAUW at the end of
+  /// \brief Cache of forward declared types to RAUW at the end of
   /// compilation.
   std::vector<std::pair<const TagType *, llvm::TrackingMDRef>> ReplaceMap;
 
@@ -93,14 +96,14 @@ class CGDebugInfo {
       FwdDeclReplaceMap;
 
   // LexicalBlockStack - Keep track of our current nested lexical block.
-  std::vector<llvm::TrackingMDNodeRef> LexicalBlockStack;
+  std::vector<llvm::TypedTrackingMDRef<llvm::DIScope>> LexicalBlockStack;
   llvm::DenseMap<const Decl *, llvm::TrackingMDRef> RegionMap;
   // FnBeginRegionCount - Keep track of LexicalBlockStack counter at the
   // beginning of a function. This is used to pop unbalanced regions at
   // the end of a function.
   std::vector<unsigned> FnBeginRegionCount;
 
-  /// DebugInfoNames - This is a storage for names that are
+  /// \brief This is a storage for names that are
   /// constructed on demand. For example, C++ destructors, C++ operators etc..
   llvm::BumpPtrAllocator DebugInfoNames;
   StringRef CWDName;
@@ -117,95 +120,94 @@ class CGDebugInfo {
 
   /// Helper functions for getOrCreateType.
   unsigned Checksum(const ObjCInterfaceDecl *InterfaceDecl);
-  llvm::DIType CreateType(const BuiltinType *Ty);
-  llvm::DIType CreateType(const ComplexType *Ty);
-  llvm::DIType CreateQualifiedType(QualType Ty, llvm::DIFile Fg);
-  llvm::DIType CreateType(const TypedefType *Ty, llvm::DIFile Fg);
-  llvm::DIType CreateType(const TemplateSpecializationType *Ty, llvm::DIFile Fg);
-  llvm::DIType CreateType(const ObjCObjectPointerType *Ty,
-                          llvm::DIFile F);
-  llvm::DIType CreateType(const PointerType *Ty, llvm::DIFile F);
-  llvm::DIType CreateType(const BlockPointerType *Ty, llvm::DIFile F);
-  llvm::DIType CreateType(const FunctionType *Ty, llvm::DIFile F);
-  llvm::DIType CreateType(const RecordType *Tyg);
-  llvm::DIType CreateTypeDefinition(const RecordType *Ty);
-  llvm::DICompositeType CreateLimitedType(const RecordType *Ty);
-  void CollectContainingType(const CXXRecordDecl *RD, llvm::DICompositeType CT);
-  llvm::DIType CreateType(const ObjCInterfaceType *Ty, llvm::DIFile F);
-  llvm::DIType CreateTypeDefinition(const ObjCInterfaceType *Ty, llvm::DIFile F);
-  llvm::DIType CreateType(const ObjCObjectType *Ty, llvm::DIFile F);
-  llvm::DIType CreateType(const VectorType *Ty, llvm::DIFile F);
-  llvm::DIType CreateType(const ArrayType *Ty, llvm::DIFile F);
-  llvm::DIType CreateType(const LValueReferenceType *Ty, llvm::DIFile F);
-  llvm::DIType CreateType(const RValueReferenceType *Ty, llvm::DIFile Unit);
-  llvm::DIType CreateType(const MemberPointerType *Ty, llvm::DIFile F);
-  llvm::DIType CreateType(const AtomicType *Ty, llvm::DIFile F);
-  llvm::DIType CreateEnumType(const EnumType *Ty);
-  llvm::DIType CreateTypeDefinition(const EnumType *Ty);
-  llvm::DIType CreateSelfType(const QualType &QualTy, llvm::DIType Ty);
-  llvm::DIType getTypeOrNull(const QualType);
-  llvm::DICompositeType getOrCreateMethodType(const CXXMethodDecl *Method,
-                                              llvm::DIFile F);
-  llvm::DICompositeType getOrCreateInstanceMethodType(
-      QualType ThisPtr, const FunctionProtoType *Func, llvm::DIFile Unit);
-  llvm::DICompositeType getOrCreateFunctionType(const Decl *D, QualType FnType,
-                                                llvm::DIFile F);
-  llvm::DIType getOrCreateVTablePtrType(llvm::DIFile F);
-  llvm::DINameSpace getOrCreateNameSpace(const NamespaceDecl *N);
-  llvm::DIType getOrCreateTypeDeclaration(QualType PointeeTy, llvm::DIFile F);
-  llvm::DIType CreatePointerLikeType(llvm::dwarf::Tag Tag,
-                                     const Type *Ty, QualType PointeeTy,
-                                     llvm::DIFile F);
+  llvm::DIType *CreateType(const BuiltinType *Ty);
+  llvm::DIType *CreateType(const ComplexType *Ty);
+  llvm::DIType *CreateQualifiedType(QualType Ty, llvm::DIFile *Fg);
+  llvm::DIType *CreateType(const TypedefType *Ty, llvm::DIFile *Fg);
+  llvm::DIType *CreateType(const TemplateSpecializationType *Ty,
+                           llvm::DIFile *Fg);
+  llvm::DIType *CreateType(const ObjCObjectPointerType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateType(const PointerType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateType(const BlockPointerType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateType(const FunctionType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateType(const RecordType *Tyg);
+  llvm::DIType *CreateTypeDefinition(const RecordType *Ty);
+  llvm::DICompositeType *CreateLimitedType(const RecordType *Ty);
+  void CollectContainingType(const CXXRecordDecl *RD,
+                             llvm::DICompositeType *CT);
+  llvm::DIType *CreateType(const ObjCInterfaceType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateTypeDefinition(const ObjCInterfaceType *Ty,
+                                     llvm::DIFile *F);
+  llvm::DIType *CreateType(const ObjCObjectType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateType(const VectorType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateType(const ArrayType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateType(const LValueReferenceType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateType(const RValueReferenceType *Ty, llvm::DIFile *Unit);
+  llvm::DIType *CreateType(const MemberPointerType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateType(const AtomicType *Ty, llvm::DIFile *F);
+  llvm::DIType *CreateEnumType(const EnumType *Ty);
+  llvm::DIType *CreateTypeDefinition(const EnumType *Ty);
+  llvm::DIType *CreateSelfType(const QualType &QualTy, llvm::DIType *Ty);
+  llvm::DIType *getTypeOrNull(const QualType);
+  llvm::DISubroutineType *getOrCreateMethodType(const CXXMethodDecl *Method,
+                                                llvm::DIFile *F);
+  llvm::DISubroutineType *
+  getOrCreateInstanceMethodType(QualType ThisPtr, const FunctionProtoType *Func,
+                                llvm::DIFile *Unit);
+  llvm::DISubroutineType *
+  getOrCreateFunctionType(const Decl *D, QualType FnType, llvm::DIFile *F);
+  llvm::DIType *getOrCreateVTablePtrType(llvm::DIFile *F);
+  llvm::DINamespace *getOrCreateNameSpace(const NamespaceDecl *N);
+  llvm::DIType *getOrCreateTypeDeclaration(QualType PointeeTy, llvm::DIFile *F);
+  llvm::DIType *CreatePointerLikeType(llvm::dwarf::Tag Tag, const Type *Ty,
+                                      QualType PointeeTy, llvm::DIFile *F);
 
   llvm::Value *getCachedInterfaceTypeOrNull(const QualType Ty);
-  llvm::DIType getOrCreateStructPtrType(StringRef Name, llvm::DIType &Cache);
+  llvm::DIType *getOrCreateStructPtrType(StringRef Name, llvm::DIType *&Cache);
 
-  llvm::DISubprogram CreateCXXMemberFunction(const CXXMethodDecl *Method,
-                                             llvm::DIFile F,
-                                             llvm::DIType RecordTy);
+  llvm::DISubprogram *CreateCXXMemberFunction(const CXXMethodDecl *Method,
+                                              llvm::DIFile *F,
+                                              llvm::DIType *RecordTy);
 
-  void CollectCXXMemberFunctions(const CXXRecordDecl *Decl, llvm::DIFile F,
+  void CollectCXXMemberFunctions(const CXXRecordDecl *Decl, llvm::DIFile *F,
                                  SmallVectorImpl<llvm::Metadata *> &E,
-                                 llvm::DIType T);
+                                 llvm::DIType *T);
 
-  void CollectCXXBases(const CXXRecordDecl *Decl, llvm::DIFile F,
+  void CollectCXXBases(const CXXRecordDecl *Decl, llvm::DIFile *F,
                        SmallVectorImpl<llvm::Metadata *> &EltTys,
-                       llvm::DIType RecordTy);
-
-  llvm::DIArray
-  CollectTemplateParams(const TemplateParameterList *TPList,
-                        ArrayRef<TemplateArgument> TAList,
-                        llvm::DIFile Unit);
-  llvm::DIArray
-  CollectFunctionTemplateParams(const FunctionDecl *FD, llvm::DIFile Unit);
-  llvm::DIArray
+                       llvm::DIType *RecordTy);
+
+  llvm::DINodeArray CollectTemplateParams(const TemplateParameterList *TPList,
+                                          ArrayRef<TemplateArgument> TAList,
+                                          llvm::DIFile *Unit);
+  llvm::DINodeArray CollectFunctionTemplateParams(const FunctionDecl *FD,
+                                                  llvm::DIFile *Unit);
+  llvm::DINodeArray
   CollectCXXTemplateParams(const ClassTemplateSpecializationDecl *TS,
-                           llvm::DIFile F);
+                           llvm::DIFile *F);
 
-  llvm::DIType createFieldType(StringRef name, QualType type,
-                               uint64_t sizeInBitsOverride, SourceLocation loc,
-                               AccessSpecifier AS,
-                               uint64_t offsetInBits,
-                               llvm::DIFile tunit,
-                               llvm::DIScope scope,
-                               const RecordDecl* RD = nullptr);
+  llvm::DIType *createFieldType(StringRef name, QualType type,
+                                uint64_t sizeInBitsOverride, SourceLocation loc,
+                                AccessSpecifier AS, uint64_t offsetInBits,
+                                llvm::DIFile *tunit, llvm::DIScope *scope,
+                                const RecordDecl *RD = nullptr);
 
   // Helpers for collecting fields of a record.
   void CollectRecordLambdaFields(const CXXRecordDecl *CXXDecl,
                                  SmallVectorImpl<llvm::Metadata *> &E,
-                                 llvm::DIType RecordTy);
-  llvm::DIDerivedType CreateRecordStaticField(const VarDecl *Var,
-                                              llvm::DIType RecordTy,
-                                              const RecordDecl* RD);
+                                 llvm::DIType *RecordTy);
+  llvm::DIDerivedType *CreateRecordStaticField(const VarDecl *Var,
+                                               llvm::DIType *RecordTy,
+                                               const RecordDecl *RD);
   void CollectRecordNormalField(const FieldDecl *Field, uint64_t OffsetInBits,
-                                llvm::DIFile F,
+                                llvm::DIFile *F,
                                 SmallVectorImpl<llvm::Metadata *> &E,
-                                llvm::DIType RecordTy, const RecordDecl *RD);
-  void CollectRecordFields(const RecordDecl *Decl, llvm::DIFile F,
+                                llvm::DIType *RecordTy, const RecordDecl *RD);
+  void CollectRecordFields(const RecordDecl *Decl, llvm::DIFile *F,
                            SmallVectorImpl<llvm::Metadata *> &E,
-                           llvm::DICompositeType RecordTy);
+                           llvm::DICompositeType *RecordTy);
 
-  void CollectVTableInfo(const CXXRecordDecl *Decl, llvm::DIFile F,
+  void CollectVTableInfo(const CXXRecordDecl *Decl, llvm::DIFile *F,
                          SmallVectorImpl<llvm::Metadata *> &EltTys);
 
   // CreateLexicalBlock - Create a new lexical block node and push it on
@@ -218,20 +220,15 @@ public:
 
   void finalize();
 
-  /// setLocation - Update the current source location. If \arg loc is
+  /// \brief Update the current source location. If \arg loc is
   /// invalid it is ignored.
   void setLocation(SourceLocation Loc);
 
-  /// getLocation - Return the current source location.
-  SourceLocation getLocation() const { return CurLoc; }
-
-  /// EmitLocation - Emit metadata to indicate a change in line/column
+  /// \brief Emit metadata to indicate a change in line/column
   /// information in the source file.
-  /// \param ForceColumnInfo  Assume DebugColumnInfo option is true.
-  void EmitLocation(CGBuilderTy &Builder, SourceLocation Loc,
-                    bool ForceColumnInfo = false);
+  void EmitLocation(CGBuilderTy &Builder, SourceLocation Loc);
 
-  /// EmitFunctionStart - Emit a call to llvm.dbg.function.start to indicate
+  /// \brief Emit a call to llvm.dbg.function.start to indicate
   /// start of a new function.
   /// \param Loc       The location of the function header.
   /// \param ScopeLoc  The location of the function body.
@@ -240,23 +237,23 @@ public:
                          QualType FnType, llvm::Function *Fn,
                          CGBuilderTy &Builder);
 
-  /// EmitFunctionEnd - Constructs the debug code for exiting a function.
+  /// \brief Constructs the debug code for exiting a function.
   void EmitFunctionEnd(CGBuilderTy &Builder);
 
-  /// EmitLexicalBlockStart - Emit metadata to indicate the beginning of a
+  /// \brief Emit metadata to indicate the beginning of a
   /// new lexical block and push the block onto the stack.
   void EmitLexicalBlockStart(CGBuilderTy &Builder, SourceLocation Loc);
 
-  /// EmitLexicalBlockEnd - Emit metadata to indicate the end of a new lexical
+  /// \brief Emit metadata to indicate the end of a new lexical
   /// block and pop the current block.
   void EmitLexicalBlockEnd(CGBuilderTy &Builder, SourceLocation Loc);
 
-  /// EmitDeclareOfAutoVariable - Emit call to llvm.dbg.declare for an automatic
+  /// \brief Emit call to llvm.dbg.declare for an automatic
   /// variable declaration.
   void EmitDeclareOfAutoVariable(const VarDecl *Decl, llvm::Value *AI,
                                  CGBuilderTy &Builder);
 
-  /// EmitDeclareOfBlockDeclRefVariable - Emit call to llvm.dbg.declare for an
+  /// \brief Emit call to llvm.dbg.declare for an
   /// imported variable declaration in a block.
   void EmitDeclareOfBlockDeclRefVariable(const VarDecl *variable,
                                          llvm::Value *storage,
@@ -264,12 +261,12 @@ public:
                                          const CGBlockInfo &blockInfo,
                                          llvm::Instruction *InsertPoint = 0);
 
-  /// EmitDeclareOfArgVariable - Emit call to llvm.dbg.declare for an argument
+  /// \brief Emit call to llvm.dbg.declare for an argument
   /// variable declaration.
   void EmitDeclareOfArgVariable(const VarDecl *Decl, llvm::Value *AI,
                                 unsigned ArgNo, CGBuilderTy &Builder);
 
-  /// EmitDeclareOfBlockLiteralArgVariable - Emit call to
+  /// \brief Emit call to
   /// llvm.dbg.declare for the block-literal argument to a block
   /// invocation function.
   void EmitDeclareOfBlockLiteralArgVariable(const CGBlockInfo &block,
@@ -277,31 +274,30 @@ public:
                                             llvm::Value *LocalAddr,
                                             CGBuilderTy &Builder);
 
-  /// EmitGlobalVariable - Emit information about a global variable.
+  /// \brief Emit information about a global variable.
   void EmitGlobalVariable(llvm::GlobalVariable *GV, const VarDecl *Decl);
 
-  /// EmitGlobalVariable - Emit global variable's debug info.
+  /// \brief Emit global variable's debug info.
   void EmitGlobalVariable(const ValueDecl *VD, llvm::Constant *Init);
 
-  /// \brief - Emit C++ using directive.
+  /// \brief Emit C++ using directive.
   void EmitUsingDirective(const UsingDirectiveDecl &UD);
 
-  /// EmitExplicitCastType - Emit the type explicitly casted to.
+  /// \brief Emit the type explicitly casted to.
   void EmitExplicitCastType(QualType Ty);
 
-  /// \brief - Emit C++ using declaration.
+  /// \brief Emit C++ using declaration.
   void EmitUsingDecl(const UsingDecl &UD);
 
-  /// \brief - Emit C++ namespace alias.
-  llvm::DIImportedEntity EmitNamespaceAlias(const NamespaceAliasDecl &NA);
+  /// \brief Emit C++ namespace alias.
+  llvm::DIImportedEntity *EmitNamespaceAlias(const NamespaceAliasDecl &NA);
 
-  /// getOrCreateRecordType - Emit record type's standalone debug info.
-  llvm::DIType getOrCreateRecordType(QualType Ty, SourceLocation L);
+  /// \brief Emit record type's standalone debug info.
+  llvm::DIType *getOrCreateRecordType(QualType Ty, SourceLocation L);
 
-  /// getOrCreateInterfaceType - Emit an objective c interface type standalone
+  /// \brief Emit an objective c interface type standalone
   /// debug info.
-  llvm::DIType getOrCreateInterfaceType(QualType Ty,
-                                        SourceLocation Loc);
+  llvm::DIType *getOrCreateInterfaceType(QualType Ty, SourceLocation Loc);
 
   void completeType(const EnumDecl *ED);
   void completeType(const RecordDecl *RD);
@@ -311,133 +307,129 @@ public:
   void completeTemplateDefinition(const ClassTemplateSpecializationDecl &SD);
 
 private:
-  /// EmitDeclare - Emit call to llvm.dbg.declare for a variable declaration.
+  /// \brief Emit call to llvm.dbg.declare for a variable declaration.
   /// Tag accepts custom types DW_TAG_arg_variable and DW_TAG_auto_variable,
   /// otherwise would be of type llvm::dwarf::Tag.
-  void EmitDeclare(const VarDecl *decl, llvm::dwarf::LLVMConstants Tag,
-                   llvm::Value *AI, unsigned ArgNo, CGBuilderTy &Builder);
+  void EmitDeclare(const VarDecl *decl, llvm::dwarf::Tag Tag, llvm::Value *AI,
+                   unsigned ArgNo, CGBuilderTy &Builder);
 
   // EmitTypeForVarWithBlocksAttr - Build up structure info for the byref.
   // See BuildByRefType.
-  llvm::DIType EmitTypeForVarWithBlocksAttr(const VarDecl *VD,
-                                            uint64_t *OffSet);
+  llvm::DIType *EmitTypeForVarWithBlocksAttr(const VarDecl *VD,
+                                             uint64_t *OffSet);
 
-  /// getContextDescriptor - Get context info for the decl.
-  llvm::DIScope getContextDescriptor(const Decl *Decl);
+  /// \brief Get context info for the decl.
+  llvm::DIScope *getContextDescriptor(const Decl *Decl);
 
-  llvm::DIScope getCurrentContextDescriptor(const Decl *Decl);
+  llvm::DIScope *getCurrentContextDescriptor(const Decl *Decl);
 
   /// \brief Create a forward decl for a RecordType in a given context.
-  llvm::DICompositeType getOrCreateRecordFwdDecl(const RecordType *,
-                                                 llvm::DIDescriptor);
-
-  /// createContextChain - Create a set of decls for the context chain.
-  llvm::DIDescriptor createContextChain(const Decl *Decl);
+  llvm::DICompositeType *getOrCreateRecordFwdDecl(const RecordType *,
+                                                  llvm::DIScope *);
 
-  /// getCurrentDirname - Return current directory name.
+  /// \brief Return current directory name.
   StringRef getCurrentDirname();
 
-  /// CreateCompileUnit - Create new compile unit.
+  /// \brief Create new compile unit.
   void CreateCompileUnit();
 
-  /// getOrCreateFile - Get the file debug info descriptor for the input
+  /// \brief Get the file debug info descriptor for the input
   /// location.
-  llvm::DIFile getOrCreateFile(SourceLocation Loc);
+  llvm::DIFile *getOrCreateFile(SourceLocation Loc);
 
-  /// getOrCreateMainFile - Get the file info for main compile unit.
-  llvm::DIFile getOrCreateMainFile();
+  /// \brief Get the file info for main compile unit.
+  llvm::DIFile *getOrCreateMainFile();
 
-  /// getOrCreateType - Get the type from the cache or create a new type if
+  /// \brief Get the type from the cache or create a new type if
   /// necessary.
-  llvm::DIType getOrCreateType(QualType Ty, llvm::DIFile Fg);
+  llvm::DIType *getOrCreateType(QualType Ty, llvm::DIFile *Fg);
 
-  /// getOrCreateLimitedType - Get the type from the cache or create a new
+  /// \brief Get the type from the cache or create a new
   /// partial type if necessary.
-  llvm::DIType getOrCreateLimitedType(const RecordType *Ty, llvm::DIFile F);
+  llvm::DIType *getOrCreateLimitedType(const RecordType *Ty, llvm::DIFile *F);
 
-  /// CreateTypeNode - Create type metadata for a source language type.
-  llvm::DIType CreateTypeNode(QualType Ty, llvm::DIFile Fg);
+  /// \brief Create type metadata for a source language type.
+  llvm::DIType *CreateTypeNode(QualType Ty, llvm::DIFile *Fg);
 
-  /// getObjCInterfaceDecl - return the underlying ObjCInterfaceDecl
+  /// \brief return the underlying ObjCInterfaceDecl
   /// if Ty is an ObjCInterface or a pointer to one.
   ObjCInterfaceDecl* getObjCInterfaceDecl(QualType Ty);
 
-  /// CreateMemberType - Create new member and increase Offset by FType's size.
-  llvm::DIType CreateMemberType(llvm::DIFile Unit, QualType FType,
-                                StringRef Name, uint64_t *Offset);
+  /// \brief Create new member and increase Offset by FType's size.
+  llvm::DIType *CreateMemberType(llvm::DIFile *Unit, QualType FType,
+                                 StringRef Name, uint64_t *Offset);
 
   /// \brief Retrieve the DIDescriptor, if any, for the canonical form of this
   /// declaration.
-  llvm::DIDescriptor getDeclarationOrDefinition(const Decl *D);
+  llvm::DINode *getDeclarationOrDefinition(const Decl *D);
 
-  /// getFunctionDeclaration - Return debug info descriptor to describe method
+  /// \brief Return debug info descriptor to describe method
   /// declaration for the given method definition.
-  llvm::DISubprogram getFunctionDeclaration(const Decl *D);
+  llvm::DISubprogram *getFunctionDeclaration(const Decl *D);
 
   /// Return debug info descriptor to describe in-class static data member
   /// declaration for the given out-of-class definition.
-  llvm::DIDerivedType
+  llvm::DIDerivedType *
   getOrCreateStaticDataMemberDeclarationOrNull(const VarDecl *D);
 
-  /// \brief Create a DISubprogram describing the forward
+  /// \brief Create a subprogram describing the forward
   /// decalration represented in the given FunctionDecl.
-  llvm::DISubprogram getFunctionForwardDeclaration(const FunctionDecl *FD);
+  llvm::DISubprogram *getFunctionForwardDeclaration(const FunctionDecl *FD);
 
-  /// \brief Create a DIGlobalVariable describing the forward
-  /// decalration represented in the given VarDecl.
-  llvm::DIGlobalVariable getGlobalVariableForwardDeclaration(const VarDecl *VD);
+  /// \brief Create a global variable describing the forward decalration
+  /// represented in the given VarDecl.
+  llvm::DIGlobalVariable *
+  getGlobalVariableForwardDeclaration(const VarDecl *VD);
 
   /// Return a global variable that represents one of the collection of
   /// global variables created for an anonmyous union.
-  llvm::DIGlobalVariable
-  CollectAnonRecordDecls(const RecordDecl *RD, llvm::DIFile Unit, unsigned LineNo,
-                         StringRef LinkageName, llvm::GlobalVariable *Var,
-                         llvm::DIDescriptor DContext);
+  llvm::DIGlobalVariable *
+  CollectAnonRecordDecls(const RecordDecl *RD, llvm::DIFile *Unit,
+                         unsigned LineNo, StringRef LinkageName,
+                         llvm::GlobalVariable *Var, llvm::DIScope *DContext);
 
-  /// getFunctionName - Get function name for the given FunctionDecl. If the
+  /// \brief Get function name for the given FunctionDecl. If the
   /// name is constructed on demand (e.g. C++ destructor) then the name
   /// is stored on the side.
   StringRef getFunctionName(const FunctionDecl *FD);
 
-  /// getObjCMethodName - Returns the unmangled name of an Objective-C method.
+  /// \brief Returns the unmangled name of an Objective-C method.
   /// This is the display name for the debugging info.
   StringRef getObjCMethodName(const ObjCMethodDecl *FD);
 
-  /// getSelectorName - Return selector name. This is used for debugging
+  /// \brief Return selector name. This is used for debugging
   /// info.
   StringRef getSelectorName(Selector S);
 
-  /// getClassName - Get class name including template argument list.
+  /// \brief Get class name including template argument list.
   StringRef getClassName(const RecordDecl *RD);
 
-  /// getVTableName - Get vtable name for the given Class.
+  /// \brief Get vtable name for the given Class.
   StringRef getVTableName(const CXXRecordDecl *Decl);
 
-  /// getLineNumber - Get line number for the location. If location is invalid
+  /// \brief Get line number for the location. If location is invalid
   /// then use current location.
   unsigned getLineNumber(SourceLocation Loc);
 
-  /// getColumnNumber - Get column number for the location. If location is
+  /// \brief Get column number for the location. If location is
   /// invalid then use current location.
   /// \param Force  Assume DebugColumnInfo option is true.
   unsigned getColumnNumber(SourceLocation Loc, bool Force=false);
 
   /// \brief Collect various properties of a FunctionDecl.
   /// \param GD  A GlobalDecl whose getDecl() must return a FunctionDecl.
-  void collectFunctionDeclProps(GlobalDecl GD,
-                                llvm::DIFile Unit,
+  void collectFunctionDeclProps(GlobalDecl GD, llvm::DIFile *Unit,
                                 StringRef &Name, StringRef &LinkageName,
-                                llvm::DIDescriptor &FDContext,
-                                llvm::DIArray &TParamsArray,
+                                llvm::DIScope *&FDContext,
+                                llvm::DINodeArray &TParamsArray,
                                 unsigned &Flags);
 
   /// \brief Collect various properties of a VarDecl.
-  void collectVarDeclProps(const VarDecl *VD, llvm::DIFile &Unit,
-                           unsigned &LineNo, QualType &T,
-                           StringRef &Name, StringRef &LinkageName,
-                           llvm::DIDescriptor &VDContext);
+  void collectVarDeclProps(const VarDecl *VD, llvm::DIFile *&Unit,
+                           unsigned &LineNo, QualType &T, StringRef &Name,
+                           StringRef &LinkageName, llvm::DIScope *&VDContext);
 
-  /// internString - Allocate a copy of \p A using the DebugInfoNames allocator
+  /// \brief Allocate a copy of \p A using the DebugInfoNames allocator
   /// and return a reference to it. If multiple arguments are given the strings
   /// are concatenated.
   StringRef internString(StringRef A, StringRef B = StringRef()) {
@@ -448,33 +440,60 @@ private:
   }
 };
 
+/// \brief A scoped helper to set the current debug location to the specified
+/// location or preferred location of the specified Expr.
 class ApplyDebugLocation {
-protected:
+private:
+  void init(SourceLocation TemporaryLocation, bool DefaultToEmpty = false);
+  ApplyDebugLocation(CodeGenFunction &CGF, bool DefaultToEmpty,
+                     SourceLocation TemporaryLocation);
+
   llvm::DebugLoc OriginalLocation;
   CodeGenFunction &CGF;
-
 public:
-  ApplyDebugLocation(CodeGenFunction &CGF,
-                     SourceLocation TemporaryLocation = SourceLocation(),
-                     bool ForceColumnInfo = false);
+
+  /// \brief Set the location to the (valid) TemporaryLocation.
+  ApplyDebugLocation(CodeGenFunction &CGF, SourceLocation TemporaryLocation);
+  ApplyDebugLocation(CodeGenFunction &CGF, const Expr *E);
   ApplyDebugLocation(CodeGenFunction &CGF, llvm::DebugLoc Loc);
+
   ~ApplyDebugLocation();
-};
 
-/// ArtificialLocation - An RAII object that temporarily switches to
-/// an artificial debug location that has a valid scope, but no line
-/// information. This is useful when emitting compiler-generated
-/// helper functions that have no source location associated with
-/// them. The DWARF specification allows the compiler to use the
-/// special line number 0 to indicate code that can not be attributed
-/// to any source location.
-///
-/// This is necessary because passing an empty SourceLocation to
-/// CGDebugInfo::setLocation() will result in the last valid location
-/// being reused.
-class ArtificialLocation : public ApplyDebugLocation {
-public:
-  ArtificialLocation(CodeGenFunction &CGF);
+  /// \brief Apply TemporaryLocation if it is valid. Otherwise switch to an
+  /// artificial debug location that has a valid scope, but no line information.
+  ///
+  /// Artificial locations are useful when emitting compiler-generated helper
+  /// functions that have no source location associated with them. The DWARF
+  /// specification allows the compiler to use the special line number 0 to
+  /// indicate code that can not be attributed to any source location. Note that
+  /// passing an empty SourceLocation to CGDebugInfo::setLocation() will result
+  /// in the last valid location being reused.
+  static ApplyDebugLocation CreateArtificial(CodeGenFunction &CGF) {
+    return ApplyDebugLocation(CGF, false, SourceLocation());
+  }
+  /// \brief Apply TemporaryLocation if it is valid. Otherwise switch to an
+  /// artificial debug location that has a valid scope, but no line information.
+  static ApplyDebugLocation CreateDefaultArtificial(CodeGenFunction &CGF,
+                                             SourceLocation TemporaryLocation) {
+    return ApplyDebugLocation(CGF, false, TemporaryLocation);
+  }
+
+  /// \brief Set the IRBuilder to not attach debug locations.  Note that passing
+  /// an empty SourceLocation to CGDebugInfo::setLocation() will result in the
+  /// last valid location being reused.  Note that all instructions that do not
+  /// have a location at the beginning of a function are counted towards to
+  /// funciton prologue.
+  static ApplyDebugLocation CreateEmpty(CodeGenFunction &CGF) {
+    return ApplyDebugLocation(CGF, true, SourceLocation());
+  }
+
+  /// \brief Apply TemporaryLocation if it is valid. Otherwise set the IRBuilder
+  /// to not attach debug locations.
+  static ApplyDebugLocation CreateDefaultEmpty(CodeGenFunction &CGF,
+                                             SourceLocation TemporaryLocation) {
+    return ApplyDebugLocation(CGF, true, TemporaryLocation);
+  }
+
 };
 
 
diff --git a/lib/CodeGen/CGDecl.cpp b/lib/CodeGen/CGDecl.cpp
index 15a1a7fb5f12..579a04145567 100644
--- a/lib/CodeGen/CGDecl.cpp
+++ b/lib/CodeGen/CGDecl.cpp
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "CodeGenFunction.h"
+#include "CGCleanup.h"
 #include "CGDebugInfo.h"
 #include "CGOpenCLRuntime.h"
 #include "CodeGenModule.h"
@@ -34,6 +35,7 @@ using namespace CodeGen;
 void CodeGenFunction::EmitDecl(const Decl &D) {
   switch (D.getKind()) {
   case Decl::TranslationUnit:
+  case Decl::ExternCContext:
   case Decl::Namespace:
   case Decl::UnresolvedUsingTypename:
   case Decl::ClassTemplateSpecialization:
@@ -154,6 +156,8 @@ static std::string getStaticDeclName(CodeGenModule &CGM, const VarDecl &D) {
   assert(!D.isExternallyVisible() && "name shouldn't matter");
   std::string ContextName;
   const DeclContext *DC = D.getDeclContext();
+  if (auto *CD = dyn_cast<CapturedDecl>(DC))
+    DC = cast<DeclContext>(CD->getNonClosureContext());
   if (const auto *FD = dyn_cast<FunctionDecl>(DC))
     ContextName = CGM.getMangledName(FD);
   else if (const auto *BD = dyn_cast<BlockDecl>(DC))
@@ -206,6 +210,9 @@ llvm::Constant *CodeGenModule::getOrCreateStaticVarDecl(
   GV->setAlignment(getContext().getDeclAlign(&D).getQuantity());
   setGlobalVisibility(GV, &D);
 
+  if (supportsCOMDAT() && GV->isWeakForLinker())
+    GV->setComdat(TheModule.getOrInsertComdat(GV->getName()));
+
   if (D.getTLSKind())
     setTLSMode(GV, D);
 
@@ -512,10 +519,7 @@ namespace {
       : Addr(addr), Size(size) {}
 
     void Emit(CodeGenFunction &CGF, Flags flags) override {
-      llvm::Value *castAddr = CGF.Builder.CreateBitCast(Addr, CGF.Int8PtrTy);
-      CGF.Builder.CreateCall2(CGF.CGM.getLLVMLifetimeEndFn(),
-                              Size, castAddr)
-        ->setDoesNotThrow();
+      CGF.EmitLifetimeEnd(Size, Addr);
     }
   };
 }
@@ -631,8 +635,9 @@ void CodeGenFunction::EmitScalarInit(const Expr *init, const ValueDecl *D,
     if (capturedByInit) {
       // We can use a simple GEP for this because it can't have been
       // moved yet.
-      tempLV.setAddress(Builder.CreateStructGEP(tempLV.getAddress(),
-                                   getByRefValueLLVMField(cast<VarDecl>(D))));
+      tempLV.setAddress(Builder.CreateStructGEP(
+          nullptr, tempLV.getAddress(),
+          getByRefValueLLVMField(cast<VarDecl>(D)).second));
     }
 
     llvm::PointerType *ty
@@ -793,8 +798,9 @@ static void emitStoresForInitAfterMemset(llvm::Constant *Init, llvm::Value *Loc,
 
       // If necessary, get a pointer to the element and emit it.
       if (!Elt->isNullValue() && !isa<llvm::UndefValue>(Elt))
-        emitStoresForInitAfterMemset(Elt, Builder.CreateConstGEP2_32(Loc, 0, i),
-                                     isVolatile, Builder);
+        emitStoresForInitAfterMemset(
+            Elt, Builder.CreateConstGEP2_32(Init->getType(), Loc, 0, i),
+            isVolatile, Builder);
     }
     return;
   }
@@ -807,8 +813,9 @@ static void emitStoresForInitAfterMemset(llvm::Constant *Init, llvm::Value *Loc,
 
     // If necessary, get a pointer to the element and emit it.
     if (!Elt->isNullValue() && !isa<llvm::UndefValue>(Elt))
-      emitStoresForInitAfterMemset(Elt, Builder.CreateConstGEP2_32(Loc, 0, i),
-                                   isVolatile, Builder);
+      emitStoresForInitAfterMemset(
+          Elt, Builder.CreateConstGEP2_32(Init->getType(), Loc, 0, i),
+          isVolatile, Builder);
   }
 }
 
@@ -833,21 +840,6 @@ static bool shouldUseMemSetPlusStoresToInitialize(llvm::Constant *Init,
          canEmitInitWithFewStoresAfterMemset(Init, StoreBudget);
 }
 
-/// Should we use the LLVM lifetime intrinsics for the given local variable?
-static bool shouldUseLifetimeMarkers(CodeGenFunction &CGF, const VarDecl &D,
-                                     unsigned Size) {
-  // For now, only in optimized builds.
-  if (CGF.CGM.getCodeGenOpts().OptimizationLevel == 0)
-    return false;
-
-  // Limit the size of marked objects to 32 bytes. We don't want to increase
-  // compile time by marking tiny objects.
-  unsigned SizeThreshold = 32;
-
-  return Size > SizeThreshold;
-}
-
-
 /// EmitAutoVarDecl - Emit code and set up an entry in LocalDeclMap for a
 /// variable declaration with auto, register, or no storage class specifier.
 /// These turn into simple stack objects, or GlobalValues depending on target.
@@ -857,6 +849,38 @@ void CodeGenFunction::EmitAutoVarDecl(const VarDecl &D) {
   EmitAutoVarCleanups(emission);
 }
 
+/// Emit a lifetime.begin marker if some criteria are satisfied.
+/// \return a pointer to the temporary size Value if a marker was emitted, null
+/// otherwise
+llvm::Value *CodeGenFunction::EmitLifetimeStart(uint64_t Size,
+                                                llvm::Value *Addr) {
+  // For now, only in optimized builds.
+  if (CGM.getCodeGenOpts().OptimizationLevel == 0)
+    return nullptr;
+
+  // Disable lifetime markers in msan builds.
+  // FIXME: Remove this when msan works with lifetime markers.
+  if (getLangOpts().Sanitize.has(SanitizerKind::Memory))
+    return nullptr;
+
+  llvm::Value *SizeV = llvm::ConstantInt::get(Int64Ty, Size);
+  llvm::Value *Args[] = {
+      SizeV,
+      new llvm::BitCastInst(Addr, Int8PtrTy, "", Builder.GetInsertBlock())};
+  llvm::CallInst *C = llvm::CallInst::Create(CGM.getLLVMLifetimeStartFn(), Args,
+                                             "", Builder.GetInsertBlock());
+  C->setDoesNotThrow();
+  return SizeV;
+}
+
+void CodeGenFunction::EmitLifetimeEnd(llvm::Value *Size, llvm::Value *Addr) {
+  llvm::Value *Args[] = {Size, new llvm::BitCastInst(Addr, Int8PtrTy, "",
+                                                     Builder.GetInsertBlock())};
+  llvm::CallInst *C = llvm::CallInst::Create(CGM.getLLVMLifetimeEndFn(), Args,
+                                             "", Builder.GetInsertBlock());
+  C->setDoesNotThrow();
+}
+
 /// EmitAutoVarAlloca - Emit the alloca and debug information for a
 /// local variable.  Does not emit initialization or destruction.
 CodeGenFunction::AutoVarEmission
@@ -952,13 +976,8 @@ CodeGenFunction::EmitAutoVarAlloca(const VarDecl &D) {
       // Emit a lifetime intrinsic if meaningful.  There's no point
       // in doing this if we don't have a valid insertion point (?).
       uint64_t size = CGM.getDataLayout().getTypeAllocSize(LTy);
-      if (HaveInsertPoint() && shouldUseLifetimeMarkers(*this, D, size)) {
-        llvm::Value *sizeV = llvm::ConstantInt::get(Int64Ty, size);
-
-        emission.SizeForLifetimeMarkers = sizeV;
-        llvm::Value *castAddr = Builder.CreateBitCast(Alloc, Int8PtrTy);
-        Builder.CreateCall2(CGM.getLLVMLifetimeStartFn(), sizeV, castAddr)
-          ->setDoesNotThrow();
+      if (HaveInsertPoint()) {
+        emission.SizeForLifetimeMarkers = EmitLifetimeStart(size, Alloc);
       } else {
         assert(!emission.useLifetimeMarkers());
       }
@@ -971,7 +990,7 @@ CodeGenFunction::EmitAutoVarAlloca(const VarDecl &D) {
       llvm::Value *Stack = CreateTempAlloca(Int8PtrTy, "saved_stack");
 
       llvm::Value *F = CGM.getIntrinsic(llvm::Intrinsic::stacksave);
-      llvm::Value *V = Builder.CreateCall(F);
+      llvm::Value *V = Builder.CreateCall(F, {});
 
       Builder.CreateStore(V, Stack);
 
@@ -1087,7 +1106,7 @@ void CodeGenFunction::EmitAutoVarInit(const AutoVarEmission &emission) {
   if (emission.wasEmittedAsGlobal()) return;
 
   const VarDecl &D = *emission.Variable;
-  ApplyDebugLocation DL(*this, D.getLocation());
+  auto DL = ApplyDebugLocation::CreateDefaultArtificial(*this, D.getLocation());
   QualType type = D.getType();
 
   // If this local has an initializer, emit it now.
@@ -1304,6 +1323,8 @@ void CodeGenFunction::EmitAutoVarCleanups(const AutoVarEmission &emission) {
     EHStack.pushCleanup<CallLifetimeEnd>(NormalCleanup,
                                          emission.getAllocatedAddress(),
                                          emission.getSizeForLifetimeMarkers());
+    EHCleanupScope &cleanup = cast<EHCleanupScope>(*EHStack.begin());
+    cleanup.setLifetimeMarker();
   }
 
   // Check the type for a cleanup.
diff --git a/lib/CodeGen/CGDeclCXX.cpp b/lib/CodeGen/CGDeclCXX.cpp
index 19e4bdd00cfe..06d157bd82e7 100644
--- a/lib/CodeGen/CGDeclCXX.cpp
+++ b/lib/CodeGen/CGDeclCXX.cpp
@@ -139,9 +139,32 @@ void CodeGenFunction::EmitCXXGlobalVarDeclInit(const VarDecl &D,
   const Expr *Init = D.getInit();
   QualType T = D.getType();
 
+  // The address space of a static local variable (DeclPtr) may be different
+  // from the address space of the "this" argument of the constructor. In that
+  // case, we need an addrspacecast before calling the constructor.
+  //
+  // struct StructWithCtor {
+  //   __device__ StructWithCtor() {...}
+  // };
+  // __device__ void foo() {
+  //   __shared__ StructWithCtor s;
+  //   ...
+  // }
+  //
+  // For example, in the above CUDA code, the static local variable s has a
+  // "shared" address space qualifier, but the constructor of StructWithCtor
+  // expects "this" in the "generic" address space.
+  unsigned ExpectedAddrSpace = getContext().getTargetAddressSpace(T);
+  unsigned ActualAddrSpace = DeclPtr->getType()->getPointerAddressSpace();
+  if (ActualAddrSpace != ExpectedAddrSpace) {
+    llvm::Type *LTy = CGM.getTypes().ConvertTypeForMem(T);
+    llvm::PointerType *PTy = llvm::PointerType::get(LTy, ExpectedAddrSpace);
+    DeclPtr = llvm::ConstantExpr::getAddrSpaceCast(DeclPtr, PTy);
+  }
+
   if (!T->isReferenceType()) {
     if (getLangOpts().OpenMP && D.hasAttr<OMPThreadPrivateDeclAttr>())
-      (void)CGM.getOpenMPRuntime().EmitOMPThreadPrivateVarDefinition(
+      (void)CGM.getOpenMPRuntime().emitThreadPrivateVarDefinition(
           &D, DeclPtr, D.getAttr<OMPThreadPrivateDeclAttr>()->getLocation(),
           PerformInit, this);
     if (PerformInit)
@@ -236,6 +259,8 @@ llvm::Function *CodeGenModule::CreateGlobalInitOrDestructFunction(
       Fn->setSection(Section);
   }
 
+  SetLLVMFunctionAttributes(nullptr, getTypes().arrangeNullaryFunction(), Fn);
+
   Fn->setCallingConv(getRuntimeCC());
 
   if (!getLangOpts().Exceptions)
@@ -267,15 +292,7 @@ void CodeGenModule::EmitPointerToInitFunc(const VarDecl *D,
   addUsedGlobal(PtrArray);
 
   // If the GV is already in a comdat group, then we have to join it.
-  llvm::Comdat *C = GV->getComdat();
-
-  // LinkOnce and Weak linkage are lowered down to a single-member comdat group.
-  // Make an explicit group so we can join it.
-  if (!C && (GV->hasWeakLinkage() || GV->hasLinkOnceLinkage())) {
-    C = TheModule.getOrInsertComdat(GV->getName());
-    GV->setComdat(C);
-  }
-  if (C)
+  if (llvm::Comdat *C = GV->getComdat())
     PtrArray->setComdat(C);
 }
 
@@ -283,6 +300,11 @@ void
 CodeGenModule::EmitCXXGlobalVarDeclInitFunc(const VarDecl *D,
                                             llvm::GlobalVariable *Addr,
                                             bool PerformInit) {
+  // Check if we've already initialized this decl.
+  auto I = DelayedCXXInitPosition.find(D);
+  if (I != DelayedCXXInitPosition.end() && I->second == ~0U)
+    return;
+
   llvm::FunctionType *FTy = llvm::FunctionType::get(VoidTy, false);
   SmallString<256> FnName;
   {
@@ -312,11 +334,9 @@ CodeGenModule::EmitCXXGlobalVarDeclInitFunc(const VarDecl *D,
     CXXThreadLocalInitVars.push_back(Addr);
   } else if (PerformInit && ISA) {
     EmitPointerToInitFunc(D, Addr, Fn, ISA);
-    DelayedCXXInitPosition.erase(D);
   } else if (auto *IPA = D->getAttr<InitPriorityAttr>()) {
     OrderGlobalInits Key(IPA->getPriority(), PrioritizedCXXGlobalInits.size());
     PrioritizedCXXGlobalInits.push_back(std::make_pair(Key, Fn));
-    DelayedCXXInitPosition.erase(D);
   } else if (isTemplateInstantiation(D->getTemplateSpecializationKind())) {
     // C++ [basic.start.init]p2:
     //   Definitions of explicitly specialized class template static data
@@ -331,24 +351,24 @@ CodeGenModule::EmitCXXGlobalVarDeclInitFunc(const VarDecl *D,
     // minor startup time optimization.  In the MS C++ ABI, there are no guard
     // variables, so this COMDAT key is required for correctness.
     AddGlobalCtor(Fn, 65535, COMDATKey);
-    DelayedCXXInitPosition.erase(D);
   } else if (D->hasAttr<SelectAnyAttr>()) {
     // SelectAny globals will be comdat-folded. Put the initializer into a
     // COMDAT group associated with the global, so the initializers get folded
     // too.
     AddGlobalCtor(Fn, 65535, COMDATKey);
-    DelayedCXXInitPosition.erase(D);
   } else {
-    llvm::DenseMap<const Decl *, unsigned>::iterator I =
-      DelayedCXXInitPosition.find(D);
+    I = DelayedCXXInitPosition.find(D); // Re-do lookup in case of re-hash.
     if (I == DelayedCXXInitPosition.end()) {
       CXXGlobalInits.push_back(Fn);
-    } else {
-      assert(CXXGlobalInits[I->second] == nullptr);
+    } else if (I->second != ~0U) {
+      assert(I->second < CXXGlobalInits.size() &&
+             CXXGlobalInits[I->second] == nullptr);
       CXXGlobalInits[I->second] = Fn;
-      DelayedCXXInitPosition.erase(I);
     }
   }
+
+  // Remember that we already emitted the initializer for this global.
+  DelayedCXXInitPosition[D] = ~0U;
 }
 
 void CodeGenModule::EmitCXXThreadLocalInitFunc() {
@@ -411,7 +431,7 @@ CodeGenModule::EmitCXXGlobalInitFunc() {
     // priority emitted above.
     FileName = llvm::sys::path::filename(MainFile->getName());
   } else {
-    FileName = SmallString<128>("<null>");
+    FileName = "<null>";
   }
 
   for (size_t i = 0; i < FileName.size(); ++i) {
@@ -477,11 +497,11 @@ CodeGenFunction::GenerateCXXGlobalInitFunc(llvm::Function *Fn,
                                            ArrayRef<llvm::Function *> Decls,
                                            llvm::GlobalVariable *Guard) {
   {
-    ApplyDebugLocation NL(*this);
+    auto NL = ApplyDebugLocation::CreateEmpty(*this);
     StartFunction(GlobalDecl(), getContext().VoidTy, Fn,
                   getTypes().arrangeNullaryFunction(), FunctionArgList());
     // Emit an artificial location for this function.
-    ArtificialLocation AL(*this);
+    auto AL = ApplyDebugLocation::CreateArtificial(*this);
 
     llvm::BasicBlock *ExitBlock = nullptr;
     if (Guard) {
@@ -528,11 +548,11 @@ void CodeGenFunction::GenerateCXXGlobalDtorsFunc(llvm::Function *Fn,
                   const std::vector<std::pair<llvm::WeakVH, llvm::Constant*> >
                                                 &DtorsAndObjects) {
   {
-    ApplyDebugLocation NL(*this);
+    auto NL = ApplyDebugLocation::CreateEmpty(*this);
     StartFunction(GlobalDecl(), getContext().VoidTy, Fn,
                   getTypes().arrangeNullaryFunction(), FunctionArgList());
     // Emit an artificial location for this function.
-    ArtificialLocation AL(*this);
+    auto AL = ApplyDebugLocation::CreateArtificial(*this);
 
     // Emit the dtors, in reverse order from construction.
     for (unsigned i = 0, e = DtorsAndObjects.size(); i != e; ++i) {
diff --git a/lib/CodeGen/CGException.cpp b/lib/CodeGen/CGException.cpp
index cb8eb8fa490c..d9a3f0b252a5 100644
--- a/lib/CodeGen/CGException.cpp
+++ b/lib/CodeGen/CGException.cpp
@@ -16,23 +16,18 @@
 #include "CGCleanup.h"
 #include "CGObjCRuntime.h"
 #include "TargetInfo.h"
+#include "clang/AST/Mangle.h"
 #include "clang/AST/StmtCXX.h"
 #include "clang/AST/StmtObjC.h"
+#include "clang/AST/StmtVisitor.h"
 #include "llvm/IR/CallSite.h"
 #include "llvm/IR/Intrinsics.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/Support/SaveAndRestore.h"
 
 using namespace clang;
 using namespace CodeGen;
 
-static llvm::Constant *getAllocateExceptionFn(CodeGenModule &CGM) {
-  // void *__cxa_allocate_exception(size_t thrown_size);
-
-  llvm::FunctionType *FTy =
-    llvm::FunctionType::get(CGM.Int8PtrTy, CGM.SizeTy, /*IsVarArgs=*/false);
-
-  return CGM.CreateRuntimeFunction(FTy, "__cxa_allocate_exception");
-}
-
 static llvm::Constant *getFreeExceptionFn(CodeGenModule &CGM) {
   // void __cxa_free_exception(void *thrown_exception);
 
@@ -42,44 +37,6 @@ static llvm::Constant *getFreeExceptionFn(CodeGenModule &CGM) {
   return CGM.CreateRuntimeFunction(FTy, "__cxa_free_exception");
 }
 
-static llvm::Constant *getThrowFn(CodeGenModule &CGM) {
-  // void __cxa_throw(void *thrown_exception, std::type_info *tinfo,
-  //                  void (*dest) (void *));
-
-  llvm::Type *Args[3] = { CGM.Int8PtrTy, CGM.Int8PtrTy, CGM.Int8PtrTy };
-  llvm::FunctionType *FTy =
-    llvm::FunctionType::get(CGM.VoidTy, Args, /*IsVarArgs=*/false);
-
-  return CGM.CreateRuntimeFunction(FTy, "__cxa_throw");
-}
-
-static llvm::Constant *getGetExceptionPtrFn(CodeGenModule &CGM) {
-  // void *__cxa_get_exception_ptr(void*);
-
-  llvm::FunctionType *FTy =
-    llvm::FunctionType::get(CGM.Int8PtrTy, CGM.Int8PtrTy, /*IsVarArgs=*/false);
-
-  return CGM.CreateRuntimeFunction(FTy, "__cxa_get_exception_ptr");
-}
-
-static llvm::Constant *getBeginCatchFn(CodeGenModule &CGM) {
-  // void *__cxa_begin_catch(void*);
-
-  llvm::FunctionType *FTy =
-    llvm::FunctionType::get(CGM.Int8PtrTy, CGM.Int8PtrTy, /*IsVarArgs=*/false);
-
-  return CGM.CreateRuntimeFunction(FTy, "__cxa_begin_catch");
-}
-
-static llvm::Constant *getEndCatchFn(CodeGenModule &CGM) {
-  // void __cxa_end_catch();
-
-  llvm::FunctionType *FTy =
-    llvm::FunctionType::get(CGM.VoidTy, /*IsVarArgs=*/false);
-
-  return CGM.CreateRuntimeFunction(FTy, "__cxa_end_catch");
-}
-
 static llvm::Constant *getUnexpectedFn(CodeGenModule &CGM) {
   // void __cxa_call_unexpected(void *thrown_exception);
 
@@ -89,23 +46,30 @@ static llvm::Constant *getUnexpectedFn(CodeGenModule &CGM) {
   return CGM.CreateRuntimeFunction(FTy, "__cxa_call_unexpected");
 }
 
-static llvm::Constant *getTerminateFn(CodeGenModule &CGM) {
+llvm::Constant *CodeGenModule::getTerminateFn() {
   // void __terminate();
 
   llvm::FunctionType *FTy =
-    llvm::FunctionType::get(CGM.VoidTy, /*IsVarArgs=*/false);
+    llvm::FunctionType::get(VoidTy, /*IsVarArgs=*/false);
 
   StringRef name;
 
   // In C++, use std::terminate().
-  if (CGM.getLangOpts().CPlusPlus)
-    name = "_ZSt9terminatev"; // FIXME: mangling!
-  else if (CGM.getLangOpts().ObjC1 &&
-           CGM.getLangOpts().ObjCRuntime.hasTerminate())
+  if (getLangOpts().CPlusPlus &&
+      getTarget().getCXXABI().isItaniumFamily()) {
+    name = "_ZSt9terminatev";
+  } else if (getLangOpts().CPlusPlus &&
+             getTarget().getCXXABI().isMicrosoft()) {
+    if (getLangOpts().isCompatibleWithMSVC(LangOptions::MSVC2015))
+      name = "__std_terminate";
+    else
+      name = "\01?terminate@@YAXXZ";
+  } else if (getLangOpts().ObjC1 &&
+             getLangOpts().ObjCRuntime.hasTerminate())
     name = "objc_terminate";
   else
     name = "abort";
-  return CGM.CreateRuntimeFunction(FTy, name);
+  return CreateRuntimeFunction(FTy, name);
 }
 
 static llvm::Constant *getCatchallRethrowFn(CodeGenModule &CGM,
@@ -126,7 +90,12 @@ namespace {
     // This function must have prototype void(void*).
     const char *CatchallRethrowFn;
 
-    static const EHPersonality &get(CodeGenModule &CGM);
+    static const EHPersonality &get(CodeGenModule &CGM,
+                                    const FunctionDecl *FD);
+    static const EHPersonality &get(CodeGenFunction &CGF) {
+      return get(CGF.CGM, dyn_cast_or_null<FunctionDecl>(CGF.CurCodeDecl));
+    }
+
     static const EHPersonality GNU_C;
     static const EHPersonality GNU_C_SJLJ;
     static const EHPersonality GNU_C_SEH;
@@ -137,6 +106,9 @@ namespace {
     static const EHPersonality GNU_CPlusPlus;
     static const EHPersonality GNU_CPlusPlus_SJLJ;
     static const EHPersonality GNU_CPlusPlus_SEH;
+    static const EHPersonality MSVC_except_handler;
+    static const EHPersonality MSVC_C_specific_handler;
+    static const EHPersonality MSVC_CxxFrameHandler3;
   };
 }
 
@@ -159,6 +131,12 @@ const EHPersonality
 EHPersonality::GNU_ObjCXX = { "__gnustep_objcxx_personality_v0", nullptr };
 const EHPersonality
 EHPersonality::GNUstep_ObjC = { "__gnustep_objc_personality_v0", nullptr };
+const EHPersonality
+EHPersonality::MSVC_except_handler = { "_except_handler3", nullptr };
+const EHPersonality
+EHPersonality::MSVC_C_specific_handler = { "__C_specific_handler", nullptr };
+const EHPersonality
+EHPersonality::MSVC_CxxFrameHandler3 = { "__CxxFrameHandler3", nullptr };
 
 /// On Win64, use libgcc's SEH personality function. We fall back to dwarf on
 /// other platforms, unless the user asked for SjLj exceptions.
@@ -231,9 +209,29 @@ static const EHPersonality &getObjCXXPersonality(const llvm::Triple &T,
   llvm_unreachable("bad runtime kind");
 }
 
-const EHPersonality &EHPersonality::get(CodeGenModule &CGM) {
+static const EHPersonality &getSEHPersonalityMSVC(const llvm::Triple &T) {
+  if (T.getArch() == llvm::Triple::x86)
+    return EHPersonality::MSVC_except_handler;
+  return EHPersonality::MSVC_C_specific_handler;
+}
+
+const EHPersonality &EHPersonality::get(CodeGenModule &CGM,
+                                        const FunctionDecl *FD) {
   const llvm::Triple &T = CGM.getTarget().getTriple();
   const LangOptions &L = CGM.getLangOpts();
+
+  // Try to pick a personality function that is compatible with MSVC if we're
+  // not compiling Obj-C. Obj-C users better have an Obj-C runtime that supports
+  // the GCC-style personality function.
+  if (T.isWindowsMSVCEnvironment() && !L.ObjC1) {
+    if (L.SjLjExceptions)
+      return EHPersonality::GNU_CPlusPlus_SJLJ;
+    else if (FD && FD->usesSEHTry())
+      return getSEHPersonalityMSVC(T);
+    else
+      return EHPersonality::MSVC_CxxFrameHandler3;
+  }
+
   if (L.CPlusPlus && L.ObjC1)
     return getObjCXXPersonality(T, L);
   else if (L.CPlusPlus)
@@ -318,7 +316,7 @@ void CodeGenModule::SimplifyPersonality() {
   if (!LangOpts.ObjCRuntime.isNeXTFamily())
     return;
 
-  const EHPersonality &ObjCXX = EHPersonality::get(*this);
+  const EHPersonality &ObjCXX = EHPersonality::get(*this, /*FD=*/nullptr);
   const EHPersonality &CXX =
       getCXXPersonality(getTarget().getTriple(), LangOpts);
   if (&ObjCXX == &CXX)
@@ -369,17 +367,16 @@ namespace {
 // differs from EmitAnyExprToMem only in that, if a final copy-ctor
 // call is required, an exception within that copy ctor causes
 // std::terminate to be invoked.
-static void EmitAnyExprToExn(CodeGenFunction &CGF, const Expr *e,
-                             llvm::Value *addr) {
+void CodeGenFunction::EmitAnyExprToExn(const Expr *e, llvm::Value *addr) {
   // Make sure the exception object is cleaned up if there's an
   // exception during initialization.
-  CGF.pushFullExprCleanup<FreeException>(EHCleanup, addr);
-  EHScopeStack::stable_iterator cleanup = CGF.EHStack.stable_begin();
+  pushFullExprCleanup<FreeException>(EHCleanup, addr);
+  EHScopeStack::stable_iterator cleanup = EHStack.stable_begin();
 
   // __cxa_allocate_exception returns a void*;  we need to cast this
   // to the appropriate type for the object.
-  llvm::Type *ty = CGF.ConvertTypeForMem(e->getType())->getPointerTo();
-  llvm::Value *typedAddr = CGF.Builder.CreateBitCast(addr, ty);
+  llvm::Type *ty = ConvertTypeForMem(e->getType())->getPointerTo();
+  llvm::Value *typedAddr = Builder.CreateBitCast(addr, ty);
 
   // FIXME: this isn't quite right!  If there's a final unelided call
   // to a copy constructor, then according to [except.terminate]p1 we
@@ -388,11 +385,11 @@ static void EmitAnyExprToExn(CodeGenFunction &CGF, const Expr *e,
   // evaluated but before the exception is caught.  But the best way
   // to handle that is to teach EmitAggExpr to do the final copy
   // differently if it can't be elided.
-  CGF.EmitAnyExprToMem(e, typedAddr, e->getType().getQualifiers(), 
-                       /*IsInit*/ true);
+  EmitAnyExprToMem(e, typedAddr, e->getType().getQualifiers(),
+                   /*IsInit*/ true);
 
   // Deactivate the cleanup block.
-  CGF.DeactivateCleanupBlock(cleanup, cast<llvm::Instruction>(typedAddr));
+  DeactivateCleanupBlock(cleanup, cast<llvm::Instruction>(typedAddr));
 }
 
 llvm::Value *CodeGenFunction::getExceptionSlot() {
@@ -417,67 +414,18 @@ llvm::Value *CodeGenFunction::getSelectorFromSlot() {
 
 void CodeGenFunction::EmitCXXThrowExpr(const CXXThrowExpr *E,
                                        bool KeepInsertionPoint) {
-  if (!E->getSubExpr()) {
-    CGM.getCXXABI().emitRethrow(*this, /*isNoReturn*/true);
-
-    // throw is an expression, and the expression emitters expect us
-    // to leave ourselves at a valid insertion point.
-    if (KeepInsertionPoint)
-      EmitBlock(createBasicBlock("throw.cont"));
-
-    return;
-  }
-
-  if (CGM.getTarget().getTriple().isKnownWindowsMSVCEnvironment()) {
-    ErrorUnsupported(E, "throw expression");
-    return;
-  }
-
-  QualType ThrowType = E->getSubExpr()->getType();
-
-  if (ThrowType->isObjCObjectPointerType()) {
-    const Stmt *ThrowStmt = E->getSubExpr();
-    const ObjCAtThrowStmt S(E->getExprLoc(),
-                            const_cast<Stmt *>(ThrowStmt));
-    CGM.getObjCRuntime().EmitThrowStmt(*this, S, false);
-    // This will clear insertion point which was not cleared in
-    // call to EmitThrowStmt.
-    if (KeepInsertionPoint)
-      EmitBlock(createBasicBlock("throw.cont"));
-    return;
-  }
-  
-  // Now allocate the exception object.
-  llvm::Type *SizeTy = ConvertType(getContext().getSizeType());
-  uint64_t TypeSize = getContext().getTypeSizeInChars(ThrowType).getQuantity();
-
-  llvm::Constant *AllocExceptionFn = getAllocateExceptionFn(CGM);
-  llvm::CallInst *ExceptionPtr =
-    EmitNounwindRuntimeCall(AllocExceptionFn,
-                            llvm::ConstantInt::get(SizeTy, TypeSize),
-                            "exception");
-  
-  EmitAnyExprToExn(*this, E->getSubExpr(), ExceptionPtr);
-
-  // Now throw the exception.
-  llvm::Constant *TypeInfo = CGM.GetAddrOfRTTIDescriptor(ThrowType, 
-                                                         /*ForEH=*/true);
-
-  // The address of the destructor.  If the exception type has a
-  // trivial destructor (or isn't a record), we just pass null.
-  llvm::Constant *Dtor = nullptr;
-  if (const RecordType *RecordTy = ThrowType->getAs<RecordType>()) {
-    CXXRecordDecl *Record = cast<CXXRecordDecl>(RecordTy->getDecl());
-    if (!Record->hasTrivialDestructor()) {
-      CXXDestructorDecl *DtorD = Record->getDestructor();
-      Dtor = CGM.getAddrOfCXXStructor(DtorD, StructorType::Complete);
-      Dtor = llvm::ConstantExpr::getBitCast(Dtor, Int8PtrTy);
+  if (const Expr *SubExpr = E->getSubExpr()) {
+    QualType ThrowType = SubExpr->getType();
+    if (ThrowType->isObjCObjectPointerType()) {
+      const Stmt *ThrowStmt = E->getSubExpr();
+      const ObjCAtThrowStmt S(E->getExprLoc(), const_cast<Stmt *>(ThrowStmt));
+      CGM.getObjCRuntime().EmitThrowStmt(*this, S, false);
+    } else {
+      CGM.getCXXABI().emitThrow(*this, E);
     }
+  } else {
+    CGM.getCXXABI().emitRethrow(*this, /*isNoReturn=*/true);
   }
-  if (!Dtor) Dtor = llvm::Constant::getNullValue(Int8PtrTy);
-
-  llvm::Value *args[] = { ExceptionPtr, TypeInfo, Dtor };
-  EmitNoreturnRuntimeCallOrInvoke(getThrowFn(CGM), args);
 
   // throw is an expression, and the expression emitters expect us
   // to leave ourselves at a valid insertion point.
@@ -509,6 +457,10 @@ void CodeGenFunction::EmitStartEHSpec(const Decl *D) {
       EHStack.pushTerminate();
     }
   } else if (EST == EST_Dynamic || EST == EST_DynamicNone) {
+    // TODO: Revisit exception specifications for the MS ABI.  There is a way to
+    // encode these in an object file but MSVC doesn't do anything with it.
+    if (getTarget().getCXXABI().isMicrosoft())
+      return;
     unsigned NumExceptions = Proto->getNumExceptions();
     EHFilterScope *Filter = EHStack.pushFilter(NumExceptions);
 
@@ -543,8 +495,9 @@ static void emitFilterDispatchBlock(CodeGenFunction &CGF,
 
     llvm::Value *zero = CGF.Builder.getInt32(0);
     llvm::Value *failsFilter =
-      CGF.Builder.CreateICmpSLT(selector, zero, "ehspec.fails");
-    CGF.Builder.CreateCondBr(failsFilter, unexpectedBB, CGF.getEHResumeBlock(false));
+        CGF.Builder.CreateICmpSLT(selector, zero, "ehspec.fails");
+    CGF.Builder.CreateCondBr(failsFilter, unexpectedBB,
+                             CGF.getEHResumeBlock(false));
 
     CGF.EmitBlock(unexpectedBB);
   }
@@ -582,6 +535,10 @@ void CodeGenFunction::EmitEndEHSpec(const Decl *D) {
       EHStack.popTerminate();
     }
   } else if (EST == EST_Dynamic || EST == EST_DynamicNone) {
+    // TODO: Revisit exception specifications for the MS ABI.  There is a way to
+    // encode these in an object file but MSVC doesn't do anything with it.
+    if (getTarget().getCXXABI().isMicrosoft())
+      return;
     EHFilterScope &filterScope = cast<EHFilterScope>(*EHStack.begin());
     emitFilterDispatchBlock(*this, filterScope);
     EHStack.popFilter();
@@ -589,11 +546,6 @@ void CodeGenFunction::EmitEndEHSpec(const Decl *D) {
 }
 
 void CodeGenFunction::EmitCXXTryStmt(const CXXTryStmt &S) {
-  if (CGM.getTarget().getTriple().isKnownWindowsMSVCEnvironment()) {
-    ErrorUnsupported(&S, "try statement");
-    return;
-  }
-
   EnterCXXTryStmt(S);
   EmitStmt(S.getTryBlock());
   ExitCXXTryStmt(S);
@@ -622,7 +574,8 @@ void CodeGenFunction::EnterCXXTryStmt(const CXXTryStmt &S, bool IsFnTryBlock) {
       if (CaughtType->isObjCObjectPointerType())
         TypeInfo = CGM.getObjCRuntime().GetEHType(CaughtType);
       else
-        TypeInfo = CGM.GetAddrOfRTTIDescriptor(CaughtType, /*ForEH=*/true);
+        TypeInfo =
+            CGM.getAddrOfCXXCatchHandlerType(CaughtType, C->getCaughtType());
       CatchScope->setHandler(I, TypeInfo, Handler);
     } else {
       // No exception decl indicates '...', a catch-all.
@@ -695,8 +648,15 @@ llvm::BasicBlock *CodeGenFunction::getInvokeDestImpl() {
   assert(EHStack.requiresLandingPad());
   assert(!EHStack.empty());
 
-  if (!CGM.getLangOpts().Exceptions)
-    return nullptr;
+  // If exceptions are disabled, there are usually no landingpads. However, when
+  // SEH is enabled, functions using SEH still get landingpads.
+  const LangOptions &LO = CGM.getLangOpts();
+  if (!LO.Exceptions) {
+    if (!LO.Borland && !LO.MicrosoftExt)
+      return nullptr;
+    if (!currentFunctionUsesSEHTry())
+      return nullptr;
+  }
 
   // Check the innermost scope for a cached landing pad.  If this is
   // a non-EH cleanup, we'll check enclosing scopes in EmitLandingPad.
@@ -734,9 +694,9 @@ llvm::BasicBlock *CodeGenFunction::EmitLandingPad() {
 
   // Save the current IR generation state.
   CGBuilderTy::InsertPoint savedIP = Builder.saveAndClearIP();
-  ApplyDebugLocation AutoRestoreLocation(*this, CurEHLocation);
+  auto DL = ApplyDebugLocation::CreateDefaultArtificial(*this, CurEHLocation);
 
-  const EHPersonality &personality = EHPersonality::get(CGM);
+  const EHPersonality &personality = EHPersonality::get(*this);
 
   // Create and configure the landing pad.
   llvm::BasicBlock *lpad = createBasicBlock("lpad");
@@ -762,8 +722,8 @@ llvm::BasicBlock *CodeGenFunction::EmitLandingPad() {
   bool hasFilter = false;
   SmallVector<llvm::Value*, 4> filterTypes;
   llvm::SmallPtrSet<llvm::Value*, 4> catchTypes;
-  for (EHScopeStack::iterator I = EHStack.begin(), E = EHStack.end();
-         I != E; ++I) {
+  for (EHScopeStack::iterator I = EHStack.begin(), E = EHStack.end(); I != E;
+       ++I) {
 
     switch (I->getKind()) {
     case EHScope::Cleanup:
@@ -857,263 +817,6 @@ llvm::BasicBlock *CodeGenFunction::EmitLandingPad() {
   return lpad;
 }
 
-namespace {
-  /// A cleanup to call __cxa_end_catch.  In many cases, the caught
-  /// exception type lets us state definitively that the thrown exception
-  /// type does not have a destructor.  In particular:
-  ///   - Catch-alls tell us nothing, so we have to conservatively
-  ///     assume that the thrown exception might have a destructor.
-  ///   - Catches by reference behave according to their base types.
-  ///   - Catches of non-record types will only trigger for exceptions
-  ///     of non-record types, which never have destructors.
-  ///   - Catches of record types can trigger for arbitrary subclasses
-  ///     of the caught type, so we have to assume the actual thrown
-  ///     exception type might have a throwing destructor, even if the
-  ///     caught type's destructor is trivial or nothrow.
-  struct CallEndCatch : EHScopeStack::Cleanup {
-    CallEndCatch(bool MightThrow) : MightThrow(MightThrow) {}
-    bool MightThrow;
-
-    void Emit(CodeGenFunction &CGF, Flags flags) override {
-      if (!MightThrow) {
-        CGF.EmitNounwindRuntimeCall(getEndCatchFn(CGF.CGM));
-        return;
-      }
-
-      CGF.EmitRuntimeCallOrInvoke(getEndCatchFn(CGF.CGM));
-    }
-  };
-}
-
-/// Emits a call to __cxa_begin_catch and enters a cleanup to call
-/// __cxa_end_catch.
-///
-/// \param EndMightThrow - true if __cxa_end_catch might throw
-static llvm::Value *CallBeginCatch(CodeGenFunction &CGF,
-                                   llvm::Value *Exn,
-                                   bool EndMightThrow) {
-  llvm::CallInst *call =
-    CGF.EmitNounwindRuntimeCall(getBeginCatchFn(CGF.CGM), Exn);
-
-  CGF.EHStack.pushCleanup<CallEndCatch>(NormalAndEHCleanup, EndMightThrow);
-
-  return call;
-}
-
-/// A "special initializer" callback for initializing a catch
-/// parameter during catch initialization.
-static void InitCatchParam(CodeGenFunction &CGF,
-                           const VarDecl &CatchParam,
-                           llvm::Value *ParamAddr,
-                           SourceLocation Loc) {
-  // Load the exception from where the landing pad saved it.
-  llvm::Value *Exn = CGF.getExceptionFromSlot();
-
-  CanQualType CatchType =
-    CGF.CGM.getContext().getCanonicalType(CatchParam.getType());
-  llvm::Type *LLVMCatchTy = CGF.ConvertTypeForMem(CatchType);
-
-  // If we're catching by reference, we can just cast the object
-  // pointer to the appropriate pointer.
-  if (isa<ReferenceType>(CatchType)) {
-    QualType CaughtType = cast<ReferenceType>(CatchType)->getPointeeType();
-    bool EndCatchMightThrow = CaughtType->isRecordType();
-
-    // __cxa_begin_catch returns the adjusted object pointer.
-    llvm::Value *AdjustedExn = CallBeginCatch(CGF, Exn, EndCatchMightThrow);
-
-    // We have no way to tell the personality function that we're
-    // catching by reference, so if we're catching a pointer,
-    // __cxa_begin_catch will actually return that pointer by value.
-    if (const PointerType *PT = dyn_cast<PointerType>(CaughtType)) {
-      QualType PointeeType = PT->getPointeeType();
-
-      // When catching by reference, generally we should just ignore
-      // this by-value pointer and use the exception object instead.
-      if (!PointeeType->isRecordType()) {
-
-        // Exn points to the struct _Unwind_Exception header, which
-        // we have to skip past in order to reach the exception data.
-        unsigned HeaderSize =
-          CGF.CGM.getTargetCodeGenInfo().getSizeOfUnwindException();
-        AdjustedExn = CGF.Builder.CreateConstGEP1_32(Exn, HeaderSize);
-
-      // However, if we're catching a pointer-to-record type that won't
-      // work, because the personality function might have adjusted
-      // the pointer.  There's actually no way for us to fully satisfy
-      // the language/ABI contract here:  we can't use Exn because it
-      // might have the wrong adjustment, but we can't use the by-value
-      // pointer because it's off by a level of abstraction.
-      //
-      // The current solution is to dump the adjusted pointer into an
-      // alloca, which breaks language semantics (because changing the
-      // pointer doesn't change the exception) but at least works.
-      // The better solution would be to filter out non-exact matches
-      // and rethrow them, but this is tricky because the rethrow
-      // really needs to be catchable by other sites at this landing
-      // pad.  The best solution is to fix the personality function.
-      } else {
-        // Pull the pointer for the reference type off.
-        llvm::Type *PtrTy =
-          cast<llvm::PointerType>(LLVMCatchTy)->getElementType();
-
-        // Create the temporary and write the adjusted pointer into it.
-        llvm::Value *ExnPtrTmp = CGF.CreateTempAlloca(PtrTy, "exn.byref.tmp");
-        llvm::Value *Casted = CGF.Builder.CreateBitCast(AdjustedExn, PtrTy);
-        CGF.Builder.CreateStore(Casted, ExnPtrTmp);
-
-        // Bind the reference to the temporary.
-        AdjustedExn = ExnPtrTmp;
-      }
-    }
-
-    llvm::Value *ExnCast =
-      CGF.Builder.CreateBitCast(AdjustedExn, LLVMCatchTy, "exn.byref");
-    CGF.Builder.CreateStore(ExnCast, ParamAddr);
-    return;
-  }
-
-  // Scalars and complexes.
-  TypeEvaluationKind TEK = CGF.getEvaluationKind(CatchType);
-  if (TEK != TEK_Aggregate) {
-    llvm::Value *AdjustedExn = CallBeginCatch(CGF, Exn, false);
-    
-    // If the catch type is a pointer type, __cxa_begin_catch returns
-    // the pointer by value.
-    if (CatchType->hasPointerRepresentation()) {
-      llvm::Value *CastExn =
-        CGF.Builder.CreateBitCast(AdjustedExn, LLVMCatchTy, "exn.casted");
-
-      switch (CatchType.getQualifiers().getObjCLifetime()) {
-      case Qualifiers::OCL_Strong:
-        CastExn = CGF.EmitARCRetainNonBlock(CastExn);
-        // fallthrough
-
-      case Qualifiers::OCL_None:
-      case Qualifiers::OCL_ExplicitNone:
-      case Qualifiers::OCL_Autoreleasing:
-        CGF.Builder.CreateStore(CastExn, ParamAddr);
-        return;
-
-      case Qualifiers::OCL_Weak:
-        CGF.EmitARCInitWeak(ParamAddr, CastExn);
-        return;
-      }
-      llvm_unreachable("bad ownership qualifier!");
-    }
-
-    // Otherwise, it returns a pointer into the exception object.
-
-    llvm::Type *PtrTy = LLVMCatchTy->getPointerTo(0); // addrspace 0 ok
-    llvm::Value *Cast = CGF.Builder.CreateBitCast(AdjustedExn, PtrTy);
-
-    LValue srcLV = CGF.MakeNaturalAlignAddrLValue(Cast, CatchType);
-    LValue destLV = CGF.MakeAddrLValue(ParamAddr, CatchType,
-                                  CGF.getContext().getDeclAlign(&CatchParam));
-    switch (TEK) {
-    case TEK_Complex:
-      CGF.EmitStoreOfComplex(CGF.EmitLoadOfComplex(srcLV, Loc), destLV,
-                             /*init*/ true);
-      return;
-    case TEK_Scalar: {
-      llvm::Value *ExnLoad = CGF.EmitLoadOfScalar(srcLV, Loc);
-      CGF.EmitStoreOfScalar(ExnLoad, destLV, /*init*/ true);
-      return;
-    }
-    case TEK_Aggregate:
-      llvm_unreachable("evaluation kind filtered out!");
-    }
-    llvm_unreachable("bad evaluation kind");
-  }
-
-  assert(isa<RecordType>(CatchType) && "unexpected catch type!");
-
-  llvm::Type *PtrTy = LLVMCatchTy->getPointerTo(0); // addrspace 0 ok
-
-  // Check for a copy expression.  If we don't have a copy expression,
-  // that means a trivial copy is okay.
-  const Expr *copyExpr = CatchParam.getInit();
-  if (!copyExpr) {
-    llvm::Value *rawAdjustedExn = CallBeginCatch(CGF, Exn, true);
-    llvm::Value *adjustedExn = CGF.Builder.CreateBitCast(rawAdjustedExn, PtrTy);
-    CGF.EmitAggregateCopy(ParamAddr, adjustedExn, CatchType);
-    return;
-  }
-
-  // We have to call __cxa_get_exception_ptr to get the adjusted
-  // pointer before copying.
-  llvm::CallInst *rawAdjustedExn =
-    CGF.EmitNounwindRuntimeCall(getGetExceptionPtrFn(CGF.CGM), Exn);
-
-  // Cast that to the appropriate type.
-  llvm::Value *adjustedExn = CGF.Builder.CreateBitCast(rawAdjustedExn, PtrTy);
-
-  // The copy expression is defined in terms of an OpaqueValueExpr.
-  // Find it and map it to the adjusted expression.
-  CodeGenFunction::OpaqueValueMapping
-    opaque(CGF, OpaqueValueExpr::findInCopyConstruct(copyExpr),
-           CGF.MakeAddrLValue(adjustedExn, CatchParam.getType()));
-
-  // Call the copy ctor in a terminate scope.
-  CGF.EHStack.pushTerminate();
-
-  // Perform the copy construction.
-  CharUnits Alignment = CGF.getContext().getDeclAlign(&CatchParam);
-  CGF.EmitAggExpr(copyExpr,
-                  AggValueSlot::forAddr(ParamAddr, Alignment, Qualifiers(),
-                                        AggValueSlot::IsNotDestructed,
-                                        AggValueSlot::DoesNotNeedGCBarriers,
-                                        AggValueSlot::IsNotAliased));
-
-  // Leave the terminate scope.
-  CGF.EHStack.popTerminate();
-
-  // Undo the opaque value mapping.
-  opaque.pop();
-
-  // Finally we can call __cxa_begin_catch.
-  CallBeginCatch(CGF, Exn, true);
-}
-
-/// Begins a catch statement by initializing the catch variable and
-/// calling __cxa_begin_catch.
-static void BeginCatch(CodeGenFunction &CGF, const CXXCatchStmt *S) {
-  // We have to be very careful with the ordering of cleanups here:
-  //   C++ [except.throw]p4:
-  //     The destruction [of the exception temporary] occurs
-  //     immediately after the destruction of the object declared in
-  //     the exception-declaration in the handler.
-  //
-  // So the precise ordering is:
-  //   1.  Construct catch variable.
-  //   2.  __cxa_begin_catch
-  //   3.  Enter __cxa_end_catch cleanup
-  //   4.  Enter dtor cleanup
-  //
-  // We do this by using a slightly abnormal initialization process.
-  // Delegation sequence:
-  //   - ExitCXXTryStmt opens a RunCleanupsScope
-  //     - EmitAutoVarAlloca creates the variable and debug info
-  //       - InitCatchParam initializes the variable from the exception
-  //       - CallBeginCatch calls __cxa_begin_catch
-  //       - CallBeginCatch enters the __cxa_end_catch cleanup
-  //     - EmitAutoVarCleanups enters the variable destructor cleanup
-  //   - EmitCXXTryStmt emits the code for the catch body
-  //   - EmitCXXTryStmt close the RunCleanupsScope
-
-  VarDecl *CatchParam = S->getExceptionDecl();
-  if (!CatchParam) {
-    llvm::Value *Exn = CGF.getExceptionFromSlot();
-    CallBeginCatch(CGF, Exn, true);
-    return;
-  }
-
-  // Emit the local.
-  CodeGenFunction::AutoVarEmission var = CGF.EmitAutoVarAlloca(*CatchParam);
-  InitCatchParam(CGF, *CatchParam, var.getObjectAddress(CGF), S->getLocStart());
-  CGF.EmitAutoVarCleanups(var);
-}
-
 /// Emit the structure of the dispatch block for the given catch scope.
 /// It is an invariant that the dispatch block already exists.
 static void emitCatchDispatchBlock(CodeGenFunction &CGF,
@@ -1252,11 +955,10 @@ void CodeGenFunction::ExitCXXTryStmt(const CXXTryStmt &S, bool IsFnTryBlock) {
     RunCleanupsScope CatchScope(*this);
 
     // Initialize the catch variable and set up the cleanups.
-    BeginCatch(*this, C);
+    CGM.getCXXABI().emitBeginCatch(*this, C);
 
     // Emit the PGO counter increment.
-    RegionCounter CatchCnt = getPGORegionCounter(C);
-    CatchCnt.beginRegion(Builder);
+    incrementProfileCounter(C);
 
     // Perform the body of the catch.
     EmitStmt(C->getHandlerBlock());
@@ -1284,9 +986,8 @@ void CodeGenFunction::ExitCXXTryStmt(const CXXTryStmt &S, bool IsFnTryBlock) {
       Builder.CreateBr(ContBB);
   }
 
-  RegionCounter ContCnt = getPGORegionCounter(&S);
   EmitBlock(ContBB);
-  ContCnt.beginRegion(Builder);
+  incrementProfileCounter(&S);
 }
 
 namespace {
@@ -1480,68 +1181,6 @@ void CodeGenFunction::FinallyInfo::exit(CodeGenFunction &CGF) {
   CGF.PopCleanupBlock();
 }
 
-/// In a terminate landing pad, should we use __clang__call_terminate
-/// or just a naked call to std::terminate?
-///
-/// __clang_call_terminate calls __cxa_begin_catch, which then allows
-/// std::terminate to usefully report something about the
-/// violating exception.
-static bool useClangCallTerminate(CodeGenModule &CGM) {
-  // Only do this for Itanium-family ABIs in C++ mode.
-  return (CGM.getLangOpts().CPlusPlus &&
-          CGM.getTarget().getCXXABI().isItaniumFamily());
-}
-
-/// Get or define the following function:
-///   void @__clang_call_terminate(i8* %exn) nounwind noreturn
-/// This code is used only in C++.
-static llvm::Constant *getClangCallTerminateFn(CodeGenModule &CGM) {
-  llvm::FunctionType *fnTy =
-    llvm::FunctionType::get(CGM.VoidTy, CGM.Int8PtrTy, /*IsVarArgs=*/false);
-  llvm::Constant *fnRef =
-    CGM.CreateRuntimeFunction(fnTy, "__clang_call_terminate");
-
-  llvm::Function *fn = dyn_cast<llvm::Function>(fnRef);
-  if (fn && fn->empty()) {
-    fn->setDoesNotThrow();
-    fn->setDoesNotReturn();
-
-    // What we really want is to massively penalize inlining without
-    // forbidding it completely.  The difference between that and
-    // 'noinline' is negligible.
-    fn->addFnAttr(llvm::Attribute::NoInline);
-
-    // Allow this function to be shared across translation units, but
-    // we don't want it to turn into an exported symbol.
-    fn->setLinkage(llvm::Function::LinkOnceODRLinkage);
-    fn->setVisibility(llvm::Function::HiddenVisibility);
-
-    // Set up the function.
-    llvm::BasicBlock *entry =
-      llvm::BasicBlock::Create(CGM.getLLVMContext(), "", fn);
-    CGBuilderTy builder(entry);
-
-    // Pull the exception pointer out of the parameter list.
-    llvm::Value *exn = &*fn->arg_begin();
-
-    // Call __cxa_begin_catch(exn).
-    llvm::CallInst *catchCall = builder.CreateCall(getBeginCatchFn(CGM), exn);
-    catchCall->setDoesNotThrow();
-    catchCall->setCallingConv(CGM.getRuntimeCC());
-
-    // Call std::terminate().
-    llvm::CallInst *termCall = builder.CreateCall(getTerminateFn(CGM));
-    termCall->setDoesNotThrow();
-    termCall->setDoesNotReturn();
-    termCall->setCallingConv(CGM.getRuntimeCC());
-
-    // std::terminate cannot return.
-    builder.CreateUnreachable();
-  }
-
-  return fnRef;
-}
-
 llvm::BasicBlock *CodeGenFunction::getTerminateLandingPad() {
   if (TerminateLandingPad)
     return TerminateLandingPad;
@@ -1553,20 +1192,17 @@ llvm::BasicBlock *CodeGenFunction::getTerminateLandingPad() {
   Builder.SetInsertPoint(TerminateLandingPad);
 
   // Tell the backend that this is a landing pad.
-  const EHPersonality &Personality = EHPersonality::get(CGM);
+  const EHPersonality &Personality = EHPersonality::get(*this);
   llvm::LandingPadInst *LPadInst =
     Builder.CreateLandingPad(llvm::StructType::get(Int8PtrTy, Int32Ty, nullptr),
                              getOpaquePersonalityFn(CGM, Personality), 0);
   LPadInst->addClause(getCatchAllValue(*this));
 
-  llvm::CallInst *terminateCall;
-  if (useClangCallTerminate(CGM)) {
-    // Extract out the exception pointer.
-    llvm::Value *exn = Builder.CreateExtractValue(LPadInst, 0);
-    terminateCall = EmitNounwindRuntimeCall(getClangCallTerminateFn(CGM), exn);
-  } else {
-    terminateCall = EmitNounwindRuntimeCall(getTerminateFn(CGM));
-  }
+  llvm::Value *Exn = 0;
+  if (getLangOpts().CPlusPlus)
+    Exn = Builder.CreateExtractValue(LPadInst, 0);
+  llvm::CallInst *terminateCall =
+      CGM.getCXXABI().emitTerminateForUnexpectedException(*this, Exn);
   terminateCall->setDoesNotReturn();
   Builder.CreateUnreachable();
 
@@ -1586,14 +1222,11 @@ llvm::BasicBlock *CodeGenFunction::getTerminateHandler() {
   // end of the function by FinishFunction.
   TerminateHandler = createBasicBlock("terminate.handler");
   Builder.SetInsertPoint(TerminateHandler);
-  llvm::CallInst *terminateCall;
-  if (useClangCallTerminate(CGM)) {
-    // Load the exception pointer.
-    llvm::Value *exn = getExceptionFromSlot();
-    terminateCall = EmitNounwindRuntimeCall(getClangCallTerminateFn(CGM), exn);
-  } else {
-    terminateCall = EmitNounwindRuntimeCall(getTerminateFn(CGM));
-  }
+  llvm::Value *Exn = 0;
+  if (getLangOpts().CPlusPlus)
+    Exn = getExceptionFromSlot();
+  llvm::CallInst *terminateCall =
+      CGM.getCXXABI().emitTerminateForUnexpectedException(*this, Exn);
   terminateCall->setDoesNotReturn();
   Builder.CreateUnreachable();
 
@@ -1612,15 +1245,14 @@ llvm::BasicBlock *CodeGenFunction::getEHResumeBlock(bool isCleanup) {
   EHResumeBlock = createBasicBlock("eh.resume");
   Builder.SetInsertPoint(EHResumeBlock);
 
-  const EHPersonality &Personality = EHPersonality::get(CGM);
+  const EHPersonality &Personality = EHPersonality::get(*this);
 
   // This can always be a call because we necessarily didn't find
   // anything on the EH stack which needs our help.
   const char *RethrowName = Personality.CatchallRethrowFn;
   if (RethrowName != nullptr && !isCleanup) {
     EmitRuntimeCall(getCatchallRethrowFn(CGM, RethrowName),
-                    getExceptionFromSlot())
-      ->setDoesNotReturn();
+                    getExceptionFromSlot())->setDoesNotReturn();
     Builder.CreateUnreachable();
     Builder.restoreIP(SavedIP);
     return EHResumeBlock;
@@ -1642,9 +1274,433 @@ llvm::BasicBlock *CodeGenFunction::getEHResumeBlock(bool isCleanup) {
 }
 
 void CodeGenFunction::EmitSEHTryStmt(const SEHTryStmt &S) {
-  CGM.ErrorUnsupported(&S, "SEH __try");
+  // FIXME: Implement SEH on other architectures.
+  const llvm::Triple &T = CGM.getTarget().getTriple();
+  if (T.getArch() != llvm::Triple::x86_64 ||
+      !T.isKnownWindowsMSVCEnvironment()) {
+    ErrorUnsupported(&S, "__try statement");
+    return;
+  }
+
+  EnterSEHTryStmt(S);
+  {
+    JumpDest TryExit = getJumpDestInCurrentScope("__try.__leave");
+
+    SEHTryEpilogueStack.push_back(&TryExit);
+    EmitStmt(S.getTryBlock());
+    SEHTryEpilogueStack.pop_back();
+
+    if (!TryExit.getBlock()->use_empty())
+      EmitBlock(TryExit.getBlock(), /*IsFinished=*/true);
+    else
+      delete TryExit.getBlock();
+  }
+  ExitSEHTryStmt(S);
+}
+
+namespace {
+struct PerformSEHFinally : EHScopeStack::Cleanup {
+  llvm::Function *OutlinedFinally;
+  PerformSEHFinally(llvm::Function *OutlinedFinally)
+      : OutlinedFinally(OutlinedFinally) {}
+
+  void Emit(CodeGenFunction &CGF, Flags F) override {
+    ASTContext &Context = CGF.getContext();
+    QualType ArgTys[2] = {Context.UnsignedCharTy, Context.VoidPtrTy};
+    FunctionProtoType::ExtProtoInfo EPI;
+    const auto *FTP = cast<FunctionType>(
+        Context.getFunctionType(Context.VoidTy, ArgTys, EPI));
+
+    CallArgList Args;
+    llvm::Value *IsForEH =
+        llvm::ConstantInt::get(CGF.ConvertType(ArgTys[0]), F.isForEHCleanup());
+    Args.add(RValue::get(IsForEH), ArgTys[0]);
+
+    CodeGenModule &CGM = CGF.CGM;
+    llvm::Value *Zero = llvm::ConstantInt::get(CGM.Int32Ty, 0);
+    llvm::Value *FrameAddr = CGM.getIntrinsic(llvm::Intrinsic::frameaddress);
+    llvm::Value *FP = CGF.Builder.CreateCall(FrameAddr, Zero);
+    Args.add(RValue::get(FP), ArgTys[1]);
+
+    const CGFunctionInfo &FnInfo =
+        CGM.getTypes().arrangeFreeFunctionCall(Args, FTP, /*chainCall=*/false);
+    CGF.EmitCall(FnInfo, OutlinedFinally, ReturnValueSlot(), Args);
+  }
+};
+}
+
+namespace {
+/// Find all local variable captures in the statement.
+struct CaptureFinder : ConstStmtVisitor<CaptureFinder> {
+  CodeGenFunction &ParentCGF;
+  const VarDecl *ParentThis;
+  SmallVector<const VarDecl *, 4> Captures;
+  CaptureFinder(CodeGenFunction &ParentCGF, const VarDecl *ParentThis)
+      : ParentCGF(ParentCGF), ParentThis(ParentThis) {}
+
+  void Visit(const Stmt *S) {
+    // See if this is a capture, then recurse.
+    ConstStmtVisitor<CaptureFinder>::Visit(S);
+    for (const Stmt *Child : S->children())
+      if (Child)
+        Visit(Child);
+  }
+
+  void VisitDeclRefExpr(const DeclRefExpr *E) {
+    // If this is already a capture, just make sure we capture 'this'.
+    if (E->refersToEnclosingVariableOrCapture()) {
+      Captures.push_back(ParentThis);
+      return;
+    }
+
+    const auto *D = dyn_cast<VarDecl>(E->getDecl());
+    if (D && D->isLocalVarDeclOrParm() && D->hasLocalStorage())
+      Captures.push_back(D);
+  }
+
+  void VisitCXXThisExpr(const CXXThisExpr *E) {
+    Captures.push_back(ParentThis);
+  }
+};
+}
+
+void CodeGenFunction::EmitCapturedLocals(CodeGenFunction &ParentCGF,
+                                         const Stmt *OutlinedStmt,
+                                         llvm::Value *ParentFP) {
+  // Find all captures in the Stmt.
+  CaptureFinder Finder(ParentCGF, ParentCGF.CXXABIThisDecl);
+  Finder.Visit(OutlinedStmt);
+
+  // Typically there are no captures and we can exit early.
+  if (Finder.Captures.empty())
+    return;
+
+  // Prepare the first two arguments to llvm.framerecover.
+  llvm::Function *FrameRecoverFn = llvm::Intrinsic::getDeclaration(
+      &CGM.getModule(), llvm::Intrinsic::framerecover);
+  llvm::Constant *ParentI8Fn =
+      llvm::ConstantExpr::getBitCast(ParentCGF.CurFn, Int8PtrTy);
+
+  // Create llvm.framerecover calls for all captures.
+  for (const VarDecl *VD : Finder.Captures) {
+    if (isa<ImplicitParamDecl>(VD)) {
+      CGM.ErrorUnsupported(VD, "'this' captured by SEH");
+      CXXThisValue = llvm::UndefValue::get(ConvertTypeForMem(VD->getType()));
+      continue;
+    }
+    if (VD->getType()->isVariablyModifiedType()) {
+      CGM.ErrorUnsupported(VD, "VLA captured by SEH");
+      continue;
+    }
+    assert((isa<ImplicitParamDecl>(VD) || VD->isLocalVarDeclOrParm()) &&
+           "captured non-local variable");
+
+    // If this decl hasn't been declared yet, it will be declared in the
+    // OutlinedStmt.
+    auto I = ParentCGF.LocalDeclMap.find(VD);
+    if (I == ParentCGF.LocalDeclMap.end())
+      continue;
+    llvm::Value *ParentVar = I->second;
+
+    llvm::CallInst *RecoverCall = nullptr;
+    CGBuilderTy Builder(AllocaInsertPt);
+    if (auto *ParentAlloca = dyn_cast<llvm::AllocaInst>(ParentVar)) {
+      // Mark the variable escaped if nobody else referenced it and compute the
+      // frameescape index.
+      auto InsertPair =
+          ParentCGF.EscapedLocals.insert(std::make_pair(ParentAlloca, -1));
+      if (InsertPair.second)
+        InsertPair.first->second = ParentCGF.EscapedLocals.size() - 1;
+      int FrameEscapeIdx = InsertPair.first->second;
+      // call i8* @llvm.framerecover(i8* bitcast(@parentFn), i8* %fp, i32 N)
+      RecoverCall = Builder.CreateCall(
+          FrameRecoverFn, {ParentI8Fn, ParentFP,
+                           llvm::ConstantInt::get(Int32Ty, FrameEscapeIdx)});
+
+    } else {
+      // If the parent didn't have an alloca, we're doing some nested outlining.
+      // Just clone the existing framerecover call, but tweak the FP argument to
+      // use our FP value. All other arguments are constants.
+      auto *ParentRecover =
+          cast<llvm::IntrinsicInst>(ParentVar->stripPointerCasts());
+      assert(ParentRecover->getIntrinsicID() == llvm::Intrinsic::framerecover &&
+             "expected alloca or framerecover in parent LocalDeclMap");
+      RecoverCall = cast<llvm::CallInst>(ParentRecover->clone());
+      RecoverCall->setArgOperand(1, ParentFP);
+      RecoverCall->insertBefore(AllocaInsertPt);
+    }
+
+    // Bitcast the variable, rename it, and insert it in the local decl map.
+    llvm::Value *ChildVar =
+        Builder.CreateBitCast(RecoverCall, ParentVar->getType());
+    ChildVar->setName(ParentVar->getName());
+    LocalDeclMap[VD] = ChildVar;
+  }
+}
+
+/// Arrange a function prototype that can be called by Windows exception
+/// handling personalities. On Win64, the prototype looks like:
+/// RetTy func(void *EHPtrs, void *ParentFP);
+void CodeGenFunction::startOutlinedSEHHelper(CodeGenFunction &ParentCGF,
+                                             StringRef Name, QualType RetTy,
+                                             FunctionArgList &Args,
+                                             const Stmt *OutlinedStmt) {
+  llvm::Function *ParentFn = ParentCGF.CurFn;
+  const CGFunctionInfo &FnInfo = CGM.getTypes().arrangeFreeFunctionDeclaration(
+      RetTy, Args, FunctionType::ExtInfo(), /*isVariadic=*/false);
+
+  llvm::FunctionType *FnTy = CGM.getTypes().GetFunctionType(FnInfo);
+  llvm::Function *Fn = llvm::Function::Create(
+      FnTy, llvm::GlobalValue::InternalLinkage, Name.str(), &CGM.getModule());
+  // The filter is either in the same comdat as the function, or it's internal.
+  if (llvm::Comdat *C = ParentFn->getComdat()) {
+    Fn->setComdat(C);
+  } else if (ParentFn->hasWeakLinkage() || ParentFn->hasLinkOnceLinkage()) {
+    llvm::Comdat *C = CGM.getModule().getOrInsertComdat(ParentFn->getName());
+    ParentFn->setComdat(C);
+    Fn->setComdat(C);
+  } else {
+    Fn->setLinkage(llvm::GlobalValue::InternalLinkage);
+  }
+
+  IsOutlinedSEHHelper = true;
+
+  StartFunction(GlobalDecl(), RetTy, Fn, FnInfo, Args,
+                OutlinedStmt->getLocStart(), OutlinedStmt->getLocStart());
+
+  CGM.SetLLVMFunctionAttributes(nullptr, FnInfo, CurFn);
+
+  auto AI = Fn->arg_begin();
+  ++AI;
+  EmitCapturedLocals(ParentCGF, OutlinedStmt, &*AI);
+}
+
+/// Create a stub filter function that will ultimately hold the code of the
+/// filter expression. The EH preparation passes in LLVM will outline the code
+/// from the main function body into this stub.
+llvm::Function *
+CodeGenFunction::GenerateSEHFilterFunction(CodeGenFunction &ParentCGF,
+                                           const SEHExceptStmt &Except) {
+  const Expr *FilterExpr = Except.getFilterExpr();
+  SourceLocation StartLoc = FilterExpr->getLocStart();
+
+  SEHPointersDecl = ImplicitParamDecl::Create(
+      getContext(), nullptr, StartLoc,
+      &getContext().Idents.get("exception_pointers"), getContext().VoidPtrTy);
+  FunctionArgList Args;
+  Args.push_back(SEHPointersDecl);
+  Args.push_back(ImplicitParamDecl::Create(
+      getContext(), nullptr, StartLoc,
+      &getContext().Idents.get("frame_pointer"), getContext().VoidPtrTy));
+
+  // Get the mangled function name.
+  SmallString<128> Name;
+  {
+    llvm::raw_svector_ostream OS(Name);
+    const Decl *ParentCodeDecl = ParentCGF.CurCodeDecl;
+    const NamedDecl *Parent = dyn_cast_or_null<NamedDecl>(ParentCodeDecl);
+    assert(Parent && "FIXME: handle unnamed decls (lambdas, blocks) with SEH");
+    CGM.getCXXABI().getMangleContext().mangleSEHFilterExpression(Parent, OS);
+  }
+
+  startOutlinedSEHHelper(ParentCGF, Name, getContext().LongTy, Args,
+                         FilterExpr);
+
+  // Mark finally block calls as nounwind and noinline to make LLVM's job a
+  // little easier.
+  // FIXME: Remove these restrictions in the future.
+  CurFn->addFnAttr(llvm::Attribute::NoUnwind);
+  CurFn->addFnAttr(llvm::Attribute::NoInline);
+
+  EmitSEHExceptionCodeSave();
+
+  // Emit the original filter expression, convert to i32, and return.
+  llvm::Value *R = EmitScalarExpr(FilterExpr);
+  R = Builder.CreateIntCast(R, ConvertType(getContext().LongTy),
+                            FilterExpr->getType()->isSignedIntegerType());
+  Builder.CreateStore(R, ReturnValue);
+
+  FinishFunction(FilterExpr->getLocEnd());
+
+  return CurFn;
+}
+
+llvm::Function *
+CodeGenFunction::GenerateSEHFinallyFunction(CodeGenFunction &ParentCGF,
+                                            const SEHFinallyStmt &Finally) {
+  const Stmt *FinallyBlock = Finally.getBlock();
+  SourceLocation StartLoc = FinallyBlock->getLocStart();
+
+  FunctionArgList Args;
+  Args.push_back(ImplicitParamDecl::Create(
+      getContext(), nullptr, StartLoc,
+      &getContext().Idents.get("abnormal_termination"),
+      getContext().UnsignedCharTy));
+  Args.push_back(ImplicitParamDecl::Create(
+      getContext(), nullptr, StartLoc,
+      &getContext().Idents.get("frame_pointer"), getContext().VoidPtrTy));
+
+  // Get the mangled function name.
+  SmallString<128> Name;
+  {
+    llvm::raw_svector_ostream OS(Name);
+    const Decl *ParentCodeDecl = ParentCGF.CurCodeDecl;
+    const NamedDecl *Parent = dyn_cast_or_null<NamedDecl>(ParentCodeDecl);
+    assert(Parent && "FIXME: handle unnamed decls (lambdas, blocks) with SEH");
+    CGM.getCXXABI().getMangleContext().mangleSEHFinallyBlock(Parent, OS);
+  }
+
+  startOutlinedSEHHelper(ParentCGF, Name, getContext().VoidTy, Args,
+                         FinallyBlock);
+
+  // Emit the original filter expression, convert to i32, and return.
+  EmitStmt(FinallyBlock);
+
+  FinishFunction(FinallyBlock->getLocEnd());
+
+  return CurFn;
+}
+
+void CodeGenFunction::EmitSEHExceptionCodeSave() {
+  // Save the exception code in the exception slot to unify exception access in
+  // the filter function and the landing pad.
+  // struct EXCEPTION_POINTERS {
+  //   EXCEPTION_RECORD *ExceptionRecord;
+  //   CONTEXT *ContextRecord;
+  // };
+  // void *exn.slot =
+  //     (void *)(uintptr_t)exception_pointers->ExceptionRecord->ExceptionCode;
+  llvm::Value *Ptrs = Builder.CreateLoad(GetAddrOfLocalVar(SEHPointersDecl));
+  llvm::Type *RecordTy = CGM.Int32Ty->getPointerTo();
+  llvm::Type *PtrsTy = llvm::StructType::get(RecordTy, CGM.VoidPtrTy, nullptr);
+  Ptrs = Builder.CreateBitCast(Ptrs, PtrsTy->getPointerTo());
+  llvm::Value *Rec = Builder.CreateStructGEP(PtrsTy, Ptrs, 0);
+  Rec = Builder.CreateLoad(Rec);
+  llvm::Value *Code = Builder.CreateLoad(Rec);
+  Code = Builder.CreateZExt(Code, CGM.IntPtrTy);
+  // FIXME: Change landing pads to produce {i32, i32} and make the exception
+  // slot an i32.
+  Code = Builder.CreateIntToPtr(Code, CGM.VoidPtrTy);
+  Builder.CreateStore(Code, getExceptionSlot());
+}
+
+llvm::Value *CodeGenFunction::EmitSEHExceptionInfo() {
+  // Sema should diagnose calling this builtin outside of a filter context, but
+  // don't crash if we screw up.
+  if (!SEHPointersDecl)
+    return llvm::UndefValue::get(Int8PtrTy);
+  return Builder.CreateLoad(GetAddrOfLocalVar(SEHPointersDecl));
+}
+
+llvm::Value *CodeGenFunction::EmitSEHExceptionCode() {
+  // If we're in a landing pad or filter function, the exception slot contains
+  // the code.
+  assert(ExceptionSlot);
+  llvm::Value *Code =
+      Builder.CreatePtrToInt(getExceptionFromSlot(), CGM.IntPtrTy);
+  return Builder.CreateTrunc(Code, CGM.Int32Ty);
+}
+
+llvm::Value *CodeGenFunction::EmitSEHAbnormalTermination() {
+  // Abnormal termination is just the first parameter to the outlined finally
+  // helper.
+  auto AI = CurFn->arg_begin();
+  return Builder.CreateZExt(&*AI, Int32Ty);
+}
+
+void CodeGenFunction::EnterSEHTryStmt(const SEHTryStmt &S) {
+  CodeGenFunction HelperCGF(CGM, /*suppressNewContext=*/true);
+  if (const SEHFinallyStmt *Finally = S.getFinallyHandler()) {
+    // Push a cleanup for __finally blocks.
+    llvm::Function *FinallyFunc =
+        HelperCGF.GenerateSEHFinallyFunction(*this, *Finally);
+    EHStack.pushCleanup<PerformSEHFinally>(NormalAndEHCleanup, FinallyFunc);
+    return;
+  }
+
+  // Otherwise, we must have an __except block.
+  const SEHExceptStmt *Except = S.getExceptHandler();
+  assert(Except);
+  EHCatchScope *CatchScope = EHStack.pushCatch(1);
+
+  // If the filter is known to evaluate to 1, then we can use the clause "catch
+  // i8* null".
+  llvm::Constant *C =
+      CGM.EmitConstantExpr(Except->getFilterExpr(), getContext().IntTy, this);
+  if (C && C->isOneValue()) {
+    CatchScope->setCatchAllHandler(0, createBasicBlock("__except"));
+    return;
+  }
+
+  // In general, we have to emit an outlined filter function. Use the function
+  // in place of the RTTI typeinfo global that C++ EH uses.
+  llvm::Function *FilterFunc =
+      HelperCGF.GenerateSEHFilterFunction(*this, *Except);
+  llvm::Constant *OpaqueFunc =
+      llvm::ConstantExpr::getBitCast(FilterFunc, Int8PtrTy);
+  CatchScope->setHandler(0, OpaqueFunc, createBasicBlock("__except"));
+}
+
+void CodeGenFunction::ExitSEHTryStmt(const SEHTryStmt &S) {
+  // Just pop the cleanup if it's a __finally block.
+  if (S.getFinallyHandler()) {
+    PopCleanupBlock();
+    return;
+  }
+
+  // Otherwise, we must have an __except block.
+  const SEHExceptStmt *Except = S.getExceptHandler();
+  assert(Except && "__try must have __finally xor __except");
+  EHCatchScope &CatchScope = cast<EHCatchScope>(*EHStack.begin());
+
+  // Don't emit the __except block if the __try block lacked invokes.
+  // TODO: Model unwind edges from instructions, either with iload / istore or
+  // a try body function.
+  if (!CatchScope.hasEHBranches()) {
+    CatchScope.clearHandlerBlocks();
+    EHStack.popCatch();
+    return;
+  }
+
+  // The fall-through block.
+  llvm::BasicBlock *ContBB = createBasicBlock("__try.cont");
+
+  // We just emitted the body of the __try; jump to the continue block.
+  if (HaveInsertPoint())
+    Builder.CreateBr(ContBB);
+
+  // Check if our filter function returned true.
+  emitCatchDispatchBlock(*this, CatchScope);
+
+  // Grab the block before we pop the handler.
+  llvm::BasicBlock *ExceptBB = CatchScope.getHandler(0).Block;
+  EHStack.popCatch();
+
+  EmitBlockAfterUses(ExceptBB);
+
+  // Emit the __except body.
+  EmitStmt(Except->getBlock());
+
+  if (HaveInsertPoint())
+    Builder.CreateBr(ContBB);
+
+  EmitBlock(ContBB);
 }
 
 void CodeGenFunction::EmitSEHLeaveStmt(const SEHLeaveStmt &S) {
-  CGM.ErrorUnsupported(&S, "SEH __leave");
+  // If this code is reachable then emit a stop point (if generating
+  // debug info). We have to do this ourselves because we are on the
+  // "simple" statement path.
+  if (HaveInsertPoint())
+    EmitStopPoint(&S);
+
+  // This must be a __leave from a __finally block, which we warn on and is UB.
+  // Just emit unreachable.
+  if (!isSEHTryScope()) {
+    Builder.CreateUnreachable();
+    Builder.ClearInsertionPoint();
+    return;
+  }
+
+  EmitBranchThroughCleanup(*SEHTryEpilogueStack.back());
 }
diff --git a/lib/CodeGen/CGExpr.cpp b/lib/CodeGen/CGExpr.cpp
index ce7679c836e4..1ed45a33d0fd 100644
--- a/lib/CodeGen/CGExpr.cpp
+++ b/lib/CodeGen/CGExpr.cpp
@@ -31,6 +31,7 @@
 #include "llvm/IR/LLVMContext.h"
 #include "llvm/IR/MDBuilder.h"
 #include "llvm/Support/ConvertUTF.h"
+#include "llvm/Support/MathExtras.h"
 
 using namespace clang;
 using namespace CodeGen;
@@ -300,9 +301,26 @@ createReferenceTemporary(CodeGenFunction &CGF,
                          const MaterializeTemporaryExpr *M, const Expr *Inner) {
   switch (M->getStorageDuration()) {
   case SD_FullExpression:
-  case SD_Automatic:
-    return CGF.CreateMemTemp(Inner->getType(), "ref.tmp");
-
+  case SD_Automatic: {
+    // If we have a constant temporary array or record try to promote it into a
+    // constant global under the same rules a normal constant would've been
+    // promoted. This is easier on the optimizer and generally emits fewer
+    // instructions.
+    QualType Ty = Inner->getType();
+    if (CGF.CGM.getCodeGenOpts().MergeAllConstants &&
+        (Ty->isArrayType() || Ty->isRecordType()) &&
+        CGF.CGM.isTypeConstant(Ty, true))
+      if (llvm::Constant *Init = CGF.CGM.EmitConstantExpr(Inner, Ty, &CGF)) {
+        auto *GV = new llvm::GlobalVariable(
+            CGF.CGM.getModule(), Init->getType(), /*isConstant=*/true,
+            llvm::GlobalValue::PrivateLinkage, Init, ".ref.tmp");
+        GV->setAlignment(
+            CGF.getContext().getTypeAlignInChars(Ty).getQuantity());
+        // FIXME: Should we put the new global into a COMDAT?
+        return GV;
+      }
+    return CGF.CreateMemTemp(Ty, "ref.tmp");
+  }
   case SD_Thread:
   case SD_Static:
     return CGF.CGM.GetAddrOfGlobalTemporary(M, Inner);
@@ -324,14 +342,15 @@ EmitMaterializeTemporaryExpr(const MaterializeTemporaryExpr *M) {
       M->getType().getObjCLifetime() != Qualifiers::OCL_None &&
       M->getType().getObjCLifetime() != Qualifiers::OCL_ExplicitNone) {
     llvm::Value *Object = createReferenceTemporary(*this, M, E);
-    LValue RefTempDst = MakeAddrLValue(Object, M->getType());
-
     if (auto *Var = dyn_cast<llvm::GlobalVariable>(Object)) {
+      Object = llvm::ConstantExpr::getBitCast(
+          Var, ConvertTypeForMem(E->getType())->getPointerTo());
       // We should not have emitted the initializer for this temporary as a
       // constant.
       assert(!Var->hasInitializer());
       Var->setInitializer(CGM.EmitNullConstant(E->getType()));
     }
+    LValue RefTempDst = MakeAddrLValue(Object, M->getType());
 
     switch (getEvaluationKind(E->getType())) {
     default: llvm_unreachable("expected scalar or aggregate expression");
@@ -370,8 +389,11 @@ EmitMaterializeTemporaryExpr(const MaterializeTemporaryExpr *M) {
   // Create and initialize the reference temporary.
   llvm::Value *Object = createReferenceTemporary(*this, M, E);
   if (auto *Var = dyn_cast<llvm::GlobalVariable>(Object)) {
-    // If the temporary is a global and has a constant initializer, we may
-    // have already initialized it.
+    Object = llvm::ConstantExpr::getBitCast(
+        Var, ConvertTypeForMem(E->getType())->getPointerTo());
+    // If the temporary is a global and has a constant initializer or is a
+    // constant temporary that we promoted to a global, we may have already
+    // initialized it.
     if (!Var->hasInitializer()) {
       Var->setInitializer(CGM.EmitNullConstant(E->getType()));
       EmitAnyExprToMem(E, Object, Qualifiers(), /*IsInit*/true);
@@ -478,7 +500,7 @@ void CodeGenFunction::EmitTypeCheck(TypeCheckKind TCK, SourceLocation Loc,
 
   SanitizerScope SanScope(this);
 
-  SmallVector<std::pair<llvm::Value *, SanitizerKind>, 3> Checks;
+  SmallVector<std::pair<llvm::Value *, SanitizerMask>, 3> Checks;
   llvm::BasicBlock *Done = nullptr;
 
   bool AllowNullPointers = TCK == TCK_DowncastPointer || TCK == TCK_Upcast ||
@@ -513,7 +535,7 @@ void CodeGenFunction::EmitTypeCheck(TypeCheckKind TCK, SourceLocation Loc,
     llvm::Value *Min = Builder.getFalse();
     llvm::Value *CastAddr = Builder.CreateBitCast(Address, Int8PtrTy);
     llvm::Value *LargeEnough =
-        Builder.CreateICmpUGE(Builder.CreateCall2(F, CastAddr, Min),
+        Builder.CreateICmpUGE(Builder.CreateCall(F, {CastAddr, Min}),
                               llvm::ConstantInt::get(IntPtrTy, Size));
     Checks.push_back(std::make_pair(LargeEnough, SanitizerKind::ObjectSize));
   }
@@ -807,6 +829,7 @@ LValue CodeGenFunction::EmitCheckedLValue(const Expr *E, TypeCheckKind TCK) {
 /// length type, this is not possible.
 ///
 LValue CodeGenFunction::EmitLValue(const Expr *E) {
+  ApplyDebugLocation DL(*this, E);
   switch (E->getStmtClass()) {
   default: return EmitUnsupportedLValue(E, "l-value expression");
 
@@ -819,10 +842,14 @@ LValue CodeGenFunction::EmitLValue(const Expr *E) {
     return EmitObjCIsaExpr(cast<ObjCIsaExpr>(E));
   case Expr::BinaryOperatorClass:
     return EmitBinaryOperatorLValue(cast<BinaryOperator>(E));
-  case Expr::CompoundAssignOperatorClass:
-    if (!E->getType()->isAnyComplexType())
+  case Expr::CompoundAssignOperatorClass: {
+    QualType Ty = E->getType();
+    if (const AtomicType *AT = Ty->getAs<AtomicType>())
+      Ty = AT->getValueType();
+    if (!Ty->isAnyComplexType())
       return EmitCompoundAssignmentLValue(cast<CompoundAssignOperator>(E));
     return EmitComplexCompoundAssignmentLValue(cast<CompoundAssignOperator>(E));
+  }
   case Expr::CallExprClass:
   case Expr::CXXMemberCallExprClass:
   case Expr::CXXOperatorCallExprClass:
@@ -1135,7 +1162,7 @@ llvm::Value *CodeGenFunction::EmitLoadOfScalar(llvm::Value *Addr, bool Volatile,
   }
 
   // Atomic operations have to be done on integral types.
-  if (Ty->isAtomicType()) {
+  if (Ty->isAtomicType() || typeIsSuitableForInlineAtomic(Ty, Volatile)) {
     LValue lvalue = LValue::MakeAddr(Addr, Ty,
                                      CharUnits::fromQuantity(Alignment),
                                      getContext(), TBAAInfo);
@@ -1178,7 +1205,7 @@ llvm::Value *CodeGenFunction::EmitLoadOfScalar(llvm::Value *Addr, bool Volatile,
         EmitCheckSourceLocation(Loc),
         EmitCheckTypeDescriptor(Ty)
       };
-      SanitizerKind Kind = NeedsEnumCheck ? SanitizerKind::Enum : SanitizerKind::Bool;
+      SanitizerMask Kind = NeedsEnumCheck ? SanitizerKind::Enum : SanitizerKind::Bool;
       EmitCheck(std::make_pair(Check, Kind), "load_invalid_value", StaticArgs,
                 EmitCheckValue(Load));
     }
@@ -1254,7 +1281,8 @@ void CodeGenFunction::EmitStoreOfScalar(llvm::Value *Value, llvm::Value *Addr,
 
   Value = EmitToMemory(Value, Ty);
 
-  if (Ty->isAtomicType()) {
+  if (Ty->isAtomicType() ||
+      (!isInit && typeIsSuitableForInlineAtomic(Ty, Volatile))) {
     EmitAtomicStore(RValue::get(Value),
                     LValue::MakeAddr(Addr, Ty,
                                      CharUnits::fromQuantity(Alignment),
@@ -1692,8 +1720,8 @@ void CodeGenFunction::EmitStoreThroughGlobalRegLValue(RValue Src, LValue Dst) {
   llvm::Value *Value = Src.getScalarVal();
   if (OrigTy->isPointerTy())
     Value = Builder.CreatePtrToInt(Value, Ty);
-  Builder.CreateCall2(F, llvm::MetadataAsValue::get(Ty->getContext(), RegName),
-                      Value);
+  Builder.CreateCall(
+      F, {llvm::MetadataAsValue::get(Ty->getContext(), RegName), Value});
 }
 
 // setObjCGCLValueClass - sets class of the lvalue for the purpose of
@@ -1807,7 +1835,7 @@ EmitBitCastOfLValueToProperType(CodeGenFunction &CGF,
 static LValue EmitThreadPrivateVarDeclLValue(
     CodeGenFunction &CGF, const VarDecl *VD, QualType T, llvm::Value *V,
     llvm::Type *RealVarTy, CharUnits Alignment, SourceLocation Loc) {
-  V = CGF.CGM.getOpenMPRuntime().getOMPAddrOfThreadPrivate(CGF, VD, V, Loc);
+  V = CGF.CGM.getOpenMPRuntime().getAddrOfThreadPrivate(CGF, VD, V, Loc);
   V = EmitBitCastOfLValueToProperType(CGF, V, RealVarTy);
   return CGF.MakeAddrLValue(V, T, Alignment);
 }
@@ -2050,9 +2078,8 @@ LValue CodeGenFunction::EmitUnaryOpLValue(const UnaryOperator *E) {
     assert(E->getSubExpr()->getType()->isAnyComplexType());
 
     unsigned Idx = E->getOpcode() == UO_Imag;
-    return MakeAddrLValue(Builder.CreateStructGEP(LV.getAddress(),
-                                                  Idx, "idx"),
-                          ExprTy);
+    return MakeAddrLValue(
+        Builder.CreateStructGEP(nullptr, LV.getAddress(), Idx, "idx"), ExprTy);
   }
   case UO_PreInc:
   case UO_PreDec: {
@@ -2217,7 +2244,8 @@ enum class CheckRecoverableKind {
 };
 }
 
-static CheckRecoverableKind getRecoverableKind(SanitizerKind Kind) {
+static CheckRecoverableKind getRecoverableKind(SanitizerMask Kind) {
+  assert(llvm::countPopulation(Kind) == 1);
   switch (Kind) {
   case SanitizerKind::Vptr:
     return CheckRecoverableKind::AlwaysRecoverable;
@@ -2264,7 +2292,7 @@ static void emitCheckHandlerCall(CodeGenFunction &CGF,
 }
 
 void CodeGenFunction::EmitCheck(
-    ArrayRef<std::pair<llvm::Value *, SanitizerKind>> Checked,
+    ArrayRef<std::pair<llvm::Value *, SanitizerMask>> Checked,
     StringRef CheckName, ArrayRef<llvm::Constant *> StaticArgs,
     ArrayRef<llvm::Value *> DynamicArgs) {
   assert(IsSanitizerScope);
@@ -2376,7 +2404,7 @@ void CodeGenFunction::EmitTrapCheck(llvm::Value *Checked) {
     Builder.CreateCondBr(Checked, Cont, TrapBB);
     EmitBlock(TrapBB);
     llvm::Value *F = CGM.getIntrinsic(llvm::Intrinsic::trap);
-    llvm::CallInst *TrapCall = Builder.CreateCall(F);
+    llvm::CallInst *TrapCall = Builder.CreateCall(F, {});
     TrapCall->setDoesNotReturn();
     TrapCall->setDoesNotThrow();
     Builder.CreateUnreachable();
@@ -2648,7 +2676,7 @@ LValue CodeGenFunction::EmitLValueForField(LValue base,
     unsigned Idx = RL.getLLVMFieldNo(field);
     if (Idx != 0)
       // For structs, we GEP to the field that the record layout suggests.
-      Addr = Builder.CreateStructGEP(Addr, Idx, field->getName());
+      Addr = Builder.CreateStructGEP(nullptr, Addr, Idx, field->getName());
     // Get the access type.
     llvm::Type *PtrTy = llvm::Type::getIntNPtrTy(
       getLLVMContext(), Info.StorageSize,
@@ -2683,7 +2711,7 @@ LValue CodeGenFunction::EmitLValueForField(LValue base,
   } else {
     // For structs, we GEP to the field that the record layout suggests.
     unsigned idx = CGM.getTypes().getCGRecordLayout(rec).getLLVMFieldNo(field);
-    addr = Builder.CreateStructGEP(addr, idx, field->getName());
+    addr = Builder.CreateStructGEP(nullptr, addr, idx, field->getName());
 
     // If this is a reference field, load the reference right now.
     if (const ReferenceType *refType = type->getAs<ReferenceType>()) {
@@ -2762,7 +2790,7 @@ CodeGenFunction::EmitLValueForFieldInitialization(LValue Base,
   const CGRecordLayout &RL =
     CGM.getTypes().getCGRecordLayout(Field->getParent());
   unsigned idx = RL.getLLVMFieldNo(Field);
-  llvm::Value *V = Builder.CreateStructGEP(Base.getAddress(), idx);
+  llvm::Value *V = Builder.CreateStructGEP(nullptr, Base.getAddress(), idx);
   assert(!FieldType.getObjCGCAttr() && "fields cannot have GC attrs");
 
   // Make sure that the address is pointing to the right type.  This is critical
@@ -2834,7 +2862,6 @@ EmitConditionalOperatorLValue(const AbstractConditionalOperator *expr) {
   }
 
   OpaqueValueMapping binding(*this, expr);
-  RegionCounter Cnt = getPGORegionCounter(expr);
 
   const Expr *condExpr = expr->getCond();
   bool CondExprBool;
@@ -2845,7 +2872,7 @@ EmitConditionalOperatorLValue(const AbstractConditionalOperator *expr) {
     if (!ContainsLabel(dead)) {
       // If the true case is live, we need to track its region.
       if (CondExprBool)
-        Cnt.beginRegion(Builder);
+        incrementProfileCounter(expr);
       return EmitLValue(live);
     }
   }
@@ -2855,11 +2882,11 @@ EmitConditionalOperatorLValue(const AbstractConditionalOperator *expr) {
   llvm::BasicBlock *contBlock = createBasicBlock("cond.end");
 
   ConditionalEvaluation eval(*this);
-  EmitBranchOnBoolExpr(condExpr, lhsBlock, rhsBlock, Cnt.getCount());
+  EmitBranchOnBoolExpr(condExpr, lhsBlock, rhsBlock, getProfileCount(expr));
 
   // Any temporaries created here are conditional.
   EmitBlock(lhsBlock);
-  Cnt.beginRegion(Builder);
+  incrementProfileCounter(expr);
   eval.begin(*this);
   Optional<LValue> lhs =
       EmitLValueOrThrowExpression(*this, expr->getTrueExpr());
@@ -3007,6 +3034,9 @@ LValue CodeGenFunction::EmitCastLValue(const CastExpr *E) {
       EmitTypeCheck(TCK_DowncastReference, E->getExprLoc(),
                     Derived, E->getType());
 
+    if (SanOpts.has(SanitizerKind::CFIDerivedCast))
+      EmitVTablePtrCheckForCast(E->getType(), Derived, /*MayBeNull=*/false);
+
     return MakeAddrLValue(Derived, E->getType());
   }
   case CK_LValueBitCast: {
@@ -3016,6 +3046,10 @@ LValue CodeGenFunction::EmitCastLValue(const CastExpr *E) {
     LValue LV = EmitLValue(E->getSubExpr());
     llvm::Value *V = Builder.CreateBitCast(LV.getAddress(),
                                            ConvertType(CE->getTypeAsWritten()));
+
+    if (SanOpts.has(SanitizerKind::CFIUnrelatedCast))
+      EmitVTablePtrCheckForCast(E->getType(), V, /*MayBeNull=*/false);
+
     return MakeAddrLValue(V, E->getType());
   }
   case CK_ObjCObjectLValueCast: {
@@ -3059,16 +3093,6 @@ RValue CodeGenFunction::EmitRValueForField(LValue LV,
 
 RValue CodeGenFunction::EmitCallExpr(const CallExpr *E,
                                      ReturnValueSlot ReturnValue) {
-  // Force column info to be generated so we can differentiate
-  // multiple call sites on the same line in the debug info.
-  // FIXME: This is insufficient. Two calls coming from the same macro
-  // expansion will still get the same line/column and break debug info. It's
-  // possible that LLVM can be fixed to not rely on this uniqueness, at which
-  // point this workaround can be removed.
-  ApplyDebugLocation DL(*this, E->getLocStart(),
-                        E->getDirectCallee() &&
-                            E->getDirectCallee()->isInlineSpecified());
-
   // Builtins never have block type.
   if (E->getCallee()->getType()->isBlockPointerType())
     return EmitBlockCallExpr(E, ReturnValue);
@@ -3202,7 +3226,7 @@ LValue CodeGenFunction::EmitCallExprLValue(const CallExpr *E) {
   if (!RV.isScalar())
     return MakeAddrLValue(RV.getAggregateAddr(), E->getType());
 
-  assert(E->getCallReturnType()->isReferenceType() &&
+  assert(E->getCallReturnType(getContext())->isReferenceType() &&
          "Can't have a scalar return unless the return type is a "
          "reference type!");
 
@@ -3328,16 +3352,6 @@ RValue CodeGenFunction::EmitCall(QualType CalleeType, llvm::Value *Callee,
   const auto *FnType =
       cast<FunctionType>(cast<PointerType>(CalleeType)->getPointeeType());
 
-  // Force column info to differentiate multiple inlined call sites on
-  // the same line, analoguous to EmitCallExpr.
-  // FIXME: This is insufficient. Two calls coming from the same macro expansion
-  // will still get the same line/column and break debug info. It's possible
-  // that LLVM can be fixed to not rely on this uniqueness, at which point this
-  // workaround can be removed.
-  bool ForceColumnInfo = false;
-  if (const FunctionDecl* FD = dyn_cast_or_null<const FunctionDecl>(TargetDecl))
-    ForceColumnInfo = FD->isInlineSpecified();
-
   if (getLangOpts().CPlusPlus && SanOpts.has(SanitizerKind::Function) &&
       (!TargetDecl || !isa<FunctionDecl>(TargetDecl))) {
     if (llvm::Constant *PrefixSig =
@@ -3355,7 +3369,7 @@ RValue CodeGenFunction::EmitCall(QualType CalleeType, llvm::Value *Callee,
       llvm::Value *CalleePrefixStruct = Builder.CreateBitCast(
           Callee, llvm::PointerType::getUnqual(PrefixStructTy));
       llvm::Value *CalleeSigPtr =
-          Builder.CreateConstGEP2_32(CalleePrefixStruct, 0, 0);
+          Builder.CreateConstGEP2_32(PrefixStructTy, CalleePrefixStruct, 0, 0);
       llvm::Value *CalleeSig = Builder.CreateLoad(CalleeSigPtr);
       llvm::Value *CalleeSigMatch = Builder.CreateICmpEQ(CalleeSig, PrefixSig);
 
@@ -3365,7 +3379,7 @@ RValue CodeGenFunction::EmitCall(QualType CalleeType, llvm::Value *Callee,
 
       EmitBlock(TypeCheck);
       llvm::Value *CalleeRTTIPtr =
-          Builder.CreateConstGEP2_32(CalleePrefixStruct, 0, 1);
+          Builder.CreateConstGEP2_32(PrefixStructTy, CalleePrefixStruct, 0, 1);
       llvm::Value *CalleeRTTI = Builder.CreateLoad(CalleeRTTIPtr);
       llvm::Value *CalleeRTTIMatch =
           Builder.CreateICmpEQ(CalleeRTTI, FTRTTIConst);
@@ -3386,8 +3400,7 @@ RValue CodeGenFunction::EmitCall(QualType CalleeType, llvm::Value *Callee,
     Args.add(RValue::get(Builder.CreateBitCast(Chain, CGM.VoidPtrTy)),
              CGM.getContext().VoidPtrTy);
   EmitCallArgs(Args, dyn_cast<FunctionProtoType>(FnType), E->arg_begin(),
-               E->arg_end(), E->getDirectCallee(), /*ParamsToSkip*/ 0,
-               ForceColumnInfo);
+               E->arg_end(), E->getDirectCallee(), /*ParamsToSkip*/ 0);
 
   const CGFunctionInfo &FnInfo = CGM.getTypes().arrangeFreeFunctionCall(
       Args, FnType, /*isChainCall=*/Chain);
diff --git a/lib/CodeGen/CGExprAgg.cpp b/lib/CodeGen/CGExprAgg.cpp
index 4cf94c033bb4..6fedf0efda9d 100644
--- a/lib/CodeGen/CGExprAgg.cpp
+++ b/lib/CodeGen/CGExprAgg.cpp
@@ -98,6 +98,11 @@ public:
   //                            Visitor Methods
   //===--------------------------------------------------------------------===//
 
+  void Visit(Expr *E) {
+    ApplyDebugLocation DL(CGF, E);
+    StmtVisitor<AggExprEmitter>::Visit(E);
+  }
+
   void VisitStmt(Stmt *S) {
     CGF.ErrorUnsupported(S, "aggregate expression");
   }
@@ -207,7 +212,7 @@ void AggExprEmitter::EmitAggLoadOfLValue(const Expr *E) {
   LValue LV = CGF.EmitLValue(E);
 
   // If the type of the l-value is atomic, then do an atomic load.
-  if (LV.getType()->isAtomicType()) {
+  if (LV.getType()->isAtomicType() || CGF.LValueIsSuitableForInlineAtomic(LV)) {
     CGF.EmitAtomicLoad(LV, E->getExprLoc(), Dest);
     return;
   }
@@ -579,7 +584,12 @@ void AggExprEmitter::VisitCastExpr(CastExpr *E) {
   }
       
   case CK_ToUnion: {
-    if (Dest.isIgnored()) break;
+    // Evaluate even if the destination is ignored.
+    if (Dest.isIgnored()) {
+      CGF.EmitAnyExpr(E->getSubExpr(), AggValueSlot::ignored(),
+                      /*ignoreResult=*/true);
+      break;
+    }
 
     // GCC union extension
     QualType Ty = E->getSubExpr()->getType();
@@ -640,7 +650,7 @@ void AggExprEmitter::VisitCastExpr(CastExpr *E) {
 
         // Build a GEP to refer to the subobject.
         llvm::Value *valueAddr =
-            CGF.Builder.CreateStructGEP(valueDest.getAddr(), 0);
+            CGF.Builder.CreateStructGEP(nullptr, valueDest.getAddr(), 0);
         valueDest = AggValueSlot::forAddr(valueAddr,
                                           valueDest.getAlignment(),
                                           valueDest.getQualifiers(),
@@ -661,7 +671,7 @@ void AggExprEmitter::VisitCastExpr(CastExpr *E) {
     CGF.EmitAggExpr(E->getSubExpr(), atomicSlot);
 
     llvm::Value *valueAddr =
-      Builder.CreateStructGEP(atomicSlot.getAddr(), 0);
+        Builder.CreateStructGEP(nullptr, atomicSlot.getAddr(), 0);
     RValue rvalue = RValue::getAggregate(valueAddr, atomicSlot.isVolatile());
     return EmitFinalDestCopy(valueType, rvalue);
   }
@@ -736,7 +746,7 @@ void AggExprEmitter::VisitCastExpr(CastExpr *E) {
 }
 
 void AggExprEmitter::VisitCallExpr(const CallExpr *E) {
-  if (E->getCallReturnType()->isReferenceType()) {
+  if (E->getCallReturnType(CGF.getContext())->isReferenceType()) {
     EmitAggLoadOfLValue(E);
     return;
   }
@@ -860,7 +870,8 @@ void AggExprEmitter::VisitBinAssign(const BinaryOperator *E) {
     LValue LHS = CGF.EmitCheckedLValue(E->getLHS(), CodeGenFunction::TCK_Store);
 
     // That copy is an atomic copy if the LHS is atomic.
-    if (LHS.getType()->isAtomicType()) {
+    if (LHS.getType()->isAtomicType() ||
+        CGF.LValueIsSuitableForInlineAtomic(LHS)) {
       CGF.EmitAtomicStore(Dest.asRValue(), LHS, /*isInit*/ false);
       return;
     }
@@ -877,7 +888,8 @@ void AggExprEmitter::VisitBinAssign(const BinaryOperator *E) {
 
   // If we have an atomic type, evaluate into the destination and then
   // do an atomic copy.
-  if (LHS.getType()->isAtomicType()) {
+  if (LHS.getType()->isAtomicType() ||
+      CGF.LValueIsSuitableForInlineAtomic(LHS)) {
     EnsureDest(E->getRHS()->getType());
     Visit(E->getRHS());
     CGF.EmitAtomicStore(Dest.asRValue(), LHS, /*isInit*/ false);
@@ -909,16 +921,16 @@ VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
   // Bind the common expression if necessary.
   CodeGenFunction::OpaqueValueMapping binding(CGF, E);
 
-  RegionCounter Cnt = CGF.getPGORegionCounter(E);
   CodeGenFunction::ConditionalEvaluation eval(CGF);
-  CGF.EmitBranchOnBoolExpr(E->getCond(), LHSBlock, RHSBlock, Cnt.getCount());
+  CGF.EmitBranchOnBoolExpr(E->getCond(), LHSBlock, RHSBlock,
+                           CGF.getProfileCount(E));
 
   // Save whether the destination's lifetime is externally managed.
   bool isExternallyDestructed = Dest.isExternallyDestructed();
 
   eval.begin(CGF);
   CGF.EmitBlock(LHSBlock);
-  Cnt.beginRegion(Builder);
+  CGF.incrementProfileCounter(E);
   Visit(E->getTrueExpr());
   eval.end(CGF);
 
@@ -1408,7 +1420,8 @@ void CodeGenFunction::EmitAggregateCopy(llvm::Value *DestPtr,
       assert((Record->hasTrivialCopyConstructor() || 
               Record->hasTrivialCopyAssignment() ||
               Record->hasTrivialMoveConstructor() ||
-              Record->hasTrivialMoveAssignment()) &&
+              Record->hasTrivialMoveAssignment() ||
+              Record->isUnion()) &&
              "Trying to aggregate-copy a type without a trivial copy/move "
              "constructor or assignment operator");
       // Ignore empty classes in C++.
@@ -1439,7 +1452,34 @@ void CodeGenFunction::EmitAggregateCopy(llvm::Value *DestPtr,
   if (alignment.isZero())
     alignment = TypeInfo.second;
 
-  // FIXME: Handle variable sized types.
+  llvm::Value *SizeVal = nullptr;
+  if (TypeInfo.first.isZero()) {
+    // But note that getTypeInfo returns 0 for a VLA.
+    if (auto *VAT = dyn_cast_or_null<VariableArrayType>(
+            getContext().getAsArrayType(Ty))) {
+      QualType BaseEltTy;
+      SizeVal = emitArrayLength(VAT, BaseEltTy, DestPtr);
+      TypeInfo = getContext().getTypeInfoDataSizeInChars(BaseEltTy);
+      std::pair<CharUnits, CharUnits> LastElementTypeInfo;
+      if (!isAssignment)
+        LastElementTypeInfo = getContext().getTypeInfoInChars(BaseEltTy);
+      assert(!TypeInfo.first.isZero());
+      SizeVal = Builder.CreateNUWMul(
+          SizeVal,
+          llvm::ConstantInt::get(SizeTy, TypeInfo.first.getQuantity()));
+      if (!isAssignment) {
+        SizeVal = Builder.CreateNUWSub(
+            SizeVal,
+            llvm::ConstantInt::get(SizeTy, TypeInfo.first.getQuantity()));
+        SizeVal = Builder.CreateNUWAdd(
+            SizeVal, llvm::ConstantInt::get(
+                         SizeTy, LastElementTypeInfo.first.getQuantity()));
+      }
+    }
+  }
+  if (!SizeVal) {
+    SizeVal = llvm::ConstantInt::get(SizeTy, TypeInfo.first.getQuantity());
+  }
 
   // FIXME: If we have a volatile struct, the optimizer can remove what might
   // appear to be `extra' memory ops:
@@ -1470,9 +1510,6 @@ void CodeGenFunction::EmitAggregateCopy(llvm::Value *DestPtr,
   } else if (const RecordType *RecordTy = Ty->getAs<RecordType>()) {
     RecordDecl *Record = RecordTy->getDecl();
     if (Record->hasObjectMember()) {
-      CharUnits size = TypeInfo.first;
-      llvm::Type *SizeTy = ConvertType(getContext().getSizeType());
-      llvm::Value *SizeVal = llvm::ConstantInt::get(SizeTy, size.getQuantity());
       CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this, DestPtr, SrcPtr, 
                                                     SizeVal);
       return;
@@ -1481,10 +1518,6 @@ void CodeGenFunction::EmitAggregateCopy(llvm::Value *DestPtr,
     QualType BaseType = getContext().getBaseElementType(Ty);
     if (const RecordType *RecordTy = BaseType->getAs<RecordType>()) {
       if (RecordTy->getDecl()->hasObjectMember()) {
-        CharUnits size = TypeInfo.first;
-        llvm::Type *SizeTy = ConvertType(getContext().getSizeType());
-        llvm::Value *SizeVal = 
-          llvm::ConstantInt::get(SizeTy, size.getQuantity());
         CGM.getObjCRuntime().EmitGCMemmoveCollectable(*this, DestPtr, SrcPtr, 
                                                       SizeVal);
         return;
@@ -1497,9 +1530,6 @@ void CodeGenFunction::EmitAggregateCopy(llvm::Value *DestPtr,
   // the optimizer wishes to expand it in to scalar memory operations.
   llvm::MDNode *TBAAStructTag = CGM.getTBAAStructInfo(Ty);
 
-  Builder.CreateMemCpy(DestPtr, SrcPtr,
-                       llvm::ConstantInt::get(IntPtrTy, 
-                                              TypeInfo.first.getQuantity()),
-                       alignment.getQuantity(), isVolatile,
-                       /*TBAATag=*/nullptr, TBAAStructTag);
+  Builder.CreateMemCpy(DestPtr, SrcPtr, SizeVal, alignment.getQuantity(),
+                       isVolatile, /*TBAATag=*/nullptr, TBAAStructTag);
 }
diff --git a/lib/CodeGen/CGExprCXX.cpp b/lib/CodeGen/CGExprCXX.cpp
index 6d63b3ae9cf8..13dfbb38816f 100644
--- a/lib/CodeGen/CGExprCXX.cpp
+++ b/lib/CodeGen/CGExprCXX.cpp
@@ -173,7 +173,7 @@ RValue CodeGenFunction::EmitCXXMemberOrOperatorMemberCallExpr(
     This = EmitLValue(Base).getAddress();
 
 
-  if (MD->isTrivial()) {
+  if (MD->isTrivial() || (MD->isDefaulted() && MD->getParent()->isUnion())) {
     if (isa<CXXDestructorDecl>(MD)) return RValue::get(nullptr);
     if (isa<CXXConstructorDecl>(MD) && 
         cast<CXXConstructorDecl>(MD)->isDefaultConstructor())
@@ -256,6 +256,12 @@ RValue CodeGenFunction::EmitCXXMemberOrOperatorMemberCallExpr(
   } else if (UseVirtualCall) {
     Callee = CGM.getCXXABI().getVirtualFunctionPointer(*this, MD, This, Ty);
   } else {
+    if (SanOpts.has(SanitizerKind::CFINVCall) &&
+        MD->getParent()->isDynamicClass()) {
+      llvm::Value *VTable = GetVTablePtr(This, Int8PtrTy);
+      EmitVTablePtrCheckForCall(MD, VTable);
+    }
+
     if (getLangOpts().AppleKext && MD->isVirtual() && HasQualifier)
       Callee = BuildAppleKextVirtualCall(MD, Qualifier, Ty);
     else if (!DevirtualizedMethod)
@@ -684,7 +690,7 @@ static llvm::Value *EmitCXXNewAllocSize(CodeGenFunction &CGF,
       llvm::Value *tsmV =
         llvm::ConstantInt::get(CGF.SizeTy, typeSizeMultiplier);
       llvm::Value *result =
-        CGF.Builder.CreateCall2(umul_with_overflow, size, tsmV);
+          CGF.Builder.CreateCall(umul_with_overflow, {size, tsmV});
 
       llvm::Value *overflowed = CGF.Builder.CreateExtractValue(result, 1);
       if (hasOverflow)
@@ -723,7 +729,7 @@ static llvm::Value *EmitCXXNewAllocSize(CodeGenFunction &CGF,
 
       llvm::Value *cookieSizeV = llvm::ConstantInt::get(CGF.SizeTy, cookieSize);
       llvm::Value *result =
-        CGF.Builder.CreateCall2(uadd_with_overflow, size, cookieSizeV);
+          CGF.Builder.CreateCall(uadd_with_overflow, {size, cookieSizeV});
 
       llvm::Value *overflowed = CGF.Builder.CreateExtractValue(result, 1);
       if (hasOverflow)
@@ -778,12 +784,10 @@ static void StoreAnyExprIntoOneUnit(CodeGenFunction &CGF, const Expr *Init,
   llvm_unreachable("bad evaluation kind");
 }
 
-void
-CodeGenFunction::EmitNewArrayInitializer(const CXXNewExpr *E,
-                                         QualType ElementType,
-                                         llvm::Value *BeginPtr,
-                                         llvm::Value *NumElements,
-                                         llvm::Value *AllocSizeWithoutCookie) {
+void CodeGenFunction::EmitNewArrayInitializer(
+    const CXXNewExpr *E, QualType ElementType, llvm::Type *ElementTy,
+    llvm::Value *BeginPtr, llvm::Value *NumElements,
+    llvm::Value *AllocSizeWithoutCookie) {
   // If we have a type with trivial initialization and no initializer,
   // there's nothing to do.
   if (!E->hasInitializer())
@@ -809,7 +813,8 @@ CodeGenFunction::EmitNewArrayInitializer(const CXXNewExpr *E,
     if (const ConstantArrayType *CAT = dyn_cast_or_null<ConstantArrayType>(
             AllocType->getAsArrayTypeUnsafe())) {
       unsigned AS = CurPtr->getType()->getPointerAddressSpace();
-      llvm::Type *AllocPtrTy = ConvertTypeForMem(AllocType)->getPointerTo(AS);
+      ElementTy = ConvertTypeForMem(AllocType);
+      llvm::Type *AllocPtrTy = ElementTy->getPointerTo(AS);
       CurPtr = Builder.CreateBitCast(CurPtr, AllocPtrTy);
       InitListElements *= getContext().getConstantArrayElementCount(CAT);
     }
@@ -839,7 +844,8 @@ CodeGenFunction::EmitNewArrayInitializer(const CXXNewExpr *E,
       // initialization loops.
       StoreAnyExprIntoOneUnit(*this, ILE->getInit(i),
                               ILE->getInit(i)->getType(), CurPtr);
-      CurPtr = Builder.CreateConstInBoundsGEP1_32(CurPtr, 1, "array.exp.next");
+      CurPtr = Builder.CreateConstInBoundsGEP1_32(ElementTy, CurPtr, 1,
+                                                  "array.exp.next");
     }
 
     // The remaining elements are filled with the array filler expression.
@@ -1000,7 +1006,7 @@ CodeGenFunction::EmitNewArrayInitializer(const CXXNewExpr *E,
 
   // Advance to the next element by adjusting the pointer type as necessary.
   llvm::Value *NextPtr =
-      Builder.CreateConstInBoundsGEP1_32(CurPtr, 1, "array.next");
+      Builder.CreateConstInBoundsGEP1_32(ElementTy, CurPtr, 1, "array.next");
 
   // Check whether we've gotten to the end of the array and, if so,
   // exit the loop.
@@ -1012,13 +1018,12 @@ CodeGenFunction::EmitNewArrayInitializer(const CXXNewExpr *E,
 }
 
 static void EmitNewInitializer(CodeGenFunction &CGF, const CXXNewExpr *E,
-                               QualType ElementType,
-                               llvm::Value *NewPtr,
-                               llvm::Value *NumElements,
+                               QualType ElementType, llvm::Type *ElementTy,
+                               llvm::Value *NewPtr, llvm::Value *NumElements,
                                llvm::Value *AllocSizeWithoutCookie) {
-  ApplyDebugLocation DL(CGF, E->getStartLoc());
+  ApplyDebugLocation DL(CGF, E);
   if (E->isArray())
-    CGF.EmitNewArrayInitializer(E, ElementType, NewPtr, NumElements,
+    CGF.EmitNewArrayInitializer(E, ElementType, ElementTy, NewPtr, NumElements,
                                 AllocSizeWithoutCookie);
   else if (const Expr *Init = E->getInitializer())
     StoreAnyExprIntoOneUnit(CGF, Init, E->getAllocatedType(), NewPtr);
@@ -1279,10 +1284,9 @@ llvm::Value *CodeGenFunction::EmitCXXNewExpr(const CXXNewExpr *E) {
 
   // Emit a null check on the allocation result if the allocation
   // function is allowed to return null (because it has a non-throwing
-  // exception spec; for this part, we inline
-  // CXXNewExpr::shouldNullCheckAllocation()) and we have an
+  // exception spec or is the reserved placement new) and we have an
   // interesting initializer.
-  bool nullCheck = allocatorType->isNothrow(getContext()) &&
+  bool nullCheck = E->shouldNullCheckAllocation(getContext()) &&
     (!allocType.isPODType(getContext()) || E->hasInitializer());
 
   llvm::BasicBlock *nullCheckBB = nullptr;
@@ -1327,11 +1331,11 @@ llvm::Value *CodeGenFunction::EmitCXXNewExpr(const CXXNewExpr *E) {
                                                        E, allocType);
   }
 
-  llvm::Type *elementPtrTy
-    = ConvertTypeForMem(allocType)->getPointerTo(AS);
+  llvm::Type *elementTy = ConvertTypeForMem(allocType);
+  llvm::Type *elementPtrTy = elementTy->getPointerTo(AS);
   llvm::Value *result = Builder.CreateBitCast(allocation, elementPtrTy);
 
-  EmitNewInitializer(*this, E, allocType, result, numElements,
+  EmitNewInitializer(*this, E, allocType, elementTy, result, numElements,
                      allocSizeWithoutCookie);
   if (E->isArray()) {
     // NewPtr is a pointer to the base element type.  If we're
diff --git a/lib/CodeGen/CGExprComplex.cpp b/lib/CodeGen/CGExprComplex.cpp
index 1580bbe6a294..27d1c689966b 100644
--- a/lib/CodeGen/CGExprComplex.cpp
+++ b/lib/CodeGen/CGExprComplex.cpp
@@ -95,6 +95,7 @@ public:
   //===--------------------------------------------------------------------===//
 
   ComplexPairTy Visit(Expr *E) {
+    ApplyDebugLocation DL(CGF, E);
     return StmtVisitor<ComplexExprEmitter, ComplexPairTy>::Visit(E);
   }
 
@@ -316,14 +317,14 @@ ComplexPairTy ComplexExprEmitter::EmitLoadOfLValue(LValue lvalue,
   llvm::Value *Real=nullptr, *Imag=nullptr;
 
   if (!IgnoreReal || isVolatile) {
-    llvm::Value *RealP = Builder.CreateStructGEP(SrcPtr, 0,
+    llvm::Value *RealP = Builder.CreateStructGEP(nullptr, SrcPtr, 0,
                                                  SrcPtr->getName() + ".realp");
     Real = Builder.CreateAlignedLoad(RealP, AlignR, isVolatile,
                                      SrcPtr->getName() + ".real");
   }
 
   if (!IgnoreImag || isVolatile) {
-    llvm::Value *ImagP = Builder.CreateStructGEP(SrcPtr, 1,
+    llvm::Value *ImagP = Builder.CreateStructGEP(nullptr, SrcPtr, 1,
                                                  SrcPtr->getName() + ".imagp");
     Imag = Builder.CreateAlignedLoad(ImagP, AlignI, isVolatile,
                                      SrcPtr->getName() + ".imag");
@@ -335,12 +336,13 @@ ComplexPairTy ComplexExprEmitter::EmitLoadOfLValue(LValue lvalue,
 /// specified value pointer.
 void ComplexExprEmitter::EmitStoreOfComplex(ComplexPairTy Val, LValue lvalue,
                                             bool isInit) {
-  if (lvalue.getType()->isAtomicType())
+  if (lvalue.getType()->isAtomicType() ||
+      (!isInit && CGF.LValueIsSuitableForInlineAtomic(lvalue)))
     return CGF.EmitAtomicStore(RValue::getComplex(Val), lvalue, isInit);
 
   llvm::Value *Ptr = lvalue.getAddress();
-  llvm::Value *RealPtr = Builder.CreateStructGEP(Ptr, 0, "real");
-  llvm::Value *ImagPtr = Builder.CreateStructGEP(Ptr, 1, "imag");
+  llvm::Value *RealPtr = Builder.CreateStructGEP(nullptr, Ptr, 0, "real");
+  llvm::Value *ImagPtr = Builder.CreateStructGEP(nullptr, Ptr, 1, "imag");
   unsigned AlignR = lvalue.getAlignment().getQuantity();
   ASTContext &C = CGF.getContext();
   QualType ComplexTy = lvalue.getType();
@@ -375,7 +377,7 @@ VisitImaginaryLiteral(const ImaginaryLiteral *IL) {
 
 
 ComplexPairTy ComplexExprEmitter::VisitCallExpr(const CallExpr *E) {
-  if (E->getCallReturnType()->isReferenceType())
+  if (E->getCallReturnType(CGF.getContext())->isReferenceType())
     return EmitLoadOfLValue(E);
 
   return CGF.EmitCallExpr(E).getComplexVal();
@@ -818,6 +820,8 @@ EmitCompoundAssignLValue(const CompoundAssignOperator *E,
   TestAndClearIgnoreReal();
   TestAndClearIgnoreImag();
   QualType LHSTy = E->getLHS()->getType();
+  if (const AtomicType *AT = LHSTy->getAs<AtomicType>())
+    LHSTy = AT->getValueType();
 
   BinOpInfo OpInfo;
 
@@ -945,13 +949,14 @@ VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
   // Bind the common expression if necessary.
   CodeGenFunction::OpaqueValueMapping binding(CGF, E);
 
-  RegionCounter Cnt = CGF.getPGORegionCounter(E);
+
   CodeGenFunction::ConditionalEvaluation eval(CGF);
-  CGF.EmitBranchOnBoolExpr(E->getCond(), LHSBlock, RHSBlock, Cnt.getCount());
+  CGF.EmitBranchOnBoolExpr(E->getCond(), LHSBlock, RHSBlock,
+                           CGF.getProfileCount(E));
 
   eval.begin(CGF);
   CGF.EmitBlock(LHSBlock);
-  Cnt.beginRegion(Builder);
+  CGF.incrementProfileCounter(E);
   ComplexPairTy LHS = Visit(E->getTrueExpr());
   LHSBlock = Builder.GetInsertBlock();
   CGF.EmitBranch(ContBlock);
@@ -1033,7 +1038,7 @@ ComplexPairTy CodeGenFunction::EmitComplexExpr(const Expr *E, bool IgnoreReal,
          "Invalid complex expression to emit");
 
   return ComplexExprEmitter(*this, IgnoreReal, IgnoreImag)
-    .Visit(const_cast<Expr*>(E));
+      .Visit(const_cast<Expr *>(E));
 }
 
 void CodeGenFunction::EmitComplexExprIntoLValue(const Expr *E, LValue dest,
@@ -1085,8 +1090,8 @@ EmitComplexCompoundAssignmentLValue(const CompoundAssignOperator *E) {
 }
 
 LValue CodeGenFunction::
-EmitScalarCompooundAssignWithComplex(const CompoundAssignOperator *E,
-                                     llvm::Value *&Result) {
+EmitScalarCompoundAssignWithComplex(const CompoundAssignOperator *E,
+                                    llvm::Value *&Result) {
   CompoundFunc Op = getComplexOp(E->getOpcode());
   RValue Val;
   LValue Ret = ComplexExprEmitter(*this).EmitCompoundAssignLValue(E, Op, Val);
diff --git a/lib/CodeGen/CGExprConstant.cpp b/lib/CodeGen/CGExprConstant.cpp
index 54f7eee6791e..73ca0cc1c3d5 100644
--- a/lib/CodeGen/CGExprConstant.cpp
+++ b/lib/CodeGen/CGExprConstant.cpp
@@ -383,14 +383,19 @@ bool ConstStructBuilder::Build(InitListExpr *ILE) {
 
     if (!EltInit)
       return false;
-    
+
     if (!Field->isBitField()) {
       // Handle non-bitfield members.
       AppendField(*Field, Layout.getFieldOffset(FieldNo), EltInit);
     } else {
       // Otherwise we have a bitfield.
-      AppendBitField(*Field, Layout.getFieldOffset(FieldNo),
-                     cast<llvm::ConstantInt>(EltInit));
+      if (auto *CI = dyn_cast<llvm::ConstantInt>(EltInit)) {
+        AppendBitField(*Field, Layout.getFieldOffset(FieldNo), CI);
+      } else {
+        // We are trying to initialize a bitfield with a non-trivial constant,
+        // this must require run-time code.
+        return false;
+      }
     }
   }
 
@@ -1110,7 +1115,7 @@ llvm::Constant *CodeGenModule::EmitConstantValue(const APValue &Value,
         unsigned AS = C->getType()->getPointerAddressSpace();
         llvm::Type *CharPtrTy = Int8Ty->getPointerTo(AS);
         llvm::Constant *Casted = llvm::ConstantExpr::getBitCast(C, CharPtrTy);
-        Casted = llvm::ConstantExpr::getGetElementPtr(Casted, Offset);
+        Casted = llvm::ConstantExpr::getGetElementPtr(Int8Ty, Casted, Offset);
         C = llvm::ConstantExpr::getPointerCast(Casted, C->getType());
       }
 
@@ -1403,10 +1408,6 @@ llvm::Constant *CodeGenModule::EmitNullConstant(QualType T) {
 
     llvm::Constant *Element = EmitNullConstant(ElementTy);
     unsigned NumElements = CAT->getSize().getZExtValue();
-    
-    if (Element->isNullValue())
-      return llvm::ConstantAggregateZero::get(ATy);
-    
     SmallVector<llvm::Constant *, 8> Array(NumElements, Element);
     return llvm::ConstantArray::get(ATy, Array);
   }
@@ -1416,8 +1417,7 @@ llvm::Constant *CodeGenModule::EmitNullConstant(QualType T) {
     return ::EmitNullConstant(*this, RD, /*complete object*/ true);
   }
 
-  assert(T->isMemberPointerType() && "Should only see member pointers here!");
-  assert(!T->getAs<MemberPointerType>()->getPointeeType()->isFunctionType() &&
+  assert(T->isMemberDataPointerType() &&
          "Should only see pointers to data members here!");
 
   return getCXXABI().EmitNullMemberPointer(T->castAs<MemberPointerType>());
diff --git a/lib/CodeGen/CGExprScalar.cpp b/lib/CodeGen/CGExprScalar.cpp
index 3be14c8c3ee6..08c81c012c76 100644
--- a/lib/CodeGen/CGExprScalar.cpp
+++ b/lib/CodeGen/CGExprScalar.cpp
@@ -85,7 +85,7 @@ public:
     return CGF.EmitCheckedLValue(E, TCK);
   }
 
-  void EmitBinOpCheck(ArrayRef<std::pair<Value *, SanitizerKind>> Checks,
+  void EmitBinOpCheck(ArrayRef<std::pair<Value *, SanitizerMask>> Checks,
                       const BinOpInfo &Info);
 
   Value *EmitLoadOfLValue(LValue LV, SourceLocation Loc) {
@@ -196,7 +196,7 @@ public:
   //===--------------------------------------------------------------------===//
 
   Value *Visit(Expr *E) {
-    ApplyDebugLocation DL(CGF, E->getLocStart());
+    ApplyDebugLocation DL(CGF, E);
     return StmtVisitor<ScalarExprEmitter, Value*>::Visit(E);
   }
 
@@ -320,7 +320,7 @@ public:
   Value *VisitCastExpr(CastExpr *E);
 
   Value *VisitCallExpr(const CallExpr *E) {
-    if (E->getCallReturnType()->isReferenceType())
+    if (E->getCallReturnType(CGF.getContext())->isReferenceType())
       return EmitLoadOfLValue(E);
 
     Value *V = CGF.EmitCallExpr(E).getScalarVal();
@@ -349,10 +349,9 @@ public:
     return EmitScalarPrePostIncDec(E, LV, true, true);
   }
 
-  llvm::Value *EmitAddConsiderOverflowBehavior(const UnaryOperator *E,
-                                               llvm::Value *InVal,
-                                               llvm::Value *NextVal,
-                                               bool IsInc);
+  llvm::Value *EmitIncDecConsiderOverflowBehavior(const UnaryOperator *E,
+                                                  llvm::Value *InVal,
+                                                  bool IsInc);
 
   llvm::Value *EmitScalarPrePostIncDec(const UnaryOperator *E, LValue LV,
                                        bool isInc, bool isPre);
@@ -745,23 +744,37 @@ Value *ScalarExprEmitter::EmitScalarConversion(Value *Src, QualType SrcType,
   QualType OrigSrcType = SrcType;
   llvm::Type *SrcTy = Src->getType();
 
-  // If casting to/from storage-only half FP, use special intrinsics.
-  if (SrcType->isHalfType() && !CGF.getContext().getLangOpts().NativeHalfType &&
-      !CGF.getContext().getLangOpts().HalfArgsAndReturns) {
-    Src = Builder.CreateCall(
-        CGF.CGM.getIntrinsic(llvm::Intrinsic::convert_from_fp16,
-                             CGF.CGM.FloatTy),
-        Src);
-    SrcType = CGF.getContext().FloatTy;
-    SrcTy = CGF.FloatTy;
-  }
-
   // Handle conversions to bool first, they are special: comparisons against 0.
   if (DstType->isBooleanType())
     return EmitConversionToBool(Src, SrcType);
 
   llvm::Type *DstTy = ConvertType(DstType);
 
+  // Cast from half through float if half isn't a native type.
+  if (SrcType->isHalfType() && !CGF.getContext().getLangOpts().NativeHalfType) {
+    // Cast to FP using the intrinsic if the half type itself isn't supported.
+    if (DstTy->isFloatingPointTy()) {
+      if (!CGF.getContext().getLangOpts().HalfArgsAndReturns)
+        return Builder.CreateCall(
+            CGF.CGM.getIntrinsic(llvm::Intrinsic::convert_from_fp16, DstTy),
+            Src);
+    } else {
+      // Cast to other types through float, using either the intrinsic or FPExt,
+      // depending on whether the half type itself is supported
+      // (as opposed to operations on half, available with NativeHalfType).
+      if (!CGF.getContext().getLangOpts().HalfArgsAndReturns) {
+        Src = Builder.CreateCall(
+            CGF.CGM.getIntrinsic(llvm::Intrinsic::convert_from_fp16,
+                                 CGF.CGM.FloatTy),
+            Src);
+      } else {
+        Src = Builder.CreateFPExt(Src, CGF.CGM.FloatTy, "conv");
+      }
+      SrcType = CGF.getContext().FloatTy;
+      SrcTy = CGF.FloatTy;
+    }
+  }
+
   // Ignore conversions like int -> uint.
   if (SrcTy == DstTy)
     return Src;
@@ -818,10 +831,20 @@ Value *ScalarExprEmitter::EmitScalarConversion(Value *Src, QualType SrcType,
     EmitFloatConversionCheck(OrigSrc, OrigSrcType, Src, SrcType, DstType,
                              DstTy);
 
-  // Cast to half via float
-  if (DstType->isHalfType() && !CGF.getContext().getLangOpts().NativeHalfType &&
-      !CGF.getContext().getLangOpts().HalfArgsAndReturns)
+  // Cast to half through float if half isn't a native type.
+  if (DstType->isHalfType() && !CGF.getContext().getLangOpts().NativeHalfType) {
+    // Make sure we cast in a single step if from another FP type.
+    if (SrcTy->isFloatingPointTy()) {
+      // Use the intrinsic if the half type itself isn't supported
+      // (as opposed to operations on half, available with NativeHalfType).
+      if (!CGF.getContext().getLangOpts().HalfArgsAndReturns)
+        return Builder.CreateCall(
+            CGF.CGM.getIntrinsic(llvm::Intrinsic::convert_to_fp16, SrcTy), Src);
+      // If the half type is supported, just use an fptrunc.
+      return Builder.CreateFPTrunc(Src, DstTy);
+    }
     DstTy = CGF.FloatTy;
+  }
 
   if (isa<llvm::IntegerType>(SrcTy)) {
     bool InputSigned = SrcType->isSignedIntegerOrEnumerationType();
@@ -847,10 +870,14 @@ Value *ScalarExprEmitter::EmitScalarConversion(Value *Src, QualType SrcType,
   }
 
   if (DstTy != ResTy) {
-    assert(ResTy->isIntegerTy(16) && "Only half FP requires extra conversion");
-    Res = Builder.CreateCall(
+    if (!CGF.getContext().getLangOpts().HalfArgsAndReturns) {
+      assert(ResTy->isIntegerTy(16) && "Only half FP requires extra conversion");
+      Res = Builder.CreateCall(
         CGF.CGM.getIntrinsic(llvm::Intrinsic::convert_to_fp16, CGF.CGM.FloatTy),
         Res);
+    } else {
+      Res = Builder.CreateFPTrunc(Res, ResTy, "conv");
+    }
   }
 
   return Res;
@@ -889,7 +916,7 @@ Value *ScalarExprEmitter::EmitNullValue(QualType Ty) {
 /// operation). The check passes if all values in \p Checks (which are \c i1),
 /// are \c true.
 void ScalarExprEmitter::EmitBinOpCheck(
-    ArrayRef<std::pair<Value *, SanitizerKind>> Checks, const BinOpInfo &Info) {
+    ArrayRef<std::pair<Value *, SanitizerMask>> Checks, const BinOpInfo &Info) {
   assert(CGF.IsSanitizerScope);
   StringRef CheckName;
   SmallVector<llvm::Constant *, 4> StaticData;
@@ -1355,6 +1382,13 @@ Value *ScalarExprEmitter::VisitCastExpr(CastExpr *CE) {
       llvm_unreachable("wrong cast for pointers in different address spaces"
                        "(must be an address space cast)!");
     }
+
+    if (CGF.SanOpts.has(SanitizerKind::CFIUnrelatedCast)) {
+      if (auto PT = DestTy->getAs<PointerType>())
+        CGF.EmitVTablePtrCheckForCast(PT->getPointeeType(), Src,
+                                      /*MayBeNull=*/true);
+    }
+
     return Builder.CreateBitCast(Src, DstTy);
   }
   case CK_AddressSpaceConversion: {
@@ -1384,6 +1418,10 @@ Value *ScalarExprEmitter::VisitCastExpr(CastExpr *CE) {
       CGF.EmitTypeCheck(CodeGenFunction::TCK_DowncastPointer, CE->getExprLoc(),
                         Derived, DestTy->getPointeeType());
 
+    if (CGF.SanOpts.has(SanitizerKind::CFIDerivedCast))
+      CGF.EmitVTablePtrCheckForCast(DestTy->getPointeeType(), Derived,
+                                    /*MayBeNull=*/true);
+
     return Derived;
   }
   case CK_UncheckedDerivedToBase:
@@ -1412,13 +1450,13 @@ Value *ScalarExprEmitter::VisitCastExpr(CastExpr *CE) {
     // anything here.
     if (!E->getType()->isVariableArrayType()) {
       assert(isa<llvm::PointerType>(V->getType()) && "Expected pointer");
+      llvm::Type *NewTy = ConvertType(E->getType());
       V = CGF.Builder.CreatePointerCast(
-          V, ConvertType(E->getType())->getPointerTo(
-            V->getType()->getPointerAddressSpace()));
+          V, NewTy->getPointerTo(V->getType()->getPointerAddressSpace()));
 
       assert(isa<llvm::ArrayType>(V->getType()->getPointerElementType()) &&
              "Expected pointer to array");
-      V = Builder.CreateStructGEP(V, 0, "arraydecay");
+      V = Builder.CreateStructGEP(NewTy, V, 0, "arraydecay");
     }
 
     // Make sure the array decay ends up being the right type.  This matters if
@@ -1571,26 +1609,32 @@ Value *ScalarExprEmitter::VisitStmtExpr(const StmtExpr *E) {
 //                             Unary Operators
 //===----------------------------------------------------------------------===//
 
-llvm::Value *ScalarExprEmitter::
-EmitAddConsiderOverflowBehavior(const UnaryOperator *E,
-                                llvm::Value *InVal,
-                                llvm::Value *NextVal, bool IsInc) {
+static BinOpInfo createBinOpInfoFromIncDec(const UnaryOperator *E,
+                                           llvm::Value *InVal, bool IsInc) {
+  BinOpInfo BinOp;
+  BinOp.LHS = InVal;
+  BinOp.RHS = llvm::ConstantInt::get(InVal->getType(), 1, false);
+  BinOp.Ty = E->getType();
+  BinOp.Opcode = IsInc ? BO_Add : BO_Sub;
+  BinOp.FPContractable = false;
+  BinOp.E = E;
+  return BinOp;
+}
+
+llvm::Value *ScalarExprEmitter::EmitIncDecConsiderOverflowBehavior(
+    const UnaryOperator *E, llvm::Value *InVal, bool IsInc) {
+  llvm::Value *Amount =
+      llvm::ConstantInt::get(InVal->getType(), IsInc ? 1 : -1, true);
+  StringRef Name = IsInc ? "inc" : "dec";
   switch (CGF.getLangOpts().getSignedOverflowBehavior()) {
   case LangOptions::SOB_Defined:
-    return Builder.CreateAdd(InVal, NextVal, IsInc ? "inc" : "dec");
+    return Builder.CreateAdd(InVal, Amount, Name);
   case LangOptions::SOB_Undefined:
     if (!CGF.SanOpts.has(SanitizerKind::SignedIntegerOverflow))
-      return Builder.CreateNSWAdd(InVal, NextVal, IsInc ? "inc" : "dec");
+      return Builder.CreateNSWAdd(InVal, Amount, Name);
     // Fall through.
   case LangOptions::SOB_Trapping:
-    BinOpInfo BinOp;
-    BinOp.LHS = InVal;
-    BinOp.RHS = NextVal;
-    BinOp.Ty = E->getType();
-    BinOp.Opcode = BO_Add;
-    BinOp.FPContractable = false;
-    BinOp.E = E;
-    return EmitOverflowCheckedBinOp(BinOp);
+    return EmitOverflowCheckedBinOp(createBinOpInfoFromIncDec(E, InVal, IsInc));
   }
   llvm_unreachable("Unknown SignedOverflowBehaviorTy");
 }
@@ -1668,27 +1712,20 @@ ScalarExprEmitter::EmitScalarPrePostIncDec(const UnaryOperator *E, LValue LV,
 
   // Most common case by far: integer increment.
   } else if (type->isIntegerType()) {
-
-    llvm::Value *amt = llvm::ConstantInt::get(value->getType(), amount, true);
-
     // Note that signed integer inc/dec with width less than int can't
     // overflow because of promotion rules; we're just eliding a few steps here.
     bool CanOverflow = value->getType()->getIntegerBitWidth() >=
                        CGF.IntTy->getIntegerBitWidth();
     if (CanOverflow && type->isSignedIntegerOrEnumerationType()) {
-      value = EmitAddConsiderOverflowBehavior(E, value, amt, isInc);
+      value = EmitIncDecConsiderOverflowBehavior(E, value, isInc);
     } else if (CanOverflow && type->isUnsignedIntegerType() &&
                CGF.SanOpts.has(SanitizerKind::UnsignedIntegerOverflow)) {
-      BinOpInfo BinOp;
-      BinOp.LHS = value;
-      BinOp.RHS = llvm::ConstantInt::get(value->getType(), 1, false);
-      BinOp.Ty = E->getType();
-      BinOp.Opcode = isInc ? BO_Add : BO_Sub;
-      BinOp.FPContractable = false;
-      BinOp.E = E;
-      value = EmitOverflowCheckedBinOp(BinOp);
-    } else
+      value =
+          EmitOverflowCheckedBinOp(createBinOpInfoFromIncDec(E, value, isInc));
+    } else {
+      llvm::Value *amt = llvm::ConstantInt::get(value->getType(), amount, true);
       value = Builder.CreateAdd(value, amt, isInc ? "inc" : "dec");
+    }
 
   // Next most common: pointer increment.
   } else if (const PointerType *ptr = type->getAs<PointerType>()) {
@@ -1742,13 +1779,16 @@ ScalarExprEmitter::EmitScalarPrePostIncDec(const UnaryOperator *E, LValue LV,
     // Add the inc/dec to the real part.
     llvm::Value *amt;
 
-    if (type->isHalfType() && !CGF.getContext().getLangOpts().NativeHalfType &&
-        !CGF.getContext().getLangOpts().HalfArgsAndReturns) {
+    if (type->isHalfType() && !CGF.getContext().getLangOpts().NativeHalfType) {
       // Another special case: half FP increment should be done via float
-      value = Builder.CreateCall(
-          CGF.CGM.getIntrinsic(llvm::Intrinsic::convert_from_fp16,
-                               CGF.CGM.FloatTy),
-          input);
+      if (!CGF.getContext().getLangOpts().HalfArgsAndReturns) {
+        value = Builder.CreateCall(
+            CGF.CGM.getIntrinsic(llvm::Intrinsic::convert_from_fp16,
+                                 CGF.CGM.FloatTy),
+            input, "incdec.conv");
+      } else {
+        value = Builder.CreateFPExt(input, CGF.CGM.FloatTy, "incdec.conv");
+      }
     }
 
     if (value->getType()->isFloatTy())
@@ -1758,20 +1798,29 @@ ScalarExprEmitter::EmitScalarPrePostIncDec(const UnaryOperator *E, LValue LV,
       amt = llvm::ConstantFP::get(VMContext,
                                   llvm::APFloat(static_cast<double>(amount)));
     else {
+      // Remaining types are either Half or LongDouble.  Convert from float.
       llvm::APFloat F(static_cast<float>(amount));
       bool ignored;
-      F.convert(CGF.getTarget().getLongDoubleFormat(),
+      // Don't use getFloatTypeSemantics because Half isn't
+      // necessarily represented using the "half" LLVM type.
+      F.convert(value->getType()->isHalfTy()
+                    ? CGF.getTarget().getHalfFormat()
+                    : CGF.getTarget().getLongDoubleFormat(),
                 llvm::APFloat::rmTowardZero, &ignored);
       amt = llvm::ConstantFP::get(VMContext, F);
     }
     value = Builder.CreateFAdd(value, amt, isInc ? "inc" : "dec");
 
-    if (type->isHalfType() && !CGF.getContext().getLangOpts().NativeHalfType &&
-        !CGF.getContext().getLangOpts().HalfArgsAndReturns)
-      value = Builder.CreateCall(
-          CGF.CGM.getIntrinsic(llvm::Intrinsic::convert_to_fp16,
-                               CGF.CGM.FloatTy),
-          value);
+    if (type->isHalfType() && !CGF.getContext().getLangOpts().NativeHalfType) {
+      if (!CGF.getContext().getLangOpts().HalfArgsAndReturns) {
+        value = Builder.CreateCall(
+            CGF.CGM.getIntrinsic(llvm::Intrinsic::convert_to_fp16,
+                                 CGF.CGM.FloatTy),
+            value, "incdec.conv");
+      } else {
+        value = Builder.CreateFPTrunc(value, input->getType(), "incdec.conv");
+      }
+    }
 
   // Objective-C pointer types.
   } else {
@@ -1794,10 +1843,9 @@ ScalarExprEmitter::EmitScalarPrePostIncDec(const UnaryOperator *E, LValue LV,
     llvm::BasicBlock *opBB = Builder.GetInsertBlock();
     llvm::BasicBlock *contBB = CGF.createBasicBlock("atomic_cont", CGF.CurFn);
     auto Pair = CGF.EmitAtomicCompareExchange(
-        LV, RValue::get(atomicPHI), RValue::get(CGF.EmitToMemory(value, type)),
-        E->getExprLoc());
-    llvm::Value *old = Pair.first.getScalarVal();
-    llvm::Value *success = Pair.second.getScalarVal();
+        LV, RValue::get(atomicPHI), RValue::get(value), E->getExprLoc());
+    llvm::Value *old = CGF.EmitToMemory(Pair.first.getScalarVal(), type);
+    llvm::Value *success = Pair.second;
     atomicPHI->addIncoming(old, opBB);
     Builder.CreateCondBr(success, contBB, opBB);
     Builder.SetInsertPoint(contBB);
@@ -2056,7 +2104,7 @@ LValue ScalarExprEmitter::EmitCompoundAssignLValue(
   BinOpInfo OpInfo;
 
   if (E->getComputationResultType()->isAnyComplexType())
-    return CGF.EmitScalarCompooundAssignWithComplex(E, Result);
+    return CGF.EmitScalarCompoundAssignWithComplex(E, Result);
 
   // Emit the RHS first.  __block variables need to have the rhs evaluated
   // first, plus this should improve codegen a little.
@@ -2138,10 +2186,9 @@ LValue ScalarExprEmitter::EmitCompoundAssignLValue(
     llvm::BasicBlock *opBB = Builder.GetInsertBlock();
     llvm::BasicBlock *contBB = CGF.createBasicBlock("atomic_cont", CGF.CurFn);
     auto Pair = CGF.EmitAtomicCompareExchange(
-        LHSLV, RValue::get(atomicPHI),
-        RValue::get(CGF.EmitToMemory(Result, LHSTy)), E->getExprLoc());
-    llvm::Value *old = Pair.first.getScalarVal();
-    llvm::Value *success = Pair.second.getScalarVal();
+        LHSLV, RValue::get(atomicPHI), RValue::get(Result), E->getExprLoc());
+    llvm::Value *old = CGF.EmitToMemory(Pair.first.getScalarVal(), LHSTy);
+    llvm::Value *success = Pair.second;
     atomicPHI->addIncoming(old, opBB);
     Builder.CreateCondBr(success, contBB, opBB);
     Builder.SetInsertPoint(contBB);
@@ -2184,7 +2231,7 @@ Value *ScalarExprEmitter::EmitCompoundAssign(const CompoundAssignOperator *E,
 
 void ScalarExprEmitter::EmitUndefinedBehaviorIntegerDivAndRemCheck(
     const BinOpInfo &Ops, llvm::Value *Zero, bool isDiv) {
-  SmallVector<std::pair<llvm::Value *, SanitizerKind>, 2> Checks;
+  SmallVector<std::pair<llvm::Value *, SanitizerMask>, 2> Checks;
 
   if (CGF.SanOpts.has(SanitizerKind::IntegerDivideByZero)) {
     Checks.push_back(std::make_pair(Builder.CreateICmpNE(Ops.RHS, Zero),
@@ -2296,7 +2343,7 @@ Value *ScalarExprEmitter::EmitOverflowCheckedBinOp(const BinOpInfo &Ops) {
 
   llvm::Function *intrinsic = CGF.CGM.getIntrinsic(IID, opTy);
 
-  Value *resultAndOverflow = Builder.CreateCall2(intrinsic, Ops.LHS, Ops.RHS);
+  Value *resultAndOverflow = Builder.CreateCall(intrinsic, {Ops.LHS, Ops.RHS});
   Value *result = Builder.CreateExtractValue(resultAndOverflow, 0);
   Value *overflow = Builder.CreateExtractValue(resultAndOverflow, 1);
 
@@ -2309,7 +2356,7 @@ Value *ScalarExprEmitter::EmitOverflowCheckedBinOp(const BinOpInfo &Ops) {
     if (!isSigned || CGF.SanOpts.has(SanitizerKind::SignedIntegerOverflow)) {
       CodeGenFunction::SanitizerScope SanScope(&CGF);
       llvm::Value *NotOverflow = Builder.CreateNot(overflow);
-      SanitizerKind Kind = isSigned ? SanitizerKind::SignedIntegerOverflow
+      SanitizerMask Kind = isSigned ? SanitizerKind::SignedIntegerOverflow
                               : SanitizerKind::UnsignedIntegerOverflow;
       EmitBinOpCheck(std::make_pair(NotOverflow, Kind), Ops);
     } else
@@ -2476,10 +2523,9 @@ static Value* buildFMulAdd(llvm::BinaryOperator *MulOp, Value *Addend,
         "neg");
   }
 
-  Value *FMulAdd =
-    Builder.CreateCall3(
+  Value *FMulAdd = Builder.CreateCall(
       CGF.CGM.getIntrinsic(llvm::Intrinsic::fmuladd, Addend->getType()),
-                           MulOp0, MulOp1, Addend);
+      {MulOp0, MulOp1, Addend});
    MulOp->eraseFromParent();
 
    return FMulAdd;
@@ -2664,21 +2710,34 @@ Value *ScalarExprEmitter::EmitShl(const BinOpInfo &Ops) {
   if (Ops.LHS->getType() != RHS->getType())
     RHS = Builder.CreateIntCast(RHS, Ops.LHS->getType(), false, "sh_prom");
 
-  if (CGF.SanOpts.has(SanitizerKind::Shift) && !CGF.getLangOpts().OpenCL &&
-      isa<llvm::IntegerType>(Ops.LHS->getType())) {
+  bool SanitizeBase = CGF.SanOpts.has(SanitizerKind::ShiftBase) &&
+                      Ops.Ty->hasSignedIntegerRepresentation();
+  bool SanitizeExponent = CGF.SanOpts.has(SanitizerKind::ShiftExponent);
+  // OpenCL 6.3j: shift values are effectively % word size of LHS.
+  if (CGF.getLangOpts().OpenCL)
+    RHS =
+        Builder.CreateAnd(RHS, GetWidthMinusOneValue(Ops.LHS, RHS), "shl.mask");
+  else if ((SanitizeBase || SanitizeExponent) &&
+           isa<llvm::IntegerType>(Ops.LHS->getType())) {
     CodeGenFunction::SanitizerScope SanScope(&CGF);
+    SmallVector<std::pair<Value *, SanitizerMask>, 2> Checks;
     llvm::Value *WidthMinusOne = GetWidthMinusOneValue(Ops.LHS, RHS);
-    llvm::Value *Valid = Builder.CreateICmpULE(RHS, WidthMinusOne);
+    llvm::Value *ValidExponent = Builder.CreateICmpULE(RHS, WidthMinusOne);
 
-    if (Ops.Ty->hasSignedIntegerRepresentation()) {
-      llvm::BasicBlock *Orig = Builder.GetInsertBlock();
-      llvm::BasicBlock *Cont = CGF.createBasicBlock("cont");
-      llvm::BasicBlock *CheckBitsShifted = CGF.createBasicBlock("check");
-      Builder.CreateCondBr(Valid, CheckBitsShifted, Cont);
+    if (SanitizeExponent) {
+      Checks.push_back(
+          std::make_pair(ValidExponent, SanitizerKind::ShiftExponent));
+    }
 
+    if (SanitizeBase) {
       // Check whether we are shifting any non-zero bits off the top of the
-      // integer.
-      CGF.EmitBlock(CheckBitsShifted);
+      // integer. We only emit this check if exponent is valid - otherwise
+      // instructions below will have undefined behavior themselves.
+      llvm::BasicBlock *Orig = Builder.GetInsertBlock();
+      llvm::BasicBlock *Cont = CGF.createBasicBlock("cont");
+      llvm::BasicBlock *CheckShiftBase = CGF.createBasicBlock("check");
+      Builder.CreateCondBr(ValidExponent, CheckShiftBase, Cont);
+      CGF.EmitBlock(CheckShiftBase);
       llvm::Value *BitsShiftedOff =
         Builder.CreateLShr(Ops.LHS,
                            Builder.CreateSub(WidthMinusOne, RHS, "shl.zeros",
@@ -2693,19 +2752,17 @@ Value *ScalarExprEmitter::EmitShl(const BinOpInfo &Ops) {
         BitsShiftedOff = Builder.CreateLShr(BitsShiftedOff, One);
       }
       llvm::Value *Zero = llvm::ConstantInt::get(BitsShiftedOff->getType(), 0);
-      llvm::Value *SecondCheck = Builder.CreateICmpEQ(BitsShiftedOff, Zero);
+      llvm::Value *ValidBase = Builder.CreateICmpEQ(BitsShiftedOff, Zero);
       CGF.EmitBlock(Cont);
-      llvm::PHINode *P = Builder.CreatePHI(Valid->getType(), 2);
-      P->addIncoming(Valid, Orig);
-      P->addIncoming(SecondCheck, CheckBitsShifted);
-      Valid = P;
+      llvm::PHINode *BaseCheck = Builder.CreatePHI(ValidBase->getType(), 2);
+      BaseCheck->addIncoming(Builder.getTrue(), Orig);
+      BaseCheck->addIncoming(ValidBase, CheckShiftBase);
+      Checks.push_back(std::make_pair(BaseCheck, SanitizerKind::ShiftBase));
     }
 
-    EmitBinOpCheck(std::make_pair(Valid, SanitizerKind::Shift), Ops);
+    assert(!Checks.empty());
+    EmitBinOpCheck(Checks, Ops);
   }
-  // OpenCL 6.3j: shift values are effectively % word size of LHS.
-  if (CGF.getLangOpts().OpenCL)
-    RHS = Builder.CreateAnd(RHS, GetWidthMinusOneValue(Ops.LHS, RHS), "shl.mask");
 
   return Builder.CreateShl(Ops.LHS, RHS, "shl");
 }
@@ -2717,18 +2774,18 @@ Value *ScalarExprEmitter::EmitShr(const BinOpInfo &Ops) {
   if (Ops.LHS->getType() != RHS->getType())
     RHS = Builder.CreateIntCast(RHS, Ops.LHS->getType(), false, "sh_prom");
 
-  if (CGF.SanOpts.has(SanitizerKind::Shift) && !CGF.getLangOpts().OpenCL &&
-      isa<llvm::IntegerType>(Ops.LHS->getType())) {
+  // OpenCL 6.3j: shift values are effectively % word size of LHS.
+  if (CGF.getLangOpts().OpenCL)
+    RHS =
+        Builder.CreateAnd(RHS, GetWidthMinusOneValue(Ops.LHS, RHS), "shr.mask");
+  else if (CGF.SanOpts.has(SanitizerKind::ShiftExponent) &&
+           isa<llvm::IntegerType>(Ops.LHS->getType())) {
     CodeGenFunction::SanitizerScope SanScope(&CGF);
     llvm::Value *Valid =
         Builder.CreateICmpULE(RHS, GetWidthMinusOneValue(Ops.LHS, RHS));
-    EmitBinOpCheck(std::make_pair(Valid, SanitizerKind::Shift), Ops);
+    EmitBinOpCheck(std::make_pair(Valid, SanitizerKind::ShiftExponent), Ops);
   }
 
-  // OpenCL 6.3j: shift values are effectively % word size of LHS.
-  if (CGF.getLangOpts().OpenCL)
-    RHS = Builder.CreateAnd(RHS, GetWidthMinusOneValue(Ops.LHS, RHS), "shr.mask");
-
   if (Ops.Ty->hasUnsignedIntegerRepresentation())
     return Builder.CreateLShr(Ops.LHS, RHS, "shr");
   return Builder.CreateAShr(Ops.LHS, RHS, "shr");
@@ -2846,7 +2903,7 @@ Value *ScalarExprEmitter::EmitCompare(const BinaryOperator *E,unsigned UICmpOpc,
 
       Value *CR6Param = Builder.getInt32(CR6);
       llvm::Function *F = CGF.CGM.getIntrinsic(ID);
-      Result = Builder.CreateCall3(F, CR6Param, FirstVecArg, SecondVecArg, "");
+      Result = Builder.CreateCall(F, {CR6Param, FirstVecArg, SecondVecArg});
       return EmitScalarConversion(Result, CGF.getContext().BoolTy, E->getType());
     }
 
@@ -2975,11 +3032,9 @@ Value *ScalarExprEmitter::VisitBinAssign(const BinaryOperator *E) {
 }
 
 Value *ScalarExprEmitter::VisitBinLAnd(const BinaryOperator *E) {
-  RegionCounter Cnt = CGF.getPGORegionCounter(E);
-
   // Perform vector logical and on comparisons with zero vectors.
   if (E->getType()->isVectorType()) {
-    Cnt.beginRegion(Builder);
+    CGF.incrementProfileCounter(E);
 
     Value *LHS = Visit(E->getLHS());
     Value *RHS = Visit(E->getRHS());
@@ -3002,7 +3057,7 @@ Value *ScalarExprEmitter::VisitBinLAnd(const BinaryOperator *E) {
   bool LHSCondVal;
   if (CGF.ConstantFoldsToSimpleInteger(E->getLHS(), LHSCondVal)) {
     if (LHSCondVal) { // If we have 1 && X, just emit X.
-      Cnt.beginRegion(Builder);
+      CGF.incrementProfileCounter(E);
 
       Value *RHSCond = CGF.EvaluateExprAsBool(E->getRHS());
       // ZExt result to int or bool.
@@ -3020,7 +3075,8 @@ Value *ScalarExprEmitter::VisitBinLAnd(const BinaryOperator *E) {
   CodeGenFunction::ConditionalEvaluation eval(CGF);
 
   // Branch on the LHS first.  If it is false, go to the failure (cont) block.
-  CGF.EmitBranchOnBoolExpr(E->getLHS(), RHSBlock, ContBlock, Cnt.getCount());
+  CGF.EmitBranchOnBoolExpr(E->getLHS(), RHSBlock, ContBlock,
+                           CGF.getProfileCount(E->getRHS()));
 
   // Any edges into the ContBlock are now from an (indeterminate number of)
   // edges from this first condition.  All of these values will be false.  Start
@@ -3033,7 +3089,7 @@ Value *ScalarExprEmitter::VisitBinLAnd(const BinaryOperator *E) {
 
   eval.begin(CGF);
   CGF.EmitBlock(RHSBlock);
-  Cnt.beginRegion(Builder);
+  CGF.incrementProfileCounter(E);
   Value *RHSCond = CGF.EvaluateExprAsBool(E->getRHS());
   eval.end(CGF);
 
@@ -3043,7 +3099,7 @@ Value *ScalarExprEmitter::VisitBinLAnd(const BinaryOperator *E) {
   // Emit an unconditional branch from this block to ContBlock.
   {
     // There is no need to emit line number for unconditional branch.
-    ApplyDebugLocation DL(CGF);
+    auto NL = ApplyDebugLocation::CreateEmpty(CGF);
     CGF.EmitBlock(ContBlock);
   }
   // Insert an entry into the phi node for the edge with the value of RHSCond.
@@ -3054,11 +3110,9 @@ Value *ScalarExprEmitter::VisitBinLAnd(const BinaryOperator *E) {
 }
 
 Value *ScalarExprEmitter::VisitBinLOr(const BinaryOperator *E) {
-  RegionCounter Cnt = CGF.getPGORegionCounter(E);
-
   // Perform vector logical or on comparisons with zero vectors.
   if (E->getType()->isVectorType()) {
-    Cnt.beginRegion(Builder);
+    CGF.incrementProfileCounter(E);
 
     Value *LHS = Visit(E->getLHS());
     Value *RHS = Visit(E->getRHS());
@@ -3081,7 +3135,7 @@ Value *ScalarExprEmitter::VisitBinLOr(const BinaryOperator *E) {
   bool LHSCondVal;
   if (CGF.ConstantFoldsToSimpleInteger(E->getLHS(), LHSCondVal)) {
     if (!LHSCondVal) { // If we have 0 || X, just emit X.
-      Cnt.beginRegion(Builder);
+      CGF.incrementProfileCounter(E);
 
       Value *RHSCond = CGF.EvaluateExprAsBool(E->getRHS());
       // ZExt result to int or bool.
@@ -3100,7 +3154,8 @@ Value *ScalarExprEmitter::VisitBinLOr(const BinaryOperator *E) {
 
   // Branch on the LHS first.  If it is true, go to the success (cont) block.
   CGF.EmitBranchOnBoolExpr(E->getLHS(), ContBlock, RHSBlock,
-                           Cnt.getParentCount() - Cnt.getCount());
+                           CGF.getCurrentProfileCount() -
+                               CGF.getProfileCount(E->getRHS()));
 
   // Any edges into the ContBlock are now from an (indeterminate number of)
   // edges from this first condition.  All of these values will be true.  Start
@@ -3115,7 +3170,7 @@ Value *ScalarExprEmitter::VisitBinLOr(const BinaryOperator *E) {
 
   // Emit the RHS condition as a bool value.
   CGF.EmitBlock(RHSBlock);
-  Cnt.beginRegion(Builder);
+  CGF.incrementProfileCounter(E);
   Value *RHSCond = CGF.EvaluateExprAsBool(E->getRHS());
 
   eval.end(CGF);
@@ -3166,7 +3221,6 @@ VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
 
   // Bind the common expression if necessary.
   CodeGenFunction::OpaqueValueMapping binding(CGF, E);
-  RegionCounter Cnt = CGF.getPGORegionCounter(E);
 
   Expr *condExpr = E->getCond();
   Expr *lhsExpr = E->getTrueExpr();
@@ -3182,7 +3236,7 @@ VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
     // If the dead side doesn't have labels we need, just emit the Live part.
     if (!CGF.ContainsLabel(dead)) {
       if (CondExprBool)
-        Cnt.beginRegion(Builder);
+        CGF.incrementProfileCounter(E);
       Value *Result = Visit(live);
 
       // If the live part is a throw expression, it acts like it has a void
@@ -3199,7 +3253,7 @@ VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
   // the select function.
   if (CGF.getLangOpts().OpenCL
       && condExpr->getType()->isVectorType()) {
-    Cnt.beginRegion(Builder);
+    CGF.incrementProfileCounter(E);
 
     llvm::Value *CondV = CGF.EmitScalarExpr(condExpr);
     llvm::Value *LHS = Visit(lhsExpr);
@@ -3244,7 +3298,7 @@ VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
   // safe to evaluate the LHS and RHS unconditionally.
   if (isCheapEnoughToEvaluateUnconditionally(lhsExpr, CGF) &&
       isCheapEnoughToEvaluateUnconditionally(rhsExpr, CGF)) {
-    Cnt.beginRegion(Builder);
+    CGF.incrementProfileCounter(E);
 
     llvm::Value *CondV = CGF.EvaluateExprAsBool(condExpr);
     llvm::Value *LHS = Visit(lhsExpr);
@@ -3262,10 +3316,11 @@ VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
   llvm::BasicBlock *ContBlock = CGF.createBasicBlock("cond.end");
 
   CodeGenFunction::ConditionalEvaluation eval(CGF);
-  CGF.EmitBranchOnBoolExpr(condExpr, LHSBlock, RHSBlock, Cnt.getCount());
+  CGF.EmitBranchOnBoolExpr(condExpr, LHSBlock, RHSBlock,
+                           CGF.getProfileCount(lhsExpr));
 
   CGF.EmitBlock(LHSBlock);
-  Cnt.beginRegion(Builder);
+  CGF.incrementProfileCounter(E);
   eval.begin(CGF);
   Value *LHS = Visit(lhsExpr);
   eval.end(CGF);
@@ -3393,14 +3448,8 @@ Value *CodeGenFunction::EmitScalarExpr(const Expr *E, bool IgnoreResultAssign) {
   assert(E && hasScalarEvaluationKind(E->getType()) &&
          "Invalid scalar expression to emit");
 
-  bool hasDebugInfo = getDebugInfo();
-  if (isa<CXXDefaultArgExpr>(E))
-    disableDebugInfo();
-  Value *V = ScalarExprEmitter(*this, IgnoreResultAssign)
-    .Visit(const_cast<Expr*>(E));
-  if (isa<CXXDefaultArgExpr>(E) && hasDebugInfo)
-    enableDebugInfo();
-  return V;
+  return ScalarExprEmitter(*this, IgnoreResultAssign)
+      .Visit(const_cast<Expr *>(E));
 }
 
 /// EmitScalarConversion - Emit a conversion from the specified type to the
diff --git a/lib/CodeGen/CGLoopInfo.cpp b/lib/CodeGen/CGLoopInfo.cpp
index 89f43c281590..011ae7e836ae 100644
--- a/lib/CodeGen/CGLoopInfo.cpp
+++ b/lib/CodeGen/CGLoopInfo.cpp
@@ -26,8 +26,8 @@ static MDNode *createMetadata(LLVMContext &Ctx, const LoopAttributes &Attrs) {
 
   SmallVector<Metadata *, 4> Args;
   // Reserve operand 0 for loop id self reference.
-  MDNode *TempNode = MDNode::getTemporary(Ctx, None);
-  Args.push_back(TempNode);
+  auto TempNode = MDNode::getTemporary(Ctx, None);
+  Args.push_back(TempNode.get());
 
   // Setting vectorizer.width
   if (Attrs.VectorizerWidth > 0) {
@@ -58,7 +58,6 @@ static MDNode *createMetadata(LLVMContext &Ctx, const LoopAttributes &Attrs) {
   // Set the first operand to itself.
   MDNode *LoopID = MDNode::get(Ctx, Args);
   LoopID->replaceOperandWith(0, LoopID);
-  MDNode::deleteTemporary(TempNode);
   return LoopID;
 }
 
diff --git a/lib/CodeGen/CGLoopInfo.h b/lib/CodeGen/CGLoopInfo.h
index b1693996507e..aee162118a2f 100644
--- a/lib/CodeGen/CGLoopInfo.h
+++ b/lib/CodeGen/CGLoopInfo.h
@@ -78,8 +78,8 @@ private:
 /// This stack can be used to prepare attributes which are applied when a loop
 /// is emitted.
 class LoopInfoStack {
-  LoopInfoStack(const LoopInfoStack &) LLVM_DELETED_FUNCTION;
-  void operator=(const LoopInfoStack &) LLVM_DELETED_FUNCTION;
+  LoopInfoStack(const LoopInfoStack &) = delete;
+  void operator=(const LoopInfoStack &) = delete;
 
 public:
   LoopInfoStack() {}
diff --git a/lib/CodeGen/CGObjC.cpp b/lib/CodeGen/CGObjC.cpp
index 34c6d94f8817..ef9a92dfc35b 100644
--- a/lib/CodeGen/CGObjC.cpp
+++ b/lib/CodeGen/CGObjC.cpp
@@ -102,8 +102,8 @@ llvm::Value *CodeGenFunction::EmitObjCCollectionLiteral(const Expr *E,
                                    ArrayType::Normal, /*IndexTypeQuals=*/0);
 
   // Allocate the temporary array(s).
-  llvm::Value *Objects = CreateMemTemp(ElementArrayType, "objects");
-  llvm::Value *Keys = nullptr;
+  llvm::AllocaInst *Objects = CreateMemTemp(ElementArrayType, "objects");
+  llvm::AllocaInst *Keys = nullptr;
   if (DLE)
     Keys = CreateMemTemp(ElementArrayType, "keys");
   
@@ -119,10 +119,9 @@ llvm::Value *CodeGenFunction::EmitObjCCollectionLiteral(const Expr *E,
     if (ALE) {
       // Emit the element and store it to the appropriate array slot.
       const Expr *Rhs = ALE->getElement(i);
-      LValue LV = LValue::MakeAddr(Builder.CreateStructGEP(Objects, i),
-                                   ElementType,
-                                   Context.getTypeAlignInChars(Rhs->getType()),
-                                   Context);
+      LValue LV = LValue::MakeAddr(
+          Builder.CreateStructGEP(Objects->getAllocatedType(), Objects, i),
+          ElementType, Context.getTypeAlignInChars(Rhs->getType()), Context);
 
       llvm::Value *value = EmitScalarExpr(Rhs);
       EmitStoreThroughLValue(RValue::get(value), LV, true);
@@ -132,19 +131,17 @@ llvm::Value *CodeGenFunction::EmitObjCCollectionLiteral(const Expr *E,
     } else {      
       // Emit the key and store it to the appropriate array slot.
       const Expr *Key = DLE->getKeyValueElement(i).Key;
-      LValue KeyLV = LValue::MakeAddr(Builder.CreateStructGEP(Keys, i),
-                                      ElementType,
-                                    Context.getTypeAlignInChars(Key->getType()),
-                                      Context);
+      LValue KeyLV = LValue::MakeAddr(
+          Builder.CreateStructGEP(Keys->getAllocatedType(), Keys, i),
+          ElementType, Context.getTypeAlignInChars(Key->getType()), Context);
       llvm::Value *keyValue = EmitScalarExpr(Key);
       EmitStoreThroughLValue(RValue::get(keyValue), KeyLV, /*isInit=*/true);
 
       // Emit the value and store it to the appropriate array slot.
-      const Expr *Value = DLE->getKeyValueElement(i).Value;  
-      LValue ValueLV = LValue::MakeAddr(Builder.CreateStructGEP(Objects, i), 
-                                        ElementType,
-                                  Context.getTypeAlignInChars(Value->getType()),
-                                        Context);
+      const Expr *Value = DLE->getKeyValueElement(i).Value;
+      LValue ValueLV = LValue::MakeAddr(
+          Builder.CreateStructGEP(Objects->getAllocatedType(), Objects, i),
+          ElementType, Context.getTypeAlignInChars(Value->getType()), Context);
       llvm::Value *valueValue = EmitScalarExpr(Value);
       EmitStoreThroughLValue(RValue::get(valueValue), ValueLV, /*isInit=*/true);
       if (TrackNeededObjects) {
@@ -472,8 +469,7 @@ void CodeGenFunction::StartObjCMethod(const ObjCMethodDecl *OMD,
   args.push_back(OMD->getSelfDecl());
   args.push_back(OMD->getCmdDecl());
 
-  for (const auto *PI : OMD->params())
-    args.push_back(PI);
+  args.append(OMD->param_begin(), OMD->param_end());
 
   CurGD = OMD;
   CurEHLocation = OMD->getLocEnd();
@@ -501,8 +497,7 @@ void CodeGenFunction::GenerateObjCMethod(const ObjCMethodDecl *OMD) {
   StartObjCMethod(OMD, OMD->getClassInterface());
   PGO.assignRegionCounters(OMD, CurFn);
   assert(isa<CompoundStmt>(OMD->getBody()));
-  RegionCounter Cnt = getPGORegionCounter(OMD->getBody());
-  Cnt.beginRegion(Builder);
+  incrementProfileCounter(OMD->getBody());
   EmitCompoundStmtWithoutScope(*cast<CompoundStmt>(OMD->getBody()));
   FinishFunction(OMD->getBodyRBrace());
 }
@@ -1435,7 +1430,7 @@ void CodeGenFunction::EmitObjCForCollectionStmt(const ObjCForCollectionStmt &S){
 
   // Fast enumeration state.
   QualType StateTy = CGM.getObjCFastEnumerationStateType();
-  llvm::Value *StatePtr = CreateMemTemp(StateTy, "state.ptr");
+  llvm::AllocaInst *StatePtr = CreateMemTemp(StateTy, "state.ptr");
   EmitNullInitialization(StatePtr, StateTy);
 
   // Number of elements in the items array.
@@ -1507,11 +1502,11 @@ void CodeGenFunction::EmitObjCForCollectionStmt(const ObjCForCollectionStmt &S){
   // If the limit pointer was zero to begin with, the collection is
   // empty; skip all this. Set the branch weight assuming this has the same
   // probability of exiting the loop as any other loop exit.
-  uint64_t EntryCount = PGO.getCurrentRegionCount();
-  RegionCounter Cnt = getPGORegionCounter(&S);
-  Builder.CreateCondBr(Builder.CreateICmpEQ(initialBufferLimit, zero, "iszero"),
-                       EmptyBB, LoopInitBB,
-                       PGO.createBranchWeights(EntryCount, Cnt.getCount()));
+  uint64_t EntryCount = getCurrentProfileCount();
+  Builder.CreateCondBr(
+      Builder.CreateICmpEQ(initialBufferLimit, zero, "iszero"), EmptyBB,
+      LoopInitBB,
+      createProfileWeights(EntryCount, getProfileCount(S.getBody())));
 
   // Otherwise, initialize the loop.
   EmitBlock(LoopInitBB);
@@ -1519,8 +1514,8 @@ void CodeGenFunction::EmitObjCForCollectionStmt(const ObjCForCollectionStmt &S){
   // Save the initial mutations value.  This is the value at an
   // address that was written into the state object by
   // countByEnumeratingWithState:objects:count:.
-  llvm::Value *StateMutationsPtrPtr =
-    Builder.CreateStructGEP(StatePtr, 2, "mutationsptr.ptr");
+  llvm::Value *StateMutationsPtrPtr = Builder.CreateStructGEP(
+      StatePtr->getAllocatedType(), StatePtr, 2, "mutationsptr.ptr");
   llvm::Value *StateMutationsPtr = Builder.CreateLoad(StateMutationsPtrPtr,
                                                       "mutationsptr");
 
@@ -1540,7 +1535,7 @@ void CodeGenFunction::EmitObjCForCollectionStmt(const ObjCForCollectionStmt &S){
   llvm::PHINode *count = Builder.CreatePHI(UnsignedLongLTy, 3, "forcoll.count");
   count->addIncoming(initialBufferLimit, LoopInitBB);
 
-  Cnt.beginRegion(Builder);
+  incrementProfileCounter(&S);
 
   // Check whether the mutations value has changed from where it was
   // at start.  StateMutationsPtr should actually be invariant between
@@ -1600,8 +1595,8 @@ void CodeGenFunction::EmitObjCForCollectionStmt(const ObjCForCollectionStmt &S){
   // Fetch the buffer out of the enumeration state.
   // TODO: this pointer should actually be invariant between
   // refreshes, which would help us do certain loop optimizations.
-  llvm::Value *StateItemsPtr =
-    Builder.CreateStructGEP(StatePtr, 1, "stateitems.ptr");
+  llvm::Value *StateItemsPtr = Builder.CreateStructGEP(
+      StatePtr->getAllocatedType(), StatePtr, 1, "stateitems.ptr");
   llvm::Value *EnumStateItems =
     Builder.CreateLoad(StateItemsPtr, "stateitems");
 
@@ -1652,9 +1647,9 @@ void CodeGenFunction::EmitObjCForCollectionStmt(const ObjCForCollectionStmt &S){
   // Set the branch weights based on the simplifying assumption that this is
   // like a while-loop, i.e., ignoring that the false branch fetches more
   // elements and then returns to the loop.
-  Builder.CreateCondBr(Builder.CreateICmpULT(indexPlusOne, count),
-                       LoopBodyBB, FetchMoreBB,
-                       PGO.createBranchWeights(Cnt.getCount(), EntryCount));
+  Builder.CreateCondBr(
+      Builder.CreateICmpULT(indexPlusOne, count), LoopBodyBB, FetchMoreBB,
+      createProfileWeights(getProfileCount(S.getBody()), EntryCount));
 
   index->addIncoming(indexPlusOne, AfterBody.getBlock());
   count->addIncoming(count, AfterBody.getBlock());
@@ -1985,7 +1980,8 @@ CodeGenFunction::EmitARCRetainAutoreleasedReturnValue(llvm::Value *value) {
   }
 
   // Call the marker asm if we made one, which we do only at -O0.
-  if (marker) Builder.CreateCall(marker);
+  if (marker)
+    Builder.CreateCall(marker, {});
 
   return emitARCValueOperation(*this, value,
                      CGM.getARCEntrypoints().objc_retainAutoreleasedReturnValue,
diff --git a/lib/CodeGen/CGObjCGNU.cpp b/lib/CodeGen/CGObjCGNU.cpp
index c0dc3b8002d8..1580c777ea47 100644
--- a/lib/CodeGen/CGObjCGNU.cpp
+++ b/lib/CodeGen/CGObjCGNU.cpp
@@ -46,54 +46,49 @@ namespace {
 /// avoids constructing the type more than once if it's used more than once.
 class LazyRuntimeFunction {
   CodeGenModule *CGM;
-  std::vector<llvm::Type*> ArgTys;
+  llvm::FunctionType *FTy;
   const char *FunctionName;
   llvm::Constant *Function;
-  public:
-    /// Constructor leaves this class uninitialized, because it is intended to
-    /// be used as a field in another class and not all of the types that are
-    /// used as arguments will necessarily be available at construction time.
-    LazyRuntimeFunction()
-      : CGM(nullptr), FunctionName(nullptr), Function(nullptr) {}
 
-    /// Initialises the lazy function with the name, return type, and the types
-    /// of the arguments.
-    LLVM_END_WITH_NULL
-    void init(CodeGenModule *Mod, const char *name,
-        llvm::Type *RetTy, ...) {
-       CGM =Mod;
-       FunctionName = name;
-       Function = nullptr;
-       ArgTys.clear();
-       va_list Args;
-       va_start(Args, RetTy);
-         while (llvm::Type *ArgTy = va_arg(Args, llvm::Type*))
-           ArgTys.push_back(ArgTy);
-       va_end(Args);
-       // Push the return type on at the end so we can pop it off easily
-       ArgTys.push_back(RetTy);
-   }
-   /// Overloaded cast operator, allows the class to be implicitly cast to an
-   /// LLVM constant.
-   operator llvm::Constant*() {
-     if (!Function) {
-       if (!FunctionName) return nullptr;
-       // We put the return type on the end of the vector, so pop it back off
-       llvm::Type *RetTy = ArgTys.back();
-       ArgTys.pop_back();
-       llvm::FunctionType *FTy = llvm::FunctionType::get(RetTy, ArgTys, false);
-       Function =
-         cast<llvm::Constant>(CGM->CreateRuntimeFunction(FTy, FunctionName));
-       // We won't need to use the types again, so we may as well clean up the
-       // vector now
-       ArgTys.resize(0);
-     }
-     return Function;
-   }
-   operator llvm::Function*() {
-     return cast<llvm::Function>((llvm::Constant*)*this);
-   }
+public:
+  /// Constructor leaves this class uninitialized, because it is intended to
+  /// be used as a field in another class and not all of the types that are
+  /// used as arguments will necessarily be available at construction time.
+  LazyRuntimeFunction()
+      : CGM(nullptr), FunctionName(nullptr), Function(nullptr) {}
 
+  /// Initialises the lazy function with the name, return type, and the types
+  /// of the arguments.
+  LLVM_END_WITH_NULL
+  void init(CodeGenModule *Mod, const char *name, llvm::Type *RetTy, ...) {
+    CGM = Mod;
+    FunctionName = name;
+    Function = nullptr;
+    std::vector<llvm::Type *> ArgTys;
+    va_list Args;
+    va_start(Args, RetTy);
+    while (llvm::Type *ArgTy = va_arg(Args, llvm::Type *))
+      ArgTys.push_back(ArgTy);
+    va_end(Args);
+    FTy = llvm::FunctionType::get(RetTy, ArgTys, false);
+  }
+
+  llvm::FunctionType *getType() { return FTy; }
+
+  /// Overloaded cast operator, allows the class to be implicitly cast to an
+  /// LLVM constant.
+  operator llvm::Constant *() {
+    if (!Function) {
+      if (!FunctionName)
+        return nullptr;
+      Function =
+          cast<llvm::Constant>(CGM->CreateRuntimeFunction(FTy, FunctionName));
+    }
+    return Function;
+  }
+  operator llvm::Function *() {
+    return cast<llvm::Function>((llvm::Constant *)*this);
+  }
 };
 
 
@@ -171,8 +166,9 @@ protected:
   /// where the C code specifies const char*.  
   llvm::Constant *MakeConstantString(const std::string &Str,
                                      const std::string &Name="") {
-    llvm::Constant *ConstStr = CGM.GetAddrOfConstantCString(Str, Name.c_str());
-    return llvm::ConstantExpr::getGetElementPtr(ConstStr, Zeros);
+    auto *ConstStr = CGM.GetAddrOfConstantCString(Str, Name.c_str());
+    return llvm::ConstantExpr::getGetElementPtr(ConstStr->getValueType(),
+                                                ConstStr, Zeros);
   }
   /// Emits a linkonce_odr string, whose name is the prefix followed by the
   /// string value.  This allows the linker to combine the strings between
@@ -181,13 +177,14 @@ protected:
   llvm::Constant *ExportUniqueString(const std::string &Str,
                                      const std::string prefix) {
     std::string name = prefix + Str;
-    llvm::Constant *ConstStr = TheModule.getGlobalVariable(name);
+    auto *ConstStr = TheModule.getGlobalVariable(name);
     if (!ConstStr) {
       llvm::Constant *value = llvm::ConstantDataArray::getString(VMContext,Str);
       ConstStr = new llvm::GlobalVariable(TheModule, value->getType(), true,
               llvm::GlobalValue::LinkOnceODRLinkage, value, prefix + Str);
     }
-    return llvm::ConstantExpr::getGetElementPtr(ConstStr, Zeros);
+    return llvm::ConstantExpr::getGetElementPtr(ConstStr->getValueType(),
+                                                ConstStr, Zeros);
   }
   /// Generates a global structure, initialized by the elements in the vector.
   /// The element types must match the types of the structure elements in the
@@ -237,8 +234,9 @@ protected:
       NameAndAttributes += TypeStr;
       NameAndAttributes += '\0';
       NameAndAttributes += PD->getNameAsString();
-      return llvm::ConstantExpr::getGetElementPtr(
-          CGM.GetAddrOfConstantCString(NameAndAttributes), Zeros);
+      auto *ConstStr = CGM.GetAddrOfConstantCString(NameAndAttributes);
+      return llvm::ConstantExpr::getGetElementPtr(ConstStr->getValueType(),
+                                                  ConstStr, Zeros);
     }
     return MakeConstantString(PD->getNameAsString());
   }
@@ -672,8 +670,8 @@ class CGObjCGNUstep : public CGObjCGNU {
       slot->setMetadata(msgSendMDKind, node);
 
       // Load the imp from the slot
-      llvm::Value *imp =
-        Builder.CreateLoad(Builder.CreateStructGEP(slot.getInstruction(), 4));
+      llvm::Value *imp = Builder.CreateLoad(
+          Builder.CreateStructGEP(nullptr, slot.getInstruction(), 4));
 
       // The lookup function may have changed the receiver, so make sure we use
       // the new one.
@@ -690,7 +688,7 @@ class CGObjCGNUstep : public CGObjCGNU {
         CGF.EmitNounwindRuntimeCall(SlotLookupSuperFn, lookupArgs);
       slot->setOnlyReadsMemory();
 
-      return Builder.CreateLoad(Builder.CreateStructGEP(slot, 4));
+      return Builder.CreateLoad(Builder.CreateStructGEP(nullptr, slot, 4));
     }
   public:
     CGObjCGNUstep(CodeGenModule &Mod) : CGObjCGNU(Mod, 9, 3) {
@@ -1013,7 +1011,7 @@ CGObjCGNU::CGObjCGNU(CodeGenModule &cgm, unsigned runtimeABIVersion,
 llvm::Value *CGObjCGNU::GetClassNamed(CodeGenFunction &CGF,
                                       const std::string &Name,
                                       bool isWeak) {
-  llvm::Value *ClassName = CGM.GetAddrOfConstantCString(Name);
+  llvm::GlobalVariable *ClassNameGV = CGM.GetAddrOfConstantCString(Name);
   // With the incompatible ABI, this will need to be replaced with a direct
   // reference to the class symbol.  For the compatible nonfragile ABI we are
   // still performing this lookup at run time but emitting the symbol for the
@@ -1023,7 +1021,8 @@ llvm::Value *CGObjCGNU::GetClassNamed(CodeGenFunction &CGF,
   // with memoized versions or with static references if it's safe to do so.
   if (!isWeak)
     EmitClassRef(Name);
-  ClassName = CGF.Builder.CreateStructGEP(ClassName, 0);
+  llvm::Value *ClassName =
+      CGF.Builder.CreateStructGEP(ClassNameGV->getValueType(), ClassNameGV, 0);
 
   llvm::Constant *ClassLookupFn =
     CGM.CreateRuntimeFunction(llvm::FunctionType::get(IdTy, PtrToInt8Ty, true),
@@ -1056,7 +1055,7 @@ llvm::Value *CGObjCGNU::GetSelector(CodeGenFunction &CGF, Selector Sel,
   }
   if (!SelValue) {
     SelValue = llvm::GlobalAlias::create(
-        SelectorTy->getElementType(), 0, llvm::GlobalValue::PrivateLinkage,
+        SelectorTy, llvm::GlobalValue::PrivateLinkage,
         ".objc_selector_" + Sel.getAsString(), &TheModule);
     Types.push_back(TypedSelector(TypeEncoding, SelValue));
   }
@@ -1143,21 +1142,22 @@ llvm::Constant *CGObjCGNUstep::GetEHType(QualType T) {
   // It's quite ugly hard-coding this.  Ideally we'd generate it using the host
   // platform's name mangling.
   const char *vtableName = "_ZTVN7gnustep7libobjc22__objc_class_type_infoE";
-  llvm::Constant *Vtable = TheModule.getGlobalVariable(vtableName);
+  auto *Vtable = TheModule.getGlobalVariable(vtableName);
   if (!Vtable) {
     Vtable = new llvm::GlobalVariable(TheModule, PtrToInt8Ty, true,
                                       llvm::GlobalValue::ExternalLinkage,
                                       nullptr, vtableName);
   }
   llvm::Constant *Two = llvm::ConstantInt::get(IntTy, 2);
-  Vtable = llvm::ConstantExpr::getGetElementPtr(Vtable, Two);
-  Vtable = llvm::ConstantExpr::getBitCast(Vtable, PtrToInt8Ty);
+  auto *BVtable = llvm::ConstantExpr::getBitCast(
+      llvm::ConstantExpr::getGetElementPtr(Vtable->getValueType(), Vtable, Two),
+      PtrToInt8Ty);
 
   llvm::Constant *typeName =
     ExportUniqueString(className, "__objc_eh_typename_");
 
   std::vector<llvm::Constant*> fields;
-  fields.push_back(Vtable);
+  fields.push_back(BVtable);
   fields.push_back(typeName);
   llvm::Constant *TI = 
       MakeGlobal(llvm::StructType::get(PtrToInt8Ty, PtrToInt8Ty,
@@ -1261,25 +1261,25 @@ CGObjCGNU::GenerateMessageSendSuper(CodeGenFunction &CGF,
     if (IsClassMessage)  {
       if (!MetaClassPtrAlias) {
         MetaClassPtrAlias = llvm::GlobalAlias::create(
-            IdTy->getElementType(), 0, llvm::GlobalValue::InternalLinkage,
+            IdTy, llvm::GlobalValue::InternalLinkage,
             ".objc_metaclass_ref" + Class->getNameAsString(), &TheModule);
       }
       ReceiverClass = MetaClassPtrAlias;
     } else {
       if (!ClassPtrAlias) {
         ClassPtrAlias = llvm::GlobalAlias::create(
-            IdTy->getElementType(), 0, llvm::GlobalValue::InternalLinkage,
+            IdTy, llvm::GlobalValue::InternalLinkage,
             ".objc_class_ref" + Class->getNameAsString(), &TheModule);
       }
       ReceiverClass = ClassPtrAlias;
     }
   }
   // Cast the pointer to a simplified version of the class structure
+  llvm::Type *CastTy = llvm::StructType::get(IdTy, IdTy, nullptr);
   ReceiverClass = Builder.CreateBitCast(ReceiverClass,
-      llvm::PointerType::getUnqual(
-        llvm::StructType::get(IdTy, IdTy, nullptr)));
+                                        llvm::PointerType::getUnqual(CastTy));
   // Get the superclass pointer
-  ReceiverClass = Builder.CreateStructGEP(ReceiverClass, 1);
+  ReceiverClass = Builder.CreateStructGEP(CastTy, ReceiverClass, 1);
   // Load the superclass pointer
   ReceiverClass = Builder.CreateLoad(ReceiverClass);
   // Construct the structure used to look up the IMP
@@ -1287,8 +1287,10 @@ CGObjCGNU::GenerateMessageSendSuper(CodeGenFunction &CGF,
       Receiver->getType(), IdTy, nullptr);
   llvm::Value *ObjCSuper = Builder.CreateAlloca(ObjCSuperTy);
 
-  Builder.CreateStore(Receiver, Builder.CreateStructGEP(ObjCSuper, 0));
-  Builder.CreateStore(ReceiverClass, Builder.CreateStructGEP(ObjCSuper, 1));
+  Builder.CreateStore(Receiver,
+                      Builder.CreateStructGEP(ObjCSuperTy, ObjCSuper, 0));
+  Builder.CreateStore(ReceiverClass,
+                      Builder.CreateStructGEP(ObjCSuperTy, ObjCSuper, 1));
 
   ObjCSuper = EnforceType(Builder, ObjCSuper, PtrToObjCSuperTy);
 
@@ -2294,7 +2296,8 @@ void CGObjCGNU::GenerateClass(const ObjCImplementationDecl *OID) {
       offsetPointerIndexes[2] = llvm::ConstantInt::get(IndexTy, ivarIndex);
       // Get the correct ivar field
       llvm::Constant *offsetValue = llvm::ConstantExpr::getGetElementPtr(
-              IvarList, offsetPointerIndexes);
+          cast<llvm::GlobalVariable>(IvarList)->getValueType(), IvarList,
+          offsetPointerIndexes);
       // Get the existing variable, if one exists.
       llvm::GlobalVariable *offset = TheModule.getNamedGlobal(Name);
       if (offset) {
@@ -2366,7 +2369,7 @@ llvm::Function *CGObjCGNU::ModuleInitFunction() {
   std::vector<llvm::Constant*> Elements;
   llvm::Constant *Statics = NULLPtr;
   // Generate statics list:
-  if (ConstantStrings.size()) {
+  if (!ConstantStrings.empty()) {
     llvm::ArrayType *StaticsArrayTy = llvm::ArrayType::get(PtrToInt8Ty,
         ConstantStrings.size() + 1);
     ConstantStrings.push_back(NULLPtr);
@@ -2439,8 +2442,8 @@ llvm::Function *CGObjCGNU::ModuleInitFunction() {
 
   // Number of static selectors
   Elements.push_back(llvm::ConstantInt::get(LongTy, SelectorCount));
-  llvm::Constant *SelectorList = MakeGlobalArray(SelStructTy, Selectors,
-          ".objc_selector_list");
+  llvm::GlobalVariable *SelectorList =
+      MakeGlobalArray(SelStructTy, Selectors, ".objc_selector_list");
   Elements.push_back(llvm::ConstantExpr::getBitCast(SelectorList,
     SelStructPtrTy));
 
@@ -2450,8 +2453,8 @@ llvm::Function *CGObjCGNU::ModuleInitFunction() {
     llvm::Constant *Idxs[] = {Zeros[0],
       llvm::ConstantInt::get(Int32Ty, i), Zeros[0]};
     // FIXME: We're generating redundant loads and stores here!
-    llvm::Constant *SelPtr = llvm::ConstantExpr::getGetElementPtr(SelectorList,
-        makeArrayRef(Idxs, 2));
+    llvm::Constant *SelPtr = llvm::ConstantExpr::getGetElementPtr(
+        SelectorList->getValueType(), SelectorList, makeArrayRef(Idxs, 2));
     // If selectors are defined as an opaque type, cast the pointer to this
     // type.
     SelPtr = llvm::ConstantExpr::getBitCast(SelPtr, SelectorTy);
@@ -2562,8 +2565,8 @@ llvm::Function *CGObjCGNU::ModuleInitFunction() {
             true);
        if (TheClass) {
          TheClass = llvm::ConstantExpr::getBitCast(TheClass, PtrTy);
-         Builder.CreateCall2(RegisterAlias, TheClass,
-            MakeConstantString(iter->second));
+         Builder.CreateCall(RegisterAlias,
+                            {TheClass, MakeConstantString(iter->second)});
        }
     }
     // Jump to end:
@@ -2679,7 +2682,7 @@ llvm::Value * CGObjCGNU::EmitObjCWeakRead(CodeGenFunction &CGF,
                                           llvm::Value *AddrWeakObj) {
   CGBuilderTy &B = CGF.Builder;
   AddrWeakObj = EnforceType(B, AddrWeakObj, PtrToIdTy);
-  return B.CreateCall(WeakReadFn, AddrWeakObj);
+  return B.CreateCall(WeakReadFn.getType(), WeakReadFn, AddrWeakObj);
 }
 
 void CGObjCGNU::EmitObjCWeakAssign(CodeGenFunction &CGF,
@@ -2687,7 +2690,7 @@ void CGObjCGNU::EmitObjCWeakAssign(CodeGenFunction &CGF,
   CGBuilderTy &B = CGF.Builder;
   src = EnforceType(B, src, IdTy);
   dst = EnforceType(B, dst, PtrToIdTy);
-  B.CreateCall2(WeakAssignFn, src, dst);
+  B.CreateCall(WeakAssignFn.getType(), WeakAssignFn, {src, dst});
 }
 
 void CGObjCGNU::EmitObjCGlobalAssign(CodeGenFunction &CGF,
@@ -2696,11 +2699,9 @@ void CGObjCGNU::EmitObjCGlobalAssign(CodeGenFunction &CGF,
   CGBuilderTy &B = CGF.Builder;
   src = EnforceType(B, src, IdTy);
   dst = EnforceType(B, dst, PtrToIdTy);
-  if (!threadlocal)
-    B.CreateCall2(GlobalAssignFn, src, dst);
-  else
-    // FIXME. Add threadloca assign API
-    llvm_unreachable("EmitObjCGlobalAssign - Threal Local API NYI");
+  // FIXME. Add threadloca assign API
+  assert(!threadlocal && "EmitObjCGlobalAssign - Threal Local API NYI");
+  B.CreateCall(GlobalAssignFn.getType(), GlobalAssignFn, {src, dst});
 }
 
 void CGObjCGNU::EmitObjCIvarAssign(CodeGenFunction &CGF,
@@ -2709,7 +2710,7 @@ void CGObjCGNU::EmitObjCIvarAssign(CodeGenFunction &CGF,
   CGBuilderTy &B = CGF.Builder;
   src = EnforceType(B, src, IdTy);
   dst = EnforceType(B, dst, IdTy);
-  B.CreateCall3(IvarAssignFn, src, dst, ivarOffset);
+  B.CreateCall(IvarAssignFn.getType(), IvarAssignFn, {src, dst, ivarOffset});
 }
 
 void CGObjCGNU::EmitObjCStrongCastAssign(CodeGenFunction &CGF,
@@ -2717,7 +2718,7 @@ void CGObjCGNU::EmitObjCStrongCastAssign(CodeGenFunction &CGF,
   CGBuilderTy &B = CGF.Builder;
   src = EnforceType(B, src, IdTy);
   dst = EnforceType(B, dst, PtrToIdTy);
-  B.CreateCall2(StrongCastAssignFn, src, dst);
+  B.CreateCall(StrongCastAssignFn.getType(), StrongCastAssignFn, {src, dst});
 }
 
 void CGObjCGNU::EmitGCMemmoveCollectable(CodeGenFunction &CGF,
@@ -2728,7 +2729,7 @@ void CGObjCGNU::EmitGCMemmoveCollectable(CodeGenFunction &CGF,
   DestPtr = EnforceType(B, DestPtr, PtrTy);
   SrcPtr = EnforceType(B, SrcPtr, PtrTy);
 
-  B.CreateCall3(MemMoveFn, DestPtr, SrcPtr, Size);
+  B.CreateCall(MemMoveFn.getType(), MemMoveFn, {DestPtr, SrcPtr, Size});
 }
 
 llvm::GlobalVariable *CGObjCGNU::ObjCIvarOffsetVariable(
diff --git a/lib/CodeGen/CGObjCMac.cpp b/lib/CodeGen/CGObjCMac.cpp
index f91e8e15b039..a45446a7065a 100644
--- a/lib/CodeGen/CGObjCMac.cpp
+++ b/lib/CodeGen/CGObjCMac.cpp
@@ -486,7 +486,6 @@ public:
   }
 
   ObjCCommonTypesHelper(CodeGen::CodeGenModule &cgm);
-  ~ObjCCommonTypesHelper(){}
 };
 
 /// ObjCTypesHelper - Helper class that encapsulates lazy
@@ -595,7 +594,6 @@ public:
 
 public:
   ObjCTypesHelper(CodeGen::CodeGenModule &cgm);
-  ~ObjCTypesHelper() {}
 };
 
 /// ObjCNonFragileABITypesHelper - will have all types needed by objective-c's
@@ -733,7 +731,6 @@ public:
   llvm::Type *EHTypePtrTy;
   
   ObjCNonFragileABITypesHelper(CodeGen::CodeGenModule &cgm);
-  ~ObjCNonFragileABITypesHelper(){}
 };
 
 class CGObjCCommonMac : public CodeGen::CGObjCRuntime {
@@ -1678,14 +1675,13 @@ struct NullReturnState {
 
 /// getConstantGEP() - Help routine to construct simple GEPs.
 static llvm::Constant *getConstantGEP(llvm::LLVMContext &VMContext,
-                                      llvm::Constant *C,
-                                      unsigned idx0,
+                                      llvm::GlobalVariable *C, unsigned idx0,
                                       unsigned idx1) {
   llvm::Value *Idxs[] = {
     llvm::ConstantInt::get(llvm::Type::getInt32Ty(VMContext), idx0),
     llvm::ConstantInt::get(llvm::Type::getInt32Ty(VMContext), idx1)
   };
-  return llvm::ConstantExpr::getGetElementPtr(C, Idxs);
+  return llvm::ConstantExpr::getGetElementPtr(C->getValueType(), C, Idxs);
 }
 
 /// hasObjCExceptionAttribute - Return true if this class or any super
@@ -1791,8 +1787,9 @@ CGObjCMac::GenerateMessageSendSuper(CodeGen::CodeGenFunction &CGF,
     CGF.CreateTempAlloca(ObjCTypes.SuperTy, "objc_super");
   llvm::Value *ReceiverAsObject =
     CGF.Builder.CreateBitCast(Receiver, ObjCTypes.ObjectPtrTy);
-  CGF.Builder.CreateStore(ReceiverAsObject,
-                          CGF.Builder.CreateStructGEP(ObjCSuper, 0));
+  CGF.Builder.CreateStore(
+      ReceiverAsObject,
+      CGF.Builder.CreateStructGEP(ObjCTypes.SuperTy, ObjCSuper, 0));
 
   // If this is a class message the metaclass is passed as the target.
   llvm::Value *Target;
@@ -1805,20 +1802,20 @@ CGObjCMac::GenerateMessageSendSuper(CodeGen::CodeGenFunction &CGF,
       // the class's "isa" pointer.  The following assumes that
       // isa" is the first ivar in a class (which it must be).
       Target = EmitClassRef(CGF, Class->getSuperClass());
-      Target = CGF.Builder.CreateStructGEP(Target, 0);
+      Target = CGF.Builder.CreateStructGEP(ObjCTypes.ClassTy, Target, 0);
       Target = CGF.Builder.CreateLoad(Target);
     } else {
-      llvm::Value *MetaClassPtr = EmitMetaClassRef(Class);
-      llvm::Value *SuperPtr = CGF.Builder.CreateStructGEP(MetaClassPtr, 1);
+      llvm::Constant *MetaClassPtr = EmitMetaClassRef(Class);
+      llvm::Value *SuperPtr =
+          CGF.Builder.CreateStructGEP(ObjCTypes.ClassTy, MetaClassPtr, 1);
       llvm::Value *Super = CGF.Builder.CreateLoad(SuperPtr);
       Target = Super;
     }
-  } 
-  else if (isCategoryImpl)
+  } else if (isCategoryImpl)
     Target = EmitClassRef(CGF, Class->getSuperClass());
   else {
     llvm::Value *ClassPtr = EmitSuperClassRef(Class);
-    ClassPtr = CGF.Builder.CreateStructGEP(ClassPtr, 1);
+    ClassPtr = CGF.Builder.CreateStructGEP(ObjCTypes.ClassTy, ClassPtr, 1);
     Target = CGF.Builder.CreateLoad(ClassPtr);
   }
   // FIXME: We shouldn't need to do this cast, rectify the ASTContext and
@@ -1826,8 +1823,8 @@ CGObjCMac::GenerateMessageSendSuper(CodeGen::CodeGenFunction &CGF,
   llvm::Type *ClassTy =
     CGM.getTypes().ConvertType(CGF.getContext().getObjCClassType());
   Target = CGF.Builder.CreateBitCast(Target, ClassTy);
-  CGF.Builder.CreateStore(Target,
-                          CGF.Builder.CreateStructGEP(ObjCSuper, 1));
+  CGF.Builder.CreateStore(
+      Target, CGF.Builder.CreateStructGEP(ObjCTypes.SuperTy, ObjCSuper, 1));
   return EmitMessageSend(CGF, Return, ResultType,
                          EmitSelector(CGF, Sel),
                          ObjCSuper, ObjCTypes.SuperPtrCTy,
@@ -3810,15 +3807,16 @@ void CGObjCMac::EmitTryOrSynchronizedStmt(CodeGen::CodeGenFunction &CGF,
   // Enter a try block:
   //  - Call objc_exception_try_enter to push ExceptionData on top of
   //    the EH stack.
-  CGF.EmitNounwindRuntimeCall(ObjCTypes.getExceptionTryEnterFn(), ExceptionData);
+  CGF.EmitNounwindRuntimeCall(ObjCTypes.getExceptionTryEnterFn(),
+                              ExceptionData);
 
   //  - Call setjmp on the exception data buffer.
   llvm::Constant *Zero = llvm::ConstantInt::get(CGF.Builder.getInt32Ty(), 0);
   llvm::Value *GEPIndexes[] = { Zero, Zero, Zero };
-  llvm::Value *SetJmpBuffer =
-    CGF.Builder.CreateGEP(ExceptionData, GEPIndexes, "setjmp_buffer");
-  llvm::CallInst *SetJmpResult =
-    CGF.EmitNounwindRuntimeCall(ObjCTypes.getSetJmpFn(), SetJmpBuffer, "setjmp_result");
+  llvm::Value *SetJmpBuffer = CGF.Builder.CreateGEP(
+      ObjCTypes.ExceptionDataTy, ExceptionData, GEPIndexes, "setjmp_buffer");
+  llvm::CallInst *SetJmpResult = CGF.EmitNounwindRuntimeCall(
+      ObjCTypes.getSetJmpFn(), SetJmpBuffer, "setjmp_result");
   SetJmpResult->setCanReturnTwice();
 
   // If setjmp returned 0, enter the protected block; otherwise,
@@ -5263,6 +5261,7 @@ ObjCNonFragileABITypesHelper::ObjCNonFragileABITypesHelper(CodeGen::CodeGenModul
   //   const uint32_t size;  // sizeof(struct _protocol_t)
   //   const uint32_t flags;  // = 0
   //   const char ** extendedMethodTypes;
+  //   const char *demangledName;
   // }
 
   // Holder for struct _protocol_list_t *
@@ -5275,6 +5274,7 @@ ObjCNonFragileABITypesHelper::ObjCNonFragileABITypesHelper(CodeGen::CodeGenModul
                              MethodListnfABIPtrTy, MethodListnfABIPtrTy,
                              MethodListnfABIPtrTy, MethodListnfABIPtrTy,
                              PropertyListPtrTy, IntTy, IntTy, Int8PtrPtrTy,
+                             Int8PtrTy,
                              nullptr);
 
   // struct _protocol_t*
@@ -6207,6 +6207,7 @@ llvm::Constant *CGObjCNonFragileABIMac::GetOrEmitProtocolRef(
 ///   const uint32_t size;  // sizeof(struct _protocol_t)
 ///   const uint32_t flags;  // = 0
 ///   const char ** extendedMethodTypes;
+///   const char *demangledName;
 /// }
 /// @endcode
 ///
@@ -6258,7 +6259,7 @@ llvm::Constant *CGObjCNonFragileABIMac::GetOrEmitProtocol(
   MethodTypesExt.insert(MethodTypesExt.end(),
                         OptMethodTypesExt.begin(), OptMethodTypesExt.end());
 
-  llvm::Constant *Values[11];
+  llvm::Constant *Values[12];
   // isa is NULL
   Values[0] = llvm::Constant::getNullValue(ObjCTypes.ObjectPtrTy);
   Values[1] = GetClassName(PD->getObjCRuntimeNameAsString());
@@ -6291,6 +6292,9 @@ llvm::Constant *CGObjCNonFragileABIMac::GetOrEmitProtocol(
   Values[10] = EmitProtocolMethodTypes("\01l_OBJC_$_PROTOCOL_METHOD_TYPES_"
                                        + PD->getObjCRuntimeNameAsString(),
                                        MethodTypesExt, ObjCTypes);
+  // const char *demangledName;
+  Values[11] = llvm::Constant::getNullValue(ObjCTypes.Int8PtrTy);
+    
   llvm::Constant *Init = llvm::ConstantStruct::get(ObjCTypes.ProtocolnfABITy,
                                                    Values);
 
@@ -6562,7 +6566,8 @@ CGObjCNonFragileABIMac::EmitVTableMessageSend(CodeGenFunction &CGF,
   args[1].RV = RValue::get(mref);
 
   // Load the function to call from the message ref table.
-  llvm::Value *callee = CGF.Builder.CreateStructGEP(mref, 0);
+  llvm::Value *callee =
+      CGF.Builder.CreateStructGEP(ObjCTypes.MessageRefTy, mref, 0);
   callee = CGF.Builder.CreateLoad(callee, "msgSend_fn");
 
   callee = CGF.Builder.CreateBitCast(callee, MSI.MessengerType);
@@ -6727,8 +6732,9 @@ CGObjCNonFragileABIMac::GenerateMessageSendSuper(CodeGen::CodeGenFunction &CGF,
 
   llvm::Value *ReceiverAsObject =
     CGF.Builder.CreateBitCast(Receiver, ObjCTypes.ObjectPtrTy);
-  CGF.Builder.CreateStore(ReceiverAsObject,
-                          CGF.Builder.CreateStructGEP(ObjCSuper, 0));
+  CGF.Builder.CreateStore(
+      ReceiverAsObject,
+      CGF.Builder.CreateStructGEP(ObjCTypes.SuperTy, ObjCSuper, 0));
 
   // If this is a class message the metaclass is passed as the target.
   llvm::Value *Target;
@@ -6742,8 +6748,8 @@ CGObjCNonFragileABIMac::GenerateMessageSendSuper(CodeGen::CodeGenFunction &CGF,
   llvm::Type *ClassTy =
     CGM.getTypes().ConvertType(CGF.getContext().getObjCClassType());
   Target = CGF.Builder.CreateBitCast(Target, ClassTy);
-  CGF.Builder.CreateStore(Target,
-                          CGF.Builder.CreateStructGEP(ObjCSuper, 1));
+  CGF.Builder.CreateStore(
+      Target, CGF.Builder.CreateStructGEP(ObjCTypes.SuperTy, ObjCSuper, 1));
 
   return (isVTableDispatchedSelector(Sel))
     ? EmitVTableMessageSend(CGF, Return, ResultType, Sel,
@@ -6992,10 +6998,10 @@ CGObjCNonFragileABIMac::GetInterfaceEHType(const ObjCInterfaceDecl *ID,
   llvm::Value *VTableIdx = llvm::ConstantInt::get(CGM.Int32Ty, 2);
 
   llvm::Constant *Values[] = {
-    llvm::ConstantExpr::getGetElementPtr(VTableGV, VTableIdx),
-    GetClassName(ID->getObjCRuntimeNameAsString()),
-    GetClassGlobal(ClassName.str())
-  };
+      llvm::ConstantExpr::getGetElementPtr(VTableGV->getValueType(), VTableGV,
+                                           VTableIdx),
+      GetClassName(ID->getObjCRuntimeNameAsString()),
+      GetClassGlobal(ClassName.str())};
   llvm::Constant *Init =
     llvm::ConstantStruct::get(ObjCTypes.EHTypeTy, Values);
 
diff --git a/lib/CodeGen/CGObjCRuntime.cpp b/lib/CodeGen/CGObjCRuntime.cpp
index 3d013da51e65..5290a87cebf2 100644
--- a/lib/CodeGen/CGObjCRuntime.cpp
+++ b/lib/CodeGen/CGObjCRuntime.cpp
@@ -160,7 +160,7 @@ namespace {
 
     void Emit(CodeGenFunction &CGF, Flags flags) override {
       if (!MightThrow) {
-        CGF.Builder.CreateCall(Fn)->setDoesNotThrow();
+        CGF.Builder.CreateCall(Fn, {})->setDoesNotThrow();
         return;
       }
 
diff --git a/lib/CodeGen/CGOpenMPRuntime.cpp b/lib/CodeGen/CGOpenMPRuntime.cpp
index 22ee00f2c7ae..1238accf42d7 100644
--- a/lib/CodeGen/CGOpenMPRuntime.cpp
+++ b/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -13,6 +13,7 @@
 
 #include "CGOpenMPRuntime.h"
 #include "CodeGenFunction.h"
+#include "CGCleanup.h"
 #include "clang/AST/Decl.h"
 #include "clang/AST/StmtOpenMP.h"
 #include "llvm/ADT/ArrayRef.h"
@@ -27,61 +28,236 @@ using namespace clang;
 using namespace CodeGen;
 
 namespace {
-/// \brief API for captured statement code generation in OpenMP constructs.
+/// \brief Base class for handling code generation inside OpenMP regions.
 class CGOpenMPRegionInfo : public CodeGenFunction::CGCapturedStmtInfo {
 public:
-  CGOpenMPRegionInfo(const OMPExecutableDirective &D, const CapturedStmt &CS,
-                     const VarDecl *ThreadIDVar)
-      : CGCapturedStmtInfo(CS, CR_OpenMP), ThreadIDVar(ThreadIDVar),
-        Directive(D) {
-    assert(ThreadIDVar != nullptr && "No ThreadID in OpenMP region.");
-  }
+  /// \brief Kinds of OpenMP regions used in codegen.
+  enum CGOpenMPRegionKind {
+    /// \brief Region with outlined function for standalone 'parallel'
+    /// directive.
+    ParallelOutlinedRegion,
+    /// \brief Region with outlined function for standalone 'task' directive.
+    TaskOutlinedRegion,
+    /// \brief Region for constructs that do not require function outlining,
+    /// like 'for', 'sections', 'atomic' etc. directives.
+    InlinedRegion,
+  };
+
+  CGOpenMPRegionInfo(const CapturedStmt &CS,
+                     const CGOpenMPRegionKind RegionKind,
+                     const RegionCodeGenTy &CodeGen)
+      : CGCapturedStmtInfo(CS, CR_OpenMP), RegionKind(RegionKind),
+        CodeGen(CodeGen) {}
+
+  CGOpenMPRegionInfo(const CGOpenMPRegionKind RegionKind,
+                     const RegionCodeGenTy &CodeGen)
+      : CGCapturedStmtInfo(CR_OpenMP), RegionKind(RegionKind),
+        CodeGen(CodeGen) {}
 
-  /// \brief Gets a variable or parameter for storing global thread id
+  /// \brief Get a variable or parameter for storing global thread id
   /// inside OpenMP construct.
-  const VarDecl *getThreadIDVariable() const { return ThreadIDVar; }
+  virtual const VarDecl *getThreadIDVariable() const = 0;
 
-  /// \brief Gets an LValue for the current ThreadID variable.
-  LValue getThreadIDVariableLValue(CodeGenFunction &CGF);
+  /// \brief Emit the captured statement body.
+  virtual void EmitBody(CodeGenFunction &CGF, const Stmt *S) override;
+
+  /// \brief Get an LValue for the current ThreadID variable.
+  /// \return LValue for thread id variable. This LValue always has type int32*.
+  virtual LValue getThreadIDVariableLValue(CodeGenFunction &CGF);
+
+  CGOpenMPRegionKind getRegionKind() const { return RegionKind; }
 
   static bool classof(const CGCapturedStmtInfo *Info) {
     return Info->getKind() == CR_OpenMP;
   }
 
-  /// \brief Emit the captured statement body.
-  void EmitBody(CodeGenFunction &CGF, Stmt *S) override;
+protected:
+  CGOpenMPRegionKind RegionKind;
+  const RegionCodeGenTy &CodeGen;
+};
+
+/// \brief API for captured statement code generation in OpenMP constructs.
+class CGOpenMPOutlinedRegionInfo : public CGOpenMPRegionInfo {
+public:
+  CGOpenMPOutlinedRegionInfo(const CapturedStmt &CS, const VarDecl *ThreadIDVar,
+                             const RegionCodeGenTy &CodeGen)
+      : CGOpenMPRegionInfo(CS, ParallelOutlinedRegion, CodeGen),
+        ThreadIDVar(ThreadIDVar) {
+    assert(ThreadIDVar != nullptr && "No ThreadID in OpenMP region.");
+  }
+  /// \brief Get a variable or parameter for storing global thread id
+  /// inside OpenMP construct.
+  const VarDecl *getThreadIDVariable() const override { return ThreadIDVar; }
 
   /// \brief Get the name of the capture helper.
   StringRef getHelperName() const override { return ".omp_outlined."; }
 
+  static bool classof(const CGCapturedStmtInfo *Info) {
+    return CGOpenMPRegionInfo::classof(Info) &&
+           cast<CGOpenMPRegionInfo>(Info)->getRegionKind() ==
+               ParallelOutlinedRegion;
+  }
+
 private:
   /// \brief A variable or parameter storing global thread id for OpenMP
   /// constructs.
   const VarDecl *ThreadIDVar;
-  /// \brief OpenMP executable directive associated with the region.
-  const OMPExecutableDirective &Directive;
 };
+
+/// \brief API for captured statement code generation in OpenMP constructs.
+class CGOpenMPTaskOutlinedRegionInfo : public CGOpenMPRegionInfo {
+public:
+  CGOpenMPTaskOutlinedRegionInfo(const CapturedStmt &CS,
+                                 const VarDecl *ThreadIDVar,
+                                 const RegionCodeGenTy &CodeGen)
+      : CGOpenMPRegionInfo(CS, TaskOutlinedRegion, CodeGen),
+        ThreadIDVar(ThreadIDVar) {
+    assert(ThreadIDVar != nullptr && "No ThreadID in OpenMP region.");
+  }
+  /// \brief Get a variable or parameter for storing global thread id
+  /// inside OpenMP construct.
+  const VarDecl *getThreadIDVariable() const override { return ThreadIDVar; }
+
+  /// \brief Get an LValue for the current ThreadID variable.
+  LValue getThreadIDVariableLValue(CodeGenFunction &CGF) override;
+
+  /// \brief Get the name of the capture helper.
+  StringRef getHelperName() const override { return ".omp_outlined."; }
+
+  static bool classof(const CGCapturedStmtInfo *Info) {
+    return CGOpenMPRegionInfo::classof(Info) &&
+           cast<CGOpenMPRegionInfo>(Info)->getRegionKind() ==
+               TaskOutlinedRegion;
+  }
+
+private:
+  /// \brief A variable or parameter storing global thread id for OpenMP
+  /// constructs.
+  const VarDecl *ThreadIDVar;
+};
+
+/// \brief API for inlined captured statement code generation in OpenMP
+/// constructs.
+class CGOpenMPInlinedRegionInfo : public CGOpenMPRegionInfo {
+public:
+  CGOpenMPInlinedRegionInfo(CodeGenFunction::CGCapturedStmtInfo *OldCSI,
+                            const RegionCodeGenTy &CodeGen)
+      : CGOpenMPRegionInfo(InlinedRegion, CodeGen), OldCSI(OldCSI),
+        OuterRegionInfo(dyn_cast_or_null<CGOpenMPRegionInfo>(OldCSI)) {}
+  // \brief Retrieve the value of the context parameter.
+  llvm::Value *getContextValue() const override {
+    if (OuterRegionInfo)
+      return OuterRegionInfo->getContextValue();
+    llvm_unreachable("No context value for inlined OpenMP region");
+  }
+  virtual void setContextValue(llvm::Value *V) override {
+    if (OuterRegionInfo) {
+      OuterRegionInfo->setContextValue(V);
+      return;
+    }
+    llvm_unreachable("No context value for inlined OpenMP region");
+  }
+  /// \brief Lookup the captured field decl for a variable.
+  const FieldDecl *lookup(const VarDecl *VD) const override {
+    if (OuterRegionInfo)
+      return OuterRegionInfo->lookup(VD);
+    // If there is no outer outlined region,no need to lookup in a list of
+    // captured variables, we can use the original one.
+    return nullptr;
+  }
+  FieldDecl *getThisFieldDecl() const override {
+    if (OuterRegionInfo)
+      return OuterRegionInfo->getThisFieldDecl();
+    return nullptr;
+  }
+  /// \brief Get a variable or parameter for storing global thread id
+  /// inside OpenMP construct.
+  const VarDecl *getThreadIDVariable() const override {
+    if (OuterRegionInfo)
+      return OuterRegionInfo->getThreadIDVariable();
+    return nullptr;
+  }
+
+  /// \brief Get the name of the capture helper.
+  StringRef getHelperName() const override {
+    if (auto *OuterRegionInfo = getOldCSI())
+      return OuterRegionInfo->getHelperName();
+    llvm_unreachable("No helper name for inlined OpenMP construct");
+  }
+
+  CodeGenFunction::CGCapturedStmtInfo *getOldCSI() const { return OldCSI; }
+
+  static bool classof(const CGCapturedStmtInfo *Info) {
+    return CGOpenMPRegionInfo::classof(Info) &&
+           cast<CGOpenMPRegionInfo>(Info)->getRegionKind() == InlinedRegion;
+  }
+
+private:
+  /// \brief CodeGen info about outer OpenMP region.
+  CodeGenFunction::CGCapturedStmtInfo *OldCSI;
+  CGOpenMPRegionInfo *OuterRegionInfo;
+};
+
+/// \brief RAII for emitting code of OpenMP constructs.
+class InlinedOpenMPRegionRAII {
+  CodeGenFunction &CGF;
+
+public:
+  /// \brief Constructs region for combined constructs.
+  /// \param CodeGen Code generation sequence for combined directives. Includes
+  /// a list of functions used for code generation of implicitly inlined
+  /// regions.
+  InlinedOpenMPRegionRAII(CodeGenFunction &CGF, const RegionCodeGenTy &CodeGen)
+      : CGF(CGF) {
+    // Start emission for the construct.
+    CGF.CapturedStmtInfo =
+        new CGOpenMPInlinedRegionInfo(CGF.CapturedStmtInfo, CodeGen);
+  }
+  ~InlinedOpenMPRegionRAII() {
+    // Restore original CapturedStmtInfo only if we're done with code emission.
+    auto *OldCSI =
+        cast<CGOpenMPInlinedRegionInfo>(CGF.CapturedStmtInfo)->getOldCSI();
+    delete CGF.CapturedStmtInfo;
+    CGF.CapturedStmtInfo = OldCSI;
+  }
+};
+
 } // namespace
 
 LValue CGOpenMPRegionInfo::getThreadIDVariableLValue(CodeGenFunction &CGF) {
   return CGF.MakeNaturalAlignAddrLValue(
-      CGF.GetAddrOfLocalVar(ThreadIDVar),
-      CGF.getContext().getPointerType(ThreadIDVar->getType()));
+      CGF.Builder.CreateAlignedLoad(
+          CGF.GetAddrOfLocalVar(getThreadIDVariable()),
+          CGF.PointerAlignInBytes),
+      getThreadIDVariable()
+          ->getType()
+          ->castAs<PointerType>()
+          ->getPointeeType());
 }
 
-void CGOpenMPRegionInfo::EmitBody(CodeGenFunction &CGF, Stmt *S) {
-  CodeGenFunction::OMPPrivateScope PrivateScope(CGF);
-  CGF.EmitOMPPrivateClause(Directive, PrivateScope);
-  CGF.EmitOMPFirstprivateClause(Directive, PrivateScope);
-  if (PrivateScope.Privatize())
-    // Emit implicit barrier to synchronize threads and avoid data races.
-    CGF.CGM.getOpenMPRuntime().EmitOMPBarrierCall(CGF, Directive.getLocStart(),
-                                                  /*IsExplicit=*/false);
-  CGCapturedStmtInfo::EmitBody(CGF, S);
+void CGOpenMPRegionInfo::EmitBody(CodeGenFunction &CGF, const Stmt * /*S*/) {
+  // 1.2.2 OpenMP Language Terminology
+  // Structured block - An executable statement with a single entry at the
+  // top and a single exit at the bottom.
+  // The point of exit cannot be a branch out of the structured block.
+  // longjmp() and throw() must not violate the entry/exit criteria.
+  CGF.EHStack.pushTerminate();
+  {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    CodeGen(CGF);
+  }
+  CGF.EHStack.popTerminate();
+}
+
+LValue CGOpenMPTaskOutlinedRegionInfo::getThreadIDVariableLValue(
+    CodeGenFunction &CGF) {
+  return CGF.MakeNaturalAlignAddrLValue(
+      CGF.GetAddrOfLocalVar(getThreadIDVariable()),
+      getThreadIDVariable()->getType());
 }
 
 CGOpenMPRuntime::CGOpenMPRuntime(CodeGenModule &CGM)
-    : CGM(CGM), DefaultOpenMPPSource(nullptr) {
+    : CGM(CGM), DefaultOpenMPPSource(nullptr), KmpRoutineEntryPtrTy(nullptr) {
   IdentTy = llvm::StructType::create(
       "ident_t", CGM.Int32Ty /* reserved_1 */, CGM.Int32Ty /* flags */,
       CGM.Int32Ty /* reserved_2 */, CGM.Int32Ty /* reserved_3 */,
@@ -93,18 +269,38 @@ CGOpenMPRuntime::CGOpenMPRuntime(CodeGenModule &CGM)
   KmpCriticalNameTy = llvm::ArrayType::get(CGM.Int32Ty, /*NumElements*/ 8);
 }
 
+void CGOpenMPRuntime::clear() {
+  InternalVars.clear();
+}
+
 llvm::Value *
-CGOpenMPRuntime::EmitOpenMPOutlinedFunction(const OMPExecutableDirective &D,
-                                            const VarDecl *ThreadIDVar) {
+CGOpenMPRuntime::emitParallelOutlinedFunction(const OMPExecutableDirective &D,
+                                              const VarDecl *ThreadIDVar,
+                                              const RegionCodeGenTy &CodeGen) {
+  assert(ThreadIDVar->getType()->isPointerType() &&
+         "thread id variable must be of type kmp_int32 *");
   const CapturedStmt *CS = cast<CapturedStmt>(D.getAssociatedStmt());
   CodeGenFunction CGF(CGM, true);
-  CGOpenMPRegionInfo CGInfo(D, *CS, ThreadIDVar);
+  CGOpenMPOutlinedRegionInfo CGInfo(*CS, ThreadIDVar, CodeGen);
   CGF.CapturedStmtInfo = &CGInfo;
   return CGF.GenerateCapturedStmtFunction(*CS);
 }
 
 llvm::Value *
-CGOpenMPRuntime::GetOrCreateDefaultOpenMPLocation(OpenMPLocationFlags Flags) {
+CGOpenMPRuntime::emitTaskOutlinedFunction(const OMPExecutableDirective &D,
+                                          const VarDecl *ThreadIDVar,
+                                          const RegionCodeGenTy &CodeGen) {
+  assert(!ThreadIDVar->getType()->isPointerType() &&
+         "thread id variable must be of type kmp_int32 for tasks");
+  auto *CS = cast<CapturedStmt>(D.getAssociatedStmt());
+  CodeGenFunction CGF(CGM, true);
+  CGOpenMPTaskOutlinedRegionInfo CGInfo(*CS, ThreadIDVar, CodeGen);
+  CGF.CapturedStmtInfo = &CGInfo;
+  return CGF.GenerateCapturedStmtFunction(*CS);
+}
+
+llvm::Value *
+CGOpenMPRuntime::getOrCreateDefaultLocation(OpenMPLocationFlags Flags) {
   llvm::Value *Entry = OpenMPDefaultLocMap.lookup(Flags);
   if (!Entry) {
     if (!DefaultOpenMPPSource) {
@@ -134,12 +330,13 @@ CGOpenMPRuntime::GetOrCreateDefaultOpenMPLocation(OpenMPLocationFlags Flags) {
   return Entry;
 }
 
-llvm::Value *CGOpenMPRuntime::EmitOpenMPUpdateLocation(
-    CodeGenFunction &CGF, SourceLocation Loc, OpenMPLocationFlags Flags) {
+llvm::Value *CGOpenMPRuntime::emitUpdateLocation(CodeGenFunction &CGF,
+                                                 SourceLocation Loc,
+                                                 OpenMPLocationFlags Flags) {
   // If no debug info is generated - return global default location.
   if (CGM.getCodeGenOpts().getDebugInfo() == CodeGenOptions::NoDebugInfo ||
       Loc.isInvalid())
-    return GetOrCreateDefaultOpenMPLocation(Flags);
+    return getOrCreateDefaultLocation(Flags);
 
   assert(CGF.CurFn && "No function in current CodeGenFunction.");
 
@@ -159,14 +356,14 @@ llvm::Value *CGOpenMPRuntime::EmitOpenMPUpdateLocation(
 
     CGBuilderTy::InsertPointGuard IPG(CGF.Builder);
     CGF.Builder.SetInsertPoint(CGF.AllocaInsertPt);
-    CGF.Builder.CreateMemCpy(LocValue, GetOrCreateDefaultOpenMPLocation(Flags),
+    CGF.Builder.CreateMemCpy(LocValue, getOrCreateDefaultLocation(Flags),
                              llvm::ConstantExpr::getSizeOf(IdentTy),
                              CGM.PointerAlignInBytes);
   }
 
   // char **psource = &.kmpc_loc_<flags>.addr.psource;
-  auto *PSource =
-      CGF.Builder.CreateConstInBoundsGEP2_32(LocValue, 0, IdentField_PSource);
+  auto *PSource = CGF.Builder.CreateConstInBoundsGEP2_32(IdentTy, LocValue, 0,
+                                                         IdentField_PSource);
 
   auto OMPDebugLoc = OpenMPDebugLocMap.lookup(Loc.getRawEncoding());
   if (OMPDebugLoc == nullptr) {
@@ -189,8 +386,8 @@ llvm::Value *CGOpenMPRuntime::EmitOpenMPUpdateLocation(
   return LocValue;
 }
 
-llvm::Value *CGOpenMPRuntime::GetOpenMPThreadID(CodeGenFunction &CGF,
-                                                SourceLocation Loc) {
+llvm::Value *CGOpenMPRuntime::getThreadID(CodeGenFunction &CGF,
+                                          SourceLocation Loc) {
   assert(CGF.CurFn && "No function in current CodeGenFunction.");
 
   llvm::Value *ThreadID = nullptr;
@@ -204,36 +401,35 @@ llvm::Value *CGOpenMPRuntime::GetOpenMPThreadID(CodeGenFunction &CGF,
   }
   if (auto OMPRegionInfo =
           dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo)) {
-    // Check if this an outlined function with thread id passed as argument.
-    auto ThreadIDVar = OMPRegionInfo->getThreadIDVariable();
-    auto LVal = OMPRegionInfo->getThreadIDVariableLValue(CGF);
-    auto RVal = CGF.EmitLoadOfLValue(LVal, Loc);
-    LVal = CGF.MakeNaturalAlignAddrLValue(RVal.getScalarVal(),
-                                          ThreadIDVar->getType());
-    ThreadID = CGF.EmitLoadOfLValue(LVal, Loc).getScalarVal();
-    // If value loaded in entry block, cache it and use it everywhere in
-    // function.
-    if (CGF.Builder.GetInsertBlock() == CGF.AllocaInsertPt->getParent()) {
-      auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
-      Elem.second.ThreadID = ThreadID;
+    if (OMPRegionInfo->getThreadIDVariable()) {
+      // Check if this an outlined function with thread id passed as argument.
+      auto LVal = OMPRegionInfo->getThreadIDVariableLValue(CGF);
+      ThreadID = CGF.EmitLoadOfLValue(LVal, Loc).getScalarVal();
+      // If value loaded in entry block, cache it and use it everywhere in
+      // function.
+      if (CGF.Builder.GetInsertBlock() == CGF.AllocaInsertPt->getParent()) {
+        auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
+        Elem.second.ThreadID = ThreadID;
+      }
+      return ThreadID;
     }
-  } else {
-    // This is not an outlined function region - need to call __kmpc_int32
-    // kmpc_global_thread_num(ident_t *loc).
-    // Generate thread id value and cache this value for use across the
-    // function.
-    CGBuilderTy::InsertPointGuard IPG(CGF.Builder);
-    CGF.Builder.SetInsertPoint(CGF.AllocaInsertPt);
-    llvm::Value *Args[] = {EmitOpenMPUpdateLocation(CGF, Loc)};
-    ThreadID = CGF.EmitRuntimeCall(
-        CreateRuntimeFunction(OMPRTL__kmpc_global_thread_num), Args);
-    auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
-    Elem.second.ThreadID = ThreadID;
   }
+
+  // This is not an outlined function region - need to call __kmpc_int32
+  // kmpc_global_thread_num(ident_t *loc).
+  // Generate thread id value and cache this value for use across the
+  // function.
+  CGBuilderTy::InsertPointGuard IPG(CGF.Builder);
+  CGF.Builder.SetInsertPoint(CGF.AllocaInsertPt);
+  ThreadID =
+      CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_global_thread_num),
+                          emitUpdateLocation(CGF, Loc));
+  auto &Elem = OpenMPLocThreadIDMap.FindAndConstruct(CGF.CurFn);
+  Elem.second.ThreadID = ThreadID;
   return ThreadID;
 }
 
-void CGOpenMPRuntime::FunctionFinished(CodeGenFunction &CGF) {
+void CGOpenMPRuntime::functionFinished(CodeGenFunction &CGF) {
   assert(CGF.CurFn && "No function in current CodeGenFunction.");
   if (OpenMPLocThreadIDMap.count(CGF.CurFn))
     OpenMPLocThreadIDMap.erase(CGF.CurFn);
@@ -248,7 +444,7 @@ llvm::Type *CGOpenMPRuntime::getKmpc_MicroPointerTy() {
 }
 
 llvm::Constant *
-CGOpenMPRuntime::CreateRuntimeFunction(OpenMPRTLFunction Function) {
+CGOpenMPRuntime::createRuntimeFunction(OpenMPRTLFunction Function) {
   llvm::Constant *RTLFn = nullptr;
   switch (Function) {
   case OMPRTL__kmpc_fork_call: {
@@ -334,87 +530,6 @@ CGOpenMPRuntime::CreateRuntimeFunction(OpenMPRTLFunction Function) {
     RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name*/ "__kmpc_cancel_barrier");
     break;
   }
-  // Build __kmpc_for_static_init*(
-  //               ident_t *loc, kmp_int32 tid, kmp_int32 schedtype,
-  //               kmp_int32 *p_lastiter, kmp_int[32|64] *p_lower,
-  //               kmp_int[32|64] *p_upper, kmp_int[32|64] *p_stride,
-  //               kmp_int[32|64] incr, kmp_int[32|64] chunk);
-  case OMPRTL__kmpc_for_static_init_4: {
-    auto ITy = CGM.Int32Ty;
-    auto PtrTy = llvm::PointerType::getUnqual(ITy);
-    llvm::Type *TypeParams[] = {
-        getIdentTyPointerTy(),                     // loc
-        CGM.Int32Ty,                               // tid
-        CGM.Int32Ty,                               // schedtype
-        llvm::PointerType::getUnqual(CGM.Int32Ty), // p_lastiter
-        PtrTy,                                     // p_lower
-        PtrTy,                                     // p_upper
-        PtrTy,                                     // p_stride
-        ITy,                                       // incr
-        ITy                                        // chunk
-    };
-    llvm::FunctionType *FnTy =
-        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
-    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_for_static_init_4");
-    break;
-  }
-  case OMPRTL__kmpc_for_static_init_4u: {
-    auto ITy = CGM.Int32Ty;
-    auto PtrTy = llvm::PointerType::getUnqual(ITy);
-    llvm::Type *TypeParams[] = {
-        getIdentTyPointerTy(),                     // loc
-        CGM.Int32Ty,                               // tid
-        CGM.Int32Ty,                               // schedtype
-        llvm::PointerType::getUnqual(CGM.Int32Ty), // p_lastiter
-        PtrTy,                                     // p_lower
-        PtrTy,                                     // p_upper
-        PtrTy,                                     // p_stride
-        ITy,                                       // incr
-        ITy                                        // chunk
-    };
-    llvm::FunctionType *FnTy =
-        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
-    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_for_static_init_4u");
-    break;
-  }
-  case OMPRTL__kmpc_for_static_init_8: {
-    auto ITy = CGM.Int64Ty;
-    auto PtrTy = llvm::PointerType::getUnqual(ITy);
-    llvm::Type *TypeParams[] = {
-        getIdentTyPointerTy(),                     // loc
-        CGM.Int32Ty,                               // tid
-        CGM.Int32Ty,                               // schedtype
-        llvm::PointerType::getUnqual(CGM.Int32Ty), // p_lastiter
-        PtrTy,                                     // p_lower
-        PtrTy,                                     // p_upper
-        PtrTy,                                     // p_stride
-        ITy,                                       // incr
-        ITy                                        // chunk
-    };
-    llvm::FunctionType *FnTy =
-        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
-    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_for_static_init_8");
-    break;
-  }
-  case OMPRTL__kmpc_for_static_init_8u: {
-    auto ITy = CGM.Int64Ty;
-    auto PtrTy = llvm::PointerType::getUnqual(ITy);
-    llvm::Type *TypeParams[] = {
-        getIdentTyPointerTy(),                     // loc
-        CGM.Int32Ty,                               // tid
-        CGM.Int32Ty,                               // schedtype
-        llvm::PointerType::getUnqual(CGM.Int32Ty), // p_lastiter
-        PtrTy,                                     // p_lower
-        PtrTy,                                     // p_upper
-        PtrTy,                                     // p_stride
-        ITy,                                       // incr
-        ITy                                        // chunk
-    };
-    llvm::FunctionType *FnTy =
-        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
-    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_for_static_init_8u");
-    break;
-  }
   case OMPRTL__kmpc_for_static_fini: {
     // Build void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid);
     llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
@@ -452,10 +567,10 @@ CGOpenMPRuntime::CreateRuntimeFunction(OpenMPRTLFunction Function) {
     break;
   }
   case OMPRTL__kmpc_flush: {
-    // Build void __kmpc_flush(ident_t *loc, ...);
+    // Build void __kmpc_flush(ident_t *loc);
     llvm::Type *TypeParams[] = {getIdentTyPointerTy()};
     llvm::FunctionType *FnTy =
-        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ true);
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
     RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_flush");
     break;
   }
@@ -475,38 +590,291 @@ CGOpenMPRuntime::CreateRuntimeFunction(OpenMPRTLFunction Function) {
     RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_end_master");
     break;
   }
+  case OMPRTL__kmpc_omp_taskyield: {
+    // Build kmp_int32 __kmpc_omp_taskyield(ident_t *, kmp_int32 global_tid,
+    // int end_part);
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.IntTy};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_omp_taskyield");
+    break;
+  }
+  case OMPRTL__kmpc_single: {
+    // Build kmp_int32 __kmpc_single(ident_t *loc, kmp_int32 global_tid);
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_single");
+    break;
+  }
+  case OMPRTL__kmpc_end_single: {
+    // Build void __kmpc_end_single(ident_t *loc, kmp_int32 global_tid);
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_end_single");
+    break;
+  }
+  case OMPRTL__kmpc_omp_task_alloc: {
+    // Build kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid,
+    // kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds,
+    // kmp_routine_entry_t *task_entry);
+    assert(KmpRoutineEntryPtrTy != nullptr &&
+           "Type kmp_routine_entry_t must be created.");
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.Int32Ty,
+                                CGM.SizeTy, CGM.SizeTy, KmpRoutineEntryPtrTy};
+    // Return void * and then cast to particular kmp_task_t type.
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidPtrTy, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_omp_task_alloc");
+    break;
+  }
+  case OMPRTL__kmpc_omp_task: {
+    // Build kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t
+    // *new_task);
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
+                                CGM.VoidPtrTy};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_omp_task");
+    break;
+  }
+  case OMPRTL__kmpc_copyprivate: {
+    // Build void __kmpc_copyprivate(ident_t *loc, kmp_int32 global_tid,
+    // size_t cpy_size, void *cpy_data, void(*cpy_func)(void *, void *),
+    // kmp_int32 didit);
+    llvm::Type *CpyTypeParams[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
+    auto *CpyFnTy =
+        llvm::FunctionType::get(CGM.VoidTy, CpyTypeParams, /*isVarArg=*/false);
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty, CGM.SizeTy,
+                                CGM.VoidPtrTy, CpyFnTy->getPointerTo(),
+                                CGM.Int32Ty};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_copyprivate");
+    break;
+  }
+  case OMPRTL__kmpc_reduce: {
+    // Build kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid,
+    // kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void
+    // (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck);
+    llvm::Type *ReduceTypeParams[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
+    auto *ReduceFnTy = llvm::FunctionType::get(CGM.VoidTy, ReduceTypeParams,
+                                               /*isVarArg=*/false);
+    llvm::Type *TypeParams[] = {
+        getIdentTyPointerTy(), CGM.Int32Ty, CGM.Int32Ty, CGM.SizeTy,
+        CGM.VoidPtrTy, ReduceFnTy->getPointerTo(),
+        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_reduce");
+    break;
+  }
+  case OMPRTL__kmpc_reduce_nowait: {
+    // Build kmp_int32 __kmpc_reduce_nowait(ident_t *loc, kmp_int32
+    // global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data,
+    // void (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name
+    // *lck);
+    llvm::Type *ReduceTypeParams[] = {CGM.VoidPtrTy, CGM.VoidPtrTy};
+    auto *ReduceFnTy = llvm::FunctionType::get(CGM.VoidTy, ReduceTypeParams,
+                                               /*isVarArg=*/false);
+    llvm::Type *TypeParams[] = {
+        getIdentTyPointerTy(), CGM.Int32Ty, CGM.Int32Ty, CGM.SizeTy,
+        CGM.VoidPtrTy, ReduceFnTy->getPointerTo(),
+        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_reduce_nowait");
+    break;
+  }
+  case OMPRTL__kmpc_end_reduce: {
+    // Build void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid,
+    // kmp_critical_name *lck);
+    llvm::Type *TypeParams[] = {
+        getIdentTyPointerTy(), CGM.Int32Ty,
+        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_end_reduce");
+    break;
+  }
+  case OMPRTL__kmpc_end_reduce_nowait: {
+    // Build __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid,
+    // kmp_critical_name *lck);
+    llvm::Type *TypeParams[] = {
+        getIdentTyPointerTy(), CGM.Int32Ty,
+        llvm::PointerType::getUnqual(KmpCriticalNameTy)};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+    RTLFn =
+        CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_end_reduce_nowait");
+    break;
+  }
+  case OMPRTL__kmpc_omp_task_begin_if0: {
+    // Build void __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t
+    // *new_task);
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
+                                CGM.VoidPtrTy};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+    RTLFn =
+        CGM.CreateRuntimeFunction(FnTy, /*Name=*/"__kmpc_omp_task_begin_if0");
+    break;
+  }
+  case OMPRTL__kmpc_omp_task_complete_if0: {
+    // Build void __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t
+    // *new_task);
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty,
+                                CGM.VoidPtrTy};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy,
+                                      /*Name=*/"__kmpc_omp_task_complete_if0");
+    break;
+  }
+  case OMPRTL__kmpc_ordered: {
+    // Build void __kmpc_ordered(ident_t *loc, kmp_int32 global_tid);
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_ordered");
+    break;
+  }
+  case OMPRTL__kmpc_end_ordered: {
+    // Build void __kmpc_ordered(ident_t *loc, kmp_int32 global_tid);
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_end_ordered");
+    break;
+  }
+  case OMPRTL__kmpc_omp_taskwait: {
+    // Build kmp_int32 __kmpc_omp_taskwait(ident_t *loc, kmp_int32 global_tid);
+    llvm::Type *TypeParams[] = {getIdentTyPointerTy(), CGM.Int32Ty};
+    llvm::FunctionType *FnTy =
+        llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg=*/false);
+    RTLFn = CGM.CreateRuntimeFunction(FnTy, "__kmpc_omp_taskwait");
+    break;
+  }
   }
   return RTLFn;
 }
 
+llvm::Constant *CGOpenMPRuntime::createForStaticInitFunction(unsigned IVSize,
+                                                             bool IVSigned) {
+  assert((IVSize == 32 || IVSize == 64) &&
+         "IV size is not compatible with the omp runtime");
+  auto Name = IVSize == 32 ? (IVSigned ? "__kmpc_for_static_init_4"
+                                       : "__kmpc_for_static_init_4u")
+                           : (IVSigned ? "__kmpc_for_static_init_8"
+                                       : "__kmpc_for_static_init_8u");
+  auto ITy = IVSize == 32 ? CGM.Int32Ty : CGM.Int64Ty;
+  auto PtrTy = llvm::PointerType::getUnqual(ITy);
+  llvm::Type *TypeParams[] = {
+    getIdentTyPointerTy(),                     // loc
+    CGM.Int32Ty,                               // tid
+    CGM.Int32Ty,                               // schedtype
+    llvm::PointerType::getUnqual(CGM.Int32Ty), // p_lastiter
+    PtrTy,                                     // p_lower
+    PtrTy,                                     // p_upper
+    PtrTy,                                     // p_stride
+    ITy,                                       // incr
+    ITy                                        // chunk
+  };
+  llvm::FunctionType *FnTy =
+      llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
+  return CGM.CreateRuntimeFunction(FnTy, Name);
+}
+
+llvm::Constant *CGOpenMPRuntime::createDispatchInitFunction(unsigned IVSize,
+                                                            bool IVSigned) {
+  assert((IVSize == 32 || IVSize == 64) &&
+         "IV size is not compatible with the omp runtime");
+  auto Name =
+      IVSize == 32
+          ? (IVSigned ? "__kmpc_dispatch_init_4" : "__kmpc_dispatch_init_4u")
+          : (IVSigned ? "__kmpc_dispatch_init_8" : "__kmpc_dispatch_init_8u");
+  auto ITy = IVSize == 32 ? CGM.Int32Ty : CGM.Int64Ty;
+  llvm::Type *TypeParams[] = { getIdentTyPointerTy(), // loc
+                               CGM.Int32Ty,           // tid
+                               CGM.Int32Ty,           // schedtype
+                               ITy,                   // lower
+                               ITy,                   // upper
+                               ITy,                   // stride
+                               ITy                    // chunk
+  };
+  llvm::FunctionType *FnTy =
+      llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg*/ false);
+  return CGM.CreateRuntimeFunction(FnTy, Name);
+}
+
+llvm::Constant *CGOpenMPRuntime::createDispatchFiniFunction(unsigned IVSize,
+                                                            bool IVSigned) {
+  assert((IVSize == 32 || IVSize == 64) &&
+         "IV size is not compatible with the omp runtime");
+  auto Name =
+      IVSize == 32
+          ? (IVSigned ? "__kmpc_dispatch_fini_4" : "__kmpc_dispatch_fini_4u")
+          : (IVSigned ? "__kmpc_dispatch_fini_8" : "__kmpc_dispatch_fini_8u");
+  llvm::Type *TypeParams[] = {
+      getIdentTyPointerTy(), // loc
+      CGM.Int32Ty,           // tid
+  };
+  llvm::FunctionType *FnTy =
+      llvm::FunctionType::get(CGM.VoidTy, TypeParams, /*isVarArg=*/false);
+  return CGM.CreateRuntimeFunction(FnTy, Name);
+}
+
+llvm::Constant *CGOpenMPRuntime::createDispatchNextFunction(unsigned IVSize,
+                                                            bool IVSigned) {
+  assert((IVSize == 32 || IVSize == 64) &&
+         "IV size is not compatible with the omp runtime");
+  auto Name =
+      IVSize == 32
+          ? (IVSigned ? "__kmpc_dispatch_next_4" : "__kmpc_dispatch_next_4u")
+          : (IVSigned ? "__kmpc_dispatch_next_8" : "__kmpc_dispatch_next_8u");
+  auto ITy = IVSize == 32 ? CGM.Int32Ty : CGM.Int64Ty;
+  auto PtrTy = llvm::PointerType::getUnqual(ITy);
+  llvm::Type *TypeParams[] = {
+    getIdentTyPointerTy(),                     // loc
+    CGM.Int32Ty,                               // tid
+    llvm::PointerType::getUnqual(CGM.Int32Ty), // p_lastiter
+    PtrTy,                                     // p_lower
+    PtrTy,                                     // p_upper
+    PtrTy                                      // p_stride
+  };
+  llvm::FunctionType *FnTy =
+      llvm::FunctionType::get(CGM.Int32Ty, TypeParams, /*isVarArg*/ false);
+  return CGM.CreateRuntimeFunction(FnTy, Name);
+}
+
 llvm::Constant *
 CGOpenMPRuntime::getOrCreateThreadPrivateCache(const VarDecl *VD) {
   // Lookup the entry, lazily creating it if necessary.
-  return GetOrCreateInternalVariable(CGM.Int8PtrPtrTy,
+  return getOrCreateInternalVariable(CGM.Int8PtrPtrTy,
                                      Twine(CGM.getMangledName(VD)) + ".cache.");
 }
 
-llvm::Value *CGOpenMPRuntime::getOMPAddrOfThreadPrivate(CodeGenFunction &CGF,
-                                                        const VarDecl *VD,
-                                                        llvm::Value *VDAddr,
-                                                        SourceLocation Loc) {
+llvm::Value *CGOpenMPRuntime::getAddrOfThreadPrivate(CodeGenFunction &CGF,
+                                                     const VarDecl *VD,
+                                                     llvm::Value *VDAddr,
+                                                     SourceLocation Loc) {
   auto VarTy = VDAddr->getType()->getPointerElementType();
-  llvm::Value *Args[] = {EmitOpenMPUpdateLocation(CGF, Loc),
-                         GetOpenMPThreadID(CGF, Loc),
+  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
                          CGF.Builder.CreatePointerCast(VDAddr, CGM.Int8PtrTy),
                          CGM.getSize(CGM.GetTargetTypeStoreSize(VarTy)),
                          getOrCreateThreadPrivateCache(VD)};
   return CGF.EmitRuntimeCall(
-      CreateRuntimeFunction(OMPRTL__kmpc_threadprivate_cached), Args);
+      createRuntimeFunction(OMPRTL__kmpc_threadprivate_cached), Args);
 }
 
-void CGOpenMPRuntime::EmitOMPThreadPrivateVarInit(
+void CGOpenMPRuntime::emitThreadPrivateVarInit(
     CodeGenFunction &CGF, llvm::Value *VDAddr, llvm::Value *Ctor,
     llvm::Value *CopyCtor, llvm::Value *Dtor, SourceLocation Loc) {
   // Call kmp_int32 __kmpc_global_thread_num(&loc) to init OpenMP runtime
   // library.
-  auto OMPLoc = EmitOpenMPUpdateLocation(CGF, Loc);
-  CGF.EmitRuntimeCall(CreateRuntimeFunction(OMPRTL__kmpc_global_thread_num),
+  auto OMPLoc = emitUpdateLocation(CGF, Loc);
+  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_global_thread_num),
                       OMPLoc);
   // Call __kmpc_threadprivate_register(&loc, &var, ctor, cctor/*NULL*/, dtor)
   // to register constructor/destructor for variable.
@@ -514,10 +882,10 @@ void CGOpenMPRuntime::EmitOMPThreadPrivateVarInit(
                          CGF.Builder.CreatePointerCast(VDAddr, CGM.VoidPtrTy),
                          Ctor, CopyCtor, Dtor};
   CGF.EmitRuntimeCall(
-      CreateRuntimeFunction(OMPRTL__kmpc_threadprivate_register), Args);
+      createRuntimeFunction(OMPRTL__kmpc_threadprivate_register), Args);
 }
 
-llvm::Function *CGOpenMPRuntime::EmitOMPThreadPrivateVarDefinition(
+llvm::Function *CGOpenMPRuntime::emitThreadPrivateVarDefinition(
     const VarDecl *VD, llvm::Value *VDAddr, SourceLocation Loc,
     bool PerformInit, CodeGenFunction *CGF) {
   VD = VD->getDefinition(CGM.getContext());
@@ -620,54 +988,121 @@ llvm::Function *CGOpenMPRuntime::EmitOMPThreadPrivateVarDefinition(
       InitCGF.StartFunction(GlobalDecl(), CGM.getContext().VoidTy, InitFunction,
                             CGM.getTypes().arrangeNullaryFunction(), ArgList,
                             Loc);
-      EmitOMPThreadPrivateVarInit(InitCGF, VDAddr, Ctor, CopyCtor, Dtor, Loc);
+      emitThreadPrivateVarInit(InitCGF, VDAddr, Ctor, CopyCtor, Dtor, Loc);
       InitCGF.FinishFunction();
       return InitFunction;
     }
-    EmitOMPThreadPrivateVarInit(*CGF, VDAddr, Ctor, CopyCtor, Dtor, Loc);
+    emitThreadPrivateVarInit(*CGF, VDAddr, Ctor, CopyCtor, Dtor, Loc);
   }
   return nullptr;
 }
 
-void CGOpenMPRuntime::EmitOMPParallelCall(CodeGenFunction &CGF,
-                                          SourceLocation Loc,
-                                          llvm::Value *OutlinedFn,
-                                          llvm::Value *CapturedStruct) {
-  // Build call __kmpc_fork_call(loc, 1, microtask, captured_struct/*context*/)
-  llvm::Value *Args[] = {
-      EmitOpenMPUpdateLocation(CGF, Loc),
-      CGF.Builder.getInt32(1), // Number of arguments after 'microtask' argument
-      // (there is only one additional argument - 'context')
-      CGF.Builder.CreateBitCast(OutlinedFn, getKmpc_MicroPointerTy()),
-      CGF.EmitCastToVoidPtr(CapturedStruct)};
-  auto RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_fork_call);
-  CGF.EmitRuntimeCall(RTLFn, Args);
-}
-
-void CGOpenMPRuntime::EmitOMPSerialCall(CodeGenFunction &CGF,
-                                        SourceLocation Loc,
-                                        llvm::Value *OutlinedFn,
-                                        llvm::Value *CapturedStruct) {
-  auto ThreadID = GetOpenMPThreadID(CGF, Loc);
-  // Build calls:
-  // __kmpc_serialized_parallel(&Loc, GTid);
-  llvm::Value *SerArgs[] = {EmitOpenMPUpdateLocation(CGF, Loc), ThreadID};
-  auto RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_serialized_parallel);
-  CGF.EmitRuntimeCall(RTLFn, SerArgs);
-
-  // OutlinedFn(&GTid, &zero, CapturedStruct);
-  auto ThreadIDAddr = EmitThreadIDAddress(CGF, Loc);
-  auto Int32Ty =
-      CGF.getContext().getIntTypeForBitwidth(/*DestWidth*/ 32, /*Signed*/ true);
-  auto ZeroAddr = CGF.CreateMemTemp(Int32Ty, /*Name*/ ".zero.addr");
-  CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
-  llvm::Value *OutlinedFnArgs[] = {ThreadIDAddr, ZeroAddr, CapturedStruct};
-  CGF.EmitCallOrInvoke(OutlinedFn, OutlinedFnArgs);
+/// \brief Emits code for OpenMP 'if' clause using specified \a CodeGen
+/// function. Here is the logic:
+/// if (Cond) {
+///   ThenGen();
+/// } else {
+///   ElseGen();
+/// }
+static void emitOMPIfClause(CodeGenFunction &CGF, const Expr *Cond,
+                            const RegionCodeGenTy &ThenGen,
+                            const RegionCodeGenTy &ElseGen) {
+  CodeGenFunction::LexicalScope ConditionScope(CGF, Cond->getSourceRange());
+
+  // If the condition constant folds and can be elided, try to avoid emitting
+  // the condition and the dead arm of the if/else.
+  bool CondConstant;
+  if (CGF.ConstantFoldsToSimpleInteger(Cond, CondConstant)) {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    if (CondConstant) {
+      ThenGen(CGF);
+    } else {
+      ElseGen(CGF);
+    }
+    return;
+  }
+
+  // Otherwise, the condition did not fold, or we couldn't elide it.  Just
+  // emit the conditional branch.
+  auto ThenBlock = CGF.createBasicBlock("omp_if.then");
+  auto ElseBlock = CGF.createBasicBlock("omp_if.else");
+  auto ContBlock = CGF.createBasicBlock("omp_if.end");
+  CGF.EmitBranchOnBoolExpr(Cond, ThenBlock, ElseBlock, /*TrueCount=*/0);
 
-  // __kmpc_end_serialized_parallel(&Loc, GTid);
-  llvm::Value *EndSerArgs[] = {EmitOpenMPUpdateLocation(CGF, Loc), ThreadID};
-  RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_end_serialized_parallel);
-  CGF.EmitRuntimeCall(RTLFn, EndSerArgs);
+  // Emit the 'then' code.
+  CGF.EmitBlock(ThenBlock);
+  {
+    CodeGenFunction::RunCleanupsScope ThenScope(CGF);
+    ThenGen(CGF);
+  }
+  CGF.EmitBranch(ContBlock);
+  // Emit the 'else' code if present.
+  {
+    // There is no need to emit line number for unconditional branch.
+    auto NL = ApplyDebugLocation::CreateEmpty(CGF);
+    CGF.EmitBlock(ElseBlock);
+  }
+  {
+    CodeGenFunction::RunCleanupsScope ThenScope(CGF);
+    ElseGen(CGF);
+  }
+  {
+    // There is no need to emit line number for unconditional branch.
+    auto NL = ApplyDebugLocation::CreateEmpty(CGF);
+    CGF.EmitBranch(ContBlock);
+  }
+  // Emit the continuation block for code after the if.
+  CGF.EmitBlock(ContBlock, /*IsFinished=*/true);
+}
+
+void CGOpenMPRuntime::emitParallelCall(CodeGenFunction &CGF, SourceLocation Loc,
+                                       llvm::Value *OutlinedFn,
+                                       llvm::Value *CapturedStruct,
+                                       const Expr *IfCond) {
+  auto *RTLoc = emitUpdateLocation(CGF, Loc);
+  auto &&ThenGen =
+      [this, OutlinedFn, CapturedStruct, RTLoc](CodeGenFunction &CGF) {
+        // Build call __kmpc_fork_call(loc, 1, microtask,
+        // captured_struct/*context*/)
+        llvm::Value *Args[] = {
+            RTLoc,
+            CGF.Builder.getInt32(
+                1), // Number of arguments after 'microtask' argument
+            // (there is only one additional argument - 'context')
+            CGF.Builder.CreateBitCast(OutlinedFn, getKmpc_MicroPointerTy()),
+            CGF.EmitCastToVoidPtr(CapturedStruct)};
+        auto RTLFn = createRuntimeFunction(OMPRTL__kmpc_fork_call);
+        CGF.EmitRuntimeCall(RTLFn, Args);
+      };
+  auto &&ElseGen = [this, OutlinedFn, CapturedStruct, RTLoc, Loc](
+      CodeGenFunction &CGF) {
+    auto ThreadID = getThreadID(CGF, Loc);
+    // Build calls:
+    // __kmpc_serialized_parallel(&Loc, GTid);
+    llvm::Value *Args[] = {RTLoc, ThreadID};
+    CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_serialized_parallel),
+                        Args);
+
+    // OutlinedFn(&GTid, &zero, CapturedStruct);
+    auto ThreadIDAddr = emitThreadIDAddress(CGF, Loc);
+    auto Int32Ty = CGF.getContext().getIntTypeForBitwidth(/*DestWidth*/ 32,
+                                                          /*Signed*/ true);
+    auto ZeroAddr = CGF.CreateMemTemp(Int32Ty, /*Name*/ ".zero.addr");
+    CGF.InitTempAlloca(ZeroAddr, CGF.Builder.getInt32(/*C*/ 0));
+    llvm::Value *OutlinedFnArgs[] = {ThreadIDAddr, ZeroAddr, CapturedStruct};
+    CGF.EmitCallOrInvoke(OutlinedFn, OutlinedFnArgs);
+
+    // __kmpc_end_serialized_parallel(&Loc, GTid);
+    llvm::Value *EndArgs[] = {emitUpdateLocation(CGF, Loc), ThreadID};
+    CGF.EmitRuntimeCall(
+        createRuntimeFunction(OMPRTL__kmpc_end_serialized_parallel), EndArgs);
+  };
+  if (IfCond) {
+    emitOMPIfClause(CGF, IfCond, ThenGen, ElseGen);
+  } else {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    ThenGen(CGF);
+  }
 }
 
 // If we're inside an (outlined) parallel region, use the region info's
@@ -676,13 +1111,14 @@ void CGOpenMPRuntime::EmitOMPSerialCall(CodeGenFunction &CGF,
 // regular serial code region, get thread ID by calling kmp_int32
 // kmpc_global_thread_num(ident_t *loc), stash this thread ID in a temporary and
 // return the address of that temp.
-llvm::Value *CGOpenMPRuntime::EmitThreadIDAddress(CodeGenFunction &CGF,
+llvm::Value *CGOpenMPRuntime::emitThreadIDAddress(CodeGenFunction &CGF,
                                                   SourceLocation Loc) {
   if (auto OMPRegionInfo =
           dyn_cast_or_null<CGOpenMPRegionInfo>(CGF.CapturedStmtInfo))
-    return CGF.EmitLoadOfLValue(OMPRegionInfo->getThreadIDVariableLValue(CGF),
-                                SourceLocation()).getScalarVal();
-  auto ThreadID = GetOpenMPThreadID(CGF, Loc);
+    if (OMPRegionInfo->getThreadIDVariable())
+      return OMPRegionInfo->getThreadIDVariableLValue(CGF).getAddress();
+
+  auto ThreadID = getThreadID(CGF, Loc);
   auto Int32Ty =
       CGF.getContext().getIntTypeForBitwidth(/*DestWidth*/ 32, /*Signed*/ true);
   auto ThreadIDTemp = CGF.CreateMemTemp(Int32Ty, /*Name*/ ".threadid_temp.");
@@ -693,7 +1129,7 @@ llvm::Value *CGOpenMPRuntime::EmitThreadIDAddress(CodeGenFunction &CGF,
 }
 
 llvm::Constant *
-CGOpenMPRuntime::GetOrCreateInternalVariable(llvm::Type *Ty,
+CGOpenMPRuntime::getOrCreateInternalVariable(llvm::Type *Ty,
                                              const llvm::Twine &Name) {
   SmallString<256> Buffer;
   llvm::raw_svector_ostream Out(Buffer);
@@ -712,31 +1148,51 @@ CGOpenMPRuntime::GetOrCreateInternalVariable(llvm::Type *Ty,
              Elem.first());
 }
 
-llvm::Value *CGOpenMPRuntime::GetCriticalRegionLock(StringRef CriticalName) {
+llvm::Value *CGOpenMPRuntime::getCriticalRegionLock(StringRef CriticalName) {
   llvm::Twine Name(".gomp_critical_user_", CriticalName);
-  return GetOrCreateInternalVariable(KmpCriticalNameTy, Name.concat(".var"));
+  return getOrCreateInternalVariable(KmpCriticalNameTy, Name.concat(".var"));
 }
 
-void CGOpenMPRuntime::EmitOMPCriticalRegion(
-    CodeGenFunction &CGF, StringRef CriticalName,
-    const std::function<void()> &CriticalOpGen, SourceLocation Loc) {
-  auto RegionLock = GetCriticalRegionLock(CriticalName);
+namespace {
+template <size_t N> class CallEndCleanup : public EHScopeStack::Cleanup {
+  llvm::Value *Callee;
+  llvm::Value *Args[N];
+
+public:
+  CallEndCleanup(llvm::Value *Callee, ArrayRef<llvm::Value *> CleanupArgs)
+      : Callee(Callee) {
+    assert(CleanupArgs.size() == N);
+    std::copy(CleanupArgs.begin(), CleanupArgs.end(), std::begin(Args));
+  }
+  void Emit(CodeGenFunction &CGF, Flags /*flags*/) override {
+    CGF.EmitRuntimeCall(Callee, Args);
+  }
+};
+} // namespace
+
+void CGOpenMPRuntime::emitCriticalRegion(CodeGenFunction &CGF,
+                                         StringRef CriticalName,
+                                         const RegionCodeGenTy &CriticalOpGen,
+                                         SourceLocation Loc) {
   // __kmpc_critical(ident_t *, gtid, Lock);
   // CriticalOpGen();
   // __kmpc_end_critical(ident_t *, gtid, Lock);
   // Prepare arguments and build a call to __kmpc_critical
-  llvm::Value *Args[] = {EmitOpenMPUpdateLocation(CGF, Loc),
-                         GetOpenMPThreadID(CGF, Loc), RegionLock};
-  auto RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_critical);
-  CGF.EmitRuntimeCall(RTLFn, Args);
-  CriticalOpGen();
-  // Build a call to __kmpc_end_critical
-  RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_end_critical);
-  CGF.EmitRuntimeCall(RTLFn, Args);
-}
-
-static void EmitOMPIfStmt(CodeGenFunction &CGF, llvm::Value *IfCond,
-                          const std::function<void()> &BodyOpGen) {
+  {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
+                           getCriticalRegionLock(CriticalName)};
+    CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_critical), Args);
+    // Build a call to __kmpc_end_critical
+    CGF.EHStack.pushCleanup<CallEndCleanup<std::extent<decltype(Args)>::value>>(
+        NormalAndEHCleanup, createRuntimeFunction(OMPRTL__kmpc_end_critical),
+        llvm::makeArrayRef(Args));
+    emitInlinedDirective(CGF, CriticalOpGen);
+  }
+}
+
+static void emitIfStmt(CodeGenFunction &CGF, llvm::Value *IfCond,
+                       const RegionCodeGenTy &BodyOpGen) {
   llvm::Value *CallBool = CGF.EmitScalarConversion(
       IfCond,
       CGF.getContext().getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/true),
@@ -747,61 +1203,234 @@ static void EmitOMPIfStmt(CodeGenFunction &CGF, llvm::Value *IfCond,
   // Generate the branch (If-stmt)
   CGF.Builder.CreateCondBr(CallBool, ThenBlock, ContBlock);
   CGF.EmitBlock(ThenBlock);
-  BodyOpGen();
+  CGF.CGM.getOpenMPRuntime().emitInlinedDirective(CGF, BodyOpGen);
   // Emit the rest of bblocks/branches
   CGF.EmitBranch(ContBlock);
   CGF.EmitBlock(ContBlock, true);
 }
 
-void CGOpenMPRuntime::EmitOMPMasterRegion(
-    CodeGenFunction &CGF, const std::function<void()> &MasterOpGen,
-    SourceLocation Loc) {
+void CGOpenMPRuntime::emitMasterRegion(CodeGenFunction &CGF,
+                                       const RegionCodeGenTy &MasterOpGen,
+                                       SourceLocation Loc) {
   // if(__kmpc_master(ident_t *, gtid)) {
   //   MasterOpGen();
   //   __kmpc_end_master(ident_t *, gtid);
   // }
   // Prepare arguments and build a call to __kmpc_master
-  llvm::Value *Args[] = {EmitOpenMPUpdateLocation(CGF, Loc),
-                         GetOpenMPThreadID(CGF, Loc)};
-  auto RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_master);
-  auto *IsMaster = CGF.EmitRuntimeCall(RTLFn, Args);
-  EmitOMPIfStmt(CGF, IsMaster, [&]() -> void {
-    MasterOpGen();
-    // Build a call to __kmpc_end_master.
-    // OpenMP [1.2.2 OpenMP Language Terminology]
-    // For C/C++, an executable statement, possibly compound, with a single
-    // entry at the top and a single exit at the bottom, or an OpenMP construct.
-    // * Access to the structured block must not be the result of a branch.
-    // * The point of exit cannot be a branch out of the structured block.
-    // * The point of entry must not be a call to setjmp().
-    // * longjmp() and throw() must not violate the entry/exit criteria.
-    // * An expression statement, iteration statement, selection statement, or
-    // try block is considered to be a structured block if the corresponding
-    // compound statement obtained by enclosing it in { and } would be a
-    // structured block.
-    // It is analyzed in Sema, so we can just call __kmpc_end_master() on
-    // fallthrough rather than pushing a normal cleanup for it.
-    RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_end_master);
-    CGF.EmitRuntimeCall(RTLFn, Args);
+  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
+  auto *IsMaster =
+      CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_master), Args);
+  typedef CallEndCleanup<std::extent<decltype(Args)>::value>
+      MasterCallEndCleanup;
+  emitIfStmt(CGF, IsMaster, [&](CodeGenFunction &CGF) -> void {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    CGF.EHStack.pushCleanup<MasterCallEndCleanup>(
+        NormalAndEHCleanup, createRuntimeFunction(OMPRTL__kmpc_end_master),
+        llvm::makeArrayRef(Args));
+    MasterOpGen(CGF);
   });
 }
 
-void CGOpenMPRuntime::EmitOMPBarrierCall(CodeGenFunction &CGF,
-                                         SourceLocation Loc, bool IsExplicit) {
+void CGOpenMPRuntime::emitTaskyieldCall(CodeGenFunction &CGF,
+                                        SourceLocation Loc) {
+  // Build call __kmpc_omp_taskyield(loc, thread_id, 0);
+  llvm::Value *Args[] = {
+      emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
+      llvm::ConstantInt::get(CGM.IntTy, /*V=*/0, /*isSigned=*/true)};
+  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_omp_taskyield), Args);
+}
+
+static llvm::Value *emitCopyprivateCopyFunction(
+    CodeGenModule &CGM, llvm::Type *ArgsType,
+    ArrayRef<const Expr *> CopyprivateVars, ArrayRef<const Expr *> DestExprs,
+    ArrayRef<const Expr *> SrcExprs, ArrayRef<const Expr *> AssignmentOps) {
+  auto &C = CGM.getContext();
+  // void copy_func(void *LHSArg, void *RHSArg);
+  FunctionArgList Args;
+  ImplicitParamDecl LHSArg(C, /*DC=*/nullptr, SourceLocation(), /*Id=*/nullptr,
+                           C.VoidPtrTy);
+  ImplicitParamDecl RHSArg(C, /*DC=*/nullptr, SourceLocation(), /*Id=*/nullptr,
+                           C.VoidPtrTy);
+  Args.push_back(&LHSArg);
+  Args.push_back(&RHSArg);
+  FunctionType::ExtInfo EI;
+  auto &CGFI = CGM.getTypes().arrangeFreeFunctionDeclaration(
+      C.VoidTy, Args, EI, /*isVariadic=*/false);
+  auto *Fn = llvm::Function::Create(
+      CGM.getTypes().GetFunctionType(CGFI), llvm::GlobalValue::InternalLinkage,
+      ".omp.copyprivate.copy_func", &CGM.getModule());
+  CGM.SetLLVMFunctionAttributes(/*D=*/nullptr, CGFI, Fn);
+  CodeGenFunction CGF(CGM);
+  CGF.StartFunction(GlobalDecl(), C.VoidTy, Fn, CGFI, Args);
+  // Dest = (void*[n])(LHSArg);
+  // Src = (void*[n])(RHSArg);
+  auto *LHS = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+      CGF.Builder.CreateAlignedLoad(CGF.GetAddrOfLocalVar(&LHSArg),
+                                    CGF.PointerAlignInBytes),
+      ArgsType);
+  auto *RHS = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+      CGF.Builder.CreateAlignedLoad(CGF.GetAddrOfLocalVar(&RHSArg),
+                                    CGF.PointerAlignInBytes),
+      ArgsType);
+  // *(Type0*)Dst[0] = *(Type0*)Src[0];
+  // *(Type1*)Dst[1] = *(Type1*)Src[1];
+  // ...
+  // *(Typen*)Dst[n] = *(Typen*)Src[n];
+  for (unsigned I = 0, E = AssignmentOps.size(); I < E; ++I) {
+    auto *DestAddr = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+        CGF.Builder.CreateAlignedLoad(
+            CGF.Builder.CreateStructGEP(nullptr, LHS, I),
+            CGM.PointerAlignInBytes),
+        CGF.ConvertTypeForMem(C.getPointerType(SrcExprs[I]->getType())));
+    auto *SrcAddr = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+        CGF.Builder.CreateAlignedLoad(
+            CGF.Builder.CreateStructGEP(nullptr, RHS, I),
+            CGM.PointerAlignInBytes),
+        CGF.ConvertTypeForMem(C.getPointerType(SrcExprs[I]->getType())));
+    auto *VD = cast<DeclRefExpr>(CopyprivateVars[I])->getDecl();
+    QualType Type = VD->getType();
+    CGF.EmitOMPCopy(CGF, Type, DestAddr, SrcAddr,
+                    cast<VarDecl>(cast<DeclRefExpr>(DestExprs[I])->getDecl()),
+                    cast<VarDecl>(cast<DeclRefExpr>(SrcExprs[I])->getDecl()),
+                    AssignmentOps[I]);
+  }
+  CGF.FinishFunction();
+  return Fn;
+}
+
+void CGOpenMPRuntime::emitSingleRegion(CodeGenFunction &CGF,
+                                       const RegionCodeGenTy &SingleOpGen,
+                                       SourceLocation Loc,
+                                       ArrayRef<const Expr *> CopyprivateVars,
+                                       ArrayRef<const Expr *> SrcExprs,
+                                       ArrayRef<const Expr *> DstExprs,
+                                       ArrayRef<const Expr *> AssignmentOps) {
+  assert(CopyprivateVars.size() == SrcExprs.size() &&
+         CopyprivateVars.size() == DstExprs.size() &&
+         CopyprivateVars.size() == AssignmentOps.size());
+  auto &C = CGM.getContext();
+  // int32 did_it = 0;
+  // if(__kmpc_single(ident_t *, gtid)) {
+  //   SingleOpGen();
+  //   __kmpc_end_single(ident_t *, gtid);
+  //   did_it = 1;
+  // }
+  // call __kmpc_copyprivate(ident_t *, gtid, <buf_size>, <copyprivate list>,
+  // <copy_func>, did_it);
+
+  llvm::AllocaInst *DidIt = nullptr;
+  if (!CopyprivateVars.empty()) {
+    // int32 did_it = 0;
+    auto KmpInt32Ty = C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1);
+    DidIt = CGF.CreateMemTemp(KmpInt32Ty, ".omp.copyprivate.did_it");
+    CGF.Builder.CreateAlignedStore(CGF.Builder.getInt32(0), DidIt,
+                                   DidIt->getAlignment());
+  }
+  // Prepare arguments and build a call to __kmpc_single
+  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
+  auto *IsSingle =
+      CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_single), Args);
+  typedef CallEndCleanup<std::extent<decltype(Args)>::value>
+      SingleCallEndCleanup;
+  emitIfStmt(CGF, IsSingle, [&](CodeGenFunction &CGF) -> void {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    CGF.EHStack.pushCleanup<SingleCallEndCleanup>(
+        NormalAndEHCleanup, createRuntimeFunction(OMPRTL__kmpc_end_single),
+        llvm::makeArrayRef(Args));
+    SingleOpGen(CGF);
+    if (DidIt) {
+      // did_it = 1;
+      CGF.Builder.CreateAlignedStore(CGF.Builder.getInt32(1), DidIt,
+                                     DidIt->getAlignment());
+    }
+  });
+  // call __kmpc_copyprivate(ident_t *, gtid, <buf_size>, <copyprivate list>,
+  // <copy_func>, did_it);
+  if (DidIt) {
+    llvm::APInt ArraySize(/*unsigned int numBits=*/32, CopyprivateVars.size());
+    auto CopyprivateArrayTy =
+        C.getConstantArrayType(C.VoidPtrTy, ArraySize, ArrayType::Normal,
+                               /*IndexTypeQuals=*/0);
+    // Create a list of all private variables for copyprivate.
+    auto *CopyprivateList =
+        CGF.CreateMemTemp(CopyprivateArrayTy, ".omp.copyprivate.cpr_list");
+    for (unsigned I = 0, E = CopyprivateVars.size(); I < E; ++I) {
+      auto *Elem = CGF.Builder.CreateStructGEP(
+          CopyprivateList->getAllocatedType(), CopyprivateList, I);
+      CGF.Builder.CreateAlignedStore(
+          CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+              CGF.EmitLValue(CopyprivateVars[I]).getAddress(), CGF.VoidPtrTy),
+          Elem, CGM.PointerAlignInBytes);
+    }
+    // Build function that copies private values from single region to all other
+    // threads in the corresponding parallel region.
+    auto *CpyFn = emitCopyprivateCopyFunction(
+        CGM, CGF.ConvertTypeForMem(CopyprivateArrayTy)->getPointerTo(),
+        CopyprivateVars, SrcExprs, DstExprs, AssignmentOps);
+    auto *BufSize = llvm::ConstantInt::get(
+        CGM.SizeTy, C.getTypeSizeInChars(CopyprivateArrayTy).getQuantity());
+    auto *CL = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(CopyprivateList,
+                                                               CGF.VoidPtrTy);
+    auto *DidItVal =
+        CGF.Builder.CreateAlignedLoad(DidIt, CGF.PointerAlignInBytes);
+    llvm::Value *Args[] = {
+        emitUpdateLocation(CGF, Loc), // ident_t *<loc>
+        getThreadID(CGF, Loc),        // i32 <gtid>
+        BufSize,                      // size_t <buf_size>
+        CL,                           // void *<copyprivate list>
+        CpyFn,                        // void (*) (void *, void *) <copy_func>
+        DidItVal                      // i32 did_it
+    };
+    CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_copyprivate), Args);
+  }
+}
+
+void CGOpenMPRuntime::emitOrderedRegion(CodeGenFunction &CGF,
+                                        const RegionCodeGenTy &OrderedOpGen,
+                                        SourceLocation Loc) {
+  // __kmpc_ordered(ident_t *, gtid);
+  // OrderedOpGen();
+  // __kmpc_end_ordered(ident_t *, gtid);
+  // Prepare arguments and build a call to __kmpc_ordered
+  {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
+    CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_ordered), Args);
+    // Build a call to __kmpc_end_ordered
+    CGF.EHStack.pushCleanup<CallEndCleanup<std::extent<decltype(Args)>::value>>(
+        NormalAndEHCleanup, createRuntimeFunction(OMPRTL__kmpc_end_ordered),
+        llvm::makeArrayRef(Args));
+    emitInlinedDirective(CGF, OrderedOpGen);
+  }
+}
+
+void CGOpenMPRuntime::emitBarrierCall(CodeGenFunction &CGF, SourceLocation Loc,
+                                      OpenMPDirectiveKind Kind) {
   // Build call __kmpc_cancel_barrier(loc, thread_id);
-  auto Flags = static_cast<OpenMPLocationFlags>(
-      OMP_IDENT_KMPC |
-      (IsExplicit ? OMP_IDENT_BARRIER_EXPL : OMP_IDENT_BARRIER_IMPL));
+  OpenMPLocationFlags Flags = OMP_IDENT_KMPC;
+  if (Kind == OMPD_for) {
+    Flags =
+        static_cast<OpenMPLocationFlags>(Flags | OMP_IDENT_BARRIER_IMPL_FOR);
+  } else if (Kind == OMPD_sections) {
+    Flags = static_cast<OpenMPLocationFlags>(Flags |
+                                             OMP_IDENT_BARRIER_IMPL_SECTIONS);
+  } else if (Kind == OMPD_single) {
+    Flags =
+        static_cast<OpenMPLocationFlags>(Flags | OMP_IDENT_BARRIER_IMPL_SINGLE);
+  } else if (Kind == OMPD_barrier) {
+    Flags = static_cast<OpenMPLocationFlags>(Flags | OMP_IDENT_BARRIER_EXPL);
+  } else {
+    Flags = static_cast<OpenMPLocationFlags>(Flags | OMP_IDENT_BARRIER_IMPL);
+  }
   // Build call __kmpc_cancel_barrier(loc, thread_id);
   // Replace __kmpc_barrier() function by __kmpc_cancel_barrier() because this
   // one provides the same functionality and adds initial support for
   // cancellation constructs introduced in OpenMP 4.0. __kmpc_cancel_barrier()
   // is provided default by the runtime library so it safe to make such
   // replacement.
-  llvm::Value *Args[] = {EmitOpenMPUpdateLocation(CGF, Loc, Flags),
-                         GetOpenMPThreadID(CGF, Loc)};
-  auto RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_cancel_barrier);
-  CGF.EmitRuntimeCall(RTLFn, Args);
+  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc, Flags),
+                         getThreadID(CGF, Loc)};
+  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_cancel_barrier), Args);
 }
 
 /// \brief Schedule types for 'omp for' loops (these enumerators are taken from
@@ -817,106 +1446,1017 @@ enum OpenMPSchedType {
   OMP_sch_auto = 38,
   /// \brief Lower bound for 'ordered' versions.
   OMP_ord_lower = 64,
-  /// \brief Lower bound for 'nomerge' versions.
-  OMP_nm_lower = 160,
+  OMP_ord_static_chunked = 65,
+  OMP_ord_static = 66,
+  OMP_ord_dynamic_chunked = 67,
+  OMP_ord_guided_chunked = 68,
+  OMP_ord_runtime = 69,
+  OMP_ord_auto = 70,
+  OMP_sch_default = OMP_sch_static,
 };
 
 /// \brief Map the OpenMP loop schedule to the runtime enumeration.
 static OpenMPSchedType getRuntimeSchedule(OpenMPScheduleClauseKind ScheduleKind,
-                                          bool Chunked) {
+                                          bool Chunked, bool Ordered) {
   switch (ScheduleKind) {
   case OMPC_SCHEDULE_static:
-    return Chunked ? OMP_sch_static_chunked : OMP_sch_static;
+    return Chunked ? (Ordered ? OMP_ord_static_chunked : OMP_sch_static_chunked)
+                   : (Ordered ? OMP_ord_static : OMP_sch_static);
   case OMPC_SCHEDULE_dynamic:
-    return OMP_sch_dynamic_chunked;
+    return Ordered ? OMP_ord_dynamic_chunked : OMP_sch_dynamic_chunked;
   case OMPC_SCHEDULE_guided:
-    return OMP_sch_guided_chunked;
-  case OMPC_SCHEDULE_auto:
-    return OMP_sch_auto;
+    return Ordered ? OMP_ord_guided_chunked : OMP_sch_guided_chunked;
   case OMPC_SCHEDULE_runtime:
-    return OMP_sch_runtime;
+    return Ordered ? OMP_ord_runtime : OMP_sch_runtime;
+  case OMPC_SCHEDULE_auto:
+    return Ordered ? OMP_ord_auto : OMP_sch_auto;
   case OMPC_SCHEDULE_unknown:
     assert(!Chunked && "chunk was specified but schedule kind not known");
-    return OMP_sch_static;
+    return Ordered ? OMP_ord_static : OMP_sch_static;
   }
   llvm_unreachable("Unexpected runtime schedule");
 }
 
 bool CGOpenMPRuntime::isStaticNonchunked(OpenMPScheduleClauseKind ScheduleKind,
                                          bool Chunked) const {
-  auto Schedule = getRuntimeSchedule(ScheduleKind, Chunked);
+  auto Schedule = getRuntimeSchedule(ScheduleKind, Chunked, /*Ordered=*/false);
   return Schedule == OMP_sch_static;
 }
 
-void CGOpenMPRuntime::EmitOMPForInit(CodeGenFunction &CGF, SourceLocation Loc,
-                                     OpenMPScheduleClauseKind ScheduleKind,
-                                     unsigned IVSize, bool IVSigned,
-                                     llvm::Value *IL, llvm::Value *LB,
-                                     llvm::Value *UB, llvm::Value *ST,
-                                     llvm::Value *Chunk) {
-  OpenMPSchedType Schedule = getRuntimeSchedule(ScheduleKind, Chunk != nullptr);
-  // Call __kmpc_for_static_init(
-  //          ident_t *loc, kmp_int32 tid, kmp_int32 schedtype,
-  //          kmp_int32 *p_lastiter, kmp_int[32|64] *p_lower,
-  //          kmp_int[32|64] *p_upper, kmp_int[32|64] *p_stride,
-  //          kmp_int[32|64] incr, kmp_int[32|64] chunk);
-  // TODO: Implement dynamic schedule.
-
-  // If the Chunk was not specified in the clause - use default value 1.
-  if (Chunk == nullptr)
-    Chunk = CGF.Builder.getIntN(IVSize, /*C*/ 1);
+bool CGOpenMPRuntime::isDynamic(OpenMPScheduleClauseKind ScheduleKind) const {
+  auto Schedule =
+      getRuntimeSchedule(ScheduleKind, /*Chunked=*/false, /*Ordered=*/false);
+  assert(Schedule != OMP_sch_static_chunked && "cannot be chunked here");
+  return Schedule != OMP_sch_static;
+}
+
+void CGOpenMPRuntime::emitForInit(CodeGenFunction &CGF, SourceLocation Loc,
+                                  OpenMPScheduleClauseKind ScheduleKind,
+                                  unsigned IVSize, bool IVSigned, bool Ordered,
+                                  llvm::Value *IL, llvm::Value *LB,
+                                  llvm::Value *UB, llvm::Value *ST,
+                                  llvm::Value *Chunk) {
+  OpenMPSchedType Schedule =
+      getRuntimeSchedule(ScheduleKind, Chunk != nullptr, Ordered);
+  if (Ordered ||
+      (Schedule != OMP_sch_static && Schedule != OMP_sch_static_chunked &&
+       Schedule != OMP_ord_static && Schedule != OMP_ord_static_chunked)) {
+    // Call __kmpc_dispatch_init(
+    //          ident_t *loc, kmp_int32 tid, kmp_int32 schedule,
+    //          kmp_int[32|64] lower, kmp_int[32|64] upper,
+    //          kmp_int[32|64] stride, kmp_int[32|64] chunk);
+
+    // If the Chunk was not specified in the clause - use default value 1.
+    if (Chunk == nullptr)
+      Chunk = CGF.Builder.getIntN(IVSize, 1);
+    llvm::Value *Args[] = { emitUpdateLocation(CGF, Loc, OMP_IDENT_KMPC),
+                            getThreadID(CGF, Loc),
+                            CGF.Builder.getInt32(Schedule), // Schedule type
+                            CGF.Builder.getIntN(IVSize, 0), // Lower
+                            UB,                             // Upper
+                            CGF.Builder.getIntN(IVSize, 1), // Stride
+                            Chunk                           // Chunk
+    };
+    CGF.EmitRuntimeCall(createDispatchInitFunction(IVSize, IVSigned), Args);
+  } else {
+    // Call __kmpc_for_static_init(
+    //          ident_t *loc, kmp_int32 tid, kmp_int32 schedtype,
+    //          kmp_int32 *p_lastiter, kmp_int[32|64] *p_lower,
+    //          kmp_int[32|64] *p_upper, kmp_int[32|64] *p_stride,
+    //          kmp_int[32|64] incr, kmp_int[32|64] chunk);
+    if (Chunk == nullptr) {
+      assert((Schedule == OMP_sch_static || Schedule == OMP_ord_static) &&
+             "expected static non-chunked schedule");
+      // If the Chunk was not specified in the clause - use default value 1.
+      Chunk = CGF.Builder.getIntN(IVSize, 1);
+    } else
+      assert((Schedule == OMP_sch_static_chunked ||
+              Schedule == OMP_ord_static_chunked) &&
+             "expected static chunked schedule");
+    llvm::Value *Args[] = { emitUpdateLocation(CGF, Loc, OMP_IDENT_KMPC),
+                            getThreadID(CGF, Loc),
+                            CGF.Builder.getInt32(Schedule), // Schedule type
+                            IL,                             // &isLastIter
+                            LB,                             // &LB
+                            UB,                             // &UB
+                            ST,                             // &Stride
+                            CGF.Builder.getIntN(IVSize, 1), // Incr
+                            Chunk                           // Chunk
+    };
+    CGF.EmitRuntimeCall(createForStaticInitFunction(IVSize, IVSigned), Args);
+  }
+}
 
+void CGOpenMPRuntime::emitForStaticFinish(CodeGenFunction &CGF,
+                                          SourceLocation Loc) {
+  // Call __kmpc_for_static_fini(ident_t *loc, kmp_int32 tid);
+  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc, OMP_IDENT_KMPC),
+                         getThreadID(CGF, Loc)};
+  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_for_static_fini),
+                      Args);
+}
+
+void CGOpenMPRuntime::emitForOrderedIterationEnd(CodeGenFunction &CGF,
+                                                 SourceLocation Loc,
+                                                 unsigned IVSize,
+                                                 bool IVSigned) {
+  // Call __kmpc_for_dynamic_fini_(4|8)[u](ident_t *loc, kmp_int32 tid);
+  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc, OMP_IDENT_KMPC),
+                         getThreadID(CGF, Loc)};
+  CGF.EmitRuntimeCall(createDispatchFiniFunction(IVSize, IVSigned), Args);
+}
+
+llvm::Value *CGOpenMPRuntime::emitForNext(CodeGenFunction &CGF,
+                                          SourceLocation Loc, unsigned IVSize,
+                                          bool IVSigned, llvm::Value *IL,
+                                          llvm::Value *LB, llvm::Value *UB,
+                                          llvm::Value *ST) {
+  // Call __kmpc_dispatch_next(
+  //          ident_t *loc, kmp_int32 tid, kmp_int32 *p_lastiter,
+  //          kmp_int[32|64] *p_lower, kmp_int[32|64] *p_upper,
+  //          kmp_int[32|64] *p_stride);
   llvm::Value *Args[] = {
-      EmitOpenMPUpdateLocation(CGF, Loc, OMP_IDENT_KMPC),
-      GetOpenMPThreadID(CGF, Loc),
-      CGF.Builder.getInt32(Schedule), // Schedule type
-      IL,                             // &isLastIter
-      LB,                             // &LB
-      UB,                             // &UB
-      ST,                             // &Stride
-      CGF.Builder.getIntN(IVSize, 1), // Incr
-      Chunk                           // Chunk
+      emitUpdateLocation(CGF, Loc, OMP_IDENT_KMPC), getThreadID(CGF, Loc),
+      IL, // &isLastIter
+      LB, // &Lower
+      UB, // &Upper
+      ST  // &Stride
   };
-  assert((IVSize == 32 || IVSize == 64) &&
-         "Index size is not compatible with the omp runtime");
-  auto F = IVSize == 32 ? (IVSigned ? OMPRTL__kmpc_for_static_init_4
-                                    : OMPRTL__kmpc_for_static_init_4u)
-                        : (IVSigned ? OMPRTL__kmpc_for_static_init_8
-                                    : OMPRTL__kmpc_for_static_init_8u);
-  auto RTLFn = CreateRuntimeFunction(F);
-  CGF.EmitRuntimeCall(RTLFn, Args);
-}
-
-void CGOpenMPRuntime::EmitOMPForFinish(CodeGenFunction &CGF, SourceLocation Loc,
-                                       OpenMPScheduleClauseKind ScheduleKind) {
-  assert((ScheduleKind == OMPC_SCHEDULE_static ||
-          ScheduleKind == OMPC_SCHEDULE_unknown) &&
-         "Non-static schedule kinds are not yet implemented");
-  // Call __kmpc_for_static_fini(ident_t *loc, kmp_int32 tid);
-  llvm::Value *Args[] = {EmitOpenMPUpdateLocation(CGF, Loc, OMP_IDENT_KMPC),
-                         GetOpenMPThreadID(CGF, Loc)};
-  auto RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_for_static_fini);
-  CGF.EmitRuntimeCall(RTLFn, Args);
+  llvm::Value *Call =
+      CGF.EmitRuntimeCall(createDispatchNextFunction(IVSize, IVSigned), Args);
+  return CGF.EmitScalarConversion(
+      Call, CGF.getContext().getIntTypeForBitwidth(32, /* Signed */ true),
+      CGF.getContext().BoolTy);
 }
 
-void CGOpenMPRuntime::EmitOMPNumThreadsClause(CodeGenFunction &CGF,
-                                              llvm::Value *NumThreads,
-                                              SourceLocation Loc) {
+void CGOpenMPRuntime::emitNumThreadsClause(CodeGenFunction &CGF,
+                                           llvm::Value *NumThreads,
+                                           SourceLocation Loc) {
   // Build call __kmpc_push_num_threads(&loc, global_tid, num_threads)
   llvm::Value *Args[] = {
-      EmitOpenMPUpdateLocation(CGF, Loc), GetOpenMPThreadID(CGF, Loc),
+      emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc),
       CGF.Builder.CreateIntCast(NumThreads, CGF.Int32Ty, /*isSigned*/ true)};
-  llvm::Constant *RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_push_num_threads);
-  CGF.EmitRuntimeCall(RTLFn, Args);
+  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_push_num_threads),
+                      Args);
+}
+
+void CGOpenMPRuntime::emitFlush(CodeGenFunction &CGF, ArrayRef<const Expr *>,
+                                SourceLocation Loc) {
+  // Build call void __kmpc_flush(ident_t *loc)
+  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_flush),
+                      emitUpdateLocation(CGF, Loc));
+}
+
+namespace {
+/// \brief Indexes of fields for type kmp_task_t.
+enum KmpTaskTFields {
+  /// \brief List of shared variables.
+  KmpTaskTShareds,
+  /// \brief Task routine.
+  KmpTaskTRoutine,
+  /// \brief Partition id for the untied tasks.
+  KmpTaskTPartId,
+  /// \brief Function with call of destructors for private variables.
+  KmpTaskTDestructors,
+};
+} // namespace
+
+void CGOpenMPRuntime::emitKmpRoutineEntryT(QualType KmpInt32Ty) {
+  if (!KmpRoutineEntryPtrTy) {
+    // Build typedef kmp_int32 (* kmp_routine_entry_t)(kmp_int32, void *); type.
+    auto &C = CGM.getContext();
+    QualType KmpRoutineEntryTyArgs[] = {KmpInt32Ty, C.VoidPtrTy};
+    FunctionProtoType::ExtProtoInfo EPI;
+    KmpRoutineEntryPtrQTy = C.getPointerType(
+        C.getFunctionType(KmpInt32Ty, KmpRoutineEntryTyArgs, EPI));
+    KmpRoutineEntryPtrTy = CGM.getTypes().ConvertType(KmpRoutineEntryPtrQTy);
+  }
+}
+
+static void addFieldToRecordDecl(ASTContext &C, DeclContext *DC,
+                                 QualType FieldTy) {
+  auto *Field = FieldDecl::Create(
+      C, DC, SourceLocation(), SourceLocation(), /*Id=*/nullptr, FieldTy,
+      C.getTrivialTypeSourceInfo(FieldTy, SourceLocation()),
+      /*BW=*/nullptr, /*Mutable=*/false, /*InitStyle=*/ICIS_NoInit);
+  Field->setAccess(AS_public);
+  DC->addDecl(Field);
+}
+
+namespace {
+struct PrivateHelpersTy {
+  PrivateHelpersTy(const VarDecl *Original, const VarDecl *PrivateCopy,
+                   const VarDecl *PrivateElemInit)
+      : Original(Original), PrivateCopy(PrivateCopy),
+        PrivateElemInit(PrivateElemInit) {}
+  const VarDecl *Original;
+  const VarDecl *PrivateCopy;
+  const VarDecl *PrivateElemInit;
+};
+typedef std::pair<CharUnits /*Align*/, PrivateHelpersTy> PrivateDataTy;
+} // namespace
+
+static RecordDecl *
+createPrivatesRecordDecl(CodeGenModule &CGM,
+                         const ArrayRef<PrivateDataTy> Privates) {
+  if (!Privates.empty()) {
+    auto &C = CGM.getContext();
+    // Build struct .kmp_privates_t. {
+    //         /*  private vars  */
+    //       };
+    auto *RD = C.buildImplicitRecord(".kmp_privates.t");
+    RD->startDefinition();
+    for (auto &&Pair : Privates) {
+      auto Type = Pair.second.Original->getType();
+      Type = Type.getNonReferenceType();
+      addFieldToRecordDecl(C, RD, Type);
+    }
+    RD->completeDefinition();
+    return RD;
+  }
+  return nullptr;
 }
 
-void CGOpenMPRuntime::EmitOMPFlush(CodeGenFunction &CGF, ArrayRef<const Expr *>,
-                                   SourceLocation Loc) {
-  // Build call void __kmpc_flush(ident_t *loc, ...)
-  // FIXME: List of variables is ignored by libiomp5 runtime, no need to
-  // generate it, just request full memory fence.
-  llvm::Value *Args[] = {EmitOpenMPUpdateLocation(CGF, Loc),
-                         llvm::ConstantInt::get(CGM.Int32Ty, 0)};
-  auto *RTLFn = CreateRuntimeFunction(OMPRTL__kmpc_flush);
-  CGF.EmitRuntimeCall(RTLFn, Args);
+static RecordDecl *
+createKmpTaskTRecordDecl(CodeGenModule &CGM, QualType KmpInt32Ty,
+                         QualType KmpRoutineEntryPointerQTy) {
+  auto &C = CGM.getContext();
+  // Build struct kmp_task_t {
+  //         void *              shareds;
+  //         kmp_routine_entry_t routine;
+  //         kmp_int32           part_id;
+  //         kmp_routine_entry_t destructors;
+  //       };
+  auto *RD = C.buildImplicitRecord("kmp_task_t");
+  RD->startDefinition();
+  addFieldToRecordDecl(C, RD, C.VoidPtrTy);
+  addFieldToRecordDecl(C, RD, KmpRoutineEntryPointerQTy);
+  addFieldToRecordDecl(C, RD, KmpInt32Ty);
+  addFieldToRecordDecl(C, RD, KmpRoutineEntryPointerQTy);
+  RD->completeDefinition();
+  return RD;
+}
+
+static RecordDecl *
+createKmpTaskTWithPrivatesRecordDecl(CodeGenModule &CGM, QualType KmpTaskTQTy,
+                                     const ArrayRef<PrivateDataTy> Privates) {
+  auto &C = CGM.getContext();
+  // Build struct kmp_task_t_with_privates {
+  //         kmp_task_t task_data;
+  //         .kmp_privates_t. privates;
+  //       };
+  auto *RD = C.buildImplicitRecord("kmp_task_t_with_privates");
+  RD->startDefinition();
+  addFieldToRecordDecl(C, RD, KmpTaskTQTy);
+  if (auto *PrivateRD = createPrivatesRecordDecl(CGM, Privates)) {
+    addFieldToRecordDecl(C, RD, C.getRecordType(PrivateRD));
+  }
+  RD->completeDefinition();
+  return RD;
 }
+
+/// \brief Emit a proxy function which accepts kmp_task_t as the second
+/// argument.
+/// \code
+/// kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) {
+///   TaskFunction(gtid, tt->part_id, &tt->privates, task_privates_map,
+///   tt->shareds);
+///   return 0;
+/// }
+/// \endcode
+static llvm::Value *
+emitProxyTaskFunction(CodeGenModule &CGM, SourceLocation Loc,
+                      QualType KmpInt32Ty, QualType KmpTaskTWithPrivatesPtrQTy,
+                      QualType KmpTaskTWithPrivatesQTy, QualType KmpTaskTQTy,
+                      QualType SharedsPtrTy, llvm::Value *TaskFunction,
+                      llvm::Value *TaskPrivatesMap) {
+  auto &C = CGM.getContext();
+  FunctionArgList Args;
+  ImplicitParamDecl GtidArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpInt32Ty);
+  ImplicitParamDecl TaskTypeArg(C, /*DC=*/nullptr, Loc,
+                                /*Id=*/nullptr, KmpTaskTWithPrivatesPtrQTy);
+  Args.push_back(&GtidArg);
+  Args.push_back(&TaskTypeArg);
+  FunctionType::ExtInfo Info;
+  auto &TaskEntryFnInfo =
+      CGM.getTypes().arrangeFreeFunctionDeclaration(KmpInt32Ty, Args, Info,
+                                                    /*isVariadic=*/false);
+  auto *TaskEntryTy = CGM.getTypes().GetFunctionType(TaskEntryFnInfo);
+  auto *TaskEntry =
+      llvm::Function::Create(TaskEntryTy, llvm::GlobalValue::InternalLinkage,
+                             ".omp_task_entry.", &CGM.getModule());
+  CGM.SetLLVMFunctionAttributes(/*D=*/nullptr, TaskEntryFnInfo, TaskEntry);
+  CodeGenFunction CGF(CGM);
+  CGF.disableDebugInfo();
+  CGF.StartFunction(GlobalDecl(), KmpInt32Ty, TaskEntry, TaskEntryFnInfo, Args);
+
+  // TaskFunction(gtid, tt->task_data.part_id, &tt->privates, task_privates_map,
+  // tt->task_data.shareds);
+  auto *GtidParam = CGF.EmitLoadOfScalar(
+      CGF.GetAddrOfLocalVar(&GtidArg), /*Volatile=*/false,
+      C.getTypeAlignInChars(KmpInt32Ty).getQuantity(), KmpInt32Ty, Loc);
+  auto *TaskTypeArgAddr = CGF.Builder.CreateAlignedLoad(
+      CGF.GetAddrOfLocalVar(&TaskTypeArg), CGM.PointerAlignInBytes);
+  LValue TDBase =
+      CGF.MakeNaturalAlignAddrLValue(TaskTypeArgAddr, KmpTaskTWithPrivatesQTy);
+  auto *KmpTaskTWithPrivatesQTyRD =
+      cast<RecordDecl>(KmpTaskTWithPrivatesQTy->getAsTagDecl());
+  LValue Base =
+      CGF.EmitLValueForField(TDBase, *KmpTaskTWithPrivatesQTyRD->field_begin());
+  auto *KmpTaskTQTyRD = cast<RecordDecl>(KmpTaskTQTy->getAsTagDecl());
+  auto PartIdFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTPartId);
+  auto PartIdLVal = CGF.EmitLValueForField(Base, *PartIdFI);
+  auto *PartidParam = CGF.EmitLoadOfLValue(PartIdLVal, Loc).getScalarVal();
+
+  auto SharedsFI = std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTShareds);
+  auto SharedsLVal = CGF.EmitLValueForField(Base, *SharedsFI);
+  auto *SharedsParam = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+      CGF.EmitLoadOfLValue(SharedsLVal, Loc).getScalarVal(),
+      CGF.ConvertTypeForMem(SharedsPtrTy));
+
+  auto PrivatesFI = std::next(KmpTaskTWithPrivatesQTyRD->field_begin(), 1);
+  llvm::Value *PrivatesParam;
+  if (PrivatesFI != KmpTaskTWithPrivatesQTyRD->field_end()) {
+    auto PrivatesLVal = CGF.EmitLValueForField(TDBase, *PrivatesFI);
+    PrivatesParam = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+        PrivatesLVal.getAddress(), CGF.VoidPtrTy);
+  } else {
+    PrivatesParam = llvm::ConstantPointerNull::get(CGF.VoidPtrTy);
+  }
+
+  llvm::Value *CallArgs[] = {GtidParam, PartidParam, PrivatesParam,
+                             TaskPrivatesMap, SharedsParam};
+  CGF.EmitCallOrInvoke(TaskFunction, CallArgs);
+  CGF.EmitStoreThroughLValue(
+      RValue::get(CGF.Builder.getInt32(/*C=*/0)),
+      CGF.MakeNaturalAlignAddrLValue(CGF.ReturnValue, KmpInt32Ty));
+  CGF.FinishFunction();
+  return TaskEntry;
+}
+
+static llvm::Value *emitDestructorsFunction(CodeGenModule &CGM,
+                                            SourceLocation Loc,
+                                            QualType KmpInt32Ty,
+                                            QualType KmpTaskTWithPrivatesPtrQTy,
+                                            QualType KmpTaskTWithPrivatesQTy) {
+  auto &C = CGM.getContext();
+  FunctionArgList Args;
+  ImplicitParamDecl GtidArg(C, /*DC=*/nullptr, Loc, /*Id=*/nullptr, KmpInt32Ty);
+  ImplicitParamDecl TaskTypeArg(C, /*DC=*/nullptr, Loc,
+                                /*Id=*/nullptr, KmpTaskTWithPrivatesPtrQTy);
+  Args.push_back(&GtidArg);
+  Args.push_back(&TaskTypeArg);
+  FunctionType::ExtInfo Info;
+  auto &DestructorFnInfo =
+      CGM.getTypes().arrangeFreeFunctionDeclaration(KmpInt32Ty, Args, Info,
+                                                    /*isVariadic=*/false);
+  auto *DestructorFnTy = CGM.getTypes().GetFunctionType(DestructorFnInfo);
+  auto *DestructorFn =
+      llvm::Function::Create(DestructorFnTy, llvm::GlobalValue::InternalLinkage,
+                             ".omp_task_destructor.", &CGM.getModule());
+  CGM.SetLLVMFunctionAttributes(/*D=*/nullptr, DestructorFnInfo, DestructorFn);
+  CodeGenFunction CGF(CGM);
+  CGF.disableDebugInfo();
+  CGF.StartFunction(GlobalDecl(), KmpInt32Ty, DestructorFn, DestructorFnInfo,
+                    Args);
+
+  auto *TaskTypeArgAddr = CGF.Builder.CreateAlignedLoad(
+      CGF.GetAddrOfLocalVar(&TaskTypeArg), CGM.PointerAlignInBytes);
+  LValue Base =
+      CGF.MakeNaturalAlignAddrLValue(TaskTypeArgAddr, KmpTaskTWithPrivatesQTy);
+  auto *KmpTaskTWithPrivatesQTyRD =
+      cast<RecordDecl>(KmpTaskTWithPrivatesQTy->getAsTagDecl());
+  auto FI = std::next(KmpTaskTWithPrivatesQTyRD->field_begin());
+  Base = CGF.EmitLValueForField(Base, *FI);
+  for (auto *Field :
+       cast<RecordDecl>(FI->getType()->getAsTagDecl())->fields()) {
+    if (auto DtorKind = Field->getType().isDestructedType()) {
+      auto FieldLValue = CGF.EmitLValueForField(Base, Field);
+      CGF.pushDestroy(DtorKind, FieldLValue.getAddress(), Field->getType());
+    }
+  }
+  CGF.FinishFunction();
+  return DestructorFn;
+}
+
+/// \brief Emit a privates mapping function for correct handling of private and
+/// firstprivate variables.
+/// \code
+/// void .omp_task_privates_map.(const .privates. *noalias privs, <ty1>
+/// **noalias priv1,...,  <tyn> **noalias privn) {
+///   *priv1 = &.privates.priv1;
+///   ...;
+///   *privn = &.privates.privn;
+/// }
+/// \endcode
+static llvm::Value *
+emitTaskPrivateMappingFunction(CodeGenModule &CGM, SourceLocation Loc,
+                               const ArrayRef<const Expr *> PrivateVars,
+                               const ArrayRef<const Expr *> FirstprivateVars,
+                               QualType PrivatesQTy,
+                               const ArrayRef<PrivateDataTy> Privates) {
+  auto &C = CGM.getContext();
+  FunctionArgList Args;
+  ImplicitParamDecl TaskPrivatesArg(
+      C, /*DC=*/nullptr, Loc, /*Id=*/nullptr,
+      C.getPointerType(PrivatesQTy).withConst().withRestrict());
+  Args.push_back(&TaskPrivatesArg);
+  llvm::DenseMap<const VarDecl *, unsigned> PrivateVarsPos;
+  unsigned Counter = 1;
+  for (auto *E: PrivateVars) {
+    Args.push_back(ImplicitParamDecl::Create(
+        C, /*DC=*/nullptr, Loc,
+        /*Id=*/nullptr, C.getPointerType(C.getPointerType(E->getType()))
+                            .withConst()
+                            .withRestrict()));
+    auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
+    PrivateVarsPos[VD] = Counter;
+    ++Counter;
+  }
+  for (auto *E : FirstprivateVars) {
+    Args.push_back(ImplicitParamDecl::Create(
+        C, /*DC=*/nullptr, Loc,
+        /*Id=*/nullptr, C.getPointerType(C.getPointerType(E->getType()))
+                            .withConst()
+                            .withRestrict()));
+    auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
+    PrivateVarsPos[VD] = Counter;
+    ++Counter;
+  }
+  FunctionType::ExtInfo Info;
+  auto &TaskPrivatesMapFnInfo =
+      CGM.getTypes().arrangeFreeFunctionDeclaration(C.VoidTy, Args, Info,
+                                                    /*isVariadic=*/false);
+  auto *TaskPrivatesMapTy =
+      CGM.getTypes().GetFunctionType(TaskPrivatesMapFnInfo);
+  auto *TaskPrivatesMap = llvm::Function::Create(
+      TaskPrivatesMapTy, llvm::GlobalValue::InternalLinkage,
+      ".omp_task_privates_map.", &CGM.getModule());
+  CGM.SetLLVMFunctionAttributes(/*D=*/nullptr, TaskPrivatesMapFnInfo,
+                                TaskPrivatesMap);
+  TaskPrivatesMap->addFnAttr(llvm::Attribute::AlwaysInline);
+  CodeGenFunction CGF(CGM);
+  CGF.disableDebugInfo();
+  CGF.StartFunction(GlobalDecl(), C.VoidTy, TaskPrivatesMap,
+                    TaskPrivatesMapFnInfo, Args);
+
+  // *privi = &.privates.privi;
+  auto *TaskPrivatesArgAddr = CGF.Builder.CreateAlignedLoad(
+      CGF.GetAddrOfLocalVar(&TaskPrivatesArg), CGM.PointerAlignInBytes);
+  LValue Base =
+      CGF.MakeNaturalAlignAddrLValue(TaskPrivatesArgAddr, PrivatesQTy);
+  auto *PrivatesQTyRD = cast<RecordDecl>(PrivatesQTy->getAsTagDecl());
+  Counter = 0;
+  for (auto *Field : PrivatesQTyRD->fields()) {
+    auto FieldLVal = CGF.EmitLValueForField(Base, Field);
+    auto *VD = Args[PrivateVarsPos[Privates[Counter].second.Original]];
+    auto RefLVal = CGF.MakeNaturalAlignAddrLValue(CGF.GetAddrOfLocalVar(VD),
+                                                  VD->getType());
+    auto RefLoadRVal = CGF.EmitLoadOfLValue(RefLVal, Loc);
+    CGF.EmitStoreOfScalar(
+        FieldLVal.getAddress(),
+        CGF.MakeNaturalAlignAddrLValue(RefLoadRVal.getScalarVal(),
+                                       RefLVal.getType()->getPointeeType()));
+    ++Counter;
+  }
+  CGF.FinishFunction();
+  return TaskPrivatesMap;
+}
+
+static int array_pod_sort_comparator(const PrivateDataTy *P1,
+                                     const PrivateDataTy *P2) {
+  return P1->first < P2->first ? 1 : (P2->first < P1->first ? -1 : 0);
+}
+
+void CGOpenMPRuntime::emitTaskCall(
+    CodeGenFunction &CGF, SourceLocation Loc, const OMPExecutableDirective &D,
+    bool Tied, llvm::PointerIntPair<llvm::Value *, 1, bool> Final,
+    llvm::Value *TaskFunction, QualType SharedsTy, llvm::Value *Shareds,
+    const Expr *IfCond, const ArrayRef<const Expr *> PrivateVars,
+    const ArrayRef<const Expr *> PrivateCopies,
+    const ArrayRef<const Expr *> FirstprivateVars,
+    const ArrayRef<const Expr *> FirstprivateCopies,
+    const ArrayRef<const Expr *> FirstprivateInits) {
+  auto &C = CGM.getContext();
+  llvm::SmallVector<PrivateDataTy, 8> Privates;
+  // Aggregate privates and sort them by the alignment.
+  auto I = PrivateCopies.begin();
+  for (auto *E : PrivateVars) {
+    auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
+    Privates.push_back(std::make_pair(
+        C.getTypeAlignInChars(VD->getType()),
+        PrivateHelpersTy(VD, cast<VarDecl>(cast<DeclRefExpr>(*I)->getDecl()),
+                         /*PrivateElemInit=*/nullptr)));
+    ++I;
+  }
+  I = FirstprivateCopies.begin();
+  auto IElemInitRef = FirstprivateInits.begin();
+  for (auto *E : FirstprivateVars) {
+    auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
+    Privates.push_back(std::make_pair(
+        C.getTypeAlignInChars(VD->getType()),
+        PrivateHelpersTy(
+            VD, cast<VarDecl>(cast<DeclRefExpr>(*I)->getDecl()),
+            cast<VarDecl>(cast<DeclRefExpr>(*IElemInitRef)->getDecl()))));
+    ++I, ++IElemInitRef;
+  }
+  llvm::array_pod_sort(Privates.begin(), Privates.end(),
+                       array_pod_sort_comparator);
+  auto KmpInt32Ty = C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1);
+  // Build type kmp_routine_entry_t (if not built yet).
+  emitKmpRoutineEntryT(KmpInt32Ty);
+  // Build type kmp_task_t (if not built yet).
+  if (KmpTaskTQTy.isNull()) {
+    KmpTaskTQTy = C.getRecordType(
+        createKmpTaskTRecordDecl(CGM, KmpInt32Ty, KmpRoutineEntryPtrQTy));
+  }
+  auto *KmpTaskTQTyRD = cast<RecordDecl>(KmpTaskTQTy->getAsTagDecl());
+  // Build particular struct kmp_task_t for the given task.
+  auto *KmpTaskTWithPrivatesQTyRD =
+      createKmpTaskTWithPrivatesRecordDecl(CGM, KmpTaskTQTy, Privates);
+  auto KmpTaskTWithPrivatesQTy = C.getRecordType(KmpTaskTWithPrivatesQTyRD);
+  QualType KmpTaskTWithPrivatesPtrQTy =
+      C.getPointerType(KmpTaskTWithPrivatesQTy);
+  auto *KmpTaskTWithPrivatesTy = CGF.ConvertType(KmpTaskTWithPrivatesQTy);
+  auto *KmpTaskTWithPrivatesPtrTy = KmpTaskTWithPrivatesTy->getPointerTo();
+  auto KmpTaskTWithPrivatesTySize =
+      CGM.getSize(C.getTypeSizeInChars(KmpTaskTWithPrivatesQTy));
+  QualType SharedsPtrTy = C.getPointerType(SharedsTy);
+
+  // Emit initial values for private copies (if any).
+  llvm::Value *TaskPrivatesMap = nullptr;
+  auto *TaskPrivatesMapTy =
+      std::next(cast<llvm::Function>(TaskFunction)->getArgumentList().begin(),
+                3)
+          ->getType();
+  if (!Privates.empty()) {
+    auto FI = std::next(KmpTaskTWithPrivatesQTyRD->field_begin());
+    TaskPrivatesMap = emitTaskPrivateMappingFunction(
+        CGM, Loc, PrivateVars, FirstprivateVars, FI->getType(), Privates);
+    TaskPrivatesMap = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+        TaskPrivatesMap, TaskPrivatesMapTy);
+  } else {
+    TaskPrivatesMap = llvm::ConstantPointerNull::get(
+        cast<llvm::PointerType>(TaskPrivatesMapTy));
+  }
+  // Build a proxy function kmp_int32 .omp_task_entry.(kmp_int32 gtid,
+  // kmp_task_t *tt);
+  auto *TaskEntry = emitProxyTaskFunction(
+      CGM, Loc, KmpInt32Ty, KmpTaskTWithPrivatesPtrQTy, KmpTaskTWithPrivatesQTy,
+      KmpTaskTQTy, SharedsPtrTy, TaskFunction, TaskPrivatesMap);
+
+  // Build call kmp_task_t * __kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid,
+  // kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds,
+  // kmp_routine_entry_t *task_entry);
+  // Task flags. Format is taken from
+  // http://llvm.org/svn/llvm-project/openmp/trunk/runtime/src/kmp.h,
+  // description of kmp_tasking_flags struct.
+  const unsigned TiedFlag = 0x1;
+  const unsigned FinalFlag = 0x2;
+  unsigned Flags = Tied ? TiedFlag : 0;
+  auto *TaskFlags =
+      Final.getPointer()
+          ? CGF.Builder.CreateSelect(Final.getPointer(),
+                                     CGF.Builder.getInt32(FinalFlag),
+                                     CGF.Builder.getInt32(/*C=*/0))
+          : CGF.Builder.getInt32(Final.getInt() ? FinalFlag : 0);
+  TaskFlags = CGF.Builder.CreateOr(TaskFlags, CGF.Builder.getInt32(Flags));
+  auto SharedsSize = C.getTypeSizeInChars(SharedsTy);
+  llvm::Value *AllocArgs[] = {
+      emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc), TaskFlags,
+      KmpTaskTWithPrivatesTySize, CGM.getSize(SharedsSize),
+      CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(TaskEntry,
+                                                      KmpRoutineEntryPtrTy)};
+  auto *NewTask = CGF.EmitRuntimeCall(
+      createRuntimeFunction(OMPRTL__kmpc_omp_task_alloc), AllocArgs);
+  auto *NewTaskNewTaskTTy = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+      NewTask, KmpTaskTWithPrivatesPtrTy);
+  LValue Base = CGF.MakeNaturalAlignAddrLValue(NewTaskNewTaskTTy,
+                                               KmpTaskTWithPrivatesQTy);
+  LValue TDBase =
+      CGF.EmitLValueForField(Base, *KmpTaskTWithPrivatesQTyRD->field_begin());
+  // Fill the data in the resulting kmp_task_t record.
+  // Copy shareds if there are any.
+  llvm::Value *KmpTaskSharedsPtr = nullptr;
+  if (!SharedsTy->getAsStructureType()->getDecl()->field_empty()) {
+    KmpTaskSharedsPtr = CGF.EmitLoadOfScalar(
+        CGF.EmitLValueForField(
+            TDBase, *std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTShareds)),
+        Loc);
+    CGF.EmitAggregateCopy(KmpTaskSharedsPtr, Shareds, SharedsTy);
+  }
+  // Emit initial values for private copies (if any).
+  bool NeedsCleanup = false;
+  if (!Privates.empty()) {
+    auto FI = std::next(KmpTaskTWithPrivatesQTyRD->field_begin());
+    auto PrivatesBase = CGF.EmitLValueForField(Base, *FI);
+    FI = cast<RecordDecl>(FI->getType()->getAsTagDecl())->field_begin();
+    LValue SharedsBase;
+    if (!FirstprivateVars.empty()) {
+      SharedsBase = CGF.MakeNaturalAlignAddrLValue(
+          CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+              KmpTaskSharedsPtr, CGF.ConvertTypeForMem(SharedsPtrTy)),
+          SharedsTy);
+    }
+    CodeGenFunction::CGCapturedStmtInfo CapturesInfo(
+        cast<CapturedStmt>(*D.getAssociatedStmt()));
+    for (auto &&Pair : Privates) {
+      auto *VD = Pair.second.PrivateCopy;
+      auto *Init = VD->getAnyInitializer();
+      LValue PrivateLValue = CGF.EmitLValueForField(PrivatesBase, *FI);
+      if (Init) {
+        if (auto *Elem = Pair.second.PrivateElemInit) {
+          auto *OriginalVD = Pair.second.Original;
+          auto *SharedField = CapturesInfo.lookup(OriginalVD);
+          auto SharedRefLValue =
+              CGF.EmitLValueForField(SharedsBase, SharedField);
+          QualType Type = OriginalVD->getType();
+          if (Type->isArrayType()) {
+            // Initialize firstprivate array.
+            if (!isa<CXXConstructExpr>(Init) ||
+                CGF.isTrivialInitializer(Init)) {
+              // Perform simple memcpy.
+              CGF.EmitAggregateAssign(PrivateLValue.getAddress(),
+                                      SharedRefLValue.getAddress(), Type);
+            } else {
+              // Initialize firstprivate array using element-by-element
+              // intialization.
+              CGF.EmitOMPAggregateAssign(
+                  PrivateLValue.getAddress(), SharedRefLValue.getAddress(),
+                  Type, [&CGF, Elem, Init, &CapturesInfo](
+                            llvm::Value *DestElement, llvm::Value *SrcElement) {
+                    // Clean up any temporaries needed by the initialization.
+                    CodeGenFunction::OMPPrivateScope InitScope(CGF);
+                    InitScope.addPrivate(Elem, [SrcElement]() -> llvm::Value *{
+                      return SrcElement;
+                    });
+                    (void)InitScope.Privatize();
+                    // Emit initialization for single element.
+                    auto *OldCapturedStmtInfo = CGF.CapturedStmtInfo;
+                    CGF.CapturedStmtInfo = &CapturesInfo;
+                    CGF.EmitAnyExprToMem(Init, DestElement,
+                                         Init->getType().getQualifiers(),
+                                         /*IsInitializer=*/false);
+                    CGF.CapturedStmtInfo = OldCapturedStmtInfo;
+                  });
+            }
+          } else {
+            CodeGenFunction::OMPPrivateScope InitScope(CGF);
+            InitScope.addPrivate(Elem, [SharedRefLValue]() -> llvm::Value *{
+              return SharedRefLValue.getAddress();
+            });
+            (void)InitScope.Privatize();
+            auto *OldCapturedStmtInfo = CGF.CapturedStmtInfo;
+            CGF.CapturedStmtInfo = &CapturesInfo;
+            CGF.EmitExprAsInit(Init, VD, PrivateLValue,
+                               /*capturedByInit=*/false);
+            CGF.CapturedStmtInfo = OldCapturedStmtInfo;
+          }
+        } else {
+          CGF.EmitExprAsInit(Init, VD, PrivateLValue, /*capturedByInit=*/false);
+        }
+      }
+      NeedsCleanup = NeedsCleanup || FI->getType().isDestructedType();
+      ++FI;
+    }
+  }
+  // Provide pointer to function with destructors for privates.
+  llvm::Value *DestructorFn =
+      NeedsCleanup ? emitDestructorsFunction(CGM, Loc, KmpInt32Ty,
+                                             KmpTaskTWithPrivatesPtrQTy,
+                                             KmpTaskTWithPrivatesQTy)
+                   : llvm::ConstantPointerNull::get(
+                         cast<llvm::PointerType>(KmpRoutineEntryPtrTy));
+  LValue Destructor = CGF.EmitLValueForField(
+      TDBase, *std::next(KmpTaskTQTyRD->field_begin(), KmpTaskTDestructors));
+  CGF.EmitStoreOfScalar(CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+                            DestructorFn, KmpRoutineEntryPtrTy),
+                        Destructor);
+  // NOTE: routine and part_id fields are intialized by __kmpc_omp_task_alloc()
+  // libcall.
+  // Build kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t
+  // *new_task);
+  auto *ThreadID = getThreadID(CGF, Loc);
+  llvm::Value *TaskArgs[] = {emitUpdateLocation(CGF, Loc), ThreadID, NewTask};
+  auto &&ThenCodeGen = [this, &TaskArgs](CodeGenFunction &CGF) {
+    // TODO: add check for untied tasks.
+    CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_omp_task), TaskArgs);
+  };
+  typedef CallEndCleanup<std::extent<decltype(TaskArgs)>::value>
+      IfCallEndCleanup;
+  auto &&ElseCodeGen =
+      [this, &TaskArgs, ThreadID, NewTaskNewTaskTTy, TaskEntry](
+          CodeGenFunction &CGF) {
+        CodeGenFunction::RunCleanupsScope LocalScope(CGF);
+        CGF.EmitRuntimeCall(
+            createRuntimeFunction(OMPRTL__kmpc_omp_task_begin_if0), TaskArgs);
+        // Build void __kmpc_omp_task_complete_if0(ident_t *, kmp_int32 gtid,
+        // kmp_task_t *new_task);
+        CGF.EHStack.pushCleanup<IfCallEndCleanup>(
+            NormalAndEHCleanup,
+            createRuntimeFunction(OMPRTL__kmpc_omp_task_complete_if0),
+            llvm::makeArrayRef(TaskArgs));
+
+        // Call proxy_task_entry(gtid, new_task);
+        llvm::Value *OutlinedFnArgs[] = {ThreadID, NewTaskNewTaskTTy};
+        CGF.EmitCallOrInvoke(TaskEntry, OutlinedFnArgs);
+      };
+  if (IfCond) {
+    emitOMPIfClause(CGF, IfCond, ThenCodeGen, ElseCodeGen);
+  } else {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    ThenCodeGen(CGF);
+  }
+}
+
+static llvm::Value *emitReductionFunction(CodeGenModule &CGM,
+                                          llvm::Type *ArgsType,
+                                          ArrayRef<const Expr *> LHSExprs,
+                                          ArrayRef<const Expr *> RHSExprs,
+                                          ArrayRef<const Expr *> ReductionOps) {
+  auto &C = CGM.getContext();
+
+  // void reduction_func(void *LHSArg, void *RHSArg);
+  FunctionArgList Args;
+  ImplicitParamDecl LHSArg(C, /*DC=*/nullptr, SourceLocation(), /*Id=*/nullptr,
+                           C.VoidPtrTy);
+  ImplicitParamDecl RHSArg(C, /*DC=*/nullptr, SourceLocation(), /*Id=*/nullptr,
+                           C.VoidPtrTy);
+  Args.push_back(&LHSArg);
+  Args.push_back(&RHSArg);
+  FunctionType::ExtInfo EI;
+  auto &CGFI = CGM.getTypes().arrangeFreeFunctionDeclaration(
+      C.VoidTy, Args, EI, /*isVariadic=*/false);
+  auto *Fn = llvm::Function::Create(
+      CGM.getTypes().GetFunctionType(CGFI), llvm::GlobalValue::InternalLinkage,
+      ".omp.reduction.reduction_func", &CGM.getModule());
+  CGM.SetLLVMFunctionAttributes(/*D=*/nullptr, CGFI, Fn);
+  CodeGenFunction CGF(CGM);
+  CGF.StartFunction(GlobalDecl(), C.VoidTy, Fn, CGFI, Args);
+
+  // Dst = (void*[n])(LHSArg);
+  // Src = (void*[n])(RHSArg);
+  auto *LHS = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+      CGF.Builder.CreateAlignedLoad(CGF.GetAddrOfLocalVar(&LHSArg),
+                                    CGF.PointerAlignInBytes),
+      ArgsType);
+  auto *RHS = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+      CGF.Builder.CreateAlignedLoad(CGF.GetAddrOfLocalVar(&RHSArg),
+                                    CGF.PointerAlignInBytes),
+      ArgsType);
+
+  //  ...
+  //  *(Type<i>*)lhs[i] = RedOp<i>(*(Type<i>*)lhs[i], *(Type<i>*)rhs[i]);
+  //  ...
+  CodeGenFunction::OMPPrivateScope Scope(CGF);
+  for (unsigned I = 0, E = ReductionOps.size(); I < E; ++I) {
+    Scope.addPrivate(
+        cast<VarDecl>(cast<DeclRefExpr>(RHSExprs[I])->getDecl()),
+        [&]() -> llvm::Value *{
+          return CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+              CGF.Builder.CreateAlignedLoad(
+                  CGF.Builder.CreateStructGEP(/*Ty=*/nullptr, RHS, I),
+                  CGM.PointerAlignInBytes),
+              CGF.ConvertTypeForMem(C.getPointerType(RHSExprs[I]->getType())));
+        });
+    Scope.addPrivate(
+        cast<VarDecl>(cast<DeclRefExpr>(LHSExprs[I])->getDecl()),
+        [&]() -> llvm::Value *{
+          return CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+              CGF.Builder.CreateAlignedLoad(
+                  CGF.Builder.CreateStructGEP(/*Ty=*/nullptr, LHS, I),
+                  CGM.PointerAlignInBytes),
+              CGF.ConvertTypeForMem(C.getPointerType(LHSExprs[I]->getType())));
+        });
+  }
+  Scope.Privatize();
+  for (auto *E : ReductionOps) {
+    CGF.EmitIgnoredExpr(E);
+  }
+  Scope.ForceCleanup();
+  CGF.FinishFunction();
+  return Fn;
+}
+
+void CGOpenMPRuntime::emitReduction(CodeGenFunction &CGF, SourceLocation Loc,
+                                    ArrayRef<const Expr *> LHSExprs,
+                                    ArrayRef<const Expr *> RHSExprs,
+                                    ArrayRef<const Expr *> ReductionOps,
+                                    bool WithNowait) {
+  // Next code should be emitted for reduction:
+  //
+  // static kmp_critical_name lock = { 0 };
+  //
+  // void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
+  //  *(Type0*)lhs[0] = ReductionOperation0(*(Type0*)lhs[0], *(Type0*)rhs[0]);
+  //  ...
+  //  *(Type<n>-1*)lhs[<n>-1] = ReductionOperation<n>-1(*(Type<n>-1*)lhs[<n>-1],
+  //  *(Type<n>-1*)rhs[<n>-1]);
+  // }
+  //
+  // ...
+  // void *RedList[<n>] = {&<RHSExprs>[0], ..., &<RHSExprs>[<n>-1]};
+  // switch (__kmpc_reduce{_nowait}(<loc>, <gtid>, <n>, sizeof(RedList),
+  // RedList, reduce_func, &<lock>)) {
+  // case 1:
+  //  ...
+  //  <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]);
+  //  ...
+  // __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>);
+  // break;
+  // case 2:
+  //  ...
+  //  Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]));
+  //  ...
+  // [__kmpc_end_reduce(<loc>, <gtid>, &<lock>);]
+  // break;
+  // default:;
+  // }
+
+  auto &C = CGM.getContext();
+
+  // 1. Build a list of reduction variables.
+  // void *RedList[<n>] = {<ReductionVars>[0], ..., <ReductionVars>[<n>-1]};
+  llvm::APInt ArraySize(/*unsigned int numBits=*/32, RHSExprs.size());
+  QualType ReductionArrayTy =
+      C.getConstantArrayType(C.VoidPtrTy, ArraySize, ArrayType::Normal,
+                             /*IndexTypeQuals=*/0);
+  auto *ReductionList =
+      CGF.CreateMemTemp(ReductionArrayTy, ".omp.reduction.red_list");
+  for (unsigned I = 0, E = RHSExprs.size(); I < E; ++I) {
+    auto *Elem = CGF.Builder.CreateStructGEP(/*Ty=*/nullptr, ReductionList, I);
+    CGF.Builder.CreateAlignedStore(
+        CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
+            CGF.EmitLValue(RHSExprs[I]).getAddress(), CGF.VoidPtrTy),
+        Elem, CGM.PointerAlignInBytes);
+  }
+
+  // 2. Emit reduce_func().
+  auto *ReductionFn = emitReductionFunction(
+      CGM, CGF.ConvertTypeForMem(ReductionArrayTy)->getPointerTo(), LHSExprs,
+      RHSExprs, ReductionOps);
+
+  // 3. Create static kmp_critical_name lock = { 0 };
+  auto *Lock = getCriticalRegionLock(".reduction");
+
+  // 4. Build res = __kmpc_reduce{_nowait}(<loc>, <gtid>, <n>, sizeof(RedList),
+  // RedList, reduce_func, &<lock>);
+  auto *IdentTLoc = emitUpdateLocation(
+      CGF, Loc,
+      static_cast<OpenMPLocationFlags>(OMP_IDENT_KMPC | OMP_ATOMIC_REDUCE));
+  auto *ThreadId = getThreadID(CGF, Loc);
+  auto *ReductionArrayTySize = llvm::ConstantInt::get(
+      CGM.SizeTy, C.getTypeSizeInChars(ReductionArrayTy).getQuantity());
+  auto *RL = CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(ReductionList,
+                                                             CGF.VoidPtrTy);
+  llvm::Value *Args[] = {
+      IdentTLoc,                             // ident_t *<loc>
+      ThreadId,                              // i32 <gtid>
+      CGF.Builder.getInt32(RHSExprs.size()), // i32 <n>
+      ReductionArrayTySize,                  // size_type sizeof(RedList)
+      RL,                                    // void *RedList
+      ReductionFn, // void (*) (void *, void *) <reduce_func>
+      Lock         // kmp_critical_name *&<lock>
+  };
+  auto Res = CGF.EmitRuntimeCall(
+      createRuntimeFunction(WithNowait ? OMPRTL__kmpc_reduce_nowait
+                                       : OMPRTL__kmpc_reduce),
+      Args);
+
+  // 5. Build switch(res)
+  auto *DefaultBB = CGF.createBasicBlock(".omp.reduction.default");
+  auto *SwInst = CGF.Builder.CreateSwitch(Res, DefaultBB, /*NumCases=*/2);
+
+  // 6. Build case 1:
+  //  ...
+  //  <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]);
+  //  ...
+  // __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>);
+  // break;
+  auto *Case1BB = CGF.createBasicBlock(".omp.reduction.case1");
+  SwInst->addCase(CGF.Builder.getInt32(1), Case1BB);
+  CGF.EmitBlock(Case1BB);
+
+  {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    // Add emission of __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>);
+    llvm::Value *EndArgs[] = {
+        IdentTLoc, // ident_t *<loc>
+        ThreadId,  // i32 <gtid>
+        Lock       // kmp_critical_name *&<lock>
+    };
+    CGF.EHStack
+        .pushCleanup<CallEndCleanup<std::extent<decltype(EndArgs)>::value>>(
+            NormalAndEHCleanup,
+            createRuntimeFunction(WithNowait ? OMPRTL__kmpc_end_reduce_nowait
+                                             : OMPRTL__kmpc_end_reduce),
+            llvm::makeArrayRef(EndArgs));
+    for (auto *E : ReductionOps) {
+      CGF.EmitIgnoredExpr(E);
+    }
+  }
+
+  CGF.EmitBranch(DefaultBB);
+
+  // 7. Build case 2:
+  //  ...
+  //  Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]));
+  //  ...
+  // break;
+  auto *Case2BB = CGF.createBasicBlock(".omp.reduction.case2");
+  SwInst->addCase(CGF.Builder.getInt32(2), Case2BB);
+  CGF.EmitBlock(Case2BB);
+
+  {
+    CodeGenFunction::RunCleanupsScope Scope(CGF);
+    if (!WithNowait) {
+      // Add emission of __kmpc_end_reduce(<loc>, <gtid>, &<lock>);
+      llvm::Value *EndArgs[] = {
+          IdentTLoc, // ident_t *<loc>
+          ThreadId,  // i32 <gtid>
+          Lock       // kmp_critical_name *&<lock>
+      };
+      CGF.EHStack
+          .pushCleanup<CallEndCleanup<std::extent<decltype(EndArgs)>::value>>(
+              NormalAndEHCleanup,
+              createRuntimeFunction(OMPRTL__kmpc_end_reduce),
+              llvm::makeArrayRef(EndArgs));
+    }
+    auto I = LHSExprs.begin();
+    for (auto *E : ReductionOps) {
+      const Expr *XExpr = nullptr;
+      const Expr *EExpr = nullptr;
+      const Expr *UpExpr = nullptr;
+      BinaryOperatorKind BO = BO_Comma;
+      if (auto *BO = dyn_cast<BinaryOperator>(E)) {
+        if (BO->getOpcode() == BO_Assign) {
+          XExpr = BO->getLHS();
+          UpExpr = BO->getRHS();
+        }
+      }
+      // Try to emit update expression as a simple atomic.
+      auto *RHSExpr = UpExpr;
+      if (RHSExpr) {
+        // Analyze RHS part of the whole expression.
+        if (auto *ACO = dyn_cast<AbstractConditionalOperator>(
+                RHSExpr->IgnoreParenImpCasts())) {
+          // If this is a conditional operator, analyze its condition for
+          // min/max reduction operator.
+          RHSExpr = ACO->getCond();
+        }
+        if (auto *BORHS =
+                dyn_cast<BinaryOperator>(RHSExpr->IgnoreParenImpCasts())) {
+          EExpr = BORHS->getRHS();
+          BO = BORHS->getOpcode();
+        }
+      }
+      if (XExpr) {
+        auto *VD = cast<VarDecl>(cast<DeclRefExpr>(*I)->getDecl());
+        LValue X = CGF.EmitLValue(XExpr);
+        RValue E;
+        if (EExpr)
+          E = CGF.EmitAnyExpr(EExpr);
+        CGF.EmitOMPAtomicSimpleUpdateExpr(
+            X, E, BO, /*IsXLHSInRHSPart=*/true, llvm::Monotonic, Loc,
+            [&CGF, UpExpr, VD](RValue XRValue) {
+              CodeGenFunction::OMPPrivateScope PrivateScope(CGF);
+              PrivateScope.addPrivate(
+                  VD, [&CGF, VD, XRValue]() -> llvm::Value *{
+                    auto *LHSTemp = CGF.CreateMemTemp(VD->getType());
+                    CGF.EmitStoreThroughLValue(
+                        XRValue,
+                        CGF.MakeNaturalAlignAddrLValue(LHSTemp, VD->getType()));
+                    return LHSTemp;
+                  });
+              (void)PrivateScope.Privatize();
+              return CGF.EmitAnyExpr(UpExpr);
+            });
+      } else {
+        // Emit as a critical region.
+        emitCriticalRegion(CGF, ".atomic_reduction", [E](CodeGenFunction &CGF) {
+          CGF.EmitIgnoredExpr(E);
+        }, Loc);
+      }
+      ++I;
+    }
+  }
+
+  CGF.EmitBranch(DefaultBB);
+  CGF.EmitBlock(DefaultBB, /*IsFinished=*/true);
+}
+
+void CGOpenMPRuntime::emitTaskwaitCall(CodeGenFunction &CGF,
+                                       SourceLocation Loc) {
+  // Build call kmp_int32 __kmpc_omp_taskwait(ident_t *loc, kmp_int32
+  // global_tid);
+  llvm::Value *Args[] = {emitUpdateLocation(CGF, Loc), getThreadID(CGF, Loc)};
+  // Ignore return result until untied tasks are supported.
+  CGF.EmitRuntimeCall(createRuntimeFunction(OMPRTL__kmpc_omp_taskwait), Args);
+}
+
+void CGOpenMPRuntime::emitInlinedDirective(CodeGenFunction &CGF,
+                                           const RegionCodeGenTy &CodeGen) {
+  InlinedOpenMPRegionRAII Region(CGF, CodeGen);
+  CGF.CapturedStmtInfo->EmitBody(CGF, /*S=*/nullptr);
+}
+
diff --git a/lib/CodeGen/CGOpenMPRuntime.h b/lib/CodeGen/CGOpenMPRuntime.h
index 6daf8179c14e..f5aa4a51df93 100644
--- a/lib/CodeGen/CGOpenMPRuntime.h
+++ b/lib/CodeGen/CGOpenMPRuntime.h
@@ -14,6 +14,7 @@
 #ifndef LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIME_H
 #define LLVM_CLANG_LIB_CODEGEN_CGOPENMPRUNTIME_H
 
+#include "clang/AST/Type.h"
 #include "clang/Basic/OpenMPKinds.h"
 #include "clang/Basic/SourceLocation.h"
 #include "llvm/ADT/DenseMap.h"
@@ -42,9 +43,9 @@ namespace CodeGen {
 class CodeGenFunction;
 class CodeGenModule;
 
-class CGOpenMPRuntime {
-public:
+typedef llvm::function_ref<void(CodeGenFunction &)> RegionCodeGenTy;
 
+class CGOpenMPRuntime {
 private:
   enum OpenMPRTLFunction {
     /// \brief Call to void __kmpc_fork_call(ident_t *loc, kmp_int32 argc,
@@ -67,11 +68,7 @@ private:
     // Call to kmp_int32 __kmpc_cancel_barrier(ident_t *loc, kmp_int32
     // global_tid);
     OMPRTL__kmpc_cancel_barrier,
-    // Calls for static scheduling 'omp for' loops.
-    OMPRTL__kmpc_for_static_init_4,
-    OMPRTL__kmpc_for_static_init_4u,
-    OMPRTL__kmpc_for_static_init_8,
-    OMPRTL__kmpc_for_static_init_8u,
+    // Call to void __kmpc_for_static_fini(ident_t *loc, kmp_int32 global_tid);
     OMPRTL__kmpc_for_static_fini,
     // Call to void __kmpc_serialized_parallel(ident_t *loc, kmp_int32
     // global_tid);
@@ -82,12 +79,58 @@ private:
     // Call to void __kmpc_push_num_threads(ident_t *loc, kmp_int32 global_tid,
     // kmp_int32 num_threads);
     OMPRTL__kmpc_push_num_threads,
-    // Call to void __kmpc_flush(ident_t *loc, ...);
+    // Call to void __kmpc_flush(ident_t *loc);
     OMPRTL__kmpc_flush,
     // Call to kmp_int32 __kmpc_master(ident_t *, kmp_int32 global_tid);
     OMPRTL__kmpc_master,
     // Call to void __kmpc_end_master(ident_t *, kmp_int32 global_tid);
     OMPRTL__kmpc_end_master,
+    // Call to kmp_int32 __kmpc_omp_taskyield(ident_t *, kmp_int32 global_tid,
+    // int end_part);
+    OMPRTL__kmpc_omp_taskyield,
+    // Call to kmp_int32 __kmpc_single(ident_t *, kmp_int32 global_tid);
+    OMPRTL__kmpc_single,
+    // Call to void __kmpc_end_single(ident_t *, kmp_int32 global_tid);
+    OMPRTL__kmpc_end_single,
+    // Call to kmp_task_t * __kmpc_omp_task_alloc(ident_t *, kmp_int32 gtid,
+    // kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds,
+    // kmp_routine_entry_t *task_entry);
+    OMPRTL__kmpc_omp_task_alloc,
+    // Call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid, kmp_task_t *
+    // new_task);
+    OMPRTL__kmpc_omp_task,
+    // Call to void __kmpc_copyprivate(ident_t *loc, kmp_int32 global_tid,
+    // size_t cpy_size, void *cpy_data, void(*cpy_func)(void *, void *),
+    // kmp_int32 didit);
+    OMPRTL__kmpc_copyprivate,
+    // Call to kmp_int32 __kmpc_reduce(ident_t *loc, kmp_int32 global_tid,
+    // kmp_int32 num_vars, size_t reduce_size, void *reduce_data, void
+    // (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name *lck);
+    OMPRTL__kmpc_reduce,
+    // Call to kmp_int32 __kmpc_reduce_nowait(ident_t *loc, kmp_int32
+    // global_tid, kmp_int32 num_vars, size_t reduce_size, void *reduce_data,
+    // void (*reduce_func)(void *lhs_data, void *rhs_data), kmp_critical_name
+    // *lck);
+    OMPRTL__kmpc_reduce_nowait,
+    // Call to void __kmpc_end_reduce(ident_t *loc, kmp_int32 global_tid,
+    // kmp_critical_name *lck);
+    OMPRTL__kmpc_end_reduce,
+    // Call to void __kmpc_end_reduce_nowait(ident_t *loc, kmp_int32 global_tid,
+    // kmp_critical_name *lck);
+    OMPRTL__kmpc_end_reduce_nowait,
+    // Call to void __kmpc_omp_task_begin_if0(ident_t *, kmp_int32 gtid,
+    // kmp_task_t * new_task);
+    OMPRTL__kmpc_omp_task_begin_if0,
+    // Call to void __kmpc_omp_task_complete_if0(ident_t *, kmp_int32 gtid,
+    // kmp_task_t * new_task);
+    OMPRTL__kmpc_omp_task_complete_if0,
+    // Call to void __kmpc_ordered(ident_t *loc, kmp_int32 global_tid);
+    OMPRTL__kmpc_ordered,
+    // Call to void __kmpc_end_ordered(ident_t *loc, kmp_int32 global_tid);
+    OMPRTL__kmpc_end_ordered,
+    // Call to kmp_int32 __kmpc_omp_taskwait(ident_t *loc, kmp_int32
+    // global_tid);
+    OMPRTL__kmpc_omp_taskwait,
   };
 
   /// \brief Values for bit flags used in the ident_t to describe the fields.
@@ -118,7 +161,7 @@ private:
   /// \brief Map of flags and corresponding default locations.
   typedef llvm::DenseMap<unsigned, llvm::Value *> OpenMPDefaultLocMapTy;
   OpenMPDefaultLocMapTy OpenMPDefaultLocMap;
-  llvm::Value *GetOrCreateDefaultOpenMPLocation(OpenMPLocationFlags Flags);
+  llvm::Value *getOrCreateDefaultLocation(OpenMPLocationFlags Flags);
   /// \brief Describes ident structure that describes a source location.
   /// All descriptions are taken from
   /// http://llvm.org/svn/llvm-project/openmp/trunk/runtime/src/kmp.h
@@ -186,13 +229,28 @@ private:
   /// variables.
   llvm::StringMap<llvm::AssertingVH<llvm::Constant>, llvm::BumpPtrAllocator>
       InternalVars;
+  /// \brief Type typedef kmp_int32 (* kmp_routine_entry_t)(kmp_int32, void *);
+  llvm::Type *KmpRoutineEntryPtrTy;
+  QualType KmpRoutineEntryPtrQTy;
+  /// \brief Type typedef struct kmp_task {
+  ///    void *              shareds; /**< pointer to block of pointers to
+  ///    shared vars   */
+  ///    kmp_routine_entry_t routine; /**< pointer to routine to call for
+  ///    executing task */
+  ///    kmp_int32           part_id; /**< part id for the task */
+  ///    kmp_routine_entry_t destructors; /* pointer to function to invoke
+  ///    deconstructors of firstprivate C++ objects */
+  /// } kmp_task_t;
+  QualType KmpTaskTQTy;
+
+  /// \brief Build type kmp_routine_entry_t (if not built yet).
+  void emitKmpRoutineEntryT(QualType KmpInt32Ty);
 
   /// \brief Emits object of ident_t type with info for source location.
   /// \param Flags Flags for OpenMP location.
   ///
-  llvm::Value *
-  EmitOpenMPUpdateLocation(CodeGenFunction &CGF, SourceLocation Loc,
-                           OpenMPLocationFlags Flags = OMP_IDENT_KMPC);
+  llvm::Value *emitUpdateLocation(CodeGenFunction &CGF, SourceLocation Loc,
+                                  OpenMPLocationFlags Flags = OMP_IDENT_KMPC);
 
   /// \brief Returns pointer to ident_t type.
   llvm::Type *getIdentTyPointerTy();
@@ -203,7 +261,23 @@ private:
   /// \brief Returns specified OpenMP runtime function.
   /// \param Function OpenMP runtime function.
   /// \return Specified function.
-  llvm::Constant *CreateRuntimeFunction(OpenMPRTLFunction Function);
+  llvm::Constant *createRuntimeFunction(OpenMPRTLFunction Function);
+
+  /// \brief Returns __kmpc_for_static_init_* runtime function for the specified
+  /// size \a IVSize and sign \a IVSigned.
+  llvm::Constant *createForStaticInitFunction(unsigned IVSize, bool IVSigned);
+
+  /// \brief Returns __kmpc_dispatch_init_* runtime function for the specified
+  /// size \a IVSize and sign \a IVSigned.
+  llvm::Constant *createDispatchInitFunction(unsigned IVSize, bool IVSigned);
+
+  /// \brief Returns __kmpc_dispatch_next_* runtime function for the specified
+  /// size \a IVSize and sign \a IVSigned.
+  llvm::Constant *createDispatchNextFunction(unsigned IVSize, bool IVSigned);
+
+  /// \brief Returns __kmpc_dispatch_fini_* runtime function for the specified
+  /// size \a IVSize and sign \a IVSigned.
+  llvm::Constant *createDispatchFiniFunction(unsigned IVSize, bool IVSigned);
 
   /// \brief If the specified mangled name is not in the module, create and
   /// return threadprivate cache object. This object is a pointer's worth of
@@ -214,12 +288,12 @@ private:
 
   /// \brief Emits address of the word in a memory where current thread id is
   /// stored.
-  virtual llvm::Value *EmitThreadIDAddress(CodeGenFunction &CGF,
+  virtual llvm::Value *emitThreadIDAddress(CodeGenFunction &CGF,
                                            SourceLocation Loc);
 
   /// \brief Gets thread id value for the current thread.
   ///
-  llvm::Value *GetOpenMPThreadID(CodeGenFunction &CGF, SourceLocation Loc);
+  llvm::Value *getThreadID(CodeGenFunction &CGF, SourceLocation Loc);
 
   /// \brief Gets (if variable with the given name already exist) or creates
   /// internal global variable with the specified Name. The created variable has
@@ -227,7 +301,7 @@ private:
   /// \param Ty Type of the global variable. If it is exist already the type
   /// must be the same.
   /// \param Name Name of the variable.
-  llvm::Constant *GetOrCreateInternalVariable(llvm::Type *Ty,
+  llvm::Constant *getOrCreateInternalVariable(llvm::Type *Ty,
                                               const llvm::Twine &Name);
 
   /// \brief Set of threadprivate variables with the generated initializer.
@@ -239,78 +313,105 @@ private:
   /// \param CopyCtor Pointer to a global copy function for \a VD.
   /// \param Dtor Pointer to a global destructor function for \a VD.
   /// \param Loc Location of threadprivate declaration.
-  void EmitOMPThreadPrivateVarInit(CodeGenFunction &CGF, llvm::Value *VDAddr,
-                                   llvm::Value *Ctor, llvm::Value *CopyCtor,
-                                   llvm::Value *Dtor, SourceLocation Loc);
+  void emitThreadPrivateVarInit(CodeGenFunction &CGF, llvm::Value *VDAddr,
+                                llvm::Value *Ctor, llvm::Value *CopyCtor,
+                                llvm::Value *Dtor, SourceLocation Loc);
 
   /// \brief Returns corresponding lock object for the specified critical region
   /// name. If the lock object does not exist it is created, otherwise the
   /// reference to the existing copy is returned.
   /// \param CriticalName Name of the critical region.
   ///
-  llvm::Value *GetCriticalRegionLock(StringRef CriticalName);
+  llvm::Value *getCriticalRegionLock(StringRef CriticalName);
 
 public:
   explicit CGOpenMPRuntime(CodeGenModule &CGM);
   virtual ~CGOpenMPRuntime() {}
+  virtual void clear();
 
-  /// \brief Emits outlined function for the specified OpenMP directive \a D
-  /// (required for parallel and task directives). This outlined function has
-  /// type void(*)(kmp_int32 /*ThreadID*/, kmp_int32 /*BoundID*/, struct
-  /// context_vars*).
+  /// \brief Emits outlined function for the specified OpenMP parallel directive
+  /// \a D. This outlined function has type void(*)(kmp_int32 *ThreadID,
+  /// kmp_int32 BoundID, struct context_vars*).
   /// \param D OpenMP directive.
   /// \param ThreadIDVar Variable for thread id in the current OpenMP region.
-  ///
+  /// \param CodeGen Code generation sequence for the \a D directive.
   virtual llvm::Value *
-  EmitOpenMPOutlinedFunction(const OMPExecutableDirective &D,
-                             const VarDecl *ThreadIDVar);
+  emitParallelOutlinedFunction(const OMPExecutableDirective &D,
+                               const VarDecl *ThreadIDVar,
+                               const RegionCodeGenTy &CodeGen);
 
-  /// \brief Cleans up references to the objects in finished function.
+  /// \brief Emits outlined function for the OpenMP task directive \a D. This
+  /// outlined function has type void(*)(kmp_int32 ThreadID, kmp_int32
+  /// PartID, struct context_vars*).
+  /// \param D OpenMP directive.
+  /// \param ThreadIDVar Variable for thread id in the current OpenMP region.
+  /// \param CodeGen Code generation sequence for the \a D directive.
   ///
-  void FunctionFinished(CodeGenFunction &CGF);
+  virtual llvm::Value *emitTaskOutlinedFunction(const OMPExecutableDirective &D,
+                                                const VarDecl *ThreadIDVar,
+                                                const RegionCodeGenTy &CodeGen);
 
-  /// \brief Emits code for parallel call of the \a OutlinedFn with variables
-  /// captured in a record which address is stored in \a CapturedStruct.
-  /// \param OutlinedFn Outlined function to be run in parallel threads. Type of
-  /// this function is void(*)(kmp_int32, kmp_int32, struct context_vars*).
-  /// \param CapturedStruct A pointer to the record with the references to
-  /// variables used in \a OutlinedFn function.
+  /// \brief Cleans up references to the objects in finished function.
   ///
-  virtual void EmitOMPParallelCall(CodeGenFunction &CGF, SourceLocation Loc,
-                                   llvm::Value *OutlinedFn,
-                                   llvm::Value *CapturedStruct);
+  void functionFinished(CodeGenFunction &CGF);
 
-  /// \brief Emits code for serial call of the \a OutlinedFn with variables
-  /// captured in a record which address is stored in \a CapturedStruct.
-  /// \param OutlinedFn Outlined function to be run in serial mode.
+  /// \brief Emits code for parallel or serial call of the \a OutlinedFn with
+  /// variables captured in a record which address is stored in \a
+  /// CapturedStruct.
+  /// \param OutlinedFn Outlined function to be run in parallel threads. Type of
+  /// this function is void(*)(kmp_int32 *, kmp_int32, struct context_vars*).
   /// \param CapturedStruct A pointer to the record with the references to
   /// variables used in \a OutlinedFn function.
+  /// \param IfCond Condition in the associated 'if' clause, if it was
+  /// specified, nullptr otherwise.
   ///
-  virtual void EmitOMPSerialCall(CodeGenFunction &CGF, SourceLocation Loc,
-                                 llvm::Value *OutlinedFn,
-                                 llvm::Value *CapturedStruct);
+  virtual void emitParallelCall(CodeGenFunction &CGF, SourceLocation Loc,
+                                llvm::Value *OutlinedFn,
+                                llvm::Value *CapturedStruct,
+                                const Expr *IfCond);
 
   /// \brief Emits a critical region.
   /// \param CriticalName Name of the critical region.
   /// \param CriticalOpGen Generator for the statement associated with the given
   /// critical region.
-  virtual void EmitOMPCriticalRegion(CodeGenFunction &CGF,
-                                     StringRef CriticalName,
-                                     const std::function<void()> &CriticalOpGen,
-                                     SourceLocation Loc);
+  virtual void emitCriticalRegion(CodeGenFunction &CGF, StringRef CriticalName,
+                                  const RegionCodeGenTy &CriticalOpGen,
+                                  SourceLocation Loc);
 
   /// \brief Emits a master region.
   /// \param MasterOpGen Generator for the statement associated with the given
   /// master region.
-  virtual void EmitOMPMasterRegion(CodeGenFunction &CGF,
-                                   const std::function<void()> &MasterOpGen,
-                                   SourceLocation Loc);
+  virtual void emitMasterRegion(CodeGenFunction &CGF,
+                                const RegionCodeGenTy &MasterOpGen,
+                                SourceLocation Loc);
+
+  /// \brief Emits code for a taskyield directive.
+  virtual void emitTaskyieldCall(CodeGenFunction &CGF, SourceLocation Loc);
+
+  /// \brief Emits a single region.
+  /// \param SingleOpGen Generator for the statement associated with the given
+  /// single region.
+  virtual void emitSingleRegion(CodeGenFunction &CGF,
+                                const RegionCodeGenTy &SingleOpGen,
+                                SourceLocation Loc,
+                                ArrayRef<const Expr *> CopyprivateVars,
+                                ArrayRef<const Expr *> DestExprs,
+                                ArrayRef<const Expr *> SrcExprs,
+                                ArrayRef<const Expr *> AssignmentOps);
+
+  /// \brief Emit an ordered region.
+  /// \param OrderedOpGen Generator for the statement associated with the given
+  /// critical region.
+  virtual void emitOrderedRegion(CodeGenFunction &CGF,
+                                 const RegionCodeGenTy &OrderedOpGen,
+                                 SourceLocation Loc);
 
-  /// \brief Emits explicit barrier for OpenMP threads.
-  /// \param IsExplicit true, if it is explicitly specified barrier.
+  /// \brief Emit an implicit/explicit barrier for OpenMP threads.
+  /// \param Kind Directive for which this implicit barrier call must be
+  /// generated. Must be OMPD_barrier for explicit barrier generation.
   ///
-  virtual void EmitOMPBarrierCall(CodeGenFunction &CGF, SourceLocation Loc,
-                                  bool IsExplicit = true);
+  virtual void emitBarrierCall(CodeGenFunction &CGF, SourceLocation Loc,
+                               OpenMPDirectiveKind Kind);
 
   /// \brief Check if the specified \a ScheduleKind is static non-chunked.
   /// This kind of worksharing directive is emitted without outer loop.
@@ -320,6 +421,12 @@ public:
   virtual bool isStaticNonchunked(OpenMPScheduleClauseKind ScheduleKind,
                                   bool Chunked) const;
 
+  /// \brief Check if the specified \a ScheduleKind is dynamic.
+  /// This kind of worksharing directive is emitted without outer loop.
+  /// \param ScheduleKind Schedule Kind specified in the 'schedule' clause.
+  ///
+  virtual bool isDynamic(OpenMPScheduleClauseKind ScheduleKind) const;
+
   /// \brief Call the appropriate runtime routine to initialize it before start
   /// of loop.
   ///
@@ -332,6 +439,7 @@ public:
   /// \param SchedKind Schedule kind, specified by the 'schedule' clause.
   /// \param IVSize Size of the iteration variable in bits.
   /// \param IVSigned Sign of the interation variable.
+  /// \param Ordered true if loop is ordered, false otherwise.
   /// \param IL Address of the output variable in which the flag of the
   /// last iteration is returned.
   /// \param LB Address of the output variable in which the lower iteration
@@ -343,29 +451,58 @@ public:
   /// \param Chunk Value of the chunk for the static_chunked scheduled loop.
   /// For the default (nullptr) value, the chunk 1 will be used.
   ///
-  virtual void EmitOMPForInit(CodeGenFunction &CGF, SourceLocation Loc,
-                              OpenMPScheduleClauseKind SchedKind,
-                              unsigned IVSize, bool IVSigned, llvm::Value *IL,
-                              llvm::Value *LB, llvm::Value *UB, llvm::Value *ST,
-                              llvm::Value *Chunk = nullptr);
+  virtual void emitForInit(CodeGenFunction &CGF, SourceLocation Loc,
+                           OpenMPScheduleClauseKind SchedKind, unsigned IVSize,
+                           bool IVSigned, bool Ordered, llvm::Value *IL,
+                           llvm::Value *LB, llvm::Value *UB, llvm::Value *ST,
+                           llvm::Value *Chunk = nullptr);
+
+  /// \brief Call the appropriate runtime routine to notify that we finished
+  /// iteration of the ordered loop with the dynamic scheduling.
+  ///
+  /// \param CGF Reference to current CodeGenFunction.
+  /// \param Loc Clang source location.
+  /// \param IVSize Size of the iteration variable in bits.
+  /// \param IVSigned Sign of the interation variable.
+  ///
+  virtual void emitForOrderedIterationEnd(CodeGenFunction &CGF,
+                                          SourceLocation Loc, unsigned IVSize,
+                                          bool IVSigned);
 
   /// \brief Call the appropriate runtime routine to notify that we finished
   /// all the work with current loop.
   ///
   /// \param CGF Reference to current CodeGenFunction.
   /// \param Loc Clang source location.
-  /// \param ScheduleKind Schedule kind, specified by the 'schedule' clause.
   ///
-  virtual void EmitOMPForFinish(CodeGenFunction &CGF, SourceLocation Loc,
-                                OpenMPScheduleClauseKind ScheduleKind);
+  virtual void emitForStaticFinish(CodeGenFunction &CGF, SourceLocation Loc);
+
+  /// Call __kmpc_dispatch_next(
+  ///          ident_t *loc, kmp_int32 tid, kmp_int32 *p_lastiter,
+  ///          kmp_int[32|64] *p_lower, kmp_int[32|64] *p_upper,
+  ///          kmp_int[32|64] *p_stride);
+  /// \param IVSize Size of the iteration variable in bits.
+  /// \param IVSigned Sign of the interation variable.
+  /// \param IL Address of the output variable in which the flag of the
+  /// last iteration is returned.
+  /// \param LB Address of the output variable in which the lower iteration
+  /// number is returned.
+  /// \param UB Address of the output variable in which the upper iteration
+  /// number is returned.
+  /// \param ST Address of the output variable in which the stride value is
+  /// returned.
+  virtual llvm::Value *emitForNext(CodeGenFunction &CGF, SourceLocation Loc,
+                                   unsigned IVSize, bool IVSigned,
+                                   llvm::Value *IL, llvm::Value *LB,
+                                   llvm::Value *UB, llvm::Value *ST);
 
   /// \brief Emits call to void __kmpc_push_num_threads(ident_t *loc, kmp_int32
   /// global_tid, kmp_int32 num_threads) to generate code for 'num_threads'
   /// clause.
   /// \param NumThreads An integer value of threads.
-  virtual void EmitOMPNumThreadsClause(CodeGenFunction &CGF,
-                                       llvm::Value *NumThreads,
-                                       SourceLocation Loc);
+  virtual void emitNumThreadsClause(CodeGenFunction &CGF,
+                                    llvm::Value *NumThreads,
+                                    SourceLocation Loc);
 
   /// \brief Returns address of the threadprivate variable for the current
   /// thread.
@@ -373,10 +510,10 @@ public:
   /// \param VDAddr Address of the global variable \a VD.
   /// \param Loc Location of the reference to threadprivate var.
   /// \return Address of the threadprivate variable for the current thread.
-  virtual llvm::Value *getOMPAddrOfThreadPrivate(CodeGenFunction &CGF,
-                                                 const VarDecl *VD,
-                                                 llvm::Value *VDAddr,
-                                                 SourceLocation Loc);
+  virtual llvm::Value *getAddrOfThreadPrivate(CodeGenFunction &CGF,
+                                              const VarDecl *VD,
+                                              llvm::Value *VDAddr,
+                                              SourceLocation Loc);
 
   /// \brief Emit a code for initialization of threadprivate variable. It emits
   /// a call to runtime library which adds initial value to the newly created
@@ -387,15 +524,120 @@ public:
   /// \param Loc Location of threadprivate declaration.
   /// \param PerformInit true if initialization expression is not constant.
   virtual llvm::Function *
-  EmitOMPThreadPrivateVarDefinition(const VarDecl *VD, llvm::Value *VDAddr,
-                                    SourceLocation Loc, bool PerformInit,
-                                    CodeGenFunction *CGF = nullptr);
+  emitThreadPrivateVarDefinition(const VarDecl *VD, llvm::Value *VDAddr,
+                                 SourceLocation Loc, bool PerformInit,
+                                 CodeGenFunction *CGF = nullptr);
 
   /// \brief Emit flush of the variables specified in 'omp flush' directive.
   /// \param Vars List of variables to flush.
-  virtual void EmitOMPFlush(CodeGenFunction &CGF, ArrayRef<const Expr *> Vars,
-                            SourceLocation Loc);
+  virtual void emitFlush(CodeGenFunction &CGF, ArrayRef<const Expr *> Vars,
+                         SourceLocation Loc);
+
+  /// \brief Emit task region for the task directive. The task region is
+  /// emitted in several steps:
+  /// 1. Emit a call to kmp_task_t *__kmpc_omp_task_alloc(ident_t *, kmp_int32
+  /// gtid, kmp_int32 flags, size_t sizeof_kmp_task_t, size_t sizeof_shareds,
+  /// kmp_routine_entry_t *task_entry). Here task_entry is a pointer to the
+  /// function:
+  /// kmp_int32 .omp_task_entry.(kmp_int32 gtid, kmp_task_t *tt) {
+  ///   TaskFunction(gtid, tt->part_id, tt->shareds);
+  ///   return 0;
+  /// }
+  /// 2. Copy a list of shared variables to field shareds of the resulting
+  /// structure kmp_task_t returned by the previous call (if any).
+  /// 3. Copy a pointer to destructions function to field destructions of the
+  /// resulting structure kmp_task_t.
+  /// 4. Emit a call to kmp_int32 __kmpc_omp_task(ident_t *, kmp_int32 gtid,
+  /// kmp_task_t *new_task), where new_task is a resulting structure from
+  /// previous items.
+  /// \param D Current task directive.
+  /// \param Tied true if the task is tied (the task is tied to the thread that
+  /// can suspend its task region), false - untied (the task is not tied to any
+  /// thread).
+  /// \param Final Contains either constant bool value, or llvm::Value * of i1
+  /// type for final clause. If the value is true, the task forces all of its
+  /// child tasks to become final and included tasks.
+  /// \param TaskFunction An LLVM function with type void (*)(i32 /*gtid*/, i32
+  /// /*part_id*/, captured_struct */*__context*/);
+  /// \param SharedsTy A type which contains references the shared variables.
+  /// \param Shareds Context with the list of shared variables from the \a
+  /// TaskFunction.
+  /// \param IfCond Not a nullptr if 'if' clause was specified, nullptr
+  /// otherwise.
+  /// \param PrivateVars List of references to private variables for the task
+  /// directive.
+  /// \param PrivateCopies List of private copies for each private variable in
+  /// \p PrivateVars.
+  /// \param FirstprivateVars List of references to private variables for the
+  /// task directive.
+  /// \param FirstprivateCopies List of private copies for each private variable
+  /// in \p FirstprivateVars.
+  /// \param FirstprivateInits List of references to auto generated variables
+  /// used for initialization of a single array element. Used if firstprivate
+  /// variable is of array type.
+  virtual void emitTaskCall(CodeGenFunction &CGF, SourceLocation Loc,
+                            const OMPExecutableDirective &D, bool Tied,
+                            llvm::PointerIntPair<llvm::Value *, 1, bool> Final,
+                            llvm::Value *TaskFunction, QualType SharedsTy,
+                            llvm::Value *Shareds, const Expr *IfCond,
+                            const ArrayRef<const Expr *> PrivateVars,
+                            const ArrayRef<const Expr *> PrivateCopies,
+                            const ArrayRef<const Expr *> FirstprivateVars,
+                            const ArrayRef<const Expr *> FirstprivateCopies,
+                            const ArrayRef<const Expr *> FirstprivateInits);
+
+  /// \brief Emit code for the directive that does not require outlining.
+  ///
+  /// \param CodeGen Code generation sequence for the \a D directive.
+  virtual void emitInlinedDirective(CodeGenFunction &CGF,
+                                    const RegionCodeGenTy &CodeGen);
+  /// \brief Emit a code for reduction clause. Next code should be emitted for
+  /// reduction:
+  /// \code
+  ///
+  /// static kmp_critical_name lock = { 0 };
+  ///
+  /// void reduce_func(void *lhs[<n>], void *rhs[<n>]) {
+  ///  ...
+  ///  *(Type<i>*)lhs[i] = RedOp<i>(*(Type<i>*)lhs[i], *(Type<i>*)rhs[i]);
+  ///  ...
+  /// }
+  ///
+  /// ...
+  /// void *RedList[<n>] = {&<RHSExprs>[0], ..., &<RHSExprs>[<n>-1]};
+  /// switch (__kmpc_reduce{_nowait}(<loc>, <gtid>, <n>, sizeof(RedList),
+  /// RedList, reduce_func, &<lock>)) {
+  /// case 1:
+  ///  ...
+  ///  <LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]);
+  ///  ...
+  /// __kmpc_end_reduce{_nowait}(<loc>, <gtid>, &<lock>);
+  /// break;
+  /// case 2:
+  ///  ...
+  ///  Atomic(<LHSExprs>[i] = RedOp<i>(*<LHSExprs>[i], *<RHSExprs>[i]));
+  ///  ...
+  /// break;
+  /// default:;
+  /// }
+  /// \endcode
+  ///
+  /// \param LHSExprs List of LHS in \a ReductionOps reduction operations.
+  /// \param RHSExprs List of RHS in \a ReductionOps reduction operations.
+  /// \param ReductionOps List of reduction operations in form 'LHS binop RHS'
+  /// or 'operator binop(LHS, RHS)'.
+  /// \param WithNowait true if parent directive has also nowait clause, false
+  /// otherwise.
+  virtual void emitReduction(CodeGenFunction &CGF, SourceLocation Loc,
+                             ArrayRef<const Expr *> LHSExprs,
+                             ArrayRef<const Expr *> RHSExprs,
+                             ArrayRef<const Expr *> ReductionOps,
+                             bool WithNowait);
+
+  /// \brief Emit code for 'taskwait' directive.
+  virtual void emitTaskwaitCall(CodeGenFunction &CGF, SourceLocation Loc);
 };
+
 } // namespace CodeGen
 } // namespace clang
 
diff --git a/lib/CodeGen/CGRecordLayout.h b/lib/CodeGen/CGRecordLayout.h
index 2de0b2f87fc1..c15f9fde06f5 100644
--- a/lib/CodeGen/CGRecordLayout.h
+++ b/lib/CodeGen/CGRecordLayout.h
@@ -109,8 +109,8 @@ struct CGBitFieldInfo {
 class CGRecordLayout {
   friend class CodeGenTypes;
 
-  CGRecordLayout(const CGRecordLayout &) LLVM_DELETED_FUNCTION;
-  void operator=(const CGRecordLayout &) LLVM_DELETED_FUNCTION;
+  CGRecordLayout(const CGRecordLayout &) = delete;
+  void operator=(const CGRecordLayout &) = delete;
 
 private:
   /// The LLVM type corresponding to this record layout; used when
diff --git a/lib/CodeGen/CGRecordLayoutBuilder.cpp b/lib/CodeGen/CGRecordLayoutBuilder.cpp
index 7ad394b5eeb0..72ecd65c28a9 100644
--- a/lib/CodeGen/CGRecordLayoutBuilder.cpp
+++ b/lib/CodeGen/CGRecordLayoutBuilder.cpp
@@ -99,10 +99,25 @@ struct CGRecordLowering {
   MemberInfo StorageInfo(CharUnits Offset, llvm::Type *Data) {
     return MemberInfo(Offset, MemberInfo::Field, Data);
   }
-  bool useMSABI() {
+
+  /// The Microsoft bitfield layout rule allocates discrete storage
+  /// units of the field's formal type and only combines adjacent
+  /// fields of the same formal type.  We want to emit a layout with
+  /// these discrete storage units instead of combining them into a
+  /// continuous run.
+  bool isDiscreteBitFieldABI() {
     return Context.getTargetInfo().getCXXABI().isMicrosoft() ||
            D->isMsStruct(Context);
   }
+
+  /// The Itanium base layout rule allows virtual bases to overlap
+  /// other bases, which complicates layout in specific ways.
+  ///
+  /// Note specifically that the ms_struct attribute doesn't change this.
+  bool isOverlappingVBaseABI() {
+    return !Context.getTargetInfo().getCXXABI().isMicrosoft();
+  }
+
   /// \brief Wraps llvm::Type::getIntNTy with some implicit arguments.
   llvm::Type *getIntNType(uint64_t NumBits) {
     return llvm::Type::getIntNTy(Types.getLLVMContext(),
@@ -119,8 +134,9 @@ struct CGRecordLowering {
   /// for itanium bitfields that are smaller than their declared type.
   llvm::Type *getStorageType(const FieldDecl *FD) {
     llvm::Type *Type = Types.ConvertTypeForMem(FD->getType());
-    return useMSABI() || !FD->isBitField() ? Type :
-        getIntNType(std::min(FD->getBitWidthValue(Context),
+    if (!FD->isBitField()) return Type;
+    if (isDiscreteBitFieldABI()) return Type;
+    return getIntNType(std::min(FD->getBitWidthValue(Context),
                              (unsigned)Context.toBits(getSize(Type))));
   }
   /// \brief Gets the llvm Basesubobject type from a CXXRecordDecl.
@@ -137,15 +153,10 @@ struct CGRecordLowering {
     return CharUnits::fromQuantity(DataLayout.getABITypeAlignment(Type));
   }
   bool isZeroInitializable(const FieldDecl *FD) {
-    const Type *Type = FD->getType()->getBaseElementTypeUnsafe();
-    if (const MemberPointerType *MPT = Type->getAs<MemberPointerType>())
-      return Types.getCXXABI().isZeroInitializable(MPT);
-    if (const RecordType *RT = Type->getAs<RecordType>())
-      return isZeroInitializable(RT->getDecl());
-    return true;
+    return Types.isZeroInitializable(FD->getType());
   }
   bool isZeroInitializable(const RecordDecl *RD) {
-    return Types.getCGRecordLayout(RD).isZeroInitializable();
+    return Types.isZeroInitializable(RD);
   }
   void appendPaddingBytes(CharUnits Size) {
     if (!Size.isZero())
@@ -198,8 +209,8 @@ struct CGRecordLowering {
   bool IsZeroInitializableAsBase : 1;
   bool Packed : 1;
 private:
-  CGRecordLowering(const CGRecordLowering &) LLVM_DELETED_FUNCTION;
-  void operator =(const CGRecordLowering &) LLVM_DELETED_FUNCTION;
+  CGRecordLowering(const CGRecordLowering &) = delete;
+  void operator =(const CGRecordLowering &) = delete;
 };
 } // namespace {
 
@@ -365,7 +376,7 @@ CGRecordLowering::accumulateBitFields(RecordDecl::field_iterator Field,
   // used to determine if the ASTRecordLayout is treating these two bitfields as
   // contiguous.  StartBitOffset is offset of the beginning of the Run.
   uint64_t StartBitOffset, Tail = 0;
-  if (useMSABI()) {
+  if (isDiscreteBitFieldABI()) {
     for (; Field != FieldEnd; ++Field) {
       uint64_t BitOffset = getFieldBitOffset(*Field);
       // Zero-width bitfields end runs.
@@ -438,8 +449,12 @@ void CGRecordLowering::accumulateBases() {
   for (const auto &Base : RD->bases()) {
     if (Base.isVirtual())
       continue;
+
+    // Bases can be zero-sized even if not technically empty if they
+    // contain only a trailing array member.
     const CXXRecordDecl *BaseDecl = Base.getType()->getAsCXXRecordDecl();
-    if (!BaseDecl->isEmpty())
+    if (!BaseDecl->isEmpty() &&
+        !Context.getASTRecordLayout(BaseDecl).getSize().isZero())
       Members.push_back(MemberInfo(Layout.getBaseClassOffset(BaseDecl),
           MemberInfo::Base, getStorageType(BaseDecl), BaseDecl));
   }
@@ -461,7 +476,7 @@ void CGRecordLowering::accumulateVBases() {
   // smaller than the nvsize.  Here we check to see if such a base is placed
   // before the nvsize and set the scissor offset to that, instead of the
   // nvsize.
-  if (!useMSABI())
+  if (isOverlappingVBaseABI())
     for (const auto &Base : RD->vbases()) {
       const CXXRecordDecl *BaseDecl = Base.getType()->getAsCXXRecordDecl();
       if (BaseDecl->isEmpty())
@@ -482,7 +497,8 @@ void CGRecordLowering::accumulateVBases() {
     CharUnits Offset = Layout.getVBaseClassOffset(BaseDecl);
     // If the vbase is a primary virtual base of some base, then it doesn't
     // get its own storage location but instead lives inside of that base.
-    if (!useMSABI() && Context.isNearlyEmpty(BaseDecl) &&
+    if (isOverlappingVBaseABI() &&
+        Context.isNearlyEmpty(BaseDecl) &&
         !hasOwnStorage(RD, BaseDecl)) {
       Members.push_back(MemberInfo(Offset, MemberInfo::VBase, nullptr,
                                    BaseDecl));
diff --git a/lib/CodeGen/CGStmt.cpp b/lib/CodeGen/CGStmt.cpp
index e3bdf8661ef7..c879750015bf 100644
--- a/lib/CodeGen/CGStmt.cpp
+++ b/lib/CodeGen/CGStmt.cpp
@@ -88,6 +88,7 @@ void CodeGenFunction::EmitStmt(const Stmt *S) {
   case Stmt::ContinueStmtClass:
   case Stmt::DefaultStmtClass:
   case Stmt::CaseStmtClass:
+  case Stmt::SEHLeaveStmtClass:
     llvm_unreachable("should have emitted these statements as simple");
 
 #define STMT(Type, Base)
@@ -173,9 +174,6 @@ void CodeGenFunction::EmitStmt(const Stmt *S) {
   case Stmt::SEHTryStmtClass:
     EmitSEHTryStmt(cast<SEHTryStmt>(*S));
     break;
-  case Stmt::SEHLeaveStmtClass:
-    EmitSEHLeaveStmt(cast<SEHLeaveStmt>(*S));
-    break;
   case Stmt::OMPParallelDirectiveClass:
     EmitOMPParallelDirective(cast<OMPParallelDirective>(*S));
     break;
@@ -256,6 +254,7 @@ bool CodeGenFunction::EmitSimpleStmt(const Stmt *S) {
   case Stmt::ContinueStmtClass: EmitContinueStmt(cast<ContinueStmt>(*S)); break;
   case Stmt::DefaultStmtClass:  EmitDefaultStmt(cast<DefaultStmt>(*S));   break;
   case Stmt::CaseStmtClass:     EmitCaseStmt(cast<CaseStmt>(*S));         break;
+  case Stmt::SEHLeaveStmtClass: EmitSEHLeaveStmt(cast<SEHLeaveStmt>(*S)); break;
   }
 
   return true;
@@ -424,9 +423,8 @@ void CodeGenFunction::EmitLabel(const LabelDecl *D) {
     ResolveBranchFixups(Dest.getBlock());
   }
 
-  RegionCounter Cnt = getPGORegionCounter(D->getStmt());
   EmitBlock(Dest.getBlock());
-  Cnt.beginRegion(Builder);
+  incrementProfileCounter(D->getStmt());
 }
 
 /// Change the cleanup scope of the labels in this lexical scope to
@@ -514,7 +512,6 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
   // C99 6.8.4.1: The first substatement is executed if the expression compares
   // unequal to 0.  The condition must be a scalar type.
   LexicalScope ConditionScope(*this, S.getCond()->getSourceRange());
-  RegionCounter Cnt = getPGORegionCounter(&S);
 
   if (S.getConditionVariable())
     EmitAutoVarDecl(*S.getConditionVariable());
@@ -533,7 +530,7 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
     // This avoids emitting dead code and simplifies the CFG substantially.
     if (!ContainsLabel(Skipped)) {
       if (CondConstant)
-        Cnt.beginRegion(Builder);
+        incrementProfileCounter(&S);
       if (Executed) {
         RunCleanupsScope ExecutedScope(*this);
         EmitStmt(Executed);
@@ -550,11 +547,12 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
   if (S.getElse())
     ElseBlock = createBasicBlock("if.else");
 
-  EmitBranchOnBoolExpr(S.getCond(), ThenBlock, ElseBlock, Cnt.getCount());
+  EmitBranchOnBoolExpr(S.getCond(), ThenBlock, ElseBlock,
+                       getProfileCount(S.getThen()));
 
   // Emit the 'then' code.
   EmitBlock(ThenBlock);
-  Cnt.beginRegion(Builder);
+  incrementProfileCounter(&S);
   {
     RunCleanupsScope ThenScope(*this);
     EmitStmt(S.getThen());
@@ -564,8 +562,8 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
   // Emit the 'else' code if present.
   if (const Stmt *Else = S.getElse()) {
     {
-      // There is no need to emit line number for unconditional branch.
-      ApplyDebugLocation DL(*this);
+      // There is no need to emit line number for an unconditional branch.
+      auto NL = ApplyDebugLocation::CreateEmpty(*this);
       EmitBlock(ElseBlock);
     }
     {
@@ -573,8 +571,8 @@ void CodeGenFunction::EmitIfStmt(const IfStmt &S) {
       EmitStmt(Else);
     }
     {
-      // There is no need to emit line number for unconditional branch.
-      ApplyDebugLocation DL(*this);
+      // There is no need to emit line number for an unconditional branch.
+      auto NL = ApplyDebugLocation::CreateEmpty(*this);
       EmitBranch(ContBlock);
     }
   }
@@ -679,8 +677,6 @@ void CodeGenFunction::EmitCondBrHints(llvm::LLVMContext &Context,
 
 void CodeGenFunction::EmitWhileStmt(const WhileStmt &S,
                                     ArrayRef<const Attr *> WhileAttrs) {
-  RegionCounter Cnt = getPGORegionCounter(&S);
-
   // Emit the header for the loop, which will also become
   // the continue target.
   JumpDest LoopHeader = getJumpDestInCurrentScope("while.cond");
@@ -725,9 +721,9 @@ void CodeGenFunction::EmitWhileStmt(const WhileStmt &S,
     llvm::BasicBlock *ExitBlock = LoopExit.getBlock();
     if (ConditionScope.requiresCleanups())
       ExitBlock = createBasicBlock("while.exit");
-    llvm::BranchInst *CondBr =
-        Builder.CreateCondBr(BoolCondVal, LoopBody, ExitBlock,
-                             PGO.createLoopWeights(S.getCond(), Cnt));
+    llvm::BranchInst *CondBr = Builder.CreateCondBr(
+        BoolCondVal, LoopBody, ExitBlock,
+        createProfileWeightsForLoop(S.getCond(), getProfileCount(S.getBody())));
 
     if (ExitBlock != LoopExit.getBlock()) {
       EmitBlock(ExitBlock);
@@ -743,7 +739,7 @@ void CodeGenFunction::EmitWhileStmt(const WhileStmt &S,
   {
     RunCleanupsScope BodyScope(*this);
     EmitBlock(LoopBody);
-    Cnt.beginRegion(Builder);
+    incrementProfileCounter(&S);
     EmitStmt(S.getBody());
   }
 
@@ -772,7 +768,7 @@ void CodeGenFunction::EmitDoStmt(const DoStmt &S,
   JumpDest LoopExit = getJumpDestInCurrentScope("do.end");
   JumpDest LoopCond = getJumpDestInCurrentScope("do.cond");
 
-  RegionCounter Cnt = getPGORegionCounter(&S);
+  uint64_t ParentCount = getCurrentProfileCount();
 
   // Store the blocks to use for break and continue.
   BreakContinueStack.push_back(BreakContinue(LoopExit, LoopCond));
@@ -782,7 +778,7 @@ void CodeGenFunction::EmitDoStmt(const DoStmt &S,
 
   LoopStack.push(LoopBody);
 
-  EmitBlockWithFallThrough(LoopBody, Cnt);
+  EmitBlockWithFallThrough(LoopBody, &S);
   {
     RunCleanupsScope BodyScope(*this);
     EmitStmt(S.getBody());
@@ -809,9 +805,10 @@ void CodeGenFunction::EmitDoStmt(const DoStmt &S,
 
   // As long as the condition is true, iterate the loop.
   if (EmitBoolCondBranch) {
-    llvm::BranchInst *CondBr =
-        Builder.CreateCondBr(BoolCondVal, LoopBody, LoopExit.getBlock(),
-                             PGO.createLoopWeights(S.getCond(), Cnt));
+    uint64_t BackedgeCount = getProfileCount(S.getBody()) - ParentCount;
+    llvm::BranchInst *CondBr = Builder.CreateCondBr(
+        BoolCondVal, LoopBody, LoopExit.getBlock(),
+        createProfileWeightsForLoop(S.getCond(), BackedgeCount));
 
     // Attach metadata to loop body conditional branch.
     EmitCondBrHints(LoopBody->getContext(), CondBr, DoAttrs);
@@ -838,8 +835,6 @@ void CodeGenFunction::EmitForStmt(const ForStmt &S,
   if (S.getInit())
     EmitStmt(S.getInit());
 
-  RegionCounter Cnt = getPGORegionCounter(&S);
-
   // Start the loop with a block that tests the condition.
   // If there's an increment, the continue scope will be overwritten
   // later.
@@ -881,9 +876,9 @@ void CodeGenFunction::EmitForStmt(const ForStmt &S,
     // C99 6.8.5p2/p4: The first substatement is executed if the expression
     // compares unequal to 0.  The condition must be a scalar type.
     llvm::Value *BoolCondVal = EvaluateExprAsBool(S.getCond());
-    llvm::BranchInst *CondBr =
-        Builder.CreateCondBr(BoolCondVal, ForBody, ExitBlock,
-                             PGO.createLoopWeights(S.getCond(), Cnt));
+    llvm::BranchInst *CondBr = Builder.CreateCondBr(
+        BoolCondVal, ForBody, ExitBlock,
+        createProfileWeightsForLoop(S.getCond(), getProfileCount(S.getBody())));
 
     // Attach metadata to loop body conditional branch.
     EmitCondBrHints(ForBody->getContext(), CondBr, ForAttrs);
@@ -898,7 +893,7 @@ void CodeGenFunction::EmitForStmt(const ForStmt &S,
     // Treat it as a non-zero constant.  Don't even create a new block for the
     // body, just fall into it.
   }
-  Cnt.beginRegion(Builder);
+  incrementProfileCounter(&S);
 
   {
     // Create a separate cleanup scope for the body, in case it is not
@@ -939,8 +934,6 @@ CodeGenFunction::EmitCXXForRangeStmt(const CXXForRangeStmt &S,
   EmitStmt(S.getRangeStmt());
   EmitStmt(S.getBeginEndStmt());
 
-  RegionCounter Cnt = getPGORegionCounter(&S);
-
   // Start the loop with a block that tests the condition.
   // If there's an increment, the continue scope will be overwritten
   // later.
@@ -962,7 +955,8 @@ CodeGenFunction::EmitCXXForRangeStmt(const CXXForRangeStmt &S,
   // to bool, is true.
   llvm::Value *BoolCondVal = EvaluateExprAsBool(S.getCond());
   llvm::BranchInst *CondBr = Builder.CreateCondBr(
-      BoolCondVal, ForBody, ExitBlock, PGO.createLoopWeights(S.getCond(), Cnt));
+      BoolCondVal, ForBody, ExitBlock,
+      createProfileWeightsForLoop(S.getCond(), getProfileCount(S.getBody())));
 
   // Attach metadata to loop body conditional branch.
   EmitCondBrHints(ForBody->getContext(), CondBr, ForAttrs);
@@ -973,7 +967,7 @@ CodeGenFunction::EmitCXXForRangeStmt(const CXXForRangeStmt &S,
   }
 
   EmitBlock(ForBody);
-  Cnt.beginRegion(Builder);
+  incrementProfileCounter(&S);
 
   // Create a block for the increment. In case of a 'continue', we jump there.
   JumpDest Continue = getJumpDestInCurrentScope("for.inc");
@@ -1022,6 +1016,12 @@ void CodeGenFunction::EmitReturnOfRValue(RValue RV, QualType Ty) {
 /// if the function returns void, or may be missing one if the function returns
 /// non-void.  Fun stuff :).
 void CodeGenFunction::EmitReturnStmt(const ReturnStmt &S) {
+  // Returning from an outlined SEH helper is UB, and we already warn on it.
+  if (IsOutlinedSEHHelper) {
+    Builder.CreateUnreachable();
+    Builder.ClearInsertionPoint();
+  }
+
   // Emit the result value, even if unused, to evalute the side effects.
   const Expr *RV = S.getRetValue();
 
@@ -1133,13 +1133,11 @@ void CodeGenFunction::EmitCaseStmtRange(const CaseStmt &S) {
   llvm::APSInt LHS = S.getLHS()->EvaluateKnownConstInt(getContext());
   llvm::APSInt RHS = S.getRHS()->EvaluateKnownConstInt(getContext());
 
-  RegionCounter CaseCnt = getPGORegionCounter(&S);
-
   // Emit the code for this case. We do this first to make sure it is
   // properly chained from our predecessor before generating the
   // switch machinery to enter this block.
   llvm::BasicBlock *CaseDest = createBasicBlock("sw.bb");
-  EmitBlockWithFallThrough(CaseDest, CaseCnt);
+  EmitBlockWithFallThrough(CaseDest, &S);
   EmitStmt(S.getSubStmt());
 
   // If range is empty, do nothing.
@@ -1150,7 +1148,7 @@ void CodeGenFunction::EmitCaseStmtRange(const CaseStmt &S) {
   // FIXME: parameters such as this should not be hardcoded.
   if (Range.ult(llvm::APInt(Range.getBitWidth(), 64))) {
     // Range is small enough to add multiple switch instruction cases.
-    uint64_t Total = CaseCnt.getCount();
+    uint64_t Total = getProfileCount(&S);
     unsigned NCases = Range.getZExtValue() + 1;
     // We only have one region counter for the entire set of cases here, so we
     // need to divide the weights evenly between the generated cases, ensuring
@@ -1189,9 +1187,9 @@ void CodeGenFunction::EmitCaseStmtRange(const CaseStmt &S) {
 
   llvm::MDNode *Weights = nullptr;
   if (SwitchWeights) {
-    uint64_t ThisCount = CaseCnt.getCount();
+    uint64_t ThisCount = getProfileCount(&S);
     uint64_t DefaultCount = (*SwitchWeights)[0];
-    Weights = PGO.createBranchWeights(ThisCount, DefaultCount);
+    Weights = createProfileWeights(ThisCount, DefaultCount);
 
     // Since we're chaining the switch default through each large case range, we
     // need to update the weight for the default, ie, the first case, to include
@@ -1224,7 +1222,6 @@ void CodeGenFunction::EmitCaseStmt(const CaseStmt &S) {
     return;
   }
 
-  RegionCounter CaseCnt = getPGORegionCounter(&S);
   llvm::ConstantInt *CaseVal =
     Builder.getInt(S.getLHS()->EvaluateKnownConstInt(getContext()));
 
@@ -1239,7 +1236,7 @@ void CodeGenFunction::EmitCaseStmt(const CaseStmt &S) {
     // Only do this optimization if there are no cleanups that need emitting.
     if (isObviouslyBranchWithoutCleanups(Block)) {
       if (SwitchWeights)
-        SwitchWeights->push_back(CaseCnt.getCount());
+        SwitchWeights->push_back(getProfileCount(&S));
       SwitchInsn->addCase(CaseVal, Block.getBlock());
 
       // If there was a fallthrough into this case, make sure to redirect it to
@@ -1253,9 +1250,9 @@ void CodeGenFunction::EmitCaseStmt(const CaseStmt &S) {
   }
 
   llvm::BasicBlock *CaseDest = createBasicBlock("sw.bb");
-  EmitBlockWithFallThrough(CaseDest, CaseCnt);
+  EmitBlockWithFallThrough(CaseDest, &S);
   if (SwitchWeights)
-    SwitchWeights->push_back(CaseCnt.getCount());
+    SwitchWeights->push_back(getProfileCount(&S));
   SwitchInsn->addCase(CaseVal, CaseDest);
 
   // Recursively emitting the statement is acceptable, but is not wonderful for
@@ -1276,12 +1273,11 @@ void CodeGenFunction::EmitCaseStmt(const CaseStmt &S) {
     llvm::ConstantInt *CaseVal =
       Builder.getInt(CurCase->getLHS()->EvaluateKnownConstInt(getContext()));
 
-    CaseCnt = getPGORegionCounter(NextCase);
     if (SwitchWeights)
-      SwitchWeights->push_back(CaseCnt.getCount());
+      SwitchWeights->push_back(getProfileCount(NextCase));
     if (CGM.getCodeGenOpts().ProfileInstrGenerate) {
       CaseDest = createBasicBlock("sw.bb");
-      EmitBlockWithFallThrough(CaseDest, CaseCnt);
+      EmitBlockWithFallThrough(CaseDest, &S);
     }
 
     SwitchInsn->addCase(CaseVal, CaseDest);
@@ -1297,8 +1293,7 @@ void CodeGenFunction::EmitDefaultStmt(const DefaultStmt &S) {
   assert(DefaultBlock->empty() &&
          "EmitDefaultStmt: Default block already defined?");
 
-  RegionCounter Cnt = getPGORegionCounter(&S);
-  EmitBlockWithFallThrough(DefaultBlock, Cnt);
+  EmitBlockWithFallThrough(DefaultBlock, &S);
 
   EmitStmt(S.getSubStmt());
 }
@@ -1520,10 +1515,8 @@ void CodeGenFunction::EmitSwitchStmt(const SwitchStmt &S) {
     const SwitchCase *Case = nullptr;
     if (FindCaseStatementsForValue(S, ConstantCondValue, CaseStmts,
                                    getContext(), Case)) {
-      if (Case) {
-        RegionCounter CaseCnt = getPGORegionCounter(Case);
-        CaseCnt.beginRegion(Builder);
-      }
+      if (Case)
+        incrementProfileCounter(Case);
       RunCleanupsScope ExecutedScope(*this);
 
       // Emit the condition variable if needed inside the entire cleanup scope
@@ -1540,8 +1533,7 @@ void CodeGenFunction::EmitSwitchStmt(const SwitchStmt &S) {
       // specified series of statements and we're good.
       for (unsigned i = 0, e = CaseStmts.size(); i != e; ++i)
         EmitStmt(CaseStmts[i]);
-      RegionCounter ExitCnt = getPGORegionCounter(&S);
-      ExitCnt.beginRegion(Builder);
+      incrementProfileCounter(&S);
 
       // Now we want to restore the saved switch instance so that nested
       // switches continue to function properly
@@ -1572,7 +1564,7 @@ void CodeGenFunction::EmitSwitchStmt(const SwitchStmt &S) {
          Case;
          Case = Case->getNextSwitchCase()) {
       if (isa<DefaultStmt>(Case))
-        DefaultCount = getPGORegionCounter(Case).getCount();
+        DefaultCount = getProfileCount(Case);
       NumCases += 1;
     }
     SwitchWeights = new SmallVector<uint64_t, 16>();
@@ -1621,8 +1613,7 @@ void CodeGenFunction::EmitSwitchStmt(const SwitchStmt &S) {
 
   // Emit continuation.
   EmitBlock(SwitchExit.getBlock(), true);
-  RegionCounter ExitCnt = getPGORegionCounter(&S);
-  ExitCnt.beginRegion(Builder);
+  incrementProfileCounter(&S);
 
   if (SwitchWeights) {
     assert(SwitchWeights->size() == 1 + SwitchInsn->getNumCases() &&
@@ -1630,7 +1621,7 @@ void CodeGenFunction::EmitSwitchStmt(const SwitchStmt &S) {
     // If there's only one jump destination there's no sense weighting it.
     if (SwitchWeights->size() > 1)
       SwitchInsn->setMetadata(llvm::LLVMContext::MD_prof,
-                              PGO.createBranchWeights(*SwitchWeights));
+                              createProfileWeights(*SwitchWeights));
     delete SwitchWeights;
   }
   SwitchInsn = SavedSwitchInsn;
@@ -1696,7 +1687,7 @@ SimplifyConstraint(const char *Constraint, const TargetInfo &Target,
 static std::string
 AddVariableConstraints(const std::string &Constraint, const Expr &AsmExpr,
                        const TargetInfo &Target, CodeGenModule &CGM,
-                       const AsmStmt &Stmt) {
+                       const AsmStmt &Stmt, const bool EarlyClobber) {
   const DeclRefExpr *AsmDeclRef = dyn_cast<DeclRefExpr>(&AsmExpr);
   if (!AsmDeclRef)
     return Constraint;
@@ -1721,7 +1712,7 @@ AddVariableConstraints(const std::string &Constraint, const Expr &AsmExpr,
   }
   // Canonicalize the register here before returning it.
   Register = Target.getNormalizedGCCRegisterName(Register);
-  return "{" + Register.str() + "}";
+  return (EarlyClobber ? "&{" : "{") + Register.str() + "}";
 }
 
 llvm::Value*
@@ -1854,7 +1845,8 @@ void CodeGenFunction::EmitAsmStmt(const AsmStmt &S) {
     OutExpr = OutExpr->IgnoreParenNoopCasts(getContext());
 
     OutputConstraint = AddVariableConstraints(OutputConstraint, *OutExpr,
-                                              getTarget(), CGM, S);
+                                              getTarget(), CGM, S,
+                                              Info.earlyClobber());
 
     LValue Dest = EmitLValue(OutExpr);
     if (!Constraints.empty())
@@ -1956,10 +1948,9 @@ void CodeGenFunction::EmitAsmStmt(const AsmStmt &S) {
     InputConstraint = SimplifyConstraint(InputConstraint.c_str(), getTarget(),
                                          &OutputConstraintInfos);
 
-    InputConstraint =
-      AddVariableConstraints(InputConstraint,
-                            *InputExpr->IgnoreParenNoopCasts(getContext()),
-                            getTarget(), CGM, S);
+    InputConstraint = AddVariableConstraints(
+        InputConstraint, *InputExpr->IgnoreParenNoopCasts(getContext()),
+        getTarget(), CGM, S, false /* No EarlyClobber */);
 
     llvm::Value *Arg = EmitAsmInput(Info, InputExpr, Constraints);
 
@@ -2187,6 +2178,8 @@ CodeGenFunction::GenerateCapturedStmtFunction(const CapturedStmt &S) {
     llvm::Function::Create(FuncLLVMTy, llvm::GlobalValue::InternalLinkage,
                            CapturedStmtInfo->getHelperName(), &CGM.getModule());
   CGM.SetInternalFunctionAttributes(CD, F, FuncInfo);
+  if (CD->isNothrow())
+    F->addFnAttr(llvm::Attribute::NoUnwind);
 
   // Generate the function.
   StartFunction(CD, Ctx.VoidTy, F, FuncInfo, Args,
diff --git a/lib/CodeGen/CGStmtOpenMP.cpp b/lib/CodeGen/CGStmtOpenMP.cpp
index 78fd37ce6562..07fc6e966af6 100644
--- a/lib/CodeGen/CGStmtOpenMP.cpp
+++ b/lib/CodeGen/CGStmtOpenMP.cpp
@@ -20,258 +20,473 @@
 using namespace clang;
 using namespace CodeGen;
 
-namespace {
-/// \brief RAII for emitting code of CapturedStmt without function outlining.
-class InlinedOpenMPRegion {
-  CodeGenFunction &CGF;
-  CodeGenFunction::CGCapturedStmtInfo *PrevCapturedStmtInfo;
-  const Decl *StoredCurCodeDecl;
-
-  /// \brief A class to emit CapturedStmt construct as inlined statement without
-  /// generating a function for outlined code.
-  class CGInlinedOpenMPRegionInfo : public CodeGenFunction::CGCapturedStmtInfo {
-  public:
-    CGInlinedOpenMPRegionInfo() : CGCapturedStmtInfo() {}
-  };
-
-public:
-  InlinedOpenMPRegion(CodeGenFunction &CGF, const Stmt *S)
-      : CGF(CGF), PrevCapturedStmtInfo(CGF.CapturedStmtInfo),
-        StoredCurCodeDecl(CGF.CurCodeDecl) {
-    CGF.CurCodeDecl = cast<CapturedStmt>(S)->getCapturedDecl();
-    CGF.CapturedStmtInfo = new CGInlinedOpenMPRegionInfo();
-  }
-  ~InlinedOpenMPRegion() {
-    delete CGF.CapturedStmtInfo;
-    CGF.CapturedStmtInfo = PrevCapturedStmtInfo;
-    CGF.CurCodeDecl = StoredCurCodeDecl;
-  }
-};
-} // namespace
-
 //===----------------------------------------------------------------------===//
 //                              OpenMP Directive Emission
 //===----------------------------------------------------------------------===//
+void CodeGenFunction::EmitOMPAggregateAssign(
+    llvm::Value *DestAddr, llvm::Value *SrcAddr, QualType OriginalType,
+    const llvm::function_ref<void(llvm::Value *, llvm::Value *)> &CopyGen) {
+  // Perform element-by-element initialization.
+  QualType ElementTy;
+  auto SrcBegin = SrcAddr;
+  auto DestBegin = DestAddr;
+  auto ArrayTy = OriginalType->getAsArrayTypeUnsafe();
+  auto NumElements = emitArrayLength(ArrayTy, ElementTy, DestBegin);
+  // Cast from pointer to array type to pointer to single element.
+  SrcBegin = Builder.CreatePointerBitCastOrAddrSpaceCast(SrcBegin,
+                                                         DestBegin->getType());
+  auto DestEnd = Builder.CreateGEP(DestBegin, NumElements);
+  // The basic structure here is a while-do loop.
+  auto BodyBB = createBasicBlock("omp.arraycpy.body");
+  auto DoneBB = createBasicBlock("omp.arraycpy.done");
+  auto IsEmpty =
+      Builder.CreateICmpEQ(DestBegin, DestEnd, "omp.arraycpy.isempty");
+  Builder.CreateCondBr(IsEmpty, DoneBB, BodyBB);
+
+  // Enter the loop body, making that address the current address.
+  auto EntryBB = Builder.GetInsertBlock();
+  EmitBlock(BodyBB);
+  auto SrcElementCurrent =
+      Builder.CreatePHI(SrcBegin->getType(), 2, "omp.arraycpy.srcElementPast");
+  SrcElementCurrent->addIncoming(SrcBegin, EntryBB);
+  auto DestElementCurrent = Builder.CreatePHI(DestBegin->getType(), 2,
+                                              "omp.arraycpy.destElementPast");
+  DestElementCurrent->addIncoming(DestBegin, EntryBB);
+
+  // Emit copy.
+  CopyGen(DestElementCurrent, SrcElementCurrent);
+
+  // Shift the address forward by one element.
+  auto DestElementNext = Builder.CreateConstGEP1_32(
+      DestElementCurrent, /*Idx0=*/1, "omp.arraycpy.dest.element");
+  auto SrcElementNext = Builder.CreateConstGEP1_32(
+      SrcElementCurrent, /*Idx0=*/1, "omp.arraycpy.src.element");
+  // Check whether we've reached the end.
+  auto Done =
+      Builder.CreateICmpEQ(DestElementNext, DestEnd, "omp.arraycpy.done");
+  Builder.CreateCondBr(Done, DoneBB, BodyBB);
+  DestElementCurrent->addIncoming(DestElementNext, Builder.GetInsertBlock());
+  SrcElementCurrent->addIncoming(SrcElementNext, Builder.GetInsertBlock());
+
+  // Done.
+  EmitBlock(DoneBB, /*IsFinished=*/true);
+}
 
-/// \brief Emits code for OpenMP 'if' clause using specified \a CodeGen
-/// function. Here is the logic:
-/// if (Cond) {
-///   CodeGen(true);
-/// } else {
-///   CodeGen(false);
-/// }
-static void EmitOMPIfClause(CodeGenFunction &CGF, const Expr *Cond,
-                            const std::function<void(bool)> &CodeGen) {
-  CodeGenFunction::LexicalScope ConditionScope(CGF, Cond->getSourceRange());
-
-  // If the condition constant folds and can be elided, try to avoid emitting
-  // the condition and the dead arm of the if/else.
-  bool CondConstant;
-  if (CGF.ConstantFoldsToSimpleInteger(Cond, CondConstant)) {
-    CodeGen(CondConstant);
-    return;
-  }
-
-  // Otherwise, the condition did not fold, or we couldn't elide it.  Just
-  // emit the conditional branch.
-  auto ThenBlock = CGF.createBasicBlock(/*name*/ "omp_if.then");
-  auto ElseBlock = CGF.createBasicBlock(/*name*/ "omp_if.else");
-  auto ContBlock = CGF.createBasicBlock(/*name*/ "omp_if.end");
-  CGF.EmitBranchOnBoolExpr(Cond, ThenBlock, ElseBlock, /*TrueCount*/ 0);
-
-  // Emit the 'then' code.
-  CGF.EmitBlock(ThenBlock);
-  CodeGen(/*ThenBlock*/ true);
-  CGF.EmitBranch(ContBlock);
-  // Emit the 'else' code if present.
-  {
-    // There is no need to emit line number for unconditional branch.
-    ApplyDebugLocation DL(CGF);
-    CGF.EmitBlock(ElseBlock);
-  }
-  CodeGen(/*ThenBlock*/ false);
-  {
-    // There is no need to emit line number for unconditional branch.
-    ApplyDebugLocation DL(CGF);
-    CGF.EmitBranch(ContBlock);
-  }
-  // Emit the continuation block for code after the if.
-  CGF.EmitBlock(ContBlock, /*IsFinished*/ true);
-}
-
-void CodeGenFunction::EmitOMPAggregateAssign(LValue OriginalAddr,
-                                             llvm::Value *PrivateAddr,
-                                             const Expr *AssignExpr,
-                                             QualType OriginalType,
-                                             const VarDecl *VDInit) {
-  EmitBlock(createBasicBlock(".omp.assign.begin."));
-  if (!isa<CXXConstructExpr>(AssignExpr) || isTrivialInitializer(AssignExpr)) {
-    // Perform simple memcpy.
-    EmitAggregateAssign(PrivateAddr, OriginalAddr.getAddress(),
-                        AssignExpr->getType());
-  } else {
-    // Perform element-by-element initialization.
-    QualType ElementTy;
-    auto SrcBegin = OriginalAddr.getAddress();
-    auto DestBegin = PrivateAddr;
-    auto ArrayTy = OriginalType->getAsArrayTypeUnsafe();
-    auto SrcNumElements = emitArrayLength(ArrayTy, ElementTy, SrcBegin);
-    auto DestNumElements = emitArrayLength(ArrayTy, ElementTy, DestBegin);
-    auto SrcEnd = Builder.CreateGEP(SrcBegin, SrcNumElements);
-    auto DestEnd = Builder.CreateGEP(DestBegin, DestNumElements);
-    // The basic structure here is a do-while loop, because we don't
-    // need to check for the zero-element case.
-    auto BodyBB = createBasicBlock("omp.arraycpy.body");
-    auto DoneBB = createBasicBlock("omp.arraycpy.done");
-    auto IsEmpty =
-        Builder.CreateICmpEQ(DestBegin, DestEnd, "omp.arraycpy.isempty");
-    Builder.CreateCondBr(IsEmpty, DoneBB, BodyBB);
-
-    // Enter the loop body, making that address the current address.
-    auto EntryBB = Builder.GetInsertBlock();
-    EmitBlock(BodyBB);
-    auto SrcElementPast = Builder.CreatePHI(SrcBegin->getType(), 2,
-                                            "omp.arraycpy.srcElementPast");
-    SrcElementPast->addIncoming(SrcEnd, EntryBB);
-    auto DestElementPast = Builder.CreatePHI(DestBegin->getType(), 2,
-                                             "omp.arraycpy.destElementPast");
-    DestElementPast->addIncoming(DestEnd, EntryBB);
-
-    // Shift the address back by one element.
-    auto NegativeOne = llvm::ConstantInt::get(SizeTy, -1, true);
-    auto DestElement = Builder.CreateGEP(DestElementPast, NegativeOne,
-                                         "omp.arraycpy.dest.element");
-    auto SrcElement = Builder.CreateGEP(SrcElementPast, NegativeOne,
-                                        "omp.arraycpy.src.element");
-    {
-      // Create RunCleanScope to cleanup possible temps.
-      CodeGenFunction::RunCleanupsScope Init(*this);
-      // Emit initialization for single element.
-      LocalDeclMap[VDInit] = SrcElement;
-      EmitAnyExprToMem(AssignExpr, DestElement,
-                       AssignExpr->getType().getQualifiers(),
-                       /*IsInitializer*/ false);
-      LocalDeclMap.erase(VDInit);
+void CodeGenFunction::EmitOMPCopy(CodeGenFunction &CGF,
+                                  QualType OriginalType, llvm::Value *DestAddr,
+                                  llvm::Value *SrcAddr, const VarDecl *DestVD,
+                                  const VarDecl *SrcVD, const Expr *Copy) {
+  if (OriginalType->isArrayType()) {
+    auto *BO = dyn_cast<BinaryOperator>(Copy);
+    if (BO && BO->getOpcode() == BO_Assign) {
+      // Perform simple memcpy for simple copying.
+      CGF.EmitAggregateAssign(DestAddr, SrcAddr, OriginalType);
+    } else {
+      // For arrays with complex element types perform element by element
+      // copying.
+      CGF.EmitOMPAggregateAssign(
+          DestAddr, SrcAddr, OriginalType,
+          [&CGF, Copy, SrcVD, DestVD](llvm::Value *DestElement,
+                                          llvm::Value *SrcElement) {
+            // Working with the single array element, so have to remap
+            // destination and source variables to corresponding array
+            // elements.
+            CodeGenFunction::OMPPrivateScope Remap(CGF);
+            Remap.addPrivate(DestVD, [DestElement]() -> llvm::Value *{
+              return DestElement;
+            });
+            Remap.addPrivate(
+                SrcVD, [SrcElement]() -> llvm::Value *{ return SrcElement; });
+            (void)Remap.Privatize();
+            CGF.EmitIgnoredExpr(Copy);
+          });
     }
-
-    // Check whether we've reached the end.
-    auto Done =
-        Builder.CreateICmpEQ(DestElement, DestBegin, "omp.arraycpy.done");
-    Builder.CreateCondBr(Done, DoneBB, BodyBB);
-    DestElementPast->addIncoming(DestElement, Builder.GetInsertBlock());
-    SrcElementPast->addIncoming(SrcElement, Builder.GetInsertBlock());
-
-    // Done.
-    EmitBlock(DoneBB, true);
+  } else {
+    // Remap pseudo source variable to private copy.
+    CodeGenFunction::OMPPrivateScope Remap(CGF);
+    Remap.addPrivate(SrcVD, [SrcAddr]() -> llvm::Value *{ return SrcAddr; });
+    Remap.addPrivate(DestVD, [DestAddr]() -> llvm::Value *{ return DestAddr; });
+    (void)Remap.Privatize();
+    // Emit copying of the whole variable.
+    CGF.EmitIgnoredExpr(Copy);
   }
-  EmitBlock(createBasicBlock(".omp.assign.end."));
 }
 
-void CodeGenFunction::EmitOMPFirstprivateClause(
-    const OMPExecutableDirective &D,
-    CodeGenFunction::OMPPrivateScope &PrivateScope) {
-  auto PrivateFilter = [](const OMPClause *C) -> bool {
-    return C->getClauseKind() == OMPC_firstprivate;
-  };
-  for (OMPExecutableDirective::filtered_clause_iterator<decltype(PrivateFilter)>
-           I(D.clauses(), PrivateFilter); I; ++I) {
+bool CodeGenFunction::EmitOMPFirstprivateClause(const OMPExecutableDirective &D,
+                                                OMPPrivateScope &PrivateScope) {
+  llvm::DenseSet<const VarDecl *> EmittedAsFirstprivate;
+  for (auto &&I = D.getClausesOfKind(OMPC_firstprivate); I; ++I) {
     auto *C = cast<OMPFirstprivateClause>(*I);
     auto IRef = C->varlist_begin();
     auto InitsRef = C->inits().begin();
     for (auto IInit : C->private_copies()) {
       auto *OrigVD = cast<VarDecl>(cast<DeclRefExpr>(*IRef)->getDecl());
-      auto *VD = cast<VarDecl>(cast<DeclRefExpr>(IInit)->getDecl());
-      bool IsRegistered;
-      if (*InitsRef != nullptr) {
-        // Emit VarDecl with copy init for arrays.
-        auto *FD = CapturedStmtInfo->lookup(OrigVD);
-        LValue Base = MakeNaturalAlignAddrLValue(
-            CapturedStmtInfo->getContextValue(),
-            getContext().getTagDeclType(FD->getParent()));
-        auto OriginalAddr = EmitLValueForField(Base, FD);
-        auto VDInit = cast<VarDecl>(cast<DeclRefExpr>(*InitsRef)->getDecl());
-        IsRegistered = PrivateScope.addPrivate(OrigVD, [&]() -> llvm::Value * {
-          auto Emission = EmitAutoVarAlloca(*VD);
-          // Emit initialization of aggregate firstprivate vars.
-          EmitOMPAggregateAssign(OriginalAddr, Emission.getAllocatedAddress(),
-                                 VD->getInit(), (*IRef)->getType(), VDInit);
-          EmitAutoVarCleanups(Emission);
-          return Emission.getAllocatedAddress();
-        });
-      } else
-        IsRegistered = PrivateScope.addPrivate(OrigVD, [&]() -> llvm::Value * {
-          // Emit private VarDecl with copy init.
-          EmitDecl(*VD);
-          return GetAddrOfLocalVar(VD);
-        });
-      assert(IsRegistered && "counter already registered as private");
-      // Silence the warning about unused variable.
-      (void)IsRegistered;
+      if (EmittedAsFirstprivate.count(OrigVD) == 0) {
+        EmittedAsFirstprivate.insert(OrigVD);
+        auto *VD = cast<VarDecl>(cast<DeclRefExpr>(IInit)->getDecl());
+        auto *VDInit = cast<VarDecl>(cast<DeclRefExpr>(*InitsRef)->getDecl());
+        bool IsRegistered;
+        DeclRefExpr DRE(
+            const_cast<VarDecl *>(OrigVD),
+            /*RefersToEnclosingVariableOrCapture=*/CapturedStmtInfo->lookup(
+                OrigVD) != nullptr,
+            (*IRef)->getType(), VK_LValue, (*IRef)->getExprLoc());
+        auto *OriginalAddr = EmitLValue(&DRE).getAddress();
+        QualType Type = OrigVD->getType();
+        if (Type->isArrayType()) {
+          // Emit VarDecl with copy init for arrays.
+          // Get the address of the original variable captured in current
+          // captured region.
+          IsRegistered = PrivateScope.addPrivate(OrigVD, [&]() -> llvm::Value *{
+            auto Emission = EmitAutoVarAlloca(*VD);
+            auto *Init = VD->getInit();
+            if (!isa<CXXConstructExpr>(Init) || isTrivialInitializer(Init)) {
+              // Perform simple memcpy.
+              EmitAggregateAssign(Emission.getAllocatedAddress(), OriginalAddr,
+                                  Type);
+            } else {
+              EmitOMPAggregateAssign(
+                  Emission.getAllocatedAddress(), OriginalAddr, Type,
+                  [this, VDInit, Init](llvm::Value *DestElement,
+                                       llvm::Value *SrcElement) {
+                    // Clean up any temporaries needed by the initialization.
+                    RunCleanupsScope InitScope(*this);
+                    // Emit initialization for single element.
+                    LocalDeclMap[VDInit] = SrcElement;
+                    EmitAnyExprToMem(Init, DestElement,
+                                     Init->getType().getQualifiers(),
+                                     /*IsInitializer*/ false);
+                    LocalDeclMap.erase(VDInit);
+                  });
+            }
+            EmitAutoVarCleanups(Emission);
+            return Emission.getAllocatedAddress();
+          });
+        } else {
+          IsRegistered = PrivateScope.addPrivate(OrigVD, [&]() -> llvm::Value *{
+            // Emit private VarDecl with copy init.
+            // Remap temp VDInit variable to the address of the original
+            // variable
+            // (for proper handling of captured global variables).
+            LocalDeclMap[VDInit] = OriginalAddr;
+            EmitDecl(*VD);
+            LocalDeclMap.erase(VDInit);
+            return GetAddrOfLocalVar(VD);
+          });
+        }
+        assert(IsRegistered &&
+               "firstprivate var already registered as private");
+        // Silence the warning about unused variable.
+        (void)IsRegistered;
+      }
       ++IRef, ++InitsRef;
     }
   }
+  return !EmittedAsFirstprivate.empty();
 }
 
 void CodeGenFunction::EmitOMPPrivateClause(
     const OMPExecutableDirective &D,
     CodeGenFunction::OMPPrivateScope &PrivateScope) {
-  auto PrivateFilter = [](const OMPClause *C) -> bool {
-    return C->getClauseKind() == OMPC_private;
-  };
-  for (OMPExecutableDirective::filtered_clause_iterator<decltype(PrivateFilter)>
-           I(D.clauses(), PrivateFilter); I; ++I) {
+  llvm::DenseSet<const VarDecl *> EmittedAsPrivate;
+  for (auto &&I = D.getClausesOfKind(OMPC_private); I; ++I) {
     auto *C = cast<OMPPrivateClause>(*I);
     auto IRef = C->varlist_begin();
     for (auto IInit : C->private_copies()) {
       auto *OrigVD = cast<VarDecl>(cast<DeclRefExpr>(*IRef)->getDecl());
-      auto VD = cast<VarDecl>(cast<DeclRefExpr>(IInit)->getDecl());
+      if (EmittedAsPrivate.insert(OrigVD->getCanonicalDecl()).second) {
+        auto VD = cast<VarDecl>(cast<DeclRefExpr>(IInit)->getDecl());
+        bool IsRegistered =
+            PrivateScope.addPrivate(OrigVD, [&]() -> llvm::Value *{
+              // Emit private VarDecl with copy init.
+              EmitDecl(*VD);
+              return GetAddrOfLocalVar(VD);
+            });
+        assert(IsRegistered && "private var already registered as private");
+        // Silence the warning about unused variable.
+        (void)IsRegistered;
+      }
+      ++IRef;
+    }
+  }
+}
+
+bool CodeGenFunction::EmitOMPCopyinClause(const OMPExecutableDirective &D) {
+  // threadprivate_var1 = master_threadprivate_var1;
+  // operator=(threadprivate_var2, master_threadprivate_var2);
+  // ...
+  // __kmpc_barrier(&loc, global_tid);
+  llvm::DenseSet<const VarDecl *> CopiedVars;
+  llvm::BasicBlock *CopyBegin = nullptr, *CopyEnd = nullptr;
+  for (auto &&I = D.getClausesOfKind(OMPC_copyin); I; ++I) {
+    auto *C = cast<OMPCopyinClause>(*I);
+    auto IRef = C->varlist_begin();
+    auto ISrcRef = C->source_exprs().begin();
+    auto IDestRef = C->destination_exprs().begin();
+    for (auto *AssignOp : C->assignment_ops()) {
+      auto *VD = cast<VarDecl>(cast<DeclRefExpr>(*IRef)->getDecl());
+      QualType Type = VD->getType();
+      if (CopiedVars.insert(VD->getCanonicalDecl()).second) {
+        // Get the address of the master variable.
+        auto *MasterAddr = VD->isStaticLocal()
+                               ? CGM.getStaticLocalDeclAddress(VD)
+                               : CGM.GetAddrOfGlobal(VD);
+        // Get the address of the threadprivate variable.
+        auto *PrivateAddr = EmitLValue(*IRef).getAddress();
+        if (CopiedVars.size() == 1) {
+          // At first check if current thread is a master thread. If it is, no
+          // need to copy data.
+          CopyBegin = createBasicBlock("copyin.not.master");
+          CopyEnd = createBasicBlock("copyin.not.master.end");
+          Builder.CreateCondBr(
+              Builder.CreateICmpNE(
+                  Builder.CreatePtrToInt(MasterAddr, CGM.IntPtrTy),
+                  Builder.CreatePtrToInt(PrivateAddr, CGM.IntPtrTy)),
+              CopyBegin, CopyEnd);
+          EmitBlock(CopyBegin);
+        }
+        auto *SrcVD = cast<VarDecl>(cast<DeclRefExpr>(*ISrcRef)->getDecl());
+        auto *DestVD = cast<VarDecl>(cast<DeclRefExpr>(*IDestRef)->getDecl());
+        EmitOMPCopy(*this, Type, PrivateAddr, MasterAddr, DestVD, SrcVD,
+                    AssignOp);
+      }
+      ++IRef;
+      ++ISrcRef;
+      ++IDestRef;
+    }
+  }
+  if (CopyEnd) {
+    // Exit out of copying procedure for non-master thread.
+    EmitBlock(CopyEnd, /*IsFinished=*/true);
+    return true;
+  }
+  return false;
+}
+
+bool CodeGenFunction::EmitOMPLastprivateClauseInit(
+    const OMPExecutableDirective &D, OMPPrivateScope &PrivateScope) {
+  bool HasAtLeastOneLastprivate = false;
+  llvm::DenseSet<const VarDecl *> AlreadyEmittedVars;
+  for (auto &&I = D.getClausesOfKind(OMPC_lastprivate); I; ++I) {
+    HasAtLeastOneLastprivate = true;
+    auto *C = cast<OMPLastprivateClause>(*I);
+    auto IRef = C->varlist_begin();
+    auto IDestRef = C->destination_exprs().begin();
+    for (auto *IInit : C->private_copies()) {
+      // Keep the address of the original variable for future update at the end
+      // of the loop.
+      auto *OrigVD = cast<VarDecl>(cast<DeclRefExpr>(*IRef)->getDecl());
+      if (AlreadyEmittedVars.insert(OrigVD->getCanonicalDecl()).second) {
+        auto *DestVD = cast<VarDecl>(cast<DeclRefExpr>(*IDestRef)->getDecl());
+        PrivateScope.addPrivate(DestVD, [this, OrigVD, IRef]() -> llvm::Value *{
+          DeclRefExpr DRE(
+              const_cast<VarDecl *>(OrigVD),
+              /*RefersToEnclosingVariableOrCapture=*/CapturedStmtInfo->lookup(
+                  OrigVD) != nullptr,
+              (*IRef)->getType(), VK_LValue, (*IRef)->getExprLoc());
+          return EmitLValue(&DRE).getAddress();
+        });
+        // Check if the variable is also a firstprivate: in this case IInit is
+        // not generated. Initialization of this variable will happen in codegen
+        // for 'firstprivate' clause.
+        if (IInit) {
+          auto *VD = cast<VarDecl>(cast<DeclRefExpr>(IInit)->getDecl());
+          bool IsRegistered =
+              PrivateScope.addPrivate(OrigVD, [&]() -> llvm::Value *{
+                // Emit private VarDecl with copy init.
+                EmitDecl(*VD);
+                return GetAddrOfLocalVar(VD);
+              });
+          assert(IsRegistered &&
+                 "lastprivate var already registered as private");
+          (void)IsRegistered;
+        }
+      }
+      ++IRef, ++IDestRef;
+    }
+  }
+  return HasAtLeastOneLastprivate;
+}
+
+void CodeGenFunction::EmitOMPLastprivateClauseFinal(
+    const OMPExecutableDirective &D, llvm::Value *IsLastIterCond) {
+  // Emit following code:
+  // if (<IsLastIterCond>) {
+  //   orig_var1 = private_orig_var1;
+  //   ...
+  //   orig_varn = private_orig_varn;
+  // }
+  auto *ThenBB = createBasicBlock(".omp.lastprivate.then");
+  auto *DoneBB = createBasicBlock(".omp.lastprivate.done");
+  Builder.CreateCondBr(IsLastIterCond, ThenBB, DoneBB);
+  EmitBlock(ThenBB);
+  llvm::DenseMap<const Decl *, const Expr *> LoopCountersAndUpdates;
+  const Expr *LastIterVal = nullptr;
+  const Expr *IVExpr = nullptr;
+  const Expr *IncExpr = nullptr;
+  if (auto *LoopDirective = dyn_cast<OMPLoopDirective>(&D)) {
+    LastIterVal =
+        cast<VarDecl>(cast<DeclRefExpr>(LoopDirective->getUpperBoundVariable())
+                          ->getDecl())
+            ->getAnyInitializer();
+    IVExpr = LoopDirective->getIterationVariable();
+    IncExpr = LoopDirective->getInc();
+    auto IUpdate = LoopDirective->updates().begin();
+    for (auto *E : LoopDirective->counters()) {
+      auto *D = cast<DeclRefExpr>(E)->getDecl()->getCanonicalDecl();
+      LoopCountersAndUpdates[D] = *IUpdate;
+      ++IUpdate;
+    }
+  }
+  {
+    llvm::DenseSet<const VarDecl *> AlreadyEmittedVars;
+    bool FirstLCV = true;
+    for (auto &&I = D.getClausesOfKind(OMPC_lastprivate); I; ++I) {
+      auto *C = cast<OMPLastprivateClause>(*I);
+      auto IRef = C->varlist_begin();
+      auto ISrcRef = C->source_exprs().begin();
+      auto IDestRef = C->destination_exprs().begin();
+      for (auto *AssignOp : C->assignment_ops()) {
+        auto *PrivateVD = cast<VarDecl>(cast<DeclRefExpr>(*IRef)->getDecl());
+        QualType Type = PrivateVD->getType();
+        auto *CanonicalVD = PrivateVD->getCanonicalDecl();
+        if (AlreadyEmittedVars.insert(CanonicalVD).second) {
+          // If lastprivate variable is a loop control variable for loop-based
+          // directive, update its value before copyin back to original
+          // variable.
+          if (auto *UpExpr = LoopCountersAndUpdates.lookup(CanonicalVD)) {
+            if (FirstLCV) {
+              EmitAnyExprToMem(LastIterVal, EmitLValue(IVExpr).getAddress(),
+                               IVExpr->getType().getQualifiers(),
+                               /*IsInitializer=*/false);
+              EmitIgnoredExpr(IncExpr);
+              FirstLCV = false;
+            }
+            EmitIgnoredExpr(UpExpr);
+          }
+          auto *SrcVD = cast<VarDecl>(cast<DeclRefExpr>(*ISrcRef)->getDecl());
+          auto *DestVD = cast<VarDecl>(cast<DeclRefExpr>(*IDestRef)->getDecl());
+          // Get the address of the original variable.
+          auto *OriginalAddr = GetAddrOfLocalVar(DestVD);
+          // Get the address of the private variable.
+          auto *PrivateAddr = GetAddrOfLocalVar(PrivateVD);
+          EmitOMPCopy(*this, Type, OriginalAddr, PrivateAddr, DestVD, SrcVD,
+                      AssignOp);
+        }
+        ++IRef;
+        ++ISrcRef;
+        ++IDestRef;
+      }
+    }
+  }
+  EmitBlock(DoneBB, /*IsFinished=*/true);
+}
+
+void CodeGenFunction::EmitOMPReductionClauseInit(
+    const OMPExecutableDirective &D,
+    CodeGenFunction::OMPPrivateScope &PrivateScope) {
+  for (auto &&I = D.getClausesOfKind(OMPC_reduction); I; ++I) {
+    auto *C = cast<OMPReductionClause>(*I);
+    auto ILHS = C->lhs_exprs().begin();
+    auto IRHS = C->rhs_exprs().begin();
+    for (auto IRef : C->varlists()) {
+      auto *OrigVD = cast<VarDecl>(cast<DeclRefExpr>(IRef)->getDecl());
+      auto *LHSVD = cast<VarDecl>(cast<DeclRefExpr>(*ILHS)->getDecl());
+      auto *PrivateVD = cast<VarDecl>(cast<DeclRefExpr>(*IRHS)->getDecl());
+      // Store the address of the original variable associated with the LHS
+      // implicit variable.
+      PrivateScope.addPrivate(LHSVD, [this, OrigVD, IRef]() -> llvm::Value *{
+        DeclRefExpr DRE(const_cast<VarDecl *>(OrigVD),
+                        CapturedStmtInfo->lookup(OrigVD) != nullptr,
+                        IRef->getType(), VK_LValue, IRef->getExprLoc());
+        return EmitLValue(&DRE).getAddress();
+      });
+      // Emit reduction copy.
       bool IsRegistered =
-          PrivateScope.addPrivate(OrigVD, [&]() -> llvm::Value * {
-            // Emit private VarDecl with copy init.
-            EmitDecl(*VD);
-            return GetAddrOfLocalVar(VD);
+          PrivateScope.addPrivate(OrigVD, [this, PrivateVD]() -> llvm::Value *{
+            // Emit private VarDecl with reduction init.
+            EmitDecl(*PrivateVD);
+            return GetAddrOfLocalVar(PrivateVD);
           });
-      assert(IsRegistered && "counter already registered as private");
+      assert(IsRegistered && "private var already registered as private");
       // Silence the warning about unused variable.
       (void)IsRegistered;
-      ++IRef;
+      ++ILHS, ++IRHS;
     }
   }
 }
 
-/// \brief Emits code for OpenMP parallel directive in the parallel region.
-static void EmitOMPParallelCall(CodeGenFunction &CGF,
-                                const OMPParallelDirective &S,
-                                llvm::Value *OutlinedFn,
-                                llvm::Value *CapturedStruct) {
-  if (auto C = S.getSingleClause(/*K*/ OMPC_num_threads)) {
+void CodeGenFunction::EmitOMPReductionClauseFinal(
+    const OMPExecutableDirective &D) {
+  llvm::SmallVector<const Expr *, 8> LHSExprs;
+  llvm::SmallVector<const Expr *, 8> RHSExprs;
+  llvm::SmallVector<const Expr *, 8> ReductionOps;
+  bool HasAtLeastOneReduction = false;
+  for (auto &&I = D.getClausesOfKind(OMPC_reduction); I; ++I) {
+    HasAtLeastOneReduction = true;
+    auto *C = cast<OMPReductionClause>(*I);
+    LHSExprs.append(C->lhs_exprs().begin(), C->lhs_exprs().end());
+    RHSExprs.append(C->rhs_exprs().begin(), C->rhs_exprs().end());
+    ReductionOps.append(C->reduction_ops().begin(), C->reduction_ops().end());
+  }
+  if (HasAtLeastOneReduction) {
+    // Emit nowait reduction if nowait clause is present or directive is a
+    // parallel directive (it always has implicit barrier).
+    CGM.getOpenMPRuntime().emitReduction(
+        *this, D.getLocEnd(), LHSExprs, RHSExprs, ReductionOps,
+        D.getSingleClause(OMPC_nowait) ||
+            isOpenMPParallelDirective(D.getDirectiveKind()));
+  }
+}
+
+static void emitCommonOMPParallelDirective(CodeGenFunction &CGF,
+                                           const OMPExecutableDirective &S,
+                                           const RegionCodeGenTy &CodeGen) {
+  auto CS = cast<CapturedStmt>(S.getAssociatedStmt());
+  auto CapturedStruct = CGF.GenerateCapturedStmtArgument(*CS);
+  auto OutlinedFn = CGF.CGM.getOpenMPRuntime().emitParallelOutlinedFunction(
+      S, *CS->getCapturedDecl()->param_begin(), CodeGen);
+  if (auto C = S.getSingleClause(OMPC_num_threads)) {
     CodeGenFunction::RunCleanupsScope NumThreadsScope(CGF);
     auto NumThreadsClause = cast<OMPNumThreadsClause>(C);
     auto NumThreads = CGF.EmitScalarExpr(NumThreadsClause->getNumThreads(),
                                          /*IgnoreResultAssign*/ true);
-    CGF.CGM.getOpenMPRuntime().EmitOMPNumThreadsClause(
+    CGF.CGM.getOpenMPRuntime().emitNumThreadsClause(
         CGF, NumThreads, NumThreadsClause->getLocStart());
   }
-  CGF.CGM.getOpenMPRuntime().EmitOMPParallelCall(CGF, S.getLocStart(),
-                                                 OutlinedFn, CapturedStruct);
+  const Expr *IfCond = nullptr;
+  if (auto C = S.getSingleClause(OMPC_if)) {
+    IfCond = cast<OMPIfClause>(C)->getCondition();
+  }
+  CGF.CGM.getOpenMPRuntime().emitParallelCall(CGF, S.getLocStart(), OutlinedFn,
+                                              CapturedStruct, IfCond);
 }
 
 void CodeGenFunction::EmitOMPParallelDirective(const OMPParallelDirective &S) {
-  auto CS = cast<CapturedStmt>(S.getAssociatedStmt());
-  auto CapturedStruct = GenerateCapturedStmtArgument(*CS);
-  auto OutlinedFn = CGM.getOpenMPRuntime().EmitOpenMPOutlinedFunction(
-      S, *CS->getCapturedDecl()->param_begin());
-  if (auto C = S.getSingleClause(/*K*/ OMPC_if)) {
-    auto Cond = cast<OMPIfClause>(C)->getCondition();
-    EmitOMPIfClause(*this, Cond, [&](bool ThenBlock) {
-      if (ThenBlock)
-        EmitOMPParallelCall(*this, S, OutlinedFn, CapturedStruct);
-      else
-        CGM.getOpenMPRuntime().EmitOMPSerialCall(*this, S.getLocStart(),
-                                                 OutlinedFn, CapturedStruct);
-    });
-  } else
-    EmitOMPParallelCall(*this, S, OutlinedFn, CapturedStruct);
+  LexicalScope Scope(*this, S.getSourceRange());
+  // Emit parallel region as a standalone region.
+  auto &&CodeGen = [&S](CodeGenFunction &CGF) {
+    OMPPrivateScope PrivateScope(CGF);
+    bool Copyins = CGF.EmitOMPCopyinClause(S);
+    bool Firstprivates = CGF.EmitOMPFirstprivateClause(S, PrivateScope);
+    if (Copyins || Firstprivates) {
+      // Emit implicit barrier to synchronize threads and avoid data races on
+      // initialization of firstprivate variables or propagation master's thread
+      // values of threadprivate variables to local instances of that variables
+      // of all other implicit threads.
+      CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
+                                                 OMPD_unknown);
+    }
+    CGF.EmitOMPPrivateClause(S, PrivateScope);
+    CGF.EmitOMPReductionClauseInit(S, PrivateScope);
+    (void)PrivateScope.Privatize();
+    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EmitOMPReductionClauseFinal(S);
+    // Emit implicit barrier at the end of the 'parallel' directive.
+    CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
+                                               OMPD_unknown);
+  };
+  emitCommonOMPParallelDirective(*this, S, CodeGen);
 }
 
 void CodeGenFunction::EmitOMPLoopBody(const OMPLoopDirective &S,
@@ -281,6 +496,14 @@ void CodeGenFunction::EmitOMPLoopBody(const OMPLoopDirective &S,
   for (auto I : S.updates()) {
     EmitIgnoredExpr(I);
   }
+  // Update the linear variables.
+  for (auto &&I = S.getClausesOfKind(OMPC_linear); I; ++I) {
+    auto *C = cast<OMPLinearClause>(*I);
+    for (auto U : C->updates()) {
+      EmitIgnoredExpr(U);
+    }
+  }
+
   // On a continue in the body, jump to the end.
   auto Continue = getJumpDestInCurrentScope("omp.body.continue");
   BreakContinueStack.push_back(BreakContinue(JumpDest(), Continue));
@@ -297,11 +520,12 @@ void CodeGenFunction::EmitOMPLoopBody(const OMPLoopDirective &S,
   }
 }
 
-void CodeGenFunction::EmitOMPInnerLoop(const OMPLoopDirective &S,
-                                       OMPPrivateScope &LoopScope,
-                                       bool SeparateIter) {
+void CodeGenFunction::EmitOMPInnerLoop(
+    const Stmt &S, bool RequiresCleanup, const Expr *LoopCond,
+    const Expr *IncExpr,
+    const llvm::function_ref<void(CodeGenFunction &)> &BodyGen,
+    const llvm::function_ref<void(CodeGenFunction &)> &PostIncGen) {
   auto LoopExit = getJumpDestInCurrentScope("omp.inner.for.end");
-  auto Cnt = getPGORegionCounter(&S);
 
   // Start the loop with a block that tests the condition.
   auto CondBlock = createBasicBlock("omp.inner.for.cond");
@@ -311,35 +535,31 @@ void CodeGenFunction::EmitOMPInnerLoop(const OMPLoopDirective &S,
   // If there are any cleanups between here and the loop-exit scope,
   // create a block to stage a loop exit along.
   auto ExitBlock = LoopExit.getBlock();
-  if (LoopScope.requiresCleanups())
+  if (RequiresCleanup)
     ExitBlock = createBasicBlock("omp.inner.for.cond.cleanup");
 
   auto LoopBody = createBasicBlock("omp.inner.for.body");
 
-  // Emit condition: "IV < LastIteration + 1 [ - 1]"
-  // ("- 1" when lastprivate clause is present - separate one iteration).
-  llvm::Value *BoolCondVal = EvaluateExprAsBool(S.getCond(SeparateIter));
-  Builder.CreateCondBr(BoolCondVal, LoopBody, ExitBlock,
-                       PGO.createLoopWeights(S.getCond(SeparateIter), Cnt));
-
+  // Emit condition.
+  EmitBranchOnBoolExpr(LoopCond, LoopBody, ExitBlock, getProfileCount(&S));
   if (ExitBlock != LoopExit.getBlock()) {
     EmitBlock(ExitBlock);
     EmitBranchThroughCleanup(LoopExit);
   }
 
   EmitBlock(LoopBody);
-  Cnt.beginRegion(Builder);
+  incrementProfileCounter(&S);
 
   // Create a block for the increment.
   auto Continue = getJumpDestInCurrentScope("omp.inner.for.inc");
   BreakContinueStack.push_back(BreakContinue(LoopExit, Continue));
 
-  EmitOMPLoopBody(S);
-  EmitStopPoint(&S);
+  BodyGen(*this);
 
   // Emit "IV = IV + 1" and a back-edge to the condition block.
   EmitBlock(Continue.getBlock());
-  EmitIgnoredExpr(S.getInc());
+  EmitIgnoredExpr(IncExpr);
+  PostIncGen(*this);
   BreakContinueStack.pop_back();
   EmitBranch(CondBlock);
   LoopStack.pop();
@@ -350,11 +570,38 @@ void CodeGenFunction::EmitOMPInnerLoop(const OMPLoopDirective &S,
 void CodeGenFunction::EmitOMPSimdFinal(const OMPLoopDirective &S) {
   auto IC = S.counters().begin();
   for (auto F : S.finals()) {
-    if (LocalDeclMap.lookup(cast<DeclRefExpr>((*IC))->getDecl())) {
+    auto *OrigVD = cast<VarDecl>(cast<DeclRefExpr>((*IC))->getDecl());
+    if (LocalDeclMap.lookup(OrigVD)) {
+      DeclRefExpr DRE(const_cast<VarDecl *>(OrigVD),
+                      CapturedStmtInfo->lookup(OrigVD) != nullptr,
+                      (*IC)->getType(), VK_LValue, (*IC)->getExprLoc());
+      auto *OrigAddr = EmitLValue(&DRE).getAddress();
+      OMPPrivateScope VarScope(*this);
+      VarScope.addPrivate(OrigVD,
+                          [OrigAddr]() -> llvm::Value *{ return OrigAddr; });
+      (void)VarScope.Privatize();
       EmitIgnoredExpr(F);
     }
     ++IC;
   }
+  // Emit the final values of the linear variables.
+  for (auto &&I = S.getClausesOfKind(OMPC_linear); I; ++I) {
+    auto *C = cast<OMPLinearClause>(*I);
+    auto IC = C->varlist_begin();
+    for (auto F : C->finals()) {
+      auto *OrigVD = cast<VarDecl>(cast<DeclRefExpr>(*IC)->getDecl());
+      DeclRefExpr DRE(const_cast<VarDecl *>(OrigVD),
+                      CapturedStmtInfo->lookup(OrigVD) != nullptr,
+                      (*IC)->getType(), VK_LValue, (*IC)->getExprLoc());
+      auto *OrigAddr = EmitLValue(&DRE).getAddress();
+      OMPPrivateScope VarScope(*this);
+      VarScope.addPrivate(OrigVD,
+                          [OrigAddr]() -> llvm::Value *{ return OrigAddr; });
+      (void)VarScope.Privatize();
+      EmitIgnoredExpr(F);
+      ++IC;
+    }
+  }
 }
 
 static void EmitOMPAlignedClause(CodeGenFunction &CGF, CodeGenModule &CGM,
@@ -388,116 +635,348 @@ static void EmitPrivateLoopCounters(CodeGenFunction &CGF,
                                     ArrayRef<Expr *> Counters) {
   for (auto *E : Counters) {
     auto VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
-    bool IsRegistered = LoopScope.addPrivate(VD, [&]() -> llvm::Value * {
+    (void)LoopScope.addPrivate(VD, [&]() -> llvm::Value *{
       // Emit var without initialization.
       auto VarEmission = CGF.EmitAutoVarAlloca(*VD);
       CGF.EmitAutoVarCleanups(VarEmission);
       return VarEmission.getAllocatedAddress();
     });
-    assert(IsRegistered && "counter already registered as private");
-    // Silence the warning about unused variable.
-    (void)IsRegistered;
   }
-  (void)LoopScope.Privatize();
+}
+
+static void emitPreCond(CodeGenFunction &CGF, const OMPLoopDirective &S,
+                        const Expr *Cond, llvm::BasicBlock *TrueBlock,
+                        llvm::BasicBlock *FalseBlock, uint64_t TrueCount) {
+  CodeGenFunction::OMPPrivateScope PreCondScope(CGF);
+  EmitPrivateLoopCounters(CGF, PreCondScope, S.counters());
+  const VarDecl *IVDecl =
+      cast<VarDecl>(cast<DeclRefExpr>(S.getIterationVariable())->getDecl());
+  bool IsRegistered = PreCondScope.addPrivate(IVDecl, [&]() -> llvm::Value *{
+    // Emit var without initialization.
+    auto VarEmission = CGF.EmitAutoVarAlloca(*IVDecl);
+    CGF.EmitAutoVarCleanups(VarEmission);
+    return VarEmission.getAllocatedAddress();
+  });
+  assert(IsRegistered && "counter already registered as private");
+  // Silence the warning about unused variable.
+  (void)IsRegistered;
+  (void)PreCondScope.Privatize();
+  // Initialize internal counter to 0 to calculate initial values of real
+  // counters.
+  LValue IV = CGF.EmitLValue(S.getIterationVariable());
+  CGF.EmitStoreOfScalar(
+      llvm::ConstantInt::getNullValue(
+          IV.getAddress()->getType()->getPointerElementType()),
+      CGF.EmitLValue(S.getIterationVariable()), /*isInit=*/true);
+  // Get initial values of real counters.
+  for (auto I : S.updates()) {
+    CGF.EmitIgnoredExpr(I);
+  }
+  // Check that loop is executed at least one time.
+  CGF.EmitBranchOnBoolExpr(Cond, TrueBlock, FalseBlock, TrueCount);
+}
+
+static void
+EmitPrivateLinearVars(CodeGenFunction &CGF, const OMPExecutableDirective &D,
+                      CodeGenFunction::OMPPrivateScope &PrivateScope) {
+  for (auto &&I = D.getClausesOfKind(OMPC_linear); I; ++I) {
+    auto *C = cast<OMPLinearClause>(*I);
+    for (auto *E : C->varlists()) {
+      auto VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
+      bool IsRegistered = PrivateScope.addPrivate(VD, [&]()->llvm::Value * {
+        // Emit var without initialization.
+        auto VarEmission = CGF.EmitAutoVarAlloca(*VD);
+        CGF.EmitAutoVarCleanups(VarEmission);
+        return VarEmission.getAllocatedAddress();
+      });
+      assert(IsRegistered && "linear var already registered as private");
+      // Silence the warning about unused variable.
+      (void)IsRegistered;
+    }
+  }
 }
 
 void CodeGenFunction::EmitOMPSimdDirective(const OMPSimdDirective &S) {
-  // Pragma 'simd' code depends on presence of 'lastprivate'.
-  // If present, we have to separate last iteration of the loop:
+  auto &&CodeGen = [&S](CodeGenFunction &CGF) {
+    // Pragma 'simd' code depends on presence of 'lastprivate'.
+    // If present, we have to separate last iteration of the loop:
+    //
+    // if (PreCond) {
+    //   for (IV in 0..LastIteration-1) BODY;
+    //   BODY with updates of lastprivate vars;
+    //   <Final counter/linear vars updates>;
+    // }
+    //
+    // otherwise (when there's no lastprivate):
+    //
+    // if (PreCond) {
+    //   for (IV in 0..LastIteration) BODY;
+    //   <Final counter/linear vars updates>;
+    // }
+    //
+
+    // Emit: if (PreCond) - begin.
+    // If the condition constant folds and can be elided, avoid emitting the
+    // whole loop.
+    bool CondConstant;
+    llvm::BasicBlock *ContBlock = nullptr;
+    if (CGF.ConstantFoldsToSimpleInteger(S.getPreCond(), CondConstant)) {
+      if (!CondConstant)
+        return;
+    } else {
+      auto *ThenBlock = CGF.createBasicBlock("simd.if.then");
+      ContBlock = CGF.createBasicBlock("simd.if.end");
+      emitPreCond(CGF, S, S.getPreCond(), ThenBlock, ContBlock,
+                  CGF.getProfileCount(&S));
+      CGF.EmitBlock(ThenBlock);
+      CGF.incrementProfileCounter(&S);
+    }
+    // Walk clauses and process safelen/lastprivate.
+    bool SeparateIter = false;
+    CGF.LoopStack.setParallel();
+    CGF.LoopStack.setVectorizerEnable(true);
+    for (auto C : S.clauses()) {
+      switch (C->getClauseKind()) {
+      case OMPC_safelen: {
+        RValue Len = CGF.EmitAnyExpr(cast<OMPSafelenClause>(C)->getSafelen(),
+                                     AggValueSlot::ignored(), true);
+        llvm::ConstantInt *Val = cast<llvm::ConstantInt>(Len.getScalarVal());
+        CGF.LoopStack.setVectorizerWidth(Val->getZExtValue());
+        // In presence of finite 'safelen', it may be unsafe to mark all
+        // the memory instructions parallel, because loop-carried
+        // dependences of 'safelen' iterations are possible.
+        CGF.LoopStack.setParallel(false);
+        break;
+      }
+      case OMPC_aligned:
+        EmitOMPAlignedClause(CGF, CGF.CGM, cast<OMPAlignedClause>(*C));
+        break;
+      case OMPC_lastprivate:
+        SeparateIter = true;
+        break;
+      default:
+        // Not handled yet
+        ;
+      }
+    }
+
+    // Emit inits for the linear variables.
+    for (auto &&I = S.getClausesOfKind(OMPC_linear); I; ++I) {
+      auto *C = cast<OMPLinearClause>(*I);
+      for (auto Init : C->inits()) {
+        auto *D = cast<VarDecl>(cast<DeclRefExpr>(Init)->getDecl());
+        CGF.EmitVarDecl(*D);
+      }
+    }
+
+    // Emit the loop iteration variable.
+    const Expr *IVExpr = S.getIterationVariable();
+    const VarDecl *IVDecl = cast<VarDecl>(cast<DeclRefExpr>(IVExpr)->getDecl());
+    CGF.EmitVarDecl(*IVDecl);
+    CGF.EmitIgnoredExpr(S.getInit());
+
+    // Emit the iterations count variable.
+    // If it is not a variable, Sema decided to calculate iterations count on
+    // each iteration (e.g., it is foldable into a constant).
+    if (auto LIExpr = dyn_cast<DeclRefExpr>(S.getLastIteration())) {
+      CGF.EmitVarDecl(*cast<VarDecl>(LIExpr->getDecl()));
+      // Emit calculation of the iterations count.
+      CGF.EmitIgnoredExpr(S.getCalcLastIteration());
+    }
+
+    // Emit the linear steps for the linear clauses.
+    // If a step is not constant, it is pre-calculated before the loop.
+    for (auto &&I = S.getClausesOfKind(OMPC_linear); I; ++I) {
+      auto *C = cast<OMPLinearClause>(*I);
+      if (auto CS = cast_or_null<BinaryOperator>(C->getCalcStep()))
+        if (auto SaveRef = cast<DeclRefExpr>(CS->getLHS())) {
+          CGF.EmitVarDecl(*cast<VarDecl>(SaveRef->getDecl()));
+          // Emit calculation of the linear step.
+          CGF.EmitIgnoredExpr(CS);
+        }
+    }
+
+    {
+      OMPPrivateScope LoopScope(CGF);
+      EmitPrivateLoopCounters(CGF, LoopScope, S.counters());
+      EmitPrivateLinearVars(CGF, S, LoopScope);
+      CGF.EmitOMPPrivateClause(S, LoopScope);
+      (void)LoopScope.Privatize();
+      CGF.EmitOMPInnerLoop(S, LoopScope.requiresCleanups(),
+                           S.getCond(SeparateIter), S.getInc(),
+                           [&S](CodeGenFunction &CGF) {
+                             CGF.EmitOMPLoopBody(S);
+                             CGF.EmitStopPoint(&S);
+                           },
+                           [](CodeGenFunction &) {});
+      if (SeparateIter) {
+        CGF.EmitOMPLoopBody(S, /*SeparateIter=*/true);
+      }
+    }
+    CGF.EmitOMPSimdFinal(S);
+    // Emit: if (PreCond) - end.
+    if (ContBlock) {
+      CGF.EmitBranch(ContBlock);
+      CGF.EmitBlock(ContBlock, true);
+    }
+  };
+  CGM.getOpenMPRuntime().emitInlinedDirective(*this, CodeGen);
+}
+
+void CodeGenFunction::EmitOMPForOuterLoop(OpenMPScheduleClauseKind ScheduleKind,
+                                          const OMPLoopDirective &S,
+                                          OMPPrivateScope &LoopScope,
+                                          bool Ordered, llvm::Value *LB,
+                                          llvm::Value *UB, llvm::Value *ST,
+                                          llvm::Value *IL, llvm::Value *Chunk) {
+  auto &RT = CGM.getOpenMPRuntime();
+
+  // Dynamic scheduling of the outer loop (dynamic, guided, auto, runtime).
+  const bool DynamicOrOrdered = Ordered || RT.isDynamic(ScheduleKind);
+
+  assert((Ordered ||
+          !RT.isStaticNonchunked(ScheduleKind, /*Chunked=*/Chunk != nullptr)) &&
+         "static non-chunked schedule does not need outer loop");
+
+  // Emit outer loop.
+  //
+  // OpenMP [2.7.1, Loop Construct, Description, table 2-1]
+  // When schedule(dynamic,chunk_size) is specified, the iterations are
+  // distributed to threads in the team in chunks as the threads request them.
+  // Each thread executes a chunk of iterations, then requests another chunk,
+  // until no chunks remain to be distributed. Each chunk contains chunk_size
+  // iterations, except for the last chunk to be distributed, which may have
+  // fewer iterations. When no chunk_size is specified, it defaults to 1.
   //
-  // if (LastIteration != 0) {
-  //   for (IV in 0..LastIteration-1) BODY;
-  //   BODY with updates of lastprivate vars;
-  //   <Final counter/linear vars updates>;
+  // When schedule(guided,chunk_size) is specified, the iterations are assigned
+  // to threads in the team in chunks as the executing threads request them.
+  // Each thread executes a chunk of iterations, then requests another chunk,
+  // until no chunks remain to be assigned. For a chunk_size of 1, the size of
+  // each chunk is proportional to the number of unassigned iterations divided
+  // by the number of threads in the team, decreasing to 1. For a chunk_size
+  // with value k (greater than 1), the size of each chunk is determined in the
+  // same way, with the restriction that the chunks do not contain fewer than k
+  // iterations (except for the last chunk to be assigned, which may have fewer
+  // than k iterations).
+  //
+  // When schedule(auto) is specified, the decision regarding scheduling is
+  // delegated to the compiler and/or runtime system. The programmer gives the
+  // implementation the freedom to choose any possible mapping of iterations to
+  // threads in the team.
+  //
+  // When schedule(runtime) is specified, the decision regarding scheduling is
+  // deferred until run time, and the schedule and chunk size are taken from the
+  // run-sched-var ICV. If the ICV is set to auto, the schedule is
+  // implementation defined
+  //
+  // while(__kmpc_dispatch_next(&LB, &UB)) {
+  //   idx = LB;
+  //   while (idx <= UB) { BODY; ++idx;
+  //   __kmpc_dispatch_fini_(4|8)[u](); // For ordered loops only.
+  //   } // inner loop
   // }
   //
-  // otherwise (when there's no lastprivate):
+  // OpenMP [2.7.1, Loop Construct, Description, table 2-1]
+  // When schedule(static, chunk_size) is specified, iterations are divided into
+  // chunks of size chunk_size, and the chunks are assigned to the threads in
+  // the team in a round-robin fashion in the order of the thread number.
   //
-  //   for (IV in 0..LastIteration) BODY;
-  //   <Final counter/linear vars updates>;
+  // while(UB = min(UB, GlobalUB), idx = LB, idx < UB) {
+  //   while (idx <= UB) { BODY; ++idx; } // inner loop
+  //   LB = LB + ST;
+  //   UB = UB + ST;
+  // }
   //
 
-  // Walk clauses and process safelen/lastprivate.
-  bool SeparateIter = false;
-  LoopStack.setParallel();
-  LoopStack.setVectorizerEnable(true);
-  for (auto C : S.clauses()) {
-    switch (C->getClauseKind()) {
-    case OMPC_safelen: {
-      RValue Len = EmitAnyExpr(cast<OMPSafelenClause>(C)->getSafelen(),
-                               AggValueSlot::ignored(), true);
-      llvm::ConstantInt *Val = cast<llvm::ConstantInt>(Len.getScalarVal());
-      LoopStack.setVectorizerWidth(Val->getZExtValue());
-      // In presence of finite 'safelen', it may be unsafe to mark all
-      // the memory instructions parallel, because loop-carried
-      // dependences of 'safelen' iterations are possible.
-      LoopStack.setParallel(false);
-      break;
-    }
-    case OMPC_aligned:
-      EmitOMPAlignedClause(*this, CGM, cast<OMPAlignedClause>(*C));
-      break;
-    case OMPC_lastprivate:
-      SeparateIter = true;
-      break;
-    default:
-      // Not handled yet
-      ;
-    }
-  }
+  const Expr *IVExpr = S.getIterationVariable();
+  const unsigned IVSize = getContext().getTypeSize(IVExpr->getType());
+  const bool IVSigned = IVExpr->getType()->hasSignedIntegerRepresentation();
 
-  InlinedOpenMPRegion Region(*this, S.getAssociatedStmt());
-  RunCleanupsScope DirectiveScope(*this);
+  RT.emitForInit(
+      *this, S.getLocStart(), ScheduleKind, IVSize, IVSigned, Ordered, IL, LB,
+      (DynamicOrOrdered ? EmitAnyExpr(S.getLastIteration()).getScalarVal()
+                        : UB),
+      ST, Chunk);
 
-  CGDebugInfo *DI = getDebugInfo();
-  if (DI)
-    DI->EmitLexicalBlockStart(Builder, S.getSourceRange().getBegin());
+  auto LoopExit = getJumpDestInCurrentScope("omp.dispatch.end");
 
-  // Emit the loop iteration variable.
-  const Expr *IVExpr = S.getIterationVariable();
-  const VarDecl *IVDecl = cast<VarDecl>(cast<DeclRefExpr>(IVExpr)->getDecl());
-  EmitVarDecl(*IVDecl);
-  EmitIgnoredExpr(S.getInit());
+  // Start the loop with a block that tests the condition.
+  auto CondBlock = createBasicBlock("omp.dispatch.cond");
+  EmitBlock(CondBlock);
+  LoopStack.push(CondBlock);
 
-  // Emit the iterations count variable.
-  // If it is not a variable, Sema decided to calculate iterations count on each
-  // iteration (e.g., it is foldable into a constant).
-  if (auto LIExpr = dyn_cast<DeclRefExpr>(S.getLastIteration())) {
-    EmitVarDecl(*cast<VarDecl>(LIExpr->getDecl()));
-    // Emit calculation of the iterations count.
-    EmitIgnoredExpr(S.getCalcLastIteration());
+  llvm::Value *BoolCondVal = nullptr;
+  if (!DynamicOrOrdered) {
+    // UB = min(UB, GlobalUB)
+    EmitIgnoredExpr(S.getEnsureUpperBound());
+    // IV = LB
+    EmitIgnoredExpr(S.getInit());
+    // IV < UB
+    BoolCondVal = EvaluateExprAsBool(S.getCond(false));
+  } else {
+    BoolCondVal = RT.emitForNext(*this, S.getLocStart(), IVSize, IVSigned,
+                                    IL, LB, UB, ST);
   }
 
-  if (SeparateIter) {
-    // Emit: if (LastIteration > 0) - begin.
-    RegionCounter Cnt = getPGORegionCounter(&S);
-    auto ThenBlock = createBasicBlock("simd.if.then");
-    auto ContBlock = createBasicBlock("simd.if.end");
-    EmitBranchOnBoolExpr(S.getPreCond(), ThenBlock, ContBlock, Cnt.getCount());
-    EmitBlock(ThenBlock);
-    Cnt.beginRegion(Builder);
-    // Emit 'then' code.
-    {
-      OMPPrivateScope LoopScope(*this);
-      EmitPrivateLoopCounters(*this, LoopScope, S.counters());
-      EmitOMPInnerLoop(S, LoopScope, /* SeparateIter */ true);
-      EmitOMPLoopBody(S, /* SeparateIter */ true);
-    }
-    EmitOMPSimdFinal(S);
-    // Emit: if (LastIteration != 0) - end.
-    EmitBranch(ContBlock);
-    EmitBlock(ContBlock, true);
-  } else {
-    {
-      OMPPrivateScope LoopScope(*this);
-      EmitPrivateLoopCounters(*this, LoopScope, S.counters());
-      EmitOMPInnerLoop(S, LoopScope);
-    }
-    EmitOMPSimdFinal(S);
+  // If there are any cleanups between here and the loop-exit scope,
+  // create a block to stage a loop exit along.
+  auto ExitBlock = LoopExit.getBlock();
+  if (LoopScope.requiresCleanups())
+    ExitBlock = createBasicBlock("omp.dispatch.cleanup");
+
+  auto LoopBody = createBasicBlock("omp.dispatch.body");
+  Builder.CreateCondBr(BoolCondVal, LoopBody, ExitBlock);
+  if (ExitBlock != LoopExit.getBlock()) {
+    EmitBlock(ExitBlock);
+    EmitBranchThroughCleanup(LoopExit);
+  }
+  EmitBlock(LoopBody);
+
+  // Emit "IV = LB" (in case of static schedule, we have already calculated new
+  // LB for loop condition and emitted it above).
+  if (DynamicOrOrdered)
+    EmitIgnoredExpr(S.getInit());
+
+  // Create a block for the increment.
+  auto Continue = getJumpDestInCurrentScope("omp.dispatch.inc");
+  BreakContinueStack.push_back(BreakContinue(LoopExit, Continue));
+
+  SourceLocation Loc = S.getLocStart();
+  // Generate !llvm.loop.parallel metadata for loads and stores for loops with
+  // dynamic/guided scheduling and without ordered clause.
+  LoopStack.setParallel((ScheduleKind == OMPC_SCHEDULE_dynamic ||
+                         ScheduleKind == OMPC_SCHEDULE_guided) &&
+                        !Ordered);
+  EmitOMPInnerLoop(
+      S, LoopScope.requiresCleanups(), S.getCond(/*SeparateIter=*/false),
+      S.getInc(),
+      [&S](CodeGenFunction &CGF) {
+        CGF.EmitOMPLoopBody(S);
+        CGF.EmitStopPoint(&S);
+      },
+      [Ordered, IVSize, IVSigned, Loc](CodeGenFunction &CGF) {
+        if (Ordered) {
+          CGF.CGM.getOpenMPRuntime().emitForOrderedIterationEnd(
+              CGF, Loc, IVSize, IVSigned);
+        }
+      });
+
+  EmitBlock(Continue.getBlock());
+  BreakContinueStack.pop_back();
+  if (!DynamicOrOrdered) {
+    // Emit "LB = LB + Stride", "UB = UB + Stride".
+    EmitIgnoredExpr(S.getNextLowerBound());
+    EmitIgnoredExpr(S.getNextUpperBound());
   }
 
-  if (DI)
-    DI->EmitLexicalBlockEnd(Builder, S.getSourceRange().getEnd());
+  EmitBranch(CondBlock);
+  LoopStack.pop();
+  // Emit the fall-through block.
+  EmitBlock(LoopExit.getBlock());
+
+  // Tell the runtime we are done.
+  if (!DynamicOrOrdered)
+    RT.emitForStaticFinish(*this, S.getLocEnd());
 }
 
 /// \brief Emit a helper variable and return corresponding lvalue.
@@ -508,7 +987,39 @@ static LValue EmitOMPHelperVar(CodeGenFunction &CGF,
   return CGF.EmitLValue(Helper);
 }
 
-void CodeGenFunction::EmitOMPWorksharingLoop(const OMPLoopDirective &S) {
+static std::pair<llvm::Value * /*Chunk*/, OpenMPScheduleClauseKind>
+emitScheduleClause(CodeGenFunction &CGF, const OMPLoopDirective &S,
+                   bool OuterRegion) {
+  // Detect the loop schedule kind and chunk.
+  auto ScheduleKind = OMPC_SCHEDULE_unknown;
+  llvm::Value *Chunk = nullptr;
+  if (auto *C =
+          cast_or_null<OMPScheduleClause>(S.getSingleClause(OMPC_schedule))) {
+    ScheduleKind = C->getScheduleKind();
+    if (const auto *Ch = C->getChunkSize()) {
+      if (auto *ImpRef = cast_or_null<DeclRefExpr>(C->getHelperChunkSize())) {
+        if (OuterRegion) {
+          const VarDecl *ImpVar = cast<VarDecl>(ImpRef->getDecl());
+          CGF.EmitVarDecl(*ImpVar);
+          CGF.EmitStoreThroughLValue(
+              CGF.EmitAnyExpr(Ch),
+              CGF.MakeNaturalAlignAddrLValue(CGF.GetAddrOfLocalVar(ImpVar),
+                                             ImpVar->getType()));
+        } else {
+          Ch = ImpRef;
+        }
+      }
+      if (!C->getHelperChunkSize() || !OuterRegion) {
+        Chunk = CGF.EmitScalarExpr(Ch);
+        Chunk = CGF.EmitScalarConversion(Chunk, Ch->getType(),
+                                         S.getIterationVariable()->getType());
+      }
+    }
+  }
+  return std::make_pair(Chunk, ScheduleKind);
+}
+
+bool CodeGenFunction::EmitOMPWorksharingLoop(const OMPLoopDirective &S) {
   // Emit the loop iteration variable.
   auto IVExpr = cast<DeclRefExpr>(S.getIterationVariable());
   auto IVDecl = cast<VarDecl>(IVExpr->getDecl());
@@ -525,15 +1036,25 @@ void CodeGenFunction::EmitOMPWorksharingLoop(const OMPLoopDirective &S) {
 
   auto &RT = CGM.getOpenMPRuntime();
 
+  bool HasLastprivateClause;
   // Check pre-condition.
   {
     // Skip the entire loop if we don't meet the precondition.
-    RegionCounter Cnt = getPGORegionCounter(&S);
-    auto ThenBlock = createBasicBlock("omp.precond.then");
-    auto ContBlock = createBasicBlock("omp.precond.end");
-    EmitBranchOnBoolExpr(S.getPreCond(), ThenBlock, ContBlock, Cnt.getCount());
-    EmitBlock(ThenBlock);
-    Cnt.beginRegion(Builder);
+    // If the condition constant folds and can be elided, avoid emitting the
+    // whole loop.
+    bool CondConstant;
+    llvm::BasicBlock *ContBlock = nullptr;
+    if (ConstantFoldsToSimpleInteger(S.getPreCond(), CondConstant)) {
+      if (!CondConstant)
+        return false;
+    } else {
+      auto *ThenBlock = createBasicBlock("omp.precond.then");
+      ContBlock = createBasicBlock("omp.precond.end");
+      emitPreCond(*this, S, S.getPreCond(), ThenBlock, ContBlock,
+                  getProfileCount(&S));
+      EmitBlock(ThenBlock);
+      incrementProfileCounter(&S);
+    }
     // Emit 'then' code.
     {
       // Emit helper vars inits.
@@ -547,105 +1068,342 @@ void CodeGenFunction::EmitOMPWorksharingLoop(const OMPLoopDirective &S) {
           EmitOMPHelperVar(*this, cast<DeclRefExpr>(S.getIsLastIterVariable()));
 
       OMPPrivateScope LoopScope(*this);
+      if (EmitOMPFirstprivateClause(S, LoopScope)) {
+        // Emit implicit barrier to synchronize threads and avoid data races on
+        // initialization of firstprivate variables.
+        CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(),
+                                               OMPD_unknown);
+      }
+      EmitOMPPrivateClause(S, LoopScope);
+      HasLastprivateClause = EmitOMPLastprivateClauseInit(S, LoopScope);
+      EmitOMPReductionClauseInit(S, LoopScope);
       EmitPrivateLoopCounters(*this, LoopScope, S.counters());
+      (void)LoopScope.Privatize();
 
       // Detect the loop schedule kind and chunk.
-      auto ScheduleKind = OMPC_SCHEDULE_unknown;
-      llvm::Value *Chunk = nullptr;
-      if (auto C = cast_or_null<OMPScheduleClause>(
-              S.getSingleClause(OMPC_schedule))) {
-        ScheduleKind = C->getScheduleKind();
-        if (auto Ch = C->getChunkSize()) {
-          Chunk = EmitScalarExpr(Ch);
-          Chunk = EmitScalarConversion(Chunk, Ch->getType(),
-                                       S.getIterationVariable()->getType());
-        }
-      }
+      llvm::Value *Chunk;
+      OpenMPScheduleClauseKind ScheduleKind;
+      auto ScheduleInfo =
+          emitScheduleClause(*this, S, /*OuterRegion=*/false);
+      Chunk = ScheduleInfo.first;
+      ScheduleKind = ScheduleInfo.second;
       const unsigned IVSize = getContext().getTypeSize(IVExpr->getType());
       const bool IVSigned = IVExpr->getType()->hasSignedIntegerRepresentation();
+      const bool Ordered = S.getSingleClause(OMPC_ordered) != nullptr;
       if (RT.isStaticNonchunked(ScheduleKind,
-                                /* Chunked */ Chunk != nullptr)) {
+                                /* Chunked */ Chunk != nullptr) &&
+          !Ordered) {
         // OpenMP [2.7.1, Loop Construct, Description, table 2-1]
         // When no chunk_size is specified, the iteration space is divided into
         // chunks that are approximately equal in size, and at most one chunk is
         // distributed to each thread. Note that the size of the chunks is
         // unspecified in this case.
-        RT.EmitOMPForInit(*this, S.getLocStart(), ScheduleKind, IVSize, IVSigned,
-                          IL.getAddress(), LB.getAddress(), UB.getAddress(),
-                          ST.getAddress());
+        RT.emitForInit(*this, S.getLocStart(), ScheduleKind, IVSize, IVSigned,
+                       Ordered, IL.getAddress(), LB.getAddress(),
+                       UB.getAddress(), ST.getAddress());
         // UB = min(UB, GlobalUB);
         EmitIgnoredExpr(S.getEnsureUpperBound());
         // IV = LB;
         EmitIgnoredExpr(S.getInit());
         // while (idx <= UB) { BODY; ++idx; }
-        EmitOMPInnerLoop(S, LoopScope);
+        EmitOMPInnerLoop(S, LoopScope.requiresCleanups(),
+                         S.getCond(/*SeparateIter=*/false), S.getInc(),
+                         [&S](CodeGenFunction &CGF) {
+                           CGF.EmitOMPLoopBody(S);
+                           CGF.EmitStopPoint(&S);
+                         },
+                         [](CodeGenFunction &) {});
         // Tell the runtime we are done.
-        RT.EmitOMPForFinish(*this, S.getLocStart(), ScheduleKind);
-      } else
-        ErrorUnsupported(&S, "OpenMP loop with requested schedule");
+        RT.emitForStaticFinish(*this, S.getLocStart());
+      } else {
+        // Emit the outer loop, which requests its work chunk [LB..UB] from
+        // runtime and runs the inner loop to process it.
+        EmitOMPForOuterLoop(ScheduleKind, S, LoopScope, Ordered,
+                            LB.getAddress(), UB.getAddress(), ST.getAddress(),
+                            IL.getAddress(), Chunk);
+      }
+      EmitOMPReductionClauseFinal(S);
+      // Emit final copy of the lastprivate variables if IsLastIter != 0.
+      if (HasLastprivateClause)
+        EmitOMPLastprivateClauseFinal(
+            S, Builder.CreateIsNotNull(EmitLoadOfScalar(IL, S.getLocStart())));
     }
     // We're now done with the loop, so jump to the continuation block.
-    EmitBranch(ContBlock);
-    EmitBlock(ContBlock, true);
+    if (ContBlock) {
+      EmitBranch(ContBlock);
+      EmitBlock(ContBlock, true);
+    }
   }
+  return HasLastprivateClause;
 }
 
 void CodeGenFunction::EmitOMPForDirective(const OMPForDirective &S) {
-  InlinedOpenMPRegion Region(*this, S.getAssociatedStmt());
-  RunCleanupsScope DirectiveScope(*this);
-
-  CGDebugInfo *DI = getDebugInfo();
-  if (DI)
-    DI->EmitLexicalBlockStart(Builder, S.getSourceRange().getBegin());
-
-  EmitOMPWorksharingLoop(S);
+  LexicalScope Scope(*this, S.getSourceRange());
+  bool HasLastprivates = false;
+  auto &&CodeGen = [&S, &HasLastprivates](CodeGenFunction &CGF) {
+    HasLastprivates = CGF.EmitOMPWorksharingLoop(S);
+  };
+  CGM.getOpenMPRuntime().emitInlinedDirective(*this, CodeGen);
 
   // Emit an implicit barrier at the end.
-  CGM.getOpenMPRuntime().EmitOMPBarrierCall(*this, S.getLocStart(),
-                                            /*IsExplicit*/ false);
-  if (DI)
-    DI->EmitLexicalBlockEnd(Builder, S.getSourceRange().getEnd());
+  if (!S.getSingleClause(OMPC_nowait) || HasLastprivates) {
+    CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(), OMPD_for);
+  }
 }
 
 void CodeGenFunction::EmitOMPForSimdDirective(const OMPForSimdDirective &) {
   llvm_unreachable("CodeGen for 'omp for simd' is not supported yet.");
 }
 
-void CodeGenFunction::EmitOMPSectionsDirective(const OMPSectionsDirective &) {
-  llvm_unreachable("CodeGen for 'omp sections' is not supported yet.");
+static LValue createSectionLVal(CodeGenFunction &CGF, QualType Ty,
+                                const Twine &Name,
+                                llvm::Value *Init = nullptr) {
+  auto LVal = CGF.MakeNaturalAlignAddrLValue(CGF.CreateMemTemp(Ty, Name), Ty);
+  if (Init)
+    CGF.EmitScalarInit(Init, LVal);
+  return LVal;
 }
 
-void CodeGenFunction::EmitOMPSectionDirective(const OMPSectionDirective &) {
-  llvm_unreachable("CodeGen for 'omp section' is not supported yet.");
+static OpenMPDirectiveKind emitSections(CodeGenFunction &CGF,
+                                        const OMPExecutableDirective &S) {
+  auto *Stmt = cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt();
+  auto *CS = dyn_cast<CompoundStmt>(Stmt);
+  if (CS && CS->size() > 1) {
+    bool HasLastprivates = false;
+    auto &&CodeGen = [&S, CS, &HasLastprivates](CodeGenFunction &CGF) {
+      auto &C = CGF.CGM.getContext();
+      auto KmpInt32Ty = C.getIntTypeForBitwidth(/*DestWidth=*/32, /*Signed=*/1);
+      // Emit helper vars inits.
+      LValue LB = createSectionLVal(CGF, KmpInt32Ty, ".omp.sections.lb.",
+                                    CGF.Builder.getInt32(0));
+      auto *GlobalUBVal = CGF.Builder.getInt32(CS->size() - 1);
+      LValue UB =
+          createSectionLVal(CGF, KmpInt32Ty, ".omp.sections.ub.", GlobalUBVal);
+      LValue ST = createSectionLVal(CGF, KmpInt32Ty, ".omp.sections.st.",
+                                    CGF.Builder.getInt32(1));
+      LValue IL = createSectionLVal(CGF, KmpInt32Ty, ".omp.sections.il.",
+                                    CGF.Builder.getInt32(0));
+      // Loop counter.
+      LValue IV = createSectionLVal(CGF, KmpInt32Ty, ".omp.sections.iv.");
+      OpaqueValueExpr IVRefExpr(S.getLocStart(), KmpInt32Ty, VK_LValue);
+      CodeGenFunction::OpaqueValueMapping OpaqueIV(CGF, &IVRefExpr, IV);
+      OpaqueValueExpr UBRefExpr(S.getLocStart(), KmpInt32Ty, VK_LValue);
+      CodeGenFunction::OpaqueValueMapping OpaqueUB(CGF, &UBRefExpr, UB);
+      // Generate condition for loop.
+      BinaryOperator Cond(&IVRefExpr, &UBRefExpr, BO_LE, C.BoolTy, VK_RValue,
+                          OK_Ordinary, S.getLocStart(),
+                          /*fpContractable=*/false);
+      // Increment for loop counter.
+      UnaryOperator Inc(&IVRefExpr, UO_PreInc, KmpInt32Ty, VK_RValue,
+                        OK_Ordinary, S.getLocStart());
+      auto BodyGen = [CS, &S, &IV](CodeGenFunction &CGF) {
+        // Iterate through all sections and emit a switch construct:
+        // switch (IV) {
+        //   case 0:
+        //     <SectionStmt[0]>;
+        //     break;
+        // ...
+        //   case <NumSection> - 1:
+        //     <SectionStmt[<NumSection> - 1]>;
+        //     break;
+        // }
+        // .omp.sections.exit:
+        auto *ExitBB = CGF.createBasicBlock(".omp.sections.exit");
+        auto *SwitchStmt = CGF.Builder.CreateSwitch(
+            CGF.EmitLoadOfLValue(IV, S.getLocStart()).getScalarVal(), ExitBB,
+            CS->size());
+        unsigned CaseNumber = 0;
+        for (auto C = CS->children(); C; ++C, ++CaseNumber) {
+          auto CaseBB = CGF.createBasicBlock(".omp.sections.case");
+          CGF.EmitBlock(CaseBB);
+          SwitchStmt->addCase(CGF.Builder.getInt32(CaseNumber), CaseBB);
+          CGF.EmitStmt(*C);
+          CGF.EmitBranch(ExitBB);
+        }
+        CGF.EmitBlock(ExitBB, /*IsFinished=*/true);
+      };
+
+      CodeGenFunction::OMPPrivateScope LoopScope(CGF);
+      if (CGF.EmitOMPFirstprivateClause(S, LoopScope)) {
+        // Emit implicit barrier to synchronize threads and avoid data races on
+        // initialization of firstprivate variables.
+        CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
+                                                   OMPD_unknown);
+      }
+      CGF.EmitOMPPrivateClause(S, LoopScope);
+      HasLastprivates = CGF.EmitOMPLastprivateClauseInit(S, LoopScope);
+      CGF.EmitOMPReductionClauseInit(S, LoopScope);
+      (void)LoopScope.Privatize();
+
+      // Emit static non-chunked loop.
+      CGF.CGM.getOpenMPRuntime().emitForInit(
+          CGF, S.getLocStart(), OMPC_SCHEDULE_static, /*IVSize=*/32,
+          /*IVSigned=*/true, /*Ordered=*/false, IL.getAddress(),
+          LB.getAddress(), UB.getAddress(), ST.getAddress());
+      // UB = min(UB, GlobalUB);
+      auto *UBVal = CGF.EmitLoadOfScalar(UB, S.getLocStart());
+      auto *MinUBGlobalUB = CGF.Builder.CreateSelect(
+          CGF.Builder.CreateICmpSLT(UBVal, GlobalUBVal), UBVal, GlobalUBVal);
+      CGF.EmitStoreOfScalar(MinUBGlobalUB, UB);
+      // IV = LB;
+      CGF.EmitStoreOfScalar(CGF.EmitLoadOfScalar(LB, S.getLocStart()), IV);
+      // while (idx <= UB) { BODY; ++idx; }
+      CGF.EmitOMPInnerLoop(S, /*RequiresCleanup=*/false, &Cond, &Inc, BodyGen,
+                           [](CodeGenFunction &) {});
+      // Tell the runtime we are done.
+      CGF.CGM.getOpenMPRuntime().emitForStaticFinish(CGF, S.getLocStart());
+      CGF.EmitOMPReductionClauseFinal(S);
+
+      // Emit final copy of the lastprivate variables if IsLastIter != 0.
+      if (HasLastprivates)
+        CGF.EmitOMPLastprivateClauseFinal(
+            S, CGF.Builder.CreateIsNotNull(
+                   CGF.EmitLoadOfScalar(IL, S.getLocStart())));
+    };
+
+    CGF.CGM.getOpenMPRuntime().emitInlinedDirective(CGF, CodeGen);
+    // Emit barrier for lastprivates only if 'sections' directive has 'nowait'
+    // clause. Otherwise the barrier will be generated by the codegen for the
+    // directive.
+    if (HasLastprivates && S.getSingleClause(OMPC_nowait)) {
+      // Emit implicit barrier to synchronize threads and avoid data races on
+      // initialization of firstprivate variables.
+      CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
+                                                 OMPD_unknown);
+    }
+    return OMPD_sections;
+  }
+  // If only one section is found - no need to generate loop, emit as a single
+  // region.
+  bool HasFirstprivates;
+  // No need to generate reductions for sections with single section region, we
+  // can use original shared variables for all operations.
+  bool HasReductions = !S.getClausesOfKind(OMPC_reduction).empty();
+  // No need to generate lastprivates for sections with single section region,
+  // we can use original shared variable for all calculations with barrier at
+  // the end of the sections.
+  bool HasLastprivates = !S.getClausesOfKind(OMPC_lastprivate).empty();
+  auto &&CodeGen = [Stmt, &S, &HasFirstprivates](CodeGenFunction &CGF) {
+    CodeGenFunction::OMPPrivateScope SingleScope(CGF);
+    HasFirstprivates = CGF.EmitOMPFirstprivateClause(S, SingleScope);
+    CGF.EmitOMPPrivateClause(S, SingleScope);
+    (void)SingleScope.Privatize();
+
+    CGF.EmitStmt(Stmt);
+    CGF.EnsureInsertPoint();
+  };
+  CGF.CGM.getOpenMPRuntime().emitSingleRegion(CGF, CodeGen, S.getLocStart(),
+                                              llvm::None, llvm::None,
+                                              llvm::None, llvm::None);
+  // Emit barrier for firstprivates, lastprivates or reductions only if
+  // 'sections' directive has 'nowait' clause. Otherwise the barrier will be
+  // generated by the codegen for the directive.
+  if ((HasFirstprivates || HasLastprivates || HasReductions) &&
+      S.getSingleClause(OMPC_nowait)) {
+    // Emit implicit barrier to synchronize threads and avoid data races on
+    // initialization of firstprivate variables.
+    CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
+                                               OMPD_unknown);
+  }
+  return OMPD_single;
+}
+
+void CodeGenFunction::EmitOMPSectionsDirective(const OMPSectionsDirective &S) {
+  LexicalScope Scope(*this, S.getSourceRange());
+  OpenMPDirectiveKind EmittedAs = emitSections(*this, S);
+  // Emit an implicit barrier at the end.
+  if (!S.getSingleClause(OMPC_nowait)) {
+    CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(), EmittedAs);
+  }
 }
 
-void CodeGenFunction::EmitOMPSingleDirective(const OMPSingleDirective &) {
-  llvm_unreachable("CodeGen for 'omp single' is not supported yet.");
+void CodeGenFunction::EmitOMPSectionDirective(const OMPSectionDirective &S) {
+  LexicalScope Scope(*this, S.getSourceRange());
+  auto &&CodeGen = [&S](CodeGenFunction &CGF) {
+    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EnsureInsertPoint();
+  };
+  CGM.getOpenMPRuntime().emitInlinedDirective(*this, CodeGen);
+}
+
+void CodeGenFunction::EmitOMPSingleDirective(const OMPSingleDirective &S) {
+  llvm::SmallVector<const Expr *, 8> CopyprivateVars;
+  llvm::SmallVector<const Expr *, 8> DestExprs;
+  llvm::SmallVector<const Expr *, 8> SrcExprs;
+  llvm::SmallVector<const Expr *, 8> AssignmentOps;
+  // Check if there are any 'copyprivate' clauses associated with this
+  // 'single'
+  // construct.
+  // Build a list of copyprivate variables along with helper expressions
+  // (<source>, <destination>, <destination>=<source> expressions)
+  for (auto &&I = S.getClausesOfKind(OMPC_copyprivate); I; ++I) {
+    auto *C = cast<OMPCopyprivateClause>(*I);
+    CopyprivateVars.append(C->varlists().begin(), C->varlists().end());
+    DestExprs.append(C->destination_exprs().begin(),
+                     C->destination_exprs().end());
+    SrcExprs.append(C->source_exprs().begin(), C->source_exprs().end());
+    AssignmentOps.append(C->assignment_ops().begin(),
+                         C->assignment_ops().end());
+  }
+  LexicalScope Scope(*this, S.getSourceRange());
+  // Emit code for 'single' region along with 'copyprivate' clauses
+  bool HasFirstprivates;
+  auto &&CodeGen = [&S, &HasFirstprivates](CodeGenFunction &CGF) {
+    CodeGenFunction::OMPPrivateScope SingleScope(CGF);
+    HasFirstprivates = CGF.EmitOMPFirstprivateClause(S, SingleScope);
+    CGF.EmitOMPPrivateClause(S, SingleScope);
+    (void)SingleScope.Privatize();
+
+    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EnsureInsertPoint();
+  };
+  CGM.getOpenMPRuntime().emitSingleRegion(*this, CodeGen, S.getLocStart(),
+                                          CopyprivateVars, DestExprs, SrcExprs,
+                                          AssignmentOps);
+  // Emit an implicit barrier at the end (to avoid data race on firstprivate
+  // init or if no 'nowait' clause was specified and no 'copyprivate' clause).
+  if ((!S.getSingleClause(OMPC_nowait) || HasFirstprivates) &&
+      CopyprivateVars.empty()) {
+    CGM.getOpenMPRuntime().emitBarrierCall(
+        *this, S.getLocStart(),
+        S.getSingleClause(OMPC_nowait) ? OMPD_unknown : OMPD_single);
+  }
 }
 
 void CodeGenFunction::EmitOMPMasterDirective(const OMPMasterDirective &S) {
-  CGM.getOpenMPRuntime().EmitOMPMasterRegion(*this, [&]() -> void {
-    InlinedOpenMPRegion Region(*this, S.getAssociatedStmt());
-    RunCleanupsScope Scope(*this);
-    EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
-    EnsureInsertPoint();
-  }, S.getLocStart());
+  LexicalScope Scope(*this, S.getSourceRange());
+  auto &&CodeGen = [&S](CodeGenFunction &CGF) {
+    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EnsureInsertPoint();
+  };
+  CGM.getOpenMPRuntime().emitMasterRegion(*this, CodeGen, S.getLocStart());
 }
 
 void CodeGenFunction::EmitOMPCriticalDirective(const OMPCriticalDirective &S) {
-  CGM.getOpenMPRuntime().EmitOMPCriticalRegion(
-      *this, S.getDirectiveName().getAsString(), [&]() -> void {
-    InlinedOpenMPRegion Region(*this, S.getAssociatedStmt());
-    RunCleanupsScope Scope(*this);
-    EmitStmt(
-        cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
-    EnsureInsertPoint();
-  }, S.getLocStart());
+  LexicalScope Scope(*this, S.getSourceRange());
+  auto &&CodeGen = [&S](CodeGenFunction &CGF) {
+    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EnsureInsertPoint();
+  };
+  CGM.getOpenMPRuntime().emitCriticalRegion(
+      *this, S.getDirectiveName().getAsString(), CodeGen, S.getLocStart());
 }
 
-void
-CodeGenFunction::EmitOMPParallelForDirective(const OMPParallelForDirective &) {
-  llvm_unreachable("CodeGen for 'omp parallel for' is not supported yet.");
+void CodeGenFunction::EmitOMPParallelForDirective(
+    const OMPParallelForDirective &S) {
+  // Emit directive as a combined directive that consists of two implicit
+  // directives: 'parallel' with 'for' directive.
+  LexicalScope Scope(*this, S.getSourceRange());
+  (void)emitScheduleClause(*this, S, /*OuterRegion=*/true);
+  auto &&CodeGen = [&S](CodeGenFunction &CGF) {
+    CGF.EmitOMPWorksharingLoop(S);
+    // Emit implicit barrier at the end of parallel region, but this barrier
+    // is at the end of 'for' directive, so emit it as the implicit barrier for
+    // this 'for' directive.
+    CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
+                                               OMPD_parallel);
+  };
+  emitCommonOMPParallelDirective(*this, S, CodeGen);
 }
 
 void CodeGenFunction::EmitOMPParallelForSimdDirective(
@@ -654,45 +1412,600 @@ void CodeGenFunction::EmitOMPParallelForSimdDirective(
 }
 
 void CodeGenFunction::EmitOMPParallelSectionsDirective(
-    const OMPParallelSectionsDirective &) {
-  llvm_unreachable("CodeGen for 'omp parallel sections' is not supported yet.");
+    const OMPParallelSectionsDirective &S) {
+  // Emit directive as a combined directive that consists of two implicit
+  // directives: 'parallel' with 'sections' directive.
+  LexicalScope Scope(*this, S.getSourceRange());
+  auto &&CodeGen = [&S](CodeGenFunction &CGF) {
+    (void)emitSections(CGF, S);
+    // Emit implicit barrier at the end of parallel region.
+    CGF.CGM.getOpenMPRuntime().emitBarrierCall(CGF, S.getLocStart(),
+                                               OMPD_parallel);
+  };
+  emitCommonOMPParallelDirective(*this, S, CodeGen);
 }
 
-void CodeGenFunction::EmitOMPTaskDirective(const OMPTaskDirective &) {
-  llvm_unreachable("CodeGen for 'omp task' is not supported yet.");
+void CodeGenFunction::EmitOMPTaskDirective(const OMPTaskDirective &S) {
+  // Emit outlined function for task construct.
+  LexicalScope Scope(*this, S.getSourceRange());
+  auto CS = cast<CapturedStmt>(S.getAssociatedStmt());
+  auto CapturedStruct = GenerateCapturedStmtArgument(*CS);
+  auto *I = CS->getCapturedDecl()->param_begin();
+  auto *PartId = std::next(I);
+  // The first function argument for tasks is a thread id, the second one is a
+  // part id (0 for tied tasks, >=0 for untied task).
+  llvm::DenseSet<const VarDecl *> EmittedAsPrivate;
+  // Get list of private variables.
+  llvm::SmallVector<const Expr *, 8> PrivateVars;
+  llvm::SmallVector<const Expr *, 8> PrivateCopies;
+  for (auto &&I = S.getClausesOfKind(OMPC_private); I; ++I) {
+    auto *C = cast<OMPPrivateClause>(*I);
+    auto IRef = C->varlist_begin();
+    for (auto *IInit : C->private_copies()) {
+      auto *OrigVD = cast<VarDecl>(cast<DeclRefExpr>(*IRef)->getDecl());
+      if (EmittedAsPrivate.insert(OrigVD->getCanonicalDecl()).second) {
+        PrivateVars.push_back(*IRef);
+        PrivateCopies.push_back(IInit);
+      }
+      ++IRef;
+    }
+  }
+  EmittedAsPrivate.clear();
+  // Get list of firstprivate variables.
+  llvm::SmallVector<const Expr *, 8> FirstprivateVars;
+  llvm::SmallVector<const Expr *, 8> FirstprivateCopies;
+  llvm::SmallVector<const Expr *, 8> FirstprivateInits;
+  for (auto &&I = S.getClausesOfKind(OMPC_firstprivate); I; ++I) {
+    auto *C = cast<OMPFirstprivateClause>(*I);
+    auto IRef = C->varlist_begin();
+    auto IElemInitRef = C->inits().begin();
+    for (auto *IInit : C->private_copies()) {
+      auto *OrigVD = cast<VarDecl>(cast<DeclRefExpr>(*IRef)->getDecl());
+      if (EmittedAsPrivate.insert(OrigVD->getCanonicalDecl()).second) {
+        FirstprivateVars.push_back(*IRef);
+        FirstprivateCopies.push_back(IInit);
+        FirstprivateInits.push_back(*IElemInitRef);
+      }
+      ++IRef, ++IElemInitRef;
+    }
+  }
+  auto &&CodeGen = [PartId, &S, &PrivateVars, &FirstprivateVars](
+      CodeGenFunction &CGF) {
+    // Set proper addresses for generated private copies.
+    auto *CS = cast<CapturedStmt>(S.getAssociatedStmt());
+    OMPPrivateScope Scope(CGF);
+    if (!PrivateVars.empty() || !FirstprivateVars.empty()) {
+      auto *CopyFn = CGF.Builder.CreateAlignedLoad(
+          CGF.GetAddrOfLocalVar(CS->getCapturedDecl()->getParam(3)),
+          CGF.PointerAlignInBytes);
+      auto *PrivatesPtr = CGF.Builder.CreateAlignedLoad(
+          CGF.GetAddrOfLocalVar(CS->getCapturedDecl()->getParam(2)),
+          CGF.PointerAlignInBytes);
+      // Map privates.
+      llvm::SmallVector<std::pair<const VarDecl *, llvm::Value *>, 16>
+          PrivatePtrs;
+      llvm::SmallVector<llvm::Value *, 16> CallArgs;
+      CallArgs.push_back(PrivatesPtr);
+      for (auto *E : PrivateVars) {
+        auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
+        auto *PrivatePtr =
+            CGF.CreateMemTemp(CGF.getContext().getPointerType(E->getType()));
+        PrivatePtrs.push_back(std::make_pair(VD, PrivatePtr));
+        CallArgs.push_back(PrivatePtr);
+      }
+      for (auto *E : FirstprivateVars) {
+        auto *VD = cast<VarDecl>(cast<DeclRefExpr>(E)->getDecl());
+        auto *PrivatePtr =
+            CGF.CreateMemTemp(CGF.getContext().getPointerType(E->getType()));
+        PrivatePtrs.push_back(std::make_pair(VD, PrivatePtr));
+        CallArgs.push_back(PrivatePtr);
+      }
+      CGF.EmitRuntimeCall(CopyFn, CallArgs);
+      for (auto &&Pair : PrivatePtrs) {
+        auto *Replacement =
+            CGF.Builder.CreateAlignedLoad(Pair.second, CGF.PointerAlignInBytes);
+        Scope.addPrivate(Pair.first, [Replacement]() { return Replacement; });
+      }
+    }
+    (void)Scope.Privatize();
+    if (*PartId) {
+      // TODO: emit code for untied tasks.
+    }
+    CGF.EmitStmt(CS->getCapturedStmt());
+  };
+  auto OutlinedFn =
+      CGM.getOpenMPRuntime().emitTaskOutlinedFunction(S, *I, CodeGen);
+  // Check if we should emit tied or untied task.
+  bool Tied = !S.getSingleClause(OMPC_untied);
+  // Check if the task is final
+  llvm::PointerIntPair<llvm::Value *, 1, bool> Final;
+  if (auto *Clause = S.getSingleClause(OMPC_final)) {
+    // If the condition constant folds and can be elided, try to avoid emitting
+    // the condition and the dead arm of the if/else.
+    auto *Cond = cast<OMPFinalClause>(Clause)->getCondition();
+    bool CondConstant;
+    if (ConstantFoldsToSimpleInteger(Cond, CondConstant))
+      Final.setInt(CondConstant);
+    else
+      Final.setPointer(EvaluateExprAsBool(Cond));
+  } else {
+    // By default the task is not final.
+    Final.setInt(/*IntVal=*/false);
+  }
+  auto SharedsTy = getContext().getRecordType(CS->getCapturedRecordDecl());
+  const Expr *IfCond = nullptr;
+  if (auto C = S.getSingleClause(OMPC_if)) {
+    IfCond = cast<OMPIfClause>(C)->getCondition();
+  }
+  CGM.getOpenMPRuntime().emitTaskCall(
+      *this, S.getLocStart(), S, Tied, Final, OutlinedFn, SharedsTy,
+      CapturedStruct, IfCond, PrivateVars, PrivateCopies, FirstprivateVars,
+      FirstprivateCopies, FirstprivateInits);
 }
 
-void CodeGenFunction::EmitOMPTaskyieldDirective(const OMPTaskyieldDirective &) {
-  llvm_unreachable("CodeGen for 'omp taskyield' is not supported yet.");
+void CodeGenFunction::EmitOMPTaskyieldDirective(
+    const OMPTaskyieldDirective &S) {
+  CGM.getOpenMPRuntime().emitTaskyieldCall(*this, S.getLocStart());
 }
 
 void CodeGenFunction::EmitOMPBarrierDirective(const OMPBarrierDirective &S) {
-  CGM.getOpenMPRuntime().EmitOMPBarrierCall(*this, S.getLocStart());
+  CGM.getOpenMPRuntime().emitBarrierCall(*this, S.getLocStart(), OMPD_barrier);
 }
 
-void CodeGenFunction::EmitOMPTaskwaitDirective(const OMPTaskwaitDirective &) {
-  llvm_unreachable("CodeGen for 'omp taskwait' is not supported yet.");
+void CodeGenFunction::EmitOMPTaskwaitDirective(const OMPTaskwaitDirective &S) {
+  CGM.getOpenMPRuntime().emitTaskwaitCall(*this, S.getLocStart());
 }
 
 void CodeGenFunction::EmitOMPFlushDirective(const OMPFlushDirective &S) {
-  CGM.getOpenMPRuntime().EmitOMPFlush(
-      *this, [&]() -> ArrayRef<const Expr *> {
-               if (auto C = S.getSingleClause(/*K*/ OMPC_flush)) {
-                 auto FlushClause = cast<OMPFlushClause>(C);
-                 return llvm::makeArrayRef(FlushClause->varlist_begin(),
-                                           FlushClause->varlist_end());
-               }
-               return llvm::None;
-             }(),
-      S.getLocStart());
+  CGM.getOpenMPRuntime().emitFlush(*this, [&]() -> ArrayRef<const Expr *> {
+    if (auto C = S.getSingleClause(/*K*/ OMPC_flush)) {
+      auto FlushClause = cast<OMPFlushClause>(C);
+      return llvm::makeArrayRef(FlushClause->varlist_begin(),
+                                FlushClause->varlist_end());
+    }
+    return llvm::None;
+  }(), S.getLocStart());
+}
+
+void CodeGenFunction::EmitOMPOrderedDirective(const OMPOrderedDirective &S) {
+  LexicalScope Scope(*this, S.getSourceRange());
+  auto &&CodeGen = [&S](CodeGenFunction &CGF) {
+    CGF.EmitStmt(cast<CapturedStmt>(S.getAssociatedStmt())->getCapturedStmt());
+    CGF.EnsureInsertPoint();
+  };
+  CGM.getOpenMPRuntime().emitOrderedRegion(*this, CodeGen, S.getLocStart());
+}
+
+static llvm::Value *convertToScalarValue(CodeGenFunction &CGF, RValue Val,
+                                         QualType SrcType, QualType DestType) {
+  assert(CGF.hasScalarEvaluationKind(DestType) &&
+         "DestType must have scalar evaluation kind.");
+  assert(!Val.isAggregate() && "Must be a scalar or complex.");
+  return Val.isScalar()
+             ? CGF.EmitScalarConversion(Val.getScalarVal(), SrcType, DestType)
+             : CGF.EmitComplexToScalarConversion(Val.getComplexVal(), SrcType,
+                                                 DestType);
+}
+
+static CodeGenFunction::ComplexPairTy
+convertToComplexValue(CodeGenFunction &CGF, RValue Val, QualType SrcType,
+                      QualType DestType) {
+  assert(CGF.getEvaluationKind(DestType) == TEK_Complex &&
+         "DestType must have complex evaluation kind.");
+  CodeGenFunction::ComplexPairTy ComplexVal;
+  if (Val.isScalar()) {
+    // Convert the input element to the element type of the complex.
+    auto DestElementType = DestType->castAs<ComplexType>()->getElementType();
+    auto ScalarVal =
+        CGF.EmitScalarConversion(Val.getScalarVal(), SrcType, DestElementType);
+    ComplexVal = CodeGenFunction::ComplexPairTy(
+        ScalarVal, llvm::Constant::getNullValue(ScalarVal->getType()));
+  } else {
+    assert(Val.isComplex() && "Must be a scalar or complex.");
+    auto SrcElementType = SrcType->castAs<ComplexType>()->getElementType();
+    auto DestElementType = DestType->castAs<ComplexType>()->getElementType();
+    ComplexVal.first = CGF.EmitScalarConversion(
+        Val.getComplexVal().first, SrcElementType, DestElementType);
+    ComplexVal.second = CGF.EmitScalarConversion(
+        Val.getComplexVal().second, SrcElementType, DestElementType);
+  }
+  return ComplexVal;
+}
+
+static void emitSimpleAtomicStore(CodeGenFunction &CGF, bool IsSeqCst,
+                                  LValue LVal, RValue RVal) {
+  if (LVal.isGlobalReg()) {
+    CGF.EmitStoreThroughGlobalRegLValue(RVal, LVal);
+  } else {
+    CGF.EmitAtomicStore(RVal, LVal, IsSeqCst ? llvm::SequentiallyConsistent
+                                             : llvm::Monotonic,
+                        LVal.isVolatile(), /*IsInit=*/false);
+  }
+}
+
+static void emitSimpleStore(CodeGenFunction &CGF, LValue LVal, RValue RVal,
+                            QualType RValTy) {
+  switch (CGF.getEvaluationKind(LVal.getType())) {
+  case TEK_Scalar:
+    CGF.EmitStoreThroughLValue(
+        RValue::get(convertToScalarValue(CGF, RVal, RValTy, LVal.getType())),
+        LVal);
+    break;
+  case TEK_Complex:
+    CGF.EmitStoreOfComplex(
+        convertToComplexValue(CGF, RVal, RValTy, LVal.getType()), LVal,
+        /*isInit=*/false);
+    break;
+  case TEK_Aggregate:
+    llvm_unreachable("Must be a scalar or complex.");
+  }
 }
 
-void CodeGenFunction::EmitOMPOrderedDirective(const OMPOrderedDirective &) {
-  llvm_unreachable("CodeGen for 'omp ordered' is not supported yet.");
+static void EmitOMPAtomicReadExpr(CodeGenFunction &CGF, bool IsSeqCst,
+                                  const Expr *X, const Expr *V,
+                                  SourceLocation Loc) {
+  // v = x;
+  assert(V->isLValue() && "V of 'omp atomic read' is not lvalue");
+  assert(X->isLValue() && "X of 'omp atomic read' is not lvalue");
+  LValue XLValue = CGF.EmitLValue(X);
+  LValue VLValue = CGF.EmitLValue(V);
+  RValue Res = XLValue.isGlobalReg()
+                   ? CGF.EmitLoadOfLValue(XLValue, Loc)
+                   : CGF.EmitAtomicLoad(XLValue, Loc,
+                                        IsSeqCst ? llvm::SequentiallyConsistent
+                                                 : llvm::Monotonic,
+                                        XLValue.isVolatile());
+  // OpenMP, 2.12.6, atomic Construct
+  // Any atomic construct with a seq_cst clause forces the atomically
+  // performed operation to include an implicit flush operation without a
+  // list.
+  if (IsSeqCst)
+    CGF.CGM.getOpenMPRuntime().emitFlush(CGF, llvm::None, Loc);
+  emitSimpleStore(CGF,VLValue, Res, X->getType().getNonReferenceType());
 }
 
-void CodeGenFunction::EmitOMPAtomicDirective(const OMPAtomicDirective &) {
-  llvm_unreachable("CodeGen for 'omp atomic' is not supported yet.");
+static void EmitOMPAtomicWriteExpr(CodeGenFunction &CGF, bool IsSeqCst,
+                                   const Expr *X, const Expr *E,
+                                   SourceLocation Loc) {
+  // x = expr;
+  assert(X->isLValue() && "X of 'omp atomic write' is not lvalue");
+  emitSimpleAtomicStore(CGF, IsSeqCst, CGF.EmitLValue(X), CGF.EmitAnyExpr(E));
+  // OpenMP, 2.12.6, atomic Construct
+  // Any atomic construct with a seq_cst clause forces the atomically
+  // performed operation to include an implicit flush operation without a
+  // list.
+  if (IsSeqCst)
+    CGF.CGM.getOpenMPRuntime().emitFlush(CGF, llvm::None, Loc);
+}
+
+static std::pair<bool, RValue> emitOMPAtomicRMW(CodeGenFunction &CGF, LValue X,
+                                                RValue Update,
+                                                BinaryOperatorKind BO,
+                                                llvm::AtomicOrdering AO,
+                                                bool IsXLHSInRHSPart) {
+  auto &Context = CGF.CGM.getContext();
+  // Allow atomicrmw only if 'x' and 'update' are integer values, lvalue for 'x'
+  // expression is simple and atomic is allowed for the given type for the
+  // target platform.
+  if (BO == BO_Comma || !Update.isScalar() ||
+      !Update.getScalarVal()->getType()->isIntegerTy() ||
+      !X.isSimple() || (!isa<llvm::ConstantInt>(Update.getScalarVal()) &&
+                        (Update.getScalarVal()->getType() !=
+                         X.getAddress()->getType()->getPointerElementType())) ||
+      !X.getAddress()->getType()->getPointerElementType()->isIntegerTy() ||
+      !Context.getTargetInfo().hasBuiltinAtomic(
+          Context.getTypeSize(X.getType()), Context.toBits(X.getAlignment())))
+    return std::make_pair(false, RValue::get(nullptr));
+
+  llvm::AtomicRMWInst::BinOp RMWOp;
+  switch (BO) {
+  case BO_Add:
+    RMWOp = llvm::AtomicRMWInst::Add;
+    break;
+  case BO_Sub:
+    if (!IsXLHSInRHSPart)
+      return std::make_pair(false, RValue::get(nullptr));
+    RMWOp = llvm::AtomicRMWInst::Sub;
+    break;
+  case BO_And:
+    RMWOp = llvm::AtomicRMWInst::And;
+    break;
+  case BO_Or:
+    RMWOp = llvm::AtomicRMWInst::Or;
+    break;
+  case BO_Xor:
+    RMWOp = llvm::AtomicRMWInst::Xor;
+    break;
+  case BO_LT:
+    RMWOp = X.getType()->hasSignedIntegerRepresentation()
+                ? (IsXLHSInRHSPart ? llvm::AtomicRMWInst::Min
+                                   : llvm::AtomicRMWInst::Max)
+                : (IsXLHSInRHSPart ? llvm::AtomicRMWInst::UMin
+                                   : llvm::AtomicRMWInst::UMax);
+    break;
+  case BO_GT:
+    RMWOp = X.getType()->hasSignedIntegerRepresentation()
+                ? (IsXLHSInRHSPart ? llvm::AtomicRMWInst::Max
+                                   : llvm::AtomicRMWInst::Min)
+                : (IsXLHSInRHSPart ? llvm::AtomicRMWInst::UMax
+                                   : llvm::AtomicRMWInst::UMin);
+    break;
+  case BO_Assign:
+    RMWOp = llvm::AtomicRMWInst::Xchg;
+    break;
+  case BO_Mul:
+  case BO_Div:
+  case BO_Rem:
+  case BO_Shl:
+  case BO_Shr:
+  case BO_LAnd:
+  case BO_LOr:
+    return std::make_pair(false, RValue::get(nullptr));
+  case BO_PtrMemD:
+  case BO_PtrMemI:
+  case BO_LE:
+  case BO_GE:
+  case BO_EQ:
+  case BO_NE:
+  case BO_AddAssign:
+  case BO_SubAssign:
+  case BO_AndAssign:
+  case BO_OrAssign:
+  case BO_XorAssign:
+  case BO_MulAssign:
+  case BO_DivAssign:
+  case BO_RemAssign:
+  case BO_ShlAssign:
+  case BO_ShrAssign:
+  case BO_Comma:
+    llvm_unreachable("Unsupported atomic update operation");
+  }
+  auto *UpdateVal = Update.getScalarVal();
+  if (auto *IC = dyn_cast<llvm::ConstantInt>(UpdateVal)) {
+    UpdateVal = CGF.Builder.CreateIntCast(
+        IC, X.getAddress()->getType()->getPointerElementType(),
+        X.getType()->hasSignedIntegerRepresentation());
+  }
+  auto *Res = CGF.Builder.CreateAtomicRMW(RMWOp, X.getAddress(), UpdateVal, AO);
+  return std::make_pair(true, RValue::get(Res));
+}
+
+std::pair<bool, RValue> CodeGenFunction::EmitOMPAtomicSimpleUpdateExpr(
+    LValue X, RValue E, BinaryOperatorKind BO, bool IsXLHSInRHSPart,
+    llvm::AtomicOrdering AO, SourceLocation Loc,
+    const llvm::function_ref<RValue(RValue)> &CommonGen) {
+  // Update expressions are allowed to have the following forms:
+  // x binop= expr; -> xrval + expr;
+  // x++, ++x -> xrval + 1;
+  // x--, --x -> xrval - 1;
+  // x = x binop expr; -> xrval binop expr
+  // x = expr Op x; - > expr binop xrval;
+  auto Res = emitOMPAtomicRMW(*this, X, E, BO, AO, IsXLHSInRHSPart);
+  if (!Res.first) {
+    if (X.isGlobalReg()) {
+      // Emit an update expression: 'xrval' binop 'expr' or 'expr' binop
+      // 'xrval'.
+      EmitStoreThroughLValue(CommonGen(EmitLoadOfLValue(X, Loc)), X);
+    } else {
+      // Perform compare-and-swap procedure.
+      EmitAtomicUpdate(X, AO, CommonGen, X.getType().isVolatileQualified());
+    }
+  }
+  return Res;
+}
+
+static void EmitOMPAtomicUpdateExpr(CodeGenFunction &CGF, bool IsSeqCst,
+                                    const Expr *X, const Expr *E,
+                                    const Expr *UE, bool IsXLHSInRHSPart,
+                                    SourceLocation Loc) {
+  assert(isa<BinaryOperator>(UE->IgnoreImpCasts()) &&
+         "Update expr in 'atomic update' must be a binary operator.");
+  auto *BOUE = cast<BinaryOperator>(UE->IgnoreImpCasts());
+  // Update expressions are allowed to have the following forms:
+  // x binop= expr; -> xrval + expr;
+  // x++, ++x -> xrval + 1;
+  // x--, --x -> xrval - 1;
+  // x = x binop expr; -> xrval binop expr
+  // x = expr Op x; - > expr binop xrval;
+  assert(X->isLValue() && "X of 'omp atomic update' is not lvalue");
+  LValue XLValue = CGF.EmitLValue(X);
+  RValue ExprRValue = CGF.EmitAnyExpr(E);
+  auto AO = IsSeqCst ? llvm::SequentiallyConsistent : llvm::Monotonic;
+  auto *LHS = cast<OpaqueValueExpr>(BOUE->getLHS()->IgnoreImpCasts());
+  auto *RHS = cast<OpaqueValueExpr>(BOUE->getRHS()->IgnoreImpCasts());
+  auto *XRValExpr = IsXLHSInRHSPart ? LHS : RHS;
+  auto *ERValExpr = IsXLHSInRHSPart ? RHS : LHS;
+  auto Gen =
+      [&CGF, UE, ExprRValue, XRValExpr, ERValExpr](RValue XRValue) -> RValue {
+        CodeGenFunction::OpaqueValueMapping MapExpr(CGF, ERValExpr, ExprRValue);
+        CodeGenFunction::OpaqueValueMapping MapX(CGF, XRValExpr, XRValue);
+        return CGF.EmitAnyExpr(UE);
+      };
+  (void)CGF.EmitOMPAtomicSimpleUpdateExpr(
+      XLValue, ExprRValue, BOUE->getOpcode(), IsXLHSInRHSPart, AO, Loc, Gen);
+  // OpenMP, 2.12.6, atomic Construct
+  // Any atomic construct with a seq_cst clause forces the atomically
+  // performed operation to include an implicit flush operation without a
+  // list.
+  if (IsSeqCst)
+    CGF.CGM.getOpenMPRuntime().emitFlush(CGF, llvm::None, Loc);
+}
+
+static RValue convertToType(CodeGenFunction &CGF, RValue Value,
+                            QualType SourceType, QualType ResType) {
+  switch (CGF.getEvaluationKind(ResType)) {
+  case TEK_Scalar:
+    return RValue::get(convertToScalarValue(CGF, Value, SourceType, ResType));
+  case TEK_Complex: {
+    auto Res = convertToComplexValue(CGF, Value, SourceType, ResType);
+    return RValue::getComplex(Res.first, Res.second);
+  }
+  case TEK_Aggregate:
+    break;
+  }
+  llvm_unreachable("Must be a scalar or complex.");
+}
+
+static void EmitOMPAtomicCaptureExpr(CodeGenFunction &CGF, bool IsSeqCst,
+                                     bool IsPostfixUpdate, const Expr *V,
+                                     const Expr *X, const Expr *E,
+                                     const Expr *UE, bool IsXLHSInRHSPart,
+                                     SourceLocation Loc) {
+  assert(X->isLValue() && "X of 'omp atomic capture' is not lvalue");
+  assert(V->isLValue() && "V of 'omp atomic capture' is not lvalue");
+  RValue NewVVal;
+  LValue VLValue = CGF.EmitLValue(V);
+  LValue XLValue = CGF.EmitLValue(X);
+  RValue ExprRValue = CGF.EmitAnyExpr(E);
+  auto AO = IsSeqCst ? llvm::SequentiallyConsistent : llvm::Monotonic;
+  QualType NewVValType;
+  if (UE) {
+    // 'x' is updated with some additional value.
+    assert(isa<BinaryOperator>(UE->IgnoreImpCasts()) &&
+           "Update expr in 'atomic capture' must be a binary operator.");
+    auto *BOUE = cast<BinaryOperator>(UE->IgnoreImpCasts());
+    // Update expressions are allowed to have the following forms:
+    // x binop= expr; -> xrval + expr;
+    // x++, ++x -> xrval + 1;
+    // x--, --x -> xrval - 1;
+    // x = x binop expr; -> xrval binop expr
+    // x = expr Op x; - > expr binop xrval;
+    auto *LHS = cast<OpaqueValueExpr>(BOUE->getLHS()->IgnoreImpCasts());
+    auto *RHS = cast<OpaqueValueExpr>(BOUE->getRHS()->IgnoreImpCasts());
+    auto *XRValExpr = IsXLHSInRHSPart ? LHS : RHS;
+    NewVValType = XRValExpr->getType();
+    auto *ERValExpr = IsXLHSInRHSPart ? RHS : LHS;
+    auto &&Gen = [&CGF, &NewVVal, UE, ExprRValue, XRValExpr, ERValExpr,
+                  IsSeqCst, IsPostfixUpdate](RValue XRValue) -> RValue {
+      CodeGenFunction::OpaqueValueMapping MapExpr(CGF, ERValExpr, ExprRValue);
+      CodeGenFunction::OpaqueValueMapping MapX(CGF, XRValExpr, XRValue);
+      RValue Res = CGF.EmitAnyExpr(UE);
+      NewVVal = IsPostfixUpdate ? XRValue : Res;
+      return Res;
+    };
+    auto Res = CGF.EmitOMPAtomicSimpleUpdateExpr(
+        XLValue, ExprRValue, BOUE->getOpcode(), IsXLHSInRHSPart, AO, Loc, Gen);
+    if (Res.first) {
+      // 'atomicrmw' instruction was generated.
+      if (IsPostfixUpdate) {
+        // Use old value from 'atomicrmw'.
+        NewVVal = Res.second;
+      } else {
+        // 'atomicrmw' does not provide new value, so evaluate it using old
+        // value of 'x'.
+        CodeGenFunction::OpaqueValueMapping MapExpr(CGF, ERValExpr, ExprRValue);
+        CodeGenFunction::OpaqueValueMapping MapX(CGF, XRValExpr, Res.second);
+        NewVVal = CGF.EmitAnyExpr(UE);
+      }
+    }
+  } else {
+    // 'x' is simply rewritten with some 'expr'.
+    NewVValType = X->getType().getNonReferenceType();
+    ExprRValue = convertToType(CGF, ExprRValue, E->getType(),
+                               X->getType().getNonReferenceType());
+    auto &&Gen = [&CGF, &NewVVal, ExprRValue](RValue XRValue) -> RValue {
+      NewVVal = XRValue;
+      return ExprRValue;
+    };
+    // Try to perform atomicrmw xchg, otherwise simple exchange.
+    auto Res = CGF.EmitOMPAtomicSimpleUpdateExpr(
+        XLValue, ExprRValue, /*BO=*/BO_Assign, /*IsXLHSInRHSPart=*/false, AO,
+        Loc, Gen);
+    if (Res.first) {
+      // 'atomicrmw' instruction was generated.
+      NewVVal = IsPostfixUpdate ? Res.second : ExprRValue;
+    }
+  }
+  // Emit post-update store to 'v' of old/new 'x' value.
+  emitSimpleStore(CGF, VLValue, NewVVal, NewVValType);
+  // OpenMP, 2.12.6, atomic Construct
+  // Any atomic construct with a seq_cst clause forces the atomically
+  // performed operation to include an implicit flush operation without a
+  // list.
+  if (IsSeqCst)
+    CGF.CGM.getOpenMPRuntime().emitFlush(CGF, llvm::None, Loc);
+}
+
+static void EmitOMPAtomicExpr(CodeGenFunction &CGF, OpenMPClauseKind Kind,
+                              bool IsSeqCst, bool IsPostfixUpdate,
+                              const Expr *X, const Expr *V, const Expr *E,
+                              const Expr *UE, bool IsXLHSInRHSPart,
+                              SourceLocation Loc) {
+  switch (Kind) {
+  case OMPC_read:
+    EmitOMPAtomicReadExpr(CGF, IsSeqCst, X, V, Loc);
+    break;
+  case OMPC_write:
+    EmitOMPAtomicWriteExpr(CGF, IsSeqCst, X, E, Loc);
+    break;
+  case OMPC_unknown:
+  case OMPC_update:
+    EmitOMPAtomicUpdateExpr(CGF, IsSeqCst, X, E, UE, IsXLHSInRHSPart, Loc);
+    break;
+  case OMPC_capture:
+    EmitOMPAtomicCaptureExpr(CGF, IsSeqCst, IsPostfixUpdate, V, X, E, UE,
+                             IsXLHSInRHSPart, Loc);
+    break;
+  case OMPC_if:
+  case OMPC_final:
+  case OMPC_num_threads:
+  case OMPC_private:
+  case OMPC_firstprivate:
+  case OMPC_lastprivate:
+  case OMPC_reduction:
+  case OMPC_safelen:
+  case OMPC_collapse:
+  case OMPC_default:
+  case OMPC_seq_cst:
+  case OMPC_shared:
+  case OMPC_linear:
+  case OMPC_aligned:
+  case OMPC_copyin:
+  case OMPC_copyprivate:
+  case OMPC_flush:
+  case OMPC_proc_bind:
+  case OMPC_schedule:
+  case OMPC_ordered:
+  case OMPC_nowait:
+  case OMPC_untied:
+  case OMPC_threadprivate:
+  case OMPC_mergeable:
+    llvm_unreachable("Clause is not allowed in 'omp atomic'.");
+  }
+}
+
+void CodeGenFunction::EmitOMPAtomicDirective(const OMPAtomicDirective &S) {
+  bool IsSeqCst = S.getSingleClause(/*K=*/OMPC_seq_cst);
+  OpenMPClauseKind Kind = OMPC_unknown;
+  for (auto *C : S.clauses()) {
+    // Find first clause (skip seq_cst clause, if it is first).
+    if (C->getClauseKind() != OMPC_seq_cst) {
+      Kind = C->getClauseKind();
+      break;
+    }
+  }
+
+  const auto *CS =
+      S.getAssociatedStmt()->IgnoreContainers(/*IgnoreCaptured=*/true);
+  if (const auto *EWC = dyn_cast<ExprWithCleanups>(CS)) {
+    enterFullExpression(EWC);
+  }
+  // Processing for statements under 'atomic capture'.
+  if (const auto *Compound = dyn_cast<CompoundStmt>(CS)) {
+    for (const auto *C : Compound->body()) {
+      if (const auto *EWC = dyn_cast<ExprWithCleanups>(C)) {
+        enterFullExpression(EWC);
+      }
+    }
+  }
+
+  LexicalScope Scope(*this, S.getSourceRange());
+  auto &&CodeGen = [&S, Kind, IsSeqCst](CodeGenFunction &CGF) {
+    EmitOMPAtomicExpr(CGF, Kind, IsSeqCst, S.isPostfixUpdate(), S.getX(),
+                      S.getV(), S.getExpr(), S.getUpdateExpr(),
+                      S.isXLHSInRHSPart(), S.getLocStart());
+  };
+  CGM.getOpenMPRuntime().emitInlinedDirective(*this, CodeGen);
 }
 
 void CodeGenFunction::EmitOMPTargetDirective(const OMPTargetDirective &) {
@@ -702,4 +2015,3 @@ void CodeGenFunction::EmitOMPTargetDirective(const OMPTargetDirective &) {
 void CodeGenFunction::EmitOMPTeamsDirective(const OMPTeamsDirective &) {
   llvm_unreachable("CodeGen for 'omp teams' is not supported yet.");
 }
-
diff --git a/lib/CodeGen/CGVTT.cpp b/lib/CodeGen/CGVTT.cpp
index bd280ea14e8d..e3df5a4c05e3 100644
--- a/lib/CodeGen/CGVTT.cpp
+++ b/lib/CodeGen/CGVTT.cpp
@@ -18,7 +18,7 @@
 using namespace clang;
 using namespace CodeGen;
 
-static llvm::Constant *
+static llvm::GlobalVariable *
 GetAddrOfVTTVTable(CodeGenVTables &CGVT, CodeGenModule &CGM,
                    const CXXRecordDecl *MostDerivedClass,
                    const VTTVTable &VTable,
@@ -47,8 +47,8 @@ CodeGenVTables::EmitVTTDefinition(llvm::GlobalVariable *VTT,
   llvm::Type *Int8PtrTy = CGM.Int8PtrTy, *Int64Ty = CGM.Int64Ty;
   llvm::ArrayType *ArrayType = 
     llvm::ArrayType::get(Int8PtrTy, Builder.getVTTComponents().size());
-  
-  SmallVector<llvm::Constant *, 8> VTables;
+
+  SmallVector<llvm::GlobalVariable *, 8> VTables;
   SmallVector<VTableAddressPointsMapTy, 8> VTableAddressPoints;
   for (const VTTVTable *i = Builder.getVTTVTables().begin(),
                        *e = Builder.getVTTVTables().end(); i != e; ++i) {
@@ -61,7 +61,7 @@ CodeGenVTables::EmitVTTDefinition(llvm::GlobalVariable *VTT,
   for (const VTTComponent *i = Builder.getVTTComponents().begin(),
                           *e = Builder.getVTTComponents().end(); i != e; ++i) {
     const VTTVTable &VTTVT = Builder.getVTTVTables()[i->VTableIndex];
-    llvm::Constant *VTable = VTables[i->VTableIndex];
+    llvm::GlobalVariable *VTable = VTables[i->VTableIndex];
     uint64_t AddressPoint;
     if (VTTVT.getBase() == RD) {
       // Just get the address point for the regular vtable.
@@ -79,8 +79,8 @@ CodeGenVTables::EmitVTTDefinition(llvm::GlobalVariable *VTT,
        llvm::ConstantInt::get(Int64Ty, AddressPoint)
      };
 
-     llvm::Constant *Init = 
-       llvm::ConstantExpr::getInBoundsGetElementPtr(VTable, Idxs);
+     llvm::Constant *Init = llvm::ConstantExpr::getInBoundsGetElementPtr(
+         VTable->getValueType(), VTable, Idxs);
 
      Init = llvm::ConstantExpr::getBitCast(Init, Int8PtrTy);
 
@@ -94,6 +94,9 @@ CodeGenVTables::EmitVTTDefinition(llvm::GlobalVariable *VTT,
   // Set the correct linkage.
   VTT->setLinkage(Linkage);
 
+  if (CGM.supportsCOMDAT() && VTT->isWeakForLinker())
+    VTT->setComdat(CGM.getModule().getOrInsertComdat(VTT->getName()));
+
   // Set the right visibility.
   CGM.setGlobalVisibility(VTT, RD);
 }
@@ -174,4 +177,3 @@ CodeGenVTables::getSecondaryVirtualPointerIndex(const CXXRecordDecl *RD,
   
   return I->second;
 }
-
diff --git a/lib/CodeGen/CGVTables.cpp b/lib/CodeGen/CGVTables.cpp
index acb2a56fab30..57370a6faa2c 100644
--- a/lib/CodeGen/CGVTables.cpp
+++ b/lib/CodeGen/CGVTables.cpp
@@ -377,7 +377,10 @@ void CodeGenFunction::GenerateThunk(llvm::Function *Fn,
 
   // Set the right linkage.
   CGM.setFunctionLinkage(GD, Fn);
-  
+
+  if (CGM.supportsCOMDAT() && Fn->isWeakForLinker())
+    Fn->setComdat(CGM.getModule().getOrInsertComdat(Fn->getName()));
+
   // Set the right visibility.
   const CXXMethodDecl *MD = cast<CXXMethodDecl>(GD.getDecl());
   setThunkVisibility(CGM, MD, Thunk, Fn);
@@ -666,6 +669,8 @@ CodeGenVTables::GenerateConstructionVTable(const CXXRecordDecl *RD,
       VTLayout->getNumVTableThunks(), RTTI);
   VTable->setInitializer(Init);
   
+  CGM.EmitVTableBitSetEntries(VTable, *VTLayout.get());
+
   return VTable;
 }
 
@@ -738,7 +743,7 @@ CodeGenModule::getVTableLinkage(const CXXRecordDecl *RD) {
     return DiscardableODRLinkage;
 
   case TSK_ExplicitInstantiationDeclaration:
-    llvm_unreachable("Should not have been asked to emit this");
+    return llvm::GlobalVariable::ExternalLinkage;
 
   case TSK_ExplicitInstantiationDefinition:
     return NonDiscardableODRLinkage;
@@ -747,19 +752,13 @@ CodeGenModule::getVTableLinkage(const CXXRecordDecl *RD) {
   llvm_unreachable("Invalid TemplateSpecializationKind!");
 }
 
-/// This is a callback from Sema to tell us that it believes that a
-/// particular v-table is required to be emitted in this translation
-/// unit.
+/// This is a callback from Sema to tell us that that a particular v-table is
+/// required to be emitted in this translation unit.
 ///
-/// The reason we don't simply trust this callback is because Sema
-/// will happily report that something is used even when it's used
-/// only in code that we don't actually have to emit.
-///
-/// \param isRequired - if true, the v-table is mandatory, e.g.
-///   because the translation unit defines the key function
-void CodeGenModule::EmitVTable(CXXRecordDecl *theClass, bool isRequired) {
-  if (!isRequired) return;
-
+/// This is only called for vtables that _must_ be emitted (mainly due to key
+/// functions).  For weak vtables, CodeGen tracks when they are needed and
+/// emits them as-needed.
+void CodeGenModule::EmitVTable(CXXRecordDecl *theClass) {
   VTables.GenerateClassData(theClass);
 }
 
@@ -840,3 +839,68 @@ void CodeGenModule::EmitDeferredVTables() {
          "deferred extra v-tables during v-table emission?");
   DeferredVTables.clear();
 }
+
+void CodeGenModule::EmitVTableBitSetEntries(llvm::GlobalVariable *VTable,
+                                            const VTableLayout &VTLayout) {
+  if (!LangOpts.Sanitize.has(SanitizerKind::CFIVCall) &&
+      !LangOpts.Sanitize.has(SanitizerKind::CFINVCall) &&
+      !LangOpts.Sanitize.has(SanitizerKind::CFIDerivedCast) &&
+      !LangOpts.Sanitize.has(SanitizerKind::CFIUnrelatedCast))
+    return;
+
+  llvm::Metadata *VTableMD = llvm::ConstantAsMetadata::get(VTable);
+
+  std::vector<llvm::MDTuple *> BitsetEntries;
+  // Create a bit set entry for each address point.
+  for (auto &&AP : VTLayout.getAddressPoints()) {
+    // FIXME: Add blacklisting scheme.
+    if (AP.first.getBase()->isInStdNamespace())
+      continue;
+
+    std::string OutName;
+    llvm::raw_string_ostream Out(OutName);
+    getCXXABI().getMangleContext().mangleCXXVTableBitSet(AP.first.getBase(),
+                                                         Out);
+
+    CharUnits PointerWidth =
+        Context.toCharUnitsFromBits(Context.getTargetInfo().getPointerWidth(0));
+    uint64_t AddrPointOffset = AP.second * PointerWidth.getQuantity();
+
+    llvm::Metadata *BitsetOps[] = {
+        llvm::MDString::get(getLLVMContext(), Out.str()),
+        VTableMD,
+        llvm::ConstantAsMetadata::get(
+            llvm::ConstantInt::get(Int64Ty, AddrPointOffset))};
+    llvm::MDTuple *BitsetEntry =
+        llvm::MDTuple::get(getLLVMContext(), BitsetOps);
+    BitsetEntries.push_back(BitsetEntry);
+  }
+
+  // Sort the bit set entries for determinism.
+  std::sort(BitsetEntries.begin(), BitsetEntries.end(), [](llvm::MDTuple *T1,
+                                                           llvm::MDTuple *T2) {
+    if (T1 == T2)
+      return false;
+
+    StringRef S1 = cast<llvm::MDString>(T1->getOperand(0))->getString();
+    StringRef S2 = cast<llvm::MDString>(T2->getOperand(0))->getString();
+    if (S1 < S2)
+      return true;
+    if (S1 != S2)
+      return false;
+
+    uint64_t Offset1 = cast<llvm::ConstantInt>(
+                           cast<llvm::ConstantAsMetadata>(T1->getOperand(2))
+                               ->getValue())->getZExtValue();
+    uint64_t Offset2 = cast<llvm::ConstantInt>(
+                           cast<llvm::ConstantAsMetadata>(T2->getOperand(2))
+                               ->getValue())->getZExtValue();
+    assert(Offset1 != Offset2);
+    return Offset1 < Offset2;
+  });
+
+  llvm::NamedMDNode *BitsetsMD =
+      getModule().getOrInsertNamedMetadata("llvm.bitsets");
+  for (auto BitsetEntry : BitsetEntries)
+    BitsetsMD->addOperand(BitsetEntry);
+}
diff --git a/lib/CodeGen/CGValue.h b/lib/CodeGen/CGValue.h
index 82cd9496e00c..92055917dba9 100644
--- a/lib/CodeGen/CGValue.h
+++ b/lib/CodeGen/CGValue.h
@@ -19,6 +19,7 @@
 #include "clang/AST/CharUnits.h"
 #include "clang/AST/Type.h"
 #include "llvm/IR/Value.h"
+#include "llvm/IR/Type.h"
 
 namespace llvm {
   class Constant;
@@ -299,6 +300,7 @@ public:
 
     LValue R;
     R.LVType = Simple;
+    assert(address->getType()->isPointerTy());
     R.V = address;
     R.Initialize(type, qs, alignment, TBAAInfo);
     return R;
diff --git a/lib/CodeGen/CMakeLists.txt b/lib/CodeGen/CMakeLists.txt
index 10c2409f6bff..18f505d8c592 100644
--- a/lib/CodeGen/CMakeLists.txt
+++ b/lib/CodeGen/CMakeLists.txt
@@ -1,4 +1,5 @@
 set(LLVM_LINK_COMPONENTS
+  Analysis
   BitReader
   BitWriter
   Core
@@ -12,7 +13,6 @@ set(LLVM_LINK_COMPONENTS
   ProfileData
   ScalarOpts
   Support
-  Target
   TransformUtils
   )
 
diff --git a/lib/CodeGen/CodeGenAction.cpp b/lib/CodeGen/CodeGenAction.cpp
index a6f6fdef335c..7e82fcc4b22a 100644
--- a/lib/CodeGen/CodeGenAction.cpp
+++ b/lib/CodeGen/CodeGenAction.cpp
@@ -46,7 +46,7 @@ namespace clang {
     const CodeGenOptions &CodeGenOpts;
     const TargetOptions &TargetOpts;
     const LangOptions &LangOpts;
-    raw_ostream *AsmOutStream;
+    raw_pwrite_stream *AsmOutStream;
     ASTContext *Context;
 
     Timer LLVMIRGeneration;
@@ -56,18 +56,17 @@ namespace clang {
     std::unique_ptr<llvm::Module> TheModule, LinkModule;
 
   public:
-    BackendConsumer(BackendAction action, DiagnosticsEngine &_Diags,
-                    const CodeGenOptions &compopts,
-                    const TargetOptions &targetopts,
-                    const LangOptions &langopts, bool TimePasses,
-                    const std::string &infile, llvm::Module *LinkModule,
-                    raw_ostream *OS, LLVMContext &C,
+    BackendConsumer(BackendAction Action, DiagnosticsEngine &Diags,
+                    const CodeGenOptions &CodeGenOpts,
+                    const TargetOptions &TargetOpts,
+                    const LangOptions &LangOpts, bool TimePasses,
+                    const std::string &InFile, llvm::Module *LinkModule,
+                    raw_pwrite_stream *OS, LLVMContext &C,
                     CoverageSourceInfo *CoverageInfo = nullptr)
-        : Diags(_Diags), Action(action), CodeGenOpts(compopts),
-          TargetOpts(targetopts), LangOpts(langopts), AsmOutStream(OS),
+        : Diags(Diags), Action(Action), CodeGenOpts(CodeGenOpts),
+          TargetOpts(TargetOpts), LangOpts(LangOpts), AsmOutStream(OS),
           Context(nullptr), LLVMIRGeneration("LLVM IR Generation Time"),
-          Gen(CreateLLVMCodeGen(Diags, infile, compopts,
-                                targetopts, C, CoverageInfo)),
+          Gen(CreateLLVMCodeGen(Diags, InFile, CodeGenOpts, C, CoverageInfo)),
           LinkModule(LinkModule) {
       llvm::TimePassesIsEnabled = TimePasses;
     }
@@ -80,6 +79,11 @@ namespace clang {
     }
 
     void Initialize(ASTContext &Ctx) override {
+      if (Context) {
+        assert(Context == &Ctx);
+        return;
+      }
+        
       Context = &Ctx;
 
       if (llvm::TimePassesIsEnabled)
@@ -196,8 +200,8 @@ namespace clang {
       Gen->CompleteTentativeDefinition(D);
     }
 
-    void HandleVTable(CXXRecordDecl *RD, bool DefinitionRequired) override {
-      Gen->HandleVTable(RD, DefinitionRequired);
+    void HandleVTable(CXXRecordDecl *RD) override {
+      Gen->HandleVTable(RD);
     }
 
     void HandleLinkerOptionPragma(llvm::StringRef Opts) override {
@@ -430,13 +434,16 @@ void BackendConsumer::EmitOptimizationMessage(
   FileManager &FileMgr = SourceMgr.getFileManager();
   StringRef Filename;
   unsigned Line, Column;
-  D.getLocation(&Filename, &Line, &Column);
   SourceLocation DILoc;
-  const FileEntry *FE = FileMgr.getFile(Filename);
-  if (FE && Line > 0) {
-    // If -gcolumn-info was not used, Column will be 0. This upsets the
-    // source manager, so pass 1 if Column is not set.
-    DILoc = SourceMgr.translateFileLineCol(FE, Line, Column ? Column : 1);
+
+  if (D.isLocationAvailable()) {
+    D.getLocation(&Filename, &Line, &Column);
+    const FileEntry *FE = FileMgr.getFile(Filename);
+    if (FE && Line > 0) {
+      // If -gcolumn-info was not used, Column will be 0. This upsets the
+      // source manager, so pass 1 if Column is not set.
+      DILoc = SourceMgr.translateFileLineCol(FE, Line, Column ? Column : 1);
+    }
   }
 
   // If a location isn't available, try to approximate it using the associated
@@ -451,7 +458,7 @@ void BackendConsumer::EmitOptimizationMessage(
       << AddFlagValue(D.getPassName() ? D.getPassName() : "")
       << D.getMsg().str();
 
-  if (DILoc.isInvalid())
+  if (DILoc.isInvalid() && D.isLocationAvailable())
     // If we were not able to translate the file:line:col information
     // back to a SourceLocation, at least emit a note stating that
     // we could not translate this location. This can happen in the
@@ -602,9 +609,8 @@ llvm::LLVMContext *CodeGenAction::takeLLVMContext() {
   return VMContext;
 }
 
-static raw_ostream *GetOutputStream(CompilerInstance &CI,
-                                    StringRef InFile,
-                                    BackendAction Action) {
+static raw_pwrite_stream *
+GetOutputStream(CompilerInstance &CI, StringRef InFile, BackendAction Action) {
   switch (Action) {
   case Backend_EmitAssembly:
     return CI.createDefaultOutputFile(false, InFile, "s");
@@ -626,7 +632,7 @@ static raw_ostream *GetOutputStream(CompilerInstance &CI,
 std::unique_ptr<ASTConsumer>
 CodeGenAction::CreateASTConsumer(CompilerInstance &CI, StringRef InFile) {
   BackendAction BA = static_cast<BackendAction>(Act);
-  std::unique_ptr<raw_ostream> OS(GetOutputStream(CI, InFile, BA));
+  raw_pwrite_stream *OS = GetOutputStream(CI, InFile, BA);
   if (BA != Backend_EmitNothing && !OS)
     return nullptr;
 
@@ -663,17 +669,23 @@ CodeGenAction::CreateASTConsumer(CompilerInstance &CI, StringRef InFile) {
   std::unique_ptr<BackendConsumer> Result(new BackendConsumer(
       BA, CI.getDiagnostics(), CI.getCodeGenOpts(), CI.getTargetOpts(),
       CI.getLangOpts(), CI.getFrontendOpts().ShowTimers, InFile,
-      LinkModuleToUse, OS.release(), *VMContext, CoverageInfo));
+      LinkModuleToUse, OS, *VMContext, CoverageInfo));
   BEConsumer = Result.get();
   return std::move(Result);
 }
 
+static void BitcodeInlineAsmDiagHandler(const llvm::SMDiagnostic &SM,
+                                         void *Context,
+                                         unsigned LocCookie) {
+  SM.print(nullptr, llvm::errs());
+}
+
 void CodeGenAction::ExecuteAction() {
   // If this is an IR file, we have to treat it specially.
   if (getCurrentFileKind() == IK_LLVM_IR) {
     BackendAction BA = static_cast<BackendAction>(Act);
     CompilerInstance &CI = getCompilerInstance();
-    raw_ostream *OS = GetOutputStream(CI, getCurrentFile(), BA);
+    raw_pwrite_stream *OS = GetOutputStream(CI, getCurrentFile(), BA);
     if (BA != Backend_EmitNothing && !OS)
       return;
 
@@ -710,14 +722,14 @@ void CodeGenAction::ExecuteAction() {
     }
     const TargetOptions &TargetOpts = CI.getTargetOpts();
     if (TheModule->getTargetTriple() != TargetOpts.Triple) {
-      unsigned DiagID = CI.getDiagnostics().getCustomDiagID(
-          DiagnosticsEngine::Warning,
-          "overriding the module target triple with %0");
-
-      CI.getDiagnostics().Report(SourceLocation(), DiagID) << TargetOpts.Triple;
+      CI.getDiagnostics().Report(SourceLocation(),
+                                 diag::warn_fe_override_module)
+          << TargetOpts.Triple;
       TheModule->setTargetTriple(TargetOpts.Triple);
     }
 
+    LLVMContext &Ctx = TheModule->getContext();
+    Ctx.setInlineAsmDiagnosticHandler(BitcodeInlineAsmDiagHandler);
     EmitBackendOutput(CI.getDiagnostics(), CI.getCodeGenOpts(), TargetOpts,
                       CI.getLangOpts(), CI.getTarget().getTargetDescription(),
                       TheModule.get(), BA, OS);
diff --git a/lib/CodeGen/CodeGenFunction.cpp b/lib/CodeGen/CodeGenFunction.cpp
index 826171a46e23..01da75005610 100644
--- a/lib/CodeGen/CodeGenFunction.cpp
+++ b/lib/CodeGen/CodeGenFunction.cpp
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "CodeGenFunction.h"
+#include "CGCleanup.h"
 #include "CGCUDARuntime.h"
 #include "CGCXXABI.h"
 #include "CGDebugInfo.h"
@@ -40,10 +41,11 @@ CodeGenFunction::CodeGenFunction(CodeGenModule &cgm, bool suppressNewContext)
       CurFn(nullptr), CapturedStmtInfo(nullptr),
       SanOpts(CGM.getLangOpts().Sanitize), IsSanitizerScope(false),
       CurFuncIsThunk(false), AutoreleaseResult(false), SawAsmBlock(false),
-      BlockInfo(nullptr), BlockPointer(nullptr),
+      IsOutlinedSEHHelper(false), BlockInfo(nullptr), BlockPointer(nullptr),
       LambdaThisCaptureField(nullptr), NormalCleanupDest(nullptr),
       NextCleanupDestIndex(1), FirstBlockInfo(nullptr), EHResumeBlock(nullptr),
       ExceptionSlot(nullptr), EHSelectorSlot(nullptr),
+      AbnormalTerminationSlot(nullptr), SEHPointersDecl(nullptr),
       DebugInfo(CGM.getModuleDebugInfo()), DisableDebugInfo(false),
       DidCallStackSave(false), IndirectBranch(nullptr), PGO(cgm),
       SwitchInsn(nullptr), SwitchWeights(nullptr), CaseRangeBlock(nullptr),
@@ -69,6 +71,9 @@ CodeGenFunction::CodeGenFunction(CodeGenModule &cgm, bool suppressNewContext)
   if (CGM.getCodeGenOpts().NoSignedZeros) {
     FMF.setNoSignedZeros();
   }
+  if (CGM.getCodeGenOpts().ReciprocalMath) {
+    FMF.setAllowReciprocal();
+  }
   Builder.SetFastMathFlags(FMF);
 }
 
@@ -82,7 +87,7 @@ CodeGenFunction::~CodeGenFunction() {
     destroyBlockInfos(FirstBlockInfo);
 
   if (getLangOpts().OpenMP) {
-    CGM.getOpenMPRuntime().FunctionFinished(*this);
+    CGM.getOpenMPRuntime().functionFinished(*this);
   }
 }
 
@@ -239,17 +244,18 @@ void CodeGenFunction::FinishFunction(SourceLocation EndLoc) {
   // parameters.  Do this in whatever block we're currently in; it's
   // important to do this before we enter the return block or return
   // edges will be *really* confused.
-  bool EmitRetDbgLoc = true;
-  if (EHStack.stable_begin() != PrologueCleanupDepth) {
-    PopCleanupBlocks(PrologueCleanupDepth);
-
+  bool HasCleanups = EHStack.stable_begin() != PrologueCleanupDepth;
+  bool HasOnlyLifetimeMarkers =
+      HasCleanups && EHStack.containsOnlyLifetimeMarkers(PrologueCleanupDepth);
+  bool EmitRetDbgLoc = !HasCleanups || HasOnlyLifetimeMarkers;
+  if (HasCleanups) {
     // Make sure the line table doesn't jump back into the body for
     // the ret after it's been at EndLoc.
-    EmitRetDbgLoc = false;
-
     if (CGDebugInfo *DI = getDebugInfo())
       if (OnlySimpleReturnStmts)
         DI->EmitLocation(Builder, EndLoc);
+
+    PopCleanupBlocks(PrologueCleanupDepth);
   }
 
   // Emit function epilog (to return).
@@ -278,6 +284,20 @@ void CodeGenFunction::FinishFunction(SourceLocation EndLoc) {
     Builder.ClearInsertionPoint();
   }
 
+  // If some of our locals escaped, insert a call to llvm.frameescape in the
+  // entry block.
+  if (!EscapedLocals.empty()) {
+    // Invert the map from local to index into a simple vector. There should be
+    // no holes.
+    SmallVector<llvm::Value *, 4> EscapeArgs;
+    EscapeArgs.resize(EscapedLocals.size());
+    for (auto &Pair : EscapedLocals)
+      EscapeArgs[Pair.second] = Pair.first;
+    llvm::Function *FrameEscapeFn = llvm::Intrinsic::getDeclaration(
+        &CGM.getModule(), llvm::Intrinsic::frameescape);
+    CGBuilderTy(AllocaInsertPt).CreateCall(FrameEscapeFn, EscapeArgs);
+  }
+
   // Remove the AllocaInsertPt instruction, which is just a convenience for us.
   llvm::Instruction *Ptr = AllocaInsertPt;
   AllocaInsertPt = nullptr;
@@ -588,6 +608,20 @@ void CodeGenFunction::StartFunction(GlobalDecl GD,
   if (CGM.isInSanitizerBlacklist(Fn, Loc))
     SanOpts.clear();
 
+  if (D) {
+    // Apply the no_sanitize* attributes to SanOpts.
+    for (auto Attr : D->specific_attrs<NoSanitizeAttr>())
+      SanOpts.Mask &= ~Attr->getMask();
+  }
+
+  // Apply sanitizer attributes to the function.
+  if (SanOpts.has(SanitizerKind::Address))
+    Fn->addFnAttr(llvm::Attribute::SanitizeAddress);
+  if (SanOpts.has(SanitizerKind::Thread))
+    Fn->addFnAttr(llvm::Attribute::SanitizeThread);
+  if (SanOpts.has(SanitizerKind::Memory))
+    Fn->addFnAttr(llvm::Attribute::SanitizeMemory);
+
   // Pass inline keyword to optimizer if it appears explicitly on any
   // declaration. Also, in the case of -fno-inline attach NoInline
   // attribute to all function that are not marked AlwaysInline.
@@ -679,7 +713,7 @@ void CodeGenFunction::StartFunction(GlobalDecl GD,
     unsigned Idx = CurFnInfo->getReturnInfo().getInAllocaFieldIndex();
     llvm::Function::arg_iterator EI = CurFn->arg_end();
     --EI;
-    llvm::Value *Addr = Builder.CreateStructGEP(EI, Idx);
+    llvm::Value *Addr = Builder.CreateStructGEP(nullptr, EI, Idx);
     ReturnValue = Builder.CreateLoad(Addr, "agg.result");
   } else {
     ReturnValue = CreateIRTemp(RetTy, "retval");
@@ -753,8 +787,7 @@ void CodeGenFunction::StartFunction(GlobalDecl GD,
 
 void CodeGenFunction::EmitFunctionBody(FunctionArgList &Args,
                                        const Stmt *Body) {
-  RegionCounter Cnt = getPGORegionCounter(Body);
-  Cnt.beginRegion(Builder);
+  incrementProfileCounter(Body);
   if (const CompoundStmt *S = dyn_cast<CompoundStmt>(Body))
     EmitCompoundStmtWithoutScope(*S);
   else
@@ -766,7 +799,7 @@ void CodeGenFunction::EmitFunctionBody(FunctionArgList &Args,
 /// emit a branch around the instrumentation code. When not instrumenting,
 /// this just calls EmitBlock().
 void CodeGenFunction::EmitBlockWithFallThrough(llvm::BasicBlock *BB,
-                                               RegionCounter &Cnt) {
+                                               const Stmt *S) {
   llvm::BasicBlock *SkipCountBB = nullptr;
   if (HaveInsertPoint() && CGM.getCodeGenOpts().ProfileInstrGenerate) {
     // When instrumenting for profiling, the fallthrough to certain
@@ -776,7 +809,9 @@ void CodeGenFunction::EmitBlockWithFallThrough(llvm::BasicBlock *BB,
     EmitBranch(SkipCountBB);
   }
   EmitBlock(BB);
-  Cnt.beginRegion(Builder, /*AddIncomingFallThrough=*/true);
+  uint64_t CurrentCount = getCurrentProfileCount();
+  incrementProfileCounter(S);
+  setCurrentProfileCount(getCurrentProfileCount() + CurrentCount);
   if (SkipCountBB)
     EmitBlock(SkipCountBB);
 }
@@ -801,17 +836,6 @@ static void TryMarkNoThrow(llvm::Function *F) {
   F->setDoesNotThrow();
 }
 
-static void EmitSizedDeallocationFunction(CodeGenFunction &CGF,
-                                          const FunctionDecl *UnsizedDealloc) {
-  // This is a weak discardable definition of the sized deallocation function.
-  CGF.CurFn->setLinkage(llvm::Function::LinkOnceAnyLinkage);
-
-  // Call the unsized deallocation function and forward the first argument
-  // unchanged.
-  llvm::Constant *Unsized = CGF.CGM.GetAddrOfFunction(UnsizedDealloc);
-  CGF.Builder.CreateCall(Unsized, &*CGF.CurFn->arg_begin());
-}
-
 void CodeGenFunction::GenerateCode(GlobalDecl GD, llvm::Function *Fn,
                                    const CGFunctionInfo &FnInfo) {
   const FunctionDecl *FD = cast<FunctionDecl>(GD.getDecl());
@@ -832,7 +856,7 @@ void CodeGenFunction::GenerateCode(GlobalDecl GD, llvm::Function *Fn,
       ResTy = CGM.getContext().VoidPtrTy;
     CGM.getCXXABI().buildThisParam(*this, Args);
   }
-  
+
   Args.append(FD->param_begin(), FD->param_end());
 
   if (MD && (isa<CXXConstructorDecl>(MD) || isa<CXXDestructorDecl>(MD)))
@@ -866,9 +890,9 @@ void CodeGenFunction::GenerateCode(GlobalDecl GD, llvm::Function *Fn,
   else if (isa<CXXConstructorDecl>(FD))
     EmitConstructorBody(Args);
   else if (getLangOpts().CUDA &&
-           !CGM.getCodeGenOpts().CUDAIsDevice &&
+           !getLangOpts().CUDAIsDevice &&
            FD->hasAttr<CUDAGlobalAttr>())
-    CGM.getCUDARuntime().EmitDeviceStubBody(*this, Args);
+    CGM.getCUDARuntime().emitDeviceStub(*this, Args);
   else if (isa<CXXConversionDecl>(FD) &&
            cast<CXXConversionDecl>(FD)->isLambdaToBlockPointerConversion()) {
     // The lambda conversion to block pointer is special; the semantics can't be
@@ -887,11 +911,6 @@ void CodeGenFunction::GenerateCode(GlobalDecl GD, llvm::Function *Fn,
     emitImplicitAssignmentOperatorBody(Args);
   } else if (Stmt *Body = FD->getBody()) {
     EmitFunctionBody(Args, Body);
-  } else if (FunctionDecl *UnsizedDealloc =
-                 FD->getCorrespondingUnsizedGlobalDeallocationFunction()) {
-    // Global sized deallocation functions get an implicit weak definition if
-    // they don't have an explicit definition.
-    EmitSizedDeallocationFunction(*this, UnsizedDealloc);
   } else
     llvm_unreachable("no definition for emitted function");
 
@@ -910,7 +929,7 @@ void CodeGenFunction::GenerateCode(GlobalDecl GD, llvm::Function *Fn,
                 "missing_return", EmitCheckSourceLocation(FD->getLocation()),
                 None);
     } else if (CGM.getCodeGenOpts().OptimizationLevel == 0)
-      Builder.CreateCall(CGM.getIntrinsic(llvm::Intrinsic::trap));
+      Builder.CreateCall(CGM.getIntrinsic(llvm::Intrinsic::trap), {});
     Builder.CreateUnreachable();
     Builder.ClearInsertionPoint();
   }
@@ -1028,15 +1047,13 @@ void CodeGenFunction::EmitBranchOnBoolExpr(const Expr *Cond,
 
     // Handle X && Y in a condition.
     if (CondBOp->getOpcode() == BO_LAnd) {
-      RegionCounter Cnt = getPGORegionCounter(CondBOp);
-
       // If we have "1 && X", simplify the code.  "0 && X" would have constant
       // folded if the case was simple enough.
       bool ConstantBool = false;
       if (ConstantFoldsToSimpleInteger(CondBOp->getLHS(), ConstantBool) &&
           ConstantBool) {
         // br(1 && X) -> br(X).
-        Cnt.beginRegion(Builder);
+        incrementProfileCounter(CondBOp);
         return EmitBranchOnBoolExpr(CondBOp->getRHS(), TrueBlock, FalseBlock,
                                     TrueCount);
       }
@@ -1055,14 +1072,19 @@ void CodeGenFunction::EmitBranchOnBoolExpr(const Expr *Cond,
       llvm::BasicBlock *LHSTrue = createBasicBlock("land.lhs.true");
       // The counter tells us how often we evaluate RHS, and all of TrueCount
       // can be propagated to that branch.
-      uint64_t RHSCount = Cnt.getCount();
+      uint64_t RHSCount = getProfileCount(CondBOp->getRHS());
 
       ConditionalEvaluation eval(*this);
-      EmitBranchOnBoolExpr(CondBOp->getLHS(), LHSTrue, FalseBlock, RHSCount);
-      EmitBlock(LHSTrue);
+      {
+        ApplyDebugLocation DL(*this, Cond);
+        EmitBranchOnBoolExpr(CondBOp->getLHS(), LHSTrue, FalseBlock, RHSCount);
+        EmitBlock(LHSTrue);
+      }
+
+      incrementProfileCounter(CondBOp);
+      setCurrentProfileCount(getProfileCount(CondBOp->getRHS()));
 
       // Any temporaries created here are conditional.
-      Cnt.beginRegion(Builder);
       eval.begin(*this);
       EmitBranchOnBoolExpr(CondBOp->getRHS(), TrueBlock, FalseBlock, TrueCount);
       eval.end(*this);
@@ -1071,15 +1093,13 @@ void CodeGenFunction::EmitBranchOnBoolExpr(const Expr *Cond,
     }
 
     if (CondBOp->getOpcode() == BO_LOr) {
-      RegionCounter Cnt = getPGORegionCounter(CondBOp);
-
       // If we have "0 || X", simplify the code.  "1 || X" would have constant
       // folded if the case was simple enough.
       bool ConstantBool = false;
       if (ConstantFoldsToSimpleInteger(CondBOp->getLHS(), ConstantBool) &&
           !ConstantBool) {
         // br(0 || X) -> br(X).
-        Cnt.beginRegion(Builder);
+        incrementProfileCounter(CondBOp);
         return EmitBranchOnBoolExpr(CondBOp->getRHS(), TrueBlock, FalseBlock,
                                     TrueCount);
       }
@@ -1099,15 +1119,21 @@ void CodeGenFunction::EmitBranchOnBoolExpr(const Expr *Cond,
       // We have the count for entry to the RHS and for the whole expression
       // being true, so we can divy up True count between the short circuit and
       // the RHS.
-      uint64_t LHSCount = Cnt.getParentCount() - Cnt.getCount();
+      uint64_t LHSCount =
+          getCurrentProfileCount() - getProfileCount(CondBOp->getRHS());
       uint64_t RHSCount = TrueCount - LHSCount;
 
       ConditionalEvaluation eval(*this);
-      EmitBranchOnBoolExpr(CondBOp->getLHS(), TrueBlock, LHSFalse, LHSCount);
-      EmitBlock(LHSFalse);
+      {
+        ApplyDebugLocation DL(*this, Cond);
+        EmitBranchOnBoolExpr(CondBOp->getLHS(), TrueBlock, LHSFalse, LHSCount);
+        EmitBlock(LHSFalse);
+      }
+
+      incrementProfileCounter(CondBOp);
+      setCurrentProfileCount(getProfileCount(CondBOp->getRHS()));
 
       // Any temporaries created here are conditional.
-      Cnt.beginRegion(Builder);
       eval.begin(*this);
       EmitBranchOnBoolExpr(CondBOp->getRHS(), TrueBlock, FalseBlock, RHSCount);
 
@@ -1121,7 +1147,7 @@ void CodeGenFunction::EmitBranchOnBoolExpr(const Expr *Cond,
     // br(!x, t, f) -> br(x, f, t)
     if (CondUOp->getOpcode() == UO_LNot) {
       // Negate the count.
-      uint64_t FalseCount = PGO.getCurrentRegionCount() - TrueCount;
+      uint64_t FalseCount = getCurrentProfileCount() - TrueCount;
       // Negate the condition and swap the destination blocks.
       return EmitBranchOnBoolExpr(CondUOp->getSubExpr(), FalseBlock, TrueBlock,
                                   FalseCount);
@@ -1133,9 +1159,9 @@ void CodeGenFunction::EmitBranchOnBoolExpr(const Expr *Cond,
     llvm::BasicBlock *LHSBlock = createBasicBlock("cond.true");
     llvm::BasicBlock *RHSBlock = createBasicBlock("cond.false");
 
-    RegionCounter Cnt = getPGORegionCounter(CondOp);
     ConditionalEvaluation cond(*this);
-    EmitBranchOnBoolExpr(CondOp->getCond(), LHSBlock, RHSBlock, Cnt.getCount());
+    EmitBranchOnBoolExpr(CondOp->getCond(), LHSBlock, RHSBlock,
+                         getProfileCount(CondOp));
 
     // When computing PGO branch weights, we only know the overall count for
     // the true block. This code is essentially doing tail duplication of the
@@ -1144,15 +1170,19 @@ void CodeGenFunction::EmitBranchOnBoolExpr(const Expr *Cond,
     // the conditional operator.
     uint64_t LHSScaledTrueCount = 0;
     if (TrueCount) {
-      double LHSRatio = Cnt.getCount() / (double) Cnt.getParentCount();
+      double LHSRatio =
+          getProfileCount(CondOp) / (double)getCurrentProfileCount();
       LHSScaledTrueCount = TrueCount * LHSRatio;
     }
 
     cond.begin(*this);
     EmitBlock(LHSBlock);
-    Cnt.beginRegion(Builder);
-    EmitBranchOnBoolExpr(CondOp->getLHS(), TrueBlock, FalseBlock,
-                         LHSScaledTrueCount);
+    incrementProfileCounter(CondOp);
+    {
+      ApplyDebugLocation DL(*this, Cond);
+      EmitBranchOnBoolExpr(CondOp->getLHS(), TrueBlock, FalseBlock,
+                           LHSScaledTrueCount);
+    }
     cond.end(*this);
 
     cond.begin(*this);
@@ -1176,12 +1206,16 @@ void CodeGenFunction::EmitBranchOnBoolExpr(const Expr *Cond,
 
   // Create branch weights based on the number of times we get here and the
   // number of times the condition should be true.
-  uint64_t CurrentCount = std::max(PGO.getCurrentRegionCount(), TrueCount);
-  llvm::MDNode *Weights = PGO.createBranchWeights(TrueCount,
-                                                  CurrentCount - TrueCount);
+  uint64_t CurrentCount = std::max(getCurrentProfileCount(), TrueCount);
+  llvm::MDNode *Weights =
+      createProfileWeights(TrueCount, CurrentCount - TrueCount);
 
   // Emit the code with the fully general case.
-  llvm::Value *CondV = EvaluateExprAsBool(Cond);
+  llvm::Value *CondV;
+  {
+    ApplyDebugLocation DL(*this, Cond);
+    CondV = EvaluateExprAsBool(Cond);
+  }
   Builder.CreateCondBr(CondV, TrueBlock, FalseBlock, Weights);
 }
 
@@ -1231,7 +1265,8 @@ static void emitNonZeroVLAInit(CodeGenFunction &CGF, QualType baseType,
                        /*volatile*/ false);
 
   // Go to the next element.
-  llvm::Value *next = Builder.CreateConstInBoundsGEP1_32(cur, 1, "vla.next");
+  llvm::Value *next = Builder.CreateConstInBoundsGEP1_32(Builder.getInt8Ty(),
+                                                         cur, 1, "vla.next");
 
   // Leave if that's the end of the VLA.
   llvm::Value *done = Builder.CreateICmpEQ(next, end, "vla-init.isdone");
diff --git a/lib/CodeGen/CodeGenFunction.h b/lib/CodeGen/CodeGenFunction.h
index 3a990d21490d..650ad7b04009 100644
--- a/lib/CodeGen/CodeGenFunction.h
+++ b/lib/CodeGen/CodeGenFunction.h
@@ -27,6 +27,7 @@
 #include "clang/AST/Type.h"
 #include "clang/Basic/ABI.h"
 #include "clang/Basic/CapturedStmt.h"
+#include "clang/Basic/OpenMPKinds.h"
 #include "clang/Basic/TargetInfo.h"
 #include "clang/Frontend/CodeGenOptions.h"
 #include "llvm/ADT/ArrayRef.h"
@@ -96,8 +97,8 @@ enum TypeEvaluationKind {
 /// CodeGenFunction - This class organizes the per-function state that is used
 /// while generating LLVM code.
 class CodeGenFunction : public CodeGenTypeCache {
-  CodeGenFunction(const CodeGenFunction &) LLVM_DELETED_FUNCTION;
-  void operator=(const CodeGenFunction &) LLVM_DELETED_FUNCTION;
+  CodeGenFunction(const CodeGenFunction &) = delete;
+  void operator=(const CodeGenFunction &) = delete;
 
   friend class CGCXXABI;
 public:
@@ -191,26 +192,25 @@ public:
 
     CapturedRegionKind getKind() const { return Kind; }
 
-    void setContextValue(llvm::Value *V) { ThisValue = V; }
+    virtual void setContextValue(llvm::Value *V) { ThisValue = V; }
     // \brief Retrieve the value of the context parameter.
-    llvm::Value *getContextValue() const { return ThisValue; }
+    virtual llvm::Value *getContextValue() const { return ThisValue; }
 
     /// \brief Lookup the captured field decl for a variable.
-    const FieldDecl *lookup(const VarDecl *VD) const {
+    virtual const FieldDecl *lookup(const VarDecl *VD) const {
       return CaptureFields.lookup(VD);
     }
 
-    bool isCXXThisExprCaptured() const { return CXXThisFieldDecl != nullptr; }
-    FieldDecl *getThisFieldDecl() const { return CXXThisFieldDecl; }
+    bool isCXXThisExprCaptured() const { return getThisFieldDecl() != nullptr; }
+    virtual FieldDecl *getThisFieldDecl() const { return CXXThisFieldDecl; }
 
     static bool classof(const CGCapturedStmtInfo *) {
       return true;
     }
 
     /// \brief Emit the captured statement body.
-    virtual void EmitBody(CodeGenFunction &CGF, Stmt *S) {
-      RegionCounter Cnt = CGF.getPGORegionCounter(S);
-      Cnt.beginRegion(CGF.Builder);
+    virtual void EmitBody(CodeGenFunction &CGF, const Stmt *S) {
+      CGF.incrementProfileCounter(S);
       CGF.EmitStmt(S);
     }
 
@@ -262,6 +262,10 @@ public:
   /// potentially set the return value.
   bool SawAsmBlock;
 
+  /// True if the current function is an outlined SEH helper. This can be a
+  /// finally block or filter expression.
+  bool IsOutlinedSEHHelper;
+
   const CodeGen::CGBlockInfo *BlockInfo;
   llvm::Value *BlockPointer;
 
@@ -274,6 +278,7 @@ public:
 
   EHScopeStack EHStack;
   llvm::SmallVector<char, 256> LifetimeExtendedCleanupStack;
+  llvm::SmallVector<const JumpDest *, 2> SEHTryEpilogueStack;
 
   /// Header for data within LifetimeExtendedCleanupStack.
   struct LifetimeExtendedCleanupHeader {
@@ -305,6 +310,12 @@ public:
   /// write the current selector value into this alloca.
   llvm::AllocaInst *EHSelectorSlot;
 
+  llvm::AllocaInst *AbnormalTerminationSlot;
+
+  /// The implicit parameter to SEH filter functions of type
+  /// 'EXCEPTION_POINTERS*'.
+  ImplicitParamDecl *SEHPointersDecl;
+
   /// Emits a landing pad for the current EH stack.
   llvm::BasicBlock *EmitLandingPad();
 
@@ -343,87 +354,32 @@ public:
     void exit(CodeGenFunction &CGF);
   };
 
-  /// pushFullExprCleanup - Push a cleanup to be run at the end of the
-  /// current full-expression.  Safe against the possibility that
-  /// we're currently inside a conditionally-evaluated expression.
-  template <class T, class A0>
-  void pushFullExprCleanup(CleanupKind kind, A0 a0) {
-    // If we're not in a conditional branch, or if none of the
-    // arguments requires saving, then use the unconditional cleanup.
-    if (!isInConditionalBranch())
-      return EHStack.pushCleanup<T>(kind, a0);
-
-    typename DominatingValue<A0>::saved_type a0_saved = saveValueInCond(a0);
-
-    typedef EHScopeStack::ConditionalCleanup1<T, A0> CleanupType;
-    EHStack.pushCleanup<CleanupType>(kind, a0_saved);
-    initFullExprCleanup();
-  }
+  /// Returns true inside SEH __try blocks.
+  bool isSEHTryScope() const { return !SEHTryEpilogueStack.empty(); }
 
   /// pushFullExprCleanup - Push a cleanup to be run at the end of the
   /// current full-expression.  Safe against the possibility that
   /// we're currently inside a conditionally-evaluated expression.
-  template <class T, class A0, class A1>
-  void pushFullExprCleanup(CleanupKind kind, A0 a0, A1 a1) {
+  template <class T, class... As>
+  void pushFullExprCleanup(CleanupKind kind, As... A) {
     // If we're not in a conditional branch, or if none of the
     // arguments requires saving, then use the unconditional cleanup.
     if (!isInConditionalBranch())
-      return EHStack.pushCleanup<T>(kind, a0, a1);
+      return EHStack.pushCleanup<T>(kind, A...);
 
-    typename DominatingValue<A0>::saved_type a0_saved = saveValueInCond(a0);
-    typename DominatingValue<A1>::saved_type a1_saved = saveValueInCond(a1);
+    // Stash values in a tuple so we can guarantee the order of saves.
+    typedef std::tuple<typename DominatingValue<As>::saved_type...> SavedTuple;
+    SavedTuple Saved{saveValueInCond(A)...};
 
-    typedef EHScopeStack::ConditionalCleanup2<T, A0, A1> CleanupType;
-    EHStack.pushCleanup<CleanupType>(kind, a0_saved, a1_saved);
-    initFullExprCleanup();
-  }
-
-  /// pushFullExprCleanup - Push a cleanup to be run at the end of the
-  /// current full-expression.  Safe against the possibility that
-  /// we're currently inside a conditionally-evaluated expression.
-  template <class T, class A0, class A1, class A2>
-  void pushFullExprCleanup(CleanupKind kind, A0 a0, A1 a1, A2 a2) {
-    // If we're not in a conditional branch, or if none of the
-    // arguments requires saving, then use the unconditional cleanup.
-    if (!isInConditionalBranch()) {
-      return EHStack.pushCleanup<T>(kind, a0, a1, a2);
-    }
-    
-    typename DominatingValue<A0>::saved_type a0_saved = saveValueInCond(a0);
-    typename DominatingValue<A1>::saved_type a1_saved = saveValueInCond(a1);
-    typename DominatingValue<A2>::saved_type a2_saved = saveValueInCond(a2);
-    
-    typedef EHScopeStack::ConditionalCleanup3<T, A0, A1, A2> CleanupType;
-    EHStack.pushCleanup<CleanupType>(kind, a0_saved, a1_saved, a2_saved);
-    initFullExprCleanup();
-  }
-
-  /// pushFullExprCleanup - Push a cleanup to be run at the end of the
-  /// current full-expression.  Safe against the possibility that
-  /// we're currently inside a conditionally-evaluated expression.
-  template <class T, class A0, class A1, class A2, class A3>
-  void pushFullExprCleanup(CleanupKind kind, A0 a0, A1 a1, A2 a2, A3 a3) {
-    // If we're not in a conditional branch, or if none of the
-    // arguments requires saving, then use the unconditional cleanup.
-    if (!isInConditionalBranch()) {
-      return EHStack.pushCleanup<T>(kind, a0, a1, a2, a3);
-    }
-    
-    typename DominatingValue<A0>::saved_type a0_saved = saveValueInCond(a0);
-    typename DominatingValue<A1>::saved_type a1_saved = saveValueInCond(a1);
-    typename DominatingValue<A2>::saved_type a2_saved = saveValueInCond(a2);
-    typename DominatingValue<A3>::saved_type a3_saved = saveValueInCond(a3);
-    
-    typedef EHScopeStack::ConditionalCleanup4<T, A0, A1, A2, A3> CleanupType;
-    EHStack.pushCleanup<CleanupType>(kind, a0_saved, a1_saved,
-                                     a2_saved, a3_saved);
+    typedef EHScopeStack::ConditionalCleanup<T, As...> CleanupType;
+    EHStack.pushCleanupTuple<CleanupType>(kind, Saved);
     initFullExprCleanup();
   }
 
   /// \brief Queue a cleanup to be pushed after finishing the current
   /// full-expression.
-  template <class T, class A0, class A1, class A2, class A3>
-  void pushCleanupAfterFullExpr(CleanupKind Kind, A0 a0, A1 a1, A2 a2, A3 a3) {
+  template <class T, class... As>
+  void pushCleanupAfterFullExpr(CleanupKind Kind, As... A) {
     assert(!isInConditionalBranch() && "can't defer conditional cleanup");
 
     LifetimeExtendedCleanupHeader Header = { sizeof(T), Kind };
@@ -434,7 +390,7 @@ public:
 
     char *Buffer = &LifetimeExtendedCleanupStack[OldSize];
     new (Buffer) LifetimeExtendedCleanupHeader(Header);
-    new (Buffer + sizeof(Header)) T(a0, a1, a2, a3);
+    new (Buffer + sizeof(Header)) T(A...);
   }
 
   /// Set up the last cleaup that was pushed as a conditional
@@ -488,8 +444,8 @@ public:
     bool PerformCleanup;
   private:
 
-    RunCleanupsScope(const RunCleanupsScope &) LLVM_DELETED_FUNCTION;
-    void operator=(const RunCleanupsScope &) LLVM_DELETED_FUNCTION;
+    RunCleanupsScope(const RunCleanupsScope &) = delete;
+    void operator=(const RunCleanupsScope &) = delete;
 
   protected:
     CodeGenFunction& CGF;
@@ -537,8 +493,8 @@ public:
     SmallVector<const LabelDecl*, 4> Labels;
     LexicalScope *ParentScope;
 
-    LexicalScope(const LexicalScope &) LLVM_DELETED_FUNCTION;
-    void operator=(const LexicalScope &) LLVM_DELETED_FUNCTION;
+    LexicalScope(const LexicalScope &) = delete;
+    void operator=(const LexicalScope &) = delete;
 
   public:
     /// \brief Enter a new cleanup scope.
@@ -562,7 +518,10 @@ public:
 
       // If we should perform a cleanup, force them now.  Note that
       // this ends the cleanup scope before rescoping any labels.
-      if (PerformCleanup) ForceCleanup();
+      if (PerformCleanup) {
+        ApplyDebugLocation DL(CGF, Range.getEnd());
+        ForceCleanup();
+      }
     }
 
     /// \brief Force the emission of cleanups now, instead of waiting
@@ -587,8 +546,8 @@ public:
     VarDeclMapTy SavedPrivates;
 
   private:
-    OMPPrivateScope(const OMPPrivateScope &) LLVM_DELETED_FUNCTION;
-    void operator=(const OMPPrivateScope &) LLVM_DELETED_FUNCTION;
+    OMPPrivateScope(const OMPPrivateScope &) = delete;
+    void operator=(const OMPPrivateScope &) = delete;
 
   public:
     /// \brief Enter a new OpenMP private scope.
@@ -637,7 +596,10 @@ public:
     }
 
     /// \brief Exit scope - all the mapped variables are restored.
-    ~OMPPrivateScope() { ForceCleanup(); }
+    ~OMPPrivateScope() {
+      if (PerformCleanup)
+        ForceCleanup();
+    }
   };
 
   /// \brief Takes the old cleanup stack size and emits the cleanup blocks
@@ -881,7 +843,8 @@ public:
   
   /// getByrefValueFieldNumber - Given a declaration, returns the LLVM field
   /// number that holds the value.
-  unsigned getByRefValueLLVMField(const ValueDecl *VD) const;
+  std::pair<llvm::Type *, unsigned>
+  getByRefValueLLVMField(const ValueDecl *VD) const;
 
   /// BuildBlockByrefAddress - Computes address location of the
   /// variable which is declared as __block.
@@ -906,6 +869,10 @@ private:
   typedef llvm::DenseMap<const Decl*, llvm::Value*> DeclMapTy;
   DeclMapTy LocalDeclMap;
 
+  /// Track escaped local variables with auto storage. Used during SEH
+  /// outlining to produce a call to llvm.frameescape.
+  llvm::DenseMap<llvm::AllocaInst *, int> EscapedLocals;
+
   /// LabelMap - This keeps track of the LLVM basic block for each C label.
   llvm::DenseMap<const LabelDecl*, JumpDest> LabelMap;
 
@@ -922,12 +889,39 @@ private:
 
   CodeGenPGO PGO;
 
+  /// Calculate branch weights appropriate for PGO data
+  llvm::MDNode *createProfileWeights(uint64_t TrueCount, uint64_t FalseCount);
+  llvm::MDNode *createProfileWeights(ArrayRef<uint64_t> Weights);
+  llvm::MDNode *createProfileWeightsForLoop(const Stmt *Cond,
+                                            uint64_t LoopCount);
+
 public:
-  /// Get a counter for instrumentation of the region associated with the given
-  /// statement.
-  RegionCounter getPGORegionCounter(const Stmt *S) {
-    return RegionCounter(PGO, S);
+  /// Increment the profiler's counter for the given statement.
+  void incrementProfileCounter(const Stmt *S) {
+    if (CGM.getCodeGenOpts().ProfileInstrGenerate)
+      PGO.emitCounterIncrement(Builder, S);
+    PGO.setCurrentStmt(S);
+  }
+
+  /// Get the profiler's count for the given statement.
+  uint64_t getProfileCount(const Stmt *S) {
+    Optional<uint64_t> Count = PGO.getStmtCount(S);
+    if (!Count.hasValue())
+      return 0;
+    return *Count;
+  }
+
+  /// Set the profiler's current count.
+  void setCurrentProfileCount(uint64_t Count) {
+    PGO.setCurrentRegionCount(Count);
+  }
+
+  /// Get the profiler's current count. This is generally the count for the most
+  /// recently incremented counter.
+  uint64_t getCurrentProfileCount() {
+    return PGO.getCurrentRegionCount();
   }
+
 private:
 
   /// SwitchInsn - This is nearest current switch instruction. It is null if
@@ -1097,6 +1091,11 @@ public:
     return getInvokeDestImpl();
   }
 
+  bool currentFunctionUsesSEHTry() const {
+    const auto *FD = dyn_cast_or_null<FunctionDecl>(CurCodeDecl);
+    return FD && FD->usesSEHTry();
+  }
+
   const TargetInfo &getTarget() const { return Target; }
   llvm::LLVMContext &getLLVMContext() { return CGM.getLLVMContext(); }
 
@@ -1248,7 +1247,7 @@ public:
   void EmitDestructorBody(FunctionArgList &Args);
   void emitImplicitAssignmentOperatorBody(FunctionArgList &Args);
   void EmitFunctionBody(FunctionArgList &Args, const Stmt *Body);
-  void EmitBlockWithFallThrough(llvm::BasicBlock *BB, RegionCounter &Cnt);
+  void EmitBlockWithFallThrough(llvm::BasicBlock *BB, const Stmt *S);
 
   void EmitForwardingCallToLambda(const CXXMethodDecl *LambdaCallOperator,
                                   CallArgList &CallArgs);
@@ -1258,15 +1257,18 @@ public:
   void EmitLambdaStaticInvokeFunction(const CXXMethodDecl *MD);
   void EmitAsanPrologueOrEpilogue(bool Prologue);
 
-  /// EmitReturnBlock - Emit the unified return block, trying to avoid its
-  /// emission when possible.
+  /// \brief Emit the unified return block, trying to avoid its emission when
+  /// possible.
+  /// \return The debug location of the user written return statement if the
+  /// return block is is avoided.
   llvm::DebugLoc EmitReturnBlock();
 
   /// FinishFunction - Complete IR generation of the current function. It is
   /// legal to call this function even if there is no current insertion point.
   void FinishFunction(SourceLocation EndLoc=SourceLocation());
 
-  void StartThunk(llvm::Function *Fn, GlobalDecl GD, const CGFunctionInfo &FnInfo);
+  void StartThunk(llvm::Function *Fn, GlobalDecl GD,
+                  const CGFunctionInfo &FnInfo);
 
   void EmitCallAndReturnForThunk(llvm::Value *Callee, const ThunkInfo *Thunk);
 
@@ -1309,6 +1311,19 @@ public:
   /// to by This.
   llvm::Value *GetVTablePtr(llvm::Value *This, llvm::Type *Ty);
 
+  /// \brief Derived is the presumed address of an object of type T after a
+  /// cast. If T is a polymorphic class type, emit a check that the virtual
+  /// table for Derived belongs to a class derived from T.
+  void EmitVTablePtrCheckForCast(QualType T, llvm::Value *Derived,
+                                 bool MayBeNull);
+
+  /// EmitVTablePtrCheckForCall - Virtual method MD is being called via VTable.
+  /// If vptr CFI is enabled, emit a check that VTable is valid.
+  void EmitVTablePtrCheckForCall(const CXXMethodDecl *MD, llvm::Value *VTable);
+
+  /// EmitVTablePtrCheck - Emit a check that VTable is a valid virtual table for
+  /// RD using llvm.bitset.test.
+  void EmitVTablePtrCheck(const CXXRecordDecl *RD, llvm::Value *VTable);
 
   /// CanDevirtualizeMemberFunctionCalls - Checks whether virtual calls on given
   /// expr can be devirtualized.
@@ -1527,6 +1542,8 @@ public:
   void EmitAnyExprToMem(const Expr *E, llvm::Value *Location,
                         Qualifiers Quals, bool IsInitializer);
 
+  void EmitAnyExprToExn(const Expr *E, llvm::Value *Addr);
+
   /// EmitExprAsInit - Emits the code necessary to initialize a
   /// location in memory with the given initializer.
   void EmitExprAsInit(const Expr *init, const ValueDecl *D, LValue lvalue,
@@ -1552,6 +1569,15 @@ public:
                       true);
   }
 
+  void EmitAggregateCopyCtor(llvm::Value *DestPtr, llvm::Value *SrcPtr,
+                           QualType DestTy, QualType SrcTy) {
+    CharUnits DestTypeAlign = getContext().getTypeAlignInChars(DestTy);
+    CharUnits SrcTypeAlign = getContext().getTypeAlignInChars(SrcTy);
+    EmitAggregateCopy(DestPtr, SrcPtr, SrcTy, /*IsVolatile=*/false,
+                      std::min(DestTypeAlign, SrcTypeAlign),
+                      /*IsAssignment=*/false);
+  }
+
   /// EmitAggregateCopy - Emit an aggregate copy.
   ///
   /// \param isVolatile - True iff either the source or the destination is
@@ -1723,12 +1749,16 @@ public:
                              llvm::Value *This);
 
   void EmitNewArrayInitializer(const CXXNewExpr *E, QualType elementType,
-                               llvm::Value *NewPtr, llvm::Value *NumElements,
+                               llvm::Type *ElementTy, llvm::Value *NewPtr,
+                               llvm::Value *NumElements,
                                llvm::Value *AllocSizeWithoutCookie);
 
   void EmitCXXTemporary(const CXXTemporary *Temporary, QualType TempType,
                         llvm::Value *Ptr);
 
+  llvm::Value *EmitLifetimeStart(uint64_t Size, llvm::Value *Addr);
+  void EmitLifetimeEnd(llvm::Value *Size, llvm::Value *Addr);
+
   llvm::Value *EmitCXXNewExpr(const CXXNewExpr *E);
   void EmitCXXDeleteExpr(const CXXDeleteExpr *E);
 
@@ -1889,8 +1919,8 @@ public:
     llvm::Value *getObjectAddress(CodeGenFunction &CGF) const {
       if (!IsByRef) return Address;
 
-      return CGF.Builder.CreateStructGEP(Address,
-                                         CGF.getByRefValueLLVMField(Variable),
+      auto F = CGF.getByRefValueLLVMField(Variable);
+      return CGF.Builder.CreateStructGEP(F.first, Address, F.second,
                                          Variable->getNameAsString());
     }
   };
@@ -1988,6 +2018,30 @@ public:
   void EmitCXXTryStmt(const CXXTryStmt &S);
   void EmitSEHTryStmt(const SEHTryStmt &S);
   void EmitSEHLeaveStmt(const SEHLeaveStmt &S);
+  void EnterSEHTryStmt(const SEHTryStmt &S);
+  void ExitSEHTryStmt(const SEHTryStmt &S);
+
+  void startOutlinedSEHHelper(CodeGenFunction &ParentCGF, StringRef Name,
+                              QualType RetTy, FunctionArgList &Args,
+                              const Stmt *OutlinedStmt);
+
+  llvm::Function *GenerateSEHFilterFunction(CodeGenFunction &ParentCGF,
+                                            const SEHExceptStmt &Except);
+
+  llvm::Function *GenerateSEHFinallyFunction(CodeGenFunction &ParentCGF,
+                                             const SEHFinallyStmt &Finally);
+
+  void EmitSEHExceptionCodeSave();
+  llvm::Value *EmitSEHExceptionCode();
+  llvm::Value *EmitSEHExceptionInfo();
+  llvm::Value *EmitSEHAbnormalTermination();
+
+  /// Scan the outlined statement for captures from the parent function. For
+  /// each capture, mark the capture as escaped and emit a call to
+  /// llvm.framerecover. Insert the framerecover result into the LocalDeclMap.
+  void EmitCapturedLocals(CodeGenFunction &ParentCGF, const Stmt *OutlinedStmt,
+                          llvm::Value *ParentFP);
+
   void EmitCXXForRangeStmt(const CXXForRangeStmt &S,
                            ArrayRef<const Attr *> Attrs = None);
 
@@ -1997,13 +2051,102 @@ public:
   llvm::Function *GenerateCapturedStmtFunctionEpilog(const CapturedStmt &S);
   llvm::Function *GenerateCapturedStmtFunction(const CapturedStmt &S);
   llvm::Value *GenerateCapturedStmtArgument(const CapturedStmt &S);
-  void EmitOMPAggregateAssign(LValue OriginalAddr, llvm::Value *PrivateAddr,
-                              const Expr *AssignExpr, QualType Type,
-                              const VarDecl *VDInit);
-  void EmitOMPFirstprivateClause(const OMPExecutableDirective &D,
+  /// \brief Perform element by element copying of arrays with type \a
+  /// OriginalType from \a SrcAddr to \a DestAddr using copying procedure
+  /// generated by \a CopyGen.
+  ///
+  /// \param DestAddr Address of the destination array.
+  /// \param SrcAddr Address of the source array.
+  /// \param OriginalType Type of destination and source arrays.
+  /// \param CopyGen Copying procedure that copies value of single array element
+  /// to another single array element.
+  void EmitOMPAggregateAssign(
+      llvm::Value *DestAddr, llvm::Value *SrcAddr, QualType OriginalType,
+      const llvm::function_ref<void(llvm::Value *, llvm::Value *)> &CopyGen);
+  /// \brief Emit proper copying of data from one variable to another.
+  ///
+  /// \param OriginalType Original type of the copied variables.
+  /// \param DestAddr Destination address.
+  /// \param SrcAddr Source address.
+  /// \param DestVD Destination variable used in \a CopyExpr (for arrays, has
+  /// type of the base array element).
+  /// \param SrcVD Source variable used in \a CopyExpr (for arrays, has type of
+  /// the base array element).
+  /// \param Copy Actual copygin expression for copying data from \a SrcVD to \a
+  /// DestVD.
+  void EmitOMPCopy(CodeGenFunction &CGF, QualType OriginalType,
+                   llvm::Value *DestAddr, llvm::Value *SrcAddr,
+                   const VarDecl *DestVD, const VarDecl *SrcVD,
+                   const Expr *Copy);
+  /// \brief Emit atomic update code for constructs: \a X = \a X \a BO \a E or
+  /// \a X = \a E \a BO \a E.
+  ///
+  /// \param X Value to be updated.
+  /// \param E Update value.
+  /// \param BO Binary operation for update operation.
+  /// \param IsXLHSInRHSPart true if \a X is LHS in RHS part of the update
+  /// expression, false otherwise.
+  /// \param AO Atomic ordering of the generated atomic instructions.
+  /// \param CommonGen Code generator for complex expressions that cannot be
+  /// expressed through atomicrmw instruction.
+  /// \returns <true, OldAtomicValue> if simple 'atomicrmw' instruction was
+  /// generated, <false, RValue::get(nullptr)> otherwise.
+  std::pair<bool, RValue> EmitOMPAtomicSimpleUpdateExpr(
+      LValue X, RValue E, BinaryOperatorKind BO, bool IsXLHSInRHSPart,
+      llvm::AtomicOrdering AO, SourceLocation Loc,
+      const llvm::function_ref<RValue(RValue)> &CommonGen);
+  bool EmitOMPFirstprivateClause(const OMPExecutableDirective &D,
                                  OMPPrivateScope &PrivateScope);
   void EmitOMPPrivateClause(const OMPExecutableDirective &D,
                             OMPPrivateScope &PrivateScope);
+  /// \brief Emit code for copyin clause in \a D directive. The next code is
+  /// generated at the start of outlined functions for directives:
+  /// \code
+  /// threadprivate_var1 = master_threadprivate_var1;
+  /// operator=(threadprivate_var2, master_threadprivate_var2);
+  /// ...
+  /// __kmpc_barrier(&loc, global_tid);
+  /// \endcode
+  ///
+  /// \param D OpenMP directive possibly with 'copyin' clause(s).
+  /// \returns true if at least one copyin variable is found, false otherwise.
+  bool EmitOMPCopyinClause(const OMPExecutableDirective &D);
+  /// \brief Emit initial code for lastprivate variables. If some variable is
+  /// not also firstprivate, then the default initialization is used. Otherwise
+  /// initialization of this variable is performed by EmitOMPFirstprivateClause
+  /// method.
+  ///
+  /// \param D Directive that may have 'lastprivate' directives.
+  /// \param PrivateScope Private scope for capturing lastprivate variables for
+  /// proper codegen in internal captured statement.
+  ///
+  /// \returns true if there is at least one lastprivate variable, false
+  /// otherwise.
+  bool EmitOMPLastprivateClauseInit(const OMPExecutableDirective &D,
+                                    OMPPrivateScope &PrivateScope);
+  /// \brief Emit final copying of lastprivate values to original variables at
+  /// the end of the worksharing or simd directive.
+  ///
+  /// \param D Directive that has at least one 'lastprivate' directives.
+  /// \param IsLastIterCond Boolean condition that must be set to 'i1 true' if
+  /// it is the last iteration of the loop code in associated directive, or to
+  /// 'i1 false' otherwise.
+  void EmitOMPLastprivateClauseFinal(const OMPExecutableDirective &D,
+                                     llvm::Value *IsLastIterCond);
+  /// \brief Emit initial code for reduction variables. Creates reduction copies
+  /// and initializes them with the values according to OpenMP standard.
+  ///
+  /// \param D Directive (possibly) with the 'reduction' clause.
+  /// \param PrivateScope Private scope for capturing reduction variables for
+  /// proper codegen in internal captured statement.
+  ///
+  void EmitOMPReductionClauseInit(const OMPExecutableDirective &D,
+                                  OMPPrivateScope &PrivateScope);
+  /// \brief Emit final update of reduction values to original variables at
+  /// the end of the directive.
+  ///
+  /// \param D Directive that has at least one 'reduction' directives.
+  void EmitOMPReductionClauseFinal(const OMPExecutableDirective &D);
 
   void EmitOMPParallelDirective(const OMPParallelDirective &S);
   void EmitOMPSimdDirective(const OMPSimdDirective &S);
@@ -2027,15 +2170,37 @@ public:
   void EmitOMPTargetDirective(const OMPTargetDirective &S);
   void EmitOMPTeamsDirective(const OMPTeamsDirective &S);
 
+  /// \brief Emit inner loop of the worksharing/simd construct.
+  ///
+  /// \param S Directive, for which the inner loop must be emitted.
+  /// \param RequiresCleanup true, if directive has some associated private
+  /// variables.
+  /// \param LoopCond Bollean condition for loop continuation.
+  /// \param IncExpr Increment expression for loop control variable.
+  /// \param BodyGen Generator for the inner body of the inner loop.
+  /// \param PostIncGen Genrator for post-increment code (required for ordered
+  /// loop directvies).
+  void EmitOMPInnerLoop(
+      const Stmt &S, bool RequiresCleanup, const Expr *LoopCond,
+      const Expr *IncExpr,
+      const llvm::function_ref<void(CodeGenFunction &)> &BodyGen,
+      const llvm::function_ref<void(CodeGenFunction &)> &PostIncGen);
+
 private:
 
   /// Helpers for the OpenMP loop directives.
   void EmitOMPLoopBody(const OMPLoopDirective &Directive,
                        bool SeparateIter = false);
-  void EmitOMPInnerLoop(const OMPLoopDirective &S, OMPPrivateScope &LoopScope,
-                        bool SeparateIter = false);
   void EmitOMPSimdFinal(const OMPLoopDirective &S);
-  void EmitOMPWorksharingLoop(const OMPLoopDirective &S);
+  /// \brief Emit code for the worksharing loop-based directive.
+  /// \return true, if this construct has any lastprivate clause, false -
+  /// otherwise.
+  bool EmitOMPWorksharingLoop(const OMPLoopDirective &S);
+  void EmitOMPForOuterLoop(OpenMPScheduleClauseKind ScheduleKind,
+                           const OMPLoopDirective &S,
+                           OMPPrivateScope &LoopScope, bool Ordered,
+                           llvm::Value *LB, llvm::Value *UB, llvm::Value *ST,
+                           llvm::Value *IL, llvm::Value *Chunk);
 
 public:
 
@@ -2085,17 +2250,31 @@ public:
 
   void EmitAtomicInit(Expr *E, LValue lvalue);
 
+  bool LValueIsSuitableForInlineAtomic(LValue Src);
+  bool typeIsSuitableForInlineAtomic(QualType Ty, bool IsVolatile) const;
+
+  RValue EmitAtomicLoad(LValue LV, SourceLocation SL,
+                        AggValueSlot Slot = AggValueSlot::ignored());
+
   RValue EmitAtomicLoad(LValue lvalue, SourceLocation loc,
+                        llvm::AtomicOrdering AO, bool IsVolatile = false,
                         AggValueSlot slot = AggValueSlot::ignored());
 
   void EmitAtomicStore(RValue rvalue, LValue lvalue, bool isInit);
 
-  std::pair<RValue, RValue> EmitAtomicCompareExchange(
+  void EmitAtomicStore(RValue rvalue, LValue lvalue, llvm::AtomicOrdering AO,
+                       bool IsVolatile, bool isInit);
+
+  std::pair<RValue, llvm::Value *> EmitAtomicCompareExchange(
       LValue Obj, RValue Expected, RValue Desired, SourceLocation Loc,
       llvm::AtomicOrdering Success = llvm::SequentiallyConsistent,
       llvm::AtomicOrdering Failure = llvm::SequentiallyConsistent,
       bool IsWeak = false, AggValueSlot Slot = AggValueSlot::ignored());
 
+  void EmitAtomicUpdate(LValue LVal, llvm::AtomicOrdering AO,
+                        const llvm::function_ref<RValue(RValue)> &UpdateOp,
+                        bool IsVolatile);
+
   /// EmitToMemory - Change a scalar value from its value
   /// representation to its in-memory representation.
   llvm::Value *EmitToMemory(llvm::Value *Value, QualType Ty);
@@ -2163,8 +2342,8 @@ public:
   /// Emit an l-value for an assignment (simple or compound) of complex type.
   LValue EmitComplexAssignmentLValue(const BinaryOperator *E);
   LValue EmitComplexCompoundAssignmentLValue(const CompoundAssignOperator *E);
-  LValue EmitScalarCompooundAssignWithComplex(const CompoundAssignOperator *E,
-                                              llvm::Value *&Result);
+  LValue EmitScalarCompoundAssignWithComplex(const CompoundAssignOperator *E,
+                                             llvm::Value *&Result);
 
   // Note: only available for agg return types
   LValue EmitBinaryOperatorLValue(const BinaryOperator *E);
@@ -2208,7 +2387,7 @@ public:
       return ConstantEmission(C, false);
     }
 
-    LLVM_EXPLICIT operator bool() const {
+    explicit operator bool() const {
       return ValueAndIsReference.getOpaqueValue() != nullptr;
     }
 
@@ -2397,6 +2576,7 @@ public:
   llvm::Value *EmitX86BuiltinExpr(unsigned BuiltinID, const CallExpr *E);
   llvm::Value *EmitPPCBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
   llvm::Value *EmitR600BuiltinExpr(unsigned BuiltinID, const CallExpr *E);
+  llvm::Value *EmitSystemZBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
 
   llvm::Value *EmitObjCProtocolExpr(const ObjCProtocolExpr *E);
   llvm::Value *EmitObjCStringLiteral(const ObjCStringLiteral *E);
@@ -2655,7 +2835,7 @@ public:
   /// \brief Create a basic block that will call a handler function in a
   /// sanitizer runtime with the provided arguments, and create a conditional
   /// branch to it.
-  void EmitCheck(ArrayRef<std::pair<llvm::Value *, SanitizerKind>> Checked,
+  void EmitCheck(ArrayRef<std::pair<llvm::Value *, SanitizerMask>> Checked,
                  StringRef CheckName, ArrayRef<llvm::Constant *> StaticArgs,
                  ArrayRef<llvm::Value *> DynamicArgs);
 
@@ -2714,7 +2894,7 @@ public:
                     CallExpr::const_arg_iterator ArgBeg,
                     CallExpr::const_arg_iterator ArgEnd,
                     const FunctionDecl *CalleeDecl = nullptr,
-                    unsigned ParamsToSkip = 0, bool ForceColumnInfo = false) {
+                    unsigned ParamsToSkip = 0) {
     SmallVector<QualType, 16> ArgTypes;
     CallExpr::const_arg_iterator Arg = ArgBeg;
 
@@ -2747,15 +2927,14 @@ public:
     for (; Arg != ArgEnd; ++Arg)
       ArgTypes.push_back(getVarArgType(*Arg));
 
-    EmitCallArgs(Args, ArgTypes, ArgBeg, ArgEnd, CalleeDecl, ParamsToSkip,
-                 ForceColumnInfo);
+    EmitCallArgs(Args, ArgTypes, ArgBeg, ArgEnd, CalleeDecl, ParamsToSkip);
   }
 
   void EmitCallArgs(CallArgList &Args, ArrayRef<QualType> ArgTypes,
                     CallExpr::const_arg_iterator ArgBeg,
                     CallExpr::const_arg_iterator ArgEnd,
                     const FunctionDecl *CalleeDecl = nullptr,
-                    unsigned ParamsToSkip = 0, bool ForceColumnInfo = false);
+                    unsigned ParamsToSkip = 0);
 
 private:
   QualType getVarArgType(const Expr *Arg);
diff --git a/lib/CodeGen/CodeGenModule.cpp b/lib/CodeGen/CodeGenModule.cpp
index d653130df3d0..400506129e64 100644
--- a/lib/CodeGen/CodeGenModule.cpp
+++ b/lib/CodeGen/CodeGenModule.cpp
@@ -141,12 +141,14 @@ CodeGenModule::CodeGenModule(ASTContext &C, const CodeGenOptions &CGO,
   RRData = new RREntrypoints();
 
   if (!CodeGenOpts.InstrProfileInput.empty()) {
-    if (std::error_code EC = llvm::IndexedInstrProfReader::create(
-            CodeGenOpts.InstrProfileInput, PGOReader)) {
+    auto ReaderOrErr =
+        llvm::IndexedInstrProfReader::create(CodeGenOpts.InstrProfileInput);
+    if (std::error_code EC = ReaderOrErr.getError()) {
       unsigned DiagID = Diags.getCustomDiagID(DiagnosticsEngine::Error,
                                               "Could not read profile: %0");
       getDiags().Report(DiagID) << EC.message();
-    }
+    } else
+      PGOReader = std::move(ReaderOrErr.get());
   }
 
   // If coverage mapping generation is enabled, create the
@@ -321,6 +323,8 @@ void CodeGenModule::checkAliases() {
 
 void CodeGenModule::clear() {
   DeferredDeclsToEmit.clear();
+  if (OpenMPRuntime)
+    OpenMPRuntime->clear();
 }
 
 void InstrProfStats::reportDiagnostics(DiagnosticsEngine &Diags,
@@ -346,6 +350,13 @@ void CodeGenModule::Release() {
   if (ObjCRuntime)
     if (llvm::Function *ObjCInitFunction = ObjCRuntime->ModuleInitFunction())
       AddGlobalCtor(ObjCInitFunction);
+  if (Context.getLangOpts().CUDA && !Context.getLangOpts().CUDAIsDevice &&
+      CUDARuntime) {
+    if (llvm::Function *CudaCtorFunction = CUDARuntime->makeModuleCtorFunction())
+      AddGlobalCtor(CudaCtorFunction);
+    if (llvm::Function *CudaDtorFunction = CUDARuntime->makeModuleDtorFunction())
+      AddGlobalDtor(CudaDtorFunction);
+  }
   if (PGOReader && PGOStats.hasDiagnostics())
     PGOStats.reportDiagnostics(getDiags(), getCodeGenOpts().MainFileName);
   EmitCtorList(GlobalCtors, "llvm.global_ctors");
@@ -741,23 +752,6 @@ void CodeGenModule::SetLLVMFunctionAttributesForDefinition(const Decl *D,
   else if (LangOpts.getStackProtector() == LangOptions::SSPReq)
     B.addAttribute(llvm::Attribute::StackProtectReq);
 
-  // Add sanitizer attributes if function is not blacklisted.
-  if (!isInSanitizerBlacklist(F, D->getLocation())) {
-    // When AddressSanitizer is enabled, set SanitizeAddress attribute
-    // unless __attribute__((no_sanitize_address)) is used.
-    if (LangOpts.Sanitize.has(SanitizerKind::Address) &&
-        !D->hasAttr<NoSanitizeAddressAttr>())
-      B.addAttribute(llvm::Attribute::SanitizeAddress);
-    // Same for ThreadSanitizer and __attribute__((no_sanitize_thread))
-    if (LangOpts.Sanitize.has(SanitizerKind::Thread) &&
-        !D->hasAttr<NoSanitizeThreadAttr>())
-      B.addAttribute(llvm::Attribute::SanitizeThread);
-    // Same for MemorySanitizer and __attribute__((no_sanitize_memory))
-    if (LangOpts.Sanitize.has(SanitizerKind::Memory) &&
-        !D->hasAttr<NoSanitizeMemoryAttr>())
-      B.addAttribute(llvm::Attribute::SanitizeMemory);
-  }
-
   F->addAttributes(llvm::AttributeSet::FunctionIndex,
                    llvm::AttributeSet::get(
                        F->getContext(), llvm::AttributeSet::FunctionIndex, B));
@@ -865,11 +859,10 @@ static void setLinkageAndVisibilityForGV(llvm::GlobalValue *GV,
 void CodeGenModule::SetFunctionAttributes(GlobalDecl GD, llvm::Function *F,
                                           bool IsIncompleteFunction,
                                           bool IsThunk) {
-  if (unsigned IID = F->getIntrinsicID()) {
+  if (llvm::Intrinsic::ID IID = F->getIntrinsicID()) {
     // If this is an intrinsic function, set the function's attributes
     // to the intrinsic's attributes.
-    F->setAttributes(llvm::Intrinsic::getAttributes(getLLVMContext(),
-                                                    (llvm::Intrinsic::ID)IID));
+    F->setAttributes(llvm::Intrinsic::getAttributes(getLLVMContext(), IID));
     return;
   }
 
@@ -1085,29 +1078,33 @@ void CodeGenModule::EmitDeferred() {
   // previously unused static decl may become used during the generation of code
   // for a static function, iterate until no changes are made.
 
-  while (true) {
-    if (!DeferredVTables.empty()) {
-      EmitDeferredVTables();
+  if (!DeferredVTables.empty()) {
+    EmitDeferredVTables();
 
-      // Emitting a v-table doesn't directly cause more v-tables to
-      // become deferred, although it can cause functions to be
-      // emitted that then need those v-tables.
-      assert(DeferredVTables.empty());
-    }
+    // Emitting a v-table doesn't directly cause more v-tables to
+    // become deferred, although it can cause functions to be
+    // emitted that then need those v-tables.
+    assert(DeferredVTables.empty());
+  }
+
+  // Stop if we're out of both deferred v-tables and deferred declarations.
+  if (DeferredDeclsToEmit.empty())
+    return;
 
-    // Stop if we're out of both deferred v-tables and deferred declarations.
-    if (DeferredDeclsToEmit.empty()) break;
+  // Grab the list of decls to emit. If EmitGlobalDefinition schedules more
+  // work, it will not interfere with this.
+  std::vector<DeferredGlobal> CurDeclsToEmit;
+  CurDeclsToEmit.swap(DeferredDeclsToEmit);
 
-    DeferredGlobal &G = DeferredDeclsToEmit.back();
+  for (DeferredGlobal &G : CurDeclsToEmit) {
     GlobalDecl D = G.GD;
     llvm::GlobalValue *GV = G.GV;
-    DeferredDeclsToEmit.pop_back();
+    G.GV = nullptr;
 
     assert(!GV || GV == GetGlobalValue(getMangledName(D)));
     if (!GV)
       GV = GetGlobalValue(getMangledName(D));
 
-
     // Check to see if we've already emitted this.  This is necessary
     // for a couple of reasons: first, decls can end up in the
     // deferred-decls queue multiple times, and second, decls can end
@@ -1119,6 +1116,14 @@ void CodeGenModule::EmitDeferred() {
 
     // Otherwise, emit the definition and move on to the next one.
     EmitGlobalDefinition(D, GV);
+
+    // If we found out that we need to emit more decls, do that recursively.
+    // This has the advantage that the decls are emitted in a DFS and related
+    // ones are close together, which is convenient for testing.
+    if (!DeferredVTables.empty() || !DeferredDeclsToEmit.empty()) {
+      EmitDeferred();
+      assert(DeferredVTables.empty() && DeferredDeclsToEmit.empty());
+    }
   }
 }
 
@@ -1275,6 +1280,8 @@ llvm::Constant *CodeGenModule::GetAddrOfUuidDescriptor(
   auto *GV = new llvm::GlobalVariable(
       getModule(), Init->getType(),
       /*isConstant=*/true, llvm::GlobalValue::LinkOnceODRLinkage, Init, Name);
+  if (supportsCOMDAT())
+    GV->setComdat(TheModule.getOrInsertComdat(GV->getName()));
   return GV;
 }
 
@@ -1322,7 +1329,7 @@ void CodeGenModule::EmitGlobal(GlobalDecl GD) {
 
   // If this is CUDA, be selective about which declarations we emit.
   if (LangOpts.CUDA) {
-    if (CodeGenOpts.CUDAIsDevice) {
+    if (LangOpts.CUDAIsDevice) {
       if (!Global->hasAttr<CUDADeviceAttr>() &&
           !Global->hasAttr<CUDAGlobalAttr>() &&
           !Global->hasAttr<CUDAConstantAttr>() &&
@@ -1603,13 +1610,6 @@ CodeGenModule::GetOrCreateLLVMFunction(StringRef MangledName,
       addDeferredDeclToEmit(F, DDI->second);
       DeferredDecls.erase(DDI);
 
-      // Otherwise, if this is a sized deallocation function, emit a weak
-      // definition
-      // for it at the end of the translation unit.
-    } else if (D && cast<FunctionDecl>(D)
-                        ->getCorrespondingUnsizedGlobalDeallocationFunction()) {
-      addDeferredDeclToEmit(F, GD);
-
       // Otherwise, there are cases we have to worry about where we're
       // using a declaration for which we must emit a definition but where
       // we might not find a top-level definition:
@@ -1766,6 +1766,8 @@ CodeGenModule::GetOrCreateLLVMGlobal(StringRef MangledName,
     // handling.
     GV->setConstant(isTypeConstant(D->getType(), false));
 
+    GV->setAlignment(getContext().getDeclAlign(D).getQuantity());
+
     setLinkageAndVisibilityForGV(GV, D);
 
     if (D->getTLSKind()) {
@@ -1829,7 +1831,11 @@ CodeGenModule::CreateOrReplaceCXXRuntimeVariable(StringRef Name,
     
     OldGV->eraseFromParent();
   }
-  
+
+  if (supportsCOMDAT() && GV->isWeakForLinker() &&
+      !GV->hasAvailableExternallyLinkage())
+    GV->setComdat(TheModule.getOrInsertComdat(GV->getName()));
+
   return GV;
 }
 
@@ -1884,7 +1890,7 @@ CharUnits CodeGenModule::GetTargetTypeStoreSize(llvm::Type *Ty) const {
 
 unsigned CodeGenModule::GetGlobalVarAddressSpace(const VarDecl *D,
                                                  unsigned AddrSpace) {
-  if (LangOpts.CUDA && CodeGenOpts.CUDAIsDevice) {
+  if (LangOpts.CUDA && LangOpts.CUDAIsDevice) {
     if (D->hasAttr<CUDAConstantAttr>())
       AddrSpace = getContext().getTargetAddressSpace(LangAS::cuda_constant);
     else if (D->hasAttr<CUDASharedAttr>())
@@ -1929,6 +1935,38 @@ void CodeGenModule::MaybeHandleStaticInExternC(const SomeDecl *D,
     R.first->second = nullptr;
 }
 
+static bool shouldBeInCOMDAT(CodeGenModule &CGM, const Decl &D) {
+  if (!CGM.supportsCOMDAT())
+    return false;
+
+  if (D.hasAttr<SelectAnyAttr>())
+    return true;
+
+  GVALinkage Linkage;
+  if (auto *VD = dyn_cast<VarDecl>(&D))
+    Linkage = CGM.getContext().GetGVALinkageForVariable(VD);
+  else
+    Linkage = CGM.getContext().GetGVALinkageForFunction(cast<FunctionDecl>(&D));
+
+  switch (Linkage) {
+  case GVA_Internal:
+  case GVA_AvailableExternally:
+  case GVA_StrongExternal:
+    return false;
+  case GVA_DiscardableODR:
+  case GVA_StrongODR:
+    return true;
+  }
+  llvm_unreachable("No such linkage");
+}
+
+void CodeGenModule::maybeSetTrivialComdat(const Decl &D,
+                                          llvm::GlobalObject &GO) {
+  if (!shouldBeInCOMDAT(*this, D))
+    return;
+  GO.setComdat(TheModule.getOrInsertComdat(GO.getName()));
+}
+
 void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D) {
   llvm::Constant *Init = nullptr;
   QualType ASTTy = D->getType();
@@ -2072,6 +2110,8 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D) {
     setTLSMode(GV, *D);
   }
 
+  maybeSetTrivialComdat(*D, *GV);
+
   // Emit the initializer function if necessary.
   if (NeedsGlobalCtor || NeedsGlobalDtor)
     EmitCXXGlobalVarDeclInitFunc(D, GV, NeedsGlobalCtor);
@@ -2085,7 +2125,8 @@ void CodeGenModule::EmitGlobalVarDefinition(const VarDecl *D) {
 }
 
 static bool isVarDeclStrongDefinition(const ASTContext &Context,
-                                      const VarDecl *D, bool NoCommon) {
+                                      CodeGenModule &CGM, const VarDecl *D,
+                                      bool NoCommon) {
   // Don't give variables common linkage if -fno-common was specified unless it
   // was overridden by a NoCommon attribute.
   if ((NoCommon || D->hasAttr<NoCommonAttr>()) && !D->hasAttr<CommonAttr>())
@@ -2110,11 +2151,31 @@ static bool isVarDeclStrongDefinition(const ASTContext &Context,
   if (D->hasAttr<WeakImportAttr>())
     return true;
 
+  // A variable cannot be both common and exist in a comdat.
+  if (shouldBeInCOMDAT(CGM, *D))
+    return true;
+
   // Declarations with a required alignment do not have common linakge in MSVC
   // mode.
-  if (Context.getLangOpts().MSVCCompat &&
-      (Context.isAlignmentRequired(D->getType()) || D->hasAttr<AlignedAttr>()))
-    return true;
+  if (Context.getLangOpts().MSVCCompat) {
+    if (D->hasAttr<AlignedAttr>())
+      return true;
+    QualType VarType = D->getType();
+    if (Context.isAlignmentRequired(VarType))
+      return true;
+
+    if (const auto *RT = VarType->getAs<RecordType>()) {
+      const RecordDecl *RD = RT->getDecl();
+      for (const FieldDecl *FD : RD->fields()) {
+        if (FD->isBitField())
+          continue;
+        if (FD->hasAttr<AlignedAttr>())
+          return true;
+        if (Context.isAlignmentRequired(FD->getType()))
+          return true;
+      }
+    }
+  }
 
   return false;
 }
@@ -2162,7 +2223,7 @@ llvm::GlobalValue::LinkageTypes CodeGenModule::getLLVMLinkageForDeclarator(
   // C++ doesn't have tentative definitions and thus cannot have common
   // linkage.
   if (!getLangOpts().CPlusPlus && isa<VarDecl>(D) &&
-      !isVarDeclStrongDefinition(Context, cast<VarDecl>(D),
+      !isVarDeclStrongDefinition(Context, *this, cast<VarDecl>(D),
                                  CodeGenOpts.NoCommon))
     return llvm::GlobalVariable::CommonLinkage;
 
@@ -2285,7 +2346,7 @@ static void replaceUsesOfNonProtoConstant(llvm::Constant *old,
       callSite->replaceAllUsesWith(newCall.getInstruction());
 
     // Copy debug location attached to CI.
-    if (!callSite->getDebugLoc().isUnknown())
+    if (callSite->getDebugLoc())
       newCall->setDebugLoc(callSite->getDebugLoc());
     callSite->eraseFromParent();
   }
@@ -2406,6 +2467,8 @@ void CodeGenModule::EmitGlobalFunctionDefinition(GlobalDecl GD,
 
   MaybeHandleStaticInExternC(D, Fn);
 
+  maybeSetTrivialComdat(*D, *Fn);
+
   CodeGenFunction(*this).GenerateCode(D, Fn, FI);
 
   setFunctionDefinitionAttributes(D, Fn);
@@ -2449,7 +2512,7 @@ void CodeGenModule::EmitAliasDefinition(GlobalDecl GD) {
 
   // Create the new alias itself, but don't set a name yet.
   auto *GA = llvm::GlobalAlias::create(
-      cast<llvm::PointerType>(Aliasee->getType())->getElementType(), 0,
+      cast<llvm::PointerType>(Aliasee->getType()),
       llvm::Function::ExternalLinkage, "", Aliasee, &getModule());
 
   if (Entry) {
@@ -2497,12 +2560,10 @@ llvm::Function *CodeGenModule::getIntrinsic(unsigned IID,
                                          Tys);
 }
 
-static llvm::StringMapEntry<llvm::Constant*> &
-GetConstantCFStringEntry(llvm::StringMap<llvm::Constant*> &Map,
-                         const StringLiteral *Literal,
-                         bool TargetIsLSB,
-                         bool &IsUTF16,
-                         unsigned &StringLength) {
+static llvm::StringMapEntry<llvm::GlobalVariable *> &
+GetConstantCFStringEntry(llvm::StringMap<llvm::GlobalVariable *> &Map,
+                         const StringLiteral *Literal, bool TargetIsLSB,
+                         bool &IsUTF16, unsigned &StringLength) {
   StringRef String = Literal->getString();
   unsigned NumBytes = String.size();
 
@@ -2534,10 +2595,9 @@ GetConstantCFStringEntry(llvm::StringMap<llvm::Constant*> &Map,
                          nullptr)).first;
 }
 
-static llvm::StringMapEntry<llvm::Constant*> &
-GetConstantStringEntry(llvm::StringMap<llvm::Constant*> &Map,
-                       const StringLiteral *Literal,
-                       unsigned &StringLength) {
+static llvm::StringMapEntry<llvm::GlobalVariable *> &
+GetConstantStringEntry(llvm::StringMap<llvm::GlobalVariable *> &Map,
+                       const StringLiteral *Literal, unsigned &StringLength) {
   StringRef String = Literal->getString();
   StringLength = String.size();
   return *Map.insert(std::make_pair(String, nullptr)).first;
@@ -2547,10 +2607,10 @@ llvm::Constant *
 CodeGenModule::GetAddrOfConstantCFString(const StringLiteral *Literal) {
   unsigned StringLength = 0;
   bool isUTF16 = false;
-  llvm::StringMapEntry<llvm::Constant*> &Entry =
-    GetConstantCFStringEntry(CFConstantStringMap, Literal,
-                             getDataLayout().isLittleEndian(),
-                             isUTF16, StringLength);
+  llvm::StringMapEntry<llvm::GlobalVariable *> &Entry =
+      GetConstantCFStringEntry(CFConstantStringMap, Literal,
+                               getDataLayout().isLittleEndian(), isUTF16,
+                               StringLength);
 
   if (auto *C = Entry.second)
     return C;
@@ -2566,7 +2626,7 @@ CodeGenModule::GetAddrOfConstantCFString(const StringLiteral *Literal) {
     llvm::Constant *GV = CreateRuntimeVariable(Ty,
                                            "__CFConstantStringClassReference");
     // Decay array -> ptr
-    V = llvm::ConstantExpr::getGetElementPtr(GV, Zeros);
+    V = llvm::ConstantExpr::getGetElementPtr(Ty, GV, Zeros);
     CFConstantStringClassRef = V;
   }
   else
@@ -2619,7 +2679,8 @@ CodeGenModule::GetAddrOfConstantCFString(const StringLiteral *Literal) {
   }
 
   // String.
-  Fields[2] = llvm::ConstantExpr::getGetElementPtr(GV, Zeros);
+  Fields[2] =
+      llvm::ConstantExpr::getGetElementPtr(GV->getValueType(), GV, Zeros);
 
   if (isUTF16)
     // Cast the UTF16 string to the correct type.
@@ -2640,11 +2701,11 @@ CodeGenModule::GetAddrOfConstantCFString(const StringLiteral *Literal) {
   return GV;
 }
 
-llvm::Constant *
+llvm::GlobalVariable *
 CodeGenModule::GetAddrOfConstantString(const StringLiteral *Literal) {
   unsigned StringLength = 0;
-  llvm::StringMapEntry<llvm::Constant*> &Entry =
-    GetConstantStringEntry(CFConstantStringMap, Literal, StringLength);
+  llvm::StringMapEntry<llvm::GlobalVariable *> &Entry =
+      GetConstantStringEntry(CFConstantStringMap, Literal, StringLength);
 
   if (auto *C = Entry.second)
     return C;
@@ -2673,11 +2734,10 @@ CodeGenModule::GetAddrOfConstantString(const StringLiteral *Literal) {
       llvm::Type *PTy = llvm::ArrayType::get(Ty, 0);
       GV = CreateRuntimeVariable(PTy, str);
       // Decay array -> ptr
-      V = llvm::ConstantExpr::getGetElementPtr(GV, Zeros);
+      V = llvm::ConstantExpr::getGetElementPtr(PTy, GV, Zeros);
       ConstantStringClassRef = V;
     }
-  }
-  else
+  } else
     V = ConstantStringClassRef;
 
   if (!NSConstantStringType) {
@@ -2733,8 +2793,9 @@ CodeGenModule::GetAddrOfConstantString(const StringLiteral *Literal) {
   // of the string is via this class initializer.
   CharUnits Align = getContext().getTypeAlignInChars(getContext().CharTy);
   GV->setAlignment(Align.getQuantity());
-  Fields[1] = llvm::ConstantExpr::getGetElementPtr(GV, Zeros);
-  
+  Fields[1] =
+      llvm::ConstantExpr::getGetElementPtr(GV->getValueType(), GV, Zeros);
+
   // String length.
   llvm::Type *Ty = getTypes().ConvertType(getContext().UnsignedIntTy);
   Fields[2] = llvm::ConstantInt::get(Ty, StringLength);
@@ -2838,12 +2899,18 @@ GenerateStringLiteral(llvm::Constant *C, llvm::GlobalValue::LinkageTypes LT,
   if (CGM.getLangOpts().OpenCL)
     AddrSpace = CGM.getContext().getTargetAddressSpace(LangAS::opencl_constant);
 
+  llvm::Module &M = CGM.getModule();
   // Create a global variable for this string
   auto *GV = new llvm::GlobalVariable(
-      CGM.getModule(), C->getType(), !CGM.getLangOpts().WritableStrings, LT, C,
-      GlobalName, nullptr, llvm::GlobalVariable::NotThreadLocal, AddrSpace);
+      M, C->getType(), !CGM.getLangOpts().WritableStrings, LT, C, GlobalName,
+      nullptr, llvm::GlobalVariable::NotThreadLocal, AddrSpace);
   GV->setAlignment(Alignment);
   GV->setUnnamedAddr(true);
+  if (GV->isWeakForLinker()) {
+    assert(CGM.supportsCOMDAT() && "Only COFF uses weak string literals");
+    GV->setComdat(M.getOrInsertComdat(GV->getName()));
+  }
+
   return GV;
 }
 
@@ -3002,10 +3069,19 @@ llvm::Constant *CodeGenModule::GetAddrOfGlobalTemporary(
   // Create a global variable for this lifetime-extended temporary.
   llvm::GlobalValue::LinkageTypes Linkage =
       getLLVMLinkageVarDefinition(VD, Constant);
-  // There is no need for this temporary to have global linkage if the global
-  // variable has external linkage.
-  if (Linkage == llvm::GlobalVariable::ExternalLinkage)
-    Linkage = llvm::GlobalVariable::PrivateLinkage;
+  if (Linkage == llvm::GlobalVariable::ExternalLinkage) {
+    const VarDecl *InitVD;
+    if (VD->isStaticDataMember() && VD->getAnyInitializer(InitVD) &&
+        isa<CXXRecordDecl>(InitVD->getLexicalDeclContext())) {
+      // Temporaries defined inside a class get linkonce_odr linkage because the
+      // class can be defined in multipe translation units.
+      Linkage = llvm::GlobalVariable::LinkOnceODRLinkage;
+    } else {
+      // There is no need for this temporary to have external linkage if the
+      // VarDecl has external linkage.
+      Linkage = llvm::GlobalVariable::InternalLinkage;
+    }
+  }
   unsigned AddrSpace = GetGlobalVarAddressSpace(
       VD, getContext().getTargetAddressSpace(MaterializedType));
   auto *GV = new llvm::GlobalVariable(
@@ -3015,6 +3091,8 @@ llvm::Constant *CodeGenModule::GetAddrOfGlobalTemporary(
   setGlobalVisibility(GV, VD);
   GV->setAlignment(
       getContext().getTypeAlignInChars(MaterializedType).getQuantity());
+  if (supportsCOMDAT() && GV->isWeakForLinker())
+    GV->setComdat(TheModule.getOrInsertComdat(GV->getName()));
   if (VD->getTLSKind())
     setTLSMode(GV, *VD);
   Slot = GV;
@@ -3268,16 +3346,11 @@ void CodeGenModule::EmitTopLevelDecl(Decl *D) {
     break;
 
   case Decl::FileScopeAsm: {
+    // File-scope asm is ignored during device-side CUDA compilation.
+    if (LangOpts.CUDA && LangOpts.CUDAIsDevice)
+      break;
     auto *AD = cast<FileScopeAsmDecl>(D);
-    StringRef AsmString = AD->getAsmString()->getString();
-
-    const std::string &S = getModule().getModuleInlineAsm();
-    if (S.empty())
-      getModule().setModuleInlineAsm(AsmString);
-    else if (S.end()[-1] == '\n')
-      getModule().setModuleInlineAsm(S + AsmString.str());
-    else
-      getModule().setModuleInlineAsm(S + '\n' + AsmString.str());
+    getModule().appendModuleInlineAsm(AD->getAsmString()->getString());
     break;
   }
 
@@ -3285,7 +3358,7 @@ void CodeGenModule::EmitTopLevelDecl(Decl *D) {
     auto *Import = cast<ImportDecl>(D);
 
     // Ignore import declarations that come from imported modules.
-    if (clang::Module *Owner = Import->getOwningModule()) {
+    if (clang::Module *Owner = Import->getImportedOwningModule()) {
       if (getLangOpts().CurrentModule.empty() ||
           Owner->getTopLevelModule()->Name == getLangOpts().CurrentModule)
         break;
@@ -3357,7 +3430,7 @@ void CodeGenModule::ClearUnusedCoverageMapping(const Decl *D) {
 
 void CodeGenModule::EmitDeferredUnusedCoverageMappings() {
   std::vector<const Decl *> DeferredDecls;
-  for (const auto I : DeferredEmptyCoverageMappingDecls) {
+  for (const auto &I : DeferredEmptyCoverageMappingDecls) {
     if (!I.second)
       continue;
     DeferredDecls.push_back(I.first);
@@ -3561,6 +3634,12 @@ llvm::Constant *CodeGenModule::EmitUuidofInitializer(StringRef Uuid) {
   return llvm::ConstantStruct::getAnon(Fields);
 }
 
+llvm::Constant *
+CodeGenModule::getAddrOfCXXCatchHandlerType(QualType Ty,
+                                            QualType CatchHandlerType) {
+  return getCXXABI().getAddrOfCXXCatchHandlerType(Ty, CatchHandlerType);
+}
+
 llvm::Constant *CodeGenModule::GetAddrOfRTTIDescriptor(QualType Ty,
                                                        bool ForEH) {
   // Return a bogus pointer if RTTI is disabled, unless it's for EH.
@@ -3583,11 +3662,8 @@ void CodeGenModule::EmitOMPThreadPrivateDecl(const OMPThreadPrivateDecl *D) {
         VD->getAnyInitializer() &&
         !VD->getAnyInitializer()->isConstantInitializer(getContext(),
                                                         /*ForRef=*/false);
-    if (auto InitFunction =
-            getOpenMPRuntime().EmitOMPThreadPrivateVarDefinition(
-                VD, GetAddrOfGlobalVar(VD), RefExpr->getLocStart(),
-                PerformInit))
+    if (auto InitFunction = getOpenMPRuntime().emitThreadPrivateVarDefinition(
+            VD, GetAddrOfGlobalVar(VD), RefExpr->getLocStart(), PerformInit))
       CXXGlobalInits.push_back(InitFunction);
   }
 }
-
diff --git a/lib/CodeGen/CodeGenModule.h b/lib/CodeGen/CodeGenModule.h
index 2aafe7e88bfe..feef6c2583ec 100644
--- a/lib/CodeGen/CodeGenModule.h
+++ b/lib/CodeGen/CodeGenModule.h
@@ -258,8 +258,8 @@ public:
 /// This class organizes the cross-function state that is used while generating
 /// LLVM code.
 class CodeGenModule : public CodeGenTypeCache {
-  CodeGenModule(const CodeGenModule &) LLVM_DELETED_FUNCTION;
-  void operator=(const CodeGenModule &) LLVM_DELETED_FUNCTION;
+  CodeGenModule(const CodeGenModule &) = delete;
+  void operator=(const CodeGenModule &) = delete;
 
 public:
   struct Structor {
@@ -366,7 +366,7 @@ private:
   /// Map used to get unique annotation strings.
   llvm::StringMap<llvm::Constant*> AnnotationStrings;
 
-  llvm::StringMap<llvm::Constant*> CFConstantStringMap;
+  llvm::StringMap<llvm::GlobalVariable *> CFConstantStringMap;
 
   llvm::DenseMap<llvm::Constant *, llvm::GlobalVariable *> ConstantStringMap;
   llvm::DenseMap<const Decl*, llvm::Constant *> StaticLocalDeclMap;
@@ -400,7 +400,8 @@ private:
   /// When a C++ decl with an initializer is deferred, null is
   /// appended to CXXGlobalInits, and the index of that null is placed
   /// here so that the initializer will be performed in the correct
-  /// order.
+  /// order. Once the decl is emitted, the index is replaced with ~0U to ensure
+  /// that we don't re-emit the initializer.
   llvm::DenseMap<const Decl*, unsigned> DelayedCXXInitPosition;
   
   typedef std::pair<OrderGlobalInits, llvm::Function*> GlobalInitData;
@@ -606,6 +607,7 @@ public:
   const TargetInfo &getTarget() const { return Target; }
   const llvm::Triple &getTriple() const;
   bool supportsCOMDAT() const;
+  void maybeSetTrivialComdat(const Decl &D, llvm::GlobalObject &GO);
 
   CGCXXABI &getCXXABI() const { return *ABI; }
   llvm::LLVMContext &getLLVMContext() { return VMContext; }
@@ -718,6 +720,9 @@ public:
   /// Get the address of the RTTI descriptor for the given type.
   llvm::Constant *GetAddrOfRTTIDescriptor(QualType Ty, bool ForEH = false);
 
+  llvm::Constant *getAddrOfCXXCatchHandlerType(QualType Ty,
+                                               QualType CatchHandlerType);
+
   /// Get the address of a uuid descriptor .
   llvm::Constant *GetAddrOfUuidDescriptor(const CXXUuidofExpr* E);
 
@@ -782,7 +787,7 @@ public:
   /// Return a pointer to a constant NSString object for the given string. Or a
   /// user defined String object as defined via
   /// -fconstant-string-class=class_name option.
-  llvm::Constant *GetAddrOfConstantString(const StringLiteral *Literal);
+  llvm::GlobalVariable *GetAddrOfConstantString(const StringLiteral *Literal);
 
   /// Return a constant array for the given string.
   llvm::Constant *GetConstantArrayFromStringLiteral(const StringLiteral *E);
@@ -993,7 +998,7 @@ public:
 
   void EmitTentativeDefinition(const VarDecl *D);
 
-  void EmitVTable(CXXRecordDecl *Class, bool DefinitionRequired);
+  void EmitVTable(CXXRecordDecl *Class);
 
   /// Emit the RTTI descriptors for the builtin types.
   void EmitFundamentalRTTIDescriptors();
@@ -1102,6 +1107,14 @@ public:
   /// \param D Threadprivate declaration.
   void EmitOMPThreadPrivateDecl(const OMPThreadPrivateDecl *D);
 
+  /// Emit bit set entries for the given vtable using the given layout if
+  /// vptr CFI is enabled.
+  void EmitVTableBitSetEntries(llvm::GlobalVariable *VTable,
+                               const VTableLayout &VTLayout);
+
+  /// \breif Get the declaration of std::terminate for the platform.
+  llvm::Constant *getTerminateFn();
+
 private:
   llvm::Constant *
   GetOrCreateLLVMFunction(StringRef MangledName, llvm::Type *Ty, GlobalDecl D,
diff --git a/lib/CodeGen/CodeGenPGO.cpp b/lib/CodeGen/CodeGenPGO.cpp
index 24b035d67598..c97244328d37 100644
--- a/lib/CodeGen/CodeGenPGO.cpp
+++ b/lib/CodeGen/CodeGenPGO.cpp
@@ -58,12 +58,16 @@ void CodeGenPGO::setFuncName(llvm::Function *Fn) {
 }
 
 void CodeGenPGO::createFuncNameVar(llvm::GlobalValue::LinkageTypes Linkage) {
-  // Usually, we want to match the function's linkage, but
-  // available_externally and extern_weak both have the wrong semantics.
+  // We generally want to match the function's linkage, but available_externally
+  // and extern_weak both have the wrong semantics, and anything that doesn't
+  // need to link across compilation units doesn't need to be visible at all.
   if (Linkage == llvm::GlobalValue::ExternalWeakLinkage)
     Linkage = llvm::GlobalValue::LinkOnceAnyLinkage;
   else if (Linkage == llvm::GlobalValue::AvailableExternallyLinkage)
     Linkage = llvm::GlobalValue::LinkOnceODRLinkage;
+  else if (Linkage == llvm::GlobalValue::InternalLinkage ||
+           Linkage == llvm::GlobalValue::ExternalLinkage)
+    Linkage = llvm::GlobalValue::PrivateLinkage;
 
   auto *Value =
       llvm::ConstantDataArray::getString(CGM.getLLVMContext(), FuncName, false);
@@ -138,482 +142,469 @@ const int PGOHash::NumBitsPerType;
 const unsigned PGOHash::NumTypesPerWord;
 const unsigned PGOHash::TooBig;
 
-  /// A RecursiveASTVisitor that fills a map of statements to PGO counters.
-  struct MapRegionCounters : public RecursiveASTVisitor<MapRegionCounters> {
-    /// The next counter value to assign.
-    unsigned NextCounter;
-    /// The function hash.
-    PGOHash Hash;
-    /// The map of statements to counters.
-    llvm::DenseMap<const Stmt *, unsigned> &CounterMap;
-
-    MapRegionCounters(llvm::DenseMap<const Stmt *, unsigned> &CounterMap)
-        : NextCounter(0), CounterMap(CounterMap) {}
-
-    // Blocks and lambdas are handled as separate functions, so we need not
-    // traverse them in the parent context.
-    bool TraverseBlockExpr(BlockExpr *BE) { return true; }
-    bool TraverseLambdaBody(LambdaExpr *LE) { return true; }
-    bool TraverseCapturedStmt(CapturedStmt *CS) { return true; }
-
-    bool VisitDecl(const Decl *D) {
-      switch (D->getKind()) {
-      default:
-        break;
-      case Decl::Function:
-      case Decl::CXXMethod:
-      case Decl::CXXConstructor:
-      case Decl::CXXDestructor:
-      case Decl::CXXConversion:
-      case Decl::ObjCMethod:
-      case Decl::Block:
-      case Decl::Captured:
-        CounterMap[D->getBody()] = NextCounter++;
-        break;
-      }
-      return true;
+/// A RecursiveASTVisitor that fills a map of statements to PGO counters.
+struct MapRegionCounters : public RecursiveASTVisitor<MapRegionCounters> {
+  /// The next counter value to assign.
+  unsigned NextCounter;
+  /// The function hash.
+  PGOHash Hash;
+  /// The map of statements to counters.
+  llvm::DenseMap<const Stmt *, unsigned> &CounterMap;
+
+  MapRegionCounters(llvm::DenseMap<const Stmt *, unsigned> &CounterMap)
+      : NextCounter(0), CounterMap(CounterMap) {}
+
+  // Blocks and lambdas are handled as separate functions, so we need not
+  // traverse them in the parent context.
+  bool TraverseBlockExpr(BlockExpr *BE) { return true; }
+  bool TraverseLambdaBody(LambdaExpr *LE) { return true; }
+  bool TraverseCapturedStmt(CapturedStmt *CS) { return true; }
+
+  bool VisitDecl(const Decl *D) {
+    switch (D->getKind()) {
+    default:
+      break;
+    case Decl::Function:
+    case Decl::CXXMethod:
+    case Decl::CXXConstructor:
+    case Decl::CXXDestructor:
+    case Decl::CXXConversion:
+    case Decl::ObjCMethod:
+    case Decl::Block:
+    case Decl::Captured:
+      CounterMap[D->getBody()] = NextCounter++;
+      break;
     }
+    return true;
+  }
 
-    bool VisitStmt(const Stmt *S) {
-      auto Type = getHashType(S);
-      if (Type == PGOHash::None)
-        return true;
-
-      CounterMap[S] = NextCounter++;
-      Hash.combine(Type);
+  bool VisitStmt(const Stmt *S) {
+    auto Type = getHashType(S);
+    if (Type == PGOHash::None)
       return true;
+
+    CounterMap[S] = NextCounter++;
+    Hash.combine(Type);
+    return true;
+  }
+  PGOHash::HashType getHashType(const Stmt *S) {
+    switch (S->getStmtClass()) {
+    default:
+      break;
+    case Stmt::LabelStmtClass:
+      return PGOHash::LabelStmt;
+    case Stmt::WhileStmtClass:
+      return PGOHash::WhileStmt;
+    case Stmt::DoStmtClass:
+      return PGOHash::DoStmt;
+    case Stmt::ForStmtClass:
+      return PGOHash::ForStmt;
+    case Stmt::CXXForRangeStmtClass:
+      return PGOHash::CXXForRangeStmt;
+    case Stmt::ObjCForCollectionStmtClass:
+      return PGOHash::ObjCForCollectionStmt;
+    case Stmt::SwitchStmtClass:
+      return PGOHash::SwitchStmt;
+    case Stmt::CaseStmtClass:
+      return PGOHash::CaseStmt;
+    case Stmt::DefaultStmtClass:
+      return PGOHash::DefaultStmt;
+    case Stmt::IfStmtClass:
+      return PGOHash::IfStmt;
+    case Stmt::CXXTryStmtClass:
+      return PGOHash::CXXTryStmt;
+    case Stmt::CXXCatchStmtClass:
+      return PGOHash::CXXCatchStmt;
+    case Stmt::ConditionalOperatorClass:
+      return PGOHash::ConditionalOperator;
+    case Stmt::BinaryConditionalOperatorClass:
+      return PGOHash::BinaryConditionalOperator;
+    case Stmt::BinaryOperatorClass: {
+      const BinaryOperator *BO = cast<BinaryOperator>(S);
+      if (BO->getOpcode() == BO_LAnd)
+        return PGOHash::BinaryOperatorLAnd;
+      if (BO->getOpcode() == BO_LOr)
+        return PGOHash::BinaryOperatorLOr;
+      break;
     }
-    PGOHash::HashType getHashType(const Stmt *S) {
-      switch (S->getStmtClass()) {
-      default:
-        break;
-      case Stmt::LabelStmtClass:
-        return PGOHash::LabelStmt;
-      case Stmt::WhileStmtClass:
-        return PGOHash::WhileStmt;
-      case Stmt::DoStmtClass:
-        return PGOHash::DoStmt;
-      case Stmt::ForStmtClass:
-        return PGOHash::ForStmt;
-      case Stmt::CXXForRangeStmtClass:
-        return PGOHash::CXXForRangeStmt;
-      case Stmt::ObjCForCollectionStmtClass:
-        return PGOHash::ObjCForCollectionStmt;
-      case Stmt::SwitchStmtClass:
-        return PGOHash::SwitchStmt;
-      case Stmt::CaseStmtClass:
-        return PGOHash::CaseStmt;
-      case Stmt::DefaultStmtClass:
-        return PGOHash::DefaultStmt;
-      case Stmt::IfStmtClass:
-        return PGOHash::IfStmt;
-      case Stmt::CXXTryStmtClass:
-        return PGOHash::CXXTryStmt;
-      case Stmt::CXXCatchStmtClass:
-        return PGOHash::CXXCatchStmt;
-      case Stmt::ConditionalOperatorClass:
-        return PGOHash::ConditionalOperator;
-      case Stmt::BinaryConditionalOperatorClass:
-        return PGOHash::BinaryConditionalOperator;
-      case Stmt::BinaryOperatorClass: {
-        const BinaryOperator *BO = cast<BinaryOperator>(S);
-        if (BO->getOpcode() == BO_LAnd)
-          return PGOHash::BinaryOperatorLAnd;
-        if (BO->getOpcode() == BO_LOr)
-          return PGOHash::BinaryOperatorLOr;
-        break;
-      }
-      }
-      return PGOHash::None;
     }
+    return PGOHash::None;
+  }
+};
+
+/// A StmtVisitor that propagates the raw counts through the AST and
+/// records the count at statements where the value may change.
+struct ComputeRegionCounts : public ConstStmtVisitor<ComputeRegionCounts> {
+  /// PGO state.
+  CodeGenPGO &PGO;
+
+  /// A flag that is set when the current count should be recorded on the
+  /// next statement, such as at the exit of a loop.
+  bool RecordNextStmtCount;
+
+  /// The count at the current location in the traversal.
+  uint64_t CurrentCount;
+
+  /// The map of statements to count values.
+  llvm::DenseMap<const Stmt *, uint64_t> &CountMap;
+
+  /// BreakContinueStack - Keep counts of breaks and continues inside loops.
+  struct BreakContinue {
+    uint64_t BreakCount;
+    uint64_t ContinueCount;
+    BreakContinue() : BreakCount(0), ContinueCount(0) {}
   };
+  SmallVector<BreakContinue, 8> BreakContinueStack;
 
-  /// A StmtVisitor that propagates the raw counts through the AST and
-  /// records the count at statements where the value may change.
-  struct ComputeRegionCounts : public ConstStmtVisitor<ComputeRegionCounts> {
-    /// PGO state.
-    CodeGenPGO &PGO;
-
-    /// A flag that is set when the current count should be recorded on the
-    /// next statement, such as at the exit of a loop.
-    bool RecordNextStmtCount;
-
-    /// The map of statements to count values.
-    llvm::DenseMap<const Stmt *, uint64_t> &CountMap;
-
-    /// BreakContinueStack - Keep counts of breaks and continues inside loops.
-    struct BreakContinue {
-      uint64_t BreakCount;
-      uint64_t ContinueCount;
-      BreakContinue() : BreakCount(0), ContinueCount(0) {}
-    };
-    SmallVector<BreakContinue, 8> BreakContinueStack;
-
-    ComputeRegionCounts(llvm::DenseMap<const Stmt *, uint64_t> &CountMap,
-                        CodeGenPGO &PGO)
-        : PGO(PGO), RecordNextStmtCount(false), CountMap(CountMap) {}
-
-    void RecordStmtCount(const Stmt *S) {
-      if (RecordNextStmtCount) {
-        CountMap[S] = PGO.getCurrentRegionCount();
-        RecordNextStmtCount = false;
-      }
-    }
+  ComputeRegionCounts(llvm::DenseMap<const Stmt *, uint64_t> &CountMap,
+                      CodeGenPGO &PGO)
+      : PGO(PGO), RecordNextStmtCount(false), CountMap(CountMap) {}
 
-    void VisitStmt(const Stmt *S) {
-      RecordStmtCount(S);
-      for (Stmt::const_child_range I = S->children(); I; ++I) {
-        if (*I)
-         this->Visit(*I);
-      }
+  void RecordStmtCount(const Stmt *S) {
+    if (RecordNextStmtCount) {
+      CountMap[S] = CurrentCount;
+      RecordNextStmtCount = false;
     }
+  }
 
-    void VisitFunctionDecl(const FunctionDecl *D) {
-      // Counter tracks entry to the function body.
-      RegionCounter Cnt(PGO, D->getBody());
-      Cnt.beginRegion();
-      CountMap[D->getBody()] = PGO.getCurrentRegionCount();
-      Visit(D->getBody());
-    }
+  /// Set and return the current count.
+  uint64_t setCount(uint64_t Count) {
+    CurrentCount = Count;
+    return Count;
+  }
 
-    // Skip lambda expressions. We visit these as FunctionDecls when we're
-    // generating them and aren't interested in the body when generating a
-    // parent context.
-    void VisitLambdaExpr(const LambdaExpr *LE) {}
-
-    void VisitCapturedDecl(const CapturedDecl *D) {
-      // Counter tracks entry to the capture body.
-      RegionCounter Cnt(PGO, D->getBody());
-      Cnt.beginRegion();
-      CountMap[D->getBody()] = PGO.getCurrentRegionCount();
-      Visit(D->getBody());
+  void VisitStmt(const Stmt *S) {
+    RecordStmtCount(S);
+    for (Stmt::const_child_range I = S->children(); I; ++I) {
+      if (*I)
+        this->Visit(*I);
     }
+  }
 
-    void VisitObjCMethodDecl(const ObjCMethodDecl *D) {
-      // Counter tracks entry to the method body.
-      RegionCounter Cnt(PGO, D->getBody());
-      Cnt.beginRegion();
-      CountMap[D->getBody()] = PGO.getCurrentRegionCount();
-      Visit(D->getBody());
-    }
+  void VisitFunctionDecl(const FunctionDecl *D) {
+    // Counter tracks entry to the function body.
+    uint64_t BodyCount = setCount(PGO.getRegionCount(D->getBody()));
+    CountMap[D->getBody()] = BodyCount;
+    Visit(D->getBody());
+  }
 
-    void VisitBlockDecl(const BlockDecl *D) {
-      // Counter tracks entry to the block body.
-      RegionCounter Cnt(PGO, D->getBody());
-      Cnt.beginRegion();
-      CountMap[D->getBody()] = PGO.getCurrentRegionCount();
-      Visit(D->getBody());
-    }
+  // Skip lambda expressions. We visit these as FunctionDecls when we're
+  // generating them and aren't interested in the body when generating a
+  // parent context.
+  void VisitLambdaExpr(const LambdaExpr *LE) {}
 
-    void VisitReturnStmt(const ReturnStmt *S) {
-      RecordStmtCount(S);
-      if (S->getRetValue())
-        Visit(S->getRetValue());
-      PGO.setCurrentRegionUnreachable();
-      RecordNextStmtCount = true;
-    }
+  void VisitCapturedDecl(const CapturedDecl *D) {
+    // Counter tracks entry to the capture body.
+    uint64_t BodyCount = setCount(PGO.getRegionCount(D->getBody()));
+    CountMap[D->getBody()] = BodyCount;
+    Visit(D->getBody());
+  }
 
-    void VisitGotoStmt(const GotoStmt *S) {
-      RecordStmtCount(S);
-      PGO.setCurrentRegionUnreachable();
-      RecordNextStmtCount = true;
-    }
+  void VisitObjCMethodDecl(const ObjCMethodDecl *D) {
+    // Counter tracks entry to the method body.
+    uint64_t BodyCount = setCount(PGO.getRegionCount(D->getBody()));
+    CountMap[D->getBody()] = BodyCount;
+    Visit(D->getBody());
+  }
 
-    void VisitLabelStmt(const LabelStmt *S) {
-      RecordNextStmtCount = false;
-      // Counter tracks the block following the label.
-      RegionCounter Cnt(PGO, S);
-      Cnt.beginRegion();
-      CountMap[S] = PGO.getCurrentRegionCount();
-      Visit(S->getSubStmt());
-    }
+  void VisitBlockDecl(const BlockDecl *D) {
+    // Counter tracks entry to the block body.
+    uint64_t BodyCount = setCount(PGO.getRegionCount(D->getBody()));
+    CountMap[D->getBody()] = BodyCount;
+    Visit(D->getBody());
+  }
 
-    void VisitBreakStmt(const BreakStmt *S) {
-      RecordStmtCount(S);
-      assert(!BreakContinueStack.empty() && "break not in a loop or switch!");
-      BreakContinueStack.back().BreakCount += PGO.getCurrentRegionCount();
-      PGO.setCurrentRegionUnreachable();
-      RecordNextStmtCount = true;
-    }
+  void VisitReturnStmt(const ReturnStmt *S) {
+    RecordStmtCount(S);
+    if (S->getRetValue())
+      Visit(S->getRetValue());
+    CurrentCount = 0;
+    RecordNextStmtCount = true;
+  }
 
-    void VisitContinueStmt(const ContinueStmt *S) {
-      RecordStmtCount(S);
-      assert(!BreakContinueStack.empty() && "continue stmt not in a loop!");
-      BreakContinueStack.back().ContinueCount += PGO.getCurrentRegionCount();
-      PGO.setCurrentRegionUnreachable();
-      RecordNextStmtCount = true;
-    }
+  void VisitCXXThrowExpr(const CXXThrowExpr *E) {
+    RecordStmtCount(E);
+    if (E->getSubExpr())
+      Visit(E->getSubExpr());
+    CurrentCount = 0;
+    RecordNextStmtCount = true;
+  }
 
-    void VisitWhileStmt(const WhileStmt *S) {
-      RecordStmtCount(S);
-      // Counter tracks the body of the loop.
-      RegionCounter Cnt(PGO, S);
-      BreakContinueStack.push_back(BreakContinue());
-      // Visit the body region first so the break/continue adjustments can be
-      // included when visiting the condition.
-      Cnt.beginRegion();
-      CountMap[S->getBody()] = PGO.getCurrentRegionCount();
-      Visit(S->getBody());
-      Cnt.adjustForControlFlow();
-
-      // ...then go back and propagate counts through the condition. The count
-      // at the start of the condition is the sum of the incoming edges,
-      // the backedge from the end of the loop body, and the edges from
-      // continue statements.
-      BreakContinue BC = BreakContinueStack.pop_back_val();
-      Cnt.setCurrentRegionCount(Cnt.getParentCount() +
-                                Cnt.getAdjustedCount() + BC.ContinueCount);
-      CountMap[S->getCond()] = PGO.getCurrentRegionCount();
-      Visit(S->getCond());
-      Cnt.adjustForControlFlow();
-      Cnt.applyAdjustmentsToRegion(BC.BreakCount + BC.ContinueCount);
-      RecordNextStmtCount = true;
-    }
+  void VisitGotoStmt(const GotoStmt *S) {
+    RecordStmtCount(S);
+    CurrentCount = 0;
+    RecordNextStmtCount = true;
+  }
 
-    void VisitDoStmt(const DoStmt *S) {
-      RecordStmtCount(S);
-      // Counter tracks the body of the loop.
-      RegionCounter Cnt(PGO, S);
-      BreakContinueStack.push_back(BreakContinue());
-      Cnt.beginRegion(/*AddIncomingFallThrough=*/true);
-      CountMap[S->getBody()] = PGO.getCurrentRegionCount();
-      Visit(S->getBody());
-      Cnt.adjustForControlFlow();
-
-      BreakContinue BC = BreakContinueStack.pop_back_val();
-      // The count at the start of the condition is equal to the count at the
-      // end of the body. The adjusted count does not include either the
-      // fall-through count coming into the loop or the continue count, so add
-      // both of those separately. This is coincidentally the same equation as
-      // with while loops but for different reasons.
-      Cnt.setCurrentRegionCount(Cnt.getParentCount() +
-                                Cnt.getAdjustedCount() + BC.ContinueCount);
-      CountMap[S->getCond()] = PGO.getCurrentRegionCount();
-      Visit(S->getCond());
-      Cnt.adjustForControlFlow();
-      Cnt.applyAdjustmentsToRegion(BC.BreakCount + BC.ContinueCount);
-      RecordNextStmtCount = true;
-    }
+  void VisitLabelStmt(const LabelStmt *S) {
+    RecordNextStmtCount = false;
+    // Counter tracks the block following the label.
+    uint64_t BlockCount = setCount(PGO.getRegionCount(S));
+    CountMap[S] = BlockCount;
+    Visit(S->getSubStmt());
+  }
 
-    void VisitForStmt(const ForStmt *S) {
-      RecordStmtCount(S);
-      if (S->getInit())
-        Visit(S->getInit());
-      // Counter tracks the body of the loop.
-      RegionCounter Cnt(PGO, S);
-      BreakContinueStack.push_back(BreakContinue());
-      // Visit the body region first. (This is basically the same as a while
-      // loop; see further comments in VisitWhileStmt.)
-      Cnt.beginRegion();
-      CountMap[S->getBody()] = PGO.getCurrentRegionCount();
-      Visit(S->getBody());
-      Cnt.adjustForControlFlow();
-
-      // The increment is essentially part of the body but it needs to include
-      // the count for all the continue statements.
-      if (S->getInc()) {
-        Cnt.setCurrentRegionCount(PGO.getCurrentRegionCount() +
-                                  BreakContinueStack.back().ContinueCount);
-        CountMap[S->getInc()] = PGO.getCurrentRegionCount();
-        Visit(S->getInc());
-        Cnt.adjustForControlFlow();
-      }
-
-      BreakContinue BC = BreakContinueStack.pop_back_val();
-
-      // ...then go back and propagate counts through the condition.
-      if (S->getCond()) {
-        Cnt.setCurrentRegionCount(Cnt.getParentCount() +
-                                  Cnt.getAdjustedCount() +
-                                  BC.ContinueCount);
-        CountMap[S->getCond()] = PGO.getCurrentRegionCount();
-        Visit(S->getCond());
-        Cnt.adjustForControlFlow();
-      }
-      Cnt.applyAdjustmentsToRegion(BC.BreakCount + BC.ContinueCount);
-      RecordNextStmtCount = true;
-    }
+  void VisitBreakStmt(const BreakStmt *S) {
+    RecordStmtCount(S);
+    assert(!BreakContinueStack.empty() && "break not in a loop or switch!");
+    BreakContinueStack.back().BreakCount += CurrentCount;
+    CurrentCount = 0;
+    RecordNextStmtCount = true;
+  }
 
-    void VisitCXXForRangeStmt(const CXXForRangeStmt *S) {
-      RecordStmtCount(S);
-      Visit(S->getRangeStmt());
-      Visit(S->getBeginEndStmt());
-      // Counter tracks the body of the loop.
-      RegionCounter Cnt(PGO, S);
-      BreakContinueStack.push_back(BreakContinue());
-      // Visit the body region first. (This is basically the same as a while
-      // loop; see further comments in VisitWhileStmt.)
-      Cnt.beginRegion();
-      CountMap[S->getLoopVarStmt()] = PGO.getCurrentRegionCount();
-      Visit(S->getLoopVarStmt());
-      Visit(S->getBody());
-      Cnt.adjustForControlFlow();
-
-      // The increment is essentially part of the body but it needs to include
-      // the count for all the continue statements.
-      Cnt.setCurrentRegionCount(PGO.getCurrentRegionCount() +
-                                BreakContinueStack.back().ContinueCount);
-      CountMap[S->getInc()] = PGO.getCurrentRegionCount();
-      Visit(S->getInc());
-      Cnt.adjustForControlFlow();
+  void VisitContinueStmt(const ContinueStmt *S) {
+    RecordStmtCount(S);
+    assert(!BreakContinueStack.empty() && "continue stmt not in a loop!");
+    BreakContinueStack.back().ContinueCount += CurrentCount;
+    CurrentCount = 0;
+    RecordNextStmtCount = true;
+  }
 
-      BreakContinue BC = BreakContinueStack.pop_back_val();
+  void VisitWhileStmt(const WhileStmt *S) {
+    RecordStmtCount(S);
+    uint64_t ParentCount = CurrentCount;
+
+    BreakContinueStack.push_back(BreakContinue());
+    // Visit the body region first so the break/continue adjustments can be
+    // included when visiting the condition.
+    uint64_t BodyCount = setCount(PGO.getRegionCount(S));
+    CountMap[S->getBody()] = CurrentCount;
+    Visit(S->getBody());
+    uint64_t BackedgeCount = CurrentCount;
+
+    // ...then go back and propagate counts through the condition. The count
+    // at the start of the condition is the sum of the incoming edges,
+    // the backedge from the end of the loop body, and the edges from
+    // continue statements.
+    BreakContinue BC = BreakContinueStack.pop_back_val();
+    uint64_t CondCount =
+        setCount(ParentCount + BackedgeCount + BC.ContinueCount);
+    CountMap[S->getCond()] = CondCount;
+    Visit(S->getCond());
+    setCount(BC.BreakCount + CondCount - BodyCount);
+    RecordNextStmtCount = true;
+  }
 
-      // ...then go back and propagate counts through the condition.
-      Cnt.setCurrentRegionCount(Cnt.getParentCount() +
-                                Cnt.getAdjustedCount() +
-                                BC.ContinueCount);
-      CountMap[S->getCond()] = PGO.getCurrentRegionCount();
-      Visit(S->getCond());
-      Cnt.adjustForControlFlow();
-      Cnt.applyAdjustmentsToRegion(BC.BreakCount + BC.ContinueCount);
-      RecordNextStmtCount = true;
-    }
+  void VisitDoStmt(const DoStmt *S) {
+    RecordStmtCount(S);
+    uint64_t LoopCount = PGO.getRegionCount(S);
+
+    BreakContinueStack.push_back(BreakContinue());
+    // The count doesn't include the fallthrough from the parent scope. Add it.
+    uint64_t BodyCount = setCount(LoopCount + CurrentCount);
+    CountMap[S->getBody()] = BodyCount;
+    Visit(S->getBody());
+    uint64_t BackedgeCount = CurrentCount;
+
+    BreakContinue BC = BreakContinueStack.pop_back_val();
+    // The count at the start of the condition is equal to the count at the
+    // end of the body, plus any continues.
+    uint64_t CondCount = setCount(BackedgeCount + BC.ContinueCount);
+    CountMap[S->getCond()] = CondCount;
+    Visit(S->getCond());
+    setCount(BC.BreakCount + CondCount - LoopCount);
+    RecordNextStmtCount = true;
+  }
 
-    void VisitObjCForCollectionStmt(const ObjCForCollectionStmt *S) {
-      RecordStmtCount(S);
-      Visit(S->getElement());
-      // Counter tracks the body of the loop.
-      RegionCounter Cnt(PGO, S);
-      BreakContinueStack.push_back(BreakContinue());
-      Cnt.beginRegion();
-      CountMap[S->getBody()] = PGO.getCurrentRegionCount();
-      Visit(S->getBody());
-      BreakContinue BC = BreakContinueStack.pop_back_val();
-      Cnt.adjustForControlFlow();
-      Cnt.applyAdjustmentsToRegion(BC.BreakCount + BC.ContinueCount);
-      RecordNextStmtCount = true;
+  void VisitForStmt(const ForStmt *S) {
+    RecordStmtCount(S);
+    if (S->getInit())
+      Visit(S->getInit());
+
+    uint64_t ParentCount = CurrentCount;
+
+    BreakContinueStack.push_back(BreakContinue());
+    // Visit the body region first. (This is basically the same as a while
+    // loop; see further comments in VisitWhileStmt.)
+    uint64_t BodyCount = setCount(PGO.getRegionCount(S));
+    CountMap[S->getBody()] = BodyCount;
+    Visit(S->getBody());
+    uint64_t BackedgeCount = CurrentCount;
+    BreakContinue BC = BreakContinueStack.pop_back_val();
+
+    // The increment is essentially part of the body but it needs to include
+    // the count for all the continue statements.
+    if (S->getInc()) {
+      uint64_t IncCount = setCount(BackedgeCount + BC.ContinueCount);
+      CountMap[S->getInc()] = IncCount;
+      Visit(S->getInc());
     }
 
-    void VisitSwitchStmt(const SwitchStmt *S) {
-      RecordStmtCount(S);
+    // ...then go back and propagate counts through the condition.
+    uint64_t CondCount =
+        setCount(ParentCount + BackedgeCount + BC.ContinueCount);
+    if (S->getCond()) {
+      CountMap[S->getCond()] = CondCount;
       Visit(S->getCond());
-      PGO.setCurrentRegionUnreachable();
-      BreakContinueStack.push_back(BreakContinue());
-      Visit(S->getBody());
-      // If the switch is inside a loop, add the continue counts.
-      BreakContinue BC = BreakContinueStack.pop_back_val();
-      if (!BreakContinueStack.empty())
-        BreakContinueStack.back().ContinueCount += BC.ContinueCount;
-      // Counter tracks the exit block of the switch.
-      RegionCounter ExitCnt(PGO, S);
-      ExitCnt.beginRegion();
-      RecordNextStmtCount = true;
     }
+    setCount(BC.BreakCount + CondCount - BodyCount);
+    RecordNextStmtCount = true;
+  }
 
-    void VisitCaseStmt(const CaseStmt *S) {
-      RecordNextStmtCount = false;
-      // Counter for this particular case. This counts only jumps from the
-      // switch header and does not include fallthrough from the case before
-      // this one.
-      RegionCounter Cnt(PGO, S);
-      Cnt.beginRegion(/*AddIncomingFallThrough=*/true);
-      CountMap[S] = Cnt.getCount();
-      RecordNextStmtCount = true;
-      Visit(S->getSubStmt());
-    }
+  void VisitCXXForRangeStmt(const CXXForRangeStmt *S) {
+    RecordStmtCount(S);
+    Visit(S->getLoopVarStmt());
+    Visit(S->getRangeStmt());
+    Visit(S->getBeginEndStmt());
+
+    uint64_t ParentCount = CurrentCount;
+    BreakContinueStack.push_back(BreakContinue());
+    // Visit the body region first. (This is basically the same as a while
+    // loop; see further comments in VisitWhileStmt.)
+    uint64_t BodyCount = setCount(PGO.getRegionCount(S));
+    CountMap[S->getBody()] = BodyCount;
+    Visit(S->getBody());
+    uint64_t BackedgeCount = CurrentCount;
+    BreakContinue BC = BreakContinueStack.pop_back_val();
+
+    // The increment is essentially part of the body but it needs to include
+    // the count for all the continue statements.
+    uint64_t IncCount = setCount(BackedgeCount + BC.ContinueCount);
+    CountMap[S->getInc()] = IncCount;
+    Visit(S->getInc());
+
+    // ...then go back and propagate counts through the condition.
+    uint64_t CondCount =
+        setCount(ParentCount + BackedgeCount + BC.ContinueCount);
+    CountMap[S->getCond()] = CondCount;
+    Visit(S->getCond());
+    setCount(BC.BreakCount + CondCount - BodyCount);
+    RecordNextStmtCount = true;
+  }
 
-    void VisitDefaultStmt(const DefaultStmt *S) {
-      RecordNextStmtCount = false;
-      // Counter for this default case. This does not include fallthrough from
-      // the previous case.
-      RegionCounter Cnt(PGO, S);
-      Cnt.beginRegion(/*AddIncomingFallThrough=*/true);
-      CountMap[S] = Cnt.getCount();
-      RecordNextStmtCount = true;
-      Visit(S->getSubStmt());
-    }
+  void VisitObjCForCollectionStmt(const ObjCForCollectionStmt *S) {
+    RecordStmtCount(S);
+    Visit(S->getElement());
+    uint64_t ParentCount = CurrentCount;
+    BreakContinueStack.push_back(BreakContinue());
+    // Counter tracks the body of the loop.
+    uint64_t BodyCount = setCount(PGO.getRegionCount(S));
+    CountMap[S->getBody()] = BodyCount;
+    Visit(S->getBody());
+    uint64_t BackedgeCount = CurrentCount;
+    BreakContinue BC = BreakContinueStack.pop_back_val();
+
+    setCount(BC.BreakCount + ParentCount + BackedgeCount + BC.ContinueCount -
+             BodyCount);
+    RecordNextStmtCount = true;
+  }
 
-    void VisitIfStmt(const IfStmt *S) {
-      RecordStmtCount(S);
-      // Counter tracks the "then" part of an if statement. The count for
-      // the "else" part, if it exists, will be calculated from this counter.
-      RegionCounter Cnt(PGO, S);
-      Visit(S->getCond());
+  void VisitSwitchStmt(const SwitchStmt *S) {
+    RecordStmtCount(S);
+    Visit(S->getCond());
+    CurrentCount = 0;
+    BreakContinueStack.push_back(BreakContinue());
+    Visit(S->getBody());
+    // If the switch is inside a loop, add the continue counts.
+    BreakContinue BC = BreakContinueStack.pop_back_val();
+    if (!BreakContinueStack.empty())
+      BreakContinueStack.back().ContinueCount += BC.ContinueCount;
+    // Counter tracks the exit block of the switch.
+    setCount(PGO.getRegionCount(S));
+    RecordNextStmtCount = true;
+  }
 
-      Cnt.beginRegion();
-      CountMap[S->getThen()] = PGO.getCurrentRegionCount();
-      Visit(S->getThen());
-      Cnt.adjustForControlFlow();
-
-      if (S->getElse()) {
-        Cnt.beginElseRegion();
-        CountMap[S->getElse()] = PGO.getCurrentRegionCount();
-        Visit(S->getElse());
-        Cnt.adjustForControlFlow();
-      }
-      Cnt.applyAdjustmentsToRegion(0);
-      RecordNextStmtCount = true;
-    }
+  void VisitSwitchCase(const SwitchCase *S) {
+    RecordNextStmtCount = false;
+    // Counter for this particular case. This counts only jumps from the
+    // switch header and does not include fallthrough from the case before
+    // this one.
+    uint64_t CaseCount = PGO.getRegionCount(S);
+    setCount(CurrentCount + CaseCount);
+    // We need the count without fallthrough in the mapping, so it's more useful
+    // for branch probabilities.
+    CountMap[S] = CaseCount;
+    RecordNextStmtCount = true;
+    Visit(S->getSubStmt());
+  }
 
-    void VisitCXXTryStmt(const CXXTryStmt *S) {
-      RecordStmtCount(S);
-      Visit(S->getTryBlock());
-      for (unsigned I = 0, E = S->getNumHandlers(); I < E; ++I)
-        Visit(S->getHandler(I));
-      // Counter tracks the continuation block of the try statement.
-      RegionCounter Cnt(PGO, S);
-      Cnt.beginRegion();
-      RecordNextStmtCount = true;
-    }
+  void VisitIfStmt(const IfStmt *S) {
+    RecordStmtCount(S);
+    uint64_t ParentCount = CurrentCount;
+    Visit(S->getCond());
+
+    // Counter tracks the "then" part of an if statement. The count for
+    // the "else" part, if it exists, will be calculated from this counter.
+    uint64_t ThenCount = setCount(PGO.getRegionCount(S));
+    CountMap[S->getThen()] = ThenCount;
+    Visit(S->getThen());
+    uint64_t OutCount = CurrentCount;
+
+    uint64_t ElseCount = ParentCount - ThenCount;
+    if (S->getElse()) {
+      setCount(ElseCount);
+      CountMap[S->getElse()] = ElseCount;
+      Visit(S->getElse());
+      OutCount += CurrentCount;
+    } else
+      OutCount += ElseCount;
+    setCount(OutCount);
+    RecordNextStmtCount = true;
+  }
 
-    void VisitCXXCatchStmt(const CXXCatchStmt *S) {
-      RecordNextStmtCount = false;
-      // Counter tracks the catch statement's handler block.
-      RegionCounter Cnt(PGO, S);
-      Cnt.beginRegion();
-      CountMap[S] = PGO.getCurrentRegionCount();
-      Visit(S->getHandlerBlock());
-    }
+  void VisitCXXTryStmt(const CXXTryStmt *S) {
+    RecordStmtCount(S);
+    Visit(S->getTryBlock());
+    for (unsigned I = 0, E = S->getNumHandlers(); I < E; ++I)
+      Visit(S->getHandler(I));
+    // Counter tracks the continuation block of the try statement.
+    setCount(PGO.getRegionCount(S));
+    RecordNextStmtCount = true;
+  }
 
-    void VisitAbstractConditionalOperator(
-        const AbstractConditionalOperator *E) {
-      RecordStmtCount(E);
-      // Counter tracks the "true" part of a conditional operator. The
-      // count in the "false" part will be calculated from this counter.
-      RegionCounter Cnt(PGO, E);
-      Visit(E->getCond());
-
-      Cnt.beginRegion();
-      CountMap[E->getTrueExpr()] = PGO.getCurrentRegionCount();
-      Visit(E->getTrueExpr());
-      Cnt.adjustForControlFlow();
-
-      Cnt.beginElseRegion();
-      CountMap[E->getFalseExpr()] = PGO.getCurrentRegionCount();
-      Visit(E->getFalseExpr());
-      Cnt.adjustForControlFlow();
-
-      Cnt.applyAdjustmentsToRegion(0);
-      RecordNextStmtCount = true;
-    }
+  void VisitCXXCatchStmt(const CXXCatchStmt *S) {
+    RecordNextStmtCount = false;
+    // Counter tracks the catch statement's handler block.
+    uint64_t CatchCount = setCount(PGO.getRegionCount(S));
+    CountMap[S] = CatchCount;
+    Visit(S->getHandlerBlock());
+  }
 
-    void VisitBinLAnd(const BinaryOperator *E) {
-      RecordStmtCount(E);
-      // Counter tracks the right hand side of a logical and operator.
-      RegionCounter Cnt(PGO, E);
-      Visit(E->getLHS());
-      Cnt.beginRegion();
-      CountMap[E->getRHS()] = PGO.getCurrentRegionCount();
-      Visit(E->getRHS());
-      Cnt.adjustForControlFlow();
-      Cnt.applyAdjustmentsToRegion(0);
-      RecordNextStmtCount = true;
-    }
+  void VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
+    RecordStmtCount(E);
+    uint64_t ParentCount = CurrentCount;
+    Visit(E->getCond());
+
+    // Counter tracks the "true" part of a conditional operator. The
+    // count in the "false" part will be calculated from this counter.
+    uint64_t TrueCount = setCount(PGO.getRegionCount(E));
+    CountMap[E->getTrueExpr()] = TrueCount;
+    Visit(E->getTrueExpr());
+    uint64_t OutCount = CurrentCount;
+
+    uint64_t FalseCount = setCount(ParentCount - TrueCount);
+    CountMap[E->getFalseExpr()] = FalseCount;
+    Visit(E->getFalseExpr());
+    OutCount += CurrentCount;
+
+    setCount(OutCount);
+    RecordNextStmtCount = true;
+  }
 
-    void VisitBinLOr(const BinaryOperator *E) {
-      RecordStmtCount(E);
-      // Counter tracks the right hand side of a logical or operator.
-      RegionCounter Cnt(PGO, E);
-      Visit(E->getLHS());
-      Cnt.beginRegion();
-      CountMap[E->getRHS()] = PGO.getCurrentRegionCount();
-      Visit(E->getRHS());
-      Cnt.adjustForControlFlow();
-      Cnt.applyAdjustmentsToRegion(0);
-      RecordNextStmtCount = true;
-    }
-  };
+  void VisitBinLAnd(const BinaryOperator *E) {
+    RecordStmtCount(E);
+    uint64_t ParentCount = CurrentCount;
+    Visit(E->getLHS());
+    // Counter tracks the right hand side of a logical and operator.
+    uint64_t RHSCount = setCount(PGO.getRegionCount(E));
+    CountMap[E->getRHS()] = RHSCount;
+    Visit(E->getRHS());
+    setCount(ParentCount + RHSCount - CurrentCount);
+    RecordNextStmtCount = true;
+  }
+
+  void VisitBinLOr(const BinaryOperator *E) {
+    RecordStmtCount(E);
+    uint64_t ParentCount = CurrentCount;
+    Visit(E->getLHS());
+    // Counter tracks the right hand side of a logical or operator.
+    uint64_t RHSCount = setCount(PGO.getRegionCount(E));
+    CountMap[E->getRHS()] = RHSCount;
+    Visit(E->getRHS());
+    setCount(ParentCount + RHSCount - CurrentCount);
+    RecordNextStmtCount = true;
+  }
+};
 }
 
 void PGOHash::combine(HashType Type) {
@@ -728,12 +719,10 @@ void CodeGenPGO::emitCounterRegionMapping(const Decl *D) {
 }
 
 void
-CodeGenPGO::emitEmptyCounterMapping(const Decl *D, StringRef FuncName,
+CodeGenPGO::emitEmptyCounterMapping(const Decl *D, StringRef Name,
                                     llvm::GlobalValue::LinkageTypes Linkage) {
   if (SkipCoverageMapping)
     return;
-  setFuncName(FuncName, Linkage);
-
   // Don't map the functions inside the system headers
   auto Loc = D->getBody()->getLocStart();
   if (CGM.getContext().getSourceManager().isInSystemHeader(Loc))
@@ -750,6 +739,7 @@ CodeGenPGO::emitEmptyCounterMapping(const Decl *D, StringRef FuncName,
   if (CoverageMapping.empty())
     return;
 
+  setFuncName(Name, Linkage);
   CGM.getCoverageMapping()->addFunctionMappingRecord(
       FuncNameVar, FuncName, FunctionHash, CoverageMapping);
 }
@@ -785,17 +775,19 @@ CodeGenPGO::applyFunctionAttributes(llvm::IndexedInstrProfReader *PGOReader,
     Fn->addFnAttr(llvm::Attribute::Cold);
 }
 
-void CodeGenPGO::emitCounterIncrement(CGBuilderTy &Builder, unsigned Counter) {
+void CodeGenPGO::emitCounterIncrement(CGBuilderTy &Builder, const Stmt *S) {
   if (!CGM.getCodeGenOpts().ProfileInstrGenerate || !RegionCounterMap)
     return;
   if (!Builder.GetInsertPoint())
     return;
+
+  unsigned Counter = (*RegionCounterMap)[S];
   auto *I8PtrTy = llvm::Type::getInt8PtrTy(CGM.getLLVMContext());
-  Builder.CreateCall4(CGM.getIntrinsic(llvm::Intrinsic::instrprof_increment),
-                      llvm::ConstantExpr::getBitCast(FuncNameVar, I8PtrTy),
+  Builder.CreateCall(CGM.getIntrinsic(llvm::Intrinsic::instrprof_increment),
+                     {llvm::ConstantExpr::getBitCast(FuncNameVar, I8PtrTy),
                       Builder.getInt64(FunctionHash),
                       Builder.getInt32(NumRegionCounters),
-                      Builder.getInt32(Counter));
+                      Builder.getInt32(Counter)});
 }
 
 void CodeGenPGO::loadRegionCounts(llvm::IndexedInstrProfReader *PGOReader,
@@ -839,8 +831,8 @@ static uint32_t scaleBranchWeight(uint64_t Weight, uint64_t Scale) {
   return Scaled;
 }
 
-llvm::MDNode *CodeGenPGO::createBranchWeights(uint64_t TrueCount,
-                                              uint64_t FalseCount) {
+llvm::MDNode *CodeGenFunction::createProfileWeights(uint64_t TrueCount,
+                                                    uint64_t FalseCount) {
   // Check for empty weights.
   if (!TrueCount && !FalseCount)
     return nullptr;
@@ -853,7 +845,8 @@ llvm::MDNode *CodeGenPGO::createBranchWeights(uint64_t TrueCount,
                                       scaleBranchWeight(FalseCount, Scale));
 }
 
-llvm::MDNode *CodeGenPGO::createBranchWeights(ArrayRef<uint64_t> Weights) {
+llvm::MDNode *
+CodeGenFunction::createProfileWeights(ArrayRef<uint64_t> Weights) {
   // We need at least two elements to create meaningful weights.
   if (Weights.size() < 2)
     return nullptr;
@@ -875,17 +868,14 @@ llvm::MDNode *CodeGenPGO::createBranchWeights(ArrayRef<uint64_t> Weights) {
   return MDHelper.createBranchWeights(ScaledWeights);
 }
 
-llvm::MDNode *CodeGenPGO::createLoopWeights(const Stmt *Cond,
-                                            RegionCounter &Cnt) {
-  if (!haveRegionCounts())
+llvm::MDNode *CodeGenFunction::createProfileWeightsForLoop(const Stmt *Cond,
+                                                           uint64_t LoopCount) {
+  if (!PGO.haveRegionCounts())
     return nullptr;
-  uint64_t LoopCount = Cnt.getCount();
-  uint64_t CondCount = 0;
-  bool Found = getStmtCount(Cond, CondCount);
-  assert(Found && "missing expected loop condition count");
-  (void)Found;
-  if (CondCount == 0)
+  Optional<uint64_t> CondCount = PGO.getStmtCount(Cond);
+  assert(CondCount.hasValue() && "missing expected loop condition count");
+  if (*CondCount == 0)
     return nullptr;
-  return createBranchWeights(LoopCount,
-                             std::max(CondCount, LoopCount) - LoopCount);
+  return createProfileWeights(LoopCount,
+                              std::max(*CondCount, LoopCount) - LoopCount);
 }
diff --git a/lib/CodeGen/CodeGenPGO.h b/lib/CodeGen/CodeGenPGO.h
index 431c850ef81e..de6f369fb351 100644
--- a/lib/CodeGen/CodeGenPGO.h
+++ b/lib/CodeGen/CodeGenPGO.h
@@ -24,10 +24,8 @@
 
 namespace clang {
 namespace CodeGen {
-class RegionCounter;
 
-/// Per-function PGO state. This class should generally not be used directly,
-/// but instead through the CodeGenFunction and RegionCounter types.
+/// Per-function PGO state.
 class CodeGenPGO {
 private:
   CodeGenModule &CGM;
@@ -62,37 +60,24 @@ public:
   /// exits.
   void setCurrentRegionCount(uint64_t Count) { CurrentRegionCount = Count; }
 
-  /// Indicate that the current region is never reached, and thus should have a
-  /// counter value of zero. This is important so that subsequent regions can
-  /// correctly track their parent counts.
-  void setCurrentRegionUnreachable() { setCurrentRegionCount(0); }
-
   /// Check if an execution count is known for a given statement. If so, return
   /// true and put the value in Count; else return false.
-  bool getStmtCount(const Stmt *S, uint64_t &Count) {
+  Optional<uint64_t> getStmtCount(const Stmt *S) {
     if (!StmtCountMap)
-      return false;
-    llvm::DenseMap<const Stmt*, uint64_t>::const_iterator
-      I = StmtCountMap->find(S);
+      return None;
+    auto I = StmtCountMap->find(S);
     if (I == StmtCountMap->end())
-      return false;
-    Count = I->second;
-    return true;
+      return None;
+    return I->second;
   }
 
   /// If the execution count for the current statement is known, record that
   /// as the current count.
   void setCurrentStmt(const Stmt *S) {
-    uint64_t Count;
-    if (getStmtCount(S, Count))
-      setCurrentRegionCount(Count);
+    if (auto Count = getStmtCount(S))
+      setCurrentRegionCount(*Count);
   }
 
-  /// Calculate branch weights appropriate for PGO data
-  llvm::MDNode *createBranchWeights(uint64_t TrueCount, uint64_t FalseCount);
-  llvm::MDNode *createBranchWeights(ArrayRef<uint64_t> Weights);
-  llvm::MDNode *createLoopWeights(const Stmt *Cond, RegionCounter &Cnt);
-
   /// Check if we need to emit coverage mapping for a given declaration
   void checkGlobalDecl(GlobalDecl GD);
   /// Assign counters to regions and configure them for PGO of a given
@@ -117,110 +102,16 @@ private:
   void emitCounterVariables();
   void emitCounterRegionMapping(const Decl *D);
 
-  /// Emit code to increment the counter at the given index
-  void emitCounterIncrement(CGBuilderTy &Builder, unsigned Counter);
+public:
+  void emitCounterIncrement(CGBuilderTy &Builder, const Stmt *S);
 
-  /// Return the region counter for the given statement. This should only be
-  /// called on statements that have a dedicated counter.
-  unsigned getRegionCounter(const Stmt *S) {
+  /// Return the region count for the counter at the given index.
+  uint64_t getRegionCount(const Stmt *S) {
     if (!RegionCounterMap)
       return 0;
-    return (*RegionCounterMap)[S];
-  }
-
-  /// Return the region count for the counter at the given index.
-  uint64_t getRegionCount(unsigned Counter) {
     if (!haveRegionCounts())
       return 0;
-    return RegionCounts[Counter];
-  }
-
-  friend class RegionCounter;
-};
-
-/// A counter for a particular region. This is the primary interface through
-/// which clients manage PGO counters and their values.
-class RegionCounter {
-  CodeGenPGO *PGO;
-  unsigned Counter;
-  uint64_t Count;
-  uint64_t ParentCount;
-  uint64_t RegionCount;
-  int64_t Adjust;
-
-  RegionCounter(CodeGenPGO &PGO, unsigned CounterIndex)
-    : PGO(&PGO), Counter(CounterIndex), Count(PGO.getRegionCount(Counter)),
-      ParentCount(PGO.getCurrentRegionCount()), Adjust(0) {}
-
-public:
-  RegionCounter(CodeGenPGO &PGO, const Stmt *S)
-    : PGO(&PGO), Counter(PGO.getRegionCounter(S)),
-      Count(PGO.getRegionCount(Counter)),
-      ParentCount(PGO.getCurrentRegionCount()), Adjust(0) {}
-
-  /// Get the value of the counter. In most cases this is the number of times
-  /// the region of the counter was entered, but for switch labels it's the
-  /// number of direct jumps to that label.
-  uint64_t getCount() const { return Count; }
-
-  /// Get the value of the counter with adjustments applied. Adjustments occur
-  /// when control enters or leaves the region abnormally; i.e., if there is a
-  /// jump to a label within the region, or if the function can return from
-  /// within the region. The adjusted count, then, is the value of the counter
-  /// at the end of the region.
-  uint64_t getAdjustedCount() const {
-    return Count + Adjust;
-  }
-
-  /// Get the value of the counter in this region's parent, i.e., the region
-  /// that was active when this region began. This is useful for deriving
-  /// counts in implicitly counted regions, like the false case of a condition
-  /// or the normal exits of a loop.
-  uint64_t getParentCount() const { return ParentCount; }
-
-  /// Activate the counter by emitting an increment and starting to track
-  /// adjustments. If AddIncomingFallThrough is true, the current region count
-  /// will be added to the counter for the purposes of tracking the region.
-  void beginRegion(CGBuilderTy &Builder, bool AddIncomingFallThrough=false) {
-    beginRegion(AddIncomingFallThrough);
-    PGO->emitCounterIncrement(Builder, Counter);
-  }
-  void beginRegion(bool AddIncomingFallThrough=false) {
-    RegionCount = Count;
-    if (AddIncomingFallThrough)
-      RegionCount += PGO->getCurrentRegionCount();
-    PGO->setCurrentRegionCount(RegionCount);
-  }
-
-  /// For counters on boolean branches, begins tracking adjustments for the
-  /// uncounted path.
-  void beginElseRegion() {
-    RegionCount = ParentCount - Count;
-    PGO->setCurrentRegionCount(RegionCount);
-  }
-
-  /// Reset the current region count.
-  void setCurrentRegionCount(uint64_t CurrentCount) {
-    RegionCount = CurrentCount;
-    PGO->setCurrentRegionCount(RegionCount);
-  }
-
-  /// Adjust for non-local control flow after emitting a subexpression or
-  /// substatement. This must be called to account for constructs such as gotos,
-  /// labels, and returns, so that we can ensure that our region's count is
-  /// correct in the code that follows.
-  void adjustForControlFlow() {
-    Adjust += PGO->getCurrentRegionCount() - RegionCount;
-    // Reset the region count in case this is called again later.
-    RegionCount = PGO->getCurrentRegionCount();
-  }
-
-  /// Commit all adjustments to the current region. If the region is a loop,
-  /// the LoopAdjust value should be the count of all the breaks and continues
-  /// from the loop, to compensate for those counts being deducted from the
-  /// adjustments for the body of the loop.
-  void applyAdjustmentsToRegion(uint64_t LoopAdjust) {
-    PGO->setCurrentRegionCount(ParentCount + Adjust + LoopAdjust);
+    return RegionCounts[(*RegionCounterMap)[S]];
   }
 };
 
diff --git a/lib/CodeGen/CodeGenTypes.cpp b/lib/CodeGen/CodeGenTypes.cpp
index 67a9fbec2690..e0f926cabd71 100644
--- a/lib/CodeGen/CodeGenTypes.cpp
+++ b/lib/CodeGen/CodeGenTypes.cpp
@@ -715,9 +715,16 @@ bool CodeGenTypes::isZeroInitializable(QualType T) {
   // No need to check for member pointers when not compiling C++.
   if (!Context.getLangOpts().CPlusPlus)
     return true;
-  
-  T = Context.getBaseElementType(T);
-  
+
+  if (const auto *AT = Context.getAsArrayType(T)) {
+    if (isa<IncompleteArrayType>(AT))
+      return true;
+    if (const auto *CAT = dyn_cast<ConstantArrayType>(AT))
+      if (Context.getConstantArrayElementCount(CAT) == 0)
+        return true;
+    T = Context.getBaseElementType(T);
+  }
+
   // Records are non-zero-initializable if they contain any
   // non-zero-initializable subobjects.
   if (const RecordType *RT = T->getAs<RecordType>()) {
@@ -733,6 +740,6 @@ bool CodeGenTypes::isZeroInitializable(QualType T) {
   return true;
 }
 
-bool CodeGenTypes::isZeroInitializable(const CXXRecordDecl *RD) {
+bool CodeGenTypes::isZeroInitializable(const RecordDecl *RD) {
   return getCGRecordLayout(RD).isZeroInitializable();
 }
diff --git a/lib/CodeGen/CodeGenTypes.h b/lib/CodeGen/CodeGenTypes.h
index 64c5799ccec9..1580e21d11dc 100644
--- a/lib/CodeGen/CodeGenTypes.h
+++ b/lib/CodeGen/CodeGenTypes.h
@@ -82,6 +82,9 @@ inline StructorType getFromCtorType(CXXCtorType T) {
     return StructorType::Base;
   case Ctor_Comdat:
     llvm_unreachable("not expecting a COMDAT");
+  case Ctor_CopyingClosure:
+  case Ctor_DefaultClosure:
+    llvm_unreachable("not expecting a closure");
   }
   llvm_unreachable("not a CXXCtorType");
 }
@@ -112,8 +115,8 @@ inline StructorType getFromDtorType(CXXDtorType T) {
   llvm_unreachable("not a CXXDtorType");
 }
 
-/// CodeGenTypes - This class organizes the cross-module state that is used
-/// while lowering AST types to LLVM types.
+/// This class organizes the cross-module state that is used while lowering
+/// AST types to LLVM types.
 class CodeGenTypes {
   CodeGenModule &CGM;
   // Some of this stuff should probably be left on the CGM.
@@ -133,34 +136,32 @@ class CodeGenTypes {
   /// types are never refined.
   llvm::DenseMap<const ObjCInterfaceType*, llvm::Type *> InterfaceTypes;
 
-  /// CGRecordLayouts - This maps llvm struct type with corresponding
-  /// record layout info.
+  /// Maps clang struct type with corresponding record layout info.
   llvm::DenseMap<const Type*, CGRecordLayout *> CGRecordLayouts;
 
-  /// RecordDeclTypes - This contains the LLVM IR type for any converted
-  /// RecordDecl.
+  /// Contains the LLVM IR type for any converted RecordDecl.
   llvm::DenseMap<const Type*, llvm::StructType *> RecordDeclTypes;
   
-  /// FunctionInfos - Hold memoized CGFunctionInfo results.
+  /// Hold memoized CGFunctionInfo results.
   llvm::FoldingSet<CGFunctionInfo> FunctionInfos;
 
-  /// RecordsBeingLaidOut - This set keeps track of records that we're currently
-  /// converting to an IR type.  For example, when converting:
+  /// This set keeps track of records that we're currently converting
+  /// to an IR type.  For example, when converting:
   /// struct A { struct B { int x; } } when processing 'x', the 'A' and 'B'
   /// types will be in this set.
   llvm::SmallPtrSet<const Type*, 4> RecordsBeingLaidOut;
   
   llvm::SmallPtrSet<const CGFunctionInfo*, 4> FunctionsBeingProcessed;
   
-  /// SkippedLayout - True if we didn't layout a function due to a being inside
+  /// True if we didn't layout a function due to a being inside
   /// a recursive struct conversion, set this to true.
   bool SkippedLayout;
 
   SmallVector<const RecordDecl *, 8> DeferredRecords;
   
 private:
-  /// TypeCache - This map keeps cache of llvm::Types
-  /// and maps clang::Type to corresponding llvm::Type.
+  /// This map keeps cache of llvm::Types and maps clang::Type to
+  /// corresponding llvm::Type.
   llvm::DenseMap<const Type *, llvm::Type *> TypeCache;
 
 public:
@@ -261,6 +262,8 @@ public:
                                              const FunctionProtoType *type,
                                              RequiredArgs required);
   const CGFunctionInfo &arrangeMSMemberPointerThunk(const CXXMethodDecl *MD);
+  const CGFunctionInfo &arrangeMSCtorClosure(const CXXConstructorDecl *CD,
+                                                 CXXCtorType CT);
 
   const CGFunctionInfo &arrangeFreeFunctionType(CanQual<FunctionProtoType> Ty);
   const CGFunctionInfo &arrangeFreeFunctionType(CanQual<FunctionNoProtoType> Ty);
@@ -305,7 +308,7 @@ public:  // These are internal details of CGT that shouldn't be used externally.
 
   /// IsZeroInitializable - Return whether a record type can be
   /// zero-initialized (in the C++ sense) with an LLVM zeroinitializer.
-  bool isZeroInitializable(const CXXRecordDecl *RD);
+  bool isZeroInitializable(const RecordDecl *RD);
   
   bool isRecordLayoutComplete(const Type *Ty) const;
   bool noRecordsBeingLaidOut() const {
diff --git a/lib/CodeGen/CoverageMappingGen.cpp b/lib/CodeGen/CoverageMappingGen.cpp
index 6f159d47375e..024a45dba2b7 100644
--- a/lib/CodeGen/CoverageMappingGen.cpp
+++ b/lib/CodeGen/CoverageMappingGen.cpp
@@ -15,6 +15,7 @@
 #include "CodeGenFunction.h"
 #include "clang/AST/StmtVisitor.h"
 #include "clang/Lex/Lexer.h"
+#include "llvm/ADT/Optional.h"
 #include "llvm/ProfileData/CoverageMapping.h"
 #include "llvm/ProfileData/CoverageMappingReader.h"
 #include "llvm/ProfileData/CoverageMappingWriter.h"
@@ -33,87 +34,50 @@ namespace {
 
 /// \brief A region of source code that can be mapped to a counter.
 class SourceMappingRegion {
-public:
-  enum RegionFlags {
-    /// \brief This region won't be emitted if it wasn't extended.
-    /// This is useful so that we won't emit source ranges for single tokens
-    /// that we don't really care that much about, like:
-    ///   the '(' token in #define MACRO (
-    IgnoreIfNotExtended = 0x0001,
-  };
-
-private:
-  FileID File, MacroArgumentFile;
-
   Counter Count;
 
-  /// \brief A statement that initiated the count of Zero.
-  ///
-  /// This initiator statement is useful to prevent merging of unreachable
-  /// regions with different statements that caused the counter to become
-  /// unreachable.
-  const Stmt *UnreachableInitiator;
-
-  /// \brief A statement that separates certain mapping regions into groups.
-  ///
-  /// The group statement is sometimes useful when we are emitting the source
-  /// regions not in their correct lexical order, e.g. the regions for the
-  /// incrementation expression in the 'for' construct. By marking the regions
-  /// in the incrementation expression with the group statement, we avoid the
-  /// merging of the regions from the incrementation expression and the loop's
-  /// body.
-  const Stmt *Group;
-
   /// \brief The region's starting location.
-  SourceLocation LocStart;
+  Optional<SourceLocation> LocStart;
 
   /// \brief The region's ending location.
-  SourceLocation LocEnd, AlternativeLocEnd;
-  unsigned Flags;
+  Optional<SourceLocation> LocEnd;
 
 public:
-  SourceMappingRegion(FileID File, FileID MacroArgumentFile, Counter Count,
-                      const Stmt *UnreachableInitiator, const Stmt *Group,
-                      SourceLocation LocStart, SourceLocation LocEnd,
-                      unsigned Flags = 0)
-      : File(File), MacroArgumentFile(MacroArgumentFile), Count(Count),
-        UnreachableInitiator(UnreachableInitiator), Group(Group),
-        LocStart(LocStart), LocEnd(LocEnd), AlternativeLocEnd(LocStart),
-        Flags(Flags) {}
+  SourceMappingRegion(Counter Count, Optional<SourceLocation> LocStart,
+                      Optional<SourceLocation> LocEnd)
+      : Count(Count), LocStart(LocStart), LocEnd(LocEnd) {}
+
+  SourceMappingRegion(SourceMappingRegion &&Region)
+      : Count(std::move(Region.Count)), LocStart(std::move(Region.LocStart)),
+        LocEnd(std::move(Region.LocEnd)) {}
 
-  const FileID &getFile() const { return File; }
+  SourceMappingRegion &operator=(SourceMappingRegion &&RHS) {
+    Count = std::move(RHS.Count);
+    LocStart = std::move(RHS.LocStart);
+    LocEnd = std::move(RHS.LocEnd);
+    return *this;
+  }
 
   const Counter &getCounter() const { return Count; }
 
-  const SourceLocation &getStartLoc() const { return LocStart; }
+  void setCounter(Counter C) { Count = C; }
 
-  const SourceLocation &getEndLoc(const SourceManager &SM) const {
-    if (SM.getFileID(LocEnd) != File)
-      return AlternativeLocEnd;
-    return LocEnd;
-  }
+  bool hasStartLoc() const { return LocStart.hasValue(); }
 
-  bool hasFlag(RegionFlags Flag) const { return (Flags & Flag) != 0; }
+  void setStartLoc(SourceLocation Loc) { LocStart = Loc; }
 
-  void setFlag(RegionFlags Flag) { Flags |= Flag; }
+  const SourceLocation &getStartLoc() const {
+    assert(LocStart && "Region has no start location");
+    return *LocStart;
+  }
 
-  void clearFlag(RegionFlags Flag) { Flags &= ~Flag; }
+  bool hasEndLoc() const { return LocEnd.hasValue(); }
 
-  /// \brief Return true if two regions can be merged together.
-  bool isMergeable(SourceMappingRegion &R) {
-    // FIXME: We allow merging regions with a gap in between them. Should we?
-    return File == R.File && MacroArgumentFile == R.MacroArgumentFile &&
-           Count == R.Count && UnreachableInitiator == R.UnreachableInitiator &&
-           Group == R.Group;
-  }
+  void setEndLoc(SourceLocation Loc) { LocEnd = Loc; }
 
-  /// \brief A comparison that sorts such that mergeable regions are adjacent.
-  friend bool operator<(const SourceMappingRegion &LHS,
-                        const SourceMappingRegion &RHS) {
-    return std::tie(LHS.File, LHS.MacroArgumentFile, LHS.Count,
-                    LHS.UnreachableInitiator, LHS.Group) <
-           std::tie(RHS.File, RHS.MacroArgumentFile, RHS.Count,
-                    RHS.UnreachableInitiator, RHS.Group);
+  const SourceLocation &getEndLoc() const {
+    assert(LocEnd && "Region has no end location");
+    return *LocEnd;
   }
 };
 
@@ -126,26 +90,11 @@ public:
   const LangOptions &LangOpts;
 
 private:
-  struct FileInfo {
-    /// \brief The file id that will be used by the coverage mapping system.
-    unsigned CovMappingFileID;
-    const FileEntry *Entry;
-
-    FileInfo(unsigned CovMappingFileID, const FileEntry *Entry)
-        : CovMappingFileID(CovMappingFileID), Entry(Entry) {}
-  };
-
-  /// \brief This mapping maps clang's FileIDs to file ids used
-  /// by the coverage mapping system and clang's file entries.
-  llvm::SmallDenseMap<FileID, FileInfo, 8> FileIDMapping;
+  /// \brief Map of clang's FileIDs to IDs used for coverage mapping.
+  llvm::SmallDenseMap<FileID, std::pair<unsigned, SourceLocation>, 8>
+      FileIDMapping;
 
 public:
-  /// \brief The statement that corresponds to the current source group.
-  const Stmt *CurrentSourceGroup;
-
-  /// \brief The statement the initiated the current unreachable region.
-  const Stmt *CurrentUnreachableRegionInitiator;
-
   /// \brief The coverage mapping regions for this function
   llvm::SmallVector<CounterMappingRegion, 32> MappingRegions;
   /// \brief The source mapping regions for this function.
@@ -153,60 +102,104 @@ public:
 
   CoverageMappingBuilder(CoverageMappingModuleGen &CVM, SourceManager &SM,
                          const LangOptions &LangOpts)
-      : CVM(CVM), SM(SM), LangOpts(LangOpts),
-        CurrentSourceGroup(nullptr),
-        CurrentUnreachableRegionInitiator(nullptr) {}
+      : CVM(CVM), SM(SM), LangOpts(LangOpts) {}
 
   /// \brief Return the precise end location for the given token.
   SourceLocation getPreciseTokenLocEnd(SourceLocation Loc) {
-    return Lexer::getLocForEndOfToken(SM.getSpellingLoc(Loc), 0, SM, LangOpts);
+    // We avoid getLocForEndOfToken here, because it doesn't do what we want for
+    // macro locations, which we just treat as expanded files.
+    unsigned TokLen =
+        Lexer::MeasureTokenLength(SM.getSpellingLoc(Loc), SM, LangOpts);
+    return Loc.getLocWithOffset(TokLen);
   }
 
-  /// \brief Create the mapping that maps from the function's file ids to
-  /// the indices for the translation unit's filenames.
-  void createFileIDMapping(SmallVectorImpl<unsigned> &Mapping) {
-    Mapping.resize(FileIDMapping.size(), 0);
-    for (const auto &I : FileIDMapping)
-      Mapping[I.second.CovMappingFileID] = CVM.getFileID(I.second.Entry);
+  /// \brief Return the start location of an included file or expanded macro.
+  SourceLocation getStartOfFileOrMacro(SourceLocation Loc) {
+    if (Loc.isMacroID())
+      return Loc.getLocWithOffset(-SM.getFileOffset(Loc));
+    return SM.getLocForStartOfFile(SM.getFileID(Loc));
   }
 
-  /// \brief Get the coverage mapping file id that corresponds to the given
-  /// clang file id. If such file id doesn't exist, it gets added to the
-  /// mapping that maps from clang's file ids to coverage mapping file ids.
-  /// Return true if there was an error getting the coverage mapping file id.
-  /// An example of an when this function fails is when the region tries
-  /// to get a coverage file id for a location in a built-in macro.
-  bool getCoverageFileID(SourceLocation LocStart, FileID File,
-                         FileID SpellingFile, unsigned &Result) {
-    auto Mapping = FileIDMapping.find(File);
-    if (Mapping != FileIDMapping.end()) {
-      Result = Mapping->second.CovMappingFileID;
-      return false;
-    }
+  /// \brief Return the end location of an included file or expanded macro.
+  SourceLocation getEndOfFileOrMacro(SourceLocation Loc) {
+    if (Loc.isMacroID())
+      return Loc.getLocWithOffset(SM.getFileIDSize(SM.getFileID(Loc)) -
+                                  SM.getFileOffset(Loc));
+    return SM.getLocForEndOfFile(SM.getFileID(Loc));
+  }
+
+  /// \brief Find out where the current file is included or macro is expanded.
+  SourceLocation getIncludeOrExpansionLoc(SourceLocation Loc) {
+    return Loc.isMacroID() ? SM.getImmediateExpansionRange(Loc).first
+                           : SM.getIncludeLoc(SM.getFileID(Loc));
+  }
+
+  /// \brief Return true if \c Loc is a location in a built-in macro.
+  bool isInBuiltin(SourceLocation Loc) {
+    return strcmp(SM.getBufferName(SM.getSpellingLoc(Loc)), "<built-in>") == 0;
+  }
 
-    auto Entry = SM.getFileEntryForID(SpellingFile);
-    if (!Entry)
-      return true;
+  /// \brief Get the start of \c S ignoring macro arguments and builtin macros.
+  SourceLocation getStart(const Stmt *S) {
+    SourceLocation Loc = S->getLocStart();
+    while (SM.isMacroArgExpansion(Loc) || isInBuiltin(Loc))
+      Loc = SM.getImmediateExpansionRange(Loc).first;
+    return Loc;
+  }
 
-    Result = FileIDMapping.size();
-    FileIDMapping.insert(std::make_pair(File, FileInfo(Result, Entry)));
-    createFileExpansionRegion(LocStart, File);
-    return false;
+  /// \brief Get the end of \c S ignoring macro arguments and builtin macros.
+  SourceLocation getEnd(const Stmt *S) {
+    SourceLocation Loc = S->getLocEnd();
+    while (SM.isMacroArgExpansion(Loc) || isInBuiltin(Loc))
+      Loc = SM.getImmediateExpansionRange(Loc).first;
+    return getPreciseTokenLocEnd(Loc);
   }
 
-  /// \brief Get the coverage mapping file id that corresponds to the given
-  /// clang file id.
-  /// Return true if there was an error getting the coverage mapping file id.
-  bool getExistingCoverageFileID(FileID File, unsigned &Result) {
-    // Make sure that the file is valid.
-    if (File.isInvalid())
-      return true;
-    auto Mapping = FileIDMapping.find(File);
-    if (Mapping != FileIDMapping.end()) {
-      Result = Mapping->second.CovMappingFileID;
-      return false;
+  /// \brief Find the set of files we have regions for and assign IDs
+  ///
+  /// Fills \c Mapping with the virtual file mapping needed to write out
+  /// coverage and collects the necessary file information to emit source and
+  /// expansion regions.
+  void gatherFileIDs(SmallVectorImpl<unsigned> &Mapping) {
+    FileIDMapping.clear();
+
+    SmallVector<FileID, 8> Visited;
+    SmallVector<std::pair<SourceLocation, unsigned>, 8> FileLocs;
+    for (const auto &Region : SourceRegions) {
+      SourceLocation Loc = Region.getStartLoc();
+      FileID File = SM.getFileID(Loc);
+      if (std::find(Visited.begin(), Visited.end(), File) != Visited.end())
+        continue;
+      Visited.push_back(File);
+
+      unsigned Depth = 0;
+      for (SourceLocation Parent = getIncludeOrExpansionLoc(Loc);
+           !Parent.isInvalid(); Parent = getIncludeOrExpansionLoc(Parent))
+        ++Depth;
+      FileLocs.push_back(std::make_pair(Loc, Depth));
     }
-    return true;
+    std::stable_sort(FileLocs.begin(), FileLocs.end(), llvm::less_second());
+
+    for (const auto &FL : FileLocs) {
+      SourceLocation Loc = FL.first;
+      FileID SpellingFile = SM.getDecomposedSpellingLoc(Loc).first;
+      auto Entry = SM.getFileEntryForID(SpellingFile);
+      if (!Entry)
+        continue;
+
+      FileIDMapping[SM.getFileID(Loc)] = std::make_pair(Mapping.size(), Loc);
+      Mapping.push_back(CVM.getFileID(Entry));
+    }
+  }
+
+  /// \brief Get the coverage mapping file ID for \c Loc.
+  ///
+  /// If such file id doesn't exist, return None.
+  Optional<unsigned> getCoverageFileID(SourceLocation Loc) {
+    auto Mapping = FileIDMapping.find(SM.getFileID(Loc));
+    if (Mapping != FileIDMapping.end())
+      return Mapping->second.first;
+    return None;
   }
 
   /// \brief Return true if the given clang's file id has a corresponding
@@ -235,154 +228,83 @@ public:
     for (const auto &I : SkippedRanges) {
       auto LocStart = I.getBegin();
       auto LocEnd = I.getEnd();
-      auto FileStart = SM.getFileID(LocStart);
-      if (!hasExistingCoverageFileID(FileStart))
-        continue;
-      auto ActualFileStart = SM.getDecomposedSpellingLoc(LocStart).first;
-      if (ActualFileStart != SM.getDecomposedSpellingLoc(LocEnd).first)
-        // Ignore regions that span across multiple files.
-        continue;
+      assert(SM.isWrittenInSameFile(LocStart, LocEnd) &&
+             "region spans multiple files");
 
-      unsigned CovFileID;
-      if (getCoverageFileID(LocStart, FileStart, ActualFileStart, CovFileID))
+      auto CovFileID = getCoverageFileID(LocStart);
+      if (!CovFileID)
         continue;
       unsigned LineStart = SM.getSpellingLineNumber(LocStart);
       unsigned ColumnStart = SM.getSpellingColumnNumber(LocStart);
       unsigned LineEnd = SM.getSpellingLineNumber(LocEnd);
       unsigned ColumnEnd = SM.getSpellingColumnNumber(LocEnd);
-      CounterMappingRegion Region(Counter(), CovFileID, LineStart, ColumnStart,
-                                  LineEnd, ColumnEnd, false,
-                                  CounterMappingRegion::SkippedRegion);
+      auto Region = CounterMappingRegion::makeSkipped(
+          *CovFileID, LineStart, ColumnStart, LineEnd, ColumnEnd);
       // Make sure that we only collect the regions that are inside
       // the souce code of this function.
-      if (Region.LineStart >= FileLineRanges[CovFileID].first &&
-          Region.LineEnd <= FileLineRanges[CovFileID].second)
+      if (Region.LineStart >= FileLineRanges[*CovFileID].first &&
+          Region.LineEnd <= FileLineRanges[*CovFileID].second)
         MappingRegions.push_back(Region);
     }
   }
 
-  /// \brief Create a mapping region that correponds to an expansion of
-  /// a macro or an embedded include.
-  void createFileExpansionRegion(SourceLocation Loc, FileID ExpandedFile) {
-    SourceLocation LocStart;
-    if (Loc.isMacroID())
-      LocStart = SM.getImmediateExpansionRange(Loc).first;
-    else {
-      LocStart = SM.getIncludeLoc(ExpandedFile);
-      if (LocStart.isInvalid())
-        return; // This file has no expansion region.
-    }
-
-    auto File = SM.getFileID(LocStart);
-    auto SpellingFile = SM.getDecomposedSpellingLoc(LocStart).first;
-    unsigned CovFileID, ExpandedFileID;
-    if (getExistingCoverageFileID(ExpandedFile, ExpandedFileID))
-      return;
-    if (getCoverageFileID(LocStart, File, SpellingFile, CovFileID))
-      return;
-    unsigned LineStart = SM.getSpellingLineNumber(LocStart);
-    unsigned ColumnStart = SM.getSpellingColumnNumber(LocStart);
-    unsigned LineEnd = LineStart;
-    // Compute the end column manually as Lexer::getLocForEndOfToken doesn't
-    // give the correct result in all cases.
-    unsigned ColumnEnd =
-        ColumnStart +
-        Lexer::MeasureTokenLength(SM.getSpellingLoc(LocStart), SM, LangOpts);
-
-    MappingRegions.push_back(CounterMappingRegion(
-        Counter(), CovFileID, LineStart, ColumnStart, LineEnd, ColumnEnd,
-        false, CounterMappingRegion::ExpansionRegion));
-    MappingRegions.back().ExpandedFileID = ExpandedFileID;
-  }
-
-  /// \brief Enter a source region group that is identified by the given
-  /// statement.
-  /// It's not possible to enter a group when there is already
-  /// another group present.
-  void beginSourceRegionGroup(const Stmt *Group) {
-    assert(!CurrentSourceGroup);
-    CurrentSourceGroup = Group;
-  }
-
-  /// \brief Exit the current source region group.
-  void endSourceRegionGroup() { CurrentSourceGroup = nullptr; }
-
-  /// \brief Associate a counter with a given source code range.
-  void mapSourceCodeRange(SourceLocation LocStart, SourceLocation LocEnd,
-                          Counter Count, const Stmt *UnreachableInitiator,
-                          const Stmt *SourceGroup, unsigned Flags = 0,
-                          FileID MacroArgumentFile = FileID()) {
-    if (SM.isMacroArgExpansion(LocStart)) {
-      // Map the code range with the macro argument's value.
-      mapSourceCodeRange(SM.getImmediateSpellingLoc(LocStart),
-                         SM.getImmediateSpellingLoc(LocEnd), Count,
-                         UnreachableInitiator, SourceGroup, Flags,
-                         SM.getFileID(LocStart));
-      // Map the code range where the macro argument is referenced.
-      SourceLocation RefLocStart(SM.getImmediateExpansionRange(LocStart).first);
-      SourceLocation RefLocEnd(RefLocStart);
-      if (SM.isMacroArgExpansion(RefLocStart))
-        mapSourceCodeRange(RefLocStart, RefLocEnd, Count, UnreachableInitiator,
-                           SourceGroup, 0, SM.getFileID(RefLocStart));
-      else
-        mapSourceCodeRange(RefLocStart, RefLocEnd, Count, UnreachableInitiator,
-                           SourceGroup);
-      return;
-    }
-    auto File = SM.getFileID(LocStart);
-    // Make sure that the file id is valid.
-    if (File.isInvalid())
-      return;
-    SourceRegions.emplace_back(File, MacroArgumentFile, Count,
-                               UnreachableInitiator, SourceGroup, LocStart,
-                               LocEnd, Flags);
-  }
-
-  void mapSourceCodeRange(SourceLocation LocStart, SourceLocation LocEnd,
-                          Counter Count, unsigned Flags = 0) {
-    mapSourceCodeRange(LocStart, LocEnd, Count,
-                       CurrentUnreachableRegionInitiator, CurrentSourceGroup,
-                       Flags);
-  }
-
   /// \brief Generate the coverage counter mapping regions from collected
   /// source regions.
   void emitSourceRegions() {
-    std::sort(SourceRegions.begin(), SourceRegions.end());
-
-    for (auto I = SourceRegions.begin(), E = SourceRegions.end(); I != E; ++I) {
-      // Keep the original start location of this region.
-      SourceLocation LocStart = I->getStartLoc();
-      SourceLocation LocEnd = I->getEndLoc(SM);
-
-      bool Ignore = I->hasFlag(SourceMappingRegion::IgnoreIfNotExtended);
-      // We need to handle mergeable regions together.
-      for (auto Next = I + 1; Next != E && Next->isMergeable(*I); ++Next) {
-        ++I;
-        LocStart = std::min(LocStart, I->getStartLoc());
-        LocEnd = std::max(LocEnd, I->getEndLoc(SM));
-        // FIXME: Should we && together the Ignore flag of multiple regions?
-        Ignore = false;
-      }
-      if (Ignore)
+    for (const auto &Region : SourceRegions) {
+      assert(Region.hasEndLoc() && "incomplete region");
+
+      SourceLocation LocStart = Region.getStartLoc();
+      assert(!SM.getFileID(LocStart).isInvalid() && "region in invalid file");
+
+      auto CovFileID = getCoverageFileID(LocStart);
+      // Ignore regions that don't have a file, such as builtin macros.
+      if (!CovFileID)
         continue;
 
+      SourceLocation LocEnd = Region.getEndLoc();
+      assert(SM.isWrittenInSameFile(LocStart, LocEnd) &&
+             "region spans multiple files");
+
       // Find the spilling locations for the mapping region.
-      LocEnd = getPreciseTokenLocEnd(LocEnd);
       unsigned LineStart = SM.getSpellingLineNumber(LocStart);
       unsigned ColumnStart = SM.getSpellingColumnNumber(LocStart);
       unsigned LineEnd = SM.getSpellingLineNumber(LocEnd);
       unsigned ColumnEnd = SM.getSpellingColumnNumber(LocEnd);
 
-      auto SpellingFile = SM.getDecomposedSpellingLoc(LocStart).first;
-      unsigned CovFileID;
-      if (getCoverageFileID(LocStart, I->getFile(), SpellingFile, CovFileID))
+      assert(LineStart <= LineEnd && "region start and end out of order");
+      MappingRegions.push_back(CounterMappingRegion::makeRegion(
+          Region.getCounter(), *CovFileID, LineStart, ColumnStart, LineEnd,
+          ColumnEnd));
+    }
+  }
+
+  /// \brief Generate expansion regions for each virtual file we've seen.
+  void emitExpansionRegions() {
+    for (const auto &FM : FileIDMapping) {
+      SourceLocation ExpandedLoc = FM.second.second;
+      SourceLocation ParentLoc = getIncludeOrExpansionLoc(ExpandedLoc);
+      if (ParentLoc.isInvalid())
+        continue;
+
+      auto ParentFileID = getCoverageFileID(ParentLoc);
+      if (!ParentFileID)
         continue;
+      auto ExpandedFileID = getCoverageFileID(ExpandedLoc);
+      assert(ExpandedFileID && "expansion in uncovered file");
+
+      SourceLocation LocEnd = getPreciseTokenLocEnd(ParentLoc);
+      assert(SM.isWrittenInSameFile(ParentLoc, LocEnd) &&
+             "region spans multiple files");
 
-      assert(LineStart <= LineEnd);
-      MappingRegions.push_back(CounterMappingRegion(
-          I->getCounter(), CovFileID, LineStart, ColumnStart, LineEnd,
-          ColumnEnd, false, CounterMappingRegion::CodeRegion));
+      unsigned LineStart = SM.getSpellingLineNumber(ParentLoc);
+      unsigned ColumnStart = SM.getSpellingColumnNumber(ParentLoc);
+      unsigned LineEnd = SM.getSpellingLineNumber(LocEnd);
+      unsigned ColumnEnd = SM.getSpellingColumnNumber(LocEnd);
+
+      MappingRegions.push_back(CounterMappingRegion::makeExpansion(
+          *ParentFileID, *ExpandedFileID, LineStart, ColumnStart, LineEnd,
+          ColumnEnd));
     }
   }
 };
@@ -398,14 +320,14 @@ struct EmptyCoverageMappingBuilder : public CoverageMappingBuilder {
     if (!D->hasBody())
       return;
     auto Body = D->getBody();
-    mapSourceCodeRange(Body->getLocStart(), Body->getLocEnd(), Counter());
+    SourceRegions.emplace_back(Counter(), getStart(Body), getEnd(Body));
   }
 
   /// \brief Write the mapping data to the output stream
   void write(llvm::raw_ostream &OS) {
-    emitSourceRegions();
     SmallVector<unsigned, 16> FileIDMapping;
-    createFileIDMapping(FileIDMapping);
+    gatherFileIDs(FileIDMapping);
+    emitSourceRegions();
 
     CoverageMappingWriter Writer(FileIDMapping, None, MappingRegions);
     Writer.write(OS);
@@ -420,136 +342,215 @@ struct CounterCoverageMappingBuilder
   /// \brief The map of statements to count values.
   llvm::DenseMap<const Stmt *, unsigned> &CounterMap;
 
-  Counter CurrentRegionCount;
+  /// \brief A stack of currently live regions.
+  std::vector<SourceMappingRegion> RegionStack;
 
   CounterExpressionBuilder Builder;
 
-  /// \brief Return a counter that represents the
-  /// expression that subracts rhs from lhs.
+  /// \brief A location in the most recently visited file or macro.
+  ///
+  /// This is used to adjust the active source regions appropriately when
+  /// expressions cross file or macro boundaries.
+  SourceLocation MostRecentLocation;
+
+  /// \brief Return a counter for the subtraction of \c RHS from \c LHS
   Counter subtractCounters(Counter LHS, Counter RHS) {
     return Builder.subtract(LHS, RHS);
   }
 
-  /// \brief Return a counter that represents the
-  /// the exression that adds lhs and rhs.
+  /// \brief Return a counter for the sum of \c LHS and \c RHS.
   Counter addCounters(Counter LHS, Counter RHS) {
     return Builder.add(LHS, RHS);
   }
 
+  Counter addCounters(Counter C1, Counter C2, Counter C3) {
+    return addCounters(addCounters(C1, C2), C3);
+  }
+
+  Counter addCounters(Counter C1, Counter C2, Counter C3, Counter C4) {
+    return addCounters(addCounters(C1, C2, C3), C4);
+  }
+
   /// \brief Return the region counter for the given statement.
+  ///
   /// This should only be called on statements that have a dedicated counter.
-  unsigned getRegionCounter(const Stmt *S) { return CounterMap[S]; }
+  Counter getRegionCounter(const Stmt *S) {
+    return Counter::getCounter(CounterMap[S]);
+  }
+
+  /// \brief Push a region onto the stack.
+  ///
+  /// Returns the index on the stack where the region was pushed. This can be
+  /// used with popRegions to exit a "scope", ending the region that was pushed.
+  size_t pushRegion(Counter Count, Optional<SourceLocation> StartLoc = None,
+                    Optional<SourceLocation> EndLoc = None) {
+    if (StartLoc)
+      MostRecentLocation = *StartLoc;
+    RegionStack.emplace_back(Count, StartLoc, EndLoc);
 
-  /// \brief Return the region count for the counter at the given index.
-  Counter getRegionCount(unsigned CounterId) {
-    return Counter::getCounter(CounterId);
+    return RegionStack.size() - 1;
   }
 
-  /// \brief Return the counter value of the current region.
-  Counter getCurrentRegionCount() { return CurrentRegionCount; }
+  /// \brief Pop regions from the stack into the function's list of regions.
+  ///
+  /// Adds all regions from \c ParentIndex to the top of the stack to the
+  /// function's \c SourceRegions.
+  void popRegions(size_t ParentIndex) {
+    assert(RegionStack.size() >= ParentIndex && "parent not in stack");
+    while (RegionStack.size() > ParentIndex) {
+      SourceMappingRegion &Region = RegionStack.back();
+      if (Region.hasStartLoc()) {
+        SourceLocation StartLoc = Region.getStartLoc();
+        SourceLocation EndLoc = Region.hasEndLoc()
+                                    ? Region.getEndLoc()
+                                    : RegionStack[ParentIndex].getEndLoc();
+        while (!SM.isWrittenInSameFile(StartLoc, EndLoc)) {
+          // The region ends in a nested file or macro expansion. Create a
+          // separate region for each expansion.
+          SourceLocation NestedLoc = getStartOfFileOrMacro(EndLoc);
+          assert(SM.isWrittenInSameFile(NestedLoc, EndLoc));
+
+          SourceRegions.emplace_back(Region.getCounter(), NestedLoc, EndLoc);
+
+          EndLoc = getPreciseTokenLocEnd(getIncludeOrExpansionLoc(EndLoc));
+          assert(!EndLoc.isInvalid() &&
+                 "File exit was not handled before popRegions");
+        }
+        Region.setEndLoc(EndLoc);
+
+        MostRecentLocation = EndLoc;
+        // If this region happens to span an entire expansion, we need to make
+        // sure we don't overlap the parent region with it.
+        if (StartLoc == getStartOfFileOrMacro(StartLoc) &&
+            EndLoc == getEndOfFileOrMacro(EndLoc))
+          MostRecentLocation = getIncludeOrExpansionLoc(EndLoc);
+
+        assert(SM.isWrittenInSameFile(Region.getStartLoc(), EndLoc));
+        SourceRegions.push_back(std::move(Region));
+      }
+      RegionStack.pop_back();
+    }
+  }
 
-  /// \brief Set the counter value for the current region.
-  /// This is used to keep track of changes to the most recent counter
-  /// from control flow and non-local exits.
-  void setCurrentRegionCount(Counter Count) {
-    CurrentRegionCount = Count;
-    CurrentUnreachableRegionInitiator = nullptr;
+  /// \brief Return the currently active region.
+  SourceMappingRegion &getRegion() {
+    assert(!RegionStack.empty() && "statement has no region");
+    return RegionStack.back();
   }
 
-  /// \brief Indicate that the current region is never reached,
-  /// and thus should have a counter value of zero.
-  /// This is important so that subsequent regions can correctly track
-  /// their parent counts.
-  void setCurrentRegionUnreachable(const Stmt *Initiator) {
-    CurrentRegionCount = Counter::getZero();
-    CurrentUnreachableRegionInitiator = Initiator;
+  /// \brief Propagate counts through the children of \c S.
+  Counter propagateCounts(Counter TopCount, const Stmt *S) {
+    size_t Index = pushRegion(TopCount, getStart(S), getEnd(S));
+    Visit(S);
+    Counter ExitCount = getRegion().getCounter();
+    popRegions(Index);
+    return ExitCount;
   }
 
-  /// \brief A counter for a particular region.
-  /// This is the primary interface through
-  /// which the coverage mapping builder manages counters and their values.
-  class RegionMapper {
-    CounterCoverageMappingBuilder &Mapping;
-    Counter Count;
-    Counter ParentCount;
-    Counter RegionCount;
-    Counter Adjust;
-
-  public:
-    RegionMapper(CounterCoverageMappingBuilder *Mapper, const Stmt *S)
-        : Mapping(*Mapper),
-          Count(Mapper->getRegionCount(Mapper->getRegionCounter(S))),
-          ParentCount(Mapper->getCurrentRegionCount()) {}
-
-    /// Get the value of the counter. In most cases this is the number of times
-    /// the region of the counter was entered, but for switch labels it's the
-    /// number of direct jumps to that label.
-    Counter getCount() const { return Count; }
-
-    /// Get the value of the counter with adjustments applied. Adjustments occur
-    /// when control enters or leaves the region abnormally; i.e., if there is a
-    /// jump to a label within the region, or if the function can return from
-    /// within the region. The adjusted count, then, is the value of the counter
-    /// at the end of the region.
-    Counter getAdjustedCount() const {
-      return Mapping.addCounters(Count, Adjust);
-    }
+  /// \brief Adjust the most recently visited location to \c EndLoc.
+  ///
+  /// This should be used after visiting any statements in non-source order.
+  void adjustForOutOfOrderTraversal(SourceLocation EndLoc) {
+    MostRecentLocation = EndLoc;
+    // Avoid adding duplicate regions if we have a completed region on the top
+    // of the stack and are adjusting to the end of a virtual file.
+    if (getRegion().hasEndLoc() &&
+        MostRecentLocation == getEndOfFileOrMacro(MostRecentLocation))
+      MostRecentLocation = getIncludeOrExpansionLoc(MostRecentLocation);
+  }
+
+  /// \brief Check whether \c Loc is included or expanded from \c Parent.
+  bool isNestedIn(SourceLocation Loc, FileID Parent) {
+    do {
+      Loc = getIncludeOrExpansionLoc(Loc);
+      if (Loc.isInvalid())
+        return false;
+    } while (!SM.isInFileID(Loc, Parent));
+    return true;
+  }
 
-    /// Get the value of the counter in this region's parent, i.e., the region
-    /// that was active when this region began. This is useful for deriving
-    /// counts in implicitly counted regions, like the false case of a condition
-    /// or the normal exits of a loop.
-    Counter getParentCount() const { return ParentCount; }
-
-    /// Activate the counter by emitting an increment and starting to track
-    /// adjustments. If AddIncomingFallThrough is true, the current region count
-    /// will be added to the counter for the purposes of tracking the region.
-    void beginRegion(bool AddIncomingFallThrough = false) {
-      RegionCount = Count;
-      if (AddIncomingFallThrough)
-        RegionCount =
-            Mapping.addCounters(RegionCount, Mapping.getCurrentRegionCount());
-      Mapping.setCurrentRegionCount(RegionCount);
-    }
+  /// \brief Adjust regions and state when \c NewLoc exits a file.
+  ///
+  /// If moving from our most recently tracked location to \c NewLoc exits any
+  /// files, this adjusts our current region stack and creates the file regions
+  /// for the exited file.
+  void handleFileExit(SourceLocation NewLoc) {
+    if (SM.isWrittenInSameFile(MostRecentLocation, NewLoc))
+      return;
 
-    /// For counters on boolean branches, begins tracking adjustments for the
-    /// uncounted path.
-    void beginElseRegion() {
-      RegionCount = Mapping.subtractCounters(ParentCount, Count);
-      Mapping.setCurrentRegionCount(RegionCount);
+    // If NewLoc is not in a file that contains MostRecentLocation, walk up to
+    // find the common ancestor.
+    SourceLocation LCA = NewLoc;
+    FileID ParentFile = SM.getFileID(LCA);
+    while (!isNestedIn(MostRecentLocation, ParentFile)) {
+      LCA = getIncludeOrExpansionLoc(LCA);
+      if (LCA.isInvalid() || SM.isWrittenInSameFile(LCA, MostRecentLocation)) {
+        // Since there isn't a common ancestor, no file was exited. We just need
+        // to adjust our location to the new file.
+        MostRecentLocation = NewLoc;
+        return;
+      }
+      ParentFile = SM.getFileID(LCA);
     }
 
-    /// Reset the current region count.
-    void setCurrentRegionCount(Counter CurrentCount) {
-      RegionCount = CurrentCount;
-      Mapping.setCurrentRegionCount(RegionCount);
-    }
+    llvm::SmallSet<SourceLocation, 8> StartLocs;
+    Optional<Counter> ParentCounter;
+    for (auto I = RegionStack.rbegin(), E = RegionStack.rend(); I != E; ++I) {
+      if (!I->hasStartLoc())
+        continue;
+      SourceLocation Loc = I->getStartLoc();
+      if (!isNestedIn(Loc, ParentFile)) {
+        ParentCounter = I->getCounter();
+        break;
+      }
 
-    /// Adjust for non-local control flow after emitting a subexpression or
-    /// substatement. This must be called to account for constructs such as
-    /// gotos,
-    /// labels, and returns, so that we can ensure that our region's count is
-    /// correct in the code that follows.
-    void adjustForControlFlow() {
-      Adjust = Mapping.addCounters(
-          Adjust, Mapping.subtractCounters(Mapping.getCurrentRegionCount(),
-                                           RegionCount));
-      // Reset the region count in case this is called again later.
-      RegionCount = Mapping.getCurrentRegionCount();
+      while (!SM.isInFileID(Loc, ParentFile)) {
+        // The most nested region for each start location is the one with the
+        // correct count. We avoid creating redundant regions by stopping once
+        // we've seen this region.
+        if (StartLocs.insert(Loc).second)
+          SourceRegions.emplace_back(I->getCounter(), Loc,
+                                     getEndOfFileOrMacro(Loc));
+        Loc = getIncludeOrExpansionLoc(Loc);
+      }
+      I->setStartLoc(getPreciseTokenLocEnd(Loc));
     }
 
-    /// Commit all adjustments to the current region. If the region is a loop,
-    /// the LoopAdjust value should be the count of all the breaks and continues
-    /// from the loop, to compensate for those counts being deducted from the
-    /// adjustments for the body of the loop.
-    void applyAdjustmentsToRegion() {
-      Mapping.setCurrentRegionCount(Mapping.addCounters(ParentCount, Adjust));
-    }
-    void applyAdjustmentsToRegion(Counter LoopAdjust) {
-      Mapping.setCurrentRegionCount(Mapping.addCounters(
-          Mapping.addCounters(ParentCount, Adjust), LoopAdjust));
+    if (ParentCounter) {
+      // If the file is contained completely by another region and doesn't
+      // immediately start its own region, the whole file gets a region
+      // corresponding to the parent.
+      SourceLocation Loc = MostRecentLocation;
+      while (isNestedIn(Loc, ParentFile)) {
+        SourceLocation FileStart = getStartOfFileOrMacro(Loc);
+        if (StartLocs.insert(FileStart).second)
+          SourceRegions.emplace_back(*ParentCounter, FileStart,
+                                     getEndOfFileOrMacro(Loc));
+        Loc = getIncludeOrExpansionLoc(Loc);
+      }
     }
-  };
+
+    MostRecentLocation = NewLoc;
+  }
+
+  /// \brief Ensure that \c S is included in the current region.
+  void extendRegion(const Stmt *S) {
+    SourceMappingRegion &Region = getRegion();
+    SourceLocation StartLoc = getStart(S);
+
+    handleFileExit(StartLoc);
+    if (!Region.hasStartLoc())
+      Region.setStartLoc(StartLoc);
+  }
+
+  /// \brief Mark \c S as a terminator, starting a zero region.
+  void terminateRegion(const Stmt *S) {
+    extendRegion(S);
+    SourceMappingRegion &Region = getRegion();
+    if (!Region.hasEndLoc())
+      Region.setEndLoc(getEnd(S));
+    pushRegion(Counter::getZero());
+  }
 
   /// \brief Keep counts of breaks and continues inside loops.
   struct BreakContinue {
@@ -566,437 +567,318 @@ struct CounterCoverageMappingBuilder
 
   /// \brief Write the mapping data to the output stream
   void write(llvm::raw_ostream &OS) {
-    emitSourceRegions();
     llvm::SmallVector<unsigned, 8> VirtualFileMapping;
-    createFileIDMapping(VirtualFileMapping);
+    gatherFileIDs(VirtualFileMapping);
+    emitSourceRegions();
+    emitExpansionRegions();
     gatherSkippedRegions();
 
-    CoverageMappingWriter Writer(
-        VirtualFileMapping, Builder.getExpressions(), MappingRegions);
+    CoverageMappingWriter Writer(VirtualFileMapping, Builder.getExpressions(),
+                                 MappingRegions);
     Writer.write(OS);
   }
 
-  /// \brief Associate the source code range with the current region count.
-  void mapSourceCodeRange(SourceLocation LocStart, SourceLocation LocEnd,
-                          unsigned Flags = 0) {
-    CoverageMappingBuilder::mapSourceCodeRange(LocStart, LocEnd,
-                                               CurrentRegionCount, Flags);
-  }
-
-  void mapSourceCodeRange(SourceLocation LocStart) {
-    CoverageMappingBuilder::mapSourceCodeRange(LocStart, LocStart,
-                                               CurrentRegionCount);
-  }
-
-  /// \brief Associate the source range of a token with the current region
-  /// count.
-  /// Ignore the source range for this token if it produces a distinct
-  /// mapping region with no other source ranges.
-  void mapToken(SourceLocation LocStart) {
-    CoverageMappingBuilder::mapSourceCodeRange(
-        LocStart, LocStart, CurrentRegionCount,
-        SourceMappingRegion::IgnoreIfNotExtended);
-  }
-
   void VisitStmt(const Stmt *S) {
-    mapSourceCodeRange(S->getLocStart());
+    if (!S->getLocStart().isInvalid())
+      extendRegion(S);
     for (Stmt::const_child_range I = S->children(); I; ++I) {
       if (*I)
         this->Visit(*I);
     }
+    handleFileExit(getEnd(S));
   }
 
   void VisitDecl(const Decl *D) {
-    if (!D->hasBody())
-      return;
-    // Counter tracks entry to the function body.
-    auto Body = D->getBody();
-    RegionMapper Cnt(this, Body);
-    Cnt.beginRegion();
-    Visit(Body);
-  }
-
-  void VisitDeclStmt(const DeclStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
-    for (Stmt::const_child_range I = static_cast<const Stmt *>(S)->children();
-         I; ++I) {
-      if (*I)
-        this->Visit(*I);
-    }
-  }
-
-  void VisitCompoundStmt(const CompoundStmt *S) {
-    mapSourceCodeRange(S->getLBracLoc());
-    mapSourceCodeRange(S->getRBracLoc());
-    for (Stmt::const_child_range I = S->children(); I; ++I) {
-      if (*I)
-        this->Visit(*I);
-    }
+    Stmt *Body = D->getBody();
+    propagateCounts(getRegionCounter(Body), Body);
   }
 
   void VisitReturnStmt(const ReturnStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
+    extendRegion(S);
     if (S->getRetValue())
       Visit(S->getRetValue());
-    setCurrentRegionUnreachable(S);
+    terminateRegion(S);
   }
 
-  void VisitGotoStmt(const GotoStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
-    mapToken(S->getLabelLoc());
-    setCurrentRegionUnreachable(S);
+  void VisitCXXThrowExpr(const CXXThrowExpr *E) {
+    extendRegion(E);
+    if (E->getSubExpr())
+      Visit(E->getSubExpr());
+    terminateRegion(E);
   }
 
+  void VisitGotoStmt(const GotoStmt *S) { terminateRegion(S); }
+
   void VisitLabelStmt(const LabelStmt *S) {
-    // Counter tracks the block following the label.
-    RegionMapper Cnt(this, S);
-    Cnt.beginRegion();
-    mapSourceCodeRange(S->getLocStart());
-    // Can't map the ':' token as its location isn't known.
+    SourceLocation Start = getStart(S);
+    // We can't extendRegion here or we risk overlapping with our new region.
+    handleFileExit(Start);
+    pushRegion(getRegionCounter(S), Start);
     Visit(S->getSubStmt());
   }
 
   void VisitBreakStmt(const BreakStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
     assert(!BreakContinueStack.empty() && "break not in a loop or switch!");
     BreakContinueStack.back().BreakCount = addCounters(
-        BreakContinueStack.back().BreakCount, getCurrentRegionCount());
-    setCurrentRegionUnreachable(S);
+        BreakContinueStack.back().BreakCount, getRegion().getCounter());
+    terminateRegion(S);
   }
 
   void VisitContinueStmt(const ContinueStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
     assert(!BreakContinueStack.empty() && "continue stmt not in a loop!");
     BreakContinueStack.back().ContinueCount = addCounters(
-        BreakContinueStack.back().ContinueCount, getCurrentRegionCount());
-    setCurrentRegionUnreachable(S);
+        BreakContinueStack.back().ContinueCount, getRegion().getCounter());
+    terminateRegion(S);
   }
 
   void VisitWhileStmt(const WhileStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
-    // Counter tracks the body of the loop.
-    RegionMapper Cnt(this, S);
+    extendRegion(S);
+
+    Counter ParentCount = getRegion().getCounter();
+    Counter BodyCount = getRegionCounter(S);
+
+    // Handle the body first so that we can get the backedge count.
     BreakContinueStack.push_back(BreakContinue());
-    // Visit the body region first so the break/continue adjustments can be
-    // included when visiting the condition.
-    Cnt.beginRegion();
-    Visit(S->getBody());
-    Cnt.adjustForControlFlow();
-
-    // ...then go back and propagate counts through the condition. The count
-    // at the start of the condition is the sum of the incoming edges,
-    // the backedge from the end of the loop body, and the edges from
-    // continue statements.
+    extendRegion(S->getBody());
+    Counter BackedgeCount = propagateCounts(BodyCount, S->getBody());
     BreakContinue BC = BreakContinueStack.pop_back_val();
-    Cnt.setCurrentRegionCount(
-        addCounters(Cnt.getParentCount(),
-                    addCounters(Cnt.getAdjustedCount(), BC.ContinueCount)));
-    beginSourceRegionGroup(S->getCond());
-    Visit(S->getCond());
-    endSourceRegionGroup();
-    Cnt.adjustForControlFlow();
-    Cnt.applyAdjustmentsToRegion(addCounters(BC.BreakCount, BC.ContinueCount));
+
+    // Go back to handle the condition.
+    Counter CondCount =
+        addCounters(ParentCount, BackedgeCount, BC.ContinueCount);
+    propagateCounts(CondCount, S->getCond());
+    adjustForOutOfOrderTraversal(getEnd(S));
+
+    Counter OutCount =
+        addCounters(BC.BreakCount, subtractCounters(CondCount, BodyCount));
+    if (OutCount != ParentCount)
+      pushRegion(OutCount);
   }
 
   void VisitDoStmt(const DoStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
-    // Counter tracks the body of the loop.
-    RegionMapper Cnt(this, S);
-    BreakContinueStack.push_back(BreakContinue());
-    Cnt.beginRegion(/*AddIncomingFallThrough=*/true);
-    Visit(S->getBody());
-    Cnt.adjustForControlFlow();
+    extendRegion(S);
+
+    Counter ParentCount = getRegion().getCounter();
+    Counter BodyCount = getRegionCounter(S);
 
+    BreakContinueStack.push_back(BreakContinue());
+    extendRegion(S->getBody());
+    Counter BackedgeCount =
+        propagateCounts(addCounters(ParentCount, BodyCount), S->getBody());
     BreakContinue BC = BreakContinueStack.pop_back_val();
-    // The count at the start of the condition is equal to the count at the
-    // end of the body. The adjusted count does not include either the
-    // fall-through count coming into the loop or the continue count, so add
-    // both of those separately. This is coincidentally the same equation as
-    // with while loops but for different reasons.
-    Cnt.setCurrentRegionCount(
-        addCounters(Cnt.getParentCount(),
-                    addCounters(Cnt.getAdjustedCount(), BC.ContinueCount)));
-    Visit(S->getCond());
-    Cnt.adjustForControlFlow();
-    Cnt.applyAdjustmentsToRegion(addCounters(BC.BreakCount, BC.ContinueCount));
+
+    Counter CondCount = addCounters(BackedgeCount, BC.ContinueCount);
+    propagateCounts(CondCount, S->getCond());
+
+    Counter OutCount =
+        addCounters(BC.BreakCount, subtractCounters(CondCount, BodyCount));
+    if (OutCount != ParentCount)
+      pushRegion(OutCount);
   }
 
   void VisitForStmt(const ForStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
+    extendRegion(S);
     if (S->getInit())
       Visit(S->getInit());
 
-    // Counter tracks the body of the loop.
-    RegionMapper Cnt(this, S);
+    Counter ParentCount = getRegion().getCounter();
+    Counter BodyCount = getRegionCounter(S);
+
+    // Handle the body first so that we can get the backedge count.
     BreakContinueStack.push_back(BreakContinue());
-    // Visit the body region first. (This is basically the same as a while
-    // loop; see further comments in VisitWhileStmt.)
-    Cnt.beginRegion();
-    Visit(S->getBody());
-    Cnt.adjustForControlFlow();
+    extendRegion(S->getBody());
+    Counter BackedgeCount = propagateCounts(BodyCount, S->getBody());
+    BreakContinue BC = BreakContinueStack.pop_back_val();
 
     // The increment is essentially part of the body but it needs to include
     // the count for all the continue statements.
-    if (S->getInc()) {
-      Cnt.setCurrentRegionCount(addCounters(
-          getCurrentRegionCount(), BreakContinueStack.back().ContinueCount));
-      beginSourceRegionGroup(S->getInc());
-      Visit(S->getInc());
-      endSourceRegionGroup();
-      Cnt.adjustForControlFlow();
+    if (const Stmt *Inc = S->getInc())
+      propagateCounts(addCounters(BackedgeCount, BC.ContinueCount), Inc);
+
+    // Go back to handle the condition.
+    Counter CondCount =
+        addCounters(ParentCount, BackedgeCount, BC.ContinueCount);
+    if (const Expr *Cond = S->getCond()) {
+      propagateCounts(CondCount, Cond);
+      adjustForOutOfOrderTraversal(getEnd(S));
     }
 
-    BreakContinue BC = BreakContinueStack.pop_back_val();
-
-    // ...then go back and propagate counts through the condition.
-    if (S->getCond()) {
-      Cnt.setCurrentRegionCount(
-          addCounters(addCounters(Cnt.getParentCount(), Cnt.getAdjustedCount()),
-                      BC.ContinueCount));
-      beginSourceRegionGroup(S->getCond());
-      Visit(S->getCond());
-      endSourceRegionGroup();
-      Cnt.adjustForControlFlow();
-    }
-    Cnt.applyAdjustmentsToRegion(addCounters(BC.BreakCount, BC.ContinueCount));
+    Counter OutCount =
+        addCounters(BC.BreakCount, subtractCounters(CondCount, BodyCount));
+    if (OutCount != ParentCount)
+      pushRegion(OutCount);
   }
 
   void VisitCXXForRangeStmt(const CXXForRangeStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
+    extendRegion(S);
+    Visit(S->getLoopVarStmt());
     Visit(S->getRangeStmt());
-    Visit(S->getBeginEndStmt());
-    // Counter tracks the body of the loop.
-    RegionMapper Cnt(this, S);
+
+    Counter ParentCount = getRegion().getCounter();
+    Counter BodyCount = getRegionCounter(S);
+
     BreakContinueStack.push_back(BreakContinue());
-    // Visit the body region first. (This is basically the same as a while
-    // loop; see further comments in VisitWhileStmt.)
-    Cnt.beginRegion();
-    Visit(S->getBody());
-    Cnt.adjustForControlFlow();
+    extendRegion(S->getBody());
+    Counter BackedgeCount = propagateCounts(BodyCount, S->getBody());
     BreakContinue BC = BreakContinueStack.pop_back_val();
-    Cnt.applyAdjustmentsToRegion(addCounters(BC.BreakCount, BC.ContinueCount));
+
+    Counter LoopCount =
+        addCounters(ParentCount, BackedgeCount, BC.ContinueCount);
+    Counter OutCount =
+        addCounters(BC.BreakCount, subtractCounters(LoopCount, BodyCount));
+    if (OutCount != ParentCount)
+      pushRegion(OutCount);
   }
 
   void VisitObjCForCollectionStmt(const ObjCForCollectionStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
+    extendRegion(S);
     Visit(S->getElement());
-    // Counter tracks the body of the loop.
-    RegionMapper Cnt(this, S);
+
+    Counter ParentCount = getRegion().getCounter();
+    Counter BodyCount = getRegionCounter(S);
+
     BreakContinueStack.push_back(BreakContinue());
-    Cnt.beginRegion();
-    Visit(S->getBody());
+    extendRegion(S->getBody());
+    Counter BackedgeCount = propagateCounts(BodyCount, S->getBody());
     BreakContinue BC = BreakContinueStack.pop_back_val();
-    Cnt.adjustForControlFlow();
-    Cnt.applyAdjustmentsToRegion(addCounters(BC.BreakCount, BC.ContinueCount));
+
+    Counter LoopCount =
+        addCounters(ParentCount, BackedgeCount, BC.ContinueCount);
+    Counter OutCount =
+        addCounters(BC.BreakCount, subtractCounters(LoopCount, BodyCount));
+    if (OutCount != ParentCount)
+      pushRegion(OutCount);
   }
 
   void VisitSwitchStmt(const SwitchStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
+    extendRegion(S);
     Visit(S->getCond());
+
     BreakContinueStack.push_back(BreakContinue());
-    // Map the '}' for the body to have the same count as the regions after
-    // the switch.
-    SourceLocation RBracLoc;
-    if (const auto *CS = dyn_cast<CompoundStmt>(S->getBody())) {
-      mapSourceCodeRange(CS->getLBracLoc());
-      setCurrentRegionUnreachable(S);
-      for (Stmt::const_child_range I = CS->children(); I; ++I) {
-        if (*I)
-          this->Visit(*I);
+
+    const Stmt *Body = S->getBody();
+    extendRegion(Body);
+    if (const auto *CS = dyn_cast<CompoundStmt>(Body)) {
+      if (!CS->body_empty()) {
+        // The body of the switch needs a zero region so that fallthrough counts
+        // behave correctly, but it would be misleading to include the braces of
+        // the compound statement in the zeroed area, so we need to handle this
+        // specially.
+        size_t Index =
+            pushRegion(Counter::getZero(), getStart(CS->body_front()),
+                       getEnd(CS->body_back()));
+        for (const auto *Child : CS->children())
+          Visit(Child);
+        popRegions(Index);
       }
-      RBracLoc = CS->getRBracLoc();
-    } else {
-      setCurrentRegionUnreachable(S);
-      Visit(S->getBody());
-    }
-    // If the switch is inside a loop, add the continue counts.
+    } else
+      propagateCounts(Counter::getZero(), Body);
     BreakContinue BC = BreakContinueStack.pop_back_val();
+
     if (!BreakContinueStack.empty())
       BreakContinueStack.back().ContinueCount = addCounters(
           BreakContinueStack.back().ContinueCount, BC.ContinueCount);
-    // Counter tracks the exit block of the switch.
-    RegionMapper ExitCnt(this, S);
-    ExitCnt.beginRegion();
-    if (RBracLoc.isValid())
-      mapSourceCodeRange(RBracLoc);
-  }
 
-  void VisitCaseStmt(const CaseStmt *S) {
-    // Counter for this particular case. This counts only jumps from the
-    // switch header and does not include fallthrough from the case before
-    // this one.
-    RegionMapper Cnt(this, S);
-    Cnt.beginRegion(/*AddIncomingFallThrough=*/true);
-    mapSourceCodeRange(S->getLocStart());
-    mapToken(S->getColonLoc());
-    Visit(S->getSubStmt());
+    Counter ExitCount = getRegionCounter(S);
+    pushRegion(ExitCount);
   }
 
-  void VisitDefaultStmt(const DefaultStmt *S) {
-    // Counter for this default case. This does not include fallthrough from
-    // the previous case.
-    RegionMapper Cnt(this, S);
-    Cnt.beginRegion(/*AddIncomingFallThrough=*/true);
-    mapSourceCodeRange(S->getLocStart());
-    mapToken(S->getColonLoc());
+  void VisitSwitchCase(const SwitchCase *S) {
+    extendRegion(S);
+
+    SourceMappingRegion &Parent = getRegion();
+
+    Counter Count = addCounters(Parent.getCounter(), getRegionCounter(S));
+    // Reuse the existing region if it starts at our label. This is typical of
+    // the first case in a switch.
+    if (Parent.hasStartLoc() && Parent.getStartLoc() == getStart(S))
+      Parent.setCounter(Count);
+    else
+      pushRegion(Count, getStart(S));
+
+    if (const CaseStmt *CS = dyn_cast<CaseStmt>(S)) {
+      Visit(CS->getLHS());
+      if (const Expr *RHS = CS->getRHS())
+        Visit(RHS);
+    }
     Visit(S->getSubStmt());
   }
 
   void VisitIfStmt(const IfStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
-    Visit(S->getCond());
-    mapToken(S->getElseLoc());
-
-    // Counter tracks the "then" part of an if statement. The count for
-    // the "else" part, if it exists, will be calculated from this counter.
-    RegionMapper Cnt(this, S);
-    Cnt.beginRegion();
-    Visit(S->getThen());
-    Cnt.adjustForControlFlow();
-
-    if (S->getElse()) {
-      Cnt.beginElseRegion();
-      Visit(S->getElse());
-      Cnt.adjustForControlFlow();
-    }
-    Cnt.applyAdjustmentsToRegion();
+    extendRegion(S);
+
+    Counter ParentCount = getRegion().getCounter();
+    Counter ThenCount = getRegionCounter(S);
+
+    // Emitting a counter for the condition makes it easier to interpret the
+    // counter for the body when looking at the coverage.
+    propagateCounts(ParentCount, S->getCond());
+
+    extendRegion(S->getThen());
+    Counter OutCount = propagateCounts(ThenCount, S->getThen());
+
+    Counter ElseCount = subtractCounters(ParentCount, ThenCount);
+    if (const Stmt *Else = S->getElse()) {
+      extendRegion(S->getElse());
+      OutCount = addCounters(OutCount, propagateCounts(ElseCount, Else));
+    } else
+      OutCount = addCounters(OutCount, ElseCount);
+
+    if (OutCount != ParentCount)
+      pushRegion(OutCount);
   }
 
   void VisitCXXTryStmt(const CXXTryStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
+    extendRegion(S);
     Visit(S->getTryBlock());
     for (unsigned I = 0, E = S->getNumHandlers(); I < E; ++I)
       Visit(S->getHandler(I));
-    // Counter tracks the continuation block of the try statement.
-    RegionMapper Cnt(this, S);
-    Cnt.beginRegion();
+
+    Counter ExitCount = getRegionCounter(S);
+    pushRegion(ExitCount);
   }
 
   void VisitCXXCatchStmt(const CXXCatchStmt *S) {
-    mapSourceCodeRange(S->getLocStart());
-    // Counter tracks the catch statement's handler block.
-    RegionMapper Cnt(this, S);
-    Cnt.beginRegion();
-    Visit(S->getHandlerBlock());
+    extendRegion(S);
+    propagateCounts(getRegionCounter(S), S->getHandlerBlock());
   }
 
   void VisitAbstractConditionalOperator(const AbstractConditionalOperator *E) {
-    Visit(E->getCond());
-    mapToken(E->getQuestionLoc());
-    mapToken(E->getColonLoc());
+    extendRegion(E);
 
-    // Counter tracks the "true" part of a conditional operator. The
-    // count in the "false" part will be calculated from this counter.
-    RegionMapper Cnt(this, E);
-    Cnt.beginRegion();
-    Visit(E->getTrueExpr());
-    Cnt.adjustForControlFlow();
+    Counter ParentCount = getRegion().getCounter();
+    Counter TrueCount = getRegionCounter(E);
 
-    Cnt.beginElseRegion();
-    Visit(E->getFalseExpr());
-    Cnt.adjustForControlFlow();
+    Visit(E->getCond());
 
-    Cnt.applyAdjustmentsToRegion();
+    if (!isa<BinaryConditionalOperator>(E)) {
+      extendRegion(E->getTrueExpr());
+      propagateCounts(TrueCount, E->getTrueExpr());
+    }
+    extendRegion(E->getFalseExpr());
+    propagateCounts(subtractCounters(ParentCount, TrueCount),
+                    E->getFalseExpr());
   }
 
   void VisitBinLAnd(const BinaryOperator *E) {
+    extendRegion(E);
     Visit(E->getLHS());
-    mapToken(E->getOperatorLoc());
-    // Counter tracks the right hand side of a logical and operator.
-    RegionMapper Cnt(this, E);
-    Cnt.beginRegion();
-    Visit(E->getRHS());
-    Cnt.adjustForControlFlow();
-    Cnt.applyAdjustmentsToRegion();
-  }
-
-  void VisitBinLOr(const BinaryOperator *E) {
-    Visit(E->getLHS());
-    mapToken(E->getOperatorLoc());
-    // Counter tracks the right hand side of a logical or operator.
-    RegionMapper Cnt(this, E);
-    Cnt.beginRegion();
-    Visit(E->getRHS());
-    Cnt.adjustForControlFlow();
-    Cnt.applyAdjustmentsToRegion();
-  }
 
-  void VisitParenExpr(const ParenExpr *E) {
-    mapToken(E->getLParen());
-    Visit(E->getSubExpr());
-    mapToken(E->getRParen());
+    extendRegion(E->getRHS());
+    propagateCounts(getRegionCounter(E), E->getRHS());
   }
 
-  void VisitBinaryOperator(const BinaryOperator *E) {
-    Visit(E->getLHS());
-    mapToken(E->getOperatorLoc());
-    Visit(E->getRHS());
-  }
-
-  void VisitUnaryOperator(const UnaryOperator *E) {
-    bool Postfix = E->isPostfix();
-    if (!Postfix)
-      mapToken(E->getOperatorLoc());
-    Visit(E->getSubExpr());
-    if (Postfix)
-      mapToken(E->getOperatorLoc());
-  }
-
-  void VisitMemberExpr(const MemberExpr *E) {
-    Visit(E->getBase());
-    mapToken(E->getMemberLoc());
-  }
-
-  void VisitCallExpr(const CallExpr *E) {
-    Visit(E->getCallee());
-    for (const auto &Arg : E->arguments())
-      Visit(Arg);
-    mapToken(E->getRParenLoc());
-  }
-
-  void VisitArraySubscriptExpr(const ArraySubscriptExpr *E) {
+  void VisitBinLOr(const BinaryOperator *E) {
+    extendRegion(E);
     Visit(E->getLHS());
-    Visit(E->getRHS());
-    mapToken(E->getRBracketLoc());
-  }
-
-  void VisitCStyleCastExpr(const CStyleCastExpr *E) {
-    mapToken(E->getLParenLoc());
-    mapToken(E->getRParenLoc());
-    Visit(E->getSubExpr());
-  }
-
-  // Map literals as tokens so that the macros like #define PI 3.14
-  // won't generate coverage mapping regions.
 
-  void VisitIntegerLiteral(const IntegerLiteral *E) {
-    mapToken(E->getLocStart());
+    extendRegion(E->getRHS());
+    propagateCounts(getRegionCounter(E), E->getRHS());
   }
 
-  void VisitFloatingLiteral(const FloatingLiteral *E) {
-    mapToken(E->getLocStart());
-  }
-
-  void VisitCharacterLiteral(const CharacterLiteral *E) {
-    mapToken(E->getLocStart());
-  }
-
-  void VisitStringLiteral(const StringLiteral *E) {
-    mapToken(E->getLocStart());
-  }
-
-  void VisitImaginaryLiteral(const ImaginaryLiteral *E) {
-    mapToken(E->getLocStart());
-  }
-
-  void VisitObjCMessageExpr(const ObjCMessageExpr *E) {
-    mapToken(E->getLeftLoc());
-    for (Stmt::const_child_range I = static_cast<const Stmt*>(E)->children(); I;
-         ++I) {
-      if (*I)
-        this->Visit(*I);
-    }
-    mapToken(E->getRightLoc());
+  void VisitLambdaExpr(const LambdaExpr *LE) {
+    // Lambdas are treated as their own functions for now, so we shouldn't
+    // propagate counts into them.
   }
 };
 }
@@ -1009,10 +891,12 @@ static StringRef getCoverageSection(const CodeGenModule &CGM) {
   return isMachO(CGM) ? "__DATA,__llvm_covmap" : "__llvm_covmap";
 }
 
-static void dump(llvm::raw_ostream &OS, const CoverageMappingRecord &Function) {
-  OS << Function.FunctionName << ":\n";
-  CounterMappingContext Ctx(Function.Expressions);
-  for (const auto &R : Function.MappingRegions) {
+static void dump(llvm::raw_ostream &OS, StringRef FunctionName,
+                 ArrayRef<CounterExpression> Expressions,
+                 ArrayRef<CounterMappingRegion> Regions) {
+  OS << FunctionName << ":\n";
+  CounterMappingContext Ctx(Expressions);
+  for (const auto &R : Regions) {
     OS.indent(2);
     switch (R.Kind) {
     case CounterMappingRegion::CodeRegion:
@@ -1025,15 +909,12 @@ static void dump(llvm::raw_ostream &OS, const CoverageMappingRecord &Function) {
       break;
     }
 
-    OS << "File " << R.FileID << ", " << R.LineStart << ":"
-           << R.ColumnStart << " -> " << R.LineEnd << ":" << R.ColumnEnd
-           << " = ";
-    Ctx.dump(R.Count);
-    OS << " (HasCodeBefore = " << R.HasCodeBefore;
+    OS << "File " << R.FileID << ", " << R.LineStart << ":" << R.ColumnStart
+       << " -> " << R.LineEnd << ":" << R.ColumnEnd << " = ";
+    Ctx.dump(R.Count, OS);
     if (R.Kind == CounterMappingRegion::ExpansionRegion)
-      OS << ", Expanded file = " << R.ExpandedFileID;
-
-    OS << ")\n";
+      OS << " (Expanded file = " << R.ExpandedFileID << ")";
+    OS << "\n";
   }
 }
 
@@ -1072,13 +953,11 @@ void CoverageMappingModuleGen::addFunctionMappingRecord(
     FilenameRefs.resize(FileEntries.size());
     for (const auto &Entry : FileEntries)
       FilenameRefs[Entry.second] = Entry.first->getName();
-    RawCoverageMappingReader Reader(FunctionNameValue, CoverageMapping,
-                                    FilenameRefs,
-                                    Filenames, Expressions, Regions);
-    CoverageMappingRecord FunctionRecord;
-    if (Reader.read(FunctionRecord))
+    RawCoverageMappingReader Reader(CoverageMapping, FilenameRefs, Filenames,
+                                    Expressions, Regions);
+    if (Reader.read())
       return;
-    dump(llvm::outs(), FunctionRecord);
+    dump(llvm::outs(), FunctionNameValue, Expressions, Regions);
   }
 }
 
@@ -1098,7 +977,7 @@ void CoverageMappingModuleGen::emit() {
     llvm::sys::fs::make_absolute(Path);
 
     auto I = Entry.second;
-    FilenameStrs[I] = std::move(std::string(Path.begin(), Path.end()));
+    FilenameStrs[I] = std::string(Path.begin(), Path.end());
     FilenameRefs[I] = FilenameStrs[I];
   }
 
diff --git a/lib/CodeGen/EHScopeStack.h b/lib/CodeGen/EHScopeStack.h
index e69584840990..a7951888c825 100644
--- a/lib/CodeGen/EHScopeStack.h
+++ b/lib/CodeGen/EHScopeStack.h
@@ -17,6 +17,7 @@
 #define LLVM_CLANG_LIB_CODEGEN_EHSCOPESTACK_H
 
 #include "clang/Basic/LLVM.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/IR/BasicBlock.h"
 #include "llvm/IR/Instructions.h"
@@ -75,8 +76,14 @@ template <class T> struct DominatingPointer<T,false> : InvariantValue<T*> {};
 template <class T> struct DominatingValue<T*> : DominatingPointer<T> {};
 
 enum CleanupKind : unsigned {
+  /// Denotes a cleanup that should run when a scope is exited using exceptional
+  /// control flow (a throw statement leading to stack unwinding, ).
   EHCleanup = 0x1,
+
+  /// Denotes a cleanup that should run when a scope is exited using normal
+  /// control flow (falling off the end of the scope, return, goto, ...).
   NormalCleanup = 0x2,
+
   NormalAndEHCleanup = EHCleanup | NormalCleanup,
 
   InactiveCleanup = 0x4,
@@ -175,84 +182,28 @@ public:
     virtual void Emit(CodeGenFunction &CGF, Flags flags) = 0;
   };
 
-  /// ConditionalCleanupN stores the saved form of its N parameters,
+  /// ConditionalCleanup stores the saved form of its parameters,
   /// then restores them and performs the cleanup.
-  template <class T, class A0>
-  class ConditionalCleanup1 : public Cleanup {
-    typedef typename DominatingValue<A0>::saved_type A0_saved;
-    A0_saved a0_saved;
-
-    void Emit(CodeGenFunction &CGF, Flags flags) override {
-      A0 a0 = DominatingValue<A0>::restore(CGF, a0_saved);
-      T(a0).Emit(CGF, flags);
-    }
-
-  public:
-    ConditionalCleanup1(A0_saved a0)
-      : a0_saved(a0) {}
-  };
-
-  template <class T, class A0, class A1>
-  class ConditionalCleanup2 : public Cleanup {
-    typedef typename DominatingValue<A0>::saved_type A0_saved;
-    typedef typename DominatingValue<A1>::saved_type A1_saved;
-    A0_saved a0_saved;
-    A1_saved a1_saved;
-
-    void Emit(CodeGenFunction &CGF, Flags flags) override {
-      A0 a0 = DominatingValue<A0>::restore(CGF, a0_saved);
-      A1 a1 = DominatingValue<A1>::restore(CGF, a1_saved);
-      T(a0, a1).Emit(CGF, flags);
+  template <class T, class... As> class ConditionalCleanup : public Cleanup {
+    typedef std::tuple<typename DominatingValue<As>::saved_type...> SavedTuple;
+    SavedTuple Saved;
+
+    template <std::size_t... Is>
+    T restore(CodeGenFunction &CGF, llvm::index_sequence<Is...>) {
+      // It's important that the restores are emitted in order. The braced init
+      // list guarentees that.
+      return T{DominatingValue<As>::restore(CGF, std::get<Is>(Saved))...};
     }
 
-  public:
-    ConditionalCleanup2(A0_saved a0, A1_saved a1)
-      : a0_saved(a0), a1_saved(a1) {}
-  };
-
-  template <class T, class A0, class A1, class A2>
-  class ConditionalCleanup3 : public Cleanup {
-    typedef typename DominatingValue<A0>::saved_type A0_saved;
-    typedef typename DominatingValue<A1>::saved_type A1_saved;
-    typedef typename DominatingValue<A2>::saved_type A2_saved;
-    A0_saved a0_saved;
-    A1_saved a1_saved;
-    A2_saved a2_saved;
-
     void Emit(CodeGenFunction &CGF, Flags flags) override {
-      A0 a0 = DominatingValue<A0>::restore(CGF, a0_saved);
-      A1 a1 = DominatingValue<A1>::restore(CGF, a1_saved);
-      A2 a2 = DominatingValue<A2>::restore(CGF, a2_saved);
-      T(a0, a1, a2).Emit(CGF, flags);
+      restore(CGF, llvm::index_sequence_for<As...>()).Emit(CGF, flags);
     }
 
   public:
-    ConditionalCleanup3(A0_saved a0, A1_saved a1, A2_saved a2)
-      : a0_saved(a0), a1_saved(a1), a2_saved(a2) {}
-  };
-
-  template <class T, class A0, class A1, class A2, class A3>
-  class ConditionalCleanup4 : public Cleanup {
-    typedef typename DominatingValue<A0>::saved_type A0_saved;
-    typedef typename DominatingValue<A1>::saved_type A1_saved;
-    typedef typename DominatingValue<A2>::saved_type A2_saved;
-    typedef typename DominatingValue<A3>::saved_type A3_saved;
-    A0_saved a0_saved;
-    A1_saved a1_saved;
-    A2_saved a2_saved;
-    A3_saved a3_saved;
+    ConditionalCleanup(typename DominatingValue<As>::saved_type... A)
+        : Saved(A...) {}
 
-    void Emit(CodeGenFunction &CGF, Flags flags) override {
-      A0 a0 = DominatingValue<A0>::restore(CGF, a0_saved);
-      A1 a1 = DominatingValue<A1>::restore(CGF, a1_saved);
-      A2 a2 = DominatingValue<A2>::restore(CGF, a2_saved);
-      A3 a3 = DominatingValue<A3>::restore(CGF, a3_saved);
-      T(a0, a1, a2, a3).Emit(CGF, flags);
-    }
-
-  public:
-    ConditionalCleanup4(A0_saved a0, A1_saved a1, A2_saved a2, A3_saved a3)
-      : a0_saved(a0), a1_saved(a1), a2_saved(a2), a3_saved(a3) {}
+    ConditionalCleanup(SavedTuple Tuple) : Saved(std::move(Tuple)) {}
   };
 
 private:
@@ -306,53 +257,18 @@ public:
                    InnermostEHScope(stable_end()) {}
   ~EHScopeStack() { delete[] StartOfBuffer; }
 
-  // Variadic templates would make this not terrible.
-
   /// Push a lazily-created cleanup on the stack.
-  template <class T>
-  void pushCleanup(CleanupKind Kind) {
+  template <class T, class... As> void pushCleanup(CleanupKind Kind, As... A) {
     void *Buffer = pushCleanup(Kind, sizeof(T));
-    Cleanup *Obj = new(Buffer) T();
+    Cleanup *Obj = new (Buffer) T(A...);
     (void) Obj;
   }
 
-  /// Push a lazily-created cleanup on the stack.
-  template <class T, class A0>
-  void pushCleanup(CleanupKind Kind, A0 a0) {
-    void *Buffer = pushCleanup(Kind, sizeof(T));
-    Cleanup *Obj = new(Buffer) T(a0);
-    (void) Obj;
-  }
-
-  /// Push a lazily-created cleanup on the stack.
-  template <class T, class A0, class A1>
-  void pushCleanup(CleanupKind Kind, A0 a0, A1 a1) {
-    void *Buffer = pushCleanup(Kind, sizeof(T));
-    Cleanup *Obj = new(Buffer) T(a0, a1);
-    (void) Obj;
-  }
-
-  /// Push a lazily-created cleanup on the stack.
-  template <class T, class A0, class A1, class A2>
-  void pushCleanup(CleanupKind Kind, A0 a0, A1 a1, A2 a2) {
+  /// Push a lazily-created cleanup on the stack. Tuple version.
+  template <class T, class... As>
+  void pushCleanupTuple(CleanupKind Kind, std::tuple<As...> A) {
     void *Buffer = pushCleanup(Kind, sizeof(T));
-    Cleanup *Obj = new(Buffer) T(a0, a1, a2);
-    (void) Obj;
-  }
-
-  /// Push a lazily-created cleanup on the stack.
-  template <class T, class A0, class A1, class A2, class A3>
-  void pushCleanup(CleanupKind Kind, A0 a0, A1 a1, A2 a2, A3 a3) {
-    void *Buffer = pushCleanup(Kind, sizeof(T));
-    Cleanup *Obj = new(Buffer) T(a0, a1, a2, a3);
-    (void) Obj;
-  }
-
-  /// Push a lazily-created cleanup on the stack.
-  template <class T, class A0, class A1, class A2, class A3, class A4>
-  void pushCleanup(CleanupKind Kind, A0 a0, A1 a1, A2 a2, A3 a3, A4 a4) {
-    void *Buffer = pushCleanup(Kind, sizeof(T));
-    Cleanup *Obj = new(Buffer) T(a0, a1, a2, a3, a4);
+    Cleanup *Obj = new (Buffer) T(std::move(A));
     (void) Obj;
   }
 
@@ -369,10 +285,10 @@ public:
   ///
   /// The pointer returned from this method is valid until the cleanup
   /// stack is modified.
-  template <class T, class A0, class A1, class A2>
-  T *pushCleanupWithExtra(CleanupKind Kind, size_t N, A0 a0, A1 a1, A2 a2) {
+  template <class T, class... As>
+  T *pushCleanupWithExtra(CleanupKind Kind, size_t N, As... A) {
     void *Buffer = pushCleanup(Kind, sizeof(T) + T::getExtraSize(N));
-    return new (Buffer) T(N, a0, a1, a2);
+    return new (Buffer) T(N, A...);
   }
 
   void pushCopyOfCleanup(CleanupKind Kind, const void *Cleanup, size_t Size) {
@@ -403,6 +319,10 @@ public:
   /// Pops a terminate handler off the stack.
   void popTerminate();
 
+  // Returns true iff the current scope is either empty or contains only
+  // lifetime markers, i.e. no real cleanup code
+  bool containsOnlyLifetimeMarkers(stable_iterator Old) const;
+
   /// Determines whether the exception-scopes stack is empty.
   bool empty() const { return StartOfData == EndOfBuffer; }
 
diff --git a/lib/CodeGen/ItaniumCXXABI.cpp b/lib/CodeGen/ItaniumCXXABI.cpp
index 9e88b1ebbadc..0a1a4ceea3ef 100644
--- a/lib/CodeGen/ItaniumCXXABI.cpp
+++ b/lib/CodeGen/ItaniumCXXABI.cpp
@@ -19,14 +19,18 @@
 //===----------------------------------------------------------------------===//
 
 #include "CGCXXABI.h"
+#include "CGCleanup.h"
 #include "CGRecordLayout.h"
 #include "CGVTables.h"
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
+#include "TargetInfo.h"
 #include "clang/AST/Mangle.h"
 #include "clang/AST/Type.h"
+#include "clang/AST/StmtCXX.h"
 #include "llvm/IR/CallSite.h"
 #include "llvm/IR/DataLayout.h"
+#include "llvm/IR/Instructions.h"
 #include "llvm/IR/Intrinsics.h"
 #include "llvm/IR/Value.h"
 
@@ -111,10 +115,22 @@ public:
                                const CXXDestructorDecl *Dtor) override;
 
   void emitRethrow(CodeGenFunction &CGF, bool isNoReturn) override;
+  void emitThrow(CodeGenFunction &CGF, const CXXThrowExpr *E) override;
+
+  void emitBeginCatch(CodeGenFunction &CGF, const CXXCatchStmt *C) override;
+
+  llvm::CallInst *
+  emitTerminateForUnexpectedException(CodeGenFunction &CGF,
+                                      llvm::Value *Exn) override;
 
   void EmitFundamentalRTTIDescriptor(QualType Type);
   void EmitFundamentalRTTIDescriptors();
   llvm::Constant *getAddrOfRTTIDescriptor(QualType Ty) override;
+  llvm::Constant *
+  getAddrOfCXXCatchHandlerType(QualType Ty,
+                               QualType CatchHandlerType) override {
+    return getAddrOfRTTIDescriptor(Ty);
+  }
 
   bool shouldTypeidBeNullChecked(bool IsDeref, QualType SrcRecordTy) override;
   void EmitBadTypeidCall(CodeGenFunction &CGF) override;
@@ -849,7 +865,7 @@ bool ItaniumCXXABI::classifyReturnType(CGFunctionInfo &FI) const {
 /// The Itanium ABI requires non-zero initialization only for data
 /// member pointers, for which '0' is a valid offset.
 bool ItaniumCXXABI::isZeroInitializable(const MemberPointerType *MPT) {
-  return MPT->getPointeeType()->isFunctionType();
+  return MPT->isMemberFunctionPointer();
 }
 
 /// The Itanium ABI always places an offset to the complete object
@@ -906,6 +922,59 @@ void ItaniumCXXABI::emitRethrow(CodeGenFunction &CGF, bool isNoReturn) {
     CGF.EmitRuntimeCallOrInvoke(Fn);
 }
 
+static llvm::Constant *getAllocateExceptionFn(CodeGenModule &CGM) {
+  // void *__cxa_allocate_exception(size_t thrown_size);
+
+  llvm::FunctionType *FTy =
+    llvm::FunctionType::get(CGM.Int8PtrTy, CGM.SizeTy, /*IsVarArgs=*/false);
+
+  return CGM.CreateRuntimeFunction(FTy, "__cxa_allocate_exception");
+}
+
+static llvm::Constant *getThrowFn(CodeGenModule &CGM) {
+  // void __cxa_throw(void *thrown_exception, std::type_info *tinfo,
+  //                  void (*dest) (void *));
+
+  llvm::Type *Args[3] = { CGM.Int8PtrTy, CGM.Int8PtrTy, CGM.Int8PtrTy };
+  llvm::FunctionType *FTy =
+    llvm::FunctionType::get(CGM.VoidTy, Args, /*IsVarArgs=*/false);
+
+  return CGM.CreateRuntimeFunction(FTy, "__cxa_throw");
+}
+
+void ItaniumCXXABI::emitThrow(CodeGenFunction &CGF, const CXXThrowExpr *E) {
+  QualType ThrowType = E->getSubExpr()->getType();
+  // Now allocate the exception object.
+  llvm::Type *SizeTy = CGF.ConvertType(getContext().getSizeType());
+  uint64_t TypeSize = getContext().getTypeSizeInChars(ThrowType).getQuantity();
+
+  llvm::Constant *AllocExceptionFn = getAllocateExceptionFn(CGM);
+  llvm::CallInst *ExceptionPtr = CGF.EmitNounwindRuntimeCall(
+      AllocExceptionFn, llvm::ConstantInt::get(SizeTy, TypeSize), "exception");
+
+  CGF.EmitAnyExprToExn(E->getSubExpr(), ExceptionPtr);
+
+  // Now throw the exception.
+  llvm::Constant *TypeInfo = CGM.GetAddrOfRTTIDescriptor(ThrowType,
+                                                         /*ForEH=*/true);
+
+  // The address of the destructor.  If the exception type has a
+  // trivial destructor (or isn't a record), we just pass null.
+  llvm::Constant *Dtor = nullptr;
+  if (const RecordType *RecordTy = ThrowType->getAs<RecordType>()) {
+    CXXRecordDecl *Record = cast<CXXRecordDecl>(RecordTy->getDecl());
+    if (!Record->hasTrivialDestructor()) {
+      CXXDestructorDecl *DtorD = Record->getDestructor();
+      Dtor = CGM.getAddrOfCXXStructor(DtorD, StructorType::Complete);
+      Dtor = llvm::ConstantExpr::getBitCast(Dtor, CGM.Int8PtrTy);
+    }
+  }
+  if (!Dtor) Dtor = llvm::Constant::getNullValue(CGM.Int8PtrTy);
+
+  llvm::Value *args[] = { ExceptionPtr, TypeInfo, Dtor };
+  CGF.EmitNoreturnRuntimeCallOrInvoke(getThrowFn(CGM), args);
+}
+
 static llvm::Constant *getItaniumDynamicCastFn(CodeGenFunction &CGF) {
   // void *__dynamic_cast(const void *sub,
   //                      const abi::__class_type_info *src,
@@ -1259,6 +1328,9 @@ void ItaniumCXXABI::emitVTableDefinitions(CodeGenVTables &CGVT,
   // Set the correct linkage.
   VTable->setLinkage(Linkage);
 
+  if (CGM.supportsCOMDAT() && VTable->isWeakForLinker())
+    VTable->setComdat(CGM.getModule().getOrInsertComdat(VTable->getName()));
+
   // Set the right visibility.
   CGM.setGlobalVisibility(VTable, RD);
 
@@ -1278,6 +1350,8 @@ void ItaniumCXXABI::emitVTableDefinitions(CodeGenVTables &CGVT,
       cast<NamespaceDecl>(DC)->getIdentifier()->isStr("__cxxabiv1") &&
       DC->getParent()->isTranslationUnit())
     EmitFundamentalRTTIDescriptors();
+
+  CGM.EmitVTableBitSetEntries(VTable, VTLayout);
 }
 
 llvm::Value *ItaniumCXXABI::getVTableAddressPointInStructor(
@@ -1314,7 +1388,7 @@ llvm::Value *ItaniumCXXABI::getVTableAddressPointInStructor(
 
 llvm::Constant *ItaniumCXXABI::getVTableAddressPointForConstExpr(
     BaseSubobject Base, const CXXRecordDecl *VTableClass) {
-  llvm::Constant *VTable = getAddrOfVTable(VTableClass, CharUnits());
+  auto *VTable = getAddrOfVTable(VTableClass, CharUnits());
 
   // Find the appropriate vtable within the vtable group.
   uint64_t AddressPoint = CGM.getItaniumVTableContext()
@@ -1325,7 +1399,8 @@ llvm::Constant *ItaniumCXXABI::getVTableAddressPointForConstExpr(
     llvm::ConstantInt::get(CGM.Int64Ty, AddressPoint)
   };
 
-  return llvm::ConstantExpr::getInBoundsGetElementPtr(VTable, Indices);
+  return llvm::ConstantExpr::getInBoundsGetElementPtr(VTable->getValueType(),
+                                                      VTable, Indices);
 }
 
 llvm::GlobalVariable *ItaniumCXXABI::getAddrOfVTable(const CXXRecordDecl *RD,
@@ -1369,6 +1444,9 @@ llvm::Value *ItaniumCXXABI::getVirtualFunctionPointer(CodeGenFunction &CGF,
   Ty = Ty->getPointerTo()->getPointerTo();
   llvm::Value *VTable = CGF.GetVTablePtr(This, Ty);
 
+  if (CGF.SanOpts.has(SanitizerKind::CFIVCall))
+    CGF.EmitVTablePtrCheckForCall(cast<CXXMethodDecl>(GD.getDecl()), VTable);
+
   uint64_t VTableIndex = CGM.getItaniumVTableContext().getMethodVTableIndex(GD);
   llvm::Value *VFuncPtr =
       CGF.Builder.CreateConstInBoundsGEP1_64(VTable, VTableIndex, "vfn");
@@ -1591,7 +1669,7 @@ llvm::Value *ARMCXXABI::InitializeArrayCookie(CodeGenFunction &CGF,
   CGF.Builder.CreateStore(elementSize, cookie);
 
   // The second element is the element count.
-  cookie = CGF.Builder.CreateConstInBoundsGEP1_32(cookie, 1);
+  cookie = CGF.Builder.CreateConstInBoundsGEP1_32(CGF.SizeTy, cookie, 1);
   CGF.Builder.CreateStore(numElements, cookie);
 
   // Finally, compute a pointer to the actual data buffer by skipping
@@ -1714,11 +1792,12 @@ void ItaniumCXXABI::EmitGuardedInit(CodeGenFunction &CGF,
 
     // The ABI says: It is suggested that it be emitted in the same COMDAT group
     // as the associated data object
-    if (!D.isLocalVarDecl() && var->isWeakForLinker() && CGM.supportsCOMDAT()) {
-      llvm::Comdat *C = CGM.getModule().getOrInsertComdat(var->getName());
+    llvm::Comdat *C = var->getComdat();
+    if (!D.isLocalVarDecl() && C) {
       guard->setComdat(C);
-      var->setComdat(C);
       CGF.CurFn->setComdat(C);
+    } else if (CGM.supportsCOMDAT() && guard->isWeakForLinker()) {
+      guard->setComdat(CGM.getModule().getOrInsertComdat(guard->getName()));
     }
 
     CGM.setStaticLocalDeclGuardAddress(&D, guard);
@@ -2011,7 +2090,7 @@ void ItaniumCXXABI::EmitThreadLocalInitFuncs(
     CGBuilderTy Builder(Entry);
     if (InitIsInitFunc) {
       if (Init)
-        Builder.CreateCall(Init);
+        Builder.CreateCall(Init, {});
     } else {
       // Don't know whether we have an init function. Call it if it exists.
       llvm::Value *Have = Builder.CreateIsNotNull(Init);
@@ -2020,7 +2099,7 @@ void ItaniumCXXABI::EmitThreadLocalInitFuncs(
       Builder.CreateCondBr(Have, InitBB, ExitBB);
 
       Builder.SetInsertPoint(InitBB);
-      Builder.CreateCall(Init);
+      Builder.CreateCall(Init, {});
       Builder.CreateBr(ExitBB);
 
       Builder.SetInsertPoint(ExitBB);
@@ -2049,7 +2128,7 @@ LValue ItaniumCXXABI::EmitThreadLocalVarDeclLValue(CodeGenFunction &CGF,
   llvm::Value *Val = CGF.CGM.GetAddrOfGlobalVar(VD, Ty);
   llvm::Function *Wrapper = getOrCreateThreadLocalWrapper(VD, Val);
 
-  Val = CGF.Builder.CreateCall(Wrapper);
+  Val = CGF.Builder.CreateCall(Wrapper, {});
 
   LValue LV;
   if (VD->getType()->isReferenceType())
@@ -2528,7 +2607,8 @@ void ItaniumRTTIBuilder::BuildVTablePointer(const Type *Ty) {
 
   // The vtable address point is 2.
   llvm::Constant *Two = llvm::ConstantInt::get(PtrDiffTy, 2);
-  VTable = llvm::ConstantExpr::getInBoundsGetElementPtr(VTable, Two);
+  VTable =
+      llvm::ConstantExpr::getInBoundsGetElementPtr(CGM.Int8PtrTy, VTable, Two);
   VTable = llvm::ConstantExpr::getBitCast(VTable, CGM.Int8PtrTy);
 
   Fields.push_back(VTable);
@@ -2718,9 +2798,13 @@ llvm::Constant *ItaniumRTTIBuilder::BuildTypeInfo(QualType Ty, bool Force) {
 
   llvm::Constant *Init = llvm::ConstantStruct::getAnon(Fields);
 
+  llvm::Module &M = CGM.getModule();
   llvm::GlobalVariable *GV =
-    new llvm::GlobalVariable(CGM.getModule(), Init->getType(),
-                             /*Constant=*/true, Linkage, Init, Name);
+      new llvm::GlobalVariable(M, Init->getType(),
+                               /*Constant=*/true, Linkage, Init, Name);
+
+  if (CGM.supportsCOMDAT() && GV->isWeakForLinker())
+    GV->setComdat(M.getOrInsertComdat(GV->getName()));
 
   // If there's already an old global variable, replace it with the new one.
   if (OldGV) {
@@ -3136,8 +3220,8 @@ static void emitConstructorDestructorAlias(CodeGenModule &CGM,
   llvm::PointerType *AliasType = Aliasee->getType();
 
   // Create the alias with no name.
-  auto *Alias = llvm::GlobalAlias::create(
-      AliasType->getElementType(), 0, Linkage, "", Aliasee, &CGM.getModule());
+  auto *Alias = llvm::GlobalAlias::create(AliasType, Linkage, "", Aliasee,
+                                          &CGM.getModule());
 
   // Switch any previous uses to the alias.
   if (Entry) {
@@ -3204,5 +3288,352 @@ void ItaniumCXXABI::emitCXXStructor(const CXXMethodDecl *MD,
       getMangleContext().mangleCXXCtorComdat(CD, Out);
     llvm::Comdat *C = CGM.getModule().getOrInsertComdat(Out.str());
     Fn->setComdat(C);
+  } else {
+    CGM.maybeSetTrivialComdat(*MD, *Fn);
+  }
+}
+
+static llvm::Constant *getBeginCatchFn(CodeGenModule &CGM) {
+  // void *__cxa_begin_catch(void*);
+  llvm::FunctionType *FTy = llvm::FunctionType::get(
+      CGM.Int8PtrTy, CGM.Int8PtrTy, /*IsVarArgs=*/false);
+
+  return CGM.CreateRuntimeFunction(FTy, "__cxa_begin_catch");
+}
+
+static llvm::Constant *getEndCatchFn(CodeGenModule &CGM) {
+  // void __cxa_end_catch();
+  llvm::FunctionType *FTy =
+      llvm::FunctionType::get(CGM.VoidTy, /*IsVarArgs=*/false);
+
+  return CGM.CreateRuntimeFunction(FTy, "__cxa_end_catch");
+}
+
+static llvm::Constant *getGetExceptionPtrFn(CodeGenModule &CGM) {
+  // void *__cxa_get_exception_ptr(void*);
+  llvm::FunctionType *FTy = llvm::FunctionType::get(
+      CGM.Int8PtrTy, CGM.Int8PtrTy, /*IsVarArgs=*/false);
+
+  return CGM.CreateRuntimeFunction(FTy, "__cxa_get_exception_ptr");
+}
+
+namespace {
+  /// A cleanup to call __cxa_end_catch.  In many cases, the caught
+  /// exception type lets us state definitively that the thrown exception
+  /// type does not have a destructor.  In particular:
+  ///   - Catch-alls tell us nothing, so we have to conservatively
+  ///     assume that the thrown exception might have a destructor.
+  ///   - Catches by reference behave according to their base types.
+  ///   - Catches of non-record types will only trigger for exceptions
+  ///     of non-record types, which never have destructors.
+  ///   - Catches of record types can trigger for arbitrary subclasses
+  ///     of the caught type, so we have to assume the actual thrown
+  ///     exception type might have a throwing destructor, even if the
+  ///     caught type's destructor is trivial or nothrow.
+  struct CallEndCatch : EHScopeStack::Cleanup {
+    CallEndCatch(bool MightThrow) : MightThrow(MightThrow) {}
+    bool MightThrow;
+
+    void Emit(CodeGenFunction &CGF, Flags flags) override {
+      if (!MightThrow) {
+        CGF.EmitNounwindRuntimeCall(getEndCatchFn(CGF.CGM));
+        return;
+      }
+
+      CGF.EmitRuntimeCallOrInvoke(getEndCatchFn(CGF.CGM));
+    }
+  };
+}
+
+/// Emits a call to __cxa_begin_catch and enters a cleanup to call
+/// __cxa_end_catch.
+///
+/// \param EndMightThrow - true if __cxa_end_catch might throw
+static llvm::Value *CallBeginCatch(CodeGenFunction &CGF,
+                                   llvm::Value *Exn,
+                                   bool EndMightThrow) {
+  llvm::CallInst *call =
+    CGF.EmitNounwindRuntimeCall(getBeginCatchFn(CGF.CGM), Exn);
+
+  CGF.EHStack.pushCleanup<CallEndCatch>(NormalAndEHCleanup, EndMightThrow);
+
+  return call;
+}
+
+/// A "special initializer" callback for initializing a catch
+/// parameter during catch initialization.
+static void InitCatchParam(CodeGenFunction &CGF,
+                           const VarDecl &CatchParam,
+                           llvm::Value *ParamAddr,
+                           SourceLocation Loc) {
+  // Load the exception from where the landing pad saved it.
+  llvm::Value *Exn = CGF.getExceptionFromSlot();
+
+  CanQualType CatchType =
+    CGF.CGM.getContext().getCanonicalType(CatchParam.getType());
+  llvm::Type *LLVMCatchTy = CGF.ConvertTypeForMem(CatchType);
+
+  // If we're catching by reference, we can just cast the object
+  // pointer to the appropriate pointer.
+  if (isa<ReferenceType>(CatchType)) {
+    QualType CaughtType = cast<ReferenceType>(CatchType)->getPointeeType();
+    bool EndCatchMightThrow = CaughtType->isRecordType();
+
+    // __cxa_begin_catch returns the adjusted object pointer.
+    llvm::Value *AdjustedExn = CallBeginCatch(CGF, Exn, EndCatchMightThrow);
+
+    // We have no way to tell the personality function that we're
+    // catching by reference, so if we're catching a pointer,
+    // __cxa_begin_catch will actually return that pointer by value.
+    if (const PointerType *PT = dyn_cast<PointerType>(CaughtType)) {
+      QualType PointeeType = PT->getPointeeType();
+
+      // When catching by reference, generally we should just ignore
+      // this by-value pointer and use the exception object instead.
+      if (!PointeeType->isRecordType()) {
+
+        // Exn points to the struct _Unwind_Exception header, which
+        // we have to skip past in order to reach the exception data.
+        unsigned HeaderSize =
+          CGF.CGM.getTargetCodeGenInfo().getSizeOfUnwindException();
+        AdjustedExn = CGF.Builder.CreateConstGEP1_32(Exn, HeaderSize);
+
+      // However, if we're catching a pointer-to-record type that won't
+      // work, because the personality function might have adjusted
+      // the pointer.  There's actually no way for us to fully satisfy
+      // the language/ABI contract here:  we can't use Exn because it
+      // might have the wrong adjustment, but we can't use the by-value
+      // pointer because it's off by a level of abstraction.
+      //
+      // The current solution is to dump the adjusted pointer into an
+      // alloca, which breaks language semantics (because changing the
+      // pointer doesn't change the exception) but at least works.
+      // The better solution would be to filter out non-exact matches
+      // and rethrow them, but this is tricky because the rethrow
+      // really needs to be catchable by other sites at this landing
+      // pad.  The best solution is to fix the personality function.
+      } else {
+        // Pull the pointer for the reference type off.
+        llvm::Type *PtrTy =
+          cast<llvm::PointerType>(LLVMCatchTy)->getElementType();
+
+        // Create the temporary and write the adjusted pointer into it.
+        llvm::Value *ExnPtrTmp = CGF.CreateTempAlloca(PtrTy, "exn.byref.tmp");
+        llvm::Value *Casted = CGF.Builder.CreateBitCast(AdjustedExn, PtrTy);
+        CGF.Builder.CreateStore(Casted, ExnPtrTmp);
+
+        // Bind the reference to the temporary.
+        AdjustedExn = ExnPtrTmp;
+      }
+    }
+
+    llvm::Value *ExnCast =
+      CGF.Builder.CreateBitCast(AdjustedExn, LLVMCatchTy, "exn.byref");
+    CGF.Builder.CreateStore(ExnCast, ParamAddr);
+    return;
+  }
+
+  // Scalars and complexes.
+  TypeEvaluationKind TEK = CGF.getEvaluationKind(CatchType);
+  if (TEK != TEK_Aggregate) {
+    llvm::Value *AdjustedExn = CallBeginCatch(CGF, Exn, false);
+
+    // If the catch type is a pointer type, __cxa_begin_catch returns
+    // the pointer by value.
+    if (CatchType->hasPointerRepresentation()) {
+      llvm::Value *CastExn =
+        CGF.Builder.CreateBitCast(AdjustedExn, LLVMCatchTy, "exn.casted");
+
+      switch (CatchType.getQualifiers().getObjCLifetime()) {
+      case Qualifiers::OCL_Strong:
+        CastExn = CGF.EmitARCRetainNonBlock(CastExn);
+        // fallthrough
+
+      case Qualifiers::OCL_None:
+      case Qualifiers::OCL_ExplicitNone:
+      case Qualifiers::OCL_Autoreleasing:
+        CGF.Builder.CreateStore(CastExn, ParamAddr);
+        return;
+
+      case Qualifiers::OCL_Weak:
+        CGF.EmitARCInitWeak(ParamAddr, CastExn);
+        return;
+      }
+      llvm_unreachable("bad ownership qualifier!");
+    }
+
+    // Otherwise, it returns a pointer into the exception object.
+
+    llvm::Type *PtrTy = LLVMCatchTy->getPointerTo(0); // addrspace 0 ok
+    llvm::Value *Cast = CGF.Builder.CreateBitCast(AdjustedExn, PtrTy);
+
+    LValue srcLV = CGF.MakeNaturalAlignAddrLValue(Cast, CatchType);
+    LValue destLV = CGF.MakeAddrLValue(ParamAddr, CatchType,
+                                  CGF.getContext().getDeclAlign(&CatchParam));
+    switch (TEK) {
+    case TEK_Complex:
+      CGF.EmitStoreOfComplex(CGF.EmitLoadOfComplex(srcLV, Loc), destLV,
+                             /*init*/ true);
+      return;
+    case TEK_Scalar: {
+      llvm::Value *ExnLoad = CGF.EmitLoadOfScalar(srcLV, Loc);
+      CGF.EmitStoreOfScalar(ExnLoad, destLV, /*init*/ true);
+      return;
+    }
+    case TEK_Aggregate:
+      llvm_unreachable("evaluation kind filtered out!");
+    }
+    llvm_unreachable("bad evaluation kind");
+  }
+
+  assert(isa<RecordType>(CatchType) && "unexpected catch type!");
+
+  llvm::Type *PtrTy = LLVMCatchTy->getPointerTo(0); // addrspace 0 ok
+
+  // Check for a copy expression.  If we don't have a copy expression,
+  // that means a trivial copy is okay.
+  const Expr *copyExpr = CatchParam.getInit();
+  if (!copyExpr) {
+    llvm::Value *rawAdjustedExn = CallBeginCatch(CGF, Exn, true);
+    llvm::Value *adjustedExn = CGF.Builder.CreateBitCast(rawAdjustedExn, PtrTy);
+    CGF.EmitAggregateCopy(ParamAddr, adjustedExn, CatchType);
+    return;
+  }
+
+  // We have to call __cxa_get_exception_ptr to get the adjusted
+  // pointer before copying.
+  llvm::CallInst *rawAdjustedExn =
+    CGF.EmitNounwindRuntimeCall(getGetExceptionPtrFn(CGF.CGM), Exn);
+
+  // Cast that to the appropriate type.
+  llvm::Value *adjustedExn = CGF.Builder.CreateBitCast(rawAdjustedExn, PtrTy);
+
+  // The copy expression is defined in terms of an OpaqueValueExpr.
+  // Find it and map it to the adjusted expression.
+  CodeGenFunction::OpaqueValueMapping
+    opaque(CGF, OpaqueValueExpr::findInCopyConstruct(copyExpr),
+           CGF.MakeAddrLValue(adjustedExn, CatchParam.getType()));
+
+  // Call the copy ctor in a terminate scope.
+  CGF.EHStack.pushTerminate();
+
+  // Perform the copy construction.
+  CharUnits Alignment = CGF.getContext().getDeclAlign(&CatchParam);
+  CGF.EmitAggExpr(copyExpr,
+                  AggValueSlot::forAddr(ParamAddr, Alignment, Qualifiers(),
+                                        AggValueSlot::IsNotDestructed,
+                                        AggValueSlot::DoesNotNeedGCBarriers,
+                                        AggValueSlot::IsNotAliased));
+
+  // Leave the terminate scope.
+  CGF.EHStack.popTerminate();
+
+  // Undo the opaque value mapping.
+  opaque.pop();
+
+  // Finally we can call __cxa_begin_catch.
+  CallBeginCatch(CGF, Exn, true);
+}
+
+/// Begins a catch statement by initializing the catch variable and
+/// calling __cxa_begin_catch.
+void ItaniumCXXABI::emitBeginCatch(CodeGenFunction &CGF,
+                                   const CXXCatchStmt *S) {
+  // We have to be very careful with the ordering of cleanups here:
+  //   C++ [except.throw]p4:
+  //     The destruction [of the exception temporary] occurs
+  //     immediately after the destruction of the object declared in
+  //     the exception-declaration in the handler.
+  //
+  // So the precise ordering is:
+  //   1.  Construct catch variable.
+  //   2.  __cxa_begin_catch
+  //   3.  Enter __cxa_end_catch cleanup
+  //   4.  Enter dtor cleanup
+  //
+  // We do this by using a slightly abnormal initialization process.
+  // Delegation sequence:
+  //   - ExitCXXTryStmt opens a RunCleanupsScope
+  //     - EmitAutoVarAlloca creates the variable and debug info
+  //       - InitCatchParam initializes the variable from the exception
+  //       - CallBeginCatch calls __cxa_begin_catch
+  //       - CallBeginCatch enters the __cxa_end_catch cleanup
+  //     - EmitAutoVarCleanups enters the variable destructor cleanup
+  //   - EmitCXXTryStmt emits the code for the catch body
+  //   - EmitCXXTryStmt close the RunCleanupsScope
+
+  VarDecl *CatchParam = S->getExceptionDecl();
+  if (!CatchParam) {
+    llvm::Value *Exn = CGF.getExceptionFromSlot();
+    CallBeginCatch(CGF, Exn, true);
+    return;
+  }
+
+  // Emit the local.
+  CodeGenFunction::AutoVarEmission var = CGF.EmitAutoVarAlloca(*CatchParam);
+  InitCatchParam(CGF, *CatchParam, var.getObjectAddress(CGF), S->getLocStart());
+  CGF.EmitAutoVarCleanups(var);
+}
+
+/// Get or define the following function:
+///   void @__clang_call_terminate(i8* %exn) nounwind noreturn
+/// This code is used only in C++.
+static llvm::Constant *getClangCallTerminateFn(CodeGenModule &CGM) {
+  llvm::FunctionType *fnTy =
+    llvm::FunctionType::get(CGM.VoidTy, CGM.Int8PtrTy, /*IsVarArgs=*/false);
+  llvm::Constant *fnRef =
+    CGM.CreateRuntimeFunction(fnTy, "__clang_call_terminate");
+
+  llvm::Function *fn = dyn_cast<llvm::Function>(fnRef);
+  if (fn && fn->empty()) {
+    fn->setDoesNotThrow();
+    fn->setDoesNotReturn();
+
+    // What we really want is to massively penalize inlining without
+    // forbidding it completely.  The difference between that and
+    // 'noinline' is negligible.
+    fn->addFnAttr(llvm::Attribute::NoInline);
+
+    // Allow this function to be shared across translation units, but
+    // we don't want it to turn into an exported symbol.
+    fn->setLinkage(llvm::Function::LinkOnceODRLinkage);
+    fn->setVisibility(llvm::Function::HiddenVisibility);
+    if (CGM.supportsCOMDAT())
+      fn->setComdat(CGM.getModule().getOrInsertComdat(fn->getName()));
+
+    // Set up the function.
+    llvm::BasicBlock *entry =
+      llvm::BasicBlock::Create(CGM.getLLVMContext(), "", fn);
+    CGBuilderTy builder(entry);
+
+    // Pull the exception pointer out of the parameter list.
+    llvm::Value *exn = &*fn->arg_begin();
+
+    // Call __cxa_begin_catch(exn).
+    llvm::CallInst *catchCall = builder.CreateCall(getBeginCatchFn(CGM), exn);
+    catchCall->setDoesNotThrow();
+    catchCall->setCallingConv(CGM.getRuntimeCC());
+
+    // Call std::terminate().
+    llvm::CallInst *termCall = builder.CreateCall(CGM.getTerminateFn(), {});
+    termCall->setDoesNotThrow();
+    termCall->setDoesNotReturn();
+    termCall->setCallingConv(CGM.getRuntimeCC());
+
+    // std::terminate cannot return.
+    builder.CreateUnreachable();
+  }
+
+  return fnRef;
+}
+
+llvm::CallInst *
+ItaniumCXXABI::emitTerminateForUnexpectedException(CodeGenFunction &CGF,
+                                                   llvm::Value *Exn) {
+  // In C++, we want to call __cxa_begin_catch() before terminating.
+  if (Exn) {
+    assert(CGF.CGM.getLangOpts().CPlusPlus);
+    return CGF.EmitNounwindRuntimeCall(getClangCallTerminateFn(CGF.CGM), Exn);
   }
+  return CGF.EmitNounwindRuntimeCall(CGF.CGM.getTerminateFn());
 }
diff --git a/lib/CodeGen/MicrosoftCXXABI.cpp b/lib/CodeGen/MicrosoftCXXABI.cpp
index c067fab1e921..de30883c545f 100644
--- a/lib/CodeGen/MicrosoftCXXABI.cpp
+++ b/lib/CodeGen/MicrosoftCXXABI.cpp
@@ -17,12 +17,16 @@
 #include "CGCXXABI.h"
 #include "CGVTables.h"
 #include "CodeGenModule.h"
+#include "CodeGenTypes.h"
+#include "TargetInfo.h"
 #include "clang/AST/Decl.h"
 #include "clang/AST/DeclCXX.h"
+#include "clang/AST/StmtCXX.h"
 #include "clang/AST/VTableBuilder.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringSet.h"
 #include "llvm/IR/CallSite.h"
+#include "llvm/IR/Intrinsics.h"
 
 using namespace clang;
 using namespace CodeGen;
@@ -40,7 +44,8 @@ public:
   MicrosoftCXXABI(CodeGenModule &CGM)
       : CGCXXABI(CGM), BaseClassDescriptorType(nullptr),
         ClassHierarchyDescriptorType(nullptr),
-        CompleteObjectLocatorType(nullptr) {}
+        CompleteObjectLocatorType(nullptr), CatchableTypeType(nullptr),
+        ThrowInfoType(nullptr), CatchHandlerTypeType(nullptr) {}
 
   bool HasThisReturn(GlobalDecl GD) const override;
   bool hasMostDerivedReturn(GlobalDecl GD) const override;
@@ -71,11 +76,16 @@ public:
                                const CXXDestructorDecl *Dtor) override;
 
   void emitRethrow(CodeGenFunction &CGF, bool isNoReturn) override;
+  void emitThrow(CodeGenFunction &CGF, const CXXThrowExpr *E) override;
+
+  void emitBeginCatch(CodeGenFunction &CGF, const CXXCatchStmt *C) override;
 
   llvm::GlobalVariable *getMSCompleteObjectLocator(const CXXRecordDecl *RD,
                                                    const VPtrInfo *Info);
 
   llvm::Constant *getAddrOfRTTIDescriptor(QualType Ty) override;
+  llvm::Constant *
+  getAddrOfCXXCatchHandlerType(QualType Ty, QualType CatchHandlerType) override;
 
   bool shouldTypeidBeNullChecked(bool IsDeref, QualType SrcRecordTy) override;
   void EmitBadTypeidCall(CodeGenFunction &CGF) override;
@@ -225,7 +235,7 @@ public:
     assert(GD.getDtorType() == Dtor_Deleting &&
            "Only deleting destructor thunks are available in this ABI");
     CallArgs.add(RValue::get(getStructorImplicitParamValue(CGF)),
-                             CGM.getContext().IntTy);
+                 getContext().IntTy);
   }
 
   void emitVirtualInheritanceTables(const CXXRecordDecl *RD) override;
@@ -410,6 +420,9 @@ public:
     if (!isImageRelative())
       return PtrVal;
 
+    if (PtrVal->isNullValue())
+      return llvm::Constant::getNullValue(CGM.IntTy);
+
     llvm::Constant *ImageBaseAsInt =
         llvm::ConstantExpr::getPtrToInt(getImageBase(), CGM.IntPtrTy);
     llvm::Constant *PtrValAsInt =
@@ -467,6 +480,10 @@ private:
     return GetVBaseOffsetFromVBPtr(CGF, Base, VBPOffset, VBTOffset, VBPtr);
   }
 
+  std::pair<llvm::Value *, llvm::Value *>
+  performBaseAdjustment(CodeGenFunction &CGF, llvm::Value *Value,
+                        QualType SrcRecordTy);
+
   /// \brief Performs a full virtual base adjustment.  Used to dereference
   /// pointers to members of virtual bases.
   llvm::Value *AdjustVirtualBase(CodeGenFunction &CGF, const Expr *E,
@@ -479,7 +496,8 @@ private:
   llvm::Constant *EmitFullMemberPointer(llvm::Constant *FirstField,
                                         bool IsMemberFunction,
                                         const CXXRecordDecl *RD,
-                                        CharUnits NonVirtualBaseAdjustment);
+                                        CharUnits NonVirtualBaseAdjustment,
+                                        unsigned VBTableIndex);
 
   llvm::Constant *BuildMemberPointer(const CXXRecordDecl *RD,
                                      const CXXMethodDecl *MD,
@@ -556,6 +574,94 @@ public:
 
   void emitCXXStructor(const CXXMethodDecl *MD, StructorType Type) override;
 
+  llvm::StructType *getCatchHandlerTypeType() {
+    if (!CatchHandlerTypeType) {
+      llvm::Type *FieldTypes[] = {
+          CGM.IntTy,     // Flags
+          CGM.Int8PtrTy, // TypeDescriptor
+      };
+      CatchHandlerTypeType = llvm::StructType::create(
+          CGM.getLLVMContext(), FieldTypes, "eh.CatchHandlerType");
+    }
+    return CatchHandlerTypeType;
+  }
+
+  llvm::StructType *getCatchableTypeType() {
+    if (CatchableTypeType)
+      return CatchableTypeType;
+    llvm::Type *FieldTypes[] = {
+        CGM.IntTy,                           // Flags
+        getImageRelativeType(CGM.Int8PtrTy), // TypeDescriptor
+        CGM.IntTy,                           // NonVirtualAdjustment
+        CGM.IntTy,                           // OffsetToVBPtr
+        CGM.IntTy,                           // VBTableIndex
+        CGM.IntTy,                           // Size
+        getImageRelativeType(CGM.Int8PtrTy)  // CopyCtor
+    };
+    CatchableTypeType = llvm::StructType::create(
+        CGM.getLLVMContext(), FieldTypes, "eh.CatchableType");
+    return CatchableTypeType;
+  }
+
+  llvm::StructType *getCatchableTypeArrayType(uint32_t NumEntries) {
+    llvm::StructType *&CatchableTypeArrayType =
+        CatchableTypeArrayTypeMap[NumEntries];
+    if (CatchableTypeArrayType)
+      return CatchableTypeArrayType;
+
+    llvm::SmallString<23> CTATypeName("eh.CatchableTypeArray.");
+    CTATypeName += llvm::utostr(NumEntries);
+    llvm::Type *CTType =
+        getImageRelativeType(getCatchableTypeType()->getPointerTo());
+    llvm::Type *FieldTypes[] = {
+        CGM.IntTy,                               // NumEntries
+        llvm::ArrayType::get(CTType, NumEntries) // CatchableTypes
+    };
+    CatchableTypeArrayType =
+        llvm::StructType::create(CGM.getLLVMContext(), FieldTypes, CTATypeName);
+    return CatchableTypeArrayType;
+  }
+
+  llvm::StructType *getThrowInfoType() {
+    if (ThrowInfoType)
+      return ThrowInfoType;
+    llvm::Type *FieldTypes[] = {
+        CGM.IntTy,                           // Flags
+        getImageRelativeType(CGM.Int8PtrTy), // CleanupFn
+        getImageRelativeType(CGM.Int8PtrTy), // ForwardCompat
+        getImageRelativeType(CGM.Int8PtrTy)  // CatchableTypeArray
+    };
+    ThrowInfoType = llvm::StructType::create(CGM.getLLVMContext(), FieldTypes,
+                                             "eh.ThrowInfo");
+    return ThrowInfoType;
+  }
+
+  llvm::Constant *getThrowFn() {
+    // _CxxThrowException is passed an exception object and a ThrowInfo object
+    // which describes the exception.
+    llvm::Type *Args[] = {CGM.Int8PtrTy, getThrowInfoType()->getPointerTo()};
+    llvm::FunctionType *FTy =
+        llvm::FunctionType::get(CGM.VoidTy, Args, /*IsVarArgs=*/false);
+    auto *Fn = cast<llvm::Function>(
+        CGM.CreateRuntimeFunction(FTy, "_CxxThrowException"));
+    // _CxxThrowException is stdcall on 32-bit x86 platforms.
+    if (CGM.getTarget().getTriple().getArch() == llvm::Triple::x86)
+      Fn->setCallingConv(llvm::CallingConv::X86_StdCall);
+    return Fn;
+  }
+
+  llvm::Function *getAddrOfCXXCtorClosure(const CXXConstructorDecl *CD,
+                                          CXXCtorType CT);
+
+  llvm::Constant *getCatchableType(QualType T,
+                                   uint32_t NVOffset = 0,
+                                   int32_t VBPtrOffset = -1,
+                                   uint32_t VBIndex = 0);
+
+  llvm::GlobalVariable *getCatchableTypeArray(QualType T);
+
+  llvm::GlobalVariable *getThrowInfo(QualType T) override;
+
 private:
   typedef std::pair<const CXXRecordDecl *, CharUnits> VFTableIdTy;
   typedef llvm::DenseMap<VFTableIdTy, llvm::GlobalVariable *> VTablesMapTy;
@@ -582,11 +688,20 @@ private:
   /// Map from DeclContext to the current guard variable.  We assume that the
   /// AST is visited in source code order.
   llvm::DenseMap<const DeclContext *, GuardInfo> GuardVariableMap;
+  llvm::DenseMap<const DeclContext *, GuardInfo> ThreadLocalGuardVariableMap;
+  llvm::DenseMap<const DeclContext *, unsigned> ThreadSafeGuardNumMap;
 
   llvm::DenseMap<size_t, llvm::StructType *> TypeDescriptorTypeMap;
   llvm::StructType *BaseClassDescriptorType;
   llvm::StructType *ClassHierarchyDescriptorType;
   llvm::StructType *CompleteObjectLocatorType;
+
+  llvm::DenseMap<QualType, llvm::GlobalVariable *> CatchableTypeArrays;
+
+  llvm::StructType *CatchableTypeType;
+  llvm::DenseMap<uint32_t, llvm::StructType *> CatchableTypeArrayTypeMap;
+  llvm::StructType *ThrowInfoType;
+  llvm::StructType *CatchHandlerTypeType;
 };
 
 }
@@ -667,55 +782,73 @@ void MicrosoftCXXABI::emitVirtualObjectDelete(CodeGenFunction &CGF,
     CGF.EmitDeleteCall(DE->getOperatorDelete(), MDThis, ElementType);
 }
 
-static llvm::Function *getRethrowFn(CodeGenModule &CGM) {
-  // _CxxThrowException takes two pointer width arguments: a value and a context
-  // object which points to a TypeInfo object.
-  llvm::Type *ArgTypes[] = {CGM.Int8PtrTy, CGM.Int8PtrTy};
-  llvm::FunctionType *FTy =
-      llvm::FunctionType::get(CGM.VoidTy, ArgTypes, false);
-  auto *Fn = cast<llvm::Function>(
-      CGM.CreateRuntimeFunction(FTy, "_CxxThrowException"));
-  // _CxxThrowException is stdcall on 32-bit x86 platforms.
-  if (CGM.getTarget().getTriple().getArch() == llvm::Triple::x86)
-    Fn->setCallingConv(llvm::CallingConv::X86_StdCall);
-  return Fn;
-}
-
 void MicrosoftCXXABI::emitRethrow(CodeGenFunction &CGF, bool isNoReturn) {
-  llvm::Value *Args[] = {llvm::ConstantPointerNull::get(CGM.Int8PtrTy),
-                         llvm::ConstantPointerNull::get(CGM.Int8PtrTy)};
-  auto *Fn = getRethrowFn(CGM);
+  llvm::Value *Args[] = {
+      llvm::ConstantPointerNull::get(CGM.Int8PtrTy),
+      llvm::ConstantPointerNull::get(getThrowInfoType()->getPointerTo())};
+  auto *Fn = getThrowFn();
   if (isNoReturn)
     CGF.EmitNoreturnRuntimeCallOrInvoke(Fn, Args);
   else
     CGF.EmitRuntimeCallOrInvoke(Fn, Args);
 }
 
-/// \brief Gets the offset to the virtual base that contains the vfptr for
-/// MS-ABI polymorphic types.
-static llvm::Value *getPolymorphicOffset(CodeGenFunction &CGF,
-                                         const CXXRecordDecl *RD,
-                                         llvm::Value *Value) {
-  const ASTContext &Context = RD->getASTContext();
-  for (const CXXBaseSpecifier &Base : RD->vbases())
-    if (Context.getASTRecordLayout(Base.getType()->getAsCXXRecordDecl())
-            .hasExtendableVFPtr())
-      return CGF.CGM.getCXXABI().GetVirtualBaseClassOffset(
-          CGF, Value, RD, Base.getType()->getAsCXXRecordDecl());
-  llvm_unreachable("One of our vbases should be polymorphic.");
+namespace {
+struct CallEndCatchMSVC : EHScopeStack::Cleanup {
+  CallEndCatchMSVC() {}
+  void Emit(CodeGenFunction &CGF, Flags flags) override {
+    CGF.EmitNounwindRuntimeCall(
+        CGF.CGM.getIntrinsic(llvm::Intrinsic::eh_endcatch));
+  }
+};
 }
 
-static std::pair<llvm::Value *, llvm::Value *>
-performBaseAdjustment(CodeGenFunction &CGF, llvm::Value *Value,
-                      QualType SrcRecordTy) {
+void MicrosoftCXXABI::emitBeginCatch(CodeGenFunction &CGF,
+                                     const CXXCatchStmt *S) {
+  // In the MS ABI, the runtime handles the copy, and the catch handler is
+  // responsible for destruction.
+  VarDecl *CatchParam = S->getExceptionDecl();
+  llvm::Value *Exn = CGF.getExceptionFromSlot();
+  llvm::Function *BeginCatch =
+      CGF.CGM.getIntrinsic(llvm::Intrinsic::eh_begincatch);
+
+  // If this is a catch-all or the catch parameter is unnamed, we don't need to
+  // emit an alloca to the object.
+  if (!CatchParam || !CatchParam->getDeclName()) {
+    llvm::Value *Args[2] = {Exn, llvm::Constant::getNullValue(CGF.Int8PtrTy)};
+    CGF.EmitNounwindRuntimeCall(BeginCatch, Args);
+    CGF.EHStack.pushCleanup<CallEndCatchMSVC>(NormalCleanup);
+    return;
+  }
+
+  CodeGenFunction::AutoVarEmission var = CGF.EmitAutoVarAlloca(*CatchParam);
+  llvm::Value *ParamAddr =
+      CGF.Builder.CreateBitCast(var.getObjectAddress(CGF), CGF.Int8PtrTy);
+  llvm::Value *Args[2] = {Exn, ParamAddr};
+  CGF.EmitNounwindRuntimeCall(BeginCatch, Args);
+  CGF.EHStack.pushCleanup<CallEndCatchMSVC>(NormalCleanup);
+  CGF.EmitAutoVarCleanups(var);
+}
+
+std::pair<llvm::Value *, llvm::Value *>
+MicrosoftCXXABI::performBaseAdjustment(CodeGenFunction &CGF, llvm::Value *Value,
+                                       QualType SrcRecordTy) {
   Value = CGF.Builder.CreateBitCast(Value, CGF.Int8PtrTy);
   const CXXRecordDecl *SrcDecl = SrcRecordTy->getAsCXXRecordDecl();
+  const ASTContext &Context = getContext();
 
-  if (CGF.getContext().getASTRecordLayout(SrcDecl).hasExtendableVFPtr())
+  if (Context.getASTRecordLayout(SrcDecl).hasExtendableVFPtr())
     return std::make_pair(Value, llvm::ConstantInt::get(CGF.Int32Ty, 0));
 
   // Perform a base adjustment.
-  llvm::Value *Offset = getPolymorphicOffset(CGF, SrcDecl, Value);
+  const CXXBaseSpecifier *PolymorphicBase = std::find_if(
+      SrcDecl->vbases_begin(), SrcDecl->vbases_end(),
+      [&](const CXXBaseSpecifier &Base) {
+        const CXXRecordDecl *BaseDecl = Base.getType()->getAsCXXRecordDecl();
+        return Context.getASTRecordLayout(BaseDecl).hasExtendableVFPtr();
+      });
+  llvm::Value *Offset = GetVirtualBaseClassOffset(
+      CGF, Value, SrcDecl, PolymorphicBase->getType()->getAsCXXRecordDecl());
   Value = CGF.Builder.CreateInBoundsGEP(Value, Offset);
   Offset = CGF.Builder.CreateTrunc(Offset, CGF.Int32Ty);
   return std::make_pair(Value, Offset);
@@ -725,7 +858,7 @@ bool MicrosoftCXXABI::shouldTypeidBeNullChecked(bool IsDeref,
                                                 QualType SrcRecordTy) {
   const CXXRecordDecl *SrcDecl = SrcRecordTy->getAsCXXRecordDecl();
   return IsDeref &&
-         !CGM.getContext().getASTRecordLayout(SrcDecl).hasExtendableVFPtr();
+         !getContext().getASTRecordLayout(SrcDecl).hasExtendableVFPtr();
 }
 
 static llvm::CallSite emitRTtypeidCall(CodeGenFunction &CGF,
@@ -759,7 +892,7 @@ bool MicrosoftCXXABI::shouldDynamicCastCallBeNullChecked(bool SrcIsPtr,
                                                          QualType SrcRecordTy) {
   const CXXRecordDecl *SrcDecl = SrcRecordTy->getAsCXXRecordDecl();
   return SrcIsPtr &&
-         !CGM.getContext().getASTRecordLayout(SrcDecl).hasExtendableVFPtr();
+         !getContext().getASTRecordLayout(SrcDecl).hasExtendableVFPtr();
 }
 
 llvm::Value *MicrosoftCXXABI::EmitDynamicCastCall(
@@ -817,10 +950,11 @@ bool MicrosoftCXXABI::EmitBadCastCall(CodeGenFunction &CGF) {
 llvm::Value *MicrosoftCXXABI::GetVirtualBaseClassOffset(
     CodeGenFunction &CGF, llvm::Value *This, const CXXRecordDecl *ClassDecl,
     const CXXRecordDecl *BaseClassDecl) {
+  const ASTContext &Context = getContext();
   int64_t VBPtrChars =
-      getContext().getASTRecordLayout(ClassDecl).getVBPtrOffset().getQuantity();
+      Context.getASTRecordLayout(ClassDecl).getVBPtrOffset().getQuantity();
   llvm::Value *VBPtrOffset = llvm::ConstantInt::get(CGM.PtrDiffTy, VBPtrChars);
-  CharUnits IntSize = getContext().getTypeSizeInChars(getContext().IntTy);
+  CharUnits IntSize = Context.getTypeSizeInChars(Context.IntTy);
   CharUnits VBTableChars =
       IntSize *
       CGM.getMicrosoftVTableContext().getVBTableIndex(ClassDecl, BaseClassDecl);
@@ -947,30 +1081,52 @@ void MicrosoftCXXABI::initializeHiddenVirtualInheritanceMembers(
   }
 }
 
+static bool hasDefaultCXXMethodCC(ASTContext &Context,
+                                  const CXXMethodDecl *MD) {
+  CallingConv ExpectedCallingConv = Context.getDefaultCallingConvention(
+      /*IsVariadic=*/false, /*IsCXXMethod=*/true);
+  CallingConv ActualCallingConv =
+      MD->getType()->getAs<FunctionProtoType>()->getCallConv();
+  return ExpectedCallingConv == ActualCallingConv;
+}
+
 void MicrosoftCXXABI::EmitCXXConstructors(const CXXConstructorDecl *D) {
   // There's only one constructor type in this ABI.
   CGM.EmitGlobal(GlobalDecl(D, Ctor_Complete));
+
+  // Exported default constructors either have a simple call-site where they use
+  // the typical calling convention and have a single 'this' pointer for an
+  // argument -or- they get a wrapper function which appropriately thunks to the
+  // real default constructor.  This thunk is the default constructor closure.
+  if (D->hasAttr<DLLExportAttr>() && D->isDefaultConstructor())
+    if (!hasDefaultCXXMethodCC(getContext(), D) || D->getNumParams() != 0) {
+      llvm::Function *Fn = getAddrOfCXXCtorClosure(D, Ctor_DefaultClosure);
+      Fn->setLinkage(llvm::GlobalValue::WeakODRLinkage);
+      Fn->setDLLStorageClass(llvm::GlobalValue::DLLExportStorageClass);
+    }
 }
 
 void MicrosoftCXXABI::EmitVBPtrStores(CodeGenFunction &CGF,
                                       const CXXRecordDecl *RD) {
   llvm::Value *ThisInt8Ptr =
     CGF.Builder.CreateBitCast(getThisValue(CGF), CGM.Int8PtrTy, "this.int8");
-  const ASTRecordLayout &Layout = CGM.getContext().getASTRecordLayout(RD);
+  const ASTContext &Context = getContext();
+  const ASTRecordLayout &Layout = Context.getASTRecordLayout(RD);
 
   const VBTableGlobals &VBGlobals = enumerateVBTables(RD);
   for (unsigned I = 0, E = VBGlobals.VBTables->size(); I != E; ++I) {
     const VPtrInfo *VBT = (*VBGlobals.VBTables)[I];
     llvm::GlobalVariable *GV = VBGlobals.Globals[I];
     const ASTRecordLayout &SubobjectLayout =
-        CGM.getContext().getASTRecordLayout(VBT->BaseWithVPtr);
+        Context.getASTRecordLayout(VBT->BaseWithVPtr);
     CharUnits Offs = VBT->NonVirtualOffset;
     Offs += SubobjectLayout.getVBPtrOffset();
     if (VBT->getVBaseWithVPtr())
       Offs += Layout.getVBaseClassOffset(VBT->getVBaseWithVPtr());
     llvm::Value *VBPtr =
         CGF.Builder.CreateConstInBoundsGEP1_64(ThisInt8Ptr, Offs.getQuantity());
-    llvm::Value *GVPtr = CGF.Builder.CreateConstInBoundsGEP2_32(GV, 0, 0);
+    llvm::Value *GVPtr =
+        CGF.Builder.CreateConstInBoundsGEP2_32(GV->getValueType(), GV, 0, 0);
     VBPtr = CGF.Builder.CreateBitCast(VBPtr, GVPtr->getType()->getPointerTo(0),
                                       "vbptr." + VBT->ReusingBase->getName());
     CGF.Builder.CreateStore(GVPtr, VBPtr);
@@ -983,7 +1139,7 @@ MicrosoftCXXABI::buildStructorSignature(const CXXMethodDecl *MD, StructorType T,
   // TODO: 'for base' flag
   if (T == StructorType::Deleting) {
     // The scalar deleting destructor takes an implicit int parameter.
-    ArgTys.push_back(CGM.getContext().IntTy);
+    ArgTys.push_back(getContext().IntTy);
   }
   auto *CD = dyn_cast<CXXConstructorDecl>(MD);
   if (!CD)
@@ -996,9 +1152,9 @@ MicrosoftCXXABI::buildStructorSignature(const CXXMethodDecl *MD, StructorType T,
   const FunctionProtoType *FPT = CD->getType()->castAs<FunctionProtoType>();
   if (Class->getNumVBases()) {
     if (FPT->isVariadic())
-      ArgTys.insert(ArgTys.begin() + 1, CGM.getContext().IntTy);
+      ArgTys.insert(ArgTys.begin() + 1, getContext().IntTy);
     else
-      ArgTys.push_back(CGM.getContext().IntTy);
+      ArgTys.push_back(getContext().IntTy);
   }
 }
 
@@ -1038,7 +1194,7 @@ MicrosoftCXXABI::getVirtualFunctionPrologueThisAdjustment(GlobalDecl GD) {
 
   if (ML.VBase) {
     const ASTRecordLayout &DerivedLayout =
-        CGM.getContext().getASTRecordLayout(MD->getParent());
+        getContext().getASTRecordLayout(MD->getParent());
     Adjustment += DerivedLayout.getVBaseClassOffset(ML.VBase);
   }
 
@@ -1104,7 +1260,7 @@ llvm::Value *MicrosoftCXXABI::adjustThisArgumentForVirtualFunctionCall(
       // FIXME: Update the code that emits this adjustment in thunks prologues.
       This = CGF.Builder.CreateConstGEP1_32(This, StaticOffset.getQuantity());
     } else {
-      This = CGF.Builder.CreateConstInBoundsGEP1_32(This,
+      This = CGF.Builder.CreateConstInBoundsGEP1_32(CGF.Int8Ty, This,
                                                     StaticOffset.getQuantity());
     }
   }
@@ -1159,8 +1315,8 @@ llvm::Value *MicrosoftCXXABI::adjustThisParameterInVirtualFunctionPrologue(
 
   This = CGF.Builder.CreateBitCast(This, charPtrTy);
   assert(Adjustment.isPositive());
-  This =
-      CGF.Builder.CreateConstInBoundsGEP1_32(This, -Adjustment.getQuantity());
+  This = CGF.Builder.CreateConstInBoundsGEP1_32(CGF.Int8Ty, This,
+                                                -Adjustment.getQuantity());
   return CGF.Builder.CreateBitCast(This, thisTy);
 }
 
@@ -1280,7 +1436,7 @@ llvm::Value *MicrosoftCXXABI::getVTableAddressPointInStructor(
   llvm::GlobalValue *VTableAddressPoint = VFTablesMap[ID];
   if (!VTableAddressPoint) {
     assert(Base.getBase()->getNumVBases() &&
-           !CGM.getContext().getASTRecordLayout(Base.getBase()).hasOwnVFPtr());
+           !getContext().getASTRecordLayout(Base.getBase()).hasOwnVFPtr());
   }
   return VTableAddressPoint;
 }
@@ -1336,98 +1492,96 @@ llvm::GlobalVariable *MicrosoftCXXABI::getAddrOfVTable(const CXXRecordDecl *RD,
 #endif
   }
 
-  for (size_t J = 0, F = VFPtrs.size(); J != F; ++J) {
-    if (VFPtrs[J]->FullOffsetInMDC != VPtrOffset)
-      continue;
-    SmallString<256> VFTableName;
-    mangleVFTableName(getMangleContext(), RD, VFPtrs[J], VFTableName);
-    StringRef VTableName = VFTableName;
-
-    uint64_t NumVTableSlots =
-        VTContext.getVFTableLayout(RD, VFPtrs[J]->FullOffsetInMDC)
-            .getNumVTableComponents();
-    llvm::GlobalValue::LinkageTypes VTableLinkage =
-        llvm::GlobalValue::ExternalLinkage;
-    llvm::ArrayType *VTableType =
-        llvm::ArrayType::get(CGM.Int8PtrTy, NumVTableSlots);
-    if (getContext().getLangOpts().RTTIData) {
-      VTableLinkage = llvm::GlobalValue::PrivateLinkage;
-      VTableName = "";
-    }
-
-    VTable = CGM.getModule().getNamedGlobal(VFTableName);
-    if (!VTable) {
-      // Create a backing variable for the contents of VTable.  The VTable may
-      // or may not include space for a pointer to RTTI data.
-      llvm::GlobalValue *VFTable = VTable = new llvm::GlobalVariable(
-          CGM.getModule(), VTableType, /*isConstant=*/true, VTableLinkage,
-          /*Initializer=*/nullptr, VTableName);
-      VTable->setUnnamedAddr(true);
-
-      // Only insert a pointer into the VFTable for RTTI data if we are not
-      // importing it.  We never reference the RTTI data directly so there is no
-      // need to make room for it.
-      if (getContext().getLangOpts().RTTIData &&
-          !RD->hasAttr<DLLImportAttr>()) {
-        llvm::Value *GEPIndices[] = {llvm::ConstantInt::get(CGM.IntTy, 0),
-                                     llvm::ConstantInt::get(CGM.IntTy, 1)};
-        // Create a GEP which points just after the first entry in the VFTable,
-        // this should be the location of the first virtual method.
-        llvm::Constant *VTableGEP =
-            llvm::ConstantExpr::getInBoundsGetElementPtr(VTable, GEPIndices);
-        // The symbol for the VFTable is an alias to the GEP.  It is
-        // transparent, to other modules, what the nature of this symbol is; all
-        // that matters is that the alias be the address of the first virtual
-        // method.
-        VFTable = llvm::GlobalAlias::create(
-            cast<llvm::SequentialType>(VTableGEP->getType())->getElementType(),
-            /*AddressSpace=*/0, llvm::GlobalValue::ExternalLinkage,
-            VFTableName.str(), VTableGEP, &CGM.getModule());
-      } else {
-        // We don't need a GlobalAlias to be a symbol for the VTable if we won't
-        // be referencing any RTTI data.  The GlobalVariable will end up being
-        // an appropriate definition of the VFTable.
-        VTable->setName(VFTableName.str());
-      }
-
-      VFTable->setUnnamedAddr(true);
-      if (RD->hasAttr<DLLImportAttr>())
-        VFTable->setDLLStorageClass(llvm::GlobalValue::DLLImportStorageClass);
-      else if (RD->hasAttr<DLLExportAttr>())
-        VFTable->setDLLStorageClass(llvm::GlobalValue::DLLExportStorageClass);
-
-      llvm::GlobalValue::LinkageTypes VFTableLinkage = CGM.getVTableLinkage(RD);
-      if (VFTable != VTable) {
-        if (llvm::GlobalValue::isAvailableExternallyLinkage(VFTableLinkage)) {
-          // AvailableExternally implies that we grabbed the data from another
-          // executable.  No need to stick the alias in a Comdat.
-        } else if (llvm::GlobalValue::isInternalLinkage(VFTableLinkage) ||
-                   llvm::GlobalValue::isWeakODRLinkage(VFTableLinkage) ||
-                   llvm::GlobalValue::isLinkOnceODRLinkage(VFTableLinkage)) {
-          // The alias is going to be dropped into a Comdat, no need to make it
-          // weak.
-          if (!llvm::GlobalValue::isInternalLinkage(VFTableLinkage))
-            VFTableLinkage = llvm::GlobalValue::ExternalLinkage;
-          llvm::Comdat *C =
-              CGM.getModule().getOrInsertComdat(VFTable->getName());
-          // We must indicate which VFTable is larger to support linking between
-          // translation units which do and do not have RTTI data.  The largest
-          // VFTable contains the RTTI data; translation units which reference
-          // the smaller VFTable always reference it relative to the first
-          // virtual method.
-          C->setSelectionKind(llvm::Comdat::Largest);
-          VTable->setComdat(C);
-        } else {
-          llvm_unreachable("unexpected linkage for vftable!");
-        }
-      }
-      VFTable->setLinkage(VFTableLinkage);
-      CGM.setGlobalVisibility(VFTable, RD);
-      VFTablesMap[ID] = VFTable;
+  VPtrInfo *const *VFPtrI =
+      std::find_if(VFPtrs.begin(), VFPtrs.end(), [&](VPtrInfo *VPI) {
+        return VPI->FullOffsetInMDC == VPtrOffset;
+      });
+  if (VFPtrI == VFPtrs.end()) {
+    VFTablesMap[ID] = nullptr;
+    return nullptr;
+  }
+  VPtrInfo *VFPtr = *VFPtrI;
+
+  SmallString<256> VFTableName;
+  mangleVFTableName(getMangleContext(), RD, VFPtr, VFTableName);
+
+  llvm::GlobalValue::LinkageTypes VFTableLinkage = CGM.getVTableLinkage(RD);
+  bool VFTableComesFromAnotherTU =
+      llvm::GlobalValue::isAvailableExternallyLinkage(VFTableLinkage) ||
+      llvm::GlobalValue::isExternalLinkage(VFTableLinkage);
+  bool VTableAliasIsRequred =
+      !VFTableComesFromAnotherTU && getContext().getLangOpts().RTTIData;
+
+  if (llvm::GlobalValue *VFTable =
+          CGM.getModule().getNamedGlobal(VFTableName)) {
+    VFTablesMap[ID] = VFTable;
+    return VTableAliasIsRequred
+               ? cast<llvm::GlobalVariable>(
+                     cast<llvm::GlobalAlias>(VFTable)->getBaseObject())
+               : cast<llvm::GlobalVariable>(VFTable);
+  }
+
+  uint64_t NumVTableSlots =
+      VTContext.getVFTableLayout(RD, VFPtr->FullOffsetInMDC)
+          .getNumVTableComponents();
+  llvm::GlobalValue::LinkageTypes VTableLinkage =
+      VTableAliasIsRequred ? llvm::GlobalValue::PrivateLinkage : VFTableLinkage;
+
+  StringRef VTableName = VTableAliasIsRequred ? StringRef() : VFTableName.str();
+
+  llvm::ArrayType *VTableType =
+      llvm::ArrayType::get(CGM.Int8PtrTy, NumVTableSlots);
+
+  // Create a backing variable for the contents of VTable.  The VTable may
+  // or may not include space for a pointer to RTTI data.
+  llvm::GlobalValue *VFTable;
+  VTable = new llvm::GlobalVariable(CGM.getModule(), VTableType,
+                                    /*isConstant=*/true, VTableLinkage,
+                                    /*Initializer=*/nullptr, VTableName);
+  VTable->setUnnamedAddr(true);
+
+  llvm::Comdat *C = nullptr;
+  if (!VFTableComesFromAnotherTU &&
+      (llvm::GlobalValue::isWeakForLinker(VFTableLinkage) ||
+       (llvm::GlobalValue::isLocalLinkage(VFTableLinkage) &&
+        VTableAliasIsRequred)))
+    C = CGM.getModule().getOrInsertComdat(VFTableName.str());
+
+  // Only insert a pointer into the VFTable for RTTI data if we are not
+  // importing it.  We never reference the RTTI data directly so there is no
+  // need to make room for it.
+  if (VTableAliasIsRequred) {
+    llvm::Value *GEPIndices[] = {llvm::ConstantInt::get(CGM.IntTy, 0),
+                                 llvm::ConstantInt::get(CGM.IntTy, 1)};
+    // Create a GEP which points just after the first entry in the VFTable,
+    // this should be the location of the first virtual method.
+    llvm::Constant *VTableGEP = llvm::ConstantExpr::getInBoundsGetElementPtr(
+        VTable->getValueType(), VTable, GEPIndices);
+    if (llvm::GlobalValue::isWeakForLinker(VFTableLinkage)) {
+      VFTableLinkage = llvm::GlobalValue::ExternalLinkage;
+      if (C)
+        C->setSelectionKind(llvm::Comdat::Largest);
     }
-    break;
+    VFTable = llvm::GlobalAlias::create(
+        cast<llvm::PointerType>(VTableGEP->getType()), VFTableLinkage,
+        VFTableName.str(), VTableGEP, &CGM.getModule());
+    VFTable->setUnnamedAddr(true);
+  } else {
+    // We don't need a GlobalAlias to be a symbol for the VTable if we won't
+    // be referencing any RTTI data.
+    // The GlobalVariable will end up being an appropriate definition of the
+    // VFTable.
+    VFTable = VTable;
   }
+  if (C)
+    VTable->setComdat(C);
+
+  if (RD->hasAttr<DLLImportAttr>())
+    VFTable->setDLLStorageClass(llvm::GlobalValue::DLLImportStorageClass);
+  else if (RD->hasAttr<DLLExportAttr>())
+    VFTable->setDLLStorageClass(llvm::GlobalValue::DLLExportStorageClass);
 
+  VFTablesMap[ID] = VFTable;
   return VTable;
 }
 
@@ -1464,7 +1618,7 @@ llvm::Value *MicrosoftCXXABI::EmitVirtualDestructorCall(
   llvm::Type *Ty = CGF.CGM.getTypes().GetFunctionType(*FInfo);
   llvm::Value *Callee = getVirtualFunctionPointer(CGF, GD, This, Ty);
 
-  ASTContext &Context = CGF.getContext();
+  ASTContext &Context = getContext();
   llvm::Value *ImplicitParam = llvm::ConstantInt::get(
       llvm::IntegerType::getInt32Ty(CGF.getLLVMContext()),
       DtorType == Dtor_Deleting);
@@ -1530,6 +1684,8 @@ llvm::Function *MicrosoftCXXABI::EmitVirtualMemPtrThunk(
   ThunkFn->setLinkage(MD->isExternallyVisible()
                           ? llvm::GlobalValue::LinkOnceODRLinkage
                           : llvm::GlobalValue::InternalLinkage);
+  if (MD->isExternallyVisible())
+    ThunkFn->setComdat(CGM.getModule().getOrInsertComdat(ThunkFn->getName()));
 
   CGM.SetLLVMFunctionAttributes(MD, FnInfo, ThunkFn);
   CGM.SetLLVMFunctionAttributesForDefinition(MD, ThunkFn);
@@ -1576,7 +1732,8 @@ void MicrosoftCXXABI::emitVirtualInheritanceTables(const CXXRecordDecl *RD) {
   for (unsigned I = 0, E = VBGlobals.VBTables->size(); I != E; ++I) {
     const VPtrInfo *VBT = (*VBGlobals.VBTables)[I];
     llvm::GlobalVariable *GV = VBGlobals.Globals[I];
-    emitVBTableDefinition(*VBT, RD, GV);
+    if (GV->isDeclaration())
+      emitVBTableDefinition(*VBT, RD, GV);
   }
 }
 
@@ -1603,6 +1760,9 @@ MicrosoftCXXABI::getAddrOfVBTable(const VPtrInfo &VBT, const CXXRecordDecl *RD,
   else if (RD->hasAttr<DLLExportAttr>())
     GV->setDLLStorageClass(llvm::GlobalValue::DLLExportStorageClass);
 
+  if (!GV->hasExternalLinkage())
+    emitVBTableDefinition(VBT, RD, GV);
+
   return GV;
 }
 
@@ -1615,9 +1775,8 @@ void MicrosoftCXXABI::emitVBTableDefinition(const VPtrInfo &VBT,
          "should only emit vbtables for classes with vbtables");
 
   const ASTRecordLayout &BaseLayout =
-      CGM.getContext().getASTRecordLayout(VBT.BaseWithVPtr);
-  const ASTRecordLayout &DerivedLayout =
-    CGM.getContext().getASTRecordLayout(RD);
+      getContext().getASTRecordLayout(VBT.BaseWithVPtr);
+  const ASTRecordLayout &DerivedLayout = getContext().getASTRecordLayout(RD);
 
   SmallVector<llvm::Constant *, 4> Offsets(1 + ReusingBase->getNumVBases(),
                                            nullptr);
@@ -1651,9 +1810,6 @@ void MicrosoftCXXABI::emitVBTableDefinition(const VPtrInfo &VBT,
     llvm::ArrayType::get(CGM.IntTy, Offsets.size());
   llvm::Constant *Init = llvm::ConstantArray::get(VBTableType, Offsets);
   GV->setInitializer(Init);
-
-  // Set the right visibility.
-  CGM.setGlobalVisibility(GV, RD);
 }
 
 llvm::Value *MicrosoftCXXABI::performThisAdjustment(CodeGenFunction &CGF,
@@ -1709,8 +1865,8 @@ MicrosoftCXXABI::performReturnAdjustment(CodeGenFunction &CGF, llvm::Value *Ret,
 
   if (RA.Virtual.Microsoft.VBIndex) {
     assert(RA.Virtual.Microsoft.VBIndex > 0);
-    int32_t IntSize =
-        getContext().getTypeSizeInChars(getContext().IntTy).getQuantity();
+    const ASTContext &Context = getContext();
+    int32_t IntSize = Context.getTypeSizeInChars(Context.IntTy).getQuantity();
     llvm::Value *VBPtr;
     llvm::Value *VBaseOffset =
         GetVBaseOffsetFromVBPtr(CGF, V, RA.Virtual.Microsoft.VBPtrOffset,
@@ -1719,7 +1875,7 @@ MicrosoftCXXABI::performReturnAdjustment(CodeGenFunction &CGF, llvm::Value *Ret,
   }
 
   if (RA.NonVirtual)
-    V = CGF.Builder.CreateConstInBoundsGEP1_32(V, RA.NonVirtual);
+    V = CGF.Builder.CreateConstInBoundsGEP1_32(CGF.Int8Ty, V, RA.NonVirtual);
 
   // Cast back to the original type.
   return CGF.Builder.CreateBitCast(V, Ret->getType());
@@ -1835,18 +1991,10 @@ void MicrosoftCXXABI::EmitThreadLocalInitFuncs(
     llvm::Function *F = CXXThreadLocalInits[I];
 
     // If the GV is already in a comdat group, then we have to join it.
-    llvm::Comdat *C = GV->getComdat();
-
-    // LinkOnce and Weak linkage are lowered down to a single-member comdat
-    // group.
-    // Make an explicit group so we can join it.
-    if (!C && (GV->hasWeakLinkage() || GV->hasLinkOnceLinkage())) {
-      C = CGM.getModule().getOrInsertComdat(GV->getName());
-      GV->setComdat(C);
+    if (llvm::Comdat *C = GV->getComdat())
       AddToXDU(F)->setComdat(C);
-    } else {
+    else
       NonComdatInits.push_back(F);
-    }
   }
 
   if (!NonComdatInits.empty()) {
@@ -1868,6 +2016,81 @@ LValue MicrosoftCXXABI::EmitThreadLocalVarDeclLValue(CodeGenFunction &CGF,
   return LValue();
 }
 
+static llvm::GlobalVariable *getInitThreadEpochPtr(CodeGenModule &CGM) {
+  StringRef VarName("_Init_thread_epoch");
+  if (auto *GV = CGM.getModule().getNamedGlobal(VarName))
+    return GV;
+  auto *GV = new llvm::GlobalVariable(
+      CGM.getModule(), CGM.IntTy,
+      /*Constant=*/false, llvm::GlobalVariable::ExternalLinkage,
+      /*Initializer=*/nullptr, VarName,
+      /*InsertBefore=*/nullptr, llvm::GlobalVariable::GeneralDynamicTLSModel);
+  GV->setAlignment(CGM.getTarget().getIntAlign() / 8);
+  return GV;
+}
+
+static llvm::Constant *getInitThreadHeaderFn(CodeGenModule &CGM) {
+  llvm::FunctionType *FTy =
+      llvm::FunctionType::get(llvm::Type::getVoidTy(CGM.getLLVMContext()),
+                              CGM.IntTy->getPointerTo(), /*isVarArg=*/false);
+  return CGM.CreateRuntimeFunction(
+      FTy, "_Init_thread_header",
+      llvm::AttributeSet::get(CGM.getLLVMContext(),
+                              llvm::AttributeSet::FunctionIndex,
+                              llvm::Attribute::NoUnwind));
+}
+
+static llvm::Constant *getInitThreadFooterFn(CodeGenModule &CGM) {
+  llvm::FunctionType *FTy =
+      llvm::FunctionType::get(llvm::Type::getVoidTy(CGM.getLLVMContext()),
+                              CGM.IntTy->getPointerTo(), /*isVarArg=*/false);
+  return CGM.CreateRuntimeFunction(
+      FTy, "_Init_thread_footer",
+      llvm::AttributeSet::get(CGM.getLLVMContext(),
+                              llvm::AttributeSet::FunctionIndex,
+                              llvm::Attribute::NoUnwind));
+}
+
+static llvm::Constant *getInitThreadAbortFn(CodeGenModule &CGM) {
+  llvm::FunctionType *FTy =
+      llvm::FunctionType::get(llvm::Type::getVoidTy(CGM.getLLVMContext()),
+                              CGM.IntTy->getPointerTo(), /*isVarArg=*/false);
+  return CGM.CreateRuntimeFunction(
+      FTy, "_Init_thread_abort",
+      llvm::AttributeSet::get(CGM.getLLVMContext(),
+                              llvm::AttributeSet::FunctionIndex,
+                              llvm::Attribute::NoUnwind));
+}
+
+namespace {
+struct ResetGuardBit : EHScopeStack::Cleanup {
+  llvm::GlobalVariable *Guard;
+  unsigned GuardNum;
+  ResetGuardBit(llvm::GlobalVariable *Guard, unsigned GuardNum)
+      : Guard(Guard), GuardNum(GuardNum) {}
+
+  void Emit(CodeGenFunction &CGF, Flags flags) override {
+    // Reset the bit in the mask so that the static variable may be
+    // reinitialized.
+    CGBuilderTy &Builder = CGF.Builder;
+    llvm::LoadInst *LI = Builder.CreateLoad(Guard);
+    llvm::ConstantInt *Mask =
+        llvm::ConstantInt::get(CGF.IntTy, ~(1U << GuardNum));
+    Builder.CreateStore(Builder.CreateAnd(LI, Mask), Guard);
+  }
+};
+
+struct CallInitThreadAbort : EHScopeStack::Cleanup {
+  llvm::GlobalVariable *Guard;
+  CallInitThreadAbort(llvm::GlobalVariable *Guard) : Guard(Guard) {}
+
+  void Emit(CodeGenFunction &CGF, Flags flags) override {
+    // Calling _Init_thread_abort will reset the guard's state.
+    CGF.EmitNounwindRuntimeCall(getInitThreadAbortFn(CGF.CGM), Guard);
+  }
+};
+}
+
 void MicrosoftCXXABI::EmitGuardedInit(CodeGenFunction &CGF, const VarDecl &D,
                                       llvm::GlobalVariable *GV,
                                       bool PerformInit) {
@@ -1875,91 +2098,161 @@ void MicrosoftCXXABI::EmitGuardedInit(CodeGenFunction &CGF, const VarDecl &D,
   if (!D.isStaticLocal()) {
     assert(GV->hasWeakLinkage() || GV->hasLinkOnceLinkage());
     // GlobalOpt is allowed to discard the initializer, so use linkonce_odr.
-    CGF.CurFn->setLinkage(llvm::GlobalValue::LinkOnceODRLinkage);
+    llvm::Function *F = CGF.CurFn;
+    F->setLinkage(llvm::GlobalValue::LinkOnceODRLinkage);
+    F->setComdat(CGM.getModule().getOrInsertComdat(F->getName()));
     CGF.EmitCXXGlobalVarDeclInit(D, GV, PerformInit);
     return;
   }
 
-  // MSVC always uses an i32 bitfield to guard initialization, which is *not*
-  // threadsafe.  Since the user may be linking in inline functions compiled by
-  // cl.exe, there's no reason to provide a false sense of security by using
-  // critical sections here.
+  bool ThreadlocalStatic = D.getTLSKind();
+  bool ThreadsafeStatic = getContext().getLangOpts().ThreadsafeStatics;
 
-  if (D.getTLSKind())
-    CGM.ErrorUnsupported(&D, "dynamic TLS initialization");
+  // Thread-safe static variables which aren't thread-specific have a
+  // per-variable guard.
+  bool HasPerVariableGuard = ThreadsafeStatic && !ThreadlocalStatic;
 
   CGBuilderTy &Builder = CGF.Builder;
   llvm::IntegerType *GuardTy = CGF.Int32Ty;
   llvm::ConstantInt *Zero = llvm::ConstantInt::get(GuardTy, 0);
 
   // Get the guard variable for this function if we have one already.
-  GuardInfo *GI = &GuardVariableMap[D.getDeclContext()];
-
-  unsigned BitIndex;
-  if (D.isStaticLocal() && D.isExternallyVisible()) {
+  GuardInfo *GI = nullptr;
+  if (ThreadlocalStatic)
+    GI = &ThreadLocalGuardVariableMap[D.getDeclContext()];
+  else if (!ThreadsafeStatic)
+    GI = &GuardVariableMap[D.getDeclContext()];
+
+  llvm::GlobalVariable *GuardVar = GI ? GI->Guard : nullptr;
+  unsigned GuardNum;
+  if (D.isExternallyVisible()) {
     // Externally visible variables have to be numbered in Sema to properly
     // handle unreachable VarDecls.
-    BitIndex = getContext().getStaticLocalNumber(&D);
-    assert(BitIndex > 0);
-    BitIndex--;
+    GuardNum = getContext().getStaticLocalNumber(&D);
+    assert(GuardNum > 0);
+    GuardNum--;
+  } else if (HasPerVariableGuard) {
+    GuardNum = ThreadSafeGuardNumMap[D.getDeclContext()]++;
   } else {
     // Non-externally visible variables are numbered here in CodeGen.
-    BitIndex = GI->BitIndex++;
+    GuardNum = GI->BitIndex++;
   }
 
-  if (BitIndex >= 32) {
+  if (!HasPerVariableGuard && GuardNum >= 32) {
     if (D.isExternallyVisible())
       ErrorUnsupportedABI(CGF, "more than 32 guarded initializations");
-    BitIndex %= 32;
-    GI->Guard = nullptr;
+    GuardNum %= 32;
+    GuardVar = nullptr;
   }
 
-  // Lazily create the i32 bitfield for this function.
-  if (!GI->Guard) {
+  if (!GuardVar) {
     // Mangle the name for the guard.
     SmallString<256> GuardName;
     {
       llvm::raw_svector_ostream Out(GuardName);
-      getMangleContext().mangleStaticGuardVariable(&D, Out);
+      if (HasPerVariableGuard)
+        getMangleContext().mangleThreadSafeStaticGuardVariable(&D, GuardNum,
+                                                               Out);
+      else
+        getMangleContext().mangleStaticGuardVariable(&D, Out);
       Out.flush();
     }
 
     // Create the guard variable with a zero-initializer. Just absorb linkage,
     // visibility and dll storage class from the guarded variable.
-    GI->Guard =
-        new llvm::GlobalVariable(CGM.getModule(), GuardTy, false,
+    GuardVar =
+        new llvm::GlobalVariable(CGM.getModule(), GuardTy, /*isConstant=*/false,
                                  GV->getLinkage(), Zero, GuardName.str());
-    GI->Guard->setVisibility(GV->getVisibility());
-    GI->Guard->setDLLStorageClass(GV->getDLLStorageClass());
-  } else {
-    assert(GI->Guard->getLinkage() == GV->getLinkage() &&
-           "static local from the same function had different linkage");
-  }
-
-  // Pseudo code for the test:
-  // if (!(GuardVar & MyGuardBit)) {
-  //   GuardVar |= MyGuardBit;
-  //   ... initialize the object ...;
-  // }
-
-  // Test our bit from the guard variable.
-  llvm::ConstantInt *Bit = llvm::ConstantInt::get(GuardTy, 1U << BitIndex);
-  llvm::LoadInst *LI = Builder.CreateLoad(GI->Guard);
-  llvm::Value *IsInitialized =
-      Builder.CreateICmpNE(Builder.CreateAnd(LI, Bit), Zero);
-  llvm::BasicBlock *InitBlock = CGF.createBasicBlock("init");
-  llvm::BasicBlock *EndBlock = CGF.createBasicBlock("init.end");
-  Builder.CreateCondBr(IsInitialized, EndBlock, InitBlock);
+    GuardVar->setVisibility(GV->getVisibility());
+    GuardVar->setDLLStorageClass(GV->getDLLStorageClass());
+    if (GuardVar->isWeakForLinker())
+      GuardVar->setComdat(
+          CGM.getModule().getOrInsertComdat(GuardVar->getName()));
+    if (D.getTLSKind())
+      GuardVar->setThreadLocal(true);
+    if (GI && !HasPerVariableGuard)
+      GI->Guard = GuardVar;
+  }
+
+  assert(GuardVar->getLinkage() == GV->getLinkage() &&
+         "static local from the same function had different linkage");
+
+  if (!HasPerVariableGuard) {
+    // Pseudo code for the test:
+    // if (!(GuardVar & MyGuardBit)) {
+    //   GuardVar |= MyGuardBit;
+    //   ... initialize the object ...;
+    // }
+
+    // Test our bit from the guard variable.
+    llvm::ConstantInt *Bit = llvm::ConstantInt::get(GuardTy, 1U << GuardNum);
+    llvm::LoadInst *LI = Builder.CreateLoad(GuardVar);
+    llvm::Value *IsInitialized =
+        Builder.CreateICmpNE(Builder.CreateAnd(LI, Bit), Zero);
+    llvm::BasicBlock *InitBlock = CGF.createBasicBlock("init");
+    llvm::BasicBlock *EndBlock = CGF.createBasicBlock("init.end");
+    Builder.CreateCondBr(IsInitialized, EndBlock, InitBlock);
+
+    // Set our bit in the guard variable and emit the initializer and add a global
+    // destructor if appropriate.
+    CGF.EmitBlock(InitBlock);
+    Builder.CreateStore(Builder.CreateOr(LI, Bit), GuardVar);
+    CGF.EHStack.pushCleanup<ResetGuardBit>(EHCleanup, GuardVar, GuardNum);
+    CGF.EmitCXXGlobalVarDeclInit(D, GV, PerformInit);
+    CGF.PopCleanupBlock();
+    Builder.CreateBr(EndBlock);
 
-  // Set our bit in the guard variable and emit the initializer and add a global
-  // destructor if appropriate.
-  CGF.EmitBlock(InitBlock);
-  Builder.CreateStore(Builder.CreateOr(LI, Bit), GI->Guard);
-  CGF.EmitCXXGlobalVarDeclInit(D, GV, PerformInit);
-  Builder.CreateBr(EndBlock);
+    // Continue.
+    CGF.EmitBlock(EndBlock);
+  } else {
+    // Pseudo code for the test:
+    // if (TSS > _Init_thread_epoch) {
+    //   _Init_thread_header(&TSS);
+    //   if (TSS == -1) {
+    //     ... initialize the object ...;
+    //     _Init_thread_footer(&TSS);
+    //   }
+    // }
+    //
+    // The algorithm is almost identical to what can be found in the appendix
+    // found in N2325.
+
+    unsigned IntAlign = CGM.getTarget().getIntAlign() / 8;
+
+    // This BasicBLock determines whether or not we have any work to do.
+    llvm::LoadInst *FirstGuardLoad =
+        Builder.CreateAlignedLoad(GuardVar, IntAlign);
+    FirstGuardLoad->setOrdering(llvm::AtomicOrdering::Unordered);
+    llvm::LoadInst *InitThreadEpoch =
+        Builder.CreateLoad(getInitThreadEpochPtr(CGM));
+    llvm::Value *IsUninitialized =
+        Builder.CreateICmpSGT(FirstGuardLoad, InitThreadEpoch);
+    llvm::BasicBlock *AttemptInitBlock = CGF.createBasicBlock("init.attempt");
+    llvm::BasicBlock *EndBlock = CGF.createBasicBlock("init.end");
+    Builder.CreateCondBr(IsUninitialized, AttemptInitBlock, EndBlock);
+
+    // This BasicBlock attempts to determine whether or not this thread is
+    // responsible for doing the initialization.
+    CGF.EmitBlock(AttemptInitBlock);
+    CGF.EmitNounwindRuntimeCall(getInitThreadHeaderFn(CGM), GuardVar);
+    llvm::LoadInst *SecondGuardLoad =
+        Builder.CreateAlignedLoad(GuardVar, IntAlign);
+    SecondGuardLoad->setOrdering(llvm::AtomicOrdering::Unordered);
+    llvm::Value *ShouldDoInit =
+        Builder.CreateICmpEQ(SecondGuardLoad, getAllOnesInt());
+    llvm::BasicBlock *InitBlock = CGF.createBasicBlock("init");
+    Builder.CreateCondBr(ShouldDoInit, InitBlock, EndBlock);
+
+    // Ok, we ended up getting selected as the initializing thread.
+    CGF.EmitBlock(InitBlock);
+    CGF.EHStack.pushCleanup<CallInitThreadAbort>(EHCleanup, GuardVar);
+    CGF.EmitCXXGlobalVarDeclInit(D, GV, PerformInit);
+    CGF.PopCleanupBlock();
+    CGF.EmitNounwindRuntimeCall(getInitThreadFooterFn(CGM), GuardVar);
+    Builder.CreateBr(EndBlock);
 
-  // Continue.
-  CGF.EmitBlock(EndBlock);
+    CGF.EmitBlock(EndBlock);
+  }
 }
 
 bool MicrosoftCXXABI::isZeroInitializable(const MemberPointerType *MPT) {
@@ -2040,8 +2333,8 @@ llvm::Constant *
 MicrosoftCXXABI::EmitFullMemberPointer(llvm::Constant *FirstField,
                                        bool IsMemberFunction,
                                        const CXXRecordDecl *RD,
-                                       CharUnits NonVirtualBaseAdjustment)
-{
+                                       CharUnits NonVirtualBaseAdjustment,
+                                       unsigned VBTableIndex) {
   MSInheritanceAttr::Spelling Inheritance = RD->getMSInheritanceModel();
 
   // Single inheritance class member pointer are represented as scalars instead
@@ -2065,7 +2358,7 @@ MicrosoftCXXABI::EmitFullMemberPointer(llvm::Constant *FirstField,
 
   // The rest of the fields are adjusted by conversions to a more derived class.
   if (MSInheritanceAttr::hasVBTableOffsetField(Inheritance))
-    fields.push_back(getZeroInt());
+    fields.push_back(llvm::ConstantInt::get(CGM.IntTy, VBTableIndex));
 
   return llvm::ConstantStruct::getAnon(fields);
 }
@@ -2077,7 +2370,7 @@ MicrosoftCXXABI::EmitMemberDataPointer(const MemberPointerType *MPT,
   llvm::Constant *FirstField =
     llvm::ConstantInt::get(CGM.IntTy, offset.getQuantity());
   return EmitFullMemberPointer(FirstField, /*IsMemberFunction=*/false, RD,
-                               CharUnits::Zero());
+                               CharUnits::Zero(), /*VBTableIndex=*/0);
 }
 
 llvm::Constant *MicrosoftCXXABI::EmitMemberPointer(const CXXMethodDecl *MD) {
@@ -2113,6 +2406,7 @@ MicrosoftCXXABI::BuildMemberPointer(const CXXRecordDecl *RD,
   RD = RD->getMostRecentDecl();
   CodeGenTypes &Types = CGM.getTypes();
 
+  unsigned VBTableIndex = 0;
   llvm::Constant *FirstField;
   const FunctionProtoType *FPT = MD->getType()->castAs<FunctionProtoType>();
   if (!MD->isVirtual()) {
@@ -2129,8 +2423,6 @@ MicrosoftCXXABI::BuildMemberPointer(const CXXRecordDecl *RD,
     FirstField = CGM.GetAddrOfFunction(MD, Ty);
     FirstField = llvm::ConstantExpr::getBitCast(FirstField, CGM.VoidPtrTy);
   } else {
-    MicrosoftVTableContext::MethodVFTableLocation ML =
-        CGM.getMicrosoftVTableContext().getMethodVFTableLocation(MD);
     if (!CGM.getTypes().isFuncTypeConvertible(
             MD->getType()->castAs<FunctionType>())) {
       CGM.ErrorUnsupported(MD, "pointer to virtual member function with "
@@ -2139,21 +2431,22 @@ MicrosoftCXXABI::BuildMemberPointer(const CXXRecordDecl *RD,
     } else if (FPT->getCallConv() == CC_X86FastCall) {
       CGM.ErrorUnsupported(MD, "pointer to fastcall virtual member function");
       FirstField = llvm::Constant::getNullValue(CGM.VoidPtrTy);
-    } else if (ML.VBase) {
-      CGM.ErrorUnsupported(MD, "pointer to virtual member function overriding "
-                               "member function in virtual base class");
-      FirstField = llvm::Constant::getNullValue(CGM.VoidPtrTy);
     } else {
+      auto &VTableContext = CGM.getMicrosoftVTableContext();
+      MicrosoftVTableContext::MethodVFTableLocation ML =
+          VTableContext.getMethodVFTableLocation(MD);
       llvm::Function *Thunk = EmitVirtualMemPtrThunk(MD, ML);
       FirstField = llvm::ConstantExpr::getBitCast(Thunk, CGM.VoidPtrTy);
       // Include the vfptr adjustment if the method is in a non-primary vftable.
       NonVirtualBaseAdjustment += ML.VFPtrOffset;
+      if (ML.VBase)
+        VBTableIndex = VTableContext.getVBTableIndex(RD, ML.VBase) * 4;
     }
   }
 
   // The rest of the fields are common with data member pointers.
   return EmitFullMemberPointer(FirstField, /*IsMemberFunction=*/true, RD,
-                               NonVirtualBaseAdjustment);
+                               NonVirtualBaseAdjustment, VBTableIndex);
 }
 
 /// Member pointers are the same if they're either bitwise identical *or* both
@@ -2867,7 +3160,11 @@ llvm::GlobalVariable *MSRTTIBuilder::getClassHierarchyDescriptor() {
   auto Type = ABI.getClassHierarchyDescriptorType();
   auto CHD = new llvm::GlobalVariable(Module, Type, /*Constant=*/true, Linkage,
                                       /*Initializer=*/nullptr,
-                                      MangledName.c_str());
+                                      StringRef(MangledName));
+  if (CHD->isWeakForLinker())
+    CHD->setComdat(CGM.getModule().getOrInsertComdat(CHD->getName()));
+
+  auto *Bases = getBaseClassArray(Classes);
 
   // Initialize the base class ClassHierarchyDescriptor.
   llvm::Constant *Fields[] = {
@@ -2875,7 +3172,7 @@ llvm::GlobalVariable *MSRTTIBuilder::getClassHierarchyDescriptor() {
       llvm::ConstantInt::get(CGM.IntTy, Flags),
       llvm::ConstantInt::get(CGM.IntTy, Classes.size()),
       ABI.getImageRelativeConstant(llvm::ConstantExpr::getInBoundsGetElementPtr(
-          getBaseClassArray(Classes),
+          Bases->getValueType(), Bases,
           llvm::ArrayRef<llvm::Value *>(GEPIndices))),
   };
   CHD->setInitializer(llvm::ConstantStruct::get(Type, Fields));
@@ -2898,9 +3195,12 @@ MSRTTIBuilder::getBaseClassArray(SmallVectorImpl<MSRTTIClass> &Classes) {
   llvm::Type *PtrType = ABI.getImageRelativeType(
       ABI.getBaseClassDescriptorType()->getPointerTo());
   auto *ArrType = llvm::ArrayType::get(PtrType, Classes.size() + 1);
-  auto *BCA = new llvm::GlobalVariable(
-      Module, ArrType,
-      /*Constant=*/true, Linkage, /*Initializer=*/nullptr, MangledName.c_str());
+  auto *BCA =
+      new llvm::GlobalVariable(Module, ArrType,
+                               /*Constant=*/true, Linkage,
+                               /*Initializer=*/nullptr, StringRef(MangledName));
+  if (BCA->isWeakForLinker())
+    BCA->setComdat(CGM.getModule().getOrInsertComdat(BCA->getName()));
 
   // Initialize the BaseClassArray.
   SmallVector<llvm::Constant *, 8> BaseClassArrayData;
@@ -2938,9 +3238,11 @@ MSRTTIBuilder::getBaseClassDescriptor(const MSRTTIClass &Class) {
 
   // Forward-declare the base class descriptor.
   auto Type = ABI.getBaseClassDescriptorType();
-  auto BCD = new llvm::GlobalVariable(Module, Type, /*Constant=*/true, Linkage,
-                                      /*Initializer=*/nullptr,
-                                      MangledName.c_str());
+  auto BCD =
+      new llvm::GlobalVariable(Module, Type, /*Constant=*/true, Linkage,
+                               /*Initializer=*/nullptr, StringRef(MangledName));
+  if (BCD->isWeakForLinker())
+    BCD->setComdat(CGM.getModule().getOrInsertComdat(BCD->getName()));
 
   // Initialize the BaseClassDescriptor.
   llvm::Constant *Fields[] = {
@@ -2984,7 +3286,7 @@ MSRTTIBuilder::getCompleteObjectLocator(const VPtrInfo *Info) {
   // Forward-declare the complete object locator.
   llvm::StructType *Type = ABI.getCompleteObjectLocatorType();
   auto COL = new llvm::GlobalVariable(Module, Type, /*Constant=*/true, Linkage,
-    /*Initializer=*/nullptr, MangledName.c_str());
+    /*Initializer=*/nullptr, StringRef(MangledName));
 
   // Initialize the CompleteObjectLocator.
   llvm::Constant *Fields[] = {
@@ -3000,15 +3302,91 @@ MSRTTIBuilder::getCompleteObjectLocator(const VPtrInfo *Info) {
   if (!ABI.isImageRelative())
     FieldsRef = FieldsRef.drop_back();
   COL->setInitializer(llvm::ConstantStruct::get(Type, FieldsRef));
+  if (COL->isWeakForLinker())
+    COL->setComdat(CGM.getModule().getOrInsertComdat(COL->getName()));
   return COL;
 }
 
+static QualType decomposeTypeForEH(ASTContext &Context, QualType T,
+                                   bool &IsConst, bool &IsVolatile) {
+  T = Context.getExceptionObjectType(T);
+
+  // C++14 [except.handle]p3:
+  //   A handler is a match for an exception object of type E if [...]
+  //     - the handler is of type cv T or const T& where T is a pointer type and
+  //       E is a pointer type that can be converted to T by [...]
+  //         - a qualification conversion
+  IsConst = false;
+  IsVolatile = false;
+  QualType PointeeType = T->getPointeeType();
+  if (!PointeeType.isNull()) {
+    IsConst = PointeeType.isConstQualified();
+    IsVolatile = PointeeType.isVolatileQualified();
+  }
+
+  // Member pointer types like "const int A::*" are represented by having RTTI
+  // for "int A::*" and separately storing the const qualifier.
+  if (const auto *MPTy = T->getAs<MemberPointerType>())
+    T = Context.getMemberPointerType(PointeeType.getUnqualifiedType(),
+                                     MPTy->getClass());
+
+  // Pointer types like "const int * const *" are represented by having RTTI
+  // for "const int **" and separately storing the const qualifier.
+  if (T->isPointerType())
+    T = Context.getPointerType(PointeeType.getUnqualifiedType());
+
+  return T;
+}
+
+llvm::Constant *
+MicrosoftCXXABI::getAddrOfCXXCatchHandlerType(QualType Type,
+                                              QualType CatchHandlerType) {
+  // TypeDescriptors for exceptions never have qualified pointer types,
+  // qualifiers are stored seperately in order to support qualification
+  // conversions.
+  bool IsConst, IsVolatile;
+  Type = decomposeTypeForEH(getContext(), Type, IsConst, IsVolatile);
+
+  bool IsReference = CatchHandlerType->isReferenceType();
+
+  uint32_t Flags = 0;
+  if (IsConst)
+    Flags |= 1;
+  if (IsVolatile)
+    Flags |= 2;
+  if (IsReference)
+    Flags |= 8;
+
+  SmallString<256> MangledName;
+  {
+    llvm::raw_svector_ostream Out(MangledName);
+    getMangleContext().mangleCXXCatchHandlerType(Type, Flags, Out);
+  }
+
+  if (llvm::GlobalVariable *GV = CGM.getModule().getNamedGlobal(MangledName))
+    return llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy);
+
+  llvm::Constant *Fields[] = {
+      llvm::ConstantInt::get(CGM.IntTy, Flags), // Flags
+      getAddrOfRTTIDescriptor(Type),            // TypeDescriptor
+  };
+  llvm::StructType *CatchHandlerTypeType = getCatchHandlerTypeType();
+  auto *Var = new llvm::GlobalVariable(
+      CGM.getModule(), CatchHandlerTypeType, /*Constant=*/true,
+      llvm::GlobalValue::PrivateLinkage,
+      llvm::ConstantStruct::get(CatchHandlerTypeType, Fields),
+      StringRef(MangledName));
+  Var->setUnnamedAddr(true);
+  Var->setSection("llvm.metadata");
+  return Var;
+}
+
 /// \brief Gets a TypeDescriptor.  Returns a llvm::Constant * rather than a
 /// llvm::GlobalVariable * because different type descriptors have different
 /// types, and need to be abstracted.  They are abstracting by casting the
 /// address to an Int8PtrTy.
 llvm::Constant *MicrosoftCXXABI::getAddrOfRTTIDescriptor(QualType Type) {
-  SmallString<256> MangledName, TypeInfoString;
+  SmallString<256> MangledName;
   {
     llvm::raw_svector_ostream Out(MangledName);
     getMangleContext().mangleCXXRTTI(Type, Out);
@@ -3019,6 +3397,7 @@ llvm::Constant *MicrosoftCXXABI::getAddrOfRTTIDescriptor(QualType Type) {
     return llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy);
 
   // Compute the fields for the TypeDescriptor.
+  SmallString<256> TypeInfoString;
   {
     llvm::raw_svector_ostream Out(TypeInfoString);
     getMangleContext().mangleCXXRTTIName(Type, Out);
@@ -3031,13 +3410,14 @@ llvm::Constant *MicrosoftCXXABI::getAddrOfRTTIDescriptor(QualType Type) {
     llvm::ConstantDataArray::getString(CGM.getLLVMContext(), TypeInfoString)};
   llvm::StructType *TypeDescriptorType =
       getTypeDescriptorType(TypeInfoString);
-  return llvm::ConstantExpr::getBitCast(
-      new llvm::GlobalVariable(
-          CGM.getModule(), TypeDescriptorType, /*Constant=*/false,
-          getLinkageForRTTI(Type),
-          llvm::ConstantStruct::get(TypeDescriptorType, Fields),
-          MangledName.c_str()),
-      CGM.Int8PtrTy);
+  auto *Var = new llvm::GlobalVariable(
+      CGM.getModule(), TypeDescriptorType, /*Constant=*/false,
+      getLinkageForRTTI(Type),
+      llvm::ConstantStruct::get(TypeDescriptorType, Fields),
+      StringRef(MangledName));
+  if (Var->isWeakForLinker())
+    Var->setComdat(CGM.getModule().getOrInsertComdat(Var->getName()));
+  return llvm::ConstantExpr::getBitCast(Var, CGM.Int8PtrTy);
 }
 
 /// \brief Gets or a creates a Microsoft CompleteObjectLocator.
@@ -3051,7 +3431,8 @@ static void emitCXXConstructor(CodeGenModule &CGM,
                                const CXXConstructorDecl *ctor,
                                StructorType ctorType) {
   // There are no constructor variants, always emit the complete destructor.
-  CGM.codegenCXXStructor(ctor, StructorType::Complete);
+  llvm::Function *Fn = CGM.codegenCXXStructor(ctor, StructorType::Complete);
+  CGM.maybeSetTrivialComdat(*ctor, *Fn);
 }
 
 static void emitCXXDestructor(CodeGenModule &CGM, const CXXDestructorDecl *dtor,
@@ -3077,7 +3458,9 @@ static void emitCXXDestructor(CodeGenModule &CGM, const CXXDestructorDecl *dtor,
   if (dtorType == StructorType::Base && !CGM.TryEmitBaseDestructorAsAlias(dtor))
     return;
 
-  CGM.codegenCXXStructor(dtor, dtorType);
+  llvm::Function *Fn = CGM.codegenCXXStructor(dtor, dtorType);
+  if (Fn->isWeakForLinker())
+    Fn->setComdat(CGM.getModule().getOrInsertComdat(Fn->getName()));
 }
 
 void MicrosoftCXXABI::emitCXXStructor(const CXXMethodDecl *MD,
@@ -3088,3 +3471,400 @@ void MicrosoftCXXABI::emitCXXStructor(const CXXMethodDecl *MD,
   }
   emitCXXDestructor(CGM, cast<CXXDestructorDecl>(MD), Type);
 }
+
+llvm::Function *
+MicrosoftCXXABI::getAddrOfCXXCtorClosure(const CXXConstructorDecl *CD,
+                                         CXXCtorType CT) {
+  assert(CT == Ctor_CopyingClosure || CT == Ctor_DefaultClosure);
+
+  // Calculate the mangled name.
+  SmallString<256> ThunkName;
+  llvm::raw_svector_ostream Out(ThunkName);
+  getMangleContext().mangleCXXCtor(CD, CT, Out);
+  Out.flush();
+
+  // If the thunk has been generated previously, just return it.
+  if (llvm::GlobalValue *GV = CGM.getModule().getNamedValue(ThunkName))
+    return cast<llvm::Function>(GV);
+
+  // Create the llvm::Function.
+  const CGFunctionInfo &FnInfo = CGM.getTypes().arrangeMSCtorClosure(CD, CT);
+  llvm::FunctionType *ThunkTy = CGM.getTypes().GetFunctionType(FnInfo);
+  const CXXRecordDecl *RD = CD->getParent();
+  QualType RecordTy = getContext().getRecordType(RD);
+  llvm::Function *ThunkFn = llvm::Function::Create(
+      ThunkTy, getLinkageForRTTI(RecordTy), ThunkName.str(), &CGM.getModule());
+  ThunkFn->setCallingConv(static_cast<llvm::CallingConv::ID>(
+      FnInfo.getEffectiveCallingConvention()));
+  bool IsCopy = CT == Ctor_CopyingClosure;
+
+  // Start codegen.
+  CodeGenFunction CGF(CGM);
+  CGF.CurGD = GlobalDecl(CD, Ctor_Complete);
+
+  // Build FunctionArgs.
+  FunctionArgList FunctionArgs;
+
+  // A constructor always starts with a 'this' pointer as its first argument.
+  buildThisParam(CGF, FunctionArgs);
+
+  // Following the 'this' pointer is a reference to the source object that we
+  // are copying from.
+  ImplicitParamDecl SrcParam(
+      getContext(), nullptr, SourceLocation(), &getContext().Idents.get("src"),
+      getContext().getLValueReferenceType(RecordTy,
+                                          /*SpelledAsLValue=*/true));
+  if (IsCopy)
+    FunctionArgs.push_back(&SrcParam);
+
+  // Constructors for classes which utilize virtual bases have an additional
+  // parameter which indicates whether or not it is being delegated to by a more
+  // derived constructor.
+  ImplicitParamDecl IsMostDerived(getContext(), nullptr, SourceLocation(),
+                                  &getContext().Idents.get("is_most_derived"),
+                                  getContext().IntTy);
+  // Only add the parameter to the list if thie class has virtual bases.
+  if (RD->getNumVBases() > 0)
+    FunctionArgs.push_back(&IsMostDerived);
+
+  // Start defining the function.
+  CGF.StartFunction(GlobalDecl(), FnInfo.getReturnType(), ThunkFn, FnInfo,
+                    FunctionArgs, CD->getLocation(), SourceLocation());
+  EmitThisParam(CGF);
+  llvm::Value *This = getThisValue(CGF);
+
+  llvm::Value *SrcVal =
+      IsCopy ? CGF.Builder.CreateLoad(CGF.GetAddrOfLocalVar(&SrcParam), "src")
+             : nullptr;
+
+  CallArgList Args;
+
+  // Push the this ptr.
+  Args.add(RValue::get(This), CD->getThisType(getContext()));
+
+  // Push the src ptr.
+  if (SrcVal)
+    Args.add(RValue::get(SrcVal), SrcParam.getType());
+
+  // Add the rest of the default arguments.
+  std::vector<Stmt *> ArgVec;
+  for (unsigned I = IsCopy ? 1 : 0, E = CD->getNumParams(); I != E; ++I) {
+    Stmt *DefaultArg = getContext().getDefaultArgExprForConstructor(CD, I);
+    assert(DefaultArg && "sema forgot to instantiate default args");
+    ArgVec.push_back(DefaultArg);
+  }
+
+  CodeGenFunction::RunCleanupsScope Cleanups(CGF);
+
+  const auto *FPT = CD->getType()->castAs<FunctionProtoType>();
+  ConstExprIterator ArgBegin(ArgVec.data()),
+      ArgEnd(ArgVec.data() + ArgVec.size());
+  CGF.EmitCallArgs(Args, FPT, ArgBegin, ArgEnd, CD, IsCopy ? 1 : 0);
+
+  // Insert any ABI-specific implicit constructor arguments.
+  unsigned ExtraArgs = addImplicitConstructorArgs(CGF, CD, Ctor_Complete,
+                                                  /*ForVirtualBase=*/false,
+                                                  /*Delegating=*/false, Args);
+
+  // Call the destructor with our arguments.
+  llvm::Value *CalleeFn = CGM.getAddrOfCXXStructor(CD, StructorType::Complete);
+  const CGFunctionInfo &CalleeInfo = CGM.getTypes().arrangeCXXConstructorCall(
+      Args, CD, Ctor_Complete, ExtraArgs);
+  CGF.EmitCall(CalleeInfo, CalleeFn, ReturnValueSlot(), Args, CD);
+
+  Cleanups.ForceCleanup();
+
+  // Emit the ret instruction, remove any temporary instructions created for the
+  // aid of CodeGen.
+  CGF.FinishFunction(SourceLocation());
+
+  return ThunkFn;
+}
+
+llvm::Constant *MicrosoftCXXABI::getCatchableType(QualType T,
+                                                  uint32_t NVOffset,
+                                                  int32_t VBPtrOffset,
+                                                  uint32_t VBIndex) {
+  assert(!T->isReferenceType());
+
+  CXXRecordDecl *RD = T->getAsCXXRecordDecl();
+  const CXXConstructorDecl *CD =
+      RD ? CGM.getContext().getCopyConstructorForExceptionObject(RD) : nullptr;
+  CXXCtorType CT = Ctor_Complete;
+  if (CD)
+    if (!hasDefaultCXXMethodCC(getContext(), CD) || CD->getNumParams() != 1)
+      CT = Ctor_CopyingClosure;
+
+  uint32_t Size = getContext().getTypeSizeInChars(T).getQuantity();
+  SmallString<256> MangledName;
+  {
+    llvm::raw_svector_ostream Out(MangledName);
+    getMangleContext().mangleCXXCatchableType(T, CD, CT, Size, NVOffset,
+                                              VBPtrOffset, VBIndex, Out);
+  }
+  if (llvm::GlobalVariable *GV = CGM.getModule().getNamedGlobal(MangledName))
+    return getImageRelativeConstant(GV);
+
+  // The TypeDescriptor is used by the runtime to determine if a catch handler
+  // is appropriate for the exception object.
+  llvm::Constant *TD = getImageRelativeConstant(getAddrOfRTTIDescriptor(T));
+
+  // The runtime is responsible for calling the copy constructor if the
+  // exception is caught by value.
+  llvm::Constant *CopyCtor;
+  if (CD) {
+    if (CT == Ctor_CopyingClosure)
+      CopyCtor = getAddrOfCXXCtorClosure(CD, Ctor_CopyingClosure);
+    else
+      CopyCtor = CGM.getAddrOfCXXStructor(CD, StructorType::Complete);
+
+    CopyCtor = llvm::ConstantExpr::getBitCast(CopyCtor, CGM.Int8PtrTy);
+  } else {
+    CopyCtor = llvm::Constant::getNullValue(CGM.Int8PtrTy);
+  }
+  CopyCtor = getImageRelativeConstant(CopyCtor);
+
+  bool IsScalar = !RD;
+  bool HasVirtualBases = false;
+  bool IsStdBadAlloc = false; // std::bad_alloc is special for some reason.
+  QualType PointeeType = T;
+  if (T->isPointerType())
+    PointeeType = T->getPointeeType();
+  if (const CXXRecordDecl *RD = PointeeType->getAsCXXRecordDecl()) {
+    HasVirtualBases = RD->getNumVBases() > 0;
+    if (IdentifierInfo *II = RD->getIdentifier())
+      IsStdBadAlloc = II->isStr("bad_alloc") && RD->isInStdNamespace();
+  }
+
+  // Encode the relevant CatchableType properties into the Flags bitfield.
+  // FIXME: Figure out how bits 2 or 8 can get set.
+  uint32_t Flags = 0;
+  if (IsScalar)
+    Flags |= 1;
+  if (HasVirtualBases)
+    Flags |= 4;
+  if (IsStdBadAlloc)
+    Flags |= 16;
+
+  llvm::Constant *Fields[] = {
+      llvm::ConstantInt::get(CGM.IntTy, Flags),       // Flags
+      TD,                                             // TypeDescriptor
+      llvm::ConstantInt::get(CGM.IntTy, NVOffset),    // NonVirtualAdjustment
+      llvm::ConstantInt::get(CGM.IntTy, VBPtrOffset), // OffsetToVBPtr
+      llvm::ConstantInt::get(CGM.IntTy, VBIndex),     // VBTableIndex
+      llvm::ConstantInt::get(CGM.IntTy, Size),        // Size
+      CopyCtor                                        // CopyCtor
+  };
+  llvm::StructType *CTType = getCatchableTypeType();
+  auto *GV = new llvm::GlobalVariable(
+      CGM.getModule(), CTType, /*Constant=*/true, getLinkageForRTTI(T),
+      llvm::ConstantStruct::get(CTType, Fields), StringRef(MangledName));
+  GV->setUnnamedAddr(true);
+  GV->setSection(".xdata");
+  if (GV->isWeakForLinker())
+    GV->setComdat(CGM.getModule().getOrInsertComdat(GV->getName()));
+  return getImageRelativeConstant(GV);
+}
+
+llvm::GlobalVariable *MicrosoftCXXABI::getCatchableTypeArray(QualType T) {
+  assert(!T->isReferenceType());
+
+  // See if we've already generated a CatchableTypeArray for this type before.
+  llvm::GlobalVariable *&CTA = CatchableTypeArrays[T];
+  if (CTA)
+    return CTA;
+
+  // Ensure that we don't have duplicate entries in our CatchableTypeArray by
+  // using a SmallSetVector.  Duplicates may arise due to virtual bases
+  // occurring more than once in the hierarchy.
+  llvm::SmallSetVector<llvm::Constant *, 2> CatchableTypes;
+
+  // C++14 [except.handle]p3:
+  //   A handler is a match for an exception object of type E if [...]
+  //     - the handler is of type cv T or cv T& and T is an unambiguous public
+  //       base class of E, or
+  //     - the handler is of type cv T or const T& where T is a pointer type and
+  //       E is a pointer type that can be converted to T by [...]
+  //         - a standard pointer conversion (4.10) not involving conversions to
+  //           pointers to private or protected or ambiguous classes
+  const CXXRecordDecl *MostDerivedClass = nullptr;
+  bool IsPointer = T->isPointerType();
+  if (IsPointer)
+    MostDerivedClass = T->getPointeeType()->getAsCXXRecordDecl();
+  else
+    MostDerivedClass = T->getAsCXXRecordDecl();
+
+  // Collect all the unambiguous public bases of the MostDerivedClass.
+  if (MostDerivedClass) {
+    const ASTContext &Context = getContext();
+    const ASTRecordLayout &MostDerivedLayout =
+        Context.getASTRecordLayout(MostDerivedClass);
+    MicrosoftVTableContext &VTableContext = CGM.getMicrosoftVTableContext();
+    SmallVector<MSRTTIClass, 8> Classes;
+    serializeClassHierarchy(Classes, MostDerivedClass);
+    Classes.front().initialize(/*Parent=*/nullptr, /*Specifier=*/nullptr);
+    detectAmbiguousBases(Classes);
+    for (const MSRTTIClass &Class : Classes) {
+      // Skip any ambiguous or private bases.
+      if (Class.Flags &
+          (MSRTTIClass::IsPrivateOnPath | MSRTTIClass::IsAmbiguous))
+        continue;
+      // Write down how to convert from a derived pointer to a base pointer.
+      uint32_t OffsetInVBTable = 0;
+      int32_t VBPtrOffset = -1;
+      if (Class.VirtualRoot) {
+        OffsetInVBTable =
+          VTableContext.getVBTableIndex(MostDerivedClass, Class.VirtualRoot)*4;
+        VBPtrOffset = MostDerivedLayout.getVBPtrOffset().getQuantity();
+      }
+
+      // Turn our record back into a pointer if the exception object is a
+      // pointer.
+      QualType RTTITy = QualType(Class.RD->getTypeForDecl(), 0);
+      if (IsPointer)
+        RTTITy = Context.getPointerType(RTTITy);
+      CatchableTypes.insert(getCatchableType(RTTITy, Class.OffsetInVBase,
+                                             VBPtrOffset, OffsetInVBTable));
+    }
+  }
+
+  // C++14 [except.handle]p3:
+  //   A handler is a match for an exception object of type E if
+  //     - The handler is of type cv T or cv T& and E and T are the same type
+  //       (ignoring the top-level cv-qualifiers)
+  CatchableTypes.insert(getCatchableType(T));
+
+  // C++14 [except.handle]p3:
+  //   A handler is a match for an exception object of type E if
+  //     - the handler is of type cv T or const T& where T is a pointer type and
+  //       E is a pointer type that can be converted to T by [...]
+  //         - a standard pointer conversion (4.10) not involving conversions to
+  //           pointers to private or protected or ambiguous classes
+  //
+  // C++14 [conv.ptr]p2:
+  //   A prvalue of type "pointer to cv T," where T is an object type, can be
+  //   converted to a prvalue of type "pointer to cv void".
+  if (IsPointer && T->getPointeeType()->isObjectType())
+    CatchableTypes.insert(getCatchableType(getContext().VoidPtrTy));
+
+  // C++14 [except.handle]p3:
+  //   A handler is a match for an exception object of type E if [...]
+  //     - the handler is of type cv T or const T& where T is a pointer or
+  //       pointer to member type and E is std::nullptr_t.
+  //
+  // We cannot possibly list all possible pointer types here, making this
+  // implementation incompatible with the standard.  However, MSVC includes an
+  // entry for pointer-to-void in this case.  Let's do the same.
+  if (T->isNullPtrType())
+    CatchableTypes.insert(getCatchableType(getContext().VoidPtrTy));
+
+  uint32_t NumEntries = CatchableTypes.size();
+  llvm::Type *CTType =
+      getImageRelativeType(getCatchableTypeType()->getPointerTo());
+  llvm::ArrayType *AT = llvm::ArrayType::get(CTType, NumEntries);
+  llvm::StructType *CTAType = getCatchableTypeArrayType(NumEntries);
+  llvm::Constant *Fields[] = {
+      llvm::ConstantInt::get(CGM.IntTy, NumEntries),    // NumEntries
+      llvm::ConstantArray::get(
+          AT, llvm::makeArrayRef(CatchableTypes.begin(),
+                                 CatchableTypes.end())) // CatchableTypes
+  };
+  SmallString<256> MangledName;
+  {
+    llvm::raw_svector_ostream Out(MangledName);
+    getMangleContext().mangleCXXCatchableTypeArray(T, NumEntries, Out);
+  }
+  CTA = new llvm::GlobalVariable(
+      CGM.getModule(), CTAType, /*Constant=*/true, getLinkageForRTTI(T),
+      llvm::ConstantStruct::get(CTAType, Fields), StringRef(MangledName));
+  CTA->setUnnamedAddr(true);
+  CTA->setSection(".xdata");
+  if (CTA->isWeakForLinker())
+    CTA->setComdat(CGM.getModule().getOrInsertComdat(CTA->getName()));
+  return CTA;
+}
+
+llvm::GlobalVariable *MicrosoftCXXABI::getThrowInfo(QualType T) {
+  bool IsConst, IsVolatile;
+  T = decomposeTypeForEH(getContext(), T, IsConst, IsVolatile);
+
+  // The CatchableTypeArray enumerates the various (CV-unqualified) types that
+  // the exception object may be caught as.
+  llvm::GlobalVariable *CTA = getCatchableTypeArray(T);
+  // The first field in a CatchableTypeArray is the number of CatchableTypes.
+  // This is used as a component of the mangled name which means that we need to
+  // know what it is in order to see if we have previously generated the
+  // ThrowInfo.
+  uint32_t NumEntries =
+      cast<llvm::ConstantInt>(CTA->getInitializer()->getAggregateElement(0U))
+          ->getLimitedValue();
+
+  SmallString<256> MangledName;
+  {
+    llvm::raw_svector_ostream Out(MangledName);
+    getMangleContext().mangleCXXThrowInfo(T, IsConst, IsVolatile, NumEntries,
+                                          Out);
+  }
+
+  // Reuse a previously generated ThrowInfo if we have generated an appropriate
+  // one before.
+  if (llvm::GlobalVariable *GV = CGM.getModule().getNamedGlobal(MangledName))
+    return GV;
+
+  // The RTTI TypeDescriptor uses an unqualified type but catch clauses must
+  // be at least as CV qualified.  Encode this requirement into the Flags
+  // bitfield.
+  uint32_t Flags = 0;
+  if (IsConst)
+    Flags |= 1;
+  if (IsVolatile)
+    Flags |= 2;
+
+  // The cleanup-function (a destructor) must be called when the exception
+  // object's lifetime ends.
+  llvm::Constant *CleanupFn = llvm::Constant::getNullValue(CGM.Int8PtrTy);
+  if (const CXXRecordDecl *RD = T->getAsCXXRecordDecl())
+    if (CXXDestructorDecl *DtorD = RD->getDestructor())
+      if (!DtorD->isTrivial())
+        CleanupFn = llvm::ConstantExpr::getBitCast(
+            CGM.getAddrOfCXXStructor(DtorD, StructorType::Complete),
+            CGM.Int8PtrTy);
+  // This is unused as far as we can tell, initialize it to null.
+  llvm::Constant *ForwardCompat =
+      getImageRelativeConstant(llvm::Constant::getNullValue(CGM.Int8PtrTy));
+  llvm::Constant *PointerToCatchableTypes = getImageRelativeConstant(
+      llvm::ConstantExpr::getBitCast(CTA, CGM.Int8PtrTy));
+  llvm::StructType *TIType = getThrowInfoType();
+  llvm::Constant *Fields[] = {
+      llvm::ConstantInt::get(CGM.IntTy, Flags), // Flags
+      getImageRelativeConstant(CleanupFn),      // CleanupFn
+      ForwardCompat,                            // ForwardCompat
+      PointerToCatchableTypes                   // CatchableTypeArray
+  };
+  auto *GV = new llvm::GlobalVariable(
+      CGM.getModule(), TIType, /*Constant=*/true, getLinkageForRTTI(T),
+      llvm::ConstantStruct::get(TIType, Fields), StringRef(MangledName));
+  GV->setUnnamedAddr(true);
+  GV->setSection(".xdata");
+  if (GV->isWeakForLinker())
+    GV->setComdat(CGM.getModule().getOrInsertComdat(GV->getName()));
+  return GV;
+}
+
+void MicrosoftCXXABI::emitThrow(CodeGenFunction &CGF, const CXXThrowExpr *E) {
+  const Expr *SubExpr = E->getSubExpr();
+  QualType ThrowType = SubExpr->getType();
+  // The exception object lives on the stack and it's address is passed to the
+  // runtime function.
+  llvm::AllocaInst *AI = CGF.CreateMemTemp(ThrowType);
+  CGF.EmitAnyExprToMem(SubExpr, AI, ThrowType.getQualifiers(),
+                       /*IsInit=*/true);
+
+  // The so-called ThrowInfo is used to describe how the exception object may be
+  // caught.
+  llvm::GlobalVariable *TI = getThrowInfo(ThrowType);
+
+  // Call into the runtime to throw the exception.
+  llvm::Value *Args[] = {CGF.Builder.CreateBitCast(AI, CGM.Int8PtrTy), TI};
+  CGF.EmitNoreturnRuntimeCallOrInvoke(getThrowFn(), Args);
+}
diff --git a/lib/CodeGen/ModuleBuilder.cpp b/lib/CodeGen/ModuleBuilder.cpp
index 4f1a82e0248d..25e57401fd57 100644
--- a/lib/CodeGen/ModuleBuilder.cpp
+++ b/lib/CodeGen/ModuleBuilder.cpp
@@ -63,7 +63,7 @@ namespace {
         CoverageInfo(CoverageInfo),
         M(new llvm::Module(ModuleName, C)) {}
 
-    virtual ~CodeGeneratorImpl() {
+    ~CodeGeneratorImpl() override {
       // There should normally not be any leftover inline method definitions.
       assert(DeferredInlineMethodDefinitions.empty() ||
              Diags.hasErrorOccurred());
@@ -95,7 +95,8 @@ namespace {
 
       M->setTargetTriple(Ctx->getTargetInfo().getTriple().getTriple());
       M->setDataLayout(Ctx->getTargetInfo().getTargetDescription());
-      TD.reset(new llvm::DataLayout(Ctx->getTargetInfo().getTargetDescription()));
+      TD.reset(
+          new llvm::DataLayout(Ctx->getTargetInfo().getTargetDescription()));
       Builder.reset(new CodeGen::CodeGenModule(Context, CodeGenOpts, *M, *TD,
                                                Diags, CoverageInfo));
 
@@ -211,11 +212,11 @@ namespace {
       Builder->EmitTentativeDefinition(D);
     }
 
-    void HandleVTable(CXXRecordDecl *RD, bool DefinitionRequired) override {
+    void HandleVTable(CXXRecordDecl *RD) override {
       if (Diags.hasErrorOccurred())
         return;
 
-      Builder->EmitVTable(RD, DefinitionRequired);
+      Builder->EmitVTable(RD);
     }
 
     void HandleLinkerOptionPragma(llvm::StringRef Opts) override {
@@ -238,7 +239,6 @@ void CodeGenerator::anchor() { }
 CodeGenerator *clang::CreateLLVMCodeGen(DiagnosticsEngine &Diags,
                                         const std::string& ModuleName,
                                         const CodeGenOptions &CGO,
-                                        const TargetOptions &/*TO*/,
                                         llvm::LLVMContext& C,
                                         CoverageSourceInfo *CoverageInfo) {
   return new CodeGeneratorImpl(Diags, ModuleName, CGO, C, CoverageInfo);
diff --git a/lib/CodeGen/SanitizerMetadata.h b/lib/CodeGen/SanitizerMetadata.h
index d2f0651159a2..166f0e6c9b58 100644
--- a/lib/CodeGen/SanitizerMetadata.h
+++ b/lib/CodeGen/SanitizerMetadata.h
@@ -31,8 +31,8 @@ namespace CodeGen {
 class CodeGenModule;
 
 class SanitizerMetadata {
-  SanitizerMetadata(const SanitizerMetadata &) LLVM_DELETED_FUNCTION;
-  void operator=(const SanitizerMetadata &) LLVM_DELETED_FUNCTION;
+  SanitizerMetadata(const SanitizerMetadata &) = delete;
+  void operator=(const SanitizerMetadata &) = delete;
 
   CodeGenModule &CGM;
 public:
diff --git a/lib/CodeGen/TargetInfo.cpp b/lib/CodeGen/TargetInfo.cpp
index c05b23a32e86..53154b513eb7 100644
--- a/lib/CodeGen/TargetInfo.cpp
+++ b/lib/CodeGen/TargetInfo.cpp
@@ -37,7 +37,8 @@ static void AssignToArrayRange(CodeGen::CGBuilderTy &Builder,
                                unsigned LastIndex) {
   // Alternatively, we could emit this as a loop in the source.
   for (unsigned I = FirstIndex; I <= LastIndex; ++I) {
-    llvm::Value *Cell = Builder.CreateConstInBoundsGEP1_32(Array, I);
+    llvm::Value *Cell =
+        Builder.CreateConstInBoundsGEP1_32(Builder.getInt8Ty(), Array, I);
     Builder.CreateStore(Value, Cell);
   }
 }
@@ -107,6 +108,10 @@ bool ABIInfo::isHomogeneousAggregateSmallEnough(const Type *Base,
   return false;
 }
 
+bool ABIInfo::shouldSignExtUnsignedType(QualType Ty) const {
+  return false;
+}
+
 void ABIArgInfo::dump() const {
   raw_ostream &OS = llvm::errs();
   OS << "(ABIArgInfo Kind=";
@@ -238,7 +243,7 @@ static bool isEmptyRecord(ASTContext &Context, QualType T, bool AllowArrays) {
 /// \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();
+  const RecordType *RT = T->getAs<RecordType>();
   if (!RT)
     return nullptr;
 
@@ -339,9 +344,15 @@ static bool canExpandIndirectArgument(QualType Ty, ASTContext &Context) {
   //
   // FIXME: This needs to be generalized to handle classes as well.
   const RecordDecl *RD = RT->getDecl();
-  if (!RD->isStruct() || isa<CXXRecordDecl>(RD))
+  if (!RD->isStruct())
     return false;
 
+  // We try to expand CLike CXXRecordDecl.
+  if (const CXXRecordDecl *CXXRD = dyn_cast<CXXRecordDecl>(RD)) {
+    if (!CXXRD->isCLike())
+      return false;
+  }
+
   uint64_t Size = 0;
 
   for (const auto *FD : RD->fields()) {
@@ -399,8 +410,16 @@ llvm::Value *DefaultABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
 }
 
 ABIArgInfo DefaultABIInfo::classifyArgumentType(QualType Ty) const {
-  if (isAggregateTypeForABI(Ty))
+  Ty = useFirstFieldIfTransparentUnion(Ty);
+
+  if (isAggregateTypeForABI(Ty)) {
+    // Records with non-trivial destructors/copy-constructors should not be
+    // passed by value.
+    if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI()))
+      return ABIArgInfo::getIndirect(0, RAA == CGCXXABI::RAA_DirectInMemory);
+
     return ABIArgInfo::getIndirect(0);
+  }
 
   // Treat an enum type as its underlying type.
   if (const EnumType *EnumTy = Ty->getAs<EnumType>())
@@ -1352,7 +1371,8 @@ bool X86_32TargetCodeGenInfo::initDwarfEHRegSizeTable(
   } else {
     // 9 is %eflags, which doesn't get a size on Darwin for some
     // reason.
-    Builder.CreateStore(Four8, Builder.CreateConstInBoundsGEP1_32(Address, 9));
+    Builder.CreateStore(
+        Four8, Builder.CreateConstInBoundsGEP1_32(CGF.Int8Ty, Address, 9));
 
     // 11-16 are st(0..5).  Not sure why we stop at 5.
     // These have size 12, which is sizeof(long double) on
@@ -1475,14 +1495,13 @@ class X86_64ABIInfo : public ABIInfo {
     return !getTarget().getTriple().isOSDarwin();
   }
 
-  bool HasAVX;
   // Some ABIs (e.g. X32 ABI and Native Client OS) use 32 bit pointers on
   // 64-bit hardware.
   bool Has64BitPointers;
 
 public:
-  X86_64ABIInfo(CodeGen::CodeGenTypes &CGT, bool hasavx) :
-      ABIInfo(CGT), HasAVX(hasavx),
+  X86_64ABIInfo(CodeGen::CodeGenTypes &CGT) :
+      ABIInfo(CGT),
       Has64BitPointers(CGT.getDataLayout().getPointerSize(0) == 8) {
   }
 
@@ -1503,6 +1522,14 @@ public:
 
   llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
                          CodeGenFunction &CGF) const override;
+
+  bool has64BitPointers() const {
+    return Has64BitPointers;
+  }
+
+  bool hasAVX() const {
+    return getTarget().getABI() == "avx";
+  }
 };
 
 /// WinX86_64ABIInfo - The Windows X86_64 ABI information.
@@ -1532,10 +1559,9 @@ public:
 };
 
 class X86_64TargetCodeGenInfo : public TargetCodeGenInfo {
-  bool HasAVX;
 public:
-  X86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, bool HasAVX)
-      : TargetCodeGenInfo(new X86_64ABIInfo(CGT, HasAVX)), HasAVX(HasAVX) {}
+  X86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT)
+      : TargetCodeGenInfo(new X86_64ABIInfo(CGT)) {}
 
   const X86_64ABIInfo &getABIInfo() const {
     return static_cast<const X86_64ABIInfo&>(TargetCodeGenInfo::getABIInfo());
@@ -1588,24 +1614,47 @@ public:
 
   llvm::Constant *
   getUBSanFunctionSignature(CodeGen::CodeGenModule &CGM) const override {
-    unsigned Sig = (0xeb << 0) |  // jmp rel8
-                   (0x0a << 8) |  //           .+0x0c
-                   ('F' << 16) |
-                   ('T' << 24);
+    unsigned Sig;
+    if (getABIInfo().has64BitPointers())
+      Sig = (0xeb << 0) |  // jmp rel8
+            (0x0a << 8) |  //           .+0x0c
+            ('F' << 16) |
+            ('T' << 24);
+    else
+      Sig = (0xeb << 0) |  // jmp rel8
+            (0x06 << 8) |  //           .+0x08
+            ('F' << 16) |
+            ('T' << 24);
     return llvm::ConstantInt::get(CGM.Int32Ty, Sig);
   }
 
   unsigned getOpenMPSimdDefaultAlignment(QualType) const override {
-    return HasAVX ? 32 : 16;
+    return getABIInfo().hasAVX() ? 32 : 16;
+  }
+};
+
+class PS4TargetCodeGenInfo : public X86_64TargetCodeGenInfo {
+public:
+  PS4TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT)
+    : X86_64TargetCodeGenInfo(CGT) {}
+
+  void getDependentLibraryOption(llvm::StringRef Lib,
+                                 llvm::SmallString<24> &Opt) const override {
+    Opt = "\01";
+    Opt += Lib;
   }
 };
 
 static std::string qualifyWindowsLibrary(llvm::StringRef Lib) {
-  // If the argument does not end in .lib, automatically add the suffix. This
-  // matches the behavior of MSVC.
-  std::string ArgStr = Lib;
+  // If the argument does not end in .lib, automatically add the suffix.
+  // If the argument contains a space, enclose it in quotes.
+  // This matches the behavior of MSVC.
+  bool Quote = (Lib.find(" ") != StringRef::npos);
+  std::string ArgStr = Quote ? "\"" : "";
+  ArgStr += Lib;
   if (!Lib.endswith_lower(".lib"))
     ArgStr += ".lib";
+  ArgStr += Quote ? "\"" : "";
   return ArgStr;
 }
 
@@ -1615,6 +1664,9 @@ public:
         bool d, bool p, bool w, unsigned RegParms)
     : X86_32TargetCodeGenInfo(CGT, d, p, w, RegParms) {}
 
+  void SetTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
+                           CodeGen::CodeGenModule &CGM) const override;
+
   void getDependentLibraryOption(llvm::StringRef Lib,
                                  llvm::SmallString<24> &Opt) const override {
     Opt = "/DEFAULTLIB:";
@@ -1628,11 +1680,35 @@ public:
   }
 };
 
+static void addStackProbeSizeTargetAttribute(const Decl *D,
+                                             llvm::GlobalValue *GV,
+                                             CodeGen::CodeGenModule &CGM) {
+  if (isa<FunctionDecl>(D)) {
+    if (CGM.getCodeGenOpts().StackProbeSize != 4096) {
+      llvm::Function *Fn = cast<llvm::Function>(GV);
+
+      Fn->addFnAttr("stack-probe-size", llvm::utostr(CGM.getCodeGenOpts().StackProbeSize));
+    }
+  }
+}
+
+void WinX86_32TargetCodeGenInfo::SetTargetAttributes(const Decl *D,
+                                                     llvm::GlobalValue *GV,
+                                            CodeGen::CodeGenModule &CGM) const {
+  X86_32TargetCodeGenInfo::SetTargetAttributes(D, GV, CGM);
+
+  addStackProbeSizeTargetAttribute(D, GV, CGM);
+}
+
 class WinX86_64TargetCodeGenInfo : public TargetCodeGenInfo {
-  bool HasAVX;
+  bool hasAVX() const { return getABIInfo().getTarget().getABI() == "avx"; }
+
 public:
-  WinX86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, bool HasAVX)
-    : TargetCodeGenInfo(new WinX86_64ABIInfo(CGT)), HasAVX(HasAVX) {}
+  WinX86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT)
+    : TargetCodeGenInfo(new WinX86_64ABIInfo(CGT)) {}
+
+  void SetTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
+                           CodeGen::CodeGenModule &CGM) const override;
 
   int getDwarfEHStackPointer(CodeGen::CodeGenModule &CGM) const override {
     return 7;
@@ -1661,10 +1737,17 @@ public:
   }
 
   unsigned getOpenMPSimdDefaultAlignment(QualType) const override {
-    return HasAVX ? 32 : 16;
+    return hasAVX() ? 32 : 16;
   }
 };
 
+void WinX86_64TargetCodeGenInfo::SetTargetAttributes(const Decl *D,
+                                                     llvm::GlobalValue *GV,
+                                            CodeGen::CodeGenModule &CGM) const {
+  TargetCodeGenInfo::SetTargetAttributes(D, GV, CGM);
+
+  addStackProbeSizeTargetAttribute(D, GV, CGM);
+}
 }
 
 void X86_64ABIInfo::postMerge(unsigned AggregateSize, Class &Lo,
@@ -1843,7 +1926,7 @@ void X86_64ABIInfo::classify(QualType Ty, uint64_t OffsetBase,
       // split.
       if (OffsetBase && OffsetBase != 64)
         Hi = Lo;
-    } else if (Size == 128 || (HasAVX && isNamedArg && Size == 256)) {
+    } else if (Size == 128 || (hasAVX() && isNamedArg && Size == 256)) {
       // Arguments of 256-bits are split into four eightbyte chunks. The
       // least significant one belongs to class SSE and all the others to class
       // SSEUP. The original Lo and Hi design considers that types can't be
@@ -2065,7 +2148,7 @@ ABIArgInfo X86_64ABIInfo::getIndirectReturnResult(QualType Ty) const {
 bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
   if (const VectorType *VecTy = Ty->getAs<VectorType>()) {
     uint64_t Size = getContext().getTypeSize(VecTy);
-    unsigned LargestVector = HasAVX ? 256 : 128;
+    unsigned LargestVector = hasAVX() ? 256 : 128;
     if (Size <= 64 || Size > LargestVector)
       return true;
   }
@@ -2142,20 +2225,9 @@ llvm::Type *X86_64ABIInfo::GetByteVectorType(QualType Ty) const {
     Ty = QualType(InnerTy, 0);
 
   llvm::Type *IRType = CGT.ConvertType(Ty);
-
-  // If the preferred type is a 16-byte vector, prefer to pass it.
-  if (llvm::VectorType *VT = dyn_cast<llvm::VectorType>(IRType)){
-    llvm::Type *EltTy = VT->getElementType();
-    unsigned BitWidth = VT->getBitWidth();
-    if ((BitWidth >= 128 && BitWidth <= 256) &&
-        (EltTy->isFloatTy() || EltTy->isDoubleTy() ||
-         EltTy->isIntegerTy(8) || EltTy->isIntegerTy(16) ||
-         EltTy->isIntegerTy(32) || EltTy->isIntegerTy(64) ||
-         EltTy->isIntegerTy(128)))
-      return VT;
-  }
-
-  return llvm::VectorType::get(llvm::Type::getDoubleTy(getVMContext()), 2);
+  assert(isa<llvm::VectorType>(IRType) &&
+         "Trying to return a non-vector type in a vector register!");
+  return IRType;
 }
 
 /// BitsContainNoUserData - Return true if the specified [start,end) bit range
@@ -2717,8 +2789,8 @@ void X86_64ABIInfo::computeInfo(CGFunctionInfo &FI) const {
 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_p = CGF.Builder.CreateStructGEP(
+      nullptr, VAListAddr, 2, "overflow_arg_area_p");
   llvm::Value *overflow_arg_area =
     CGF.Builder.CreateLoad(overflow_arg_area_p, "overflow_arg_area");
 
@@ -2798,14 +2870,16 @@ llvm::Value *X86_64ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
   llvm::Value *gp_offset_p = nullptr, *gp_offset = nullptr;
   llvm::Value *fp_offset_p = nullptr, *fp_offset = nullptr;
   if (neededInt) {
-    gp_offset_p = CGF.Builder.CreateStructGEP(VAListAddr, 0, "gp_offset_p");
+    gp_offset_p =
+        CGF.Builder.CreateStructGEP(nullptr, VAListAddr, 0, "gp_offset_p");
     gp_offset = CGF.Builder.CreateLoad(gp_offset_p, "gp_offset");
     InRegs = llvm::ConstantInt::get(CGF.Int32Ty, 48 - neededInt * 8);
     InRegs = CGF.Builder.CreateICmpULE(gp_offset, InRegs, "fits_in_gp");
   }
 
   if (neededSSE) {
-    fp_offset_p = CGF.Builder.CreateStructGEP(VAListAddr, 1, "fp_offset_p");
+    fp_offset_p =
+        CGF.Builder.CreateStructGEP(nullptr, VAListAddr, 1, "fp_offset_p");
     fp_offset = CGF.Builder.CreateLoad(fp_offset_p, "fp_offset");
     llvm::Value *FitsInFP =
       llvm::ConstantInt::get(CGF.Int32Ty, 176 - neededSSE * 16);
@@ -2833,9 +2907,8 @@ llvm::Value *X86_64ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
   // simple assembling of a structure from scattered addresses has many more
   // loads than necessary. Can we clean this up?
   llvm::Type *LTy = CGF.ConvertTypeForMem(Ty);
-  llvm::Value *RegAddr =
-    CGF.Builder.CreateLoad(CGF.Builder.CreateStructGEP(VAListAddr, 3),
-                           "reg_save_area");
+  llvm::Value *RegAddr = CGF.Builder.CreateLoad(
+      CGF.Builder.CreateStructGEP(nullptr, VAListAddr, 3), "reg_save_area");
   if (neededInt && neededSSE) {
     // FIXME: Cleanup.
     assert(AI.isDirect() && "Unexpected ABI info for mixed regs");
@@ -2855,9 +2928,9 @@ llvm::Value *X86_64ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
     llvm::Value *RegHiAddr = TyLo->isFPOrFPVectorTy() ? GPAddr : FPAddr;
     llvm::Value *V =
       CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegLoAddr, PTyLo));
-    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 0));
+    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(ST, Tmp, 0));
     V = CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegHiAddr, PTyHi));
-    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 1));
+    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(ST, Tmp, 1));
 
     RegAddr = CGF.Builder.CreateBitCast(Tmp,
                                         llvm::PointerType::getUnqual(LTy));
@@ -2894,10 +2967,10 @@ llvm::Value *X86_64ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
     Tmp = CGF.Builder.CreateBitCast(Tmp, ST->getPointerTo());
     V = CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegAddrLo,
                                                          DblPtrTy));
-    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 0));
+    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(ST, Tmp, 0));
     V = CGF.Builder.CreateLoad(CGF.Builder.CreateBitCast(RegAddrHi,
                                                          DblPtrTy));
-    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 1));
+    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(ST, Tmp, 1));
     RegAddr = CGF.Builder.CreateBitCast(Tmp,
                                         llvm::PointerType::getUnqual(LTy));
   }
@@ -2984,7 +3057,7 @@ ABIArgInfo WinX86_64ABIInfo::classify(QualType Ty, unsigned &FreeSSERegs,
       return ABIArgInfo::getDirect();
   }
 
-  if (RT || Ty->isMemberPointerType()) {
+  if (RT || Ty->isAnyComplexType() || Ty->isMemberPointerType()) {
     // MS x64 ABI requirement: "Any argument that doesn't fit in 8 bytes, or is
     // not 1, 2, 4, or 8 bytes, must be passed by reference."
     if (Width > 64 || !llvm::isPowerOf2_64(Width))
@@ -3040,48 +3113,6 @@ llvm::Value *WinX86_64ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
   return AddrTyped;
 }
 
-namespace {
-
-class NaClX86_64ABIInfo : public ABIInfo {
- public:
-  NaClX86_64ABIInfo(CodeGen::CodeGenTypes &CGT, bool HasAVX)
-      : ABIInfo(CGT), PInfo(CGT), NInfo(CGT, HasAVX) {}
-  void computeInfo(CGFunctionInfo &FI) const override;
-  llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                         CodeGenFunction &CGF) const override;
- private:
-  PNaClABIInfo PInfo;  // Used for generating calls with pnaclcall callingconv.
-  X86_64ABIInfo NInfo; // Used for everything else.
-};
-
-class NaClX86_64TargetCodeGenInfo : public TargetCodeGenInfo  {
-  bool HasAVX;
- public:
-   NaClX86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, bool HasAVX)
-       : TargetCodeGenInfo(new NaClX86_64ABIInfo(CGT, HasAVX)), HasAVX(HasAVX) {
-   }
-   unsigned getOpenMPSimdDefaultAlignment(QualType) const override {
-     return HasAVX ? 32 : 16;
-   }
-};
-
-}
-
-void NaClX86_64ABIInfo::computeInfo(CGFunctionInfo &FI) const {
-  if (FI.getASTCallingConvention() == CC_PnaclCall)
-    PInfo.computeInfo(FI);
-  else
-    NInfo.computeInfo(FI);
-}
-
-llvm::Value *NaClX86_64ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                          CodeGenFunction &CGF) const {
-  // Always use the native convention; calling pnacl-style varargs functions
-  // is unuspported.
-  return NInfo.EmitVAArg(VAListAddr, Ty, CGF);
-}
-
-
 // PowerPC-32
 namespace {
 /// PPC32_SVR4_ABIInfo - The 32-bit PowerPC ELF (SVR4) ABI information.
@@ -3257,13 +3288,42 @@ public:
 private:
   static const unsigned GPRBits = 64;
   ABIKind Kind;
+  bool HasQPX;
+
+  // A vector of float or double will be promoted to <4 x f32> or <4 x f64> and
+  // will be passed in a QPX register.
+  bool IsQPXVectorTy(const Type *Ty) const {
+    if (!HasQPX)
+      return false;
+
+    if (const VectorType *VT = Ty->getAs<VectorType>()) {
+      unsigned NumElements = VT->getNumElements();
+      if (NumElements == 1)
+        return false;
+
+      if (VT->getElementType()->isSpecificBuiltinType(BuiltinType::Double)) {
+        if (getContext().getTypeSize(Ty) <= 256)
+          return true;
+      } else if (VT->getElementType()->
+                   isSpecificBuiltinType(BuiltinType::Float)) {
+        if (getContext().getTypeSize(Ty) <= 128)
+          return true;
+      }
+    }
+
+    return false;
+  }
+
+  bool IsQPXVectorTy(QualType Ty) const {
+    return IsQPXVectorTy(Ty.getTypePtr());
+  }
 
 public:
-  PPC64_SVR4_ABIInfo(CodeGen::CodeGenTypes &CGT, ABIKind Kind)
-    : DefaultABIInfo(CGT), Kind(Kind) {}
+  PPC64_SVR4_ABIInfo(CodeGen::CodeGenTypes &CGT, ABIKind Kind, bool HasQPX)
+    : DefaultABIInfo(CGT), Kind(Kind), HasQPX(HasQPX) {}
 
   bool isPromotableTypeForABI(QualType Ty) const;
-  bool isAlignedParamType(QualType Ty) const;
+  bool isAlignedParamType(QualType Ty, bool &Align32) const;
 
   ABIArgInfo classifyReturnType(QualType RetTy) const;
   ABIArgInfo classifyArgumentType(QualType Ty) const;
@@ -3288,7 +3348,8 @@ public:
       const Type *T = isSingleElementStruct(I.type, getContext());
       if (T) {
         const BuiltinType *BT = T->getAs<BuiltinType>();
-        if ((T->isVectorType() && getContext().getTypeSize(T) == 128) ||
+        if (IsQPXVectorTy(T) ||
+            (T->isVectorType() && getContext().getTypeSize(T) == 128) ||
             (BT && BT->isFloatingPoint())) {
           QualType QT(T, 0);
           I.info = ABIArgInfo::getDirectInReg(CGT.ConvertType(QT));
@@ -3304,10 +3365,13 @@ public:
 };
 
 class PPC64_SVR4_TargetCodeGenInfo : public TargetCodeGenInfo {
+  bool HasQPX;
+
 public:
   PPC64_SVR4_TargetCodeGenInfo(CodeGenTypes &CGT,
-                               PPC64_SVR4_ABIInfo::ABIKind Kind)
-    : TargetCodeGenInfo(new PPC64_SVR4_ABIInfo(CGT, Kind)) {}
+                               PPC64_SVR4_ABIInfo::ABIKind Kind, bool HasQPX)
+    : TargetCodeGenInfo(new PPC64_SVR4_ABIInfo(CGT, Kind, HasQPX)),
+      HasQPX(HasQPX) {}
 
   int getDwarfEHStackPointer(CodeGen::CodeGenModule &M) const override {
     // This is recovered from gcc output.
@@ -3317,7 +3381,12 @@ public:
   bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                                llvm::Value *Address) const override;
 
-  unsigned getOpenMPSimdDefaultAlignment(QualType) const override {
+  unsigned getOpenMPSimdDefaultAlignment(QualType QT) const override {
+    if (HasQPX)
+      if (const PointerType *PT = QT->getAs<PointerType>())
+        if (PT->getPointeeType()->isSpecificBuiltinType(BuiltinType::Double))
+          return 32; // Natural alignment for QPX doubles.
+
     return 16; // Natural alignment for Altivec and VSX vectors.
   }
 };
@@ -3370,15 +3439,23 @@ PPC64_SVR4_ABIInfo::isPromotableTypeForABI(QualType Ty) const {
 /// isAlignedParamType - Determine whether a type requires 16-byte
 /// alignment in the parameter area.
 bool
-PPC64_SVR4_ABIInfo::isAlignedParamType(QualType Ty) const {
+PPC64_SVR4_ABIInfo::isAlignedParamType(QualType Ty, bool &Align32) const {
+  Align32 = false;
+
   // Complex types are passed just like their elements.
   if (const ComplexType *CTy = Ty->getAs<ComplexType>())
     Ty = CTy->getElementType();
 
   // Only vector types of size 16 bytes need alignment (larger types are
   // passed via reference, smaller types are not aligned).
-  if (Ty->isVectorType())
+  if (IsQPXVectorTy(Ty)) {
+    if (getContext().getTypeSize(Ty) > 128)
+      Align32 = true;
+
+    return true;
+  } else if (Ty->isVectorType()) {
     return getContext().getTypeSize(Ty) == 128;
+  }
 
   // For single-element float/vector structs, we consider the whole type
   // to have the same alignment requirements as its single element.
@@ -3386,7 +3463,7 @@ PPC64_SVR4_ABIInfo::isAlignedParamType(QualType Ty) const {
   const Type *EltType = isSingleElementStruct(Ty, getContext());
   if (EltType) {
     const BuiltinType *BT = EltType->getAs<BuiltinType>();
-    if ((EltType->isVectorType() &&
+    if (IsQPXVectorTy(EltType) || (EltType->isVectorType() &&
          getContext().getTypeSize(EltType) == 128) ||
         (BT && BT->isFloatingPoint()))
       AlignAsType = EltType;
@@ -3400,13 +3477,22 @@ PPC64_SVR4_ABIInfo::isAlignedParamType(QualType Ty) const {
     AlignAsType = Base;
 
   // With special case aggregates, only vector base types need alignment.
-  if (AlignAsType)
+  if (AlignAsType && IsQPXVectorTy(AlignAsType)) {
+    if (getContext().getTypeSize(AlignAsType) > 128)
+      Align32 = true;
+
+    return true;
+  } else if (AlignAsType) {
     return AlignAsType->isVectorType();
+  }
 
   // Otherwise, we only need alignment for any aggregate type that
   // has an alignment requirement of >= 16 bytes.
-  if (isAggregateTypeForABI(Ty) && getContext().getTypeAlign(Ty) >= 128)
+  if (isAggregateTypeForABI(Ty) && getContext().getTypeAlign(Ty) >= 128) {
+    if (HasQPX && getContext().getTypeAlign(Ty) >= 256)
+      Align32 = true;
     return true;
+  }
 
   return false;
 }
@@ -3512,7 +3598,7 @@ bool PPC64_SVR4_ABIInfo::isHomogeneousAggregateBaseType(QualType Ty) const {
       return true;
   }
   if (const VectorType *VT = Ty->getAs<VectorType>()) {
-    if (getContext().getTypeSize(VT) == 128)
+    if (getContext().getTypeSize(VT) == 128 || IsQPXVectorTy(Ty))
       return true;
   }
   return false;
@@ -3538,7 +3624,7 @@ PPC64_SVR4_ABIInfo::classifyArgumentType(QualType Ty) const {
 
   // Non-Altivec vector types are passed in GPRs (smaller than 16 bytes)
   // or via reference (larger than 16 bytes).
-  if (Ty->isVectorType()) {
+  if (Ty->isVectorType() && !IsQPXVectorTy(Ty)) {
     uint64_t Size = getContext().getTypeSize(Ty);
     if (Size > 128)
       return ABIArgInfo::getIndirect(0, /*ByVal=*/false);
@@ -3552,7 +3638,9 @@ PPC64_SVR4_ABIInfo::classifyArgumentType(QualType Ty) const {
     if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI()))
       return ABIArgInfo::getIndirect(0, RAA == CGCXXABI::RAA_DirectInMemory);
 
-    uint64_t ABIAlign = isAlignedParamType(Ty)? 16 : 8;
+    bool Align32;
+    uint64_t ABIAlign = isAlignedParamType(Ty, Align32) ?
+                          (Align32 ? 32 : 16) : 8;
     uint64_t TyAlign = getContext().getTypeAlign(Ty) / 8;
 
     // ELFv2 homogeneous aggregates are passed as array types.
@@ -3609,7 +3697,7 @@ PPC64_SVR4_ABIInfo::classifyReturnType(QualType RetTy) const {
 
   // Non-Altivec vector types are returned in GPRs (smaller than 16 bytes)
   // or via reference (larger than 16 bytes).
-  if (RetTy->isVectorType()) {
+  if (RetTy->isVectorType() && !IsQPXVectorTy(RetTy)) {
     uint64_t Size = getContext().getTypeSize(RetTy);
     if (Size > 128)
       return ABIArgInfo::getIndirect(0);
@@ -3666,10 +3754,13 @@ llvm::Value *PPC64_SVR4_ABIInfo::EmitVAArg(llvm::Value *VAListAddr,
   llvm::Value *Addr = Builder.CreateLoad(VAListAddrAsBPP, "ap.cur");
 
   // Handle types that require 16-byte alignment in the parameter save area.
-  if (isAlignedParamType(Ty)) {
+  bool Align32;
+  if (isAlignedParamType(Ty, Align32)) {
     llvm::Value *AddrAsInt = Builder.CreatePtrToInt(Addr, CGF.Int64Ty);
-    AddrAsInt = Builder.CreateAdd(AddrAsInt, Builder.getInt64(15));
-    AddrAsInt = Builder.CreateAnd(AddrAsInt, Builder.getInt64(-16));
+    AddrAsInt = Builder.CreateAdd(AddrAsInt,
+                                  Builder.getInt64(Align32 ? 31 : 15));
+    AddrAsInt = Builder.CreateAnd(AddrAsInt,
+                                  Builder.getInt64(Align32 ? -32 : -16));
     Addr = Builder.CreateIntToPtr(AddrAsInt, BP, "ap.align");
   }
 
@@ -3714,10 +3805,12 @@ llvm::Value *PPC64_SVR4_ABIInfo::EmitVAArg(llvm::Value *VAListAddr,
     ImagAddr = Builder.CreateIntToPtr(ImagAddr, PBaseTy);
     llvm::Value *Real = Builder.CreateLoad(RealAddr, false, ".vareal");
     llvm::Value *Imag = Builder.CreateLoad(ImagAddr, false, ".vaimag");
-    llvm::Value *Ptr = CGF.CreateTempAlloca(CGT.ConvertTypeForMem(Ty),
-                                            "vacplx");
-    llvm::Value *RealPtr = Builder.CreateStructGEP(Ptr, 0, ".real");
-    llvm::Value *ImagPtr = Builder.CreateStructGEP(Ptr, 1, ".imag");
+    llvm::AllocaInst *Ptr =
+        CGF.CreateTempAlloca(CGT.ConvertTypeForMem(Ty), "vacplx");
+    llvm::Value *RealPtr =
+        Builder.CreateStructGEP(Ptr->getAllocatedType(), Ptr, 0, ".real");
+    llvm::Value *ImagPtr =
+        Builder.CreateStructGEP(Ptr->getAllocatedType(), Ptr, 1, ".imag");
     Builder.CreateStore(Real, RealPtr, false);
     Builder.CreateStore(Imag, ImagPtr, false);
     return Ptr;
@@ -3837,8 +3930,8 @@ private:
   llvm::Value *EmitAAPCSVAArg(llvm::Value *VAListAddr, QualType Ty,
                               CodeGenFunction &CGF) const;
 
-  virtual llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                 CodeGenFunction &CGF) const override {
+  llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
+                         CodeGenFunction &CGF) const override {
     return isDarwinPCS() ? EmitDarwinVAArg(VAListAddr, Ty, CGF)
                          : EmitAAPCSVAArg(VAListAddr, Ty, CGF);
   }
@@ -3849,13 +3942,15 @@ public:
   AArch64TargetCodeGenInfo(CodeGenTypes &CGT, AArch64ABIInfo::ABIKind Kind)
       : TargetCodeGenInfo(new AArch64ABIInfo(CGT, Kind)) {}
 
-  StringRef getARCRetainAutoreleasedReturnValueMarker() const {
+  StringRef getARCRetainAutoreleasedReturnValueMarker() const override {
     return "mov\tfp, fp\t\t; marker for objc_retainAutoreleaseReturnValue";
   }
 
-  int getDwarfEHStackPointer(CodeGen::CodeGenModule &M) const { return 31; }
+  int getDwarfEHStackPointer(CodeGen::CodeGenModule &M) const override {
+    return 31;
+  }
 
-  virtual bool doesReturnSlotInterfereWithArgs() const { return false; }
+  bool doesReturnSlotInterfereWithArgs() const override { return false; }
 };
 }
 
@@ -3964,7 +4059,15 @@ ABIArgInfo AArch64ABIInfo::classifyReturnType(QualType RetTy) const {
   // Aggregates <= 16 bytes are returned directly in registers or on the stack.
   uint64_t Size = getContext().getTypeSize(RetTy);
   if (Size <= 128) {
+    unsigned Alignment = getContext().getTypeAlign(RetTy);
     Size = 64 * ((Size + 63) / 64); // round up to multiple of 8 bytes
+
+    // We use a pair of i64 for 16-byte aggregate with 8-byte alignment.
+    // For aggregates with 16-byte alignment, we use i128.
+    if (Alignment < 128 && Size == 128) {
+      llvm::Type *BaseTy = llvm::Type::getInt64Ty(getVMContext());
+      return ABIArgInfo::getDirect(llvm::ArrayType::get(BaseTy, Size / 64));
+    }
     return ABIArgInfo::getDirect(llvm::IntegerType::get(getVMContext(), Size));
   }
 
@@ -4047,13 +4150,15 @@ llvm::Value *AArch64ABIInfo::EmitAAPCSVAArg(llvm::Value *VAListAddr,
   int RegSize = IsIndirect ? 8 : getContext().getTypeSize(Ty) / 8;
   if (!IsFPR) {
     // 3 is the field number of __gr_offs
-    reg_offs_p = CGF.Builder.CreateStructGEP(VAListAddr, 3, "gr_offs_p");
+    reg_offs_p =
+        CGF.Builder.CreateStructGEP(nullptr, VAListAddr, 3, "gr_offs_p");
     reg_offs = CGF.Builder.CreateLoad(reg_offs_p, "gr_offs");
     reg_top_index = 1; // field number for __gr_top
     RegSize = llvm::RoundUpToAlignment(RegSize, 8);
   } else {
     // 4 is the field number of __vr_offs.
-    reg_offs_p = CGF.Builder.CreateStructGEP(VAListAddr, 4, "vr_offs_p");
+    reg_offs_p =
+        CGF.Builder.CreateStructGEP(nullptr, VAListAddr, 4, "vr_offs_p");
     reg_offs = CGF.Builder.CreateLoad(reg_offs_p, "vr_offs");
     reg_top_index = 2; // field number for __vr_top
     RegSize = 16 * NumRegs;
@@ -4114,8 +4219,8 @@ llvm::Value *AArch64ABIInfo::EmitAAPCSVAArg(llvm::Value *VAListAddr,
   CGF.EmitBlock(InRegBlock);
 
   llvm::Value *reg_top_p = nullptr, *reg_top = nullptr;
-  reg_top_p =
-      CGF.Builder.CreateStructGEP(VAListAddr, reg_top_index, "reg_top_p");
+  reg_top_p = CGF.Builder.CreateStructGEP(nullptr, VAListAddr, reg_top_index,
+                                          "reg_top_p");
   reg_top = CGF.Builder.CreateLoad(reg_top_p, "reg_top");
   llvm::Value *BaseAddr = CGF.Builder.CreateGEP(reg_top, reg_offs);
   llvm::Value *RegAddr = nullptr;
@@ -4138,7 +4243,7 @@ llvm::Value *AArch64ABIInfo::EmitAAPCSVAArg(llvm::Value *VAListAddr,
     assert(!IsIndirect && "Homogeneous aggregates should be passed directly");
     llvm::Type *BaseTy = CGF.ConvertType(QualType(Base, 0));
     llvm::Type *HFATy = llvm::ArrayType::get(BaseTy, NumMembers);
-    llvm::Value *Tmp = CGF.CreateTempAlloca(HFATy);
+    llvm::AllocaInst *Tmp = CGF.CreateTempAlloca(HFATy);
     int Offset = 0;
 
     if (CGF.CGM.getDataLayout().isBigEndian() && Ctx.getTypeSize(Base) < 128)
@@ -4149,7 +4254,8 @@ llvm::Value *AArch64ABIInfo::EmitAAPCSVAArg(llvm::Value *VAListAddr,
       llvm::Value *LoadAddr = CGF.Builder.CreateGEP(BaseAddr, BaseOffset);
       LoadAddr = CGF.Builder.CreateBitCast(
           LoadAddr, llvm::PointerType::getUnqual(BaseTy));
-      llvm::Value *StoreAddr = CGF.Builder.CreateStructGEP(Tmp, i);
+      llvm::Value *StoreAddr =
+          CGF.Builder.CreateStructGEP(Tmp->getAllocatedType(), Tmp, i);
 
       llvm::Value *Elem = CGF.Builder.CreateLoad(LoadAddr);
       CGF.Builder.CreateStore(Elem, StoreAddr);
@@ -4182,7 +4288,7 @@ llvm::Value *AArch64ABIInfo::EmitAAPCSVAArg(llvm::Value *VAListAddr,
   CGF.EmitBlock(OnStackBlock);
 
   llvm::Value *stack_p = nullptr, *OnStackAddr = nullptr;
-  stack_p = CGF.Builder.CreateStructGEP(VAListAddr, 0, "stack_p");
+  stack_p = CGF.Builder.CreateStructGEP(nullptr, VAListAddr, 0, "stack_p");
   OnStackAddr = CGF.Builder.CreateLoad(stack_p, "stack");
 
   // Again, stack arguments may need realigmnent. In this case both integer and
@@ -4324,17 +4430,10 @@ public:
 
 private:
   ABIKind Kind;
-  mutable int VFPRegs[16];
-  const unsigned NumVFPs;
-  const unsigned NumGPRs;
-  mutable unsigned AllocatedGPRs;
-  mutable unsigned AllocatedVFPs;
 
 public:
-  ARMABIInfo(CodeGenTypes &CGT, ABIKind _Kind) : ABIInfo(CGT), Kind(_Kind),
-    NumVFPs(16), NumGPRs(4) {
+  ARMABIInfo(CodeGenTypes &CGT, ABIKind _Kind) : ABIInfo(CGT), Kind(_Kind) {
     setCCs();
-    resetAllocatedRegs();
   }
 
   bool isEABI() const {
@@ -4364,8 +4463,7 @@ public:
 
 private:
   ABIArgInfo classifyReturnType(QualType RetTy, bool isVariadic) const;
-  ABIArgInfo classifyArgumentType(QualType RetTy, bool isVariadic,
-                                  bool &IsCPRC) const;
+  ABIArgInfo classifyArgumentType(QualType RetTy, bool isVariadic) const;
   bool isIllegalVectorType(QualType Ty) const;
 
   bool isHomogeneousAggregateBaseType(QualType Ty) const override;
@@ -4380,10 +4478,6 @@ private:
   llvm::CallingConv::ID getLLVMDefaultCC() const;
   llvm::CallingConv::ID getABIDefaultCC() const;
   void setCCs();
-
-  void markAllocatedGPRs(unsigned Alignment, unsigned NumRequired) const;
-  void markAllocatedVFPs(unsigned Alignment, unsigned NumRequired) const;
-  void resetAllocatedRegs(void) const;
 };
 
 class ARMTargetCodeGenInfo : public TargetCodeGenInfo {
@@ -4456,55 +4550,43 @@ public:
   }
 };
 
+class WindowsARMTargetCodeGenInfo : public ARMTargetCodeGenInfo {
+  void addStackProbeSizeTargetAttribute(const Decl *D, llvm::GlobalValue *GV,
+                                        CodeGen::CodeGenModule &CGM) const;
+
+public:
+  WindowsARMTargetCodeGenInfo(CodeGenTypes &CGT, ARMABIInfo::ABIKind K)
+      : ARMTargetCodeGenInfo(CGT, K) {}
+
+  void SetTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
+                           CodeGen::CodeGenModule &CGM) const override;
+};
+
+void WindowsARMTargetCodeGenInfo::addStackProbeSizeTargetAttribute(
+    const Decl *D, llvm::GlobalValue *GV, CodeGen::CodeGenModule &CGM) const {
+  if (!isa<FunctionDecl>(D))
+    return;
+  if (CGM.getCodeGenOpts().StackProbeSize == 4096)
+    return;
+
+  llvm::Function *F = cast<llvm::Function>(GV);
+  F->addFnAttr("stack-probe-size",
+               llvm::utostr(CGM.getCodeGenOpts().StackProbeSize));
+}
+
+void WindowsARMTargetCodeGenInfo::SetTargetAttributes(
+    const Decl *D, llvm::GlobalValue *GV, CodeGen::CodeGenModule &CGM) const {
+  ARMTargetCodeGenInfo::SetTargetAttributes(D, GV, CGM);
+  addStackProbeSizeTargetAttribute(D, GV, CGM);
+}
 }
 
 void ARMABIInfo::computeInfo(CGFunctionInfo &FI) const {
-  // To correctly handle Homogeneous Aggregate, we need to keep track of the
-  // VFP registers allocated so far.
-  // C.1.vfp If the argument is a VFP CPRC and there are sufficient consecutive
-  // VFP registers of the appropriate type unallocated then the argument is
-  // allocated to the lowest-numbered sequence of such registers.
-  // C.2.vfp If the argument is a VFP CPRC then any VFP registers that are
-  // unallocated are marked as unavailable. 
-  resetAllocatedRegs();
-
-  if (getCXXABI().classifyReturnType(FI)) {
-    if (FI.getReturnInfo().isIndirect())
-      markAllocatedGPRs(1, 1);
-  } else {
+  if (!getCXXABI().classifyReturnType(FI))
     FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), FI.isVariadic());
-  }
-  for (auto &I : FI.arguments()) {
-    unsigned PreAllocationVFPs = AllocatedVFPs;
-    unsigned PreAllocationGPRs = AllocatedGPRs;
-    bool IsCPRC = false;
-    // 6.1.2.3 There is one VFP co-processor register class using registers
-    // s0-s15 (d0-d7) for passing arguments.
-    I.info = classifyArgumentType(I.type, FI.isVariadic(), IsCPRC);
-
-    // If we have allocated some arguments onto the stack (due to running
-    // out of VFP registers), we cannot split an argument between GPRs and
-    // the stack. If this situation occurs, we add padding to prevent the
-    // GPRs from being used. In this situation, the current argument could
-    // only be allocated by rule C.8, so rule C.6 would mark these GPRs as
-    // unusable anyway.
-    // We do not have to do this if the argument is being passed ByVal, as the
-    // backend can handle that situation correctly.
-    const bool StackUsed = PreAllocationGPRs > NumGPRs || PreAllocationVFPs > NumVFPs;
-    const bool IsByVal = I.info.isIndirect() && I.info.getIndirectByVal();
-    if (!IsCPRC && PreAllocationGPRs < NumGPRs && AllocatedGPRs > NumGPRs &&
-        StackUsed && !IsByVal) {
-      llvm::Type *PaddingTy = llvm::ArrayType::get(
-          llvm::Type::getInt32Ty(getVMContext()), NumGPRs - PreAllocationGPRs);
-      if (I.info.canHaveCoerceToType()) {
-        I.info = ABIArgInfo::getDirect(I.info.getCoerceToType() /* type */,
-                                       0 /* offset */, PaddingTy, true);
-      } else {
-        I.info = ABIArgInfo::getDirect(nullptr /* type */, 0 /* offset */,
-                                       PaddingTy, true);
-      }
-    }
-  }
+
+  for (auto &I : FI.arguments())
+    I.info = classifyArgumentType(I.type, FI.isVariadic());
 
   // Always honor user-specified calling convention.
   if (FI.getCallingConvention() != llvm::CallingConv::C)
@@ -4512,7 +4594,7 @@ void ARMABIInfo::computeInfo(CGFunctionInfo &FI) const {
 
   llvm::CallingConv::ID cc = getRuntimeCC();
   if (cc != llvm::CallingConv::C)
-    FI.setEffectiveCallingConvention(cc);    
+    FI.setEffectiveCallingConvention(cc);
 }
 
 /// Return the default calling convention that LLVM will use.
@@ -4550,64 +4632,8 @@ void ARMABIInfo::setCCs() {
                llvm::CallingConv::ARM_APCS : llvm::CallingConv::ARM_AAPCS);
 }
 
-/// markAllocatedVFPs - update VFPRegs according to the alignment and
-/// number of VFP registers (unit is S register) requested.
-void ARMABIInfo::markAllocatedVFPs(unsigned Alignment,
-                                   unsigned NumRequired) const {
-  // Early Exit.
-  if (AllocatedVFPs >= 16) {
-    // We use AllocatedVFP > 16 to signal that some CPRCs were allocated on
-    // the stack.
-    AllocatedVFPs = 17;
-    return;
-  }
-  // C.1.vfp If the argument is a VFP CPRC and there are sufficient consecutive
-  // VFP registers of the appropriate type unallocated then the argument is
-  // allocated to the lowest-numbered sequence of such registers.
-  for (unsigned I = 0; I < 16; I += Alignment) {
-    bool FoundSlot = true;
-    for (unsigned J = I, JEnd = I + NumRequired; J < JEnd; J++)
-      if (J >= 16 || VFPRegs[J]) {
-         FoundSlot = false;
-         break;
-      }
-    if (FoundSlot) {
-      for (unsigned J = I, JEnd = I + NumRequired; J < JEnd; J++)
-        VFPRegs[J] = 1;
-      AllocatedVFPs += NumRequired;
-      return;
-    }
-  }
-  // C.2.vfp If the argument is a VFP CPRC then any VFP registers that are
-  // unallocated are marked as unavailable.
-  for (unsigned I = 0; I < 16; I++)
-    VFPRegs[I] = 1;
-  AllocatedVFPs = 17; // We do not have enough VFP registers.
-}
-
-/// Update AllocatedGPRs to record the number of general purpose registers
-/// which have been allocated. It is valid for AllocatedGPRs to go above 4,
-/// this represents arguments being stored on the stack.
-void ARMABIInfo::markAllocatedGPRs(unsigned Alignment,
-                                   unsigned NumRequired) const {
-  assert((Alignment == 1 || Alignment == 2) && "Alignment must be 4 or 8 bytes");
-
-  if (Alignment == 2 && AllocatedGPRs & 0x1)
-    AllocatedGPRs += 1;
-
-  AllocatedGPRs += NumRequired;
-}
-
-void ARMABIInfo::resetAllocatedRegs(void) const {
-  AllocatedGPRs = 0;
-  AllocatedVFPs = 0;
-  for (unsigned i = 0; i < NumVFPs; ++i)
-    VFPRegs[i] = 0;
-}
-
-ABIArgInfo ARMABIInfo::classifyArgumentType(QualType Ty, bool isVariadic,
-                                            bool &IsCPRC) const {
-  // We update number of allocated VFPs according to
+ABIArgInfo ARMABIInfo::classifyArgumentType(QualType Ty,
+                                            bool isVariadic) const {
   // 6.1.2.1 The following argument types are VFP CPRCs:
   //   A single-precision floating-point type (including promoted
   //   half-precision types); A double-precision floating-point type;
@@ -4625,58 +4651,20 @@ ABIArgInfo ARMABIInfo::classifyArgumentType(QualType Ty, bool isVariadic,
     if (Size <= 32) {
       llvm::Type *ResType =
           llvm::Type::getInt32Ty(getVMContext());
-      markAllocatedGPRs(1, 1);
       return ABIArgInfo::getDirect(ResType);
     }
     if (Size == 64) {
       llvm::Type *ResType = llvm::VectorType::get(
           llvm::Type::getInt32Ty(getVMContext()), 2);
-      if (getABIKind() == ARMABIInfo::AAPCS || isVariadic){
-        markAllocatedGPRs(2, 2);
-      } else {
-        markAllocatedVFPs(2, 2);
-        IsCPRC = true;
-      }
       return ABIArgInfo::getDirect(ResType);
     }
     if (Size == 128) {
       llvm::Type *ResType = llvm::VectorType::get(
           llvm::Type::getInt32Ty(getVMContext()), 4);
-      if (getABIKind() == ARMABIInfo::AAPCS || isVariadic) {
-        markAllocatedGPRs(2, 4);
-      } else {
-        markAllocatedVFPs(4, 4);
-        IsCPRC = true;
-      }
       return ABIArgInfo::getDirect(ResType);
     }
-    markAllocatedGPRs(1, 1);
     return ABIArgInfo::getIndirect(0, /*ByVal=*/false);
   }
-  // Update VFPRegs for legal vector types.
-  if (getABIKind() == ARMABIInfo::AAPCS_VFP && !isVariadic) {
-    if (const VectorType *VT = Ty->getAs<VectorType>()) {
-      uint64_t Size = getContext().getTypeSize(VT);
-      // Size of a legal vector should be power of 2 and above 64.
-      markAllocatedVFPs(Size >= 128 ? 4 : 2, Size / 32);
-      IsCPRC = true;
-    }
-  }
-  // Update VFPRegs for floating point types.
-  if (getABIKind() == ARMABIInfo::AAPCS_VFP && !isVariadic) {
-    if (const BuiltinType *BT = Ty->getAs<BuiltinType>()) {
-      if (BT->getKind() == BuiltinType::Half ||
-          BT->getKind() == BuiltinType::Float) {
-        markAllocatedVFPs(1, 1);
-        IsCPRC = true;
-      }
-      if (BT->getKind() == BuiltinType::Double ||
-          BT->getKind() == BuiltinType::LongDouble) {
-        markAllocatedVFPs(2, 2);
-        IsCPRC = true;
-      }
-    }
-  }
 
   if (!isAggregateTypeForABI(Ty)) {
     // Treat an enum type as its underlying type.
@@ -4684,15 +4672,11 @@ ABIArgInfo ARMABIInfo::classifyArgumentType(QualType Ty, bool isVariadic,
       Ty = EnumTy->getDecl()->getIntegerType();
     }
 
-    unsigned Size = getContext().getTypeSize(Ty);
-    if (!IsCPRC)
-      markAllocatedGPRs(Size > 32 ? 2 : 1, (Size + 31) / 32);
     return (Ty->isPromotableIntegerType() ? ABIArgInfo::getExtend()
                                           : ABIArgInfo::getDirect());
   }
 
   if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI())) {
-    markAllocatedGPRs(1, 1);
     return ABIArgInfo::getIndirect(0, RAA == CGCXXABI::RAA_DirectInMemory);
   }
 
@@ -4708,19 +4692,6 @@ ABIArgInfo ARMABIInfo::classifyArgumentType(QualType Ty, bool isVariadic,
     if (isHomogeneousAggregate(Ty, Base, Members)) {
       assert(Base && "Base class should be set for homogeneous aggregate");
       // Base can be a floating-point or a vector.
-      if (Base->isVectorType()) {
-        // ElementSize is in number of floats.
-        unsigned ElementSize = getContext().getTypeSize(Base) == 64 ? 2 : 4;
-        markAllocatedVFPs(ElementSize,
-                          Members * ElementSize);
-      } else if (Base->isSpecificBuiltinType(BuiltinType::Float))
-        markAllocatedVFPs(1, Members);
-      else {
-        assert(Base->isSpecificBuiltinType(BuiltinType::Double) ||
-               Base->isSpecificBuiltinType(BuiltinType::LongDouble));
-        markAllocatedVFPs(2, Members * 2);
-      }
-      IsCPRC = true;
       return ABIArgInfo::getDirect(nullptr, 0, nullptr, false);
     }
   }
@@ -4732,15 +4703,11 @@ ABIArgInfo ARMABIInfo::classifyArgumentType(QualType Ty, bool isVariadic,
   uint64_t ABIAlign = 4;
   uint64_t TyAlign = getContext().getTypeAlign(Ty) / 8;
   if (getABIKind() == ARMABIInfo::AAPCS_VFP ||
-      getABIKind() == ARMABIInfo::AAPCS)
+       getABIKind() == ARMABIInfo::AAPCS)
     ABIAlign = std::min(std::max(TyAlign, (uint64_t)4), (uint64_t)8);
+
   if (getContext().getTypeSizeInChars(Ty) > CharUnits::fromQuantity(64)) {
-    // Update Allocated GPRs. Since this is only used when the size of the
-    // argument is greater than 64 bytes, this will always use up any available
-    // registers (of which there are 4). We also don't care about getting the
-    // alignment right, because general-purpose registers cannot be back-filled.
-    markAllocatedGPRs(1, 4);
-    return ABIArgInfo::getIndirect(TyAlign, /*ByVal=*/true,
+    return ABIArgInfo::getIndirect(ABIAlign, /*ByVal=*/true,
            /*Realign=*/TyAlign > ABIAlign);
   }
 
@@ -4752,11 +4719,9 @@ ABIArgInfo ARMABIInfo::classifyArgumentType(QualType Ty, bool isVariadic,
   if (getContext().getTypeAlign(Ty) <= 32) {
     ElemTy = llvm::Type::getInt32Ty(getVMContext());
     SizeRegs = (getContext().getTypeSize(Ty) + 31) / 32;
-    markAllocatedGPRs(1, SizeRegs);
   } else {
     ElemTy = llvm::Type::getInt64Ty(getVMContext());
     SizeRegs = (getContext().getTypeSize(Ty) + 63) / 64;
-    markAllocatedGPRs(2, SizeRegs * 2);
   }
 
   return ABIArgInfo::getDirect(llvm::ArrayType::get(ElemTy, SizeRegs));
@@ -4856,7 +4821,6 @@ ABIArgInfo ARMABIInfo::classifyReturnType(QualType RetTy,
 
   // Large vector types should be returned via memory.
   if (RetTy->isVectorType() && getContext().getTypeSize(RetTy) > 128) {
-    markAllocatedGPRs(1, 1);
     return ABIArgInfo::getIndirect(0);
   }
 
@@ -4894,7 +4858,6 @@ ABIArgInfo ARMABIInfo::classifyReturnType(QualType RetTy,
     }
 
     // Otherwise return in memory.
-    markAllocatedGPRs(1, 1);
     return ABIArgInfo::getIndirect(0);
   }
 
@@ -4930,7 +4893,6 @@ ABIArgInfo ARMABIInfo::classifyReturnType(QualType RetTy,
     return ABIArgInfo::getDirect(llvm::Type::getInt32Ty(getVMContext()));
   }
 
-  markAllocatedGPRs(1, 1);
   return ABIArgInfo::getIndirect(0);
 }
 
@@ -5046,42 +5008,6 @@ llvm::Value *ARMABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
   return AddrTyped;
 }
 
-namespace {
-
-class NaClARMABIInfo : public ABIInfo {
- public:
-  NaClARMABIInfo(CodeGen::CodeGenTypes &CGT, ARMABIInfo::ABIKind Kind)
-      : ABIInfo(CGT), PInfo(CGT), NInfo(CGT, Kind) {}
-  void computeInfo(CGFunctionInfo &FI) const override;
-  llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                         CodeGenFunction &CGF) const override;
- private:
-  PNaClABIInfo PInfo; // Used for generating calls with pnaclcall callingconv.
-  ARMABIInfo NInfo; // Used for everything else.
-};
-
-class NaClARMTargetCodeGenInfo : public TargetCodeGenInfo  {
- public:
-  NaClARMTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, ARMABIInfo::ABIKind Kind)
-      : TargetCodeGenInfo(new NaClARMABIInfo(CGT, Kind)) {}
-};
-
-}
-
-void NaClARMABIInfo::computeInfo(CGFunctionInfo &FI) const {
-  if (FI.getASTCallingConvention() == CC_PnaclCall)
-    PInfo.computeInfo(FI);
-  else
-    static_cast<const ABIInfo&>(NInfo).computeInfo(FI);
-}
-
-llvm::Value *NaClARMABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
-                                       CodeGenFunction &CGF) const {
-  // Always use the native convention; calling pnacl-style varargs functions
-  // is unsupported.
-  return static_cast<const ABIInfo&>(NInfo).EmitVAArg(VAListAddr, Ty, CGF);
-}
-
 //===----------------------------------------------------------------------===//
 // NVPTX ABI Implementation
 //===----------------------------------------------------------------------===//
@@ -5190,18 +5116,22 @@ SetTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
       // Create !{<func-ref>, metadata !"kernel", i32 1} node
       addNVVMMetadata(F, "kernel", 1);
     }
-    if (FD->hasAttr<CUDALaunchBoundsAttr>()) {
+    if (CUDALaunchBoundsAttr *Attr = FD->getAttr<CUDALaunchBoundsAttr>()) {
       // Create !{<func-ref>, metadata !"maxntidx", i32 <val>} node
-      addNVVMMetadata(F, "maxntidx",
-                      FD->getAttr<CUDALaunchBoundsAttr>()->getMaxThreads());
-      // min blocks is a default argument for CUDALaunchBoundsAttr, so getting a
-      // zero value from getMinBlocks either means it was not specified in
-      // __launch_bounds__ or the user specified a 0 value. In both cases, we
-      // don't have to add a PTX directive.
-      int MinCTASM = FD->getAttr<CUDALaunchBoundsAttr>()->getMinBlocks();
-      if (MinCTASM > 0) {
-        // Create !{<func-ref>, metadata !"minctasm", i32 <val>} node
-        addNVVMMetadata(F, "minctasm", MinCTASM);
+      llvm::APSInt MaxThreads(32);
+      MaxThreads = Attr->getMaxThreads()->EvaluateKnownConstInt(M.getContext());
+      if (MaxThreads > 0)
+        addNVVMMetadata(F, "maxntidx", MaxThreads.getExtValue());
+
+      // min blocks is an optional argument for CUDALaunchBoundsAttr. If it was
+      // not specified in __launch_bounds__ or if the user specified a 0 value,
+      // we don't have to add a PTX directive.
+      if (Attr->getMinBlocks()) {
+        llvm::APSInt MinBlocks(32);
+        MinBlocks = Attr->getMinBlocks()->EvaluateKnownConstInt(M.getContext());
+        if (MinBlocks > 0)
+          // Create !{<func-ref>, metadata !"minctasm", i32 <val>} node
+          addNVVMMetadata(F, "minctasm", MinBlocks.getExtValue());
       }
     }
   }
@@ -5231,12 +5161,17 @@ void NVPTXTargetCodeGenInfo::addNVVMMetadata(llvm::Function *F, StringRef Name,
 namespace {
 
 class SystemZABIInfo : public ABIInfo {
+  bool HasVector;
+
 public:
-  SystemZABIInfo(CodeGenTypes &CGT) : ABIInfo(CGT) {}
+  SystemZABIInfo(CodeGenTypes &CGT, bool HV)
+    : ABIInfo(CGT), HasVector(HV) {}
 
   bool isPromotableIntegerType(QualType Ty) const;
   bool isCompoundType(QualType Ty) const;
+  bool isVectorArgumentType(QualType Ty) const;
   bool isFPArgumentType(QualType Ty) const;
+  QualType GetSingleElementType(QualType Ty) const;
 
   ABIArgInfo classifyReturnType(QualType RetTy) const;
   ABIArgInfo classifyArgumentType(QualType ArgTy) const;
@@ -5254,8 +5189,8 @@ public:
 
 class SystemZTargetCodeGenInfo : public TargetCodeGenInfo {
 public:
-  SystemZTargetCodeGenInfo(CodeGenTypes &CGT)
-    : TargetCodeGenInfo(new SystemZABIInfo(CGT)) {}
+  SystemZTargetCodeGenInfo(CodeGenTypes &CGT, bool HasVector)
+    : TargetCodeGenInfo(new SystemZABIInfo(CGT, HasVector)) {}
 };
 
 }
@@ -5282,7 +5217,15 @@ bool SystemZABIInfo::isPromotableIntegerType(QualType Ty) const {
 }
 
 bool SystemZABIInfo::isCompoundType(QualType Ty) const {
-  return Ty->isAnyComplexType() || isAggregateTypeForABI(Ty);
+  return (Ty->isAnyComplexType() ||
+          Ty->isVectorType() ||
+          isAggregateTypeForABI(Ty));
+}
+
+bool SystemZABIInfo::isVectorArgumentType(QualType Ty) const {
+  return (HasVector &&
+          Ty->isVectorType() &&
+          getContext().getTypeSize(Ty) <= 128);
 }
 
 bool SystemZABIInfo::isFPArgumentType(QualType Ty) const {
@@ -5295,9 +5238,13 @@ bool SystemZABIInfo::isFPArgumentType(QualType Ty) const {
       return false;
     }
 
+  return false;
+}
+
+QualType SystemZABIInfo::GetSingleElementType(QualType Ty) const {
   if (const RecordType *RT = Ty->getAsStructureType()) {
     const RecordDecl *RD = RT->getDecl();
-    bool Found = false;
+    QualType Found;
 
     // If this is a C++ record, check the bases first.
     if (const CXXRecordDecl *CXXRD = dyn_cast<CXXRecordDecl>(RD))
@@ -5308,36 +5255,34 @@ bool SystemZABIInfo::isFPArgumentType(QualType Ty) const {
         if (isEmptyRecord(getContext(), Base, true))
           continue;
 
-        if (Found)
-          return false;
-        Found = isFPArgumentType(Base);
-        if (!Found)
-          return false;
+        if (!Found.isNull())
+          return Ty;
+        Found = GetSingleElementType(Base);
       }
 
     // Check the fields.
     for (const auto *FD : RD->fields()) {
-      // Empty bitfields don't affect things either way.
+      // For compatibility with GCC, ignore empty bitfields in C++ mode.
       // Unlike isSingleElementStruct(), empty structure and array fields
       // do count.  So do anonymous bitfields that aren't zero-sized.
-      if (FD->isBitField() && FD->getBitWidthValue(getContext()) == 0)
-        return true;
+      if (getContext().getLangOpts().CPlusPlus &&
+          FD->isBitField() && FD->getBitWidthValue(getContext()) == 0)
+        continue;
 
       // Unlike isSingleElementStruct(), arrays do not count.
-      // Nested isFPArgumentType structures still do though.
-      if (Found)
-        return false;
-      Found = isFPArgumentType(FD->getType());
-      if (!Found)
-        return false;
+      // Nested structures still do though.
+      if (!Found.isNull())
+        return Ty;
+      Found = GetSingleElementType(FD->getType());
     }
 
     // Unlike isSingleElementStruct(), trailing padding is allowed.
     // An 8-byte aligned struct s { float f; } is passed as a double.
-    return Found;
+    if (!Found.isNull())
+      return Found;
   }
 
-  return false;
+  return Ty;
 }
 
 llvm::Value *SystemZABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
@@ -5350,26 +5295,56 @@ llvm::Value *SystemZABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
   //   i8 *__reg_save_area;
   // };
 
-  // Every argument occupies 8 bytes and is passed by preference in either
-  // GPRs or FPRs.
+  // Every non-vector argument occupies 8 bytes and is passed by preference
+  // in either GPRs or FPRs.  Vector arguments occupy 8 or 16 bytes and are
+  // always passed on the stack.
   Ty = CGF.getContext().getCanonicalType(Ty);
+  llvm::Type *ArgTy = CGF.ConvertTypeForMem(Ty);
+  llvm::Type *APTy = llvm::PointerType::getUnqual(ArgTy);
   ABIArgInfo AI = classifyArgumentType(Ty);
-  bool InFPRs = isFPArgumentType(Ty);
-
-  llvm::Type *APTy = llvm::PointerType::getUnqual(CGF.ConvertTypeForMem(Ty));
   bool IsIndirect = AI.isIndirect();
+  bool InFPRs = false;
+  bool IsVector = false;
   unsigned UnpaddedBitSize;
   if (IsIndirect) {
     APTy = llvm::PointerType::getUnqual(APTy);
     UnpaddedBitSize = 64;
-  } else
+  } else {
+    if (AI.getCoerceToType())
+      ArgTy = AI.getCoerceToType();
+    InFPRs = ArgTy->isFloatTy() || ArgTy->isDoubleTy();
+    IsVector = ArgTy->isVectorTy();
     UnpaddedBitSize = getContext().getTypeSize(Ty);
-  unsigned PaddedBitSize = 64;
+  }
+  unsigned PaddedBitSize = (IsVector && UnpaddedBitSize > 64) ? 128 : 64;
   assert((UnpaddedBitSize <= PaddedBitSize) && "Invalid argument size.");
 
   unsigned PaddedSize = PaddedBitSize / 8;
   unsigned Padding = (PaddedBitSize - UnpaddedBitSize) / 8;
 
+  llvm::Type *IndexTy = CGF.Int64Ty;
+  llvm::Value *PaddedSizeV = llvm::ConstantInt::get(IndexTy, PaddedSize);
+
+  if (IsVector) {
+    // Work out the address of a vector argument on the stack.
+    // Vector arguments are always passed in the high bits of a
+    // single (8 byte) or double (16 byte) stack slot.
+    llvm::Value *OverflowArgAreaPtr =
+      CGF.Builder.CreateStructGEP(nullptr, VAListAddr, 2,
+                                  "overflow_arg_area_ptr");
+    llvm::Value *OverflowArgArea =
+      CGF.Builder.CreateLoad(OverflowArgAreaPtr, "overflow_arg_area");
+    llvm::Value *MemAddr =
+      CGF.Builder.CreateBitCast(OverflowArgArea, APTy, "mem_addr");
+
+    // Update overflow_arg_area_ptr pointer
+    llvm::Value *NewOverflowArgArea =
+      CGF.Builder.CreateGEP(OverflowArgArea, PaddedSizeV, "overflow_arg_area");
+    CGF.Builder.CreateStore(NewOverflowArgArea, OverflowArgAreaPtr);
+
+    return MemAddr;
+  }
+
   unsigned MaxRegs, RegCountField, RegSaveIndex, RegPadding;
   if (InFPRs) {
     MaxRegs = 4; // Maximum of 4 FPR arguments
@@ -5383,10 +5358,9 @@ llvm::Value *SystemZABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
     RegPadding = Padding; // values are passed in the low bits of a GPR
   }
 
-  llvm::Value *RegCountPtr =
-    CGF.Builder.CreateStructGEP(VAListAddr, RegCountField, "reg_count_ptr");
+  llvm::Value *RegCountPtr = CGF.Builder.CreateStructGEP(
+      nullptr, VAListAddr, RegCountField, "reg_count_ptr");
   llvm::Value *RegCount = CGF.Builder.CreateLoad(RegCountPtr, "reg_count");
-  llvm::Type *IndexTy = RegCount->getType();
   llvm::Value *MaxRegsV = llvm::ConstantInt::get(IndexTy, MaxRegs);
   llvm::Value *InRegs = CGF.Builder.CreateICmpULT(RegCount, MaxRegsV,
                                                  "fits_in_regs");
@@ -5400,7 +5374,6 @@ llvm::Value *SystemZABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
   CGF.EmitBlock(InRegBlock);
 
   // Work out the address of an argument register.
-  llvm::Value *PaddedSizeV = llvm::ConstantInt::get(IndexTy, PaddedSize);
   llvm::Value *ScaledRegCount =
     CGF.Builder.CreateMul(RegCount, PaddedSizeV, "scaled_reg_count");
   llvm::Value *RegBase =
@@ -5408,7 +5381,7 @@ llvm::Value *SystemZABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
   llvm::Value *RegOffset =
     CGF.Builder.CreateAdd(ScaledRegCount, RegBase, "reg_offset");
   llvm::Value *RegSaveAreaPtr =
-    CGF.Builder.CreateStructGEP(VAListAddr, 3, "reg_save_area_ptr");
+      CGF.Builder.CreateStructGEP(nullptr, VAListAddr, 3, "reg_save_area_ptr");
   llvm::Value *RegSaveArea =
     CGF.Builder.CreateLoad(RegSaveAreaPtr, "reg_save_area");
   llvm::Value *RawRegAddr =
@@ -5427,8 +5400,8 @@ llvm::Value *SystemZABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
   CGF.EmitBlock(InMemBlock);
 
   // Work out the address of a stack argument.
-  llvm::Value *OverflowArgAreaPtr =
-    CGF.Builder.CreateStructGEP(VAListAddr, 2, "overflow_arg_area_ptr");
+  llvm::Value *OverflowArgAreaPtr = CGF.Builder.CreateStructGEP(
+      nullptr, VAListAddr, 2, "overflow_arg_area_ptr");
   llvm::Value *OverflowArgArea =
     CGF.Builder.CreateLoad(OverflowArgAreaPtr, "overflow_arg_area");
   llvm::Value *PaddingV = llvm::ConstantInt::get(IndexTy, Padding);
@@ -5458,6 +5431,8 @@ llvm::Value *SystemZABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
 ABIArgInfo SystemZABIInfo::classifyReturnType(QualType RetTy) const {
   if (RetTy->isVoidType())
     return ABIArgInfo::getIgnore();
+  if (isVectorArgumentType(RetTy))
+    return ABIArgInfo::getDirect();
   if (isCompoundType(RetTy) || getContext().getTypeSize(RetTy) > 64)
     return ABIArgInfo::getIndirect(0);
   return (isPromotableIntegerType(RetTy) ?
@@ -5473,8 +5448,16 @@ ABIArgInfo SystemZABIInfo::classifyArgumentType(QualType Ty) const {
   if (isPromotableIntegerType(Ty))
     return ABIArgInfo::getExtend();
 
-  // Values that are not 1, 2, 4 or 8 bytes in size are passed indirectly.
+  // Handle vector types and vector-like structure types.  Note that
+  // as opposed to float-like structure types, we do not allow any
+  // padding for vector-like structures, so verify the sizes match.
   uint64_t Size = getContext().getTypeSize(Ty);
+  QualType SingleElementTy = GetSingleElementType(Ty);
+  if (isVectorArgumentType(SingleElementTy) &&
+      getContext().getTypeSize(SingleElementTy) == Size)
+    return ABIArgInfo::getDirect(CGT.ConvertType(SingleElementTy));
+
+  // Values that are not 1, 2, 4 or 8 bytes in size are passed indirectly.
   if (Size != 8 && Size != 16 && Size != 32 && Size != 64)
     return ABIArgInfo::getIndirect(0, /*ByVal=*/false);
 
@@ -5488,7 +5471,7 @@ ABIArgInfo SystemZABIInfo::classifyArgumentType(QualType Ty) const {
 
     // The structure is passed as an unextended integer, a float, or a double.
     llvm::Type *PassTy;
-    if (isFPArgumentType(Ty)) {
+    if (isFPArgumentType(SingleElementTy)) {
       assert(Size == 32 || Size == 64);
       if (Size == 32)
         PassTy = llvm::Type::getFloatTy(getVMContext());
@@ -5568,6 +5551,7 @@ public:
   void computeInfo(CGFunctionInfo &FI) const override;
   llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
                          CodeGenFunction &CGF) const override;
+  bool shouldSignExtUnsignedType(QualType Ty) const override;
 };
 
 class MIPSTargetCodeGenInfo : public TargetCodeGenInfo {
@@ -5870,6 +5854,16 @@ llvm::Value* MipsABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
   return AddrTyped;
 }
 
+bool MipsABIInfo::shouldSignExtUnsignedType(QualType Ty) const {
+  int TySize = getContext().getTypeSize(Ty);
+  
+  // MIPS64 ABI requires unsigned 32 bit integers to be sign extended.
+  if (Ty->isUnsignedIntegerOrEnumerationType() && TySize == 32)
+    return true;
+  
+  return false;
+}
+
 bool
 MIPSTargetCodeGenInfo::initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                                                llvm::Value *Address) const {
@@ -6264,12 +6258,7 @@ private:
 
     // Check if Ty is a usable substitute for the coercion type.
     bool isUsableType(llvm::StructType *Ty) const {
-      if (Ty->getNumElements() != Elems.size())
-        return false;
-      for (unsigned i = 0, e = Elems.size(); i != e; ++i)
-        if (Elems[i] != Ty->getElementType(i))
-          return false;
-      return true;
+      return llvm::makeArrayRef(Elems) == Ty->elements();
     }
 
     // Get the coercion type as a literal struct type.
@@ -7102,6 +7091,12 @@ const TargetCodeGenInfo &CodeGenModule::getTargetCodeGenInfo() {
   case llvm::Triple::thumb:
   case llvm::Triple::thumbeb:
     {
+      if (Triple.getOS() == llvm::Triple::Win32) {
+        TheTargetCodeGenInfo =
+            new WindowsARMTargetCodeGenInfo(Types, ARMABIInfo::AAPCS_VFP);
+        return *TheTargetCodeGenInfo;
+      }
+
       ARMABIInfo::ABIKind Kind = ARMABIInfo::AAPCS;
       if (getTarget().getABI() == "apcs-gnu")
         Kind = ARMABIInfo::APCS;
@@ -7110,14 +7105,7 @@ const TargetCodeGenInfo &CodeGenModule::getTargetCodeGenInfo() {
                 Triple.getEnvironment() == llvm::Triple::GNUEABIHF))
         Kind = ARMABIInfo::AAPCS_VFP;
 
-      switch (Triple.getOS()) {
-        case llvm::Triple::NaCl:
-          return *(TheTargetCodeGenInfo =
-                   new NaClARMTargetCodeGenInfo(Types, Kind));
-        default:
-          return *(TheTargetCodeGenInfo =
-                   new ARMTargetCodeGenInfo(Types, Kind));
-      }
+      return *(TheTargetCodeGenInfo = new ARMTargetCodeGenInfo(Types, Kind));
     }
 
   case llvm::Triple::ppc:
@@ -7127,19 +7115,21 @@ const TargetCodeGenInfo &CodeGenModule::getTargetCodeGenInfo() {
       PPC64_SVR4_ABIInfo::ABIKind Kind = PPC64_SVR4_ABIInfo::ELFv1;
       if (getTarget().getABI() == "elfv2")
         Kind = PPC64_SVR4_ABIInfo::ELFv2;
+      bool HasQPX = getTarget().getABI() == "elfv1-qpx";
 
       return *(TheTargetCodeGenInfo =
-               new PPC64_SVR4_TargetCodeGenInfo(Types, Kind));
+               new PPC64_SVR4_TargetCodeGenInfo(Types, Kind, HasQPX));
     } else
       return *(TheTargetCodeGenInfo = new PPC64TargetCodeGenInfo(Types));
   case llvm::Triple::ppc64le: {
     assert(Triple.isOSBinFormatELF() && "PPC64 LE non-ELF not supported!");
     PPC64_SVR4_ABIInfo::ABIKind Kind = PPC64_SVR4_ABIInfo::ELFv2;
-    if (getTarget().getABI() == "elfv1")
+    if (getTarget().getABI() == "elfv1" || getTarget().getABI() == "elfv1-qpx")
       Kind = PPC64_SVR4_ABIInfo::ELFv1;
+    bool HasQPX = getTarget().getABI() == "elfv1-qpx";
 
     return *(TheTargetCodeGenInfo =
-             new PPC64_SVR4_TargetCodeGenInfo(Types, Kind));
+             new PPC64_SVR4_TargetCodeGenInfo(Types, Kind, HasQPX));
   }
 
   case llvm::Triple::nvptx:
@@ -7149,8 +7139,11 @@ const TargetCodeGenInfo &CodeGenModule::getTargetCodeGenInfo() {
   case llvm::Triple::msp430:
     return *(TheTargetCodeGenInfo = new MSP430TargetCodeGenInfo(Types));
 
-  case llvm::Triple::systemz:
-    return *(TheTargetCodeGenInfo = new SystemZTargetCodeGenInfo(Types));
+  case llvm::Triple::systemz: {
+    bool HasVector = getTarget().getABI() == "vector";
+    return *(TheTargetCodeGenInfo = new SystemZTargetCodeGenInfo(Types,
+                                                                 HasVector));
+  }
 
   case llvm::Triple::tce:
     return *(TheTargetCodeGenInfo = new TCETargetCodeGenInfo(Types));
@@ -7177,18 +7170,13 @@ const TargetCodeGenInfo &CodeGenModule::getTargetCodeGenInfo() {
   }
 
   case llvm::Triple::x86_64: {
-    bool HasAVX = getTarget().getABI() == "avx";
-
     switch (Triple.getOS()) {
     case llvm::Triple::Win32:
-      return *(TheTargetCodeGenInfo =
-                   new WinX86_64TargetCodeGenInfo(Types, HasAVX));
-    case llvm::Triple::NaCl:
-      return *(TheTargetCodeGenInfo =
-                   new NaClX86_64TargetCodeGenInfo(Types, HasAVX));
+      return *(TheTargetCodeGenInfo = new WinX86_64TargetCodeGenInfo(Types));
+    case llvm::Triple::PS4:
+      return *(TheTargetCodeGenInfo = new PS4TargetCodeGenInfo(Types));
     default:
-      return *(TheTargetCodeGenInfo =
-                   new X86_64TargetCodeGenInfo(Types, HasAVX));
+      return *(TheTargetCodeGenInfo = new X86_64TargetCodeGenInfo(Types));
     }
   }
   case llvm::Triple::hexagon:
diff --git a/lib/Driver/CrossWindowsToolChain.cpp b/lib/Driver/CrossWindowsToolChain.cpp
index 03fe41b74fe0..82456e76c75a 100644
--- a/lib/Driver/CrossWindowsToolChain.cpp
+++ b/lib/Driver/CrossWindowsToolChain.cpp
@@ -60,7 +60,7 @@ AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
   if (!DriverArgs.hasArg(options::OPT_nobuiltininc)) {
     SmallString<128> ResourceDir(D.ResourceDir);
     llvm::sys::path::append(ResourceDir, "include");
-    addSystemInclude(DriverArgs, CC1Args, ResourceDir.str());
+    addSystemInclude(DriverArgs, CC1Args, ResourceDir);
   }
   addExternCSystemInclude(DriverArgs, CC1Args, SysRoot + "/usr/include");
 }
diff --git a/lib/Driver/Driver.cpp b/lib/Driver/Driver.cpp
index 1664d0d49d46..65d0049b0b49 100644
--- a/lib/Driver/Driver.cpp
+++ b/lib/Driver/Driver.cpp
@@ -17,6 +17,7 @@
 #include "clang/Driver/DriverDiagnostic.h"
 #include "clang/Driver/Job.h"
 #include "clang/Driver/Options.h"
+#include "clang/Driver/SanitizerArgs.h"
 #include "clang/Driver/Tool.h"
 #include "clang/Driver/ToolChain.h"
 #include "llvm/ADT/ArrayRef.h"
@@ -44,21 +45,19 @@ using namespace clang::driver;
 using namespace clang;
 using namespace llvm::opt;
 
-Driver::Driver(StringRef ClangExecutable,
-               StringRef DefaultTargetTriple,
+Driver::Driver(StringRef ClangExecutable, StringRef DefaultTargetTriple,
                DiagnosticsEngine &Diags)
-  : Opts(createDriverOptTable()), Diags(Diags), Mode(GCCMode),
-    ClangExecutable(ClangExecutable), SysRoot(DEFAULT_SYSROOT),
-    UseStdLib(true), DefaultTargetTriple(DefaultTargetTriple),
-    DriverTitle("clang LLVM compiler"),
-    CCPrintOptionsFilename(nullptr), CCPrintHeadersFilename(nullptr),
-    CCLogDiagnosticsFilename(nullptr),
-    CCCPrintBindings(false),
-    CCPrintHeaders(false), CCLogDiagnostics(false),
-    CCGenDiagnostics(false), CCCGenericGCCName(""), CheckInputsExist(true),
-    CCCUsePCH(true), SuppressMissingInputWarning(false) {
-
-  Name = llvm::sys::path::stem(ClangExecutable);
+    : Opts(createDriverOptTable()), Diags(Diags), Mode(GCCMode),
+      SaveTemps(SaveTempsNone), ClangExecutable(ClangExecutable),
+      SysRoot(DEFAULT_SYSROOT), UseStdLib(true),
+      DefaultTargetTriple(DefaultTargetTriple),
+      DriverTitle("clang LLVM compiler"), CCPrintOptionsFilename(nullptr),
+      CCPrintHeadersFilename(nullptr), CCLogDiagnosticsFilename(nullptr),
+      CCCPrintBindings(false), CCPrintHeaders(false), CCLogDiagnostics(false),
+      CCGenDiagnostics(false), CCCGenericGCCName(""), CheckInputsExist(true),
+      CCCUsePCH(true), SuppressMissingInputWarning(false) {
+
+  Name = llvm::sys::path::filename(ClangExecutable);
   Dir  = llvm::sys::path::parent_path(ClangExecutable);
 
   // Compute the path to the resource directory.
@@ -364,6 +363,13 @@ Compilation *Driver::BuildCompilation(ArrayRef<const char *> ArgList) {
   if (const Arg *A = Args->getLastArg(options::OPT_resource_dir))
     ResourceDir = A->getValue();
 
+  if (const Arg *A = Args->getLastArg(options::OPT_save_temps_EQ)) {
+    SaveTemps = llvm::StringSwitch<SaveTempsMode>(A->getValue())
+                    .Case("cwd", SaveTempsCwd)
+                    .Case("obj", SaveTempsObj)
+                    .Default(SaveTempsCwd);
+  }
+
   // Perform the default argument translations.
   DerivedArgList *TranslatedArgs = TranslateInputArgs(*Args);
 
@@ -504,7 +510,7 @@ void Driver::generateCompilationDiagnostics(Compilation &C,
 
   // If any of the preprocessing commands failed, clean up and exit.
   if (!FailingCommands.empty()) {
-    if (!C.getArgs().hasArg(options::OPT_save_temps))
+    if (!isSaveTempsEnabled())
       C.CleanupFileList(C.getTempFiles(), true);
 
     Diag(clang::diag::note_drv_command_failed_diag_msg)
@@ -545,6 +551,9 @@ void Driver::generateCompilationDiagnostics(Compilation &C,
     Diag(clang::diag::note_drv_command_failed_diag_msg)
         << "Error generating run script: " + Script + " " + EC.message();
   } else {
+    ScriptOS << "# Crash reproducer for " << getClangFullVersion() << "\n"
+             << "# Original command: ";
+    Cmd.Print(ScriptOS, "\n", /*Quote=*/true);
     Cmd.Print(ScriptOS, "\n", /*Quote=*/true, &CrashInfo);
     Diag(clang::diag::note_drv_command_failed_diag_msg) << Script;
   }
@@ -612,7 +621,7 @@ int Driver::ExecuteCompilation(Compilation &C,
     const Command *FailingCommand = it->second;
 
     // Remove result files if we're not saving temps.
-    if (!C.getArgs().hasArg(options::OPT_save_temps)) {
+    if (!isSaveTempsEnabled()) {
       const JobAction *JA = cast<JobAction>(&FailingCommand->getSource());
       C.CleanupFileMap(C.getResultFiles(), JA, true);
 
@@ -970,7 +979,7 @@ static bool DiagnoseInputExistence(const Driver &D, const DerivedArgList &Args,
 
   SmallString<64> Path(Value);
   if (Arg *WorkDir = Args.getLastArg(options::OPT_working_directory)) {
-    if (!llvm::sys::path::is_absolute(Path.str())) {
+    if (!llvm::sys::path::is_absolute(Path)) {
       SmallString<64> Directory(WorkDir->getValue());
       llvm::sys::path::append(Directory, Value);
       Path.assign(Directory);
@@ -980,10 +989,11 @@ static bool DiagnoseInputExistence(const Driver &D, const DerivedArgList &Args,
   if (llvm::sys::fs::exists(Twine(Path)))
     return true;
 
-  if (D.IsCLMode() && llvm::sys::Process::FindInEnvPath("LIB", Value))
+  if (D.IsCLMode() && !llvm::sys::path::is_absolute(Twine(Path)) &&
+      llvm::sys::Process::FindInEnvPath("LIB", Value))
     return true;
 
-  D.Diag(clang::diag::err_drv_no_such_file) << Path.str();
+  D.Diag(clang::diag::err_drv_no_such_file) << Path;
   return false;
 }
 
@@ -1264,7 +1274,7 @@ void Driver::BuildActions(const ToolChain &TC, DerivedArgList &Args,
         continue;
 
       // Otherwise construct the appropriate action.
-      Current = ConstructPhaseAction(Args, Phase, std::move(Current));
+      Current = ConstructPhaseAction(TC, Args, Phase, std::move(Current));
       if (Current->getType() == types::TY_Nothing)
         break;
     }
@@ -1290,7 +1300,8 @@ void Driver::BuildActions(const ToolChain &TC, DerivedArgList &Args,
 }
 
 std::unique_ptr<Action>
-Driver::ConstructPhaseAction(const ArgList &Args, phases::ID Phase,
+Driver::ConstructPhaseAction(const ToolChain &TC, const ArgList &Args,
+                             phases::ID Phase,
                              std::unique_ptr<Action> Input) const {
   llvm::PrettyStackTraceString CrashInfo("Constructing phase actions");
   // Build the appropriate action.
@@ -1349,7 +1360,7 @@ Driver::ConstructPhaseAction(const ArgList &Args, phases::ID Phase,
                                                types::TY_LLVM_BC);
   }
   case phases::Backend: {
-    if (IsUsingLTO(Args)) {
+    if (IsUsingLTO(TC, Args)) {
       types::ID Output =
         Args.hasArg(options::OPT_S) ? types::TY_LTO_IR : types::TY_LTO_BC;
       return llvm::make_unique<BackendJobAction>(std::move(Input), Output);
@@ -1370,7 +1381,10 @@ Driver::ConstructPhaseAction(const ArgList &Args, phases::ID Phase,
   llvm_unreachable("invalid phase in ConstructPhaseAction");
 }
 
-bool Driver::IsUsingLTO(const ArgList &Args) const {
+bool Driver::IsUsingLTO(const ToolChain &TC, const ArgList &Args) const {
+  if (TC.getSanitizerArgs().needsLTO())
+    return true;
+
   if (Args.hasFlag(options::OPT_flto, options::OPT_fno_lto, false))
     return true;
 
@@ -1471,8 +1485,8 @@ void Driver::BuildJobs(Compilation &C) const {
   }
 }
 
-static const Tool *SelectToolForJob(Compilation &C, const ToolChain *TC,
-                                    const JobAction *JA,
+static const Tool *SelectToolForJob(Compilation &C, bool SaveTemps,
+                                    const ToolChain *TC, const JobAction *JA,
                                     const ActionList *&Inputs) {
   const Tool *ToolForJob = nullptr;
 
@@ -1481,7 +1495,7 @@ static const Tool *SelectToolForJob(Compilation &C, const ToolChain *TC,
   // compiler input.
 
   if (TC->useIntegratedAs() &&
-      !C.getArgs().hasArg(options::OPT_save_temps) &&
+      !SaveTemps &&
       !C.getArgs().hasArg(options::OPT_via_file_asm) &&
       !C.getArgs().hasArg(options::OPT__SLASH_FA) &&
       !C.getArgs().hasArg(options::OPT__SLASH_Fa) &&
@@ -1512,8 +1526,7 @@ static const Tool *SelectToolForJob(Compilation &C, const ToolChain *TC,
     const Tool *Compiler = TC->SelectTool(*CompileJA);
     if (!Compiler)
       return nullptr;
-    if (!Compiler->canEmitIR() ||
-        !C.getArgs().hasArg(options::OPT_save_temps)) {
+    if (!Compiler->canEmitIR() || !SaveTemps) {
       Inputs = &(*Inputs)[0]->getInputs();
       ToolForJob = Compiler;
     }
@@ -1529,7 +1542,7 @@ static const Tool *SelectToolForJob(Compilation &C, const ToolChain *TC,
   if (Inputs->size() == 1 && isa<PreprocessJobAction>(*Inputs->begin()) &&
       !C.getArgs().hasArg(options::OPT_no_integrated_cpp) &&
       !C.getArgs().hasArg(options::OPT_traditional_cpp) &&
-      !C.getArgs().hasArg(options::OPT_save_temps) &&
+      !SaveTemps &&
       !C.getArgs().hasArg(options::OPT_rewrite_objc) &&
       ToolForJob->hasIntegratedCPP())
     Inputs = &(*Inputs)[0]->getInputs();
@@ -1577,7 +1590,7 @@ void Driver::BuildJobsForAction(Compilation &C,
   const ActionList *Inputs = &A->getInputs();
 
   const JobAction *JA = cast<JobAction>(A);
-  const Tool *T = SelectToolForJob(C, TC, JA, Inputs);
+  const Tool *T = SelectToolForJob(C, isSaveTempsEnabled(), TC, JA, Inputs);
   if (!T)
     return;
 
@@ -1708,7 +1721,7 @@ const char *Driver::GetNamedOutputPath(Compilation &C,
   }
 
   // Output to a temporary file?
-  if ((!AtTopLevel && !C.getArgs().hasArg(options::OPT_save_temps) &&
+  if ((!AtTopLevel && !isSaveTempsEnabled() &&
         !C.getArgs().hasArg(options::OPT__SLASH_Fo)) ||
       CCGenDiagnostics) {
     StringRef Name = llvm::sys::path::filename(BaseInput);
@@ -1780,11 +1793,20 @@ const char *Driver::GetNamedOutputPath(Compilation &C,
     NamedOutput = C.getArgs().MakeArgString(Suffixed.c_str());
   }
 
+  // Prepend object file path if -save-temps=obj
+  if (!AtTopLevel && isSaveTempsObj() && C.getArgs().hasArg(options::OPT_o) &&
+      JA.getType() != types::TY_PCH) {
+    Arg *FinalOutput = C.getArgs().getLastArg(options::OPT_o);
+    SmallString<128> TempPath(FinalOutput->getValue());
+    llvm::sys::path::remove_filename(TempPath);
+    StringRef OutputFileName = llvm::sys::path::filename(NamedOutput);
+    llvm::sys::path::append(TempPath, OutputFileName);
+    NamedOutput = C.getArgs().MakeArgString(TempPath.c_str());
+  }
+
   // If we're saving temps and the temp file conflicts with the input file,
   // then avoid overwriting input file.
-  if (!AtTopLevel && C.getArgs().hasArg(options::OPT_save_temps) &&
-      NamedOutput == BaseName) {
-
+  if (!AtTopLevel && isSaveTempsEnabled() && NamedOutput == BaseName) {
     bool SameFile = false;
     SmallString<256> Result;
     llvm::sys::fs::current_path(Result);
@@ -2003,12 +2025,15 @@ static llvm::Triple computeTargetTriple(StringRef DefaultTargetTriple,
 
 const ToolChain &Driver::getToolChain(const ArgList &Args,
                                       StringRef DarwinArchName) const {
-  llvm::Triple Target = computeTargetTriple(DefaultTargetTriple, Args,
-                                            DarwinArchName);
+  llvm::Triple Target =
+      computeTargetTriple(DefaultTargetTriple, Args, DarwinArchName);
 
   ToolChain *&TC = ToolChains[Target.str()];
   if (!TC) {
     switch (Target.getOS()) {
+    case llvm::Triple::CloudABI:
+      TC = new toolchains::CloudABI(*this, Target, Args);
+      break;
     case llvm::Triple::Darwin:
     case llvm::Triple::MacOSX:
     case llvm::Triple::IOS:
@@ -2038,6 +2063,9 @@ const ToolChain &Driver::getToolChain(const ArgList &Args,
       else
         TC = new toolchains::Linux(*this, Target, Args);
       break;
+    case llvm::Triple::NaCl:
+      TC = new toolchains::NaCl_TC(*this, Target, Args);
+      break;
     case llvm::Triple::Solaris:
       TC = new toolchains::Solaris(*this, Target, Args);
       break;
@@ -2069,29 +2097,20 @@ const ToolChain &Driver::getToolChain(const ArgList &Args,
       }
       break;
     default:
-      // TCE is an OSless target
-      if (Target.getArchName() == "tce") {
+      // Of these targets, Hexagon is the only one that might have
+      // an OS of Linux, in which case it got handled above already.
+      if (Target.getArchName() == "tce")
         TC = new toolchains::TCEToolChain(*this, Target, Args);
-        break;
-      }
-      // If Hexagon is configured as an OSless target
-      if (Target.getArch() == llvm::Triple::hexagon) {
+      else if (Target.getArch() == llvm::Triple::hexagon)
         TC = new toolchains::Hexagon_TC(*this, Target, Args);
-        break;
-      }
-      if (Target.getArch() == llvm::Triple::xcore) {
+      else if (Target.getArch() == llvm::Triple::xcore)
         TC = new toolchains::XCore(*this, Target, Args);
-        break;
-      }
-      if (Target.isOSBinFormatELF()) {
+      else if (Target.isOSBinFormatELF())
         TC = new toolchains::Generic_ELF(*this, Target, Args);
-        break;
-      }
-      if (Target.isOSBinFormatMachO()) {
+      else if (Target.isOSBinFormatMachO())
         TC = new toolchains::MachO(*this, Target, Args);
-        break;
-      }
-      TC = new toolchains::Generic_GCC(*this, Target, Args);
+      else
+        TC = new toolchains::Generic_GCC(*this, Target, Args);
       break;
     }
   }
@@ -2125,7 +2144,7 @@ bool Driver::GetReleaseVersion(const char *Str, unsigned &Major,
 
   Major = Minor = Micro = 0;
   if (*Str == '\0')
-    return true;
+    return false;
 
   char *End;
   Major = (unsigned) strtol(Str, &End, 10);
diff --git a/lib/Driver/Job.cpp b/lib/Driver/Job.cpp
index c5b3f5a307b9..6d18a41cad08 100644
--- a/lib/Driver/Job.cpp
+++ b/lib/Driver/Job.cpp
@@ -34,7 +34,7 @@ Command::Command(const Action &_Source, const Tool &_Creator,
       Executable(_Executable), Arguments(_Arguments),
       ResponseFile(nullptr) {}
 
-static int skipArgs(const char *Flag) {
+static int skipArgs(const char *Flag, bool HaveCrashVFS) {
   // These flags are all of the form -Flag <Arg> and are treated as two
   // arguments.  Therefore, we need to skip the flag and the next argument.
   bool Res = llvm::StringSwitch<bool>(Flag)
@@ -43,9 +43,11 @@ static int skipArgs(const char *Flag) {
     .Cases("-fdebug-compilation-dir", "-idirafter", true)
     .Cases("-include", "-include-pch", "-internal-isystem", true)
     .Cases("-internal-externc-isystem", "-iprefix", "-iwithprefix", true)
-    .Cases("-iwithprefixbefore", "-isysroot", "-isystem", "-iquote", true)
+    .Cases("-iwithprefixbefore", "-isystem", "-iquote", true)
     .Cases("-resource-dir", "-serialize-diagnostic-file", true)
     .Cases("-dwarf-debug-flags", "-ivfsoverlay", true)
+    // Some include flags shouldn't be skipped if we have a crash VFS
+    .Case("-isysroot", !HaveCrashVFS)
     .Default(false);
 
   // Match found.
@@ -164,11 +166,12 @@ void Command::Print(raw_ostream &OS, const char *Terminator, bool Quote,
       if (StringRef(Args[I]).equals("-main-file-name"))
         MainFilename = Args[I + 1];
 
+  bool HaveCrashVFS = CrashInfo && !CrashInfo->VFSPath.empty();
   for (size_t i = 0, e = Args.size(); i < e; ++i) {
     const char *const Arg = Args[i];
 
     if (CrashInfo) {
-      if (int Skip = skipArgs(Arg)) {
+      if (int Skip = skipArgs(Arg, HaveCrashVFS)) {
         i += Skip - 1;
         continue;
       } else if (llvm::sys::path::filename(Arg) == MainFilename &&
@@ -185,7 +188,7 @@ void Command::Print(raw_ostream &OS, const char *Terminator, bool Quote,
     PrintArg(OS, Arg, Quote);
   }
 
-  if (CrashInfo && !CrashInfo->VFSPath.empty()) {
+  if (CrashInfo && HaveCrashVFS) {
     OS << ' ';
     PrintArg(OS, "-ivfsoverlay", Quote);
     OS << ' ';
@@ -217,8 +220,7 @@ int Command::Execute(const StringRef **Redirects, std::string *ErrMsg,
 
   if (ResponseFile == nullptr) {
     Argv.push_back(Executable);
-    for (size_t i = 0, e = Arguments.size(); i != e; ++i)
-      Argv.push_back(Arguments[i]);
+    Argv.append(Arguments.begin(), Arguments.end());
     Argv.push_back(nullptr);
 
     return llvm::sys::ExecuteAndWait(Executable, Argv.data(), /*env*/ nullptr,
diff --git a/lib/Driver/MSVCToolChain.cpp b/lib/Driver/MSVCToolChain.cpp
index d6bd5c31ae6e..7739cb0cf5c7 100644
--- a/lib/Driver/MSVCToolChain.cpp
+++ b/lib/Driver/MSVCToolChain.cpp
@@ -21,6 +21,7 @@
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FileSystem.h"
 #include "llvm/Support/Process.h"
+#include <cstdio>
 
 // Include the necessary headers to interface with the Windows registry and
 // environment.
@@ -212,7 +213,7 @@ bool MSVCToolChain::getWindowsSDKDir(std::string &path, int &major,
       "SOFTWARE\\Microsoft\\Microsoft SDKs\\Windows\\$VERSION",
       "InstallationFolder", path, &sdkVersion);
   if (!sdkVersion.empty())
-    ::sscanf(sdkVersion.c_str(), "v%d.%d", &major, &minor);
+    std::sscanf(sdkVersion.c_str(), "v%d.%d", &major, &minor);
   return hasSDKDir && !path.empty();
 }
 
@@ -423,7 +424,7 @@ void MSVCToolChain::AddSystemIncludeWithSubfolder(const ArgList &DriverArgs,
                                                   const char *subfolder) const {
   llvm::SmallString<128> path(folder);
   llvm::sys::path::append(path, subfolder);
-  addSystemInclude(DriverArgs, CC1Args, path.str());
+  addSystemInclude(DriverArgs, CC1Args, path);
 }
 
 void MSVCToolChain::AddClangSystemIncludeArgs(const ArgList &DriverArgs,
@@ -434,7 +435,7 @@ void MSVCToolChain::AddClangSystemIncludeArgs(const ArgList &DriverArgs,
   if (!DriverArgs.hasArg(options::OPT_nobuiltininc)) {
     SmallString<128> P(getDriver().ResourceDir);
     llvm::sys::path::append(P, "include");
-    addSystemInclude(DriverArgs, CC1Args, P.str());
+    addSystemInclude(DriverArgs, CC1Args, P);
   }
 
   if (DriverArgs.hasArg(options::OPT_nostdlibinc))
diff --git a/lib/Driver/Multilib.cpp b/lib/Driver/Multilib.cpp
index 1f5d62f247c4..8acda6794d72 100644
--- a/lib/Driver/Multilib.cpp
+++ b/lib/Driver/Multilib.cpp
@@ -151,41 +151,23 @@ MultilibSet &MultilibSet::Maybe(const Multilib &M) {
 }
 
 MultilibSet &MultilibSet::Either(const Multilib &M1, const Multilib &M2) {
-  std::vector<Multilib> Ms;
-  Ms.push_back(M1);
-  Ms.push_back(M2);
-  return Either(Ms);
+  return Either({M1, M2});
 }
 
 MultilibSet &MultilibSet::Either(const Multilib &M1, const Multilib &M2,
                                  const Multilib &M3) {
-  std::vector<Multilib> Ms;
-  Ms.push_back(M1);
-  Ms.push_back(M2);
-  Ms.push_back(M3);
-  return Either(Ms);
+  return Either({M1, M2, M3});
 }
 
 MultilibSet &MultilibSet::Either(const Multilib &M1, const Multilib &M2,
                                  const Multilib &M3, const Multilib &M4) {
-  std::vector<Multilib> Ms;
-  Ms.push_back(M1);
-  Ms.push_back(M2);
-  Ms.push_back(M3);
-  Ms.push_back(M4);
-  return Either(Ms);
+  return Either({M1, M2, M3, M4});
 }
 
 MultilibSet &MultilibSet::Either(const Multilib &M1, const Multilib &M2,
                                  const Multilib &M3, const Multilib &M4,
                                  const Multilib &M5) {
-  std::vector<Multilib> Ms;
-  Ms.push_back(M1);
-  Ms.push_back(M2);
-  Ms.push_back(M3);
-  Ms.push_back(M4);
-  Ms.push_back(M5);
-  return Either(Ms);
+  return Either({M1, M2, M3, M4, M5});
 }
 
 static Multilib compose(const Multilib &Base, const Multilib &New) {
@@ -197,7 +179,7 @@ static Multilib compose(const Multilib &Base, const Multilib &New) {
   llvm::sys::path::append(IncludeSuffix, "/", Base.includeSuffix(),
                           New.includeSuffix());
 
-  Multilib Composed(GCCSuffix.str(), OSSuffix.str(), IncludeSuffix.str());
+  Multilib Composed(GCCSuffix, OSSuffix, IncludeSuffix);
 
   Multilib::flags_list &Flags = Composed.flags();
 
@@ -207,8 +189,7 @@ static Multilib compose(const Multilib &Base, const Multilib &New) {
   return Composed;
 }
 
-MultilibSet &
-MultilibSet::Either(const std::vector<Multilib> &MultilibSegments) {
+MultilibSet &MultilibSet::Either(ArrayRef<Multilib> MultilibSegments) {
   multilib_list Composed;
 
   if (Multilibs.empty())
@@ -229,30 +210,23 @@ MultilibSet::Either(const std::vector<Multilib> &MultilibSegments) {
   return *this;
 }
 
-MultilibSet &MultilibSet::FilterOut(const MultilibSet::FilterCallback &F) {
+MultilibSet &MultilibSet::FilterOut(FilterCallback F) {
   filterInPlace(F, Multilibs);
   return *this;
 }
 
-MultilibSet &MultilibSet::FilterOut(std::string Regex) {
-  class REFilter : public MultilibSet::FilterCallback {
-    mutable llvm::Regex R;
-
-  public:
-    REFilter(std::string Regex) : R(Regex) {}
-    bool operator()(const Multilib &M) const override {
-      std::string Error;
-      if (!R.isValid(Error)) {
-        llvm::errs() << Error;
-        assert(false);
-        return false;
-      }
-      return R.match(M.gccSuffix());
-    }
-  };
+MultilibSet &MultilibSet::FilterOut(const char *Regex) {
+  llvm::Regex R(Regex);
+#ifndef NDEBUG
+  std::string Error;
+  if (!R.isValid(Error)) {
+    llvm::errs() << Error;
+    llvm_unreachable("Invalid regex!");
+  }
+#endif
 
-  REFilter REF(Regex);
-  filterInPlace(REF, Multilibs);
+  filterInPlace([&R](const Multilib &M) { return R.match(M.gccSuffix()); },
+                Multilibs);
   return *this;
 }
 
@@ -262,38 +236,29 @@ void MultilibSet::combineWith(const MultilibSet &Other) {
   Multilibs.insert(Multilibs.end(), Other.begin(), Other.end());
 }
 
+static bool isFlagEnabled(StringRef Flag) {
+  char Indicator = Flag.front();
+  assert(Indicator == '+' || Indicator == '-');
+  return Indicator == '+';
+}
+
 bool MultilibSet::select(const Multilib::flags_list &Flags, Multilib &M) const {
-  class FilterFlagsMismatch : public MultilibSet::FilterCallback {
-    llvm::StringMap<bool> FlagSet;
-
-  public:
-    FilterFlagsMismatch(const std::vector<std::string> &Flags) {
-      // Stuff all of the flags into the FlagSet such that a true mappend
-      // indicates the flag was enabled, and a false mappend indicates the
-      // flag was disabled
-      for (StringRef Flag : Flags)
-        FlagSet[Flag.substr(1)] = isFlagEnabled(Flag);
-    }
-    bool operator()(const Multilib &M) const override {
-      for (StringRef Flag : M.flags()) {
-        llvm::StringMap<bool>::const_iterator SI = FlagSet.find(Flag.substr(1));
-        if (SI != FlagSet.end())
-          if (SI->getValue() != isFlagEnabled(Flag))
-            return true;
-      }
-      return false;
-    }
-  private:
-    bool isFlagEnabled(StringRef Flag) const {
-      char Indicator = Flag.front();
-      assert(Indicator == '+' || Indicator == '-');
-      return Indicator == '+';
+  llvm::StringMap<bool> FlagSet;
+
+  // Stuff all of the flags into the FlagSet such that a true mappend indicates
+  // the flag was enabled, and a false mappend indicates the flag was disabled.
+  for (StringRef Flag : Flags)
+    FlagSet[Flag.substr(1)] = isFlagEnabled(Flag);
+
+  multilib_list Filtered = filterCopy([&FlagSet](const Multilib &M) {
+    for (StringRef Flag : M.flags()) {
+      llvm::StringMap<bool>::const_iterator SI = FlagSet.find(Flag.substr(1));
+      if (SI != FlagSet.end())
+        if (SI->getValue() != isFlagEnabled(Flag))
+          return true;
     }
-  };
-
-  FilterFlagsMismatch FlagsMismatch(Flags);
-
-  multilib_list Filtered = filterCopy(FlagsMismatch, Multilibs);
+    return false;
+  }, Multilibs);
 
   if (Filtered.size() == 0) {
     return false;
@@ -313,19 +278,15 @@ void MultilibSet::print(raw_ostream &OS) const {
     OS << M << "\n";
 }
 
-MultilibSet::multilib_list
-MultilibSet::filterCopy(const MultilibSet::FilterCallback &F,
-                        const multilib_list &Ms) {
+MultilibSet::multilib_list MultilibSet::filterCopy(FilterCallback F,
+                                                   const multilib_list &Ms) {
   multilib_list Copy(Ms);
   filterInPlace(F, Copy);
   return Copy;
 }
 
-void MultilibSet::filterInPlace(const MultilibSet::FilterCallback &F,
-                                multilib_list &Ms) {
-  Ms.erase(std::remove_if(Ms.begin(), Ms.end(),
-                          [&F](const Multilib &M) { return F(M); }),
-           Ms.end());
+void MultilibSet::filterInPlace(FilterCallback F, multilib_list &Ms) {
+  Ms.erase(std::remove_if(Ms.begin(), Ms.end(), F), Ms.end());
 }
 
 raw_ostream &clang::driver::operator<<(raw_ostream &OS, const MultilibSet &MS) {
diff --git a/lib/Driver/SanitizerArgs.cpp b/lib/Driver/SanitizerArgs.cpp
index bd7bc218e3bf..72530b442f52 100644
--- a/lib/Driver/SanitizerArgs.cpp
+++ b/lib/Driver/SanitizerArgs.cpp
@@ -7,6 +7,7 @@
 //
 //===----------------------------------------------------------------------===//
 #include "clang/Driver/SanitizerArgs.h"
+#include "clang/Basic/Sanitizers.h"
 #include "clang/Driver/Driver.h"
 #include "clang/Driver/DriverDiagnostic.h"
 #include "clang/Driver/Options.h"
@@ -18,74 +19,41 @@
 #include "llvm/Support/SpecialCaseList.h"
 #include <memory>
 
+using namespace clang;
+using namespace clang::SanitizerKind;
 using namespace clang::driver;
 using namespace llvm::opt;
 
-namespace {
-/// Assign ordinals to possible values of -fsanitize= flag.
-/// We use the ordinal values as bit positions within \c SanitizeKind.
-enum SanitizeOrdinal {
-#define SANITIZER(NAME, ID) SO_##ID,
-#define SANITIZER_GROUP(NAME, ID, ALIAS) SO_##ID##Group,
-#include "clang/Basic/Sanitizers.def"
-  SO_Count
-};
-
-/// Represents a set of sanitizer kinds. It is also used to define:
-/// 1) set of sanitizers each sanitizer group expands into.
-/// 2) set of sanitizers sharing a specific property (e.g.
-///    all sanitizers with zero-base shadow).
-enum SanitizeKind {
-#define SANITIZER(NAME, ID) ID = 1 << SO_##ID,
-#define SANITIZER_GROUP(NAME, ID, ALIAS)                                       \
-ID = ALIAS, ID##Group = 1 << SO_##ID##Group,
-#include "clang/Basic/Sanitizers.def"
+enum : SanitizerMask {
   NeedsUbsanRt = Undefined | Integer,
   NotAllowedWithTrap = Vptr,
   RequiresPIE = Memory | DataFlow,
   NeedsUnwindTables = Address | Thread | Memory | DataFlow,
-  SupportsCoverage = Address | Memory | Leak | Undefined | Integer,
+  SupportsCoverage = Address | Memory | Leak | Undefined | Integer | DataFlow,
   RecoverableByDefault = Undefined | Integer,
   Unrecoverable = Address | Unreachable | Return,
-  LegacyFsanitizeRecoverMask = Undefined | Integer
+  LegacyFsanitizeRecoverMask = Undefined | Integer,
+  NeedsLTO = CFI,
 };
-}
-
-/// Returns true if set of \p Sanitizers contain at least one sanitizer from
-/// \p Kinds.
-static bool hasOneOf(const clang::SanitizerSet &Sanitizers, unsigned Kinds) {
-#define SANITIZER(NAME, ID)                                                    \
-  if (Sanitizers.has(clang::SanitizerKind::ID) && (Kinds & ID))                \
-    return true;
-#include "clang/Basic/Sanitizers.def"
-  return false;
-}
-
-/// Adds all sanitizers from \p Kinds to \p Sanitizers.
-static void addAllOf(clang::SanitizerSet &Sanitizers, unsigned Kinds) {
-#define SANITIZER(NAME, ID) \
-  if (Kinds & ID) \
-    Sanitizers.set(clang::SanitizerKind::ID, true);
-#include "clang/Basic/Sanitizers.def"
-}
-
-static unsigned toSanitizeKind(clang::SanitizerKind K) {
-#define SANITIZER(NAME, ID) \
-  if (K == clang::SanitizerKind::ID) \
-    return ID;
-#include "clang/Basic/Sanitizers.def"
-  llvm_unreachable("Invalid SanitizerKind!");
-}
 
-/// Parse a single value from a -fsanitize= or -fno-sanitize= value list.
-/// Returns a member of the \c SanitizeKind enumeration, or \c 0
-/// if \p Value is not known.
-static unsigned parseValue(const char *Value);
+enum CoverageFeature {
+  CoverageFunc = 1 << 0,
+  CoverageBB = 1 << 1,
+  CoverageEdge = 1 << 2,
+  CoverageIndirCall = 1 << 3,
+  CoverageTraceBB = 1 << 4,
+  CoverageTraceCmp = 1 << 5,
+  Coverage8bitCounters = 1 << 6,
+};
 
 /// Parse a -fsanitize= or -fno-sanitize= argument's values, diagnosing any
-/// invalid components. Returns OR of members of \c SanitizeKind enumeration.
-static unsigned parseArgValues(const Driver &D, const llvm::opt::Arg *A,
-                               bool DiagnoseErrors);
+/// invalid components. Returns a SanitizerMask.
+static SanitizerMask parseArgValues(const Driver &D, const llvm::opt::Arg *A,
+                                    bool DiagnoseErrors);
+
+/// Parse -f(no-)?sanitize-coverage= flag values, diagnosing any invalid
+/// components. Returns OR of members of \c CoverageFeature enumeration.
+static int parseCoverageFeatures(const Driver &D, const llvm::opt::Arg *A);
 
 /// Produce an argument string from ArgList \p Args, which shows how it
 /// provides some sanitizer kind from \p Mask. For example, the argument list
@@ -93,39 +61,32 @@ static unsigned parseArgValues(const Driver &D, const llvm::opt::Arg *A,
 /// would produce "-fsanitize=vptr".
 static std::string lastArgumentForMask(const Driver &D,
                                        const llvm::opt::ArgList &Args,
-                                       unsigned Mask);
-
-static std::string lastArgumentForKind(const Driver &D,
-                                       const llvm::opt::ArgList &Args,
-                                       clang::SanitizerKind K) {
-  return lastArgumentForMask(D, Args, toSanitizeKind(K));
-}
+                                       SanitizerMask Mask);
 
 /// Produce an argument string from argument \p A, which shows how it provides
 /// a value in \p Mask. For instance, the argument
 /// "-fsanitize=address,alignment" with mask \c NeedsUbsanRt would produce
 /// "-fsanitize=alignment".
-static std::string describeSanitizeArg(const llvm::opt::Arg *A, unsigned Mask);
+static std::string describeSanitizeArg(const llvm::opt::Arg *A,
+                                       SanitizerMask Mask);
 
 /// Produce a string containing comma-separated names of sanitizers in \p
 /// Sanitizers set.
 static std::string toString(const clang::SanitizerSet &Sanitizers);
 
-/// For each sanitizer group bit set in \p Kinds, set the bits for sanitizers
-/// this group enables.
-static unsigned expandGroups(unsigned Kinds);
-
-static unsigned getToolchainUnsupportedKinds(const ToolChain &TC) {
+static SanitizerMask getToolchainUnsupportedKinds(const ToolChain &TC) {
   bool IsFreeBSD = TC.getTriple().getOS() == llvm::Triple::FreeBSD;
   bool IsLinux = TC.getTriple().getOS() == llvm::Triple::Linux;
   bool IsX86 = TC.getTriple().getArch() == llvm::Triple::x86;
   bool IsX86_64 = TC.getTriple().getArch() == llvm::Triple::x86_64;
+  bool IsMIPS64 = TC.getTriple().getArch() == llvm::Triple::mips64 ||
+                  TC.getTriple().getArch() == llvm::Triple::mips64el;
 
-  unsigned Unsupported = 0;
-  if (!(IsLinux && IsX86_64)) {
+  SanitizerMask Unsupported = 0;
+  if (!(IsLinux && (IsX86_64 || IsMIPS64))) {
     Unsupported |= Memory | DataFlow;
   }
-  if (!((IsLinux || IsFreeBSD) && IsX86_64)) {
+  if (!((IsLinux || IsFreeBSD) && (IsX86_64 || IsMIPS64))) {
     Unsupported |= Thread;
   }
   if (!(IsLinux && (IsX86 || IsX86_64))) {
@@ -134,23 +95,51 @@ static unsigned getToolchainUnsupportedKinds(const ToolChain &TC) {
   return Unsupported;
 }
 
+static bool getDefaultBlacklist(const Driver &D, SanitizerMask Kinds,
+                                std::string &BLPath) {
+  const char *BlacklistFile = nullptr;
+  if (Kinds & Address)
+    BlacklistFile = "asan_blacklist.txt";
+  else if (Kinds & Memory)
+    BlacklistFile = "msan_blacklist.txt";
+  else if (Kinds & Thread)
+    BlacklistFile = "tsan_blacklist.txt";
+  else if (Kinds & DataFlow)
+    BlacklistFile = "dfsan_abilist.txt";
+
+  if (BlacklistFile) {
+    clang::SmallString<64> Path(D.ResourceDir);
+    llvm::sys::path::append(Path, BlacklistFile);
+    BLPath = Path.str();
+    return true;
+  }
+  return false;
+}
+
 bool SanitizerArgs::needsUbsanRt() const {
-  return !UbsanTrapOnError && hasOneOf(Sanitizers, NeedsUbsanRt);
+  return !UbsanTrapOnError && (Sanitizers.Mask & NeedsUbsanRt) &&
+         !Sanitizers.has(Address) &&
+         !Sanitizers.has(Memory) &&
+         !Sanitizers.has(Thread);
 }
 
 bool SanitizerArgs::requiresPIE() const {
-  return AsanZeroBaseShadow || hasOneOf(Sanitizers, RequiresPIE);
+  return AsanZeroBaseShadow || (Sanitizers.Mask & RequiresPIE);
 }
 
 bool SanitizerArgs::needsUnwindTables() const {
-  return hasOneOf(Sanitizers, NeedsUnwindTables);
+  return Sanitizers.Mask & NeedsUnwindTables;
+}
+
+bool SanitizerArgs::needsLTO() const {
+  return Sanitizers.Mask & NeedsLTO;
 }
 
 void SanitizerArgs::clear() {
   Sanitizers.clear();
   RecoverableSanitizers.clear();
-  BlacklistFile = "";
-  SanitizeCoverage = 0;
+  BlacklistFiles.clear();
+  CoverageFeatures = 0;
   MsanTrackOrigins = 0;
   AsanFieldPadding = 0;
   AsanZeroBaseShadow = false;
@@ -162,27 +151,36 @@ void SanitizerArgs::clear() {
 SanitizerArgs::SanitizerArgs(const ToolChain &TC,
                              const llvm::opt::ArgList &Args) {
   clear();
-  unsigned AllRemove = 0;  // During the loop below, the accumulated set of
-                           // sanitizers disabled by the current sanitizer
-                           // argument or any argument after it.
-  unsigned DiagnosedKinds = 0;  // All Kinds we have diagnosed up to now.
-                                // Used to deduplicate diagnostics.
-  unsigned Kinds = 0;
-  unsigned NotSupported = getToolchainUnsupportedKinds(TC);
+  SanitizerMask AllRemove = 0;  // During the loop below, the accumulated set of
+                                // sanitizers disabled by the current sanitizer
+                                // argument or any argument after it.
+  SanitizerMask AllAddedKinds = 0;  // Mask of all sanitizers ever enabled by
+                                    // -fsanitize= flags (directly or via group
+                                    // expansion), some of which may be disabled
+                                    // later. Used to carefully prune
+                                    // unused-argument diagnostics.
+  SanitizerMask DiagnosedKinds = 0;  // All Kinds we have diagnosed up to now.
+                                     // Used to deduplicate diagnostics.
+  SanitizerMask Kinds = 0;
+  SanitizerMask NotSupported = getToolchainUnsupportedKinds(TC);
+  ToolChain::RTTIMode RTTIMode = TC.getRTTIMode();
+
   const Driver &D = TC.getDriver();
   for (ArgList::const_reverse_iterator I = Args.rbegin(), E = Args.rend();
        I != E; ++I) {
     const auto *Arg = *I;
     if (Arg->getOption().matches(options::OPT_fsanitize_EQ)) {
       Arg->claim();
-      unsigned Add = parseArgValues(D, Arg, true);
+      SanitizerMask Add = parseArgValues(D, Arg, true);
+      AllAddedKinds |= expandSanitizerGroups(Add);
 
       // Avoid diagnosing any sanitizer which is disabled later.
       Add &= ~AllRemove;
       // At this point we have not expanded groups, so any unsupported
       // sanitizers in Add are those which have been explicitly enabled.
       // Diagnose them.
-      if (unsigned KindsToDiagnose = Add & NotSupported & ~DiagnosedKinds) {
+      if (SanitizerMask KindsToDiagnose =
+              Add & NotSupported & ~DiagnosedKinds) {
         // Only diagnose the new kinds.
         std::string Desc = describeSanitizeArg(*I, KindsToDiagnose);
         D.Diag(diag::err_drv_unsupported_opt_for_target)
@@ -191,7 +189,29 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
       }
       Add &= ~NotSupported;
 
-      Add = expandGroups(Add);
+      // Test for -fno-rtti + explicit -fsanitizer=vptr before expanding groups
+      // so we don't error out if -fno-rtti and -fsanitize=undefined were
+      // passed.
+      if (Add & Vptr &&
+          (RTTIMode == ToolChain::RM_DisabledImplicitly ||
+           RTTIMode == ToolChain::RM_DisabledExplicitly)) {
+        if (RTTIMode == ToolChain::RM_DisabledImplicitly)
+          // Warn about not having rtti enabled if the vptr sanitizer is
+          // explicitly enabled
+          D.Diag(diag::warn_drv_disabling_vptr_no_rtti_default);
+        else {
+          const llvm::opt::Arg *NoRTTIArg = TC.getRTTIArg();
+          assert(NoRTTIArg &&
+                 "RTTI disabled explicitly but we have no argument!");
+          D.Diag(diag::err_drv_argument_not_allowed_with)
+              << "-fsanitize=vptr" << NoRTTIArg->getAsString(Args);
+        }
+
+        // Take out the Vptr sanitizer from the enabled sanitizers
+        AllRemove |= Vptr;
+      }
+
+      Add = expandSanitizerGroups(Add);
       // Group expansion may have enabled a sanitizer which is disabled later.
       Add &= ~AllRemove;
       // Silently discard any unsupported sanitizers implicitly enabled through
@@ -201,122 +221,128 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
       Kinds |= Add;
     } else if (Arg->getOption().matches(options::OPT_fno_sanitize_EQ)) {
       Arg->claim();
-      unsigned Remove = parseArgValues(D, Arg, true);
-      AllRemove |= expandGroups(Remove);
+      SanitizerMask Remove = parseArgValues(D, Arg, true);
+      AllRemove |= expandSanitizerGroups(Remove);
+    }
+  }
+
+  // We disable the vptr sanitizer if it was enabled by group expansion but RTTI
+  // is disabled.
+  if ((Kinds & Vptr) &&
+      (RTTIMode == ToolChain::RM_DisabledImplicitly ||
+       RTTIMode == ToolChain::RM_DisabledExplicitly)) {
+    Kinds &= ~Vptr;
+  }
+
+  // Warn about undefined sanitizer options that require runtime support.
+  UbsanTrapOnError =
+    Args.hasFlag(options::OPT_fsanitize_undefined_trap_on_error,
+                 options::OPT_fno_sanitize_undefined_trap_on_error, false);
+  if (UbsanTrapOnError && (Kinds & NotAllowedWithTrap)) {
+    D.Diag(clang::diag::err_drv_argument_not_allowed_with)
+        << lastArgumentForMask(D, Args, NotAllowedWithTrap)
+        << "-fsanitize-undefined-trap-on-error";
+    Kinds &= ~NotAllowedWithTrap;
+  }
+
+  // Warn about incompatible groups of sanitizers.
+  std::pair<SanitizerMask, SanitizerMask> IncompatibleGroups[] = {
+      std::make_pair(Address, Thread), std::make_pair(Address, Memory),
+      std::make_pair(Thread, Memory), std::make_pair(Leak, Thread),
+      std::make_pair(Leak, Memory)};
+  for (auto G : IncompatibleGroups) {
+    SanitizerMask Group = G.first;
+    if (Kinds & Group) {
+      if (SanitizerMask Incompatible = Kinds & G.second) {
+        D.Diag(clang::diag::err_drv_argument_not_allowed_with)
+            << lastArgumentForMask(D, Args, Group)
+            << lastArgumentForMask(D, Args, Incompatible);
+        Kinds &= ~Incompatible;
+      }
     }
   }
-  addAllOf(Sanitizers, Kinds);
+  // FIXME: Currently -fsanitize=leak is silently ignored in the presence of
+  // -fsanitize=address. Perhaps it should print an error, or perhaps
+  // -f(-no)sanitize=leak should change whether leak detection is enabled by
+  // default in ASan?
 
   // Parse -f(no-)?sanitize-recover flags.
-  unsigned RecoverableKinds = RecoverableByDefault;
-  unsigned DiagnosedUnrecoverableKinds = 0;
+  SanitizerMask RecoverableKinds = RecoverableByDefault;
+  SanitizerMask DiagnosedUnrecoverableKinds = 0;
   for (const auto *Arg : Args) {
+    const char *DeprecatedReplacement = nullptr;
     if (Arg->getOption().matches(options::OPT_fsanitize_recover)) {
-      // FIXME: Add deprecation notice, and then remove this flag.
-      RecoverableKinds |= expandGroups(LegacyFsanitizeRecoverMask);
+      DeprecatedReplacement = "-fsanitize-recover=undefined,integer";
+      RecoverableKinds |= expandSanitizerGroups(LegacyFsanitizeRecoverMask);
       Arg->claim();
     } else if (Arg->getOption().matches(options::OPT_fno_sanitize_recover)) {
-      // FIXME: Add deprecation notice, and then remove this flag.
-      RecoverableKinds &= ~expandGroups(LegacyFsanitizeRecoverMask);
+      DeprecatedReplacement = "-fno-sanitize-recover=undefined,integer";
+      RecoverableKinds &= ~expandSanitizerGroups(LegacyFsanitizeRecoverMask);
       Arg->claim();
     } else if (Arg->getOption().matches(options::OPT_fsanitize_recover_EQ)) {
-      unsigned Add = parseArgValues(D, Arg, true);
+      SanitizerMask Add = parseArgValues(D, Arg, true);
       // Report error if user explicitly tries to recover from unrecoverable
       // sanitizer.
-      if (unsigned KindsToDiagnose =
+      if (SanitizerMask KindsToDiagnose =
               Add & Unrecoverable & ~DiagnosedUnrecoverableKinds) {
         SanitizerSet SetToDiagnose;
-        addAllOf(SetToDiagnose, KindsToDiagnose);
+        SetToDiagnose.Mask |= KindsToDiagnose;
         D.Diag(diag::err_drv_unsupported_option_argument)
             << Arg->getOption().getName() << toString(SetToDiagnose);
         DiagnosedUnrecoverableKinds |= KindsToDiagnose;
       }
-      RecoverableKinds |= expandGroups(Add);
+      RecoverableKinds |= expandSanitizerGroups(Add);
       Arg->claim();
     } else if (Arg->getOption().matches(options::OPT_fno_sanitize_recover_EQ)) {
-      RecoverableKinds &= ~expandGroups(parseArgValues(D, Arg, true));
+      RecoverableKinds &= ~expandSanitizerGroups(parseArgValues(D, Arg, true));
       Arg->claim();
     }
+    if (DeprecatedReplacement) {
+      D.Diag(diag::warn_drv_deprecated_arg) << Arg->getAsString(Args)
+                                            << DeprecatedReplacement;
+    }
   }
   RecoverableKinds &= Kinds;
   RecoverableKinds &= ~Unrecoverable;
-  addAllOf(RecoverableSanitizers, RecoverableKinds);
-
-  UbsanTrapOnError =
-    Args.hasFlag(options::OPT_fsanitize_undefined_trap_on_error,
-                 options::OPT_fno_sanitize_undefined_trap_on_error, false);
 
-  // Warn about undefined sanitizer options that require runtime support.
-  if (UbsanTrapOnError && hasOneOf(Sanitizers, NotAllowedWithTrap)) {
-    D.Diag(clang::diag::err_drv_argument_not_allowed_with)
-      << lastArgumentForMask(D, Args, NotAllowedWithTrap)
-      << "-fsanitize-undefined-trap-on-error";
+  // Setup blacklist files.
+  // Add default blacklist from resource directory.
+  {
+    std::string BLPath;
+    if (getDefaultBlacklist(D, Kinds, BLPath) && llvm::sys::fs::exists(BLPath))
+      BlacklistFiles.push_back(BLPath);
   }
-
-  // Check for incompatible sanitizers.
-  bool NeedsAsan = Sanitizers.has(SanitizerKind::Address);
-  bool NeedsTsan = Sanitizers.has(SanitizerKind::Thread);
-  bool NeedsMsan = Sanitizers.has(SanitizerKind::Memory);
-  bool NeedsLsan = Sanitizers.has(SanitizerKind::Leak);
-  if (NeedsAsan && NeedsTsan)
-    D.Diag(clang::diag::err_drv_argument_not_allowed_with)
-      << lastArgumentForKind(D, Args, SanitizerKind::Address)
-      << lastArgumentForKind(D, Args, SanitizerKind::Thread);
-  if (NeedsAsan && NeedsMsan)
-    D.Diag(clang::diag::err_drv_argument_not_allowed_with)
-      << lastArgumentForKind(D, Args, SanitizerKind::Address)
-      << lastArgumentForKind(D, Args, SanitizerKind::Memory);
-  if (NeedsTsan && NeedsMsan)
-    D.Diag(clang::diag::err_drv_argument_not_allowed_with)
-      << lastArgumentForKind(D, Args, SanitizerKind::Thread)
-      << lastArgumentForKind(D, Args, SanitizerKind::Memory);
-  if (NeedsLsan && NeedsTsan)
-    D.Diag(clang::diag::err_drv_argument_not_allowed_with)
-      << lastArgumentForKind(D, Args, SanitizerKind::Leak)
-      << lastArgumentForKind(D, Args, SanitizerKind::Thread);
-  if (NeedsLsan && NeedsMsan)
-    D.Diag(clang::diag::err_drv_argument_not_allowed_with)
-      << lastArgumentForKind(D, Args, SanitizerKind::Leak)
-      << lastArgumentForKind(D, Args, SanitizerKind::Memory);
-  // FIXME: Currently -fsanitize=leak is silently ignored in the presence of
-  // -fsanitize=address. Perhaps it should print an error, or perhaps
-  // -f(-no)sanitize=leak should change whether leak detection is enabled by
-  // default in ASan?
-
   // Parse -f(no-)sanitize-blacklist options.
-  if (Arg *BLArg = Args.getLastArg(options::OPT_fsanitize_blacklist,
-                                   options::OPT_fno_sanitize_blacklist)) {
-    if (BLArg->getOption().matches(options::OPT_fsanitize_blacklist)) {
-      std::string BLPath = BLArg->getValue();
-      if (llvm::sys::fs::exists(BLPath)) {
-        // Validate the blacklist format.
-        std::string BLError;
-        std::unique_ptr<llvm::SpecialCaseList> SCL(
-            llvm::SpecialCaseList::create(BLPath, BLError));
-        if (!SCL.get())
-          D.Diag(clang::diag::err_drv_malformed_sanitizer_blacklist) << BLError;
-        else
-          BlacklistFile = BLPath;
-      } else {
+  for (const auto *Arg : Args) {
+    if (Arg->getOption().matches(options::OPT_fsanitize_blacklist)) {
+      Arg->claim();
+      std::string BLPath = Arg->getValue();
+      if (llvm::sys::fs::exists(BLPath))
+        BlacklistFiles.push_back(BLPath);
+      else
         D.Diag(clang::diag::err_drv_no_such_file) << BLPath;
-      }
+    } else if (Arg->getOption().matches(options::OPT_fno_sanitize_blacklist)) {
+      Arg->claim();
+      BlacklistFiles.clear();
     }
-  } else {
-    // If no -fsanitize-blacklist option is specified, try to look up for
-    // blacklist in the resource directory.
-    std::string BLPath;
-    if (getDefaultBlacklist(D, BLPath) && llvm::sys::fs::exists(BLPath))
-      BlacklistFile = BLPath;
+  }
+  // Validate blacklists format.
+  {
+    std::string BLError;
+    std::unique_ptr<llvm::SpecialCaseList> SCL(
+        llvm::SpecialCaseList::create(BlacklistFiles, BLError));
+    if (!SCL.get())
+      D.Diag(clang::diag::err_drv_malformed_sanitizer_blacklist) << BLError;
   }
 
   // Parse -f[no-]sanitize-memory-track-origins[=level] options.
-  if (NeedsMsan) {
+  if (AllAddedKinds & Memory) {
     if (Arg *A =
             Args.getLastArg(options::OPT_fsanitize_memory_track_origins_EQ,
                             options::OPT_fsanitize_memory_track_origins,
                             options::OPT_fno_sanitize_memory_track_origins)) {
       if (A->getOption().matches(options::OPT_fsanitize_memory_track_origins)) {
-        MsanTrackOrigins = 1;
+        MsanTrackOrigins = 2;
       } else if (A->getOption().matches(
                      options::OPT_fno_sanitize_memory_track_origins)) {
         MsanTrackOrigins = 0;
@@ -330,18 +356,72 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
     }
   }
 
-  // Parse -fsanitize-coverage=N. Currently one of asan/msan/lsan is required.
-  if (hasOneOf(Sanitizers, SupportsCoverage)) {
-    if (Arg *A = Args.getLastArg(options::OPT_fsanitize_coverage)) {
-      StringRef S = A->getValue();
-      // Legal values are 0..4.
-      if (S.getAsInteger(0, SanitizeCoverage) || SanitizeCoverage < 0 ||
-          SanitizeCoverage > 4)
-        D.Diag(clang::diag::err_drv_invalid_value) << A->getAsString(Args) << S;
+  // Parse -f(no-)?sanitize-coverage flags if coverage is supported by the
+  // enabled sanitizers.
+  if (AllAddedKinds & SupportsCoverage) {
+    for (const auto *Arg : Args) {
+      if (Arg->getOption().matches(options::OPT_fsanitize_coverage)) {
+        Arg->claim();
+        int LegacySanitizeCoverage;
+        if (Arg->getNumValues() == 1 &&
+            !StringRef(Arg->getValue(0))
+                 .getAsInteger(0, LegacySanitizeCoverage) &&
+            LegacySanitizeCoverage >= 0 && LegacySanitizeCoverage <= 4) {
+          // TODO: Add deprecation notice for this form.
+          switch (LegacySanitizeCoverage) {
+          case 0:
+            CoverageFeatures = 0;
+            break;
+          case 1:
+            CoverageFeatures = CoverageFunc;
+            break;
+          case 2:
+            CoverageFeatures = CoverageBB;
+            break;
+          case 3:
+            CoverageFeatures = CoverageEdge;
+            break;
+          case 4:
+            CoverageFeatures = CoverageEdge | CoverageIndirCall;
+            break;
+          }
+          continue;
+        }
+        CoverageFeatures |= parseCoverageFeatures(D, Arg);
+      } else if (Arg->getOption().matches(options::OPT_fno_sanitize_coverage)) {
+        Arg->claim();
+        CoverageFeatures &= ~parseCoverageFeatures(D, Arg);
+      }
     }
   }
-
-  if (NeedsAsan) {
+  // Choose at most one coverage type: function, bb, or edge.
+  if ((CoverageFeatures & CoverageFunc) && (CoverageFeatures & CoverageBB))
+    D.Diag(clang::diag::err_drv_argument_not_allowed_with)
+        << "-fsanitize-coverage=func"
+        << "-fsanitize-coverage=bb";
+  if ((CoverageFeatures & CoverageFunc) && (CoverageFeatures & CoverageEdge))
+    D.Diag(clang::diag::err_drv_argument_not_allowed_with)
+        << "-fsanitize-coverage=func"
+        << "-fsanitize-coverage=edge";
+  if ((CoverageFeatures & CoverageBB) && (CoverageFeatures & CoverageEdge))
+    D.Diag(clang::diag::err_drv_argument_not_allowed_with)
+        << "-fsanitize-coverage=bb"
+        << "-fsanitize-coverage=edge";
+  // Basic block tracing and 8-bit counters require some type of coverage
+  // enabled.
+  int CoverageTypes = CoverageFunc | CoverageBB | CoverageEdge;
+  if ((CoverageFeatures & CoverageTraceBB) &&
+      !(CoverageFeatures & CoverageTypes))
+    D.Diag(clang::diag::err_drv_argument_only_allowed_with)
+        << "-fsanitize-coverage=trace-bb"
+        << "-fsanitize-coverage=(func|bb|edge)";
+  if ((CoverageFeatures & Coverage8bitCounters) &&
+      !(CoverageFeatures & CoverageTypes))
+    D.Diag(clang::diag::err_drv_argument_only_allowed_with)
+        << "-fsanitize-coverage=8bit-counters"
+        << "-fsanitize-coverage=(func|bb|edge)";
+
+  if (AllAddedKinds & Address) {
     AsanSharedRuntime =
         Args.hasArg(options::OPT_shared_libasan) ||
         (TC.getTriple().getEnvironment() == llvm::Triple::Android);
@@ -367,7 +447,7 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
       case options::OPT__SLASH_LDd:
         D.Diag(clang::diag::err_drv_argument_not_allowed_with)
             << WindowsDebugRTArg->getAsString(Args)
-            << lastArgumentForKind(D, Args, SanitizerKind::Address);
+            << lastArgumentForMask(D, Args, Address);
         D.Diag(clang::diag::note_drv_address_sanitizer_debug_runtime);
       }
     }
@@ -376,12 +456,16 @@ SanitizerArgs::SanitizerArgs(const ToolChain &TC,
   // Parse -link-cxx-sanitizer flag.
   LinkCXXRuntimes =
       Args.hasArg(options::OPT_fsanitize_link_cxx_runtime) || D.CCCIsCXX();
+
+  // Finally, initialize the set of available and recoverable sanitizers.
+  Sanitizers.Mask |= Kinds;
+  RecoverableSanitizers.Mask |= RecoverableKinds;
 }
 
 static std::string toString(const clang::SanitizerSet &Sanitizers) {
   std::string Res;
 #define SANITIZER(NAME, ID)                                                    \
-  if (Sanitizers.has(clang::SanitizerKind::ID)) {                              \
+  if (Sanitizers.has(ID)) {                                                    \
     if (!Res.empty())                                                          \
       Res += ",";                                                              \
     Res += NAME;                                                               \
@@ -403,9 +487,9 @@ void SanitizerArgs::addArgs(const llvm::opt::ArgList &Args,
   if (UbsanTrapOnError)
     CmdArgs.push_back("-fsanitize-undefined-trap-on-error");
 
-  if (!BlacklistFile.empty()) {
+  for (const auto &BLPath : BlacklistFiles) {
     SmallString<64> BlacklistOpt("-fsanitize-blacklist=");
-    BlacklistOpt += BlacklistFile;
+    BlacklistOpt += BLPath;
     CmdArgs.push_back(Args.MakeArgString(BlacklistOpt));
   }
 
@@ -415,67 +499,47 @@ void SanitizerArgs::addArgs(const llvm::opt::ArgList &Args,
   if (AsanFieldPadding)
     CmdArgs.push_back(Args.MakeArgString("-fsanitize-address-field-padding=" +
                                          llvm::utostr(AsanFieldPadding)));
-  if (SanitizeCoverage)
-    CmdArgs.push_back(Args.MakeArgString("-fsanitize-coverage=" +
-                                         llvm::utostr(SanitizeCoverage)));
-  // Workaround for PR16386.
-  if (Sanitizers.has(SanitizerKind::Memory))
-    CmdArgs.push_back(Args.MakeArgString("-fno-assume-sane-operator-new"));
-}
-
-bool SanitizerArgs::getDefaultBlacklist(const Driver &D, std::string &BLPath) {
-  const char *BlacklistFile = nullptr;
-  if (Sanitizers.has(SanitizerKind::Address))
-    BlacklistFile = "asan_blacklist.txt";
-  else if (Sanitizers.has(SanitizerKind::Memory))
-    BlacklistFile = "msan_blacklist.txt";
-  else if (Sanitizers.has(SanitizerKind::Thread))
-    BlacklistFile = "tsan_blacklist.txt";
-  else if (Sanitizers.has(SanitizerKind::DataFlow))
-    BlacklistFile = "dfsan_abilist.txt";
-
-  if (BlacklistFile) {
-    SmallString<64> Path(D.ResourceDir);
-    llvm::sys::path::append(Path, BlacklistFile);
-    BLPath = Path.str();
-    return true;
+  // Translate available CoverageFeatures to corresponding clang-cc1 flags.
+  std::pair<int, const char *> CoverageFlags[] = {
+    std::make_pair(CoverageFunc, "-fsanitize-coverage-type=1"),
+    std::make_pair(CoverageBB, "-fsanitize-coverage-type=2"),
+    std::make_pair(CoverageEdge, "-fsanitize-coverage-type=3"),
+    std::make_pair(CoverageIndirCall, "-fsanitize-coverage-indirect-calls"),
+    std::make_pair(CoverageTraceBB, "-fsanitize-coverage-trace-bb"),
+    std::make_pair(CoverageTraceCmp, "-fsanitize-coverage-trace-cmp"),
+    std::make_pair(Coverage8bitCounters, "-fsanitize-coverage-8bit-counters")};
+  for (auto F : CoverageFlags) {
+    if (CoverageFeatures & F.first)
+      CmdArgs.push_back(Args.MakeArgString(F.second));
   }
-  return false;
-}
 
-unsigned parseValue(const char *Value) {
-  unsigned ParsedKind = llvm::StringSwitch<SanitizeKind>(Value)
-#define SANITIZER(NAME, ID) .Case(NAME, ID)
-#define SANITIZER_GROUP(NAME, ID, ALIAS) .Case(NAME, ID##Group)
-#include "clang/Basic/Sanitizers.def"
-    .Default(SanitizeKind());
-  return ParsedKind;
-}
 
-unsigned expandGroups(unsigned Kinds) {
-#define SANITIZER(NAME, ID)
-#define SANITIZER_GROUP(NAME, ID, ALIAS) if (Kinds & ID##Group) Kinds |= ID;
-#include "clang/Basic/Sanitizers.def"
-  return Kinds;
+  // MSan: Workaround for PR16386.
+  // ASan: This is mainly to help LSan with cases such as
+  // https://code.google.com/p/address-sanitizer/issues/detail?id=373
+  // We can't make this conditional on -fsanitize=leak, as that flag shouldn't
+  // affect compilation.
+  if (Sanitizers.has(Memory) || Sanitizers.has(Address))
+    CmdArgs.push_back(Args.MakeArgString("-fno-assume-sane-operator-new"));
 }
 
-unsigned parseArgValues(const Driver &D, const llvm::opt::Arg *A,
-                        bool DiagnoseErrors) {
+SanitizerMask parseArgValues(const Driver &D, const llvm::opt::Arg *A,
+                             bool DiagnoseErrors) {
   assert((A->getOption().matches(options::OPT_fsanitize_EQ) ||
           A->getOption().matches(options::OPT_fno_sanitize_EQ) ||
           A->getOption().matches(options::OPT_fsanitize_recover_EQ) ||
           A->getOption().matches(options::OPT_fno_sanitize_recover_EQ)) &&
          "Invalid argument in parseArgValues!");
-  unsigned Kinds = 0;
-  for (unsigned I = 0, N = A->getNumValues(); I != N; ++I) {
-    const char *Value = A->getValue(I);
-    unsigned Kind;
+  SanitizerMask Kinds = 0;
+  for (int i = 0, n = A->getNumValues(); i != n; ++i) {
+    const char *Value = A->getValue(i);
+    SanitizerMask Kind;
     // Special case: don't accept -fsanitize=all.
     if (A->getOption().matches(options::OPT_fsanitize_EQ) &&
         0 == strcmp("all", Value))
       Kind = 0;
     else
-      Kind = parseValue(Value);
+      Kind = parseSanitizerValue(Value, /*AllowGroups=*/true);
 
     if (Kind)
       Kinds |= Kind;
@@ -486,34 +550,61 @@ unsigned parseArgValues(const Driver &D, const llvm::opt::Arg *A,
   return Kinds;
 }
 
+int parseCoverageFeatures(const Driver &D, const llvm::opt::Arg *A) {
+  assert(A->getOption().matches(options::OPT_fsanitize_coverage) ||
+         A->getOption().matches(options::OPT_fno_sanitize_coverage));
+  int Features = 0;
+  for (int i = 0, n = A->getNumValues(); i != n; ++i) {
+    const char *Value = A->getValue(i);
+    int F = llvm::StringSwitch<int>(Value)
+        .Case("func", CoverageFunc)
+        .Case("bb", CoverageBB)
+        .Case("edge", CoverageEdge)
+        .Case("indirect-calls", CoverageIndirCall)
+        .Case("trace-bb", CoverageTraceBB)
+        .Case("trace-cmp", CoverageTraceCmp)
+        .Case("8bit-counters", Coverage8bitCounters)
+        .Default(0);
+    if (F == 0)
+      D.Diag(clang::diag::err_drv_unsupported_option_argument)
+          << A->getOption().getName() << Value;
+    Features |= F;
+  }
+  return Features;
+}
+
 std::string lastArgumentForMask(const Driver &D, const llvm::opt::ArgList &Args,
-                                unsigned Mask) {
+                                SanitizerMask Mask) {
   for (llvm::opt::ArgList::const_reverse_iterator I = Args.rbegin(),
                                                   E = Args.rend();
        I != E; ++I) {
     const auto *Arg = *I;
     if (Arg->getOption().matches(options::OPT_fsanitize_EQ)) {
-      unsigned AddKinds = expandGroups(parseArgValues(D, Arg, false));
+      SanitizerMask AddKinds =
+          expandSanitizerGroups(parseArgValues(D, Arg, false));
       if (AddKinds & Mask)
         return describeSanitizeArg(Arg, Mask);
     } else if (Arg->getOption().matches(options::OPT_fno_sanitize_EQ)) {
-      unsigned RemoveKinds = expandGroups(parseArgValues(D, Arg, false));
+      SanitizerMask RemoveKinds =
+          expandSanitizerGroups(parseArgValues(D, Arg, false));
       Mask &= ~RemoveKinds;
     }
   }
   llvm_unreachable("arg list didn't provide expected value");
 }
 
-std::string describeSanitizeArg(const llvm::opt::Arg *A, unsigned Mask) {
+std::string describeSanitizeArg(const llvm::opt::Arg *A, SanitizerMask Mask) {
   assert(A->getOption().matches(options::OPT_fsanitize_EQ)
          && "Invalid argument in describeSanitizerArg!");
 
   std::string Sanitizers;
-  for (unsigned I = 0, N = A->getNumValues(); I != N; ++I) {
-    if (expandGroups(parseValue(A->getValue(I))) & Mask) {
+  for (int i = 0, n = A->getNumValues(); i != n; ++i) {
+    if (expandSanitizerGroups(
+            parseSanitizerValue(A->getValue(i), /*AllowGroups=*/true)) &
+        Mask) {
       if (!Sanitizers.empty())
         Sanitizers += ",";
-      Sanitizers += A->getValue(I);
+      Sanitizers += A->getValue(i);
     }
   }
 
diff --git a/lib/Driver/ToolChain.cpp b/lib/Driver/ToolChain.cpp
index 2bcfecf6bde6..f7b7402215a6 100644
--- a/lib/Driver/ToolChain.cpp
+++ b/lib/Driver/ToolChain.cpp
@@ -26,14 +26,47 @@ using namespace clang::driver;
 using namespace clang;
 using namespace llvm::opt;
 
+static llvm::opt::Arg *GetRTTIArgument(const ArgList &Args) {
+  return Args.getLastArg(options::OPT_mkernel, options::OPT_fapple_kext,
+                         options::OPT_fno_rtti, options::OPT_frtti);
+}
+
+static ToolChain::RTTIMode CalculateRTTIMode(const ArgList &Args,
+                                             const llvm::Triple &Triple,
+                                             const Arg *CachedRTTIArg) {
+  // Explicit rtti/no-rtti args
+  if (CachedRTTIArg) {
+    if (CachedRTTIArg->getOption().matches(options::OPT_frtti))
+      return ToolChain::RM_EnabledExplicitly;
+    else
+      return ToolChain::RM_DisabledExplicitly;
+  }
+
+  // -frtti is default, except for the PS4 CPU.
+  if (!Triple.isPS4CPU())
+    return ToolChain::RM_EnabledImplicitly;
+
+  // On the PS4, turning on c++ exceptions turns on rtti.
+  // We're assuming that, if we see -fexceptions, rtti gets turned on.
+  Arg *Exceptions = Args.getLastArgNoClaim(
+      options::OPT_fcxx_exceptions, options::OPT_fno_cxx_exceptions,
+      options::OPT_fexceptions, options::OPT_fno_exceptions);
+  if (Exceptions &&
+      (Exceptions->getOption().matches(options::OPT_fexceptions) ||
+       Exceptions->getOption().matches(options::OPT_fcxx_exceptions)))
+    return ToolChain::RM_EnabledImplicitly;
+
+  return ToolChain::RM_DisabledImplicitly;
+}
+
 ToolChain::ToolChain(const Driver &D, const llvm::Triple &T,
                      const ArgList &Args)
-  : D(D), Triple(T), Args(Args) {
+    : D(D), Triple(T), Args(Args), CachedRTTIArg(GetRTTIArgument(Args)),
+      CachedRTTIMode(CalculateRTTIMode(Args, Triple, CachedRTTIArg)) {
   if (Arg *A = Args.getLastArg(options::OPT_mthread_model))
     if (!isThreadModelSupported(A->getValue()))
       D.Diag(diag::err_drv_invalid_thread_model_for_target)
-          << A->getValue()
-          << A->getAsString(Args);
+          << A->getValue() << A->getAsString(Args);
 }
 
 ToolChain::~ToolChain() {
@@ -264,18 +297,18 @@ std::string ToolChain::ComputeLLVMTriple(const ArgList &Args,
     // '-mbig-endian'/'-EB'.
     if (Arg *A = Args.getLastArg(options::OPT_mlittle_endian,
                                  options::OPT_mbig_endian)) {
-      if (A->getOption().matches(options::OPT_mlittle_endian))
-        IsBigEndian = false;
-      else
-        IsBigEndian = true;
+      IsBigEndian = !A->getOption().matches(options::OPT_mlittle_endian);
     }
 
     // Thumb2 is the default for V7 on Darwin.
     //
     // FIXME: Thumb should just be another -target-feaure, not in the triple.
-    StringRef Suffix = Triple.isOSBinFormatMachO()
-      ? tools::arm::getLLVMArchSuffixForARM(tools::arm::getARMCPUForMArch(Args, Triple))
-      : tools::arm::getLLVMArchSuffixForARM(tools::arm::getARMTargetCPU(Args, Triple));
+    StringRef CPU = Triple.isOSBinFormatMachO()
+      ? tools::arm::getARMCPUForMArch(Args, Triple)
+      : tools::arm::getARMTargetCPU(Args, Triple);
+    StringRef Suffix = 
+      tools::arm::getLLVMArchSuffixForARM(CPU,
+                                          tools::arm::getARMArch(Args, Triple));
     bool ThumbDefault = Suffix.startswith("v6m") || Suffix.startswith("v7m") ||
       Suffix.startswith("v7em") ||
       (Suffix.startswith("v7") && getTriple().isOSBinFormatMachO());
diff --git a/lib/Driver/ToolChains.cpp b/lib/Driver/ToolChains.cpp
index 8bd8298031f5..6d52ab9d71a5 100644
--- a/lib/Driver/ToolChains.cpp
+++ b/lib/Driver/ToolChains.cpp
@@ -108,6 +108,7 @@ bool Darwin::hasBlocksRuntime() const {
   }
 }
 
+// FIXME: Use ARMTargetParser.
 static const char *GetArmArchForMArch(StringRef Value) {
   return llvm::StringSwitch<const char*>(Value)
     .Case("armv6k", "armv6")
@@ -125,6 +126,7 @@ static const char *GetArmArchForMArch(StringRef Value) {
     .Default(nullptr);
 }
 
+// FIXME: Use ARMTargetParser.
 static const char *GetArmArchForMCpu(StringRef Value) {
   return llvm::StringSwitch<const char *>(Value)
     .Cases("arm9e", "arm946e-s", "arm966e-s", "arm968e-s", "arm926ej-s","armv5")
@@ -132,11 +134,11 @@ static const char *GetArmArchForMCpu(StringRef Value) {
     .Cases("arm1020t", "arm1020e", "arm1022e", "arm1026ej-s", "armv5")
     .Case("xscale", "xscale")
     .Cases("arm1136j-s", "arm1136jf-s", "arm1176jz-s", "arm1176jzf-s", "armv6")
-    .Case("cortex-m0", "armv6m")
+    .Cases("sc000", "cortex-m0", "cortex-m0plus", "cortex-m1", "armv6m")
     .Cases("cortex-a5", "cortex-a7", "cortex-a8", "armv7")
     .Cases("cortex-a9", "cortex-a12", "cortex-a15", "cortex-a17", "krait", "armv7")
-    .Cases("cortex-r4", "cortex-r5", "armv7r")
-    .Case("cortex-m3", "armv7m")
+    .Cases("cortex-r4", "cortex-r4f", "cortex-r5", "cortex-r7", "armv7r")
+    .Cases("sc300", "cortex-m3", "armv7m")
     .Cases("cortex-m4", "cortex-m7", "armv7em")
     .Case("swift", "armv7s")
     .Default(nullptr);
@@ -302,8 +304,8 @@ void MachO::AddLinkRuntimeLib(const ArgList &Args, ArgStringList &CmdArgs,
   // For now, allow missing resource libraries to support developers who may
   // not have compiler-rt checked out or integrated into their build (unless
   // we explicitly force linking with this library).
-  if (AlwaysLink || llvm::sys::fs::exists(P.str()))
-    CmdArgs.push_back(Args.MakeArgString(P.str()));
+  if (AlwaysLink || llvm::sys::fs::exists(P))
+    CmdArgs.push_back(Args.MakeArgString(P));
 
   // Adding the rpaths might negatively interact when other rpaths are involved,
   // so we should make sure we add the rpaths last, after all user-specified
@@ -320,7 +322,49 @@ void MachO::AddLinkRuntimeLib(const ArgList &Args, ArgStringList &CmdArgs,
     // Add the path to the resource dir to rpath to support using the dylib
     // from the default location without copying.
     CmdArgs.push_back("-rpath");
-    CmdArgs.push_back(Args.MakeArgString(Dir.str()));
+    CmdArgs.push_back(Args.MakeArgString(Dir));
+  }
+}
+
+void Darwin::addProfileRTLibs(const ArgList &Args,
+                             ArgStringList &CmdArgs) const {
+  if (!(Args.hasFlag(options::OPT_fprofile_arcs, options::OPT_fno_profile_arcs,
+                     false) ||
+        Args.hasArg(options::OPT_fprofile_generate) ||
+        Args.hasArg(options::OPT_fprofile_instr_generate) ||
+        Args.hasArg(options::OPT_fprofile_instr_generate_EQ) ||
+        Args.hasArg(options::OPT_fcreate_profile) ||
+        Args.hasArg(options::OPT_coverage)))
+    return;
+
+  // Select the appropriate runtime library for the target.
+  if (isTargetIOSBased())
+    AddLinkRuntimeLib(Args, CmdArgs, "libclang_rt.profile_ios.a",
+                      /*AlwaysLink*/ true);
+  else
+    AddLinkRuntimeLib(Args, CmdArgs, "libclang_rt.profile_osx.a",
+                      /*AlwaysLink*/ true);
+}
+
+void DarwinClang::AddLinkSanitizerLibArgs(const ArgList &Args,
+                                          ArgStringList &CmdArgs,
+                                          StringRef Sanitizer) const {
+  if (!Args.hasArg(options::OPT_dynamiclib) &&
+      !Args.hasArg(options::OPT_bundle)) {
+    // Sanitizer runtime libraries requires C++.
+    AddCXXStdlibLibArgs(Args, CmdArgs);
+  }
+  assert(isTargetMacOS() || isTargetIOSSimulator());
+  StringRef OS = isTargetMacOS() ? "osx" : "iossim";
+  AddLinkRuntimeLib(Args, CmdArgs, (Twine("libclang_rt.") + Sanitizer + "_" +
+                                    OS + "_dynamic.dylib").str(),
+                    /*AlwaysLink*/ true, /*IsEmbedded*/ false,
+                    /*AddRPath*/ true);
+
+  if (GetCXXStdlibType(Args) == ToolChain::CST_Libcxx) {
+    // Add explicit dependcy on -lc++abi, as -lc++ doesn't re-export
+    // all RTTI-related symbols that UBSan uses.
+    CmdArgs.push_back("-lc++abi");
   }
 }
 
@@ -352,63 +396,26 @@ void DarwinClang::AddLinkRuntimeLibArgs(const ArgList &Args,
     return;
   }
 
-  // If we are building profile support, link that library in.
-  if (Args.hasFlag(options::OPT_fprofile_arcs, options::OPT_fno_profile_arcs,
-                   false) ||
-      Args.hasArg(options::OPT_fprofile_generate) ||
-      Args.hasArg(options::OPT_fprofile_instr_generate) ||
-      Args.hasArg(options::OPT_fcreate_profile) ||
-      Args.hasArg(options::OPT_coverage)) {
-    // Select the appropriate runtime library for the target.
-    if (isTargetIOSBased())
-      AddLinkRuntimeLib(Args, CmdArgs, "libclang_rt.profile_ios.a");
-    else
-      AddLinkRuntimeLib(Args, CmdArgs, "libclang_rt.profile_osx.a");
-  }
 
   const SanitizerArgs &Sanitize = getSanitizerArgs();
 
-  // Add Ubsan runtime library, if required.
-  if (Sanitize.needsUbsanRt()) {
-    // FIXME: Move this check to SanitizerArgs::filterUnsupportedKinds.
-    if (isTargetIOSBased()) {
+  if (Sanitize.needsAsanRt()) {
+    if (!isTargetMacOS() && !isTargetIOSSimulator()) {
+      // FIXME: Move this check to SanitizerArgs::filterUnsupportedKinds.
       getDriver().Diag(diag::err_drv_clang_unsupported_per_platform)
-        << "-fsanitize=undefined";
+          << "-fsanitize=address";
     } else {
-      assert(isTargetMacOS() && "unexpected non OS X target");
-      AddLinkRuntimeLib(Args, CmdArgs, "libclang_rt.ubsan_osx.a", true);
-
-      // The Ubsan runtime library requires C++.
-      AddCXXStdlibLibArgs(Args, CmdArgs);
+      AddLinkSanitizerLibArgs(Args, CmdArgs, "asan");
     }
   }
 
-  // Add ASAN runtime library, if required. Dynamic libraries and bundles
-  // should not be linked with the runtime library.
-  if (Sanitize.needsAsanRt()) {
-    // FIXME: Move this check to SanitizerArgs::filterUnsupportedKinds.
-    if (isTargetIPhoneOS()) {
+  if (Sanitize.needsUbsanRt()) {
+    if (!isTargetMacOS() && !isTargetIOSSimulator()) {
+      // FIXME: Move this check to SanitizerArgs::filterUnsupportedKinds.
       getDriver().Diag(diag::err_drv_clang_unsupported_per_platform)
-        << "-fsanitize=address";
+          << "-fsanitize=undefined";
     } else {
-      if (!Args.hasArg(options::OPT_dynamiclib) &&
-          !Args.hasArg(options::OPT_bundle)) {
-        // The ASAN runtime library requires C++.
-        AddCXXStdlibLibArgs(Args, CmdArgs);
-      }
-      if (isTargetMacOS()) {
-        AddLinkRuntimeLib(Args, CmdArgs,
-                          "libclang_rt.asan_osx_dynamic.dylib",
-                          /*AlwaysLink*/ true, /*IsEmbedded*/ false,
-                          /*AddRPath*/ true);
-      } else {
-        if (isTargetIOSSimulator()) {
-          AddLinkRuntimeLib(Args, CmdArgs,
-                            "libclang_rt.asan_iossim_dynamic.dylib",
-                            /*AlwaysLink*/ true, /*IsEmbedded*/ false,
-                            /*AddRPath*/ true);
-        }
-      }
+      AddLinkSanitizerLibArgs(Args, CmdArgs, "ubsan");
     }
   }
 
@@ -599,11 +606,11 @@ void DarwinClang::AddCXXStdlibLibArgs(const ArgList &Args,
       SmallString<128> P(A->getValue());
       llvm::sys::path::append(P, "usr", "lib", "libstdc++.dylib");
 
-      if (!llvm::sys::fs::exists(P.str())) {
+      if (!llvm::sys::fs::exists(P)) {
         llvm::sys::path::remove_filename(P);
         llvm::sys::path::append(P, "libstdc++.6.dylib");
-        if (llvm::sys::fs::exists(P.str())) {
-          CmdArgs.push_back(Args.MakeArgString(P.str()));
+        if (llvm::sys::fs::exists(P)) {
+          CmdArgs.push_back(Args.MakeArgString(P));
           return;
         }
       }
@@ -646,8 +653,8 @@ void DarwinClang::AddCCKextLibArgs(const ArgList &Args,
 
   // For now, allow missing resource libraries to support developers who may
   // not have compiler-rt checked out or integrated into their build.
-  if (llvm::sys::fs::exists(P.str()))
-    CmdArgs.push_back(Args.MakeArgString(P.str()));
+  if (llvm::sys::fs::exists(P))
+    CmdArgs.push_back(Args.MakeArgString(P));
 }
 
 DerivedArgList *MachO::TranslateArgs(const DerivedArgList &Args,
@@ -1498,11 +1505,12 @@ bool Generic_GCC::GCCInstallationDetector::getBiarchSibling(Multilib &M) const {
 
 namespace {
 // Filter to remove Multilibs that don't exist as a suffix to Path
-class FilterNonExistent : public MultilibSet::FilterCallback {
-  std::string Base;
+class FilterNonExistent {
+  StringRef Base;
+
 public:
-  FilterNonExistent(std::string Base) : Base(Base) {}
-  bool operator()(const Multilib &M) const override {
+  FilterNonExistent(StringRef Base) : Base(Base) {}
+  bool operator()(const Multilib &M) {
     return !llvm::sys::fs::exists(Base + M.gccSuffix() + "/crtbegin.o");
   }
 };
@@ -1783,10 +1791,13 @@ static bool findMIPSMultilibs(const llvm::Triple &TargetTriple, StringRef Path,
   addMultilibFlag(isMips64(TargetArch), "m64", Flags);
   addMultilibFlag(isMips16(Args), "mips16", Flags);
   addMultilibFlag(CPUName == "mips32", "march=mips32", Flags);
-  addMultilibFlag(CPUName == "mips32r2", "march=mips32r2", Flags);
+  addMultilibFlag(CPUName == "mips32r2" || CPUName == "mips32r3" ||
+                      CPUName == "mips32r5",
+                  "march=mips32r2", Flags);
   addMultilibFlag(CPUName == "mips32r6", "march=mips32r6", Flags);
   addMultilibFlag(CPUName == "mips64", "march=mips64", Flags);
-  addMultilibFlag(CPUName == "mips64r2" || CPUName == "octeon",
+  addMultilibFlag(CPUName == "mips64r2" || CPUName == "mips64r3" ||
+                      CPUName == "mips64r5" || CPUName == "octeon",
                   "march=mips64r2", Flags);
   addMultilibFlag(isMicroMips(Args), "mmicromips", Flags);
   addMultilibFlag(tools::mips::isUCLibc(Args), "muclibc", Flags);
@@ -2073,6 +2084,7 @@ bool Generic_GCC::IsIntegratedAssemblerDefault() const {
          getTriple().getArch() == llvm::Triple::ppc64 ||
          getTriple().getArch() == llvm::Triple::ppc64le ||
          getTriple().getArch() == llvm::Triple::sparc ||
+         getTriple().getArch() == llvm::Triple::sparcel ||
          getTriple().getArch() == llvm::Triple::sparcv9 ||
          getTriple().getArch() == llvm::Triple::systemz;
 }
@@ -2085,7 +2097,8 @@ void Generic_ELF::addClangTargetOptions(const ArgList &DriverArgs,
       getTriple().getArch() == llvm::Triple::aarch64_be ||
       (getTriple().getOS() == llvm::Triple::Linux &&
        (!V.isOlderThan(4, 7, 0) ||
-        getTriple().getEnvironment() == llvm::Triple::Android));
+        getTriple().getEnvironment() == llvm::Triple::Android)) ||
+      getTriple().getOS() == llvm::Triple::NaCl;
 
   if (DriverArgs.hasFlag(options::OPT_fuse_init_array,
                          options::OPT_fno_use_init_array,
@@ -2115,6 +2128,30 @@ std::string Hexagon_TC::GetGnuDir(const std::string &InstalledDir,
   return InstallRelDir;
 }
 
+const char *Hexagon_TC::GetSmallDataThreshold(const ArgList &Args)
+{
+  Arg *A;
+
+  A = Args.getLastArg(options::OPT_G,
+                      options::OPT_G_EQ,
+                      options::OPT_msmall_data_threshold_EQ);
+  if (A)
+    return A->getValue();
+
+  A = Args.getLastArg(options::OPT_shared,
+                      options::OPT_fpic,
+                      options::OPT_fPIC);
+  if (A)
+    return "0";
+
+  return 0;
+}
+
+bool Hexagon_TC::UsesG0(const char* smallDataThreshold)
+{
+  return smallDataThreshold && smallDataThreshold[0] == '0';
+}
+
 static void GetHexagonLibraryPaths(
   const ArgList &Args,
   const std::string &Ver,
@@ -2246,7 +2283,7 @@ void Hexagon_TC::AddClangCXXStdlibIncludeArgs(const ArgList &DriverArgs,
 
   llvm::sys::path::append(IncludeDir, "hexagon/include/c++/");
   llvm::sys::path::append(IncludeDir, Ver);
-  addSystemInclude(DriverArgs, CC1Args, IncludeDir.str());
+  addSystemInclude(DriverArgs, CC1Args, IncludeDir);
 }
 
 ToolChain::CXXStdlibType
@@ -2308,6 +2345,159 @@ StringRef Hexagon_TC::GetTargetCPU(const ArgList &Args)
 }
 // End Hexagon
 
+/// NaCl Toolchain
+NaCl_TC::NaCl_TC(const Driver &D, const llvm::Triple &Triple,
+                 const ArgList &Args)
+  : Generic_ELF(D, Triple, Args) {
+
+  // Remove paths added by Generic_GCC. NaCl Toolchain cannot use the
+  // default paths, and must instead only use the paths provided
+  // with this toolchain based on architecture.
+  path_list& file_paths = getFilePaths();
+  path_list& prog_paths = getProgramPaths();
+
+  file_paths.clear();
+  prog_paths.clear();
+
+  // Path for library files (libc.a, ...)
+  std::string FilePath(getDriver().Dir + "/../");
+
+  // Path for tools (clang, ld, etc..)
+  std::string ProgPath(getDriver().Dir + "/../");
+
+  // Path for toolchain libraries (libgcc.a, ...)
+  std::string ToolPath(getDriver().ResourceDir + "/lib/");
+
+  switch(Triple.getArch()) {
+    case llvm::Triple::x86: {
+      file_paths.push_back(FilePath + "x86_64-nacl/lib32");
+      file_paths.push_back(FilePath + "x86_64-nacl/usr/lib32");
+      prog_paths.push_back(ProgPath + "x86_64-nacl/bin");
+      file_paths.push_back(ToolPath + "i686-nacl");
+      break;
+    }
+    case llvm::Triple::x86_64: {
+      file_paths.push_back(FilePath + "x86_64-nacl/lib");
+      file_paths.push_back(FilePath + "x86_64-nacl/usr/lib");
+      prog_paths.push_back(ProgPath + "x86_64-nacl/bin");
+      file_paths.push_back(ToolPath + "x86_64-nacl");
+      break;
+    }
+    case llvm::Triple::arm: {
+      file_paths.push_back(FilePath + "arm-nacl/lib");
+      file_paths.push_back(FilePath + "arm-nacl/usr/lib");
+      prog_paths.push_back(ProgPath + "arm-nacl/bin");
+      file_paths.push_back(ToolPath + "arm-nacl");
+      break;
+    }
+    default:
+      break;
+  }
+
+  // Use provided linker, not system linker
+  Linker = GetProgramPath("ld");
+  NaClArmMacrosPath = GetFilePath("nacl-arm-macros.s");
+}
+
+void NaCl_TC::AddClangSystemIncludeArgs(const ArgList &DriverArgs,
+                                        ArgStringList &CC1Args) const {
+  const Driver &D = getDriver();
+  if (DriverArgs.hasArg(options::OPT_nostdinc))
+    return;
+
+  if (!DriverArgs.hasArg(options::OPT_nobuiltininc)) {
+    SmallString<128> P(D.ResourceDir);
+    llvm::sys::path::append(P, "include");
+    addSystemInclude(DriverArgs, CC1Args, P.str());
+  }
+
+  if (DriverArgs.hasArg(options::OPT_nostdlibinc))
+    return;
+
+  SmallString<128> P(D.Dir + "/../");
+  if (getTriple().getArch() == llvm::Triple::arm) {
+    llvm::sys::path::append(P, "arm-nacl/usr/include");
+  } else if (getTriple().getArch() == llvm::Triple::x86) {
+    llvm::sys::path::append(P, "x86_64-nacl/usr/include");
+  } else if (getTriple().getArch() == llvm::Triple::x86_64) {
+    llvm::sys::path::append(P, "x86_64-nacl/usr/include");
+  } else {
+    return;
+  }
+
+  addSystemInclude(DriverArgs, CC1Args, P.str());
+  llvm::sys::path::remove_filename(P);
+  llvm::sys::path::remove_filename(P);
+  llvm::sys::path::append(P, "include");
+  addSystemInclude(DriverArgs, CC1Args, P.str());
+}
+
+void NaCl_TC::AddCXXStdlibLibArgs(const ArgList &Args,
+                                  ArgStringList &CmdArgs) const {
+  // Check for -stdlib= flags. We only support libc++ but this consumes the arg
+  // if the value is libc++, and emits an error for other values.
+  GetCXXStdlibType(Args);
+  CmdArgs.push_back("-lc++");
+}
+
+void NaCl_TC::AddClangCXXStdlibIncludeArgs(const ArgList &DriverArgs,
+                                           ArgStringList &CC1Args) const {
+  const Driver &D = getDriver();
+  if (DriverArgs.hasArg(options::OPT_nostdlibinc) ||
+      DriverArgs.hasArg(options::OPT_nostdincxx))
+    return;
+
+  // Check for -stdlib= flags. We only support libc++ but this consumes the arg
+  // if the value is libc++, and emits an error for other values.
+  GetCXXStdlibType(DriverArgs);
+
+  if (getTriple().getArch() == llvm::Triple::arm) {
+    SmallString<128> P(D.Dir + "/../");
+    llvm::sys::path::append(P, "arm-nacl/include/c++/v1");
+    addSystemInclude(DriverArgs, CC1Args, P.str());
+  } else if (getTriple().getArch() == llvm::Triple::x86) {
+    SmallString<128> P(D.Dir + "/../");
+    llvm::sys::path::append(P, "x86_64-nacl/include/c++/v1");
+    addSystemInclude(DriverArgs, CC1Args, P.str());
+  } else if (getTriple().getArch() == llvm::Triple::x86_64) {
+    SmallString<128> P(D.Dir + "/../");
+    llvm::sys::path::append(P, "x86_64-nacl/include/c++/v1");
+    addSystemInclude(DriverArgs, CC1Args, P.str());
+  }
+}
+
+ToolChain::CXXStdlibType NaCl_TC::GetCXXStdlibType(const ArgList &Args) const {
+  if (Arg *A = Args.getLastArg(options::OPT_stdlib_EQ)) {
+    StringRef Value = A->getValue();
+    if (Value == "libc++")
+      return ToolChain::CST_Libcxx;
+    getDriver().Diag(diag::err_drv_invalid_stdlib_name)
+      << A->getAsString(Args);
+  }
+
+  return ToolChain::CST_Libcxx;
+}
+
+std::string NaCl_TC::ComputeEffectiveClangTriple(
+    const ArgList &Args, types::ID InputType) const {
+  llvm::Triple TheTriple(ComputeLLVMTriple(Args, InputType));
+  if (TheTriple.getArch() == llvm::Triple::arm &&
+      TheTriple.getEnvironment() == llvm::Triple::UnknownEnvironment)
+    TheTriple.setEnvironment(llvm::Triple::GNUEABIHF);
+  return TheTriple.getTriple();
+}
+
+Tool *NaCl_TC::buildLinker() const {
+  return new tools::nacltools::Link(*this);
+}
+
+Tool *NaCl_TC::buildAssembler() const {
+  if (getTriple().getArch() == llvm::Triple::arm)
+    return new tools::nacltools::AssembleARM(*this);
+  return new tools::gnutools::Assemble(*this);
+}
+// End NaCl
+
 /// TCEToolChain - A tool chain using the llvm bitcode tools to perform
 /// all subcommands. See http://tce.cs.tut.fi for our peculiar target.
 /// Currently does not support anything else but compilation.
@@ -2341,6 +2531,36 @@ bool TCEToolChain::isPICDefaultForced() const {
   return false;
 }
 
+// CloudABI - CloudABI tool chain which can call ld(1) directly.
+
+CloudABI::CloudABI(const Driver &D, const llvm::Triple &Triple,
+                   const ArgList &Args)
+    : Generic_ELF(D, Triple, Args) {
+  SmallString<128> P(getDriver().Dir);
+  llvm::sys::path::append(P, "..", getTriple().str(), "lib");
+  getFilePaths().push_back(P.str());
+}
+
+void CloudABI::AddClangCXXStdlibIncludeArgs(const ArgList &DriverArgs,
+                                            ArgStringList &CC1Args) const {
+  if (DriverArgs.hasArg(options::OPT_nostdlibinc) &&
+      DriverArgs.hasArg(options::OPT_nostdincxx))
+    return;
+
+  SmallString<128> P(getDriver().Dir);
+  llvm::sys::path::append(P, "..", getTriple().str(), "include/c++/v1");
+  addSystemInclude(DriverArgs, CC1Args, P.str());
+}
+
+void CloudABI::AddCXXStdlibLibArgs(const ArgList &Args,
+                                   ArgStringList &CmdArgs) const {
+  CmdArgs.push_back("-lc++");
+  CmdArgs.push_back("-lc++abi");
+  CmdArgs.push_back("-lunwind");
+}
+
+Tool *CloudABI::buildLinker() const { return new tools::cloudabi::Link(*this); }
+
 /// OpenBSD - OpenBSD tool chain which can call as(1) and ld(1) directly.
 
 OpenBSD::OpenBSD(const Driver &D, const llvm::Triple& Triple, const ArgList &Args)
@@ -2677,10 +2897,12 @@ enum Distro {
   DebianSqueeze,
   DebianWheezy,
   DebianJessie,
+  DebianStretch,
   Exherbo,
   RHEL4,
   RHEL5,
   RHEL6,
+  RHEL7,
   Fedora,
   OpenSUSE,
   UbuntuHardy,
@@ -2696,11 +2918,13 @@ enum Distro {
   UbuntuRaring,
   UbuntuSaucy,
   UbuntuTrusty,
+  UbuntuUtopic,
+  UbuntuVivid,
   UnknownDistro
 };
 
 static bool IsRedhat(enum Distro Distro) {
-  return Distro == Fedora || (Distro >= RHEL4 && Distro <= RHEL6);
+  return Distro == Fedora || (Distro >= RHEL4 && Distro <= RHEL7);
 }
 
 static bool IsOpenSUSE(enum Distro Distro) {
@@ -2708,11 +2932,11 @@ static bool IsOpenSUSE(enum Distro Distro) {
 }
 
 static bool IsDebian(enum Distro Distro) {
-  return Distro >= DebianLenny && Distro <= DebianJessie;
+  return Distro >= DebianLenny && Distro <= DebianStretch;
 }
 
 static bool IsUbuntu(enum Distro Distro) {
-  return Distro >= UbuntuHardy && Distro <= UbuntuTrusty;
+  return Distro >= UbuntuHardy && Distro <= UbuntuVivid;
 }
 
 static Distro DetectDistro(llvm::Triple::ArchType Arch) {
@@ -2739,6 +2963,8 @@ static Distro DetectDistro(llvm::Triple::ArchType Arch) {
           .Case("raring", UbuntuRaring)
           .Case("saucy", UbuntuSaucy)
           .Case("trusty", UbuntuTrusty)
+          .Case("utopic", UbuntuUtopic)
+          .Case("vivid", UbuntuVivid)
           .Default(UnknownDistro);
     return Version;
   }
@@ -2750,7 +2976,9 @@ static Distro DetectDistro(llvm::Triple::ArchType Arch) {
       return Fedora;
     if (Data.startswith("Red Hat Enterprise Linux") ||
         Data.startswith("CentOS")) {
-      if (Data.find("release 6") != StringRef::npos)
+      if (Data.find("release 7") != StringRef::npos)
+        return RHEL7;
+      else if (Data.find("release 6") != StringRef::npos)
         return RHEL6;
       else if (Data.find("release 5") != StringRef::npos)
         return RHEL5;
@@ -2771,6 +2999,8 @@ static Distro DetectDistro(llvm::Triple::ArchType Arch) {
       return DebianWheezy;
     else if (Data.startswith("jessie/sid")  || Data[0] == '8')
       return DebianJessie;
+    else if (Data.startswith("stretch/sid") || Data[0] == '9')
+      return DebianStretch;
     return UnknownDistro;
   }
 
@@ -2967,8 +3197,7 @@ Linux::Linux(const Driver &D, const llvm::Triple &Triple, const ArgList &Args)
   if (IsRedhat(Distro))
     ExtraOpts.push_back("--no-add-needed");
 
-  if (Distro == DebianSqueeze || Distro == DebianWheezy ||
-      Distro == DebianJessie || IsOpenSUSE(Distro) ||
+  if ((IsDebian(Distro) && Distro >= DebianSqueeze) || IsOpenSUSE(Distro) ||
       (IsRedhat(Distro) && Distro != RHEL4 && Distro != RHEL5) ||
       (IsUbuntu(Distro) && Distro >= UbuntuKarmic))
     ExtraOpts.push_back("--build-id");
@@ -3144,7 +3373,7 @@ void Linux::AddClangSystemIncludeArgs(const ArgList &DriverArgs,
   if (!DriverArgs.hasArg(options::OPT_nobuiltininc)) {
     SmallString<128> P(D.ResourceDir);
     llvm::sys::path::append(P, "include");
-    addSystemInclude(DriverArgs, CC1Args, P.str());
+    addSystemInclude(DriverArgs, CC1Args, P);
   }
 
   if (DriverArgs.hasArg(options::OPT_nostdlibinc))
@@ -3168,7 +3397,7 @@ void Linux::AddClangSystemIncludeArgs(const ArgList &DriverArgs,
 
   // Add include directories specific to the selected multilib set and multilib.
   if (GCCInstallation.isValid()) {
-    auto Callback = Multilibs.includeDirsCallback();
+    const auto &Callback = Multilibs.includeDirsCallback();
     if (Callback) {
       const auto IncludePaths = Callback(GCCInstallation.getInstallPath(),
                                          GCCInstallation.getTriple().str(),
diff --git a/lib/Driver/ToolChains.h b/lib/Driver/ToolChains.h
index 47fb10d3f6c5..8906e2101ea0 100644
--- a/lib/Driver/ToolChains.h
+++ b/lib/Driver/ToolChains.h
@@ -152,7 +152,7 @@ protected:
 public:
   Generic_GCC(const Driver &D, const llvm::Triple &Triple,
               const llvm::opt::ArgList &Args);
-  ~Generic_GCC();
+  ~Generic_GCC() override;
 
   void printVerboseInfo(raw_ostream &OS) const override;
 
@@ -196,7 +196,7 @@ private:
 public:
   MachO(const Driver &D, const llvm::Triple &Triple,
              const llvm::opt::ArgList &Args);
-  ~MachO();
+  ~MachO() override;
 
   /// @name MachO specific toolchain API
   /// {
@@ -239,6 +239,13 @@ public:
                          bool IsEmbedded = false,
                          bool AddRPath = false) const;
 
+  /// Add any profiling runtime libraries that are needed. This is essentially a
+  /// MachO specific version of addProfileRT in Tools.cpp.
+  virtual void addProfileRTLibs(const llvm::opt::ArgList &Args,
+                                llvm::opt::ArgStringList &CmdArgs) const {
+    // There aren't any profiling libs for embedded targets currently.
+  }
+
   /// }
   /// @name ToolChain Implementation
   /// {
@@ -345,7 +352,7 @@ private:
 public:
   Darwin(const Driver &D, const llvm::Triple &Triple,
          const llvm::opt::ArgList &Args);
-  ~Darwin();
+  ~Darwin() override;
 
   std::string ComputeEffectiveClangTriple(const llvm::opt::ArgList &Args,
                                           types::ID InputType) const override;
@@ -362,10 +369,12 @@ public:
                          llvm::opt::ArgStringList &CmdArgs) const override;
 
   bool isKernelStatic() const override {
-    return !isTargetIPhoneOS() || isIPhoneOSVersionLT(6, 0) ||
-           getTriple().getArch() == llvm::Triple::aarch64;
+    return !isTargetIPhoneOS() || isIPhoneOSVersionLT(6, 0);
   }
 
+  void addProfileRTLibs(const llvm::opt::ArgList &Args,
+                        llvm::opt::ArgStringList &CmdArgs) const override;
+
 protected:
   /// }
   /// @name Darwin specific Toolchain functions
@@ -488,13 +497,17 @@ public:
   AddCCKextLibArgs(const llvm::opt::ArgList &Args,
                    llvm::opt::ArgStringList &CmdArgs) const override;
 
-  virtual void addClangWarningOptions(llvm::opt::ArgStringList &CC1Args)
-                                                      const override;
+  void addClangWarningOptions(llvm::opt::ArgStringList &CC1Args) const override;
 
   void
   AddLinkARCArgs(const llvm::opt::ArgList &Args,
                  llvm::opt::ArgStringList &CmdArgs) const override;
   /// }
+
+private:
+  void AddLinkSanitizerLibArgs(const llvm::opt::ArgList &Args,
+                               llvm::opt::ArgStringList &CmdArgs,
+                               StringRef Sanitizer) const;
 };
 
 class LLVM_LIBRARY_VISIBILITY Generic_ELF : public Generic_GCC {
@@ -508,6 +521,31 @@ public:
                              llvm::opt::ArgStringList &CC1Args) const override;
 };
 
+class LLVM_LIBRARY_VISIBILITY CloudABI : public Generic_ELF {
+public:
+  CloudABI(const Driver &D, const llvm::Triple &Triple,
+           const llvm::opt::ArgList &Args);
+  bool HasNativeLLVMSupport() const override { return true; }
+
+  bool IsMathErrnoDefault() const override { return false; }
+  bool IsObjCNonFragileABIDefault() const override { return true; }
+
+  CXXStdlibType GetCXXStdlibType(const llvm::opt::ArgList &Args)
+      const override {
+    return ToolChain::CST_Libcxx;
+  }
+  void AddClangCXXStdlibIncludeArgs(
+      const llvm::opt::ArgList &DriverArgs,
+      llvm::opt::ArgStringList &CC1Args) const override;
+  void AddCXXStdlibLibArgs(const llvm::opt::ArgList &Args,
+                           llvm::opt::ArgStringList &CmdArgs) const override;
+
+  bool isPIEDefault() const override { return false; }
+
+protected:
+  Tool *buildLinker() const override;
+};
+
 class LLVM_LIBRARY_VISIBILITY Solaris : public Generic_GCC {
 public:
   Solaris(const Driver &D, const llvm::Triple &Triple,
@@ -670,7 +708,7 @@ protected:
 public:
   Hexagon_TC(const Driver &D, const llvm::Triple &Triple,
              const llvm::opt::ArgList &Args);
-  ~Hexagon_TC();
+  ~Hexagon_TC() override;
 
   void
   AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
@@ -686,6 +724,49 @@ public:
                                const llvm::opt::ArgList &Args);
 
   static StringRef GetTargetCPU(const llvm::opt::ArgList &Args);
+
+  static const char *GetSmallDataThreshold(const llvm::opt::ArgList &Args);
+
+  static bool UsesG0(const char* smallDataThreshold);
+};
+
+class LLVM_LIBRARY_VISIBILITY NaCl_TC : public Generic_ELF {
+public:
+  NaCl_TC(const Driver &D, const llvm::Triple &Triple,
+          const llvm::opt::ArgList &Args);
+
+  void
+  AddClangSystemIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                            llvm::opt::ArgStringList &CC1Args) const override;
+  void
+  AddClangCXXStdlibIncludeArgs(const llvm::opt::ArgList &DriverArgs,
+                               llvm::opt::ArgStringList &CC1Args) const override;
+
+  CXXStdlibType
+  GetCXXStdlibType(const llvm::opt::ArgList &Args) const override;
+
+  void
+  AddCXXStdlibLibArgs(const llvm::opt::ArgList &Args,
+                      llvm::opt::ArgStringList &CmdArgs) const override;
+
+  bool
+  IsIntegratedAssemblerDefault() const override { return false; }
+
+  // Get the path to the file containing NaCl's ARM macros. It lives in NaCl_TC
+  // because the AssembleARM tool needs a const char * that it can pass around
+  // and the toolchain outlives all the jobs.
+  const char *GetNaClArmMacrosPath() const { return NaClArmMacrosPath.c_str(); }
+
+  std::string ComputeEffectiveClangTriple(const llvm::opt::ArgList &Args,
+                                          types::ID InputType) const override;
+  std::string Linker;
+
+protected:
+  Tool *buildLinker() const override;
+  Tool *buildAssembler() const override;
+
+private:
+  std::string NaClArmMacrosPath;
 };
 
 /// TCEToolChain - A tool chain using the llvm bitcode tools to perform
@@ -694,7 +775,7 @@ class LLVM_LIBRARY_VISIBILITY TCEToolChain : public ToolChain {
 public:
   TCEToolChain(const Driver &D, const llvm::Triple &Triple,
                const llvm::opt::ArgList &Args);
-  ~TCEToolChain();
+  ~TCEToolChain() override;
 
   bool IsMathErrnoDefault() const override;
   bool isPICDefault() const override;
diff --git a/lib/Driver/Tools.cpp b/lib/Driver/Tools.cpp
index 75eef9ecdf09..527038f9650b 100644
--- a/lib/Driver/Tools.cpp
+++ b/lib/Driver/Tools.cpp
@@ -10,6 +10,7 @@
 #include "Tools.h"
 #include "InputInfo.h"
 #include "ToolChains.h"
+#include "clang/Basic/CharInfo.h"
 #include "clang/Basic/LangOptions.h"
 #include "clang/Basic/ObjCRuntime.h"
 #include "clang/Basic/Version.h"
@@ -23,6 +24,7 @@
 #include "clang/Driver/SanitizerArgs.h"
 #include "clang/Driver/ToolChain.h"
 #include "clang/Driver/Util.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringSwitch.h"
@@ -30,16 +32,20 @@
 #include "llvm/Option/Arg.h"
 #include "llvm/Option/ArgList.h"
 #include "llvm/Option/Option.h"
+#include "llvm/Support/TargetParser.h"
 #include "llvm/Support/Compression.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/FileSystem.h"
-#include "llvm/Support/Format.h"
 #include "llvm/Support/Host.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/Process.h"
 #include "llvm/Support/Program.h"
 #include "llvm/Support/raw_ostream.h"
 
+#ifdef LLVM_ON_UNIX
+#include <unistd.h> // For getuid().
+#endif
+
 using namespace clang::driver;
 using namespace clang::driver::tools;
 using namespace clang;
@@ -319,8 +325,9 @@ void Clang::AddPreprocessingOptions(Compilation &C,
     if (A->getOption().matches(options::OPT_M) ||
         A->getOption().matches(options::OPT_MD))
       CmdArgs.push_back("-sys-header-deps");
-
-    if (isa<PrecompileJobAction>(JA))
+    if ((isa<PrecompileJobAction>(JA) &&
+         !Args.hasArg(options::OPT_fno_module_file_deps)) ||
+        Args.hasArg(options::OPT_fmodule_file_deps))
       CmdArgs.push_back("-module-file-deps");
   }
 
@@ -332,6 +339,7 @@ void Clang::AddPreprocessingOptions(Compilation &C,
   }
 
   Args.AddLastArg(CmdArgs, options::OPT_MP);
+  Args.AddLastArg(CmdArgs, options::OPT_MV);
 
   // Convert all -MQ <target> args to -MT <quoted target>
   for (arg_iterator it = Args.filtered_begin(options::OPT_MT,
@@ -377,19 +385,19 @@ void Clang::AddPreprocessingOptions(Compilation &C,
       P += ".dummy";
       if (UsePCH) {
         llvm::sys::path::replace_extension(P, "pch");
-        if (llvm::sys::fs::exists(P.str()))
+        if (llvm::sys::fs::exists(P))
           FoundPCH = true;
       }
 
       if (!FoundPCH) {
         llvm::sys::path::replace_extension(P, "pth");
-        if (llvm::sys::fs::exists(P.str()))
+        if (llvm::sys::fs::exists(P))
           FoundPTH = true;
       }
 
       if (!FoundPCH && !FoundPTH) {
         llvm::sys::path::replace_extension(P, "gch");
-        if (llvm::sys::fs::exists(P.str())) {
+        if (llvm::sys::fs::exists(P)) {
           FoundPCH = UsePCH;
           FoundPTH = !UsePCH;
         }
@@ -402,12 +410,12 @@ void Clang::AddPreprocessingOptions(Compilation &C,
             CmdArgs.push_back("-include-pch");
           else
             CmdArgs.push_back("-include-pth");
-          CmdArgs.push_back(Args.MakeArgString(P.str()));
+          CmdArgs.push_back(Args.MakeArgString(P));
           continue;
         } else {
           // Ignore the PCH if not first on command line and emit warning.
           D.Diag(diag::warn_drv_pch_not_first_include)
-              << P.str() << A->getAsString(Args);
+              << P << A->getAsString(Args);
         }
       }
     }
@@ -489,6 +497,7 @@ static bool isSignedCharDefault(const llvm::Triple &Triple) {
       return true;
     return false;
 
+  case llvm::Triple::hexagon:
   case llvm::Triple::ppc64le:
   case llvm::Triple::systemz:
   case llvm::Triple::xcore:
@@ -536,75 +545,88 @@ static void getARMFPUFeatures(const Driver &D, const Arg *A,
                               std::vector<const char *> &Features) {
   StringRef FPU = A->getValue();
 
-  // Set the target features based on the FPU.
-  if (FPU == "fpa" || FPU == "fpe2" || FPU == "fpe3" || FPU == "maverick") {
-    // Disable any default FPU support.
+  // FIXME: Why does "none" disable more than "invalid"?
+  if (FPU == "none") {
     Features.push_back("-vfp2");
     Features.push_back("-vfp3");
+    Features.push_back("-vfp4");
+    Features.push_back("-fp-armv8");
+    Features.push_back("-crypto");
     Features.push_back("-neon");
-  } else if (FPU == "vfp") {
+    return;
+  }
+
+  // FIXME: Make sure we differentiate sp-only.
+  if (FPU.find("-sp-") != StringRef::npos) {
+    Features.push_back("+fp-only-sp");
+  }
+
+  // All other FPU types, valid or invalid.
+  switch(llvm::ARMTargetParser::parseFPU(FPU)) {
+  case llvm::ARM::FK_INVALID:
+  case llvm::ARM::FK_SOFTVFP:
+    Features.push_back("-vfp2");
+    Features.push_back("-vfp3");
+    Features.push_back("-neon");
+    break;
+  case llvm::ARM::FK_VFP:
+  case llvm::ARM::FK_VFPV2:
     Features.push_back("+vfp2");
     Features.push_back("-neon");
-  } else if (FPU == "vfp3-d16" || FPU == "vfpv3-d16") {
-    Features.push_back("+vfp3");
+    break;
+  case llvm::ARM::FK_VFPV3_D16:
     Features.push_back("+d16");
-    Features.push_back("-neon");
-  } else if (FPU == "vfp3" || FPU == "vfpv3") {
+    // fall-through
+  case llvm::ARM::FK_VFPV3:
     Features.push_back("+vfp3");
     Features.push_back("-neon");
-  } else if (FPU == "vfp4-d16" || FPU == "vfpv4-d16") {
-    Features.push_back("+vfp4");
+    break;
+  case llvm::ARM::FK_VFPV4_D16:
     Features.push_back("+d16");
-    Features.push_back("-neon");
-  } else if (FPU == "vfp4" || FPU == "vfpv4") {
+    // fall-through
+  case llvm::ARM::FK_VFPV4:
     Features.push_back("+vfp4");
     Features.push_back("-neon");
-  } else if (FPU == "fp4-sp-d16" || FPU == "fpv4-sp-d16") {
-    Features.push_back("+vfp4");
-    Features.push_back("+d16");
-    Features.push_back("+fp-only-sp");
-    Features.push_back("-neon");
-  } else if (FPU == "fp5-sp-d16" || FPU == "fpv5-sp-d16") {
-    Features.push_back("+fp-armv8");
-    Features.push_back("+fp-only-sp");
+    break;
+  case llvm::ARM::FK_FPV5_D16:
     Features.push_back("+d16");
-    Features.push_back("-neon");
-    Features.push_back("-crypto");
-  } else if (FPU == "fp5-dp-d16" || FPU == "fpv5-dp-d16" ||
-             FPU == "fp5-d16" || FPU == "fpv5-d16") {
+    // fall-through
+  case llvm::ARM::FK_FP_ARMV8:
     Features.push_back("+fp-armv8");
-    Features.push_back("+d16");
     Features.push_back("-neon");
     Features.push_back("-crypto");
-  } else if (FPU == "fp-armv8") {
-    Features.push_back("+fp-armv8");
-    Features.push_back("-neon");
-    Features.push_back("-crypto");
-  } else if (FPU == "neon-fp-armv8") {
+    break;
+  case llvm::ARM::FK_NEON_FP_ARMV8:
     Features.push_back("+fp-armv8");
     Features.push_back("+neon");
     Features.push_back("-crypto");
-  } else if (FPU == "crypto-neon-fp-armv8") {
+    break;
+  case llvm::ARM::FK_CRYPTO_NEON_FP_ARMV8:
     Features.push_back("+fp-armv8");
     Features.push_back("+neon");
     Features.push_back("+crypto");
-  } else if (FPU == "neon") {
-    Features.push_back("+neon");
-  } else if (FPU == "neon-vfpv3") {
-    Features.push_back("+vfp3");
+    break;
+  case llvm::ARM::FK_NEON:
     Features.push_back("+neon");
-  } else if (FPU == "neon-vfpv4") {
+    break;
+  case llvm::ARM::FK_NEON_VFPV4:
     Features.push_back("+neon");
     Features.push_back("+vfp4");
-  } else if (FPU == "none") {
-    Features.push_back("-vfp2");
-    Features.push_back("-vfp3");
-    Features.push_back("-vfp4");
-    Features.push_back("-fp-armv8");
-    Features.push_back("-crypto");
-    Features.push_back("-neon");
-  } else
+    break;
+  default:
     D.Diag(diag::err_drv_clang_unsupported) << A->getAsString(Args);
+  }
+}
+
+static int getARMSubArchVersionNumber(const llvm::Triple &Triple) {
+  llvm::StringRef Arch = Triple.getArchName();
+  return llvm::ARMTargetParser::parseArchVersion(Arch);
+}
+
+static bool isARMMProfile(const llvm::Triple &Triple) {
+  llvm::StringRef Arch = Triple.getArchName();
+  unsigned Profile = llvm::ARMTargetParser::parseArchProfile(Arch);
+  return Profile == llvm::ARM::PK_M;
 }
 
 // Select the float ABI as determined by -msoft-float, -mhard-float, and
@@ -637,11 +659,8 @@ StringRef tools::arm::getARMFloatABI(const Driver &D, const ArgList &Args,
     case llvm::Triple::IOS: {
       // Darwin defaults to "softfp" for v6 and v7.
       //
-      // FIXME: Factor out an ARM class so we can cache the arch somewhere.
-      std::string ArchName =
-        arm::getLLVMArchSuffixForARM(arm::getARMTargetCPU(Args, Triple));
-      if (StringRef(ArchName).startswith("v6") ||
-          StringRef(ArchName).startswith("v7"))
+      if (getARMSubArchVersionNumber(Triple) == 6 ||
+          getARMSubArchVersionNumber(Triple) == 7)
         FloatABI = "softfp";
       else
         FloatABI = "soft";
@@ -681,9 +700,7 @@ StringRef tools::arm::getARMFloatABI(const Driver &D, const ArgList &Args,
         FloatABI = "softfp";
         break;
       case llvm::Triple::Android: {
-        std::string ArchName =
-          arm::getLLVMArchSuffixForARM(arm::getARMTargetCPU(Args, Triple));
-        if (StringRef(ArchName).startswith("v7"))
+        if (getARMSubArchVersionNumber(Triple) == 7)
           FloatABI = "softfp";
         else
           FloatABI = "soft";
@@ -736,6 +753,25 @@ static void getARMTargetFeatures(const Driver &D, const llvm::Triple &Triple,
   if (const Arg *A = Args.getLastArg(options::OPT_mhwdiv_EQ))
     getARMHWDivFeatures(D, A, Args, Features);
 
+  // Check if -march is valid by checking if it can be canonicalised. getARMArch
+  // is used here instead of just checking the -march value in order to handle
+  // -march=native correctly.
+  if (const Arg *A = Args.getLastArg(options::OPT_march_EQ)) {
+    StringRef Arch = arm::getARMArch(Args, Triple);
+    if (llvm::ARMTargetParser::getCanonicalArchName(Arch).empty())
+      D.Diag(diag::err_drv_clang_unsupported) << A->getAsString(Args);
+  }
+
+  // We do a similar thing with -mcpu, but here things are complicated because
+  // the only function we have to check if a cpu is valid is
+  // getLLVMArchSuffixForARM which also needs an architecture.
+  if (const Arg *A = Args.getLastArg(options::OPT_mcpu_EQ)) {
+    StringRef CPU = arm::getARMTargetCPU(Args, Triple);
+    StringRef Arch = arm::getARMArch(Args, Triple);
+    if (strcmp(arm::getLLVMArchSuffixForARM(CPU, Arch), "") == 0)
+      D.Diag(diag::err_drv_clang_unsupported) << A->getAsString(Args);
+  }
+
   // Setting -msoft-float effectively disables NEON because of the GCC
   // implementation, although the same isn't true of VFP or VFP3.
   if (FloatABI == "soft") {
@@ -744,14 +780,17 @@ static void getARMTargetFeatures(const Driver &D, const llvm::Triple &Triple,
     Features.push_back("-crypto");
   }
 
-  // En/disable crc
-  if (Arg *A = Args.getLastArg(options::OPT_mcrc,
-                               options::OPT_mnocrc)) {
+  // En/disable crc code generation.
+  if (Arg *A = Args.getLastArg(options::OPT_mcrc, options::OPT_mnocrc)) {
     if (A->getOption().matches(options::OPT_mcrc))
       Features.push_back("+crc");
     else
       Features.push_back("-crc");
   }
+
+  if (Triple.getSubArch() == llvm::Triple::SubArchType::ARMSubArch_v8_1a) {
+    Features.insert(Features.begin(), "+v8.1a");
+  }
 }
 
 void Clang::AddARMTargetArgs(const ArgList &Args,
@@ -761,7 +800,6 @@ void Clang::AddARMTargetArgs(const ArgList &Args,
   // Get the effective triple, which takes into account the deployment target.
   std::string TripleStr = getToolChain().ComputeEffectiveClangTriple(Args);
   llvm::Triple Triple(TripleStr);
-  std::string CPUName = arm::getARMTargetCPU(Args, Triple);
 
   // Select the ABI to use.
   //
@@ -775,7 +813,7 @@ void Clang::AddARMTargetArgs(const ArgList &Args,
     // the frontend matches that.
     if (Triple.getEnvironment() == llvm::Triple::EABI ||
         Triple.getOS() == llvm::Triple::UnknownOS ||
-        StringRef(CPUName).startswith("cortex-m")) {
+        isARMMProfile(Triple)) {
       ABIName = "aapcs";
     } else {
       ABIName = "apcs-gnu";
@@ -856,12 +894,14 @@ void Clang::AddARMTargetArgs(const ArgList &Args,
     }
   }
 
-  // Setting -mno-global-merge disables the codegen global merge pass. Setting 
-  // -mglobal-merge has no effect as the pass is enabled by default.
+  // Forward the -mglobal-merge option for explicit control over the pass.
   if (Arg *A = Args.getLastArg(options::OPT_mglobal_merge,
                                options::OPT_mno_global_merge)) {
+    CmdArgs.push_back("-backend-option");
     if (A->getOption().matches(options::OPT_mno_global_merge))
-      CmdArgs.push_back("-mno-global-merge");
+      CmdArgs.push_back("-arm-global-merge=false");
+    else
+      CmdArgs.push_back("-arm-global-merge=true");
   }
 
   if (!Args.hasFlag(options::OPT_mimplicit_float,
@@ -952,12 +992,14 @@ void Clang::AddAArch64TargetArgs(const ArgList &Args,
     CmdArgs.push_back("-aarch64-fix-cortex-a53-835769=1");
   }
 
-  // Setting -mno-global-merge disables the codegen global merge pass. Setting
-  // -mglobal-merge has no effect as the pass is enabled by default.
+  // Forward the -mglobal-merge option for explicit control over the pass.
   if (Arg *A = Args.getLastArg(options::OPT_mglobal_merge,
                                options::OPT_mno_global_merge)) {
+    CmdArgs.push_back("-backend-option");
     if (A->getOption().matches(options::OPT_mno_global_merge))
-      CmdArgs.push_back("-mno-global-merge");
+      CmdArgs.push_back("-aarch64-global-merge=false");
+    else
+      CmdArgs.push_back("-aarch64-global-merge=true");
   }
 
   if (Args.hasArg(options::OPT_ffixed_x18)) {
@@ -1096,17 +1138,6 @@ static void getMIPSTargetFeatures(const Driver &D, const llvm::Triple &Triple,
   mips::getMipsCPUAndABI(Args, Triple, CPUName, ABIName);
   ABIName = getGnuCompatibleMipsABIName(ABIName);
 
-  // Always override the backend's default ABI.
-  std::string ABIFeature = llvm::StringSwitch<StringRef>(ABIName)
-                               .Case("32", "+o32")
-                               .Case("n32", "+n32")
-                               .Case("64", "+n64")
-                               .Case("eabi", "+eabi")
-                               .Default(("+" + ABIName).str());
-  Features.push_back("-o32");
-  Features.push_back("-n64");
-  Features.push_back(Args.MakeArgString(ABIFeature));
-
   AddTargetFeature(Args, Features, options::OPT_mno_abicalls,
                    options::OPT_mabicalls, "noabicalls");
 
@@ -1120,11 +1151,21 @@ static void getMIPSTargetFeatures(const Driver &D, const llvm::Triple &Triple,
 
   if (Arg *A = Args.getLastArg(options::OPT_mnan_EQ)) {
     StringRef Val = StringRef(A->getValue());
-    if (Val == "2008")
-      Features.push_back("+nan2008");
-    else if (Val == "legacy")
-      Features.push_back("-nan2008");
-    else
+    if (Val == "2008") {
+      if (mips::getSupportedNanEncoding(CPUName) & mips::Nan2008)
+        Features.push_back("+nan2008");
+      else {
+        Features.push_back("-nan2008");
+        D.Diag(diag::warn_target_unsupported_nan2008) << CPUName;
+      }
+    } else if (Val == "legacy") {
+      if (mips::getSupportedNanEncoding(CPUName) & mips::NanLegacy)
+        Features.push_back("-nan2008");
+      else {
+        Features.push_back("+nan2008");
+        D.Diag(diag::warn_target_unsupported_nanlegacy) << CPUName;
+      }
+    } else
       D.Diag(diag::err_drv_unsupported_option_argument)
           << A->getOption().getName() << Val;
   }
@@ -1322,9 +1363,22 @@ void Clang::AddPPCTargetArgs(const ArgList &Args,
     ABIName = A->getValue();
   } else if (getToolChain().getTriple().isOSLinux())
     switch(getToolChain().getArch()) {
-    case llvm::Triple::ppc64:
+    case llvm::Triple::ppc64: {
+      // When targeting a processor that supports QPX, or if QPX is
+      // specifically enabled, default to using the ABI that supports QPX (so
+      // long as it is not specifically disabled).
+      bool HasQPX = false;
+      if (Arg *A = Args.getLastArg(options::OPT_mcpu_EQ))
+        HasQPX = A->getValue() == StringRef("a2q");
+      HasQPX = Args.hasFlag(options::OPT_mqpx, options::OPT_mno_qpx, HasQPX);
+      if (HasQPX) {
+        ABIName = "elfv1-qpx";
+        break;
+      }
+
       ABIName = "elfv1";
       break;
+    }
     case llvm::Triple::ppc64le:
       ABIName = "elfv2";
       break;
@@ -1410,6 +1464,26 @@ static const char *getSystemZTargetCPU(const ArgList &Args) {
   return "z10";
 }
 
+static void getSystemZTargetFeatures(const ArgList &Args,
+                                     std::vector<const char *> &Features) {
+  // -m(no-)htm overrides use of the transactional-execution facility.
+  if (Arg *A = Args.getLastArg(options::OPT_mhtm,
+                               options::OPT_mno_htm)) {
+    if (A->getOption().matches(options::OPT_mhtm))
+      Features.push_back("+transactional-execution");
+    else
+      Features.push_back("-transactional-execution");
+  }
+  // -m(no-)vx overrides use of the vector facility.
+  if (Arg *A = Args.getLastArg(options::OPT_mvx,
+                               options::OPT_mno_vx)) {
+    if (A->getOption().matches(options::OPT_mvx))
+      Features.push_back("+vector");
+    else
+      Features.push_back("-vector");
+  }
+}
+
 static const char *getX86TargetCPU(const ArgList &Args,
                                    const llvm::Triple &Triple) {
   if (const Arg *A = Args.getLastArg(options::OPT_march_EQ)) {
@@ -1445,6 +1519,10 @@ static const char *getX86TargetCPU(const ArgList &Args,
     return Is64Bit ? "core2" : "yonah";
   }
 
+  // Set up default CPU name for PS4 compilers.
+  if (Triple.isPS4CPU())
+    return "btver2";
+
   // On Android use targets compatible with gcc
   if (Triple.getEnvironment() == llvm::Triple::Android)
     return Is64Bit ? "x86-64" : "i686";
@@ -1512,6 +1590,7 @@ static std::string getCPUName(const ArgList &Args, const llvm::Triple &T) {
   }
 
   case llvm::Triple::sparc:
+  case llvm::Triple::sparcel:
   case llvm::Triple::sparcv9:
     if (const Arg *A = Args.getLastArg(options::OPT_mcpu_EQ))
       return A->getValue();
@@ -1551,10 +1630,20 @@ static void AddGoldPlugin(const ToolChain &ToolChain, const ArgList &Args,
     CmdArgs.push_back(Args.MakeArgString(Twine("-plugin-opt=mcpu=") + CPU));
 }
 
-static void getX86TargetFeatures(const Driver & D,
-                                 const llvm::Triple &Triple,
+static void getX86TargetFeatures(const Driver &D, const llvm::Triple &Triple,
                                  const ArgList &Args,
                                  std::vector<const char *> &Features) {
+  // If -march=native, autodetect the feature list.
+  if (const Arg *A = Args.getLastArg(options::OPT_march_EQ)) {
+    if (StringRef(A->getValue()) == "native") {
+      llvm::StringMap<bool> HostFeatures;
+      if (llvm::sys::getHostCPUFeatures(HostFeatures))
+        for (auto &F : HostFeatures)
+          Features.push_back(Args.MakeArgString((F.second ? "+" : "-") +
+                                                F.first()));
+    }
+  }
+
   if (Triple.getArchName() == "x86_64h") {
     // x86_64h implies quite a few of the more modern subtarget features
     // for Haswell class CPUs, but not all of them. Opt-out of a few.
@@ -1566,7 +1655,7 @@ static void getX86TargetFeatures(const Driver & D,
     Features.push_back("-fsgsbase");
   }
 
-  // Add features to comply with gcc on Android
+  // Add features to be compatible with gcc for Android.
   if (Triple.getEnvironment() == llvm::Triple::Android) {
     if (Triple.getArch() == llvm::Triple::x86_64) {
       Features.push_back("+sse4.2");
@@ -1575,7 +1664,7 @@ static void getX86TargetFeatures(const Driver & D,
       Features.push_back("+ssse3");
   }
 
-  // Set features according to the -arch flag on MSVC
+  // Set features according to the -arch flag on MSVC.
   if (Arg *A = Args.getLastArg(options::OPT__SLASH_arch)) {
     StringRef Arch = A->getValue();
     bool ArchUsed = false;
@@ -1656,39 +1745,16 @@ void Clang::AddX86TargetArgs(const ArgList &Args,
   }
 }
 
-static inline bool HasPICArg(const ArgList &Args) {
-  return Args.hasArg(options::OPT_fPIC)
-    || Args.hasArg(options::OPT_fpic);
-}
-
-static Arg *GetLastSmallDataThresholdArg(const ArgList &Args) {
-  return Args.getLastArg(options::OPT_G,
-                         options::OPT_G_EQ,
-                         options::OPT_msmall_data_threshold_EQ);
-}
-
-static std::string GetHexagonSmallDataThresholdValue(const ArgList &Args) {
-  std::string value;
-  if (HasPICArg(Args))
-    value = "0";
-  else if (Arg *A = GetLastSmallDataThresholdArg(Args)) {
-    value = A->getValue();
-    A->claim();
-  }
-  return value;
-}
-
 void Clang::AddHexagonTargetArgs(const ArgList &Args,
                                  ArgStringList &CmdArgs) const {
-  CmdArgs.push_back("-fno-signed-char");
   CmdArgs.push_back("-mqdsp6-compat");
   CmdArgs.push_back("-Wreturn-type");
 
-  std::string SmallDataThreshold = GetHexagonSmallDataThresholdValue(Args);
-  if (!SmallDataThreshold.empty()) {
+  if (const char* v = toolchains::Hexagon_TC::GetSmallDataThreshold(Args)) {
+    std::string SmallDataThreshold="-hexagon-small-data-threshold=";
+    SmallDataThreshold += v;
     CmdArgs.push_back ("-mllvm");
-    CmdArgs.push_back(Args.MakeArgString(
-                        "-hexagon-small-data-threshold=" + SmallDataThreshold));
+    CmdArgs.push_back(Args.MakeArgString(SmallDataThreshold));
   }
 
   if (!Args.hasArg(options::OPT_fno_short_enums))
@@ -1734,7 +1800,7 @@ static bool DecodeAArch64Mcpu(const Driver &D, StringRef Mcpu, StringRef &CPU,
                               std::vector<const char *> &Features) {
   std::pair<StringRef, StringRef> Split = Mcpu.split("+");
   CPU = Split.first;
-  if (CPU == "cyclone" || CPU == "cortex-a53" || CPU == "cortex-a57") {
+  if (CPU == "cyclone" || CPU == "cortex-a53" || CPU == "cortex-a57" || CPU == "cortex-a72") {
     Features.push_back("+neon");
     Features.push_back("+crc");
     Features.push_back("+crypto");
@@ -1755,8 +1821,17 @@ getAArch64ArchFeaturesFromMarch(const Driver &D, StringRef March,
                                 const ArgList &Args,
                                 std::vector<const char *> &Features) {
   std::pair<StringRef, StringRef> Split = March.split("+");
-  if (Split.first != "armv8-a")
+
+  if (Split.first == "armv8-a" ||
+      Split.first == "armv8a") {
+    // ok, no additional features.
+  } else if (
+      Split.first == "armv8.1-a" ||
+      Split.first == "armv8.1a" ) {
+    Features.push_back("+v8.1a");
+  } else {
     return false;
+  }
 
   if (Split.second.size() && !DecodeAArch64Features(D, Split.second, Features))
     return false;
@@ -1871,9 +1946,13 @@ static void getTargetFeatures(const Driver &D, const llvm::Triple &Triple,
     getPPCTargetFeatures(Args, Features);
     break;
   case llvm::Triple::sparc:
+  case llvm::Triple::sparcel:
   case llvm::Triple::sparcv9:
     getSparcTargetFeatures(Args, Features);
     break;
+  case llvm::Triple::systemz:
+    getSystemZTargetFeatures(Args, Features);
+    break;
   case llvm::Triple::aarch64:
   case llvm::Triple::aarch64_be:
     getAArch64TargetFeatures(D, Args, Features);
@@ -1934,16 +2013,17 @@ static bool exceptionSettings(const ArgList &Args, const llvm::Triple &Triple) {
   return false;
 }
 
-/// addExceptionArgs - Adds exception related arguments to the driver command
-/// arguments. There's a master flag, -fexceptions and also language specific
-/// flags to enable/disable C++ and Objective-C exceptions.
-/// This makes it possible to for example disable C++ exceptions but enable
-/// Objective-C exceptions.
+/// Adds exception related arguments to the driver command arguments. There's a
+/// master flag, -fexceptions and also language specific flags to enable/disable
+/// C++ and Objective-C exceptions. This makes it possible to for example
+/// disable C++ exceptions but enable Objective-C exceptions.
 static void addExceptionArgs(const ArgList &Args, types::ID InputType,
-                             const llvm::Triple &Triple,
-                             bool KernelOrKext,
+                             const ToolChain &TC, bool KernelOrKext,
                              const ObjCRuntime &objcRuntime,
                              ArgStringList &CmdArgs) {
+  const Driver &D = TC.getDriver();
+  const llvm::Triple &Triple = TC.getTriple();
+
   if (KernelOrKext) {
     // -mkernel and -fapple-kext imply no exceptions, so claim exception related
     // arguments now to avoid warnings about unused arguments.
@@ -1971,16 +2051,32 @@ static void addExceptionArgs(const ArgList &Args, types::ID InputType,
   }
 
   if (types::isCXX(InputType)) {
-    bool CXXExceptionsEnabled = Triple.getArch() != llvm::Triple::xcore;
-    if (Arg *A = Args.getLastArg(options::OPT_fcxx_exceptions,
-                                 options::OPT_fno_cxx_exceptions,
-                                 options::OPT_fexceptions,
-                                 options::OPT_fno_exceptions))
+    bool CXXExceptionsEnabled =
+        Triple.getArch() != llvm::Triple::xcore && !Triple.isPS4CPU();
+    Arg *ExceptionArg = Args.getLastArg(
+        options::OPT_fcxx_exceptions, options::OPT_fno_cxx_exceptions,
+        options::OPT_fexceptions, options::OPT_fno_exceptions);
+    if (ExceptionArg)
       CXXExceptionsEnabled =
-          A->getOption().matches(options::OPT_fcxx_exceptions) ||
-          A->getOption().matches(options::OPT_fexceptions);
+          ExceptionArg->getOption().matches(options::OPT_fcxx_exceptions) ||
+          ExceptionArg->getOption().matches(options::OPT_fexceptions);
 
     if (CXXExceptionsEnabled) {
+      if (Triple.isPS4CPU()) {
+        ToolChain::RTTIMode RTTIMode = TC.getRTTIMode();
+        assert(ExceptionArg &&
+               "On the PS4 exceptions should only be enabled if passing "
+               "an argument");
+        if (RTTIMode == ToolChain::RM_DisabledExplicitly) {
+          const Arg *RTTIArg = TC.getRTTIArg();
+          assert(RTTIArg && "RTTI disabled explicitly but no RTTIArg!");
+          D.Diag(diag::err_drv_argument_not_allowed_with)
+              << RTTIArg->getAsString(Args) << ExceptionArg->getAsString(Args);
+        } else if (RTTIMode == ToolChain::RM_EnabledImplicitly)
+          D.Diag(diag::warn_drv_enabling_rtti_with_exceptions);
+      } else
+        assert(TC.getRTTIMode() != ToolChain::RM_DisabledImplicitly);
+
       CmdArgs.push_back("-fcxx-exceptions");
 
       EH = true;
@@ -2135,8 +2231,11 @@ static SmallString<128> getCompilerRTLibDir(const ToolChain &TC) {
 }
 
 static SmallString<128> getCompilerRT(const ToolChain &TC, StringRef Component,
-                                      bool Shared = false,
-                                      const char *Env = "") {
+                                      bool Shared = false) {
+  const char *Env = TC.getTriple().getEnvironment() == llvm::Triple::Android
+                        ? "-android"
+                        : "";
+
   bool IsOSWindows = TC.getTriple().isOSWindows();
   StringRef Arch = getArchNameForCompilerRTLib(TC);
   const char *Prefix = IsOSWindows ? "" : "lib";
@@ -2171,6 +2270,7 @@ static void addProfileRT(const ToolChain &TC, const ArgList &Args,
                      false) ||
         Args.hasArg(options::OPT_fprofile_generate) ||
         Args.hasArg(options::OPT_fprofile_instr_generate) ||
+        Args.hasArg(options::OPT_fprofile_instr_generate_EQ) ||
         Args.hasArg(options::OPT_fcreate_profile) ||
         Args.hasArg(options::OPT_coverage)))
     return;
@@ -2181,16 +2281,11 @@ static void addProfileRT(const ToolChain &TC, const ArgList &Args,
 static void addSanitizerRuntime(const ToolChain &TC, const ArgList &Args,
                                 ArgStringList &CmdArgs, StringRef Sanitizer,
                                 bool IsShared) {
-  const char *Env = TC.getTriple().getEnvironment() == llvm::Triple::Android
-                        ? "-android"
-                        : "";
-
   // Static runtimes must be forced into executable, so we wrap them in
   // whole-archive.
   if (!IsShared)
     CmdArgs.push_back("-whole-archive");
-  CmdArgs.push_back(Args.MakeArgString(getCompilerRT(TC, Sanitizer, IsShared,
-                                                     Env)));
+  CmdArgs.push_back(Args.MakeArgString(getCompilerRT(TC, Sanitizer, IsShared)));
   if (!IsShared)
     CmdArgs.push_back("-no-whole-archive");
 }
@@ -2251,19 +2346,20 @@ collectSanitizerRuntimes(const ToolChain &TC, const ArgList &Args,
     StaticRuntimes.push_back("dfsan");
   if (SanArgs.needsLsanRt())
     StaticRuntimes.push_back("lsan");
-  if (SanArgs.needsMsanRt())
+  if (SanArgs.needsMsanRt()) {
     StaticRuntimes.push_back("msan");
-  if (SanArgs.needsTsanRt())
+    if (SanArgs.linkCXXRuntimes())
+      StaticRuntimes.push_back("msan_cxx");
+  }
+  if (SanArgs.needsTsanRt()) {
     StaticRuntimes.push_back("tsan");
-  // WARNING: UBSan should always go last.
+    if (SanArgs.linkCXXRuntimes())
+      StaticRuntimes.push_back("tsan_cxx");
+  }
   if (SanArgs.needsUbsanRt()) {
-    // If UBSan is not combined with another sanitizer, we need to pull in
-    // sanitizer_common explicitly.
-    if (StaticRuntimes.empty())
-      HelperStaticRuntimes.push_back("san");
-    StaticRuntimes.push_back("ubsan");
+    StaticRuntimes.push_back("ubsan_standalone");
     if (SanArgs.linkCXXRuntimes())
-      StaticRuntimes.push_back("ubsan_cxx");
+      StaticRuntimes.push_back("ubsan_standalone_cxx");
   }
 }
 
@@ -2291,27 +2387,49 @@ static bool addSanitizerRuntimes(const ToolChain &TC, const ArgList &Args,
   return !StaticRuntimes.empty();
 }
 
+static bool areOptimizationsEnabled(const ArgList &Args) {
+  // Find the last -O arg and see if it is non-zero.
+  if (Arg *A = Args.getLastArg(options::OPT_O_Group))
+    return !A->getOption().matches(options::OPT_O0);
+  // Defaults to -O0.
+  return false;
+}
+
 static bool shouldUseFramePointerForTarget(const ArgList &Args,
                                            const llvm::Triple &Triple) {
-  switch (Triple.getArch()) {
-  // Don't use a frame pointer on linux if optimizing for certain targets.
-  case llvm::Triple::mips64:
-  case llvm::Triple::mips64el:
-  case llvm::Triple::mips:
-  case llvm::Triple::mipsel:
-  case llvm::Triple::systemz:
-  case llvm::Triple::x86:
-  case llvm::Triple::x86_64:
-    if (Triple.isOSLinux())
-      if (Arg *A = Args.getLastArg(options::OPT_O_Group))
-        if (!A->getOption().matches(options::OPT_O0))
-          return false;
-    return true;
-  case llvm::Triple::xcore:
+  // XCore never wants frame pointers, regardless of OS.
+  if (Triple.getArch() == llvm::Triple::xcore) {
     return false;
-  default:
-    return true;
   }
+
+  if (Triple.isOSLinux()) {
+    switch (Triple.getArch()) {
+    // Don't use a frame pointer on linux if optimizing for certain targets.
+    case llvm::Triple::mips64:
+    case llvm::Triple::mips64el:
+    case llvm::Triple::mips:
+    case llvm::Triple::mipsel:
+    case llvm::Triple::systemz:
+    case llvm::Triple::x86:
+    case llvm::Triple::x86_64:
+      return !areOptimizationsEnabled(Args);
+    default:
+      return true;
+    }
+  }
+
+  if (Triple.isOSWindows()) {
+    switch (Triple.getArch()) {
+    case llvm::Triple::x86:
+      return !areOptimizationsEnabled(Args);
+    default:
+      // All other supported Windows ISAs use xdata unwind information, so frame
+      // pointers are not generally useful.
+      return false;
+    }
+  }
+
+  return true;
 }
 
 static bool shouldUseFramePointer(const ArgList &Args,
@@ -2329,6 +2447,9 @@ static bool shouldUseLeafFramePointer(const ArgList &Args,
                                options::OPT_momit_leaf_frame_pointer))
     return A->getOption().matches(options::OPT_mno_omit_leaf_frame_pointer);
 
+  if (Triple.isPS4CPU())
+    return false;
+
   return shouldUseFramePointerForTarget(Args, Triple);
 }
 
@@ -2342,7 +2463,7 @@ static void addDebugCompDirArg(const ArgList &Args, ArgStringList &CmdArgs) {
 }
 
 static const char *SplitDebugName(const ArgList &Args,
-                                  const InputInfoList &Inputs) {
+                                  const InputInfo &Input) {
   Arg *FinalOutput = Args.getLastArg(options::OPT_o);
   if (FinalOutput && Args.hasArg(options::OPT_c)) {
     SmallString<128> T(FinalOutput->getValue());
@@ -2352,7 +2473,7 @@ static const char *SplitDebugName(const ArgList &Args,
     // Use the compilation dir.
     SmallString<128> T(
         Args.getLastArgValue(options::OPT_fdebug_compilation_dir));
-    SmallString<128> F(llvm::sys::path::stem(Inputs[0].getBaseInput()));
+    SmallString<128> F(llvm::sys::path::stem(Input.getBaseInput()));
     llvm::sys::path::replace_extension(F, "dwo");
     T += F;
     return Args.MakeArgString(F);
@@ -2431,24 +2552,17 @@ static void addDashXForInput(const ArgList &Args, const InputInfo &Input,
     CmdArgs.push_back(types::getTypeName(Input.getType()));
 }
 
-static std::string getMSCompatibilityVersion(const char *VersionStr) {
-  unsigned Version;
-  if (StringRef(VersionStr).getAsInteger(10, Version))
-    return "0";
-
+static VersionTuple getMSCompatibilityVersion(unsigned Version) {
   if (Version < 100)
-    return llvm::utostr_32(Version) + ".0";
+    return VersionTuple(Version);
 
   if (Version < 10000)
-    return llvm::utostr_32(Version / 100) + "." +
-        llvm::utostr_32(Version % 100);
+    return VersionTuple(Version / 100, Version % 100);
 
   unsigned Build = 0, Factor = 1;
   for ( ; Version > 10000; Version = Version / 10, Factor = Factor * 10)
     Build = Build + (Version % 10) * Factor;
-  return llvm::utostr_32(Version / 100) + "." +
-      llvm::utostr_32(Version % 100) + "." +
-      llvm::utostr_32(Build);
+  return VersionTuple(Version / 100, Version % 100, Build);
 }
 
 // Claim options we don't want to warn if they are unused. We do this for
@@ -2461,6 +2575,38 @@ static void claimNoWarnArgs(const ArgList &Args) {
   Args.ClaimAllArgs(options::OPT_fno_lto);
 }
 
+static void appendUserToPath(SmallVectorImpl<char> &Result) {
+#ifdef LLVM_ON_UNIX
+  const char *Username = getenv("LOGNAME");
+#else
+  const char *Username = getenv("USERNAME");
+#endif
+  if (Username) {
+    // Validate that LoginName can be used in a path, and get its length.
+    size_t Len = 0;
+    for (const char *P = Username; *P; ++P, ++Len) {
+      if (!isAlphanumeric(*P) && *P != '_') {
+        Username = nullptr;
+        break;
+      }
+    }
+
+    if (Username && Len > 0) {
+      Result.append(Username, Username + Len);
+      return;
+    }
+  }
+
+  // Fallback to user id.
+#ifdef LLVM_ON_UNIX
+  std::string UID = llvm::utostr(getuid());
+#else
+  // FIXME: Windows seems to have an 'SID' that might work.
+  std::string UID = "9999";
+#endif
+  Result.append(UID.begin(), UID.end());
+}
+
 void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                          const InputInfo &Output,
                          const InputInfoList &Inputs,
@@ -2477,6 +2623,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   bool IsWindowsMSVC = getToolChain().getTriple().isWindowsMSVCEnvironment();
 
   assert(Inputs.size() == 1 && "Unable to handle multiple inputs.");
+  const InputInfo &Input = Inputs[0];
 
   // Invoke ourselves in -cc1 mode.
   //
@@ -2568,6 +2715,13 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       assert(JA.getType() == types::TY_PP_Asm &&
              "Unexpected output type!");
     }
+
+    // Preserve use-list order by default when emitting bitcode, so that
+    // loading the bitcode up in 'opt' or 'llc' and running passes gives the
+    // same result as running passes here.  For LTO, we don't need to preserve
+    // the use-list order, since serialization to bitcode is part of the flow.
+    if (JA.getType() == types::TY_LLVM_BC)
+      CmdArgs.push_back("-emit-llvm-uselists");
   }
 
   // We normally speed up the clang process a bit by skipping destructors at
@@ -2584,7 +2738,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   // Set the main file name, so that debug info works even with
   // -save-temps.
   CmdArgs.push_back("-main-file-name");
-  CmdArgs.push_back(getBaseInputName(Args, Inputs));
+  CmdArgs.push_back(getBaseInputName(Args, Input));
 
   // Some flags which affect the language (via preprocessor
   // defines).
@@ -2612,7 +2766,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       
       CmdArgs.push_back("-analyzer-checker=deadcode");
       
-      if (types::isCXX(Inputs[0].getType()))
+      if (types::isCXX(Input.getType()))
         CmdArgs.push_back("-analyzer-checker=cplusplus");
 
       // Enable the following experimental checkers for testing.
@@ -2680,6 +2834,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     case llvm::Triple::mips64:
     case llvm::Triple::mips64el:
     case llvm::Triple::sparc:
+    case llvm::Triple::sparcel:
     case llvm::Triple::x86:
     case llvm::Triple::x86_64:
       IsPICLevelTwo = false; // "-fpie"
@@ -2781,6 +2936,8 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   else
     CmdArgs.push_back(Args.MakeArgString(getToolChain().getThreadModel()));
 
+  Args.AddLastArg(CmdArgs, options::OPT_fveclib);
+
   if (!Args.hasFlag(options::OPT_fmerge_all_constants,
                     options::OPT_fno_merge_all_constants))
     CmdArgs.push_back("-fno-merge-all-constants");
@@ -2961,6 +3118,11 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       !TrappingMath)
     CmdArgs.push_back("-menable-unsafe-fp-math");
 
+  if (!SignedZeros)
+    CmdArgs.push_back("-fno-signed-zeros");
+
+  if (ReciprocalMath)
+    CmdArgs.push_back("-freciprocal-math");
 
   // Validate and pass through -fp-contract option. 
   if (Arg *A = Args.getLastArg(options::OPT_ffast_math, FastMathAliasOption,
@@ -3060,9 +3222,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   }
 
   // Add the target cpu
-  std::string ETripleStr = getToolChain().ComputeEffectiveClangTriple(Args);
-  llvm::Triple ETriple(ETripleStr);
-  std::string CPU = getCPUName(Args, ETriple);
+  std::string CPU = getCPUName(Args, Triple);
   if (!CPU.empty()) {
     CmdArgs.push_back("-target-cpu");
     CmdArgs.push_back(Args.MakeArgString(CPU));
@@ -3074,7 +3234,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   }
 
   // Add the target features
-  getTargetFeatures(D, ETriple, Args, CmdArgs, false);
+  getTargetFeatures(D, Triple, Args, CmdArgs, false);
 
   // Add target specific flags.
   switch(getToolChain().getArch()) {
@@ -3107,6 +3267,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     break;
 
   case llvm::Triple::sparc:
+  case llvm::Triple::sparcel:
   case llvm::Triple::sparcv9:
     AddSparcTargetArgs(Args, CmdArgs);
     break;
@@ -3136,7 +3297,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
 
   // Explicitly error on some things we know we don't support and can't just
   // ignore.
-  types::ID InputType = Inputs[0].getType();
+  types::ID InputType = Input.getType();
   if (!Args.hasArg(options::OPT_fallow_unsupported)) {
     Arg *Unsupported;
     if (types::isCXX(InputType) &&
@@ -3235,25 +3396,36 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     CmdArgs.push_back("-generate-type-units");
   }
 
+  // CloudABI uses -ffunction-sections and -fdata-sections by default.
+  bool UseSeparateSections = Triple.getOS() == llvm::Triple::CloudABI;
+
   if (Args.hasFlag(options::OPT_ffunction_sections,
-                   options::OPT_fno_function_sections, false)) {
+                   options::OPT_fno_function_sections, UseSeparateSections)) {
     CmdArgs.push_back("-ffunction-sections");
   }
 
   if (Args.hasFlag(options::OPT_fdata_sections,
-                   options::OPT_fno_data_sections, false)) {
+                   options::OPT_fno_data_sections, UseSeparateSections)) {
     CmdArgs.push_back("-fdata-sections");
   }
 
+  if (!Args.hasFlag(options::OPT_funique_section_names,
+                    options::OPT_fno_unique_section_names, true))
+    CmdArgs.push_back("-fno-unique-section-names");
+
   Args.AddAllArgs(CmdArgs, options::OPT_finstrument_functions);
 
-  if (Args.hasArg(options::OPT_fprofile_instr_generate) &&
+  if ((Args.hasArg(options::OPT_fprofile_instr_generate) ||
+       Args.hasArg(options::OPT_fprofile_instr_generate_EQ)) &&
       (Args.hasArg(options::OPT_fprofile_instr_use) ||
        Args.hasArg(options::OPT_fprofile_instr_use_EQ)))
     D.Diag(diag::err_drv_argument_not_allowed_with)
       << "-fprofile-instr-generate" << "-fprofile-instr-use";
 
-  Args.AddAllArgs(CmdArgs, options::OPT_fprofile_instr_generate);
+  if (Arg *A = Args.getLastArg(options::OPT_fprofile_instr_generate_EQ))
+    A->render(Args, CmdArgs);
+  else
+    Args.AddAllArgs(CmdArgs, options::OPT_fprofile_instr_generate);
 
   if (Arg *A = Args.getLastArg(options::OPT_fprofile_instr_use_EQ))
     A->render(Args, CmdArgs);
@@ -3269,7 +3441,8 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     CmdArgs.push_back("-femit-coverage-data");
 
   if (Args.hasArg(options::OPT_fcoverage_mapping) &&
-      !Args.hasArg(options::OPT_fprofile_instr_generate))
+      !(Args.hasArg(options::OPT_fprofile_instr_generate) ||
+        Args.hasArg(options::OPT_fprofile_instr_generate_EQ)))
     D.Diag(diag::err_drv_argument_only_allowed_with)
       << "-fcoverage-mapping" << "-fprofile-instr-generate";
 
@@ -3286,10 +3459,10 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       } else {
         CoverageFilename = llvm::sys::path::filename(Output.getBaseInput());
       }
-      if (llvm::sys::path::is_relative(CoverageFilename.str())) {
+      if (llvm::sys::path::is_relative(CoverageFilename)) {
         SmallString<128> Pwd;
         if (!llvm::sys::fs::current_path(Pwd)) {
-          llvm::sys::path::append(Pwd, CoverageFilename.str());
+          llvm::sys::path::append(Pwd, CoverageFilename);
           CoverageFilename.swap(Pwd);
         }
       }
@@ -3372,6 +3545,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     Args.AddLastArg(CmdArgs, options::OPT_objcmt_migrate_all);
     Args.AddLastArg(CmdArgs, options::OPT_objcmt_migrate_readonly_property);
     Args.AddLastArg(CmdArgs, options::OPT_objcmt_migrate_readwrite_property);
+    Args.AddLastArg(CmdArgs, options::OPT_objcmt_migrate_property_dot_syntax);
     Args.AddLastArg(CmdArgs, options::OPT_objcmt_migrate_annotation);
     Args.AddLastArg(CmdArgs, options::OPT_objcmt_migrate_instancetype);
     Args.AddLastArg(CmdArgs, options::OPT_objcmt_migrate_nsmacros);
@@ -3427,6 +3601,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   //
   // If a std is supplied, only add -trigraphs if it follows the
   // option.
+  bool ImplyVCPPCXXVer = false;
   if (Arg *Std = Args.getLastArg(options::OPT_std_EQ, options::OPT_ansi)) {
     if (Std->getOption().matches(options::OPT_ansi))
       if (types::isCXX(InputType))
@@ -3453,7 +3628,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       Args.AddAllArgsTranslated(CmdArgs, options::OPT_std_default_EQ,
                                 "-std=", /*Joined=*/true);
     else if (IsWindowsMSVC)
-      CmdArgs.push_back("-std=c++11");
+      ImplyVCPPCXXVer = true;
 
     Args.AddLastArg(CmdArgs, options::OPT_ftrigraphs,
                     options::OPT_fno_trigraphs);
@@ -3628,6 +3803,12 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   Args.AddLastArg(CmdArgs, options::OPT_fdiagnostics_show_template_tree);
   Args.AddLastArg(CmdArgs, options::OPT_fno_elide_type);
 
+  // Forward flags for OpenMP
+  if (Args.hasArg(options::OPT_fopenmp_EQ) ||
+      Args.hasArg(options::OPT_fopenmp)) {
+    CmdArgs.push_back("-fopenmp");
+  }
+
   const SanitizerArgs &Sanitize = getToolChain().getSanitizerArgs();
   Sanitize.addArgs(Args, CmdArgs);
 
@@ -3739,6 +3920,15 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     CmdArgs.push_back(Args.MakeArgString("-mstack-alignment=" + alignment));
   }
 
+  if (Args.hasArg(options::OPT_mstack_probe_size)) {
+    StringRef Size = Args.getLastArgValue(options::OPT_mstack_probe_size);
+
+    if (!Size.empty())
+      CmdArgs.push_back(Args.MakeArgString("-mstack-probe-size=" + Size));
+    else
+      CmdArgs.push_back("-mstack-probe-size=0");
+  }
+
   if (getToolChain().getTriple().getArch() == llvm::Triple::aarch64 ||
       getToolChain().getTriple().getArch() == llvm::Triple::aarch64_be)
     CmdArgs.push_back("-fallow-half-arguments-and-returns");
@@ -3845,6 +4035,12 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     CmdArgs.push_back("-fmodules-strict-decluse");
   }
 
+  // -fno-implicit-modules turns off implicitly compiling modules on demand.
+  if (!Args.hasFlag(options::OPT_fimplicit_modules,
+                    options::OPT_fno_implicit_modules)) {
+    CmdArgs.push_back("-fno-implicit-modules");
+  }
+
   // -fmodule-name specifies the module that is currently being built (or
   // used for header checking by -fmodule-maps).
   Args.AddLastArg(CmdArgs, options::OPT_fmodule_name);
@@ -3872,7 +4068,8 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       // No module path was provided: use the default.
       llvm::sys::path::system_temp_directory(/*erasedOnReboot=*/false,
                                              ModuleCachePath);
-      llvm::sys::path::append(ModuleCachePath, "org.llvm.clang");
+      llvm::sys::path::append(ModuleCachePath, "org.llvm.clang.");
+      appendUserToPath(ModuleCachePath);
       llvm::sys::path::append(ModuleCachePath, "ModuleCache");
     }
     const char Arg[] = "-fmodules-cache-path=";
@@ -3911,10 +4108,9 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     llvm::sys::fs::file_status Status;
     if (llvm::sys::fs::status(A->getValue(), Status))
       D.Diag(diag::err_drv_no_such_file) << A->getValue();
-    char TimeStamp[48];
-    snprintf(TimeStamp, sizeof(TimeStamp), "-fbuild-session-timestamp=%" PRIu64,
-             (uint64_t)Status.getLastModificationTime().toEpochTime());
-    CmdArgs.push_back(Args.MakeArgString(TimeStamp));
+    CmdArgs.push_back(Args.MakeArgString(
+        "-fbuild-session-timestamp=" +
+        Twine((uint64_t)Status.getLastModificationTime().toEpochTime())));
   }
 
   if (Args.getLastArg(options::OPT_fmodules_validate_once_per_build_session)) {
@@ -3940,21 +4136,12 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                    false))
     CmdArgs.push_back("-fno-elide-constructors");
 
-  // -frtti is default.
-  if (!Args.hasFlag(options::OPT_frtti, options::OPT_fno_rtti) ||
-      KernelOrKext) {
-    CmdArgs.push_back("-fno-rtti");
+  ToolChain::RTTIMode RTTIMode = getToolChain().getRTTIMode();
 
-    // -fno-rtti cannot usefully be combined with -fsanitize=vptr.
-    if (Sanitize.sanitizesVptr()) {
-      std::string NoRttiArg =
-        Args.getLastArg(options::OPT_mkernel,
-                        options::OPT_fapple_kext,
-                        options::OPT_fno_rtti)->getAsString(Args);
-      D.Diag(diag::err_drv_argument_not_allowed_with)
-        << "-fsanitize=vptr" << NoRttiArg;
-    }
-  }
+  if (KernelOrKext || (types::isCXX(InputType) &&
+                       (RTTIMode == ToolChain::RM_DisabledExplicitly ||
+                        RTTIMode == ToolChain::RM_DisabledImplicitly)))
+    CmdArgs.push_back("-fno-rtti");
 
   // -fshort-enums=0 is default for all architectures except Hexagon.
   if (Args.hasFlag(options::OPT_fshort_enums,
@@ -3964,14 +4151,16 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     CmdArgs.push_back("-fshort-enums");
 
   // -fsigned-char is default.
-  if (!Args.hasFlag(options::OPT_fsigned_char, options::OPT_funsigned_char,
-                    isSignedCharDefault(getToolChain().getTriple())))
+  if (Arg *A = Args.getLastArg(
+          options::OPT_fsigned_char, options::OPT_fno_signed_char,
+          options::OPT_funsigned_char, options::OPT_fno_unsigned_char)) {
+    if (A->getOption().matches(options::OPT_funsigned_char) ||
+        A->getOption().matches(options::OPT_fno_signed_char)) {
+      CmdArgs.push_back("-fno-signed-char");
+    }
+  } else if (!isSignedCharDefault(getToolChain().getTriple())) {
     CmdArgs.push_back("-fno-signed-char");
-
-  // -fthreadsafe-static is default.
-  if (!Args.hasFlag(options::OPT_fthreadsafe_statics,
-                    options::OPT_fno_threadsafe_statics))
-    CmdArgs.push_back("-fno-threadsafe-statics");
+  }
 
   // -fuse-cxa-atexit is default.
   if (!Args.hasFlag(options::OPT_fuse_cxa_atexit,
@@ -3987,6 +4176,11 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                    IsWindowsMSVC))
     CmdArgs.push_back("-fms-extensions");
 
+  // -fno-use-line-directives is default.
+  if (Args.hasFlag(options::OPT_fuse_line_directives,
+                   options::OPT_fno_use_line_directives, false))
+    CmdArgs.push_back("-fuse-line-directives");
+
   // -fms-compatibility=0 is default.
   if (Args.hasFlag(options::OPT_fms_compatibility, 
                    options::OPT_fno_ms_compatibility,
@@ -3995,9 +4189,11 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                                                   true))))
     CmdArgs.push_back("-fms-compatibility");
 
-  // -fms-compatibility-version=17.00 is default.
+  // -fms-compatibility-version=18.00 is default.
+  VersionTuple MSVT;
   if (Args.hasFlag(options::OPT_fms_extensions, options::OPT_fno_ms_extensions,
-                   IsWindowsMSVC) || Args.hasArg(options::OPT_fmsc_version) ||
+                   IsWindowsMSVC) ||
+      Args.hasArg(options::OPT_fmsc_version) ||
       Args.hasArg(options::OPT_fms_compatibility_version)) {
     const Arg *MSCVersion = Args.getLastArg(options::OPT_fmsc_version);
     const Arg *MSCompatibilityVersion =
@@ -4008,16 +4204,31 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
           << MSCVersion->getAsString(Args)
           << MSCompatibilityVersion->getAsString(Args);
 
-    std::string Ver;
-    if (MSCompatibilityVersion)
-      Ver = Args.getLastArgValue(options::OPT_fms_compatibility_version);
-    else if (MSCVersion)
-      Ver = getMSCompatibilityVersion(MSCVersion->getValue());
+    if (MSCompatibilityVersion) {
+      if (MSVT.tryParse(MSCompatibilityVersion->getValue()))
+        D.Diag(diag::err_drv_invalid_value)
+            << MSCompatibilityVersion->getAsString(Args)
+            << MSCompatibilityVersion->getValue();
+    } else if (MSCVersion) {
+      unsigned Version = 0;
+      if (StringRef(MSCVersion->getValue()).getAsInteger(10, Version))
+        D.Diag(diag::err_drv_invalid_value) << MSCVersion->getAsString(Args)
+                                            << MSCVersion->getValue();
+      MSVT = getMSCompatibilityVersion(Version);
+    } else {
+      MSVT = VersionTuple(18);
+    }
+
+    CmdArgs.push_back(
+        Args.MakeArgString("-fms-compatibility-version=" + MSVT.getAsString()));
+  }
 
-    if (Ver.empty())
-      CmdArgs.push_back("-fms-compatibility-version=17.00");
+  bool IsMSVC2015Compatible = MSVT.getMajor() >= 19;
+  if (ImplyVCPPCXXVer) {
+    if (IsMSVC2015Compatible)
+      CmdArgs.push_back("-std=c++14");
     else
-      CmdArgs.push_back(Args.MakeArgString("-fms-compatibility-version=" + Ver));
+      CmdArgs.push_back("-std=c++11");
   }
 
   // -fno-borland-extensions is default.
@@ -4025,6 +4236,13 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                    options::OPT_fno_borland_extensions, false))
     CmdArgs.push_back("-fborland-extensions");
 
+  // -fthreadsafe-static is default, except for MSVC compatibility versions less
+  // than 19.
+  if (!Args.hasFlag(options::OPT_fthreadsafe_statics,
+                    options::OPT_fno_threadsafe_statics,
+                    !IsWindowsMSVC || IsMSVC2015Compatible))
+    CmdArgs.push_back("-fno-threadsafe-statics");
+
   // -fno-delayed-template-parsing is default, except for Windows where MSVC STL
   // needs it.
   if (Args.hasFlag(options::OPT_fdelayed_template_parsing,
@@ -4130,9 +4348,13 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
     }
   }
 
+  if (Args.hasFlag(options::OPT_fapplication_extension,
+                   options::OPT_fno_application_extension, false))
+    CmdArgs.push_back("-fapplication-extension");
+
   // Handle GCC-style exception args.
   if (!C.getDriver().IsCLMode())
-    addExceptionArgs(Args, InputType, getToolChain().getTriple(), KernelOrKext,
+    addExceptionArgs(Args, InputType, getToolChain(), KernelOrKext,
                      objcRuntime, CmdArgs);
 
   if (getToolChain().UseSjLjExceptions())
@@ -4143,6 +4365,12 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                     options::OPT_fno_assume_sane_operator_new))
     CmdArgs.push_back("-fno-assume-sane-operator-new");
 
+  // -fsized-deallocation is off by default, as it is an ABI-breaking change for
+  // most platforms.
+  if (Args.hasFlag(options::OPT_fsized_deallocation,
+                   options::OPT_fno_sized_deallocation, false))
+    CmdArgs.push_back("-fsized-deallocation");
+
   // -fconstant-cfstrings is default, and may be subject to argument translation
   // on Darwin.
   if (!Args.hasFlag(options::OPT_fconstant_cfstrings,
@@ -4326,6 +4554,11 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
                    false))
     CmdArgs.push_back("-fasm-blocks");
 
+  // -fgnu-inline-asm is default.
+  if (!Args.hasFlag(options::OPT_fgnu_inline_asm,
+                    options::OPT_fno_gnu_inline_asm, true))
+    CmdArgs.push_back("-fno-gnu-inline-asm");
+
   // Enable vectorization per default according to the optimization level
   // selected. For optimization levels that want vectorization we use the alias
   // option to simplify the hasFlag logic.
@@ -4450,7 +4683,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   // With -save-temps, we want to save the unoptimized bitcode output from the
   // CompileJobAction, so disable optimizations if they are not already
   // disabled.
-  if (Args.hasArg(options::OPT_save_temps) && !OptDisabled &&
+  if (C.getDriver().isSaveTempsEnabled() && !OptDisabled &&
       isa<CompileJobAction>(JA))
     CmdArgs.push_back("-disable-llvm-optzns");
 
@@ -4492,7 +4725,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
       Flags += EscapedArg;
     }
     CmdArgs.push_back("-dwarf-debug-flags");
-    CmdArgs.push_back(Args.MakeArgString(Flags.str()));
+    CmdArgs.push_back(Args.MakeArgString(Flags));
   }
 
   // Add the split debug info name to the command lines here so we
@@ -4504,7 +4737,7 @@ void Clang::ConstructJob(Compilation &C, const JobAction &JA,
   const char *SplitDwarfOut;
   if (SplitDwarf) {
     CmdArgs.push_back("-split-dwarf-file");
-    SplitDwarfOut = SplitDebugName(Args, Inputs);
+    SplitDwarfOut = SplitDebugName(Args, Input);
     CmdArgs.push_back(SplitDwarfOut);
   }
 
@@ -4772,8 +5005,8 @@ void Clang::AddClangCLArgs(const ArgList &Args, ArgStringList &CmdArgs) const {
   EHFlags EH = parseClangCLEHFlags(D, Args);
   // FIXME: Do something with NoExceptC.
   if (EH.Synch || EH.Asynch) {
-    CmdArgs.push_back("-fexceptions");
     CmdArgs.push_back("-fcxx-exceptions");
+    CmdArgs.push_back("-fexceptions");
   }
 
   // /EP should expand to -E -P.
@@ -4782,6 +5015,19 @@ void Clang::AddClangCLArgs(const ArgList &Args, ArgStringList &CmdArgs) const {
     CmdArgs.push_back("-P");
   }
 
+  unsigned VolatileOptionID;
+  if (getToolChain().getTriple().getArch() == llvm::Triple::x86_64 ||
+      getToolChain().getTriple().getArch() == llvm::Triple::x86)
+    VolatileOptionID = options::OPT__SLASH_volatile_ms;
+  else
+    VolatileOptionID = options::OPT__SLASH_volatile_iso;
+
+  if (Arg *A = Args.getLastArg(options::OPT__SLASH_volatile_Group))
+    VolatileOptionID = A->getOption().getID();
+
+  if (VolatileOptionID == options::OPT__SLASH_volatile_ms)
+    CmdArgs.push_back("-fms-volatile");
+
   Arg *MostGeneralArg = Args.getLastArg(options::OPT__SLASH_vmg);
   Arg *BestCaseArg = Args.getLastArg(options::OPT__SLASH_vmb);
   if (MostGeneralArg && BestCaseArg)
@@ -4826,6 +5072,17 @@ visualstudio::Compile *Clang::getCLFallback() const {
   return CLFallback.get();
 }
 
+void ClangAs::AddMIPSTargetArgs(const ArgList &Args,
+                                ArgStringList &CmdArgs) const {
+  StringRef CPUName;
+  StringRef ABIName;
+  const llvm::Triple &Triple = getToolChain().getTriple();
+  mips::getMipsCPUAndABI(Args, Triple, CPUName, ABIName);
+
+  CmdArgs.push_back("-target-abi");
+  CmdArgs.push_back(ABIName.data());
+}
+
 void ClangAs::ConstructJob(Compilation &C, const JobAction &JA,
                            const InputInfo &Output,
                            const InputInfoList &Inputs,
@@ -4862,10 +5119,10 @@ void ClangAs::ConstructJob(Compilation &C, const JobAction &JA,
   // Set the main file name, so that debug info works even with
   // -save-temps or preprocessed assembly.
   CmdArgs.push_back("-main-file-name");
-  CmdArgs.push_back(Clang::getBaseInputName(Args, Inputs));
+  CmdArgs.push_back(Clang::getBaseInputName(Args, Input));
 
   // Add the target cpu
-  const llvm::Triple &Triple = getToolChain().getTriple();
+  const llvm::Triple Triple(TripleStr);
   std::string CPU = getCPUName(Args, Triple);
   if (!CPU.empty()) {
     CmdArgs.push_back("-target-cpu");
@@ -4928,11 +5185,24 @@ void ClangAs::ConstructJob(Compilation &C, const JobAction &JA,
       Flags += EscapedArg;
     }
     CmdArgs.push_back("-dwarf-debug-flags");
-    CmdArgs.push_back(Args.MakeArgString(Flags.str()));
+    CmdArgs.push_back(Args.MakeArgString(Flags));
   }
 
   // FIXME: Add -static support, once we have it.
 
+  // Add target specific flags.
+  switch(getToolChain().getArch()) {
+  default:
+    break;
+
+  case llvm::Triple::mips:
+  case llvm::Triple::mipsel:
+  case llvm::Triple::mips64:
+  case llvm::Triple::mips64el:
+    AddMIPSTargetArgs(Args, CmdArgs);
+    break;
+  }
+
   // Consume all the warning flags. Usually this would be handled more
   // gracefully by -cc1 (warning about unknown warning flags, etc) but -cc1as
   // doesn't handle that so rather than warning about unused flags that are
@@ -4964,7 +5234,7 @@ void ClangAs::ConstructJob(Compilation &C, const JobAction &JA,
   if (Args.hasArg(options::OPT_gsplit_dwarf) &&
       getToolChain().getTriple().isOSLinux())
     SplitDebugInfo(getToolChain(), C, *this, JA, Args, Output,
-                   SplitDebugName(Args, Inputs));
+                   SplitDebugName(Args, Input));
 }
 
 void GnuTool::anchor() {}
@@ -5095,16 +5365,22 @@ void gcc::Compile::RenderExtraToolArgs(const JobAction &JA,
                                        ArgStringList &CmdArgs) const {
   const Driver &D = getToolChain().getDriver();
 
+  switch (JA.getType()) {
   // If -flto, etc. are present then make sure not to force assembly output.
-  if (JA.getType() == types::TY_LLVM_IR || JA.getType() == types::TY_LTO_IR ||
-      JA.getType() == types::TY_LLVM_BC || JA.getType() == types::TY_LTO_BC)
+  case types::TY_LLVM_IR:
+  case types::TY_LTO_IR:
+  case types::TY_LLVM_BC:
+  case types::TY_LTO_BC:
     CmdArgs.push_back("-c");
-  else {
-    if (JA.getType() != types::TY_PP_Asm)
-      D.Diag(diag::err_drv_invalid_gcc_output_type)
-        << getTypeName(JA.getType());
-
+    break;
+  case types::TY_PP_Asm:
     CmdArgs.push_back("-S");
+    break;
+  case types::TY_Nothing:
+    CmdArgs.push_back("-fsyntax-only");
+    break;
+  default:
+    D.Diag(diag::err_drv_invalid_gcc_output_type) << getTypeName(JA.getType());
   }
 }
 
@@ -5142,10 +5418,8 @@ void hexagon::Assemble::ConstructJob(Compilation &C, const JobAction &JA,
     CmdArgs.push_back("-fsyntax-only");
   }
 
-  std::string SmallDataThreshold = GetHexagonSmallDataThresholdValue(Args);
-  if (!SmallDataThreshold.empty())
-    CmdArgs.push_back(
-      Args.MakeArgString(std::string("-G") + SmallDataThreshold));
+  if (const char* v = toolchains::Hexagon_TC::GetSmallDataThreshold(Args))
+    CmdArgs.push_back(Args.MakeArgString(std::string("-G") + v));
 
   Args.AddAllArgValues(CmdArgs, options::OPT_Wa_COMMA,
                        options::OPT_Xassembler);
@@ -5188,17 +5462,16 @@ void hexagon::Link::RenderExtraToolArgs(const JobAction &JA,
   // The types are (hopefully) good enough.
 }
 
-void hexagon::Link::ConstructJob(Compilation &C, const JobAction &JA,
-                               const InputInfo &Output,
-                               const InputInfoList &Inputs,
-                               const ArgList &Args,
-                               const char *LinkingOutput) const {
+static void constructHexagonLinkArgs(Compilation &C, const JobAction &JA,
+                              const toolchains::Hexagon_TC& ToolChain,
+                              const InputInfo &Output,
+                              const InputInfoList &Inputs,
+                              const ArgList &Args,
+                              ArgStringList &CmdArgs,
+                              const char *LinkingOutput) {
 
-  const toolchains::Hexagon_TC& ToolChain =
-    static_cast<const toolchains::Hexagon_TC&>(getToolChain());
   const Driver &D = ToolChain.getDriver();
 
-  ArgStringList CmdArgs;
 
   //----------------------------------------------------------------------------
   //
@@ -5209,6 +5482,7 @@ void hexagon::Link::ConstructJob(Compilation &C, const JobAction &JA,
   bool incStdLib = !Args.hasArg(options::OPT_nostdlib);
   bool incStartFiles = !Args.hasArg(options::OPT_nostartfiles);
   bool incDefLibs = !Args.hasArg(options::OPT_nodefaultlibs);
+  bool useG0 = false;
   bool useShared = buildingLib && !hasStaticArg;
 
   //----------------------------------------------------------------------------
@@ -5242,10 +5516,9 @@ void hexagon::Link::ConstructJob(Compilation &C, const JobAction &JA,
   if (buildPIE && !buildingLib)
     CmdArgs.push_back("-pie");
 
-  std::string SmallDataThreshold = GetHexagonSmallDataThresholdValue(Args);
-  if (!SmallDataThreshold.empty()) {
-    CmdArgs.push_back(
-      Args.MakeArgString(std::string("-G") + SmallDataThreshold));
+  if (const char* v = toolchains::Hexagon_TC::GetSmallDataThreshold(Args)) {
+    CmdArgs.push_back(Args.MakeArgString(std::string("-G") + v));
+    useG0 = toolchains::Hexagon_TC::UsesG0(v);
   }
 
   //----------------------------------------------------------------------------
@@ -5261,8 +5534,7 @@ void hexagon::Link::ConstructJob(Compilation &C, const JobAction &JA,
       toolchains::Hexagon_TC::GetGnuDir(D.InstalledDir, Args) + "/";
   const std::string StartFilesDir = RootDir
                                     + "hexagon/lib"
-                                    + (buildingLib
-                                       ? MarchG0Suffix : MarchSuffix);
+                                    + (useG0 ? MarchG0Suffix : MarchSuffix);
 
   //----------------------------------------------------------------------------
   // moslib
@@ -5344,6 +5616,20 @@ void hexagon::Link::ConstructJob(Compilation &C, const JobAction &JA,
     std::string finiObj = useShared ? "/finiS.o" : "/fini.o";
     CmdArgs.push_back(Args.MakeArgString(StartFilesDir + finiObj));
   }
+}
+
+void hexagon::Link::ConstructJob(Compilation &C, const JobAction &JA,
+                               const InputInfo &Output,
+                               const InputInfoList &Inputs,
+                               const ArgList &Args,
+                               const char *LinkingOutput) const {
+
+  const toolchains::Hexagon_TC& ToolChain =
+    static_cast<const toolchains::Hexagon_TC&>(getToolChain());
+
+  ArgStringList CmdArgs;
+  constructHexagonLinkArgs(C, JA, ToolChain, Output, Inputs, Args, CmdArgs,
+                           LinkingOutput);
 
   std::string Linker = ToolChain.GetProgramPath("hexagon-ld");
   C.addCommand(llvm::make_unique<Command>(JA, *this, Args.MakeArgString(Linker),
@@ -5351,9 +5637,8 @@ void hexagon::Link::ConstructJob(Compilation &C, const JobAction &JA,
 }
 // Hexagon tools end.
 
-/// Get the (LLVM) name of the minimum ARM CPU for the arch we are targeting.
-const char *arm::getARMCPUForMArch(const ArgList &Args,
-                                   const llvm::Triple &Triple) {
+const StringRef arm::getARMArch(const ArgList &Args,
+                                const llvm::Triple &Triple) {
   StringRef MArch;
   if (Arg *A = Args.getLastArg(options::OPT_march_EQ)) {
     // Otherwise, if we have -march= choose the base CPU for that arch.
@@ -5367,13 +5652,35 @@ const char *arm::getARMCPUForMArch(const ArgList &Args,
   if (MArch == "native") {
     std::string CPU = llvm::sys::getHostCPUName();
     if (CPU != "generic") {
-      // Translate the native cpu into the architecture. The switch below will
-      // then chose the minimum cpu for that arch.
-      MArch = std::string("arm") + arm::getLLVMArchSuffixForARM(CPU);
+      // Translate the native cpu into the architecture suffix for that CPU.
+      const char *Suffix = arm::getLLVMArchSuffixForARM(CPU, MArch);
+      // If there is no valid architecture suffix for this CPU we don't know how
+      // to handle it, so return no architecture.
+      if (strcmp(Suffix,"") == 0)
+        MArch = "";
+      else
+        MArch = std::string("arm") + Suffix;
     }
   }
 
-  return Triple.getARMCPUForArch(MArch);
+  return MArch;
+}
+/// Get the (LLVM) name of the minimum ARM CPU for the arch we are targeting.
+const char *arm::getARMCPUForMArch(const ArgList &Args,
+                                   const llvm::Triple &Triple) {
+  StringRef MArch = getARMArch(Args, Triple);
+  // getARMCPUForArch defaults to the triple if MArch is empty, but empty MArch
+  // here means an -march=native that we can't handle, so instead return no CPU.
+  if (MArch.empty())
+    return "";
+
+  // We need to return an empty string here on invalid MArch values as the
+  // various places that call this function can't cope with a null result.
+  const char *result = Triple.getARMCPUForArch(MArch);
+  if (result)
+    return result;
+  else
+    return "";
 }
 
 /// getARMTargetCPU - Get the (LLVM) name of the ARM cpu we are targeting.
@@ -5394,13 +5701,24 @@ StringRef arm::getARMTargetCPU(const ArgList &Args,
 }
 
 /// getLLVMArchSuffixForARM - Get the LLVM arch name to use for a particular
-/// CPU.
+/// CPU  (or Arch, if CPU is generic).
 //
 // FIXME: This is redundant with -mcpu, why does LLVM use this.
 // FIXME: tblgen this, or kill it!
-const char *arm::getLLVMArchSuffixForARM(StringRef CPU) {
+// FIXME: Use ARMTargetParser.
+const char *arm::getLLVMArchSuffixForARM(StringRef CPU, StringRef Arch) {
+  // FIXME: Use ARMTargetParser
+  if (CPU == "generic") {
+    if (Arch == "armv8.1a" || Arch == "armv8.1-a" ||
+        Arch == "armebv8.1a" || Arch == "armebv8.1-a") {
+      return "v8.1a";
+    }
+  }
+
+  // FIXME: Use ARMTargetParser
   return llvm::StringSwitch<const char *>(CPU)
-    .Case("strongarm", "v4")
+    .Cases("arm8", "arm810", "v4")
+    .Cases("strongarm", "strongarm110", "strongarm1100", "strongarm1110", "v4")
     .Cases("arm7tdmi", "arm7tdmi-s", "arm710t", "v4t")
     .Cases("arm720t", "arm9", "arm9tdmi", "v4t")
     .Cases("arm920", "arm920t", "arm922t", "v4t")
@@ -5409,33 +5727,51 @@ const char *arm::getLLVMArchSuffixForARM(StringRef CPU) {
     .Cases("arm9e",  "arm926ej-s",  "arm946e-s", "v5e")
     .Cases("arm966e-s",  "arm968e-s",  "arm10e", "v5e")
     .Cases("arm1020e",  "arm1022e",  "xscale", "iwmmxt", "v5e")
-    .Cases("arm1136j-s",  "arm1136jf-s",  "arm1176jz-s", "v6")
-    .Cases("arm1176jzf-s",  "mpcorenovfp",  "mpcore", "v6")
+    .Cases("arm1136j-s",  "arm1136jf-s", "v6")
+    .Cases("arm1176jz-s", "arm1176jzf-s", "v6k")
+    .Cases("mpcorenovfp",  "mpcore", "v6k")
     .Cases("arm1156t2-s",  "arm1156t2f-s", "v6t2")
     .Cases("cortex-a5", "cortex-a7", "cortex-a8", "v7")
     .Cases("cortex-a9", "cortex-a12", "cortex-a15", "cortex-a17", "krait", "v7")
-    .Cases("cortex-r4", "cortex-r5", "v7r")
-    .Case("cortex-m0", "v6m")
-    .Case("cortex-m3", "v7m")
+    .Cases("cortex-r4", "cortex-r4f", "cortex-r5", "cortex-r7", "v7r")
+    .Cases("sc000", "cortex-m0", "cortex-m0plus", "cortex-m1", "v6m")
+    .Cases("sc300", "cortex-m3", "v7m")
     .Cases("cortex-m4", "cortex-m7", "v7em")
     .Case("swift", "v7s")
     .Case("cyclone", "v8")
-    .Cases("cortex-a53", "cortex-a57", "v8")
+    .Cases("cortex-a53", "cortex-a57", "cortex-a72", "v8")
     .Default("");
 }
 
-void arm::appendEBLinkFlags(const ArgList &Args, ArgStringList &CmdArgs, const llvm::Triple &Triple) {
+void arm::appendEBLinkFlags(const ArgList &Args, ArgStringList &CmdArgs, 
+                            const llvm::Triple &Triple) {
   if (Args.hasArg(options::OPT_r))
     return;
 
-  StringRef Suffix = getLLVMArchSuffixForARM(getARMCPUForMArch(Args, Triple));
-  const char *LinkFlag = llvm::StringSwitch<const char *>(Suffix)
-    .Cases("v4", "v4t", "v5", "v5e", nullptr)
-    .Cases("v6", "v6t2", nullptr)
-    .Default("--be8");
-
-  if (LinkFlag)
-    CmdArgs.push_back(LinkFlag);
+  // ARMv7 (and later) and ARMv6-M do not support BE-32, so instruct the linker
+  // to generate BE-8 executables.
+  if (getARMSubArchVersionNumber(Triple) >= 7 || isARMMProfile(Triple))
+    CmdArgs.push_back("--be8");
+}
+
+mips::NanEncoding mips::getSupportedNanEncoding(StringRef &CPU) {
+  return (NanEncoding)llvm::StringSwitch<int>(CPU)
+      .Case("mips1", NanLegacy)
+      .Case("mips2", NanLegacy)
+      .Case("mips3", NanLegacy)
+      .Case("mips4", NanLegacy)
+      .Case("mips5", NanLegacy)
+      .Case("mips32", NanLegacy)
+      .Case("mips32r2", NanLegacy)
+      .Case("mips32r3", NanLegacy | Nan2008)
+      .Case("mips32r5", NanLegacy | Nan2008)
+      .Case("mips32r6", Nan2008)
+      .Case("mips64", NanLegacy)
+      .Case("mips64r2", NanLegacy)
+      .Case("mips64r3", NanLegacy | Nan2008)
+      .Case("mips64r5", NanLegacy | Nan2008)
+      .Case("mips64r6", Nan2008)
+      .Default(NanLegacy);
 }
 
 bool mips::hasMipsAbiArg(const ArgList &Args, const char *Value) {
@@ -5474,8 +5810,8 @@ bool mips::isFPXXDefault(const llvm::Triple &Triple, StringRef CPUName,
 
   return llvm::StringSwitch<bool>(CPUName)
              .Cases("mips2", "mips3", "mips4", "mips5", true)
-             .Cases("mips32", "mips32r2", true)
-             .Cases("mips64", "mips64r2", true)
+             .Cases("mips32", "mips32r2", "mips32r3", "mips32r5", true)
+             .Cases("mips64", "mips64r2", "mips64r3", "mips64r5", true)
              .Default(false);
 }
 
@@ -5527,14 +5863,13 @@ void darwin::setTripleTypeForMachOArchName(llvm::Triple &T, StringRef Str) {
 }
 
 const char *Clang::getBaseInputName(const ArgList &Args,
-                                    const InputInfoList &Inputs) {
-  return Args.MakeArgString(
-    llvm::sys::path::filename(Inputs[0].getBaseInput()));
+                                    const InputInfo &Input) {
+  return Args.MakeArgString(llvm::sys::path::filename(Input.getBaseInput()));
 }
 
 const char *Clang::getBaseInputStem(const ArgList &Args,
                                     const InputInfoList &Inputs) {
-  const char *Str = getBaseInputName(Args, Inputs);
+  const char *Str = getBaseInputName(Args, Inputs[0]);
 
   if (const char *End = strrchr(Str, '.'))
     return Args.MakeArgString(std::string(Str, End));
@@ -5556,6 +5891,76 @@ const char *Clang::getDependencyFileName(const ArgList &Args,
   return Args.MakeArgString(Res + ".d");
 }
 
+void cloudabi::Link::ConstructJob(Compilation &C, const JobAction &JA,
+                                  const InputInfo &Output,
+                                  const InputInfoList &Inputs,
+                                  const ArgList &Args,
+                                  const char *LinkingOutput) const {
+  const ToolChain &ToolChain = getToolChain();
+  const Driver &D = ToolChain.getDriver();
+  ArgStringList CmdArgs;
+
+  // Silence warning for "clang -g foo.o -o foo"
+  Args.ClaimAllArgs(options::OPT_g_Group);
+  // and "clang -emit-llvm foo.o -o foo"
+  Args.ClaimAllArgs(options::OPT_emit_llvm);
+  // and for "clang -w foo.o -o foo". Other warning options are already
+  // handled somewhere else.
+  Args.ClaimAllArgs(options::OPT_w);
+
+  if (!D.SysRoot.empty())
+    CmdArgs.push_back(Args.MakeArgString("--sysroot=" + D.SysRoot));
+
+  // CloudABI only supports static linkage.
+  CmdArgs.push_back("-Bstatic");
+  CmdArgs.push_back("--eh-frame-hdr");
+  CmdArgs.push_back("--gc-sections");
+
+  if (Output.isFilename()) {
+    CmdArgs.push_back("-o");
+    CmdArgs.push_back(Output.getFilename());
+  } else {
+    assert(Output.isNothing() && "Invalid output.");
+  }
+
+  if (!Args.hasArg(options::OPT_nostdlib) &&
+      !Args.hasArg(options::OPT_nostartfiles)) {
+    CmdArgs.push_back(Args.MakeArgString(ToolChain.GetFilePath("crt0.o")));
+    CmdArgs.push_back(Args.MakeArgString(ToolChain.GetFilePath("crtbegin.o")));
+  }
+
+  Args.AddAllArgs(CmdArgs, options::OPT_L);
+  const ToolChain::path_list &Paths = ToolChain.getFilePaths();
+  for (const auto &Path : Paths)
+    CmdArgs.push_back(Args.MakeArgString(StringRef("-L") + Path));
+  Args.AddAllArgs(CmdArgs, options::OPT_T_Group);
+  Args.AddAllArgs(CmdArgs, options::OPT_e);
+  Args.AddAllArgs(CmdArgs, options::OPT_s);
+  Args.AddAllArgs(CmdArgs, options::OPT_t);
+  Args.AddAllArgs(CmdArgs, options::OPT_Z_Flag);
+  Args.AddAllArgs(CmdArgs, options::OPT_r);
+
+  if (D.IsUsingLTO(ToolChain, Args))
+    AddGoldPlugin(ToolChain, Args, CmdArgs);
+
+  AddLinkerInputs(ToolChain, Inputs, Args, CmdArgs);
+
+  if (!Args.hasArg(options::OPT_nostdlib) &&
+      !Args.hasArg(options::OPT_nodefaultlibs)) {
+    if (D.CCCIsCXX())
+      ToolChain.AddCXXStdlibLibArgs(Args, CmdArgs);
+    CmdArgs.push_back("-lc");
+    CmdArgs.push_back("-lcompiler_rt");
+  }
+
+  if (!Args.hasArg(options::OPT_nostdlib) &&
+      !Args.hasArg(options::OPT_nostartfiles))
+    CmdArgs.push_back(Args.MakeArgString(ToolChain.GetFilePath("crtend.o")));
+
+  const char *Exec = Args.MakeArgString(ToolChain.GetLinkerPath());
+  C.addCommand(llvm::make_unique<Command>(JA, *this, Exec, CmdArgs));
+}
+
 void darwin::Assemble::ConstructJob(Compilation &C, const JobAction &JA,
                                     const InputInfo &Output,
                                     const InputInfoList &Inputs,
@@ -5677,10 +6082,17 @@ void darwin::Link::AddLinkArgs(Compilation &C,
   if (Args.hasArg(options::OPT_rdynamic) && Version[0] >= 137)
     CmdArgs.push_back("-export_dynamic");
 
+  // If we are using App Extension restrictions, pass a flag to the linker
+  // telling it that the compiled code has been audited.
+  if (Args.hasFlag(options::OPT_fapplication_extension,
+                   options::OPT_fno_application_extension, false))
+    CmdArgs.push_back("-application_extension");
+
   // If we are using LTO, then automatically create a temporary file path for
   // the linker to use, so that it's lifetime will extend past a possible
   // dsymutil step.
-  if (Version[0] >= 116 && D.IsUsingLTO(Args) && NeedsTempPath(Inputs)) {
+  if (Version[0] >= 116 && D.IsUsingLTO(getToolChain(), Args) &&
+      NeedsTempPath(Inputs)) {
     const char *TmpPath = C.getArgs().MakeArgString(
       D.GetTemporaryPath("cc", types::getTypeTempSuffix(types::TY_Object)));
     C.addTempFile(TmpPath);
@@ -5832,6 +6244,27 @@ enum LibOpenMP {
   LibIOMP5
 };
 
+/// Map a -fopenmp=<blah> macro to the corresponding library.
+static LibOpenMP getOpenMPLibByName(StringRef Name) {
+  return llvm::StringSwitch<LibOpenMP>(Name).Case("libgomp", LibGOMP)
+                                            .Case("libiomp5", LibIOMP5)
+                                            .Default(LibUnknown);
+}
+
+/// Get the default -l<blah> flag to use for -fopenmp, if no library is
+/// specified. This can be overridden at configure time.
+static const char *getDefaultOpenMPLibFlag() {
+#ifndef OPENMP_DEFAULT_LIB
+#define OPENMP_DEFAULT_LIB iomp5
+#endif
+
+#define STR2(lib) #lib
+#define STR(lib) STR2(lib)
+  return "-l" STR(OPENMP_DEFAULT_LIB);
+#undef STR
+#undef STR2
+}
+
 void darwin::Link::ConstructJob(Compilation &C, const JobAction &JA,
                                 const InputInfo &Output,
                                 const InputInfoList &Inputs,
@@ -5889,27 +6322,21 @@ void darwin::Link::ConstructJob(Compilation &C, const JobAction &JA,
 
   Args.AddAllArgs(CmdArgs, options::OPT_L);
 
-  LibOpenMP UsedOpenMPLib = LibUnknown;
-  if (Args.hasArg(options::OPT_fopenmp)) {
-    UsedOpenMPLib = LibGOMP;
-  } else if (const Arg *A = Args.getLastArg(options::OPT_fopenmp_EQ)) {
-    UsedOpenMPLib = llvm::StringSwitch<LibOpenMP>(A->getValue())
-        .Case("libgomp",  LibGOMP)
-        .Case("libiomp5", LibIOMP5)
-        .Default(LibUnknown);
-    if (UsedOpenMPLib == LibUnknown)
+  if (const Arg *A = Args.getLastArg(options::OPT_fopenmp_EQ)) {
+    switch (getOpenMPLibByName(A->getValue())) {
+    case LibGOMP:
+      CmdArgs.push_back("-lgomp");
+      break;
+    case LibIOMP5:
+      CmdArgs.push_back("-liomp5");
+      break;
+    case LibUnknown:
       getToolChain().getDriver().Diag(diag::err_drv_unsupported_option_argument)
         << A->getOption().getName() << A->getValue();
-  }
-  switch (UsedOpenMPLib) {
-  case LibGOMP:
-    CmdArgs.push_back("-lgomp");
-    break;
-  case LibIOMP5:
-    CmdArgs.push_back("-liomp5");
-    break;
-  case LibUnknown:
-    break;
+      break;
+    }
+  } else if (Args.hasArg(options::OPT_fopenmp)) {
+    CmdArgs.push_back(getDefaultOpenMPLibFlag());
   }
 
   AddLinkerInputs(getToolChain(), Inputs, Args, CmdArgs);
@@ -5951,6 +6378,11 @@ void darwin::Link::ConstructJob(Compilation &C, const JobAction &JA,
   if (Args.hasArg(options::OPT_fnested_functions))
     CmdArgs.push_back("-allow_stack_execute");
 
+  // TODO: It would be nice to use addProfileRT() here, but darwin's compiler-rt
+  // paths are different enough from other toolchains that this needs a fair
+  // amount of refactoring done first.
+  getMachOToolChain().addProfileRTLibs(Args, CmdArgs);
+
   if (!Args.hasArg(options::OPT_nostdlib) &&
       !Args.hasArg(options::OPT_nodefaultlibs)) {
     if (getToolChain().getDriver().CCCIsCXX())
@@ -5970,6 +6402,22 @@ void darwin::Link::ConstructJob(Compilation &C, const JobAction &JA,
   Args.AddAllArgs(CmdArgs, options::OPT_T_Group);
   Args.AddAllArgs(CmdArgs, options::OPT_F);
 
+  // -iframework should be forwarded as -F.
+  for (auto it = Args.filtered_begin(options::OPT_iframework),
+         ie = Args.filtered_end(); it != ie; ++it)
+    CmdArgs.push_back(Args.MakeArgString(std::string("-F") +
+                                         (*it)->getValue()));
+
+  if (!Args.hasArg(options::OPT_nostdlib) &&
+      !Args.hasArg(options::OPT_nodefaultlibs)) {
+    if (Arg *A = Args.getLastArg(options::OPT_fveclib)) {
+      if (A->getValue() == StringRef("Accelerate")) {
+        CmdArgs.push_back("-framework");
+        CmdArgs.push_back("Accelerate");
+      }
+    }
+  }
+
   const char *Exec =
     Args.MakeArgString(getToolChain().GetLinkerPath());
   std::unique_ptr<Command> Cmd =
@@ -6191,6 +6639,7 @@ void openbsd::Assemble::ConstructJob(Compilation &C, const JobAction &JA,
     break;
 
   case llvm::Triple::sparc:
+  case llvm::Triple::sparcel:
     CmdArgs.push_back("-32");
     NeedsKPIC = true;
     break;
@@ -6569,6 +7018,7 @@ void freebsd::Assemble::ConstructJob(Compilation &C, const JobAction &JA,
       CmdArgs.push_back("-matpcs");
     }
   } else if (getToolChain().getArch() == llvm::Triple::sparc ||
+             getToolChain().getArch() == llvm::Triple::sparcel ||
              getToolChain().getArch() == llvm::Triple::sparcv9) {
     if (getToolChain().getArch() == llvm::Triple::sparc)
       CmdArgs.push_back("-Av8plusa");
@@ -6697,7 +7147,7 @@ void freebsd::Link::ConstructJob(Compilation &C, const JobAction &JA,
   Args.AddAllArgs(CmdArgs, options::OPT_Z_Flag);
   Args.AddAllArgs(CmdArgs, options::OPT_r);
 
-  if (D.IsUsingLTO(Args))
+  if (D.IsUsingLTO(getToolChain(), Args))
     AddGoldPlugin(ToolChain, Args, CmdArgs);
 
   bool NeedsSanitizerDeps = addSanitizerRuntimes(ToolChain, Args, CmdArgs);
@@ -6823,6 +7273,7 @@ void netbsd::Assemble::ConstructJob(Compilation &C, const JobAction &JA,
   }
 
   case llvm::Triple::sparc:
+  case llvm::Triple::sparcel:
     CmdArgs.push_back("-32");
     addAssemblerKPIC(Args, CmdArgs);
     break;
@@ -6901,7 +7352,8 @@ void netbsd::Link::ConstructJob(Compilation &C, const JobAction &JA,
     break;
   case llvm::Triple::armeb:
   case llvm::Triple::thumbeb:
-    arm::appendEBLinkFlags(Args, CmdArgs, getToolChain().getTriple());
+    arm::appendEBLinkFlags(Args, CmdArgs,
+        llvm::Triple(getToolChain().ComputeEffectiveClangTriple(Args)));
     CmdArgs.push_back("-m");
     switch (getToolChain().getTriple().getEnvironment()) {
     case llvm::Triple::EABI:
@@ -7069,47 +7521,66 @@ void gnutools::Assemble::ConstructJob(Compilation &C, const JobAction &JA,
   ArgStringList CmdArgs;
   bool NeedsKPIC = false;
 
+  switch (getToolChain().getArch()) {
+  default:
+    break;
   // Add --32/--64 to make sure we get the format we want.
   // This is incomplete
-  if (getToolChain().getArch() == llvm::Triple::x86) {
+  case llvm::Triple::x86:
     CmdArgs.push_back("--32");
-  } else if (getToolChain().getArch() == llvm::Triple::x86_64) {
+    break;
+  case llvm::Triple::x86_64:
     if (getToolChain().getTriple().getEnvironment() == llvm::Triple::GNUX32)
       CmdArgs.push_back("--x32");
     else
       CmdArgs.push_back("--64");
-  } else if (getToolChain().getArch() == llvm::Triple::ppc) {
+    break;
+  case llvm::Triple::ppc:
     CmdArgs.push_back("-a32");
     CmdArgs.push_back("-mppc");
     CmdArgs.push_back("-many");
-  } else if (getToolChain().getArch() == llvm::Triple::ppc64) {
+    break;
+  case llvm::Triple::ppc64:
     CmdArgs.push_back("-a64");
     CmdArgs.push_back("-mppc64");
     CmdArgs.push_back("-many");
-  } else if (getToolChain().getArch() == llvm::Triple::ppc64le) {
+    break;
+  case llvm::Triple::ppc64le:
     CmdArgs.push_back("-a64");
     CmdArgs.push_back("-mppc64");
     CmdArgs.push_back("-many");
     CmdArgs.push_back("-mlittle-endian");
-  } else if (getToolChain().getArch() == llvm::Triple::sparc) {
+    break;
+  case llvm::Triple::sparc:
+  case llvm::Triple::sparcel:
     CmdArgs.push_back("-32");
     CmdArgs.push_back("-Av8plusa");
     NeedsKPIC = true;
-  } else if (getToolChain().getArch() == llvm::Triple::sparcv9) {
+    break;
+  case llvm::Triple::sparcv9:
     CmdArgs.push_back("-64");
     CmdArgs.push_back("-Av9a");
     NeedsKPIC = true;
-  } else if (getToolChain().getArch() == llvm::Triple::arm ||
-             getToolChain().getArch() == llvm::Triple::armeb) {
-    StringRef MArch = getToolChain().getArchName();
-    if (MArch == "armv7" || MArch == "armv7a" || MArch == "armv7-a")
+    break;
+  case llvm::Triple::arm:
+  case llvm::Triple::armeb:
+  case llvm::Triple::thumb:
+  case llvm::Triple::thumbeb: {
+    const llvm::Triple &Triple = getToolChain().getTriple();
+    switch (Triple.getSubArch()) {
+    case llvm::Triple::ARMSubArch_v7:
       CmdArgs.push_back("-mfpu=neon");
-    if (MArch == "armv8" || MArch == "armv8a" || MArch == "armv8-a" ||
-        MArch == "armebv8" || MArch == "armebv8a" || MArch == "armebv8-a")
+      break;
+    case llvm::Triple::ARMSubArch_v8:
       CmdArgs.push_back("-mfpu=crypto-neon-fp-armv8");
+      break;
+    default:
+      break;
+    }
 
     StringRef ARMFloatABI = tools::arm::getARMFloatABI(
-        getToolChain().getDriver(), Args, getToolChain().getTriple());
+        getToolChain().getDriver(), Args,
+        llvm::Triple(getToolChain().ComputeEffectiveClangTriple(Args)));
     CmdArgs.push_back(Args.MakeArgString("-mfloat-abi=" + ARMFloatABI));
 
     Args.AddLastArg(CmdArgs, options::OPT_march_EQ);
@@ -7124,10 +7595,12 @@ void gnutools::Assemble::ConstructJob(Compilation &C, const JobAction &JA,
     else
       Args.AddLastArg(CmdArgs, options::OPT_mcpu_EQ);
     Args.AddLastArg(CmdArgs, options::OPT_mfpu_EQ);
-  } else if (getToolChain().getArch() == llvm::Triple::mips ||
-             getToolChain().getArch() == llvm::Triple::mipsel ||
-             getToolChain().getArch() == llvm::Triple::mips64 ||
-             getToolChain().getArch() == llvm::Triple::mips64el) {
+    break;
+  }
+  case llvm::Triple::mips:
+  case llvm::Triple::mipsel:
+  case llvm::Triple::mips64:
+  case llvm::Triple::mips64el: {
     StringRef CPUName;
     StringRef ABIName;
     mips::getMipsCPUAndABI(Args, getToolChain().getTriple(), CPUName, ABIName);
@@ -7211,11 +7684,15 @@ void gnutools::Assemble::ConstructJob(Compilation &C, const JobAction &JA,
                     options::OPT_mno_odd_spreg);
 
     NeedsKPIC = true;
-  } else if (getToolChain().getArch() == llvm::Triple::systemz) {
+    break;
+  }
+  case llvm::Triple::systemz: {
     // Always pass an -march option, since our default of z10 is later
     // than the GNU assembler's default.
     StringRef CPUName = getSystemZTargetCPU(Args);
     CmdArgs.push_back(Args.MakeArgString("-march=" + CPUName));
+    break;
+  }
   }
 
   if (NeedsKPIC)
@@ -7239,7 +7716,7 @@ void gnutools::Assemble::ConstructJob(Compilation &C, const JobAction &JA,
   if (Args.hasArg(options::OPT_gsplit_dwarf) &&
       getToolChain().getTriple().isOSLinux())
     SplitDebugInfo(getToolChain(), C, *this, JA, Args, Output,
-                   SplitDebugName(Args, Inputs));
+                   SplitDebugName(Args, Inputs[0]));
 }
 
 static void AddLibgcc(const llvm::Triple &Triple, const Driver &D,
@@ -7283,7 +7760,8 @@ static std::string getLinuxDynamicLinker(const ArgList &Args,
     else
       return "/system/bin/linker";
   } else if (ToolChain.getArch() == llvm::Triple::x86 ||
-             ToolChain.getArch() == llvm::Triple::sparc)
+             ToolChain.getArch() == llvm::Triple::sparc ||
+             ToolChain.getArch() == llvm::Triple::sparcel)
     return "/lib/ld-linux.so.2";
   else if (ToolChain.getArch() == llvm::Triple::aarch64)
     return "/lib/ld-linux-aarch64.so.1";
@@ -7385,6 +7863,7 @@ static const char *getLDMOption(const llvm::Triple &T, const ArgList &Args) {
   case llvm::Triple::ppc64le:
     return "elf64lppc";
   case llvm::Triple::sparc:
+  case llvm::Triple::sparcel:
     return "elf32_sparc";
   case llvm::Triple::sparcv9:
     return "elf64_sparc";
@@ -7424,11 +7903,7 @@ void gnutools::Link::ConstructJob(Compilation &C, const JobAction &JA,
   const bool IsPIE =
     !Args.hasArg(options::OPT_shared) &&
     !Args.hasArg(options::OPT_static) &&
-    (Args.hasArg(options::OPT_pie) || ToolChain.isPIEDefault() ||
-     // On Android every code is PIC so every executable is PIE
-     // Cannot use isPIEDefault here since otherwise
-     // PIE only logic will be enabled during compilation
-     isAndroid);
+    (Args.hasArg(options::OPT_pie) || ToolChain.isPIEDefault());
 
   ArgStringList CmdArgs;
 
@@ -7454,7 +7929,8 @@ void gnutools::Link::ConstructJob(Compilation &C, const JobAction &JA,
 
   if (ToolChain.getArch() == llvm::Triple::armeb ||
       ToolChain.getArch() == llvm::Triple::thumbeb)
-    arm::appendEBLinkFlags(Args, CmdArgs, getToolChain().getTriple());
+    arm::appendEBLinkFlags(Args, CmdArgs,
+        llvm::Triple(getToolChain().ComputeEffectiveClangTriple(Args)));
 
   for (const auto &Opt : ToolChain.ExtraOpts)
     CmdArgs.push_back(Opt.c_str());
@@ -7533,7 +8009,7 @@ void gnutools::Link::ConstructJob(Compilation &C, const JobAction &JA,
   for (const auto &Path : Paths)
     CmdArgs.push_back(Args.MakeArgString(StringRef("-L") + Path));
 
-  if (D.IsUsingLTO(Args))
+  if (D.IsUsingLTO(getToolChain(), Args))
     AddGoldPlugin(ToolChain, Args, CmdArgs);
 
   if (Args.hasArg(options::OPT_Z_Xlinker__no_demangle))
@@ -7556,6 +8032,8 @@ void gnutools::Link::ConstructJob(Compilation &C, const JobAction &JA,
       CmdArgs.push_back("-Bdynamic");
     CmdArgs.push_back("-lm");
   }
+  // Silence warnings when linking C code with a C++ '-stdlib' argument.
+  Args.ClaimAllArgs(options::OPT_stdlib_EQ);
 
   if (!Args.hasArg(options::OPT_nostdlib)) {
     if (!Args.hasArg(options::OPT_nodefaultlibs)) {
@@ -7565,37 +8043,33 @@ void gnutools::Link::ConstructJob(Compilation &C, const JobAction &JA,
       if (NeedsSanitizerDeps)
         linkSanitizerRuntimeDeps(ToolChain, CmdArgs);
 
-      LibOpenMP UsedOpenMPLib = LibUnknown;
-      if (Args.hasArg(options::OPT_fopenmp)) {
-        UsedOpenMPLib = LibGOMP;
-      } else if (const Arg *A = Args.getLastArg(options::OPT_fopenmp_EQ)) {
-        UsedOpenMPLib = llvm::StringSwitch<LibOpenMP>(A->getValue())
-            .Case("libgomp",  LibGOMP)
-            .Case("libiomp5", LibIOMP5)
-            .Default(LibUnknown);
-        if (UsedOpenMPLib == LibUnknown)
+      bool WantPthread = true;
+      if (const Arg *A = Args.getLastArg(options::OPT_fopenmp_EQ)) {
+        switch (getOpenMPLibByName(A->getValue())) {
+        case LibGOMP:
+          CmdArgs.push_back("-lgomp");
+
+          // FIXME: Exclude this for platforms with libgomp that don't require
+          // librt. Most modern Linux platforms require it, but some may not.
+          CmdArgs.push_back("-lrt");
+          break;
+        case LibIOMP5:
+          CmdArgs.push_back("-liomp5");
+          break;
+        case LibUnknown:
           D.Diag(diag::err_drv_unsupported_option_argument)
-            << A->getOption().getName() << A->getValue();
-      }
-      switch (UsedOpenMPLib) {
-      case LibGOMP:
-        CmdArgs.push_back("-lgomp");
-
-        // FIXME: Exclude this for platforms with libgomp that don't require
-        // librt. Most modern Linux platforms require it, but some may not.
-        CmdArgs.push_back("-lrt");
-        break;
-      case LibIOMP5:
-        CmdArgs.push_back("-liomp5");
-        break;
-      case LibUnknown:
-        break;
+              << A->getOption().getName() << A->getValue();
+          break;
+        }
+      } else if (Args.hasArg(options::OPT_fopenmp)) {
+        CmdArgs.push_back(getDefaultOpenMPLibFlag());
+      } else {
+        WantPthread = Args.hasArg(options::OPT_pthread) ||
+                      Args.hasArg(options::OPT_pthreads);
       }
       AddRunTimeLibs(ToolChain, D, CmdArgs, Args);
 
-      if ((Args.hasArg(options::OPT_pthread) ||
-           Args.hasArg(options::OPT_pthreads) || UsedOpenMPLib != LibUnknown) &&
-          !isAndroid)
+      if (WantPthread && !isAndroid)
         CmdArgs.push_back("-lpthread");
 
       CmdArgs.push_back("-lc");
@@ -7625,6 +8099,172 @@ void gnutools::Link::ConstructJob(Compilation &C, const JobAction &JA,
       llvm::make_unique<Command>(JA, *this, ToolChain.Linker.c_str(), CmdArgs));
 }
 
+
+// NaCl ARM assembly (inline or standalone) can be written with a set of macros
+// for the various SFI requirements like register masking. The assembly tool
+// inserts the file containing the macros as an input into all the assembly
+// jobs.
+void nacltools::AssembleARM::ConstructJob(Compilation &C, const JobAction &JA,
+                                          const InputInfo &Output,
+                                          const InputInfoList &Inputs,
+                                          const ArgList &Args,
+                                          const char *LinkingOutput) const {
+  const toolchains::NaCl_TC& ToolChain =
+    static_cast<const toolchains::NaCl_TC&>(getToolChain());
+  InputInfo NaClMacros(ToolChain.GetNaClArmMacrosPath(), types::TY_PP_Asm,
+                       "nacl-arm-macros.s");
+  InputInfoList NewInputs;
+  NewInputs.push_back(NaClMacros);
+  NewInputs.append(Inputs.begin(), Inputs.end());
+  gnutools::Assemble::ConstructJob(C, JA, Output, NewInputs, Args,
+                                   LinkingOutput);
+}
+
+
+// This is quite similar to gnutools::link::ConstructJob with changes that
+// we use static by default, do not yet support sanitizers or LTO, and a few
+// others. Eventually we can support more of that and hopefully migrate back
+// to gnutools::link.
+void nacltools::Link::ConstructJob(Compilation &C, const JobAction &JA,
+                                  const InputInfo &Output,
+                                  const InputInfoList &Inputs,
+                                  const ArgList &Args,
+                                  const char *LinkingOutput) const {
+
+  const toolchains::NaCl_TC& ToolChain =
+    static_cast<const toolchains::NaCl_TC&>(getToolChain());
+  const Driver &D = ToolChain.getDriver();
+  const bool IsStatic =
+    !Args.hasArg(options::OPT_dynamic) &&
+    !Args.hasArg(options::OPT_shared);
+
+  ArgStringList CmdArgs;
+
+  // Silence warning for "clang -g foo.o -o foo"
+  Args.ClaimAllArgs(options::OPT_g_Group);
+  // and "clang -emit-llvm foo.o -o foo"
+  Args.ClaimAllArgs(options::OPT_emit_llvm);
+  // and for "clang -w foo.o -o foo". Other warning options are already
+  // handled somewhere else.
+  Args.ClaimAllArgs(options::OPT_w);
+
+  if (!D.SysRoot.empty())
+    CmdArgs.push_back(Args.MakeArgString("--sysroot=" + D.SysRoot));
+
+  if (Args.hasArg(options::OPT_rdynamic))
+    CmdArgs.push_back("-export-dynamic");
+
+  if (Args.hasArg(options::OPT_s))
+    CmdArgs.push_back("-s");
+
+  // NaCl_TC doesn't have ExtraOpts like Linux; the only relevant flag from
+  // there is --build-id, which we do want.
+  CmdArgs.push_back("--build-id");
+
+  if (!IsStatic)
+    CmdArgs.push_back("--eh-frame-hdr");
+
+  CmdArgs.push_back("-m");
+  if (ToolChain.getArch() == llvm::Triple::x86)
+    CmdArgs.push_back("elf_i386_nacl");
+  else if (ToolChain.getArch() == llvm::Triple::arm)
+    CmdArgs.push_back("armelf_nacl");
+  else if (ToolChain.getArch() == llvm::Triple::x86_64)
+    CmdArgs.push_back("elf_x86_64_nacl");
+  else
+    D.Diag(diag::err_target_unsupported_arch) << ToolChain.getArchName() <<
+        "Native Client";
+
+
+  if (IsStatic)
+    CmdArgs.push_back("-static");
+  else if (Args.hasArg(options::OPT_shared))
+    CmdArgs.push_back("-shared");
+
+  CmdArgs.push_back("-o");
+  CmdArgs.push_back(Output.getFilename());
+  if (!Args.hasArg(options::OPT_nostdlib) &&
+      !Args.hasArg(options::OPT_nostartfiles)) {
+    if (!Args.hasArg(options::OPT_shared))
+      CmdArgs.push_back(Args.MakeArgString(ToolChain.GetFilePath("crt1.o")));
+    CmdArgs.push_back(Args.MakeArgString(ToolChain.GetFilePath("crti.o")));
+
+    const char *crtbegin;
+    if (IsStatic)
+      crtbegin = "crtbeginT.o";
+    else if (Args.hasArg(options::OPT_shared))
+      crtbegin = "crtbeginS.o";
+    else
+      crtbegin = "crtbegin.o";
+    CmdArgs.push_back(Args.MakeArgString(ToolChain.GetFilePath(crtbegin)));
+  }
+
+  Args.AddAllArgs(CmdArgs, options::OPT_L);
+  Args.AddAllArgs(CmdArgs, options::OPT_u);
+
+  const ToolChain::path_list &Paths = ToolChain.getFilePaths();
+
+  for (const auto &Path : Paths)
+    CmdArgs.push_back(Args.MakeArgString(StringRef("-L") + Path));
+
+  if (Args.hasArg(options::OPT_Z_Xlinker__no_demangle))
+    CmdArgs.push_back("--no-demangle");
+
+  AddLinkerInputs(ToolChain, Inputs, Args, CmdArgs);
+
+  if (D.CCCIsCXX() &&
+      !Args.hasArg(options::OPT_nostdlib) &&
+      !Args.hasArg(options::OPT_nodefaultlibs)) {
+    bool OnlyLibstdcxxStatic = Args.hasArg(options::OPT_static_libstdcxx) &&
+      !IsStatic;
+    if (OnlyLibstdcxxStatic)
+      CmdArgs.push_back("-Bstatic");
+    ToolChain.AddCXXStdlibLibArgs(Args, CmdArgs);
+    if (OnlyLibstdcxxStatic)
+      CmdArgs.push_back("-Bdynamic");
+    CmdArgs.push_back("-lm");
+  }
+
+  if (!Args.hasArg(options::OPT_nostdlib)) {
+    if (!Args.hasArg(options::OPT_nodefaultlibs)) {
+      // Always use groups, since it has no effect on dynamic libraries.
+      CmdArgs.push_back("--start-group");
+      CmdArgs.push_back("-lc");
+      // NaCl's libc++ currently requires libpthread, so just always include it
+      // in the group for C++.
+      if (Args.hasArg(options::OPT_pthread) ||
+          Args.hasArg(options::OPT_pthreads) ||
+          D.CCCIsCXX()) {
+        CmdArgs.push_back("-lpthread");
+      }
+
+      CmdArgs.push_back("-lgcc");
+      CmdArgs.push_back("--as-needed");
+      if (IsStatic)
+        CmdArgs.push_back("-lgcc_eh");
+      else
+        CmdArgs.push_back("-lgcc_s");
+      CmdArgs.push_back("--no-as-needed");
+      CmdArgs.push_back("--end-group");
+    }
+
+    if (!Args.hasArg(options::OPT_nostartfiles)) {
+      const char *crtend;
+      if (Args.hasArg(options::OPT_shared))
+        crtend = "crtendS.o";
+      else
+        crtend = "crtend.o";
+
+      CmdArgs.push_back(Args.MakeArgString(ToolChain.GetFilePath(crtend)));
+      CmdArgs.push_back(Args.MakeArgString(ToolChain.GetFilePath("crtn.o")));
+    }
+  }
+
+  C.addCommand(llvm::make_unique<Command>(JA, *this,
+                                          ToolChain.Linker.c_str(), CmdArgs));
+}
+
+
 void minix::Assemble::ConstructJob(Compilation &C, const JobAction &JA,
                                    const InputInfo &Output,
                                    const InputInfoList &Inputs,
@@ -7915,8 +8555,8 @@ void visualstudio::Link::ConstructJob(Compilation &C, const JobAction &JA,
   if (!llvm::sys::Process::GetEnv("LIB")) {
     // If the VC environment hasn't been configured (perhaps because the user
     // did not run vcvarsall), try to build a consistent link environment.  If
-    // the environment variable is set however, assume the user knows what he's
-    // doing.
+    // the environment variable is set however, assume the user knows what
+    // they're doing.
     std::string VisualStudioDir;
     const auto &MSVC = static_cast<const toolchains::MSVCToolChain &>(TC);
     if (MSVC.getVisualStudioInstallDir(VisualStudioDir)) {
@@ -7950,14 +8590,16 @@ void visualstudio::Link::ConstructJob(Compilation &C, const JobAction &JA,
   if (Args.hasArg(options::OPT_g_Group))
     CmdArgs.push_back("-debug");
 
-  bool DLL = Args.hasArg(options::OPT__SLASH_LD, options::OPT__SLASH_LDd);
+  bool DLL = Args.hasArg(options::OPT__SLASH_LD,
+                         options::OPT__SLASH_LDd,
+                         options::OPT_shared);
   if (DLL) {
     CmdArgs.push_back(Args.MakeArgString("-dll"));
 
     SmallString<128> ImplibName(Output.getFilename());
     llvm::sys::path::replace_extension(ImplibName, "lib");
     CmdArgs.push_back(Args.MakeArgString(std::string("-implib:") +
-                                         ImplibName.str()));
+                                         ImplibName));
   }
 
   if (TC.getSanitizerArgs().needsAsanRt()) {
@@ -8074,7 +8716,7 @@ std::unique_ptr<Command> visualstudio::Compile::GetCommand(
     }
   }
 
-  // Flags for which clang-cl have an alias.
+  // Flags for which clang-cl has an alias.
   // FIXME: How can we ensure this stays in sync with relevant clang-cl options?
 
   if (Args.hasFlag(options::OPT__SLASH_GR_, options::OPT__SLASH_GR,
@@ -8094,7 +8736,8 @@ std::unique_ptr<Command> visualstudio::Compile::GetCommand(
   if (Args.hasArg(options::OPT_g_Flag, options::OPT_gline_tables_only))
     CmdArgs.push_back("/Z7");
 
-  std::vector<std::string> Includes = Args.getAllArgValues(options::OPT_include);
+  std::vector<std::string> Includes =
+      Args.getAllArgValues(options::OPT_include);
   for (const auto &Include : Includes)
     CmdArgs.push_back(Args.MakeArgString(std::string("/FI") + Include));
 
diff --git a/lib/Driver/Tools.h b/lib/Driver/Tools.h
index 5aea8258d251..25fe063f7f12 100644
--- a/lib/Driver/Tools.h
+++ b/lib/Driver/Tools.h
@@ -40,7 +40,7 @@ using llvm::opt::ArgStringList;
   class LLVM_LIBRARY_VISIBILITY Clang : public Tool {
   public:
     static const char *getBaseInputName(const llvm::opt::ArgList &Args,
-                                        const InputInfoList &Inputs);
+                                        const InputInfo &Input);
     static const char *getBaseInputStem(const llvm::opt::ArgList &Args,
                                         const InputInfoList &Inputs);
     static const char *getDependencyFileName(const llvm::opt::ArgList &Args,
@@ -109,7 +109,8 @@ using llvm::opt::ArgStringList;
     ClangAs(const ToolChain &TC) : Tool("clang::as",
                                         "clang integrated assembler", TC,
                                         RF_Full) {}
-
+    void AddMIPSTargetArgs(const llvm::opt::ArgList &Args,
+                           llvm::opt::ArgStringList &CmdArgs) const;
     bool hasGoodDiagnostics() const override { return true; }
     bool hasIntegratedAssembler() const override { return false; }
     bool hasIntegratedCPP() const override { return false; }
@@ -225,14 +226,21 @@ namespace hexagon {
 namespace arm {
   StringRef getARMTargetCPU(const llvm::opt::ArgList &Args,
                             const llvm::Triple &Triple);
+  const StringRef getARMArch(const llvm::opt::ArgList &Args,
+                             const llvm::Triple &Triple);
   const char* getARMCPUForMArch(const llvm::opt::ArgList &Args,
                                 const llvm::Triple &Triple);
-  const char* getLLVMArchSuffixForARM(StringRef CPU);
+  const char* getLLVMArchSuffixForARM(StringRef CPU, StringRef Arch);
 
   void appendEBLinkFlags(const llvm::opt::ArgList &Args, ArgStringList &CmdArgs, const llvm::Triple &Triple);
 }
 
 namespace mips {
+  typedef enum {
+    NanLegacy = 1,
+    Nan2008 = 2
+  } NanEncoding;
+  NanEncoding getSupportedNanEncoding(StringRef &CPU);
   void getMipsCPUAndABI(const llvm::opt::ArgList &Args,
                         const llvm::Triple &Triple, StringRef &CPUName,
                         StringRef &ABIName);
@@ -247,6 +255,22 @@ namespace ppc {
   bool hasPPCAbiArg(const llvm::opt::ArgList &Args, const char *Value);
 }
 
+  /// cloudabi -- Directly call GNU Binutils linker
+namespace cloudabi {
+class LLVM_LIBRARY_VISIBILITY Link : public GnuTool {
+public:
+  Link(const ToolChain &TC) : GnuTool("cloudabi::Link", "linker", TC) {}
+
+  bool hasIntegratedCPP() const override { return false; }
+  bool isLinkJob() const override { return true; }
+
+  void ConstructJob(Compilation &C, const JobAction &JA,
+                    const InputInfo &Output, const InputInfoList &Inputs,
+                    const llvm::opt::ArgList &TCArgs,
+                    const char *LinkingOutput) const override;
+};
+} // end namespace cloudabi
+
 namespace darwin {
   llvm::Triple::ArchType getArchTypeForMachOArchName(StringRef Str);
   void setTripleTypeForMachOArchName(llvm::Triple &T, StringRef Str);
@@ -490,6 +514,33 @@ namespace gnutools {
                       const char *LinkingOutput) const override;
   };
 }
+
+namespace nacltools {
+  class LLVM_LIBRARY_VISIBILITY AssembleARM : public gnutools::Assemble  {
+  public:
+    AssembleARM(const ToolChain &TC) : gnutools::Assemble(TC) {}
+
+    void ConstructJob(Compilation &C, const JobAction &JA,
+                      const InputInfo &Output,
+                      const InputInfoList &Inputs,
+                      const llvm::opt::ArgList &TCArgs,
+                      const char *LinkingOutput) const override;
+  };
+  class LLVM_LIBRARY_VISIBILITY Link : public Tool  {
+  public:
+    Link(const ToolChain &TC) : Tool("NaCl::Link", "linker", TC) {}
+
+    bool hasIntegratedCPP() const override { return false; }
+    bool isLinkJob() const override { return true; }
+
+    void ConstructJob(Compilation &C, const JobAction &JA,
+                              const InputInfo &Output,
+                              const InputInfoList &Inputs,
+                              const llvm::opt::ArgList &TCArgs,
+                              const char *LinkingOutput) const override;
+  };
+}
+
   /// minix -- Directly call GNU Binutils assembler and linker
 namespace minix {
   class LLVM_LIBRARY_VISIBILITY Assemble : public GnuTool  {
diff --git a/lib/Driver/Types.cpp b/lib/Driver/Types.cpp
index 6ee764c64e34..7b281457557b 100644
--- a/lib/Driver/Types.cpp
+++ b/lib/Driver/Types.cpp
@@ -85,7 +85,7 @@ bool types::isAcceptedByClang(ID Id) {
   case TY_Asm:
   case TY_C: case TY_PP_C:
   case TY_CL:
-  case TY_CUDA:
+  case TY_CUDA: case TY_PP_CUDA:
   case TY_ObjC: case TY_PP_ObjC: case TY_PP_ObjC_Alias:
   case TY_CXX: case TY_PP_CXX:
   case TY_ObjCXX: case TY_PP_ObjCXX: case TY_PP_ObjCXX_Alias:
@@ -122,7 +122,7 @@ bool types::isCXX(ID Id) {
   case TY_ObjCXX: case TY_PP_ObjCXX: case TY_PP_ObjCXX_Alias:
   case TY_CXXHeader: case TY_PP_CXXHeader:
   case TY_ObjCXXHeader: case TY_PP_ObjCXXHeader:
-  case TY_CUDA:
+  case TY_CUDA: case TY_PP_CUDA:
     return true;
   }
 }
@@ -153,6 +153,7 @@ types::ID types::lookupTypeForExtension(const char *Ext) {
            .Case("cl", TY_CL)
            .Case("cp", TY_CXX)
            .Case("cu", TY_CUDA)
+           .Case("cui", TY_PP_CUDA)
            .Case("hh", TY_CXXHeader)
            .Case("ll", TY_LLVM_IR)
            .Case("hpp", TY_CXXHeader)
diff --git a/lib/Edit/EditedSource.cpp b/lib/Edit/EditedSource.cpp
index 1c66cb827706..e557de92410e 100644
--- a/lib/Edit/EditedSource.cpp
+++ b/lib/Edit/EditedSource.cpp
@@ -135,7 +135,7 @@ bool EditedSource::commitInsertFromRange(SourceLocation OrigLoc,
     StrVec += text;
   }
 
-  return commitInsert(OrigLoc, Offs, StrVec.str(), beforePreviousInsertions);
+  return commitInsert(OrigLoc, Offs, StrVec, beforePreviousInsertions);
 }
 
 void EditedSource::commitRemove(SourceLocation OrigLoc,
@@ -295,9 +295,11 @@ static void adjustRemoval(const SourceManager &SM, const LangOptions &LangOpts,
   }
 
   if (buffer[end] == ' ') {
+    assert((end + 1 != buffer.size() || buffer.data()[end + 1] == 0) &&
+           "buffer not zero-terminated!");
     if (canRemoveWhitespace(/*left=*/buffer[begin-1],
                             /*beforeWSpace=*/buffer[end-1],
-                            /*right=*/buffer[end+1],
+                            /*right=*/buffer.data()[end + 1], // zero-terminated
                             LangOpts))
       ++len;
     return;
@@ -360,14 +362,14 @@ void EditedSource::applyRewrites(EditsReceiver &receiver) {
       continue;
     }
 
-    applyRewrite(receiver, StrVec.str(), CurOffs, CurLen, SourceMgr, LangOpts);
+    applyRewrite(receiver, StrVec, CurOffs, CurLen, SourceMgr, LangOpts);
     CurOffs = offs;
     StrVec = act.Text;
     CurLen = act.RemoveLen;
     CurEnd = CurOffs.getWithOffset(CurLen);
   }
 
-  applyRewrite(receiver, StrVec.str(), CurOffs, CurLen, SourceMgr, LangOpts);
+  applyRewrite(receiver, StrVec, CurOffs, CurLen, SourceMgr, LangOpts);
 }
 
 void EditedSource::clearRewrites() {
diff --git a/lib/Format/BreakableToken.cpp b/lib/Format/BreakableToken.cpp
index 26f1371b4092..66e935abdf55 100644
--- a/lib/Format/BreakableToken.cpp
+++ b/lib/Format/BreakableToken.cpp
@@ -106,7 +106,7 @@ getStringSplit(StringRef Text, unsigned UsedColumns, unsigned ColumnLimit,
           Text.substr(0, Advance), UsedColumns + Chars, TabWidth, Encoding);
     }
 
-    if (Chars > MaxSplit || Text.size() == Advance)
+    if (Chars > MaxSplit || Text.size() <= Advance)
       break;
 
     if (IsBlank(Text[0]))
@@ -277,6 +277,8 @@ BreakableBlockComment::BreakableBlockComment(
     // If the last line is empty, the closing "*/" will have a star.
     if (i + 1 == e && Lines[i].empty())
       break;
+    if (!Lines[i].empty() && i + 1 != e && Decoration.startswith(Lines[i]))
+      continue;
     while (!Lines[i].startswith(Decoration))
       Decoration = Decoration.substr(0, Decoration.size() - 1);
   }
@@ -297,14 +299,18 @@ BreakableBlockComment::BreakableBlockComment(
       }
       continue;
     }
+
     // The first line already excludes the star.
     // For all other lines, adjust the line to exclude the star and
     // (optionally) the first whitespace.
-    StartOfLineColumn[i] += Decoration.size();
-    Lines[i] = Lines[i].substr(Decoration.size());
-    LeadingWhitespace[i] += Decoration.size();
-    IndentAtLineBreak =
-        std::min<int>(IndentAtLineBreak, std::max(0, StartOfLineColumn[i]));
+    unsigned DecorationSize =
+        Decoration.startswith(Lines[i]) ? Lines[i].size() : Decoration.size();
+    StartOfLineColumn[i] += DecorationSize;
+    Lines[i] = Lines[i].substr(DecorationSize);
+    LeadingWhitespace[i] += DecorationSize;
+    if (!Decoration.startswith(Lines[i]))
+      IndentAtLineBreak =
+          std::min<int>(IndentAtLineBreak, std::max(0, StartOfLineColumn[i]));
   }
   IndentAtLineBreak = std::max<unsigned>(IndentAtLineBreak, Decoration.size());
   DEBUG({
diff --git a/lib/Format/ContinuationIndenter.cpp b/lib/Format/ContinuationIndenter.cpp
index 4cc92b02a9e5..4e8f5af263d2 100644
--- a/lib/Format/ContinuationIndenter.cpp
+++ b/lib/Format/ContinuationIndenter.cpp
@@ -143,11 +143,10 @@ bool ContinuationIndenter::mustBreak(const LineState &State) {
   if (Previous.is(tok::semi) && State.LineContainsContinuedForLoopSection)
     return true;
   if ((startsNextParameter(Current, Style) || Previous.is(tok::semi) ||
-       (Style.BreakBeforeTernaryOperators &&
-        (Current.is(tok::question) ||
-         (Current.is(TT_ConditionalExpr) && Previous.isNot(tok::question)))) ||
+       (Style.BreakBeforeTernaryOperators && Current.is(TT_ConditionalExpr) &&
+        Previous.isNot(tok::question)) ||
        (!Style.BreakBeforeTernaryOperators &&
-        (Previous.is(tok::question) || Previous.is(TT_ConditionalExpr)))) &&
+        Previous.is(TT_ConditionalExpr))) &&
       State.Stack.back().BreakBeforeParameter && !Current.isTrailingComment() &&
       !Current.isOneOf(tok::r_paren, tok::r_brace))
     return true;
@@ -166,10 +165,17 @@ bool ContinuationIndenter::mustBreak(const LineState &State) {
       ((Style.AllowShortFunctionsOnASingleLine != FormatStyle::SFS_All) ||
        Style.BreakConstructorInitializersBeforeComma || Style.ColumnLimit != 0))
     return true;
+  if (Current.is(TT_SelectorName) && State.Stack.back().ObjCSelectorNameFound &&
+      State.Stack.back().BreakBeforeParameter)
+    return true;
 
   if (State.Column < getNewLineColumn(State))
     return false;
-  if (Style.BreakBeforeBinaryOperators == FormatStyle::BOS_None) {
+
+  // Using CanBreakBefore here and below takes care of the decision whether the
+  // current style uses wrapping before or after operators for the given
+  // operator.
+  if (Previous.is(TT_BinaryOperator) && Current.CanBreakBefore) {
     // If we need to break somewhere inside the LHS of a binary expression, we
     // should also break after the operator. Otherwise, the formatting would
     // hide the operator precedence, e.g. in:
@@ -185,16 +191,13 @@ bool ContinuationIndenter::mustBreak(const LineState &State) {
                         Previous.Previous->isNot(TT_BinaryOperator); // For >>.
     bool LHSIsBinaryExpr =
         Previous.Previous && Previous.Previous->EndsBinaryExpression;
-    if (Previous.is(TT_BinaryOperator) && (!IsComparison || LHSIsBinaryExpr) &&
-        Current.isNot(TT_BinaryOperator) && // For >>.
-        !Current.isTrailingComment() && !Previous.is(tok::lessless) &&
+    if ((!IsComparison || LHSIsBinaryExpr) && !Current.isTrailingComment() &&
         Previous.getPrecedence() != prec::Assignment &&
         State.Stack.back().BreakBeforeParameter)
       return true;
-  } else {
-    if (Current.is(TT_BinaryOperator) && Previous.EndsBinaryExpression &&
-        State.Stack.back().BreakBeforeParameter)
-      return true;
+  } else if (Current.is(TT_BinaryOperator) && Current.CanBreakBefore &&
+             State.Stack.back().BreakBeforeParameter) {
+    return true;
   }
 
   // Same as above, but for the first "<<" operator.
@@ -203,12 +206,14 @@ bool ContinuationIndenter::mustBreak(const LineState &State) {
       State.Stack.back().FirstLessLess == 0)
     return true;
 
-  if (Current.is(TT_SelectorName) && State.Stack.back().ObjCSelectorNameFound &&
-      State.Stack.back().BreakBeforeParameter)
-    return true;
   if (Current.NestingLevel == 0 && !Current.isTrailingComment()) {
+    // Always break after "template <...>" and leading annotations. This is only
+    // for cases where the entire line does not fit on a single line as a
+    // different LineFormatter would be used otherwise.
     if (Previous.ClosesTemplateDeclaration)
       return true;
+    if (Previous.is(TT_FunctionAnnotationRParen))
+      return true;
     if (Previous.is(TT_LeadingJavaAnnotation) && Current.isNot(tok::l_paren) &&
         Current.isNot(TT_LeadingJavaAnnotation))
       return true;
@@ -221,8 +226,7 @@ bool ContinuationIndenter::mustBreak(const LineState &State) {
 
   if (startsSegmentOfBuilderTypeCall(Current) &&
       (State.Stack.back().CallContinuation != 0 ||
-       (State.Stack.back().BreakBeforeParameter &&
-        State.Stack.back().ContainsUnwrappedBuilder)))
+       State.Stack.back().BreakBeforeParameter))
     return true;
 
   // The following could be precomputed as they do not depend on the state.
@@ -232,6 +236,10 @@ bool ContinuationIndenter::mustBreak(const LineState &State) {
       Previous.is(tok::l_brace) && !Current.isOneOf(tok::r_brace, tok::comment))
     return true;
 
+  if (Current.is(tok::lessless) && Previous.is(tok::identifier) &&
+      Previous.TokenText == "endl")
+    return true;
+
   return false;
 }
 
@@ -245,12 +253,18 @@ unsigned ContinuationIndenter::addTokenToState(LineState &State, bool Newline,
        (Current.Previous->Tok.getIdentifierInfo() == nullptr ||
         Current.Previous->Tok.getIdentifierInfo()->getPPKeywordID() ==
             tok::pp_not_keyword))) {
-    // FIXME: Is this correct?
-    int WhitespaceLength = SourceMgr.getSpellingColumnNumber(
-                               State.NextToken->WhitespaceRange.getEnd()) -
-                           SourceMgr.getSpellingColumnNumber(
-                               State.NextToken->WhitespaceRange.getBegin());
-    State.Column += WhitespaceLength;
+    unsigned EndColumn =
+        SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getEnd());
+    if (Current.LastNewlineOffset != 0) {
+      // If there is a newline within this token, the final column will solely
+      // determined by the current end column.
+      State.Column = EndColumn;
+    } else {
+      unsigned StartColumn =
+          SourceMgr.getSpellingColumnNumber(Current.WhitespaceRange.getBegin());
+      assert(EndColumn >= StartColumn);
+      State.Column += EndColumn - StartColumn;
+    }
     moveStateToNextToken(State, DryRun, /*Newline=*/false);
     return 0;
   }
@@ -297,7 +311,9 @@ void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
     else if (State.Stack.back().Indent + Current.LongestObjCSelectorName >
              State.Column + Spaces + Current.ColumnWidth)
       State.Stack.back().ColonPos =
-          State.Stack.back().Indent + Current.LongestObjCSelectorName;
+          std::max(State.FirstIndent + Style.ContinuationIndentWidth,
+                   State.Stack.back().Indent) +
+          Current.LongestObjCSelectorName;
     else
       State.Stack.back().ColonPos = State.Column + Spaces + Current.ColumnWidth;
   }
@@ -308,9 +324,12 @@ void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
     State.Stack.back().Indent = State.Column + Spaces;
   if (State.Stack.back().AvoidBinPacking && startsNextParameter(Current, Style))
     State.Stack.back().NoLineBreak = true;
-  if (startsSegmentOfBuilderTypeCall(Current))
+  if (startsSegmentOfBuilderTypeCall(Current) &&
+      State.Column > getNewLineColumn(State))
     State.Stack.back().ContainsUnwrappedBuilder = true;
 
+  if (Current.is(TT_LambdaArrow))
+    State.Stack.back().NoLineBreak = true;
   if (Current.isMemberAccess() && Previous.is(tok::r_paren) &&
       (Previous.MatchingParen &&
        (Previous.TotalLength - Previous.MatchingParen->TotalLength > 10))) {
@@ -359,7 +378,7 @@ void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
       const FormatToken *Next = Previous.MatchingParen->getNextNonComment();
       HasTrailingCall = Next && Next->isMemberAccess();
     }
-    if (HasTrailingCall &&
+    if (HasTrailingCall && State.Stack.size() > 1 &&
         State.Stack[State.Stack.size() - 2].CallContinuation == 0)
       State.Stack.back().LastSpace = State.Column;
   }
@@ -406,7 +425,11 @@ unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
         State.Stack.back().AlignColons = false;
       } else {
         State.Stack.back().ColonPos =
-            State.Stack.back().Indent + NextNonComment->LongestObjCSelectorName;
+            (Style.IndentWrappedFunctionNames
+                 ? std::max(State.Stack.back().Indent,
+                            State.FirstIndent + Style.ContinuationIndentWidth)
+                 : State.Stack.back().Indent) +
+            NextNonComment->LongestObjCSelectorName;
       }
     } else if (State.Stack.back().AlignColons &&
                State.Stack.back().ColonPos <= NextNonComment->ColumnWidth) {
@@ -468,8 +491,9 @@ unsigned ContinuationIndenter::addTokenOnNewLine(LineState &State,
       !PreviousNonComment->isOneOf(tok::comma, tok::semi) &&
       (PreviousNonComment->isNot(TT_TemplateCloser) ||
        Current.NestingLevel != 0) &&
-      !PreviousNonComment->isOneOf(TT_BinaryOperator, TT_JavaAnnotation,
-                                   TT_LeadingJavaAnnotation) &&
+      !PreviousNonComment->isOneOf(
+          TT_BinaryOperator, TT_FunctionAnnotationRParen, TT_JavaAnnotation,
+          TT_LeadingJavaAnnotation) &&
       Current.isNot(TT_BinaryOperator) && !PreviousNonComment->opensScope())
     State.Stack.back().BreakBeforeParameter = true;
 
@@ -516,7 +540,7 @@ unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
   if (NextNonComment->is(tok::l_brace) && NextNonComment->BlockKind == BK_Block)
     return Current.NestingLevel == 0 ? State.FirstIndent
                                      : State.Stack.back().Indent;
-  if (Current.isOneOf(tok::r_brace, tok::r_square)) {
+  if (Current.isOneOf(tok::r_brace, tok::r_square) && State.Stack.size() > 1) {
     if (Current.closesBlockTypeList(Style))
       return State.Stack[State.Stack.size() - 2].NestedBlockIndent;
     if (Current.MatchingParen &&
@@ -529,6 +553,9 @@ unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
     return State.Stack.back().Indent;
   if (NextNonComment->isStringLiteral() && State.StartOfStringLiteral != 0)
     return State.StartOfStringLiteral;
+  if (NextNonComment->is(TT_ObjCStringLiteral) &&
+      State.StartOfStringLiteral != 0)
+    return State.StartOfStringLiteral - 1;
   if (NextNonComment->is(tok::lessless) &&
       State.Stack.back().FirstLessLess != 0)
     return State.Stack.back().FirstLessLess;
@@ -546,8 +573,9 @@ unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
     return State.Stack.back().VariablePos;
   if ((PreviousNonComment &&
        (PreviousNonComment->ClosesTemplateDeclaration ||
-        PreviousNonComment->isOneOf(TT_AttributeParen, TT_JavaAnnotation,
-                                    TT_LeadingJavaAnnotation))) ||
+        PreviousNonComment->isOneOf(
+            TT_AttributeParen, TT_FunctionAnnotationRParen, TT_JavaAnnotation,
+            TT_LeadingJavaAnnotation))) ||
       (!Style.IndentWrappedFunctionNames &&
        NextNonComment->isOneOf(tok::kw_operator, TT_FunctionDeclarationName)))
     return std::max(State.Stack.back().LastSpace, State.Stack.back().Indent);
@@ -555,7 +583,10 @@ unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
     if (!State.Stack.back().ObjCSelectorNameFound) {
       if (NextNonComment->LongestObjCSelectorName == 0)
         return State.Stack.back().Indent;
-      return State.Stack.back().Indent +
+      return (Style.IndentWrappedFunctionNames
+                  ? std::max(State.Stack.back().Indent,
+                             State.FirstIndent + Style.ContinuationIndentWidth)
+                  : State.Stack.back().Indent) +
              NextNonComment->LongestObjCSelectorName -
              NextNonComment->ColumnWidth;
     }
@@ -570,10 +601,16 @@ unsigned ContinuationIndenter::getNewLineColumn(const LineState &State) {
       return State.Stack.back().StartOfArraySubscripts;
     return ContinuationIndent;
   }
-  if (NextNonComment->is(TT_StartOfName) ||
-      Previous.isOneOf(tok::coloncolon, tok::equal)) {
+
+  // This ensure that we correctly format ObjC methods calls without inputs,
+  // i.e. where the last element isn't selector like: [callee method];
+  if (NextNonComment->is(tok::identifier) && NextNonComment->FakeRParens == 0 &&
+      NextNonComment->Next && NextNonComment->Next->is(TT_ObjCMethodExpr))
+    return State.Stack.back().Indent;
+
+  if (NextNonComment->isOneOf(TT_StartOfName, TT_PointerOrReference) ||
+      Previous.isOneOf(tok::coloncolon, tok::equal))
     return ContinuationIndent;
-  }
   if (PreviousNonComment && PreviousNonComment->is(tok::colon) &&
       PreviousNonComment->isOneOf(TT_ObjCMethodExpr, TT_DictLiteral))
     return ContinuationIndent;
@@ -621,7 +658,7 @@ unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
         std::min(State.LowestLevelOnLine, Current.NestingLevel);
   if (Current.isMemberAccess())
     State.Stack.back().StartOfFunctionCall =
-        Current.LastOperator ? 0 : State.Column + Current.ColumnWidth;
+        Current.LastOperator ? 0 : State.Column;
   if (Current.is(TT_SelectorName))
     State.Stack.back().ObjCSelectorNameFound = true;
   if (Current.is(TT_CtorInitializerColon)) {
@@ -637,12 +674,9 @@ unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
       State.Stack.back().AvoidBinPacking = true;
     State.Stack.back().BreakBeforeParameter = false;
   }
-
-  // In ObjC method declaration we align on the ":" of parameters, but we need
-  // to ensure that we indent parameters on subsequent lines by at least our
-  // continuation indent width.
-  if (Current.is(TT_ObjCMethodSpecifier))
-    State.Stack.back().Indent += Style.ContinuationIndentWidth;
+  if (Current.isOneOf(TT_BinaryOperator, TT_ConditionalExpr) && Newline)
+    State.Stack.back().NestedBlockIndent =
+        State.Column + Current.ColumnWidth + 1;
 
   // Insert scopes created by fake parenthesis.
   const FormatToken *Previous = Current.getPreviousNonComment();
@@ -675,12 +709,13 @@ unsigned ContinuationIndenter::moveStateToNextToken(LineState &State,
   moveStatePastScopeCloser(State);
   moveStatePastFakeRParens(State);
 
-  if (Current.isStringLiteral() && State.StartOfStringLiteral == 0) {
+  if (Current.isStringLiteral() && State.StartOfStringLiteral == 0)
     State.StartOfStringLiteral = State.Column;
-  } else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) &&
-             !Current.isStringLiteral()) {
+  if (Current.is(TT_ObjCStringLiteral) && State.StartOfStringLiteral == 0)
+    State.StartOfStringLiteral = State.Column + 1;
+  else if (!Current.isOneOf(tok::comment, tok::identifier, tok::hash) &&
+             !Current.isStringLiteral())
     State.StartOfStringLiteral = 0;
-  }
 
   State.Column += Current.ColumnWidth;
   State.NextToken = State.NextToken->Next;
@@ -712,7 +747,8 @@ void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
   // 'return', assignments or opening <({[. The indentation for these cases
   // is special cased.
   bool SkipFirstExtraIndent =
-      (Previous && (Previous->opensScope() || Previous->is(tok::kw_return) ||
+      (Previous && (Previous->opensScope() ||
+                    Previous->isOneOf(tok::semi, tok::kw_return) ||
                     (Previous->getPrecedence() == prec::Assignment &&
                      Style.AlignOperands) ||
                     Previous->is(TT_ObjCMethodExpr)));
@@ -783,7 +819,6 @@ void ContinuationIndenter::moveStatePastFakeLParens(LineState &State,
 void ContinuationIndenter::moveStatePastFakeRParens(LineState &State) {
   for (unsigned i = 0, e = State.NextToken->FakeRParens; i != e; ++i) {
     unsigned VariablePos = State.Stack.back().VariablePos;
-    assert(State.Stack.size() > 1);
     if (State.Stack.size() == 1) {
       // Do not pop the last element.
       break;
@@ -806,6 +841,7 @@ void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
 
   unsigned NewIndent;
   unsigned NewIndentLevel = State.Stack.back().IndentLevel;
+  unsigned LastSpace = State.Stack.back().LastSpace;
   bool AvoidBinPacking;
   bool BreakBeforeParameter = false;
   if (Current.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare)) {
@@ -815,17 +851,28 @@ void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
       ++NewIndentLevel;
     } else {
       NewIndent = State.Stack.back().LastSpace + Style.ContinuationIndentWidth;
-      NewIndent = std::min(State.Column + 1, NewIndent);
     }
     const FormatToken *NextNoComment = Current.getNextNonComment();
     AvoidBinPacking =
         Current.isOneOf(TT_ArrayInitializerLSquare, TT_DictLiteral) ||
-        Style.Language == FormatStyle::LK_Proto || !Style.BinPackParameters ||
+        Style.Language == FormatStyle::LK_Proto || !Style.BinPackArguments ||
         (NextNoComment && NextNoComment->is(TT_DesignatedInitializerPeriod));
   } else {
     NewIndent = Style.ContinuationIndentWidth +
                 std::max(State.Stack.back().LastSpace,
                          State.Stack.back().StartOfFunctionCall);
+
+    // Ensure that different different brackets force relative alignment, e.g.:
+    // void SomeFunction(vector<  // break
+    //                       int> v);
+    // FIXME: We likely want to do this for more combinations of brackets.
+    // Verify that it is wanted for ObjC, too.
+    if (Current.Tok.getKind() == tok::less &&
+        Current.ParentBracket == tok::l_paren) {
+      NewIndent = std::max(NewIndent, State.Stack.back().Indent);
+      LastSpace = std::max(LastSpace, State.Stack.back().Indent);
+    }
+
     AvoidBinPacking =
         (State.Line->MustBeDeclaration && !Style.BinPackParameters) ||
         (!State.Line->MustBeDeclaration && !Style.BinPackArguments) ||
@@ -833,19 +880,33 @@ void ContinuationIndenter::moveStatePastScopeOpener(LineState &State,
          (Current.PackingKind == PPK_OnePerLine ||
           (!BinPackInconclusiveFunctions &&
            Current.PackingKind == PPK_Inconclusive)));
-    // If this '[' opens an ObjC call, determine whether all parameters fit
-    // into one line and put one per line if they don't.
-    if (Current.is(TT_ObjCMethodExpr) && Style.ColumnLimit != 0 &&
-        getLengthToMatchingParen(Current) + State.Column >
+    if (Current.is(TT_ObjCMethodExpr) && Current.MatchingParen) {
+      if (Style.ColumnLimit) {
+        // If this '[' opens an ObjC call, determine whether all parameters fit
+        // into one line and put one per line if they don't.
+        if (getLengthToMatchingParen(Current) + State.Column >
             getColumnLimit(State))
-      BreakBeforeParameter = true;
+          BreakBeforeParameter = true;
+      } else {
+        // For ColumnLimit = 0, we have to figure out whether there is or has to
+        // be a line break within this call.
+        for (const FormatToken *Tok = &Current;
+             Tok && Tok != Current.MatchingParen; Tok = Tok->Next) {
+          if (Tok->MustBreakBefore || 
+              (Tok->CanBreakBefore && Tok->NewlinesBefore > 0)) {
+            BreakBeforeParameter = true;
+            break;
+          }
+        }
+      }
+    }
   }
   bool NoLineBreak = State.Stack.back().NoLineBreak ||
                      (Current.is(TT_TemplateOpener) &&
                       State.Stack.back().ContainsUnwrappedBuilder);
-  unsigned NestedBlockIndent = State.Stack.back().NestedBlockIndent;
-  State.Stack.push_back(ParenState(NewIndent, NewIndentLevel,
-                                   State.Stack.back().LastSpace,
+  unsigned NestedBlockIndent = std::max(State.Stack.back().StartOfFunctionCall,
+                                        State.Stack.back().NestedBlockIndent);
+  State.Stack.push_back(ParenState(NewIndent, NewIndentLevel, LastSpace,
                                    AvoidBinPacking, NoLineBreak));
   State.Stack.back().NestedBlockIndent = NestedBlockIndent;
   State.Stack.back().BreakBeforeParameter = BreakBeforeParameter;
@@ -1082,8 +1143,9 @@ bool ContinuationIndenter::nextIsMultilineString(const LineState &State) {
   if (Current.getNextNonComment() &&
       Current.getNextNonComment()->isStringLiteral())
     return true; // Implicit concatenation.
-  if (State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
-      Style.ColumnLimit)
+  if (Style.ColumnLimit != 0 &&
+      State.Column + Current.ColumnWidth + Current.UnbreakableTailLength >
+          Style.ColumnLimit)
     return true; // String will be split.
   return false;
 }
diff --git a/lib/Format/Format.cpp b/lib/Format/Format.cpp
index 2a4721f2b3b7..10c68f9da617 100644
--- a/lib/Format/Format.cpp
+++ b/lib/Format/Format.cpp
@@ -109,10 +109,8 @@ struct ScalarEnumerationTraits<FormatStyle::NamespaceIndentationKind> {
   }
 };
 
-template <>
-struct ScalarEnumerationTraits<FormatStyle::PointerAlignmentStyle> {
-  static void enumeration(IO &IO,
-                          FormatStyle::PointerAlignmentStyle &Value) {
+template <> struct ScalarEnumerationTraits<FormatStyle::PointerAlignmentStyle> {
+  static void enumeration(IO &IO, FormatStyle::PointerAlignmentStyle &Value) {
     IO.enumCase(Value, "Middle", FormatStyle::PAS_Middle);
     IO.enumCase(Value, "Left", FormatStyle::PAS_Left);
     IO.enumCase(Value, "Right", FormatStyle::PAS_Right);
@@ -144,8 +142,8 @@ template <> struct MappingTraits<FormatStyle> {
     IO.mapOptional("Language", Style.Language);
 
     if (IO.outputting()) {
-      StringRef StylesArray[] = { "LLVM",    "Google", "Chromium",
-                                  "Mozilla", "WebKit", "GNU" };
+      StringRef StylesArray[] = {"LLVM",    "Google", "Chromium",
+                                 "Mozilla", "WebKit", "GNU"};
       ArrayRef<StringRef> Styles(StylesArray);
       for (size_t i = 0, e = Styles.size(); i < e; ++i) {
         StringRef StyleName(Styles[i]);
@@ -176,6 +174,7 @@ template <> struct MappingTraits<FormatStyle> {
     IO.mapOptional("AlignEscapedNewlinesLeft", Style.AlignEscapedNewlinesLeft);
     IO.mapOptional("AlignOperands", Style.AlignOperands);
     IO.mapOptional("AlignTrailingComments", Style.AlignTrailingComments);
+    IO.mapOptional("AlignConsecutiveAssignments", Style.AlignConsecutiveAssignments);
     IO.mapOptional("AllowAllParametersOfDeclarationOnNextLine",
                    Style.AllowAllParametersOfDeclarationOnNextLine);
     IO.mapOptional("AllowShortBlocksOnASingleLine",
@@ -273,7 +272,7 @@ template <> struct MappingTraits<FormatStyle> {
 // will be used to get default values for missing keys.
 // If the first element has no Language specified, it will be treated as the
 // default one for the following elements.
-template <> struct DocumentListTraits<std::vector<FormatStyle> > {
+template <> struct DocumentListTraits<std::vector<FormatStyle>> {
   static size_t size(IO &IO, std::vector<FormatStyle> &Seq) {
     return Seq.size();
   }
@@ -331,6 +330,7 @@ FormatStyle getLLVMStyle() {
   LLVMStyle.AlignAfterOpenBracket = true;
   LLVMStyle.AlignOperands = true;
   LLVMStyle.AlignTrailingComments = true;
+  LLVMStyle.AlignConsecutiveAssignments = false;
   LLVMStyle.AllowAllParametersOfDeclarationOnNextLine = true;
   LLVMStyle.AllowShortFunctionsOnASingleLine = FormatStyle::SFS_All;
   LLVMStyle.AllowShortBlocksOnASingleLine = false;
@@ -600,10 +600,10 @@ public:
   FormatTokenLexer(SourceManager &SourceMgr, FileID ID, FormatStyle &Style,
                    encoding::Encoding Encoding)
       : FormatTok(nullptr), IsFirstToken(true), GreaterStashed(false),
-        Column(0), TrailingWhitespace(0), SourceMgr(SourceMgr), ID(ID),
-        Style(Style), IdentTable(getFormattingLangOpts(Style)),
-        Keywords(IdentTable), Encoding(Encoding), FirstInLineIndex(0),
-        FormattingDisabled(false) {
+        LessStashed(false), Column(0), TrailingWhitespace(0),
+        SourceMgr(SourceMgr), ID(ID), Style(Style),
+        IdentTable(getFormattingLangOpts(Style)), Keywords(IdentTable),
+        Encoding(Encoding), FirstInLineIndex(0), FormattingDisabled(false) {
     Lex.reset(new Lexer(ID, SourceMgr.getBuffer(ID), SourceMgr,
                         getFormattingLangOpts(Style)));
     Lex->SetKeepWhitespaceMode(true);
@@ -619,7 +619,7 @@ public:
     do {
       Tokens.push_back(getNextToken());
       tryMergePreviousTokens();
-      if (Tokens.back()->NewlinesBefore > 0)
+      if (Tokens.back()->NewlinesBefore > 0 || Tokens.back()->IsMultiline)
         FirstInLineIndex = Tokens.size() - 1;
     } while (Tokens.back()->Tok.isNot(tok::eof));
     return Tokens;
@@ -633,32 +633,62 @@ private:
       return;
     if (tryMergeConflictMarkers())
       return;
+    if (tryMergeLessLess())
+      return;
 
     if (Style.Language == FormatStyle::LK_JavaScript) {
       if (tryMergeJSRegexLiteral())
         return;
       if (tryMergeEscapeSequence())
         return;
+      if (tryMergeTemplateString())
+        return;
 
-      static tok::TokenKind JSIdentity[] = { tok::equalequal, tok::equal };
-      static tok::TokenKind JSNotIdentity[] = { tok::exclaimequal, tok::equal };
-      static tok::TokenKind JSShiftEqual[] = { tok::greater, tok::greater,
-                                               tok::greaterequal };
-      static tok::TokenKind JSRightArrow[] = { tok::equal, tok::greater };
-      // FIXME: We probably need to change token type to mimic operator with the
-      // correct priority.
-      if (tryMergeTokens(JSIdentity))
+      static const tok::TokenKind JSIdentity[] = {tok::equalequal, tok::equal};
+      static const tok::TokenKind JSNotIdentity[] = {tok::exclaimequal,
+                                                     tok::equal};
+      static const tok::TokenKind JSShiftEqual[] = {tok::greater, tok::greater,
+                                                    tok::greaterequal};
+      static const tok::TokenKind JSRightArrow[] = {tok::equal, tok::greater};
+      // FIXME: Investigate what token type gives the correct operator priority.
+      if (tryMergeTokens(JSIdentity, TT_BinaryOperator))
         return;
-      if (tryMergeTokens(JSNotIdentity))
+      if (tryMergeTokens(JSNotIdentity, TT_BinaryOperator))
         return;
-      if (tryMergeTokens(JSShiftEqual))
+      if (tryMergeTokens(JSShiftEqual, TT_BinaryOperator))
         return;
-      if (tryMergeTokens(JSRightArrow))
+      if (tryMergeTokens(JSRightArrow, TT_JsFatArrow))
         return;
     }
   }
 
-  bool tryMergeTokens(ArrayRef<tok::TokenKind> Kinds) {
+  bool tryMergeLessLess() {
+    // Merge X,less,less,Y into X,lessless,Y unless X or Y is less.
+    if (Tokens.size() < 3)
+      return false;
+
+    bool FourthTokenIsLess = false;
+    if (Tokens.size() > 3)
+      FourthTokenIsLess = (Tokens.end() - 4)[0]->is(tok::less);
+
+    auto First = Tokens.end() - 3;
+    if (First[2]->is(tok::less) || First[1]->isNot(tok::less) ||
+        First[0]->isNot(tok::less) || FourthTokenIsLess)
+      return false;
+
+    // Only merge if there currently is no whitespace between the two "<".
+    if (First[1]->WhitespaceRange.getBegin() !=
+        First[1]->WhitespaceRange.getEnd())
+      return false;
+
+    First[0]->Tok.setKind(tok::lessless);
+    First[0]->TokenText = "<<";
+    First[0]->ColumnWidth += 1;
+    Tokens.erase(Tokens.end() - 2);
+    return true;
+  }
+
+  bool tryMergeTokens(ArrayRef<tok::TokenKind> Kinds, TokenType NewType) {
     if (Tokens.size() < Kinds.size())
       return false;
 
@@ -668,8 +698,9 @@ private:
       return false;
     unsigned AddLength = 0;
     for (unsigned i = 1; i < Kinds.size(); ++i) {
-      if (!First[i]->is(Kinds[i]) || First[i]->WhitespaceRange.getBegin() !=
-                                         First[i]->WhitespaceRange.getEnd())
+      if (!First[i]->is(Kinds[i]) ||
+          First[i]->WhitespaceRange.getBegin() !=
+              First[i]->WhitespaceRange.getEnd())
         return false;
       AddLength += First[i]->TokenText.size();
     }
@@ -677,6 +708,7 @@ private:
     First[0]->TokenText = StringRef(First[0]->TokenText.data(),
                                     First[0]->TokenText.size() + AddLength);
     First[0]->ColumnWidth += AddLength;
+    First[0]->Type = NewType;
     return true;
   }
 
@@ -720,7 +752,7 @@ private:
     unsigned LastColumn = Tokens.back()->OriginalColumn;
     for (auto I = Tokens.rbegin() + 1, E = Tokens.rend(); I != E; ++I) {
       ++TokenCount;
-      if (I[0]->is(tok::slash) && I + 1 != E &&
+      if (I[0]->isOneOf(tok::slash, tok::slashequal) && I + 1 != E &&
           (I[1]->isOneOf(tok::l_paren, tok::semi, tok::l_brace, tok::r_brace,
                          tok::exclaim, tok::l_square, tok::colon, tok::comma,
                          tok::question, tok::kw_return) ||
@@ -745,6 +777,91 @@ private:
     return false;
   }
 
+  bool tryMergeTemplateString() {
+    if (Tokens.size() < 2)
+      return false;
+
+    FormatToken *EndBacktick = Tokens.back();
+    // Backticks get lexed as tok::unknown tokens. If a template string contains
+    // a comment start, it gets lexed as a tok::comment, or tok::unknown if
+    // unterminated.
+    if (!EndBacktick->isOneOf(tok::comment, tok::unknown))
+      return false;
+    size_t CommentBacktickPos = EndBacktick->TokenText.find('`');
+    // Unknown token that's not actually a backtick, or a comment that doesn't
+    // contain a backtick.
+    if (CommentBacktickPos == StringRef::npos)
+      return false;
+
+    unsigned TokenCount = 0;
+    bool IsMultiline = false;
+    unsigned EndColumnInFirstLine =
+        EndBacktick->OriginalColumn + EndBacktick->ColumnWidth;
+    for (auto I = Tokens.rbegin() + 1, E = Tokens.rend(); I != E; I++) {
+      ++TokenCount;
+      if (I[0]->NewlinesBefore > 0 || I[0]->IsMultiline)
+        IsMultiline = true;
+
+      // If there was a preceding template string, this must be the start of a
+      // template string, not the end.
+      if (I[0]->is(TT_TemplateString))
+        return false;
+
+      if (I[0]->isNot(tok::unknown) || I[0]->TokenText != "`") {
+        // Keep track of the rhs offset of the last token to wrap across lines -
+        // its the rhs offset of the first line of the template string, used to
+        // determine its width.
+        if (I[0]->IsMultiline)
+          EndColumnInFirstLine = I[0]->OriginalColumn + I[0]->ColumnWidth;
+        // If the token has newlines, the token before it (if it exists) is the
+        // rhs end of the previous line.
+        if (I[0]->NewlinesBefore > 0 && (I + 1 != E))
+          EndColumnInFirstLine = I[1]->OriginalColumn + I[1]->ColumnWidth;
+
+        continue;
+      }
+
+      Tokens.resize(Tokens.size() - TokenCount);
+      Tokens.back()->Type = TT_TemplateString;
+      const char *EndOffset =
+          EndBacktick->TokenText.data() + 1 + CommentBacktickPos;
+      if (CommentBacktickPos != 0) {
+        // If the backtick was not the first character (e.g. in a comment),
+        // re-lex after the backtick position.
+        SourceLocation Loc = EndBacktick->Tok.getLocation();
+        resetLexer(SourceMgr.getFileOffset(Loc) + CommentBacktickPos + 1);
+      }
+      Tokens.back()->TokenText =
+          StringRef(Tokens.back()->TokenText.data(),
+                    EndOffset - Tokens.back()->TokenText.data());
+
+      unsigned EndOriginalColumn = EndBacktick->OriginalColumn;
+      if (EndOriginalColumn == 0) {
+        SourceLocation Loc = EndBacktick->Tok.getLocation();
+        EndOriginalColumn = SourceMgr.getSpellingColumnNumber(Loc);
+      }
+      // If the ` is further down within the token (e.g. in a comment).
+      EndOriginalColumn += CommentBacktickPos;
+
+      if (IsMultiline) {
+        // ColumnWidth is from backtick to last token in line.
+        // LastLineColumnWidth is 0 to backtick.
+        // x = `some content
+        //     until here`;
+        Tokens.back()->ColumnWidth =
+            EndColumnInFirstLine - Tokens.back()->OriginalColumn;
+        Tokens.back()->LastLineColumnWidth = EndOriginalColumn;
+        Tokens.back()->IsMultiline = true;
+      } else {
+        // Token simply spans from start to end, +1 for the ` itself.
+        Tokens.back()->ColumnWidth =
+            EndOriginalColumn - Tokens.back()->OriginalColumn + 1;
+      }
+      return true;
+    }
+    return false;
+  }
+
   bool tryMerge_TMacro() {
     if (Tokens.size() < 4)
       return false;
@@ -772,6 +889,8 @@ private:
     String->OriginalColumn = Macro->OriginalColumn;
     String->ColumnWidth = encoding::columnWidthWithTabs(
         String->TokenText, String->OriginalColumn, Style.TabWidth, Encoding);
+    String->NewlinesBefore = Macro->NewlinesBefore;
+    String->HasUnescapedNewline = Macro->HasUnescapedNewline;
 
     Tokens.pop_back();
     Tokens.pop_back();
@@ -842,21 +961,33 @@ private:
     return false;
   }
 
+  FormatToken *getStashedToken() {
+    // Create a synthesized second '>' or '<' token.
+    Token Tok = FormatTok->Tok;
+    StringRef TokenText = FormatTok->TokenText;
+
+    unsigned OriginalColumn = FormatTok->OriginalColumn;
+    FormatTok = new (Allocator.Allocate()) FormatToken;
+    FormatTok->Tok = Tok;
+    SourceLocation TokLocation =
+        FormatTok->Tok.getLocation().getLocWithOffset(Tok.getLength() - 1);
+    FormatTok->Tok.setLocation(TokLocation);
+    FormatTok->WhitespaceRange = SourceRange(TokLocation, TokLocation);
+    FormatTok->TokenText = TokenText;
+    FormatTok->ColumnWidth = 1;
+    FormatTok->OriginalColumn = OriginalColumn + 1;
+
+    return FormatTok;
+  }
+
   FormatToken *getNextToken() {
     if (GreaterStashed) {
-      // Create a synthesized second '>' token.
-      // FIXME: Increment Column and set OriginalColumn.
-      Token Greater = FormatTok->Tok;
-      FormatTok = new (Allocator.Allocate()) FormatToken;
-      FormatTok->Tok = Greater;
-      SourceLocation GreaterLocation =
-          FormatTok->Tok.getLocation().getLocWithOffset(1);
-      FormatTok->WhitespaceRange =
-          SourceRange(GreaterLocation, GreaterLocation);
-      FormatTok->TokenText = ">";
-      FormatTok->ColumnWidth = 1;
       GreaterStashed = false;
-      return FormatTok;
+      return getStashedToken();
+    }
+    if (LessStashed) {
+      LessStashed = false;
+      return getStashedToken();
     }
 
     FormatTok = new (Allocator.Allocate()) FormatToken;
@@ -869,20 +1000,32 @@ private:
     // Consume and record whitespace until we find a significant token.
     unsigned WhitespaceLength = TrailingWhitespace;
     while (FormatTok->Tok.is(tok::unknown)) {
-      for (int i = 0, e = FormatTok->TokenText.size(); i != e; ++i) {
-        switch (FormatTok->TokenText[i]) {
+      StringRef Text = FormatTok->TokenText;
+      auto EscapesNewline = [&](int pos) {
+        // A '\r' here is just part of '\r\n'. Skip it.
+        if (pos >= 0 && Text[pos] == '\r')
+          --pos;
+        // See whether there is an odd number of '\' before this.
+        unsigned count = 0;
+        for (; pos >= 0; --pos, ++count)
+          if (Text[pos] != '\\')
+            break;
+        return count & 1;
+      };
+      // FIXME: This miscounts tok:unknown tokens that are not just
+      // whitespace, e.g. a '`' character.
+      for (int i = 0, e = Text.size(); i != e; ++i) {
+        switch (Text[i]) {
         case '\n':
           ++FormatTok->NewlinesBefore;
-          // FIXME: This is technically incorrect, as it could also
-          // be a literal backslash at the end of the line.
-          if (i == 0 || (FormatTok->TokenText[i - 1] != '\\' &&
-                         (FormatTok->TokenText[i - 1] != '\r' || i == 1 ||
-                          FormatTok->TokenText[i - 2] != '\\')))
-            FormatTok->HasUnescapedNewline = true;
+          FormatTok->HasUnescapedNewline = !EscapesNewline(i - 1);
           FormatTok->LastNewlineOffset = WhitespaceLength + i + 1;
           Column = 0;
           break;
         case '\r':
+          FormatTok->LastNewlineOffset = WhitespaceLength + i + 1;
+          Column = 0;
+          break;
         case '\f':
         case '\v':
           Column = 0;
@@ -894,8 +1037,7 @@ private:
           Column += Style.TabWidth - Column % Style.TabWidth;
           break;
         case '\\':
-          if (i + 1 == e || (FormatTok->TokenText[i + 1] != '\r' &&
-                             FormatTok->TokenText[i + 1] != '\n'))
+          if (i + 1 == e || (Text[i + 1] != '\r' && Text[i + 1] != '\n'))
             FormatTok->Type = TT_ImplicitStringLiteral;
           break;
         default:
@@ -920,6 +1062,7 @@ private:
            FormatTok->TokenText[1] == '\n') {
       ++FormatTok->NewlinesBefore;
       WhitespaceLength += 2;
+      FormatTok->LastNewlineOffset = 2;
       Column = 0;
       FormatTok->TokenText = FormatTok->TokenText.substr(2);
     }
@@ -948,6 +1091,10 @@ private:
       FormatTok->Tok.setKind(tok::greater);
       FormatTok->TokenText = FormatTok->TokenText.substr(0, 1);
       GreaterStashed = true;
+    } else if (FormatTok->Tok.is(tok::lessless)) {
+      FormatTok->Tok.setKind(tok::less);
+      FormatTok->TokenText = FormatTok->TokenText.substr(0, 1);
+      LessStashed = true;
     }
 
     // Now FormatTok is the next non-whitespace token.
@@ -975,16 +1122,16 @@ private:
       Column = FormatTok->LastLineColumnWidth;
     }
 
-    FormatTok->IsForEachMacro =
-        std::binary_search(ForEachMacros.begin(), ForEachMacros.end(),
-                           FormatTok->Tok.getIdentifierInfo());
+    if (std::find(ForEachMacros.begin(), ForEachMacros.end(),
+                  FormatTok->Tok.getIdentifierInfo()) != ForEachMacros.end())
+      FormatTok->Type = TT_ForEachMacro;
 
     return FormatTok;
   }
 
   FormatToken *FormatTok;
   bool IsFirstToken;
-  bool GreaterStashed;
+  bool GreaterStashed, LessStashed;
   unsigned Column;
   unsigned TrailingWhitespace;
   std::unique_ptr<Lexer> Lex;
@@ -1072,13 +1219,13 @@ public:
                        << "\n");
   }
 
-  tooling::Replacements format() {
+  tooling::Replacements format(bool *IncompleteFormat) {
     tooling::Replacements Result;
     FormatTokenLexer Tokens(SourceMgr, ID, Style, Encoding);
 
     UnwrappedLineParser Parser(Style, Tokens.getKeywords(), Tokens.lex(),
                                *this);
-    bool StructuralError = Parser.parse();
+    Parser.parse();
     assert(UnwrappedLines.rbegin()->empty());
     for (unsigned Run = 0, RunE = UnwrappedLines.size(); Run + 1 != RunE;
          ++Run) {
@@ -1088,7 +1235,7 @@ public:
         AnnotatedLines.push_back(new AnnotatedLine(UnwrappedLines[Run][i]));
       }
       tooling::Replacements RunResult =
-          format(AnnotatedLines, StructuralError, Tokens);
+          format(AnnotatedLines, Tokens, IncompleteFormat);
       DEBUG({
         llvm::dbgs() << "Replacements for run " << Run << ":\n";
         for (tooling::Replacements::iterator I = RunResult.begin(),
@@ -1107,7 +1254,7 @@ public:
   }
 
   tooling::Replacements format(SmallVectorImpl<AnnotatedLine *> &AnnotatedLines,
-                               bool StructuralError, FormatTokenLexer &Tokens) {
+                               FormatTokenLexer &Tokens, bool *IncompleteFormat) {
     TokenAnnotator Annotator(Style, Tokens.getKeywords());
     for (unsigned i = 0, e = AnnotatedLines.size(); i != e; ++i) {
       Annotator.annotate(*AnnotatedLines[i]);
@@ -1122,8 +1269,9 @@ public:
     ContinuationIndenter Indenter(Style, Tokens.getKeywords(), SourceMgr,
                                   Whitespaces, Encoding,
                                   BinPackInconclusiveFunctions);
-    UnwrappedLineFormatter Formatter(&Indenter, &Whitespaces, Style);
-    Formatter.format(AnnotatedLines, /*DryRun=*/false);
+    UnwrappedLineFormatter(&Indenter, &Whitespaces, Style, Tokens.getKeywords(),
+                           IncompleteFormat)
+        .format(AnnotatedLines);
     return Whitespaces.generateReplacements();
   }
 
@@ -1340,27 +1488,20 @@ private:
 
 } // end anonymous namespace
 
-tooling::Replacements reformat(const FormatStyle &Style, Lexer &Lex,
-                               SourceManager &SourceMgr,
-                               ArrayRef<CharSourceRange> Ranges) {
-  if (Style.DisableFormat)
-    return tooling::Replacements();
-  return reformat(Style, SourceMgr,
-                  SourceMgr.getFileID(Lex.getSourceLocation()), Ranges);
-}
-
 tooling::Replacements reformat(const FormatStyle &Style,
                                SourceManager &SourceMgr, FileID ID,
-                               ArrayRef<CharSourceRange> Ranges) {
+                               ArrayRef<CharSourceRange> Ranges,
+                               bool *IncompleteFormat) {
   if (Style.DisableFormat)
     return tooling::Replacements();
   Formatter formatter(Style, SourceMgr, ID, Ranges);
-  return formatter.format();
+  return formatter.format(IncompleteFormat);
 }
 
 tooling::Replacements reformat(const FormatStyle &Style, StringRef Code,
                                ArrayRef<tooling::Range> Ranges,
-                               StringRef FileName) {
+                               StringRef FileName,
+                               bool *IncompleteFormat) {
   if (Style.DisableFormat)
     return tooling::Replacements();
 
@@ -1383,7 +1524,7 @@ tooling::Replacements reformat(const FormatStyle &Style, StringRef Code,
     SourceLocation End = Start.getLocWithOffset(Range.getLength());
     CharRanges.push_back(CharSourceRange::getCharRange(Start, End));
   }
-  return reformat(Style, SourceMgr, ID, CharRanges);
+  return reformat(Style, SourceMgr, ID, CharRanges, IncompleteFormat);
 }
 
 LangOptions getFormattingLangOpts(const FormatStyle &Style) {
@@ -1392,12 +1533,12 @@ LangOptions getFormattingLangOpts(const FormatStyle &Style) {
   LangOpts.CPlusPlus11 = Style.Standard == FormatStyle::LS_Cpp03 ? 0 : 1;
   LangOpts.CPlusPlus14 = Style.Standard == FormatStyle::LS_Cpp03 ? 0 : 1;
   LangOpts.LineComment = 1;
-  bool AlternativeOperators = Style.Language != FormatStyle::LK_JavaScript &&
-                              Style.Language != FormatStyle::LK_Java;
+  bool AlternativeOperators = Style.Language == FormatStyle::LK_Cpp;
   LangOpts.CXXOperatorNames = AlternativeOperators ? 1 : 0;
   LangOpts.Bool = 1;
   LangOpts.ObjC1 = 1;
   LangOpts.ObjC2 = 1;
+  LangOpts.MicrosoftExt = 1; // To get kw___try, kw___finally.
   return LangOpts;
 }
 
@@ -1415,7 +1556,8 @@ const char *StyleOptionHelpDescription =
 static FormatStyle::LanguageKind getLanguageByFileName(StringRef FileName) {
   if (FileName.endswith(".java")) {
     return FormatStyle::LK_Java;
-  } else if (FileName.endswith_lower(".js")) {
+  } else if (FileName.endswith_lower(".js") || FileName.endswith_lower(".ts")) {
+    // JavaScript or TypeScript.
     return FormatStyle::LK_JavaScript;
   } else if (FileName.endswith_lower(".proto") ||
              FileName.endswith_lower(".protodevel")) {
diff --git a/lib/Format/FormatToken.cpp b/lib/Format/FormatToken.cpp
index badb3a39c82c..88678ca1abe1 100644
--- a/lib/Format/FormatToken.cpp
+++ b/lib/Format/FormatToken.cpp
@@ -18,6 +18,7 @@
 #include "clang/Format/Format.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/Support/Debug.h"
+#include <climits>
 
 namespace clang {
 namespace format {
@@ -59,12 +60,13 @@ void TokenRole::precomputeFormattingInfos(const FormatToken *Token) {}
 unsigned CommaSeparatedList::formatAfterToken(LineState &State,
                                               ContinuationIndenter *Indenter,
                                               bool DryRun) {
-  if (!State.NextToken->Previous || !State.NextToken->Previous->Previous)
+  if (State.NextToken == nullptr || !State.NextToken->Previous)
     return 0;
 
   // Ensure that we start on the opening brace.
-  const FormatToken *LBrace = State.NextToken->Previous->Previous;
-  if (LBrace->isNot(tok::l_brace) || LBrace->BlockKind == BK_Block ||
+  const FormatToken *LBrace =
+      State.NextToken->Previous->getPreviousNonComment();
+  if (!LBrace || LBrace->isNot(tok::l_brace) || LBrace->BlockKind == BK_Block ||
       LBrace->Type == TT_DictLiteral ||
       LBrace->Next->Type == TT_DesignatedInitializerPeriod)
     return 0;
@@ -132,9 +134,9 @@ void CommaSeparatedList::precomputeFormattingInfos(const FormatToken *Token) {
     return;
 
   // In C++11 braced list style, we should not format in columns unless they
-  // have many items (20 or more) or we allow bin-packing of function
-  // parameters.
-  if (Style.Cpp11BracedListStyle && !Style.BinPackParameters &&
+  // have many items (20 or more) or we allow bin-packing of function call
+  // arguments.
+  if (Style.Cpp11BracedListStyle && !Style.BinPackArguments &&
       Commas.size() < 19)
     return;
 
@@ -143,19 +145,21 @@ void CommaSeparatedList::precomputeFormattingInfos(const FormatToken *Token) {
     return;
 
   FormatToken *ItemBegin = Token->Next;
+  while (ItemBegin->isTrailingComment())
+    ItemBegin = ItemBegin->Next;
   SmallVector<bool, 8> MustBreakBeforeItem;
 
   // The lengths of an item if it is put at the end of the line. This includes
   // trailing comments which are otherwise ignored for column alignment.
   SmallVector<unsigned, 8> EndOfLineItemLength;
 
-  unsigned MinItemLength = Style.ColumnLimit;
-  unsigned MaxItemLength = 0;
-
+  bool HasSeparatingComment = false;
   for (unsigned i = 0, e = Commas.size() + 1; i != e; ++i) {
     // Skip comments on their own line.
-    while (ItemBegin->HasUnescapedNewline && ItemBegin->isTrailingComment())
+    while (ItemBegin->HasUnescapedNewline && ItemBegin->isTrailingComment()) {
       ItemBegin = ItemBegin->Next;
+      HasSeparatingComment = i > 0;
+    }
 
     MustBreakBeforeItem.push_back(ItemBegin->MustBreakBefore);
     if (ItemBegin->is(tok::l_brace))
@@ -178,8 +182,6 @@ void CommaSeparatedList::precomputeFormattingInfos(const FormatToken *Token) {
       ItemEnd = Commas[i];
       // The comma is counted as part of the item when calculating the length.
       ItemLengths.push_back(CodePointsBetween(ItemBegin, ItemEnd));
-      MinItemLength = std::min(MinItemLength, ItemLengths.back());
-      MaxItemLength = std::max(MaxItemLength, ItemLengths.back());
 
       // Consume trailing comments so the are included in EndOfLineItemLength.
       if (ItemEnd->Next && !ItemEnd->Next->HasUnescapedNewline &&
@@ -194,12 +196,9 @@ void CommaSeparatedList::precomputeFormattingInfos(const FormatToken *Token) {
     ItemBegin = ItemEnd->Next;
   }
 
-  // If this doesn't have a nested list, we require at least 6 elements in order
-  // create a column layout. If it has a nested list, column layout ensures one
-  // list element per line. If the difference between the shortest and longest
-  // element is too large, column layout would create too much whitespace.
-  if (HasNestedBracedList || Commas.size() < 5 || Token->NestingLevel != 0 ||
-      MaxItemLength - MinItemLength > 10)
+  // Don't use column layout for nested lists, lists with few elements and in
+  // presence of separating comments.
+  if (Token->NestingLevel != 0 || Commas.size() < 5 || HasSeparatingComment)
     return;
 
   // We can never place more than ColumnLimit / 3 items in a row (because of the
@@ -208,6 +207,7 @@ void CommaSeparatedList::precomputeFormattingInfos(const FormatToken *Token) {
     ColumnFormat Format;
     Format.Columns = Columns;
     Format.ColumnSizes.resize(Columns);
+    std::vector<unsigned> MinSizeInColumn(Columns, UINT_MAX);
     Format.LineCount = 1;
     bool HasRowWithSufficientColumns = false;
     unsigned Column = 0;
@@ -219,9 +219,10 @@ void CommaSeparatedList::precomputeFormattingInfos(const FormatToken *Token) {
       }
       if (Column == Columns - 1)
         HasRowWithSufficientColumns = true;
-      unsigned length =
+      unsigned Length =
           (Column == Columns - 1) ? EndOfLineItemLength[i] : ItemLengths[i];
-      Format.ColumnSizes[Column] = std::max(Format.ColumnSizes[Column], length);
+      Format.ColumnSizes[Column] = std::max(Format.ColumnSizes[Column], Length);
+      MinSizeInColumn[Column] = std::min(MinSizeInColumn[Column], Length);
       ++Column;
     }
     // If all rows are terminated early (e.g. by trailing comments), we don't
@@ -229,9 +230,19 @@ void CommaSeparatedList::precomputeFormattingInfos(const FormatToken *Token) {
     if (!HasRowWithSufficientColumns)
       break;
     Format.TotalWidth = Columns - 1; // Width of the N-1 spaces.
-    for (unsigned i = 0; i < Columns; ++i) {
+
+    for (unsigned i = 0; i < Columns; ++i)
       Format.TotalWidth += Format.ColumnSizes[i];
-    }
+
+    // Don't use this Format, if the difference between the longest and shortest
+    // element in a column exceeds a threshold to avoid excessive spaces.
+    if ([&] {
+          for (unsigned i = 0; i < Columns - 1; ++i)
+            if (Format.ColumnSizes[i] - MinSizeInColumn[i] > 10)
+              return true;
+          return false;
+        }())
+      continue;
 
     // Ignore layouts that are bound to violate the column limit.
     if (Format.TotalWidth > Style.ColumnLimit)
diff --git a/lib/Format/FormatToken.h b/lib/Format/FormatToken.h
index 4811e02dd228..ec0fdf4aa813 100644
--- a/lib/Format/FormatToken.h
+++ b/lib/Format/FormatToken.h
@@ -41,13 +41,19 @@ enum TokenType {
   TT_CtorInitializerComma,
   TT_DesignatedInitializerPeriod,
   TT_DictLiteral,
+  TT_ForEachMacro,
+  TT_FunctionAnnotationRParen,
   TT_FunctionDeclarationName,
   TT_FunctionLBrace,
   TT_FunctionTypeLParen,
   TT_ImplicitStringLiteral,
   TT_InheritanceColon,
+  TT_InlineASMBrace,
   TT_InlineASMColon,
   TT_JavaAnnotation,
+  TT_JsFatArrow,
+  TT_JsTypeColon,
+  TT_JsTypeOptionalQuestion,
   TT_LambdaArrow,
   TT_LambdaLSquare,
   TT_LeadingJavaAnnotation,
@@ -59,6 +65,7 @@ enum TokenType {
   TT_ObjCMethodExpr,
   TT_ObjCMethodSpecifier,
   TT_ObjCProperty,
+  TT_ObjCStringLiteral,
   TT_OverloadedOperator,
   TT_OverloadedOperatorLParen,
   TT_PointerOrReference,
@@ -69,6 +76,7 @@ enum TokenType {
   TT_StartOfName,
   TT_TemplateCloser,
   TT_TemplateOpener,
+  TT_TemplateString,
   TT_TrailingAnnotation,
   TT_TrailingReturnArrow,
   TT_TrailingUnaryOperator,
@@ -102,21 +110,7 @@ class AnnotatedLine;
 /// \brief A wrapper around a \c Token storing information about the
 /// whitespace characters preceding it.
 struct FormatToken {
-  FormatToken()
-      : NewlinesBefore(0), HasUnescapedNewline(false), LastNewlineOffset(0),
-        ColumnWidth(0), LastLineColumnWidth(0), IsMultiline(false),
-        IsFirst(false), MustBreakBefore(false), IsUnterminatedLiteral(false),
-        BlockKind(BK_Unknown), Type(TT_Unknown), SpacesRequiredBefore(0),
-        CanBreakBefore(false), ClosesTemplateDeclaration(false),
-        ParameterCount(0), BlockParameterCount(0),
-        PackingKind(PPK_Inconclusive), TotalLength(0), UnbreakableTailLength(0),
-        BindingStrength(0), NestingLevel(0), SplitPenalty(0),
-        LongestObjCSelectorName(0), FakeRParens(0),
-        StartsBinaryExpression(false), EndsBinaryExpression(false),
-        OperatorIndex(0), LastOperator(false),
-        PartOfMultiVariableDeclStmt(false), IsForEachMacro(false),
-        MatchingParen(nullptr), Previous(nullptr), Next(nullptr),
-        Decision(FD_Unformatted), Finalized(false) {}
+  FormatToken() {}
 
   /// \brief The \c Token.
   Token Tok;
@@ -125,48 +119,39 @@ struct FormatToken {
   ///
   /// This can be used to determine what the user wrote in the original code
   /// and thereby e.g. leave an empty line between two function definitions.
-  unsigned NewlinesBefore;
+  unsigned NewlinesBefore = 0;
 
   /// \brief Whether there is at least one unescaped newline before the \c
   /// Token.
-  bool HasUnescapedNewline;
+  bool HasUnescapedNewline = false;
 
   /// \brief The range of the whitespace immediately preceding the \c Token.
   SourceRange WhitespaceRange;
 
   /// \brief The offset just past the last '\n' in this token's leading
   /// whitespace (relative to \c WhiteSpaceStart). 0 if there is no '\n'.
-  unsigned LastNewlineOffset;
+  unsigned LastNewlineOffset = 0;
 
   /// \brief The width of the non-whitespace parts of the token (or its first
   /// line for multi-line tokens) in columns.
   /// We need this to correctly measure number of columns a token spans.
-  unsigned ColumnWidth;
+  unsigned ColumnWidth = 0;
 
   /// \brief Contains the width in columns of the last line of a multi-line
   /// token.
-  unsigned LastLineColumnWidth;
+  unsigned LastLineColumnWidth = 0;
 
   /// \brief Whether the token text contains newlines (escaped or not).
-  bool IsMultiline;
+  bool IsMultiline = false;
 
   /// \brief Indicates that this is the first token.
-  bool IsFirst;
+  bool IsFirst = false;
 
   /// \brief Whether there must be a line break before this token.
   ///
   /// This happens for example when a preprocessor directive ended directly
   /// before the token.
-  bool MustBreakBefore;
-
-  /// \brief Returns actual token start location without leading escaped
-  /// newlines and whitespace.
-  ///
-  /// This can be different to Tok.getLocation(), which includes leading escaped
-  /// newlines.
-  SourceLocation getStartOfNonWhitespace() const {
-    return WhitespaceRange.getEnd();
-  }
+  bool MustBreakBefore = false;
 
   /// \brief The raw text of the token.
   ///
@@ -175,69 +160,74 @@ struct FormatToken {
   StringRef TokenText;
 
   /// \brief Set to \c true if this token is an unterminated literal.
-  bool IsUnterminatedLiteral;
+  bool IsUnterminatedLiteral = 0;
 
   /// \brief Contains the kind of block if this token is a brace.
-  BraceBlockKind BlockKind;
+  BraceBlockKind BlockKind = BK_Unknown;
 
-  TokenType Type;
+  TokenType Type = TT_Unknown;
 
   /// \brief The number of spaces that should be inserted before this token.
-  unsigned SpacesRequiredBefore;
+  unsigned SpacesRequiredBefore = 0;
 
   /// \brief \c true if it is allowed to break before this token.
-  bool CanBreakBefore;
+  bool CanBreakBefore = false;
 
-  bool ClosesTemplateDeclaration;
+  /// \brief \c true if this is the ">" of "template<..>".
+  bool ClosesTemplateDeclaration = false;
 
   /// \brief Number of parameters, if this is "(", "[" or "<".
   ///
   /// This is initialized to 1 as we don't need to distinguish functions with
   /// 0 parameters from functions with 1 parameter. Thus, we can simply count
   /// the number of commas.
-  unsigned ParameterCount;
+  unsigned ParameterCount = 0;
 
   /// \brief Number of parameters that are nested blocks,
   /// if this is "(", "[" or "<".
-  unsigned BlockParameterCount;
+  unsigned BlockParameterCount = 0;
+
+  /// \brief If this is a bracket ("<", "(", "[" or "{"), contains the kind of
+  /// the surrounding bracket.
+  tok::TokenKind ParentBracket = tok::unknown;
 
   /// \brief A token can have a special role that can carry extra information
   /// about the token's formatting.
   std::unique_ptr<TokenRole> Role;
 
   /// \brief If this is an opening parenthesis, how are the parameters packed?
-  ParameterPackingKind PackingKind;
+  ParameterPackingKind PackingKind = PPK_Inconclusive;
 
   /// \brief The total length of the unwrapped line up to and including this
   /// token.
-  unsigned TotalLength;
+  unsigned TotalLength = 0;
 
   /// \brief The original 0-based column of this token, including expanded tabs.
   /// The configured TabWidth is used as tab width.
-  unsigned OriginalColumn;
+  unsigned OriginalColumn = 0;
 
   /// \brief The length of following tokens until the next natural split point,
   /// or the next token that can be broken.
-  unsigned UnbreakableTailLength;
+  unsigned UnbreakableTailLength = 0;
 
   // FIXME: Come up with a 'cleaner' concept.
   /// \brief The binding strength of a token. This is a combined value of
   /// operator precedence, parenthesis nesting, etc.
-  unsigned BindingStrength;
+  unsigned BindingStrength = 0;
 
   /// \brief The nesting level of this token, i.e. the number of surrounding (),
   /// [], {} or <>.
-  unsigned NestingLevel;
+  unsigned NestingLevel = 0;
 
   /// \brief Penalty for inserting a line break before this token.
-  unsigned SplitPenalty;
+  unsigned SplitPenalty = 0;
 
   /// \brief If this is the first ObjC selector name in an ObjC method
   /// definition or call, this contains the length of the longest name.
   ///
   /// This being set to 0 means that the selectors should not be colon-aligned,
   /// e.g. because several of them are block-type.
-  unsigned LongestObjCSelectorName;
+  unsigned LongestObjCSelectorName = 0;
 
   /// \brief Stores the number of required fake parentheses and the
   /// corresponding operator precedence.
@@ -246,29 +236,47 @@ struct FormatToken {
   /// reverse order, i.e. inner fake parenthesis first.
   SmallVector<prec::Level, 4> FakeLParens;
   /// \brief Insert this many fake ) after this token for correct indentation.
-  unsigned FakeRParens;
+  unsigned FakeRParens = 0;
 
   /// \brief \c true if this token starts a binary expression, i.e. has at least
   /// one fake l_paren with a precedence greater than prec::Unknown.
-  bool StartsBinaryExpression;
+  bool StartsBinaryExpression = false;
   /// \brief \c true if this token ends a binary expression.
-  bool EndsBinaryExpression;
+  bool EndsBinaryExpression = false;
 
   /// \brief Is this is an operator (or "."/"->") in a sequence of operators
   /// with the same precedence, contains the 0-based operator index.
-  unsigned OperatorIndex;
+  unsigned OperatorIndex = 0;
 
   /// \brief Is this the last operator (or "."/"->") in a sequence of operators
   /// with the same precedence?
-  bool LastOperator;
+  bool LastOperator = false;
 
   /// \brief Is this token part of a \c DeclStmt defining multiple variables?
   ///
   /// Only set if \c Type == \c TT_StartOfName.
-  bool PartOfMultiVariableDeclStmt;
+  bool PartOfMultiVariableDeclStmt = false;
+
+  /// \brief If this is a bracket, this points to the matching one.
+  FormatToken *MatchingParen = nullptr;
+
+  /// \brief The previous token in the unwrapped line.
+  FormatToken *Previous = nullptr;
 
-  /// \brief Is this a foreach macro?
-  bool IsForEachMacro;
+  /// \brief The next token in the unwrapped line.
+  FormatToken *Next = nullptr;
+
+  /// \brief If this token starts a block, this contains all the unwrapped lines
+  /// in it.
+  SmallVector<AnnotatedLine *, 1> Children;
+
+  /// \brief Stores the formatting decision for the token once it was made.
+  FormatDecision Decision = FD_Unformatted;
+
+  /// \brief If \c true, this token has been fully formatted (indented and
+  /// potentially re-formatted inside), and we do not allow further formatting
+  /// changes.
+  bool Finalized = false;
 
   bool is(tok::TokenKind Kind) const { return Tok.is(Kind); }
   bool is(TokenType TT) const { return Type == TT; }
@@ -278,27 +286,10 @@ struct FormatToken {
   template <typename A, typename B> bool isOneOf(A K1, B K2) const {
     return is(K1) || is(K2);
   }
-  template <typename A, typename B, typename C>
-  bool isOneOf(A K1, B K2, C K3) const {
-    return is(K1) || is(K2) || is(K3);
-  }
-  template <typename A, typename B, typename C, typename D>
-  bool isOneOf(A K1, B K2, C K3, D K4) const {
-    return is(K1) || is(K2) || is(K3) || is(K4);
+  template <typename A, typename B, typename... Ts>
+  bool isOneOf(A K1, B K2, Ts... Ks) const {
+    return is(K1) || isOneOf(K2, Ks...);
   }
-  template <typename A, typename B, typename C, typename D, typename E>
-  bool isOneOf(A K1, B K2, C K3, D K4, E K5) const {
-    return is(K1) || is(K2) || is(K3) || is(K4) || is(K5);
-  }
-  template <typename T>
-  bool isOneOf(T K1, T K2, T K3, T K4, T K5, T K6, T K7 = tok::NUM_TOKENS,
-               T K8 = tok::NUM_TOKENS, T K9 = tok::NUM_TOKENS,
-               T K10 = tok::NUM_TOKENS, T K11 = tok::NUM_TOKENS,
-               T K12 = tok::NUM_TOKENS) const {
-    return is(K1) || is(K2) || is(K3) || is(K4) || is(K5) || is(K6) || is(K7) ||
-           is(K8) || is(K9) || is(K10) || is(K11) || is(K12);
-  }
-
   template <typename T> bool isNot(T Kind) const { return !is(Kind); }
 
   bool isStringLiteral() const { return tok::isStringLiteral(Tok.getKind()); }
@@ -336,7 +327,8 @@ struct FormatToken {
   /// \brief Returns \c true if this is a "." or "->" accessing a member.
   bool isMemberAccess() const {
     return isOneOf(tok::arrow, tok::period, tok::arrowstar) &&
-           !isOneOf(TT_DesignatedInitializerPeriod, TT_TrailingReturnArrow);
+           !isOneOf(TT_DesignatedInitializerPeriod, TT_TrailingReturnArrow,
+                    TT_LambdaArrow);
   }
 
   bool isUnaryOperator() const {
@@ -385,6 +377,15 @@ struct FormatToken {
     }
   }
 
+  /// \brief Returns actual token start location without leading escaped
+  /// newlines and whitespace.
+  ///
+  /// This can be different to Tok.getLocation(), which includes leading escaped
+  /// newlines.
+  SourceLocation getStartOfNonWhitespace() const {
+    return WhitespaceRange.getEnd();
+  }
+
   prec::Level getPrecedence() const {
     return getBinOpPrecedence(Tok.getKind(), true, true);
   }
@@ -419,25 +420,10 @@ struct FormatToken {
     return MatchingParen && MatchingParen->opensBlockTypeList(Style);
   }
 
-  FormatToken *MatchingParen;
-
-  FormatToken *Previous;
-  FormatToken *Next;
-
-  SmallVector<AnnotatedLine *, 1> Children;
-
-  /// \brief Stores the formatting decision for the token once it was made.
-  FormatDecision Decision;
-
-  /// \brief If \c true, this token has been fully formatted (indented and
-  /// potentially re-formatted inside), and we do not allow further formatting
-  /// changes.
-  bool Finalized;
-
 private:
   // Disallow copying.
-  FormatToken(const FormatToken &) LLVM_DELETED_FUNCTION;
-  void operator=(const FormatToken &) LLVM_DELETED_FUNCTION;
+  FormatToken(const FormatToken &) = delete;
+  void operator=(const FormatToken &) = delete;
 };
 
 class ContinuationIndenter;
@@ -543,6 +529,7 @@ struct AdditionalKeywords {
 
     kw_finally = &IdentTable.get("finally");
     kw_function = &IdentTable.get("function");
+    kw_import = &IdentTable.get("import");
     kw_var = &IdentTable.get("var");
 
     kw_abstract = &IdentTable.get("abstract");
@@ -555,24 +542,33 @@ struct AdditionalKeywords {
     kw_package = &IdentTable.get("package");
     kw_synchronized = &IdentTable.get("synchronized");
     kw_throws = &IdentTable.get("throws");
+    kw___except = &IdentTable.get("__except");
+
+    kw_mark = &IdentTable.get("mark");
 
     kw_option = &IdentTable.get("option");
     kw_optional = &IdentTable.get("optional");
     kw_repeated = &IdentTable.get("repeated");
     kw_required = &IdentTable.get("required");
     kw_returns = &IdentTable.get("returns");
+
+    kw_signals = &IdentTable.get("signals");
+    kw_slots = &IdentTable.get("slots");
+    kw_qslots = &IdentTable.get("Q_SLOTS");
   }
 
-  // ObjC context sensitive keywords.
+  // Context sensitive keywords.
   IdentifierInfo *kw_in;
   IdentifierInfo *kw_CF_ENUM;
   IdentifierInfo *kw_CF_OPTIONS;
   IdentifierInfo *kw_NS_ENUM;
   IdentifierInfo *kw_NS_OPTIONS;
+  IdentifierInfo *kw___except;
 
   // JavaScript keywords.
   IdentifierInfo *kw_finally;
   IdentifierInfo *kw_function;
+  IdentifierInfo *kw_import;
   IdentifierInfo *kw_var;
 
   // Java keywords.
@@ -587,12 +583,20 @@ struct AdditionalKeywords {
   IdentifierInfo *kw_synchronized;
   IdentifierInfo *kw_throws;
 
+  // Pragma keywords.
+  IdentifierInfo *kw_mark;
+
   // Proto keywords.
   IdentifierInfo *kw_option;
   IdentifierInfo *kw_optional;
   IdentifierInfo *kw_repeated;
   IdentifierInfo *kw_required;
   IdentifierInfo *kw_returns;
+
+  // QT keywords.
+  IdentifierInfo *kw_signals;
+  IdentifierInfo *kw_slots;
+  IdentifierInfo *kw_qslots;
 };
 
 } // namespace format
diff --git a/lib/Format/TokenAnnotator.cpp b/lib/Format/TokenAnnotator.cpp
index 4ba3f9196977..98f5709b9062 100644
--- a/lib/Format/TokenAnnotator.cpp
+++ b/lib/Format/TokenAnnotator.cpp
@@ -15,6 +15,7 @@
 
 #include "TokenAnnotator.h"
 #include "clang/Basic/SourceManager.h"
+#include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/Support/Debug.h"
 
 #define DEBUG_TYPE "format-token-annotator"
@@ -43,8 +44,14 @@ private:
   bool parseAngle() {
     if (!CurrentToken)
       return false;
-    ScopedContextCreator ContextCreator(*this, tok::less, 10);
     FormatToken *Left = CurrentToken->Previous;
+    Left->ParentBracket = Contexts.back().ContextKind;
+    ScopedContextCreator ContextCreator(*this, tok::less, 10);
+
+    // If this angle is in the context of an expression, we need to be more
+    // hesitant to detect it as opening template parameters.
+    bool InExprContext = Contexts.back().IsExpression;
+
     Contexts.back().IsExpression = false;
     // If there's a template keyword before the opening angle bracket, this is a
     // template parameter, not an argument.
@@ -68,8 +75,8 @@ private:
         next();
         continue;
       }
-      if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace,
-                                tok::colon, tok::question))
+      if (CurrentToken->isOneOf(tok::r_paren, tok::r_square, tok::r_brace) ||
+          (CurrentToken->isOneOf(tok::colon, tok::question) && InExprContext))
         return false;
       // If a && or || is found and interpreted as a binary operator, this set
       // of angles is likely part of something like "a < b && c > d". If the
@@ -92,6 +99,8 @@ private:
   bool parseParens(bool LookForDecls = false) {
     if (!CurrentToken)
       return false;
+    FormatToken *Left = CurrentToken->Previous;
+    Left->ParentBracket = Contexts.back().ContextKind;
     ScopedContextCreator ContextCreator(*this, tok::l_paren, 1);
 
     // FIXME: This is a bit of a hack. Do better.
@@ -99,7 +108,6 @@ private:
         Contexts.size() == 2 && Contexts[0].ColonIsForRangeExpr;
 
     bool StartsObjCMethodExpr = false;
-    FormatToken *Left = CurrentToken->Previous;
     if (CurrentToken->is(tok::caret)) {
       // (^ can start a block type.
       Left->Type = TT_ObjCBlockLParen;
@@ -117,22 +125,22 @@ private:
          Left->Previous->is(TT_BinaryOperator))) {
       // static_assert, if and while usually contain expressions.
       Contexts.back().IsExpression = true;
-    } else if (Line.InPPDirective &&
-               (!Left->Previous ||
-                !Left->Previous->isOneOf(tok::identifier,
-                                         TT_OverloadedOperator))) {
-      Contexts.back().IsExpression = true;
     } else if (Left->Previous && Left->Previous->is(tok::r_square) &&
                Left->Previous->MatchingParen &&
                Left->Previous->MatchingParen->is(TT_LambdaLSquare)) {
       // This is a parameter list of a lambda expression.
       Contexts.back().IsExpression = false;
+    } else if (Line.InPPDirective &&
+               (!Left->Previous ||
+                !Left->Previous->isOneOf(tok::identifier,
+                                         TT_OverloadedOperator))) {
+      Contexts.back().IsExpression = true;
     } else if (Contexts[Contexts.size() - 2].CaretFound) {
       // This is the parameter list of an ObjC block.
       Contexts.back().IsExpression = false;
     } else if (Left->Previous && Left->Previous->is(tok::kw___attribute)) {
       Left->Type = TT_AttributeParen;
-    } else if (Left->Previous && Left->Previous->IsForEachMacro) {
+    } else if (Left->Previous && Left->Previous->is(TT_ForEachMacro)) {
       // The first argument to a foreach macro is a declaration.
       Contexts.back().IsForEachMacro = true;
       Contexts.back().IsExpression = false;
@@ -149,6 +157,8 @@ private:
     bool MightBeFunctionType = CurrentToken->is(tok::star);
     bool HasMultipleLines = false;
     bool HasMultipleParametersOnALine = false;
+    bool MightBeObjCForRangeLoop =
+        Left->Previous && Left->Previous->is(tok::kw_for);
     while (CurrentToken) {
       // LookForDecls is set when "if (" has been seen. Check for
       // 'identifier' '*' 'identifier' followed by not '=' -- this
@@ -210,7 +220,8 @@ private:
       }
       if (CurrentToken->isOneOf(tok::r_square, tok::r_brace))
         return false;
-      else if (CurrentToken->is(tok::l_brace))
+
+      if (CurrentToken->is(tok::l_brace))
         Left->Type = TT_Unknown; // Not TT_ObjCBlockLParen
       if (CurrentToken->is(tok::comma) && CurrentToken->Next &&
           !CurrentToken->Next->HasUnescapedNewline &&
@@ -219,6 +230,15 @@ private:
       if (CurrentToken->isOneOf(tok::kw_const, tok::kw_auto) ||
           CurrentToken->isSimpleTypeSpecifier())
         Contexts.back().IsExpression = false;
+      if (CurrentToken->isOneOf(tok::semi, tok::colon))
+        MightBeObjCForRangeLoop = false;
+      if (MightBeObjCForRangeLoop && CurrentToken->is(Keywords.kw_in))
+        CurrentToken->Type = TT_ObjCForIn;
+      // When we discover a 'new', we set CanBeExpression to 'false' in order to
+      // parse the type correctly. Reset that after a comma.
+      if (CurrentToken->is(tok::comma))
+        Contexts.back().CanBeExpression = true;
+
       FormatToken *Tok = CurrentToken;
       if (!consumeToken())
         return false;
@@ -237,6 +257,7 @@ private:
     // ')' or ']'), it could be the start of an Objective-C method
     // expression, or it could the the start of an Objective-C array literal.
     FormatToken *Left = CurrentToken->Previous;
+    Left->ParentBracket = Contexts.back().ContextKind;
     FormatToken *Parent = Left->getPreviousNonComment();
     bool StartsObjCMethodExpr =
         Contexts.back().CanBeExpression && Left->isNot(TT_LambdaLSquare) &&
@@ -316,6 +337,7 @@ private:
   bool parseBrace() {
     if (CurrentToken) {
       FormatToken *Left = CurrentToken->Previous;
+      Left->ParentBracket = Contexts.back().ContextKind;
 
       if (Contexts.back().CaretFound)
         Left->Type = TT_ObjCBlockLBrace;
@@ -342,7 +364,8 @@ private:
                Style.Language == FormatStyle::LK_Proto) &&
               Previous->is(tok::identifier))
             Previous->Type = TT_SelectorName;
-          if (CurrentToken->is(tok::colon))
+          if (CurrentToken->is(tok::colon) ||
+              Style.Language == FormatStyle::LK_JavaScript)
             Left->Type = TT_DictLiteral;
         }
         if (!consumeToken())
@@ -408,10 +431,18 @@ private:
       if (!Tok->Previous)
         return false;
       // Colons from ?: are handled in parseConditional().
-      if (Tok->Previous->is(tok::r_paren) && Contexts.size() == 1 &&
-          Line.First->isNot(tok::kw_case)) {
-        Tok->Type = TT_CtorInitializerColon;
-      } else if (Contexts.back().ColonIsDictLiteral) {
+      if (Style.Language == FormatStyle::LK_JavaScript) {
+        if (Contexts.back().ColonIsForRangeExpr || // colon in for loop
+            (Contexts.size() == 1 &&               // switch/case labels
+             !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) ||
+            Contexts.back().ContextKind == tok::l_paren ||  // function params
+            Contexts.back().ContextKind == tok::l_square || // array type
+            Line.MustBeDeclaration) { // method/property declaration
+          Tok->Type = TT_JsTypeColon;
+          break;
+        }
+      }
+      if (Contexts.back().ColonIsDictLiteral) {
         Tok->Type = TT_DictLiteral;
       } else if (Contexts.back().ColonIsObjCMethodExpr ||
                  Line.First->is(TT_ObjCMethodSpecifier)) {
@@ -429,7 +460,10 @@ private:
         Tok->Type = TT_BitFieldColon;
       } else if (Contexts.size() == 1 &&
                  !Line.First->isOneOf(tok::kw_enum, tok::kw_case)) {
-        Tok->Type = TT_InheritanceColon;
+        if (Tok->Previous->is(tok::r_paren))
+          Tok->Type = TT_CtorInitializerColon;
+        else
+          Tok->Type = TT_InheritanceColon;
       } else if (Tok->Previous->is(tok::identifier) && Tok->Next &&
                  Tok->Next->isOneOf(tok::r_paren, tok::comma)) {
         // This handles a special macro in ObjC code where selectors including
@@ -471,13 +505,15 @@ private:
         return false;
       break;
     case tok::less:
-      if ((!Tok->Previous ||
+      if (!NonTemplateLess.count(Tok) &&
+          (!Tok->Previous ||
            (!Tok->Previous->Tok.isLiteral() &&
             !(Tok->Previous->is(tok::r_paren) && Contexts.size() > 1))) &&
           parseAngle()) {
         Tok->Type = TT_TemplateOpener;
       } else {
         Tok->Type = TT_BinaryOperator;
+        NonTemplateLess.insert(Tok);
         CurrentToken = Tok;
         next();
       }
@@ -509,21 +545,34 @@ private:
       }
       break;
     case tok::question:
+      if (Style.Language == FormatStyle::LK_JavaScript && Tok->Next &&
+          Tok->Next->isOneOf(tok::semi, tok::colon, tok::r_paren,
+                             tok::r_brace)) {
+        // Question marks before semicolons, colons, etc. indicate optional
+        // types (fields, parameters), e.g.
+        //   function(x?: string, y?) {...}
+        //   class X { y?; }
+        Tok->Type = TT_JsTypeOptionalQuestion;
+        break;
+      }
+      // Declarations cannot be conditional expressions, this can only be part
+      // of a type declaration.
+      if (Line.MustBeDeclaration &&
+          Style.Language == FormatStyle::LK_JavaScript)
+        break;
       parseConditional();
       break;
     case tok::kw_template:
       parseTemplateDeclaration();
       break;
-    case tok::identifier:
-      if (Line.First->is(tok::kw_for) && Tok->is(Keywords.kw_in) &&
-          Tok->Previous->isNot(tok::colon))
-        Tok->Type = TT_ObjCForIn;
-      break;
     case tok::comma:
-      if (Contexts.back().FirstStartOfName)
-        Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true;
       if (Contexts.back().InCtorInitializer)
         Tok->Type = TT_CtorInitializerComma;
+      else if (Contexts.back().FirstStartOfName &&
+               (Contexts.size() == 1 || Line.First->is(tok::kw_for))) {
+        Contexts.back().FirstStartOfName->PartOfMultiVariableDeclStmt = true;
+        Line.IsMultiVariableDeclStmt = true;
+      }
       if (Contexts.back().IsForEachMacro)
         Contexts.back().IsExpression = true;
       break;
@@ -557,11 +606,14 @@ private:
 
   void parsePragma() {
     next(); // Consume "pragma".
-    if (CurrentToken && CurrentToken->TokenText == "mark") {
+    if (CurrentToken &&
+        CurrentToken->isOneOf(Keywords.kw_mark, Keywords.kw_option)) {
+      bool IsMark = CurrentToken->is(Keywords.kw_mark);
       next(); // Consume "mark".
       next(); // Consume first token (so we fix leading whitespace).
       while (CurrentToken) {
-        CurrentToken->Type = TT_ImplicitStringLiteral;
+        if (IsMark || CurrentToken->Previous->is(TT_BinaryOperator))
+          CurrentToken->Type = TT_ImplicitStringLiteral;
         next();
       }
     }
@@ -582,6 +634,7 @@ private:
       return Type;
     switch (CurrentToken->Tok.getIdentifierInfo()->getPPKeywordID()) {
     case tok::pp_include:
+    case tok::pp_include_next:
     case tok::pp_import:
       next();
       parseIncludeDirective();
@@ -609,9 +662,9 @@ private:
 
 public:
   LineType parseLine() {
-    if (CurrentToken->is(tok::hash)) {
+    NonTemplateLess.clear();
+    if (CurrentToken->is(tok::hash))
       return parsePreprocessorDirective();
-    }
 
     // Directly allow to 'import <string-literal>' to support protocol buffer
     // definitions (code.google.com/p/protobuf) or missing "#" (either way we
@@ -635,6 +688,15 @@ public:
       return LT_ImportStatement;
     }
 
+    // In .proto files, top-level options are very similar to import statements
+    // and should not be line-wrapped.
+    if (Style.Language == FormatStyle::LK_Proto && Line.Level == 0 &&
+        CurrentToken->is(Keywords.kw_option)) {
+      next();
+      if (CurrentToken && CurrentToken->is(tok::identifier))
+        return LT_ImportStatement;
+    }
+
     bool KeywordVirtualFound = false;
     bool ImportStatement = false;
     while (CurrentToken) {
@@ -678,11 +740,13 @@ private:
 
     // Reset token type in case we have already looked at it and then
     // recovered from an error (e.g. failure to find the matching >).
-    if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_FunctionLBrace,
-                               TT_ImplicitStringLiteral, TT_RegexLiteral,
+    if (!CurrentToken->isOneOf(TT_LambdaLSquare, TT_ForEachMacro,
+                               TT_FunctionLBrace, TT_ImplicitStringLiteral,
+                               TT_InlineASMBrace, TT_RegexLiteral,
                                TT_TrailingReturnArrow))
       CurrentToken->Type = TT_Unknown;
     CurrentToken->Role.reset();
+    CurrentToken->MatchingParen = nullptr;
     CurrentToken->FakeLParens.clear();
     CurrentToken->FakeRParens = 0;
   }
@@ -705,27 +769,22 @@ private:
     Context(tok::TokenKind ContextKind, unsigned BindingStrength,
             bool IsExpression)
         : ContextKind(ContextKind), BindingStrength(BindingStrength),
-          LongestObjCSelectorName(0), ColonIsForRangeExpr(false),
-          ColonIsDictLiteral(false), ColonIsObjCMethodExpr(false),
-          FirstObjCSelectorName(nullptr), FirstStartOfName(nullptr),
-          IsExpression(IsExpression), CanBeExpression(true),
-          InTemplateArgument(false), InCtorInitializer(false),
-          CaretFound(false), IsForEachMacro(false) {}
+          IsExpression(IsExpression) {}
 
     tok::TokenKind ContextKind;
     unsigned BindingStrength;
-    unsigned LongestObjCSelectorName;
-    bool ColonIsForRangeExpr;
-    bool ColonIsDictLiteral;
-    bool ColonIsObjCMethodExpr;
-    FormatToken *FirstObjCSelectorName;
-    FormatToken *FirstStartOfName;
     bool IsExpression;
-    bool CanBeExpression;
-    bool InTemplateArgument;
-    bool InCtorInitializer;
-    bool CaretFound;
-    bool IsForEachMacro;
+    unsigned LongestObjCSelectorName = 0;
+    bool ColonIsForRangeExpr = false;
+    bool ColonIsDictLiteral = false;
+    bool ColonIsObjCMethodExpr = false;
+    FormatToken *FirstObjCSelectorName = nullptr;
+    FormatToken *FirstStartOfName = nullptr;
+    bool CanBeExpression = true;
+    bool InTemplateArgument = false;
+    bool InCtorInitializer = false;
+    bool CaretFound = false;
+    bool IsForEachMacro = false;
   };
 
   /// \brief Puts a new \c Context onto the stack \c Contexts for the lifetime
@@ -746,23 +805,29 @@ private:
 
   void modifyContext(const FormatToken &Current) {
     if (Current.getPrecedence() == prec::Assignment &&
-        !Line.First->isOneOf(tok::kw_template, tok::kw_using,
-                             TT_UnaryOperator) &&
+        !Line.First->isOneOf(tok::kw_template, tok::kw_using) &&
         (!Current.Previous || Current.Previous->isNot(tok::kw_operator))) {
       Contexts.back().IsExpression = true;
-      for (FormatToken *Previous = Current.Previous;
-           Previous && !Previous->isOneOf(tok::comma, tok::semi);
-           Previous = Previous->Previous) {
-        if (Previous->isOneOf(tok::r_square, tok::r_paren)) {
-          Previous = Previous->MatchingParen;
-          if (!Previous)
+      if (!Line.First->is(TT_UnaryOperator)) {
+        for (FormatToken *Previous = Current.Previous;
+             Previous && !Previous->isOneOf(tok::comma, tok::semi);
+             Previous = Previous->Previous) {
+          if (Previous->isOneOf(tok::r_square, tok::r_paren)) {
+            Previous = Previous->MatchingParen;
+            if (!Previous)
+              break;
+          }
+          if (Previous->opensScope())
             break;
+          if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) &&
+              Previous->isOneOf(tok::star, tok::amp, tok::ampamp) &&
+              Previous->Previous && Previous->Previous->isNot(tok::equal))
+            Previous->Type = TT_PointerOrReference;
         }
-        if (Previous->isOneOf(TT_BinaryOperator, TT_UnaryOperator) &&
-            Previous->isOneOf(tok::star, tok::amp) && Previous->Previous &&
-            Previous->Previous->isNot(tok::equal))
-          Previous->Type = TT_PointerOrReference;
       }
+    } else if (Current.is(tok::lessless) &&
+               (!Current.Previous || !Current.Previous->is(tok::kw_operator))) {
+      Contexts.back().IsExpression = true;
     } else if (Current.isOneOf(tok::kw_return, tok::kw_throw)) {
       Contexts.back().IsExpression = true;
     } else if (Current.is(TT_TrailingReturnArrow)) {
@@ -833,30 +898,56 @@ private:
     } else if (Current.isOneOf(tok::exclaim, tok::tilde)) {
       Current.Type = TT_UnaryOperator;
     } else if (Current.is(tok::question)) {
-      Current.Type = TT_ConditionalExpr;
+      if (Style.Language == FormatStyle::LK_JavaScript &&
+          Line.MustBeDeclaration) {
+        // In JavaScript, `interface X { foo?(): bar; }` is an optional method
+        // on the interface, not a ternary expression.
+        Current.Type = TT_JsTypeOptionalQuestion;
+      } else {
+        Current.Type = TT_ConditionalExpr;
+      }
     } else if (Current.isBinaryOperator() &&
                (!Current.Previous || Current.Previous->isNot(tok::l_square))) {
       Current.Type = TT_BinaryOperator;
     } else if (Current.is(tok::comment)) {
-      if (Current.TokenText.startswith("//"))
+      if (Current.TokenText.startswith("/*")) {
+        if (Current.TokenText.endswith("*/"))
+          Current.Type = TT_BlockComment;
+        else
+          // The lexer has for some reason determined a comment here. But we
+          // cannot really handle it, if it isn't properly terminated.
+          Current.Tok.setKind(tok::unknown);
+      } else {
         Current.Type = TT_LineComment;
-      else
-        Current.Type = TT_BlockComment;
+      }
     } else if (Current.is(tok::r_paren)) {
       if (rParenEndsCast(Current))
         Current.Type = TT_CastRParen;
+      if (Current.MatchingParen && Current.Next &&
+          !Current.Next->isBinaryOperator() &&
+          !Current.Next->isOneOf(tok::semi, tok::colon, tok::l_brace))
+        if (FormatToken *BeforeParen = Current.MatchingParen->Previous)
+          if (BeforeParen->is(tok::identifier) &&
+              BeforeParen->TokenText == BeforeParen->TokenText.upper() &&
+              (!BeforeParen->Previous ||
+               BeforeParen->Previous->ClosesTemplateDeclaration))
+            Current.Type = TT_FunctionAnnotationRParen;
     } else if (Current.is(tok::at) && Current.Next) {
-      switch (Current.Next->Tok.getObjCKeywordID()) {
-      case tok::objc_interface:
-      case tok::objc_implementation:
-      case tok::objc_protocol:
-        Current.Type = TT_ObjCDecl;
-        break;
-      case tok::objc_property:
-        Current.Type = TT_ObjCProperty;
-        break;
-      default:
-        break;
+      if (Current.Next->isStringLiteral()) {
+        Current.Type = TT_ObjCStringLiteral;
+      } else {
+        switch (Current.Next->Tok.getObjCKeywordID()) {
+        case tok::objc_interface:
+        case tok::objc_implementation:
+        case tok::objc_protocol:
+          Current.Type = TT_ObjCDecl;
+          break;
+        case tok::objc_property:
+          Current.Type = TT_ObjCProperty;
+          break;
+        default:
+          break;
+        }
       }
     } else if (Current.is(tok::period)) {
       FormatToken *PreviousNoComment = Current.getPreviousNonComment();
@@ -875,7 +966,9 @@ private:
       // Line.MightBeFunctionDecl can only be true after the parentheses of a
       // function declaration have been found.
       Current.Type = TT_TrailingAnnotation;
-    } else if (Style.Language == FormatStyle::LK_Java && Current.Previous) {
+    } else if ((Style.Language == FormatStyle::LK_Java ||
+                Style.Language == FormatStyle::LK_JavaScript) &&
+               Current.Previous) {
       if (Current.Previous->is(tok::at) &&
           Current.isNot(Keywords.kw_interface)) {
         const FormatToken &AtToken = *Current.Previous;
@@ -902,7 +995,7 @@ private:
       return false;
 
     if (Tok.Previous->is(TT_LeadingJavaAnnotation))
-        return false;
+      return false;
 
     // Skip "const" as it does not have an influence on whether this is a name.
     FormatToken *PreviousNotConst = Tok.Previous;
@@ -964,8 +1057,7 @@ private:
     bool IsSizeOfOrAlignOf =
         LeftOfParens && LeftOfParens->isOneOf(tok::kw_sizeof, tok::kw_alignof);
     if (ParensAreType && !ParensCouldEndDecl && !IsSizeOfOrAlignOf &&
-        ((Contexts.size() > 1 && Contexts[Contexts.size() - 2].IsExpression) ||
-         (Tok.Next && Tok.Next->isBinaryOperator())))
+        (Contexts.size() > 1 && Contexts[Contexts.size() - 2].IsExpression))
       IsCast = true;
     else if (Tok.Next && Tok.Next->isNot(tok::string_literal) &&
              (Tok.Next->Tok.isLiteral() ||
@@ -995,7 +1087,8 @@ private:
         }
 
         for (; Prev != Tok.MatchingParen; Prev = Prev->Previous) {
-          if (!Prev || !Prev->isOneOf(tok::kw_const, tok::identifier)) {
+          if (!Prev ||
+              !Prev->isOneOf(tok::kw_const, tok::identifier, tok::coloncolon)) {
             IsCast = false;
             break;
           }
@@ -1032,7 +1125,7 @@ private:
 
     if (NextToken->is(tok::l_square) && NextToken->isNot(TT_LambdaLSquare))
       return TT_PointerOrReference;
-    if (NextToken->isOneOf(tok::kw_operator, tok::comma))
+    if (NextToken->isOneOf(tok::kw_operator, tok::comma, tok::semi))
       return TT_PointerOrReference;
 
     if (PrevToken->is(tok::r_paren) && PrevToken->MatchingParen &&
@@ -1108,10 +1201,16 @@ private:
   FormatToken *CurrentToken;
   bool AutoFound;
   const AdditionalKeywords &Keywords;
+
+  // Set of "<" tokens that do not open a template parameter list. If parseAngle
+  // determines that a specific token can't be a template opener, it will make
+  // same decision irrespective of the decisions for tokens leading up to it.
+  // Store this information to prevent this from causing exponential runtime.
+  llvm::SmallPtrSet<FormatToken *, 16> NonTemplateLess;
 };
 
-static int PrecedenceUnaryOperator = prec::PointerToMember + 1;
-static int PrecedenceArrowAndPeriod = prec::PointerToMember + 2;
+static const int PrecedenceUnaryOperator = prec::PointerToMember + 1;
+static const int PrecedenceArrowAndPeriod = prec::PointerToMember + 2;
 
 /// \brief Parses binary expressions by inserting fake parenthesis based on
 /// operator precedence.
@@ -1361,12 +1460,13 @@ static bool isFunctionDeclarationName(const FormatToken &Current) {
   assert(Next->is(tok::l_paren));
   if (Next->Next == Next->MatchingParen)
     return true;
-  for (const FormatToken *Tok = Next->Next; Tok != Next->MatchingParen;
+  for (const FormatToken *Tok = Next->Next; Tok && Tok != Next->MatchingParen;
        Tok = Tok->Next) {
     if (Tok->is(tok::kw_const) || Tok->isSimpleTypeSpecifier() ||
         Tok->isOneOf(TT_PointerOrReference, TT_StartOfName))
       return true;
-    if (Tok->isOneOf(tok::l_brace, tok::string_literal) || Tok->Tok.isLiteral())
+    if (Tok->isOneOf(tok::l_brace, tok::string_literal, TT_ObjCMethodExpr) ||
+        Tok->Tok.isLiteral())
       return false;
   }
   return false;
@@ -1502,7 +1602,7 @@ unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
     if (Left.is(tok::comma) && Left.NestingLevel == 0)
       return 3;
   } else if (Style.Language == FormatStyle::LK_JavaScript) {
-    if (Right.is(Keywords.kw_function))
+    if (Right.is(Keywords.kw_function) && Left.isNot(tok::comma))
       return 100;
   }
 
@@ -1512,6 +1612,9 @@ unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
   if (Right.is(tok::l_square)) {
     if (Style.Language == FormatStyle::LK_Proto)
       return 1;
+    // Slightly prefer formatting local lambda definitions like functions.
+    if (Right.is(TT_LambdaLSquare) && Left.is(tok::equal))
+      return 50;
     if (!Right.isOneOf(TT_ObjCMethodExpr, TT_LambdaLSquare))
       return 500;
   }
@@ -1521,11 +1624,15 @@ unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
     if (Line.First->is(tok::kw_for) && Right.PartOfMultiVariableDeclStmt)
       return 3;
     if (Left.is(TT_StartOfName))
-      return 20;
+      return 110;
     if (InFunctionDecl && Right.NestingLevel == 0)
       return Style.PenaltyReturnTypeOnItsOwnLine;
     return 200;
   }
+  if (Right.is(TT_PointerOrReference))
+    return 190;
+  if (Right.is(TT_TrailingReturnArrow))
+    return 110;
   if (Left.is(tok::equal) && Right.is(tok::l_brace))
     return 150;
   if (Left.is(TT_CastRParen))
@@ -1575,6 +1682,8 @@ unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
 
   if (Left.is(tok::l_paren) && InFunctionDecl && Style.AlignAfterOpenBracket)
     return 100;
+  if (Left.is(tok::l_paren) && Left.Previous && Left.Previous->is(tok::kw_if))
+    return 1000;
   if (Left.is(tok::equal) && InFunctionDecl)
     return 110;
   if (Right.is(tok::r_brace))
@@ -1591,7 +1700,8 @@ unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
     return 50;
 
   if (Right.is(tok::lessless)) {
-    if (Left.is(tok::string_literal)) {
+    if (Left.is(tok::string_literal) &&
+        (!Right.LastOperator || Right.OperatorIndex != 1)) {
       StringRef Content = Left.TokenText;
       if (Content.startswith("\""))
         Content = Content.drop_front(1);
@@ -1607,7 +1717,9 @@ unsigned TokenAnnotator::splitPenalty(const AnnotatedLine &Line,
   if (Left.is(TT_ConditionalExpr))
     return prec::Conditional;
   prec::Level Level = Left.getPrecedence();
-
+  if (Level != prec::Unknown)
+    return Level;
+  Level = Right.getPrecedence();
   if (Level != prec::Unknown)
     return Level;
 
@@ -1636,7 +1748,7 @@ bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
   if (Right.isOneOf(tok::semi, tok::comma))
     return false;
   if (Right.is(tok::less) &&
-      (Left.isOneOf(tok::kw_template, tok::r_paren) ||
+      (Left.is(tok::kw_template) ||
        (Line.Type == LT_ObjCDecl && Style.ObjCSpaceBeforeProtocolList)))
     return true;
   if (Left.isOneOf(tok::exclaim, tok::tilde))
@@ -1655,17 +1767,27 @@ bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
   if (Left.is(tok::l_square) && Right.is(tok::amp))
     return false;
   if (Right.is(TT_PointerOrReference))
-    return Left.Tok.isLiteral() ||
-           (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) &&
-            Style.PointerAlignment != FormatStyle::PAS_Left);
+    return !(Left.is(tok::r_paren) && Left.MatchingParen &&
+             (Left.MatchingParen->is(TT_OverloadedOperatorLParen) ||
+              (Left.MatchingParen->Previous &&
+               Left.MatchingParen->Previous->is(
+                   TT_FunctionDeclarationName)))) &&
+           (Left.Tok.isLiteral() ||
+            (!Left.isOneOf(TT_PointerOrReference, tok::l_paren) &&
+             (Style.PointerAlignment != FormatStyle::PAS_Left ||
+              Line.IsMultiVariableDeclStmt)));
   if (Right.is(TT_FunctionTypeLParen) && Left.isNot(tok::l_paren) &&
       (!Left.is(TT_PointerOrReference) ||
-       Style.PointerAlignment != FormatStyle::PAS_Right))
+       (Style.PointerAlignment != FormatStyle::PAS_Right &&
+        !Line.IsMultiVariableDeclStmt)))
     return true;
   if (Left.is(TT_PointerOrReference))
     return Right.Tok.isLiteral() || Right.is(TT_BlockComment) ||
-           (!Right.isOneOf(TT_PointerOrReference, tok::l_paren) &&
-            Style.PointerAlignment != FormatStyle::PAS_Right && Left.Previous &&
+           (!Right.isOneOf(TT_PointerOrReference, TT_ArraySubscriptLSquare,
+                           tok::l_paren) &&
+            (Style.PointerAlignment != FormatStyle::PAS_Right &&
+             !Line.IsMultiVariableDeclStmt) &&
+            Left.Previous &&
             !Left.Previous->isOneOf(tok::l_paren, tok::coloncolon));
   if (Right.is(tok::star) && Left.is(tok::l_paren))
     return false;
@@ -1700,13 +1822,12 @@ bool TokenAnnotator::spaceRequiredBetween(const AnnotatedLine &Line,
     return Line.Type == LT_ObjCDecl || Left.is(tok::semi) ||
            (Style.SpaceBeforeParens != FormatStyle::SBPO_Never &&
             (Left.isOneOf(tok::kw_if, tok::kw_for, tok::kw_while,
-                          tok::kw_switch, tok::kw_case) ||
-             (Left.isOneOf(tok::kw_try, tok::kw_catch, tok::kw_new,
-                           tok::kw_delete) &&
-              (!Left.Previous || Left.Previous->isNot(tok::period))) ||
-             Left.IsForEachMacro)) ||
+                          tok::kw_switch, tok::kw_case, TT_ForEachMacro) ||
+             (Left.isOneOf(tok::kw_try, Keywords.kw___except, tok::kw_catch,
+                           tok::kw_new, tok::kw_delete) &&
+              (!Left.Previous || Left.Previous->isNot(tok::period))))) ||
            (Style.SpaceBeforeParens == FormatStyle::SBPO_Always &&
-            (Left.is(tok::identifier) || Left.isFunctionLikeKeyword()) &&
+            (Left.is(tok::identifier) || Left.isFunctionLikeKeyword() || Left.is(tok::r_paren)) &&
             Line.Type != LT_PreprocessorDirective);
   }
   if (Left.is(tok::at) && Right.Tok.getObjCKeywordID() != tok::objc_not_keyword)
@@ -1748,6 +1869,20 @@ bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
   } else if (Style.Language == FormatStyle::LK_JavaScript) {
     if (Left.is(Keywords.kw_var))
       return true;
+    if (Right.isOneOf(TT_JsTypeColon, TT_JsTypeOptionalQuestion))
+      return false;
+    if ((Left.is(tok::l_brace) || Right.is(tok::r_brace)) &&
+        Line.First->isOneOf(Keywords.kw_import, tok::kw_export))
+      return false;
+    if (Left.is(tok::ellipsis))
+      return false;
+    if (Left.is(TT_TemplateCloser) &&
+        !Right.isOneOf(tok::equal, tok::l_brace, tok::comma, tok::l_square,
+                       Keywords.kw_implements, Keywords.kw_extends))
+      // Type assertions ('<type>expr') are not followed by whitespace. Other
+      // locations that should have whitespace following are identified by the
+      // above set of follower tokens.
+      return false;
   } else if (Style.Language == FormatStyle::LK_Java) {
     if (Left.is(tok::r_square) && Right.is(tok::l_brace))
       return true;
@@ -1789,16 +1924,29 @@ bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
     return Right.is(tok::coloncolon);
   if (Right.is(TT_OverloadedOperatorLParen))
     return false;
-  if (Right.is(tok::colon))
-    return !Line.First->isOneOf(tok::kw_case, tok::kw_default) &&
-           Right.getNextNonComment() && Right.isNot(TT_ObjCMethodExpr) &&
-           !Left.is(tok::question) &&
-           !(Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon)) &&
-           (Right.isNot(TT_DictLiteral) || Style.SpacesInContainerLiterals);
+  if (Right.is(tok::colon)) {
+    if (Line.First->isOneOf(tok::kw_case, tok::kw_default) ||
+        !Right.getNextNonComment() || Right.getNextNonComment()->is(tok::semi))
+      return false;
+    if (Right.is(TT_ObjCMethodExpr))
+      return false;
+    if (Left.is(tok::question))
+      return false;
+    if (Right.is(TT_InlineASMColon) && Left.is(tok::coloncolon))
+      return false;
+    if (Right.is(TT_DictLiteral))
+      return Style.SpacesInContainerLiterals;
+    return true;
+  }
   if (Left.is(TT_UnaryOperator))
     return Right.is(TT_BinaryOperator);
+
+  // If the next token is a binary operator or a selector name, we have
+  // incorrectly classified the parenthesis as a cast. FIXME: Detect correctly.
   if (Left.is(TT_CastRParen))
-    return Style.SpaceAfterCStyleCast || Right.is(TT_BinaryOperator);
+    return Style.SpaceAfterCStyleCast ||
+           Right.isOneOf(TT_BinaryOperator, TT_SelectorName);
+
   if (Left.is(tok::greater) && Right.is(tok::greater)) {
     return Right.is(TT_TemplateCloser) && Left.is(TT_TemplateCloser) &&
            (Style.Standard != FormatStyle::LS_Cpp11 || Style.SpacesInAngles);
@@ -1819,7 +1967,8 @@ bool TokenAnnotator::spaceRequiredBefore(const AnnotatedLine &Line,
   if ((Right.is(TT_BinaryOperator) && !Left.is(tok::l_paren)) ||
       Left.isOneOf(TT_BinaryOperator, TT_ConditionalExpr))
     return true;
-  if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren))
+  if (Left.is(TT_TemplateCloser) && Right.is(tok::l_paren) &&
+      Right.isNot(TT_FunctionTypeLParen))
     return Style.SpaceBeforeParens == FormatStyle::SBPO_Always;
   if (Right.is(TT_TemplateOpener) && Left.is(tok::r_paren) &&
       Left.MatchingParen && Left.MatchingParen->is(TT_OverloadedOperatorLParen))
@@ -1850,9 +1999,12 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
   // intention is to insert a line break after it in order to make shuffling
   // around entries easier.
   const FormatToken *BeforeClosingBrace = nullptr;
-  if (Left.is(tok::l_brace) && Left.BlockKind != BK_Block && Left.MatchingParen)
+  if (Left.isOneOf(tok::l_brace, TT_ArrayInitializerLSquare) &&
+      Left.BlockKind != BK_Block && Left.MatchingParen)
     BeforeClosingBrace = Left.MatchingParen->Previous;
-  else if (Right.is(tok::r_brace) && Right.BlockKind != BK_Block)
+  else if (Right.MatchingParen &&
+           Right.MatchingParen->isOneOf(tok::l_brace,
+                                        TT_ArrayInitializerLSquare))
     BeforeClosingBrace = &Left;
   if (BeforeClosingBrace && (BeforeClosingBrace->is(tok::comma) ||
                              BeforeClosingBrace->isTrailingComment()))
@@ -1862,8 +2014,10 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
     return Left.BlockKind != BK_BracedInit &&
            Left.isNot(TT_CtorInitializerColon) &&
            (Right.NewlinesBefore > 0 && Right.HasUnescapedNewline);
-  if (Right.Previous->isTrailingComment() ||
-      (Right.isStringLiteral() && Right.Previous->isStringLiteral()))
+  if (Left.isTrailingComment())
+   return true;
+  if (Left.isStringLiteral() &&
+      (Right.isStringLiteral() || Right.is(TT_ObjCStringLiteral)))
     return true;
   if (Right.Previous->IsUnterminatedLiteral)
     return true;
@@ -1889,6 +2043,8 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
       Style.Language == FormatStyle::LK_Proto)
     // Don't put enums onto single lines in protocol buffers.
     return true;
+  if (Right.is(TT_InlineASMBrace))
+    return Right.HasUnescapedNewline;
   if (Style.Language == FormatStyle::LK_JavaScript && Right.is(tok::r_brace) &&
       Left.is(tok::l_brace) && !Left.Children.empty())
     // Support AllowShortFunctionsOnASingleLine for JavaScript.
@@ -1903,8 +2059,12 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
     return true;
   if (Left.is(TT_ObjCBlockLBrace) && !Style.AllowShortBlocksOnASingleLine)
     return true;
-  if (Right.is(tok::lessless) && Left.is(tok::identifier) &&
-      Left.TokenText == "endl")
+
+  if ((Style.Language == FormatStyle::LK_Java ||
+       Style.Language == FormatStyle::LK_JavaScript) &&
+      Left.is(TT_LeadingJavaAnnotation) &&
+      Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) &&
+      Line.Last->is(tok::l_brace))
     return true;
 
   if (Style.Language == FormatStyle::LK_JavaScript) {
@@ -1913,13 +2073,15 @@ bool TokenAnnotator::mustBreakBefore(const AnnotatedLine &Line,
         Left.Previous->is(tok::char_constant))
       return true;
     if (Left.is(TT_DictLiteral) && Left.is(tok::l_brace) &&
-        Left.NestingLevel == 0)
+        Left.NestingLevel == 0 && Left.Previous &&
+        Left.Previous->is(tok::equal) &&
+        Line.First->isOneOf(tok::identifier, Keywords.kw_import,
+                            tok::kw_export) &&
+        // kw_var is a pseudo-token that's a tok::identifier, so matches above.
+        !Line.First->is(Keywords.kw_var))
+      // Enum style object literal.
       return true;
   } else if (Style.Language == FormatStyle::LK_Java) {
-    if (Left.is(TT_LeadingJavaAnnotation) &&
-        Right.isNot(TT_LeadingJavaAnnotation) && Right.isNot(tok::l_paren) &&
-        Line.Last->is(tok::l_brace))
-      return true;
     if (Right.is(tok::plus) && Left.is(tok::string_literal) && Right.Next &&
         Right.Next->is(tok::string_literal))
       return true;
@@ -1947,9 +2109,15 @@ bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
     return false;
   if (Left.isOneOf(TT_JavaAnnotation, TT_LeadingJavaAnnotation))
     return !Right.is(tok::l_paren);
+  if (Right.is(TT_PointerOrReference))
+    return Line.IsMultiVariableDeclStmt ||
+           (Style.PointerAlignment == FormatStyle::PAS_Right &&
+            (!Right.Next || Right.Next->isNot(TT_FunctionDeclarationName)));
   if (Right.isOneOf(TT_StartOfName, TT_FunctionDeclarationName) ||
       Right.is(tok::kw_operator))
     return true;
+  if (Left.is(TT_PointerOrReference))
+    return false;
   if (Right.isTrailingComment())
     // We rely on MustBreakBefore being set correctly here as we should not
     // change the "binding" behavior of a comment.
@@ -1970,8 +2138,9 @@ bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
     return false;
   if (Left.is(tok::colon) && (Left.isOneOf(TT_DictLiteral, TT_ObjCMethodExpr)))
     return true;
-  if (Right.is(TT_SelectorName))
-    return true;
+  if (Right.is(TT_SelectorName) || (Right.is(tok::identifier) && Right.Next &&
+                                    Right.Next->is(TT_ObjCMethodExpr)))
+    return Left.isNot(tok::period); // FIXME: Properly parse ObjC calls.
   if (Left.is(tok::r_paren) && Line.Type == LT_ObjCProperty)
     return true;
   if (Left.ClosesTemplateDeclaration)
@@ -1983,17 +2152,16 @@ bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
     return true;
   if (Right.is(TT_RangeBasedForLoopColon))
     return false;
-  if (Left.isOneOf(TT_PointerOrReference, TT_TemplateCloser,
-                   TT_UnaryOperator) ||
+  if (Left.isOneOf(TT_TemplateCloser, TT_UnaryOperator) ||
       Left.is(tok::kw_operator))
     return false;
-  if (Left.is(tok::equal) && Line.Type == LT_VirtualFunctionDecl)
+  if (Left.is(tok::equal) && !Right.isOneOf(tok::kw_default, tok::kw_delete) &&
+      Line.Type == LT_VirtualFunctionDecl)
     return false;
   if (Left.is(tok::l_paren) && Left.is(TT_AttributeParen))
     return false;
   if (Left.is(tok::l_paren) && Left.Previous &&
-      (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen) ||
-       Left.Previous->is(tok::kw_if)))
+      (Left.Previous->isOneOf(TT_BinaryOperator, TT_CastRParen)))
     return false;
   if (Right.is(TT_ImplicitStringLiteral))
     return false;
@@ -2027,8 +2195,8 @@ bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
   if (Right.is(TT_CtorInitializerComma) &&
       Style.BreakConstructorInitializersBeforeComma)
     return true;
-  if (Left.is(tok::greater) && Right.is(tok::greater) &&
-      Left.isNot(TT_TemplateCloser))
+  if ((Left.is(tok::greater) && Right.is(tok::greater)) ||
+      (Left.is(tok::less) && Right.is(tok::less)))
     return false;
   if (Right.is(TT_BinaryOperator) &&
       Style.BreakBeforeBinaryOperators != FormatStyle::BOS_None &&
@@ -2046,8 +2214,9 @@ bool TokenAnnotator::canBreakBefore(const AnnotatedLine &Line,
     return true;
   return Left.isOneOf(tok::comma, tok::coloncolon, tok::semi, tok::l_brace,
                       tok::kw_class, tok::kw_struct) ||
-         Right.isMemberAccess() || Right.is(TT_TrailingReturnArrow) ||
-         Right.isOneOf(tok::lessless, tok::colon, tok::l_square, tok::at) ||
+         Right.isMemberAccess() ||
+         Right.isOneOf(TT_TrailingReturnArrow, TT_LambdaArrow, tok::lessless,
+                       tok::colon, tok::l_square, tok::at) ||
          (Left.is(tok::r_paren) &&
           Right.isOneOf(tok::identifier, tok::kw_const)) ||
          (Left.is(tok::l_paren) && !Right.is(tok::r_paren));
diff --git a/lib/Format/TokenAnnotator.h b/lib/Format/TokenAnnotator.h
index ff8e32a56afc..a948cdb1c419 100644
--- a/lib/Format/TokenAnnotator.h
+++ b/lib/Format/TokenAnnotator.h
@@ -42,8 +42,8 @@ public:
       : First(Line.Tokens.front().Tok), Level(Line.Level),
         InPPDirective(Line.InPPDirective),
         MustBeDeclaration(Line.MustBeDeclaration), MightBeFunctionDecl(false),
-        Affected(false), LeadingEmptyLinesAffected(false),
-        ChildrenAffected(false) {
+        IsMultiVariableDeclStmt(false), Affected(false),
+        LeadingEmptyLinesAffected(false), ChildrenAffected(false) {
     assert(!Line.Tokens.empty());
 
     // Calculate Next and Previous for all tokens. Note that we must overwrite
@@ -59,11 +59,8 @@ public:
       I->Tok->Previous = Current;
       Current = Current->Next;
       Current->Children.clear();
-      for (SmallVectorImpl<UnwrappedLine>::const_iterator
-               I = Node.Children.begin(),
-               E = Node.Children.end();
-           I != E; ++I) {
-        Children.push_back(new AnnotatedLine(*I));
+      for (const auto& Child : Node.Children) {
+        Children.push_back(new AnnotatedLine(Child));
         Current->Children.push_back(Children.back());
       }
     }
@@ -75,6 +72,12 @@ public:
     for (unsigned i = 0, e = Children.size(); i != e; ++i) {
       delete Children[i];
     }
+    FormatToken *Current = First;
+    while (Current) {
+      Current->Children.clear();
+      Current->Role.reset();
+      Current = Current->Next;
+    }
   }
 
   FormatToken *First;
@@ -87,6 +90,7 @@ public:
   bool InPPDirective;
   bool MustBeDeclaration;
   bool MightBeFunctionDecl;
+  bool IsMultiVariableDeclStmt;
 
   /// \c True if this line should be formatted, i.e. intersects directly or
   /// indirectly with one of the input ranges.
@@ -101,8 +105,8 @@ public:
 
 private:
   // Disallow copying.
-  AnnotatedLine(const AnnotatedLine &) LLVM_DELETED_FUNCTION;
-  void operator=(const AnnotatedLine &) LLVM_DELETED_FUNCTION;
+  AnnotatedLine(const AnnotatedLine &) = delete;
+  void operator=(const AnnotatedLine &) = delete;
 };
 
 /// \brief Determines extra information about the tokens comprising an
diff --git a/lib/Format/UnwrappedLineFormatter.cpp b/lib/Format/UnwrappedLineFormatter.cpp
index ca66e7351641..cbf8c6c92211 100644
--- a/lib/Format/UnwrappedLineFormatter.cpp
+++ b/lib/Format/UnwrappedLineFormatter.cpp
@@ -25,19 +25,152 @@ bool startsExternCBlock(const AnnotatedLine &Line) {
          NextNext && NextNext->is(tok::l_brace);
 }
 
+/// \brief Tracks the indent level of \c AnnotatedLines across levels.
+///
+/// \c nextLine must be called for each \c AnnotatedLine, after which \c
+/// getIndent() will return the indent for the last line \c nextLine was called
+/// with.
+/// If the line is not formatted (and thus the indent does not change), calling
+/// \c adjustToUnmodifiedLine after the call to \c nextLine will cause
+/// subsequent lines on the same level to be indented at the same level as the
+/// given line.
+class LevelIndentTracker {
+public:
+  LevelIndentTracker(const FormatStyle &Style,
+                     const AdditionalKeywords &Keywords, unsigned StartLevel,
+                     int AdditionalIndent)
+      : Style(Style), Keywords(Keywords), AdditionalIndent(AdditionalIndent) {
+    for (unsigned i = 0; i != StartLevel; ++i)
+      IndentForLevel.push_back(Style.IndentWidth * i + AdditionalIndent);
+  }
+
+  /// \brief Returns the indent for the current line.
+  unsigned getIndent() const { return Indent; }
+
+  /// \brief Update the indent state given that \p Line is going to be formatted
+  /// next.
+  void nextLine(const AnnotatedLine &Line) {
+    Offset = getIndentOffset(*Line.First);
+    if (Line.InPPDirective) {
+      Indent = Line.Level * Style.IndentWidth + AdditionalIndent;
+    } else {
+      while (IndentForLevel.size() <= Line.Level)
+        IndentForLevel.push_back(-1);
+      IndentForLevel.resize(Line.Level + 1);
+      Indent = getIndent(IndentForLevel, Line.Level);
+    }
+    if (static_cast<int>(Indent) + Offset >= 0)
+      Indent += Offset;
+  }
+
+  /// \brief Update the level indent to adapt to the given \p Line.
+  ///
+  /// When a line is not formatted, we move the subsequent lines on the same
+  /// level to the same indent.
+  /// Note that \c nextLine must have been called before this method.
+  void adjustToUnmodifiedLine(const AnnotatedLine &Line) {
+    unsigned LevelIndent = Line.First->OriginalColumn;
+    if (static_cast<int>(LevelIndent) - Offset >= 0)
+      LevelIndent -= Offset;
+    if ((Line.First->isNot(tok::comment) || IndentForLevel[Line.Level] == -1) &&
+        !Line.InPPDirective)
+      IndentForLevel[Line.Level] = LevelIndent;
+  }
+
+private:
+  /// \brief Get the offset of the line relatively to the level.
+  ///
+  /// For example, 'public:' labels in classes are offset by 1 or 2
+  /// characters to the left from their level.
+  int getIndentOffset(const FormatToken &RootToken) {
+    if (Style.Language == FormatStyle::LK_Java ||
+        Style.Language == FormatStyle::LK_JavaScript)
+      return 0;
+    if (RootToken.isAccessSpecifier(false) ||
+        RootToken.isObjCAccessSpecifier() ||
+        (RootToken.is(Keywords.kw_signals) && RootToken.Next &&
+         RootToken.Next->is(tok::colon)))
+      return Style.AccessModifierOffset;
+    return 0;
+  }
+
+  /// \brief Get the indent of \p Level from \p IndentForLevel.
+  ///
+  /// \p IndentForLevel must contain the indent for the level \c l
+  /// at \p IndentForLevel[l], or a value < 0 if the indent for
+  /// that level is unknown.
+  unsigned getIndent(ArrayRef<int> IndentForLevel, unsigned Level) {
+    if (IndentForLevel[Level] != -1)
+      return IndentForLevel[Level];
+    if (Level == 0)
+      return 0;
+    return getIndent(IndentForLevel, Level - 1) + Style.IndentWidth;
+  }
+
+  const FormatStyle &Style;
+  const AdditionalKeywords &Keywords;
+  const unsigned AdditionalIndent;
+
+  /// \brief The indent in characters for each level.
+  std::vector<int> IndentForLevel;
+
+  /// \brief Offset of the current line relative to the indent level.
+  ///
+  /// For example, the 'public' keywords is often indented with a negative
+  /// offset.
+  int Offset = 0;
+
+  /// \brief The current line's indent.
+  unsigned Indent = 0;
+};
+
 class LineJoiner {
 public:
-  LineJoiner(const FormatStyle &Style) : Style(Style) {}
+  LineJoiner(const FormatStyle &Style, const AdditionalKeywords &Keywords,
+             const SmallVectorImpl<AnnotatedLine *> &Lines)
+      : Style(Style), Keywords(Keywords), End(Lines.end()),
+        Next(Lines.begin()) {}
+
+  /// \brief Returns the next line, merging multiple lines into one if possible.
+  const AnnotatedLine *getNextMergedLine(bool DryRun,
+                                         LevelIndentTracker &IndentTracker) {
+    if (Next == End)
+      return nullptr;
+    const AnnotatedLine *Current = *Next;
+    IndentTracker.nextLine(*Current);
+    unsigned MergedLines =
+        tryFitMultipleLinesInOne(IndentTracker.getIndent(), Next, End);
+    if (MergedLines > 0 && Style.ColumnLimit == 0)
+      // Disallow line merging if there is a break at the start of one of the
+      // input lines.
+      for (unsigned i = 0; i < MergedLines; ++i)
+        if (Next[i + 1]->First->NewlinesBefore > 0)
+          MergedLines = 0;
+    if (!DryRun)
+      for (unsigned i = 0; i < MergedLines; ++i)
+        join(*Next[i], *Next[i + 1]);
+    Next = Next + MergedLines + 1;
+    return Current;
+  }
 
+private:
   /// \brief Calculates how many lines can be merged into 1 starting at \p I.
   unsigned
   tryFitMultipleLinesInOne(unsigned Indent,
                            SmallVectorImpl<AnnotatedLine *>::const_iterator I,
                            SmallVectorImpl<AnnotatedLine *>::const_iterator E) {
+    // Can't join the last line with anything.
+    if (I + 1 == E)
+      return 0;
     // We can never merge stuff if there are trailing line comments.
     const AnnotatedLine *TheLine = *I;
     if (TheLine->Last->is(TT_LineComment))
       return 0;
+    if (I[1]->Type == LT_Invalid || I[1]->First->MustBreakBefore)
+      return 0;
+    if (TheLine->InPPDirective &&
+        (!I[1]->InPPDirective || I[1]->First->HasUnescapedNewline))
+      return 0;
 
     if (Style.ColumnLimit > 0 && Indent > Style.ColumnLimit)
       return 0;
@@ -50,9 +183,6 @@ public:
                 ? 0
                 : Limit - TheLine->Last->TotalLength;
 
-    if (I + 1 == E || I[1]->Type == LT_Invalid || I[1]->First->MustBreakBefore)
-      return 0;
-
     // FIXME: TheLine->Level != 0 might or might not be the right check to do.
     // If necessary, change to something smarter.
     bool MergeShortFunctions =
@@ -113,15 +243,12 @@ public:
     return 0;
   }
 
-private:
   unsigned
   tryMergeSimplePPDirective(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
                             SmallVectorImpl<AnnotatedLine *>::const_iterator E,
                             unsigned Limit) {
     if (Limit == 0)
       return 0;
-    if (!I[1]->InPPDirective || I[1]->First->HasUnescapedNewline)
-      return 0;
     if (I + 2 != E && I[2]->InPPDirective && !I[2]->First->HasUnescapedNewline)
       return 0;
     if (1 + I[1]->Last->TotalLength > Limit)
@@ -147,8 +274,8 @@ private:
       return 0;
     if (1 + I[1]->Last->TotalLength > Limit)
       return 0;
-    if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for,
-                             tok::kw_while, TT_LineComment))
+    if (I[1]->First->isOneOf(tok::semi, tok::kw_if, tok::kw_for, tok::kw_while,
+                             TT_LineComment))
       return 0;
     // Only inline simple if's (no nested if or else).
     if (I + 2 != E && Line.First->is(tok::kw_if) &&
@@ -157,9 +284,10 @@ private:
     return 1;
   }
 
-  unsigned tryMergeShortCaseLabels(
-      SmallVectorImpl<AnnotatedLine *>::const_iterator I,
-      SmallVectorImpl<AnnotatedLine *>::const_iterator E, unsigned Limit) {
+  unsigned
+  tryMergeShortCaseLabels(SmallVectorImpl<AnnotatedLine *>::const_iterator I,
+                          SmallVectorImpl<AnnotatedLine *>::const_iterator E,
+                          unsigned Limit) {
     if (Limit == 0 || I + 1 == E ||
         I[1]->First->isOneOf(tok::kw_case, tok::kw_default))
       return 0;
@@ -191,16 +319,21 @@ private:
     AnnotatedLine &Line = **I;
 
     // Don't merge ObjC @ keywords and methods.
+    // FIXME: If an option to allow short exception handling clauses on a single
+    // line is added, change this to not return for @try and friends.
     if (Style.Language != FormatStyle::LK_Java &&
         Line.First->isOneOf(tok::at, tok::minus, tok::plus))
       return 0;
 
     // Check that the current line allows merging. This depends on whether we
     // are in a control flow statements as well as several style flags.
-    if (Line.First->isOneOf(tok::kw_else, tok::kw_case))
+    if (Line.First->isOneOf(tok::kw_else, tok::kw_case) ||
+        (Line.First->Next && Line.First->Next->is(tok::kw_else)))
       return 0;
     if (Line.First->isOneOf(tok::kw_if, tok::kw_while, tok::kw_do, tok::kw_try,
-                            tok::kw_catch, tok::kw_for, tok::r_brace)) {
+                            tok::kw___try, tok::kw_catch, tok::kw___finally,
+                            tok::kw_for, tok::r_brace) ||
+        Line.First->is(Keywords.kw___except)) {
       if (!Style.AllowShortBlocksOnASingleLine)
         return 0;
       if (!Style.AllowShortIfStatementsOnASingleLine &&
@@ -211,7 +344,11 @@ private:
         return 0;
       // FIXME: Consider an option to allow short exception handling clauses on
       // a single line.
-      if (Line.First->isOneOf(tok::kw_try, tok::kw_catch))
+      // FIXME: This isn't covered by tests.
+      // FIXME: For catch, __except, __finally the first token on the line
+      // is '}', so this isn't correct here.
+      if (Line.First->isOneOf(tok::kw_try, tok::kw___try, tok::kw_catch,
+                              Keywords.kw___except, tok::kw___finally))
         return 0;
     }
 
@@ -226,7 +363,8 @@ private:
     } else if (Limit != 0 && Line.First->isNot(tok::kw_namespace) &&
                !startsExternCBlock(Line)) {
       // We don't merge short records.
-      if (Line.First->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct))
+      if (Line.First->isOneOf(tok::kw_class, tok::kw_union, tok::kw_struct,
+                              Keywords.kw_interface))
         return 0;
 
       // Check that we still have three lines and they fit into the limit.
@@ -252,6 +390,10 @@ private:
       if (Tok->isNot(tok::r_brace))
         return 0;
 
+      // Don't merge "if (a) { .. } else {".
+      if (Tok->Next && Tok->Next->is(tok::kw_else))
+        return 0;
+
       return 2;
     }
     return 0;
@@ -285,28 +427,367 @@ private:
     return false;
   }
 
+  void join(AnnotatedLine &A, const AnnotatedLine &B) {
+    assert(!A.Last->Next);
+    assert(!B.First->Previous);
+    if (B.Affected)
+      A.Affected = true;
+    A.Last->Next = B.First;
+    B.First->Previous = A.Last;
+    B.First->CanBreakBefore = true;
+    unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
+    for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
+      Tok->TotalLength += LengthA;
+      A.Last = Tok;
+    }
+  }
+
   const FormatStyle &Style;
+  const AdditionalKeywords &Keywords;
+  const SmallVectorImpl<AnnotatedLine*>::const_iterator End;
+
+  SmallVectorImpl<AnnotatedLine*>::const_iterator Next;
 };
 
-class NoColumnLimitFormatter {
+static void markFinalized(FormatToken *Tok) {
+  for (; Tok; Tok = Tok->Next) {
+    Tok->Finalized = true;
+    for (AnnotatedLine *Child : Tok->Children)
+      markFinalized(Child->First);
+  }
+}
+
+#ifndef NDEBUG
+static void printLineState(const LineState &State) {
+  llvm::dbgs() << "State: ";
+  for (const ParenState &P : State.Stack) {
+    llvm::dbgs() << P.Indent << "|" << P.LastSpace << "|" << P.NestedBlockIndent
+                 << " ";
+  }
+  llvm::dbgs() << State.NextToken->TokenText << "\n";
+}
+#endif
+
+/// \brief Base class for classes that format one \c AnnotatedLine.
+class LineFormatter {
 public:
-  NoColumnLimitFormatter(ContinuationIndenter *Indenter) : Indenter(Indenter) {}
+  LineFormatter(ContinuationIndenter *Indenter, WhitespaceManager *Whitespaces,
+                const FormatStyle &Style,
+                UnwrappedLineFormatter *BlockFormatter)
+      : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style),
+        BlockFormatter(BlockFormatter) {}
+  virtual ~LineFormatter() {}
+
+  /// \brief Formats an \c AnnotatedLine and returns the penalty.
+  ///
+  /// If \p DryRun is \c false, directly applies the changes.
+  virtual unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
+                              bool DryRun) = 0;
+
+protected:
+  /// \brief If the \p State's next token is an r_brace closing a nested block,
+  /// format the nested block before it.
+  ///
+  /// Returns \c true if all children could be placed successfully and adapts
+  /// \p Penalty as well as \p State. If \p DryRun is false, also directly
+  /// creates changes using \c Whitespaces.
+  ///
+  /// The crucial idea here is that children always get formatted upon
+  /// encountering the closing brace right after the nested block. Now, if we
+  /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
+  /// \c false), the entire block has to be kept on the same line (which is only
+  /// possible if it fits on the line, only contains a single statement, etc.
+  ///
+  /// If \p NewLine is true, we format the nested block on separate lines, i.e.
+  /// break after the "{", format all lines with correct indentation and the put
+  /// the closing "}" on yet another new line.
+  ///
+  /// This enables us to keep the simple structure of the
+  /// \c UnwrappedLineFormatter, where we only have two options for each token:
+  /// break or don't break.
+  bool formatChildren(LineState &State, bool NewLine, bool DryRun,
+                      unsigned &Penalty) {
+    const FormatToken *LBrace = State.NextToken->getPreviousNonComment();
+    FormatToken &Previous = *State.NextToken->Previous;
+    if (!LBrace || LBrace->isNot(tok::l_brace) ||
+        LBrace->BlockKind != BK_Block || Previous.Children.size() == 0)
+      // The previous token does not open a block. Nothing to do. We don't
+      // assert so that we can simply call this function for all tokens.
+      return true;
+
+    if (NewLine) {
+      int AdditionalIndent = State.Stack.back().Indent -
+                             Previous.Children[0]->Level * Style.IndentWidth;
+
+      Penalty +=
+          BlockFormatter->format(Previous.Children, DryRun, AdditionalIndent,
+                                 /*FixBadIndentation=*/true);
+      return true;
+    }
+
+    if (Previous.Children[0]->First->MustBreakBefore)
+      return false;
+
+    // Cannot merge multiple statements into a single line.
+    if (Previous.Children.size() > 1)
+      return false;
+
+    // Cannot merge into one line if this line ends on a comment.
+    if (Previous.is(tok::comment))
+      return false;
+
+    // We can't put the closing "}" on a line with a trailing comment.
+    if (Previous.Children[0]->Last->isTrailingComment())
+      return false;
+
+    // If the child line exceeds the column limit, we wouldn't want to merge it.
+    // We add +2 for the trailing " }".
+    if (Style.ColumnLimit > 0 &&
+        Previous.Children[0]->Last->TotalLength + State.Column + 2 >
+            Style.ColumnLimit)
+      return false;
+
+    if (!DryRun) {
+      Whitespaces->replaceWhitespace(
+          *Previous.Children[0]->First,
+          /*Newlines=*/0, /*IndentLevel=*/0, /*Spaces=*/1,
+          /*StartOfTokenColumn=*/State.Column, State.Line->InPPDirective);
+    }
+    Penalty += formatLine(*Previous.Children[0], State.Column + 1, DryRun);
+
+    State.Column += 1 + Previous.Children[0]->Last->TotalLength;
+    return true;
+  }
+
+  ContinuationIndenter *Indenter;
+
+private:
+  WhitespaceManager *Whitespaces;
+  const FormatStyle &Style;
+  UnwrappedLineFormatter *BlockFormatter;
+};
 
-  /// \brief Formats the line starting at \p State, simply keeping all of the
-  /// input's line breaking decisions.
-  void format(unsigned FirstIndent, const AnnotatedLine *Line) {
+/// \brief Formatter that keeps the existing line breaks.
+class NoColumnLimitLineFormatter : public LineFormatter {
+public:
+  NoColumnLimitLineFormatter(ContinuationIndenter *Indenter,
+                             WhitespaceManager *Whitespaces,
+                             const FormatStyle &Style,
+                             UnwrappedLineFormatter *BlockFormatter)
+      : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
+
+  /// \brief Formats the line, simply keeping all of the input's line breaking
+  /// decisions.
+  unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
+                      bool DryRun) override {
+    assert(!DryRun);
     LineState State =
-        Indenter->getInitialState(FirstIndent, Line, /*DryRun=*/false);
+        Indenter->getInitialState(FirstIndent, &Line, /*DryRun=*/false);
     while (State.NextToken) {
       bool Newline =
           Indenter->mustBreak(State) ||
           (Indenter->canBreak(State) && State.NextToken->NewlinesBefore > 0);
+      unsigned Penalty = 0;
+      formatChildren(State, Newline, /*DryRun=*/false, Penalty);
       Indenter->addTokenToState(State, Newline, /*DryRun=*/false);
     }
+    return 0;
+  }
+};
+
+/// \brief Formatter that puts all tokens into a single line without breaks.
+class NoLineBreakFormatter : public LineFormatter {
+public:
+  NoLineBreakFormatter(ContinuationIndenter *Indenter,
+                       WhitespaceManager *Whitespaces, const FormatStyle &Style,
+                       UnwrappedLineFormatter *BlockFormatter)
+      : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
+
+  /// \brief Puts all tokens into a single line.
+  unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
+                      bool DryRun) {
+    unsigned Penalty = 0;
+    LineState State = Indenter->getInitialState(FirstIndent, &Line, DryRun);
+    while (State.NextToken) {
+      formatChildren(State, /*Newline=*/false, DryRun, Penalty);
+      Indenter->addTokenToState(State, /*Newline=*/false, DryRun);
+    }
+    return Penalty;
+  }
+};
+
+/// \brief Finds the best way to break lines.
+class OptimizingLineFormatter : public LineFormatter {
+public:
+  OptimizingLineFormatter(ContinuationIndenter *Indenter,
+                          WhitespaceManager *Whitespaces,
+                          const FormatStyle &Style,
+                          UnwrappedLineFormatter *BlockFormatter)
+      : LineFormatter(Indenter, Whitespaces, Style, BlockFormatter) {}
+
+  /// \brief Formats the line by finding the best line breaks with line lengths
+  /// below the column limit.
+  unsigned formatLine(const AnnotatedLine &Line, unsigned FirstIndent,
+                      bool DryRun) {
+    LineState State = Indenter->getInitialState(FirstIndent, &Line, DryRun);
+
+    // If the ObjC method declaration does not fit on a line, we should format
+    // it with one arg per line.
+    if (State.Line->Type == LT_ObjCMethodDecl)
+      State.Stack.back().BreakBeforeParameter = true;
+
+    // Find best solution in solution space.
+    return analyzeSolutionSpace(State, DryRun);
   }
 
 private:
-  ContinuationIndenter *Indenter;
+  struct CompareLineStatePointers {
+    bool operator()(LineState *obj1, LineState *obj2) const {
+      return *obj1 < *obj2;
+    }
+  };
+
+  /// \brief A pair of <penalty, count> that is used to prioritize the BFS on.
+  ///
+  /// In case of equal penalties, we want to prefer states that were inserted
+  /// first. During state generation we make sure that we insert states first
+  /// that break the line as late as possible.
+  typedef std::pair<unsigned, unsigned> OrderedPenalty;
+
+  /// \brief An edge in the solution space from \c Previous->State to \c State,
+  /// inserting a newline dependent on the \c NewLine.
+  struct StateNode {
+    StateNode(const LineState &State, bool NewLine, StateNode *Previous)
+        : State(State), NewLine(NewLine), Previous(Previous) {}
+    LineState State;
+    bool NewLine;
+    StateNode *Previous;
+  };
+
+  /// \brief An item in the prioritized BFS search queue. The \c StateNode's
+  /// \c State has the given \c OrderedPenalty.
+  typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
+
+  /// \brief The BFS queue type.
+  typedef std::priority_queue<QueueItem, std::vector<QueueItem>,
+                              std::greater<QueueItem>> QueueType;
+
+  /// \brief Analyze the entire solution space starting from \p InitialState.
+  ///
+  /// This implements a variant of Dijkstra's algorithm on the graph that spans
+  /// the solution space (\c LineStates are the nodes). The algorithm tries to
+  /// find the shortest path (the one with lowest penalty) from \p InitialState
+  /// to a state where all tokens are placed. Returns the penalty.
+  ///
+  /// If \p DryRun is \c false, directly applies the changes.
+  unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun) {
+    std::set<LineState *, CompareLineStatePointers> Seen;
+
+    // Increasing count of \c StateNode items we have created. This is used to
+    // create a deterministic order independent of the container.
+    unsigned Count = 0;
+    QueueType Queue;
+
+    // Insert start element into queue.
+    StateNode *Node =
+        new (Allocator.Allocate()) StateNode(InitialState, false, nullptr);
+    Queue.push(QueueItem(OrderedPenalty(0, Count), Node));
+    ++Count;
+
+    unsigned Penalty = 0;
+
+    // While not empty, take first element and follow edges.
+    while (!Queue.empty()) {
+      Penalty = Queue.top().first.first;
+      StateNode *Node = Queue.top().second;
+      if (!Node->State.NextToken) {
+        DEBUG(llvm::dbgs() << "\n---\nPenalty for line: " << Penalty << "\n");
+        break;
+      }
+      Queue.pop();
+
+      // Cut off the analysis of certain solutions if the analysis gets too
+      // complex. See description of IgnoreStackForComparison.
+      if (Count > 10000)
+        Node->State.IgnoreStackForComparison = true;
+
+      if (!Seen.insert(&Node->State).second)
+        // State already examined with lower penalty.
+        continue;
+
+      FormatDecision LastFormat = Node->State.NextToken->Decision;
+      if (LastFormat == FD_Unformatted || LastFormat == FD_Continue)
+        addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue);
+      if (LastFormat == FD_Unformatted || LastFormat == FD_Break)
+        addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue);
+    }
+
+    if (Queue.empty()) {
+      // We were unable to find a solution, do nothing.
+      // FIXME: Add diagnostic?
+      DEBUG(llvm::dbgs() << "Could not find a solution.\n");
+      return 0;
+    }
+
+    // Reconstruct the solution.
+    if (!DryRun)
+      reconstructPath(InitialState, Queue.top().second);
+
+    DEBUG(llvm::dbgs() << "Total number of analyzed states: " << Count << "\n");
+    DEBUG(llvm::dbgs() << "---\n");
+
+    return Penalty;
+  }
+
+  /// \brief Add the following state to the analysis queue \c Queue.
+  ///
+  /// Assume the current state is \p PreviousNode and has been reached with a
+  /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
+  void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
+                           bool NewLine, unsigned *Count, QueueType *Queue) {
+    if (NewLine && !Indenter->canBreak(PreviousNode->State))
+      return;
+    if (!NewLine && Indenter->mustBreak(PreviousNode->State))
+      return;
+
+    StateNode *Node = new (Allocator.Allocate())
+        StateNode(PreviousNode->State, NewLine, PreviousNode);
+    if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
+      return;
+
+    Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
+
+    Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
+    ++(*Count);
+  }
+
+  /// \brief Applies the best formatting by reconstructing the path in the
+  /// solution space that leads to \c Best.
+  void reconstructPath(LineState &State, StateNode *Best) {
+    std::deque<StateNode *> Path;
+    // We do not need a break before the initial token.
+    while (Best->Previous) {
+      Path.push_front(Best);
+      Best = Best->Previous;
+    }
+    for (std::deque<StateNode *>::iterator I = Path.begin(), E = Path.end();
+         I != E; ++I) {
+      unsigned Penalty = 0;
+      formatChildren(State, (*I)->NewLine, /*DryRun=*/false, Penalty);
+      Penalty += Indenter->addTokenToState(State, (*I)->NewLine, false);
+
+      DEBUG({
+        printLineState((*I)->Previous->State);
+        if ((*I)->NewLine) {
+          llvm::dbgs() << "Penalty for placing "
+                       << (*I)->Previous->State.NextToken->Tok.getName() << ": "
+                       << Penalty << "\n";
+        }
+      });
+    }
+  }
+
+  llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
 };
 
 } // namespace
@@ -315,7 +796,7 @@ unsigned
 UnwrappedLineFormatter::format(const SmallVectorImpl<AnnotatedLine *> &Lines,
                                bool DryRun, int AdditionalIndent,
                                bool FixBadIndentation) {
-  LineJoiner Joiner(Style);
+  LineJoiner Joiner(Style, Keywords, Lines);
 
   // Try to look up already computed penalty in DryRun-mode.
   std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned> CacheKey(
@@ -326,151 +807,93 @@ UnwrappedLineFormatter::format(const SmallVectorImpl<AnnotatedLine *> &Lines,
 
   assert(!Lines.empty());
   unsigned Penalty = 0;
-  std::vector<int> IndentForLevel;
-  for (unsigned i = 0, e = Lines[0]->Level; i != e; ++i)
-    IndentForLevel.push_back(Style.IndentWidth * i + AdditionalIndent);
+  LevelIndentTracker IndentTracker(Style, Keywords, Lines[0]->Level,
+                                   AdditionalIndent);
   const AnnotatedLine *PreviousLine = nullptr;
-  for (SmallVectorImpl<AnnotatedLine *>::const_iterator I = Lines.begin(),
-                                                        E = Lines.end();
-       I != E; ++I) {
-    const AnnotatedLine &TheLine = **I;
-    const FormatToken *FirstTok = TheLine.First;
-    int Offset = getIndentOffset(*FirstTok);
-
-    // Determine indent and try to merge multiple unwrapped lines.
-    unsigned Indent;
-    if (TheLine.InPPDirective) {
-      Indent = TheLine.Level * Style.IndentWidth;
-    } else {
-      while (IndentForLevel.size() <= TheLine.Level)
-        IndentForLevel.push_back(-1);
-      IndentForLevel.resize(TheLine.Level + 1);
-      Indent = getIndent(IndentForLevel, TheLine.Level);
-    }
-    unsigned LevelIndent = Indent;
-    if (static_cast<int>(Indent) + Offset >= 0)
-      Indent += Offset;
-
-    // Merge multiple lines if possible.
-    unsigned MergedLines = Joiner.tryFitMultipleLinesInOne(Indent, I, E);
-    if (MergedLines > 0 && Style.ColumnLimit == 0) {
-      // Disallow line merging if there is a break at the start of one of the
-      // input lines.
-      for (unsigned i = 0; i < MergedLines; ++i) {
-        if (I[i + 1]->First->NewlinesBefore > 0)
-          MergedLines = 0;
-      }
-    }
-    if (!DryRun) {
-      for (unsigned i = 0; i < MergedLines; ++i) {
-        join(*I[i], *I[i + 1]);
-      }
-    }
-    I += MergedLines;
-
+  const AnnotatedLine *NextLine = nullptr;
+  for (const AnnotatedLine *Line =
+           Joiner.getNextMergedLine(DryRun, IndentTracker);
+       Line; Line = NextLine) {
+    const AnnotatedLine &TheLine = *Line;
+    unsigned Indent = IndentTracker.getIndent();
     bool FixIndentation =
-        FixBadIndentation && (LevelIndent != FirstTok->OriginalColumn);
-    if (TheLine.First->is(tok::eof)) {
-      if (PreviousLine && PreviousLine->Affected && !DryRun) {
-        // Remove the file's trailing whitespace.
-        unsigned Newlines = std::min(FirstTok->NewlinesBefore, 1u);
-        Whitespaces->replaceWhitespace(*TheLine.First, Newlines,
-                                       /*IndentLevel=*/0, /*Spaces=*/0,
-                                       /*TargetColumn=*/0);
-      }
-    } else if (TheLine.Type != LT_Invalid &&
-               (TheLine.Affected || FixIndentation)) {
-      if (FirstTok->WhitespaceRange.isValid()) {
-        if (!DryRun)
-          formatFirstToken(*TheLine.First, PreviousLine, TheLine.Level, Indent,
+        FixBadIndentation && (Indent != TheLine.First->OriginalColumn);
+    bool ShouldFormat = TheLine.Affected || FixIndentation;
+    // We cannot format this line; if the reason is that the line had a
+    // parsing error, remember that.
+    if (ShouldFormat && TheLine.Type == LT_Invalid && IncompleteFormat)
+      *IncompleteFormat = true;
+
+    if (ShouldFormat && TheLine.Type != LT_Invalid) {
+      if (!DryRun)
+        formatFirstToken(*TheLine.First, PreviousLine, TheLine.Level, Indent,
+                         TheLine.InPPDirective);
+
+      NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
+      unsigned ColumnLimit = getColumnLimit(TheLine.InPPDirective, NextLine);
+      bool FitsIntoOneLine =
+          TheLine.Last->TotalLength + Indent <= ColumnLimit ||
+          TheLine.Type == LT_ImportStatement;
+
+      if (Style.ColumnLimit == 0)
+        NoColumnLimitLineFormatter(Indenter, Whitespaces, Style, this)
+            .formatLine(TheLine, Indent, DryRun);
+      else if (FitsIntoOneLine)
+        Penalty += NoLineBreakFormatter(Indenter, Whitespaces, Style, this)
+                       .formatLine(TheLine, Indent, DryRun);
+      else
+        Penalty += OptimizingLineFormatter(Indenter, Whitespaces, Style, this)
+                       .formatLine(TheLine, Indent, DryRun);
+    } else {
+      // If no token in the current line is affected, we still need to format
+      // affected children.
+      if (TheLine.ChildrenAffected)
+        format(TheLine.Children, DryRun);
+
+      // Adapt following lines on the current indent level to the same level
+      // unless the current \c AnnotatedLine is not at the beginning of a line.
+      bool StartsNewLine =
+          TheLine.First->NewlinesBefore > 0 || TheLine.First->IsFirst;
+      if (StartsNewLine)
+        IndentTracker.adjustToUnmodifiedLine(TheLine);
+      if (!DryRun) {
+        bool ReformatLeadingWhitespace =
+            StartsNewLine && ((PreviousLine && PreviousLine->Affected) ||
+                              TheLine.LeadingEmptyLinesAffected);
+        // Format the first token.
+        if (ReformatLeadingWhitespace)
+          formatFirstToken(*TheLine.First, PreviousLine, TheLine.Level,
+                           TheLine.First->OriginalColumn,
                            TheLine.InPPDirective);
-      } else {
-        Indent = LevelIndent = FirstTok->OriginalColumn;
-      }
-
-      // If everything fits on a single line, just put it there.
-      unsigned ColumnLimit = Style.ColumnLimit;
-      if (I + 1 != E) {
-        AnnotatedLine *NextLine = I[1];
-        if (NextLine->InPPDirective && !NextLine->First->HasUnescapedNewline)
-          ColumnLimit = getColumnLimit(TheLine.InPPDirective);
-      }
+        else
+          Whitespaces->addUntouchableToken(*TheLine.First,
+                                           TheLine.InPPDirective);
 
-      if (TheLine.Last->TotalLength + Indent <= ColumnLimit ||
-          TheLine.Type == LT_ImportStatement) {
-        LineState State = Indenter->getInitialState(Indent, &TheLine, DryRun);
-        while (State.NextToken) {
-          formatChildren(State, /*Newline=*/false, /*DryRun=*/false, Penalty);
-          Indenter->addTokenToState(State, /*Newline=*/false, DryRun);
-        }
-      } else if (Style.ColumnLimit == 0) {
-        // FIXME: Implement nested blocks for ColumnLimit = 0.
-        NoColumnLimitFormatter Formatter(Indenter);
-        if (!DryRun)
-          Formatter.format(Indent, &TheLine);
-      } else {
-        Penalty += format(TheLine, Indent, DryRun);
-      }
-
-      if (!TheLine.InPPDirective)
-        IndentForLevel[TheLine.Level] = LevelIndent;
-    } else if (TheLine.ChildrenAffected) {
-      format(TheLine.Children, DryRun);
-    } else {
-      // Format the first token if necessary, and notify the WhitespaceManager
-      // about the unchanged whitespace.
-      for (FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next) {
-        if (Tok == TheLine.First && (Tok->NewlinesBefore > 0 || Tok->IsFirst)) {
-          unsigned LevelIndent = Tok->OriginalColumn;
-          if (!DryRun) {
-            // Remove trailing whitespace of the previous line.
-            if ((PreviousLine && PreviousLine->Affected) ||
-                TheLine.LeadingEmptyLinesAffected) {
-              formatFirstToken(*Tok, PreviousLine, TheLine.Level, LevelIndent,
-                               TheLine.InPPDirective);
-            } else {
-              Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
-            }
-          }
-
-          if (static_cast<int>(LevelIndent) - Offset >= 0)
-            LevelIndent -= Offset;
-          if (Tok->isNot(tok::comment) && !TheLine.InPPDirective)
-            IndentForLevel[TheLine.Level] = LevelIndent;
-        } else if (!DryRun) {
+        // Notify the WhitespaceManager about the unchanged whitespace.
+        for (FormatToken *Tok = TheLine.First->Next; Tok; Tok = Tok->Next)
           Whitespaces->addUntouchableToken(*Tok, TheLine.InPPDirective);
-        }
-      }
-    }
-    if (!DryRun) {
-      for (FormatToken *Tok = TheLine.First; Tok; Tok = Tok->Next) {
-        Tok->Finalized = true;
       }
+      NextLine = Joiner.getNextMergedLine(DryRun, IndentTracker);
     }
-    PreviousLine = *I;
+    if (!DryRun)
+      markFinalized(TheLine.First);
+    PreviousLine = &TheLine;
   }
   PenaltyCache[CacheKey] = Penalty;
   return Penalty;
 }
 
-unsigned UnwrappedLineFormatter::format(const AnnotatedLine &Line,
-                                        unsigned FirstIndent, bool DryRun) {
-  LineState State = Indenter->getInitialState(FirstIndent, &Line, DryRun);
-
-  // If the ObjC method declaration does not fit on a line, we should format
-  // it with one arg per line.
-  if (State.Line->Type == LT_ObjCMethodDecl)
-    State.Stack.back().BreakBeforeParameter = true;
-
-  // Find best solution in solution space.
-  return analyzeSolutionSpace(State, DryRun);
-}
-
 void UnwrappedLineFormatter::formatFirstToken(FormatToken &RootToken,
                                               const AnnotatedLine *PreviousLine,
                                               unsigned IndentLevel,
                                               unsigned Indent,
                                               bool InPPDirective) {
+  if (RootToken.is(tok::eof)) {
+    unsigned Newlines = std::min(RootToken.NewlinesBefore, 1u);
+    Whitespaces->replaceWhitespace(RootToken, Newlines, /*IndentLevel=*/0,
+                                   /*Spaces=*/0, /*TargetColumn=*/0);
+    return;
+  }
   unsigned Newlines =
       std::min(RootToken.NewlinesBefore, Style.MaxEmptyLinesToKeep + 1);
   // Remove empty lines before "}" where applicable.
@@ -496,7 +919,8 @@ void UnwrappedLineFormatter::formatFirstToken(FormatToken &RootToken,
     ++Newlines;
 
   // Remove empty lines after access specifiers.
-  if (PreviousLine && PreviousLine->First->isAccessSpecifier())
+  if (PreviousLine && PreviousLine->First->isAccessSpecifier() &&
+      (!PreviousLine->InPPDirective || !RootToken.HasUnescapedNewline))
     Newlines = std::min(1u, Newlines);
 
   Whitespaces->replaceWhitespace(RootToken, Newlines, IndentLevel, Indent,
@@ -504,202 +928,21 @@ void UnwrappedLineFormatter::formatFirstToken(FormatToken &RootToken,
                                              !RootToken.HasUnescapedNewline);
 }
 
-/// \brief Get the indent of \p Level from \p IndentForLevel.
-///
-/// \p IndentForLevel must contain the indent for the level \c l
-/// at \p IndentForLevel[l], or a value < 0 if the indent for
-/// that level is unknown.
-unsigned UnwrappedLineFormatter::getIndent(ArrayRef<int> IndentForLevel,
-                                           unsigned Level) {
-  if (IndentForLevel[Level] != -1)
-    return IndentForLevel[Level];
-  if (Level == 0)
-    return 0;
-  return getIndent(IndentForLevel, Level - 1) + Style.IndentWidth;
-}
-
-void UnwrappedLineFormatter::join(AnnotatedLine &A, const AnnotatedLine &B) {
-  assert(!A.Last->Next);
-  assert(!B.First->Previous);
-  if (B.Affected)
-    A.Affected = true;
-  A.Last->Next = B.First;
-  B.First->Previous = A.Last;
-  B.First->CanBreakBefore = true;
-  unsigned LengthA = A.Last->TotalLength + B.First->SpacesRequiredBefore;
-  for (FormatToken *Tok = B.First; Tok; Tok = Tok->Next) {
-    Tok->TotalLength += LengthA;
-    A.Last = Tok;
-  }
-}
-
-unsigned UnwrappedLineFormatter::analyzeSolutionSpace(LineState &InitialState,
-                                                      bool DryRun) {
-  std::set<LineState *, CompareLineStatePointers> Seen;
-
-  // Increasing count of \c StateNode items we have created. This is used to
-  // create a deterministic order independent of the container.
-  unsigned Count = 0;
-  QueueType Queue;
-
-  // Insert start element into queue.
-  StateNode *Node =
-      new (Allocator.Allocate()) StateNode(InitialState, false, nullptr);
-  Queue.push(QueueItem(OrderedPenalty(0, Count), Node));
-  ++Count;
-
-  unsigned Penalty = 0;
-
-  // While not empty, take first element and follow edges.
-  while (!Queue.empty()) {
-    Penalty = Queue.top().first.first;
-    StateNode *Node = Queue.top().second;
-    if (!Node->State.NextToken) {
-      DEBUG(llvm::dbgs() << "\n---\nPenalty for line: " << Penalty << "\n");
-      break;
-    }
-    Queue.pop();
-
-    // Cut off the analysis of certain solutions if the analysis gets too
-    // complex. See description of IgnoreStackForComparison.
-    if (Count > 10000)
-      Node->State.IgnoreStackForComparison = true;
-
-    if (!Seen.insert(&Node->State).second)
-      // State already examined with lower penalty.
-      continue;
-
-    FormatDecision LastFormat = Node->State.NextToken->Decision;
-    if (LastFormat == FD_Unformatted || LastFormat == FD_Continue)
-      addNextStateToQueue(Penalty, Node, /*NewLine=*/false, &Count, &Queue);
-    if (LastFormat == FD_Unformatted || LastFormat == FD_Break)
-      addNextStateToQueue(Penalty, Node, /*NewLine=*/true, &Count, &Queue);
-  }
-
-  if (Queue.empty()) {
-    // We were unable to find a solution, do nothing.
-    // FIXME: Add diagnostic?
-    DEBUG(llvm::dbgs() << "Could not find a solution.\n");
-    return 0;
-  }
-
-  // Reconstruct the solution.
-  if (!DryRun)
-    reconstructPath(InitialState, Queue.top().second);
-
-  DEBUG(llvm::dbgs() << "Total number of analyzed states: " << Count << "\n");
-  DEBUG(llvm::dbgs() << "---\n");
-
-  return Penalty;
-}
-
-#ifndef NDEBUG
-static void printLineState(const LineState &State) {
-  llvm::dbgs() << "State: ";
-  for (const ParenState &P : State.Stack) {
-    llvm::dbgs() << P.Indent << "|" << P.LastSpace << "|" << P.NestedBlockIndent
-                 << " ";
-  }
-  llvm::dbgs() << State.NextToken->TokenText << "\n";
-}
-#endif
-
-void UnwrappedLineFormatter::reconstructPath(LineState &State,
-                                             StateNode *Current) {
-  std::deque<StateNode *> Path;
-  // We do not need a break before the initial token.
-  while (Current->Previous) {
-    Path.push_front(Current);
-    Current = Current->Previous;
-  }
-  for (std::deque<StateNode *>::iterator I = Path.begin(), E = Path.end();
-       I != E; ++I) {
-    unsigned Penalty = 0;
-    formatChildren(State, (*I)->NewLine, /*DryRun=*/false, Penalty);
-    Penalty += Indenter->addTokenToState(State, (*I)->NewLine, false);
-
-    DEBUG({
-      printLineState((*I)->Previous->State);
-      if ((*I)->NewLine) {
-        llvm::dbgs() << "Penalty for placing "
-                     << (*I)->Previous->State.NextToken->Tok.getName() << ": "
-                     << Penalty << "\n";
-      }
-    });
-  }
-}
-
-void UnwrappedLineFormatter::addNextStateToQueue(unsigned Penalty,
-                                                 StateNode *PreviousNode,
-                                                 bool NewLine, unsigned *Count,
-                                                 QueueType *Queue) {
-  if (NewLine && !Indenter->canBreak(PreviousNode->State))
-    return;
-  if (!NewLine && Indenter->mustBreak(PreviousNode->State))
-    return;
-
-  StateNode *Node = new (Allocator.Allocate())
-      StateNode(PreviousNode->State, NewLine, PreviousNode);
-  if (!formatChildren(Node->State, NewLine, /*DryRun=*/true, Penalty))
-    return;
-
-  Penalty += Indenter->addTokenToState(Node->State, NewLine, true);
-
-  Queue->push(QueueItem(OrderedPenalty(Penalty, *Count), Node));
-  ++(*Count);
-}
-
-bool UnwrappedLineFormatter::formatChildren(LineState &State, bool NewLine,
-                                            bool DryRun, unsigned &Penalty) {
-  FormatToken &Previous = *State.NextToken->Previous;
-  const FormatToken *LBrace = State.NextToken->getPreviousNonComment();
-  if (!LBrace || LBrace->isNot(tok::l_brace) || LBrace->BlockKind != BK_Block ||
-      Previous.Children.size() == 0)
-    // The previous token does not open a block. Nothing to do. We don't
-    // assert so that we can simply call this function for all tokens.
-    return true;
-
-  if (NewLine) {
-    int AdditionalIndent = State.Stack.back().Indent -
-                           Previous.Children[0]->Level * Style.IndentWidth;
-
-    Penalty += format(Previous.Children, DryRun, AdditionalIndent,
-                      /*FixBadIndentation=*/true);
-    return true;
-  }
-
-  if (Previous.Children[0]->First->MustBreakBefore)
-    return false;
-
-  // Cannot merge multiple statements into a single line.
-  if (Previous.Children.size() > 1)
-    return false;
-
-  // Cannot merge into one line if this line ends on a comment.
-  if (Previous.is(tok::comment))
-    return false;
-
-  // We can't put the closing "}" on a line with a trailing comment.
-  if (Previous.Children[0]->Last->isTrailingComment())
-    return false;
-
-  // If the child line exceeds the column limit, we wouldn't want to merge it.
-  // We add +2 for the trailing " }".
-  if (Style.ColumnLimit > 0 &&
-      Previous.Children[0]->Last->TotalLength + State.Column + 2 >
-          Style.ColumnLimit)
-    return false;
-
-  if (!DryRun) {
-    Whitespaces->replaceWhitespace(
-        *Previous.Children[0]->First,
-        /*Newlines=*/0, /*IndentLevel=*/0, /*Spaces=*/1,
-        /*StartOfTokenColumn=*/State.Column, State.Line->InPPDirective);
-  }
-  Penalty += format(*Previous.Children[0], State.Column + 1, DryRun);
-
-  State.Column += 1 + Previous.Children[0]->Last->TotalLength;
-  return true;
+unsigned
+UnwrappedLineFormatter::getColumnLimit(bool InPPDirective,
+                                       const AnnotatedLine *NextLine) const {
+  // In preprocessor directives reserve two chars for trailing " \" if the
+  // next line continues the preprocessor directive.
+  bool ContinuesPPDirective =
+      InPPDirective &&
+      // If there is no next line, this is likely a child line and the parent
+      // continues the preprocessor directive.
+      (!NextLine ||
+       (NextLine->InPPDirective &&
+        // If there is an unescaped newline between this line and the next, the
+        // next line starts a new preprocessor directive.
+        !NextLine->First->HasUnescapedNewline));
+  return Style.ColumnLimit - (ContinuesPPDirective ? 2 : 0);
 }
 
 } // namespace format
diff --git a/lib/Format/UnwrappedLineFormatter.h b/lib/Format/UnwrappedLineFormatter.h
index 3ae6dbc4db0b..da9aa1c605e4 100644
--- a/lib/Format/UnwrappedLineFormatter.h
+++ b/lib/Format/UnwrappedLineFormatter.h
@@ -32,135 +32,39 @@ class UnwrappedLineFormatter {
 public:
   UnwrappedLineFormatter(ContinuationIndenter *Indenter,
                          WhitespaceManager *Whitespaces,
-                         const FormatStyle &Style)
-      : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style) {}
+                         const FormatStyle &Style,
+                         const AdditionalKeywords &Keywords,
+                         bool *IncompleteFormat)
+      : Indenter(Indenter), Whitespaces(Whitespaces), Style(Style),
+        Keywords(Keywords), IncompleteFormat(IncompleteFormat) {}
 
-  unsigned format(const SmallVectorImpl<AnnotatedLine *> &Lines, bool DryRun,
-                  int AdditionalIndent = 0, bool FixBadIndentation = false);
+  /// \brief Format the current block and return the penalty.
+  unsigned format(const SmallVectorImpl<AnnotatedLine *> &Lines,
+                  bool DryRun = false, int AdditionalIndent = 0,
+                  bool FixBadIndentation = false);
 
 private:
-  /// \brief Formats an \c AnnotatedLine and returns the penalty.
-  ///
-  /// If \p DryRun is \c false, directly applies the changes.
-  unsigned format(const AnnotatedLine &Line, unsigned FirstIndent,
-                  bool DryRun);
-
-  /// \brief An edge in the solution space from \c Previous->State to \c State,
-  /// inserting a newline dependent on the \c NewLine.
-  struct StateNode {
-    StateNode(const LineState &State, bool NewLine, StateNode *Previous)
-        : State(State), NewLine(NewLine), Previous(Previous) {}
-    LineState State;
-    bool NewLine;
-    StateNode *Previous;
-  };
-
-  /// \brief A pair of <penalty, count> that is used to prioritize the BFS on.
-  ///
-  /// In case of equal penalties, we want to prefer states that were inserted
-  /// first. During state generation we make sure that we insert states first
-  /// that break the line as late as possible.
-  typedef std::pair<unsigned, unsigned> OrderedPenalty;
-
-  /// \brief An item in the prioritized BFS search queue. The \c StateNode's
-  /// \c State has the given \c OrderedPenalty.
-  typedef std::pair<OrderedPenalty, StateNode *> QueueItem;
-
-  /// \brief The BFS queue type.
-  typedef std::priority_queue<QueueItem, std::vector<QueueItem>,
-                              std::greater<QueueItem> > QueueType;
-
-  /// \brief Get the offset of the line relatively to the level.
-  ///
-  /// For example, 'public:' labels in classes are offset by 1 or 2
-  /// characters to the left from their level.
-  int getIndentOffset(const FormatToken &RootToken) {
-    if (Style.Language == FormatStyle::LK_Java)
-      return 0;
-    if (RootToken.isAccessSpecifier(false) || RootToken.isObjCAccessSpecifier())
-      return Style.AccessModifierOffset;
-    return 0;
-  }
-
   /// \brief Add a new line and the required indent before the first Token
   /// of the \c UnwrappedLine if there was no structural parsing error.
   void formatFirstToken(FormatToken &RootToken,
                         const AnnotatedLine *PreviousLine, unsigned IndentLevel,
                         unsigned Indent, bool InPPDirective);
 
-  /// \brief Get the indent of \p Level from \p IndentForLevel.
-  ///
-  /// \p IndentForLevel must contain the indent for the level \c l
-  /// at \p IndentForLevel[l], or a value < 0 if the indent for
-  /// that level is unknown.
-  unsigned getIndent(ArrayRef<int> IndentForLevel, unsigned Level);
-
-  void join(AnnotatedLine &A, const AnnotatedLine &B);
-
-  unsigned getColumnLimit(bool InPPDirective) const {
-    // In preprocessor directives reserve two chars for trailing " \"
-    return Style.ColumnLimit - (InPPDirective ? 2 : 0);
-  }
-
-  struct CompareLineStatePointers {
-    bool operator()(LineState *obj1, LineState *obj2) const {
-      return *obj1 < *obj2;
-    }
-  };
-
-  /// \brief Analyze the entire solution space starting from \p InitialState.
-  ///
-  /// This implements a variant of Dijkstra's algorithm on the graph that spans
-  /// the solution space (\c LineStates are the nodes). The algorithm tries to
-  /// find the shortest path (the one with lowest penalty) from \p InitialState
-  /// to a state where all tokens are placed. Returns the penalty.
-  ///
-  /// If \p DryRun is \c false, directly applies the changes.
-  unsigned analyzeSolutionSpace(LineState &InitialState, bool DryRun = false);
-
-  void reconstructPath(LineState &State, StateNode *Current);
-
-  /// \brief Add the following state to the analysis queue \c Queue.
-  ///
-  /// Assume the current state is \p PreviousNode and has been reached with a
-  /// penalty of \p Penalty. Insert a line break if \p NewLine is \c true.
-  void addNextStateToQueue(unsigned Penalty, StateNode *PreviousNode,
-                           bool NewLine, unsigned *Count, QueueType *Queue);
-
-  /// \brief If the \p State's next token is an r_brace closing a nested block,
-  /// format the nested block before it.
-  ///
-  /// Returns \c true if all children could be placed successfully and adapts
-  /// \p Penalty as well as \p State. If \p DryRun is false, also directly
-  /// creates changes using \c Whitespaces.
-  ///
-  /// The crucial idea here is that children always get formatted upon
-  /// encountering the closing brace right after the nested block. Now, if we
-  /// are currently trying to keep the "}" on the same line (i.e. \p NewLine is
-  /// \c false), the entire block has to be kept on the same line (which is only
-  /// possible if it fits on the line, only contains a single statement, etc.
-  ///
-  /// If \p NewLine is true, we format the nested block on separate lines, i.e.
-  /// break after the "{", format all lines with correct indentation and the put
-  /// the closing "}" on yet another new line.
-  ///
-  /// This enables us to keep the simple structure of the
-  /// \c UnwrappedLineFormatter, where we only have two options for each token:
-  /// break or don't break.
-  bool formatChildren(LineState &State, bool NewLine, bool DryRun,
-                      unsigned &Penalty);
-
-  ContinuationIndenter *Indenter;
-  WhitespaceManager *Whitespaces;
-  FormatStyle Style;
-
-  llvm::SpecificBumpPtrAllocator<StateNode> Allocator;
+  /// \brief Returns the column limit for a line, taking into account whether we
+  /// need an escaped newline due to a continued preprocessor directive.
+  unsigned getColumnLimit(bool InPPDirective, const AnnotatedLine *NextLine) const;
 
   // Cache to store the penalty of formatting a vector of AnnotatedLines
   // starting from a specific additional offset. Improves performance if there
   // are many nested blocks.
   std::map<std::pair<const SmallVectorImpl<AnnotatedLine *> *, unsigned>,
            unsigned> PenaltyCache;
+
+  ContinuationIndenter *Indenter;
+  WhitespaceManager *Whitespaces;
+  const FormatStyle &Style;
+  const AdditionalKeywords &Keywords;
+  bool *IncompleteFormat;
 };
 } // end namespace format
 } // end namespace clang
diff --git a/lib/Format/UnwrappedLineParser.cpp b/lib/Format/UnwrappedLineParser.cpp
index ec04af5231be..939528fbffe5 100644
--- a/lib/Format/UnwrappedLineParser.cpp
+++ b/lib/Format/UnwrappedLineParser.cpp
@@ -14,7 +14,9 @@
 //===----------------------------------------------------------------------===//
 
 #include "UnwrappedLineParser.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
 
 #define DEBUG_TYPE "format-parser"
 
@@ -56,22 +58,20 @@ private:
 class ScopedMacroState : public FormatTokenSource {
 public:
   ScopedMacroState(UnwrappedLine &Line, FormatTokenSource *&TokenSource,
-                   FormatToken *&ResetToken, bool &StructuralError)
+                   FormatToken *&ResetToken)
       : Line(Line), TokenSource(TokenSource), ResetToken(ResetToken),
         PreviousLineLevel(Line.Level), PreviousTokenSource(TokenSource),
-        StructuralError(StructuralError),
-        PreviousStructuralError(StructuralError), Token(nullptr) {
+        Token(nullptr) {
     TokenSource = this;
     Line.Level = 0;
     Line.InPPDirective = true;
   }
 
-  ~ScopedMacroState() {
+  ~ScopedMacroState() override {
     TokenSource = PreviousTokenSource;
     ResetToken = Token;
     Line.InPPDirective = false;
     Line.Level = PreviousLineLevel;
-    StructuralError = PreviousStructuralError;
   }
 
   FormatToken *getNextToken() override {
@@ -110,8 +110,6 @@ private:
   FormatToken *&ResetToken;
   unsigned PreviousLineLevel;
   FormatTokenSource *PreviousTokenSource;
-  bool &StructuralError;
-  bool PreviousStructuralError;
 
   FormatToken *Token;
 };
@@ -206,9 +204,8 @@ UnwrappedLineParser::UnwrappedLineParser(const FormatStyle &Style,
                                          ArrayRef<FormatToken *> Tokens,
                                          UnwrappedLineConsumer &Callback)
     : Line(new UnwrappedLine), MustBreakBeforeNextToken(false),
-      CurrentLines(&Lines), StructuralError(false), Style(Style),
-      Keywords(Keywords), Tokens(nullptr), Callback(Callback),
-      AllTokens(Tokens), PPBranchLevel(-1) {}
+      CurrentLines(&Lines), Style(Style), Keywords(Keywords), Tokens(nullptr),
+      Callback(Callback), AllTokens(Tokens), PPBranchLevel(-1) {}
 
 void UnwrappedLineParser::reset() {
   PPBranchLevel = -1;
@@ -219,11 +216,10 @@ void UnwrappedLineParser::reset() {
   PreprocessorDirectives.clear();
   CurrentLines = &Lines;
   DeclarationScopeStack.clear();
-  StructuralError = false;
   PPStack.clear();
 }
 
-bool UnwrappedLineParser::parse() {
+void UnwrappedLineParser::parse() {
   IndexedTokenSource TokenSource(AllTokens);
   do {
     DEBUG(llvm::dbgs() << "----\n");
@@ -256,13 +252,15 @@ bool UnwrappedLineParser::parse() {
     }
   } while (!PPLevelBranchIndex.empty());
 
-  return StructuralError;
 }
 
 void UnwrappedLineParser::parseFile() {
-  ScopedDeclarationState DeclarationState(
-      *Line, DeclarationScopeStack,
-      /*MustBeDeclaration=*/ !Line->InPPDirective);
+  // The top-level context in a file always has declarations, except for pre-
+  // processor directives and JavaScript files.
+  bool MustBeDeclaration =
+      !Line->InPPDirective && Style.Language != FormatStyle::LK_JavaScript;
+  ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
+                                          MustBeDeclaration);
   parseLevel(/*HasOpeningBrace=*/false);
   // Make sure to format the remaining tokens.
   flushComments(true);
@@ -286,7 +284,6 @@ void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) {
     case tok::r_brace:
       if (HasOpeningBrace)
         return;
-      StructuralError = true;
       nextToken();
       addUnwrappedLine();
       break;
@@ -305,7 +302,7 @@ void UnwrappedLineParser::parseLevel(bool HasOpeningBrace) {
   } while (!eof());
 }
 
-void UnwrappedLineParser::calculateBraceTypes() {
+void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) {
   // We'll parse forward through the tokens until we hit
   // a closing brace or eof - note that getNextToken() will
   // parse macros, so this will magically work inside macro
@@ -328,6 +325,7 @@ void UnwrappedLineParser::calculateBraceTypes() {
 
     switch (Tok->Tok.getKind()) {
     case tok::l_brace:
+      Tok->BlockKind = BK_Unknown;
       LBraceStack.push_back(Tok);
       break;
     case tok::r_brace:
@@ -351,9 +349,11 @@ void UnwrappedLineParser::calculateBraceTypes() {
             //
             // We exclude + and - as they can be ObjC visibility modifiers.
             ProbablyBracedList =
-                NextTok->isOneOf(tok::comma, tok::semi, tok::period, tok::colon,
+                NextTok->isOneOf(tok::comma, tok::period, tok::colon,
                                  tok::r_paren, tok::r_square, tok::l_brace,
                                  tok::l_paren, tok::ellipsis) ||
+                (NextTok->is(tok::semi) &&
+                 (!ExpectClassBody || LBraceStack.size() != 1)) ||
                 (NextTok->isBinaryOperator() && !NextIsObjCMethod);
           }
           if (ProbablyBracedList) {
@@ -374,6 +374,7 @@ void UnwrappedLineParser::calculateBraceTypes() {
     case tok::kw_for:
     case tok::kw_switch:
     case tok::kw_try:
+    case tok::kw___try:
       if (!LBraceStack.empty())
         LBraceStack.back()->BlockKind = BK_Block;
       break;
@@ -407,7 +408,6 @@ void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel,
 
   if (!FormatTok->Tok.is(tok::r_brace)) {
     Line->Level = InitialLevel;
-    StructuralError = true;
     return;
   }
 
@@ -417,7 +417,7 @@ void UnwrappedLineParser::parseBlock(bool MustBeDeclaration, bool AddLevel,
   Line->Level = InitialLevel;
 }
 
-static bool IsGoogScope(const UnwrappedLine &Line) {
+static bool isGoogScope(const UnwrappedLine &Line) {
   // FIXME: Closure-library specific stuff should not be hard-coded but be
   // configurable.
   if (Line.Tokens.size() < 4)
@@ -453,12 +453,13 @@ void UnwrappedLineParser::parseChildBlock() {
   nextToken();
   {
     bool GoogScope =
-        Style.Language == FormatStyle::LK_JavaScript && IsGoogScope(*Line);
+        Style.Language == FormatStyle::LK_JavaScript && isGoogScope(*Line);
     ScopedLineState LineState(*this);
     ScopedDeclarationState DeclarationState(*Line, DeclarationScopeStack,
                                             /*MustBeDeclaration=*/false);
     Line->Level += GoogScope ? 0 : 1;
     parseLevel(/*HasOpeningBrace=*/true);
+    flushComments(isOnNewLine(*FormatTok));
     Line->Level -= GoogScope ? 0 : 1;
   }
   nextToken();
@@ -466,7 +467,7 @@ void UnwrappedLineParser::parseChildBlock() {
 
 void UnwrappedLineParser::parsePPDirective() {
   assert(FormatTok->Tok.is(tok::hash) && "'#' expected");
-  ScopedMacroState MacroState(*Line, Tokens, FormatTok, StructuralError);
+  ScopedMacroState MacroState(*Line, Tokens, FormatTok);
   nextToken();
 
   if (!FormatTok->Tok.getIdentifierInfo()) {
@@ -549,6 +550,7 @@ void UnwrappedLineParser::conditionalCompilationEnd() {
 void UnwrappedLineParser::parsePPIf(bool IfDef) {
   nextToken();
   bool IsLiteralFalse = (FormatTok->Tok.isLiteral() &&
+                         FormatTok->Tok.getLiteralData() != nullptr &&
                          StringRef(FormatTok->Tok.getLiteralData(),
                                    FormatTok->Tok.getLength()) == "0") ||
                         FormatTok->Tok.is(tok::kw_false);
@@ -602,7 +604,7 @@ void UnwrappedLineParser::parsePPUnknown() {
 // Here we blacklist certain tokens that are not usually the first token in an
 // unwrapped line. This is used in attempt to distinguish macro calls without
 // trailing semicolons from other constructs split to several lines.
-bool tokenCanStartNewLine(clang::Token Tok) {
+static bool tokenCanStartNewLine(const clang::Token &Tok) {
   // Semicolon can be a null-statement, l_square can be a start of a macro or
   // a C++11 attribute, but this doesn't seem to be common.
   return Tok.isNot(tok::semi) && Tok.isNot(tok::l_brace) &&
@@ -655,6 +657,11 @@ void UnwrappedLineParser::parseStructuralElement() {
       nextToken();
       addUnwrappedLine();
       return;
+    case tok::objc_try:
+      // This branch isn't strictly necessary (the kw_try case below would
+      // do this too after the tok::at is parsed above).  But be explicit.
+      parseTryCatch();
+      return;
     default:
       break;
     }
@@ -662,10 +669,13 @@ void UnwrappedLineParser::parseStructuralElement() {
   case tok::kw_asm:
     nextToken();
     if (FormatTok->is(tok::l_brace)) {
+      FormatTok->Type = TT_InlineASMBrace;
       nextToken();
       while (FormatTok && FormatTok->isNot(tok::eof)) {
         if (FormatTok->is(tok::r_brace)) {
+          FormatTok->Type = TT_InlineASMBrace;
           nextToken();
+          addUnwrappedLine();
           break;
         }
         FormatTok->Finalized = true;
@@ -686,7 +696,8 @@ void UnwrappedLineParser::parseStructuralElement() {
   case tok::kw_public:
   case tok::kw_protected:
   case tok::kw_private:
-    if (Style.Language == FormatStyle::LK_Java)
+    if (Style.Language == FormatStyle::LK_Java ||
+        Style.Language == FormatStyle::LK_JavaScript)
       nextToken();
     else
       parseAccessSpecifier();
@@ -712,6 +723,7 @@ void UnwrappedLineParser::parseStructuralElement() {
     parseCaseLabel();
     return;
   case tok::kw_try:
+  case tok::kw___try:
     parseTryCatch();
     return;
   case tok::kw_extern:
@@ -725,11 +737,30 @@ void UnwrappedLineParser::parseStructuralElement() {
       }
     }
     break;
+  case tok::kw_export:
+    if (Style.Language == FormatStyle::LK_JavaScript) {
+      parseJavaScriptEs6ImportExport();
+      return;
+    }
+    break;
   case tok::identifier:
-    if (FormatTok->IsForEachMacro) {
+    if (FormatTok->is(TT_ForEachMacro)) {
       parseForOrWhileLoop();
       return;
     }
+    if (Style.Language == FormatStyle::LK_JavaScript &&
+        FormatTok->is(Keywords.kw_import)) {
+      parseJavaScriptEs6ImportExport();
+      return;
+    }
+    if (FormatTok->is(Keywords.kw_signals)) {
+      nextToken();
+      if (FormatTok->is(tok::colon)) {
+        nextToken();
+        addUnwrappedLine();
+      }
+      return;
+    }
     // In all other cases, parse the declaration.
     break;
   default:
@@ -806,26 +837,42 @@ void UnwrappedLineParser::parseStructuralElement() {
       parseTryCatch();
       return;
     case tok::identifier: {
-      StringRef Text = FormatTok->TokenText;
       // Parse function literal unless 'function' is the first token in a line
       // in which case this should be treated as a free-standing function.
-      if (Style.Language == FormatStyle::LK_JavaScript && Text == "function" &&
-          Line->Tokens.size() > 0) {
+      if (Style.Language == FormatStyle::LK_JavaScript &&
+          FormatTok->is(Keywords.kw_function) && Line->Tokens.size() > 0) {
         tryToParseJSFunction();
         break;
       }
+      if ((Style.Language == FormatStyle::LK_JavaScript ||
+           Style.Language == FormatStyle::LK_Java) &&
+          FormatTok->is(Keywords.kw_interface)) {
+        parseRecord();
+        break;
+      }
+
+      StringRef Text = FormatTok->TokenText;
       nextToken();
-      if (Line->Tokens.size() == 1) {
-        if (FormatTok->Tok.is(tok::colon)) {
+      if (Line->Tokens.size() == 1 &&
+          // JS doesn't have macros, and within classes colons indicate fields,
+          // not labels.
+          Style.Language != FormatStyle::LK_JavaScript) {
+        if (FormatTok->Tok.is(tok::colon) && !Line->MustBeDeclaration) {
           parseLabel();
           return;
         }
         // Recognize function-like macro usages without trailing semicolon as
-        // well as free-standing macrose like Q_OBJECT.
+        // well as free-standing macros like Q_OBJECT.
         bool FunctionLike = FormatTok->is(tok::l_paren);
         if (FunctionLike)
           parseParens();
-        if (FormatTok->NewlinesBefore > 0 &&
+
+        bool FollowedByNewline =
+            CommentsBeforeNextToken.empty()
+                ? FormatTok->NewlinesBefore > 0
+                : CommentsBeforeNextToken.front()->NewlinesBefore > 0;
+
+        if (FollowedByNewline &&
             (Text.size() >= 5 || FunctionLike) &&
             tokenCanStartNewLine(FormatTok->Tok) && Text == Text.upper()) {
           addUnwrappedLine();
@@ -835,6 +882,17 @@ void UnwrappedLineParser::parseStructuralElement() {
       break;
     }
     case tok::equal:
+      // Fat arrows (=>) have tok::TokenKind tok::equal but TokenType
+      // TT_JsFatArrow. The always start an expression or a child block if
+      // followed by a curly.
+      if (FormatTok->is(TT_JsFatArrow)) {
+        nextToken();
+        if (FormatTok->is(tok::l_brace)) {
+          parseChildBlock();
+        }
+        break;
+      }
+
       nextToken();
       if (FormatTok->Tok.is(tok::l_brace)) {
         parseBracedList();
@@ -843,6 +901,9 @@ void UnwrappedLineParser::parseStructuralElement() {
     case tok::l_square:
       parseSquare();
       break;
+    case tok::kw_new:
+      parseNew();
+      break;
     default:
       nextToken();
       break;
@@ -952,22 +1013,48 @@ void UnwrappedLineParser::tryToParseJSFunction() {
 
   // Consume function name.
   if (FormatTok->is(tok::identifier))
-      nextToken();
+    nextToken();
 
   if (FormatTok->isNot(tok::l_paren))
     return;
-  nextToken();
-  while (FormatTok->isNot(tok::l_brace)) {
-    // Err on the side of caution in order to avoid consuming the full file in
-    // case of incomplete code.
-    if (!FormatTok->isOneOf(tok::identifier, tok::comma, tok::r_paren,
-                            tok::comment))
-      return;
+
+  // Parse formal parameter list.
+  parseBalanced(tok::l_paren, tok::r_paren);
+
+  if (FormatTok->is(tok::colon)) {
+    // Parse a type definition.
     nextToken();
+
+    // Eat the type declaration. For braced inline object types, balance braces,
+    // otherwise just parse until finding an l_brace for the function body.
+    if (FormatTok->is(tok::l_brace)) {
+      parseBalanced(tok::l_brace, tok::r_brace);
+    } else {
+      while(FormatTok->isNot(tok::l_brace) && !eof()) {
+        nextToken();
+      }
+    }
   }
+
   parseChildBlock();
 }
 
+void UnwrappedLineParser::parseBalanced(tok::TokenKind OpenKind,
+                                        tok::TokenKind CloseKind) {
+  assert(FormatTok->is(OpenKind));
+  nextToken();
+  int Depth = 1;
+  while (Depth > 0 && !eof()) {
+    // Parse the formal parameter list.
+    if (FormatTok->is(OpenKind)) {
+      ++Depth;
+    } else if (FormatTok->is(CloseKind)) {
+      --Depth;
+    }
+    nextToken();
+  }
+}
+
 bool UnwrappedLineParser::tryToParseBracedList() {
   if (FormatTok->BlockKind == BK_Unknown)
     calculateBraceTypes();
@@ -985,10 +1072,19 @@ bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons) {
   // FIXME: Once we have an expression parser in the UnwrappedLineParser,
   // replace this by using parseAssigmentExpression() inside.
   do {
-    if (Style.Language == FormatStyle::LK_JavaScript &&
-        FormatTok->is(Keywords.kw_function)) {
-      tryToParseJSFunction();
-      continue;
+    if (Style.Language == FormatStyle::LK_JavaScript) {
+      if (FormatTok->is(Keywords.kw_function)) {
+        tryToParseJSFunction();
+        continue;
+      } else if (FormatTok->is(TT_JsFatArrow)) {
+        nextToken();
+        // Fat arrows can be followed by simple expressions or by child blocks
+        // in curly braces.
+        if (FormatTok->is(tok::l_brace)){
+          parseChildBlock();
+          continue;
+        }
+      }
     }
     switch (FormatTok->Tok.getKind()) {
     case tok::caret:
@@ -1006,6 +1102,17 @@ bool UnwrappedLineParser::parseBracedList(bool ContinueOnSemicolons) {
       FormatTok->BlockKind = BK_BracedInit;
       parseBracedList();
       break;
+    case tok::r_paren:
+      // JavaScript can just have free standing methods and getters/setters in
+      // object literals. Detect them by a "{" following ")".
+      if (Style.Language == FormatStyle::LK_JavaScript) {
+        nextToken();
+        if (FormatTok->is(tok::l_brace))
+          parseChildBlock();
+        break;
+      }
+      nextToken();
+      break;
     case tok::r_brace:
       nextToken();
       return !HasError;
@@ -1046,9 +1153,8 @@ void UnwrappedLineParser::parseParens() {
       tryToParseLambda();
       break;
     case tok::l_brace:
-      if (!tryToParseBracedList()) {
+      if (!tryToParseBracedList())
         parseChildBlock();
-      }
       break;
     case tok::at:
       nextToken();
@@ -1088,9 +1194,8 @@ void UnwrappedLineParser::parseSquare() {
       parseSquare();
       break;
     case tok::l_brace: {
-      if (!tryToParseBracedList()) {
+      if (!tryToParseBracedList())
         parseChildBlock();
-      }
       break;
     }
     case tok::at:
@@ -1148,7 +1253,7 @@ void UnwrappedLineParser::parseIfThenElse() {
 }
 
 void UnwrappedLineParser::parseTryCatch() {
-  assert(FormatTok->is(tok::kw_try) && "'try' expected");
+  assert(FormatTok->isOneOf(tok::kw_try, tok::kw___try) && "'try' expected");
   nextToken();
   bool NeedsUnwrappedLine = false;
   if (FormatTok->is(tok::colon)) {
@@ -1158,8 +1263,6 @@ void UnwrappedLineParser::parseTryCatch() {
       nextToken();
       if (FormatTok->is(tok::l_paren))
         parseParens();
-      else
-        StructuralError = true;
       if (FormatTok->is(tok::comma))
         nextToken();
     }
@@ -1182,23 +1285,29 @@ void UnwrappedLineParser::parseTryCatch() {
     // The C++ standard requires a compound-statement after a try.
     // If there's none, we try to assume there's a structuralElement
     // and try to continue.
-    StructuralError = true;
     addUnwrappedLine();
     ++Line->Level;
     parseStructuralElement();
     --Line->Level;
   }
-  while (FormatTok->is(tok::kw_catch) ||
-         ((Style.Language == FormatStyle::LK_Java ||
-           Style.Language == FormatStyle::LK_JavaScript) &&
-          FormatTok->is(Keywords.kw_finally))) {
+  while (1) {
+    if (FormatTok->is(tok::at))
+      nextToken();
+    if (!(FormatTok->isOneOf(tok::kw_catch, Keywords.kw___except,
+                             tok::kw___finally) ||
+          ((Style.Language == FormatStyle::LK_Java ||
+            Style.Language == FormatStyle::LK_JavaScript) &&
+           FormatTok->is(Keywords.kw_finally)) ||
+          (FormatTok->Tok.isObjCAtKeyword(tok::objc_catch) ||
+           FormatTok->Tok.isObjCAtKeyword(tok::objc_finally))))
+      break;
     nextToken();
     while (FormatTok->isNot(tok::l_brace)) {
       if (FormatTok->is(tok::l_paren)) {
         parseParens();
         continue;
       }
-      if (FormatTok->isOneOf(tok::semi, tok::r_brace))
+      if (FormatTok->isOneOf(tok::semi, tok::r_brace, tok::eof))
         return;
       nextToken();
     }
@@ -1242,9 +1351,33 @@ void UnwrappedLineParser::parseNamespace() {
   // FIXME: Add error handling.
 }
 
+void UnwrappedLineParser::parseNew() {
+  assert(FormatTok->is(tok::kw_new) && "'new' expected");
+  nextToken();
+  if (Style.Language != FormatStyle::LK_Java)
+    return;
+
+  // In Java, we can parse everything up to the parens, which aren't optional.
+  do {
+    // There should not be a ;, { or } before the new's open paren.
+    if (FormatTok->isOneOf(tok::semi, tok::l_brace, tok::r_brace))
+      return;
+
+    // Consume the parens.
+    if (FormatTok->is(tok::l_paren)) {
+      parseParens();
+
+      // If there is a class body of an anonymous class, consume that as child.
+      if (FormatTok->is(tok::l_brace))
+        parseChildBlock();
+      return;
+    }
+    nextToken();
+  } while (!eof());
+}
+
 void UnwrappedLineParser::parseForOrWhileLoop() {
-  assert((FormatTok->Tok.is(tok::kw_for) || FormatTok->Tok.is(tok::kw_while) ||
-          FormatTok->IsForEachMacro) &&
+  assert(FormatTok->isOneOf(tok::kw_for, tok::kw_while, TT_ForEachMacro) &&
          "'for', 'while' or foreach macro expected");
   nextToken();
   if (FormatTok->Tok.is(tok::l_paren))
@@ -1304,6 +1437,8 @@ void UnwrappedLineParser::parseLabel() {
     }
     addUnwrappedLine();
   } else {
+    if (FormatTok->is(tok::semi))
+      nextToken();
     addUnwrappedLine();
   }
   Line->Level = OldLineLevel;
@@ -1338,8 +1473,7 @@ void UnwrappedLineParser::parseSwitch() {
 void UnwrappedLineParser::parseAccessSpecifier() {
   nextToken();
   // Understand Qt's slots.
-  if (FormatTok->is(tok::identifier) &&
-      (FormatTok->TokenText == "slots" || FormatTok->TokenText == "Q_SLOTS"))
+  if (FormatTok->isOneOf(Keywords.kw_slots, Keywords.kw_qslots))
     nextToken();
   // Otherwise, we don't know what it is, and we'd better keep the next token.
   if (FormatTok->Tok.is(tok::colon))
@@ -1455,37 +1589,45 @@ void UnwrappedLineParser::parseJavaEnumBody() {
 void UnwrappedLineParser::parseRecord() {
   const FormatToken &InitialToken = *FormatTok;
   nextToken();
-  if (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::kw___attribute,
-                         tok::kw___declspec, tok::kw_alignas)) {
+
+
+  // The actual identifier can be a nested name specifier, and in macros
+  // it is often token-pasted.
+  while (FormatTok->isOneOf(tok::identifier, tok::coloncolon, tok::hashhash,
+                            tok::kw___attribute, tok::kw___declspec,
+                            tok::kw_alignas) ||
+         ((Style.Language == FormatStyle::LK_Java ||
+           Style.Language == FormatStyle::LK_JavaScript) &&
+          FormatTok->isOneOf(tok::period, tok::comma))) {
+    bool IsNonMacroIdentifier =
+        FormatTok->is(tok::identifier) &&
+        FormatTok->TokenText != FormatTok->TokenText.upper();
     nextToken();
     // We can have macros or attributes in between 'class' and the class name.
-    if (FormatTok->Tok.is(tok::l_paren)) {
+    if (!IsNonMacroIdentifier && FormatTok->Tok.is(tok::l_paren))
       parseParens();
-    }
-    // The actual identifier can be a nested name specifier, and in macros
-    // it is often token-pasted.
-    while (FormatTok->is(tok::identifier) || FormatTok->is(tok::coloncolon) ||
-           FormatTok->is(tok::hashhash) ||
-           (Style.Language == FormatStyle::LK_Java &&
-            FormatTok->isOneOf(tok::period, tok::comma)))
-      nextToken();
+  }
 
-    // Note that parsing away template declarations here leads to incorrectly
-    // accepting function declarations as record declarations.
-    // In general, we cannot solve this problem. Consider:
-    // class A<int> B() {}
-    // which can be a function definition or a class definition when B() is a
-    // macro. If we find enough real-world cases where this is a problem, we
-    // can parse for the 'template' keyword in the beginning of the statement,
-    // and thus rule out the record production in case there is no template
-    // (this would still leave us with an ambiguity between template function
-    // and class declarations).
-    if (FormatTok->Tok.is(tok::colon) || FormatTok->Tok.is(tok::less)) {
-      while (!eof() && FormatTok->Tok.isNot(tok::l_brace)) {
-        if (FormatTok->Tok.is(tok::semi))
-          return;
-        nextToken();
+  // Note that parsing away template declarations here leads to incorrectly
+  // accepting function declarations as record declarations.
+  // In general, we cannot solve this problem. Consider:
+  // class A<int> B() {}
+  // which can be a function definition or a class definition when B() is a
+  // macro. If we find enough real-world cases where this is a problem, we
+  // can parse for the 'template' keyword in the beginning of the statement,
+  // and thus rule out the record production in case there is no template
+  // (this would still leave us with an ambiguity between template function
+  // and class declarations).
+  if (FormatTok->isOneOf(tok::colon, tok::less)) {
+    while (!eof()) {
+      if (FormatTok->is(tok::l_brace)) {
+        calculateBraceTypes(/*ExpectClassBody=*/true);
+        if (!tryToParseBracedList())
+          break;
       }
+      if (FormatTok->Tok.is(tok::semi))
+        return;
+      nextToken();
     }
   }
   if (FormatTok->Tok.is(tok::l_brace)) {
@@ -1498,8 +1640,9 @@ void UnwrappedLineParser::parseRecord() {
   // We fall through to parsing a structural element afterwards, so
   // class A {} n, m;
   // will end up in one unwrapped line.
-  // This does not apply for Java.
-  if (Style.Language == FormatStyle::LK_Java)
+  // This does not apply for Java and JavaScript.
+  if (Style.Language == FormatStyle::LK_Java ||
+      Style.Language == FormatStyle::LK_JavaScript)
     addUnwrappedLine();
 }
 
@@ -1578,6 +1721,35 @@ void UnwrappedLineParser::parseObjCProtocol() {
   parseObjCUntilAtEnd();
 }
 
+void UnwrappedLineParser::parseJavaScriptEs6ImportExport() {
+  assert(FormatTok->isOneOf(Keywords.kw_import, tok::kw_export));
+  nextToken();
+
+  // Consume the "default" in "export default class/function".
+  if (FormatTok->is(tok::kw_default))
+    nextToken();
+
+  // Consume "function" and "default function", so that these get parsed as
+  // free-standing JS functions, i.e. do not require a trailing semicolon.
+  if (FormatTok->is(Keywords.kw_function)) {
+    nextToken();
+    return;
+  }
+
+  if (FormatTok->isOneOf(tok::kw_const, tok::kw_class, Keywords.kw_var))
+    return; // Fall through to parsing the corresponding structure.
+
+  if (FormatTok->is(tok::l_brace)) {
+    FormatTok->BlockKind = BK_Block;
+    parseBracedList();
+  }
+
+  while (!eof() && FormatTok->isNot(tok::semi) &&
+         FormatTok->isNot(tok::l_brace)) {
+    nextToken();
+  }
+}
+
 LLVM_ATTRIBUTE_UNUSED static void printDebugInfo(const UnwrappedLine &Line,
                                                  StringRef Prefix = "") {
   llvm::dbgs() << Prefix << "Line(" << Line.Level << ")"
@@ -1634,14 +1806,12 @@ void UnwrappedLineParser::flushComments(bool NewlineBeforeNext) {
            I = CommentsBeforeNextToken.begin(),
            E = CommentsBeforeNextToken.end();
        I != E; ++I) {
-    if (isOnNewLine(**I) && JustComments) {
+    if (isOnNewLine(**I) && JustComments)
       addUnwrappedLine();
-    }
     pushToken(*I);
   }
-  if (NewlineBeforeNext && JustComments) {
+  if (NewlineBeforeNext && JustComments)
     addUnwrappedLine();
-  }
   CommentsBeforeNextToken.clear();
 }
 
@@ -1662,8 +1832,7 @@ void UnwrappedLineParser::readToken() {
            (FormatTok->HasUnescapedNewline || FormatTok->IsFirst)) {
       // If there is an unfinished unwrapped line, we flush the preprocessor
       // directives only after that unwrapped line was finished later.
-      bool SwitchToPreprocessorLines =
-          !Line->Tokens.empty() && CurrentLines == &Lines;
+      bool SwitchToPreprocessorLines = !Line->Tokens.empty();
       ScopedLineState BlockState(*this, SwitchToPreprocessorLines);
       // Comments stored before the preprocessor directive need to be output
       // before the preprocessor directive, at the same level as the
diff --git a/lib/Format/UnwrappedLineParser.h b/lib/Format/UnwrappedLineParser.h
index 3218afecad30..6a6e56fea028 100644
--- a/lib/Format/UnwrappedLineParser.h
+++ b/lib/Format/UnwrappedLineParser.h
@@ -65,8 +65,7 @@ public:
                       ArrayRef<FormatToken *> Tokens,
                       UnwrappedLineConsumer &Callback);
 
-  /// Returns true in case of a structural error.
-  bool parse();
+  void parse();
 
 private:
   void reset();
@@ -95,6 +94,7 @@ private:
   void parseCaseLabel();
   void parseSwitch();
   void parseNamespace();
+  void parseNew();
   void parseAccessSpecifier();
   void parseEnum();
   void parseJavaEnumBody();
@@ -103,16 +103,22 @@ private:
   void parseObjCUntilAtEnd();
   void parseObjCInterfaceOrImplementation();
   void parseObjCProtocol();
+  void parseJavaScriptEs6ImportExport();
   bool tryToParseLambda();
   bool tryToParseLambdaIntroducer();
   void tryToParseJSFunction();
+  /// \brief Parses tokens until encountering the CloseKind token, but balances
+  /// tokens when encountering more OpenKind tokens. Useful for e.g. parsing a
+  /// curly brace delimited block that can contain nested blocks.
+  /// The parser must be positioned on a token of OpenKind.
+  void parseBalanced(tok::TokenKind OpenKind, tok::TokenKind CloseKind);
   void addUnwrappedLine();
   bool eof() const;
   void nextToken();
   void readToken();
   void flushComments(bool NewlineBeforeNext);
   void pushToken(FormatToken *Tok);
-  void calculateBraceTypes();
+  void calculateBraceTypes(bool ExpectClassBody = false);
 
   // Marks a conditional compilation edge (for example, an '#if', '#ifdef',
   // '#else' or merge conflict marker). If 'Unreachable' is true, assumes
@@ -156,10 +162,6 @@ private:
   // whether we are in a compound statement or not.
   std::vector<bool> DeclarationScopeStack;
 
-  // Will be true if we encounter an error that leads to possibily incorrect
-  // indentation levels.
-  bool StructuralError;
-
   const FormatStyle &Style;
   const AdditionalKeywords &Keywords;
 
diff --git a/lib/Format/WhitespaceManager.cpp b/lib/Format/WhitespaceManager.cpp
index bf1207e59c90..4baaab1c9877 100644
--- a/lib/Format/WhitespaceManager.cpp
+++ b/lib/Format/WhitespaceManager.cpp
@@ -36,7 +36,9 @@ WhitespaceManager::Change::Change(
       PreviousLinePostfix(PreviousLinePostfix),
       CurrentLinePrefix(CurrentLinePrefix), Kind(Kind),
       ContinuesPPDirective(ContinuesPPDirective), IndentLevel(IndentLevel),
-      Spaces(Spaces) {}
+      Spaces(Spaces), IsTrailingComment(false), TokenLength(0),
+      PreviousEndOfTokenColumn(0), EscapedNewlineColumn(0),
+      StartOfBlockComment(nullptr), IndentationOffset(0) {}
 
 void WhitespaceManager::reset() {
   Changes.clear();
@@ -91,6 +93,7 @@ const tooling::Replacements &WhitespaceManager::generateReplacements() {
 
   std::sort(Changes.begin(), Changes.end(), Change::IsBeforeInFile(SourceMgr));
   calculateLineBreakInformation();
+  alignConsecutiveAssignments();
   alignTrailingComments();
   alignEscapedNewlines();
   generateChanges();
@@ -139,6 +142,96 @@ void WhitespaceManager::calculateLineBreakInformation() {
   }
 }
 
+// Walk through all of the changes and find sequences of "=" to align.  To do
+// so, keep track of the lines and whether or not an "=" was found on align. If
+// a "=" is found on a line, extend the current sequence. If the current line
+// cannot be part of a sequence, e.g. because there is an empty line before it
+// or it contains non-assignments, finalize the previous sequence.
+void WhitespaceManager::alignConsecutiveAssignments() {
+  if (!Style.AlignConsecutiveAssignments)
+    return;
+
+  unsigned MinColumn = 0;
+  unsigned StartOfSequence = 0;
+  unsigned EndOfSequence = 0;
+  bool FoundAssignmentOnLine = false;
+  bool FoundLeftParenOnLine = false;
+  unsigned CurrentLine = 0;
+
+  auto AlignSequence = [&] {
+    alignConsecutiveAssignments(StartOfSequence, EndOfSequence, MinColumn);
+    MinColumn = 0;
+    StartOfSequence = 0;
+    EndOfSequence = 0;
+  };
+
+  for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
+    if (Changes[i].NewlinesBefore != 0) {
+      CurrentLine += Changes[i].NewlinesBefore;
+      if (StartOfSequence > 0 &&
+          (Changes[i].NewlinesBefore > 1 || !FoundAssignmentOnLine)) {
+        EndOfSequence = i;
+        AlignSequence();
+      }
+      FoundAssignmentOnLine = false;
+      FoundLeftParenOnLine = false;
+    }
+
+    if ((Changes[i].Kind == tok::equal &&
+         (FoundAssignmentOnLine || ((Changes[i].NewlinesBefore > 0 ||
+                                     Changes[i + 1].NewlinesBefore > 0)))) ||
+        (!FoundLeftParenOnLine && Changes[i].Kind == tok::r_paren)) {
+      if (StartOfSequence > 0)
+        AlignSequence();
+    } else if (Changes[i].Kind == tok::l_paren) {
+      FoundLeftParenOnLine = true;
+      if (!FoundAssignmentOnLine && StartOfSequence > 0)
+        AlignSequence();
+    } else if (!FoundAssignmentOnLine && !FoundLeftParenOnLine &&
+               Changes[i].Kind == tok::equal) {
+      FoundAssignmentOnLine = true;
+      EndOfSequence = i;
+      if (StartOfSequence == 0)
+        StartOfSequence = i;
+
+      unsigned ChangeMinColumn = Changes[i].StartOfTokenColumn;
+      MinColumn = std::max(MinColumn, ChangeMinColumn);
+    }
+  }
+
+  if (StartOfSequence > 0) {
+    EndOfSequence = Changes.size();
+    AlignSequence();
+  }
+}
+
+void WhitespaceManager::alignConsecutiveAssignments(unsigned Start,
+                                                    unsigned End,
+                                                    unsigned Column) {
+  bool AlignedAssignment = false;
+  int PreviousShift = 0;
+  for (unsigned i = Start; i != End; ++i) {
+    int Shift = 0;
+    if (Changes[i].NewlinesBefore > 0)
+      AlignedAssignment = false;
+    if (!AlignedAssignment && Changes[i].Kind == tok::equal) {
+      Shift = Column - Changes[i].StartOfTokenColumn;
+      AlignedAssignment = true;
+      PreviousShift = Shift;
+    }
+    assert(Shift >= 0);
+    Changes[i].Spaces += Shift;
+    if (i + 1 != Changes.size())
+      Changes[i + 1].PreviousEndOfTokenColumn += Shift;
+    Changes[i].StartOfTokenColumn += Shift;
+    if (AlignedAssignment) {
+      Changes[i].StartOfTokenColumn += PreviousShift;
+      if (i + 1 != Changes.size())
+        Changes[i + 1].PreviousEndOfTokenColumn += PreviousShift;
+    }
+  }
+}
+
 void WhitespaceManager::alignTrailingComments() {
   unsigned MinColumn = 0;
   unsigned MaxColumn = UINT_MAX;
@@ -264,6 +357,11 @@ void WhitespaceManager::alignEscapedNewlines(unsigned Start, unsigned End,
 void WhitespaceManager::generateChanges() {
   for (unsigned i = 0, e = Changes.size(); i != e; ++i) {
     const Change &C = Changes[i];
+    if (i > 0) {
+      assert(Changes[i - 1].OriginalWhitespaceRange.getBegin() !=
+                 C.OriginalWhitespaceRange.getBegin() &&
+             "Generating two replacements for the same location");
+    }
     if (C.CreateReplacement) {
       std::string ReplacementText = C.PreviousLinePostfix;
       if (C.ContinuesPPDirective)
diff --git a/lib/Format/WhitespaceManager.h b/lib/Format/WhitespaceManager.h
index 28730d457eba..4bfc813b2c34 100644
--- a/lib/Format/WhitespaceManager.h
+++ b/lib/Format/WhitespaceManager.h
@@ -164,6 +164,13 @@ private:
   /// \c EscapedNewlineColumn for the first tokens or token parts in a line.
   void calculateLineBreakInformation();
 
+  /// \brief Align consecutive assignments over all \c Changes.
+  void alignConsecutiveAssignments();
+
+  /// \brief Align consecutive assignments from change \p Start to change \p End at
+  /// the specified \p Column.
+  void alignConsecutiveAssignments(unsigned Start, unsigned End, unsigned Column);
+
   /// \brief Align trailing comments over all \c Changes.
   void alignTrailingComments();
 
diff --git a/lib/Frontend/ASTConsumers.cpp b/lib/Frontend/ASTConsumers.cpp
index f53c614b0a30..52776b6a8483 100644
--- a/lib/Frontend/ASTConsumers.cpp
+++ b/lib/Frontend/ASTConsumers.cpp
@@ -21,7 +21,6 @@
 #include "clang/Basic/Diagnostic.h"
 #include "clang/Basic/FileManager.h"
 #include "clang/Basic/SourceManager.h"
-#include "llvm/IR/Module.h"
 #include "llvm/Support/Path.h"
 #include "llvm/Support/Timer.h"
 #include "llvm/Support/raw_ostream.h"
diff --git a/lib/Frontend/ASTMerge.cpp b/lib/Frontend/ASTMerge.cpp
index 216ac6a16984..b84df94ef806 100644
--- a/lib/Frontend/ASTMerge.cpp
+++ b/lib/Frontend/ASTMerge.cpp
@@ -57,6 +57,7 @@ void ASTMergeAction::ExecuteAction() {
                          /*MinimalImport=*/false);
 
     TranslationUnitDecl *TU = Unit->getASTContext().getTranslationUnitDecl();
+    CI.getASTConsumer().Initialize(CI.getASTContext());
     for (auto *D : TU->decls()) {
       // Don't re-import __va_list_tag, __builtin_va_list.
       if (const auto *ND = dyn_cast<NamedDecl>(D))
@@ -64,7 +65,12 @@ void ASTMergeAction::ExecuteAction() {
           if (II->isStr("__va_list_tag") || II->isStr("__builtin_va_list"))
             continue;
       
-      Importer.Import(D);
+      Decl *ToD = Importer.Import(D);
+    
+      if (ToD) {
+        DeclGroupRef DGR(ToD);
+        CI.getASTConsumer().HandleTopLevelDecl(DGR);
+      }
     }
   }
 
diff --git a/lib/Frontend/ASTUnit.cpp b/lib/Frontend/ASTUnit.cpp
index a3998fa351de..7226344b754b 100644
--- a/lib/Frontend/ASTUnit.cpp
+++ b/lib/Frontend/ASTUnit.cpp
@@ -120,11 +120,10 @@ static OnDiskDataMap &getOnDiskDataMap() {
 static void cleanupOnDiskMapAtExit() {
   // Use the mutex because there can be an alive thread destroying an ASTUnit.
   llvm::MutexGuard Guard(getOnDiskMutex());
-  OnDiskDataMap &M = getOnDiskDataMap();
-  for (OnDiskDataMap::iterator I = M.begin(), E = M.end(); I != E; ++I) {
+  for (const auto &I : getOnDiskDataMap()) {
     // We don't worry about freeing the memory associated with OnDiskDataMap.
     // All we care about is erasing stale files.
-    I->second->Cleanup();
+    I.second->Cleanup();
   }
 }
 
@@ -151,7 +150,7 @@ static void removeOnDiskEntry(const ASTUnit *AU) {
   OnDiskDataMap::iterator I = M.find(AU);
   if (I != M.end()) {
     I->second->Cleanup();
-    M.erase(AU);
+    M.erase(I);
   }
 }
 
@@ -164,8 +163,8 @@ static const std::string &getPreambleFile(const ASTUnit *AU) {
 }
 
 void OnDiskData::CleanTemporaryFiles() {
-  for (unsigned I = 0, N = TemporaryFiles.size(); I != N; ++I)
-    llvm::sys::fs::remove(TemporaryFiles[I]);
+  for (StringRef File : TemporaryFiles)
+    llvm::sys::fs::remove(File);
   TemporaryFiles.clear();
 }
 
@@ -354,26 +353,24 @@ void ASTUnit::CacheCodeCompletionResults() {
   
   // Translate global code completions into cached completions.
   llvm::DenseMap<CanQualType, unsigned> CompletionTypes;
-  
-  for (unsigned I = 0, N = Results.size(); I != N; ++I) {
-    switch (Results[I].Kind) {
+
+  for (Result &R : Results) {
+    switch (R.Kind) {
     case Result::RK_Declaration: {
       bool IsNestedNameSpecifier = false;
       CachedCodeCompletionResult CachedResult;
-      CachedResult.Completion = Results[I].CreateCodeCompletionString(*TheSema,
-                                                    *CachedCompletionAllocator,
-                                                    CCTUInfo,
-                                          IncludeBriefCommentsInCodeCompletion);
-      CachedResult.ShowInContexts = getDeclShowContexts(Results[I].Declaration,
-                                                        Ctx->getLangOpts(),
-                                                        IsNestedNameSpecifier);
-      CachedResult.Priority = Results[I].Priority;
-      CachedResult.Kind = Results[I].CursorKind;
-      CachedResult.Availability = Results[I].Availability;
+      CachedResult.Completion = R.CreateCodeCompletionString(
+          *TheSema, *CachedCompletionAllocator, CCTUInfo,
+          IncludeBriefCommentsInCodeCompletion);
+      CachedResult.ShowInContexts = getDeclShowContexts(
+          R.Declaration, Ctx->getLangOpts(), IsNestedNameSpecifier);
+      CachedResult.Priority = R.Priority;
+      CachedResult.Kind = R.CursorKind;
+      CachedResult.Availability = R.Availability;
 
       // Keep track of the type of this completion in an ASTContext-agnostic 
       // way.
-      QualType UsageType = getDeclUsageType(*Ctx, Results[I].Declaration);
+      QualType UsageType = getDeclUsageType(*Ctx, R.Declaration);
       if (UsageType.isNull()) {
         CachedResult.TypeClass = STC_Void;
         CachedResult.Type = 0;
@@ -398,8 +395,8 @@ void ASTUnit::CacheCodeCompletionResults() {
       CachedCompletionResults.push_back(CachedResult);
       
       /// Handle nested-name-specifiers in C++.
-      if (TheSema->Context.getLangOpts().CPlusPlus && 
-          IsNestedNameSpecifier && !Results[I].StartsNestedNameSpecifier) {
+      if (TheSema->Context.getLangOpts().CPlusPlus && IsNestedNameSpecifier &&
+          !R.StartsNestedNameSpecifier) {
         // The contexts in which a nested-name-specifier can appear in C++.
         uint64_t NNSContexts
           = (1LL << CodeCompletionContext::CCC_TopLevel)
@@ -415,8 +412,8 @@ void ASTUnit::CacheCodeCompletionResults() {
           | (1LL << CodeCompletionContext::CCC_PotentiallyQualifiedName)
           | (1LL << CodeCompletionContext::CCC_ParenthesizedExpression);
 
-        if (isa<NamespaceDecl>(Results[I].Declaration) ||
-            isa<NamespaceAliasDecl>(Results[I].Declaration))
+        if (isa<NamespaceDecl>(R.Declaration) ||
+            isa<NamespaceAliasDecl>(R.Declaration))
           NNSContexts |= (1LL << CodeCompletionContext::CCC_Namespace);
 
         if (unsigned RemainingContexts 
@@ -424,12 +421,10 @@ void ASTUnit::CacheCodeCompletionResults() {
           // If there any contexts where this completion can be a 
           // nested-name-specifier but isn't already an option, create a 
           // nested-name-specifier completion.
-          Results[I].StartsNestedNameSpecifier = true;
-          CachedResult.Completion 
-            = Results[I].CreateCodeCompletionString(*TheSema,
-                                                    *CachedCompletionAllocator,
-                                                    CCTUInfo,
-                                        IncludeBriefCommentsInCodeCompletion);
+          R.StartsNestedNameSpecifier = true;
+          CachedResult.Completion = R.CreateCodeCompletionString(
+              *TheSema, *CachedCompletionAllocator, CCTUInfo,
+              IncludeBriefCommentsInCodeCompletion);
           CachedResult.ShowInContexts = RemainingContexts;
           CachedResult.Priority = CCP_NestedNameSpecifier;
           CachedResult.TypeClass = STC_Void;
@@ -448,11 +443,9 @@ void ASTUnit::CacheCodeCompletionResults() {
       
     case Result::RK_Macro: {
       CachedCodeCompletionResult CachedResult;
-      CachedResult.Completion 
-        = Results[I].CreateCodeCompletionString(*TheSema,
-                                                *CachedCompletionAllocator,
-                                                CCTUInfo,
-                                          IncludeBriefCommentsInCodeCompletion);
+      CachedResult.Completion = R.CreateCodeCompletionString(
+          *TheSema, *CachedCompletionAllocator, CCTUInfo,
+          IncludeBriefCommentsInCodeCompletion);
       CachedResult.ShowInContexts
         = (1LL << CodeCompletionContext::CCC_TopLevel)
         | (1LL << CodeCompletionContext::CCC_ObjCInterface)
@@ -466,10 +459,10 @@ void ASTUnit::CacheCodeCompletionResults() {
         | (1LL << CodeCompletionContext::CCC_PreprocessorExpression)
         | (1LL << CodeCompletionContext::CCC_ParenthesizedExpression)
         | (1LL << CodeCompletionContext::CCC_OtherWithMacros);
-      
-      CachedResult.Priority = Results[I].Priority;
-      CachedResult.Kind = Results[I].CursorKind;
-      CachedResult.Availability = Results[I].Availability;
+
+      CachedResult.Priority = R.Priority;
+      CachedResult.Kind = R.CursorKind;
+      CachedResult.Availability = R.Availability;
       CachedResult.TypeClass = STC_Void;
       CachedResult.Type = 0;
       CachedCompletionResults.push_back(CachedResult);
@@ -520,8 +513,8 @@ public:
     return false;
   }
 
-  bool ReadTargetOptions(const TargetOptions &TargetOpts,
-                         bool Complain) override {
+  bool ReadTargetOptions(const TargetOptions &TargetOpts, bool Complain,
+                         bool AllowCompatibleDifferences) override {
     // If we've already initialized the target, don't do it again.
     if (Target)
       return false;
@@ -689,8 +682,8 @@ std::unique_ptr<ASTUnit> ASTUnit::LoadFromASTFile(
 
   PreprocessorOptions *PPOpts = new PreprocessorOptions();
 
-  for (unsigned I = 0, N = RemappedFiles.size(); I != N; ++I)
-    PPOpts->addRemappedFile(RemappedFiles[I].first, RemappedFiles[I].second);
+  for (const auto &RemappedFile : RemappedFiles)
+    PPOpts->addRemappedFile(RemappedFile.first, RemappedFile.second);
 
   // Gather Info for preprocessor construction later on.
 
@@ -721,6 +714,13 @@ std::unique_ptr<ASTUnit> ASTUnit::LoadFromASTFile(
       *AST->PP, Context, AST->ASTFileLangOpts, AST->TargetOpts, AST->Target,
       Counter));
 
+  // Attach the AST reader to the AST context as an external AST
+  // source, so that declarations will be deserialized from the
+  // AST file as needed.
+  // We need the external source to be set up before we read the AST, because
+  // eagerly-deserialized declarations may use it.
+  Context.setExternalSource(AST->Reader);
+
   switch (AST->Reader->ReadAST(Filename, serialization::MK_MainFile,
                           SourceLocation(), ASTReader::ARR_None)) {
   case ASTReader::Success:
@@ -740,11 +740,6 @@ std::unique_ptr<ASTUnit> ASTUnit::LoadFromASTFile(
 
   PP.setCounterValue(Counter);
 
-  // Attach the AST reader to the AST context as an external AST
-  // source, so that declarations will be deserialized from the
-  // AST file as needed.
-  Context.setExternalSource(AST->Reader);
-
   // Create an AST consumer, even though it isn't used.
   AST->Consumer.reset(new ASTConsumer);
   
@@ -853,8 +848,8 @@ public:
   }
 
   bool HandleTopLevelDecl(DeclGroupRef D) override {
-    for (DeclGroupRef::iterator it = D.begin(), ie = D.end(); it != ie; ++it)
-      handleTopLevelDecl(*it);
+    for (Decl *TopLevelDecl : D)
+      handleTopLevelDecl(TopLevelDecl);
     return true;
   }
 
@@ -862,8 +857,8 @@ public:
   void HandleInterestingDecl(DeclGroupRef) override {}
 
   void HandleTopLevelDeclInObjCContainer(DeclGroupRef D) override {
-    for (DeclGroupRef::iterator it = D.begin(), ie = D.end(); it != ie; ++it)
-      handleTopLevelDecl(*it);
+    for (Decl *TopLevelDecl : D)
+      handleTopLevelDecl(TopLevelDecl);
   }
 
   ASTMutationListener *GetASTMutationListener() override {
@@ -931,9 +926,8 @@ public:
     Hash = 0;
   }
 
-  bool HandleTopLevelDecl(DeclGroupRef D) override {
-    for (DeclGroupRef::iterator it = D.begin(), ie = D.end(); it != ie; ++it) {
-      Decl *D = *it;
+  bool HandleTopLevelDecl(DeclGroupRef DG) override {
+    for (Decl *D : DG) {
       // FIXME: Currently ObjC method declarations are incorrectly being
       // reported as top-level declarations, even though their DeclContext
       // is the containing ObjC @interface/@implementation.  This is a
@@ -953,8 +947,7 @@ public:
       // parsing into declaration IDs in the precompiled
       // preamble. This will allow us to deserialize those top-level
       // declarations when requested.
-      for (unsigned I = 0, N = TopLevelDecls.size(); I != N; ++I) {
-        Decl *D = TopLevelDecls[I];
+      for (Decl *D : TopLevelDecls) {
         // Invalid top-level decls may not have been serialized.
         if (D->isInvalidDecl())
           continue;
@@ -973,9 +966,9 @@ PrecompilePreambleAction::CreateASTConsumer(CompilerInstance &CI,
                                             StringRef InFile) {
   std::string Sysroot;
   std::string OutputFile;
-  raw_ostream *OS = nullptr;
-  if (GeneratePCHAction::ComputeASTConsumerArguments(CI, InFile, Sysroot,
-                                                     OutputFile, OS))
+  raw_ostream *OS = GeneratePCHAction::ComputeASTConsumerArguments(
+      CI, InFile, Sysroot, OutputFile);
+  if (!OS)
     return nullptr;
 
   if (!CI.getFrontendOpts().RelocatablePCH)
@@ -1009,10 +1002,10 @@ static void checkAndSanitizeDiags(SmallVectorImpl<StoredDiagnostic> &
   // been careful to make sure that the source manager's state
   // before and after are identical, so that we can reuse the source
   // location itself.
-  for (unsigned I = 0, N = StoredDiagnostics.size(); I < N; ++I) {
-    if (StoredDiagnostics[I].getLocation().isValid()) {
-      FullSourceLoc Loc(StoredDiagnostics[I].getLocation(), SM);
-      StoredDiagnostics[I].setLocation(Loc);
+  for (StoredDiagnostic &SD : StoredDiagnostics) {
+    if (SD.getLocation().isValid()) {
+      FullSourceLoc Loc(SD.getLocation(), SM);
+      SD.setLocation(Loc);
     }
   }
 }
@@ -1300,14 +1293,10 @@ makeStandaloneDiagnostic(const LangOptions &LangOpts,
   if (OutDiag.Filename.empty())
     return OutDiag;
   OutDiag.LocOffset = SM.getFileOffset(FileLoc);
-  for (StoredDiagnostic::range_iterator
-         I = InDiag.range_begin(), E = InDiag.range_end(); I != E; ++I) {
-    OutDiag.Ranges.push_back(makeStandaloneRange(*I, SM, LangOpts));
-  }
-  for (StoredDiagnostic::fixit_iterator I = InDiag.fixit_begin(),
-                                        E = InDiag.fixit_end();
-       I != E; ++I)
-    OutDiag.FixIts.push_back(makeStandaloneFixIt(SM, LangOpts, *I));
+  for (const CharSourceRange &Range : InDiag.getRanges())
+    OutDiag.Ranges.push_back(makeStandaloneRange(Range, SM, LangOpts));
+  for (const FixItHint &FixIt : InDiag.getFixIts())
+    OutDiag.FixIts.push_back(makeStandaloneFixIt(SM, LangOpts, FixIt));
 
   return OutDiag;
 }
@@ -1634,11 +1623,10 @@ void ASTUnit::RealizeTopLevelDeclsFromPreamble() {
   std::vector<Decl *> Resolved;
   Resolved.reserve(TopLevelDeclsInPreamble.size());
   ExternalASTSource &Source = *getASTContext().getExternalSource();
-  for (unsigned I = 0, N = TopLevelDeclsInPreamble.size(); I != N; ++I) {
+  for (serialization::DeclID TopLevelDecl : TopLevelDeclsInPreamble) {
     // Resolve the declaration ID to an actual declaration, possibly
     // deserializing the declaration in the process.
-    Decl *D = Source.GetExternalDecl(TopLevelDeclsInPreamble[I]);
-    if (D)
+    if (Decl *D = Source.GetExternalDecl(TopLevelDecl))
       Resolved.push_back(D);
   }
   TopLevelDeclsInPreamble.clear();
@@ -1943,9 +1931,9 @@ ASTUnit *ASTUnit::LoadFromCommandLine(
   }
 
   // Override any files that need remapping
-  for (unsigned I = 0, N = RemappedFiles.size(); I != N; ++I) {
-    CI->getPreprocessorOpts().addRemappedFile(RemappedFiles[I].first,
-                                              RemappedFiles[I].second);
+  for (const auto &RemappedFile : RemappedFiles) {
+    CI->getPreprocessorOpts().addRemappedFile(RemappedFile.first,
+                                              RemappedFile.second);
   }
   PreprocessorOptions &PPOpts = CI->getPreprocessorOpts();
   PPOpts.RemappedFilesKeepOriginalName = RemappedFilesKeepOriginalName;
@@ -2015,9 +2003,9 @@ bool ASTUnit::Reparse(ArrayRef<RemappedFile> RemappedFiles) {
     delete RB.second;
 
   Invocation->getPreprocessorOpts().clearRemappedFiles();
-  for (unsigned I = 0, N = RemappedFiles.size(); I != N; ++I) {
-    Invocation->getPreprocessorOpts().addRemappedFile(RemappedFiles[I].first,
-                                                      RemappedFiles[I].second);
+  for (const auto &RemappedFile : RemappedFiles) {
+    Invocation->getPreprocessorOpts().addRemappedFile(RemappedFile.first,
+                                                      RemappedFile.second);
   }
 
   // If we have a preamble file lying around, or if we might try to
@@ -2375,10 +2363,9 @@ void ASTUnit::CodeComplete(StringRef File, unsigned Line, unsigned Column,
   // Remap files.
   PreprocessorOpts.clearRemappedFiles();
   PreprocessorOpts.RetainRemappedFileBuffers = true;
-  for (unsigned I = 0, N = RemappedFiles.size(); I != N; ++I) {
-    PreprocessorOpts.addRemappedFile(RemappedFiles[I].first,
-                                     RemappedFiles[I].second);
-    OwnedBuffers.push_back(RemappedFiles[I].second);
+  for (const auto &RemappedFile : RemappedFiles) {
+    PreprocessorOpts.addRemappedFile(RemappedFile.first, RemappedFile.second);
+    OwnedBuffers.push_back(RemappedFile.second);
   }
 
   // Use the code completion consumer we were given, but adding any cached
@@ -2446,7 +2433,7 @@ bool ASTUnit::Save(StringRef File) {
   TempPath = File;
   TempPath += "-%%%%%%%%";
   int fd;
-  if (llvm::sys::fs::createUniqueFile(TempPath.str(), fd, TempPath))
+  if (llvm::sys::fs::createUniqueFile(TempPath, fd, TempPath))
     return true;
 
   // FIXME: Can we somehow regenerate the stat cache here, or do we need to 
@@ -2460,8 +2447,8 @@ bool ASTUnit::Save(StringRef File) {
     return true;
   }
 
-  if (llvm::sys::fs::rename(TempPath.str(), File)) {
-    llvm::sys::fs::remove(TempPath.str());
+  if (llvm::sys::fs::rename(TempPath, File)) {
+    llvm::sys::fs::remove(TempPath);
     return true;
   }
 
@@ -2509,9 +2496,8 @@ void ASTUnit::TranslateStoredDiagnostics(
 
   SmallVector<StoredDiagnostic, 4> Result;
   Result.reserve(Diags.size());
-  for (unsigned I = 0, N = Diags.size(); I != N; ++I) {
+  for (const StandaloneDiagnostic &SD : Diags) {
     // Rebuild the StoredDiagnostic.
-    const StandaloneDiagnostic &SD = Diags[I];
     if (SD.Filename.empty())
       continue;
     const FileEntry *FE = FileMgr.getFile(SD.Filename);
@@ -2526,23 +2512,20 @@ void ASTUnit::TranslateStoredDiagnostics(
 
     SmallVector<CharSourceRange, 4> Ranges;
     Ranges.reserve(SD.Ranges.size());
-    for (std::vector<std::pair<unsigned, unsigned> >::const_iterator
-           I = SD.Ranges.begin(), E = SD.Ranges.end(); I != E; ++I) {
-      SourceLocation BL = FileLoc.getLocWithOffset((*I).first);
-      SourceLocation EL = FileLoc.getLocWithOffset((*I).second);
+    for (const auto &Range : SD.Ranges) {
+      SourceLocation BL = FileLoc.getLocWithOffset(Range.first);
+      SourceLocation EL = FileLoc.getLocWithOffset(Range.second);
       Ranges.push_back(CharSourceRange::getCharRange(BL, EL));
     }
 
     SmallVector<FixItHint, 2> FixIts;
     FixIts.reserve(SD.FixIts.size());
-    for (std::vector<StandaloneFixIt>::const_iterator
-           I = SD.FixIts.begin(), E = SD.FixIts.end();
-         I != E; ++I) {
+    for (const StandaloneFixIt &FixIt : SD.FixIts) {
       FixIts.push_back(FixItHint());
       FixItHint &FH = FixIts.back();
-      FH.CodeToInsert = I->CodeToInsert;
-      SourceLocation BL = FileLoc.getLocWithOffset(I->RemoveRange.first);
-      SourceLocation EL = FileLoc.getLocWithOffset(I->RemoveRange.second);
+      FH.CodeToInsert = FixIt.CodeToInsert;
+      SourceLocation BL = FileLoc.getLocWithOffset(FixIt.RemoveRange.first);
+      SourceLocation EL = FileLoc.getLocWithOffset(FixIt.RemoveRange.second);
       FH.RemoveRange = CharSourceRange::getCharRange(BL, EL);
     }
 
@@ -2736,7 +2719,7 @@ SourceLocation ASTUnit::getStartOfMainFileID() {
   return SourceMgr->getLocForStartOfFile(FID);
 }
 
-std::pair<PreprocessingRecord::iterator, PreprocessingRecord::iterator>
+llvm::iterator_range<PreprocessingRecord::iterator>
 ASTUnit::getLocalPreprocessingEntities() const {
   if (isMainFileAST()) {
     serialization::ModuleFile &
@@ -2745,20 +2728,18 @@ ASTUnit::getLocalPreprocessingEntities() const {
   }
 
   if (PreprocessingRecord *PPRec = PP->getPreprocessingRecord())
-    return std::make_pair(PPRec->local_begin(), PPRec->local_end());
+    return llvm::make_range(PPRec->local_begin(), PPRec->local_end());
 
-  return std::make_pair(PreprocessingRecord::iterator(),
-                        PreprocessingRecord::iterator());
+  return llvm::make_range(PreprocessingRecord::iterator(),
+                          PreprocessingRecord::iterator());
 }
 
 bool ASTUnit::visitLocalTopLevelDecls(void *context, DeclVisitorFn Fn) {
   if (isMainFileAST()) {
     serialization::ModuleFile &
       Mod = Reader->getModuleManager().getPrimaryModule();
-    ASTReader::ModuleDeclIterator MDI, MDE;
-    std::tie(MDI, MDE) = Reader->getModuleFileLevelDecls(Mod);
-    for (; MDI != MDE; ++MDI) {
-      if (!Fn(context, *MDI))
+    for (const Decl *D : Reader->getModuleFileLevelDecls(Mod)) {
+      if (!Fn(context, D))
         return false;
     }
 
@@ -2821,11 +2802,8 @@ void ASTUnit::PreambleData::countLines() const {
   if (empty())
     return;
 
-  for (std::vector<char>::const_iterator
-         I = Buffer.begin(), E = Buffer.end(); I != E; ++I) {
-    if (*I == '\n')
-      ++NumLines;
-  }
+  NumLines = std::count(Buffer.begin(), Buffer.end(), '\n');
+
   if (Buffer.back() != '\n')
     ++NumLines;
 }
diff --git a/lib/Frontend/CacheTokens.cpp b/lib/Frontend/CacheTokens.cpp
index d909d526b518..7d2a09cd7ca0 100644
--- a/lib/Frontend/CacheTokens.cpp
+++ b/lib/Frontend/CacheTokens.cpp
@@ -183,7 +183,7 @@ class PTHWriter {
   typedef llvm::StringMap<OffsetOpt, llvm::BumpPtrAllocator> CachedStrsTy;
 
   IDMap IM;
-  llvm::raw_fd_ostream& Out;
+  raw_pwrite_stream &Out;
   Preprocessor& PP;
   uint32_t idcount;
   PTHMap PM;
@@ -236,8 +236,8 @@ class PTHWriter {
   Offset EmitCachedSpellings();
 
 public:
-  PTHWriter(llvm::raw_fd_ostream& out, Preprocessor& pp)
-    : Out(out), PP(pp), idcount(0), CurStrOffset(0) {}
+  PTHWriter(raw_pwrite_stream &out, Preprocessor &pp)
+      : Out(out), PP(pp), idcount(0), CurStrOffset(0) {}
 
   PTHMap &getPM() { return PM; }
   void GeneratePTH(const std::string &MainFile);
@@ -468,6 +468,16 @@ Offset PTHWriter::EmitCachedSpellings() {
   return SpellingsOff;
 }
 
+static uint32_t swap32le(uint32_t X) {
+  return llvm::support::endian::byte_swap<uint32_t, llvm::support::little>(X);
+}
+
+static void pwrite32le(raw_pwrite_stream &OS, uint32_t Val, uint64_t &Off) {
+  uint32_t LEVal = swap32le(Val);
+  OS.pwrite(reinterpret_cast<const char *>(&LEVal), 4, Off);
+  Off += 4;
+}
+
 void PTHWriter::GeneratePTH(const std::string &MainFile) {
   // Generate the prologue.
   Out << "cfe-pth" << '\0';
@@ -520,11 +530,11 @@ void PTHWriter::GeneratePTH(const std::string &MainFile) {
   Offset FileTableOff = EmitFileTable();
 
   // Finally, write the prologue.
-  Out.seek(PrologueOffset);
-  Emit32(IdTableOff.first);
-  Emit32(IdTableOff.second);
-  Emit32(FileTableOff);
-  Emit32(SpellingOff);
+  uint64_t Off = PrologueOffset;
+  pwrite32le(Out, IdTableOff.first, Off);
+  pwrite32le(Out, IdTableOff.second, Off);
+  pwrite32le(Out, FileTableOff, Off);
+  pwrite32le(Out, SpellingOff, Off);
 }
 
 namespace {
@@ -537,7 +547,7 @@ class StatListener : public FileSystemStatCache {
   PTHMap &PM;
 public:
   StatListener(PTHMap &pm) : PM(pm) {}
-  ~StatListener() {}
+  ~StatListener() override {}
 
   LookupResult getStat(const char *Path, FileData &Data, bool isFile,
                        std::unique_ptr<vfs::File> *F,
@@ -559,8 +569,7 @@ public:
 };
 } // end anonymous namespace
 
-
-void clang::CacheTokens(Preprocessor &PP, llvm::raw_fd_ostream* OS) {
+void clang::CacheTokens(Preprocessor &PP, raw_pwrite_stream *OS) {
   // Get the name of the main file.
   const SourceManager &SrcMgr = PP.getSourceManager();
   const FileEntry *MainFile = SrcMgr.getFileEntryForID(SrcMgr.getMainFileID());
diff --git a/lib/Frontend/ChainedIncludesSource.cpp b/lib/Frontend/ChainedIncludesSource.cpp
index cb260b4f4c69..f3677f8000c6 100644
--- a/lib/Frontend/ChainedIncludesSource.cpp
+++ b/lib/Frontend/ChainedIncludesSource.cpp
@@ -27,7 +27,7 @@ using namespace clang;
 namespace {
 class ChainedIncludesSource : public ExternalSemaSource {
 public:
-  virtual ~ChainedIncludesSource();
+  ~ChainedIncludesSource() override;
 
   ExternalSemaSource &getFinalReader() const { return *FinalReader; }
 
@@ -43,6 +43,7 @@ protected:
   Selector GetExternalSelector(uint32_t ID) override;
   uint32_t GetNumExternalSelectors() override;
   Stmt *GetExternalDeclStmt(uint64_t Offset) override;
+  CXXCtorInitializer **GetExternalCXXCtorInitializers(uint64_t Offset) override;
   CXXBaseSpecifier *GetExternalCXXBaseSpecifiers(uint64_t Offset) override;
   bool FindExternalVisibleDeclsByName(const DeclContext *DC,
                                       DeclarationName Name) override;
@@ -232,6 +233,10 @@ CXXBaseSpecifier *
 ChainedIncludesSource::GetExternalCXXBaseSpecifiers(uint64_t Offset) {
   return getFinalReader().GetExternalCXXBaseSpecifiers(Offset);
 }
+CXXCtorInitializer **
+ChainedIncludesSource::GetExternalCXXCtorInitializers(uint64_t Offset) {
+  return getFinalReader().GetExternalCXXCtorInitializers(Offset);
+}
 bool
 ChainedIncludesSource::FindExternalVisibleDeclsByName(const DeclContext *DC,
                                                       DeclarationName Name) {
diff --git a/lib/Frontend/CompilerInstance.cpp b/lib/Frontend/CompilerInstance.cpp
index 93a34b722274..9e017273cb2a 100644
--- a/lib/Frontend/CompilerInstance.cpp
+++ b/lib/Frontend/CompilerInstance.cpp
@@ -329,14 +329,8 @@ void CompilerInstance::createPreprocessor(TranslationUnitKind TUKind) {
 
   PP->setPreprocessedOutput(getPreprocessorOutputOpts().ShowCPP);
 
-  // Set up the module path, including the hash for the
-  // module-creation options.
-  SmallString<256> SpecificModuleCache(
-                           getHeaderSearchOpts().ModuleCachePath);
-  if (!getHeaderSearchOpts().DisableModuleHash)
-    llvm::sys::path::append(SpecificModuleCache,
-                            getInvocation().getModuleHash());
-  PP->getHeaderSearchInfo().setModuleCachePath(SpecificModuleCache);
+  if (PP->getLangOpts().Modules)
+    PP->getHeaderSearchInfo().setModuleCachePath(getSpecificModuleCachePath());
 
   // Handle generating dependencies, if requested.
   const DependencyOutputOptions &DepOpts = getDependencyOutputOpts();
@@ -371,14 +365,17 @@ void CompilerInstance::createPreprocessor(TranslationUnitKind TUKind) {
     AttachHeaderIncludeGen(*PP, /*ShowAllHeaders=*/false, /*OutputPath=*/"",
                            /*ShowDepth=*/true, /*MSStyle=*/true);
   }
+}
 
-  // Load all explictly-specified module map files.
-  for (const auto &Filename : getFrontendOpts().ModuleMapFiles) {
-    if (auto *File = getFileManager().getFile(Filename))
-      PP->getHeaderSearchInfo().loadModuleMapFile(File, /*IsSystem*/false);
-    else
-      getDiagnostics().Report(diag::err_module_map_not_found) << Filename;
-  }
+std::string CompilerInstance::getSpecificModuleCachePath() {
+  // Set up the module path, including the hash for the
+  // module-creation options.
+  SmallString<256> SpecificModuleCache(
+                           getHeaderSearchOpts().ModuleCachePath);
+  if (!getHeaderSearchOpts().DisableModuleHash)
+    llvm::sys::path::append(SpecificModuleCache,
+                            getInvocation().getModuleHash());
+  return SpecificModuleCache.str();
 }
 
 // ASTContext
@@ -396,32 +393,30 @@ void CompilerInstance::createASTContext() {
 void CompilerInstance::createPCHExternalASTSource(
     StringRef Path, bool DisablePCHValidation, bool AllowPCHWithCompilerErrors,
     void *DeserializationListener, bool OwnDeserializationListener) {
-  IntrusiveRefCntPtr<ExternalASTSource> Source;
   bool Preamble = getPreprocessorOpts().PrecompiledPreambleBytes.first != 0;
-  Source = createPCHExternalASTSource(
+  ModuleManager = createPCHExternalASTSource(
       Path, getHeaderSearchOpts().Sysroot, DisablePCHValidation,
       AllowPCHWithCompilerErrors, getPreprocessor(), getASTContext(),
       DeserializationListener, OwnDeserializationListener, Preamble,
       getFrontendOpts().UseGlobalModuleIndex);
-  ModuleManager = static_cast<ASTReader*>(Source.get());
-  getASTContext().setExternalSource(Source);
 }
 
-ExternalASTSource *CompilerInstance::createPCHExternalASTSource(
+IntrusiveRefCntPtr<ASTReader> CompilerInstance::createPCHExternalASTSource(
     StringRef Path, const std::string &Sysroot, bool DisablePCHValidation,
     bool AllowPCHWithCompilerErrors, Preprocessor &PP, ASTContext &Context,
     void *DeserializationListener, bool OwnDeserializationListener,
     bool Preamble, bool UseGlobalModuleIndex) {
   HeaderSearchOptions &HSOpts = PP.getHeaderSearchInfo().getHeaderSearchOpts();
 
-  std::unique_ptr<ASTReader> Reader;
-  Reader.reset(new ASTReader(PP, Context,
-                             Sysroot.empty() ? "" : Sysroot.c_str(),
-                             DisablePCHValidation,
-                             AllowPCHWithCompilerErrors,
-                             /*AllowConfigurationMismatch*/false,
-                             HSOpts.ModulesValidateSystemHeaders,
-                             UseGlobalModuleIndex));
+  IntrusiveRefCntPtr<ASTReader> Reader(
+      new ASTReader(PP, Context, Sysroot.empty() ? "" : Sysroot.c_str(),
+                    DisablePCHValidation, AllowPCHWithCompilerErrors,
+                    /*AllowConfigurationMismatch*/ false,
+                    HSOpts.ModulesValidateSystemHeaders, UseGlobalModuleIndex));
+
+  // We need the external source to be set up before we read the AST, because
+  // eagerly-deserialized declarations may use it.
+  Context.setExternalSource(Reader.get());
 
   Reader->setDeserializationListener(
       static_cast<ASTDeserializationListener *>(DeserializationListener),
@@ -435,7 +430,7 @@ ExternalASTSource *CompilerInstance::createPCHExternalASTSource(
     // Set the predefines buffer as suggested by the PCH reader. Typically, the
     // predefines buffer will be empty.
     PP.setPredefines(Reader->getSuggestedPredefines());
-    return Reader.release();
+    return Reader;
 
   case ASTReader::Failure:
     // Unrecoverable failure: don't even try to process the input file.
@@ -450,6 +445,7 @@ ExternalASTSource *CompilerInstance::createPCHExternalASTSource(
     break;
   }
 
+  Context.setExternalSource(nullptr);
   return nullptr;
 }
 
@@ -522,42 +518,43 @@ void CompilerInstance::createSema(TranslationUnitKind TUKind,
 
 // Output Files
 
-void CompilerInstance::addOutputFile(const OutputFile &OutFile) {
+void CompilerInstance::addOutputFile(OutputFile &&OutFile) {
   assert(OutFile.OS && "Attempt to add empty stream to output list!");
-  OutputFiles.push_back(OutFile);
+  OutputFiles.push_back(std::move(OutFile));
 }
 
 void CompilerInstance::clearOutputFiles(bool EraseFiles) {
-  for (std::list<OutputFile>::iterator
-         it = OutputFiles.begin(), ie = OutputFiles.end(); it != ie; ++it) {
-    delete it->OS;
-    if (!it->TempFilename.empty()) {
+  for (OutputFile &OF : OutputFiles) {
+    // Manually close the stream before we rename it.
+    OF.OS.reset();
+
+    if (!OF.TempFilename.empty()) {
       if (EraseFiles) {
-        llvm::sys::fs::remove(it->TempFilename);
+        llvm::sys::fs::remove(OF.TempFilename);
       } else {
-        SmallString<128> NewOutFile(it->Filename);
+        SmallString<128> NewOutFile(OF.Filename);
 
         // If '-working-directory' was passed, the output filename should be
         // relative to that.
         FileMgr->FixupRelativePath(NewOutFile);
         if (std::error_code ec =
-                llvm::sys::fs::rename(it->TempFilename, NewOutFile.str())) {
+                llvm::sys::fs::rename(OF.TempFilename, NewOutFile)) {
           getDiagnostics().Report(diag::err_unable_to_rename_temp)
-            << it->TempFilename << it->Filename << ec.message();
+            << OF.TempFilename << OF.Filename << ec.message();
 
-          llvm::sys::fs::remove(it->TempFilename);
+          llvm::sys::fs::remove(OF.TempFilename);
         }
       }
-    } else if (!it->Filename.empty() && EraseFiles)
-      llvm::sys::fs::remove(it->Filename);
+    } else if (!OF.Filename.empty() && EraseFiles)
+      llvm::sys::fs::remove(OF.Filename);
 
   }
   OutputFiles.clear();
+  NonSeekStream.reset();
 }
 
-llvm::raw_fd_ostream *
-CompilerInstance::createDefaultOutputFile(bool Binary,
-                                          StringRef InFile,
+raw_pwrite_stream *
+CompilerInstance::createDefaultOutputFile(bool Binary, StringRef InFile,
                                           StringRef Extension) {
   return createOutputFile(getFrontendOpts().OutputFile, Binary,
                           /*RemoveFileOnSignal=*/true, InFile, Extension,
@@ -565,21 +562,20 @@ CompilerInstance::createDefaultOutputFile(bool Binary,
 }
 
 llvm::raw_null_ostream *CompilerInstance::createNullOutputFile() {
-  llvm::raw_null_ostream *OS = new llvm::raw_null_ostream();
-  addOutputFile(OutputFile("", "", OS));
-  return OS;
+  auto OS = llvm::make_unique<llvm::raw_null_ostream>();
+  llvm::raw_null_ostream *Ret = OS.get();
+  addOutputFile(OutputFile("", "", std::move(OS)));
+  return Ret;
 }
 
-llvm::raw_fd_ostream *
-CompilerInstance::createOutputFile(StringRef OutputPath,
-                                   bool Binary, bool RemoveFileOnSignal,
-                                   StringRef InFile,
-                                   StringRef Extension,
-                                   bool UseTemporary,
+raw_pwrite_stream *
+CompilerInstance::createOutputFile(StringRef OutputPath, bool Binary,
+                                   bool RemoveFileOnSignal, StringRef InFile,
+                                   StringRef Extension, bool UseTemporary,
                                    bool CreateMissingDirectories) {
   std::string OutputPathName, TempPathName;
   std::error_code EC;
-  llvm::raw_fd_ostream *OS = createOutputFile(
+  std::unique_ptr<raw_pwrite_stream> OS = createOutputFile(
       OutputPath, EC, Binary, RemoveFileOnSignal, InFile, Extension,
       UseTemporary, CreateMissingDirectories, &OutputPathName, &TempPathName);
   if (!OS) {
@@ -588,15 +584,16 @@ CompilerInstance::createOutputFile(StringRef OutputPath,
     return nullptr;
   }
 
+  raw_pwrite_stream *Ret = OS.get();
   // Add the output file -- but don't try to remove "-", since this means we are
   // using stdin.
   addOutputFile(OutputFile((OutputPathName != "-") ? OutputPathName : "",
-                TempPathName, OS));
+                           TempPathName, std::move(OS)));
 
-  return OS;
+  return Ret;
 }
 
-llvm::raw_fd_ostream *CompilerInstance::createOutputFile(
+std::unique_ptr<llvm::raw_pwrite_stream> CompilerInstance::createOutputFile(
     StringRef OutputPath, std::error_code &Error, bool Binary,
     bool RemoveFileOnSignal, StringRef InFile, StringRef Extension,
     bool UseTemporary, bool CreateMissingDirectories,
@@ -646,14 +643,14 @@ llvm::raw_fd_ostream *CompilerInstance::createOutputFile(
     TempPath += "-%%%%%%%%";
     int fd;
     std::error_code EC =
-        llvm::sys::fs::createUniqueFile(TempPath.str(), fd, TempPath);
+        llvm::sys::fs::createUniqueFile(TempPath, fd, TempPath);
 
     if (CreateMissingDirectories &&
         EC == llvm::errc::no_such_file_or_directory) {
       StringRef Parent = llvm::sys::path::parent_path(OutputPath);
       EC = llvm::sys::fs::create_directories(Parent);
       if (!EC) {
-        EC = llvm::sys::fs::createUniqueFile(TempPath.str(), fd, TempPath);
+        EC = llvm::sys::fs::createUniqueFile(TempPath, fd, TempPath);
       }
     }
 
@@ -684,7 +681,13 @@ llvm::raw_fd_ostream *CompilerInstance::createOutputFile(
   if (TempPathName)
     *TempPathName = TempFile;
 
-  return OS.release();
+  if (!Binary || OS->supportsSeeking())
+    return std::move(OS);
+
+  auto B = llvm::make_unique<llvm::buffer_ostream>(*OS);
+  assert(!NonSeekStream);
+  NonSeekStream = std::move(OS);
+  return std::move(B);
 }
 
 // Initialization Utilities
@@ -946,16 +949,18 @@ static bool compileModuleImpl(CompilerInstance &ImportingInstance,
     FrontendOpts.Inputs.push_back(
         FrontendInputFile(ModuleMapFile->getName(), IK));
   } else {
+    SmallString<128> FakeModuleMapFile(Module->Directory->getName());
+    llvm::sys::path::append(FakeModuleMapFile, "__inferred_module.map");
+    FrontendOpts.Inputs.push_back(FrontendInputFile(FakeModuleMapFile, IK));
+
     llvm::raw_string_ostream OS(InferredModuleMapContent);
     Module->print(OS);
     OS.flush();
-    FrontendOpts.Inputs.push_back(
-        FrontendInputFile("__inferred_module.map", IK));
 
     std::unique_ptr<llvm::MemoryBuffer> ModuleMapBuffer =
         llvm::MemoryBuffer::getMemBuffer(InferredModuleMapContent);
     ModuleMapFile = Instance.getFileManager().getVirtualFile(
-        "__inferred_module.map", InferredModuleMapContent.size(), 0);
+        FakeModuleMapFile, InferredModuleMapContent.size(), 0);
     SourceMgr.overrideFileContents(ModuleMapFile, std::move(ModuleMapBuffer));
   }
 
@@ -1031,9 +1036,19 @@ static bool compileAndLoadModule(CompilerInstance &ImportingInstance,
     case llvm::LockFileManager::LFS_Shared:
       // Someone else is responsible for building the module. Wait for them to
       // finish.
-      if (Locked.waitForUnlock() == llvm::LockFileManager::Res_OwnerDied)
+      switch (Locked.waitForUnlock()) {
+      case llvm::LockFileManager::Res_Success:
+        ModuleLoadCapabilities |= ASTReader::ARR_OutOfDate;
+        break;
+      case llvm::LockFileManager::Res_OwnerDied:
         continue; // try again to get the lock.
-      ModuleLoadCapabilities |= ASTReader::ARR_OutOfDate;
+      case llvm::LockFileManager::Res_Timeout:
+        Diags.Report(ModuleNameLoc, diag::err_module_lock_timeout)
+            << Module->Name;
+        // Clear the lock file so that future invokations can make progress.
+        Locked.unsafeRemoveLockFile();
+        return false;
+      }
       break;
     }
 
@@ -1071,79 +1086,51 @@ static void checkConfigMacro(Preprocessor &PP, StringRef ConfigMacro,
   // not have changed.
   if (!Id->hadMacroDefinition())
     return;
+  auto *LatestLocalMD = PP.getLocalMacroDirectiveHistory(Id);
 
-  // If this identifier does not currently have a macro definition,
-  // check whether it had one on the command line.
-  if (!Id->hasMacroDefinition()) {
-    MacroDirective::DefInfo LatestDef =
-        PP.getMacroDirectiveHistory(Id)->getDefinition();
-    for (MacroDirective::DefInfo Def = LatestDef; Def;
-           Def = Def.getPreviousDefinition()) {
-      FileID FID = SourceMgr.getFileID(Def.getLocation());
-      if (FID.isInvalid())
-        continue;
-
-      // We only care about the predefines buffer.
-      if (FID != PP.getPredefinesFileID())
-        continue;
-
-      // This macro was defined on the command line, then #undef'd later.
-      // Complain.
-      PP.Diag(ImportLoc, diag::warn_module_config_macro_undef)
-        << true << ConfigMacro << Mod->getFullModuleName();
-      if (LatestDef.isUndefined())
-        PP.Diag(LatestDef.getUndefLocation(), diag::note_module_def_undef_here)
-          << true;
-      return;
-    }
-
-    // Okay: no definition in the predefines buffer.
-    return;
-  }
-
-  // This identifier has a macro definition. Check whether we had a definition
-  // on the command line.
-  MacroDirective::DefInfo LatestDef =
-      PP.getMacroDirectiveHistory(Id)->getDefinition();
-  MacroDirective::DefInfo PredefinedDef;
-  for (MacroDirective::DefInfo Def = LatestDef; Def;
-         Def = Def.getPreviousDefinition()) {
-    FileID FID = SourceMgr.getFileID(Def.getLocation());
-    if (FID.isInvalid())
-      continue;
-
+  // Find the macro definition from the command line.
+  MacroInfo *CmdLineDefinition = nullptr;
+  for (auto *MD = LatestLocalMD; MD; MD = MD->getPrevious()) {
     // We only care about the predefines buffer.
-    if (FID != PP.getPredefinesFileID())
+    FileID FID = SourceMgr.getFileID(MD->getLocation());
+    if (FID.isInvalid() || FID != PP.getPredefinesFileID())
       continue;
-
-    PredefinedDef = Def;
+    if (auto *DMD = dyn_cast<DefMacroDirective>(MD))
+      CmdLineDefinition = DMD->getMacroInfo();
     break;
   }
 
-  // If there was no definition for this macro in the predefines buffer,
-  // complain.
-  if (!PredefinedDef ||
-      (!PredefinedDef.getLocation().isValid() &&
-       PredefinedDef.getUndefLocation().isValid())) {
+  auto *CurrentDefinition = PP.getMacroInfo(Id);
+  if (CurrentDefinition == CmdLineDefinition) {
+    // Macro matches. Nothing to do.
+  } else if (!CurrentDefinition) {
+    // This macro was defined on the command line, then #undef'd later.
+    // Complain.
+    PP.Diag(ImportLoc, diag::warn_module_config_macro_undef)
+      << true << ConfigMacro << Mod->getFullModuleName();
+    auto LatestDef = LatestLocalMD->getDefinition();
+    assert(LatestDef.isUndefined() &&
+           "predefined macro went away with no #undef?");
+    PP.Diag(LatestDef.getUndefLocation(), diag::note_module_def_undef_here)
+      << true;
+    return;
+  } else if (!CmdLineDefinition) {
+    // There was no definition for this macro in the predefines buffer,
+    // but there was a local definition. Complain.
     PP.Diag(ImportLoc, diag::warn_module_config_macro_undef)
       << false << ConfigMacro << Mod->getFullModuleName();
-    PP.Diag(LatestDef.getLocation(), diag::note_module_def_undef_here)
+    PP.Diag(CurrentDefinition->getDefinitionLoc(),
+            diag::note_module_def_undef_here)
+      << false;
+  } else if (!CurrentDefinition->isIdenticalTo(*CmdLineDefinition, PP,
+                                               /*Syntactically=*/true)) {
+    // The macro definitions differ.
+    PP.Diag(ImportLoc, diag::warn_module_config_macro_undef)
+      << false << ConfigMacro << Mod->getFullModuleName();
+    PP.Diag(CurrentDefinition->getDefinitionLoc(),
+            diag::note_module_def_undef_here)
       << false;
-    return;
   }
-
-  // If the current macro definition is the same as the predefined macro
-  // definition, it's okay.
-  if (LatestDef.getMacroInfo() == PredefinedDef.getMacroInfo() ||
-      LatestDef.getMacroInfo()->isIdenticalTo(*PredefinedDef.getMacroInfo(),PP,
-                                              /*Syntactically=*/true))
-    return;
-
-  // The macro definitions differ.
-  PP.Diag(ImportLoc, diag::warn_module_config_macro_undef)
-    << false << ConfigMacro << Mod->getFullModuleName();
-  PP.Diag(LatestDef.getLocation(), diag::note_module_def_undef_here)
-    << false;
 }
 
 /// \brief Write a new timestamp file with the given path.
@@ -1186,8 +1173,7 @@ static void pruneModuleCache(const HeaderSearchOptions &HSOpts) {
   std::error_code EC;
   SmallString<128> ModuleCachePathNative;
   llvm::sys::path::native(HSOpts.ModuleCachePath, ModuleCachePathNative);
-  for (llvm::sys::fs::directory_iterator
-         Dir(ModuleCachePathNative.str(), EC), DirEnd;
+  for (llvm::sys::fs::directory_iterator Dir(ModuleCachePathNative, EC), DirEnd;
        Dir != DirEnd && !EC; Dir.increment(EC)) {
     // If we don't have a directory, there's nothing to look into.
     if (!llvm::sys::fs::is_directory(Dir->path()))
@@ -1235,9 +1221,10 @@ void CompilerInstance::createModuleManager() {
     if (!hasASTContext())
       createASTContext();
 
-    // If we're not recursively building a module, check whether we
-    // need to prune the module cache.
-    if (getSourceManager().getModuleBuildStack().empty() &&
+    // If we're implicitly building modules but not currently recursively
+    // building a module, check whether we need to prune the module cache.
+    if (getLangOpts().ImplicitModules &&
+        getSourceManager().getModuleBuildStack().empty() &&
         getHeaderSearchOpts().ModuleCachePruneInterval > 0 &&
         getHeaderSearchOpts().ModuleCachePruneAfter > 0) {
       pruneModuleCache(getHeaderSearchOpts());
@@ -1274,6 +1261,7 @@ bool CompilerInstance::loadModuleFile(StringRef FileName) {
   struct ReadModuleNames : ASTReaderListener {
     CompilerInstance &CI;
     std::vector<StringRef> ModuleFileStack;
+    std::vector<StringRef> ModuleNameStack;
     bool Failed;
     bool TopFileIsModule;
 
@@ -1283,21 +1271,36 @@ bool CompilerInstance::loadModuleFile(StringRef FileName) {
     bool needsImportVisitation() const override { return true; }
 
     void visitImport(StringRef FileName) override {
+      if (!CI.ExplicitlyLoadedModuleFiles.insert(FileName).second) {
+        if (ModuleFileStack.size() == 0)
+          TopFileIsModule = true;
+        return;
+      }
+
       ModuleFileStack.push_back(FileName);
+      ModuleNameStack.push_back(StringRef());
       if (ASTReader::readASTFileControlBlock(FileName, CI.getFileManager(),
                                              *this)) {
-        CI.getDiagnostics().Report(SourceLocation(),
-                                   diag::err_module_file_not_found)
+        CI.getDiagnostics().Report(
+            SourceLocation(), CI.getFileManager().getBufferForFile(FileName)
+                                  ? diag::err_module_file_invalid
+                                  : diag::err_module_file_not_found)
             << FileName;
-        // FIXME: Produce a note stack explaining how we got here.
+        for (int I = ModuleFileStack.size() - 2; I >= 0; --I)
+          CI.getDiagnostics().Report(SourceLocation(),
+                                     diag::note_module_file_imported_by)
+              << ModuleFileStack[I]
+              << !ModuleNameStack[I].empty() << ModuleNameStack[I];
         Failed = true;
       }
+      ModuleNameStack.pop_back();
       ModuleFileStack.pop_back();
     }
 
     void ReadModuleName(StringRef ModuleName) override {
       if (ModuleFileStack.size() == 1)
         TopFileIsModule = true;
+      ModuleNameStack.back() = ModuleName;
 
       auto &ModuleFile = CI.ModuleFileOverrides[ModuleName];
       if (!ModuleFile.empty() &&
@@ -1310,6 +1313,19 @@ bool CompilerInstance::loadModuleFile(StringRef FileName) {
     }
   } RMN(*this);
 
+  // If we don't already have an ASTReader, create one now.
+  if (!ModuleManager)
+    createModuleManager();
+
+  // Tell the module manager about this module file.
+  if (getModuleManager()->getModuleManager().addKnownModuleFile(FileName)) {
+    getDiagnostics().Report(SourceLocation(), diag::err_module_file_not_found)
+      << FileName;
+    return false;
+  }
+
+  // Build our mapping of module names to module files from this file
+  // and its imports.
   RMN.visitImport(FileName);
 
   if (RMN.Failed)
@@ -1343,7 +1359,7 @@ CompilerInstance::loadModule(SourceLocation ImportLoc,
     if (LastModuleImportResult && ModuleName != getLangOpts().CurrentModule &&
         ModuleName != getLangOpts().ImplementationOfModule)
       ModuleManager->makeModuleVisible(LastModuleImportResult, Visibility,
-                                       ImportLoc, /*Complain=*/false);
+                                       ImportLoc);
     return LastModuleImportResult;
   }
 
@@ -1373,6 +1389,12 @@ CompilerInstance::loadModule(SourceLocation ImportLoc,
 
     auto Override = ModuleFileOverrides.find(ModuleName);
     bool Explicit = Override != ModuleFileOverrides.end();
+    if (!Explicit && !getLangOpts().ImplicitModules) {
+      getDiagnostics().Report(ModuleNameLoc, diag::err_module_build_disabled)
+          << ModuleName;
+      ModuleBuildFailed = true;
+      return ModuleLoadResult();
+    }
 
     std::string ModuleFileName =
         Explicit ? Override->second
@@ -1580,8 +1602,7 @@ CompilerInstance::loadModule(SourceLocation ImportLoc,
       return ModuleLoadResult();
     }
 
-    ModuleManager->makeModuleVisible(Module, Visibility, ImportLoc,
-                                     /*Complain=*/true);
+    ModuleManager->makeModuleVisible(Module, Visibility, ImportLoc);
   }
 
   // Check for any configuration macros that have changed.
@@ -1591,25 +1612,6 @@ CompilerInstance::loadModule(SourceLocation ImportLoc,
                      Module, ImportLoc);
   }
 
-  // Determine whether we're in the #include buffer for a module. The #includes
-  // in that buffer do not qualify as module imports; they're just an
-  // implementation detail of us building the module.
-  bool IsInModuleIncludes = !getLangOpts().CurrentModule.empty() &&
-                            getSourceManager().getFileID(ImportLoc) ==
-                                getSourceManager().getMainFileID();
-
-  // If this module import was due to an inclusion directive, create an 
-  // implicit import declaration to capture it in the AST.
-  if (IsInclusionDirective && hasASTContext() && !IsInModuleIncludes) {
-    TranslationUnitDecl *TU = getASTContext().getTranslationUnitDecl();
-    ImportDecl *ImportD = ImportDecl::CreateImplicit(getASTContext(), TU,
-                                                     ImportLoc, Module,
-                                                     Path.back().second);
-    TU->addDecl(ImportD);
-    if (Consumer)
-      Consumer->HandleImplicitImportDecl(ImportD);
-  }
-  
   LastModuleImportLoc = ImportLoc;
   LastModuleImportResult = ModuleLoadResult(Module, false);
   return LastModuleImportResult;
@@ -1617,9 +1619,13 @@ CompilerInstance::loadModule(SourceLocation ImportLoc,
 
 void CompilerInstance::makeModuleVisible(Module *Mod,
                                          Module::NameVisibilityKind Visibility,
-                                         SourceLocation ImportLoc,
-                                         bool Complain){
-  ModuleManager->makeModuleVisible(Mod, Visibility, ImportLoc, Complain);
+                                         SourceLocation ImportLoc) {
+  if (!ModuleManager)
+    createModuleManager();
+  if (!ModuleManager)
+    return;
+
+  ModuleManager->makeModuleVisible(Mod, Visibility, ImportLoc);
 }
 
 GlobalModuleIndex *CompilerInstance::loadGlobalModuleIndex(
diff --git a/lib/Frontend/CompilerInvocation.cpp b/lib/Frontend/CompilerInvocation.cpp
index 54025b06557a..8d3d31228a70 100644
--- a/lib/Frontend/CompilerInvocation.cpp
+++ b/lib/Frontend/CompilerInvocation.cpp
@@ -253,7 +253,7 @@ static bool ParseAnalyzerArgs(AnalyzerOptions &Opts, ArgList &Args,
     for (unsigned i = 0, e = checkers.size(); i != e; ++i)
       Opts.CheckersControlList.push_back(std::make_pair(checkers[i], enable));
   }
-  
+
   // Go through the analyzer configuration options.
   for (arg_iterator it = Args.filtered_begin(OPT_analyzer_config),
        ie = Args.filtered_end(); it != ie; ++it) {
@@ -329,11 +329,8 @@ static void parseSanitizerKinds(StringRef FlagName,
                                 const std::vector<std::string> &Sanitizers,
                                 DiagnosticsEngine &Diags, SanitizerSet &S) {
   for (const auto &Sanitizer : Sanitizers) {
-    SanitizerKind K = llvm::StringSwitch<SanitizerKind>(Sanitizer)
-#define SANITIZER(NAME, ID) .Case(NAME, SanitizerKind::ID)
-#include "clang/Basic/Sanitizers.def"
-                          .Default(SanitizerKind::Unknown);
-    if (K == SanitizerKind::Unknown)
+    SanitizerMask K = parseSanitizerValue(Sanitizer, /*AllowGroups=*/false);
+    if (K == 0)
       Diags.Report(diag::err_drv_invalid_value) << FlagName << Sanitizer;
     else
       S.set(K, true);
@@ -367,6 +364,16 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
   Opts.setInlining(Args.hasArg(OPT_fno_inline_functions) ?
                      CodeGenOptions::OnlyAlwaysInlining : Opts.getInlining());
 
+  if (Arg *A = Args.getLastArg(OPT_fveclib)) {
+    StringRef Name = A->getValue();
+    if (Name == "Accelerate")
+      Opts.setVecLib(CodeGenOptions::Accelerate);
+    else if (Name == "none")
+      Opts.setVecLib(CodeGenOptions::NoLibrary);
+    else
+      Diags.Report(diag::err_drv_invalid_value) << A->getAsString(Args) << Name;
+  }
+
   if (Args.hasArg(OPT_gline_tables_only)) {
     Opts.setDebugInfo(CodeGenOptions::DebugLineTablesOnly);
   } else if (Args.hasArg(OPT_g_Flag) || Args.hasArg(OPT_gdwarf_2) ||
@@ -395,6 +402,10 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
     // Default Dwarf version is 4 if we are generating debug information.
     Opts.DwarfVersion = 4;
 
+  if (const Arg *A =
+          Args.getLastArg(OPT_emit_llvm_uselists, OPT_no_emit_llvm_uselists))
+    Opts.EmitLLVMUseLists = A->getOption().getID() == OPT_emit_llvm_uselists;
+
   Opts.DisableLLVMOpts = Args.hasArg(OPT_disable_llvm_optzns);
   Opts.DisableRedZone = Args.hasArg(OPT_disable_red_zone);
   Opts.ForbidGuardVariables = Args.hasArg(OPT_fforbid_guard_variables);
@@ -417,13 +428,14 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
   Opts.DisableIntegratedAS = Args.hasArg(OPT_fno_integrated_as);
   Opts.Autolink = !Args.hasArg(OPT_fno_autolink);
   Opts.SampleProfileFile = Args.getLastArgValue(OPT_fprofile_sample_use_EQ);
-  Opts.ProfileInstrGenerate = Args.hasArg(OPT_fprofile_instr_generate);
+  Opts.ProfileInstrGenerate = Args.hasArg(OPT_fprofile_instr_generate) ||
+      Args.hasArg(OPT_fprofile_instr_generate_EQ);;
+  Opts.InstrProfileOutput = Args.getLastArgValue(OPT_fprofile_instr_generate_EQ);
   Opts.InstrProfileInput = Args.getLastArgValue(OPT_fprofile_instr_use_EQ);
   Opts.CoverageMapping = Args.hasArg(OPT_fcoverage_mapping);
   Opts.DumpCoverageMapping = Args.hasArg(OPT_dump_coverage_mapping);
   Opts.AsmVerbose = Args.hasArg(OPT_masm_verbose);
   Opts.ObjCAutoRefCountExceptions = Args.hasArg(OPT_fobjc_arc_exceptions);
-  Opts.CUDAIsDevice = Args.hasArg(OPT_fcuda_is_device);
   Opts.CXAAtExit = !Args.hasArg(OPT_fno_use_cxa_atexit);
   Opts.CXXCtorDtorAliases = Args.hasArg(OPT_mconstructor_aliases);
   Opts.CodeModel = getCodeModel(Args, Diags);
@@ -441,11 +453,11 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
                        Args.hasArg(OPT_cl_unsafe_math_optimizations) ||
                        Args.hasArg(OPT_cl_finite_math_only) ||
                        Args.hasArg(OPT_cl_fast_relaxed_math));
-  Opts.NoSignedZeros = Args.hasArg(OPT_cl_no_signed_zeros);
+  Opts.NoSignedZeros = Args.hasArg(OPT_fno_signed_zeros);
+  Opts.ReciprocalMath = Args.hasArg(OPT_freciprocal_math);
   Opts.NoZeroInitializedInBSS = Args.hasArg(OPT_mno_zero_initialized_in_bss);
   Opts.BackendOptions = Args.getAllArgValues(OPT_backend_option);
   Opts.NumRegisterParameters = getLastArgIntValue(Args, OPT_mregparm, 0, Diags);
-  Opts.NoGlobalMerge = Args.hasArg(OPT_mno_global_merge);
   Opts.NoExecStack = Args.hasArg(OPT_mno_exec_stack);
   Opts.FatalWarnings = Args.hasArg(OPT_massembler_fatal_warnings);
   Opts.EnableSegmentedStacks = Args.hasArg(OPT_split_stacks);
@@ -472,8 +484,13 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
                                        OPT_fno_function_sections, false);
   Opts.DataSections = Args.hasFlag(OPT_fdata_sections,
                                    OPT_fno_data_sections, false);
+  Opts.UniqueSectionNames = Args.hasFlag(OPT_funique_section_names,
+                                         OPT_fno_unique_section_names, true);
+
   Opts.MergeFunctions = Args.hasArg(OPT_fmerge_functions);
 
+  Opts.MSVolatile = Args.hasArg(OPT_fms_volatile);
+
   Opts.VectorizeBB = Args.hasArg(OPT_vectorize_slp_aggressive);
   Opts.VectorizeLoop = Args.hasArg(OPT_vectorize_loops);
   Opts.VectorizeSLP = Args.hasArg(OPT_vectorize_slp);
@@ -489,6 +506,8 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
     Opts.CoverageExtraChecksum = Args.hasArg(OPT_coverage_cfg_checksum);
     Opts.CoverageNoFunctionNamesInData =
         Args.hasArg(OPT_coverage_no_function_names_in_data);
+    Opts.CoverageExitBlockBeforeBody =
+        Args.hasArg(OPT_coverage_exit_block_before_body);
     if (Args.hasArg(OPT_coverage_version_EQ)) {
       StringRef CoverageVersion = Args.getLastArgValue(OPT_coverage_version_EQ);
       if (CoverageVersion.size() != 4) {
@@ -507,8 +526,14 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
   Opts.CompressDebugSections = Args.hasArg(OPT_compress_debug_sections);
   Opts.DebugCompilationDir = Args.getLastArgValue(OPT_fdebug_compilation_dir);
   Opts.LinkBitcodeFile = Args.getLastArgValue(OPT_mlink_bitcode_file);
-  Opts.SanitizeCoverage =
-      getLastArgIntValue(Args, OPT_fsanitize_coverage, 0, Diags);
+  Opts.SanitizeCoverageType =
+      getLastArgIntValue(Args, OPT_fsanitize_coverage_type, 0, Diags);
+  Opts.SanitizeCoverageIndirectCalls =
+      Args.hasArg(OPT_fsanitize_coverage_indirect_calls);
+  Opts.SanitizeCoverageTraceBB = Args.hasArg(OPT_fsanitize_coverage_trace_bb);
+  Opts.SanitizeCoverageTraceCmp = Args.hasArg(OPT_fsanitize_coverage_trace_cmp);
+  Opts.SanitizeCoverage8bitCounters =
+      Args.hasArg(OPT_fsanitize_coverage_8bit_counters);
   Opts.SanitizeMemoryTrackOrigins =
       getLastArgIntValue(Args, OPT_fsanitize_memory_track_origins_EQ, 0, Diags);
   Opts.SanitizeUndefinedTrapOnError =
@@ -523,6 +548,13 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
     Opts.StackAlignment = StackAlignment;
   }
 
+  if (Arg *A = Args.getLastArg(OPT_mstack_probe_size)) {
+    StringRef Val = A->getValue();
+    unsigned StackProbeSize = Opts.StackProbeSize;
+    Val.getAsInteger(0, StackProbeSize);
+    Opts.StackProbeSize = StackProbeSize;
+  }
+
   if (Arg *A = Args.getLastArg(OPT_fobjc_dispatch_method_EQ)) {
     StringRef Name = A->getValue();
     unsigned Method = llvm::StringSwitch<unsigned>(Name)
@@ -616,6 +648,9 @@ static bool ParseCodeGenArgs(CodeGenOptions &Opts, ArgList &Args, InputKind IK,
                       Args.getAllArgValues(OPT_fsanitize_recover_EQ), Diags,
                       Opts.SanitizeRecover);
 
+  Opts.CudaGpuBinaryFileNames =
+      Args.getAllArgValues(OPT_fcuda_include_gpubinary);
+
   return Success;
 }
 
@@ -634,6 +669,8 @@ static void ParseDependencyOutputArgs(DependencyOutputOptions &Opts,
   Opts.DOTOutputFile = Args.getLastArgValue(OPT_dependency_dot);
   Opts.ModuleDependencyOutputDir =
       Args.getLastArgValue(OPT_module_dependency_dir);
+  if (Args.hasArg(OPT_MV))
+    Opts.OutputFormat = DependencyOutputFormat::NMake;
 }
 
 bool clang::ParseDiagnosticArgs(DiagnosticOptions &Opts, ArgList &Args,
@@ -702,9 +739,9 @@ bool clang::ParseDiagnosticArgs(DiagnosticOptions &Opts, ArgList &Args,
   if (Format == "clang")
     Opts.setFormat(DiagnosticOptions::Clang);
   else if (Format == "msvc")
-    Opts.setFormat(DiagnosticOptions::Msvc);
+    Opts.setFormat(DiagnosticOptions::MSVC);
   else if (Format == "msvc-fallback") {
-    Opts.setFormat(DiagnosticOptions::Msvc);
+    Opts.setFormat(DiagnosticOptions::MSVC);
     Opts.CLFallbackMode = true;
   } else if (Format == "vi")
     Opts.setFormat(DiagnosticOptions::Vi);
@@ -966,6 +1003,7 @@ static InputKind ParseFrontendArgs(FrontendOptions &Opts, ArgList &Args,
       .Case("cpp-output", IK_PreprocessedC)
       .Case("assembler-with-cpp", IK_Asm)
       .Case("c++-cpp-output", IK_PreprocessedCXX)
+      .Case("cuda-cpp-output", IK_PreprocessedCuda)
       .Case("objective-c-cpp-output", IK_PreprocessedObjC)
       .Case("objc-cpp-output", IK_PreprocessedObjC)
       .Case("objective-c++-cpp-output", IK_PreprocessedObjCXX)
@@ -1169,6 +1207,7 @@ void CompilerInvocation::setLangDefaults(LangOptions &Opts, InputKind IK,
       LangStd = LangStandard::lang_opencl;
       break;
     case IK_CUDA:
+    case IK_PreprocessedCuda:
       LangStd = LangStandard::lang_cuda;
       break;
     case IK_Asm:
@@ -1197,7 +1236,7 @@ void CompilerInvocation::setLangDefaults(LangOptions &Opts, InputKind IK,
   Opts.CPlusPlus1z = Std.isCPlusPlus1z();
   Opts.Digraphs = Std.hasDigraphs();
   Opts.GNUMode = Std.isGNUMode();
-  Opts.GNUInline = !Std.isC99();
+  Opts.GNUInline = Std.isC89();
   Opts.HexFloats = Std.hasHexFloats();
   Opts.ImplicitInt = Std.hasImplicitInt();
 
@@ -1221,7 +1260,8 @@ void CompilerInvocation::setLangDefaults(LangOptions &Opts, InputKind IK,
     Opts.NativeHalfType = 1;
   }
 
-  Opts.CUDA = LangStd == LangStandard::lang_cuda || IK == IK_CUDA;
+  Opts.CUDA = IK == IK_CUDA || IK == IK_PreprocessedCuda ||
+              LangStd == LangStandard::lang_cuda;
 
   // OpenCL and C++ both have bool, true, false keywords.
   Opts.Bool = Opts.OpenCL || Opts.CPlusPlus;
@@ -1236,9 +1276,6 @@ void CompilerInvocation::setLangDefaults(LangOptions &Opts, InputKind IK,
   Opts.CXXOperatorNames = Opts.CPlusPlus;
 
   Opts.DollarIdents = !Opts.AsmPreprocessor;
-
-  // C++14 onwards has sized global deallocation functions.
-  Opts.SizedDeallocation = Opts.CPlusPlus14;
 }
 
 /// Attempt to parse a visibility value out of the given argument.
@@ -1259,43 +1296,6 @@ static Visibility parseVisibility(Arg *arg, ArgList &args,
   return DefaultVisibility;
 }
 
-static unsigned parseMSCVersion(ArgList &Args, DiagnosticsEngine &Diags) {
-  auto Arg = Args.getLastArg(OPT_fms_compatibility_version);
-  if (!Arg)
-    return 0;
-
-  // The MSC versioning scheme involves four versioning components:
-  //  - Major
-  //  - Minor
-  //  - Build
-  //  - Patch
-  //
-  // We accept either the old style (_MSC_VER) value, or a _MSC_FULL_VER value.
-  // Additionally, the value may be provided in the form of a more readable
-  // MM.mm.bbbbb.pp version.
-  //
-  // Unfortunately, due to the bit-width limitations, we cannot currently encode
-  // the value for the patch level.
-
-  unsigned VC[4] = {0};
-  StringRef Value = Arg->getValue();
-  SmallVector<StringRef, 4> Components;
-
-  Value.split(Components, ".", llvm::array_lengthof(VC));
-  for (unsigned CI = 0,
-                CE = std::min(Components.size(), llvm::array_lengthof(VC));
-       CI < CE; ++CI) {
-    if (Components[CI].getAsInteger(10, VC[CI])) {
-      Diags.Report(diag::err_drv_invalid_value)
-        << Arg->getAsString(Args) << Value;
-      return 0;
-    }
-  }
-
-  // FIXME we cannot encode the patch level
-  return VC[0] * 10000000 + VC[1] * 100000 + VC[2];
-}
-
 static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
                           DiagnosticsEngine &Diags) {
   // FIXME: Cleanup per-file based stuff.
@@ -1336,6 +1336,7 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
             << A->getAsString(Args) << "OpenCL";
         break;
       case IK_CUDA:
+      case IK_PreprocessedCuda:
         if (!Std.isCPlusPlus())
           Diags.Report(diag::err_drv_argument_not_allowed_with)
             << A->getAsString(Args) << "CUDA";
@@ -1381,6 +1382,12 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
   if (Args.hasArg(OPT_fcuda_is_device))
     Opts.CUDAIsDevice = 1;
 
+  if (Args.hasArg(OPT_fcuda_allow_host_calls_from_host_device))
+    Opts.CUDAAllowHostCallsFromHostDevice = 1;
+
+  if (Args.hasArg(OPT_fcuda_disable_target_call_checks))
+    Opts.CUDADisableTargetCallChecks = 1;
+
   if (Opts.ObjC1) {
     if (Arg *arg = Args.getLastArg(OPT_fobjc_runtime_EQ)) {
       StringRef value = arg->getValue();
@@ -1412,8 +1419,13 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
         (Opts.ObjCRuntime.getKind() == ObjCRuntime::FragileMacOSX);
   }
     
-  if (Args.hasArg(OPT_fgnu89_inline))
-    Opts.GNUInline = 1;
+  if (Args.hasArg(OPT_fgnu89_inline)) {
+    if (Opts.CPlusPlus)
+      Diags.Report(diag::err_drv_argument_not_allowed_with) << "-fgnu89-inline"
+                                                            << "C++/ObjC++";
+    else
+      Opts.GNUInline = 1;
+  }
 
   if (Args.hasArg(OPT_fapple_kext)) {
     if (!Opts.CPlusPlus)
@@ -1462,7 +1474,16 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
   Opts.MSVCCompat = Args.hasArg(OPT_fms_compatibility);
   Opts.MicrosoftExt = Opts.MSVCCompat || Args.hasArg(OPT_fms_extensions);
   Opts.AsmBlocks = Args.hasArg(OPT_fasm_blocks) || Opts.MicrosoftExt;
-  Opts.MSCompatibilityVersion = parseMSCVersion(Args, Diags);
+  Opts.MSCompatibilityVersion = 0;
+  if (const Arg *A = Args.getLastArg(OPT_fms_compatibility_version)) {
+    VersionTuple VT;
+    if (VT.tryParse(A->getValue()))
+      Diags.Report(diag::err_drv_invalid_value) << A->getAsString(Args)
+                                                << A->getValue();
+    Opts.MSCompatibilityVersion = VT.getMajor() * 10000000 +
+                                  VT.getMinor().getValueOr(0) * 100000 +
+                                  VT.getSubminor().getValueOr(0);
+  }
 
   // Mimicing gcc's behavior, trigraphs are only enabled if -trigraphs
   // is specified, or -std is set to a conforming mode.
@@ -1498,12 +1519,15 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
   Opts.ModulesStrictDeclUse = Args.hasArg(OPT_fmodules_strict_decluse);
   Opts.ModulesDeclUse =
       Args.hasArg(OPT_fmodules_decluse) || Opts.ModulesStrictDeclUse;
+  Opts.ModulesLocalVisibility =
+      Args.hasArg(OPT_fmodules_local_submodule_visibility);
   Opts.ModulesSearchAll = Opts.Modules &&
     !Args.hasArg(OPT_fno_modules_search_all) &&
     Args.hasArg(OPT_fmodules_search_all);
   Opts.ModulesErrorRecovery = !Args.hasArg(OPT_fno_modules_error_recovery);
   Opts.ModulesImplicitMaps = Args.hasFlag(OPT_fmodules_implicit_maps,
                                           OPT_fno_modules_implicit_maps, true);
+  Opts.ImplicitModules = !Args.hasArg(OPT_fno_implicit_modules);
   Opts.CharIsSigned = Opts.OpenCL || !Args.hasArg(OPT_fno_signed_char);
   Opts.WChar = Opts.CPlusPlus && !Args.hasArg(OPT_fno_wchar);
   Opts.ShortWChar = Args.hasFlag(OPT_fshort_wchar, OPT_fno_short_wchar, false);
@@ -1512,7 +1536,8 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
   Opts.NoBuiltin = Args.hasArg(OPT_fno_builtin) || Opts.Freestanding;
   Opts.NoMathBuiltin = Args.hasArg(OPT_fno_math_builtin);
   Opts.AssumeSaneOperatorNew = !Args.hasArg(OPT_fno_assume_sane_operator_new);
-  Opts.SizedDeallocation |= Args.hasArg(OPT_fsized_deallocation);
+  Opts.SizedDeallocation = Args.hasArg(OPT_fsized_deallocation);
+  Opts.ConceptsTS = Args.hasArg(OPT_fconcepts_ts);
   Opts.HeinousExtensions = Args.hasArg(OPT_fheinous_gnu_extensions);
   Opts.AccessControl = !Args.hasArg(OPT_fno_access_control);
   Opts.ElideConstructors = !Args.hasArg(OPT_fno_elide_constructors);
@@ -1559,10 +1584,13 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
   Opts.DebuggerObjCLiteral = Args.hasArg(OPT_fdebugger_objc_literal);
   Opts.ApplePragmaPack = Args.hasArg(OPT_fapple_pragma_pack);
   Opts.CurrentModule = Args.getLastArgValue(OPT_fmodule_name);
+  Opts.AppExt = Args.hasArg(OPT_fapplication_extension);
   Opts.ImplementationOfModule =
       Args.getLastArgValue(OPT_fmodule_implementation_of);
-  Opts.NativeHalfType = Opts.NativeHalfType;
+  Opts.ModuleFeatures = Args.getAllArgValues(OPT_fmodule_feature);
+  Opts.NativeHalfType |= Args.hasArg(OPT_fnative_half_type);
   Opts.HalfArgsAndReturns = Args.hasArg(OPT_fallow_half_arguments_and_returns);
+  Opts.GNUAsm = !Args.hasArg(OPT_fno_gnu_inline_asm);
 
   if (!Opts.CurrentModule.empty() && !Opts.ImplementationOfModule.empty() &&
       Opts.CurrentModule != Opts.ImplementationOfModule) {
@@ -1570,6 +1598,12 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
         << Opts.CurrentModule << Opts.ImplementationOfModule;
   }
 
+  // For now, we only support local submodule visibility in C++ (because we
+  // heavily depend on the ODR for merging redefinitions).
+  if (Opts.ModulesLocalVisibility && !Opts.CPlusPlus)
+    Diags.Report(diag::err_drv_argument_not_allowed_with)
+        << "-fmodules-local-submodule-visibility" << "C";
+
   if (Arg *A = Args.getLastArg(OPT_faddress_space_map_mangling_EQ)) {
     switch (llvm::StringSwitch<unsigned>(A->getValue())
       .Case("target", LangOptions::ASMM_Target)
@@ -1610,12 +1644,8 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
     Opts.setMSPointerToMemberRepresentationMethod(InheritanceModel);
   }
 
-  // Check if -fopenmp= is specified.
-  if (const Arg *A = Args.getLastArg(options::OPT_fopenmp_EQ)) {
-    Opts.OpenMP = llvm::StringSwitch<bool>(A->getValue())
-        .Case("libiomp5", true)
-        .Default(false);
-  }
+  // Check if -fopenmp is specified.
+  Opts.OpenMP = Args.hasArg(options::OPT_fopenmp);
 
   // Record whether the __DEPRECATED define was requested.
   Opts.Deprecated = Args.hasFlag(OPT_fdeprecated_macro,
@@ -1660,7 +1690,7 @@ static void ParseLangArgs(LangOptions &Opts, ArgList &Args, InputKind IK,
   // -fsanitize-address-field-padding=N has to be a LangOpt, parse it here.
   Opts.SanitizeAddressFieldPadding =
       getLastArgIntValue(Args, OPT_fsanitize_address_field_padding, 0, Diags);
-  Opts.SanitizerBlacklistFile = Args.getLastArgValue(OPT_fsanitize_blacklist);
+  Opts.SanitizerBlacklistFiles = Args.getAllArgValues(OPT_fsanitize_blacklist);
 }
 
 static void ParsePreprocessorArgs(PreprocessorOptions &Opts, ArgList &Args,
@@ -1805,6 +1835,7 @@ static void ParsePreprocessorOutputArgs(PreprocessorOutputOptions &Opts,
   Opts.ShowMacroComments = Args.hasArg(OPT_CC);
   Opts.ShowMacros = Args.hasArg(OPT_dM) || Args.hasArg(OPT_dD);
   Opts.RewriteIncludes = Args.hasArg(OPT_frewrite_includes);
+  Opts.UseLineDirectives = Args.hasArg(OPT_fuse_line_directives);
 }
 
 static void ParseTargetArgs(TargetOptions &Opts, ArgList &Args) {
@@ -2002,7 +2033,7 @@ std::string CompilerInvocation::getModuleHash() const {
     llvm::sys::path::append(systemVersionFile, "SystemVersion.plist");
 
     llvm::ErrorOr<std::unique_ptr<llvm::MemoryBuffer>> buffer =
-        llvm::MemoryBuffer::getFile(systemVersionFile.str());
+        llvm::MemoryBuffer::getFile(systemVersionFile);
     if (buffer) {
       code = hash_combine(code, buffer.get()->getBuffer());
 
diff --git a/lib/Frontend/DependencyFile.cpp b/lib/Frontend/DependencyFile.cpp
index 6ea8f5193ef7..0995ab4bf077 100644
--- a/lib/Frontend/DependencyFile.cpp
+++ b/lib/Frontend/DependencyFile.cpp
@@ -150,6 +150,8 @@ class DFGImpl : public PPCallbacks {
   bool AddMissingHeaderDeps;
   bool SeenMissingHeader;
   bool IncludeModuleFiles;
+  DependencyOutputFormat OutputFormat;
+
 private:
   bool FileMatchesDepCriteria(const char *Filename,
                               SrcMgr::CharacteristicKind FileType);
@@ -162,7 +164,8 @@ public:
       PhonyTarget(Opts.UsePhonyTargets),
       AddMissingHeaderDeps(Opts.AddMissingHeaderDeps),
       SeenMissingHeader(false),
-      IncludeModuleFiles(Opts.IncludeModuleFiles) {}
+      IncludeModuleFiles(Opts.IncludeModuleFiles),
+      OutputFormat(Opts.OutputFormat) {}
 
   void FileChanged(SourceLocation Loc, FileChangeReason Reason,
                    SrcMgr::CharacteristicKind FileType,
@@ -289,13 +292,76 @@ void DFGImpl::AddFilename(StringRef Filename) {
     Files.push_back(Filename);
 }
 
-/// PrintFilename - GCC escapes spaces, # and $, but apparently not ' or " or
-/// other scary characters.
-static void PrintFilename(raw_ostream &OS, StringRef Filename) {
+/// Print the filename, with escaping or quoting that accommodates the three
+/// most likely tools that use dependency files: GNU Make, BSD Make, and
+/// NMake/Jom.
+///
+/// BSD Make is the simplest case: It does no escaping at all.  This means
+/// characters that are normally delimiters, i.e. space and # (the comment
+/// character) simply aren't supported in filenames.
+///
+/// GNU Make does allow space and # in filenames, but to avoid being treated
+/// as a delimiter or comment, these must be escaped with a backslash. Because
+/// backslash is itself the escape character, if a backslash appears in a
+/// filename, it should be escaped as well.  (As a special case, $ is escaped
+/// as $$, which is the normal Make way to handle the $ character.)
+/// For compatibility with BSD Make and historical practice, if GNU Make
+/// un-escapes characters in a filename but doesn't find a match, it will
+/// retry with the unmodified original string.
+///
+/// GCC tries to accommodate both Make formats by escaping any space or #
+/// characters in the original filename, but not escaping backslashes.  The
+/// apparent intent is so that filenames with backslashes will be handled
+/// correctly by BSD Make, and by GNU Make in its fallback mode of using the
+/// unmodified original string; filenames with # or space characters aren't
+/// supported by BSD Make at all, but will be handled correctly by GNU Make
+/// due to the escaping.
+///
+/// A corner case that GCC gets only partly right is when the original filename
+/// has a backslash immediately followed by space or #.  GNU Make would expect
+/// this backslash to be escaped; however GCC escapes the original backslash
+/// only when followed by space, not #.  It will therefore take a dependency
+/// from a directive such as
+///     #include "a\ b\#c.h"
+/// and emit it as
+///     a\\\ b\\#c.h
+/// which GNU Make will interpret as
+///     a\ b\
+/// followed by a comment. Failing to find this file, it will fall back to the
+/// original string, which probably doesn't exist either; in any case it won't
+/// find
+///     a\ b\#c.h
+/// which is the actual filename specified by the include directive.
+///
+/// Clang does what GCC does, rather than what GNU Make expects.
+///
+/// NMake/Jom has a different set of scary characters, but wraps filespecs in
+/// double-quotes to avoid misinterpreting them; see
+/// https://msdn.microsoft.com/en-us/library/dd9y37ha.aspx for NMake info,
+/// https://msdn.microsoft.com/en-us/library/windows/desktop/aa365247(v=vs.85).aspx
+/// for Windows file-naming info.
+static void PrintFilename(raw_ostream &OS, StringRef Filename,
+                          DependencyOutputFormat OutputFormat) {
+  if (OutputFormat == DependencyOutputFormat::NMake) {
+    // Add quotes if needed. These are the characters listed as "special" to
+    // NMake, that are legal in a Windows filespec, and that could cause
+    // misinterpretation of the dependency string.
+    if (Filename.find_first_of(" #${}^!") != StringRef::npos)
+      OS << '\"' << Filename << '\"';
+    else
+      OS << Filename;
+    return;
+  }
+  assert(OutputFormat == DependencyOutputFormat::Make);
   for (unsigned i = 0, e = Filename.size(); i != e; ++i) {
-    if (Filename[i] == ' ' || Filename[i] == '#')
+    if (Filename[i] == '#') // Handle '#' the broken gcc way.
+      OS << '\\';
+    else if (Filename[i] == ' ') { // Handle space correctly.
       OS << '\\';
-    else if (Filename[i] == '$') // $ is escaped by $$.
+      unsigned j = i;
+      while (j > 0 && Filename[--j] == '\\')
+        OS << '\\';
+    } else if (Filename[i] == '$') // $ is escaped by $$.
       OS << '$';
     OS << Filename[i];
   }
@@ -354,7 +420,7 @@ void DFGImpl::OutputDependencyFile() {
       Columns = 2;
     }
     OS << ' ';
-    PrintFilename(OS, *I);
+    PrintFilename(OS, *I, OutputFormat);
     Columns += N + 1;
   }
   OS << '\n';
@@ -365,7 +431,7 @@ void DFGImpl::OutputDependencyFile() {
     for (std::vector<std::string>::iterator I = Files.begin() + 1,
            E = Files.end(); I != E; ++I) {
       OS << '\n';
-      PrintFilename(OS, *I);
+      PrintFilename(OS, *I, OutputFormat);
       OS << ":\n";
     }
   }
diff --git a/lib/Frontend/FrontendAction.cpp b/lib/Frontend/FrontendAction.cpp
index c81c81aba4d2..9bba75541ffb 100644
--- a/lib/Frontend/FrontendAction.cpp
+++ b/lib/Frontend/FrontendAction.cpp
@@ -44,7 +44,7 @@ public:
   explicit DelegatingDeserializationListener(
       ASTDeserializationListener *Previous, bool DeletePrevious)
       : Previous(Previous), DeletePrevious(DeletePrevious) {}
-  virtual ~DelegatingDeserializationListener() {
+  ~DelegatingDeserializationListener() override {
     if (DeletePrevious)
       delete Previous;
   }
@@ -71,7 +71,7 @@ public:
       Previous->SelectorRead(ID, Sel);
   }
   void MacroDefinitionRead(serialization::PreprocessedEntityID PPID,
-                           MacroDefinition *MD) override {
+                           MacroDefinitionRecord *MD) override {
     if (Previous)
       Previous->MacroDefinitionRead(PPID, MD);
   }
@@ -262,18 +262,20 @@ bool FrontendAction::BeginSourceFile(CompilerInstance &CI,
     FileManager &FileMgr = CI.getFileManager();
     PreprocessorOptions &PPOpts = CI.getPreprocessorOpts();
     StringRef PCHInclude = PPOpts.ImplicitPCHInclude;
+    std::string SpecificModuleCachePath = CI.getSpecificModuleCachePath();
     if (const DirectoryEntry *PCHDir = FileMgr.getDirectory(PCHInclude)) {
       std::error_code EC;
       SmallString<128> DirNative;
       llvm::sys::path::native(PCHDir->getName(), DirNative);
       bool Found = false;
-      for (llvm::sys::fs::directory_iterator Dir(DirNative.str(), EC), DirEnd;
+      for (llvm::sys::fs::directory_iterator Dir(DirNative, EC), DirEnd;
            Dir != DirEnd && !EC; Dir.increment(EC)) {
         // Check whether this is an acceptable AST file.
         if (ASTReader::isAcceptableASTFile(Dir->path(), FileMgr,
                                            CI.getLangOpts(),
                                            CI.getTargetOpts(),
-                                           CI.getPreprocessorOpts())) {
+                                           CI.getPreprocessorOpts(),
+                                           SpecificModuleCachePath)) {
           PPOpts.ImplicitPCHInclude = Dir->path();
           Found = true;
           break;
@@ -383,6 +385,15 @@ bool FrontendAction::BeginSourceFile(CompilerInstance &CI,
            "doesn't support modules");
   }
 
+  // If we were asked to load any module map files, do so now.
+  for (const auto &Filename : CI.getFrontendOpts().ModuleMapFiles) {
+    if (auto *File = CI.getFileManager().getFile(Filename))
+      CI.getPreprocessor().getHeaderSearchInfo().loadModuleMapFile(
+          File, /*IsSystem*/false);
+    else
+      CI.getDiagnostics().Report(diag::err_module_map_not_found) << Filename;
+  }
+
   // If we were asked to load any module files, do so now.
   for (const auto &ModuleFile : CI.getFrontendOpts().ModuleFiles)
     if (!CI.loadModuleFile(ModuleFile))
@@ -457,16 +468,12 @@ void FrontendAction::EndSourceFile() {
   // FIXME: There is more per-file stuff we could just drop here?
   bool DisableFree = CI.getFrontendOpts().DisableFree;
   if (DisableFree) {
-    if (!isCurrentFileAST()) {
-      CI.resetAndLeakSema();
-      CI.resetAndLeakASTContext();
-    }
+    CI.resetAndLeakSema();
+    CI.resetAndLeakASTContext();
     BuryPointer(CI.takeASTConsumer().get());
   } else {
-    if (!isCurrentFileAST()) {
-      CI.setSema(nullptr);
-      CI.setASTContext(nullptr);
-    }
+    CI.setSema(nullptr);
+    CI.setASTContext(nullptr);
     CI.setASTConsumer(nullptr);
   }
 
@@ -483,13 +490,16 @@ void FrontendAction::EndSourceFile() {
   // FrontendAction.
   CI.clearOutputFiles(/*EraseFiles=*/shouldEraseOutputFiles());
 
-  // FIXME: Only do this if DisableFree is set.
   if (isCurrentFileAST()) {
-    CI.resetAndLeakSema();
-    CI.resetAndLeakASTContext();
-    CI.resetAndLeakPreprocessor();
-    CI.resetAndLeakSourceManager();
-    CI.resetAndLeakFileManager();
+    if (DisableFree) {
+      CI.resetAndLeakPreprocessor();
+      CI.resetAndLeakSourceManager();
+      CI.resetAndLeakFileManager();
+    } else {
+      CI.setPreprocessor(nullptr);
+      CI.setSourceManager(nullptr);
+      CI.setFileManager(nullptr);
+    }
   }
 
   setCompilerInstance(nullptr);
diff --git a/lib/Frontend/FrontendActions.cpp b/lib/Frontend/FrontendActions.cpp
index 701ef026d49c..46cdeebfdbeb 100644
--- a/lib/Frontend/FrontendActions.cpp
+++ b/lib/Frontend/FrontendActions.cpp
@@ -79,8 +79,9 @@ std::unique_ptr<ASTConsumer>
 GeneratePCHAction::CreateASTConsumer(CompilerInstance &CI, StringRef InFile) {
   std::string Sysroot;
   std::string OutputFile;
-  raw_ostream *OS = nullptr;
-  if (ComputeASTConsumerArguments(CI, InFile, Sysroot, OutputFile, OS))
+  raw_ostream *OS =
+      ComputeASTConsumerArguments(CI, InFile, Sysroot, OutputFile);
+  if (!OS)
     return nullptr;
 
   if (!CI.getFrontendOpts().RelocatablePCH)
@@ -89,28 +90,27 @@ GeneratePCHAction::CreateASTConsumer(CompilerInstance &CI, StringRef InFile) {
                                          nullptr, Sysroot, OS);
 }
 
-bool GeneratePCHAction::ComputeASTConsumerArguments(CompilerInstance &CI,
-                                                    StringRef InFile,
-                                                    std::string &Sysroot,
-                                                    std::string &OutputFile,
-                                                    raw_ostream *&OS) {
+raw_ostream *GeneratePCHAction::ComputeASTConsumerArguments(
+    CompilerInstance &CI, StringRef InFile, std::string &Sysroot,
+    std::string &OutputFile) {
   Sysroot = CI.getHeaderSearchOpts().Sysroot;
   if (CI.getFrontendOpts().RelocatablePCH && Sysroot.empty()) {
     CI.getDiagnostics().Report(diag::err_relocatable_without_isysroot);
-    return true;
+    return nullptr;
   }
 
   // We use createOutputFile here because this is exposed via libclang, and we
   // must disable the RemoveFileOnSignal behavior.
   // We use a temporary to avoid race conditions.
-  OS = CI.createOutputFile(CI.getFrontendOpts().OutputFile, /*Binary=*/true,
-                           /*RemoveFileOnSignal=*/false, InFile,
-                           /*Extension=*/"", /*useTemporary=*/true);
+  raw_ostream *OS =
+      CI.createOutputFile(CI.getFrontendOpts().OutputFile, /*Binary=*/true,
+                          /*RemoveFileOnSignal=*/false, InFile,
+                          /*Extension=*/"", /*useTemporary=*/true);
   if (!OS)
-    return true;
+    return nullptr;
 
   OutputFile = CI.getFrontendOpts().OutputFile;
-  return false;
+  return OS;
 }
 
 std::unique_ptr<ASTConsumer>
@@ -118,8 +118,9 @@ GenerateModuleAction::CreateASTConsumer(CompilerInstance &CI,
                                         StringRef InFile) {
   std::string Sysroot;
   std::string OutputFile;
-  raw_ostream *OS = nullptr;
-  if (ComputeASTConsumerArguments(CI, InFile, Sysroot, OutputFile, OS))
+  raw_ostream *OS =
+      ComputeASTConsumerArguments(CI, InFile, Sysroot, OutputFile);
+  if (!OS)
     return nullptr;
 
   return llvm::make_unique<PCHGenerator>(CI.getPreprocessor(), OutputFile,
@@ -151,22 +152,6 @@ static std::error_code addHeaderInclude(StringRef HeaderName,
   return std::error_code();
 }
 
-static std::error_code addHeaderInclude(const FileEntry *Header,
-                                        SmallVectorImpl<char> &Includes,
-                                        const LangOptions &LangOpts,
-                                        bool IsExternC) {
-  // Use an absolute path if we don't have a filename as written in the module
-  // map file; this ensures that we will identify the right file independent of
-  // header search paths.
-  if (llvm::sys::path::is_absolute(Header->getName()))
-    return addHeaderInclude(Header->getName(), Includes, LangOpts, IsExternC);
-
-  SmallString<256> AbsName(Header->getName());
-  if (std::error_code Err = llvm::sys::fs::make_absolute(AbsName))
-    return Err;
-  return addHeaderInclude(AbsName, Includes, LangOpts, IsExternC);
-}
-
 /// \brief Collect the set of header includes needed to construct the given 
 /// module and update the TopHeaders file set of the module.
 ///
@@ -195,21 +180,21 @@ collectModuleHeaderIncludes(const LangOptions &LangOpts, FileManager &FileMgr,
   }
   // Note that Module->PrivateHeaders will not be a TopHeader.
 
-  if (const FileEntry *UmbrellaHeader = Module->getUmbrellaHeader()) {
-    // FIXME: Track the name as written here.
-    Module->addTopHeader(UmbrellaHeader);
+  if (Module::Header UmbrellaHeader = Module->getUmbrellaHeader()) {
+    Module->addTopHeader(UmbrellaHeader.Entry);
     if (Module->Parent) {
       // Include the umbrella header for submodules.
-      if (std::error_code Err = addHeaderInclude(UmbrellaHeader, Includes,
-                                                 LangOpts, Module->IsExternC))
+      if (std::error_code Err = addHeaderInclude(UmbrellaHeader.NameAsWritten,
+                                                 Includes, LangOpts,
+                                                 Module->IsExternC))
         return Err;
     }
-  } else if (const DirectoryEntry *UmbrellaDir = Module->getUmbrellaDir()) {
+  } else if (Module::DirectoryName UmbrellaDir = Module->getUmbrellaDir()) {
     // Add all of the headers we find in this subdirectory.
     std::error_code EC;
     SmallString<128> DirNative;
-    llvm::sys::path::native(UmbrellaDir->getName(), DirNative);
-    for (llvm::sys::fs::recursive_directory_iterator Dir(DirNative.str(), EC), 
+    llvm::sys::path::native(UmbrellaDir.Entry->getName(), DirNative);
+    for (llvm::sys::fs::recursive_directory_iterator Dir(DirNative, EC), 
                                                      DirEnd;
          Dir != DirEnd && !EC; Dir.increment(EC)) {
       // Check whether this entry has an extension typically associated with 
@@ -230,11 +215,20 @@ collectModuleHeaderIncludes(const LangOptions &LangOpts, FileManager &FileMgr,
       if (ModMap.isHeaderUnavailableInModule(Header, Module))
         continue;
 
+      // Compute the relative path from the directory to this file.
+      SmallVector<StringRef, 16> Components;
+      auto PathIt = llvm::sys::path::rbegin(Dir->path());
+      for (int I = 0; I != Dir.level() + 1; ++I, ++PathIt)
+        Components.push_back(*PathIt);
+      SmallString<128> RelativeHeader(UmbrellaDir.NameAsWritten);
+      for (auto It = Components.rbegin(), End = Components.rend(); It != End;
+           ++It)
+        llvm::sys::path::append(RelativeHeader, *It);
+
       // Include this header as part of the umbrella directory.
-      // FIXME: Track the name as written through to here.
       Module->addTopHeader(Header);
-      if (std::error_code Err =
-              addHeaderInclude(Header, Includes, LangOpts, Module->IsExternC))
+      if (std::error_code Err = addHeaderInclude(RelativeHeader, Includes,
+                                                 LangOpts, Module->IsExternC))
         return Err;
     }
 
@@ -326,10 +320,9 @@ bool GenerateModuleAction::BeginSourceFileAction(CompilerInstance &CI,
   // Collect the set of #includes we need to build the module.
   SmallString<256> HeaderContents;
   std::error_code Err = std::error_code();
-  if (const FileEntry *UmbrellaHeader = Module->getUmbrellaHeader())
-    // FIXME: Track the file name as written.
-    Err = addHeaderInclude(UmbrellaHeader, HeaderContents, CI.getLangOpts(),
-                           Module->IsExternC);
+  if (Module::Header UmbrellaHeader = Module->getUmbrellaHeader())
+    Err = addHeaderInclude(UmbrellaHeader.NameAsWritten, HeaderContents,
+                           CI.getLangOpts(), Module->IsExternC);
   if (!Err)
     Err = collectModuleHeaderIncludes(
         CI.getLangOpts(), FileMgr,
@@ -355,11 +348,9 @@ bool GenerateModuleAction::BeginSourceFileAction(CompilerInstance &CI,
   return true;
 }
 
-bool GenerateModuleAction::ComputeASTConsumerArguments(CompilerInstance &CI,
-                                                       StringRef InFile,
-                                                       std::string &Sysroot,
-                                                       std::string &OutputFile,
-                                                       raw_ostream *&OS) {
+raw_ostream *GenerateModuleAction::ComputeASTConsumerArguments(
+    CompilerInstance &CI, StringRef InFile, std::string &Sysroot,
+    std::string &OutputFile) {
   // If no output file was provided, figure out where this module would go
   // in the module cache.
   if (CI.getFrontendOpts().OutputFile.empty()) {
@@ -368,19 +359,20 @@ bool GenerateModuleAction::ComputeASTConsumerArguments(CompilerInstance &CI,
         HS.getModuleFileName(CI.getLangOpts().CurrentModule,
                              ModuleMapForUniquing->getName());
   }
-  
+
   // We use createOutputFile here because this is exposed via libclang, and we
   // must disable the RemoveFileOnSignal behavior.
   // We use a temporary to avoid race conditions.
-  OS = CI.createOutputFile(CI.getFrontendOpts().OutputFile, /*Binary=*/true,
-                           /*RemoveFileOnSignal=*/false, InFile,
-                           /*Extension=*/"", /*useTemporary=*/true,
-                           /*CreateMissingDirectories=*/true);
+  raw_ostream *OS =
+      CI.createOutputFile(CI.getFrontendOpts().OutputFile, /*Binary=*/true,
+                          /*RemoveFileOnSignal=*/false, InFile,
+                          /*Extension=*/"", /*useTemporary=*/true,
+                          /*CreateMissingDirectories=*/true);
   if (!OS)
-    return true;
-  
+    return nullptr;
+
   OutputFile = CI.getFrontendOpts().OutputFile;
-  return false;
+  return OS;
 }
 
 std::unique_ptr<ASTConsumer>
@@ -462,8 +454,8 @@ namespace {
       return false;
     }
 
-    bool ReadTargetOptions(const TargetOptions &TargetOpts,
-                           bool Complain) override {
+    bool ReadTargetOptions(const TargetOptions &TargetOpts, bool Complain,
+                           bool AllowCompatibleDifferences) override {
       Out.indent(2) << "Target options:\n";
       Out.indent(4) << "  Triple: " << TargetOpts.Triple << "\n";
       Out.indent(4) << "  CPU: " << TargetOpts.CPU << "\n";
@@ -480,9 +472,8 @@ namespace {
       return false;
     }
 
-    virtual bool
-    ReadDiagnosticOptions(IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts,
-                          bool Complain) override {
+    bool ReadDiagnosticOptions(IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts,
+                               bool Complain) override {
       Out.indent(2) << "Diagnostic options:\n";
 #define DIAGOPT(Name, Bits, Default) DUMP_BOOLEAN(DiagOpts->Name, #Name);
 #define ENUM_DIAGOPT(Name, Type, Bits, Default) \
@@ -501,9 +492,11 @@ namespace {
     }
 
     bool ReadHeaderSearchOptions(const HeaderSearchOptions &HSOpts,
+                                 StringRef SpecificModuleCachePath,
                                  bool Complain) override {
       Out.indent(2) << "Header search options:\n";
       Out.indent(4) << "System root [-isysroot=]: '" << HSOpts.Sysroot << "'\n";
+      Out.indent(4) << "Module Cache: '" << SpecificModuleCachePath << "'\n";
       DUMP_BOOLEAN(HSOpts.UseBuiltinIncludes,
                    "Use builtin include directories [-nobuiltininc]");
       DUMP_BOOLEAN(HSOpts.UseStandardSystemIncludes,
@@ -598,15 +591,9 @@ void DumpTokensAction::ExecuteAction() {
 
 void GeneratePTHAction::ExecuteAction() {
   CompilerInstance &CI = getCompilerInstance();
-  if (CI.getFrontendOpts().OutputFile.empty() ||
-      CI.getFrontendOpts().OutputFile == "-") {
-    // FIXME: Don't fail this way.
-    // FIXME: Verify that we can actually seek in the given file.
-    llvm::report_fatal_error("PTH requires a seekable file for output!");
-  }
-  llvm::raw_fd_ostream *OS =
-    CI.createDefaultOutputFile(true, getCurrentFile());
-  if (!OS) return;
+  raw_pwrite_stream *OS = CI.createDefaultOutputFile(true, getCurrentFile());
+  if (!OS)
+    return;
 
   CacheTokens(CI.getPreprocessor(), OS);
 }
@@ -688,6 +675,7 @@ void PrintPreambleAction::ExecuteAction() {
   case IK_None:
   case IK_Asm:
   case IK_PreprocessedC:
+  case IK_PreprocessedCuda:
   case IK_PreprocessedCXX:
   case IK_PreprocessedObjC:
   case IK_PreprocessedObjCXX:
diff --git a/lib/Frontend/FrontendOptions.cpp b/lib/Frontend/FrontendOptions.cpp
index 1869d0c78d19..9ede674e47ea 100644
--- a/lib/Frontend/FrontendOptions.cpp
+++ b/lib/Frontend/FrontendOptions.cpp
@@ -18,6 +18,7 @@ InputKind FrontendOptions::getInputKindForExtension(StringRef Extension) {
     .Cases("S", "s", IK_Asm)
     .Case("i", IK_PreprocessedC)
     .Case("ii", IK_PreprocessedCXX)
+    .Case("cui", IK_PreprocessedCuda)
     .Case("m", IK_ObjC)
     .Case("mi", IK_PreprocessedObjC)
     .Cases("mm", "M", IK_ObjCXX)
diff --git a/lib/Frontend/HeaderIncludeGen.cpp b/lib/Frontend/HeaderIncludeGen.cpp
index 27011945712a..5732e5b3fb73 100644
--- a/lib/Frontend/HeaderIncludeGen.cpp
+++ b/lib/Frontend/HeaderIncludeGen.cpp
@@ -35,7 +35,7 @@ public:
       OwnsOutputFile(OwnsOutputFile_), ShowAllHeaders(ShowAllHeaders_),
       ShowDepth(ShowDepth_), MSStyle(MSStyle_) {}
 
-  ~HeaderIncludesCallback() {
+  ~HeaderIncludesCallback() override {
     if (OwnsOutputFile)
       delete OutputFile;
   }
diff --git a/lib/Frontend/InitHeaderSearch.cpp b/lib/Frontend/InitHeaderSearch.cpp
index a518a0a27a23..2bd999e8e530 100644
--- a/lib/Frontend/InitHeaderSearch.cpp
+++ b/lib/Frontend/InitHeaderSearch.cpp
@@ -227,10 +227,12 @@ void InitHeaderSearch::AddDefaultCIncludePaths(const llvm::Triple &triple,
 
   if (HSOpts.UseStandardSystemIncludes) {
     switch (os) {
+    case llvm::Triple::CloudABI:
     case llvm::Triple::FreeBSD:
     case llvm::Triple::NetBSD:
     case llvm::Triple::OpenBSD:
     case llvm::Triple::Bitrig:
+    case llvm::Triple::NaCl:
       break;
     default:
       // FIXME: temporary hack: hard-coded paths.
@@ -246,7 +248,7 @@ void InitHeaderSearch::AddDefaultCIncludePaths(const llvm::Triple &triple,
     // supplied path.
     SmallString<128> P = StringRef(HSOpts.ResourceDir);
     llvm::sys::path::append(P, "include");
-    AddUnmappedPath(P.str(), ExternCSystem, false);
+    AddUnmappedPath(P, ExternCSystem, false);
   }
 
   // All remaining additions are for system include directories, early exit if
@@ -270,6 +272,14 @@ void InitHeaderSearch::AddDefaultCIncludePaths(const llvm::Triple &triple,
   case llvm::Triple::Linux:
     llvm_unreachable("Include management is handled in the driver.");
 
+  case llvm::Triple::CloudABI: {
+    // <sysroot>/<triple>/include
+    SmallString<128> P = StringRef(HSOpts.ResourceDir);
+    llvm::sys::path::append(P, "../../..", triple.str(), "include");
+    AddPath(P, System, false);
+    break;
+  }
+
   case llvm::Triple::Haiku:
     AddPath("/boot/common/include", System, false);
     AddPath("/boot/develop/headers/os", System, false);
@@ -317,18 +327,18 @@ void InitHeaderSearch::AddDefaultCIncludePaths(const llvm::Triple &triple,
       // <sysroot>/i686-w64-mingw32/include
       SmallString<128> P = StringRef(HSOpts.ResourceDir);
       llvm::sys::path::append(P, "../../../i686-w64-mingw32/include");
-      AddPath(P.str(), System, false);
+      AddPath(P, System, false);
 
       // <sysroot>/x86_64-w64-mingw32/include
       P.resize(HSOpts.ResourceDir.size());
       llvm::sys::path::append(P, "../../../x86_64-w64-mingw32/include");
-      AddPath(P.str(), System, false);
+      AddPath(P, System, false);
 
       // mingw.org crt include paths
       // <sysroot>/include
       P.resize(HSOpts.ResourceDir.size());
       llvm::sys::path::append(P, "../../../include");
-      AddPath(P.str(), System, false);
+      AddPath(P, System, false);
       AddPath("/mingw/include", System, false);
 #if defined(LLVM_ON_WIN32)
       AddPath("c:/mingw/include", System, false); 
@@ -340,8 +350,15 @@ void InitHeaderSearch::AddDefaultCIncludePaths(const llvm::Triple &triple,
     break;
   }
 
-  if ( os != llvm::Triple::RTEMS )
+  switch (os) {
+  case llvm::Triple::CloudABI:
+  case llvm::Triple::RTEMS:
+  case llvm::Triple::NaCl:
+    break;
+  default:
     AddPath("/usr/include", ExternCSystem, false);
+    break;
+  }
 }
 
 void InitHeaderSearch::
@@ -488,7 +505,7 @@ void InitHeaderSearch::AddDefaultIncludePaths(const LangOptions &Lang,
 
           // Get foo/include/c++/v1
           llvm::sys::path::append(P, "include", "c++", "v1");
-          AddUnmappedPath(P.str(), CXXSystem, false);
+          AddUnmappedPath(P, CXXSystem, false);
         }
       }
       // On Solaris, include the support directory for things like xlocale and
@@ -699,7 +716,7 @@ void clang::ApplyHeaderSearchOptions(HeaderSearch &HS,
     // Set up the builtin include directory in the module map.
     SmallString<128> P = StringRef(HSOpts.ResourceDir);
     llvm::sys::path::append(P, "include");
-    if (const DirectoryEntry *Dir = HS.getFileMgr().getDirectory(P.str()))
+    if (const DirectoryEntry *Dir = HS.getFileMgr().getDirectory(P))
       HS.getModuleMap().setBuiltinIncludeDir(Dir);
   }
 
diff --git a/lib/Frontend/InitPreprocessor.cpp b/lib/Frontend/InitPreprocessor.cpp
index 3550ac25159b..dfc46f47cce3 100644
--- a/lib/Frontend/InitPreprocessor.cpp
+++ b/lib/Frontend/InitPreprocessor.cpp
@@ -133,6 +133,7 @@ static void DefineFloatMacros(MacroBuilder &Builder, StringRef Prefix,
                      "4.94065645841246544176568792868221e-324",
                      "6.47517511943802511092443895822764655e-4966");
   int Digits = PickFP(Sem, 6, 15, 18, 31, 33);
+  int DecimalDigits = PickFP(Sem, 9, 17, 21, 33, 36);
   Epsilon = PickFP(Sem, "1.19209290e-7", "2.2204460492503131e-16",
                    "1.08420217248550443401e-19",
                    "4.94065645841246544176568792868221e-324",
@@ -159,6 +160,7 @@ static void DefineFloatMacros(MacroBuilder &Builder, StringRef Prefix,
   Builder.defineMacro(DefPrefix + "DENORM_MIN__", Twine(DenormMin)+Ext);
   Builder.defineMacro(DefPrefix + "HAS_DENORM__");
   Builder.defineMacro(DefPrefix + "DIG__", Twine(Digits));
+  Builder.defineMacro(DefPrefix + "DECIMAL_DIG__", Twine(DecimalDigits));
   Builder.defineMacro(DefPrefix + "EPSILON__", Twine(Epsilon)+Ext);
   Builder.defineMacro(DefPrefix + "HAS_INFINITY__");
   Builder.defineMacro(DefPrefix + "HAS_QUIET_NAN__");
@@ -451,6 +453,8 @@ static void InitializeCPlusPlusFeatureTestMacros(const LangOptions &LangOpts,
   }
   if (LangOpts.SizedDeallocation)
     Builder.defineMacro("__cpp_sized_deallocation", "201309");
+  if (LangOpts.ConceptsTS)
+    Builder.defineMacro("__cpp_experimental_concepts", "1");
 }
 
 static void InitializePredefinedMacros(const TargetInfo &TI,
@@ -705,6 +709,10 @@ static void InitializePredefinedMacros(const TargetInfo &TI,
   Builder.defineMacro("__POINTER_WIDTH__",
                       Twine((int)TI.getPointerWidth(0)));
 
+  // Define __BIGGEST_ALIGNMENT__ to be compatible with gcc.
+  Builder.defineMacro("__BIGGEST_ALIGNMENT__",
+                      Twine(TI.getSuitableAlign() / TI.getCharWidth()) );
+
   if (!LangOpts.CharIsSigned)
     Builder.defineMacro("__CHAR_UNSIGNED__");
 
@@ -784,7 +792,7 @@ static void InitializePredefinedMacros(const TargetInfo &TI,
     Builder.defineMacro("__FINITE_MATH_ONLY__", "0");
 
   if (!LangOpts.MSVCCompat) {
-    if (LangOpts.GNUInline)
+    if (LangOpts.GNUInline || LangOpts.CPlusPlus)
       Builder.defineMacro("__GNUC_GNU_INLINE__");
     else
       Builder.defineMacro("__GNUC_STDC_INLINE__");
@@ -831,8 +839,7 @@ static void InitializePredefinedMacros(const TargetInfo &TI,
   // Macros to control C99 numerics and <float.h>
   Builder.defineMacro("__FLT_EVAL_METHOD__", Twine(TI.getFloatEvalMethod()));
   Builder.defineMacro("__FLT_RADIX__", "2");
-  int Dig = PickFP(&TI.getLongDoubleFormat(), -1/*FIXME*/, 17, 21, 33, 36);
-  Builder.defineMacro("__DECIMAL_DIG__", Twine(Dig));
+  Builder.defineMacro("__DECIMAL_DIG__", "__LDBL_DECIMAL_DIG__");
 
   if (LangOpts.getStackProtector() == LangOptions::SSPOn)
     Builder.defineMacro("__SSP__");
diff --git a/lib/Frontend/ModuleDependencyCollector.cpp b/lib/Frontend/ModuleDependencyCollector.cpp
index 62865e984f65..67852dc02036 100644
--- a/lib/Frontend/ModuleDependencyCollector.cpp
+++ b/lib/Frontend/ModuleDependencyCollector.cpp
@@ -77,10 +77,10 @@ std::error_code ModuleDependencyListener::copyToRoot(StringRef Src) {
   if (std::error_code EC = fs::create_directories(path::parent_path(Dest),
                                                    /*IgnoreExisting=*/true))
     return EC;
-  if (std::error_code EC = fs::copy_file(AbsoluteSrc.str(), Dest.str()))
+  if (std::error_code EC = fs::copy_file(AbsoluteSrc, Dest))
     return EC;
   // Use the absolute path under the root for the file mapping.
-  Collector.addFileMapping(AbsoluteSrc.str(), Dest.str());
+  Collector.addFileMapping(AbsoluteSrc, Dest);
   return std::error_code();
 }
 
diff --git a/lib/Frontend/MultiplexConsumer.cpp b/lib/Frontend/MultiplexConsumer.cpp
index 019882833d41..219e9492d28b 100644
--- a/lib/Frontend/MultiplexConsumer.cpp
+++ b/lib/Frontend/MultiplexConsumer.cpp
@@ -37,9 +37,10 @@ public:
   void DeclRead(serialization::DeclID ID, const Decl *D) override;
   void SelectorRead(serialization::SelectorID iD, Selector Sel) override;
   void MacroDefinitionRead(serialization::PreprocessedEntityID,
-                           MacroDefinition *MD) override;
+                           MacroDefinitionRecord *MD) override;
+
 private:
-  std::vector<ASTDeserializationListener*> Listeners;
+  std::vector<ASTDeserializationListener *> Listeners;
 };
 
 MultiplexASTDeserializationListener::MultiplexASTDeserializationListener(
@@ -78,7 +79,7 @@ void MultiplexASTDeserializationListener::SelectorRead(
 }
 
 void MultiplexASTDeserializationListener::MacroDefinitionRead(
-    serialization::PreprocessedEntityID ID, MacroDefinition *MD) {
+    serialization::PreprocessedEntityID ID, MacroDefinitionRecord *MD) {
   for (size_t i = 0, e = Listeners.size(); i != e; ++i)
     Listeners[i]->MacroDefinitionRead(ID, MD);
 }
@@ -99,6 +100,8 @@ public:
   void AddedCXXTemplateSpecialization(const FunctionTemplateDecl *TD,
                                       const FunctionDecl *D) override;
   void DeducedReturnType(const FunctionDecl *FD, QualType ReturnType) override;
+  void ResolvedOperatorDelete(const CXXDestructorDecl *DD,
+                              const FunctionDecl *Delete) override;
   void CompletedImplicitDefinition(const FunctionDecl *D) override;
   void StaticDataMemberInstantiated(const VarDecl *D) override;
   void AddedObjCCategoryToInterface(const ObjCCategoryDecl *CatD,
@@ -108,6 +111,7 @@ public:
                                     const ObjCCategoryDecl *ClassExt) override;
   void DeclarationMarkedUsed(const Decl *D) override;
   void DeclarationMarkedOpenMPThreadPrivate(const Decl *D) override;
+  void RedefinedHiddenDefinition(const NamedDecl *D, Module *M) override;
 
 private:
   std::vector<ASTMutationListener*> Listeners;
@@ -154,6 +158,11 @@ void MultiplexASTMutationListener::DeducedReturnType(const FunctionDecl *FD,
   for (size_t i = 0, e = Listeners.size(); i != e; ++i)
     Listeners[i]->DeducedReturnType(FD, ReturnType);
 }
+void MultiplexASTMutationListener::ResolvedOperatorDelete(
+    const CXXDestructorDecl *DD, const FunctionDecl *Delete) {
+  for (auto *L : Listeners)
+    L->ResolvedOperatorDelete(DD, Delete);
+}
 void MultiplexASTMutationListener::CompletedImplicitDefinition(
                                                         const FunctionDecl *D) {
   for (size_t i = 0, e = Listeners.size(); i != e; ++i)
@@ -186,6 +195,11 @@ void MultiplexASTMutationListener::DeclarationMarkedOpenMPThreadPrivate(
   for (size_t i = 0, e = Listeners.size(); i != e; ++i)
     Listeners[i]->DeclarationMarkedOpenMPThreadPrivate(D);
 }
+void MultiplexASTMutationListener::RedefinedHiddenDefinition(const NamedDecl *D,
+                                                             Module *M) {
+  for (auto *L : Listeners)
+    L->RedefinedHiddenDefinition(D, M);
+}
 
 }  // end namespace clang
 
@@ -292,10 +306,9 @@ void MultiplexConsumer::CompleteTentativeDefinition(VarDecl *D) {
     Consumer->CompleteTentativeDefinition(D);
 }
 
-void MultiplexConsumer::HandleVTable(
-    CXXRecordDecl *RD, bool DefinitionRequired) {
+void MultiplexConsumer::HandleVTable(CXXRecordDecl *RD) {
   for (auto &Consumer : Consumers)
-    Consumer->HandleVTable(RD, DefinitionRequired);
+    Consumer->HandleVTable(RD);
 }
 
 ASTMutationListener *MultiplexConsumer::GetASTMutationListener() {
diff --git a/lib/Frontend/PrintPreprocessedOutput.cpp b/lib/Frontend/PrintPreprocessedOutput.cpp
index 7c1d9a568831..037a6a525e2f 100644
--- a/lib/Frontend/PrintPreprocessedOutput.cpp
+++ b/lib/Frontend/PrintPreprocessedOutput.cpp
@@ -64,12 +64,11 @@ static void PrintMacroDefinition(const IdentifierInfo &II, const MacroInfo &MI,
     OS << ' ';
 
   SmallString<128> SpellingBuffer;
-  for (MacroInfo::tokens_iterator I = MI.tokens_begin(), E = MI.tokens_end();
-       I != E; ++I) {
-    if (I->hasLeadingSpace())
+  for (const auto &T : MI.tokens()) {
+    if (T.hasLeadingSpace())
       OS << ' ';
 
-    OS << PP.getSpelling(*I, SpellingBuffer);
+    OS << PP.getSpelling(T, SpellingBuffer);
   }
 }
 
@@ -94,14 +93,14 @@ private:
   bool Initialized;
   bool DisableLineMarkers;
   bool DumpDefines;
-  bool UseLineDirective;
+  bool UseLineDirectives;
   bool IsFirstFileEntered;
 public:
-  PrintPPOutputPPCallbacks(Preprocessor &pp, raw_ostream &os,
-                           bool lineMarkers, bool defines)
-     : PP(pp), SM(PP.getSourceManager()),
-       ConcatInfo(PP), OS(os), DisableLineMarkers(lineMarkers),
-       DumpDefines(defines) {
+  PrintPPOutputPPCallbacks(Preprocessor &pp, raw_ostream &os, bool lineMarkers,
+                           bool defines, bool UseLineDirectives)
+      : PP(pp), SM(PP.getSourceManager()), ConcatInfo(PP), OS(os),
+        DisableLineMarkers(lineMarkers), DumpDefines(defines),
+        UseLineDirectives(UseLineDirectives) {
     CurLine = 0;
     CurFilename += "<uninit>";
     EmittedTokensOnThisLine = false;
@@ -109,9 +108,6 @@ public:
     FileType = SrcMgr::C_User;
     Initialized = false;
     IsFirstFileEntered = false;
-
-    // If we're in microsoft mode, use normal #line instead of line markers.
-    UseLineDirective = PP.getLangOpts().MicrosoftExt;
   }
 
   void setEmittedTokensOnThisLine() { EmittedTokensOnThisLine = true; }
@@ -173,7 +169,7 @@ public:
 
   /// MacroUndefined - This hook is called whenever a macro #undef is seen.
   void MacroUndefined(const Token &MacroNameTok,
-                      const MacroDirective *MD) override;
+                      const MacroDefinition &MD) override;
 };
 }  // end anonymous namespace
 
@@ -183,7 +179,7 @@ void PrintPPOutputPPCallbacks::WriteLineInfo(unsigned LineNo,
   startNewLineIfNeeded(/*ShouldUpdateCurrentLine=*/false);
 
   // Emit #line directives or GNU line markers depending on what mode we're in.
-  if (UseLineDirective) {
+  if (UseLineDirectives) {
     OS << "#line" << ' ' << LineNo << ' ' << '"';
     OS.write_escaped(CurFilename);
     OS << '"';
@@ -364,7 +360,7 @@ void PrintPPOutputPPCallbacks::MacroDefined(const Token &MacroNameTok,
 }
 
 void PrintPPOutputPPCallbacks::MacroUndefined(const Token &MacroNameTok,
-                                              const MacroDirective *MD) {
+                                              const MacroDefinition &MD) {
   // Only print out macro definitions in -dD mode.
   if (!DumpDefines) return;
 
@@ -689,8 +685,9 @@ static void DoPrintMacros(Preprocessor &PP, raw_ostream *OS) {
   SmallVector<id_macro_pair, 128> MacrosByID;
   for (Preprocessor::macro_iterator I = PP.macro_begin(), E = PP.macro_end();
        I != E; ++I) {
-    if (I->first->hasMacroDefinition())
-      MacrosByID.push_back(id_macro_pair(I->first, I->second->getMacroInfo()));
+    auto *MD = I->second.getLatest();
+    if (MD && MD->isDefined())
+      MacrosByID.push_back(id_macro_pair(I->first, MD->getMacroInfo()));
   }
   llvm::array_pod_sort(MacrosByID.begin(), MacrosByID.end(), MacroIDCompare);
 
@@ -719,9 +716,8 @@ void clang::DoPrintPreprocessedInput(Preprocessor &PP, raw_ostream *OS,
   // to -C or -CC.
   PP.SetCommentRetentionState(Opts.ShowComments, Opts.ShowMacroComments);
 
-  PrintPPOutputPPCallbacks *Callbacks =
-      new PrintPPOutputPPCallbacks(PP, *OS, !Opts.ShowLineMarkers,
-                                   Opts.ShowMacros);
+  PrintPPOutputPPCallbacks *Callbacks = new PrintPPOutputPPCallbacks(
+      PP, *OS, !Opts.ShowLineMarkers, Opts.ShowMacros, Opts.UseLineDirectives);
   PP.AddPragmaHandler(new UnknownPragmaHandler("#pragma", Callbacks));
   PP.AddPragmaHandler("GCC", new UnknownPragmaHandler("#pragma GCC",Callbacks));
   PP.AddPragmaHandler("clang",
diff --git a/lib/Frontend/Rewrite/InclusionRewriter.cpp b/lib/Frontend/Rewrite/InclusionRewriter.cpp
index 140055735a99..b9ea051662c6 100644
--- a/lib/Frontend/Rewrite/InclusionRewriter.cpp
+++ b/lib/Frontend/Rewrite/InclusionRewriter.cpp
@@ -43,14 +43,15 @@ class InclusionRewriter : public PPCallbacks {
   StringRef MainEOL; ///< The line ending marker to use.
   const llvm::MemoryBuffer *PredefinesBuffer; ///< The preprocessor predefines.
   bool ShowLineMarkers; ///< Show #line markers.
-  bool UseLineDirective; ///< Use of line directives or line markers.
+  bool UseLineDirectives; ///< Use of line directives or line markers.
   typedef std::map<unsigned, FileChange> FileChangeMap;
   FileChangeMap FileChanges; ///< Tracks which files were included where.
   /// Used transitively for building up the FileChanges mapping over the
   /// various \c PPCallbacks callbacks.
   FileChangeMap::iterator LastInsertedFileChange;
 public:
-  InclusionRewriter(Preprocessor &PP, raw_ostream &OS, bool ShowLineMarkers);
+  InclusionRewriter(Preprocessor &PP, raw_ostream &OS, bool ShowLineMarkers,
+                    bool UseLineDirectives);
   bool Process(FileID FileId, SrcMgr::CharacteristicKind FileType);
   void setPredefinesBuffer(const llvm::MemoryBuffer *Buf) {
     PredefinesBuffer = Buf;
@@ -60,7 +61,7 @@ private:
   void FileChanged(SourceLocation Loc, FileChangeReason Reason,
                    SrcMgr::CharacteristicKind FileType,
                    FileID PrevFID) override;
-  void FileSkipped(const FileEntry &ParentFile, const Token &FilenameTok,
+  void FileSkipped(const FileEntry &SkippedFile, const Token &FilenameTok,
                    SrcMgr::CharacteristicKind FileType) override;
   void InclusionDirective(SourceLocation HashLoc, const Token &IncludeTok,
                           StringRef FileName, bool IsAngled,
@@ -89,13 +90,12 @@ private:
 
 /// Initializes an InclusionRewriter with a \p PP source and \p OS destination.
 InclusionRewriter::InclusionRewriter(Preprocessor &PP, raw_ostream &OS,
-                                     bool ShowLineMarkers)
+                                     bool ShowLineMarkers,
+                                     bool UseLineDirectives)
     : PP(PP), SM(PP.getSourceManager()), OS(OS), MainEOL("\n"),
       PredefinesBuffer(nullptr), ShowLineMarkers(ShowLineMarkers),
-      LastInsertedFileChange(FileChanges.end()) {
-  // If we're in microsoft mode, use normal #line instead of line markers.
-  UseLineDirective = PP.getLangOpts().MicrosoftExt;
-}
+      UseLineDirectives(UseLineDirectives),
+      LastInsertedFileChange(FileChanges.end()) {}
 
 /// Write appropriate line information as either #line directives or GNU line
 /// markers depending on what mode we're in, including the \p Filename and
@@ -106,7 +106,7 @@ void InclusionRewriter::WriteLineInfo(const char *Filename, int Line,
                                       StringRef Extra) {
   if (!ShowLineMarkers)
     return;
-  if (UseLineDirective) {
+  if (UseLineDirectives) {
     OS << "#line" << ' ' << Line << ' ' << '"';
     OS.write_escaped(Filename);
     OS << '"';
@@ -153,7 +153,7 @@ void InclusionRewriter::FileChanged(SourceLocation Loc,
 
 /// Called whenever an inclusion is skipped due to canonical header protection
 /// macros.
-void InclusionRewriter::FileSkipped(const FileEntry &/*ParentFile*/,
+void InclusionRewriter::FileSkipped(const FileEntry &/*SkippedFile*/,
                                     const Token &/*FilenameTok*/,
                                     SrcMgr::CharacteristicKind /*FileType*/) {
   assert(LastInsertedFileChange != FileChanges.end() && "A file, that wasn't "
@@ -561,8 +561,8 @@ bool InclusionRewriter::Process(FileID FileId,
 void clang::RewriteIncludesInInput(Preprocessor &PP, raw_ostream *OS,
                                    const PreprocessorOutputOptions &Opts) {
   SourceManager &SM = PP.getSourceManager();
-  InclusionRewriter *Rewrite = new InclusionRewriter(PP, *OS,
-                                                     Opts.ShowLineMarkers);
+  InclusionRewriter *Rewrite = new InclusionRewriter(
+      PP, *OS, Opts.ShowLineMarkers, Opts.UseLineDirectives);
   Rewrite->detectMainFileEOL();
 
   PP.addPPCallbacks(std::unique_ptr<PPCallbacks>(Rewrite));
diff --git a/lib/Frontend/Rewrite/RewriteModernObjC.cpp b/lib/Frontend/Rewrite/RewriteModernObjC.cpp
index 47f8189f2313..e13cdb3a3e55 100644
--- a/lib/Frontend/Rewrite/RewriteModernObjC.cpp
+++ b/lib/Frontend/Rewrite/RewriteModernObjC.cpp
@@ -243,8 +243,8 @@ namespace {
     RewriteModernObjC(std::string inFile, raw_ostream *OS,
                 DiagnosticsEngine &D, const LangOptions &LOpts,
                 bool silenceMacroWarn, bool LineInfo);
-    
-    ~RewriteModernObjC() {}
+
+    ~RewriteModernObjC() override {}
 
     void HandleTranslationUnit(ASTContext &C) override;
 
@@ -889,9 +889,9 @@ RewriteModernObjC::getIvarAccessString(ObjCIvarDecl *D) {
                                         IvarT, nullptr,
                                         /*BitWidth=*/nullptr, /*Mutable=*/true,
                                         ICIS_NoInit);
-      MemberExpr *ME = new (Context) MemberExpr(PE, true, FD, SourceLocation(),
-                                                FD->getType(), VK_LValue,
-                                                OK_Ordinary);
+      MemberExpr *ME = new (Context)
+          MemberExpr(PE, true, SourceLocation(), FD, SourceLocation(),
+                     FD->getType(), VK_LValue, OK_Ordinary);
       IvarT = Context->getDecltypeType(ME, ME->getType());
     }
   }
@@ -2689,7 +2689,7 @@ Stmt *RewriteModernObjC::RewriteObjCBoxedExpr(ObjCBoxedExpr *Exp) {
   MsgExprs.push_back(subExpr);
   
   SmallVector<QualType, 4> ArgTypes;
-  ArgTypes.push_back(Context->getObjCIdType());
+  ArgTypes.push_back(Context->getObjCClassType());
   ArgTypes.push_back(Context->getObjCSelType());
   for (const auto PI : BoxingMethod->parameters())
     ArgTypes.push_back(PI->getType());
@@ -2767,11 +2767,9 @@ Stmt *RewriteModernObjC::RewriteObjCArrayLiteralExpr(ObjCArrayLiteral *Exp) {
                                     Context->getPointerType(Context->VoidPtrTy),
                                     nullptr, /*BitWidth=*/nullptr,
                                     /*Mutable=*/true, ICIS_NoInit);
-  MemberExpr *ArrayLiteralME = 
-    new (Context) MemberExpr(NSArrayCallExpr, false, ARRFD, 
-                             SourceLocation(),
-                             ARRFD->getType(), VK_LValue,
-                             OK_Ordinary);
+  MemberExpr *ArrayLiteralME = new (Context)
+      MemberExpr(NSArrayCallExpr, false, SourceLocation(), ARRFD,
+                 SourceLocation(), ARRFD->getType(), VK_LValue, OK_Ordinary);
   QualType ConstIdT = Context->getObjCIdType().withConst();
   CStyleCastExpr * ArrayLiteralObjects = 
     NoTypeInfoCStyleCastExpr(Context, 
@@ -2818,7 +2816,7 @@ Stmt *RewriteModernObjC::RewriteObjCArrayLiteralExpr(ObjCArrayLiteral *Exp) {
   
   
   SmallVector<QualType, 4> ArgTypes;
-  ArgTypes.push_back(Context->getObjCIdType());
+  ArgTypes.push_back(Context->getObjCClassType());
   ArgTypes.push_back(Context->getObjCSelType());
   for (const auto *PI : ArrayMethod->params())
     ArgTypes.push_back(PI->getType());
@@ -2904,11 +2902,9 @@ Stmt *RewriteModernObjC::RewriteObjCDictionaryLiteralExpr(ObjCDictionaryLiteral
                                        Context->getPointerType(Context->VoidPtrTy),
                                        nullptr, /*BitWidth=*/nullptr,
                                        /*Mutable=*/true, ICIS_NoInit);
-  MemberExpr *DictLiteralValueME = 
-    new (Context) MemberExpr(NSValueCallExpr, false, ARRFD, 
-                             SourceLocation(),
-                             ARRFD->getType(), VK_LValue,
-                             OK_Ordinary);
+  MemberExpr *DictLiteralValueME = new (Context)
+      MemberExpr(NSValueCallExpr, false, SourceLocation(), ARRFD,
+                 SourceLocation(), ARRFD->getType(), VK_LValue, OK_Ordinary);
   QualType ConstIdT = Context->getObjCIdType().withConst();
   CStyleCastExpr * DictValueObjects = 
     NoTypeInfoCStyleCastExpr(Context, 
@@ -2919,13 +2915,11 @@ Stmt *RewriteModernObjC::RewriteObjCDictionaryLiteralExpr(ObjCDictionaryLiteral
   Expr *NSKeyCallExpr = 
     new (Context) CallExpr(*Context, NSDictDRE, KeyExprs,
                            NSDictFType, VK_LValue, SourceLocation());
-  
-  MemberExpr *DictLiteralKeyME = 
-    new (Context) MemberExpr(NSKeyCallExpr, false, ARRFD, 
-                             SourceLocation(),
-                             ARRFD->getType(), VK_LValue,
-                             OK_Ordinary);
-  
+
+  MemberExpr *DictLiteralKeyME = new (Context)
+      MemberExpr(NSKeyCallExpr, false, SourceLocation(), ARRFD,
+                 SourceLocation(), ARRFD->getType(), VK_LValue, OK_Ordinary);
+
   CStyleCastExpr * DictKeyObjects = 
     NoTypeInfoCStyleCastExpr(Context, 
                              Context->getPointerType(ConstIdT),
@@ -2975,7 +2969,7 @@ Stmt *RewriteModernObjC::RewriteObjCDictionaryLiteralExpr(ObjCDictionaryLiteral
   
   
   SmallVector<QualType, 8> ArgTypes;
-  ArgTypes.push_back(Context->getObjCIdType());
+  ArgTypes.push_back(Context->getObjCClassType());
   ArgTypes.push_back(Context->getObjCSelType());
   for (const auto *PI : DictMethod->params()) {
     QualType T = PI->getType();
@@ -3234,9 +3228,9 @@ Expr *RewriteModernObjC::SynthMsgSendStretCallExpr(FunctionDecl *MsgSendStretFla
                                     returnType, nullptr,
                                     /*BitWidth=*/nullptr,
                                     /*Mutable=*/true, ICIS_NoInit);
-  MemberExpr *ME = new (Context) MemberExpr(STCE, false, FieldD, SourceLocation(),
-                                            FieldD->getType(), VK_LValue,
-                                            OK_Ordinary);
+  MemberExpr *ME = new (Context)
+      MemberExpr(STCE, false, SourceLocation(), FieldD, SourceLocation(),
+                 FieldD->getType(), VK_LValue, OK_Ordinary);
 
   return ME;
 }
@@ -4732,11 +4726,10 @@ Stmt *RewriteModernObjC::SynthesizeBlockCall(CallExpr *Exp, const Expr *BlockExp
                                     Context->VoidPtrTy, nullptr,
                                     /*BitWidth=*/nullptr, /*Mutable=*/true,
                                     ICIS_NoInit);
-  MemberExpr *ME = new (Context) MemberExpr(PE, true, FD, SourceLocation(),
-                                            FD->getType(), VK_LValue,
-                                            OK_Ordinary);
+  MemberExpr *ME =
+      new (Context) MemberExpr(PE, true, SourceLocation(), FD, SourceLocation(),
+                               FD->getType(), VK_LValue, OK_Ordinary);
 
-  
   CastExpr *FunkCast = NoTypeInfoCStyleCastExpr(Context, PtrToFuncCastType,
                                                 CK_BitCast, ME);
   PE = new (Context) ParenExpr(SourceLocation(), SourceLocation(), FunkCast);
@@ -4781,10 +4774,9 @@ Stmt *RewriteModernObjC::RewriteBlockDeclRefExpr(DeclRefExpr *DeclRefExp) {
                                     Context->VoidPtrTy, nullptr,
                                     /*BitWidth=*/nullptr, /*Mutable=*/true,
                                     ICIS_NoInit);
-  MemberExpr *ME = new (Context) MemberExpr(DeclRefExp, isArrow,
-                                            FD, SourceLocation(),
-                                            FD->getType(), VK_LValue,
-                                            OK_Ordinary);
+  MemberExpr *ME = new (Context)
+      MemberExpr(DeclRefExp, isArrow, SourceLocation(), FD, SourceLocation(),
+                 FD->getType(), VK_LValue, OK_Ordinary);
 
   StringRef Name = VD->getName();
   FD = FieldDecl::Create(*Context, nullptr, SourceLocation(), SourceLocation(),
@@ -4792,11 +4784,10 @@ Stmt *RewriteModernObjC::RewriteBlockDeclRefExpr(DeclRefExpr *DeclRefExp) {
                          Context->VoidPtrTy, nullptr,
                          /*BitWidth=*/nullptr, /*Mutable=*/true,
                          ICIS_NoInit);
-  ME = new (Context) MemberExpr(ME, true, FD, SourceLocation(),
-                                DeclRefExp->getType(), VK_LValue, OK_Ordinary);
-  
-  
-  
+  ME =
+      new (Context) MemberExpr(ME, true, SourceLocation(), FD, SourceLocation(),
+                               DeclRefExp->getType(), VK_LValue, OK_Ordinary);
+
   // Need parens to enforce precedence.
   ParenExpr *PE = new (Context) ParenExpr(DeclRefExp->getExprLoc(), 
                                           DeclRefExp->getExprLoc(), 
@@ -7694,9 +7685,9 @@ Stmt *RewriteModernObjC::RewriteObjCIvarRefExpr(ObjCIvarRefExpr *IV) {
                                             IvarT, nullptr,
                                             /*BitWidth=*/nullptr,
                                             /*Mutable=*/true, ICIS_NoInit);
-          MemberExpr *ME = new (Context) MemberExpr(PE, true, FD, SourceLocation(),
-                                                    FD->getType(), VK_LValue,
-                                                    OK_Ordinary);
+          MemberExpr *ME = new (Context)
+              MemberExpr(PE, true, SourceLocation(), FD, SourceLocation(),
+                         FD->getType(), VK_LValue, OK_Ordinary);
           IvarT = Context->getDecltypeType(ME, ME->getType());
         }
       }
@@ -7723,9 +7714,9 @@ Stmt *RewriteModernObjC::RewriteObjCIvarRefExpr(ObjCIvarRefExpr *IV) {
                                           D->getType(), nullptr,
                                           /*BitWidth=*/D->getBitWidth(),
                                           /*Mutable=*/true, ICIS_NoInit);
-        MemberExpr *ME = new (Context) MemberExpr(PE, /*isArrow*/false, FD, SourceLocation(),
-                                                  FD->getType(), VK_LValue,
-                                                  OK_Ordinary);
+        MemberExpr *ME = new (Context)
+            MemberExpr(PE, /*isArrow*/ false, SourceLocation(), FD,
+                       SourceLocation(), FD->getType(), VK_LValue, OK_Ordinary);
         Replacement = ME;
 
       }
diff --git a/lib/Frontend/Rewrite/RewriteObjC.cpp b/lib/Frontend/Rewrite/RewriteObjC.cpp
index 519681085f56..b2a45b407bae 100644
--- a/lib/Frontend/Rewrite/RewriteObjC.cpp
+++ b/lib/Frontend/Rewrite/RewriteObjC.cpp
@@ -193,7 +193,7 @@ namespace {
                 DiagnosticsEngine &D, const LangOptions &LOpts,
                 bool silenceMacroWarn);
 
-    ~RewriteObjC() {}
+    ~RewriteObjC() override {}
 
     void HandleTranslationUnit(ASTContext &C) override;
 
@@ -511,8 +511,8 @@ namespace {
                 bool silenceMacroWarn) : RewriteObjC(inFile, OS,
                                                      D, LOpts,
                                                      silenceMacroWarn) {}
-    
-    ~RewriteObjCFragileABI() {}
+
+    ~RewriteObjCFragileABI() override {}
     void Initialize(ASTContext &context) override;
 
     // Rewriting metadata
@@ -3821,11 +3821,10 @@ Stmt *RewriteObjC::SynthesizeBlockCall(CallExpr *Exp, const Expr *BlockExp) {
                                     Context->VoidPtrTy, nullptr,
                                     /*BitWidth=*/nullptr, /*Mutable=*/true,
                                     ICIS_NoInit);
-  MemberExpr *ME = new (Context) MemberExpr(PE, true, FD, SourceLocation(),
-                                            FD->getType(), VK_LValue,
-                                            OK_Ordinary);
+  MemberExpr *ME =
+      new (Context) MemberExpr(PE, true, SourceLocation(), FD, SourceLocation(),
+                               FD->getType(), VK_LValue, OK_Ordinary);
 
-  
   CastExpr *FunkCast = NoTypeInfoCStyleCastExpr(Context, PtrToFuncCastType,
                                                 CK_BitCast, ME);
   PE = new (Context) ParenExpr(SourceLocation(), SourceLocation(), FunkCast);
@@ -3870,10 +3869,9 @@ Stmt *RewriteObjC::RewriteBlockDeclRefExpr(DeclRefExpr *DeclRefExp) {
                                     Context->VoidPtrTy, nullptr,
                                     /*BitWidth=*/nullptr, /*Mutable=*/true,
                                     ICIS_NoInit);
-  MemberExpr *ME = new (Context) MemberExpr(DeclRefExp, isArrow,
-                                            FD, SourceLocation(),
-                                            FD->getType(), VK_LValue,
-                                            OK_Ordinary);
+  MemberExpr *ME = new (Context)
+      MemberExpr(DeclRefExp, isArrow, SourceLocation(), FD, SourceLocation(),
+                 FD->getType(), VK_LValue, OK_Ordinary);
 
   StringRef Name = VD->getName();
   FD = FieldDecl::Create(*Context, nullptr, SourceLocation(), SourceLocation(),
@@ -3881,11 +3879,10 @@ Stmt *RewriteObjC::RewriteBlockDeclRefExpr(DeclRefExpr *DeclRefExp) {
                          Context->VoidPtrTy, nullptr,
                          /*BitWidth=*/nullptr, /*Mutable=*/true,
                          ICIS_NoInit);
-  ME = new (Context) MemberExpr(ME, true, FD, SourceLocation(),
-                                DeclRefExp->getType(), VK_LValue, OK_Ordinary);
-  
-  
-  
+  ME =
+      new (Context) MemberExpr(ME, true, SourceLocation(), FD, SourceLocation(),
+                               DeclRefExp->getType(), VK_LValue, OK_Ordinary);
+
   // Need parens to enforce precedence.
   ParenExpr *PE = new (Context) ParenExpr(DeclRefExp->getExprLoc(), 
                                           DeclRefExp->getExprLoc(), 
@@ -5880,10 +5877,9 @@ Stmt *RewriteObjCFragileABI::RewriteObjCIvarRefExpr(ObjCIvarRefExpr *IV) {
                                               castExpr);
       if (IV->isFreeIvar() &&
           declaresSameEntity(CurMethodDef->getClassInterface(), iFaceDecl->getDecl())) {
-        MemberExpr *ME = new (Context) MemberExpr(PE, true, D,
-                                                  IV->getLocation(),
-                                                  D->getType(),
-                                                  VK_LValue, OK_Ordinary);
+        MemberExpr *ME = new (Context)
+            MemberExpr(PE, true, SourceLocation(), D, IV->getLocation(),
+                       D->getType(), VK_LValue, OK_Ordinary);
         Replacement = ME;
       } else {
         IV->setBase(PE);
diff --git a/lib/Frontend/SerializedDiagnosticPrinter.cpp b/lib/Frontend/SerializedDiagnosticPrinter.cpp
index f701f723a320..d31b12e87a4e 100644
--- a/lib/Frontend/SerializedDiagnosticPrinter.cpp
+++ b/lib/Frontend/SerializedDiagnosticPrinter.cpp
@@ -20,6 +20,7 @@
 #include "clang/Frontend/TextDiagnosticPrinter.h"
 #include "clang/Lex/Lexer.h"
 #include "llvm/ADT/DenseSet.h"
+#include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/StringRef.h"
 #include "llvm/Support/raw_ostream.h"
@@ -60,8 +61,8 @@ public:
                  DiagnosticOptions *DiagOpts)
     : DiagnosticNoteRenderer(LangOpts, DiagOpts), Writer(Writer) {}
 
-  virtual ~SDiagsRenderer() {}
-  
+  ~SDiagsRenderer() override {}
+
 protected:
   void emitDiagnosticMessage(SourceLocation Loc,
                              PresumedLoc PLoc,
@@ -157,7 +158,7 @@ public:
     EmitPreamble();
   }
 
-  ~SDiagsWriter() {}
+  ~SDiagsWriter() override {}
 
   void HandleDiagnostic(DiagnosticsEngine::Level DiagLevel,
                         const Diagnostic &Info) override;
@@ -631,7 +632,7 @@ void SDiagsWriter::HandleDiagnostic(DiagnosticsEngine::Level DiagLevel,
          "Unexpected diagnostic with valid location outside of a source file");
   SDiagsRenderer Renderer(*this, *LangOpts, &*State->DiagOpts);
   Renderer.emitDiagnostic(Info.getLocation(), DiagLevel,
-                          State->diagBuf.str(),
+                          State->diagBuf,
                           Info.getRanges(),
                           Info.getFixItHints(),
                           &Info.getSourceManager(),
diff --git a/lib/Frontend/TextDiagnostic.cpp b/lib/Frontend/TextDiagnostic.cpp
index bbc99141f072..aaf17a983371 100644
--- a/lib/Frontend/TextDiagnostic.cpp
+++ b/lib/Frontend/TextDiagnostic.cpp
@@ -799,18 +799,18 @@ void TextDiagnostic::emitDiagnosticLoc(SourceLocation Loc, PresumedLoc PLoc,
   OS << PLoc.getFilename();
   switch (DiagOpts->getFormat()) {
   case DiagnosticOptions::Clang: OS << ':'  << LineNo; break;
-  case DiagnosticOptions::Msvc:  OS << '('  << LineNo; break;
+  case DiagnosticOptions::MSVC:  OS << '('  << LineNo; break;
   case DiagnosticOptions::Vi:    OS << " +" << LineNo; break;
   }
 
   if (DiagOpts->ShowColumn)
     // Compute the column number.
     if (unsigned ColNo = PLoc.getColumn()) {
-      if (DiagOpts->getFormat() == DiagnosticOptions::Msvc) {
+      if (DiagOpts->getFormat() == DiagnosticOptions::MSVC) {
         OS << ',';
         // Visual Studio 2010 or earlier expects column number to be off by one
         if (LangOpts.MSCompatibilityVersion &&
-            LangOpts.MSCompatibilityVersion < 170000000)
+            !LangOpts.isCompatibleWithMSVC(LangOptions::MSVC2012))
           ColNo--;
       } else
         OS << ':';
@@ -819,7 +819,7 @@ void TextDiagnostic::emitDiagnosticLoc(SourceLocation Loc, PresumedLoc PLoc,
   switch (DiagOpts->getFormat()) {
   case DiagnosticOptions::Clang:
   case DiagnosticOptions::Vi:    OS << ':';    break;
-  case DiagnosticOptions::Msvc:  OS << ") : "; break;
+  case DiagnosticOptions::MSVC:  OS << ") : "; break;
   }
 
   if (DiagOpts->ShowSourceRanges && !Ranges.empty()) {
diff --git a/lib/Frontend/VerifyDiagnosticConsumer.cpp b/lib/Frontend/VerifyDiagnosticConsumer.cpp
index 3ff6b18e2199..910e394553de 100644
--- a/lib/Frontend/VerifyDiagnosticConsumer.cpp
+++ b/lib/Frontend/VerifyDiagnosticConsumer.cpp
@@ -58,9 +58,9 @@ public:
 
   /// \brief Hook into the preprocessor and update the list of parsed
   /// files when the preprocessor indicates a new file is entered.
-  virtual void FileChanged(SourceLocation Loc, FileChangeReason Reason,
-                           SrcMgr::CharacteristicKind FileType,
-                           FileID PrevFID) {
+  void FileChanged(SourceLocation Loc, FileChangeReason Reason,
+                   SrcMgr::CharacteristicKind FileType,
+                   FileID PrevFID) override {
     Verify.UpdateParsedFileStatus(SM, SM.getFileID(Loc),
                                   VerifyDiagnosticConsumer::IsParsed);
   }
diff --git a/lib/Headers/CMakeLists.txt b/lib/Headers/CMakeLists.txt
index 080550f7c77f..29a738e7a81a 100644
--- a/lib/Headers/CMakeLists.txt
+++ b/lib/Headers/CMakeLists.txt
@@ -9,15 +9,20 @@ set(files
   avx512fintrin.h
   avx512vlbwintrin.h
   avx512vlintrin.h
+  avx512dqintrin.h
+  avx512vldqintrin.h
   avxintrin.h
   bmi2intrin.h
   bmiintrin.h
   cpuid.h
+  cuda_builtin_vars.h
   emmintrin.h
   f16cintrin.h
   float.h
   fma4intrin.h
   fmaintrin.h
+  htmintrin.h
+  htmxlintrin.h
   ia32intrin.h
   immintrin.h
   Intrin.h
@@ -34,6 +39,7 @@ set(files
   prfchwintrin.h
   rdseedintrin.h
   rtmintrin.h
+  s390intrin.h
   shaintrin.h
   smmintrin.h
   stdalign.h
diff --git a/lib/Headers/Intrin.h b/lib/Headers/Intrin.h
index 84bc4303a133..727a55e5b761 100644
--- a/lib/Headers/Intrin.h
+++ b/lib/Headers/Intrin.h
@@ -289,6 +289,7 @@ void _WriteBarrier(void);
 unsigned __int32 xbegin(void);
 void _xend(void);
 static __inline__
+#define _XCR_XFEATURE_ENABLED_MASK 0
 unsigned __int64 __cdecl _xgetbv(unsigned int);
 void __cdecl _xrstor(void const *, unsigned __int64);
 void __cdecl _xsave(void *, unsigned __int64);
@@ -780,17 +781,17 @@ _InterlockedCompareExchange64(__int64 volatile *_Destination,
 \*----------------------------------------------------------------------------*/
 #if defined(__i386__) || defined(__x86_64__)
 static __inline__ void __attribute__((__always_inline__, __nodebug__))
-__attribute__((deprecated("use other intrinsics or C++11 atomics instead")))
+__attribute__((__deprecated__("use other intrinsics or C++11 atomics instead")))
 _ReadWriteBarrier(void) {
   __asm__ volatile ("" : : : "memory");
 }
 static __inline__ void __attribute__((__always_inline__, __nodebug__))
-__attribute__((deprecated("use other intrinsics or C++11 atomics instead")))
+__attribute__((__deprecated__("use other intrinsics or C++11 atomics instead")))
 _ReadBarrier(void) {
   __asm__ volatile ("" : : : "memory");
 }
 static __inline__ void __attribute__((__always_inline__, __nodebug__))
-__attribute__((deprecated("use other intrinsics or C++11 atomics instead")))
+__attribute__((__deprecated__("use other intrinsics or C++11 atomics instead")))
 _WriteBarrier(void) {
   __asm__ volatile ("" : : : "memory");
 }
@@ -943,14 +944,14 @@ __readmsr(unsigned long __register) {
   return (((unsigned __int64)__edx) << 32) | (unsigned __int64)__eax;
 }
 
-static __inline__ unsigned long __attribute__((always_inline, __nodebug__))
+static __inline__ unsigned long __attribute__((__always_inline__, __nodebug__))
 __readcr3(void) {
   unsigned long __cr3_val;
   __asm__ __volatile__ ("mov %%cr3, %0" : "=q"(__cr3_val) : : "memory");
   return __cr3_val;
 }
 
-static __inline__ void __attribute__((always_inline, __nodebug__))
+static __inline__ void __attribute__((__always_inline__, __nodebug__))
 __writecr3(unsigned int __cr3_val) {
   __asm__ ("mov %0, %%cr3" : : "q"(__cr3_val) : "memory");
 }
diff --git a/lib/Headers/__stddef_max_align_t.h b/lib/Headers/__stddef_max_align_t.h
index a06f412c53fb..1e10ca9865c0 100644
--- a/lib/Headers/__stddef_max_align_t.h
+++ b/lib/Headers/__stddef_max_align_t.h
@@ -26,15 +26,18 @@
 #ifndef __CLANG_MAX_ALIGN_T_DEFINED
 #define __CLANG_MAX_ALIGN_T_DEFINED
 
-#ifndef _MSC_VER
+#if defined(_MSC_VER)
+typedef double max_align_t;
+#elif defined(__APPLE__)
+typedef long double max_align_t;
+#else
+// Define 'max_align_t' to match the GCC definition.
 typedef struct {
   long long __clang_max_align_nonce1
       __attribute__((__aligned__(__alignof__(long long))));
   long double __clang_max_align_nonce2
       __attribute__((__aligned__(__alignof__(long double))));
 } max_align_t;
-#else
-typedef double max_align_t;
 #endif
 
 #endif
diff --git a/lib/Headers/altivec.h b/lib/Headers/altivec.h
index 0ac0841ae482..1f8c831bb739 100644
--- a/lib/Headers/altivec.h
+++ b/lib/Headers/altivec.h
@@ -73,6 +73,18 @@ vec_perm(vector bool int __a, vector bool int __b, vector unsigned char __c);
 static vector float __ATTRS_o_ai
 vec_perm(vector float __a, vector float __b, vector unsigned char __c);
 
+#ifdef __VSX__
+static vector long long __ATTRS_o_ai
+vec_perm(vector long long __a, vector long long __b, vector unsigned char __c);
+
+static vector unsigned long long __ATTRS_o_ai
+vec_perm(vector unsigned long long __a, vector unsigned long long __b,
+         vector unsigned char __c);
+
+static vector double __ATTRS_o_ai
+vec_perm(vector double __a, vector double __b, vector unsigned char __c);
+#endif
+
 static vector unsigned char __ATTRS_o_ai
 vec_xor(vector unsigned char __a, vector unsigned char __b);
 
@@ -245,6 +257,20 @@ vec_add(vector unsigned int __a, vector bool int __b)
   return __a + (vector unsigned int)__b;
 }
 
+#if defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+static vector signed __int128 __ATTRS_o_ai
+vec_add(vector signed __int128 __a, vector signed __int128 __b)
+{
+  return __a + __b;
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_add(vector unsigned __int128 __a, vector unsigned __int128 __b)
+{
+  return __a + __b;
+}
+#endif // defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+
 static vector float __ATTRS_o_ai
 vec_add(vector float __a, vector float __b)
 {
@@ -383,12 +409,24 @@ vec_vaddfp(vector float __a, vector float __b)
 
 /* vec_addc */
 
-static vector unsigned int __attribute__((__always_inline__))
+static vector unsigned int __ATTRS_o_ai
 vec_addc(vector unsigned int __a, vector unsigned int __b)
 {
   return __builtin_altivec_vaddcuw(__a, __b);
 }
 
+#if defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+static vector signed __int128 __ATTRS_o_ai
+vec_addc(vector signed __int128 __a, vector signed __int128 __b) {
+  return __builtin_altivec_vaddcuq(__a, __b);
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_addc(vector unsigned __int128 __a, vector unsigned __int128 __b) {
+  return __builtin_altivec_vaddcuq(__a, __b);
+}
+#endif // defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+
 /* vec_vaddcuw */
 
 static vector unsigned int __attribute__((__always_inline__))
@@ -627,6 +665,64 @@ vec_vadduws(vector unsigned int __a, vector bool int __b)
   return __builtin_altivec_vadduws(__a, (vector unsigned int)__b);
 }
 
+#if defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+/* vec_vadduqm */
+
+static vector signed __int128 __ATTRS_o_ai
+vec_vadduqm(vector signed __int128 __a, vector signed __int128 __b)
+{
+  return __a + __b;
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_vadduqm(vector unsigned __int128 __a, vector unsigned __int128 __b)
+{
+  return __a + __b;
+}
+
+/* vec_vaddeuqm */
+
+static vector signed __int128 __ATTRS_o_ai
+vec_vaddeuqm(vector signed __int128 __a, vector signed __int128 __b,
+             vector signed __int128 __c) {
+  return __builtin_altivec_vaddeuqm(__a, __b, __c);
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_vaddeuqm(vector unsigned __int128 __a, vector unsigned __int128 __b,
+             vector unsigned __int128 __c) {
+  return __builtin_altivec_vaddeuqm(__a, __b, __c);
+}
+
+/* vec_vaddcuq */
+
+static vector signed __int128 __ATTRS_o_ai
+vec_vaddcuq(vector signed __int128 __a, vector signed __int128 __b)
+{
+  return __builtin_altivec_vaddcuq(__a, __b);
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_vaddcuq(vector unsigned __int128 __a, vector unsigned __int128 __b)
+{
+  return __builtin_altivec_vaddcuq(__a, __b);
+}
+
+/* vec_vaddecuq */
+
+static vector signed __int128 __ATTRS_o_ai
+vec_vaddecuq(vector signed __int128 __a, vector signed __int128 __b,
+             vector signed __int128 __c) {
+  return __builtin_altivec_vaddecuq(__a, __b, __c);
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_vaddecuq(vector unsigned __int128 __a, vector unsigned __int128 __b,
+             vector unsigned __int128 __c) {
+  return __builtin_altivec_vaddecuq(__a, __b, __c);
+}
+#endif // defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+
 /* vec_and */
 
 #define __builtin_altivec_vand vec_and
@@ -1387,6 +1483,21 @@ vec_cmpeq(vector unsigned int __a, vector unsigned int __b)
     __builtin_altivec_vcmpequw((vector int)__a, (vector int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static vector bool long long __ATTRS_o_ai
+vec_cmpeq(vector signed long long __a, vector signed long long __b) 
+{
+  return (vector bool long long) __builtin_altivec_vcmpequd(__a, __b);
+}
+
+static vector bool long long __ATTRS_o_ai
+vec_cmpeq(vector unsigned long long __a, vector unsigned long long __b) 
+{
+  return (vector bool long long) 
+    __builtin_altivec_vcmpequd((vector long long)__a, (vector long long) __b);
+}
+#endif
+
 static vector bool int __ATTRS_o_ai
 vec_cmpeq(vector float __a, vector float __b)
 {
@@ -1447,6 +1558,20 @@ vec_cmpgt(vector unsigned int __a, vector unsigned int __b)
   return (vector bool int)__builtin_altivec_vcmpgtuw(__a, __b);
 }
 
+#ifdef __POWER8_VECTOR__
+static vector bool long long __ATTRS_o_ai
+vec_cmpgt(vector signed long long __a, vector signed long long __b)
+{
+  return (vector bool long long)__builtin_altivec_vcmpgtsd(__a, __b);
+}
+
+static vector bool long long __ATTRS_o_ai
+vec_cmpgt(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return (vector bool long long)__builtin_altivec_vcmpgtud(__a, __b);
+}
+#endif
+
 static vector bool int __ATTRS_o_ai
 vec_cmpgt(vector float __a, vector float __b)
 {
@@ -2270,7 +2395,7 @@ vec_vlogefp(vector float __a)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsl(int __a, const signed char *__b)
 {
@@ -2289,7 +2414,7 @@ vec_lvsl(int __a, const signed char *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsl(int __a, const unsigned char *__b)
 {
@@ -2308,7 +2433,7 @@ vec_lvsl(int __a, const unsigned char *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsl(int __a, const short *__b)
 {
@@ -2327,7 +2452,7 @@ vec_lvsl(int __a, const short *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsl(int __a, const unsigned short *__b)
 {
@@ -2346,7 +2471,7 @@ vec_lvsl(int __a, const unsigned short *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsl(int __a, const int *__b)
 {
@@ -2365,7 +2490,7 @@ vec_lvsl(int __a, const int *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsl(int __a, const unsigned int *__b)
 {
@@ -2384,7 +2509,7 @@ vec_lvsl(int __a, const unsigned int *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsl(int __a, const float *__b)
 {
@@ -2405,7 +2530,7 @@ vec_lvsl(int __a, const float *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsr(int __a, const signed char *__b)
 {
@@ -2424,7 +2549,7 @@ vec_lvsr(int __a, const signed char *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsr(int __a, const unsigned char *__b)
 {
@@ -2443,7 +2568,7 @@ vec_lvsr(int __a, const unsigned char *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsr(int __a, const short *__b)
 {
@@ -2462,7 +2587,7 @@ vec_lvsr(int __a, const short *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsr(int __a, const unsigned short *__b)
 {
@@ -2481,7 +2606,7 @@ vec_lvsr(int __a, const unsigned short *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsr(int __a, const int *__b)
 {
@@ -2500,7 +2625,7 @@ vec_lvsr(int __a, const int *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsr(int __a, const unsigned int *__b)
 {
@@ -2519,7 +2644,7 @@ vec_lvsr(int __a, const unsigned int *__b)
 
 #ifdef __LITTLE_ENDIAN__
 static vector unsigned char __ATTRS_o_ai
-__attribute__((deprecated("use assignment for unaligned little endian \
+__attribute__((__deprecated__("use assignment for unaligned little endian \
 loads/stores")))
 vec_lvsr(int __a, const float *__b)
 {
@@ -2679,6 +2804,20 @@ vec_max(vector unsigned int __a, vector bool int __b)
   return __builtin_altivec_vmaxuw(__a, (vector unsigned int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static vector signed long long __ATTRS_o_ai
+vec_max(vector signed long long __a, vector signed long long __b) 
+{
+  return __builtin_altivec_vmaxsd(__a, __b);
+}
+
+static vector unsigned long long __ATTRS_o_ai
+vec_max(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vmaxud(__a, __b);
+}
+#endif
+
 static vector float __ATTRS_o_ai
 vec_max(vector float __a, vector float __b)
 {
@@ -3327,6 +3466,20 @@ vec_min(vector unsigned int __a, vector bool int __b)
   return __builtin_altivec_vminuw(__a, (vector unsigned int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static vector signed long long __ATTRS_o_ai
+vec_min(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vminsd(__a, __b);
+}
+
+static vector unsigned long long __ATTRS_o_ai
+vec_min(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vminud(__a, __b);
+}
+#endif
+
 static vector float __ATTRS_o_ai
 vec_min(vector float __a, vector float __b)
 {
@@ -3762,6 +3915,28 @@ vec_mule(vector unsigned short __a, vector unsigned short __b)
 #endif
 }
 
+#ifdef __POWER8_VECTOR__
+static vector signed long long __ATTRS_o_ai
+vec_mule(vector signed int __a, vector signed int __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vmulosw(__a, __b);
+#else
+  return __builtin_altivec_vmulesw(__a, __b);
+#endif
+}
+
+static vector unsigned long long __ATTRS_o_ai
+vec_mule(vector unsigned int __a, vector unsigned int __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vmulouw(__a, __b);
+#else
+  return __builtin_altivec_vmuleuw(__a, __b);
+#endif
+}
+#endif
+
 /* vec_vmulesb */
 
 static vector short __attribute__((__always_inline__))
@@ -3852,6 +4027,28 @@ vec_mulo(vector unsigned short __a, vector unsigned short __b)
 #endif
 }
 
+#ifdef __POWER8_VECTOR__
+static vector signed long long __ATTRS_o_ai
+vec_mulo(vector signed int __a, vector signed int __b) 
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vmulesw(__a, __b);
+#else
+  return __builtin_altivec_vmulosw(__a, __b);
+#endif
+}
+
+static vector unsigned long long __ATTRS_o_ai
+vec_mulo(vector unsigned int __a, vector unsigned int __b) 
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vmuleuw(__a, __b);
+#else
+  return __builtin_altivec_vmulouw(__a, __b);
+#endif
+}
+#endif
+
 /* vec_vmulosb */
 
 static vector short __attribute__((__always_inline__))
@@ -4525,6 +4722,58 @@ vec_vpkuwum(vector bool int __a, vector bool int __b)
 #endif
 }
 
+/* vec_vpkudum */
+
+#ifdef __POWER8_VECTOR__
+#define __builtin_altivec_vpkudum vec_vpkudum
+
+static vector int __ATTRS_o_ai
+vec_vpkudum(vector long long __a, vector long long __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return (vector int)vec_perm(__a, __b, (vector unsigned char)
+    (0x00, 0x01, 0x02, 0x03, 0x08, 0x09, 0x0A, 0x0B,
+     0x10, 0x11, 0x12, 0x13, 0x18, 0x19, 0x1A, 0x1B));
+#else
+  return (vector int)vec_perm(__a, __b, (vector unsigned char)
+    (0x04, 0x05, 0x06, 0x07, 0x0C, 0x0D, 0x0E, 0x0F,
+     0x14, 0x15, 0x16, 0x17, 0x1C, 0x1D, 0x1E, 0x1F));
+#endif
+}
+
+static vector unsigned int __ATTRS_o_ai
+vec_vpkudum(vector unsigned long long __a, vector unsigned long long __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return (vector unsigned int)vec_perm(__a, __b, (vector unsigned char)
+    (0x00, 0x01, 0x02, 0x03, 0x08, 0x09, 0x0A, 0x0B,
+     0x10, 0x11, 0x12, 0x13, 0x18, 0x19, 0x1A, 0x1B));
+#else
+  return (vector unsigned int)vec_perm(__a, __b, (vector unsigned char)
+    (0x04, 0x05, 0x06, 0x07, 0x0C, 0x0D, 0x0E, 0x0F,
+     0x14, 0x15, 0x16, 0x17, 0x1C, 0x1D, 0x1E, 0x1F));
+#endif
+}
+
+static vector bool int __ATTRS_o_ai
+vec_vpkudum(vector bool long long __a, vector bool long long __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return (vector bool int)vec_perm((vector long long)__a,
+                                   (vector long long)__b,
+                                   (vector unsigned char)
+    (0x00, 0x01, 0x02, 0x03, 0x08, 0x09, 0x0A, 0x0B,
+     0x10, 0x11, 0x12, 0x13, 0x18, 0x19, 0x1A, 0x1B));
+#else
+  return (vector bool int)vec_perm((vector long long)__a,
+                                   (vector long long)__b,
+                                   (vector unsigned char)
+    (0x04, 0x05, 0x06, 0x07, 0x0C, 0x0D, 0x0E, 0x0F,
+     0x14, 0x15, 0x16, 0x17, 0x1C, 0x1D, 0x1E, 0x1F));
+#endif
+}
+#endif
+
 /* vec_packpx */
 
 static vector pixel __attribute__((__always_inline__))
@@ -4591,6 +4840,28 @@ vec_packs(vector unsigned int __a, vector unsigned int __b)
 #endif
 }
 
+#ifdef __POWER8_VECTOR__
+static vector int __ATTRS_o_ai
+vec_packs(vector long long __a, vector long long __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vpksdss(__b, __a);
+#else
+  return __builtin_altivec_vpksdss(__a, __b);
+#endif
+}
+
+static vector unsigned int __ATTRS_o_ai
+vec_packs(vector unsigned long long __a, vector unsigned long long __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vpkudus(__b, __a);
+#else
+  return __builtin_altivec_vpkudus(__a, __b);
+#endif
+}
+#endif
+
 /* vec_vpkshss */
 
 static vector signed char __attribute__((__always_inline__))
@@ -4603,6 +4874,20 @@ vec_vpkshss(vector short __a, vector short __b)
 #endif
 }
 
+/* vec_vpksdss */
+
+#ifdef __POWER8_VECTOR__
+static vector int __ATTRS_o_ai
+vec_vpksdss(vector long long __a, vector long long __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vpksdss(__b, __a);
+#else
+  return __builtin_altivec_vpksdss(__a, __b);
+#endif
+}
+#endif
+
 /* vec_vpkuhus */
 
 static vector unsigned char __attribute__((__always_inline__))
@@ -4615,6 +4900,20 @@ vec_vpkuhus(vector unsigned short __a, vector unsigned short __b)
 #endif
 }
 
+/* vec_vpkudus */
+
+#ifdef __POWER8_VECTOR__
+static vector unsigned int __attribute__((__always_inline__))
+vec_vpkudus(vector unsigned long long __a, vector unsigned long long __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vpkudus(__b, __a);
+#else
+  return __builtin_altivec_vpkudus(__a, __b);
+#endif
+}
+#endif
+
 /* vec_vpkswss */
 
 static vector signed short __attribute__((__always_inline__))
@@ -4681,6 +4980,28 @@ vec_packsu(vector unsigned int __a, vector unsigned int __b)
 #endif
 }
 
+#ifdef __POWER8_VECTOR__
+static vector unsigned int __ATTRS_o_ai
+vec_packsu(vector long long __a, vector long long __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vpksdus(__b, __a);
+#else
+  return __builtin_altivec_vpksdus(__a, __b);
+#endif
+}
+
+static vector unsigned int __ATTRS_o_ai
+vec_packsu(vector unsigned long long __a, vector unsigned long long __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vpkudus(__b, __a);
+#else
+  return __builtin_altivec_vpkudus(__a, __b);
+#endif
+}
+#endif
+
 /* vec_vpkshus */
 
 static vector unsigned char __ATTRS_o_ai
@@ -4725,6 +5046,20 @@ vec_vpkswus(vector unsigned int __a, vector unsigned int __b)
 #endif
 }
 
+/* vec_vpksdus */
+
+#ifdef __POWER8_VECTOR__
+static vector unsigned int __ATTRS_o_ai
+vec_vpksdus(vector long long __a, vector long long __b)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vpksdus(__b, __a);
+#else
+  return __builtin_altivec_vpksdus(__a, __b);
+#endif
+}
+#endif
+
 /* vec_perm */
 
 // The vperm instruction is defined architecturally with a big-endian bias.
@@ -5095,6 +5430,20 @@ vec_rl(vector unsigned int __a, vector unsigned int __b)
   return (vector unsigned int)__builtin_altivec_vrlw((vector int)__a, __b);
 }
 
+#ifdef __POWER8_VECTOR__
+static vector signed long long __ATTRS_o_ai
+vec_rl(vector signed long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vrld(__a, __b);
+}
+
+static vector unsigned long long __ATTRS_o_ai
+vec_rl(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vrld(__a, __b);
+}
+#endif
+
 /* vec_vrlb */
 
 static vector signed char __ATTRS_o_ai
@@ -5465,6 +5814,20 @@ vec_sl(vector unsigned int __a, vector unsigned int __b)
   return __a << __b;
 }
 
+#ifdef __POWER8_VECTOR__
+static vector signed long long __ATTRS_o_ai
+vec_sl(vector signed long long __a, vector unsigned long long __b)
+{
+  return __a << (vector long long)__b;
+}
+
+static vector unsigned long long __ATTRS_o_ai
+vec_sl(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __a << __b;
+}
+#endif
+
 /* vec_vslb */
 
 #define __builtin_altivec_vslb vec_vslb
@@ -6566,6 +6929,20 @@ vec_sr(vector unsigned int __a, vector unsigned int __b)
   return __a >> __b;
 }
 
+#ifdef __POWER8_VECTOR__
+static vector signed long long __ATTRS_o_ai
+vec_sr(vector signed long long __a, vector unsigned long long __b)
+{
+  return __a >> (vector long long)__b;
+}
+
+static vector unsigned long long __ATTRS_o_ai
+vec_sr(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __a >> __b;
+}
+#endif
+
 /* vec_vsrb */
 
 #define __builtin_altivec_vsrb vec_vsrb
@@ -6652,6 +7029,20 @@ vec_sra(vector unsigned int __a, vector unsigned int __b)
   return (vector unsigned int)__builtin_altivec_vsraw((vector int)__a, __b);
 }
 
+#ifdef __POWER8_VECTOR__
+static vector signed long long __ATTRS_o_ai
+vec_sra(vector signed long long __a, vector unsigned long long __b)
+{
+  return __a >> __b;
+}
+
+static vector unsigned long long __ATTRS_o_ai
+vec_sra(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return (vector unsigned long long) ( (vector signed long long) __a >> __b);
+}
+#endif
+
 /* vec_vsrab */
 
 static vector signed char __ATTRS_o_ai
@@ -8224,6 +8615,20 @@ vec_sub(vector unsigned int __a, vector bool int __b)
   return __a - (vector unsigned int)__b;
 }
 
+#if defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+static vector signed __int128 __ATTRS_o_ai
+vec_sub(vector signed __int128 __a, vector signed __int128 __b)
+{
+  return __a - __b;
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_sub(vector unsigned __int128 __a, vector unsigned __int128 __b)
+{
+  return __a - __b;
+}
+#endif // defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+
 static vector float __ATTRS_o_ai
 vec_sub(vector float __a, vector float __b)
 {
@@ -8362,12 +8767,26 @@ vec_vsubfp(vector float __a, vector float __b)
 
 /* vec_subc */
 
-static vector unsigned int __attribute__((__always_inline__))
+static vector unsigned int __ATTRS_o_ai
 vec_subc(vector unsigned int __a, vector unsigned int __b)
 {
   return __builtin_altivec_vsubcuw(__a, __b);
 }
 
+#if defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+static vector unsigned __int128 __ATTRS_o_ai
+vec_subc(vector unsigned __int128 __a, vector unsigned __int128 __b)
+{
+  return __builtin_altivec_vsubcuq(__a, __b);
+}
+
+static vector signed __int128 __ATTRS_o_ai
+vec_subc(vector signed __int128 __a, vector signed __int128 __b)
+{
+  return __builtin_altivec_vsubcuq(__a, __b);
+}
+#endif // defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+
 /* vec_vsubcuw */
 
 static vector unsigned int __attribute__((__always_inline__))
@@ -8606,6 +9025,68 @@ vec_vsubuws(vector unsigned int __a, vector bool int __b)
   return __builtin_altivec_vsubuws(__a, (vector unsigned int)__b);
 }
 
+#if defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+/* vec_vsubuqm */
+
+static vector signed __int128 __ATTRS_o_ai
+vec_vsubuqm(vector signed __int128 __a, vector signed __int128 __b)
+{  
+  return __a - __b;
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_vsubuqm(vector unsigned __int128 __a, vector unsigned __int128 __b)
+{  
+  return __a - __b;
+}
+
+/* vec_vsubeuqm */
+
+static vector signed __int128 __ATTRS_o_ai
+vec_vsubeuqm(vector signed __int128 __a, vector signed __int128 __b,
+             vector signed __int128 __c)
+{  
+  return __builtin_altivec_vsubeuqm(__a, __b, __c);
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_vsubeuqm(vector unsigned __int128 __a, vector unsigned __int128 __b,
+             vector unsigned __int128 __c)
+{  
+  return __builtin_altivec_vsubeuqm(__a, __b, __c);
+}
+
+/* vec_vsubcuq */
+
+static vector signed __int128 __ATTRS_o_ai
+vec_vsubcuq(vector signed __int128 __a, vector signed __int128 __b)
+{  
+  return __builtin_altivec_vsubcuq(__a, __b);
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_vsubcuq(vector unsigned __int128 __a, vector unsigned __int128 __b)
+{  
+  return __builtin_altivec_vsubcuq(__a, __b);
+}
+
+/* vec_vsubecuq */
+
+static vector signed __int128 __ATTRS_o_ai
+vec_vsubecuq(vector signed __int128 __a, vector signed __int128 __b,
+             vector signed __int128 __c)
+{  
+  return __builtin_altivec_vsubecuq(__a, __b, __c);
+}
+
+static vector unsigned __int128 __ATTRS_o_ai
+vec_vsubecuq(vector unsigned __int128 __a, vector unsigned __int128 __b,
+             vector unsigned __int128 __c)
+{  
+  return __builtin_altivec_vsubecuq(__a, __b, __c);
+}
+#endif // defined(__POWER8_VECTOR__) && defined(__powerpc64__)
+
 /* vec_sum4s */
 
 static vector int __ATTRS_o_ai
@@ -8797,6 +9278,28 @@ vec_unpackh(vector pixel __a)
 #endif
 }
 
+#ifdef __POWER8_VECTOR__
+static vector long long __ATTRS_o_ai
+vec_unpackh(vector int __a)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vupklsw(__a);
+#else
+  return __builtin_altivec_vupkhsw(__a);
+#endif
+}
+
+static vector bool long long __ATTRS_o_ai
+vec_unpackh(vector bool int __a)
+{
+#ifdef __LITTLE_ENDIAN__
+  return (vector bool long long)__builtin_altivec_vupklsw((vector int)__a);
+#else
+  return (vector bool long long)__builtin_altivec_vupkhsw((vector int)__a);
+#endif
+}
+#endif
+
 /* vec_vupkhsb */
 
 static vector short __ATTRS_o_ai
@@ -8851,6 +9354,30 @@ vec_vupkhsh(vector pixel __a)
 #endif
 }
 
+/* vec_vupkhsw */
+
+#ifdef __POWER8_VECTOR__
+static vector long long __ATTRS_o_ai
+vec_vupkhsw(vector int __a)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vupklsw(__a);
+#else
+  return __builtin_altivec_vupkhsw(__a);
+#endif
+}
+
+static vector bool long long __ATTRS_o_ai
+vec_vupkhsw(vector bool int __a)
+{
+#ifdef __LITTLE_ENDIAN__
+  return (vector bool long long)__builtin_altivec_vupklsw((vector int)__a);
+#else
+  return (vector bool long long)__builtin_altivec_vupkhsw((vector int)__a);
+#endif
+}
+#endif
+
 /* vec_unpackl */
 
 static vector short __ATTRS_o_ai
@@ -8903,6 +9430,28 @@ vec_unpackl(vector pixel __a)
 #endif
 }
 
+#ifdef __POWER8_VECTOR__
+static vector long long __ATTRS_o_ai
+vec_unpackl(vector int __a)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vupkhsw(__a);
+#else
+  return __builtin_altivec_vupklsw(__a);
+#endif
+}
+
+static vector bool long long __ATTRS_o_ai
+vec_unpackl(vector bool int __a)
+{
+#ifdef __LITTLE_ENDIAN__
+  return (vector bool long long)__builtin_altivec_vupkhsw((vector int)__a);
+#else
+  return (vector bool long long)__builtin_altivec_vupklsw((vector int)__a);
+#endif
+}
+#endif
+
 /* vec_vupklsb */
 
 static vector short __ATTRS_o_ai
@@ -8957,6 +9506,30 @@ vec_vupklsh(vector pixel __a)
 #endif
 }
 
+/* vec_vupklsw */
+
+#ifdef __POWER8_VECTOR__
+static vector long long __ATTRS_o_ai
+vec_vupklsw(vector int __a)
+{
+#ifdef __LITTLE_ENDIAN__
+  return __builtin_altivec_vupkhsw(__a);
+#else
+  return __builtin_altivec_vupklsw(__a);
+#endif
+}
+
+static vector bool long long __ATTRS_o_ai
+vec_vupklsw(vector bool int __a)
+{
+#ifdef __LITTLE_ENDIAN__
+  return (vector bool long long)__builtin_altivec_vupkhsw((vector int)__a);
+#else
+  return (vector bool long long)__builtin_altivec_vupklsw((vector int)__a);
+#endif
+}
+#endif
+
 /* vec_vsx_ld */
 
 #ifdef __VSX__
@@ -10887,6 +11460,55 @@ vec_all_eq(vector bool int __a, vector bool int __b)
   return __builtin_altivec_vcmpequw_p(__CR6_LT, (vector int)__a, (vector int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_all_eq(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_LT, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_all_eq(vector long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_LT, __a, (vector long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_eq(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_LT, (vector long long)__a, 
+                                      (vector long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_eq(vector unsigned long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_LT, (vector long long)__a, 
+                                      (vector long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_eq(vector bool long long __a, vector long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_LT, (vector long long)__a, 
+                                      (vector long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_eq(vector bool long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_LT, (vector long long)__a, 
+                                      (vector long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_eq(vector bool long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_LT, (vector long long)__a, 
+                                      (vector long long)__b);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_all_eq(vector float __a, vector float __b)
 {
@@ -11033,6 +11655,56 @@ vec_all_ge(vector bool int __a, vector bool int __b)
                                       (vector unsigned int)__a);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_all_ge(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_EQ, __b, __a);
+}
+static int __ATTRS_o_ai
+vec_all_ge(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_EQ, (vector signed long long)__b,
+                                      __a);
+}
+
+static int __ATTRS_o_ai
+vec_all_ge(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ, __b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_all_ge(vector unsigned long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ, (vector unsigned long long)__b,
+                                      __a);
+}
+
+static int __ATTRS_o_ai
+vec_all_ge(vector bool long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ,
+                                      (vector unsigned long long)__b,
+                                      (vector unsigned long long)__a);
+}
+
+static int __ATTRS_o_ai
+vec_all_ge(vector bool long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ, __b,
+                                      (vector unsigned long long)__a);
+}
+
+static int __ATTRS_o_ai
+vec_all_ge(vector bool long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ,
+                                      (vector unsigned long long)__b,
+                                      (vector unsigned long long)__a);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_all_ge(vector float __a, vector float __b)
 {
@@ -11179,6 +11851,56 @@ vec_all_gt(vector bool int __a, vector bool int __b)
                                       (vector unsigned int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_all_gt(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_LT, __a, __b);
+}
+static int __ATTRS_o_ai
+vec_all_gt(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_LT, __a,
+                                      (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_gt(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_all_gt(vector unsigned long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT, __a, 
+                                      (vector unsigned long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_gt(vector bool long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT,
+                                      (vector unsigned long long)__a,
+                                      (vector unsigned long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_gt(vector bool long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT, (vector unsigned long long)__a,
+                                      __b);
+}
+
+static int __ATTRS_o_ai
+vec_all_gt(vector bool long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT,
+                                      (vector unsigned long long)__a,
+                                      (vector unsigned long long)__b);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_all_gt(vector float __a, vector float __b)
 {
@@ -11333,6 +12055,57 @@ vec_all_le(vector bool int __a, vector bool int __b)
                                       (vector unsigned int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_all_le(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_EQ, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_all_le(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_all_le(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_EQ, __a,
+                                      (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_le(vector unsigned long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ, __a, 
+                                      (vector unsigned long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_le(vector bool long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ,
+                                      (vector unsigned long long)__a,
+                                      (vector unsigned long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_le(vector bool long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ, (vector unsigned long long)__a,
+                                      __b);
+}
+
+static int __ATTRS_o_ai
+vec_all_le(vector bool long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ,
+                                      (vector unsigned long long)__a,
+                                      (vector unsigned long long)__b);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_all_le(vector float __a, vector float __b)
 {
@@ -11479,6 +12252,57 @@ vec_all_lt(vector bool int __a, vector bool int __b)
                                       (vector unsigned int)__a);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_all_lt(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_LT, __b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_all_lt(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT, __b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_all_lt(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_LT, (vector signed long long)__b,
+                                      __a);
+}
+
+static int __ATTRS_o_ai
+vec_all_lt(vector unsigned long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT, (vector unsigned long long)__b,
+                                      __a);
+}
+
+static int __ATTRS_o_ai
+vec_all_lt(vector bool long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT,
+                                      (vector unsigned long long)__b,
+                                      (vector unsigned long long)__a);
+}
+
+static int __ATTRS_o_ai
+vec_all_lt(vector bool long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT, __b,
+                                      (vector unsigned long long)__a);
+}
+
+static int __ATTRS_o_ai
+vec_all_lt(vector bool long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT,
+                                      (vector unsigned long long)__b,
+                                      (vector unsigned long long)__a);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_all_lt(vector float __a, vector float __b)
 {
@@ -11633,6 +12457,56 @@ vec_all_ne(vector bool int __a, vector bool int __b)
   return __builtin_altivec_vcmpequw_p(__CR6_EQ, (vector int)__a, (vector int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_all_ne(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_EQ, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_all_ne(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_EQ, (vector long long)__a, 
+                                      (vector long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_ne(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_EQ, __a,
+                                      (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_ne(vector unsigned long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_EQ, (vector signed long long)__a, 
+                                      (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_ne(vector bool long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_EQ, (vector signed long long)__a,
+                                      (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_ne(vector bool long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_EQ, (vector signed long long)__a, 
+                                      (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_all_ne(vector bool long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_EQ, (vector signed long long)__a, 
+                                      (vector signed long long)__b);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_all_ne(vector float __a, vector float __b)
 {
@@ -11837,6 +12711,61 @@ vec_any_eq(vector bool int __a, vector bool int __b)
     __builtin_altivec_vcmpequw_p(__CR6_EQ_REV, (vector int)__a, (vector int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_any_eq(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_EQ_REV, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_any_eq(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return
+    __builtin_altivec_vcmpequd_p(__CR6_EQ_REV, (vector long long)__a, 
+                                 (vector long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_eq(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_EQ_REV, __a,
+                                      (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_eq(vector unsigned long long __a, vector bool long long __b)
+{
+  return
+    __builtin_altivec_vcmpequd_p(__CR6_EQ_REV, (vector signed long long)__a, 
+                                 (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_eq(vector bool long long __a, vector signed long long __b)
+{
+  return
+    __builtin_altivec_vcmpequd_p(__CR6_EQ_REV, (vector signed long long)__a, 
+                                 (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_eq(vector bool long long __a, vector unsigned long long __b)
+{
+  return
+    __builtin_altivec_vcmpequd_p(__CR6_EQ_REV, (vector signed long long)__a, 
+                                 (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_eq(vector bool long long __a, vector bool long long __b)
+{
+  return
+    __builtin_altivec_vcmpequd_p(__CR6_EQ_REV, (vector signed long long)__a, 
+                                 (vector signed long long)__b);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_any_eq(vector float __a, vector float __b)
 {
@@ -11985,6 +12914,57 @@ vec_any_ge(vector bool int __a, vector bool int __b)
                                       (vector unsigned int)__a);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_any_ge(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_LT_REV, __b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_any_ge(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT_REV, __b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_any_ge(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_LT_REV,
+                                      (vector signed long long)__b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_any_ge(vector unsigned long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT_REV, 
+                                      (vector unsigned long long)__b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_any_ge(vector bool long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT_REV,
+                                      (vector unsigned long long)__b,
+                                      (vector unsigned long long)__a);
+}
+
+static int __ATTRS_o_ai
+vec_any_ge(vector bool long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT_REV, __b,
+                                      (vector unsigned long long)__a);
+}
+
+static int __ATTRS_o_ai
+vec_any_ge(vector bool long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT_REV,
+                                      (vector unsigned long long)__b,
+                                      (vector unsigned long long)__a);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_any_ge(vector float __a, vector float __b)
 {
@@ -12135,6 +13115,58 @@ vec_any_gt(vector bool int __a, vector bool int __b)
                                       (vector unsigned int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_any_gt(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_EQ_REV, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_any_gt(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ_REV, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_any_gt(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_EQ_REV, __a,
+                                      (vector signed long long)__b);
+}
+
+
+static int __ATTRS_o_ai
+vec_any_gt(vector unsigned long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ_REV, __a, 
+                                      (vector unsigned long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_gt(vector bool long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ_REV,
+                                      (vector unsigned long long)__a,
+                                      (vector unsigned long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_gt(vector bool long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ_REV, 
+                                      (vector unsigned long long)__a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_any_gt(vector bool long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ_REV,
+                                      (vector unsigned long long)__a,
+                                      (vector unsigned long long)__b);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_any_gt(vector float __a, vector float __b)
 {
@@ -12285,6 +13317,57 @@ vec_any_le(vector bool int __a, vector bool int __b)
                                       (vector unsigned int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_any_le(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_LT_REV, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_any_le(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT_REV, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_any_le(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_LT_REV, __a,
+                                      (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_le(vector unsigned long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT_REV, __a, 
+                                      (vector unsigned long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_le(vector bool long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT_REV,
+                                      (vector unsigned long long)__a,
+                                      (vector unsigned long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_le(vector bool long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT_REV, 
+                                      (vector unsigned long long)__a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_any_le(vector bool long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_LT_REV,
+                                      (vector unsigned long long)__a,
+                                      (vector unsigned long long)__b);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_any_le(vector float __a, vector float __b)
 {
@@ -12435,6 +13518,57 @@ vec_any_lt(vector bool int __a, vector bool int __b)
                                       (vector unsigned int)__a);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_any_lt(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_EQ_REV, __b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_any_lt(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ_REV, __b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_any_lt(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtsd_p(__CR6_EQ_REV,
+                                      (vector signed long long)__b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_any_lt(vector unsigned long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ_REV, 
+                                      (vector unsigned long long)__b, __a);
+}
+
+static int __ATTRS_o_ai
+vec_any_lt(vector bool long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ_REV,
+                                      (vector unsigned long long)__b,
+                                      (vector unsigned long long)__a);
+}
+
+static int __ATTRS_o_ai
+vec_any_lt(vector bool long long __a, vector unsigned long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ_REV, __b, 
+                                      (vector unsigned long long)__a);
+}
+
+static int __ATTRS_o_ai
+vec_any_lt(vector bool long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpgtud_p(__CR6_EQ_REV,
+                                      (vector unsigned long long)__b,
+                                      (vector unsigned long long)__a);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_any_lt(vector float __a, vector float __b)
 {
@@ -12607,6 +13741,61 @@ vec_any_ne(vector bool int __a, vector bool int __b)
     __builtin_altivec_vcmpequw_p(__CR6_LT_REV, (vector int)__a, (vector int)__b);
 }
 
+#ifdef __POWER8_VECTOR__
+static int __ATTRS_o_ai
+vec_any_ne(vector signed long long __a, vector signed long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_LT_REV, __a, __b);
+}
+
+static int __ATTRS_o_ai
+vec_any_ne(vector unsigned long long __a, vector unsigned long long __b)
+{
+  return
+    __builtin_altivec_vcmpequd_p(__CR6_LT_REV, (vector long long)__a, 
+                                 (vector long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_ne(vector signed long long __a, vector bool long long __b)
+{
+  return __builtin_altivec_vcmpequd_p(__CR6_LT_REV, __a,
+                                      (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_ne(vector unsigned long long __a, vector bool long long __b)
+{
+  return
+    __builtin_altivec_vcmpequd_p(__CR6_LT_REV, (vector signed long long)__a, 
+                                 (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_ne(vector bool long long __a, vector signed long long __b)
+{
+  return
+    __builtin_altivec_vcmpequd_p(__CR6_LT_REV, (vector signed long long)__a, 
+                                 (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_ne(vector bool long long __a, vector unsigned long long __b)
+{
+  return
+    __builtin_altivec_vcmpequd_p(__CR6_LT_REV, (vector signed long long)__a, 
+                                 (vector signed long long)__b);
+}
+
+static int __ATTRS_o_ai
+vec_any_ne(vector bool long long __a, vector bool long long __b)
+{
+  return
+    __builtin_altivec_vcmpequd_p(__CR6_LT_REV, (vector signed long long)__a, 
+                                 (vector signed long long)__b);
+}
+#endif
+
 static int __ATTRS_o_ai
 vec_any_ne(vector float __a, vector float __b)
 {
@@ -12661,6 +13850,133 @@ vec_any_out(vector float __a, vector float __b)
   return __builtin_altivec_vcmpbfp_p(__CR6_EQ_REV, __a, __b);
 }
 
+/* Power 8 Crypto functions
+Note: We diverge from the current GCC implementation with regard
+to cryptography and related functions as follows:
+- Only the SHA and AES instructions and builtins are disabled by -mno-crypto
+- The remaining ones are only available on Power8 and up so
+  require -mpower8-vector
+The justification for this is that export requirements require that
+Category:Vector.Crypto is optional (i.e. compliant hardware may not provide
+support). As a result, we need to be able to turn off support for those.
+The remaining ones (currently controlled by -mcrypto for GCC) still
+need to be provided on compliant hardware even if Vector.Crypto is not
+provided.
+FIXME: the naming convention for the builtins will be adjusted due
+to the inconsistency (__builtin_crypto_ prefix on builtins that cannot be
+removed with -mno-crypto). This is under development.
+*/
+#ifdef __CRYPTO__
+static vector unsigned long long __attribute__((__always_inline__))
+__builtin_crypto_vsbox (vector unsigned long long __a)
+{
+  return __builtin_altivec_crypto_vsbox(__a);
+}
+
+static vector unsigned long long __attribute__((__always_inline__))
+__builtin_crypto_vcipher (vector unsigned long long __a,
+                          vector unsigned long long __b)
+{
+  return __builtin_altivec_crypto_vcipher(__a, __b);
+}
+
+static vector unsigned long long __attribute__((__always_inline__))
+__builtin_crypto_vcipherlast (vector unsigned long long __a,
+                              vector unsigned long long __b)
+{
+  return __builtin_altivec_crypto_vcipherlast(__a, __b);
+}
+
+static vector unsigned long long __attribute__((__always_inline__))
+__builtin_crypto_vncipher (vector unsigned long long __a,
+                           vector unsigned long long __b)
+{
+  return __builtin_altivec_crypto_vncipher(__a, __b);
+}
+
+static vector unsigned long long __attribute__((__always_inline__))
+__builtin_crypto_vncipherlast (vector unsigned long long __a,
+                               vector unsigned long long __b)
+{
+  return __builtin_altivec_crypto_vncipherlast(__a, __b);
+}
+
+
+#define __builtin_crypto_vshasigmad __builtin_altivec_crypto_vshasigmad
+#define __builtin_crypto_vshasigmaw __builtin_altivec_crypto_vshasigmaw
+#endif
+
+#ifdef __POWER8_VECTOR__
+static vector unsigned char __ATTRS_o_ai
+__builtin_crypto_vpermxor (vector unsigned char __a,
+                           vector unsigned char __b,
+                           vector unsigned char __c)
+{
+  return __builtin_altivec_crypto_vpermxor(__a, __b, __c);
+}
+
+static vector unsigned short __ATTRS_o_ai
+__builtin_crypto_vpermxor (vector unsigned short __a,
+                           vector unsigned short __b,
+                           vector unsigned short __c)
+{
+  return (vector unsigned short)
+          __builtin_altivec_crypto_vpermxor((vector unsigned char) __a,
+                                             (vector unsigned char) __b,
+                                             (vector unsigned char) __c);
+}
+
+static vector unsigned int __ATTRS_o_ai
+__builtin_crypto_vpermxor (vector unsigned int __a,
+                           vector unsigned int __b,
+                           vector unsigned int __c)
+{
+  return (vector unsigned int)
+          __builtin_altivec_crypto_vpermxor((vector unsigned char) __a,
+                                              (vector unsigned char) __b,
+                                              (vector unsigned char) __c);
+}
+
+static vector unsigned long long __ATTRS_o_ai
+__builtin_crypto_vpermxor (vector unsigned long long __a,
+                           vector unsigned long long __b,
+                           vector unsigned long long __c)
+{
+  return (vector unsigned long long)
+          __builtin_altivec_crypto_vpermxor((vector unsigned char) __a,
+                                              (vector unsigned char) __b,
+                                              (vector unsigned char) __c);
+}
+
+static vector unsigned char __ATTRS_o_ai
+__builtin_crypto_vpmsumb (vector unsigned char __a,
+                          vector unsigned char __b)
+{
+  return __builtin_altivec_crypto_vpmsumb(__a, __b);
+}
+
+static vector unsigned short __ATTRS_o_ai
+__builtin_crypto_vpmsumb (vector unsigned short __a,
+                          vector unsigned short __b)
+{
+  return __builtin_altivec_crypto_vpmsumh(__a, __b);
+}
+
+static vector unsigned int __ATTRS_o_ai
+__builtin_crypto_vpmsumb (vector unsigned int __a,
+                          vector unsigned int __b)
+{
+  return __builtin_altivec_crypto_vpmsumw(__a, __b);
+}
+
+static vector unsigned long long __ATTRS_o_ai
+__builtin_crypto_vpmsumb (vector unsigned long long __a,
+                          vector unsigned long long __b)
+{
+  return __builtin_altivec_crypto_vpmsumd(__a, __b);
+}
+#endif
+
 #undef __ATTRS_o_ai
 
 #endif /* __ALTIVEC_H */
diff --git a/lib/Headers/arm_acle.h b/lib/Headers/arm_acle.h
index 814df2c3d782..6c56f3b77812 100644
--- a/lib/Headers/arm_acle.h
+++ b/lib/Headers/arm_acle.h
@@ -45,23 +45,23 @@ extern "C" {
 /* 8.4 Hints */
 
 #if !defined(_MSC_VER)
-static __inline__ void __attribute__((always_inline, nodebug)) __wfi(void) {
+static __inline__ void __attribute__((__always_inline__, __nodebug__)) __wfi(void) {
   __builtin_arm_wfi();
 }
 
-static __inline__ void __attribute__((always_inline, nodebug)) __wfe(void) {
+static __inline__ void __attribute__((__always_inline__, __nodebug__)) __wfe(void) {
   __builtin_arm_wfe();
 }
 
-static __inline__ void __attribute__((always_inline, nodebug)) __sev(void) {
+static __inline__ void __attribute__((__always_inline__, __nodebug__)) __sev(void) {
   __builtin_arm_sev();
 }
 
-static __inline__ void __attribute__((always_inline, nodebug)) __sevl(void) {
+static __inline__ void __attribute__((__always_inline__, __nodebug__)) __sevl(void) {
   __builtin_arm_sevl();
 }
 
-static __inline__ void __attribute__((always_inline, nodebug)) __yield(void) {
+static __inline__ void __attribute__((__always_inline__, __nodebug__)) __yield(void) {
   __builtin_arm_yield();
 }
 #endif
@@ -71,7 +71,7 @@ static __inline__ void __attribute__((always_inline, nodebug)) __yield(void) {
 #endif
 
 /* 8.5 Swap */
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __swp(uint32_t x, volatile uint32_t *p) {
   uint32_t v;
   do v = __builtin_arm_ldrex(p); while (__builtin_arm_strex(x, p));
@@ -102,28 +102,28 @@ static __inline__ uint32_t __attribute__((always_inline, nodebug))
 #endif
 
 /* 8.7 NOP */
-static __inline__ void __attribute__((always_inline, nodebug)) __nop(void) {
+static __inline__ void __attribute__((__always_inline__, __nodebug__)) __nop(void) {
   __builtin_arm_nop();
 }
 
 /* 9 DATA-PROCESSING INTRINSICS */
 /* 9.2 Miscellaneous data-processing intrinsics */
 /* ROR */
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __ror(uint32_t x, uint32_t y) {
   y %= 32;
   if (y == 0)  return x;
   return (x >> y) | (x << (32 - y));
 }
 
-static __inline__ uint64_t __attribute__((always_inline, nodebug))
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
   __rorll(uint64_t x, uint32_t y) {
   y %= 64;
   if (y == 0)  return x;
   return (x >> y) | (x << (64 - y));
 }
 
-static __inline__ unsigned long __attribute__((always_inline, nodebug))
+static __inline__ unsigned long __attribute__((__always_inline__, __nodebug__))
   __rorl(unsigned long x, uint32_t y) {
 #if __SIZEOF_LONG__ == 4
   return __ror(x, y);
@@ -134,28 +134,28 @@ static __inline__ unsigned long __attribute__((always_inline, nodebug))
 
 
 /* CLZ */
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __clz(uint32_t t) {
   return __builtin_clz(t);
 }
 
-static __inline__ unsigned long __attribute__((always_inline, nodebug))
+static __inline__ unsigned long __attribute__((__always_inline__, __nodebug__))
   __clzl(unsigned long t) {
   return __builtin_clzl(t);
 }
 
-static __inline__ uint64_t __attribute__((always_inline, nodebug))
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
   __clzll(uint64_t t) {
   return __builtin_clzll(t);
 }
 
 /* REV */
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __rev(uint32_t t) {
   return __builtin_bswap32(t);
 }
 
-static __inline__ unsigned long __attribute__((always_inline, nodebug))
+static __inline__ unsigned long __attribute__((__always_inline__, __nodebug__))
   __revl(unsigned long t) {
 #if __SIZEOF_LONG__ == 4
   return __builtin_bswap32(t);
@@ -164,40 +164,40 @@ static __inline__ unsigned long __attribute__((always_inline, nodebug))
 #endif
 }
 
-static __inline__ uint64_t __attribute__((always_inline, nodebug))
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
   __revll(uint64_t t) {
   return __builtin_bswap64(t);
 }
 
 /* REV16 */
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __rev16(uint32_t t) {
   return __ror(__rev(t), 16);
 }
 
-static __inline__ unsigned long __attribute__((always_inline, nodebug))
+static __inline__ unsigned long __attribute__((__always_inline__, __nodebug__))
   __rev16l(unsigned long t) {
     return __rorl(__revl(t), sizeof(long) / 2);
 }
 
-static __inline__ uint64_t __attribute__((always_inline, nodebug))
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
   __rev16ll(uint64_t t) {
   return __rorll(__revll(t), 32);
 }
 
 /* REVSH */
-static __inline__ int16_t __attribute__((always_inline, nodebug))
+static __inline__ int16_t __attribute__((__always_inline__, __nodebug__))
   __revsh(int16_t t) {
   return __builtin_bswap16(t);
 }
 
 /* RBIT */
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __rbit(uint32_t t) {
   return __builtin_arm_rbit(t);
 }
 
-static __inline__ uint64_t __attribute__((always_inline, nodebug))
+static __inline__ uint64_t __attribute__((__always_inline__, __nodebug__))
   __rbitll(uint64_t t) {
 #if __ARM_32BIT_STATE
   return (((uint64_t) __builtin_arm_rbit(t)) << 32) |
@@ -207,7 +207,7 @@ static __inline__ uint64_t __attribute__((always_inline, nodebug))
 #endif
 }
 
-static __inline__ unsigned long __attribute__((always_inline, nodebug))
+static __inline__ unsigned long __attribute__((__always_inline__, __nodebug__))
   __rbitl(unsigned long t) {
 #if __SIZEOF_LONG__ == 4
   return __rbit(t);
@@ -230,17 +230,17 @@ static __inline__ unsigned long __attribute__((always_inline, nodebug))
 
 /* 9.4.2 Saturating addition and subtraction intrinsics */
 #if __ARM_32BIT_STATE
-static __inline__ int32_t __attribute__((always_inline, nodebug))
+static __inline__ int32_t __attribute__((__always_inline__, __nodebug__))
   __qadd(int32_t t, int32_t v) {
   return __builtin_arm_qadd(t, v);
 }
 
-static __inline__ int32_t __attribute__((always_inline, nodebug))
+static __inline__ int32_t __attribute__((__always_inline__, __nodebug__))
   __qsub(int32_t t, int32_t v) {
   return __builtin_arm_qsub(t, v);
 }
 
-static __inline__ int32_t __attribute__((always_inline, nodebug))
+static __inline__ int32_t __attribute__((__always_inline__, __nodebug__))
 __qdbl(int32_t t) {
   return __builtin_arm_qadd(t, t);
 }
@@ -248,42 +248,42 @@ __qdbl(int32_t t) {
 
 /* 9.7 CRC32 intrinsics */
 #if __ARM_FEATURE_CRC32
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __crc32b(uint32_t a, uint8_t b) {
   return __builtin_arm_crc32b(a, b);
 }
 
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __crc32h(uint32_t a, uint16_t b) {
   return __builtin_arm_crc32h(a, b);
 }
 
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __crc32w(uint32_t a, uint32_t b) {
   return __builtin_arm_crc32w(a, b);
 }
 
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __crc32d(uint32_t a, uint64_t b) {
   return __builtin_arm_crc32d(a, b);
 }
 
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __crc32cb(uint32_t a, uint8_t b) {
   return __builtin_arm_crc32cb(a, b);
 }
 
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __crc32ch(uint32_t a, uint16_t b) {
   return __builtin_arm_crc32ch(a, b);
 }
 
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __crc32cw(uint32_t a, uint32_t b) {
   return __builtin_arm_crc32cw(a, b);
 }
 
-static __inline__ uint32_t __attribute__((always_inline, nodebug))
+static __inline__ uint32_t __attribute__((__always_inline__, __nodebug__))
   __crc32cd(uint32_t a, uint64_t b) {
   return __builtin_arm_crc32cd(a, b);
 }
diff --git a/lib/Headers/avx2intrin.h b/lib/Headers/avx2intrin.h
index 394fdfee9652..e1e639de1ba8 100644
--- a/lib/Headers/avx2intrin.h
+++ b/lib/Headers/avx2intrin.h
@@ -160,7 +160,7 @@ _mm256_blendv_epi8(__m256i __V1, __m256i __V2, __m256i __M)
 #define _mm256_blend_epi16(V1, V2, M) __extension__ ({ \
   __m256i __V1 = (V1); \
   __m256i __V2 = (V2); \
-  (__m256d)__builtin_shufflevector((__v16hi)__V1, (__v16hi)__V2, \
+  (__m256i)__builtin_shufflevector((__v16hi)__V1, (__v16hi)__V2, \
                                    (((M) & 0x01) ? 16 : 0), \
                                    (((M) & 0x02) ? 17 : 1), \
                                    (((M) & 0x04) ? 18 : 2), \
@@ -542,6 +542,8 @@ _mm256_sign_epi32(__m256i __a, __m256i __b)
   __m256i __a = (a); \
   (__m256i)__builtin_ia32_pslldqi256(__a, (count)*8); })
 
+#define _mm256_bslli_epi128(a, count) _mm256_slli_si256((a), (count))
+
 static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_slli_epi16(__m256i __a, int __count)
 {
@@ -606,6 +608,8 @@ _mm256_sra_epi32(__m256i __a, __m128i __count)
   __m256i __a = (a); \
   (__m256i)__builtin_ia32_psrldqi256(__a, (count)*8); })
 
+#define _mm256_bsrli_epi128(a, count) _mm256_srli_si256((a), (count))
+
 static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_srli_epi16(__m256i __a, int __count)
 {
@@ -756,6 +760,12 @@ _mm_broadcastss_ps(__m128 __X)
   return (__m128)__builtin_ia32_vbroadcastss_ps((__v4sf)__X);
 }
 
+static __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
+_mm_broadcastsd_pd(__m128d __a)
+{
+  return __builtin_shufflevector(__a, __a, 0, 0);
+}
+
 static __inline__ __m256 __attribute__((__always_inline__, __nodebug__))
 _mm256_broadcastss_ps(__m128 __X)
 {
@@ -771,7 +781,7 @@ _mm256_broadcastsd_pd(__m128d __X)
 static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_broadcastsi128_si256(__m128i __X)
 {
-  return (__m256i)__builtin_ia32_vbroadcastsi256(__X);
+  return (__m256i)__builtin_shufflevector(__X, __X, 0, 1, 0, 1);
 }
 
 #define _mm_blend_epi32(V1, V2, M) __extension__ ({ \
@@ -874,14 +884,21 @@ _mm256_permutevar8x32_ps(__m256 __a, __m256 __b)
   __m256i __V2 = (V2); \
   (__m256i)__builtin_ia32_permti256(__V1, __V2, (M)); })
 
-#define _mm256_extracti128_si256(A, O) __extension__ ({ \
-  __m256i __A = (A); \
-  (__m128i)__builtin_ia32_extract128i256(__A, (O)); })
-
-#define _mm256_inserti128_si256(V1, V2, O) __extension__ ({ \
-  __m256i __V1 = (V1); \
-  __m128i __V2 = (V2); \
-  (__m256i)__builtin_ia32_insert128i256(__V1, __V2, (O)); })
+#define _mm256_extracti128_si256(V, M) __extension__ ({ \
+  (__m128i)__builtin_shufflevector( \
+    (__v4di)(V), \
+    (__v4di)(_mm256_setzero_si256()), \
+    (((M) & 1) ? 2 : 0), \
+    (((M) & 1) ? 3 : 1) );})
+
+#define _mm256_inserti128_si256(V1, V2, M) __extension__ ({ \
+  (__m256i)__builtin_shufflevector( \
+    (__v4di)(V1), \
+    (__v4di)_mm256_castsi128_si256((__m128i)(V2)), \
+    (((M) & 1) ? 0 : 4), \
+    (((M) & 1) ? 1 : 5), \
+    (((M) & 1) ? 4 : 2), \
+    (((M) & 1) ? 5 : 3) );})
 
 static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_maskload_epi32(int const *__X, __m256i __M)
diff --git a/lib/Headers/avx512bwintrin.h b/lib/Headers/avx512bwintrin.h
index bc4d4ac6afdd..d0591e406f7b 100644
--- a/lib/Headers/avx512bwintrin.h
+++ b/lib/Headers/avx512bwintrin.h
@@ -21,15 +21,37 @@
  *
  *===-----------------------------------------------------------------------===
  */
+#ifndef __IMMINTRIN_H
+#error "Never use <avx512bwintrin.h> directly; include <immintrin.h> instead."
+#endif
 
 #ifndef __AVX512BWINTRIN_H
 #define __AVX512BWINTRIN_H
 
 typedef unsigned int __mmask32;
 typedef unsigned long long __mmask64;
-typedef char __v64qi __attribute__ ((vector_size (64)));
+typedef char __v64qi __attribute__ ((__vector_size__ (64)));
 typedef short __v32hi __attribute__ ((__vector_size__ (64)));
 
+static  __inline __v64qi __attribute__ ((__always_inline__, __nodebug__))
+_mm512_setzero_qi (void) {
+  return (__v64qi){ 0, 0, 0, 0, 0, 0, 0, 0,
+                       0, 0, 0, 0, 0, 0, 0, 0,
+                       0, 0, 0, 0, 0, 0, 0, 0,
+                       0, 0, 0, 0, 0, 0, 0, 0,
+                       0, 0, 0, 0, 0, 0, 0, 0,
+                       0, 0, 0, 0, 0, 0, 0, 0,
+                       0, 0, 0, 0, 0, 0, 0, 0,
+                       0, 0, 0, 0, 0, 0, 0, 0 };
+}
+
+static  __inline __v32hi __attribute__ ((__always_inline__, __nodebug__))
+_mm512_setzero_hi (void) {
+  return (__v32hi){ 0, 0, 0, 0, 0, 0, 0, 0,
+                       0, 0, 0, 0, 0, 0, 0, 0,
+                       0, 0, 0, 0, 0, 0, 0, 0,
+                       0, 0, 0, 0, 0, 0, 0, 0 };
+}
 
 /* Integer compare */
 
@@ -45,6 +67,18 @@ _mm512_mask_cmpeq_epi8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
                                                    __u);
 }
 
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpeq_epu8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 0,
+                                                 (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpeq_epu8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 0,
+                                                 __u);
+}
+
 static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
 _mm512_cmpeq_epi16_mask(__m512i __a, __m512i __b) {
   return (__mmask32)__builtin_ia32_pcmpeqw512_mask((__v32hi)__a, (__v32hi)__b,
@@ -57,4 +91,406 @@ _mm512_mask_cmpeq_epi16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
                                                    __u);
 }
 
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpeq_epu16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 0,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpeq_epu16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 0,
+                                                 __u);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpge_epi8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_cmpb512_mask((__v64qi)__a, (__v64qi)__b, 5,
+                                                (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpge_epi8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_cmpb512_mask((__v64qi)__a, (__v64qi)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpge_epu8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 5,
+                                                 (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpge_epu8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 5,
+                                                 __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpge_epi16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_cmpw512_mask((__v32hi)__a, (__v32hi)__b, 5,
+                                                (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpge_epi16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_cmpw512_mask((__v32hi)__a, (__v32hi)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpge_epu16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 5,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpge_epu16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 5,
+                                                 __u);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpgt_epi8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_pcmpgtb512_mask((__v64qi)__a, (__v64qi)__b,
+                                                   (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpgt_epi8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_pcmpgtb512_mask((__v64qi)__a, (__v64qi)__b,
+                                                   __u);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpgt_epu8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 6,
+                                                 (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpgt_epu8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 6,
+                                                 __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpgt_epi16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_pcmpgtw512_mask((__v32hi)__a, (__v32hi)__b,
+                                                   (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpgt_epi16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_pcmpgtw512_mask((__v32hi)__a, (__v32hi)__b,
+                                                   __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpgt_epu16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 6,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpgt_epu16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 6,
+                                                 __u);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmple_epi8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_cmpb512_mask((__v64qi)__a, (__v64qi)__b, 2,
+                                                (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmple_epi8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_cmpb512_mask((__v64qi)__a, (__v64qi)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmple_epu8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 2,
+                                                 (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmple_epu8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 2,
+                                                 __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmple_epi16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_cmpw512_mask((__v32hi)__a, (__v32hi)__b, 2,
+                                                (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmple_epi16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_cmpw512_mask((__v32hi)__a, (__v32hi)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmple_epu16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 2,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmple_epu16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 2,
+                                                 __u);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmplt_epi8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_cmpb512_mask((__v64qi)__a, (__v64qi)__b, 1,
+                                                (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmplt_epi8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_cmpb512_mask((__v64qi)__a, (__v64qi)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmplt_epu8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 1,
+                                                 (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmplt_epu8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 1,
+                                                 __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmplt_epi16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_cmpw512_mask((__v32hi)__a, (__v32hi)__b, 1,
+                                                (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmplt_epi16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_cmpw512_mask((__v32hi)__a, (__v32hi)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmplt_epu16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 1,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmplt_epu16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 1,
+                                                 __u);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpneq_epi8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_cmpb512_mask((__v64qi)__a, (__v64qi)__b, 4,
+                                                (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpneq_epi8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_cmpb512_mask((__v64qi)__a, (__v64qi)__b, 4,
+                                                __u);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpneq_epu8_mask(__m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 4,
+                                                 (__mmask64)-1);
+}
+
+static __inline__ __mmask64 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpneq_epu8_mask(__mmask64 __u, __m512i __a, __m512i __b) {
+  return (__mmask64)__builtin_ia32_ucmpb512_mask((__v64qi)__a, (__v64qi)__b, 4,
+                                                 __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpneq_epi16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_cmpw512_mask((__v32hi)__a, (__v32hi)__b, 4,
+                                                (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpneq_epi16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_cmpw512_mask((__v32hi)__a, (__v32hi)__b, 4,
+                                                __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpneq_epu16_mask(__m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 4,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpneq_epu16_mask(__mmask32 __u, __m512i __a, __m512i __b) {
+  return (__mmask32)__builtin_ia32_ucmpw512_mask((__v32hi)__a, (__v32hi)__b, 4,
+                                                 __u);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_add_epi8 (__m512i __A, __m512i __B) {
+  return (__m512i) ((__v64qi) __A + (__v64qi) __B);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_add_epi8 (__m512i __W, __mmask64 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_paddb512_mask ((__v64qi) __A,
+             (__v64qi) __B,
+             (__v64qi) __W,
+             (__mmask64) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_add_epi8 (__mmask64 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_paddb512_mask ((__v64qi) __A,
+             (__v64qi) __B,
+             (__v64qi)
+             _mm512_setzero_qi (),
+             (__mmask64) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_sub_epi8 (__m512i __A, __m512i __B) {
+  return (__m512i) ((__v64qi) __A - (__v64qi) __B);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_sub_epi8 (__m512i __W, __mmask64 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_psubb512_mask ((__v64qi) __A,
+             (__v64qi) __B,
+             (__v64qi) __W,
+             (__mmask64) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_sub_epi8 (__mmask64 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_psubb512_mask ((__v64qi) __A,
+             (__v64qi) __B,
+             (__v64qi)
+             _mm512_setzero_qi (),
+             (__mmask64) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_add_epi16 (__m512i __A, __m512i __B) {
+  return (__m512i) ((__v32hi) __A + (__v32hi) __B);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_add_epi16 (__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_paddw512_mask ((__v32hi) __A,
+             (__v32hi) __B,
+             (__v32hi) __W,
+             (__mmask32) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_add_epi16 (__mmask32 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_paddw512_mask ((__v32hi) __A,
+             (__v32hi) __B,
+             (__v32hi)
+             _mm512_setzero_hi (),
+             (__mmask32) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_sub_epi16 (__m512i __A, __m512i __B) {
+  return (__m512i) ((__v32hi) __A - (__v32hi) __B);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_sub_epi16 (__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_psubw512_mask ((__v32hi) __A,
+             (__v32hi) __B,
+             (__v32hi) __W,
+             (__mmask32) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_sub_epi16 (__mmask32 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_psubw512_mask ((__v32hi) __A,
+             (__v32hi) __B,
+             (__v32hi)
+             _mm512_setzero_hi (),
+             (__mmask32) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mullo_epi16 (__m512i __A, __m512i __B) {
+  return (__m512i) ((__v32hi) __A * (__v32hi) __B);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_mullo_epi16 (__m512i __W, __mmask32 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_pmullw512_mask ((__v32hi) __A,
+              (__v32hi) __B,
+              (__v32hi) __W,
+              (__mmask32) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_mullo_epi16 (__mmask32 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_pmullw512_mask ((__v32hi) __A,
+              (__v32hi) __B,
+              (__v32hi)
+              _mm512_setzero_hi (),
+              (__mmask32) __U);
+}
+
+#define _mm512_cmp_epi8_mask(a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpb512_mask((__v64qi)(__m512i)(a), \
+                                         (__v64qi)(__m512i)(b), \
+                                         (p), (__mmask64)-1); })
+
+#define _mm512_mask_cmp_epi8_mask(m, a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpb512_mask((__v64qi)(__m512i)(a), \
+                                         (__v64qi)(__m512i)(b), \
+                                         (p), (__mmask64)(m)); })
+
+#define _mm512_cmp_epu8_mask(a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_ucmpb512_mask((__v64qi)(__m512i)(a), \
+                                          (__v64qi)(__m512i)(b), \
+                                          (p), (__mmask64)-1); })
+
+#define _mm512_mask_cmp_epu8_mask(m, a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_ucmpb512_mask((__v64qi)(__m512i)(a), \
+                                          (__v64qi)(__m512i)(b), \
+                                          (p), (__mmask64)(m)); })
+
+#define _mm512_cmp_epi16_mask(a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpw512_mask((__v32hi)(__m512i)(a), \
+                                         (__v32hi)(__m512i)(b), \
+                                         (p), (__mmask32)-1); })
+
+#define _mm512_mask_cmp_epi16_mask(m, a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpw512_mask((__v32hi)(__m512i)(a), \
+                                         (__v32hi)(__m512i)(b), \
+                                         (p), (__mmask32)(m)); })
+
+#define _mm512_cmp_epu16_mask(a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_ucmpw512_mask((__v32hi)(__m512i)(a), \
+                                          (__v32hi)(__m512i)(b), \
+                                          (p), (__mmask32)-1); })
+
+#define _mm512_mask_cmp_epu16_mask(m, a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_ucmpw512_mask((__v32hi)(__m512i)(a), \
+                                          (__v32hi)(__m512i)(b), \
+                                          (p), (__mmask32)(m)); })
+
 #endif
diff --git a/lib/Headers/avx512dqintrin.h b/lib/Headers/avx512dqintrin.h
new file mode 100644
index 000000000000..fd33be2f44a6
--- /dev/null
+++ b/lib/Headers/avx512dqintrin.h
@@ -0,0 +1,237 @@
+/*===---- avx512dqintrin.h - AVX512DQ intrinsics ---------------------------===
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __IMMINTRIN_H
+#error "Never use <avx512dqintrin.h> directly; include <immintrin.h> instead."
+#endif
+
+#ifndef __AVX512DQINTRIN_H
+#define __AVX512DQINTRIN_H
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mullo_epi64 (__m512i __A, __m512i __B) {
+  return (__m512i) ((__v8di) __A * (__v8di) __B);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_mullo_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_pmullq512_mask ((__v8di) __A,
+              (__v8di) __B,
+              (__v8di) __W,
+              (__mmask8) __U);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_mullo_epi64 (__mmask8 __U, __m512i __A, __m512i __B) {
+  return (__m512i) __builtin_ia32_pmullq512_mask ((__v8di) __A,
+              (__v8di) __B,
+              (__v8di)
+              _mm512_setzero_si512 (),
+              (__mmask8) __U);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_xor_pd (__m512d __A, __m512d __B) {
+  return (__m512d) ((__v8di) __A ^ (__v8di) __B);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_xor_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
+  return (__m512d) __builtin_ia32_xorpd512_mask ((__v8df) __A,
+             (__v8df) __B,
+             (__v8df) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_xor_pd (__mmask8 __U, __m512d __A, __m512d __B) {
+  return (__m512d) __builtin_ia32_xorpd512_mask ((__v8df) __A,
+             (__v8df) __B,
+             (__v8df)
+             _mm512_setzero_pd (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_xor_ps (__m512 __A, __m512 __B) {
+  return (__m512) ((__v16si) __A ^ (__v16si) __B);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_xor_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
+  return (__m512) __builtin_ia32_xorps512_mask ((__v16sf) __A,
+            (__v16sf) __B,
+            (__v16sf) __W,
+            (__mmask16) __U);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_xor_ps (__mmask16 __U, __m512 __A, __m512 __B) {
+  return (__m512) __builtin_ia32_xorps512_mask ((__v16sf) __A,
+            (__v16sf) __B,
+            (__v16sf)
+            _mm512_setzero_ps (),
+            (__mmask16) __U);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_or_pd (__m512d __A, __m512d __B) {
+  return (__m512d) ((__v8di) __A | (__v8di) __B);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_or_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
+  return (__m512d) __builtin_ia32_orpd512_mask ((__v8df) __A,
+            (__v8df) __B,
+            (__v8df) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_or_pd (__mmask8 __U, __m512d __A, __m512d __B) {
+  return (__m512d) __builtin_ia32_orpd512_mask ((__v8df) __A,
+            (__v8df) __B,
+            (__v8df)
+            _mm512_setzero_pd (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_or_ps (__m512 __A, __m512 __B) {
+  return (__m512) ((__v16si) __A | (__v16si) __B);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_or_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
+  return (__m512) __builtin_ia32_orps512_mask ((__v16sf) __A,
+                 (__v16sf) __B,
+                 (__v16sf) __W,
+                 (__mmask16) __U);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_or_ps (__mmask16 __U, __m512 __A, __m512 __B) {
+  return (__m512) __builtin_ia32_orps512_mask ((__v16sf) __A,
+                 (__v16sf) __B,
+                 (__v16sf)
+                 _mm512_setzero_ps (),
+                 (__mmask16) __U);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_and_pd (__m512d __A, __m512d __B) {
+  return (__m512d) ((__v8di) __A & (__v8di) __B);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_and_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
+  return (__m512d) __builtin_ia32_andpd512_mask ((__v8df) __A,
+             (__v8df) __B,
+             (__v8df) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_and_pd (__mmask8 __U, __m512d __A, __m512d __B) {
+  return (__m512d) __builtin_ia32_andpd512_mask ((__v8df) __A,
+             (__v8df) __B,
+             (__v8df)
+             _mm512_setzero_pd (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_and_ps (__m512 __A, __m512 __B) {
+  return (__m512) ((__v16si) __A & (__v16si) __B);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_and_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
+  return (__m512) __builtin_ia32_andps512_mask ((__v16sf) __A,
+            (__v16sf) __B,
+            (__v16sf) __W,
+            (__mmask16) __U);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_and_ps (__mmask16 __U, __m512 __A, __m512 __B) {
+  return (__m512) __builtin_ia32_andps512_mask ((__v16sf) __A,
+            (__v16sf) __B,
+            (__v16sf)
+            _mm512_setzero_ps (),
+            (__mmask16) __U);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_andnot_pd (__m512d __A, __m512d __B) {
+  return (__m512d) __builtin_ia32_andnpd512_mask ((__v8df) __A,
+              (__v8df) __B,
+              (__v8df)
+              _mm512_setzero_pd (),
+              (__mmask8) -1);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_andnot_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
+  return (__m512d) __builtin_ia32_andnpd512_mask ((__v8df) __A,
+              (__v8df) __B,
+              (__v8df) __W,
+              (__mmask8) __U);
+}
+
+static __inline__ __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_andnot_pd (__mmask8 __U, __m512d __A, __m512d __B) {
+  return (__m512d) __builtin_ia32_andnpd512_mask ((__v8df) __A,
+              (__v8df) __B,
+              (__v8df)
+              _mm512_setzero_pd (),
+              (__mmask8) __U);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_andnot_ps (__m512 __A, __m512 __B) {
+  return (__m512) __builtin_ia32_andnps512_mask ((__v16sf) __A,
+             (__v16sf) __B,
+             (__v16sf)
+             _mm512_setzero_ps (),
+             (__mmask16) -1);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_andnot_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
+  return (__m512) __builtin_ia32_andnps512_mask ((__v16sf) __A,
+             (__v16sf) __B,
+             (__v16sf) __W,
+             (__mmask16) __U);
+}
+
+static __inline__ __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_andnot_ps (__mmask16 __U, __m512 __A, __m512 __B) {
+  return (__m512) __builtin_ia32_andnps512_mask ((__v16sf) __A,
+             (__v16sf) __B,
+             (__v16sf)
+             _mm512_setzero_ps (),
+             (__mmask16) __U);
+}
+
+#endif
diff --git a/lib/Headers/avx512erintrin.h b/lib/Headers/avx512erintrin.h
index 1a5ea153adf3..57c61aa0e112 100644
--- a/lib/Headers/avx512erintrin.h
+++ b/lib/Headers/avx512erintrin.h
@@ -28,85 +28,259 @@
 #define __AVX512ERINTRIN_H
 
 
+// exp2a23
+#define _mm512_exp2a23_round_pd(A, R) __extension__ ({ \
+  (__m512d)__builtin_ia32_exp2pd_mask((__v8df)(__m512d)(A), \
+                                      (__v8df)_mm512_setzero_pd(), \
+                                      (__mmask8)-1, (R)); })
+
+#define _mm512_mask_exp2a23_round_pd(S, M, A, R) __extension__ ({ \
+  (__m512d)__builtin_ia32_exp2pd_mask((__v8df)(__m512d)(A), \
+                                      (__v8df)(__m512d)(S), \
+                                      (__mmask8)(M), (R)); })
+
+#define _mm512_maskz_exp2a23_round_pd(M, A, R) __extension__ ({ \
+  (__m512d)__builtin_ia32_exp2pd_mask((__v8df)(__m512d)(A), \
+                                      (__v8df)_mm512_setzero_pd(), \
+                                      (__mmask8)(M), (R)); })
+
+#define _mm512_exp2a23_pd(A) \
+   _mm512_exp2a23_round_pd((A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_mask_exp2a23_pd(S, M, A) \
+   _mm512_mask_exp2a23_round_pd((S), (M), (A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_maskz_exp2a23_pd(M, A) \
+   _mm512_maskz_exp2a23_round_pd((M), (A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_exp2a23_round_ps(A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_exp2ps_mask((__v16sf)(__m512)(A), \
+                                     (__v16sf)_mm512_setzero_ps(), \
+                                     (__mmask8)-1, (R)); })
+
+#define _mm512_mask_exp2a23_round_ps(S, M, A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_exp2ps_mask((__v16sf)(__m512)(A), \
+                                     (__v16sf)(__m512)(S), \
+                                     (__mmask8)(M), (R)); })
+
+#define _mm512_maskz_exp2a23_round_ps(M, A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_exp2ps_mask((__v16sf)(__m512)(A), \
+                                     (__v16sf)_mm512_setzero_ps(), \
+                                     (__mmask8)(M), (R)); })
+
+#define _mm512_exp2a23_ps(A) \
+   _mm512_exp2a23_round_ps((A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_mask_exp2a23_ps(S, M, A) \
+   _mm512_mask_exp2a23_round_ps((S), (M), (A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_maskz_exp2a23_ps(M, A) \
+   _mm512_maskz_exp2a23_round_ps((M), (A), _MM_FROUND_CUR_DIRECTION)
+
 // rsqrt28
-static  __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
-_mm512_rsqrt28_round_pd (__m512d __A, int __R)
-{
-  return (__m512d)__builtin_ia32_rsqrt28pd_mask ((__v8df)__A,
-                                                 (__v8df)_mm512_setzero_pd(),
-                                                 (__mmask8)-1,
-                                                 __R);
-}
-static  __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
-_mm512_rsqrt28_round_ps(__m512 __A, int __R)
-{
-  return (__m512)__builtin_ia32_rsqrt28ps_mask ((__v16sf)__A,
-                                                (__v16sf)_mm512_setzero_ps(),
-                                                (__mmask16)-1,
-                                                __R);
-}
-
-static  __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
-_mm_rsqrt28_round_ss(__m128 __A, __m128 __B, int __R)
-{
-  return (__m128) __builtin_ia32_rsqrt28ss_mask ((__v4sf) __A,
-             (__v4sf) __B,
-             (__v4sf)
-             _mm_setzero_ps (),
-             (__mmask8) -1,
-             __R);
-}
-
-static  __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
-_mm_rsqrt28_round_sd (__m128d __A, __m128d __B, int __R)
-{
-  return (__m128d) __builtin_ia32_rsqrt28sd_mask ((__v2df) __A,
-              (__v2df) __B,
-              (__v2df)
-              _mm_setzero_pd (),
-              (__mmask8) -1,
-             __R);
-}
+#define _mm512_rsqrt28_round_pd(A, R) __extension__ ({ \
+  (__m512d)__builtin_ia32_rsqrt28pd_mask((__v8df)(__m512d)(A), \
+                                         (__v8df)_mm512_setzero_pd(), \
+                                         (__mmask8)-1, (R)); })
+
+#define _mm512_mask_rsqrt28_round_pd(S, M, A, R) __extension__ ({ \
+  (__m512d)__builtin_ia32_rsqrt28pd_mask((__v8df)(__m512d)(A), \
+                                         (__v8df)(__m512d)(S), \
+                                         (__mmask8)(M), (R)); })
+
+#define _mm512_maskz_rsqrt28_round_pd(M, A, R) __extension__ ({ \
+  (__m512d)__builtin_ia32_rsqrt28pd_mask((__v8df)(__m512d)(A), \
+                                         (__v8df)_mm512_setzero_pd(), \
+                                         (__mmask8)(M), (R)); })
+
+#define _mm512_rsqrt28_pd(A) \
+  _mm512_rsqrt28_round_pd((A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_mask_rsqrt28_pd(S, M, A) \
+  _mm512_mask_rsqrt28_round_pd((S), (M), (A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_maskz_rsqrt28_pd(M, A) \
+  _mm512_maskz_rsqrt28_round_pd((M), (A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_rsqrt28_round_ps(A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_rsqrt28ps_mask((__v16sf)(__m512)(A), \
+                                        (__v16sf)_mm512_setzero_ps(), \
+                                        (__mmask16)-1, (R)); })
+
+#define _mm512_mask_rsqrt28_round_ps(S, M, A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_rsqrt28ps_mask((__v16sf)(__m512)(A), \
+                                        (__v16sf)(__m512)(S), \
+                                        (__mmask16)(M), (R)); })
+
+#define _mm512_maskz_rsqrt28_round_ps(M, A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_rsqrt28ps_mask((__v16sf)(__m512)(A), \
+                                        (__v16sf)_mm512_setzero_ps(), \
+                                        (__mmask16)(M), (R)); })
+
+#define _mm512_rsqrt28_ps(A) \
+  _mm512_rsqrt28_round_ps((A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_mask_rsqrt28_ps(S, M, A) \
+  _mm512_mask_rsqrt28_round_ps((S), (M), A, _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_maskz_rsqrt28_ps(M, A) \
+  _mm512_maskz_rsqrt28_round_ps((M), (A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_rsqrt28_round_ss(A, B, R) __extension__ ({ \
+  (__m128)__builtin_ia32_rsqrt28ss_mask((__v4sf)(__m128)(A), \
+                                        (__v4sf)(__m128)(B), \
+                                        (__v4sf)_mm_setzero_ps(), \
+                                        (__mmask8)-1, (R)); })
+
+#define _mm_mask_rsqrt28_round_ss(S, M, A, B, R) __extension__ ({ \
+  (__m128)__builtin_ia32_rsqrt28ss_mask((__v4sf)(__m128)(A), \
+                                        (__v4sf)(__m128)(B), \
+                                        (__v4sf)(__m128)(S), \
+                                        (__mmask8)(M), (R)); })
+
+#define _mm_maskz_rsqrt28_round_ss(M, A, B, R) __extension__ ({ \
+  (__m128)__builtin_ia32_rsqrt28ss_mask((__v4sf)(__m128)(A), \
+                                        (__v4sf)(__m128)(B), \
+                                        (__v4sf)_mm_setzero_ps(), \
+                                        (__mmask8)(M), (R)); })
+
+#define _mm_rsqrt28_ss(A, B) \
+  _mm_rsqrt28_round_ss((A), (B), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_mask_rsqrt28_ss(S, M, A, B) \
+  _mm_mask_rsqrt28_round_ss((S), (M), (A), (B), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_maskz_rsqrt28_ss(M, A, B) \
+  _mm_maskz_rsqrt28_round_ss((M), (A), (B), _MM_FROUND_CUR_DIRECTION)
 
+#define _mm_rsqrt28_round_sd(A, B, R) __extension__ ({ \
+  (__m128d)__builtin_ia32_rsqrt28sd_mask((__v2df)(__m128d)(A), \
+                                         (__v2df)(__m128d)(B), \
+                                         (__v2df)_mm_setzero_pd(), \
+                                         (__mmask8)-1, (R)); })
+
+#define _mm_mask_rsqrt28_round_sd(S, M, A, B, R) __extension__ ({ \
+  (__m128d)__builtin_ia32_rsqrt28sd_mask((__v2df)(__m128d)(A), \
+                                         (__v2df)(__m128d)(B), \
+                                         (__v2df)(__m128d)(S), \
+                                         (__mmask8)(M), (R)); })
+
+#define _mm_maskz_rsqrt28_round_sd(M, A, B, R) __extension__ ({ \
+  (__m128d)__builtin_ia32_rsqrt28sd_mask((__v2df)(__m128d)(A), \
+                                         (__v2df)(__m128d)(B), \
+                                         (__v2df)_mm_setzero_pd(), \
+                                         (__mmask8)(M), (R)); })
+
+#define _mm_rsqrt28_sd(A, B) \
+  _mm_rsqrt28_round_sd((A), (B), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_mask_rsqrt28_sd(S, M, A, B) \
+  _mm_mask_rsqrt28_round_sd((S), (M), (A), (B), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_maskz_rsqrt28_sd(M, A, B) \
+  _mm_mask_rsqrt28_round_sd((M), (A), (B), _MM_FROUND_CUR_DIRECTION)
 
 // rcp28
-static  __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
-_mm512_rcp28_round_pd (__m512d __A, int __R)
-{
-  return (__m512d)__builtin_ia32_rcp28pd_mask ((__v8df)__A,
-                                               (__v8df)_mm512_setzero_pd(),
-                                               (__mmask8)-1,
-                                               __R);
-}
-
-static  __inline__ __m512 __attribute__((__always_inline__, __nodebug__))
-_mm512_rcp28_round_ps (__m512 __A, int __R)
-{
-  return (__m512)__builtin_ia32_rcp28ps_mask ((__v16sf)__A,
-                                              (__v16sf)_mm512_setzero_ps (),
-                                              (__mmask16)-1,
-                                              __R);
-}
-
-static  __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
-_mm_rcp28_round_ss (__m128 __A, __m128 __B, int __R)
-{
-  return (__m128) __builtin_ia32_rcp28ss_mask ((__v4sf) __A,
-             (__v4sf) __B,
-             (__v4sf)
-             _mm_setzero_ps (),
-             (__mmask8) -1,
-             __R);
-}
-static  __inline__ __m128d __attribute__((__always_inline__, __nodebug__))
-_mm_rcp28_round_sd (__m128d __A, __m128d __B, int __R)
-{
-  return (__m128d) __builtin_ia32_rcp28sd_mask ((__v2df) __A,
-              (__v2df) __B,
-              (__v2df)
-              _mm_setzero_pd (),
-              (__mmask8) -1,
-             __R);
-}
+#define _mm512_rcp28_round_pd(A, R) __extension__ ({ \
+  (__m512d)__builtin_ia32_rcp28pd_mask((__v8df)(__m512d)(A), \
+                                       (__v8df)_mm512_setzero_pd(), \
+                                       (__mmask8)-1, (R)); })
+
+#define _mm512_mask_rcp28_round_pd(S, M, A, R) __extension__ ({ \
+  (__m512d)__builtin_ia32_rcp28pd_mask((__v8df)(__m512d)(A), \
+                                       (__v8df)(__m512d)(S), \
+                                       (__mmask8)(M), (R)); })
+
+#define _mm512_maskz_rcp28_round_pd(M, A, R) __extension__ ({ \
+  (__m512d)__builtin_ia32_rcp28pd_mask((__v8df)(__m512d)(A), \
+                                       (__v8df)_mm512_setzero_pd(), \
+                                       (__mmask8)(M), (R)); })
+
+#define _mm512_rcp28_pd(A) \
+  _mm512_rcp28_round_pd((A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_mask_rcp28_pd(S, M, A) \
+  _mm512_mask_rcp28_round_pd((S), (M), (A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_maskz_rcp28_pd(M, A) \
+  _mm512_maskz_rcp28_round_pd((M), (A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_rcp28_round_ps(A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_rcp28ps_mask((__v16sf)(__m512)(A), \
+                                      (__v16sf)_mm512_setzero_ps(), \
+                                      (__mmask16)-1, (R)); })
+
+#define _mm512_mask_rcp28_round_ps(S, M, A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_rcp28ps_mask((__v16sf)(__m512)(A), \
+                                      (__v16sf)(__m512)(S), \
+                                      (__mmask16)(M), (R)); })
+
+#define _mm512_maskz_rcp28_round_ps(M, A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_rcp28ps_mask((__v16sf)(__m512)(A), \
+                                      (__v16sf)_mm512_setzero_ps(), \
+                                      (__mmask16)(M), (R)); })
+
+#define _mm512_rcp28_ps(A) \
+  _mm512_rcp28_round_ps((A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_mask_rcp28_ps(S, M, A) \
+  _mm512_mask_rcp28_round_ps((S), (M), (A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_maskz_rcp28_ps(M, A) \
+  _mm512_maskz_rcp28_round_ps((M), (A), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_rcp28_round_ss(A, B, R) __extension__ ({ \
+  (__m128)__builtin_ia32_rcp28ss_mask((__v4sf)(__m128)(A), \
+                                      (__v4sf)(__m128)(B), \
+                                      (__v4sf)_mm_setzero_ps(), \
+                                      (__mmask8)-1, (R)); })
+
+#define _mm_mask_rcp28_round_ss(S, M, A, B, R) __extension__ ({ \
+  (__m128)__builtin_ia32_rcp28ss_mask((__v4sf)(__m128)(A), \
+                                      (__v4sf)(__m128)(B), \
+                                      (__v4sf)(__m128)(S), \
+                                      (__mmask8)(M), (R)); })
+
+#define _mm_maskz_rcp28_round_ss(M, A, B, R) __extension__ ({ \
+  (__m128)__builtin_ia32_rcp28ss_mask((__v4sf)(__m128)(A), \
+                                      (__v4sf)(__m128)(B), \
+                                      (__v4sf)_mm_setzero_ps(), \
+                                      (__mmask8)(M), (R)); })
+
+#define _mm_rcp28_ss(A, B) \
+  _mm_rcp28_round_ss((A), (B), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_mask_rcp28_ss(S, M, A, B) \
+  _mm_mask_rcp28_round_ss((S), (M), (A), (B), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_maskz_rcp28_ss(M, A, B) \
+  _mm_maskz_rcp28_round_ss((M), (A), (B), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_rcp28_round_sd(A, B, R) __extension__ ({ \
+  (__m128d)__builtin_ia32_rcp28sd_mask((__v2df)(__m128d)(A), \
+                                       (__v2df)(__m128d)(B), \
+                                       (__v2df)_mm_setzero_pd(), \
+                                       (__mmask8)-1, (R)); })
+
+#define _mm_mask_rcp28_round_sd(S, M, A, B, R) __extension__ ({ \
+  (__m128d)__builtin_ia32_rcp28sd_mask((__v2df)(__m128d)(A), \
+                                       (__v2df)(__m128d)(B), \
+                                       (__v2df)(__m128d)(S), \
+                                       (__mmask8)(M), (R)); })
+
+#define _mm_maskz_rcp28_round_sd(M, A, B, R) __extension__ ({ \
+  (__m128d)__builtin_ia32_rcp28sd_mask((__v2df)(__m128d)(A), \
+                                       (__v2df)(__m128d)(B), \
+                                       (__v2df)_mm_setzero_pd(), \
+                                       (__mmask8)(M), (R)); })
+
+#define _mm_rcp28_sd(A, B) \
+  _mm_rcp28_round_sd((A), (B), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_mask_rcp28_sd(S, M, A, B) \
+  _mm_mask_rcp28_round_sd((S), (M), (A), (B), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm_maskz_rcp28_sd(M, A, B) \
+  _mm_maskz_rcp28_round_sd((M), (A), (B), _MM_FROUND_CUR_DIRECTION)
 
 #endif // __AVX512ERINTRIN_H
diff --git a/lib/Headers/avx512fintrin.h b/lib/Headers/avx512fintrin.h
index 9c80710110b0..d299704d9b2b 100644
--- a/lib/Headers/avx512fintrin.h
+++ b/lib/Headers/avx512fintrin.h
@@ -162,6 +162,224 @@ _mm512_castps512_ps128(__m512 __a)
   return __builtin_shufflevector(__a, __a, 0, 1, 2, 3);
 }
 
+/* Bitwise operators */
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_and_epi32(__m512i __a, __m512i __b)
+{
+  return __a & __b;
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_and_epi32(__m512i __src, __mmask16 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_pandd512_mask((__v16si) __a,
+              (__v16si) __b,
+              (__v16si) __src,
+              (__mmask16) __k);
+}
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_and_epi32(__mmask16 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_pandd512_mask((__v16si) __a,
+              (__v16si) __b,
+              (__v16si)
+              _mm512_setzero_si512 (),
+              (__mmask16) __k);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_and_epi64(__m512i __a, __m512i __b)
+{
+  return __a & __b;
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_and_epi64(__m512i __src, __mmask8 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_pandq512_mask ((__v8di) __a,
+              (__v8di) __b,
+              (__v8di) __src,
+              (__mmask8) __k);
+}
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_and_epi64(__mmask8 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_pandq512_mask ((__v8di) __a,
+              (__v8di) __b,
+              (__v8di)
+              _mm512_setzero_si512 (),
+              (__mmask8) __k);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_andnot_epi32 (__m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_pandnd512_mask ((__v16si) __A,
+              (__v16si) __B,
+              (__v16si)
+              _mm512_setzero_si512 (),
+              (__mmask16) -1);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_andnot_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_pandnd512_mask ((__v16si) __A,
+              (__v16si) __B,
+              (__v16si) __W,
+              (__mmask16) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_andnot_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_pandnd512_mask ((__v16si) __A,
+              (__v16si) __B,
+              (__v16si)
+              _mm512_setzero_si512 (),
+              (__mmask16) __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_andnot_epi64 (__m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_pandnq512_mask ((__v8di) __A,
+              (__v8di) __B,
+              (__v8di)
+              _mm512_setzero_si512 (),
+              (__mmask8) -1);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_andnot_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_pandnq512_mask ((__v8di) __A,
+              (__v8di) __B,
+              (__v8di) __W, __U);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_andnot_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_pandnq512_mask ((__v8di) __A,
+              (__v8di) __B,
+              (__v8di)
+              _mm512_setzero_pd (),
+              __U);
+}
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_or_epi32(__m512i __a, __m512i __b)
+{
+  return __a | __b;
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_or_epi32(__m512i __src, __mmask16 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_pord512_mask((__v16si) __a,
+              (__v16si) __b,
+              (__v16si) __src,
+              (__mmask16) __k);
+}
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_or_epi32(__mmask16 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_pord512_mask((__v16si) __a,
+              (__v16si) __b,
+              (__v16si)
+              _mm512_setzero_si512 (),
+              (__mmask16) __k);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_or_epi64(__m512i __a, __m512i __b)
+{
+  return __a | __b;
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_or_epi64(__m512i __src, __mmask8 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_porq512_mask ((__v8di) __a,
+              (__v8di) __b,
+              (__v8di) __src,
+              (__mmask8) __k);
+}
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_or_epi64(__mmask8 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_porq512_mask ((__v8di) __a,
+              (__v8di) __b,
+              (__v8di)
+              _mm512_setzero_si512 (),
+              (__mmask8) __k);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_xor_epi32(__m512i __a, __m512i __b)
+{
+  return __a ^ __b;
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_xor_epi32(__m512i __src, __mmask16 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_pxord512_mask((__v16si) __a,
+              (__v16si) __b,
+              (__v16si) __src,
+              (__mmask16) __k);
+}
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_xor_epi32(__mmask16 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_pxord512_mask((__v16si) __a,
+              (__v16si) __b,
+              (__v16si)
+              _mm512_setzero_si512 (),
+              (__mmask16) __k);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_xor_epi64(__m512i __a, __m512i __b)
+{
+  return __a ^ __b;
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_xor_epi64(__m512i __src, __mmask8 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_pxorq512_mask ((__v8di) __a,
+              (__v8di) __b,
+              (__v8di) __src,
+              (__mmask8) __k);
+}
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_maskz_xor_epi64(__mmask8 __k, __m512i __a, __m512i __b)
+{
+  return (__m512i) __builtin_ia32_pxorq512_mask ((__v8di) __a,
+              (__v8di) __b,
+              (__v8di)
+              _mm512_setzero_si512 (),
+              (__mmask8) __k);
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_and_si512(__m512i __a, __m512i __b)
+{
+  return __a & __b;
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_or_si512(__m512i __a, __m512i __b)
+{
+  return __a | __b;
+}
+
+static __inline__ __m512i __attribute__((__always_inline__, __nodebug__))
+_mm512_xor_si512(__m512i __a, __m512i __b)
+{
+  return __a ^ __b;
+}
 /* Arithmetic */
 
 static __inline __m512d __attribute__((__always_inline__, __nodebug__))
@@ -200,6 +418,106 @@ _mm512_sub_ps(__m512 __a, __m512 __b)
   return __a - __b;
 }
 
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_add_epi64 (__m512i __A, __m512i __B)
+{
+  return (__m512i) ((__v8di) __A + (__v8di) __B);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_add_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_paddq512_mask ((__v8di) __A,
+             (__v8di) __B,
+             (__v8di) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_add_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_paddq512_mask ((__v8di) __A,
+             (__v8di) __B,
+             (__v8di)
+             _mm512_setzero_si512 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_sub_epi64 (__m512i __A, __m512i __B)
+{
+  return (__m512i) ((__v8di) __A - (__v8di) __B);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_sub_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_psubq512_mask ((__v8di) __A,
+             (__v8di) __B,
+             (__v8di) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_sub_epi64 (__mmask8 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_psubq512_mask ((__v8di) __A,
+             (__v8di) __B,
+             (__v8di)
+             _mm512_setzero_si512 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_add_epi32 (__m512i __A, __m512i __B)
+{
+  return (__m512i) ((__v16si) __A + (__v16si) __B);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_add_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_paddd512_mask ((__v16si) __A,
+             (__v16si) __B,
+             (__v16si) __W,
+             (__mmask16) __U);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_add_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_paddd512_mask ((__v16si) __A,
+             (__v16si) __B,
+             (__v16si)
+             _mm512_setzero_si512 (),
+             (__mmask16) __U);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_sub_epi32 (__m512i __A, __m512i __B)
+{
+  return (__m512i) ((__v16si) __A - (__v16si) __B);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_sub_epi32 (__m512i __W, __mmask16 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_psubd512_mask ((__v16si) __A,
+             (__v16si) __B,
+             (__v16si) __W,
+             (__mmask16) __U);
+}
+
+static __inline__ __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_sub_epi32 (__mmask16 __U, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_psubd512_mask ((__v16si) __A,
+             (__v16si) __B,
+             (__v16si)
+             _mm512_setzero_si512 (),
+             (__mmask16) __U);
+}
+
 static  __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
 _mm512_max_pd(__m512d __A, __m512d __B)
 {
@@ -337,6 +655,24 @@ _mm512_mul_epi32(__m512i __X, __m512i __Y)
 }
 
 static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_mul_epi32 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y)
+{
+  return (__m512i) __builtin_ia32_pmuldq512_mask ((__v16si) __X,
+              (__v16si) __Y,
+              (__v8di) __W, __M);
+}
+
+static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_mul_epi32 (__mmask8 __M, __m512i __X, __m512i __Y)
+{
+  return (__m512i) __builtin_ia32_pmuldq512_mask ((__v16si) __X,
+              (__v16si) __Y,
+              (__v8di)
+              _mm512_setzero_si512 (),
+              __M);
+}
+
+static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
 _mm512_mul_epu32(__m512i __X, __m512i __Y)
 {
   return (__m512i) __builtin_ia32_pmuludq512_mask ((__v16si) __X,
@@ -346,6 +682,48 @@ _mm512_mul_epu32(__m512i __X, __m512i __Y)
                (__mmask8) -1);
 }
 
+static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_mul_epu32 (__m512i __W, __mmask8 __M, __m512i __X, __m512i __Y)
+{
+  return (__m512i) __builtin_ia32_pmuludq512_mask ((__v16si) __X,
+               (__v16si) __Y,
+               (__v8di) __W, __M);
+}
+
+static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_mul_epu32 (__mmask8 __M, __m512i __X, __m512i __Y)
+{
+  return (__m512i) __builtin_ia32_pmuludq512_mask ((__v16si) __X,
+               (__v16si) __Y,
+               (__v8di)
+               _mm512_setzero_si512 (),
+               __M);
+}
+
+static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mullo_epi32 (__m512i __A, __m512i __B)
+{
+  return (__m512i) ((__v16si) __A * (__v16si) __B);
+}
+
+static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_mullo_epi32 (__mmask16 __M, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_pmulld512_mask ((__v16si) __A,
+              (__v16si) __B,
+              (__v16si)
+              _mm512_setzero_si512 (),
+              __M);
+}
+
+static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_mullo_epi32 (__m512i __W, __mmask16 __M, __m512i __A, __m512i __B)
+{
+  return (__m512i) __builtin_ia32_pmulld512_mask ((__v16si) __A,
+              (__v16si) __B,
+              (__v16si) __W, __M);
+}
+
 static  __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
 _mm512_sqrt_pd(__m512d a)
 {
@@ -492,20 +870,13 @@ _mm512_abs_epi32(__m512i __A)
              (__mmask16) -1);
 }
 
-static __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
-_mm512_roundscale_ps(__m512 __A, const int __imm)
-{
-  return (__m512) __builtin_ia32_rndscaleps_mask ((__v16sf) __A, __imm,
-              (__v16sf) __A, -1,
-              _MM_FROUND_CUR_DIRECTION);
-}
-static __inline __m512d __attribute__ ((__always_inline__, __nodebug__))
-_mm512_roundscale_pd(__m512d __A, const int __imm)
-{
-  return (__m512d) __builtin_ia32_rndscalepd_mask ((__v8df) __A, __imm,
-               (__v8df) __A, -1,
-               _MM_FROUND_CUR_DIRECTION);
-}
+#define _mm512_roundscale_ps(A, B) __extension__ ({ \
+  (__m512)__builtin_ia32_rndscaleps_mask((__v16sf)(A), (B), (__v16sf)(A), \
+                                         -1, _MM_FROUND_CUR_DIRECTION); })
+
+#define _mm512_roundscale_pd(A, B) __extension__ ({ \
+  (__m512d)__builtin_ia32_rndscalepd_mask((__v8df)(A), (B), (__v8df)(A), \
+                                          -1, _MM_FROUND_CUR_DIRECTION); })
 
 static __inline__ __m512d __attribute__((__always_inline__, __nodebug__))
 _mm512_fmadd_pd(__m512d __A, __m512d __B, __m512d __C)
@@ -613,25 +984,35 @@ _mm512_permutex2var_ps(__m512 __A, __m512i __I, __m512 __B)
                                                        (__mmask16) -1);
 }
 
-static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
-_mm512_valign_epi64(__m512i __A, __m512i __B, const int __I)
-{
-  return (__m512i) __builtin_ia32_alignq512_mask((__v8di)__A,
-                                                 (__v8di)__B,
-                                                 __I,
-                                                 (__v8di)_mm512_setzero_si512(),
-                                                 (__mmask8) -1);
-}
-
-static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
-_mm512_valign_epi32(__m512i __A, __m512i __B, const int __I)
-{
-  return (__m512i)__builtin_ia32_alignd512_mask((__v16si)__A,
-                                                (__v16si)__B,
-                                                __I,
-                                                (__v16si)_mm512_setzero_si512(),
-                                                (__mmask16) -1);
-}
+#define _mm512_alignr_epi64(A, B, I) __extension__ ({ \
+  (__m512i)__builtin_ia32_alignq512_mask((__v8di)(__m512i)(A), \
+                                         (__v8di)(__m512i)(B), \
+                                         (I), (__v8di)_mm512_setzero_si512(), \
+                                         (__mmask8)-1); })
+
+#define _mm512_alignr_epi32(A, B, I) __extension__ ({ \
+  (__m512i)__builtin_ia32_alignd512_mask((__v16si)(__m512i)(A), \
+                                         (__v16si)(__m512i)(B), \
+                                         (I), (__v16si)_mm512_setzero_si512(), \
+                                         (__mmask16)-1); })
+
+/* Vector Extract */
+
+#define _mm512_extractf64x4_pd(A, I) __extension__ ({                    \
+      __m512d __A = (A);                                                 \
+      (__m256d)                                                          \
+        __builtin_ia32_extractf64x4_mask((__v8df)__A,                    \
+                                         (I),                            \
+                                         (__v4df)_mm256_setzero_si256(), \
+                                         (__mmask8) -1); })
+
+#define _mm512_extractf32x4_ps(A, I) __extension__ ({                    \
+      __m512 __A = (A);                                                  \
+      (__m128)                                                           \
+        __builtin_ia32_extractf32x4_mask((__v16sf)__A,                   \
+                                         (I),                            \
+                                         (__v4sf)_mm_setzero_ps(),       \
+                                         (__mmask8) -1); })
 
 /* Vector Blend */
 
@@ -669,22 +1050,37 @@ _mm512_mask_blend_epi32(__mmask16 __U, __m512i __A, __m512i __W)
 
 /* Compare */
 
-static __inline __mmask16 __attribute__ ((__always_inline__, __nodebug__))
-_mm512_cmp_ps_mask(__m512 a, __m512 b, const int p)
-{
-  return (__mmask16) __builtin_ia32_cmpps512_mask ((__v16sf) a,
-               (__v16sf) b, p, (__mmask16) -1,
-               _MM_FROUND_CUR_DIRECTION);
-}
+#define _mm512_cmp_round_ps_mask(A, B, P, R) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpps512_mask((__v16sf)(__m512)(A), \
+                                          (__v16sf)(__m512)(B), \
+                                          (P), (__mmask16)-1, (R)); })
 
-static __inline __mmask8 __attribute__ ((__always_inline__, __nodebug__)) 
-_mm512_cmp_pd_mask(__m512d __X, __m512d __Y, const int __P)
-{
-  return (__mmask8) __builtin_ia32_cmppd512_mask ((__v8df) __X,
-              (__v8df) __Y, __P,
-              (__mmask8) -1,
-              _MM_FROUND_CUR_DIRECTION);
-}
+#define _mm512_mask_cmp_round_ps_mask(U, A, B, P, R) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpps512_mask((__v16sf)(__m512)(A), \
+                                          (__v16sf)(__m512)(B), \
+                                          (P), (__mmask16)(U), (R)); })
+
+#define _mm512_cmp_ps_mask(A, B, P) \
+  _mm512_cmp_round_ps_mask((A), (B), (P), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_mask_cmp_ps_mask(U, A, B, P) \
+  _mm512_mask_cmp_round_ps_mask((U), (A), (B), (P), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_cmp_round_pd_mask(A, B, P, R) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmppd512_mask((__v8df)(__m512d)(A), \
+                                         (__v8df)(__m512d)(B), \
+                                         (P), (__mmask8)-1, (R)); })
+
+#define _mm512_mask_cmp_round_pd_mask(U, A, B, P, R) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmppd512_mask((__v8df)(__m512d)(A), \
+                                         (__v8df)(__m512d)(B), \
+                                         (P), (__mmask8)(U), (R)); })
+
+#define _mm512_cmp_pd_mask(A, B, P) \
+  _mm512_cmp_round_pd_mask((A), (B), (P), _MM_FROUND_CUR_DIRECTION)
+
+#define _mm512_mask_cmp_pd_mask(U, A, B, P) \
+  _mm512_mask_cmp_round_pd_mask((U), (A), (B), (P), _MM_FROUND_CUR_DIRECTION)
 
 /* Conversion */
 
@@ -698,25 +1094,15 @@ _mm512_cvttps_epu32(__m512 __A)
                   _MM_FROUND_CUR_DIRECTION);
 }
 
-static __inline __m512 __attribute__ (( __always_inline__, __nodebug__))
-_mm512_cvt_roundepi32_ps(__m512i __A, const int __R)
-{
-  return (__m512) __builtin_ia32_cvtdq2ps512_mask ((__v16si) __A,
-               (__v16sf)
-               _mm512_setzero_ps (),
-               (__mmask16) -1,
-               __R);
-}
+#define _mm512_cvt_roundepi32_ps(A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_cvtdq2ps512_mask((__v16si)(A), \
+                                          (__v16sf)_mm512_setzero_ps(), \
+                                          (__mmask16)-1, (R)); })
 
-static __inline __m512 __attribute__ (( __always_inline__, __nodebug__))
-_mm512_cvt_roundepu32_ps(__m512i __A, const int __R)
-{
-  return (__m512) __builtin_ia32_cvtudq2ps512_mask ((__v16si) __A,
-               (__v16sf)
-               _mm512_setzero_ps (),
-               (__mmask16) -1,
-               __R);
-}
+#define _mm512_cvt_roundepu32_ps(A, R) __extension__ ({ \
+  (__m512)__builtin_ia32_cvtudq2ps512_mask((__v16si)(A), \
+                                           (__v16sf)_mm512_setzero_ps(), \
+                                           (__mmask16)-1, (R)); })
 
 static __inline __m512d __attribute__ (( __always_inline__, __nodebug__))
 _mm512_cvtepi32_pd(__m256i __A)
@@ -735,25 +1121,16 @@ _mm512_cvtepu32_pd(__m256i __A)
                 _mm512_setzero_pd (),
                 (__mmask8) -1);
 }
-static __inline __m256 __attribute__ (( __always_inline__, __nodebug__))
-_mm512_cvt_roundpd_ps(__m512d __A, const int __R)
-{
-  return (__m256) __builtin_ia32_cvtpd2ps512_mask ((__v8df) __A,
-               (__v8sf)
-               _mm256_setzero_ps (),
-               (__mmask8) -1,
-               __R);
-}
 
-static  __inline __m256i __attribute__ ((__always_inline__, __nodebug__))
-_mm512_cvtps_ph(__m512 __A, const int __I)
-{
-  return (__m256i) __builtin_ia32_vcvtps2ph512_mask ((__v16sf) __A,
-                 __I,
-                 (__v16hi)
-                 _mm256_setzero_si256 (),
-                 -1);
-}
+#define _mm512_cvt_roundpd_ps(A, R) __extension__ ({ \
+  (__m256)__builtin_ia32_cvtpd2ps512_mask((__v8df)(A), \
+                                          (__v8sf)_mm256_setzero_ps(), \
+                                          (__mmask8)-1, (R)); })
+
+#define _mm512_cvtps_ph(A, I) __extension__ ({ \
+  (__m256i)__builtin_ia32_vcvtps2ph512_mask((__v16sf)(A), (I), \
+                                            (__v16hi)_mm256_setzero_si256(), \
+                                            -1); })
 
 static  __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
 _mm512_cvtph_ps(__m256i __A)
@@ -783,61 +1160,35 @@ _mm512_cvttpd_epi32(__m512d a)
                                                     _MM_FROUND_CUR_DIRECTION);
 }
 
-static __inline __m256i __attribute__ ((__always_inline__, __nodebug__))
-_mm512_cvtt_roundpd_epi32(__m512d __A, const int __R)
-{
-  return (__m256i) __builtin_ia32_cvttpd2dq512_mask ((__v8df) __A,
-                 (__v8si)
-                 _mm256_setzero_si256 (),
-                 (__mmask8) -1,
-                 __R);
-}
-static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
-_mm512_cvtt_roundps_epi32(__m512 __A, const int __R)
-{
-  return (__m512i) __builtin_ia32_cvttps2dq512_mask ((__v16sf) __A,
-                 (__v16si)
-                 _mm512_setzero_si512 (),
-                 (__mmask16) -1,
-                 __R);
-}
+#define _mm512_cvtt_roundpd_epi32(A, R) __extension__ ({ \
+  (__m256i)__builtin_ia32_cvttpd2dq512_mask((__v8df)(A), \
+                                            (__v8si)_mm256_setzero_si256(), \
+                                            (__mmask8)-1, (R)); })
 
-static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
-_mm512_cvt_roundps_epi32(__m512 __A, const int __R)
-{
-  return (__m512i) __builtin_ia32_cvtps2dq512_mask ((__v16sf) __A,
-                (__v16si)
-                _mm512_setzero_si512 (),
-                (__mmask16) -1,
-                __R);
-}
-static __inline __m256i __attribute__ ((__always_inline__, __nodebug__))
-_mm512_cvt_roundpd_epi32(__m512d __A, const int __R)
-{
-  return (__m256i) __builtin_ia32_cvtpd2dq512_mask ((__v8df) __A,
-                (__v8si)
-                _mm256_setzero_si256 (),
-                (__mmask8) -1,
-                __R);
-}
-static __inline __m512i __attribute__ ((__always_inline__, __nodebug__))
-_mm512_cvt_roundps_epu32(__m512 __A, const int __R)
-{
-  return (__m512i) __builtin_ia32_cvtps2udq512_mask ((__v16sf) __A,
-                (__v16si)
-                _mm512_setzero_si512 (),
-                (__mmask16) -1,
-                __R);
-}
-static __inline __m256i __attribute__ ((__always_inline__, __nodebug__))
-_mm512_cvt_roundpd_epu32(__m512d __A, const int __R)
-{
-  return (__m256i) __builtin_ia32_cvtpd2udq512_mask ((__v8df) __A,
-                (__v8si)
-                _mm256_setzero_si256 (),
-                (__mmask8) -1,
-                __R);
-}
+#define _mm512_cvtt_roundps_epi32(A, R) __extension__ ({ \
+  (__m512i)__builtin_ia32_cvttps2dq512_mask((__v16sf)(A), \
+                                            (__v16si)_mm512_setzero_si512(), \
+                                            (__mmask16)-1, (R)); })
+
+#define _mm512_cvt_roundps_epi32(A, R) __extension__ ({ \
+  (__m512i)__builtin_ia32_cvtps2dq512_mask((__v16sf)(A), \
+                                           (__v16si)_mm512_setzero_si512(), \
+                                           (__mmask16)-1, (R)); })
+
+#define _mm512_cvt_roundpd_epi32(A, R) __extension__ ({ \
+  (__m256i)__builtin_ia32_cvtpd2dq512_mask((__v8df)(A), \
+                                           (__v8si)_mm256_setzero_si256(), \
+                                           (__mmask8)-1, (R)); })
+
+#define _mm512_cvt_roundps_epu32(A, R) __extension__ ({ \
+  (__m512i)__builtin_ia32_cvtps2udq512_mask((__v16sf)(A), \
+                                            (__v16si)_mm512_setzero_si512(), \
+                                            (__mmask16)-1, (R)); })
+
+#define _mm512_cvt_roundpd_epu32(A, R) __extension__ ({ \
+  (__m256i)__builtin_ia32_cvtpd2udq512_mask((__v8df)(A), \
+                                            (__v8si)_mm256_setzero_si256(), \
+                                            (__mmask8) -1, (R)); })
 
 /* Unpack and Interleave */
 static __inline __m512d __attribute__((__always_inline__, __nodebug__))
@@ -928,12 +1279,30 @@ _mm512_maskz_loadu_pd(__mmask8 __U, void const *__P)
                                                    (__mmask8) __U);
 }
 
+static __inline __m512 __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_load_ps(__mmask16 __U, void const *__P)
+{
+  return (__m512) __builtin_ia32_loadaps512_mask ((const __v16sf *)__P,
+                                                  (__v16sf)
+                                                  _mm512_setzero_ps (),
+                                                  (__mmask16) __U);
+}
+
+static __inline __m512d __attribute__ ((__always_inline__, __nodebug__))
+_mm512_maskz_load_pd(__mmask8 __U, void const *__P)
+{
+  return (__m512d) __builtin_ia32_loadapd512_mask ((const __v8df *)__P,
+                                                   (__v8df)
+                                                   _mm512_setzero_pd (),
+                                                   (__mmask8) __U);
+}
+
 static __inline __m512d __attribute__((__always_inline__, __nodebug__))
 _mm512_loadu_pd(double const *__p)
 {
   struct __loadu_pd {
     __m512d __v;
-  } __attribute__((packed, may_alias));
+  } __attribute__((__packed__, __may_alias__));
   return ((struct __loadu_pd*)__p)->__v;
 }
 
@@ -942,10 +1311,28 @@ _mm512_loadu_ps(float const *__p)
 {
   struct __loadu_ps {
     __m512 __v;
-  } __attribute__((packed, may_alias));
+  } __attribute__((__packed__, __may_alias__));
   return ((struct __loadu_ps*)__p)->__v;
 }
 
+static __inline __m512 __attribute__((__always_inline__, __nodebug__))
+_mm512_load_ps(double const *__p)
+{
+  return (__m512) __builtin_ia32_loadaps512_mask ((const __v16sf *)__p,
+                                                  (__v16sf)
+                                                  _mm512_setzero_ps (),
+                                                  (__mmask16) -1);
+}
+
+static __inline __m512d __attribute__((__always_inline__, __nodebug__))
+_mm512_load_pd(float const *__p)
+{
+  return (__m512d) __builtin_ia32_loadapd512_mask ((const __v8df *)__p,
+                                                   (__v8df)
+                                                   _mm512_setzero_pd (),
+                                                   (__mmask8) -1);
+}
+
 /* SIMD store ops */
 
 static __inline void __attribute__ ((__always_inline__, __nodebug__))
@@ -988,9 +1375,9 @@ _mm512_storeu_ps(void *__P, __m512 __A)
 }
 
 static __inline void __attribute__ ((__always_inline__, __nodebug__))
-_mm512_store_ps(void *__P, __m512 __A)
+_mm512_mask_store_pd(void *__P, __mmask8 __U, __m512d __A)
 {
-  *(__m512*)__P = __A;
+  __builtin_ia32_storeapd512_mask ((__v8df *)__P, (__v8df) __A, (__mmask8) __U);
 }
 
 static __inline void __attribute__ ((__always_inline__, __nodebug__))
@@ -999,6 +1386,19 @@ _mm512_store_pd(void *__P, __m512d __A)
   *(__m512d*)__P = __A;
 }
 
+static __inline void __attribute__ ((__always_inline__, __nodebug__))
+_mm512_mask_store_ps(void *__P, __mmask16 __U, __m512 __A)
+{
+  __builtin_ia32_storeaps512_mask ((__v16sf *)__P, (__v16sf) __A,
+                                   (__mmask16) __U);
+}
+
+static __inline void __attribute__ ((__always_inline__, __nodebug__))
+_mm512_store_ps(void *__P, __m512 __A)
+{
+  *(__m512*)__P = __A;
+}
+
 /* Mask ops */
 
 static __inline __mmask16 __attribute__ ((__always_inline__, __nodebug__))
@@ -1021,6 +1421,18 @@ _mm512_mask_cmpeq_epi32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
                                                    __u);
 }
 
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpeq_epu32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 0,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpeq_epu32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 0,
+                                                 __u);
+}
+
 static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
 _mm512_mask_cmpeq_epi64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
   return (__mmask8)__builtin_ia32_pcmpeqq512_mask((__v8di)__a, (__v8di)__b,
@@ -1033,4 +1445,303 @@ _mm512_cmpeq_epi64_mask(__m512i __a, __m512i __b) {
                                                   (__mmask8)-1);
 }
 
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpeq_epu64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 0,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpeq_epu64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 0,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpge_epi32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_cmpd512_mask((__v16si)__a, (__v16si)__b, 5,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpge_epi32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_cmpd512_mask((__v16si)__a, (__v16si)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpge_epu32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 5,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpge_epu32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 5,
+                                                 __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpge_epi64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_cmpq512_mask((__v8di)__a, (__v8di)__b, 5,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpge_epi64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_cmpq512_mask((__v8di)__a, (__v8di)__b, 5,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpge_epu64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 5,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpge_epu64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpgt_epi32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_pcmpgtd512_mask((__v16si)__a, (__v16si)__b,
+                                                   (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpgt_epi32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_pcmpgtd512_mask((__v16si)__a, (__v16si)__b,
+                                                   __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpgt_epu32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 6,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpgt_epu32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 6,
+                                                 __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpgt_epi64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtq512_mask((__v8di)__a, (__v8di)__b,
+                                                  __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpgt_epi64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtq512_mask((__v8di)__a, (__v8di)__b,
+                                                  (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpgt_epu64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 6,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpgt_epu64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 6,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmple_epi32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_cmpd512_mask((__v16si)__a, (__v16si)__b, 2,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmple_epi32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_cmpd512_mask((__v16si)__a, (__v16si)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmple_epu32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 2,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmple_epu32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 2,
+                                                 __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmple_epi64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_cmpq512_mask((__v8di)__a, (__v8di)__b, 2,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmple_epi64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_cmpq512_mask((__v8di)__a, (__v8di)__b, 2,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmple_epu64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 2,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmple_epu64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmplt_epi32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_cmpd512_mask((__v16si)__a, (__v16si)__b, 1,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmplt_epi32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_cmpd512_mask((__v16si)__a, (__v16si)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmplt_epu32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 1,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmplt_epu32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 1,
+                                                 __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmplt_epi64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_cmpq512_mask((__v8di)__a, (__v8di)__b, 1,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmplt_epi64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_cmpq512_mask((__v8di)__a, (__v8di)__b, 1,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmplt_epu64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 1,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmplt_epu64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpneq_epi32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_cmpd512_mask((__v16si)__a, (__v16si)__b, 4,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpneq_epi32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_cmpd512_mask((__v16si)__a, (__v16si)__b, 4,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpneq_epu32_mask(__m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 4,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpneq_epu32_mask(__mmask16 __u, __m512i __a, __m512i __b) {
+  return (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, 4,
+                                                 __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpneq_epi64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_cmpq512_mask((__v8di)__a, (__v8di)__b, 4,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpneq_epi64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_cmpq512_mask((__v8di)__a, (__v8di)__b, 4,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_cmpneq_epu64_mask(__m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 4,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm512_mask_cmpneq_epu64_mask(__mmask8 __u, __m512i __a, __m512i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, 4,
+                                                __u);
+}
+
+#define _mm512_cmp_epi32_mask(a, b, p) __extension__ ({ \
+  __m512i __a = (a); \
+  __m512i __b = (b); \
+  (__mmask16)__builtin_ia32_cmpd512_mask((__v16si)__a, (__v16si)__b, (p), \
+                                         (__mmask16)-1); })
+
+#define _mm512_cmp_epu32_mask(a, b, p) __extension__ ({ \
+  __m512i __a = (a); \
+  __m512i __b = (b); \
+  (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, (p), \
+                                          (__mmask16)-1); })
+
+#define _mm512_cmp_epi64_mask(a, b, p) __extension__ ({ \
+  __m512i __a = (a); \
+  __m512i __b = (b); \
+  (__mmask8)__builtin_ia32_cmpq512_mask((__v8di)__a, (__v8di)__b, (p), \
+                                        (__mmask8)-1); })
+
+#define _mm512_cmp_epu64_mask(a, b, p) __extension__ ({ \
+  __m512i __a = (a); \
+  __m512i __b = (b); \
+  (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, (p), \
+                                         (__mmask8)-1); })
+
+#define _mm512_mask_cmp_epi32_mask(m, a, b, p) __extension__ ({ \
+  __m512i __a = (a); \
+  __m512i __b = (b); \
+  (__mmask16)__builtin_ia32_cmpd512_mask((__v16si)__a, (__v16si)__b, (p), \
+                                         (__mmask16)(m)); })
+
+#define _mm512_mask_cmp_epu32_mask(m, a, b, p) __extension__ ({ \
+  __m512i __a = (a); \
+  __m512i __b = (b); \
+  (__mmask16)__builtin_ia32_ucmpd512_mask((__v16si)__a, (__v16si)__b, (p), \
+                                          (__mmask16)(m)); })
+
+#define _mm512_mask_cmp_epi64_mask(m, a, b, p) __extension__ ({ \
+  __m512i __a = (a); \
+  __m512i __b = (b); \
+  (__mmask8)__builtin_ia32_cmpq512_mask((__v8di)__a, (__v8di)__b, (p), \
+                                        (__mmask8)(m)); })
+
+#define _mm512_mask_cmp_epu64_mask(m, a, b, p) __extension__ ({ \
+  __m512i __a = (a); \
+  __m512i __b = (b); \
+  (__mmask8)__builtin_ia32_ucmpq512_mask((__v8di)__a, (__v8di)__b, (p), \
+                                         (__mmask8)(m)); })
 #endif // __AVX512FINTRIN_H
diff --git a/lib/Headers/avx512vlbwintrin.h b/lib/Headers/avx512vlbwintrin.h
index 11333f851756..c3b087e303b4 100644
--- a/lib/Headers/avx512vlbwintrin.h
+++ b/lib/Headers/avx512vlbwintrin.h
@@ -42,6 +42,17 @@ _mm_mask_cmpeq_epi8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
                                                    __u);
 }
 
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpeq_epu8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 0,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpeq_epu8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 0,
+                                                 __u);
+}
 
 static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
 _mm256_cmpeq_epi8_mask(__m256i __a, __m256i __b) {
@@ -55,6 +66,18 @@ _mm256_mask_cmpeq_epi8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
                                                    __u);
 }
 
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpeq_epu8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 0,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpeq_epu8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 0,
+                                                 __u);
+}
+
 static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
 _mm_cmpeq_epi16_mask(__m128i __a, __m128i __b) {
   return (__mmask8)__builtin_ia32_pcmpeqw128_mask((__v8hi)__a, (__v8hi)__b,
@@ -67,6 +90,17 @@ _mm_mask_cmpeq_epi16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
                                                   __u);
 }
 
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpeq_epu16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 0,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpeq_epu16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 0,
+                                                __u);
+}
 
 static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
 _mm256_cmpeq_epi16_mask(__m256i __a, __m256i __b) {
@@ -80,4 +114,744 @@ _mm256_mask_cmpeq_epi16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
                                                    __u);
 }
 
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpeq_epu16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 0,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpeq_epu16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 0,
+                                                 __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpge_epi8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_cmpb128_mask((__v16qi)__a, (__v16qi)__b, 5,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpge_epi8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_cmpb128_mask((__v16qi)__a, (__v16qi)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpge_epu8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 5,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpge_epu8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 5,
+                                                 __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpge_epi8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_cmpb256_mask((__v32qi)__a, (__v32qi)__b, 5,
+                                                (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpge_epi8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_cmpb256_mask((__v32qi)__a, (__v32qi)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpge_epu8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 5,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpge_epu8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 5,
+                                                 __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpge_epi16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpw128_mask((__v8hi)__a, (__v8hi)__b, 5,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpge_epi16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpw128_mask((__v8hi)__a, (__v8hi)__b, 5,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpge_epu16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 5,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpge_epu16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpge_epi16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_cmpw256_mask((__v16hi)__a, (__v16hi)__b, 5,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpge_epi16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_cmpw256_mask((__v16hi)__a, (__v16hi)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpge_epu16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 5,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpge_epu16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 5,
+                                                 __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpgt_epi8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_pcmpgtb128_mask((__v16qi)__a, (__v16qi)__b,
+                                                   (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpgt_epi8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_pcmpgtb128_mask((__v16qi)__a, (__v16qi)__b,
+                                                   __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpgt_epu8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 6,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpgt_epu8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 6,
+                                                 __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epi8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_pcmpgtb256_mask((__v32qi)__a, (__v32qi)__b,
+                                                   (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpgt_epi8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_pcmpgtb256_mask((__v32qi)__a, (__v32qi)__b,
+                                                   __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epu8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 6,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpgt_epu8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 6,
+                                                 __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpgt_epi16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtw128_mask((__v8hi)__a, (__v8hi)__b,
+                                                  (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpgt_epi16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtw128_mask((__v8hi)__a, (__v8hi)__b,
+                                                  __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpgt_epu16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 6,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpgt_epu16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 6,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epi16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_pcmpgtw256_mask((__v16hi)__a, (__v16hi)__b,
+                                                   (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpgt_epi16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_pcmpgtw256_mask((__v16hi)__a, (__v16hi)__b,
+                                                   __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epu16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 6,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpgt_epu16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 6,
+                                                 __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmple_epi8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_cmpb128_mask((__v16qi)__a, (__v16qi)__b, 2,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmple_epi8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_cmpb128_mask((__v16qi)__a, (__v16qi)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmple_epu8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 2,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmple_epu8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 2,
+                                                 __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmple_epi8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_cmpb256_mask((__v32qi)__a, (__v32qi)__b, 2,
+                                                (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmple_epi8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_cmpb256_mask((__v32qi)__a, (__v32qi)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmple_epu8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 2,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmple_epu8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 2,
+                                                 __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmple_epi16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpw128_mask((__v8hi)__a, (__v8hi)__b, 2,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmple_epi16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpw128_mask((__v8hi)__a, (__v8hi)__b, 2,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmple_epu16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 2,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmple_epu16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmple_epi16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_cmpw256_mask((__v16hi)__a, (__v16hi)__b, 2,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmple_epi16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_cmpw256_mask((__v16hi)__a, (__v16hi)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmple_epu16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 2,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmple_epu16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 2,
+                                                 __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmplt_epi8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_cmpb128_mask((__v16qi)__a, (__v16qi)__b, 1,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmplt_epi8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_cmpb128_mask((__v16qi)__a, (__v16qi)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmplt_epu8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 1,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmplt_epu8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 1,
+                                                 __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmplt_epi8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_cmpb256_mask((__v32qi)__a, (__v32qi)__b, 1,
+                                                (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmplt_epi8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_cmpb256_mask((__v32qi)__a, (__v32qi)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmplt_epu8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 1,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmplt_epu8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 1,
+                                                 __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmplt_epi16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpw128_mask((__v8hi)__a, (__v8hi)__b, 1,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmplt_epi16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpw128_mask((__v8hi)__a, (__v8hi)__b, 1,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmplt_epu16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 1,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmplt_epu16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmplt_epi16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_cmpw256_mask((__v16hi)__a, (__v16hi)__b, 1,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmplt_epi16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_cmpw256_mask((__v16hi)__a, (__v16hi)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmplt_epu16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 1,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmplt_epu16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 1,
+                                                 __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpneq_epi8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_cmpb128_mask((__v16qi)__a, (__v16qi)__b, 4,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpneq_epi8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_cmpb128_mask((__v16qi)__a, (__v16qi)__b, 4,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpneq_epu8_mask(__m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 4,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpneq_epu8_mask(__mmask16 __u, __m128i __a, __m128i __b) {
+  return (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)__a, (__v16qi)__b, 4,
+                                                 __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpneq_epi8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_cmpb256_mask((__v32qi)__a, (__v32qi)__b, 4,
+                                                (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpneq_epi8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_cmpb256_mask((__v32qi)__a, (__v32qi)__b, 4,
+                                                __u);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpneq_epu8_mask(__m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 4,
+                                                 (__mmask32)-1);
+}
+
+static __inline__ __mmask32 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpneq_epu8_mask(__mmask32 __u, __m256i __a, __m256i __b) {
+  return (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)__a, (__v32qi)__b, 4,
+                                                 __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpneq_epi16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpw128_mask((__v8hi)__a, (__v8hi)__b, 4,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpneq_epi16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpw128_mask((__v8hi)__a, (__v8hi)__b, 4,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpneq_epu16_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 4,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpneq_epu16_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)__a, (__v8hi)__b, 4,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpneq_epi16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_cmpw256_mask((__v16hi)__a, (__v16hi)__b, 4,
+                                                (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpneq_epi16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_cmpw256_mask((__v16hi)__a, (__v16hi)__b, 4,
+                                                __u);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpneq_epu16_mask(__m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 4,
+                                                 (__mmask16)-1);
+}
+
+static __inline__ __mmask16 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpneq_epu16_mask(__mmask16 __u, __m256i __a, __m256i __b) {
+  return (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)__a, (__v16hi)__b, 4,
+                                                 __u);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_add_epi8 (__m256i __W, __mmask32 __U, __m256i __A, __m256i __B){
+  return (__m256i) __builtin_ia32_paddb256_mask ((__v32qi) __A,
+             (__v32qi) __B,
+             (__v32qi) __W,
+             (__mmask32) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_maskz_add_epi8 (__mmask32 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_paddb256_mask ((__v32qi) __A,
+             (__v32qi) __B,
+             (__v32qi)
+             _mm256_setzero_si256 (),
+             (__mmask32) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_add_epi16 (__m256i __W, __mmask16 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_paddw256_mask ((__v16hi) __A,
+             (__v16hi) __B,
+             (__v16hi) __W,
+             (__mmask16) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_maskz_add_epi16 (__mmask16 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_paddw256_mask ((__v16hi) __A,
+             (__v16hi) __B,
+             (__v16hi)
+             _mm256_setzero_si256 (),
+             (__mmask16) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_sub_epi8 (__m256i __W, __mmask32 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_psubb256_mask ((__v32qi) __A,
+             (__v32qi) __B,
+             (__v32qi) __W,
+             (__mmask32) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_maskz_sub_epi8 (__mmask32 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_psubb256_mask ((__v32qi) __A,
+             (__v32qi) __B,
+             (__v32qi)
+             _mm256_setzero_si256 (),
+             (__mmask32) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_sub_epi16 (__m256i __W, __mmask16 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_psubw256_mask ((__v16hi) __A,
+             (__v16hi) __B,
+             (__v16hi) __W,
+             (__mmask16) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_maskz_sub_epi16 (__mmask16 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_psubw256_mask ((__v16hi) __A,
+             (__v16hi) __B,
+             (__v16hi)
+             _mm256_setzero_si256 (),
+             (__mmask16) __U);
+}
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_mask_add_epi8 (__m128i __W, __mmask16 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_paddb128_mask ((__v16qi) __A,
+             (__v16qi) __B,
+             (__v16qi) __W,
+             (__mmask16) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_maskz_add_epi8 (__mmask16 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_paddb128_mask ((__v16qi) __A,
+             (__v16qi) __B,
+             (__v16qi)
+             _mm_setzero_si128 (),
+             (__mmask16) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_mask_add_epi16 (__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_paddw128_mask ((__v8hi) __A,
+             (__v8hi) __B,
+             (__v8hi) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_maskz_add_epi16 (__mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_paddw128_mask ((__v8hi) __A,
+             (__v8hi) __B,
+             (__v8hi)
+             _mm_setzero_si128 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_mask_sub_epi8 (__m128i __W, __mmask16 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_psubb128_mask ((__v16qi) __A,
+             (__v16qi) __B,
+             (__v16qi) __W,
+             (__mmask16) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_maskz_sub_epi8 (__mmask16 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_psubb128_mask ((__v16qi) __A,
+             (__v16qi) __B,
+             (__v16qi)
+             _mm_setzero_si128 (),
+             (__mmask16) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_mask_sub_epi16 (__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_psubw128_mask ((__v8hi) __A,
+             (__v8hi) __B,
+             (__v8hi) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_maskz_sub_epi16 (__mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_psubw128_mask ((__v8hi) __A,
+             (__v8hi) __B,
+             (__v8hi)
+             _mm_setzero_si128 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_mullo_epi16 (__m256i __W, __mmask16 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_pmullw256_mask ((__v16hi) __A,
+              (__v16hi) __B,
+              (__v16hi) __W,
+              (__mmask16) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_mullo_epi16 (__mmask16 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_pmullw256_mask ((__v16hi) __A,
+              (__v16hi) __B,
+              (__v16hi)
+              _mm256_setzero_si256 (),
+              (__mmask16) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_mullo_epi16 (__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_pmullw128_mask ((__v8hi) __A,
+              (__v8hi) __B,
+              (__v8hi) __W,
+              (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_mullo_epi16 (__mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_pmullw128_mask ((__v8hi) __A,
+              (__v8hi) __B,
+              (__v8hi)
+              _mm_setzero_si128 (),
+              (__mmask8) __U);
+}
+#define _mm_cmp_epi8_mask(a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpb128_mask((__v16qi)(__m128i)(a), \
+                                         (__v16qi)(__m128i)(b), \
+                                         (p), (__mmask16)-1); })
+
+#define _mm_mask_cmp_epi8_mask(m, a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpb128_mask((__v16qi)(__m128i)(a), \
+                                         (__v16qi)(__m128i)(b), \
+                                         (p), (__mmask16)(m)); })
+
+#define _mm_cmp_epu8_mask(a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)(__m128i)(a), \
+                                          (__v16qi)(__m128i)(b), \
+                                          (p), (__mmask16)-1); })
+
+#define _mm_mask_cmp_epu8_mask(m, a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_ucmpb128_mask((__v16qi)(__m128i)(a), \
+                                          (__v16qi)(__m128i)(b), \
+                                          (p), (__mmask16)(m)); })
+
+#define _mm256_cmp_epi8_mask(a, b, p) __extension__ ({ \
+  (__mmask32)__builtin_ia32_cmpb256_mask((__v32qi)(__m256i)(a), \
+                                         (__v32qi)(__m256i)(b), \
+                                         (p), (__mmask32)-1); })
+
+#define _mm256_mask_cmp_epi8_mask(m, a, b, p) __extension__ ({ \
+  (__mmask32)__builtin_ia32_cmpb256_mask((__v32qi)(__m256i)(a), \
+                                         (__v32qi)(__m256i)(b), \
+                                         (p), (__mmask32)(m)); })
+
+#define _mm256_cmp_epu8_mask(a, b, p) __extension__ ({ \
+  (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)(__m256i)(a), \
+                                          (__v32qi)(__m256i)(b), \
+                                          (p), (__mmask32)-1); })
+
+#define _mm256_mask_cmp_epu8_mask(m, a, b, p) __extension__ ({ \
+  (__mmask32)__builtin_ia32_ucmpb256_mask((__v32qi)(__m256i)(a), \
+                                          (__v32qi)(__m256i)(b), \
+                                          (p), (__mmask32)(m)); })
+
+#define _mm_cmp_epi16_mask(a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpw128_mask((__v8hi)(__m128i)(a), \
+                                        (__v8hi)(__m128i)(b), \
+                                        (p), (__mmask8)-1); })
+
+#define _mm_mask_cmp_epi16_mask(m, a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpw128_mask((__v8hi)(__m128i)(a), \
+                                        (__v8hi)(__m128i)(b), \
+                                        (p), (__mmask8)(m)); })
+
+#define _mm_cmp_epu16_mask(a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)(__m128i)(a), \
+                                         (__v8hi)(__m128i)(b), \
+                                         (p), (__mmask8)-1); })
+
+#define _mm_mask_cmp_epu16_mask(m, a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_ucmpw128_mask((__v8hi)(__m128i)(a), \
+                                         (__v8hi)(__m128i)(b), \
+                                         (p), (__mmask8)(m)); })
+
+#define _mm256_cmp_epi16_mask(a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpw256_mask((__v16hi)(__m256i)(a), \
+                                         (__v16hi)(__m256i)(b), \
+                                         (p), (__mmask16)-1); })
+
+#define _mm256_mask_cmp_epi16_mask(m, a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_cmpw256_mask((__v16hi)(__m256i)(a), \
+                                         (__v16hi)(__m256i)(b), \
+                                         (p), (__mmask16)(m)); })
+
+#define _mm256_cmp_epu16_mask(a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)(__m256i)(a), \
+                                          (__v16hi)(__m256i)(b), \
+                                          (p), (__mmask16)-1); })
+
+#define _mm256_mask_cmp_epu16_mask(m, a, b, p) __extension__ ({ \
+  (__mmask16)__builtin_ia32_ucmpw256_mask((__v16hi)(__m256i)(a), \
+                                          (__v16hi)(__m256i)(b), \
+                                          (p), (__mmask16)(m)); })
+
 #endif /* __AVX512VLBWINTRIN_H */
diff --git a/lib/Headers/avx512vldqintrin.h b/lib/Headers/avx512vldqintrin.h
new file mode 100644
index 000000000000..4024446a3a6a
--- /dev/null
+++ b/lib/Headers/avx512vldqintrin.h
@@ -0,0 +1,349 @@
+/*===---- avx512vldqintrin.h - AVX512VL and AVX512DQ intrinsics ---------------------------===
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ *===-----------------------------------------------------------------------===
+ */
+ 
+#ifndef __IMMINTRIN_H
+#error "Never use <avx512vldqintrin.h> directly; include <immintrin.h> instead."
+#endif
+
+#ifndef __AVX512VLDQINTRIN_H
+#define __AVX512VLDQINTRIN_H
+
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mullo_epi64 (__m256i __A, __m256i __B) {
+  return (__m256i) ((__v4di) __A * (__v4di) __B);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_mullo_epi64 (__m256i __W, __mmask8 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_pmullq256_mask ((__v4di) __A,
+              (__v4di) __B,
+              (__v4di) __W,
+              (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_mullo_epi64 (__mmask8 __U, __m256i __A, __m256i __B) {
+  return (__m256i) __builtin_ia32_pmullq256_mask ((__v4di) __A,
+              (__v4di) __B,
+              (__v4di)
+              _mm256_setzero_si256 (),
+              (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mullo_epi64 (__m128i __A, __m128i __B) {
+  return (__m128i) ((__v2di) __A * (__v2di) __B);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_mullo_epi64 (__m128i __W, __mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_pmullq128_mask ((__v2di) __A,
+              (__v2di) __B,
+              (__v2di) __W,
+              (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_mullo_epi64 (__mmask8 __U, __m128i __A, __m128i __B) {
+  return (__m128i) __builtin_ia32_pmullq128_mask ((__v2di) __A,
+              (__v2di) __B,
+              (__v2di)
+              _mm_setzero_si128 (),
+              (__mmask8) __U);
+}
+
+static __inline__ __m256d __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_andnot_pd (__m256d __W, __mmask8 __U, __m256d __A, __m256d __B) {
+  return (__m256d) __builtin_ia32_andnpd256_mask ((__v4df) __A,
+              (__v4df) __B,
+              (__v4df) __W,
+              (__mmask8) __U);
+}
+
+static __inline__ __m256d __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_andnot_pd (__mmask8 __U, __m256d __A, __m256d __B) {
+  return (__m256d) __builtin_ia32_andnpd256_mask ((__v4df) __A,
+              (__v4df) __B,
+              (__v4df)
+              _mm256_setzero_pd (),
+              (__mmask8) __U);
+}
+
+static __inline__ __m128d __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_andnot_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) {
+  return (__m128d) __builtin_ia32_andnpd128_mask ((__v2df) __A,
+              (__v2df) __B,
+              (__v2df) __W,
+              (__mmask8) __U);
+}
+
+static __inline__ __m128d __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_andnot_pd (__mmask8 __U, __m128d __A, __m128d __B) {
+  return (__m128d) __builtin_ia32_andnpd128_mask ((__v2df) __A,
+              (__v2df) __B,
+              (__v2df)
+              _mm_setzero_pd (),
+              (__mmask8) __U);
+}
+
+static __inline__ __m256 __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_andnot_ps (__m256 __W, __mmask8 __U, __m256 __A, __m256 __B) {
+  return (__m256) __builtin_ia32_andnps256_mask ((__v8sf) __A,
+             (__v8sf) __B,
+             (__v8sf) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m256 __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_andnot_ps (__mmask8 __U, __m256 __A, __m256 __B) {
+  return (__m256) __builtin_ia32_andnps256_mask ((__v8sf) __A,
+             (__v8sf) __B,
+             (__v8sf)
+             _mm256_setzero_ps (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128 __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_andnot_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) {
+  return (__m128) __builtin_ia32_andnps128_mask ((__v4sf) __A,
+             (__v4sf) __B,
+             (__v4sf) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128 __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_andnot_ps (__mmask8 __U, __m128 __A, __m128 __B) {
+  return (__m128) __builtin_ia32_andnps128_mask ((__v4sf) __A,
+             (__v4sf) __B,
+             (__v4sf)
+             _mm_setzero_ps (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m256d __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_and_pd (__m256d __W, __mmask8 __U, __m256d __A, __m256d __B) {
+  return (__m256d) __builtin_ia32_andpd256_mask ((__v4df) __A,
+             (__v4df) __B,
+             (__v4df) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m256d __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_and_pd (__mmask8 __U, __m256d __A, __m256d __B) {
+  return (__m256d) __builtin_ia32_andpd256_mask ((__v4df) __A,
+             (__v4df) __B,
+             (__v4df)
+             _mm256_setzero_pd (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128d __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_and_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) {
+  return (__m128d) __builtin_ia32_andpd128_mask ((__v2df) __A,
+             (__v2df) __B,
+             (__v2df) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128d __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_and_pd (__mmask8 __U, __m128d __A, __m128d __B) {
+  return (__m128d) __builtin_ia32_andpd128_mask ((__v2df) __A,
+             (__v2df) __B,
+             (__v2df)
+             _mm_setzero_pd (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m256 __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_and_ps (__m256 __W, __mmask8 __U, __m256 __A, __m256 __B) {
+  return (__m256) __builtin_ia32_andps256_mask ((__v8sf) __A,
+            (__v8sf) __B,
+            (__v8sf) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m256 __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_and_ps (__mmask8 __U, __m256 __A, __m256 __B) {
+  return (__m256) __builtin_ia32_andps256_mask ((__v8sf) __A,
+            (__v8sf) __B,
+            (__v8sf)
+            _mm256_setzero_ps (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m128 __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_and_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) {
+  return (__m128) __builtin_ia32_andps128_mask ((__v4sf) __A,
+            (__v4sf) __B,
+            (__v4sf) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m128 __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_and_ps (__mmask8 __U, __m128 __A, __m128 __B) {
+  return (__m128) __builtin_ia32_andps128_mask ((__v4sf) __A,
+            (__v4sf) __B,
+            (__v4sf)
+            _mm_setzero_ps (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m256d __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_xor_pd (__m256d __W, __mmask8 __U, __m256d __A,
+        __m256d __B) {
+  return (__m256d) __builtin_ia32_xorpd256_mask ((__v4df) __A,
+             (__v4df) __B,
+             (__v4df) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m256d __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_xor_pd (__mmask8 __U, __m256d __A, __m256d __B) {
+  return (__m256d) __builtin_ia32_xorpd256_mask ((__v4df) __A,
+             (__v4df) __B,
+             (__v4df)
+             _mm256_setzero_pd (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128d __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_xor_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) {
+  return (__m128d) __builtin_ia32_xorpd128_mask ((__v2df) __A,
+             (__v2df) __B,
+             (__v2df) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128d __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_xor_pd (__mmask8 __U, __m128d __A, __m128d __B) {
+  return (__m128d) __builtin_ia32_xorpd128_mask ((__v2df) __A,
+             (__v2df) __B,
+             (__v2df)
+             _mm_setzero_pd (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m256 __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_xor_ps (__m256 __W, __mmask8 __U, __m256 __A, __m256 __B) {
+  return (__m256) __builtin_ia32_xorps256_mask ((__v8sf) __A,
+            (__v8sf) __B,
+            (__v8sf) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m256 __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_xor_ps (__mmask8 __U, __m256 __A, __m256 __B) {
+  return (__m256) __builtin_ia32_xorps256_mask ((__v8sf) __A,
+            (__v8sf) __B,
+            (__v8sf)
+            _mm256_setzero_ps (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m128 __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_xor_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) {
+  return (__m128) __builtin_ia32_xorps128_mask ((__v4sf) __A,
+            (__v4sf) __B,
+            (__v4sf) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m128 __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_xor_ps (__mmask8 __U, __m128 __A, __m128 __B) {
+  return (__m128) __builtin_ia32_xorps128_mask ((__v4sf) __A,
+            (__v4sf) __B,
+            (__v4sf)
+            _mm_setzero_ps (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m256d __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_or_pd (__m256d __W, __mmask8 __U, __m256d __A, __m256d __B) {
+  return (__m256d) __builtin_ia32_orpd256_mask ((__v4df) __A,
+            (__v4df) __B,
+            (__v4df) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m256d __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_or_pd (__mmask8 __U, __m256d __A, __m256d __B) {
+  return (__m256d) __builtin_ia32_orpd256_mask ((__v4df) __A,
+            (__v4df) __B,
+            (__v4df)
+            _mm256_setzero_pd (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m128d __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_or_pd (__m128d __W, __mmask8 __U, __m128d __A, __m128d __B) {
+  return (__m128d) __builtin_ia32_orpd128_mask ((__v2df) __A,
+            (__v2df) __B,
+            (__v2df) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m128d __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_or_pd (__mmask8 __U, __m128d __A, __m128d __B) {
+  return (__m128d) __builtin_ia32_orpd128_mask ((__v2df) __A,
+            (__v2df) __B,
+            (__v2df)
+            _mm_setzero_pd (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m256 __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_or_ps (__m256 __W, __mmask8 __U, __m256 __A, __m256 __B) {
+  return (__m256) __builtin_ia32_orps256_mask ((__v8sf) __A,
+                 (__v8sf) __B,
+                 (__v8sf) __W,
+                 (__mmask8) __U);
+}
+
+static __inline__ __m256 __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_or_ps (__mmask8 __U, __m256 __A, __m256 __B) {
+  return (__m256) __builtin_ia32_orps256_mask ((__v8sf) __A,
+                 (__v8sf) __B,
+                 (__v8sf)
+                 _mm256_setzero_ps (),
+                 (__mmask8) __U);
+}
+
+static __inline__ __m128 __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_or_ps (__m128 __W, __mmask8 __U, __m128 __A, __m128 __B) {
+  return (__m128) __builtin_ia32_orps128_mask ((__v4sf) __A,
+                 (__v4sf) __B,
+                 (__v4sf) __W,
+                 (__mmask8) __U);
+}
+
+static __inline__ __m128 __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_or_ps (__mmask8 __U, __m128 __A, __m128 __B) {
+  return (__m128) __builtin_ia32_orps128_mask ((__v4sf) __A,
+                 (__v4sf) __B,
+                 (__v4sf)
+                 _mm_setzero_ps (),
+                 (__mmask8) __U);
+}
+
+#endif
diff --git a/lib/Headers/avx512vlintrin.h b/lib/Headers/avx512vlintrin.h
index 8a374b102676..9de0cf418b78 100644
--- a/lib/Headers/avx512vlintrin.h
+++ b/lib/Headers/avx512vlintrin.h
@@ -42,6 +42,17 @@ _mm_mask_cmpeq_epi32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
                                                   __u);
 }
 
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpeq_epu32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 0,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpeq_epu32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 0,
+                                                __u);
+}
 
 static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
 _mm256_cmpeq_epi32_mask(__m256i __a, __m256i __b) {
@@ -56,6 +67,18 @@ _mm256_mask_cmpeq_epi32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
 }
 
 static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpeq_epu32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 0,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpeq_epu32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 0,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
 _mm_cmpeq_epi64_mask(__m128i __a, __m128i __b) {
   return (__mmask8)__builtin_ia32_pcmpeqq128_mask((__v2di)__a, (__v2di)__b,
                                                   (__mmask8)-1);
@@ -67,6 +90,17 @@ _mm_mask_cmpeq_epi64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
                                                   __u);
 }
 
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpeq_epu64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 0,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpeq_epu64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 0,
+                                                __u);
+}
 
 static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
 _mm256_cmpeq_epi64_mask(__m256i __a, __m256i __b) {
@@ -80,4 +114,1206 @@ _mm256_mask_cmpeq_epi64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
                                                   __u);
 }
 
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpeq_epu64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 0,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpeq_epu64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 0,
+                                                __u);
+}
+
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpge_epi32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpd128_mask((__v4si)__a, (__v4si)__b, 5,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpge_epi32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpd128_mask((__v4si)__a, (__v4si)__b, 5,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpge_epu32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 5,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpge_epu32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpge_epi32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpd256_mask((__v8si)__a, (__v8si)__b, 5,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpge_epi32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpd256_mask((__v8si)__a, (__v8si)__b, 5,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpge_epu32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 5,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpge_epu32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpge_epi64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpq128_mask((__v2di)__a, (__v2di)__b, 5,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpge_epi64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpq128_mask((__v2di)__a, (__v2di)__b, 5,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpge_epu64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 5,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpge_epu64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 5,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpge_epi64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpq256_mask((__v4di)__a, (__v4di)__b, 5,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpge_epi64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpq256_mask((__v4di)__a, (__v4di)__b, 5,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpge_epu64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 5,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpge_epu64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 5,
+                                                __u);
+}
+
+
+
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpgt_epi32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtd128_mask((__v4si)__a, (__v4si)__b,
+                                                  (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpgt_epi32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtd128_mask((__v4si)__a, (__v4si)__b,
+                                                  __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpgt_epu32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 6,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpgt_epu32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 6,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epi32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtd256_mask((__v8si)__a, (__v8si)__b,
+                                                  (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpgt_epi32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtd256_mask((__v8si)__a, (__v8si)__b,
+                                                  __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epu32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 6,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpgt_epu32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 6,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpgt_epi64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtq128_mask((__v2di)__a, (__v2di)__b,
+                                                  (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpgt_epi64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtq128_mask((__v2di)__a, (__v2di)__b,
+                                                  __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpgt_epu64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 6,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpgt_epu64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 6,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epi64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtq256_mask((__v4di)__a, (__v4di)__b,
+                                                  (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpgt_epi64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_pcmpgtq256_mask((__v4di)__a, (__v4di)__b,
+                                                  __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpgt_epu64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 6,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpgt_epu64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 6,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmple_epi32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpd128_mask((__v4si)__a, (__v4si)__b, 2,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmple_epi32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpd128_mask((__v4si)__a, (__v4si)__b, 2,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmple_epu32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 2,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmple_epu32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmple_epi32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpd256_mask((__v8si)__a, (__v8si)__b, 2,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmple_epi32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpd256_mask((__v8si)__a, (__v8si)__b, 2,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmple_epu32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 2,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmple_epu32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmple_epi64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpq128_mask((__v2di)__a, (__v2di)__b, 2,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmple_epi64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpq128_mask((__v2di)__a, (__v2di)__b, 2,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmple_epu64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 2,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmple_epu64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmple_epi64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpq256_mask((__v4di)__a, (__v4di)__b, 2,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmple_epi64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpq256_mask((__v4di)__a, (__v4di)__b, 2,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmple_epu64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 2,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmple_epu64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 2,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmplt_epi32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpd128_mask((__v4si)__a, (__v4si)__b, 1,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmplt_epi32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpd128_mask((__v4si)__a, (__v4si)__b, 1,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmplt_epu32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 1,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmplt_epu32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmplt_epi32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpd256_mask((__v8si)__a, (__v8si)__b, 1,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmplt_epi32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpd256_mask((__v8si)__a, (__v8si)__b, 1,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmplt_epu32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 1,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmplt_epu32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmplt_epi64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpq128_mask((__v2di)__a, (__v2di)__b, 1,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmplt_epi64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpq128_mask((__v2di)__a, (__v2di)__b, 1,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmplt_epu64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 1,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmplt_epu64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmplt_epi64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpq256_mask((__v4di)__a, (__v4di)__b, 1,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmplt_epi64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpq256_mask((__v4di)__a, (__v4di)__b, 1,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmplt_epu64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 1,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmplt_epu64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 1,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpneq_epi32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpd128_mask((__v4si)__a, (__v4si)__b, 4,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpneq_epi32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpd128_mask((__v4si)__a, (__v4si)__b, 4,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpneq_epu32_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 4,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpneq_epu32_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)__a, (__v4si)__b, 4,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpneq_epi32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpd256_mask((__v8si)__a, (__v8si)__b, 4,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpneq_epi32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpd256_mask((__v8si)__a, (__v8si)__b, 4,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpneq_epu32_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 4,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpneq_epu32_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)__a, (__v8si)__b, 4,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpneq_epi64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpq128_mask((__v2di)__a, (__v2di)__b, 4,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpneq_epi64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_cmpq128_mask((__v2di)__a, (__v2di)__b, 4,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_cmpneq_epu64_mask(__m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 4,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm_mask_cmpneq_epu64_mask(__mmask8 __u, __m128i __a, __m128i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)__a, (__v2di)__b, 4,
+                                                __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpneq_epi64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpq256_mask((__v4di)__a, (__v4di)__b, 4,
+                                               (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpneq_epi64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_cmpq256_mask((__v4di)__a, (__v4di)__b, 4,
+                                               __u);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_cmpneq_epu64_mask(__m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 4,
+                                                (__mmask8)-1);
+}
+
+static __inline__ __mmask8 __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_cmpneq_epu64_mask(__mmask8 __u, __m256i __a, __m256i __b) {
+  return (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)__a, (__v4di)__b, 4,
+                                                __u);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_add_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
+           __m256i __B)
+{
+  return (__m256i) __builtin_ia32_paddd256_mask ((__v8si) __A,
+             (__v8si) __B,
+             (__v8si) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_maskz_add_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_paddd256_mask ((__v8si) __A,
+             (__v8si) __B,
+             (__v8si)
+             _mm256_setzero_si256 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_add_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
+           __m256i __B)
+{
+  return (__m256i) __builtin_ia32_paddq256_mask ((__v4di) __A,
+             (__v4di) __B,
+             (__v4di) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_maskz_add_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_paddq256_mask ((__v4di) __A,
+             (__v4di) __B,
+             (__v4di)
+             _mm256_setzero_si256 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_sub_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
+           __m256i __B)
+{
+  return (__m256i) __builtin_ia32_psubd256_mask ((__v8si) __A,
+             (__v8si) __B,
+             (__v8si) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_maskz_sub_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_psubd256_mask ((__v8si) __A,
+             (__v8si) __B,
+             (__v8si)
+             _mm256_setzero_si256 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_sub_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
+           __m256i __B)
+{
+  return (__m256i) __builtin_ia32_psubq256_mask ((__v4di) __A,
+             (__v4di) __B,
+             (__v4di) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_maskz_sub_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_psubq256_mask ((__v4di) __A,
+             (__v4di) __B,
+             (__v4di)
+             _mm256_setzero_si256 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_mask_add_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
+        __m128i __B)
+{
+  return (__m128i) __builtin_ia32_paddd128_mask ((__v4si) __A,
+             (__v4si) __B,
+             (__v4si) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_maskz_add_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_paddd128_mask ((__v4si) __A,
+             (__v4si) __B,
+             (__v4si)
+             _mm_setzero_si128 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_mask_add_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
+        __m128i __B)
+{
+  return (__m128i) __builtin_ia32_paddq128_mask ((__v2di) __A,
+             (__v2di) __B,
+             (__v2di) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_maskz_add_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_paddq128_mask ((__v2di) __A,
+             (__v2di) __B,
+             (__v2di)
+             _mm_setzero_si128 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_mask_sub_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
+        __m128i __B)
+{
+  return (__m128i) __builtin_ia32_psubd128_mask ((__v4si) __A,
+             (__v4si) __B,
+             (__v4si) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_maskz_sub_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_psubd128_mask ((__v4si) __A,
+             (__v4si) __B,
+             (__v4si)
+             _mm_setzero_si128 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_mask_sub_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
+        __m128i __B)
+{
+  return (__m128i) __builtin_ia32_psubq128_mask ((__v2di) __A,
+             (__v2di) __B,
+             (__v2di) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_maskz_sub_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_psubq128_mask ((__v2di) __A,
+             (__v2di) __B,
+             (__v2di)
+             _mm_setzero_si128 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_mul_epi32 (__m256i __W, __mmask8 __M, __m256i __X,
+           __m256i __Y)
+{
+  return (__m256i) __builtin_ia32_pmuldq256_mask ((__v8si) __X,
+              (__v8si) __Y,
+              (__v4di) __W, __M);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_maskz_mul_epi32 (__mmask8 __M, __m256i __X, __m256i __Y)
+{
+  return (__m256i) __builtin_ia32_pmuldq256_mask ((__v8si) __X,
+              (__v8si) __Y,
+              (__v4di)
+              _mm256_setzero_si256 (),
+              __M);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_mask_mul_epi32 (__m128i __W, __mmask8 __M, __m128i __X,
+        __m128i __Y)
+{
+  return (__m128i) __builtin_ia32_pmuldq128_mask ((__v4si) __X,
+              (__v4si) __Y,
+              (__v2di) __W, __M);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_maskz_mul_epi32 (__mmask8 __M, __m128i __X, __m128i __Y)
+{
+  return (__m128i) __builtin_ia32_pmuldq128_mask ((__v4si) __X,
+              (__v4si) __Y,
+              (__v2di)
+              _mm_setzero_si128 (),
+              __M);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_mask_mul_epu32 (__m256i __W, __mmask8 __M, __m256i __X,
+           __m256i __Y)
+{
+  return (__m256i) __builtin_ia32_pmuludq256_mask ((__v8si) __X,
+               (__v8si) __Y,
+               (__v4di) __W, __M);
+}
+
+static __inline__ __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_maskz_mul_epu32 (__mmask8 __M, __m256i __X, __m256i __Y)
+{
+  return (__m256i) __builtin_ia32_pmuludq256_mask ((__v8si) __X,
+               (__v8si) __Y,
+               (__v4di)
+               _mm256_setzero_si256 (),
+               __M);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_mask_mul_epu32 (__m128i __W, __mmask8 __M, __m128i __X,
+        __m128i __Y)
+{
+  return (__m128i) __builtin_ia32_pmuludq128_mask ((__v4si) __X,
+               (__v4si) __Y,
+               (__v2di) __W, __M);
+}
+
+static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
+_mm_maskz_mul_epu32 (__mmask8 __M, __m128i __X, __m128i __Y)
+{
+  return (__m128i) __builtin_ia32_pmuludq128_mask ((__v4si) __X,
+               (__v4si) __Y,
+               (__v2di)
+               _mm_setzero_si128 (),
+               __M);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_mullo_epi32 (__mmask8 __M, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pmulld256_mask ((__v8si) __A,
+              (__v8si) __B,
+              (__v8si)
+              _mm256_setzero_si256 (),
+              __M);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_mullo_epi32 (__m256i __W, __mmask8 __M, __m256i __A,
+       __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pmulld256_mask ((__v8si) __A,
+              (__v8si) __B,
+              (__v8si) __W, __M);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_mullo_epi32 (__mmask8 __M, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pmulld128_mask ((__v4si) __A,
+              (__v4si) __B,
+              (__v4si)
+              _mm_setzero_si128 (),
+              __M);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_mullo_epi32 (__m128i __W, __mmask16 __M, __m128i __A,
+          __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pmulld128_mask ((__v4si) __A,
+              (__v4si) __B,
+              (__v4si) __W, __M);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_and_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
+           __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pandd256_mask ((__v8si) __A,
+             (__v8si) __B,
+             (__v8si) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_and_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pandd256_mask ((__v8si) __A,
+             (__v8si) __B,
+             (__v8si)
+             _mm256_setzero_si256 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_and_epi32 (__m128i __W, __mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pandd128_mask ((__v4si) __A,
+             (__v4si) __B,
+             (__v4si) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_and_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pandd128_mask ((__v4si) __A,
+             (__v4si) __B,
+             (__v4si)
+             _mm_setzero_si128 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_andnot_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
+        __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pandnd256_mask ((__v8si) __A,
+              (__v8si) __B,
+              (__v8si) __W,
+              (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_andnot_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pandnd256_mask ((__v8si) __A,
+              (__v8si) __B,
+              (__v8si)
+              _mm256_setzero_si256 (),
+              (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_andnot_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
+           __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pandnd128_mask ((__v4si) __A,
+              (__v4si) __B,
+              (__v4si) __W,
+              (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_andnot_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pandnd128_mask ((__v4si) __A,
+              (__v4si) __B,
+              (__v4si)
+              _mm_setzero_si128 (),
+              (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_or_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
+          __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pord256_mask ((__v8si) __A,
+            (__v8si) __B,
+            (__v8si) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_or_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pord256_mask ((__v8si) __A,
+            (__v8si) __B,
+            (__v8si)
+            _mm256_setzero_si256 (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_or_epi32 (__m128i __W, __mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pord128_mask ((__v4si) __A,
+            (__v4si) __B,
+            (__v4si) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_or_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pord128_mask ((__v4si) __A,
+            (__v4si) __B,
+            (__v4si)
+            _mm_setzero_si128 (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_xor_epi32 (__m256i __W, __mmask8 __U, __m256i __A,
+           __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pxord256_mask ((__v8si) __A,
+             (__v8si) __B,
+             (__v8si) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_xor_epi32 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pxord256_mask ((__v8si) __A,
+             (__v8si) __B,
+             (__v8si)
+             _mm256_setzero_si256 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_xor_epi32 (__m128i __W, __mmask8 __U, __m128i __A,
+        __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pxord128_mask ((__v4si) __A,
+             (__v4si) __B,
+             (__v4si) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_xor_epi32 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pxord128_mask ((__v4si) __A,
+             (__v4si) __B,
+             (__v4si)
+             _mm_setzero_si128 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_and_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
+           __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pandq256_mask ((__v4di) __A,
+             (__v4di) __B,
+             (__v4di) __W, __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_and_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pandq256_mask ((__v4di) __A,
+             (__v4di) __B,
+             (__v4di)
+             _mm256_setzero_pd (),
+             __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_and_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
+        __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pandq128_mask ((__v2di) __A,
+             (__v2di) __B,
+             (__v2di) __W, __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_and_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pandq128_mask ((__v2di) __A,
+             (__v2di) __B,
+             (__v2di)
+             _mm_setzero_pd (),
+             __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_andnot_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
+        __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pandnq256_mask ((__v4di) __A,
+              (__v4di) __B,
+              (__v4di) __W, __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_andnot_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pandnq256_mask ((__v4di) __A,
+              (__v4di) __B,
+              (__v4di)
+              _mm256_setzero_pd (),
+              __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_andnot_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
+           __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pandnq128_mask ((__v2di) __A,
+              (__v2di) __B,
+              (__v2di) __W, __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_andnot_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pandnq128_mask ((__v2di) __A,
+              (__v2di) __B,
+              (__v2di)
+              _mm_setzero_pd (),
+              __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_or_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
+          __m256i __B)
+{
+  return (__m256i) __builtin_ia32_porq256_mask ((__v4di) __A,
+            (__v4di) __B,
+            (__v4di) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_or_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_porq256_mask ((__v4di) __A,
+            (__v4di) __B,
+            (__v4di)
+            _mm256_setzero_si256 (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_or_epi64 (__m128i __W, __mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_porq128_mask ((__v2di) __A,
+            (__v2di) __B,
+            (__v2di) __W,
+            (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_or_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_porq128_mask ((__v2di) __A,
+            (__v2di) __B,
+            (__v2di)
+            _mm_setzero_si128 (),
+            (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_mask_xor_epi64 (__m256i __W, __mmask8 __U, __m256i __A,
+           __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pxorq256_mask ((__v4di) __A,
+             (__v4di) __B,
+             (__v4di) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m256i __attribute__ ((__always_inline__, __nodebug__))
+_mm256_maskz_xor_epi64 (__mmask8 __U, __m256i __A, __m256i __B)
+{
+  return (__m256i) __builtin_ia32_pxorq256_mask ((__v4di) __A,
+             (__v4di) __B,
+             (__v4di)
+             _mm256_setzero_si256 (),
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_mask_xor_epi64 (__m128i __W, __mmask8 __U, __m128i __A,
+        __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pxorq128_mask ((__v2di) __A,
+             (__v2di) __B,
+             (__v2di) __W,
+             (__mmask8) __U);
+}
+
+static __inline__ __m128i __attribute__ ((__always_inline__, __nodebug__))
+_mm_maskz_xor_epi64 (__mmask8 __U, __m128i __A, __m128i __B)
+{
+  return (__m128i) __builtin_ia32_pxorq128_mask ((__v2di) __A,
+             (__v2di) __B,
+             (__v2di)
+             _mm_setzero_si128 (),
+             (__mmask8) __U);
+}
+
+#define _mm_cmp_epi32_mask(a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpd128_mask((__v4si)(__m128i)(a), \
+                                        (__v4si)(__m128i)(b), \
+                                        (p), (__mmask8)-1); })
+
+#define _mm_mask_cmp_epi32_mask(m, a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpd128_mask((__v4si)(__m128i)(a), \
+                                        (__v4si)(__m128i)(b), \
+                                        (p), (__mmask8)(m)); })
+
+#define _mm_cmp_epu32_mask(a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)(__m128i)(a), \
+                                         (__v4si)(__m128i)(b), \
+                                         (p), (__mmask8)-1); })
+
+#define _mm_mask_cmp_epu32_mask(m, a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_ucmpd128_mask((__v4si)(__m128i)(a), \
+                                         (__v4si)(__m128i)(b), \
+                                         (p), (__mmask8)(m)); })
+
+#define _mm256_cmp_epi32_mask(a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpd256_mask((__v8si)(__m256i)(a), \
+                                        (__v8si)(__m256i)(b), \
+                                        (p), (__mmask8)-1); })
+
+#define _mm256_mask_cmp_epi32_mask(m, a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpd256_mask((__v8si)(__m256i)(a), \
+                                        (__v8si)(__m256i)(b), \
+                                        (p), (__mmask8)(m)); })
+
+#define _mm256_cmp_epu32_mask(a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)(__m256i)(a), \
+                                         (__v8si)(__m256i)(b), \
+                                         (p), (__mmask8)-1); })
+
+#define _mm256_mask_cmp_epu32_mask(m, a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_ucmpd256_mask((__v8si)(__m256i)(a), \
+                                         (__v8si)(__m256i)(b), \
+                                         (p), (__mmask8)(m)); })
+
+#define _mm_cmp_epi64_mask(a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpq128_mask((__v2di)(__m128i)(a), \
+                                        (__v2di)(__m128i)(b), \
+                                        (p), (__mmask8)-1); })
+
+#define _mm_mask_cmp_epi64_mask(m, a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpq128_mask((__v2di)(__m128i)(a), \
+                                        (__v2di)(__m128i)(b), \
+                                        (p), (__mmask8)(m)); })
+
+#define _mm_cmp_epu64_mask(a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)(__m128i)(a), \
+                                         (__v2di)(__m128i)(b), \
+                                         (p), (__mmask8)-1); })
+
+#define _mm_mask_cmp_epu64_mask(m, a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_ucmpq128_mask((__v2di)(__m128i)(a), \
+                                         (__v2di)(__m128i)(b), \
+                                         (p), (__mmask8)(m)); })
+
+#define _mm256_cmp_epi64_mask(a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpq256_mask((__v4di)(__m256i)(a), \
+                                        (__v4di)(__m256i)(b), \
+                                        (p), (__mmask8)-1); })
+
+#define _mm256_mask_cmp_epi64_mask(m, a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpq256_mask((__v4di)(__m256i)(a), \
+                                        (__v4di)(__m256i)(b), \
+                                        (p), (__mmask8)(m)); })
+
+#define _mm256_cmp_epu64_mask(a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)(__m256i)(a), \
+                                         (__v4di)(__m256i)(b), \
+                                         (p), (__mmask8)-1); })
+
+#define _mm256_mask_cmp_epu64_mask(m, a, b, p) __extension__ ({ \
+  (__mmask8)__builtin_ia32_ucmpq256_mask((__v4di)(__m256i)(a), \
+                                         (__v4di)(__m256i)(b), \
+                                         (p), (__mmask8)(m)); })
+
+#define _mm256_cmp_ps_mask(a, b, p)  __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpps256_mask((__v8sf)(__m256)(a), \
+                                         (__v8sf)(__m256)(b), \
+                                         (p), (__mmask8)-1); })
+
+#define _mm256_mask_cmp_ps_mask(m, a, b, p)  __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpps256_mask((__v8sf)(__m256)(a), \
+                                         (__v8sf)(__m256)(b), \
+                                         (p), (__mmask8)(m)); })
+
+#define _mm256_cmp_pd_mask(a, b, p)  __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmppd256_mask((__v4df)(__m256)(a), \
+                                         (__v4df)(__m256)(b), \
+                                         (p), (__mmask8)-1); })
+
+#define _mm256_mask_cmp_pd_mask(m, a, b, p)  __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmppd256_mask((__v4df)(__m256)(a), \
+                                         (__v4df)(__m256)(b), \
+                                         (p), (__mmask8)(m)); })
+
+#define _mm128_cmp_ps_mask(a, b, p)  __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpps128_mask((__v4sf)(__m128)(a), \
+                                         (__v4sf)(__m128)(b), \
+                                         (p), (__mmask8)-1); })
+
+#define _mm128_mask_cmp_ps_mask(m, a, b, p)  __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmpps128_mask((__v4sf)(__m128)(a), \
+                                         (__v4sf)(__m128)(b), \
+                                         (p), (__mmask8)(m)); })
+
+#define _mm128_cmp_pd_mask(a, b, p)  __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmppd128_mask((__v2df)(__m128)(a), \
+                                         (__v2df)(__m128)(b), \
+                                         (p), (__mmask8)-1); })
+
+#define _mm128_mask_cmp_pd_mask(m, a, b, p)  __extension__ ({ \
+  (__mmask8)__builtin_ia32_cmppd128_mask((__v2df)(__m128)(a), \
+                                         (__v2df)(__m128)(b), \
+                                         (p), (__mmask8)(m)); })
 #endif /* __AVX512VLINTRIN_H */
diff --git a/lib/Headers/avxintrin.h b/lib/Headers/avxintrin.h
index 4e1044af5611..4907965861d4 100644
--- a/lib/Headers/avxintrin.h
+++ b/lib/Headers/avxintrin.h
@@ -257,8 +257,7 @@ _mm_permutevar_ps(__m128 __a, __m128i __c)
 static __inline __m256 __attribute__((__always_inline__, __nodebug__))
 _mm256_permutevar_ps(__m256 __a, __m256i __c)
 {
-  return (__m256)__builtin_ia32_vpermilvarps256((__v8sf)__a,
-						  (__v8si)__c);
+  return (__m256)__builtin_ia32_vpermilvarps256((__v8sf)__a, (__v8si)__c);
 }
 
 #define _mm_permute_pd(A, C) __extension__ ({ \
@@ -430,35 +429,22 @@ _mm256_blendv_ps(__m256 __a, __m256 __b, __m256 __c)
   __m128 __b = (b); \
   (__m128)__builtin_ia32_cmpss((__v4sf)__a, (__v4sf)__b, (c)); })
 
-/* Vector extract */
-#define _mm256_extractf128_pd(A, O) __extension__ ({ \
-  __m256d __A = (A); \
-  (__m128d)__builtin_ia32_vextractf128_pd256((__v4df)__A, (O)); })
-
-#define _mm256_extractf128_ps(A, O) __extension__ ({ \
-  __m256 __A = (A); \
-  (__m128)__builtin_ia32_vextractf128_ps256((__v8sf)__A, (O)); })
-
-#define _mm256_extractf128_si256(A, O) __extension__ ({ \
-  __m256i __A = (A); \
-  (__m128i)__builtin_ia32_vextractf128_si256((__v8si)__A, (O)); })
-
 static __inline int __attribute__((__always_inline__, __nodebug__))
-_mm256_extract_epi32(__m256i __a, int const __imm)
+_mm256_extract_epi32(__m256i __a, const int __imm)
 {
   __v8si __b = (__v8si)__a;
   return __b[__imm & 7];
 }
 
 static __inline int __attribute__((__always_inline__, __nodebug__))
-_mm256_extract_epi16(__m256i __a, int const __imm)
+_mm256_extract_epi16(__m256i __a, const int __imm)
 {
   __v16hi __b = (__v16hi)__a;
   return __b[__imm & 15];
 }
 
 static __inline int __attribute__((__always_inline__, __nodebug__))
-_mm256_extract_epi8(__m256i __a, int const __imm)
+_mm256_extract_epi8(__m256i __a, const int __imm)
 {
   __v32qi __b = (__v32qi)__a;
   return __b[__imm & 31];
@@ -473,22 +459,6 @@ _mm256_extract_epi64(__m256i __a, const int __imm)
 }
 #endif
 
-/* Vector insert */
-#define _mm256_insertf128_pd(V1, V2, O) __extension__ ({ \
-  __m256d __V1 = (V1); \
-  __m128d __V2 = (V2); \
-  (__m256d)__builtin_ia32_vinsertf128_pd256((__v4df)__V1, (__v2df)__V2, (O)); })
-
-#define _mm256_insertf128_ps(V1, V2, O) __extension__ ({ \
-  __m256 __V1 = (V1); \
-  __m128 __V2 = (V2); \
-  (__m256)__builtin_ia32_vinsertf128_ps256((__v8sf)__V1, (__v4sf)__V2, (O)); })
-
-#define _mm256_insertf128_si256(V1, V2, O) __extension__ ({ \
-  __m256i __V1 = (V1); \
-  __m128i __V2 = (V2); \
-  (__m256i)__builtin_ia32_vinsertf128_si256((__v8si)__V1, (__v4si)__V2, (O)); })
-
 static __inline __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_insert_epi32(__m256i __a, int __b, int const __imm)
 {
@@ -515,7 +485,7 @@ _mm256_insert_epi8(__m256i __a, int __b, int const __imm)
 
 #ifdef __x86_64__
 static __inline __m256i __attribute__((__always_inline__, __nodebug__))
-_mm256_insert_epi64(__m256i __a, int __b, int const __imm)
+_mm256_insert_epi64(__m256i __a, long long __b, int const __imm)
 {
   __v4di __c = (__v4di)__a;
   __c[__imm & 3] = __b;
@@ -785,7 +755,7 @@ _mm256_loadu_pd(double const *__p)
 {
   struct __loadu_pd {
     __m256d __v;
-  } __attribute__((packed, may_alias));
+  } __attribute__((__packed__, __may_alias__));
   return ((struct __loadu_pd*)__p)->__v;
 }
 
@@ -794,7 +764,7 @@ _mm256_loadu_ps(float const *__p)
 {
   struct __loadu_ps {
     __m256 __v;
-  } __attribute__((packed, may_alias));
+  } __attribute__((__packed__, __may_alias__));
   return ((struct __loadu_ps*)__p)->__v;
 }
 
@@ -809,7 +779,7 @@ _mm256_loadu_si256(__m256i const *__p)
 {
   struct __loadu_si256 {
     __m256i __v;
-  } __attribute__((packed, may_alias));
+  } __attribute__((__packed__, __may_alias__));
   return ((struct __loadu_si256*)__p)->__v;
 }
 
@@ -935,23 +905,23 @@ _mm256_set_pd(double __a, double __b, double __c, double __d)
 
 static __inline __m256 __attribute__((__always_inline__, __nodebug__))
 _mm256_set_ps(float __a, float __b, float __c, float __d,
-	            float __e, float __f, float __g, float __h)
+              float __e, float __f, float __g, float __h)
 {
   return (__m256){ __h, __g, __f, __e, __d, __c, __b, __a };
 }
 
 static __inline __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_set_epi32(int __i0, int __i1, int __i2, int __i3,
-		             int __i4, int __i5, int __i6, int __i7)
+                 int __i4, int __i5, int __i6, int __i7)
 {
   return (__m256i)(__v8si){ __i7, __i6, __i5, __i4, __i3, __i2, __i1, __i0 };
 }
 
 static __inline __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_set_epi16(short __w15, short __w14, short __w13, short __w12,
-		             short __w11, short __w10, short __w09, short __w08,
-		             short __w07, short __w06, short __w05, short __w04,
-		             short __w03, short __w02, short __w01, short __w00)
+                 short __w11, short __w10, short __w09, short __w08,
+                 short __w07, short __w06, short __w05, short __w04,
+                 short __w03, short __w02, short __w01, short __w00)
 {
   return (__m256i)(__v16hi){ __w00, __w01, __w02, __w03, __w04, __w05, __w06,
     __w07, __w08, __w09, __w10, __w11, __w12, __w13, __w14, __w15 };
@@ -959,13 +929,13 @@ _mm256_set_epi16(short __w15, short __w14, short __w13, short __w12,
 
 static __inline __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_set_epi8(char __b31, char __b30, char __b29, char __b28,
-		            char __b27, char __b26, char __b25, char __b24,
-		            char __b23, char __b22, char __b21, char __b20,
-		            char __b19, char __b18, char __b17, char __b16,
-		            char __b15, char __b14, char __b13, char __b12,
-		            char __b11, char __b10, char __b09, char __b08,
-		            char __b07, char __b06, char __b05, char __b04,
-		            char __b03, char __b02, char __b01, char __b00)
+                char __b27, char __b26, char __b25, char __b24,
+                char __b23, char __b22, char __b21, char __b20,
+                char __b19, char __b18, char __b17, char __b16,
+                char __b15, char __b14, char __b13, char __b12,
+                char __b11, char __b10, char __b09, char __b08,
+                char __b07, char __b06, char __b05, char __b04,
+                char __b03, char __b02, char __b01, char __b00)
 {
   return (__m256i)(__v32qi){
     __b00, __b01, __b02, __b03, __b04, __b05, __b06, __b07,
@@ -990,23 +960,23 @@ _mm256_setr_pd(double __a, double __b, double __c, double __d)
 
 static __inline __m256 __attribute__((__always_inline__, __nodebug__))
 _mm256_setr_ps(float __a, float __b, float __c, float __d,
-		           float __e, float __f, float __g, float __h)
+               float __e, float __f, float __g, float __h)
 {
   return (__m256){ __a, __b, __c, __d, __e, __f, __g, __h };
 }
 
 static __inline __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_setr_epi32(int __i0, int __i1, int __i2, int __i3,
-		              int __i4, int __i5, int __i6, int __i7)
+                  int __i4, int __i5, int __i6, int __i7)
 {
   return (__m256i)(__v8si){ __i0, __i1, __i2, __i3, __i4, __i5, __i6, __i7 };
 }
 
 static __inline __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_setr_epi16(short __w15, short __w14, short __w13, short __w12,
-		   short __w11, short __w10, short __w09, short __w08,
-		   short __w07, short __w06, short __w05, short __w04,
-		   short __w03, short __w02, short __w01, short __w00)
+       short __w11, short __w10, short __w09, short __w08,
+       short __w07, short __w06, short __w05, short __w04,
+       short __w03, short __w02, short __w01, short __w00)
 {
   return (__m256i)(__v16hi){ __w15, __w14, __w13, __w12, __w11, __w10, __w09,
     __w08, __w07, __w06, __w05, __w04, __w03, __w02, __w01, __w00 };
@@ -1014,19 +984,19 @@ _mm256_setr_epi16(short __w15, short __w14, short __w13, short __w12,
 
 static __inline __m256i __attribute__((__always_inline__, __nodebug__))
 _mm256_setr_epi8(char __b31, char __b30, char __b29, char __b28,
-		             char __b27, char __b26, char __b25, char __b24,
-		             char __b23, char __b22, char __b21, char __b20,
-		             char __b19, char __b18, char __b17, char __b16,
-		             char __b15, char __b14, char __b13, char __b12,
-		             char __b11, char __b10, char __b09, char __b08,
-		             char __b07, char __b06, char __b05, char __b04,
-		             char __b03, char __b02, char __b01, char __b00)
+                 char __b27, char __b26, char __b25, char __b24,
+                 char __b23, char __b22, char __b21, char __b20,
+                 char __b19, char __b18, char __b17, char __b16,
+                 char __b15, char __b14, char __b13, char __b12,
+                 char __b11, char __b10, char __b09, char __b08,
+                 char __b07, char __b06, char __b05, char __b04,
+                 char __b03, char __b02, char __b01, char __b00)
 {
   return (__m256i)(__v32qi){
     __b31, __b30, __b29, __b28, __b27, __b26, __b25, __b24,
-		__b23, __b22, __b21, __b20, __b19, __b18, __b17, __b16,
-		__b15, __b14, __b13, __b12, __b11, __b10, __b09, __b08,
-		__b07, __b06, __b05, __b04, __b03, __b02, __b01, __b00 };
+    __b23, __b22, __b21, __b20, __b19, __b18, __b17, __b16,
+    __b15, __b14, __b13, __b12, __b11, __b10, __b09, __b08,
+    __b07, __b06, __b05, __b04, __b03, __b02, __b01, __b00 };
 }
 
 static __inline __m256i __attribute__((__always_inline__, __nodebug__))
@@ -1167,6 +1137,70 @@ _mm256_castsi128_si256(__m128i __a)
   return __builtin_shufflevector(__a, __a, 0, 1, -1, -1);
 }
 
+/* 
+   Vector insert.
+   We use macros rather than inlines because we only want to accept
+   invocations where the immediate M is a constant expression.
+*/
+#define _mm256_insertf128_ps(V1, V2, M) __extension__ ({ \
+  (__m256)__builtin_shufflevector( \
+    (__v8sf)(V1), \
+    (__v8sf)_mm256_castps128_ps256((__m128)(V2)), \
+    (((M) & 1) ?  0 :  8), \
+    (((M) & 1) ?  1 :  9), \
+    (((M) & 1) ?  2 : 10), \
+    (((M) & 1) ?  3 : 11), \
+    (((M) & 1) ?  8 :  4), \
+    (((M) & 1) ?  9 :  5), \
+    (((M) & 1) ? 10 :  6), \
+    (((M) & 1) ? 11 :  7) );})
+
+#define _mm256_insertf128_pd(V1, V2, M) __extension__ ({ \
+  (__m256d)__builtin_shufflevector( \
+    (__v4df)(V1), \
+    (__v4df)_mm256_castpd128_pd256((__m128d)(V2)), \
+    (((M) & 1) ? 0 : 4), \
+    (((M) & 1) ? 1 : 5), \
+    (((M) & 1) ? 4 : 2), \
+    (((M) & 1) ? 5 : 3) );})
+
+#define _mm256_insertf128_si256(V1, V2, M) __extension__ ({ \
+  (__m256i)__builtin_shufflevector( \
+    (__v4di)(V1), \
+    (__v4di)_mm256_castsi128_si256((__m128i)(V2)), \
+    (((M) & 1) ? 0 : 4), \
+    (((M) & 1) ? 1 : 5), \
+    (((M) & 1) ? 4 : 2), \
+    (((M) & 1) ? 5 : 3) );})
+
+/* 
+   Vector extract.
+   We use macros rather than inlines because we only want to accept
+   invocations where the immediate M is a constant expression.
+*/
+#define _mm256_extractf128_ps(V, M) __extension__ ({ \
+  (__m128)__builtin_shufflevector( \
+    (__v8sf)(V), \
+    (__v8sf)(_mm256_setzero_ps()), \
+    (((M) & 1) ? 4 : 0), \
+    (((M) & 1) ? 5 : 1), \
+    (((M) & 1) ? 6 : 2), \
+    (((M) & 1) ? 7 : 3) );})
+
+#define _mm256_extractf128_pd(V, M) __extension__ ({ \
+  (__m128d)__builtin_shufflevector( \
+    (__v4df)(V), \
+    (__v4df)(_mm256_setzero_pd()), \
+    (((M) & 1) ? 2 : 0), \
+    (((M) & 1) ? 3 : 1) );})
+
+#define _mm256_extractf128_si256(V, M) __extension__ ({ \
+  (__m128i)__builtin_shufflevector( \
+    (__v4di)(V), \
+    (__v4di)(_mm256_setzero_si256()), \
+    (((M) & 1) ? 2 : 0), \
+    (((M) & 1) ? 3 : 1) );})
+
 /* SIMD load ops (unaligned) */
 static __inline __m256 __attribute__((__always_inline__, __nodebug__))
 _mm256_loadu2_m128(float const *__addr_hi, float const *__addr_lo)
@@ -1195,7 +1229,7 @@ _mm256_loadu2_m128i(__m128i const *__addr_hi, __m128i const *__addr_lo)
 {
   struct __loadu_si128 {
     __m128i __v;
-  } __attribute__((packed, may_alias));
+  } __attribute__((__packed__, __may_alias__));
   __m256i __v256 = _mm256_castsi128_si256(
     ((struct __loadu_si128*)__addr_lo)->__v);
   return _mm256_insertf128_si256(__v256,
@@ -1236,4 +1270,34 @@ _mm256_storeu2_m128i(__m128i *__addr_hi, __m128i *__addr_lo, __m256i __a)
   __builtin_ia32_storedqu((char *)__addr_hi, (__v16qi)__v128);
 }
 
+static __inline __m256 __attribute__((__always_inline__, __nodebug__))
+_mm256_set_m128 (__m128 __hi, __m128 __lo) {
+  return (__m256) __builtin_shufflevector(__lo, __hi, 0, 1, 2, 3, 4, 5, 6, 7);
+}
+
+static __inline __m256d __attribute__((__always_inline__, __nodebug__))
+_mm256_set_m128d (__m128d __hi, __m128d __lo) {
+  return (__m256d)_mm256_set_m128((__m128)__hi, (__m128)__lo);
+}
+
+static __inline __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_set_m128i (__m128i __hi, __m128i __lo) {
+  return (__m256i)_mm256_set_m128((__m128)__hi, (__m128)__lo);
+}
+
+static __inline __m256 __attribute__((__always_inline__, __nodebug__))
+_mm256_setr_m128 (__m128 __lo, __m128 __hi) {
+  return _mm256_set_m128(__hi, __lo);
+}
+
+static __inline __m256d __attribute__((__always_inline__, __nodebug__))
+_mm256_setr_m128d (__m128d __lo, __m128d __hi) {
+  return (__m256d)_mm256_set_m128((__m128)__hi, (__m128)__lo);
+}
+
+static __inline __m256i __attribute__((__always_inline__, __nodebug__))
+_mm256_setr_m128i (__m128i __lo, __m128i __hi) {
+  return (__m256i)_mm256_set_m128((__m128)__hi, (__m128)__lo);
+}
+
 #endif /* __AVXINTRIN_H */
diff --git a/lib/Headers/cuda_builtin_vars.h b/lib/Headers/cuda_builtin_vars.h
new file mode 100644
index 000000000000..901356b3d5ce
--- /dev/null
+++ b/lib/Headers/cuda_builtin_vars.h
@@ -0,0 +1,110 @@
+/*===---- cuda_builtin_vars.h - CUDA built-in variables ---------------------===
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __CUDA_BUILTIN_VARS_H
+#define __CUDA_BUILTIN_VARS_H
+
+// The file implements built-in CUDA variables using __declspec(property).
+// https://msdn.microsoft.com/en-us/library/yhfk0thd.aspx
+// All read accesses of built-in variable fields get converted into calls to a
+// getter function which in turn would call appropriate builtin to fetch the
+// value.
+//
+// Example:
+//    int x = threadIdx.x;
+// IR output:
+//  %0 = call i32 @llvm.ptx.read.tid.x() #3
+// PTX output:
+//  mov.u32     %r2, %tid.x;
+
+#define __CUDA_DEVICE_BUILTIN(FIELD, INTRINSIC)                                \
+  __declspec(property(get = __fetch_builtin_##FIELD)) unsigned int FIELD;      \
+  static inline __attribute__((always_inline))                                 \
+      __attribute__((device)) unsigned int __fetch_builtin_##FIELD(void) {     \
+    return INTRINSIC;                                                          \
+  }
+
+#if __cplusplus >= 201103L
+#define __DELETE =delete
+#else
+#define __DELETE
+#endif
+
+// Make sure nobody can create instances of the special varible types.  nvcc
+// also disallows taking address of special variables, so we disable address-of
+// operator as well.
+#define __CUDA_DISALLOW_BUILTINVAR_ACCESS(TypeName)                            \
+  __attribute__((device)) TypeName() __DELETE;                                 \
+  __attribute__((device)) TypeName(const TypeName &) __DELETE;                 \
+  __attribute__((device)) void operator=(const TypeName &) const __DELETE;     \
+  __attribute__((device)) TypeName *operator&() const __DELETE
+
+struct __cuda_builtin_threadIdx_t {
+  __CUDA_DEVICE_BUILTIN(x,__builtin_ptx_read_tid_x());
+  __CUDA_DEVICE_BUILTIN(y,__builtin_ptx_read_tid_y());
+  __CUDA_DEVICE_BUILTIN(z,__builtin_ptx_read_tid_z());
+private:
+  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_threadIdx_t);
+};
+
+struct __cuda_builtin_blockIdx_t {
+  __CUDA_DEVICE_BUILTIN(x,__builtin_ptx_read_ctaid_x());
+  __CUDA_DEVICE_BUILTIN(y,__builtin_ptx_read_ctaid_y());
+  __CUDA_DEVICE_BUILTIN(z,__builtin_ptx_read_ctaid_z());
+private:
+  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_blockIdx_t);
+};
+
+struct __cuda_builtin_blockDim_t {
+  __CUDA_DEVICE_BUILTIN(x,__builtin_ptx_read_ntid_x());
+  __CUDA_DEVICE_BUILTIN(y,__builtin_ptx_read_ntid_y());
+  __CUDA_DEVICE_BUILTIN(z,__builtin_ptx_read_ntid_z());
+private:
+  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_blockDim_t);
+};
+
+struct __cuda_builtin_gridDim_t {
+  __CUDA_DEVICE_BUILTIN(x,__builtin_ptx_read_nctaid_x());
+  __CUDA_DEVICE_BUILTIN(y,__builtin_ptx_read_nctaid_y());
+  __CUDA_DEVICE_BUILTIN(z,__builtin_ptx_read_nctaid_z());
+private:
+  __CUDA_DISALLOW_BUILTINVAR_ACCESS(__cuda_builtin_gridDim_t);
+};
+
+#define __CUDA_BUILTIN_VAR                                                     \
+  extern const __attribute__((device)) __attribute__((weak))
+__CUDA_BUILTIN_VAR __cuda_builtin_threadIdx_t threadIdx;
+__CUDA_BUILTIN_VAR __cuda_builtin_blockIdx_t blockIdx;
+__CUDA_BUILTIN_VAR __cuda_builtin_blockDim_t blockDim;
+__CUDA_BUILTIN_VAR __cuda_builtin_gridDim_t gridDim;
+
+// warpSize should translate to read of %WARP_SZ but there's currently no
+// builtin to do so. According to PTX v4.2 docs 'to date, all target
+// architectures have a WARP_SZ value of 32'.
+__attribute__((device)) const int warpSize = 32;
+
+#undef __CUDA_DEVICE_BUILTIN
+#undef __CUDA_BUILTIN_VAR
+#undef __CUDA_DISALLOW_BUILTINVAR_ACCESS
+
+#endif /* __CUDA_BUILTIN_VARS_H */
diff --git a/lib/Headers/emmintrin.h b/lib/Headers/emmintrin.h
index 28d004309cf4..c764d68a185d 100644
--- a/lib/Headers/emmintrin.h
+++ b/lib/Headers/emmintrin.h
@@ -489,7 +489,7 @@ _mm_loadu_pd(double const *__dp)
 {
   struct __loadu_pd {
     __m128d __v;
-  } __attribute__((packed, may_alias));
+  } __attribute__((__packed__, __may_alias__));
   return ((struct __loadu_pd*)__dp)->__v;
 }
 
@@ -825,11 +825,28 @@ _mm_xor_si128(__m128i __a, __m128i __b)
   return __a ^ __b;
 }
 
-#define _mm_slli_si128(a, count) __extension__ ({ \
-  _Pragma("clang diagnostic push") _Pragma("clang diagnostic ignored \"-Wshadow\""); \
-  __m128i __a = (a); \
-   _Pragma("clang diagnostic pop"); \
-  (__m128i)__builtin_ia32_pslldqi128(__a, (count)*8); })
+#define _mm_slli_si128(a, imm) __extension__ ({                         \
+  (__m128i)__builtin_shufflevector((__v16qi)_mm_setzero_si128(),        \
+                                   (__v16qi)(__m128i)(a),               \
+                                   ((imm)&0xF0) ? 0 : 16 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 17 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 18 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 19 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 20 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 21 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 22 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 23 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 24 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 25 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 26 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 27 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 28 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 29 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 30 - ((imm)&0xF), \
+                                   ((imm)&0xF0) ? 0 : 31 - ((imm)&0xF)); })
+
+#define _mm_bslli_si128(a, imm) \
+  _mm_slli_si128((a), (imm))
 
 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_slli_epi16(__m128i __a, int __count)
@@ -891,12 +908,28 @@ _mm_sra_epi32(__m128i __a, __m128i __count)
   return (__m128i)__builtin_ia32_psrad128((__v4si)__a, (__v4si)__count);
 }
 
-
-#define _mm_srli_si128(a, count) __extension__ ({ \
-  _Pragma("clang diagnostic push") _Pragma("clang diagnostic ignored \"-Wshadow\""); \
-  __m128i __a = (a); \
-  _Pragma("clang diagnostic pop"); \
-  (__m128i)__builtin_ia32_psrldqi128(__a, (count)*8); })
+#define _mm_srli_si128(a, imm) __extension__ ({                          \
+  (__m128i)__builtin_shufflevector((__v16qi)(__m128i)(a),                \
+                                   (__v16qi)_mm_setzero_si128(),         \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 0,  \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 1,  \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 2,  \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 3,  \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 4,  \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 5,  \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 6,  \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 7,  \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 8,  \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 9,  \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 10, \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 11, \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 12, \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 13, \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 14, \
+                                   ((imm)&0xF0) ? 16 : ((imm)&0xF) + 15); })
+
+#define _mm_bsrli_si128(a, imm) \
+  _mm_srli_si128((a), (imm))
 
 static __inline__ __m128i __attribute__((__always_inline__, __nodebug__))
 _mm_srli_epi16(__m128i __a, int __count)
@@ -1070,7 +1103,7 @@ _mm_loadu_si128(__m128i const *__p)
 {
   struct __loadu_si128 {
     __m128i __v;
-  } __attribute__((packed, may_alias));
+  } __attribute__((__packed__, __may_alias__));
   return ((struct __loadu_si128*)__p)->__v;
 }
 
@@ -1284,27 +1317,21 @@ _mm_movemask_epi8(__m128i __a)
 }
 
 #define _mm_shuffle_epi32(a, imm) __extension__ ({ \
-  _Pragma("clang diagnostic push") _Pragma("clang diagnostic ignored \"-Wshadow\""); \
-  __m128i __a = (a); \
-  _Pragma("clang diagnostic pop"); \
-  (__m128i)__builtin_shufflevector((__v4si)__a, (__v4si) _mm_set1_epi32(0), \
+  (__m128i)__builtin_shufflevector((__v4si)(__m128i)(a), \
+                                   (__v4si)_mm_set1_epi32(0), \
                                    (imm) & 0x3, ((imm) & 0xc) >> 2, \
                                    ((imm) & 0x30) >> 4, ((imm) & 0xc0) >> 6); })
 
 #define _mm_shufflelo_epi16(a, imm) __extension__ ({ \
-  _Pragma("clang diagnostic push") _Pragma("clang diagnostic ignored \"-Wshadow\""); \
-  __m128i __a = (a); \
-  _Pragma("clang diagnostic pop"); \
-  (__m128i)__builtin_shufflevector((__v8hi)__a, (__v8hi) _mm_set1_epi16(0), \
+  (__m128i)__builtin_shufflevector((__v8hi)(__m128i)(a), \
+                                   (__v8hi)_mm_set1_epi16(0), \
                                    (imm) & 0x3, ((imm) & 0xc) >> 2, \
                                    ((imm) & 0x30) >> 4, ((imm) & 0xc0) >> 6, \
                                    4, 5, 6, 7); })
 
 #define _mm_shufflehi_epi16(a, imm) __extension__ ({ \
-  _Pragma("clang diagnostic push") _Pragma("clang diagnostic ignored \"-Wshadow\""); \
-  __m128i __a = (a); \
-  _Pragma("clang diagnostic pop"); \
-  (__m128i)__builtin_shufflevector((__v8hi)__a, (__v8hi) _mm_set1_epi16(0), \
+  (__m128i)__builtin_shufflevector((__v8hi)(__m128i)(a), \
+                                   (__v8hi)_mm_set1_epi16(0), \
                                    0, 1, 2, 3, \
                                    4 + (((imm) & 0x03) >> 0), \
                                    4 + (((imm) & 0x0c) >> 2), \
@@ -1396,11 +1423,8 @@ _mm_movemask_pd(__m128d __a)
 }
 
 #define _mm_shuffle_pd(a, b, i) __extension__ ({ \
-  _Pragma("clang diagnostic push") _Pragma("clang diagnostic ignored \"-Wshadow\""); \
-  __m128d __a = (a); \
-  __m128d __b = (b); \
-  _Pragma("clang diagnostic pop"); \
-  __builtin_shufflevector(__a, __b, (i) & 1, (((i) & 2) >> 1) + 2); })
+  __builtin_shufflevector((__m128d)(a), (__m128d)(b), \
+                          (i) & 1, (((i) & 2) >> 1) + 2); })
 
 static __inline__ __m128 __attribute__((__always_inline__, __nodebug__))
 _mm_castpd_ps(__m128d __a)
diff --git a/lib/Headers/htmintrin.h b/lib/Headers/htmintrin.h
new file mode 100644
index 000000000000..0088c7ccab93
--- /dev/null
+++ b/lib/Headers/htmintrin.h
@@ -0,0 +1,226 @@
+/*===---- htmintrin.h - Standard header for PowerPC HTM ---------------===*\
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+\*===----------------------------------------------------------------------===*/
+
+#ifndef __HTMINTRIN_H
+#define __HTMINTRIN_H
+
+#ifndef __HTM__
+#error "HTM instruction set not enabled"
+#endif
+
+#ifdef __powerpc__
+
+#include <stdint.h>
+
+typedef uint64_t texasr_t;
+typedef uint32_t texasru_t;
+typedef uint32_t texasrl_t;
+typedef uintptr_t tfiar_t;
+typedef uintptr_t tfhar_t;
+
+#define _HTM_STATE(CR0) ((CR0 >> 1) & 0x3)
+#define _HTM_NONTRANSACTIONAL 0x0
+#define _HTM_SUSPENDED        0x1
+#define _HTM_TRANSACTIONAL    0x2
+
+#define _TEXASR_EXTRACT_BITS(TEXASR,BITNUM,SIZE) \
+  (((TEXASR) >> (63-(BITNUM))) & ((1<<(SIZE))-1))
+#define _TEXASRU_EXTRACT_BITS(TEXASR,BITNUM,SIZE) \
+  (((TEXASR) >> (31-(BITNUM))) & ((1<<(SIZE))-1))
+
+#define _TEXASR_FAILURE_CODE(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 7, 8)
+#define _TEXASRU_FAILURE_CODE(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 7, 8)
+
+#define _TEXASR_FAILURE_PERSISTENT(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 7, 1)
+#define _TEXASRU_FAILURE_PERSISTENT(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 7, 1)
+
+#define _TEXASR_DISALLOWED(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 8, 1)
+#define _TEXASRU_DISALLOWED(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 8, 1)
+
+#define _TEXASR_NESTING_OVERFLOW(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 9, 1)
+#define _TEXASRU_NESTING_OVERFLOW(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 9, 1)
+
+#define _TEXASR_FOOTPRINT_OVERFLOW(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 10, 1)
+#define _TEXASRU_FOOTPRINT_OVERFLOW(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 10, 1)
+
+#define _TEXASR_SELF_INDUCED_CONFLICT(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 11, 1)
+#define _TEXASRU_SELF_INDUCED_CONFLICT(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 11, 1)
+
+#define _TEXASR_NON_TRANSACTIONAL_CONFLICT(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 12, 1)
+#define _TEXASRU_NON_TRANSACTIONAL_CONFLICT(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 12, 1)
+
+#define _TEXASR_TRANSACTION_CONFLICT(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 13, 1)
+#define _TEXASRU_TRANSACTION_CONFLICT(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 13, 1)
+
+#define _TEXASR_TRANSLATION_INVALIDATION_CONFLICT(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 14, 1)
+#define _TEXASRU_TRANSLATION_INVALIDATION_CONFLICT(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 14, 1)
+
+#define _TEXASR_IMPLEMENTAION_SPECIFIC(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 15, 1)
+#define _TEXASRU_IMPLEMENTAION_SPECIFIC(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 15, 1)
+
+#define _TEXASR_INSTRUCTION_FETCH_CONFLICT(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 16, 1)
+#define _TEXASRU_INSTRUCTION_FETCH_CONFLICT(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 16, 1)
+
+#define _TEXASR_ABORT(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 31, 1)
+#define _TEXASRU_ABORT(TEXASRU) \
+  _TEXASRU_EXTRACT_BITS(TEXASRU, 31, 1)
+
+
+#define _TEXASR_SUSPENDED(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 32, 1)
+
+#define _TEXASR_PRIVILEGE(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 35, 2)
+
+#define _TEXASR_FAILURE_SUMMARY(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 36, 1)
+
+#define _TEXASR_TFIAR_EXACT(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 37, 1)
+
+#define _TEXASR_ROT(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 38, 1)
+
+#define _TEXASR_TRANSACTION_LEVEL(TEXASR) \
+  _TEXASR_EXTRACT_BITS(TEXASR, 63, 12)
+
+#endif /* __powerpc */
+
+#ifdef __s390__
+
+/* Condition codes generated by tbegin  */
+#define _HTM_TBEGIN_STARTED       0
+#define _HTM_TBEGIN_INDETERMINATE 1
+#define _HTM_TBEGIN_TRANSIENT     2
+#define _HTM_TBEGIN_PERSISTENT    3
+
+/* The abort codes below this threshold are reserved for machine use.  */
+#define _HTM_FIRST_USER_ABORT_CODE 256
+
+/* The transaction diagnostic block is it is defined in the Principles
+   of Operation chapter 5-91.  */
+
+struct __htm_tdb {
+  unsigned char format;                /*   0 */
+  unsigned char flags;
+  unsigned char reserved1[4];
+  unsigned short nesting_depth;
+  unsigned long long abort_code;       /*   8 */
+  unsigned long long conflict_token;   /*  16 */
+  unsigned long long atia;             /*  24 */
+  unsigned char eaid;                  /*  32 */
+  unsigned char dxc;
+  unsigned char reserved2[2];
+  unsigned int program_int_id;
+  unsigned long long exception_id;     /*  40 */
+  unsigned long long bea;              /*  48 */
+  unsigned char reserved3[72];         /*  56 */
+  unsigned long long gprs[16];         /* 128 */
+} __attribute__((__packed__, __aligned__ (8)));
+
+
+/* Helper intrinsics to retry tbegin in case of transient failure.  */
+
+static __inline int __attribute__((__always_inline__, __nodebug__))
+__builtin_tbegin_retry_null (int retry)
+{
+  int cc, i = 0;
+
+  while ((cc = __builtin_tbegin(0)) == _HTM_TBEGIN_TRANSIENT
+         && i++ < retry)
+    __builtin_tx_assist(i);
+
+  return cc;
+}
+
+static __inline int __attribute__((__always_inline__, __nodebug__))
+__builtin_tbegin_retry_tdb (void *tdb, int retry)
+{
+  int cc, i = 0;
+
+  while ((cc = __builtin_tbegin(tdb)) == _HTM_TBEGIN_TRANSIENT
+         && i++ < retry)
+    __builtin_tx_assist(i);
+
+  return cc;
+}
+
+#define __builtin_tbegin_retry(tdb, retry) \
+  (__builtin_constant_p(tdb == 0) && tdb == 0 ? \
+   __builtin_tbegin_retry_null(retry) : \
+   __builtin_tbegin_retry_tdb(tdb, retry))
+
+static __inline int __attribute__((__always_inline__, __nodebug__))
+__builtin_tbegin_retry_nofloat_null (int retry)
+{
+  int cc, i = 0;
+
+  while ((cc = __builtin_tbegin_nofloat(0)) == _HTM_TBEGIN_TRANSIENT
+         && i++ < retry)
+    __builtin_tx_assist(i);
+
+  return cc;
+}
+
+static __inline int __attribute__((__always_inline__, __nodebug__))
+__builtin_tbegin_retry_nofloat_tdb (void *tdb, int retry)
+{
+  int cc, i = 0;
+
+  while ((cc = __builtin_tbegin_nofloat(tdb)) == _HTM_TBEGIN_TRANSIENT
+         && i++ < retry)
+    __builtin_tx_assist(i);
+
+  return cc;
+}
+
+#define __builtin_tbegin_retry_nofloat(tdb, retry) \
+  (__builtin_constant_p(tdb == 0) && tdb == 0 ? \
+   __builtin_tbegin_retry_nofloat_null(retry) : \
+   __builtin_tbegin_retry_nofloat_tdb(tdb, retry))
+
+#endif /* __s390__ */
+
+#endif /* __HTMINTRIN_H */
diff --git a/lib/Headers/htmxlintrin.h b/lib/Headers/htmxlintrin.h
new file mode 100644
index 000000000000..30f524d5df49
--- /dev/null
+++ b/lib/Headers/htmxlintrin.h
@@ -0,0 +1,363 @@
+/*===---- htmxlintrin.h - XL compiler HTM execution intrinsics-------------===*\
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+\*===----------------------------------------------------------------------===*/
+
+#ifndef __HTMXLINTRIN_H
+#define __HTMXLINTRIN_H
+
+#ifndef __HTM__
+#error "HTM instruction set not enabled"
+#endif
+
+#include <htmintrin.h>
+
+#ifdef __powerpc__
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define _TEXASR_PTR(TM_BUF) \
+  ((texasr_t *)((TM_BUF)+0))
+#define _TEXASRU_PTR(TM_BUF) \
+  ((texasru_t *)((TM_BUF)+0))
+#define _TEXASRL_PTR(TM_BUF) \
+  ((texasrl_t *)((TM_BUF)+4))
+#define _TFIAR_PTR(TM_BUF) \
+  ((tfiar_t *)((TM_BUF)+8))
+
+typedef char TM_buff_type[16];
+
+/* This macro can be used to determine whether a transaction was successfully 
+   started from the __TM_begin() and __TM_simple_begin() intrinsic functions
+   below.  */
+#define _HTM_TBEGIN_STARTED     1
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_simple_begin (void)
+{
+  if (__builtin_expect (__builtin_tbegin (0), 1))
+    return _HTM_TBEGIN_STARTED;
+  return 0;
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_begin (void* const TM_buff)
+{
+  *_TEXASRL_PTR (TM_buff) = 0;
+  if (__builtin_expect (__builtin_tbegin (0), 1))
+    return _HTM_TBEGIN_STARTED;
+#ifdef __powerpc64__
+  *_TEXASR_PTR (TM_buff) = __builtin_get_texasr ();
+#else
+  *_TEXASRU_PTR (TM_buff) = __builtin_get_texasru ();
+  *_TEXASRL_PTR (TM_buff) = __builtin_get_texasr ();
+#endif
+  *_TFIAR_PTR (TM_buff) = __builtin_get_tfiar ();
+  return 0;
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_end (void)
+{
+  if (__builtin_expect (__builtin_tend (0), 1))
+    return 1;
+  return 0;
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_abort (void)
+{
+  __builtin_tabort (0);
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_named_abort (unsigned char const code)
+{
+  __builtin_tabort (code);
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_resume (void)
+{
+  __builtin_tresume ();
+}
+
+extern __inline void
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_suspend (void)
+{
+  __builtin_tsuspend ();
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_is_user_abort (void* const TM_buff)
+{
+  texasru_t texasru = *_TEXASRU_PTR (TM_buff);
+  return _TEXASRU_ABORT (texasru);
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_is_named_user_abort (void* const TM_buff, unsigned char *code)
+{
+  texasru_t texasru = *_TEXASRU_PTR (TM_buff);
+
+  *code = _TEXASRU_FAILURE_CODE (texasru);
+  return _TEXASRU_ABORT (texasru);
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_is_illegal (void* const TM_buff)
+{
+  texasru_t texasru = *_TEXASRU_PTR (TM_buff);
+  return _TEXASRU_DISALLOWED (texasru);
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_is_footprint_exceeded (void* const TM_buff)
+{
+  texasru_t texasru = *_TEXASRU_PTR (TM_buff);
+  return _TEXASRU_FOOTPRINT_OVERFLOW (texasru);
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_nesting_depth (void* const TM_buff)
+{
+  texasrl_t texasrl;
+
+  if (_HTM_STATE (__builtin_ttest ()) == _HTM_NONTRANSACTIONAL)
+    {
+      texasrl = *_TEXASRL_PTR (TM_buff);
+      if (!_TEXASR_FAILURE_SUMMARY (texasrl))
+        texasrl = 0;
+    }
+  else
+    texasrl = (texasrl_t) __builtin_get_texasr ();
+
+  return _TEXASR_TRANSACTION_LEVEL (texasrl);
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_is_nested_too_deep(void* const TM_buff)
+{
+  texasru_t texasru = *_TEXASRU_PTR (TM_buff);
+  return _TEXASRU_NESTING_OVERFLOW (texasru);
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_is_conflict(void* const TM_buff)
+{
+  texasru_t texasru = *_TEXASRU_PTR (TM_buff);
+  /* Return TEXASR bits 11 (Self-Induced Conflict) through
+     14 (Translation Invalidation Conflict).  */
+  return (_TEXASRU_EXTRACT_BITS (texasru, 14, 4)) ? 1 : 0;
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_is_failure_persistent(void* const TM_buff)
+{
+  texasru_t texasru = *_TEXASRU_PTR (TM_buff);
+  return _TEXASRU_FAILURE_PERSISTENT (texasru);
+}
+
+extern __inline long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_failure_address(void* const TM_buff)
+{
+  return *_TFIAR_PTR (TM_buff);
+}
+
+extern __inline long long
+__attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
+__TM_failure_code(void* const TM_buff)
+{
+  return *_TEXASR_PTR (TM_buff);
+}
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* __powerpc__ */
+
+#ifdef __s390__
+
+#include <stdint.h>
+
+/* These intrinsics are being made available for compatibility with
+   the IBM XL compiler.  For documentation please see the "z/OS XL
+   C/C++ Programming Guide" publically available on the web.  */
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_simple_begin ()
+{
+  return __builtin_tbegin_nofloat (0);
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_begin (void* const tdb)
+{
+  return __builtin_tbegin_nofloat (tdb);
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_end ()
+{
+  return __builtin_tend ();
+}
+
+static __inline void __attribute__((__always_inline__))
+__TM_abort ()
+{
+  return __builtin_tabort (_HTM_FIRST_USER_ABORT_CODE);
+}
+
+static __inline void __attribute__((__always_inline__, __nodebug__))
+__TM_named_abort (unsigned char const code)
+{
+  return __builtin_tabort ((int)_HTM_FIRST_USER_ABORT_CODE + code);
+}
+
+static __inline void __attribute__((__always_inline__, __nodebug__))
+__TM_non_transactional_store (void* const addr, long long const value)
+{
+  __builtin_non_tx_store ((uint64_t*)addr, (uint64_t)value);
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_nesting_depth (void* const tdb_ptr)
+{
+  int depth = __builtin_tx_nesting_depth ();
+  struct __htm_tdb *tdb = (struct __htm_tdb*)tdb_ptr;
+
+  if (depth != 0)
+    return depth;
+
+  if (tdb->format != 1)
+    return 0;
+  return tdb->nesting_depth;
+}
+
+/* Transaction failure diagnostics */
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_is_user_abort (void* const tdb_ptr)
+{
+  struct __htm_tdb *tdb = (struct __htm_tdb*)tdb_ptr;
+
+  if (tdb->format != 1)
+    return 0;
+
+  return !!(tdb->abort_code >= _HTM_FIRST_USER_ABORT_CODE);
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_is_named_user_abort (void* const tdb_ptr, unsigned char* code)
+{
+  struct __htm_tdb *tdb = (struct __htm_tdb*)tdb_ptr;
+
+  if (tdb->format != 1)
+    return 0;
+
+  if (tdb->abort_code >= _HTM_FIRST_USER_ABORT_CODE)
+    {
+      *code = tdb->abort_code - _HTM_FIRST_USER_ABORT_CODE;
+      return 1;
+    }
+  return 0;
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_is_illegal (void* const tdb_ptr)
+{
+  struct __htm_tdb *tdb = (struct __htm_tdb*)tdb_ptr;
+
+  return (tdb->format == 1
+	  && (tdb->abort_code == 4 /* unfiltered program interruption */
+	      || tdb->abort_code == 11 /* restricted instruction */));
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_is_footprint_exceeded (void* const tdb_ptr)
+{
+  struct __htm_tdb *tdb = (struct __htm_tdb*)tdb_ptr;
+
+  return (tdb->format == 1
+	  && (tdb->abort_code == 7 /* fetch overflow */
+	      || tdb->abort_code == 8 /* store overflow */));
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_is_nested_too_deep (void* const tdb_ptr)
+{
+  struct __htm_tdb *tdb = (struct __htm_tdb*)tdb_ptr;
+
+  return tdb->format == 1 && tdb->abort_code == 13; /* depth exceeded */
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_is_conflict (void* const tdb_ptr)
+{
+  struct __htm_tdb *tdb = (struct __htm_tdb*)tdb_ptr;
+
+  return (tdb->format == 1
+	  && (tdb->abort_code == 9 /* fetch conflict */
+	      || tdb->abort_code == 10 /* store conflict */));
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_is_failure_persistent (long const result)
+{
+  return result == _HTM_TBEGIN_PERSISTENT;
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_failure_address (void* const tdb_ptr)
+{
+  struct __htm_tdb *tdb = (struct __htm_tdb*)tdb_ptr;
+  return tdb->atia;
+}
+
+static __inline long __attribute__((__always_inline__, __nodebug__))
+__TM_failure_code (void* const tdb_ptr)
+{
+  struct __htm_tdb *tdb = (struct __htm_tdb*)tdb_ptr;
+
+  return tdb->abort_code;
+}
+
+#endif /* __s390__ */
+
+#endif /* __HTMXLINTRIN_H  */
diff --git a/lib/Headers/immintrin.h b/lib/Headers/immintrin.h
index 2400fea499bd..ac7d54a41db8 100644
--- a/lib/Headers/immintrin.h
+++ b/lib/Headers/immintrin.h
@@ -88,10 +88,18 @@
 #include <avx512bwintrin.h>
 #endif
 
+#ifdef __AVX512DQ__
+#include <avx512dqintrin.h>
+#endif
+
 #if defined (__AVX512VL__) && defined (__AVX512BW__)
 #include <avx512vlbwintrin.h>
 #endif
 
+#if defined (__AVX512VL__) && defined (__AVX512DQ__)
+#include <avx512vldqintrin.h>
+#endif
+
 #ifdef __AVX512ER__
 #include <avx512erintrin.h>
 #endif
diff --git a/lib/Headers/module.modulemap b/lib/Headers/module.modulemap
index 062464ed2e53..ac5876f8d726 100644
--- a/lib/Headers/module.modulemap
+++ b/lib/Headers/module.modulemap
@@ -49,7 +49,7 @@ module _Builtin_intrinsics [system] [extern_c] {
     explicit module sse {
       requires sse
       export mmx
-      export * // note: for hackish <emmintrin.h> dependency
+      export sse2 // note: for hackish <emmintrin.h> dependency
       header "xmmintrin.h"
     }
 
@@ -169,6 +169,19 @@ module _Builtin_intrinsics [system] [extern_c] {
       header "__wmmintrin_pclmul.h"
     }
   }
+
+  explicit module systemz {
+    requires systemz
+    export *
+
+    header "s390intrin.h"
+
+    explicit module htm {
+      requires htm
+      header "htmintrin.h"
+      header "htmxlintrin.h"
+    }
+  }
 }
 
 module _Builtin_stddef_max_align_t [system] [extern_c] {
diff --git a/lib/Headers/s390intrin.h b/lib/Headers/s390intrin.h
new file mode 100644
index 000000000000..b20989552d4d
--- /dev/null
+++ b/lib/Headers/s390intrin.h
@@ -0,0 +1,35 @@
+/*===---- s390intrin.h - SystemZ intrinsics --------------------------------===
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ *
+ *===-----------------------------------------------------------------------===
+ */
+
+#ifndef __S390INTRIN_H
+#define __S390INTRIN_H
+
+#ifndef __s390__
+#error "<s390intrin.h> is for s390 only"
+#endif
+
+#ifdef __HTM__
+#include <htmintrin.h>
+#endif
+
+#endif /* __S390INTRIN_H*/
diff --git a/lib/Headers/stdatomic.h b/lib/Headers/stdatomic.h
index e3c34767a21a..e03798766014 100644
--- a/lib/Headers/stdatomic.h
+++ b/lib/Headers/stdatomic.h
@@ -71,7 +71,7 @@ typedef enum memory_order {
 
 /* 7.17.4 Fences */
 
-// These should be provided by the libc implementation.
+/* These should be provided by the libc implementation. */
 void atomic_thread_fence(memory_order);
 void atomic_signal_fence(memory_order);
 
@@ -164,7 +164,7 @@ typedef struct atomic_flag { atomic_bool _Value; } atomic_flag;
 
 #define ATOMIC_FLAG_INIT { 0 }
 
-// These should be provided by the libc implementation.
+/* These should be provided by the libc implementation. */
 #ifdef __cplusplus
 bool atomic_flag_test_and_set(volatile atomic_flag *);
 bool atomic_flag_test_and_set_explicit(volatile atomic_flag *, memory_order);
diff --git a/lib/Headers/unwind.h b/lib/Headers/unwind.h
index 90aca16aca39..303d79288aee 100644
--- a/lib/Headers/unwind.h
+++ b/lib/Headers/unwind.h
@@ -235,9 +235,9 @@ void *_Unwind_FindEnclosingFunction(void *);
 #ifdef __APPLE__
 
 _Unwind_Ptr _Unwind_GetDataRelBase(struct _Unwind_Context *)
-    __attribute__((unavailable));
+    __attribute__((__unavailable__));
 _Unwind_Ptr _Unwind_GetTextRelBase(struct _Unwind_Context *)
-    __attribute__((unavailable));
+    __attribute__((__unavailable__));
 
 /* Darwin-specific functions */
 void __register_frame(const void *);
@@ -251,15 +251,15 @@ struct dwarf_eh_bases {
 void *_Unwind_Find_FDE(const void *, struct dwarf_eh_bases *);
 
 void __register_frame_info_bases(const void *, void *, void *, void *)
-  __attribute__((unavailable));
-void __register_frame_info(const void *, void *) __attribute__((unavailable));
+  __attribute__((__unavailable__));
+void __register_frame_info(const void *, void *) __attribute__((__unavailable__));
 void __register_frame_info_table_bases(const void *, void*, void *, void *)
-  __attribute__((unavailable));
+  __attribute__((__unavailable__));
 void __register_frame_info_table(const void *, void *)
-  __attribute__((unavailable));
-void __register_frame_table(const void *) __attribute__((unavailable));
-void __deregister_frame_info(const void *) __attribute__((unavailable));
-void __deregister_frame_info_bases(const void *)__attribute__((unavailable));
+  __attribute__((__unavailable__));
+void __register_frame_table(const void *) __attribute__((__unavailable__));
+void __deregister_frame_info(const void *) __attribute__((__unavailable__));
+void __deregister_frame_info_bases(const void *)__attribute__((__unavailable__));
 
 #else
 
diff --git a/lib/Headers/xmmintrin.h b/lib/Headers/xmmintrin.h
index d1afe81601c3..3a6b95e8bfea 100644
--- a/lib/Headers/xmmintrin.h
+++ b/lib/Headers/xmmintrin.h
@@ -994,7 +994,7 @@ do { \
 #define _m_ _mm_
 
 /* Ugly hack for backwards-compatibility (compatible with gcc) */
-#ifdef __SSE2__
+#if defined(__SSE2__) && !__has_feature(modules)
 #include <emmintrin.h>
 #endif
 
diff --git a/lib/Index/SimpleFormatContext.h b/lib/Index/SimpleFormatContext.h
index 080a4ad04ddf..b88421498959 100644
--- a/lib/Index/SimpleFormatContext.h
+++ b/lib/Index/SimpleFormatContext.h
@@ -44,8 +44,6 @@ public:
     Diagnostics->setClient(new IgnoringDiagConsumer, true);
   }
 
-  ~SimpleFormatContext() { }
-
   FileID createInMemoryFile(StringRef Name, StringRef Content) {
     std::unique_ptr<llvm::MemoryBuffer> Source =
         llvm::MemoryBuffer::getMemBuffer(Content);
diff --git a/lib/Index/USRGeneration.cpp b/lib/Index/USRGeneration.cpp
index baa166ee2802..8cdd283ba5c5 100644
--- a/lib/Index/USRGeneration.cpp
+++ b/lib/Index/USRGeneration.cpp
@@ -847,7 +847,7 @@ bool clang::index::generateUSRForDecl(const Decl *D,
   return UG.ignoreResults();
 }
 
-bool clang::index::generateUSRForMacro(const MacroDefinition *MD,
+bool clang::index::generateUSRForMacro(const MacroDefinitionRecord *MD,
                                        const SourceManager &SM,
                                        SmallVectorImpl<char> &Buf) {
   // Don't generate USRs for things with invalid locations.
diff --git a/lib/Lex/HeaderSearch.cpp b/lib/Lex/HeaderSearch.cpp
index d6b255fb014e..ad7d3449ac4a 100644
--- a/lib/Lex/HeaderSearch.cpp
+++ b/lib/Lex/HeaderSearch.cpp
@@ -18,6 +18,7 @@
 #include "clang/Lex/HeaderSearchOptions.h"
 #include "clang/Lex/LexDiagnostic.h"
 #include "clang/Lex/Lexer.h"
+#include "clang/Lex/Preprocessor.h"
 #include "llvm/ADT/APInt.h"
 #include "llvm/ADT/Hashing.h"
 #include "llvm/ADT/SmallString.h"
@@ -149,7 +150,7 @@ std::string HeaderSearch::getModuleFileName(StringRef ModuleName,
 
     SmallString<128> HashStr;
     llvm::APInt(64, size_t(Hash)).toStringUnsigned(HashStr, /*Radix*/36);
-    llvm::sys::path::append(Result, ModuleName + "-" + HashStr.str() + ".pcm");
+    llvm::sys::path::append(Result, ModuleName + "-" + HashStr + ".pcm");
   }
   return Result.str().str();
 }
@@ -297,7 +298,7 @@ const FileEntry *DirectoryLookup::LookupFile(
       RelativePath->append(Filename.begin(), Filename.end());
     }
 
-    return getFileAndSuggestModule(HS, TmpDir.str(), getDir(),
+    return getFileAndSuggestModule(HS, TmpDir, getDir(),
                                    isSystemHeaderDirectory(),
                                    SuggestedModule);
   }
@@ -438,7 +439,7 @@ const FileEntry *DirectoryLookup::DoFrameworkLookup(
     HS.IncrementFrameworkLookupCount();
 
     // If the framework dir doesn't exist, we fail.
-    const DirectoryEntry *Dir = FileMgr.getDirectory(FrameworkName.str());
+    const DirectoryEntry *Dir = FileMgr.getDirectory(FrameworkName);
     if (!Dir) return nullptr;
 
     // Otherwise, if it does, remember that this is the right direntry for this
@@ -450,7 +451,7 @@ const FileEntry *DirectoryLookup::DoFrameworkLookup(
     if (getDirCharacteristic() == SrcMgr::C_User) {
       SmallString<1024> SystemFrameworkMarker(FrameworkName);
       SystemFrameworkMarker += ".system_framework";
-      if (llvm::sys::fs::exists(SystemFrameworkMarker.str())) {
+      if (llvm::sys::fs::exists(SystemFrameworkMarker)) {
         CacheEntry.IsUserSpecifiedSystemFramework = true;
       }
     }
@@ -476,7 +477,7 @@ const FileEntry *DirectoryLookup::DoFrameworkLookup(
   }
 
   FrameworkName.append(Filename.begin()+SlashPos+1, Filename.end());
-  const FileEntry *FE = FileMgr.getFile(FrameworkName.str(),
+  const FileEntry *FE = FileMgr.getFile(FrameworkName,
                                         /*openFile=*/!SuggestedModule);
   if (!FE) {
     // Check "/System/Library/Frameworks/Cocoa.framework/PrivateHeaders/file.h"
@@ -487,7 +488,7 @@ const FileEntry *DirectoryLookup::DoFrameworkLookup(
       SearchPath->insert(SearchPath->begin()+OrigSize, Private,
                          Private+strlen(Private));
 
-    FE = FileMgr.getFile(FrameworkName.str(), /*openFile=*/!SuggestedModule);
+    FE = FileMgr.getFile(FrameworkName, /*openFile=*/!SuggestedModule);
   }
 
   // If we found the header and are allowed to suggest a module, do so now.
@@ -594,7 +595,13 @@ const FileEntry *HeaderSearch::LookupFile(
       RelativePath->append(Filename.begin(), Filename.end());
     }
     // Otherwise, just return the file.
-    return FileMgr.getFile(Filename, /*openFile=*/true);
+    const FileEntry *File = FileMgr.getFile(Filename, /*openFile=*/true);
+    if (File && SuggestedModule) {
+      // If there is a module that corresponds to this header, suggest it.
+      hasModuleMap(Filename, File->getDir(), /*SystemHeaderDir*/false);
+      *SuggestedModule = findModuleForHeader(File);
+    }
+    return File;
   }
 
   // This is the header that MSVC's header search would have found.
@@ -628,7 +635,7 @@ const FileEntry *HeaderSearch::LookupFile(
       bool IncluderIsSystemHeader =
           Includer && getFileInfo(Includer).DirInfo != SrcMgr::C_User;
       if (const FileEntry *FE = getFileAndSuggestModule(
-              *this, TmpDir.str(), IncluderAndDir.second,
+              *this, TmpDir, IncluderAndDir.second,
               IncluderIsSystemHeader, SuggestedModule)) {
         if (!Includer) {
           assert(First && "only first includer can have no file");
@@ -865,7 +872,7 @@ LookupSubframeworkHeader(StringRef Filename,
     ++NumSubFrameworkLookups;
 
     // If the framework dir doesn't exist, we fail.
-    const DirectoryEntry *Dir = FileMgr.getDirectory(FrameworkName.str());
+    const DirectoryEntry *Dir = FileMgr.getDirectory(FrameworkName);
     if (!Dir) return nullptr;
 
     // Otherwise, if it does, remember that this is the right direntry for this
@@ -890,7 +897,7 @@ LookupSubframeworkHeader(StringRef Filename,
   }
 
   HeadersFilename.append(Filename.begin()+SlashPos+1, Filename.end());
-  if (!(FE = FileMgr.getFile(HeadersFilename.str(), /*openFile=*/true))) {
+  if (!(FE = FileMgr.getFile(HeadersFilename, /*openFile=*/true))) {
 
     // Check ".../Frameworks/HIToolbox.framework/PrivateHeaders/HIToolbox.h"
     HeadersFilename = FrameworkName;
@@ -902,7 +909,7 @@ LookupSubframeworkHeader(StringRef Filename,
     }
 
     HeadersFilename.append(Filename.begin()+SlashPos+1, Filename.end());
-    if (!(FE = FileMgr.getFile(HeadersFilename.str(), /*openFile=*/true)))
+    if (!(FE = FileMgr.getFile(HeadersFilename, /*openFile=*/true)))
       return nullptr;
   }
 
@@ -1016,7 +1023,9 @@ void HeaderSearch::MarkFileModuleHeader(const FileEntry *FE,
   HFI.setHeaderRole(Role);
 }
 
-bool HeaderSearch::ShouldEnterIncludeFile(const FileEntry *File, bool isImport){
+bool HeaderSearch::ShouldEnterIncludeFile(Preprocessor &PP,
+                                          const FileEntry *File,
+                                          bool isImport) {
   ++NumIncluded; // Count # of attempted #includes.
 
   // Get information about this file.
@@ -1041,7 +1050,7 @@ bool HeaderSearch::ShouldEnterIncludeFile(const FileEntry *File, bool isImport){
   // if the macro that guards it is defined, we know the #include has no effect.
   if (const IdentifierInfo *ControllingMacro
       = FileInfo.getControllingMacro(ExternalLookup))
-    if (ControllingMacro->hasMacroDefinition()) {
+    if (PP.isMacroDefined(ControllingMacro)) {
       ++NumMultiIncludeFileOptzn;
       return false;
     }
@@ -1067,7 +1076,7 @@ StringRef HeaderSearch::getUniqueFrameworkName(StringRef Framework) {
 bool HeaderSearch::hasModuleMap(StringRef FileName, 
                                 const DirectoryEntry *Root,
                                 bool IsSystem) {
-  if (!enabledModules() || !LangOpts.ModulesImplicitMaps)
+  if (!HSOpts->ModuleMaps || !LangOpts.ModulesImplicitMaps)
     return false;
 
   SmallVector<const DirectoryEntry *, 2> FixUpDirectories;
@@ -1284,7 +1293,7 @@ void HeaderSearch::collectAllModules(SmallVectorImpl<Module *> &Modules) {
                                 DirNative);
 
         // Search each of the ".framework" directories to load them as modules.
-        for (llvm::sys::fs::directory_iterator Dir(DirNative.str(), EC), DirEnd;
+        for (llvm::sys::fs::directory_iterator Dir(DirNative, EC), DirEnd;
              Dir != DirEnd && !EC; Dir.increment(EC)) {
           if (llvm::sys::path::extension(Dir->path()) != ".framework")
             continue;
@@ -1351,10 +1360,12 @@ void HeaderSearch::loadSubdirectoryModuleMaps(DirectoryLookup &SearchDir) {
   std::error_code EC;
   SmallString<128> DirNative;
   llvm::sys::path::native(SearchDir.getDir()->getName(), DirNative);
-  for (llvm::sys::fs::directory_iterator Dir(DirNative.str(), EC), DirEnd;
+  for (llvm::sys::fs::directory_iterator Dir(DirNative, EC), DirEnd;
        Dir != DirEnd && !EC; Dir.increment(EC)) {
-    loadModuleMapFile(Dir->path(), SearchDir.isSystemHeaderDirectory(),
-                      SearchDir.isFramework());
+    bool IsFramework = llvm::sys::path::extension(Dir->path()) == ".framework";
+    if (IsFramework == SearchDir.isFramework())
+      loadModuleMapFile(Dir->path(), SearchDir.isSystemHeaderDirectory(),
+                        SearchDir.isFramework());
   }
 
   SearchDir.setSearchedAllModuleMaps(true);
diff --git a/lib/Lex/Lexer.cpp b/lib/Lex/Lexer.cpp
index ca5252e1c9ce..3f89ea649cbb 100644
--- a/lib/Lex/Lexer.cpp
+++ b/lib/Lex/Lexer.cpp
@@ -143,14 +143,8 @@ Lexer::Lexer(SourceLocation fileloc, const LangOptions &langOpts,
 /// range will outlive it, so it doesn't take ownership of it.
 Lexer::Lexer(FileID FID, const llvm::MemoryBuffer *FromFile,
              const SourceManager &SM, const LangOptions &langOpts)
-  : FileLoc(SM.getLocForStartOfFile(FID)), LangOpts(langOpts) {
-
-  InitLexer(FromFile->getBufferStart(), FromFile->getBufferStart(),
-            FromFile->getBufferEnd());
-
-  // We *are* in raw mode.
-  LexingRawMode = true;
-}
+    : Lexer(SM.getLocForStartOfFile(FID), langOpts, FromFile->getBufferStart(),
+            FromFile->getBufferStart(), FromFile->getBufferEnd()) {}
 
 /// Create_PragmaLexer: Lexer constructor - Create a new lexer object for
 /// _Pragma expansion.  This has a variety of magic semantics that this method
@@ -1860,7 +1854,7 @@ bool Lexer::LexAngledStringLiteral(Token &Result, const char *CurPtr) {
   char C = getAndAdvanceChar(CurPtr, Result);
   while (C != '>') {
     // Skip escaped characters.
-    if (C == '\\') {
+    if (C == '\\' && CurPtr < BufferEnd) {
       // Skip the escaped character.
       getAndAdvanceChar(CurPtr, Result);
     } else if (C == '\n' || C == '\r' ||             // Newline.
diff --git a/lib/Lex/LiteralSupport.cpp b/lib/Lex/LiteralSupport.cpp
index 03331fb33eb2..aed91648799b 100644
--- a/lib/Lex/LiteralSupport.cpp
+++ b/lib/Lex/LiteralSupport.cpp
@@ -144,7 +144,8 @@ static unsigned ProcessCharEscape(const char *ThisTokBegin,
       int CharVal = llvm::hexDigitValue(ThisTokBuf[0]);
       if (CharVal == -1) break;
       // About to shift out a digit?
-      Overflow |= (ResultChar & 0xF0000000) ? true : false;
+      if (ResultChar & 0xF0000000)
+        Overflow = true;
       ResultChar <<= 4;
       ResultChar |= CharVal;
     }
@@ -596,7 +597,8 @@ NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling,
       if (isFloat) break;               // LF invalid.
 
       // Check for long long.  The L's need to be adjacent and the same case.
-      if (s+1 != ThisTokEnd && s[1] == s[0]) {
+      if (s[1] == s[0]) {
+        assert(s + 1 < ThisTokEnd && "didn't maximally munch?");
         if (isFPConstant) break;        // long long invalid for floats.
         isLongLong = true;
         ++s;  // Eat both of them.
@@ -610,54 +612,45 @@ NumericLiteralParser::NumericLiteralParser(StringRef TokSpelling,
         if (isLong || isLongLong || MicrosoftInteger)
           break;
 
-        // Allow i8, i16, i32, i64, and i128.
-        if (s + 1 != ThisTokEnd) {
+        if (!isFPConstant) {
+          // Allow i8, i16, i32, i64, and i128.
           switch (s[1]) {
-            case '8':
-              if (isFPConstant) break;
-              s += 2; // i8 suffix
-              MicrosoftInteger = 8;
-              break;
-            case '1':
-              if (isFPConstant) break;
-              if (s + 2 == ThisTokEnd) break;
-              if (s[2] == '6') {
-                s += 3; // i16 suffix
-                MicrosoftInteger = 16;
-              }
-              else if (s[2] == '2') {
-                if (s + 3 == ThisTokEnd) break;
-                if (s[3] == '8') {
-                  s += 4; // i128 suffix
-                  MicrosoftInteger = 128;
-                }
-              }
-              break;
-            case '3':
-              if (isFPConstant) break;
-              if (s + 2 == ThisTokEnd) break;
-              if (s[2] == '2') {
-                s += 3; // i32 suffix
-                MicrosoftInteger = 32;
-              }
-              break;
-            case '6':
-              if (isFPConstant) break;
-              if (s + 2 == ThisTokEnd) break;
-              if (s[2] == '4') {
-                s += 3; // i64 suffix
-                MicrosoftInteger = 64;
-              }
-              break;
-            default:
-              break;
-          }
-          if (MicrosoftInteger)
+          case '8':
+            s += 2; // i8 suffix
+            MicrosoftInteger = 8;
+            break;
+          case '1':
+            if (s[2] == '6') {
+              s += 3; // i16 suffix
+              MicrosoftInteger = 16;
+            } else if (s[2] == '2' && s[3] == '8') {
+              s += 4; // i128 suffix
+              MicrosoftInteger = 128;
+            }
+            break;
+          case '3':
+            if (s[2] == '2') {
+              s += 3; // i32 suffix
+              MicrosoftInteger = 32;
+            }
+            break;
+          case '6':
+            if (s[2] == '4') {
+              s += 3; // i64 suffix
+              MicrosoftInteger = 64;
+            }
             break;
+          default:
+            break;
+          }
+        }
+        if (MicrosoftInteger) {
+          assert(s <= ThisTokEnd && "didn't maximally munch?");
+          break;
         }
       }
       // "i", "if", and "il" are user-defined suffixes in C++1y.
-      if (PP.getLangOpts().CPlusPlus14 && *s == 'i')
+      if (*s == 'i' && PP.getLangOpts().CPlusPlus14)
         break;
       // fall through.
     case 'j':
@@ -755,11 +748,11 @@ void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) {
   s++;
 
   int c1 = s[0];
-  int c2 = s[1];
 
   // Handle a hex number like 0x1234.
-  if ((c1 == 'x' || c1 == 'X') && (isHexDigit(c2) || c2 == '.')) {
+  if ((c1 == 'x' || c1 == 'X') && (isHexDigit(s[1]) || s[1] == '.')) {
     s++;
+    assert(s < ThisTokEnd && "didn't maximally munch?");
     radix = 16;
     DigitsBegin = s;
     s = SkipHexDigits(s);
@@ -811,7 +804,7 @@ void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) {
   }
 
   // Handle simple binary numbers 0b01010
-  if ((c1 == 'b' || c1 == 'B') && (c2 == '0' || c2 == '1')) {
+  if ((c1 == 'b' || c1 == 'B') && (s[1] == '0' || s[1] == '1')) {
     // 0b101010 is a C++1y / GCC extension.
     PP.Diag(TokLoc,
             PP.getLangOpts().CPlusPlus14
@@ -820,6 +813,7 @@ void NumericLiteralParser::ParseNumberStartingWithZero(SourceLocation TokLoc) {
                 ? diag::ext_binary_literal_cxx14
                 : diag::ext_binary_literal);
     ++s;
+    assert(s < ThisTokEnd && "didn't maximally munch?");
     radix = 2;
     DigitsBegin = s;
     s = SkipBinaryDigits(s);
diff --git a/lib/Lex/MacroArgs.cpp b/lib/Lex/MacroArgs.cpp
index 9967f3f0e493..1c1979d8e83d 100644
--- a/lib/Lex/MacroArgs.cpp
+++ b/lib/Lex/MacroArgs.cpp
@@ -133,12 +133,11 @@ bool MacroArgs::ArgNeedsPreexpansion(const Token *ArgTok,
   // If there are no identifiers in the argument list, or if the identifiers are
   // known to not be macros, pre-expansion won't modify it.
   for (; ArgTok->isNot(tok::eof); ++ArgTok)
-    if (IdentifierInfo *II = ArgTok->getIdentifierInfo()) {
-      if (II->hasMacroDefinition() && PP.getMacroInfo(II)->isEnabled())
+    if (IdentifierInfo *II = ArgTok->getIdentifierInfo())
+      if (II->hasMacroDefinition())
         // Return true even though the macro could be a function-like macro
-        // without a following '(' token.
+        // without a following '(' token, or could be disabled, or not visible.
         return true;
-    }
   return false;
 }
 
diff --git a/lib/Lex/MacroInfo.cpp b/lib/Lex/MacroInfo.cpp
index 5416886cc9f7..109b6c12b89b 100644
--- a/lib/Lex/MacroInfo.cpp
+++ b/lib/Lex/MacroInfo.cpp
@@ -218,13 +218,9 @@ void MacroDirective::dump() const {
   if (auto *Prev = getPrevious())
     Out << " prev " << Prev;
   if (IsFromPCH) Out << " from_pch";
-  if (IsImported) Out << " imported";
-  if (IsAmbiguous) Out << " ambiguous";
 
-  if (IsPublic)
-    Out << " public";
-  else if (isa<VisibilityMacroDirective>(this))
-    Out << " private";
+  if (isa<VisibilityMacroDirective>(this))
+    Out << (IsPublic ? " public" : " private");
 
   if (auto *DMD = dyn_cast<DefMacroDirective>(this)) {
     if (auto *Info = DMD->getInfo()) {
@@ -234,3 +230,12 @@ void MacroDirective::dump() const {
   }
   Out << "\n";
 }
+
+ModuleMacro *ModuleMacro::create(Preprocessor &PP, Module *OwningModule,
+                                 IdentifierInfo *II, MacroInfo *Macro,
+                                 ArrayRef<ModuleMacro *> Overrides) {
+  void *Mem = PP.getPreprocessorAllocator().Allocate(
+      sizeof(ModuleMacro) + sizeof(ModuleMacro *) * Overrides.size(),
+      llvm::alignOf<ModuleMacro>());
+  return new (Mem) ModuleMacro(OwningModule, II, Macro, Overrides);
+}
diff --git a/lib/Lex/ModuleMap.cpp b/lib/Lex/ModuleMap.cpp
index ef322d8cdc4c..412918362d32 100644
--- a/lib/Lex/ModuleMap.cpp
+++ b/lib/Lex/ModuleMap.cpp
@@ -89,7 +89,9 @@ ModuleMap::ModuleMap(SourceManager &SourceMgr, DiagnosticsEngine &Diags,
                      HeaderSearch &HeaderInfo)
     : SourceMgr(SourceMgr), Diags(Diags), LangOpts(LangOpts), Target(Target),
       HeaderInfo(HeaderInfo), BuiltinIncludeDir(nullptr),
-      CompilingModule(nullptr), SourceModule(nullptr) {}
+      CompilingModule(nullptr), SourceModule(nullptr), NumCreatedModules(0) {
+  MMapLangOpts.LineComment = true;
+}
 
 ModuleMap::~ModuleMap() {
   for (llvm::StringMap<Module *>::iterator I = Modules.begin(), 
@@ -203,35 +205,32 @@ ModuleMap::findHeaderInUmbrellaDirs(const FileEntry *File,
   return KnownHeader();
 }
 
-// Returns true if RequestingModule directly uses RequestedModule.
-static bool directlyUses(const Module *RequestingModule,
-                         const Module *RequestedModule) {
-  return std::find(RequestingModule->DirectUses.begin(),
-                   RequestingModule->DirectUses.end(),
-                   RequestedModule) != RequestingModule->DirectUses.end();
-}
-
 static bool violatesPrivateInclude(Module *RequestingModule,
                                    const FileEntry *IncFileEnt,
                                    ModuleMap::ModuleHeaderRole Role,
                                    Module *RequestedModule) {
   bool IsPrivateRole = Role & ModuleMap::PrivateHeader;
 #ifndef NDEBUG
-  // Check for consistency between the module header role
-  // as obtained from the lookup and as obtained from the module.
-  // This check is not cheap, so enable it only for debugging.
-  bool IsPrivate = false;
-  SmallVectorImpl<Module::Header> *HeaderList[] =
-      {&RequestedModule->Headers[Module::HK_Private],
-       &RequestedModule->Headers[Module::HK_PrivateTextual]};
-  for (auto *Hdrs : HeaderList)
-    IsPrivate |=
-        std::find_if(Hdrs->begin(), Hdrs->end(), [&](const Module::Header &H) {
-          return H.Entry == IncFileEnt;
-        }) != Hdrs->end();
-  assert(IsPrivate == IsPrivateRole && "inconsistent headers and roles");
+  if (IsPrivateRole) {
+    // Check for consistency between the module header role
+    // as obtained from the lookup and as obtained from the module.
+    // This check is not cheap, so enable it only for debugging.
+    bool IsPrivate = false;
+    SmallVectorImpl<Module::Header> *HeaderList[] = {
+        &RequestedModule->Headers[Module::HK_Private],
+        &RequestedModule->Headers[Module::HK_PrivateTextual]};
+    for (auto *Hs : HeaderList)
+      IsPrivate |=
+          std::find_if(Hs->begin(), Hs->end(), [&](const Module::Header &H) {
+            return H.Entry == IncFileEnt;
+          }) != Hs->end();
+    assert((!IsPrivateRole || IsPrivate) && "inconsistent headers and roles");
+  }
 #endif
   return IsPrivateRole &&
+         // FIXME: Should we map RequestingModule to its top-level module here
+         //        too? This check is redundant with the isSubModuleOf check in
+         //        diagnoseHeaderInclusion.
          RequestedModule->getTopLevelModule() != RequestingModule;
 }
 
@@ -259,7 +258,8 @@ void ModuleMap::diagnoseHeaderInclusion(Module *RequestingModule,
   if (Known != Headers.end()) {
     for (const KnownHeader &Header : Known->second) {
       // If 'File' is part of 'RequestingModule' we can definitely include it.
-      if (Header.getModule() == RequestingModule)
+      if (Header.getModule() &&
+          Header.getModule()->isSubModuleOf(RequestingModule))
         return;
 
       // Remember private headers for later printing of a diagnostic.
@@ -272,7 +272,7 @@ void ModuleMap::diagnoseHeaderInclusion(Module *RequestingModule,
       // If uses need to be specified explicitly, we are only allowed to return
       // modules that are explicitly used by the requesting module.
       if (RequestingModule && LangOpts.ModulesDeclUse &&
-          !directlyUses(RequestingModule, Header.getModule())) {
+          !RequestingModule->directlyUses(Header.getModule())) {
         NotUsed = Header.getModule();
         continue;
       }
@@ -286,14 +286,14 @@ void ModuleMap::diagnoseHeaderInclusion(Module *RequestingModule,
 
   // We have found a header, but it is private.
   if (Private) {
-    Diags.Report(FilenameLoc, diag::error_use_of_private_header_outside_module)
+    Diags.Report(FilenameLoc, diag::warn_use_of_private_header_outside_module)
         << Filename;
     return;
   }
 
   // We have found a module, but we don't use it.
   if (NotUsed) {
-    Diags.Report(FilenameLoc, diag::error_undeclared_use_of_module)
+    Diags.Report(FilenameLoc, diag::err_undeclared_use_of_module)
         << RequestingModule->getFullModuleName() << Filename;
     return;
   }
@@ -304,7 +304,7 @@ void ModuleMap::diagnoseHeaderInclusion(Module *RequestingModule,
   // At this point, only non-modular includes remain.
 
   if (LangOpts.ModulesStrictDeclUse) {
-    Diags.Report(FilenameLoc, diag::error_undeclared_use_of_module)
+    Diags.Report(FilenameLoc, diag::err_undeclared_use_of_module)
         << RequestingModule->getFullModuleName() << Filename;
   } else if (RequestingModule) {
     diag::kind DiagID = RequestingModule->getTopLevelModule()->IsFramework ?
@@ -314,6 +314,22 @@ void ModuleMap::diagnoseHeaderInclusion(Module *RequestingModule,
   }
 }
 
+static bool isBetterKnownHeader(const ModuleMap::KnownHeader &New,
+                                const ModuleMap::KnownHeader &Old) {
+  // Prefer a public header over a private header.
+  if ((New.getRole() & ModuleMap::PrivateHeader) !=
+      (Old.getRole() & ModuleMap::PrivateHeader))
+    return !(New.getRole() & ModuleMap::PrivateHeader);
+
+  // Prefer a non-textual header over a textual header.
+  if ((New.getRole() & ModuleMap::TextualHeader) !=
+      (Old.getRole() & ModuleMap::TextualHeader))
+    return !(New.getRole() & ModuleMap::TextualHeader);
+
+  // Don't have a reason to choose between these. Just keep the first one.
+  return false;
+}
+
 ModuleMap::KnownHeader
 ModuleMap::findModuleForHeader(const FileEntry *File,
                                Module *RequestingModule,
@@ -345,11 +361,10 @@ ModuleMap::findModuleForHeader(const FileEntry *File,
       // If uses need to be specified explicitly, we are only allowed to return
       // modules that are explicitly used by the requesting module.
       if (RequestingModule && LangOpts.ModulesDeclUse &&
-          !directlyUses(RequestingModule, I->getModule()))
+          !RequestingModule->directlyUses(I->getModule()))
         continue;
 
-      // Prefer a public header over a private header.
-      if (!Result || (Result.getRole() & ModuleMap::PrivateHeader))
+      if (!Result || isBetterKnownHeader(*I, Result))
         Result = *I;
     }
     return MakeResult(Result);
@@ -548,7 +563,7 @@ ModuleMap::findOrCreateModule(StringRef Name, Module *Parent, bool IsFramework,
   
   // Create a new module with this name.
   Module *Result = new Module(Name, SourceLocation(), Parent,
-                              IsFramework, IsExplicit);
+                              IsFramework, IsExplicit, NumCreatedModules++);
   if (LangOpts.CurrentModule == Name) {
     SourceModule = Result;
     SourceModuleName = Name;
@@ -678,7 +693,8 @@ Module *ModuleMap::inferFrameworkModule(StringRef ModuleName,
     return nullptr;
 
   Module *Result = new Module(ModuleName, SourceLocation(), Parent,
-                              /*IsFramework=*/true, /*IsExplicit=*/false);
+                              /*IsFramework=*/true, /*IsExplicit=*/false,
+                              NumCreatedModules++);
   InferredModuleAllowedBy[Result] = ModuleMapFile;
   Result->IsInferred = true;
   if (LangOpts.CurrentModule == ModuleName) {
@@ -689,14 +705,16 @@ Module *ModuleMap::inferFrameworkModule(StringRef ModuleName,
   Result->IsSystem |= Attrs.IsSystem;
   Result->IsExternC |= Attrs.IsExternC;
   Result->ConfigMacrosExhaustive |= Attrs.IsExhaustive;
+  Result->Directory = FrameworkDir;
 
   if (!Parent)
     Modules[ModuleName] = Result;
   
   // umbrella header "umbrella-header-name"
-  Result->Umbrella = UmbrellaHeader;
-  Headers[UmbrellaHeader].push_back(KnownHeader(Result, NormalHeader));
-  UmbrellaDirs[UmbrellaHeader->getDir()] = Result;
+  //
+  // The "Headers/" component of the name is implied because this is
+  // a framework module.
+  setUmbrellaHeader(Result, UmbrellaHeader, ModuleName + ".h");
   
   // export *
   Result->Exports.push_back(Module::ExportDecl(nullptr, true));
@@ -711,8 +729,7 @@ Module *ModuleMap::inferFrameworkModule(StringRef ModuleName,
     = StringRef(FrameworkDir->getName());
   llvm::sys::path::append(SubframeworksDirName, "Frameworks");
   llvm::sys::path::native(SubframeworksDirName);
-  for (llvm::sys::fs::directory_iterator 
-         Dir(SubframeworksDirName.str(), EC), DirEnd;
+  for (llvm::sys::fs::directory_iterator Dir(SubframeworksDirName, EC), DirEnd;
        Dir != DirEnd && !EC; Dir.increment(EC)) {
     if (!StringRef(Dir->path()).endswith(".framework"))
       continue;
@@ -758,14 +775,18 @@ Module *ModuleMap::inferFrameworkModule(StringRef ModuleName,
   return Result;
 }
 
-void ModuleMap::setUmbrellaHeader(Module *Mod, const FileEntry *UmbrellaHeader){
+void ModuleMap::setUmbrellaHeader(Module *Mod, const FileEntry *UmbrellaHeader,
+                                  Twine NameAsWritten) {
   Headers[UmbrellaHeader].push_back(KnownHeader(Mod, NormalHeader));
   Mod->Umbrella = UmbrellaHeader;
+  Mod->UmbrellaAsWritten = NameAsWritten.str();
   UmbrellaDirs[UmbrellaHeader->getDir()] = Mod;
 }
 
-void ModuleMap::setUmbrellaDir(Module *Mod, const DirectoryEntry *UmbrellaDir) {
+void ModuleMap::setUmbrellaDir(Module *Mod, const DirectoryEntry *UmbrellaDir,
+                               Twine NameAsWritten) {
   Mod->Umbrella = UmbrellaDir;
+  Mod->UmbrellaAsWritten = NameAsWritten.str();
   UmbrellaDirs[UmbrellaDir] = Mod;
 }
 
@@ -850,50 +871,44 @@ void ModuleMap::dump() {
 }
 
 bool ModuleMap::resolveExports(Module *Mod, bool Complain) {
-  bool HadError = false;
-  for (unsigned I = 0, N = Mod->UnresolvedExports.size(); I != N; ++I) {
-    Module::ExportDecl Export = resolveExport(Mod, Mod->UnresolvedExports[I], 
-                                              Complain);
+  auto Unresolved = std::move(Mod->UnresolvedExports);
+  Mod->UnresolvedExports.clear();
+  for (auto &UE : Unresolved) {
+    Module::ExportDecl Export = resolveExport(Mod, UE, Complain);
     if (Export.getPointer() || Export.getInt())
       Mod->Exports.push_back(Export);
     else
-      HadError = true;
+      Mod->UnresolvedExports.push_back(UE);
   }
-  Mod->UnresolvedExports.clear();
-  return HadError;
+  return !Mod->UnresolvedExports.empty();
 }
 
 bool ModuleMap::resolveUses(Module *Mod, bool Complain) {
-  bool HadError = false;
-  for (unsigned I = 0, N = Mod->UnresolvedDirectUses.size(); I != N; ++I) {
-    Module *DirectUse =
-        resolveModuleId(Mod->UnresolvedDirectUses[I], Mod, Complain);
+  auto Unresolved = std::move(Mod->UnresolvedDirectUses);
+  Mod->UnresolvedDirectUses.clear();
+  for (auto &UDU : Unresolved) {
+    Module *DirectUse = resolveModuleId(UDU, Mod, Complain);
     if (DirectUse)
       Mod->DirectUses.push_back(DirectUse);
     else
-      HadError = true;
+      Mod->UnresolvedDirectUses.push_back(UDU);
   }
-  Mod->UnresolvedDirectUses.clear();
-  return HadError;
+  return !Mod->UnresolvedDirectUses.empty();
 }
 
 bool ModuleMap::resolveConflicts(Module *Mod, bool Complain) {
-  bool HadError = false;
-  for (unsigned I = 0, N = Mod->UnresolvedConflicts.size(); I != N; ++I) {
-    Module *OtherMod = resolveModuleId(Mod->UnresolvedConflicts[I].Id,
-                                       Mod, Complain);
-    if (!OtherMod) {
-      HadError = true;
-      continue;
-    }
-
-    Module::Conflict Conflict;
-    Conflict.Other = OtherMod;
-    Conflict.Message = Mod->UnresolvedConflicts[I].Message;
-    Mod->Conflicts.push_back(Conflict);
-  }
+  auto Unresolved = std::move(Mod->UnresolvedConflicts);
   Mod->UnresolvedConflicts.clear();
-  return HadError;
+  for (auto &UC : Unresolved) {
+    if (Module *OtherMod = resolveModuleId(UC.Id, Mod, Complain)) {
+      Module::Conflict Conflict;
+      Conflict.Other = OtherMod;
+      Conflict.Message = UC.Message;
+      Mod->Conflicts.push_back(Conflict);
+    } else
+      Mod->UnresolvedConflicts.push_back(UC);
+  }
+  return !Mod->UnresolvedConflicts.empty();
 }
 
 Module *ModuleMap::inferModuleFromLocation(FullSourceLoc Loc) {
@@ -1565,7 +1580,7 @@ void ModuleMapParser::parseExternModuleDecl() {
   if (llvm::sys::path::is_relative(FileNameRef)) {
     ModuleMapFileName += Directory->getName();
     llvm::sys::path::append(ModuleMapFileName, FileName);
-    FileNameRef = ModuleMapFileName.str();
+    FileNameRef = ModuleMapFileName;
   }
   if (const FileEntry *File = SourceMgr.getFileManager().getFile(FileNameRef))
     Map.parseModuleMapFile(
@@ -1712,7 +1727,7 @@ void ModuleMapParser::parseHeaderDecl(MMToken::TokenKind LeadingToken,
       
       // Check whether this file is in the public headers.
       llvm::sys::path::append(RelativePathName, "Headers", Header.FileName);
-      llvm::sys::path::append(FullPathName, RelativePathName.str());
+      llvm::sys::path::append(FullPathName, RelativePathName);
       File = SourceMgr.getFileManager().getFile(FullPathName);
       
       if (!File) {
@@ -1722,13 +1737,13 @@ void ModuleMapParser::parseHeaderDecl(MMToken::TokenKind LeadingToken,
         FullPathName.resize(FullPathLength);
         llvm::sys::path::append(RelativePathName, "PrivateHeaders",
                                 Header.FileName);
-        llvm::sys::path::append(FullPathName, RelativePathName.str());
+        llvm::sys::path::append(FullPathName, RelativePathName);
         File = SourceMgr.getFileManager().getFile(FullPathName);
       }
     } else {
       // Lookup for normal headers.
       llvm::sys::path::append(RelativePathName, Header.FileName);
-      llvm::sys::path::append(FullPathName, RelativePathName.str());
+      llvm::sys::path::append(FullPathName, RelativePathName);
       File = SourceMgr.getFileManager().getFile(FullPathName);
 
       // If this is a system module with a top-level header, this header
@@ -1744,7 +1759,13 @@ void ModuleMapParser::parseHeaderDecl(MMToken::TokenKind LeadingToken,
         // If Clang supplies this header but the underlying system does not,
         // just silently swap in our builtin version. Otherwise, we'll end
         // up adding both (later).
-        if (!File && BuiltinFile) {
+        //
+        // For local visibility, entirely replace the system file with our
+        // one and textually include the system one. We need to pass macros
+        // from our header to the system one if we #include_next it.
+        //
+        // FIXME: Can we do this in all cases?
+        if (BuiltinFile && (!File || Map.LangOpts.ModulesLocalVisibility)) {
           File = BuiltinFile;
           RelativePathName = BuiltinPathName;
           BuiltinFile = nullptr;
@@ -1764,7 +1785,7 @@ void ModuleMapParser::parseHeaderDecl(MMToken::TokenKind LeadingToken,
         HadError = true;
       } else {
         // Record this umbrella header.
-        Map.setUmbrellaHeader(ActiveModule, File);
+        Map.setUmbrellaHeader(ActiveModule, File, RelativePathName.str());
       }
     } else if (LeadingToken == MMToken::ExcludeKeyword) {
       Module::Header H = {RelativePathName.str(), File};
@@ -1846,7 +1867,7 @@ void ModuleMapParser::parseUmbrellaDirDecl(SourceLocation UmbrellaLoc) {
   } 
   
   // Record this umbrella directory.
-  Map.setUmbrellaDir(ActiveModule, Dir);
+  Map.setUmbrellaDir(ActiveModule, Dir, DirName);
 }
 
 /// \brief Parse a module export declaration.
@@ -1897,18 +1918,21 @@ void ModuleMapParser::parseExportDecl() {
   ActiveModule->UnresolvedExports.push_back(Unresolved);
 }
 
-/// \brief Parse a module uses declaration.
+/// \brief Parse a module use declaration.
 ///
-///   uses-declaration:
-///     'uses' wildcard-module-id
+///   use-declaration:
+///     'use' wildcard-module-id
 void ModuleMapParser::parseUseDecl() {
   assert(Tok.is(MMToken::UseKeyword));
-  consumeToken();
+  auto KWLoc = consumeToken();
   // Parse the module-id.
   ModuleId ParsedModuleId;
   parseModuleId(ParsedModuleId);
 
-  ActiveModule->UnresolvedDirectUses.push_back(ParsedModuleId);
+  if (ActiveModule->Parent)
+    Diags.Report(KWLoc, diag::err_mmap_use_decl_submodule);
+  else
+    ActiveModule->UnresolvedDirectUses.push_back(ParsedModuleId);
 }
 
 /// \brief Parse a link declaration.
diff --git a/lib/Lex/PPConditionalDirectiveRecord.cpp b/lib/Lex/PPConditionalDirectiveRecord.cpp
index 99b87a0a1525..12a77849b8b3 100644
--- a/lib/Lex/PPConditionalDirectiveRecord.cpp
+++ b/lib/Lex/PPConditionalDirectiveRecord.cpp
@@ -84,14 +84,14 @@ void PPConditionalDirectiveRecord::If(SourceLocation Loc,
 
 void PPConditionalDirectiveRecord::Ifdef(SourceLocation Loc,
                                          const Token &MacroNameTok,
-                                         const MacroDirective *MD) {
+                                         const MacroDefinition &MD) {
   addCondDirectiveLoc(CondDirectiveLoc(Loc, CondDirectiveStack.back()));
   CondDirectiveStack.push_back(Loc);
 }
 
 void PPConditionalDirectiveRecord::Ifndef(SourceLocation Loc,
                                           const Token &MacroNameTok,
-                                          const MacroDirective *MD) {
+                                          const MacroDefinition &MD) {
   addCondDirectiveLoc(CondDirectiveLoc(Loc, CondDirectiveStack.back()));
   CondDirectiveStack.push_back(Loc);
 }
diff --git a/lib/Lex/PPDirectives.cpp b/lib/Lex/PPDirectives.cpp
index bf0ce72f6668..ec06e790f018 100644
--- a/lib/Lex/PPDirectives.cpp
+++ b/lib/Lex/PPDirectives.cpp
@@ -62,26 +62,14 @@ MacroInfo *Preprocessor::AllocateDeserializedMacroInfo(SourceLocation L,
   return MI;
 }
 
-DefMacroDirective *
-Preprocessor::AllocateDefMacroDirective(MacroInfo *MI, SourceLocation Loc,
-                                        unsigned ImportedFromModuleID,
-                                        ArrayRef<unsigned> Overrides) {
-  unsigned NumExtra = (ImportedFromModuleID ? 1 : 0) + Overrides.size();
-  return new (BP.Allocate(sizeof(DefMacroDirective) +
-                              sizeof(unsigned) * NumExtra,
-                          llvm::alignOf<DefMacroDirective>()))
-      DefMacroDirective(MI, Loc, ImportedFromModuleID, Overrides);
+DefMacroDirective *Preprocessor::AllocateDefMacroDirective(MacroInfo *MI,
+                                                           SourceLocation Loc) {
+  return new (BP) DefMacroDirective(MI, Loc);
 }
 
 UndefMacroDirective *
-Preprocessor::AllocateUndefMacroDirective(SourceLocation UndefLoc,
-                                          unsigned ImportedFromModuleID,
-                                          ArrayRef<unsigned> Overrides) {
-  unsigned NumExtra = (ImportedFromModuleID ? 1 : 0) + Overrides.size();
-  return new (BP.Allocate(sizeof(UndefMacroDirective) +
-                              sizeof(unsigned) * NumExtra,
-                          llvm::alignOf<UndefMacroDirective>()))
-      UndefMacroDirective(UndefLoc, ImportedFromModuleID, Overrides);
+Preprocessor::AllocateUndefMacroDirective(SourceLocation UndefLoc) {
+  return new (BP) UndefMacroDirective(UndefLoc);
 }
 
 VisibilityMacroDirective *
@@ -182,11 +170,13 @@ bool Preprocessor::CheckMacroName(Token &MacroNameTok, MacroUse isDefineUndef,
     return Diag(MacroNameTok, diag::err_defined_macro_name);
   }
 
-  if (isDefineUndef == MU_Undef && II->hasMacroDefinition() &&
-      getMacroInfo(II)->isBuiltinMacro()) {
-    // Warn if undefining "__LINE__" and other builtins, per C99 6.10.8/4
-    // and C++ [cpp.predefined]p4], but allow it as an extension.
-    Diag(MacroNameTok, diag::ext_pp_undef_builtin_macro);
+  if (isDefineUndef == MU_Undef) {
+    auto *MI = getMacroInfo(II);
+    if (MI && MI->isBuiltinMacro()) {
+      // Warn if undefining "__LINE__" and other builtins, per C99 6.10.8/4
+      // and C++ [cpp.predefined]p4], but allow it as an extension.
+      Diag(MacroNameTok, diag::ext_pp_undef_builtin_macro);
+    }
   }
 
   // If defining/undefining reserved identifier or a keyword, we need to issue
@@ -585,16 +575,16 @@ void Preprocessor::PTHSkipExcludedConditionalBlock() {
   }
 }
 
-Module *Preprocessor::getModuleForLocation(SourceLocation FilenameLoc) {
+Module *Preprocessor::getModuleForLocation(SourceLocation Loc) {
   ModuleMap &ModMap = HeaderInfo.getModuleMap();
-  if (SourceMgr.isInMainFile(FilenameLoc)) {
+  if (SourceMgr.isInMainFile(Loc)) {
     if (Module *CurMod = getCurrentModule())
       return CurMod;                               // Compiling a module.
     return HeaderInfo.getModuleMap().SourceModule; // Compiling a source.
   }
   // Try to determine the module of the include directive.
   // FIXME: Look into directly passing the FileEntry from LookupFile instead.
-  FileID IDOfIncl = SourceMgr.getFileID(SourceMgr.getExpansionLoc(FilenameLoc));
+  FileID IDOfIncl = SourceMgr.getFileID(SourceMgr.getExpansionLoc(Loc));
   if (const FileEntry *EntryOfIncl = SourceMgr.getFileEntryForID(IDOfIncl)) {
     // The include comes from a file.
     return ModMap.findModuleForHeader(EntryOfIncl).getModule();
@@ -605,6 +595,11 @@ Module *Preprocessor::getModuleForLocation(SourceLocation FilenameLoc) {
   }
 }
 
+Module *Preprocessor::getModuleContainingLocation(SourceLocation Loc) {
+  return HeaderInfo.getModuleMap().inferModuleFromLocation(
+      FullSourceLoc(Loc, SourceMgr));
+}
+
 const FileEntry *Preprocessor::LookupFile(
     SourceLocation FilenameLoc,
     StringRef Filename,
@@ -1233,7 +1228,7 @@ void Preprocessor::HandleUserDiagnosticDirective(Token &Tok,
 
   // Find the first non-whitespace character, so that we can make the
   // diagnostic more succinct.
-  StringRef Msg = Message.str().ltrim(" ");
+  StringRef Msg = StringRef(Message).ltrim(" ");
 
   if (isWarning)
     Diag(Tok, diag::pp_hash_warning) << Msg;
@@ -1290,7 +1285,7 @@ void Preprocessor::HandleMacroPublicDirective(Token &Tok) {
 
   IdentifierInfo *II = MacroNameTok.getIdentifierInfo();
   // Okay, we finally have a valid identifier to undef.
-  MacroDirective *MD = getMacroDirective(II);
+  MacroDirective *MD = getLocalMacroDirective(II);
   
   // If the macro is not defined, this is an error.
   if (!MD) {
@@ -1317,7 +1312,7 @@ void Preprocessor::HandleMacroPrivateDirective(Token &Tok) {
   
   IdentifierInfo *II = MacroNameTok.getIdentifierInfo();
   // Okay, we finally have a valid identifier to undef.
-  MacroDirective *MD = getMacroDirective(II);
+  MacroDirective *MD = getLocalMacroDirective(II);
   
   // If the macro is not defined, this is an error.
   if (!MD) {
@@ -1444,6 +1439,8 @@ bool Preprocessor::ConcatenateIncludeName(SmallString<128> &FilenameBuffer,
 static void EnterAnnotationToken(Preprocessor &PP,
                                  SourceLocation Begin, SourceLocation End,
                                  tok::TokenKind Kind, void *AnnotationVal) {
+  // FIXME: Produce this as the current token directly, rather than
+  // allocating a new token for it.
   Token *Tok = new Token[1];
   Tok[0].startToken();
   Tok[0].setKind(Kind);
@@ -1453,6 +1450,51 @@ static void EnterAnnotationToken(Preprocessor &PP,
   PP.EnterTokenStream(Tok, 1, true, true);
 }
 
+/// \brief Produce a diagnostic informing the user that a #include or similar
+/// was implicitly treated as a module import.
+static void diagnoseAutoModuleImport(
+    Preprocessor &PP, SourceLocation HashLoc, Token &IncludeTok,
+    ArrayRef<std::pair<IdentifierInfo *, SourceLocation>> Path,
+    SourceLocation PathEnd) {
+  assert(PP.getLangOpts().ObjC2 && "no import syntax available");
+
+  SmallString<128> PathString;
+  for (unsigned I = 0, N = Path.size(); I != N; ++I) {
+    if (I)
+      PathString += '.';
+    PathString += Path[I].first->getName();
+  }
+  int IncludeKind = 0;
+  
+  switch (IncludeTok.getIdentifierInfo()->getPPKeywordID()) {
+  case tok::pp_include:
+    IncludeKind = 0;
+    break;
+    
+  case tok::pp_import:
+    IncludeKind = 1;
+    break;        
+      
+  case tok::pp_include_next:
+    IncludeKind = 2;
+    break;
+      
+  case tok::pp___include_macros:
+    IncludeKind = 3;
+    break;
+      
+  default:
+    llvm_unreachable("unknown include directive kind");
+  }
+
+  CharSourceRange ReplaceRange(SourceRange(HashLoc, PathEnd),
+                               /*IsTokenRange=*/false);
+  PP.Diag(HashLoc, diag::warn_auto_module_import)
+      << IncludeKind << PathString
+      << FixItHint::CreateReplacement(ReplaceRange,
+                                      ("@import " + PathString + ";").str());
+}
+
 /// HandleIncludeDirective - The "\#include" tokens have just been read, read
 /// the file to be included from the lexer, then include it!  This is a common
 /// routine with functionality shared between \#include, \#include_next and
@@ -1491,7 +1533,7 @@ void Preprocessor::HandleIncludeDirective(SourceLocation HashLoc,
     FilenameBuffer.push_back('<');
     if (ConcatenateIncludeName(FilenameBuffer, End))
       return;   // Found <eod> but no ">"?  Diagnostic already emitted.
-    Filename = FilenameBuffer.str();
+    Filename = FilenameBuffer;
     CharEnd = End.getLocWithOffset(1);
     break;
   default:
@@ -1563,8 +1605,8 @@ void Preprocessor::HandleIncludeDirective(SourceLocation HashLoc,
       Callbacks ? &SearchPath : nullptr, Callbacks ? &RelativePath : nullptr,
       HeaderInfo.getHeaderSearchOpts().ModuleMaps ? &SuggestedModule : nullptr);
 
-  if (Callbacks) {
-    if (!File) {
+  if (!File) {
+    if (Callbacks) {
       // Give the clients a chance to recover.
       SmallString<128> RecoveryPath;
       if (Callbacks->FileNotFound(Filename, RecoveryPath)) {
@@ -1584,18 +1626,7 @@ void Preprocessor::HandleIncludeDirective(SourceLocation HashLoc,
         }
       }
     }
-    
-    if (!SuggestedModule || !getLangOpts().Modules) {
-      // Notify the callback object that we've seen an inclusion directive.
-      Callbacks->InclusionDirective(HashLoc, IncludeTok,
-                                    LangOpts.MSVCCompat ? NormalizedPath.c_str()
-                                                        : Filename,
-                                    isAngled, FilenameRange, File, SearchPath,
-                                    RelativePath, /*ImportedModule=*/nullptr);
-    }
-  }
 
-  if (!File) {
     if (!SuppressIncludeNotFoundError) {
       // If the file could not be located and it was included via angle 
       // brackets, we can attempt a lookup as though it were a quoted path to
@@ -1616,19 +1647,27 @@ void Preprocessor::HandleIncludeDirective(SourceLocation HashLoc,
             FixItHint::CreateReplacement(Range, "\"" + Filename.str() + "\"");
         }
       }
+
       // If the file is still not found, just go with the vanilla diagnostic
       if (!File)
         Diag(FilenameTok, diag::err_pp_file_not_found) << Filename;
     }
-    if (!File)
-      return;
   }
 
-  // If we are supposed to import a module rather than including the header,
-  // do so now.
-  if (SuggestedModule && getLangOpts().Modules &&
+  // Should we enter the source file? Set to false if either the source file is
+  // known to have no effect beyond its effect on module visibility -- that is,
+  // if it's got an include guard that is already defined or is a modular header
+  // we've imported or already built.
+  bool ShouldEnter = true;
+
+  // Determine whether we should try to import the module for this #include, if
+  // there is one. Don't do so if precompiled module support is disabled or we
+  // are processing this module textually (because we're building the module).
+  if (File && SuggestedModule && getLangOpts().Modules &&
       SuggestedModule.getModule()->getTopLevelModuleName() !=
-      getLangOpts().ImplementationOfModule) {
+          getLangOpts().CurrentModule &&
+      SuggestedModule.getModule()->getTopLevelModuleName() !=
+          getLangOpts().ImplementationOfModule) {
     // Compute the module access path corresponding to this module.
     // FIXME: Should we have a second loadModule() overload to avoid this
     // extra lookup step?
@@ -1639,111 +1678,57 @@ void Preprocessor::HandleIncludeDirective(SourceLocation HashLoc,
     std::reverse(Path.begin(), Path.end());
 
     // Warn that we're replacing the include/import with a module import.
-    SmallString<128> PathString;
-    for (unsigned I = 0, N = Path.size(); I != N; ++I) {
-      if (I)
-        PathString += '.';
-      PathString += Path[I].first->getName();
-    }
-    int IncludeKind = 0;
-    
-    switch (IncludeTok.getIdentifierInfo()->getPPKeywordID()) {
-    case tok::pp_include:
-      IncludeKind = 0;
-      break;
-      
-    case tok::pp_import:
-      IncludeKind = 1;
-      break;        
-        
-    case tok::pp_include_next:
-      IncludeKind = 2;
-      break;
-        
-    case tok::pp___include_macros:
-      IncludeKind = 3;
-      break;
-        
-    default:
-      llvm_unreachable("unknown include directive kind");
-    }
-
-    // Determine whether we are actually building the module that this
-    // include directive maps to.
-    bool BuildingImportedModule
-      = Path[0].first->getName() == getLangOpts().CurrentModule;
-
-    if (!BuildingImportedModule && getLangOpts().ObjC2) {
-      // If we're not building the imported module, warn that we're going
-      // to automatically turn this inclusion directive into a module import.
-      // We only do this in Objective-C, where we have a module-import syntax.
-      CharSourceRange ReplaceRange(SourceRange(HashLoc, CharEnd), 
-                                   /*IsTokenRange=*/false);
-      Diag(HashLoc, diag::warn_auto_module_import)
-        << IncludeKind << PathString 
-        << FixItHint::CreateReplacement(ReplaceRange,
-             "@import " + PathString.str().str() + ";");
-    }
+    // We only do this in Objective-C, where we have a module-import syntax.
+    if (getLangOpts().ObjC2)
+      diagnoseAutoModuleImport(*this, HashLoc, IncludeTok, Path, CharEnd);
     
-    // Load the module. Only make macros visible. We'll make the declarations
+    // Load the module to import its macros. We'll make the declarations
     // visible when the parser gets here.
-    Module::NameVisibilityKind Visibility = Module::MacrosVisible;
-    ModuleLoadResult Imported
-      = TheModuleLoader.loadModule(IncludeTok.getLocation(), Path, Visibility,
-                                   /*IsIncludeDirective=*/true);
+    // FIXME: Pass SuggestedModule in here rather than converting it to a path
+    // and making the module loader convert it back again.
+    ModuleLoadResult Imported = TheModuleLoader.loadModule(
+        IncludeTok.getLocation(), Path, Module::Hidden,
+        /*IsIncludeDirective=*/true);
     assert((Imported == nullptr || Imported == SuggestedModule.getModule()) &&
            "the imported module is different than the suggested one");
 
-    if (!Imported && hadModuleLoaderFatalFailure()) {
-      // With a fatal failure in the module loader, we abort parsing.
-      Token &Result = IncludeTok;
-      if (CurLexer) {
-        Result.startToken();
-        CurLexer->FormTokenWithChars(Result, CurLexer->BufferEnd, tok::eof);
-        CurLexer->cutOffLexing();
-      } else {
-        assert(CurPTHLexer && "#include but no current lexer set!");
-        CurPTHLexer->getEOF(Result);
-      }
-      return;
-    }
-
-    // If this header isn't part of the module we're building, we're done.
-    if (!BuildingImportedModule && Imported) {
-      if (Callbacks) {
-        Callbacks->InclusionDirective(HashLoc, IncludeTok, Filename, isAngled,
-                                      FilenameRange, File,
-                                      SearchPath, RelativePath, Imported);
-      }
-
-      if (IncludeKind != 3) {
-        // Let the parser know that we hit a module import, and it should
-        // make the module visible.
-        // FIXME: Produce this as the current token directly, rather than
-        // allocating a new token for it.
-        EnterAnnotationToken(*this, HashLoc, End, tok::annot_module_include,
-                             Imported);
+    if (Imported)
+      ShouldEnter = false;
+    else if (Imported.isMissingExpected()) {
+      // We failed to find a submodule that we assumed would exist (because it
+      // was in the directory of an umbrella header, for instance), but no
+      // actual module exists for it (because the umbrella header is
+      // incomplete).  Treat this as a textual inclusion.
+      SuggestedModule = ModuleMap::KnownHeader();
+    } else {
+      // We hit an error processing the import. Bail out.
+      if (hadModuleLoaderFatalFailure()) {
+        // With a fatal failure in the module loader, we abort parsing.
+        Token &Result = IncludeTok;
+        if (CurLexer) {
+          Result.startToken();
+          CurLexer->FormTokenWithChars(Result, CurLexer->BufferEnd, tok::eof);
+          CurLexer->cutOffLexing();
+        } else {
+          assert(CurPTHLexer && "#include but no current lexer set!");
+          CurPTHLexer->getEOF(Result);
+        }
       }
       return;
     }
-
-    // If we failed to find a submodule that we expected to find, we can
-    // continue. Otherwise, there's an error in the included file, so we
-    // don't want to include it.
-    if (!BuildingImportedModule && !Imported.isMissingExpected()) {
-      return;
-    }
   }
 
-  if (Callbacks && SuggestedModule) {
-    // We didn't notify the callback object that we've seen an inclusion
-    // directive before. Now that we are parsing the include normally and not
-    // turning it to a module import, notify the callback object.
-    Callbacks->InclusionDirective(HashLoc, IncludeTok, Filename, isAngled,
-                                  FilenameRange, File,
-                                  SearchPath, RelativePath,
-                                  /*ImportedModule=*/nullptr);
+  if (Callbacks) {
+    // Notify the callback object that we've seen an inclusion directive.
+    Callbacks->InclusionDirective(
+        HashLoc, IncludeTok,
+        LangOpts.MSVCCompat ? NormalizedPath.c_str() : Filename, isAngled,
+        FilenameRange, File, SearchPath, RelativePath,
+        ShouldEnter ? nullptr : SuggestedModule.getModule());
   }
+
+  if (!File)
+    return;
   
   // The #included file will be considered to be a system header if either it is
   // in a system include directory, or if the #includer is a system include
@@ -1752,11 +1737,28 @@ void Preprocessor::HandleIncludeDirective(SourceLocation HashLoc,
     std::max(HeaderInfo.getFileDirFlavor(File),
              SourceMgr.getFileCharacteristic(FilenameTok.getLocation()));
 
+  // FIXME: If we have a suggested module, and we've already visited this file,
+  // don't bother entering it again. We know it has no further effect.
+
   // Ask HeaderInfo if we should enter this #include file.  If not, #including
   // this file will have no effect.
-  if (!HeaderInfo.ShouldEnterIncludeFile(File, isImport)) {
+  if (ShouldEnter &&
+      !HeaderInfo.ShouldEnterIncludeFile(*this, File, isImport)) {
+    ShouldEnter = false;
     if (Callbacks)
       Callbacks->FileSkipped(*File, FilenameTok, FileCharacter);
+  }
+
+  // If we don't need to enter the file, stop now.
+  if (!ShouldEnter) {
+    // If this is a module import, make it visible if needed.
+    if (auto *M = SuggestedModule.getModule()) {
+      makeModuleVisible(M, HashLoc);
+
+      if (IncludeTok.getIdentifierInfo()->getPPKeywordID() !=
+          tok::pp___include_macros)
+        EnterAnnotationToken(*this, HashLoc, End, tok::annot_module_include, M);
+    }
     return;
   }
 
@@ -1769,26 +1771,24 @@ void Preprocessor::HandleIncludeDirective(SourceLocation HashLoc,
   FileID FID = SourceMgr.createFileID(File, IncludePos, FileCharacter);
   assert(!FID.isInvalid() && "Expected valid file ID");
 
-  // Determine if we're switching to building a new submodule, and which one.
-  ModuleMap::KnownHeader BuildingModule;
-  if (getLangOpts().Modules && !getLangOpts().CurrentModule.empty()) {
-    Module *RequestingModule = getModuleForLocation(FilenameLoc);
-    BuildingModule =
-        HeaderInfo.getModuleMap().findModuleForHeader(File, RequestingModule);
-  }
-
   // If all is good, enter the new file!
   if (EnterSourceFile(FID, CurDir, FilenameTok.getLocation()))
     return;
 
-  // If we're walking into another part of the same module, let the parser
-  // know that any future declarations are within that other submodule.
-  if (BuildingModule) {
+  // Determine if we're switching to building a new submodule, and which one.
+  if (auto *M = SuggestedModule.getModule()) {
     assert(!CurSubmodule && "should not have marked this as a module yet");
-    CurSubmodule = BuildingModule.getModule();
+    CurSubmodule = M;
 
-    EnterAnnotationToken(*this, HashLoc, End, tok::annot_module_begin,
-                         CurSubmodule);
+    // Let the macro handling code know that any future macros are within
+    // the new submodule.
+    EnterSubmodule(M, HashLoc);
+
+    // Let the parser know that any future declarations are within the new
+    // submodule.
+    // FIXME: There's no point doing this if we're handling a #__include_macros
+    // directive.
+    EnterAnnotationToken(*this, HashLoc, End, tok::annot_module_begin, M);
   }
 }
 
@@ -2290,9 +2290,9 @@ void Preprocessor::HandleUndefDirective(Token &UndefTok) {
   // Check to see if this is the last token on the #undef line.
   CheckEndOfDirective("undef");
 
-  // Okay, we finally have a valid identifier to undef.
-  MacroDirective *MD = getMacroDirective(MacroNameTok.getIdentifierInfo());
-  const MacroInfo *MI = MD ? MD->getMacroInfo() : nullptr;
+  // Okay, we have a valid identifier to undef.
+  auto *II = MacroNameTok.getIdentifierInfo();
+  auto MD = getMacroDefinition(II);
 
   // If the callbacks want to know, tell them about the macro #undef.
   // Note: no matter if the macro was defined or not.
@@ -2300,6 +2300,7 @@ void Preprocessor::HandleUndefDirective(Token &UndefTok) {
     Callbacks->MacroUndefined(MacroNameTok, MD);
 
   // If the macro is not defined, this is a noop undef, just return.
+  const MacroInfo *MI = MD.getMacroInfo();
   if (!MI)
     return;
 
@@ -2344,8 +2345,8 @@ void Preprocessor::HandleIfdefDirective(Token &Result, bool isIfndef,
   CheckEndOfDirective(isIfndef ? "ifndef" : "ifdef");
 
   IdentifierInfo *MII = MacroNameTok.getIdentifierInfo();
-  MacroDirective *MD = getMacroDirective(MII);
-  MacroInfo *MI = MD ? MD->getMacroInfo() : nullptr;
+  auto MD = getMacroDefinition(MII);
+  MacroInfo *MI = MD.getMacroInfo();
 
   if (CurPPLexer->getConditionalStackDepth() == 0) {
     // If the start of a top-level #ifdef and if the macro is not defined,
diff --git a/lib/Lex/PPExpressions.cpp b/lib/Lex/PPExpressions.cpp
index 9cf72cf8f8fb..44513023395d 100644
--- a/lib/Lex/PPExpressions.cpp
+++ b/lib/Lex/PPExpressions.cpp
@@ -108,15 +108,13 @@ static bool EvaluateDefined(PPValue &Result, Token &PeekTok, DefinedTracker &DT,
 
   // Otherwise, we got an identifier, is it defined to something?
   IdentifierInfo *II = PeekTok.getIdentifierInfo();
-  Result.Val = II->hasMacroDefinition();
-  Result.Val.setIsUnsigned(false);  // Result is signed intmax_t.
+  MacroDefinition Macro = PP.getMacroDefinition(II);
+  Result.Val = !!Macro;
+  Result.Val.setIsUnsigned(false); // Result is signed intmax_t.
 
-  MacroDirective *Macro = nullptr;
   // If there is a macro, mark it used.
-  if (Result.Val != 0 && ValueLive) {
-    Macro = PP.getMacroDirective(II);
-    PP.markMacroAsUsed(Macro->getMacroInfo());
-  }
+  if (Result.Val != 0 && ValueLive)
+    PP.markMacroAsUsed(Macro.getMacroInfo());
 
   // Save macro token for callback.
   Token macroToken(PeekTok);
@@ -144,11 +142,7 @@ static bool EvaluateDefined(PPValue &Result, Token &PeekTok, DefinedTracker &DT,
 
   // Invoke the 'defined' callback.
   if (PPCallbacks *Callbacks = PP.getPPCallbacks()) {
-    MacroDirective *MD = Macro;
-    // Pass the MacroInfo for the macro name even if the value is dead.
-    if (!MD && Result.Val != 0)
-      MD = PP.getMacroDirective(II);
-    Callbacks->Defined(macroToken, MD,
+    Callbacks->Defined(macroToken, Macro,
                        SourceRange(beginLoc, PeekTok.getLocation()));
   }
 
@@ -310,7 +304,9 @@ static bool EvaluateValue(PPValue &Result, Token &PeekTok, DefinedTracker &DT,
     // Set the value.
     Val = Literal.getValue();
     // Set the signedness. UTF-16 and UTF-32 are always unsigned
-    if (!Literal.isUTF16() && !Literal.isUTF32())
+    if (Literal.isWide())
+      Val.setIsUnsigned(!TargetInfo::isTypeSigned(TI.getWCharType()));
+    else if (!Literal.isUTF16() && !Literal.isUTF32())
       Val.setIsUnsigned(!PP.getLangOpts().CharIsSigned);
 
     if (Result.Val.getBitWidth() > Val.getBitWidth()) {
@@ -732,8 +728,7 @@ static bool EvaluateDirectiveSubExpr(PPValue &LHS, unsigned MinPrec,
 /// EvaluateDirectiveExpression - Evaluate an integer constant expression that
 /// may occur after a #if or #elif directive.  If the expression is equivalent
 /// to "!defined(X)" return X in IfNDefMacro.
-bool Preprocessor::
-EvaluateDirectiveExpression(IdentifierInfo *&IfNDefMacro) {
+bool Preprocessor::EvaluateDirectiveExpression(IdentifierInfo *&IfNDefMacro) {
   SaveAndRestore<bool> PPDir(ParsingIfOrElifDirective, true);
   // Save the current state of 'DisableMacroExpansion' and reset it to false. If
   // 'DisableMacroExpansion' is true, then we must be in a macro argument list
diff --git a/lib/Lex/PPLexerChange.cpp b/lib/Lex/PPLexerChange.cpp
index fb5e2b05808c..e68fb7df8e88 100644
--- a/lib/Lex/PPLexerChange.cpp
+++ b/lib/Lex/PPLexerChange.cpp
@@ -309,7 +309,7 @@ bool Preprocessor::HandleEndOfFile(Token &Result, bool isEndOfMacro) {
         }
         if (const IdentifierInfo *DefinedMacro =
               CurPPLexer->MIOpt.GetDefinedMacro()) {
-          if (!ControllingMacro->hasMacroDefinition() &&
+          if (!isMacroDefined(ControllingMacro) &&
               DefinedMacro != ControllingMacro &&
               HeaderInfo.FirstTimeLexingFile(FE)) {
 
@@ -400,6 +400,9 @@ bool Preprocessor::HandleEndOfFile(Token &Result, bool isEndOfMacro) {
       CurLexer->FormTokenWithChars(Result, EndPos, tok::annot_module_end);
       Result.setAnnotationEndLoc(Result.getLocation());
       Result.setAnnotationValue(CurSubmodule);
+
+      // We're done with this submodule.
+      LeaveSubmodule();
     }
 
     // We're done with the #included file.
@@ -471,7 +474,7 @@ bool Preprocessor::HandleEndOfFile(Token &Result, bool isEndOfMacro) {
       if (!getDiagnostics().isIgnored(diag::warn_uncovered_module_header,
                                       StartLoc)) {
         ModuleMap &ModMap = getHeaderSearchInfo().getModuleMap();
-        const DirectoryEntry *Dir = Mod->getUmbrellaDir();
+        const DirectoryEntry *Dir = Mod->getUmbrellaDir().Entry;
         vfs::FileSystem &FS = *FileMgr.getVirtualFileSystem();
         std::error_code EC;
         for (vfs::recursive_directory_iterator Entry(FS, Dir->getName(), EC), End;
@@ -605,3 +608,126 @@ void Preprocessor::HandleMicrosoftCommentPaste(Token &Tok) {
   // preprocessor directive mode), so just return EOF as our token.
   assert(!FoundLexer && "Lexer should return EOD before EOF in PP mode");
 }
+
+void Preprocessor::EnterSubmodule(Module *M, SourceLocation ImportLoc) {
+  if (!getLangOpts().ModulesLocalVisibility) {
+    // Just track that we entered this submodule.
+    BuildingSubmoduleStack.push_back(
+        BuildingSubmoduleInfo(M, ImportLoc, CurSubmoduleState));
+    return;
+  }
+
+  // Resolve as much of the module definition as we can now, before we enter
+  // one of its headers.
+  // FIXME: Can we enable Complain here?
+  // FIXME: Can we do this when local visibility is disabled?
+  ModuleMap &ModMap = getHeaderSearchInfo().getModuleMap();
+  ModMap.resolveExports(M, /*Complain=*/false);
+  ModMap.resolveUses(M, /*Complain=*/false);
+  ModMap.resolveConflicts(M, /*Complain=*/false);
+
+  // If this is the first time we've entered this module, set up its state.
+  auto R = Submodules.insert(std::make_pair(M, SubmoduleState()));
+  auto &State = R.first->second;
+  bool FirstTime = R.second;
+  if (FirstTime) {
+    // Determine the set of starting macros for this submodule; take these
+    // from the "null" module (the predefines buffer).
+    auto &StartingMacros = NullSubmoduleState.Macros;
+
+    // Restore to the starting state.
+    // FIXME: Do this lazily, when each macro name is first referenced.
+    for (auto &Macro : StartingMacros) {
+      MacroState MS(Macro.second.getLatest());
+      MS.setOverriddenMacros(*this, Macro.second.getOverriddenMacros());
+      State.Macros.insert(std::make_pair(Macro.first, std::move(MS)));
+    }
+  }
+
+  // Track that we entered this module.
+  BuildingSubmoduleStack.push_back(
+      BuildingSubmoduleInfo(M, ImportLoc, CurSubmoduleState));
+
+  // Switch to this submodule as the current submodule.
+  CurSubmoduleState = &State;
+
+  // This module is visible to itself.
+  if (FirstTime)
+    makeModuleVisible(M, ImportLoc);
+}
+
+void Preprocessor::LeaveSubmodule() {
+  auto &Info = BuildingSubmoduleStack.back();
+
+  Module *LeavingMod = Info.M;
+  SourceLocation ImportLoc = Info.ImportLoc;
+
+  // Create ModuleMacros for any macros defined in this submodule.
+  for (auto &Macro : CurSubmoduleState->Macros) {
+    auto *II = const_cast<IdentifierInfo*>(Macro.first);
+
+    // Find the starting point for the MacroDirective chain in this submodule.
+    MacroDirective *OldMD = nullptr;
+    if (getLangOpts().ModulesLocalVisibility) {
+      // FIXME: It'd be better to start at the state from when we most recently
+      // entered this submodule, but it doesn't really matter.
+      auto &PredefMacros = NullSubmoduleState.Macros;
+      auto PredefMacroIt = PredefMacros.find(Macro.first);
+      if (PredefMacroIt == PredefMacros.end())
+        OldMD = nullptr;
+      else
+        OldMD = PredefMacroIt->second.getLatest();
+    }
+
+    // This module may have exported a new macro. If so, create a ModuleMacro
+    // representing that fact.
+    bool ExplicitlyPublic = false;
+    for (auto *MD = Macro.second.getLatest(); MD != OldMD;
+         MD = MD->getPrevious()) {
+      assert(MD && "broken macro directive chain");
+
+      // Stop on macros defined in other submodules we #included along the way.
+      // There's no point doing this if we're tracking local submodule
+      // visibility, since there can be no such directives in our list.
+      if (!getLangOpts().ModulesLocalVisibility) {
+        Module *Mod = getModuleContainingLocation(MD->getLocation());
+        if (Mod != LeavingMod)
+          break;
+      }
+
+      if (auto *VisMD = dyn_cast<VisibilityMacroDirective>(MD)) {
+        // The latest visibility directive for a name in a submodule affects
+        // all the directives that come before it.
+        if (VisMD->isPublic())
+          ExplicitlyPublic = true;
+        else if (!ExplicitlyPublic)
+          // Private with no following public directive: not exported.
+          break;
+      } else {
+        MacroInfo *Def = nullptr;
+        if (DefMacroDirective *DefMD = dyn_cast<DefMacroDirective>(MD))
+          Def = DefMD->getInfo();
+
+        // FIXME: Issue a warning if multiple headers for the same submodule
+        // define a macro, rather than silently ignoring all but the first.
+        bool IsNew;
+        // Don't bother creating a module macro if it would represent a #undef
+        // that doesn't override anything.
+        if (Def || !Macro.second.getOverriddenMacros().empty())
+          addModuleMacro(LeavingMod, II, Def,
+                         Macro.second.getOverriddenMacros(), IsNew);
+        break;
+      }
+    }
+  }
+
+  // Put back the outer module's state, if we're tracking it.
+  if (getLangOpts().ModulesLocalVisibility)
+    CurSubmoduleState = Info.OuterSubmoduleState;
+
+  BuildingSubmoduleStack.pop_back();
+
+  // A nested #include makes the included submodule visible.
+  if (!BuildingSubmoduleStack.empty() || !getLangOpts().ModulesLocalVisibility)
+    makeModuleVisible(LeavingMod, ImportLoc);
+}
diff --git a/lib/Lex/PPMacroExpansion.cpp b/lib/Lex/PPMacroExpansion.cpp
index cd05d06633f2..9046ad51c14a 100644
--- a/lib/Lex/PPMacroExpansion.cpp
+++ b/lib/Lex/PPMacroExpansion.cpp
@@ -34,44 +34,234 @@
 using namespace clang;
 
 MacroDirective *
-Preprocessor::getMacroDirectiveHistory(const IdentifierInfo *II) const {
-  assert(II->hadMacroDefinition() && "Identifier has not been not a macro!");
-
-  macro_iterator Pos = Macros.find(II);
-  assert(Pos != Macros.end() && "Identifier macro info is missing!");
-  return Pos->second;
+Preprocessor::getLocalMacroDirectiveHistory(const IdentifierInfo *II) const {
+  if (!II->hadMacroDefinition())
+    return nullptr;
+  auto Pos = CurSubmoduleState->Macros.find(II);
+  return Pos == CurSubmoduleState->Macros.end() ? nullptr
+                                                : Pos->second.getLatest();
 }
 
 void Preprocessor::appendMacroDirective(IdentifierInfo *II, MacroDirective *MD){
   assert(MD && "MacroDirective should be non-zero!");
   assert(!MD->getPrevious() && "Already attached to a MacroDirective history.");
 
-  MacroDirective *&StoredMD = Macros[II];
-  MD->setPrevious(StoredMD);
-  StoredMD = MD;
-  // Setup the identifier as having associated macro history.
+  MacroState &StoredMD = CurSubmoduleState->Macros[II];
+  auto *OldMD = StoredMD.getLatest();
+  MD->setPrevious(OldMD);
+  StoredMD.setLatest(MD);
+  StoredMD.overrideActiveModuleMacros(*this, II);
+
+  // Set up the identifier as having associated macro history.
   II->setHasMacroDefinition(true);
-  if (!MD->isDefined())
+  if (!MD->isDefined() && LeafModuleMacros.find(II) == LeafModuleMacros.end())
     II->setHasMacroDefinition(false);
-  bool isImportedMacro = isa<DefMacroDirective>(MD) &&
-                         cast<DefMacroDirective>(MD)->isImported();
-  if (II->isFromAST() && !isImportedMacro)
+  if (II->isFromAST())
     II->setChangedSinceDeserialization();
 }
 
 void Preprocessor::setLoadedMacroDirective(IdentifierInfo *II,
                                            MacroDirective *MD) {
   assert(II && MD);
-  MacroDirective *&StoredMD = Macros[II];
-  assert(!StoredMD &&
+  MacroState &StoredMD = CurSubmoduleState->Macros[II];
+  assert(!StoredMD.getLatest() &&
          "the macro history was modified before initializing it from a pch");
   StoredMD = MD;
   // Setup the identifier as having associated macro history.
   II->setHasMacroDefinition(true);
-  if (!MD->isDefined())
+  if (!MD->isDefined() && LeafModuleMacros.find(II) == LeafModuleMacros.end())
     II->setHasMacroDefinition(false);
 }
 
+ModuleMacro *Preprocessor::addModuleMacro(Module *Mod, IdentifierInfo *II,
+                                          MacroInfo *Macro,
+                                          ArrayRef<ModuleMacro *> Overrides,
+                                          bool &New) {
+  llvm::FoldingSetNodeID ID;
+  ModuleMacro::Profile(ID, Mod, II);
+
+  void *InsertPos;
+  if (auto *MM = ModuleMacros.FindNodeOrInsertPos(ID, InsertPos)) {
+    New = false;
+    return MM;
+  }
+
+  auto *MM = ModuleMacro::create(*this, Mod, II, Macro, Overrides);
+  ModuleMacros.InsertNode(MM, InsertPos);
+
+  // Each overridden macro is now overridden by one more macro.
+  bool HidAny = false;
+  for (auto *O : Overrides) {
+    HidAny |= (O->NumOverriddenBy == 0);
+    ++O->NumOverriddenBy;
+  }
+
+  // If we were the first overrider for any macro, it's no longer a leaf.
+  auto &LeafMacros = LeafModuleMacros[II];
+  if (HidAny) {
+    LeafMacros.erase(std::remove_if(LeafMacros.begin(), LeafMacros.end(),
+                                    [](ModuleMacro *MM) {
+                                      return MM->NumOverriddenBy != 0;
+                                    }),
+                     LeafMacros.end());
+  }
+
+  // The new macro is always a leaf macro.
+  LeafMacros.push_back(MM);
+  // The identifier now has defined macros (that may or may not be visible).
+  II->setHasMacroDefinition(true);
+
+  New = true;
+  return MM;
+}
+
+ModuleMacro *Preprocessor::getModuleMacro(Module *Mod, IdentifierInfo *II) {
+  llvm::FoldingSetNodeID ID;
+  ModuleMacro::Profile(ID, Mod, II);
+
+  void *InsertPos;
+  return ModuleMacros.FindNodeOrInsertPos(ID, InsertPos);
+}
+
+void Preprocessor::updateModuleMacroInfo(const IdentifierInfo *II,
+                                         ModuleMacroInfo &Info) {
+  assert(Info.ActiveModuleMacrosGeneration !=
+             CurSubmoduleState->VisibleModules.getGeneration() &&
+         "don't need to update this macro name info");
+  Info.ActiveModuleMacrosGeneration =
+      CurSubmoduleState->VisibleModules.getGeneration();
+
+  auto Leaf = LeafModuleMacros.find(II);
+  if (Leaf == LeafModuleMacros.end()) {
+    // No imported macros at all: nothing to do.
+    return;
+  }
+
+  Info.ActiveModuleMacros.clear();
+
+  // Every macro that's locally overridden is overridden by a visible macro.
+  llvm::DenseMap<ModuleMacro *, int> NumHiddenOverrides;
+  for (auto *O : Info.OverriddenMacros)
+    NumHiddenOverrides[O] = -1;
+
+  // Collect all macros that are not overridden by a visible macro.
+  llvm::SmallVector<ModuleMacro *, 16> Worklist(Leaf->second.begin(),
+                                                Leaf->second.end());
+  while (!Worklist.empty()) {
+    auto *MM = Worklist.pop_back_val();
+    if (CurSubmoduleState->VisibleModules.isVisible(MM->getOwningModule())) {
+      // We only care about collecting definitions; undefinitions only act
+      // to override other definitions.
+      if (MM->getMacroInfo())
+        Info.ActiveModuleMacros.push_back(MM);
+    } else {
+      for (auto *O : MM->overrides())
+        if ((unsigned)++NumHiddenOverrides[O] == O->getNumOverridingMacros())
+          Worklist.push_back(O);
+    }
+  }
+  // Our reverse postorder walk found the macros in reverse order.
+  std::reverse(Info.ActiveModuleMacros.begin(), Info.ActiveModuleMacros.end());
+
+  // Determine whether the macro name is ambiguous.
+  MacroInfo *MI = nullptr;
+  bool IsSystemMacro = true;
+  bool IsAmbiguous = false;
+  if (auto *MD = Info.MD) {
+    while (MD && isa<VisibilityMacroDirective>(MD))
+      MD = MD->getPrevious();
+    if (auto *DMD = dyn_cast_or_null<DefMacroDirective>(MD)) {
+      MI = DMD->getInfo();
+      IsSystemMacro &= SourceMgr.isInSystemHeader(DMD->getLocation());
+    }
+  }
+  for (auto *Active : Info.ActiveModuleMacros) {
+    auto *NewMI = Active->getMacroInfo();
+
+    // Before marking the macro as ambiguous, check if this is a case where
+    // both macros are in system headers. If so, we trust that the system
+    // did not get it wrong. This also handles cases where Clang's own
+    // headers have a different spelling of certain system macros:
+    //   #define LONG_MAX __LONG_MAX__ (clang's limits.h)
+    //   #define LONG_MAX 0x7fffffffffffffffL (system's limits.h)
+    //
+    // FIXME: Remove the defined-in-system-headers check. clang's limits.h
+    // overrides the system limits.h's macros, so there's no conflict here.
+    if (MI && NewMI != MI &&
+        !MI->isIdenticalTo(*NewMI, *this, /*Syntactically=*/true))
+      IsAmbiguous = true;
+    IsSystemMacro &= Active->getOwningModule()->IsSystem ||
+                     SourceMgr.isInSystemHeader(NewMI->getDefinitionLoc());
+    MI = NewMI;
+  }
+  Info.IsAmbiguous = IsAmbiguous && !IsSystemMacro;
+}
+
+void Preprocessor::dumpMacroInfo(const IdentifierInfo *II) {
+  ArrayRef<ModuleMacro*> Leaf;
+  auto LeafIt = LeafModuleMacros.find(II);
+  if (LeafIt != LeafModuleMacros.end())
+    Leaf = LeafIt->second;
+  const MacroState *State = nullptr;
+  auto Pos = CurSubmoduleState->Macros.find(II);
+  if (Pos != CurSubmoduleState->Macros.end())
+    State = &Pos->second;
+
+  llvm::errs() << "MacroState " << State << " " << II->getNameStart();
+  if (State && State->isAmbiguous(*this, II))
+    llvm::errs() << " ambiguous";
+  if (State && !State->getOverriddenMacros().empty()) {
+    llvm::errs() << " overrides";
+    for (auto *O : State->getOverriddenMacros())
+      llvm::errs() << " " << O->getOwningModule()->getFullModuleName();
+  }
+  llvm::errs() << "\n";
+
+  // Dump local macro directives.
+  for (auto *MD = State ? State->getLatest() : nullptr; MD;
+       MD = MD->getPrevious()) {
+    llvm::errs() << " ";
+    MD->dump();
+  }
+
+  // Dump module macros.
+  llvm::DenseSet<ModuleMacro*> Active;
+  for (auto *MM : State ? State->getActiveModuleMacros(*this, II) : None)
+    Active.insert(MM);
+  llvm::DenseSet<ModuleMacro*> Visited;
+  llvm::SmallVector<ModuleMacro *, 16> Worklist(Leaf.begin(), Leaf.end());
+  while (!Worklist.empty()) {
+    auto *MM = Worklist.pop_back_val();
+    llvm::errs() << " ModuleMacro " << MM << " "
+                 << MM->getOwningModule()->getFullModuleName();
+    if (!MM->getMacroInfo())
+      llvm::errs() << " undef";
+
+    if (Active.count(MM))
+      llvm::errs() << " active";
+    else if (!CurSubmoduleState->VisibleModules.isVisible(
+                 MM->getOwningModule()))
+      llvm::errs() << " hidden";
+    else if (MM->getMacroInfo())
+      llvm::errs() << " overridden";
+
+    if (!MM->overrides().empty()) {
+      llvm::errs() << " overrides";
+      for (auto *O : MM->overrides()) {
+        llvm::errs() << " " << O->getOwningModule()->getFullModuleName();
+        if (Visited.insert(O).second)
+          Worklist.push_back(O);
+      }
+    }
+    llvm::errs() << "\n";
+    if (auto *MI = MM->getMacroInfo()) {
+      llvm::errs() << "  ";
+      MI->dump();
+      llvm::errs() << "\n";
+    }
+  }
+}
+
 /// RegisterBuiltinMacro - Register the specified identifier in the identifier
 /// table and mark it as a builtin macro to be expanded.
 static IdentifierInfo *RegisterBuiltinMacro(Preprocessor &PP, const char *Name){
@@ -97,7 +287,11 @@ void Preprocessor::RegisterBuiltinMacros() {
   Ident_Pragma  = RegisterBuiltinMacro(*this, "_Pragma");
 
   // C++ Standing Document Extensions.
-  Ident__has_cpp_attribute = RegisterBuiltinMacro(*this, "__has_cpp_attribute");
+  if (LangOpts.CPlusPlus)
+    Ident__has_cpp_attribute =
+        RegisterBuiltinMacro(*this, "__has_cpp_attribute");
+  else
+    Ident__has_cpp_attribute = nullptr;
 
   // GCC Extensions.
   Ident__BASE_FILE__     = RegisterBuiltinMacro(*this, "__BASE_FILE__");
@@ -156,10 +350,11 @@ static bool isTrivialSingleTokenExpansion(const MacroInfo *MI,
 
   // If the identifier is a macro, and if that macro is enabled, it may be
   // expanded so it's not a trivial expansion.
-  if (II->hasMacroDefinition() && PP.getMacroInfo(II)->isEnabled() &&
-      // Fast expanding "#define X X" is ok, because X would be disabled.
-      II != MacroIdent)
-    return false;
+  if (auto *ExpansionMI = PP.getMacroInfo(II))
+    if (ExpansionMI->isEnabled() &&
+        // Fast expanding "#define X X" is ok, because X would be disabled.
+        II != MacroIdent)
+      return false;
 
   // If this is an object-like macro invocation, it is safe to trivially expand
   // it.
@@ -222,10 +417,8 @@ bool Preprocessor::isNextPPTokenLParen() {
 /// HandleMacroExpandedIdentifier - If an identifier token is read that is to be
 /// expanded as a macro, handle it and return the next token as 'Identifier'.
 bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier,
-                                                 MacroDirective *MD) {
-  MacroDirective::DefInfo Def = MD->getDefinition();
-  assert(Def.isValid());
-  MacroInfo *MI = Def.getMacroInfo();
+                                                 const MacroDefinition &M) {
+  MacroInfo *MI = M.getMacroInfo();
 
   // If this is a macro expansion in the "#if !defined(x)" line for the file,
   // then the macro could expand to different things in other contexts, we need
@@ -234,9 +427,9 @@ bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier,
 
   // If this is a builtin macro, like __LINE__ or _Pragma, handle it specially.
   if (MI->isBuiltinMacro()) {
-    if (Callbacks) Callbacks->MacroExpands(Identifier, MD,
-                                           Identifier.getLocation(),
-                                           /*Args=*/nullptr);
+    if (Callbacks)
+      Callbacks->MacroExpands(Identifier, M, Identifier.getLocation(),
+                              /*Args=*/nullptr);
     ExpandBuiltinMacro(Identifier);
     return true;
   }
@@ -283,9 +476,9 @@ bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier,
       // MacroExpands callbacks still happen in source order, queue this
       // callback to have it happen after the function macro callback.
       DelayedMacroExpandsCallbacks.push_back(
-                              MacroExpandsInfo(Identifier, MD, ExpansionRange));
+          MacroExpandsInfo(Identifier, M, ExpansionRange));
     } else {
-      Callbacks->MacroExpands(Identifier, MD, ExpansionRange, Args);
+      Callbacks->MacroExpands(Identifier, M, ExpansionRange, Args);
       if (!DelayedMacroExpandsCallbacks.empty()) {
         for (unsigned i=0, e = DelayedMacroExpandsCallbacks.size(); i!=e; ++i) {
           MacroExpandsInfo &Info = DelayedMacroExpandsCallbacks[i];
@@ -299,20 +492,16 @@ bool Preprocessor::HandleMacroExpandedIdentifier(Token &Identifier,
   }
 
   // If the macro definition is ambiguous, complain.
-  if (Def.getDirective()->isAmbiguous()) {
+  if (M.isAmbiguous()) {
     Diag(Identifier, diag::warn_pp_ambiguous_macro)
       << Identifier.getIdentifierInfo();
     Diag(MI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_chosen)
       << Identifier.getIdentifierInfo();
-    for (MacroDirective::DefInfo PrevDef = Def.getPreviousDefinition();
-         PrevDef && !PrevDef.isUndefined();
-         PrevDef = PrevDef.getPreviousDefinition()) {
-      Diag(PrevDef.getMacroInfo()->getDefinitionLoc(),
-           diag::note_pp_ambiguous_macro_other)
-        << Identifier.getIdentifierInfo();
-      if (!PrevDef.getDirective()->isAmbiguous())
-        break;
-    }
+    M.forAllDefinitions([&](const MacroInfo *OtherMI) {
+      if (OtherMI != MI)
+        Diag(OtherMI->getDefinitionLoc(), diag::note_pp_ambiguous_macro_other)
+          << Identifier.getIdentifierInfo();
+    });
   }
 
   // If we started lexing a macro, enter the macro expansion body.
@@ -598,7 +787,7 @@ MacroArgs *Preprocessor::ReadFunctionLikeMacroArgs(Token &MacroName,
         // If this is a comment token in the argument list and we're just in
         // -C mode (not -CC mode), discard the comment.
         continue;
-      } else if (Tok.getIdentifierInfo() != nullptr) {
+      } else if (!Tok.isAnnotation() && Tok.getIdentifierInfo() != nullptr) {
         // Reading macro arguments can cause macros that we are currently
         // expanding from to be popped off the expansion stack.  Doing so causes
         // them to be reenabled for expansion.  Here we record whether any
@@ -868,6 +1057,7 @@ static bool HasFeature(const Preprocessor &PP, const IdentifierInfo *II) {
       .Case("attribute_analyzer_noreturn", true)
       .Case("attribute_availability", true)
       .Case("attribute_availability_with_message", true)
+      .Case("attribute_availability_app_extension", true)
       .Case("attribute_cf_returns_not_retained", true)
       .Case("attribute_cf_returns_retained", true)
       .Case("attribute_deprecated_with_message", true)
@@ -912,7 +1102,8 @@ static bool HasFeature(const Preprocessor &PP, const IdentifierInfo *II) {
       .Case("objc_dictionary_literals", LangOpts.ObjC2)
       .Case("objc_boxed_expressions", LangOpts.ObjC2)
       .Case("arc_cf_code_audited", true)
-      .Case("objc_bridge_id", LangOpts.ObjC2)
+      .Case("objc_bridge_id", true)
+      .Case("objc_bridge_id_on_typedefs", true)
       // C11 features
       .Case("c_alignas", LangOpts.C11)
       .Case("c_alignof", LangOpts.C11)
@@ -1050,6 +1241,7 @@ static bool HasExtension(const Preprocessor &PP, const IdentifierInfo *II) {
            .Case("cxx_range_for", LangOpts.CPlusPlus)
            .Case("cxx_reference_qualified_functions", LangOpts.CPlusPlus)
            .Case("cxx_rvalue_references", LangOpts.CPlusPlus)
+           .Case("cxx_variadic_templates", LangOpts.CPlusPlus)
            // C++1y features supported by other languages as extensions.
            .Case("cxx_binary_literals", true)
            .Case("cxx_init_captures", LangOpts.CPlusPlus11)
@@ -1071,6 +1263,9 @@ static bool EvaluateHasIncludeCommon(Token &Tok,
   // These expressions are only allowed within a preprocessor directive.
   if (!PP.isParsingIfOrElifDirective()) {
     PP.Diag(LParenLoc, diag::err_pp_directive_required) << II->getName();
+    // Return a valid identifier token.
+    assert(Tok.is(tok::identifier));
+    Tok.setIdentifierInfo(II);
     return false;
   }
 
@@ -1129,7 +1324,7 @@ static bool EvaluateHasIncludeCommon(Token &Tok,
       Tok.setKind(tok::eod);
       return false;   // Found <eod> but no ">"?  Diagnostic already emitted.
     }
-    Filename = FilenameBuffer.str();
+    Filename = FilenameBuffer;
     break;
   default:
     PP.Diag(Tok.getLocation(), diag::err_pp_expects_filename);
@@ -1179,7 +1374,7 @@ static bool EvaluateHasIncludeNext(Token &Tok,
   // __has_include_next is like __has_include, except that we start
   // searching after the current found directory.  If we can't do this,
   // issue a diagnostic.
-  // FIXME: Factor out duplication wiht
+  // FIXME: Factor out duplication with 
   // Preprocessor::HandleIncludeNextDirective.
   const DirectoryLookup *Lookup = PP.GetCurDirLookup();
   const FileEntry *LookupFromFile = nullptr;
@@ -1312,7 +1507,7 @@ void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
     if (PLoc.isValid()) {
       FN += PLoc.getFilename();
       Lexer::Stringify(FN);
-      OS << '"' << FN.str() << '"';
+      OS << '"' << FN << '"';
     }
     Tok.setKind(tok::string_literal);
   } else if (II == Ident__DATE__) {
@@ -1459,9 +1654,11 @@ void Preprocessor::ExpandBuiltinMacro(Token &Tok) {
       Value = EvaluateHasInclude(Tok, II, *this);
     else
       Value = EvaluateHasIncludeNext(Tok, II, *this);
+
+    if (Tok.isNot(tok::r_paren))
+      return;
     OS << (int)Value;
-    if (Tok.is(tok::r_paren))
-      Tok.setKind(tok::numeric_constant);
+    Tok.setKind(tok::numeric_constant);
   } else if (II == Ident__has_warning) {
     // The argument should be a parenthesized string literal.
     // The argument to these builtins should be a parenthesized identifier.
diff --git a/lib/Lex/Pragma.cpp b/lib/Lex/Pragma.cpp
index 8ed832893771..26ed674f65aa 100644
--- a/lib/Lex/Pragma.cpp
+++ b/lib/Lex/Pragma.cpp
@@ -22,6 +22,7 @@
 #include "clang/Lex/Preprocessor.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/StringSwitch.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/Support/CrashRecoveryContext.h"
 #include "llvm/Support/ErrorHandling.h"
 #include <algorithm>
@@ -400,7 +401,7 @@ void Preprocessor::HandlePragmaPoison(Token &PoisonTok) {
     if (II->isPoisoned()) continue;
 
     // If this is a macro identifier, emit a warning.
-    if (II->hasMacroDefinition())
+    if (isMacroDefined(II))
       Diag(Tok, diag::pp_poisoning_existing_macro);
 
     // Finally, poison it!
@@ -590,8 +591,7 @@ void Preprocessor::HandlePragmaPopMacro(Token &PopMacroTok) {
     PragmaPushMacroInfo.find(IdentInfo);
   if (iter != PragmaPushMacroInfo.end()) {
     // Forget the MacroInfo currently associated with IdentInfo.
-    if (MacroDirective *CurrentMD = getMacroDirective(IdentInfo)) {
-      MacroInfo *MI = CurrentMD->getMacroInfo();
+    if (MacroInfo *MI = getMacroInfo(IdentInfo)) {
       if (MI->isWarnIfUnused())
         WarnUnusedMacroLocs.erase(MI->getDefinitionLoc());
       appendMacroDirective(IdentInfo, AllocateUndefMacroDirective(MessageLoc));
@@ -600,11 +600,9 @@ void Preprocessor::HandlePragmaPopMacro(Token &PopMacroTok) {
     // Get the MacroInfo we want to reinstall.
     MacroInfo *MacroToReInstall = iter->second.back();
 
-    if (MacroToReInstall) {
+    if (MacroToReInstall)
       // Reinstall the previously pushed macro.
-      appendDefMacroDirective(IdentInfo, MacroToReInstall, MessageLoc,
-                              /*isImported=*/false, /*Overrides*/None);
-    }
+      appendDefMacroDirective(IdentInfo, MacroToReInstall, MessageLoc);
 
     // Pop PragmaPushMacroInfo stack.
     iter->second.pop_back();
@@ -648,7 +646,7 @@ void Preprocessor::HandlePragmaIncludeAlias(Token &Tok) {
     SourceLocation End;
     if (ConcatenateIncludeName(FileNameBuffer, End))
       return; // Diagnostic already emitted
-    SourceFileName = FileNameBuffer.str();
+    SourceFileName = FileNameBuffer;
   } else {
     Diag(Tok, diag::warn_pragma_include_alias_expected_filename);
     return;
@@ -679,7 +677,7 @@ void Preprocessor::HandlePragmaIncludeAlias(Token &Tok) {
     SourceLocation End;
     if (ConcatenateIncludeName(FileNameBuffer, End))
       return; // Diagnostic already emitted
-    ReplaceFileName = FileNameBuffer.str();
+    ReplaceFileName = FileNameBuffer;
   } else {
     Diag(Tok, diag::warn_pragma_include_alias_expected_filename);
     return;
@@ -870,12 +868,22 @@ struct PragmaDebugHandler : public PragmaHandler {
       LLVM_BUILTIN_TRAP;
     } else if (II->isStr("parser_crash")) {
       Token Crasher;
+      Crasher.startToken();
       Crasher.setKind(tok::annot_pragma_parser_crash);
+      Crasher.setAnnotationRange(SourceRange(Tok.getLocation()));
       PP.EnterToken(Crasher);
     } else if (II->isStr("llvm_fatal_error")) {
       llvm::report_fatal_error("#pragma clang __debug llvm_fatal_error");
     } else if (II->isStr("llvm_unreachable")) {
       llvm_unreachable("#pragma clang __debug llvm_unreachable");
+    } else if (II->isStr("macro")) {
+      Token MacroName;
+      PP.LexUnexpandedToken(MacroName);
+      auto *MacroII = MacroName.getIdentifierInfo();
+      if (MacroII)
+        PP.dumpMacroInfo(MacroII);
+      else
+        PP.Diag(MacroName, diag::warn_pragma_diagnostic_invalid);
     } else if (II->isStr("overflow_stack")) {
       DebugOverflowStack();
     } else if (II->isStr("handle_crash")) {
@@ -1029,12 +1037,8 @@ struct PragmaWarningHandler : public PragmaHandler {
 
     PP.Lex(Tok);
     IdentifierInfo *II = Tok.getIdentifierInfo();
-    if (!II) {
-      PP.Diag(Tok, diag::warn_pragma_warning_spec_invalid);
-      return;
-    }
 
-    if (II->isStr("push")) {
+    if (II && II->isStr("push")) {
       // #pragma warning( push[ ,n ] )
       int Level = -1;
       PP.Lex(Tok);
@@ -1051,7 +1055,7 @@ struct PragmaWarningHandler : public PragmaHandler {
       }
       if (Callbacks)
         Callbacks->PragmaWarningPush(DiagLoc, Level);
-    } else if (II->isStr("pop")) {
+    } else if (II && II->isStr("pop")) {
       // #pragma warning( pop )
       PP.Lex(Tok);
       if (Callbacks)
@@ -1061,23 +1065,40 @@ struct PragmaWarningHandler : public PragmaHandler {
       //                  [; warning-specifier : warning-number-list...] )
       while (true) {
         II = Tok.getIdentifierInfo();
-        if (!II) {
+        if (!II && !Tok.is(tok::numeric_constant)) {
           PP.Diag(Tok, diag::warn_pragma_warning_spec_invalid);
           return;
         }
 
         // Figure out which warning specifier this is.
-        StringRef Specifier = II->getName();
-        bool SpecifierValid =
-            llvm::StringSwitch<bool>(Specifier)
-                .Cases("1", "2", "3", "4", true)
-                .Cases("default", "disable", "error", "once", "suppress", true)
-                .Default(false);
+        bool SpecifierValid;
+        StringRef Specifier;
+        llvm::SmallString<1> SpecifierBuf;
+        if (II) {
+          Specifier = II->getName();
+          SpecifierValid = llvm::StringSwitch<bool>(Specifier)
+                               .Cases("default", "disable", "error", "once",
+                                      "suppress", true)
+                               .Default(false);
+          // If we read a correct specifier, snatch next token (that should be
+          // ":", checked later).
+          if (SpecifierValid)
+            PP.Lex(Tok);
+        } else {
+          // Token is a numeric constant. It should be either 1, 2, 3 or 4.
+          uint64_t Value;
+          Specifier = PP.getSpelling(Tok, SpecifierBuf);
+          if (PP.parseSimpleIntegerLiteral(Tok, Value)) {
+            SpecifierValid = (Value >= 1) && (Value <= 4);
+          } else
+            SpecifierValid = false;
+          // Next token already snatched by parseSimpleIntegerLiteral.
+        }
+
         if (!SpecifierValid) {
           PP.Diag(Tok, diag::warn_pragma_warning_spec_invalid);
           return;
         }
-        PP.Lex(Tok);
         if (Tok.isNot(tok::colon)) {
           PP.Diag(Tok, diag::warn_pragma_warning_expected) << ":";
           return;
diff --git a/lib/Lex/PreprocessingRecord.cpp b/lib/Lex/PreprocessingRecord.cpp
index 41bb581bb0aa..a423041a2d95 100644
--- a/lib/Lex/PreprocessingRecord.cpp
+++ b/lib/Lex/PreprocessingRecord.cpp
@@ -45,22 +45,23 @@ PreprocessingRecord::PreprocessingRecord(SourceManager &SM)
 
 /// \brief Returns a pair of [Begin, End) iterators of preprocessed entities
 /// that source range \p Range encompasses.
-std::pair<PreprocessingRecord::iterator, PreprocessingRecord::iterator>
+llvm::iterator_range<PreprocessingRecord::iterator>
 PreprocessingRecord::getPreprocessedEntitiesInRange(SourceRange Range) {
   if (Range.isInvalid())
-    return std::make_pair(iterator(), iterator());
+    return llvm::make_range(iterator(), iterator());
 
   if (CachedRangeQuery.Range == Range) {
-    return std::make_pair(iterator(this, CachedRangeQuery.Result.first),
-                          iterator(this, CachedRangeQuery.Result.second));
+    return llvm::make_range(iterator(this, CachedRangeQuery.Result.first),
+                            iterator(this, CachedRangeQuery.Result.second));
   }
 
   std::pair<int, int> Res = getPreprocessedEntitiesInRangeSlow(Range);
   
   CachedRangeQuery.Range = Range;
   CachedRangeQuery.Result = Res;
-  
-  return std::make_pair(iterator(this, Res.first), iterator(this, Res.second));
+
+  return llvm::make_range(iterator(this, Res.first),
+                          iterator(this, Res.second));
 }
 
 static bool isPreprocessedEntityIfInFileID(PreprocessedEntity *PPE, FileID FID,
@@ -72,11 +73,8 @@ static bool isPreprocessedEntityIfInFileID(PreprocessedEntity *PPE, FileID FID,
   SourceLocation Loc = PPE->getSourceRange().getBegin();
   if (Loc.isInvalid())
     return false;
-  
-  if (SM.isInFileID(SM.getFileLoc(Loc), FID))
-    return true;
-  else
-    return false;
+
+  return SM.isInFileID(SM.getFileLoc(Loc), FID);
 }
 
 /// \brief Returns true if the preprocessed entity that \arg PPEI iterator
@@ -90,7 +88,7 @@ bool PreprocessingRecord::isEntityInFileID(iterator PPEI, FileID FID) {
   if (FID.isInvalid())
     return false;
 
-  int Pos = PPEI.Position;
+  int Pos = std::distance(iterator(this, 0), PPEI);
   if (Pos < 0) {
     if (unsigned(-Pos-1) >= LoadedPreprocessedEntities.size()) {
       assert(0 && "Out-of bounds loaded preprocessed entity");
@@ -248,10 +246,11 @@ PreprocessingRecord::addPreprocessedEntity(PreprocessedEntity *Entity) {
   assert(Entity);
   SourceLocation BeginLoc = Entity->getSourceRange().getBegin();
 
-  if (isa<MacroDefinition>(Entity)) {
+  if (isa<MacroDefinitionRecord>(Entity)) {
     assert((PreprocessedEntities.empty() ||
-            !SourceMgr.isBeforeInTranslationUnit(BeginLoc,
-                   PreprocessedEntities.back()->getSourceRange().getBegin())) &&
+            !SourceMgr.isBeforeInTranslationUnit(
+                BeginLoc,
+                PreprocessedEntities.back()->getSourceRange().getBegin())) &&
            "a macro definition was encountered out-of-order");
     PreprocessedEntities.push_back(Entity);
     return getPPEntityID(PreprocessedEntities.size()-1, /*isLoaded=*/false);
@@ -320,7 +319,7 @@ unsigned PreprocessingRecord::allocateLoadedEntities(unsigned NumEntities) {
 }
 
 void PreprocessingRecord::RegisterMacroDefinition(MacroInfo *Macro,
-                                                  MacroDefinition *Def) {
+                                                  MacroDefinitionRecord *Def) {
   MacroDefinitions[Macro] = Def;
 }
 
@@ -357,9 +356,10 @@ PreprocessingRecord::getLoadedPreprocessedEntity(unsigned Index) {
   return Entity;
 }
 
-MacroDefinition *PreprocessingRecord::findMacroDefinition(const MacroInfo *MI) {
-  llvm::DenseMap<const MacroInfo *, MacroDefinition *>::iterator Pos
-    = MacroDefinitions.find(MI);
+MacroDefinitionRecord *
+PreprocessingRecord::findMacroDefinition(const MacroInfo *MI) {
+  llvm::DenseMap<const MacroInfo *, MacroDefinitionRecord *>::iterator Pos =
+      MacroDefinitions.find(MI);
   if (Pos == MacroDefinitions.end())
     return nullptr;
 
@@ -374,35 +374,34 @@ void PreprocessingRecord::addMacroExpansion(const Token &Id,
     return;
 
   if (MI->isBuiltinMacro())
-    addPreprocessedEntity(
-                      new (*this) MacroExpansion(Id.getIdentifierInfo(),Range));
-  else if (MacroDefinition *Def = findMacroDefinition(MI))
-    addPreprocessedEntity(
-                       new (*this) MacroExpansion(Def, Range));
+    addPreprocessedEntity(new (*this)
+                              MacroExpansion(Id.getIdentifierInfo(), Range));
+  else if (MacroDefinitionRecord *Def = findMacroDefinition(MI))
+    addPreprocessedEntity(new (*this) MacroExpansion(Def, Range));
 }
 
 void PreprocessingRecord::Ifdef(SourceLocation Loc, const Token &MacroNameTok,
-                                const MacroDirective *MD) {
+                                const MacroDefinition &MD) {
   // This is not actually a macro expansion but record it as a macro reference.
   if (MD)
-    addMacroExpansion(MacroNameTok, MD->getMacroInfo(),
+    addMacroExpansion(MacroNameTok, MD.getMacroInfo(),
                       MacroNameTok.getLocation());
 }
 
 void PreprocessingRecord::Ifndef(SourceLocation Loc, const Token &MacroNameTok,
-                                 const MacroDirective *MD) {
+                                 const MacroDefinition &MD) {
   // This is not actually a macro expansion but record it as a macro reference.
   if (MD)
-    addMacroExpansion(MacroNameTok, MD->getMacroInfo(),
+    addMacroExpansion(MacroNameTok, MD.getMacroInfo(),
                       MacroNameTok.getLocation());
 }
 
 void PreprocessingRecord::Defined(const Token &MacroNameTok,
-                                  const MacroDirective *MD,
+                                  const MacroDefinition &MD,
                                   SourceRange Range) {
   // This is not actually a macro expansion but record it as a macro reference.
   if (MD)
-    addMacroExpansion(MacroNameTok, MD->getMacroInfo(),
+    addMacroExpansion(MacroNameTok, MD.getMacroInfo(),
                       MacroNameTok.getLocation());
 }
 
@@ -410,27 +409,26 @@ void PreprocessingRecord::SourceRangeSkipped(SourceRange Range) {
   SkippedRanges.push_back(Range);
 }
 
-void PreprocessingRecord::MacroExpands(const Token &Id,const MacroDirective *MD,
+void PreprocessingRecord::MacroExpands(const Token &Id,
+                                       const MacroDefinition &MD,
                                        SourceRange Range,
                                        const MacroArgs *Args) {
-  addMacroExpansion(Id, MD->getMacroInfo(), Range);
+  addMacroExpansion(Id, MD.getMacroInfo(), Range);
 }
 
 void PreprocessingRecord::MacroDefined(const Token &Id,
                                        const MacroDirective *MD) {
   const MacroInfo *MI = MD->getMacroInfo();
   SourceRange R(MI->getDefinitionLoc(), MI->getDefinitionEndLoc());
-  MacroDefinition *Def
-      = new (*this) MacroDefinition(Id.getIdentifierInfo(), R);
+  MacroDefinitionRecord *Def =
+      new (*this) MacroDefinitionRecord(Id.getIdentifierInfo(), R);
   addPreprocessedEntity(Def);
   MacroDefinitions[MI] = Def;
 }
 
 void PreprocessingRecord::MacroUndefined(const Token &Id,
-                                         const MacroDirective *MD) {
-  // Note: MI may be null (when #undef'ining an undefined macro).
-  if (MD)
-    MacroDefinitions.erase(MD->getMacroInfo());
+                                         const MacroDefinition &MD) {
+  MD.forAllDefinitions([&](MacroInfo *MI) { MacroDefinitions.erase(MI); });
 }
 
 void PreprocessingRecord::InclusionDirective(
diff --git a/lib/Lex/Preprocessor.cpp b/lib/Lex/Preprocessor.cpp
index b2a6d933a0f2..7e33f1ccb487 100644
--- a/lib/Lex/Preprocessor.cpp
+++ b/lib/Lex/Preprocessor.cpp
@@ -73,7 +73,8 @@ Preprocessor::Preprocessor(IntrusiveRefCntPtr<PreprocessorOptions> PPOpts,
       ModuleImportExpectsIdentifier(false), CodeCompletionReached(0),
       MainFileDir(nullptr), SkipMainFilePreamble(0, true), CurPPLexer(nullptr),
       CurDirLookup(nullptr), CurLexerKind(CLK_Lexer), CurSubmodule(nullptr),
-      Callbacks(nullptr), MacroArgCache(nullptr), Record(nullptr),
+      Callbacks(nullptr), CurSubmoduleState(&NullSubmoduleState),
+      MacroArgCache(nullptr), Record(nullptr),
       MIChainHead(nullptr), DeserialMIChainHead(nullptr) {
   OwnsHeaderSearch = OwnsHeaders;
   
@@ -266,7 +267,9 @@ void Preprocessor::PrintStats() {
   llvm::errs() << "\n  Macro Expanded Tokens: "
                << llvm::capacity_in_bytes(MacroExpandedTokens);
   llvm::errs() << "\n  Predefines Buffer: " << Predefines.capacity();
-  llvm::errs() << "\n  Macros: " << llvm::capacity_in_bytes(Macros);
+  // FIXME: List information for all submodules.
+  llvm::errs() << "\n  Macros: "
+               << llvm::capacity_in_bytes(CurSubmoduleState->Macros);
   llvm::errs() << "\n  #pragma push_macro Info: "
                << llvm::capacity_in_bytes(PragmaPushMacroInfo);
   llvm::errs() << "\n  Poison Reasons: "
@@ -283,14 +286,16 @@ Preprocessor::macro_begin(bool IncludeExternalMacros) const {
     ExternalSource->ReadDefinedMacros();
   }
 
-  return Macros.begin();
+  return CurSubmoduleState->Macros.begin();
 }
 
 size_t Preprocessor::getTotalMemory() const {
   return BP.getTotalMemory()
     + llvm::capacity_in_bytes(MacroExpandedTokens)
     + Predefines.capacity() /* Predefines buffer. */
-    + llvm::capacity_in_bytes(Macros)
+    // FIXME: Include sizes from all submodules, and include MacroInfo sizes,
+    // and ModuleMacros.
+    + llvm::capacity_in_bytes(CurSubmoduleState->Macros)
     + llvm::capacity_in_bytes(PragmaPushMacroInfo)
     + llvm::capacity_in_bytes(PoisonReasons)
     + llvm::capacity_in_bytes(CommentHandlers);
@@ -304,7 +309,7 @@ Preprocessor::macro_end(bool IncludeExternalMacros) const {
     ExternalSource->ReadDefinedMacros();
   }
 
-  return Macros.end();
+  return CurSubmoduleState->Macros.end();
 }
 
 /// \brief Compares macro tokens with a specified token value sequence.
@@ -321,11 +326,11 @@ StringRef Preprocessor::getLastMacroWithSpelling(
   StringRef BestSpelling;
   for (Preprocessor::macro_iterator I = macro_begin(), E = macro_end();
        I != E; ++I) {
-    if (!I->second->getMacroInfo()->isObjectLike())
-      continue;
     const MacroDirective::DefInfo
-      Def = I->second->findDirectiveAtLoc(Loc, SourceMgr);
-    if (!Def)
+      Def = I->second.findDirectiveAtLoc(Loc, SourceMgr);
+    if (!Def || !Def.getMacroInfo())
+      continue;
+    if (!Def.getMacroInfo()->isObjectLike())
       continue;
     if (!MacroDefinitionEquals(Def.getMacroInfo(), Tokens))
       continue;
@@ -584,6 +589,23 @@ void Preprocessor::HandlePoisonedIdentifier(Token & Identifier) {
     Diag(Identifier,it->second) << Identifier.getIdentifierInfo();
 }
 
+/// \brief Returns a diagnostic message kind for reporting a future keyword as
+/// appropriate for the identifier and specified language.
+static diag::kind getFutureCompatDiagKind(const IdentifierInfo &II,
+                                          const LangOptions &LangOpts) {
+  assert(II.isFutureCompatKeyword() && "diagnostic should not be needed");
+
+  if (LangOpts.CPlusPlus)
+    return llvm::StringSwitch<diag::kind>(II.getName())
+#define CXX11_KEYWORD(NAME, FLAGS)                                             \
+        .Case(#NAME, diag::warn_cxx11_keyword)
+#include "clang/Basic/TokenKinds.def"
+        ;
+
+  llvm_unreachable(
+      "Keyword not known to come from a newer Standard or proposed Standard");
+}
+
 /// HandleIdentifier - This callback is invoked when the lexer reads an
 /// identifier.  This callback looks up the identifier in the map and/or
 /// potentially macro expands it or turns it into a named token (like 'for').
@@ -622,8 +644,9 @@ bool Preprocessor::HandleIdentifier(Token &Identifier) {
   }
 
   // If this is a macro to be expanded, do it.
-  if (MacroDirective *MD = getMacroDirective(&II)) {
-    MacroInfo *MI = MD->getMacroInfo();
+  if (MacroDefinition MD = getMacroDefinition(&II)) {
+    auto *MI = MD.getMacroInfo();
+    assert(MI && "macro definition with no macro info?");
     if (!DisableMacroExpansion) {
       if (!Identifier.isExpandDisabled() && MI->isEnabled()) {
         // C99 6.10.3p10: If the preprocessing token immediately after the
@@ -641,15 +664,16 @@ bool Preprocessor::HandleIdentifier(Token &Identifier) {
     }
   }
 
-  // If this identifier is a keyword in C++11, produce a warning. Don't warn if
-  // we're not considering macro expansion, since this identifier might be the
-  // name of a macro.
+  // If this identifier is a keyword in a newer Standard or proposed Standard,
+  // produce a warning. Don't warn if we're not considering macro expansion,
+  // since this identifier might be the name of a macro.
   // FIXME: This warning is disabled in cases where it shouldn't be, like
   //   "#define constexpr constexpr", "int constexpr;"
-  if (II.isCXX11CompatKeyword() && !DisableMacroExpansion) {
-    Diag(Identifier, diag::warn_cxx11_keyword) << II.getName();
+  if (II.isFutureCompatKeyword() && !DisableMacroExpansion) {
+    Diag(Identifier, getFutureCompatDiagKind(II, getLangOpts()))
+        << II.getName();
     // Don't diagnose this keyword again in this translation unit.
-    II.setIsCXX11CompatKeyword(false);
+    II.setIsFutureCompatKeyword(false);
   }
 
   // C++ 2.11p2: If this is an alternative representation of a C++ operator,
@@ -748,16 +772,36 @@ void Preprocessor::LexAfterModuleImport(Token &Result) {
   // If we have a non-empty module path, load the named module.
   if (!ModuleImportPath.empty()) {
     Module *Imported = nullptr;
-    if (getLangOpts().Modules)
+    if (getLangOpts().Modules) {
       Imported = TheModuleLoader.loadModule(ModuleImportLoc,
                                             ModuleImportPath,
-                                            Module::MacrosVisible,
+                                            Module::Hidden,
                                             /*IsIncludeDirective=*/false);
+      if (Imported)
+        makeModuleVisible(Imported, ModuleImportLoc);
+    }
     if (Callbacks && (getLangOpts().Modules || getLangOpts().DebuggerSupport))
       Callbacks->moduleImport(ModuleImportLoc, ModuleImportPath, Imported);
   }
 }
 
+void Preprocessor::makeModuleVisible(Module *M, SourceLocation Loc) {
+  CurSubmoduleState->VisibleModules.setVisible(
+      M, Loc, [](Module *) {},
+      [&](ArrayRef<Module *> Path, Module *Conflict, StringRef Message) {
+        // FIXME: Include the path in the diagnostic.
+        // FIXME: Include the import location for the conflicting module.
+        Diag(ModuleImportLoc, diag::warn_module_conflict)
+            << Path[0]->getFullModuleName()
+            << Conflict->getFullModuleName()
+            << Message;
+      });
+
+  // Add this module to the imports list of the currently-built submodule.
+  if (!BuildingSubmoduleStack.empty() && M != BuildingSubmoduleStack.back().M)
+    BuildingSubmoduleStack.back().M->Imports.insert(M);
+}
+
 bool Preprocessor::FinishLexStringLiteral(Token &Result, std::string &String,
                                           const char *DiagnosticTag,
                                           bool AllowMacroExpansion) {
diff --git a/lib/Lex/ScratchBuffer.cpp b/lib/Lex/ScratchBuffer.cpp
index 3bac8891c646..cd8a27e76c28 100644
--- a/lib/Lex/ScratchBuffer.cpp
+++ b/lib/Lex/ScratchBuffer.cpp
@@ -64,12 +64,13 @@ void ScratchBuffer::AllocScratchBuffer(unsigned RequestLen) {
   if (RequestLen < ScratchBufSize)
     RequestLen = ScratchBufSize;
 
+  // Get scratch buffer. Zero-initialize it so it can be dumped into a PCH file
+  // deterministically.
   std::unique_ptr<llvm::MemoryBuffer> OwnBuf =
       llvm::MemoryBuffer::getNewMemBuffer(RequestLen, "<scratch space>");
   llvm::MemoryBuffer &Buf = *OwnBuf;
   FileID FID = SourceMgr.createFileID(std::move(OwnBuf));
   BufferStartLoc = SourceMgr.getLocForStartOfFile(FID);
   CurBuffer = const_cast<char*>(Buf.getBufferStart());
-  BytesUsed = 1;
-  CurBuffer[0] = '0';  // Start out with a \0 for cleanliness.
+  BytesUsed = 0;
 }
diff --git a/lib/Lex/TokenLexer.cpp b/lib/Lex/TokenLexer.cpp
index 5f4705eff697..83efbab76139 100644
--- a/lib/Lex/TokenLexer.cpp
+++ b/lib/Lex/TokenLexer.cpp
@@ -521,6 +521,13 @@ bool TokenLexer::Lex(Token &Tok) {
 /// are more ## after it, chomp them iteratively.  Return the result as Tok.
 /// If this returns true, the caller should immediately return the token.
 bool TokenLexer::PasteTokens(Token &Tok) {
+  // MSVC: If previous token was pasted, this must be a recovery from an invalid
+  // paste operation. Ignore spaces before this token to mimic MSVC output.
+  // Required for generating valid UUID strings in some MS headers.
+  if (PP.getLangOpts().MicrosoftExt && (CurToken >= 2) &&
+      Tokens[CurToken - 2].is(tok::hashhash))
+    Tok.clearFlag(Token::LeadingSpace);
+  
   SmallString<128> Buffer;
   const char *ResultTokStrPtr = nullptr;
   SourceLocation StartLoc = Tok.getLocation();
@@ -637,7 +644,7 @@ bool TokenLexer::PasteTokens(Token &Tok) {
           // disabling it.
           PP.Diag(Loc, PP.getLangOpts().MicrosoftExt ? diag::ext_pp_bad_paste_ms
                                                      : diag::err_pp_bad_paste)
-              << Buffer.str();
+              << Buffer;
         }
 
         // An error has occurred so exit loop.
diff --git a/lib/Parse/ParseCXXInlineMethods.cpp b/lib/Parse/ParseCXXInlineMethods.cpp
index 59b491abfd15..5da70d025e56 100644
--- a/lib/Parse/ParseCXXInlineMethods.cpp
+++ b/lib/Parse/ParseCXXInlineMethods.cpp
@@ -26,8 +26,7 @@ NamedDecl *Parser::ParseCXXInlineMethodDef(AccessSpecifier AS,
                                       AttributeList *AccessAttrs,
                                       ParsingDeclarator &D,
                                       const ParsedTemplateInfo &TemplateInfo,
-                                      const VirtSpecifiers& VS, 
-                                      FunctionDefinitionKind DefinitionKind,
+                                      const VirtSpecifiers& VS,
                                       ExprResult& Init) {
   assert(D.isFunctionDeclarator() && "This isn't a function declarator!");
   assert((Tok.is(tok::l_brace) || Tok.is(tok::colon) || Tok.is(tok::kw_try) ||
@@ -40,7 +39,6 @@ NamedDecl *Parser::ParseCXXInlineMethodDef(AccessSpecifier AS,
       TemplateInfo.TemplateParams ? TemplateInfo.TemplateParams->size() : 0);
 
   NamedDecl *FnD;
-  D.setFunctionDefinitionKind(DefinitionKind);
   if (D.getDeclSpec().isFriendSpecified())
     FnD = Actions.ActOnFriendFunctionDecl(getCurScope(), D,
                                           TemplateParams);
@@ -71,17 +69,24 @@ NamedDecl *Parser::ParseCXXInlineMethodDef(AccessSpecifier AS,
 
     bool Delete = false;
     SourceLocation KWLoc;
+    SourceLocation KWEndLoc = Tok.getEndLoc().getLocWithOffset(-1);
     if (TryConsumeToken(tok::kw_delete, KWLoc)) {
       Diag(KWLoc, getLangOpts().CPlusPlus11
                       ? diag::warn_cxx98_compat_deleted_function
                       : diag::ext_deleted_function);
       Actions.SetDeclDeleted(FnD, KWLoc);
       Delete = true;
+      if (auto *DeclAsFunction = dyn_cast<FunctionDecl>(FnD)) {
+        DeclAsFunction->setRangeEnd(KWEndLoc);
+      }
     } else if (TryConsumeToken(tok::kw_default, KWLoc)) {
       Diag(KWLoc, getLangOpts().CPlusPlus11
                       ? diag::warn_cxx98_compat_defaulted_function
                       : diag::ext_defaulted_function);
       Actions.SetDeclDefaulted(FnD, KWLoc);
+      if (auto *DeclAsFunction = dyn_cast<FunctionDecl>(FnD)) {
+        DeclAsFunction->setRangeEnd(KWEndLoc);
+      }
     } else {
       llvm_unreachable("function definition after = not 'delete' or 'default'");
     }
@@ -97,12 +102,12 @@ NamedDecl *Parser::ParseCXXInlineMethodDef(AccessSpecifier AS,
 
     return FnD;
   }
-  
+
   // In delayed template parsing mode, if we are within a class template
   // or if we are about to parse function member template then consume
   // the tokens and store them for parsing at the end of the translation unit.
   if (getLangOpts().DelayedTemplateParsing &&
-      DefinitionKind == FDK_Definition &&
+      D.getFunctionDefinitionKind() == FDK_Definition &&
       !D.getDeclSpec().isConstexprSpecified() &&
       !(FnD && FnD->getAsFunction() &&
         FnD->getAsFunction()->getReturnType()->getContainedAutoType()) &&
@@ -306,9 +311,10 @@ void Parser::ParseLexedMethodDeclaration(LateParsedMethodDeclaration &LM) {
   ParseScope PrototypeScope(this, Scope::FunctionPrototypeScope |
                             Scope::FunctionDeclarationScope | Scope::DeclScope);
   for (unsigned I = 0, N = LM.DefaultArgs.size(); I != N; ++I) {
+    auto Param = cast<ParmVarDecl>(LM.DefaultArgs[I].Param);
     // Introduce the parameter into scope.
-    Actions.ActOnDelayedCXXMethodParameter(getCurScope(),
-                                           LM.DefaultArgs[I].Param);
+    bool HasUnparsed = Param->hasUnparsedDefaultArg();
+    Actions.ActOnDelayedCXXMethodParameter(getCurScope(), Param);
     if (CachedTokens *Toks = LM.DefaultArgs[I].Toks) {
       // Mark the end of the default argument so that we know when to stop when
       // we parse it later on.
@@ -316,9 +322,8 @@ void Parser::ParseLexedMethodDeclaration(LateParsedMethodDeclaration &LM) {
       Token DefArgEnd;
       DefArgEnd.startToken();
       DefArgEnd.setKind(tok::eof);
-      DefArgEnd.setLocation(LastDefaultArgToken.getLocation().getLocWithOffset(
-          LastDefaultArgToken.getLength()));
-      DefArgEnd.setEofData(LM.DefaultArgs[I].Param);
+      DefArgEnd.setLocation(LastDefaultArgToken.getEndLoc());
+      DefArgEnd.setEofData(Param);
       Toks->push_back(DefArgEnd);
 
       // Parse the default argument from its saved token stream.
@@ -336,7 +341,7 @@ void Parser::ParseLexedMethodDeclaration(LateParsedMethodDeclaration &LM) {
       // used.
       EnterExpressionEvaluationContext Eval(Actions,
                                             Sema::PotentiallyEvaluatedIfUsed,
-                                            LM.DefaultArgs[I].Param);
+                                            Param);
 
       ExprResult DefArgResult;
       if (getLangOpts().CPlusPlus11 && Tok.is(tok::l_brace)) {
@@ -346,11 +351,9 @@ void Parser::ParseLexedMethodDeclaration(LateParsedMethodDeclaration &LM) {
         DefArgResult = ParseAssignmentExpression();
       DefArgResult = Actions.CorrectDelayedTyposInExpr(DefArgResult);
       if (DefArgResult.isInvalid()) {
-        Actions.ActOnParamDefaultArgumentError(LM.DefaultArgs[I].Param,
-                                               EqualLoc);
+        Actions.ActOnParamDefaultArgumentError(Param, EqualLoc);
       } else {
-        if (Tok.isNot(tok::eof) ||
-            Tok.getEofData() != LM.DefaultArgs[I].Param) {
+        if (Tok.isNot(tok::eof) || Tok.getEofData() != Param) {
           // The last two tokens are the terminator and the saved value of
           // Tok; the last token in the default argument is the one before
           // those.
@@ -359,7 +362,7 @@ void Parser::ParseLexedMethodDeclaration(LateParsedMethodDeclaration &LM) {
             << SourceRange(Tok.getLocation(),
                            (*Toks)[Toks->size() - 3].getLocation());
         }
-        Actions.ActOnParamDefaultArgument(LM.DefaultArgs[I].Param, EqualLoc,
+        Actions.ActOnParamDefaultArgument(Param, EqualLoc,
                                           DefArgResult.get());
       }
 
@@ -368,11 +371,21 @@ void Parser::ParseLexedMethodDeclaration(LateParsedMethodDeclaration &LM) {
       while (Tok.isNot(tok::eof))
         ConsumeAnyToken();
 
-      if (Tok.is(tok::eof) && Tok.getEofData() == LM.DefaultArgs[I].Param)
+      if (Tok.is(tok::eof) && Tok.getEofData() == Param)
         ConsumeAnyToken();
 
       delete Toks;
       LM.DefaultArgs[I].Toks = nullptr;
+    } else if (HasUnparsed) {
+      assert(Param->hasInheritedDefaultArg());
+      FunctionDecl *Old = cast<FunctionDecl>(LM.Method)->getPreviousDecl();
+      ParmVarDecl *OldParam = Old->getParamDecl(I);
+      assert (!OldParam->hasUnparsedDefaultArg());
+      if (OldParam->hasUninstantiatedDefaultArg())
+        Param->setUninstantiatedDefaultArg(
+                                      Param->getUninstantiatedDefaultArg());
+      else
+        Param->setDefaultArg(OldParam->getInit());
     }
   }
 
@@ -383,9 +396,7 @@ void Parser::ParseLexedMethodDeclaration(LateParsedMethodDeclaration &LM) {
     Token ExceptionSpecEnd;
     ExceptionSpecEnd.startToken();
     ExceptionSpecEnd.setKind(tok::eof);
-    ExceptionSpecEnd.setLocation(
-        LastExceptionSpecToken.getLocation().getLocWithOffset(
-            LastExceptionSpecToken.getLength()));
+    ExceptionSpecEnd.setLocation(LastExceptionSpecToken.getEndLoc());
     ExceptionSpecEnd.setEofData(LM.Method);
     Toks->push_back(ExceptionSpecEnd);
 
@@ -490,8 +501,7 @@ void Parser::ParseLexedMethodDef(LexedMethod &LM) {
   Token BodyEnd;
   BodyEnd.startToken();
   BodyEnd.setKind(tok::eof);
-  BodyEnd.setLocation(
-      LastBodyToken.getLocation().getLocWithOffset(LastBodyToken.getLength()));
+  BodyEnd.setLocation(LastBodyToken.getEndLoc());
   BodyEnd.setEofData(LM.D);
   LM.Toks.push_back(BodyEnd);
   // Append the current token at the end of the new token stream so that it
diff --git a/lib/Parse/ParseDecl.cpp b/lib/Parse/ParseDecl.cpp
index 4d05e16491b0..bd114d7e34a7 100644
--- a/lib/Parse/ParseDecl.cpp
+++ b/lib/Parse/ParseDecl.cpp
@@ -529,64 +529,72 @@ bool Parser::ParseMicrosoftDeclSpecArgs(IdentifierInfo *AttrName,
 /// [MS] extended-decl-modifier-seq:
 ///             extended-decl-modifier[opt]
 ///             extended-decl-modifier extended-decl-modifier-seq
-void Parser::ParseMicrosoftDeclSpec(ParsedAttributes &Attrs) {
+void Parser::ParseMicrosoftDeclSpecs(ParsedAttributes &Attrs,
+                                     SourceLocation *End) {
+  assert((getLangOpts().MicrosoftExt || getLangOpts().Borland ||
+          getLangOpts().CUDA) &&
+         "Incorrect language options for parsing __declspec");
   assert(Tok.is(tok::kw___declspec) && "Not a declspec!");
 
-  ConsumeToken();
-  BalancedDelimiterTracker T(*this, tok::l_paren);
-  if (T.expectAndConsume(diag::err_expected_lparen_after, "__declspec",
-                         tok::r_paren))
-    return;
-
-  // An empty declspec is perfectly legal and should not warn.  Additionally,
-  // you can specify multiple attributes per declspec.
-  while (Tok.isNot(tok::r_paren)) {
-    // Attribute not present.
-    if (TryConsumeToken(tok::comma))
-      continue;
-
-    // We expect either a well-known identifier or a generic string.  Anything
-    // else is a malformed declspec.
-    bool IsString = Tok.getKind() == tok::string_literal ? true : false;
-    if (!IsString && Tok.getKind() != tok::identifier &&
-        Tok.getKind() != tok::kw_restrict) {
-      Diag(Tok, diag::err_ms_declspec_type);
-      T.skipToEnd();
+  while (Tok.is(tok::kw___declspec)) {
+    ConsumeToken();
+    BalancedDelimiterTracker T(*this, tok::l_paren);
+    if (T.expectAndConsume(diag::err_expected_lparen_after, "__declspec",
+                           tok::r_paren))
       return;
-    }
 
-    IdentifierInfo *AttrName;
-    SourceLocation AttrNameLoc;
-    if (IsString) {
-      SmallString<8> StrBuffer;
-      bool Invalid = false;
-      StringRef Str = PP.getSpelling(Tok, StrBuffer, &Invalid);
-      if (Invalid) {
+    // An empty declspec is perfectly legal and should not warn.  Additionally,
+    // you can specify multiple attributes per declspec.
+    while (Tok.isNot(tok::r_paren)) {
+      // Attribute not present.
+      if (TryConsumeToken(tok::comma))
+        continue;
+
+      // We expect either a well-known identifier or a generic string.  Anything
+      // else is a malformed declspec.
+      bool IsString = Tok.getKind() == tok::string_literal;
+      if (!IsString && Tok.getKind() != tok::identifier &&
+          Tok.getKind() != tok::kw_restrict) {
+        Diag(Tok, diag::err_ms_declspec_type);
         T.skipToEnd();
         return;
       }
-      AttrName = PP.getIdentifierInfo(Str);
-      AttrNameLoc = ConsumeStringToken();
-    } else {
-      AttrName = Tok.getIdentifierInfo();
-      AttrNameLoc = ConsumeToken();
-    }
 
-    bool AttrHandled = false;
+      IdentifierInfo *AttrName;
+      SourceLocation AttrNameLoc;
+      if (IsString) {
+        SmallString<8> StrBuffer;
+        bool Invalid = false;
+        StringRef Str = PP.getSpelling(Tok, StrBuffer, &Invalid);
+        if (Invalid) {
+          T.skipToEnd();
+          return;
+        }
+        AttrName = PP.getIdentifierInfo(Str);
+        AttrNameLoc = ConsumeStringToken();
+      } else {
+        AttrName = Tok.getIdentifierInfo();
+        AttrNameLoc = ConsumeToken();
+      }
+
+      bool AttrHandled = false;
 
-    // Parse attribute arguments.
-    if (Tok.is(tok::l_paren))
-      AttrHandled = ParseMicrosoftDeclSpecArgs(AttrName, AttrNameLoc, Attrs);
-    else if (AttrName->getName() == "property")
-      // The property attribute must have an argument list.
-      Diag(Tok.getLocation(), diag::err_expected_lparen_after)
-          << AttrName->getName();
+      // Parse attribute arguments.
+      if (Tok.is(tok::l_paren))
+        AttrHandled = ParseMicrosoftDeclSpecArgs(AttrName, AttrNameLoc, Attrs);
+      else if (AttrName->getName() == "property")
+        // The property attribute must have an argument list.
+        Diag(Tok.getLocation(), diag::err_expected_lparen_after)
+            << AttrName->getName();
 
-    if (!AttrHandled)
-      Attrs.addNew(AttrName, AttrNameLoc, nullptr, AttrNameLoc, nullptr, 0,
-                   AttributeList::AS_Declspec);
+      if (!AttrHandled)
+        Attrs.addNew(AttrName, AttrNameLoc, nullptr, AttrNameLoc, nullptr, 0,
+                     AttributeList::AS_Declspec);
+    }
+    T.consumeClose();
+    if (End)
+      *End = T.getCloseLocation();
   }
-  T.consumeClose();
 }
 
 void Parser::ParseMicrosoftTypeAttributes(ParsedAttributes &attrs) {
@@ -1360,6 +1368,46 @@ void Parser::ProhibitCXX11Attributes(ParsedAttributesWithRange &attrs) {
   }
 }
 
+// As an exception to the rule, __declspec(align(...)) before the
+// class-key affects the type instead of the variable.
+void Parser::handleDeclspecAlignBeforeClassKey(ParsedAttributesWithRange &Attrs,
+                                               DeclSpec &DS,
+                                               Sema::TagUseKind TUK) {
+  if (TUK == Sema::TUK_Reference)
+    return;
+
+  ParsedAttributes &PA = DS.getAttributes();
+  AttributeList *AL = PA.getList();
+  AttributeList *Prev = nullptr;
+  while (AL) {
+    AttributeList *Next = AL->getNext();
+
+    // We only consider attributes using the appropriate '__declspec' spelling,
+    // this behavior doesn't extend to any other spellings.
+    if (AL->getKind() == AttributeList::AT_Aligned &&
+        AL->isDeclspecAttribute()) {
+      // Stitch the attribute into the tag's attribute list.
+      AL->setNext(nullptr);
+      Attrs.add(AL);
+
+      // Remove the attribute from the variable's attribute list.
+      if (Prev) {
+        // Set the last variable attribute's next attribute to be the attribute
+        // after the current one.
+        Prev->setNext(Next);
+      } else {
+        // Removing the head of the list requires us to reset the head to the
+        // next attribute.
+        PA.set(Next);
+      }
+    } else {
+      Prev = AL;
+    }
+
+    AL = Next;
+  }
+}
+
 /// ParseDeclaration - Parse a full 'declaration', which consists of
 /// declaration-specifiers, some number of declarators, and a semicolon.
 /// 'Context' should be a Declarator::TheContext value.  This returns the
@@ -1691,11 +1739,12 @@ Parser::DeclGroupPtrTy Parser::ParseDeclGroup(ParsingDeclSpec &DS,
       }
 
       if (isDeclarationSpecifier()) {
-        // If there is an invalid declaration specifier right after the function
-        // prototype, then we must be in a missing semicolon case where this isn't
-        // actually a body.  Just fall through into the code that handles it as a
-        // prototype, and let the top-level code handle the erroneous declspec
-        // where it would otherwise expect a comma or semicolon.
+        // If there is an invalid declaration specifier right after the
+        // function prototype, then we must be in a missing semicolon case
+        // where this isn't actually a body.  Just fall through into the code
+        // that handles it as a prototype, and let the top-level code handle
+        // the erroneous declspec where it would otherwise expect a comma or
+        // semicolon.
       } else {
         Diag(Tok, diag::err_expected_fn_body);
         SkipUntil(tok::semi);
@@ -1993,7 +2042,11 @@ Decl *Parser::ParseDeclarationAfterDeclaratorAndAttributes(
       Actions.ActOnCXXEnterDeclInitializer(getCurScope(), ThisDecl);
     }
 
-    if (ParseExpressionList(Exprs, CommaLocs)) {
+    if (ParseExpressionList(Exprs, CommaLocs, [&] {
+          Actions.CodeCompleteConstructor(getCurScope(),
+                 cast<VarDecl>(ThisDecl)->getType()->getCanonicalTypeInternal(),
+                                          ThisDecl->getLocation(), Exprs);
+       })) {
       Actions.ActOnInitializerError(ThisDecl);
       SkipUntil(tok::r_paren, StopAtSemi);
 
@@ -2564,7 +2617,8 @@ void Parser::ParseDeclarationSpecifiers(DeclSpec &DS,
   bool EnteringContext = (DSContext == DSC_class || DSContext == DSC_top_level);
   bool AttrsLastTime = false;
   ParsedAttributesWithRange attrs(AttrFactory);
-  const PrintingPolicy &Policy = Actions.getASTContext().getPrintingPolicy();
+  // We use Sema's policy to get bool macros right.
+  const PrintingPolicy &Policy = Actions.getPrintingPolicy();
   while (1) {
     bool isInvalid = false;
     bool isStorageClass = false;
@@ -2871,9 +2925,9 @@ void Parser::ParseDeclarationSpecifiers(DeclSpec &DS,
                             Tok.getLocation(), getCurScope());
 
       // MSVC: If we weren't able to parse a default template argument, and it's
-      // just a simple identifier, create a DependentNameType.  This will allow us
-      // to defer the name lookup to template instantiation time, as long we forge a
-      // NestedNameSpecifier for the current context.
+      // just a simple identifier, create a DependentNameType.  This will allow
+      // us to defer the name lookup to template instantiation time, as long we
+      // forge a NestedNameSpecifier for the current context.
       if (!TypeRep && DSContext == DSC_template_type_arg &&
           getLangOpts().MSVCCompat && getCurScope()->isTemplateParamScope()) {
         TypeRep = Actions.ActOnDelayedDefaultTemplateArg(
@@ -2950,7 +3004,7 @@ void Parser::ParseDeclarationSpecifiers(DeclSpec &DS,
 
     // Microsoft declspec support.
     case tok::kw___declspec:
-      ParseMicrosoftDeclSpec(DS.getAttributes());
+      ParseMicrosoftDeclSpecs(DS.getAttributes());
       continue;
 
     // Microsoft single token adornments.
@@ -3594,10 +3648,7 @@ void Parser::ParseEnumSpecifier(SourceLocation StartLoc, DeclSpec &DS,
   ParsedAttributesWithRange attrs(AttrFactory);
   MaybeParseGNUAttributes(attrs);
   MaybeParseCXX11Attributes(attrs);
-
-  // If declspecs exist after tag, parse them.
-  while (Tok.is(tok::kw___declspec))
-    ParseMicrosoftDeclSpec(attrs);
+  MaybeParseMicrosoftDeclSpecs(attrs);
 
   SourceLocation ScopedEnumKWLoc;
   bool IsScopedUsingClassTag = false;
@@ -3616,8 +3667,7 @@ void Parser::ParseEnumSpecifier(SourceLocation StartLoc, DeclSpec &DS,
     // They are allowed afterwards, though.
     MaybeParseGNUAttributes(attrs);
     MaybeParseCXX11Attributes(attrs);
-    while (Tok.is(tok::kw___declspec))
-      ParseMicrosoftDeclSpec(attrs);
+    MaybeParseMicrosoftDeclSpecs(attrs);
   }
 
   // C++11 [temp.explicit]p12:
@@ -3644,11 +3694,12 @@ void Parser::ParseEnumSpecifier(SourceLocation StartLoc, DeclSpec &DS,
     // if a fixed underlying type is allowed.
     ColonProtectionRAIIObject X(*this, AllowFixedUnderlyingType);
 
-    if (ParseOptionalCXXScopeSpecifier(SS, ParsedType(),
+    CXXScopeSpec Spec;
+    if (ParseOptionalCXXScopeSpecifier(Spec, ParsedType(),
                                        /*EnteringContext=*/true))
       return;
 
-    if (SS.isSet() && Tok.isNot(tok::identifier)) {
+    if (Spec.isSet() && Tok.isNot(tok::identifier)) {
       Diag(Tok, diag::err_expected) << tok::identifier;
       if (Tok.isNot(tok::l_brace)) {
         // Has no name and is not a definition.
@@ -3657,6 +3708,8 @@ void Parser::ParseEnumSpecifier(SourceLocation StartLoc, DeclSpec &DS,
         return;
       }
     }
+
+    SS = Spec;
   }
 
   // Must have either 'enum name' or 'enum {...}'.
@@ -3842,6 +3895,15 @@ void Parser::ParseEnumSpecifier(SourceLocation StartLoc, DeclSpec &DS,
     return;
   }
 
+  handleDeclspecAlignBeforeClassKey(attrs, DS, TUK);
+
+  Sema::SkipBodyInfo SkipBody;
+  if (!Name && TUK == Sema::TUK_Definition && Tok.is(tok::l_brace) &&
+      NextToken().is(tok::identifier))
+    SkipBody = Actions.shouldSkipAnonEnumBody(getCurScope(),
+                                              NextToken().getIdentifierInfo(),
+                                              NextToken().getLocation());
+
   bool Owned = false;
   bool IsDependent = false;
   const char *PrevSpec = nullptr;
@@ -3851,7 +3913,22 @@ void Parser::ParseEnumSpecifier(SourceLocation StartLoc, DeclSpec &DS,
                                    AS, DS.getModulePrivateSpecLoc(), TParams,
                                    Owned, IsDependent, ScopedEnumKWLoc,
                                    IsScopedUsingClassTag, BaseType,
-                                   DSC == DSC_type_specifier);
+                                   DSC == DSC_type_specifier, &SkipBody);
+
+  if (SkipBody.ShouldSkip) {
+    assert(TUK == Sema::TUK_Definition && "can only skip a definition");
+
+    BalancedDelimiterTracker T(*this, tok::l_brace);
+    T.consumeOpen();
+    T.skipToEnd();
+
+    if (DS.SetTypeSpecType(DeclSpec::TST_enum, StartLoc,
+                           NameLoc.isValid() ? NameLoc : StartLoc,
+                           PrevSpec, DiagID, TagDecl, Owned,
+                           Actions.getASTContext().getPrintingPolicy()))
+      Diag(StartLoc, DiagID) << PrevSpec;
+    return;
+  }
 
   if (IsDependent) {
     // This enum has a dependent nested-name-specifier. Handle it as a
@@ -3923,6 +4000,7 @@ void Parser::ParseEnumBody(SourceLocation StartLoc, Decl *EnumDecl) {
     Diag(Tok, diag::error_empty_enum);
 
   SmallVector<Decl *, 32> EnumConstantDecls;
+  SmallVector<SuppressAccessChecks, 32> EnumAvailabilityDiags;
 
   Decl *LastEnumConstDecl = nullptr;
 
@@ -3953,7 +4031,7 @@ void Parser::ParseEnumBody(SourceLocation StartLoc, Decl *EnumDecl) {
 
     SourceLocation EqualLoc;
     ExprResult AssignedVal;
-    ParsingDeclRAIIObject PD(*this, ParsingDeclRAIIObject::NoParent);
+    EnumAvailabilityDiags.emplace_back(*this);
 
     if (TryConsumeToken(tok::equal, EqualLoc)) {
       AssignedVal = ParseConstantExpression();
@@ -3967,7 +4045,7 @@ void Parser::ParseEnumBody(SourceLocation StartLoc, Decl *EnumDecl) {
                                                     IdentLoc, Ident,
                                                     attrs.getList(), EqualLoc,
                                                     AssignedVal.get());
-    PD.complete(EnumConstDecl);
+    EnumAvailabilityDiags.back().done();
 
     EnumConstantDecls.push_back(EnumConstDecl);
     LastEnumConstDecl = EnumConstDecl;
@@ -4023,6 +4101,14 @@ void Parser::ParseEnumBody(SourceLocation StartLoc, Decl *EnumDecl) {
                         getCurScope(),
                         attrs.getList());
 
+  // Now handle enum constant availability diagnostics.
+  assert(EnumConstantDecls.size() == EnumAvailabilityDiags.size());
+  for (size_t i = 0, e = EnumConstantDecls.size(); i != e; ++i) {
+    ParsingDeclRAIIObject PD(*this, ParsingDeclRAIIObject::NoParent);
+    EnumAvailabilityDiags[i].redelay();
+    PD.complete(EnumConstantDecls[i]);
+  }
+
   EnumScope.Exit();
   Actions.ActOnTagFinishDefinition(getCurScope(), EnumDecl,
                                    T.getCloseLocation());
@@ -4773,7 +4859,8 @@ void Parser::ParseDeclaratorInternal(Declarator &D,
       D.AddTypeInfo(DeclaratorChunk::getPointer(DS.getTypeQualifiers(), Loc,
                                                 DS.getConstSpecLoc(),
                                                 DS.getVolatileSpecLoc(),
-                                                DS.getRestrictSpecLoc()),
+                                                DS.getRestrictSpecLoc(),
+                                                DS.getAtomicSpecLoc()),
                     DS.getAttributes(),
                     SourceLocation());
     else
@@ -4916,7 +5003,8 @@ void Parser::ParseDirectDeclarator(Declarator &D) {
     }
 
     if (D.getCXXScopeSpec().isValid()) {
-      if (Actions.ShouldEnterDeclaratorScope(getCurScope(), D.getCXXScopeSpec()))
+      if (Actions.ShouldEnterDeclaratorScope(getCurScope(),
+                                             D.getCXXScopeSpec()))
         // Change the declaration context for name lookup, until this function
         // is exited (and the declarator has been parsed).
         DeclScopeObj.EnterDeclaratorScope();
@@ -4968,6 +5056,7 @@ void Parser::ParseDirectDeclarator(Declarator &D) {
         AllowConstructorName = (D.getContext() == Declarator::MemberContext);
 
       SourceLocation TemplateKWLoc;
+      bool HadScope = D.getCXXScopeSpec().isValid();
       if (ParseUnqualifiedId(D.getCXXScopeSpec(),
                              /*EnteringContext=*/true,
                              /*AllowDestructorName=*/true,
@@ -4981,6 +5070,13 @@ void Parser::ParseDirectDeclarator(Declarator &D) {
         D.SetIdentifier(nullptr, Tok.getLocation());
         D.setInvalidType(true);
       } else {
+        // ParseUnqualifiedId might have parsed a scope specifier during error
+        // recovery. If it did so, enter that scope.
+        if (!HadScope && D.getCXXScopeSpec().isValid() &&
+            Actions.ShouldEnterDeclaratorScope(getCurScope(),
+                                               D.getCXXScopeSpec()))
+          DeclScopeObj.EnterDeclaratorScope();
+
         // Parsed the unqualified-id; update range information and move along.
         if (D.getSourceRange().getBegin().isInvalid())
           D.SetRangeBegin(D.getName().getSourceRange().getBegin());
@@ -5022,7 +5118,8 @@ void Parser::ParseDirectDeclarator(Declarator &D) {
       // If there was an error parsing parenthesized declarator, declarator
       // scope may have been entered before. Don't do it again.
       if (!D.isInvalidType() &&
-          Actions.ShouldEnterDeclaratorScope(getCurScope(), D.getCXXScopeSpec()))
+          Actions.ShouldEnterDeclaratorScope(getCurScope(),
+                                             D.getCXXScopeSpec()))
         // Change the declaration context for name lookup, until this function
         // is exited (and the declarator has been parsed).
         DeclScopeObj.EnterDeclaratorScope();
@@ -5310,7 +5407,7 @@ void Parser::ParseFunctionDeclarator(Declarator &D,
     if (getLangOpts().CPlusPlus) {
       // FIXME: Accept these components in any order, and produce fixits to
       // correct the order if the user gets it wrong. Ideally we should deal
-      // with the virt-specifier-seq and pure-specifier in the same way.
+      // with the pure-specifier in the same way.
 
       // Parse cv-qualifier-seq[opt].
       ParseTypeQualifierListOpt(DS, AR_NoAttributesParsed,
@@ -5323,15 +5420,8 @@ void Parser::ParseFunctionDeclarator(Declarator &D,
       }
 
       // Parse ref-qualifier[opt].
-      if (Tok.is(tok::amp) || Tok.is(tok::ampamp)) {
-        Diag(Tok, getLangOpts().CPlusPlus11 ?
-             diag::warn_cxx98_compat_ref_qualifier :
-             diag::ext_ref_qualifier);
-
-        RefQualifierIsLValueRef = Tok.is(tok::amp);
-        RefQualifierLoc = ConsumeToken();
+      if (ParseRefQualifier(RefQualifierIsLValueRef, RefQualifierLoc))
         EndLoc = RefQualifierLoc;
-      }
 
       // C++11 [expr.prim.general]p3:
       //   If a declaration declares a member function or member function
@@ -5427,6 +5517,22 @@ void Parser::ParseFunctionDeclarator(Declarator &D,
                 FnAttrs, EndLoc);
 }
 
+/// ParseRefQualifier - Parses a member function ref-qualifier. Returns
+/// true if a ref-qualifier is found.
+bool Parser::ParseRefQualifier(bool &RefQualifierIsLValueRef,
+                               SourceLocation &RefQualifierLoc) {
+  if (Tok.is(tok::amp) || Tok.is(tok::ampamp)) {
+    Diag(Tok, getLangOpts().CPlusPlus11 ?
+         diag::warn_cxx98_compat_ref_qualifier :
+         diag::ext_ref_qualifier);
+
+    RefQualifierIsLValueRef = Tok.is(tok::amp);
+    RefQualifierLoc = ConsumeToken();
+    return true;
+  }
+  return false;
+}
+
 /// isFunctionDeclaratorIdentifierList - This parameter list may have an
 /// identifier list form for a K&R-style function:  void foo(a,b,c)
 ///
diff --git a/lib/Parse/ParseDeclCXX.cpp b/lib/Parse/ParseDeclCXX.cpp
index 87d99096440c..53e4a419682e 100644
--- a/lib/Parse/ParseDeclCXX.cpp
+++ b/lib/Parse/ParseDeclCXX.cpp
@@ -558,6 +558,7 @@ Decl *Parser::ParseUsingDeclaration(unsigned Context,
   // Maybe this is an alias-declaration.
   TypeResult TypeAlias;
   bool IsAliasDecl = Tok.is(tok::equal);
+  Decl *DeclFromDeclSpec = nullptr;
   if (IsAliasDecl) {
     // If we had any misplaced attributes from earlier, this is where they
     // should have been written.
@@ -612,10 +613,12 @@ Decl *Parser::ParseUsingDeclaration(unsigned Context,
       Diag(SS.getBeginLoc(), diag::err_alias_declaration_not_identifier)
         << FixItHint::CreateRemoval(SS.getRange());
 
-    TypeAlias = ParseTypeName(nullptr, TemplateInfo.Kind ?
-                              Declarator::AliasTemplateContext :
-                              Declarator::AliasDeclContext, AS, OwnedType,
-                              &Attrs);
+    TypeAlias = ParseTypeName(nullptr, TemplateInfo.Kind
+                                           ? Declarator::AliasTemplateContext
+                                           : Declarator::AliasDeclContext,
+                              AS, &DeclFromDeclSpec, &Attrs);
+    if (OwnedType)
+      *OwnedType = DeclFromDeclSpec;
   } else {
     // C++11 attributes are not allowed on a using-declaration, but GNU ones
     // are.
@@ -664,7 +667,7 @@ Decl *Parser::ParseUsingDeclaration(unsigned Context,
       TemplateParams ? TemplateParams->size() : 0);
     return Actions.ActOnAliasDeclaration(getCurScope(), AS, TemplateParamsArg,
                                          UsingLoc, Name, Attrs.getList(),
-                                         TypeAlias);
+                                         TypeAlias, DeclFromDeclSpec);
   }
 
   return Actions.ActOnUsingDeclaration(getCurScope(), AS,
@@ -796,7 +799,10 @@ SourceLocation Parser::ParseDecltypeSpecifier(DeclSpec &DS) {
       //   The operand of the decltype specifier is an unevaluated operand.
       EnterExpressionEvaluationContext Unevaluated(Actions, Sema::Unevaluated,
                                                    nullptr,/*IsDecltype=*/true);
-      Result = Actions.CorrectDelayedTyposInExpr(ParseExpression());
+      Result =
+          Actions.CorrectDelayedTyposInExpr(ParseExpression(), [](Expr *E) {
+            return E->hasPlaceholderType() ? ExprError() : E;
+          });
       if (Result.isInvalid()) {
         DS.SetTypeSpecError();
         if (SkipUntil(tok::r_paren, StopAtSemi | StopBeforeMatch)) {
@@ -1223,10 +1229,7 @@ void Parser::ParseClassSpecifier(tok::TokenKind TagTokKind,
   ParsedAttributesWithRange attrs(AttrFactory);
   // If attributes exist after tag, parse them.
   MaybeParseGNUAttributes(attrs);
-
-  // If declspecs exist after tag, parse them.
-  while (Tok.is(tok::kw___declspec))
-    ParseMicrosoftDeclSpec(attrs);
+  MaybeParseMicrosoftDeclSpecs(attrs);
 
   // Parse inheritance specifiers.
   if (Tok.is(tok::kw___single_inheritance) ||
@@ -1310,11 +1313,19 @@ void Parser::ParseClassSpecifier(tok::TokenKind TagTokKind,
     // is a base-specifier-list.
     ColonProtectionRAIIObject X(*this);
 
-    if (ParseOptionalCXXScopeSpecifier(SS, ParsedType(), EnteringContext))
+    CXXScopeSpec Spec;
+    bool HasValidSpec = true;
+    if (ParseOptionalCXXScopeSpecifier(Spec, ParsedType(), EnteringContext)) {
       DS.SetTypeSpecError();
-    if (SS.isSet())
-      if (Tok.isNot(tok::identifier) && Tok.isNot(tok::annot_template_id))
+      HasValidSpec = false;
+    }
+    if (Spec.isSet())
+      if (Tok.isNot(tok::identifier) && Tok.isNot(tok::annot_template_id)) {
         Diag(Tok, diag::err_expected) << tok::identifier;
+        HasValidSpec = false;
+      }
+    if (HasValidSpec)
+      SS = Spec;
   }
 
   TemplateParameterLists *TemplateParams = TemplateInfo.TemplateParams;
@@ -1539,6 +1550,7 @@ void Parser::ParseClassSpecifier(tok::TokenKind TagTokKind,
   TypeResult TypeResult = true; // invalid
 
   bool Owned = false;
+  Sema::SkipBodyInfo SkipBody;
   if (TemplateId) {
     // Explicit specialization, class template partial specialization,
     // or explicit instantiation.
@@ -1625,7 +1637,8 @@ void Parser::ParseClassSpecifier(tok::TokenKind TagTokKind,
           *TemplateId, attrs.getList(),
           MultiTemplateParamsArg(TemplateParams ? &(*TemplateParams)[0]
                                                 : nullptr,
-                                 TemplateParams ? TemplateParams->size() : 0));
+                                 TemplateParams ? TemplateParams->size() : 0),
+          &SkipBody);
     }
   } else if (TemplateInfo.Kind == ParsedTemplateInfo::ExplicitInstantiation &&
              TUK == Sema::TUK_Declaration) {
@@ -1677,6 +1690,8 @@ void Parser::ParseClassSpecifier(tok::TokenKind TagTokKind,
       TParams =
         MultiTemplateParamsArg(&(*TemplateParams)[0], TemplateParams->size());
 
+    handleDeclspecAlignBeforeClassKey(attrs, DS, TUK);
+
     // Declaration or definition of a class type
     TagOrTempResult = Actions.ActOnTag(getCurScope(), TagType, TUK, StartLoc,
                                        SS, Name, NameLoc, attrs.getList(), AS,
@@ -1684,7 +1699,8 @@ void Parser::ParseClassSpecifier(tok::TokenKind TagTokKind,
                                        TParams, Owned, IsDependent,
                                        SourceLocation(), false,
                                        clang::TypeResult(),
-                                       DSC == DSC_type_specifier);
+                                       DSC == DSC_type_specifier,
+                                       &SkipBody);
 
     // If ActOnTag said the type was dependent, try again with the
     // less common call.
@@ -1700,7 +1716,10 @@ void Parser::ParseClassSpecifier(tok::TokenKind TagTokKind,
     assert(Tok.is(tok::l_brace) ||
            (getLangOpts().CPlusPlus && Tok.is(tok::colon)) ||
            isCXX11FinalKeyword());
-    if (getLangOpts().CPlusPlus)
+    if (SkipBody.ShouldSkip)
+      SkipCXXMemberSpecification(StartLoc, AttrFixitLoc, TagType,
+                                 TagOrTempResult.get());
+    else if (getLangOpts().CPlusPlus)
       ParseCXXMemberSpecification(StartLoc, AttrFixitLoc, attrs, TagType,
                                   TagOrTempResult.get());
     else
@@ -1882,51 +1901,40 @@ AccessSpecifier Parser::getAccessSpecifierIfPresent() const {
 /// the class definition.
 void Parser::HandleMemberFunctionDeclDelays(Declarator& DeclaratorInfo,
                                             Decl *ThisDecl) {
-  // We just declared a member function. If this member function
-  // has any default arguments or an exception-specification, we'll need to
-  // parse them later.
-  LateParsedMethodDeclaration *LateMethod = nullptr;
   DeclaratorChunk::FunctionTypeInfo &FTI
     = DeclaratorInfo.getFunctionTypeInfo();
+  // If there was a late-parsed exception-specification, we'll need a
+  // late parse
+  bool NeedLateParse = FTI.getExceptionSpecType() == EST_Unparsed;
+
+  if (!NeedLateParse) {
+    // Look ahead to see if there are any default args
+    for (unsigned ParamIdx = 0; ParamIdx < FTI.NumParams; ++ParamIdx) {
+      auto Param = cast<ParmVarDecl>(FTI.Params[ParamIdx].Param);
+      if (Param->hasUnparsedDefaultArg()) {
+        NeedLateParse = true;
+        break;
+      }
+    }
+  }
 
-  // If there was a late-parsed exception-specification, hold onto its tokens.
-  if (FTI.getExceptionSpecType() == EST_Unparsed) {
+  if (NeedLateParse) {
     // Push this method onto the stack of late-parsed method
     // declarations.
-    LateMethod = new LateParsedMethodDeclaration(this, ThisDecl);
+    auto LateMethod = new LateParsedMethodDeclaration(this, ThisDecl);
     getCurrentClass().LateParsedDeclarations.push_back(LateMethod);
     LateMethod->TemplateScope = getCurScope()->isTemplateParamScope();
 
-    // Stash the exception-specification tokens in the late-pased mthod.
+    // Stash the exception-specification tokens in the late-pased method.
     LateMethod->ExceptionSpecTokens = FTI.ExceptionSpecTokens;
     FTI.ExceptionSpecTokens = 0;
 
-    // Reserve space for the parameters.
+    // Push tokens for each parameter.  Those that do not have
+    // defaults will be NULL.
     LateMethod->DefaultArgs.reserve(FTI.NumParams);
-  }
-
-  for (unsigned ParamIdx = 0; ParamIdx < FTI.NumParams; ++ParamIdx) {
-    if (LateMethod || FTI.Params[ParamIdx].DefaultArgTokens) {
-      if (!LateMethod) {
-        // Push this method onto the stack of late-parsed method
-        // declarations.
-        LateMethod = new LateParsedMethodDeclaration(this, ThisDecl);
-        getCurrentClass().LateParsedDeclarations.push_back(LateMethod);
-        LateMethod->TemplateScope = getCurScope()->isTemplateParamScope();
-
-        // Add all of the parameters prior to this one (they don't
-        // have default arguments).
-        LateMethod->DefaultArgs.reserve(FTI.NumParams);
-        for (unsigned I = 0; I < ParamIdx; ++I)
-          LateMethod->DefaultArgs.push_back(
-              LateParsedDefaultArgument(FTI.Params[I].Param));
-      }
-
-      // Add this parameter to the list of parameters (it may or may
-      // not have a default argument).
+    for (unsigned ParamIdx = 0; ParamIdx < FTI.NumParams; ++ParamIdx)
       LateMethod->DefaultArgs.push_back(LateParsedDefaultArgument(
-          FTI.Params[ParamIdx].Param, FTI.Params[ParamIdx].DefaultArgTokens));
-    }
+        FTI.Params[ParamIdx].Param, FTI.Params[ParamIdx].DefaultArgTokens));
   }
 }
 
@@ -2019,7 +2027,7 @@ bool Parser::isCXX11FinalKeyword() const {
 
 /// \brief Parse a C++ member-declarator up to, but not including, the optional
 /// brace-or-equal-initializer or pure-specifier.
-void Parser::ParseCXXMemberDeclaratorBeforeInitializer(
+bool Parser::ParseCXXMemberDeclaratorBeforeInitializer(
     Declarator &DeclaratorInfo, VirtSpecifiers &VS, ExprResult &BitfieldSize,
     LateParsedAttrList &LateParsedAttrs) {
   // member-declarator:
@@ -2037,10 +2045,13 @@ void Parser::ParseCXXMemberDeclaratorBeforeInitializer(
     BitfieldSize = ParseConstantExpression();
     if (BitfieldSize.isInvalid())
       SkipUntil(tok::comma, StopAtSemi | StopBeforeMatch);
-  } else
+  } else {
     ParseOptionalCXX11VirtSpecifierSeq(
         VS, getCurrentClass().IsInterface,
         DeclaratorInfo.getDeclSpec().getFriendSpecLoc());
+    if (!VS.isUnset())
+      MaybeParseAndDiagnoseDeclSpecAfterCXX11VirtSpecifierSeq(DeclaratorInfo, VS);
+  }
 
   // If a simple-asm-expr is present, parse it.
   if (Tok.is(tok::kw_asm)) {
@@ -2071,6 +2082,78 @@ void Parser::ParseCXXMemberDeclaratorBeforeInitializer(
           Diag(Attr->getLoc(), diag::warn_gcc_attribute_location);
         Attr = Attr->getNext();
       }
+      MaybeParseAndDiagnoseDeclSpecAfterCXX11VirtSpecifierSeq(DeclaratorInfo, VS);
+    }
+  }
+
+  // If this has neither a name nor a bit width, something has gone seriously
+  // wrong. Skip until the semi-colon or }.
+  if (!DeclaratorInfo.hasName() && BitfieldSize.isUnset()) {
+    // If so, skip until the semi-colon or a }.
+    SkipUntil(tok::r_brace, StopAtSemi | StopBeforeMatch);
+    return true;
+  }
+  return false;
+}
+
+/// \brief Look for declaration specifiers possibly occurring after C++11
+/// virt-specifier-seq and diagnose them.
+void Parser::MaybeParseAndDiagnoseDeclSpecAfterCXX11VirtSpecifierSeq(
+    Declarator &D,
+    VirtSpecifiers &VS) {
+  DeclSpec DS(AttrFactory);
+
+  // GNU-style and C++11 attributes are not allowed here, but they will be
+  // handled by the caller.  Diagnose everything else.
+  ParseTypeQualifierListOpt(DS, AR_NoAttributesParsed, false);
+  D.ExtendWithDeclSpec(DS);
+
+  if (D.isFunctionDeclarator()) {
+    auto &Function = D.getFunctionTypeInfo();
+    if (DS.getTypeQualifiers() != DeclSpec::TQ_unspecified) {
+      auto DeclSpecCheck = [&] (DeclSpec::TQ TypeQual,
+                                const char *FixItName,
+                                SourceLocation SpecLoc,
+                                unsigned* QualifierLoc) {
+        FixItHint Insertion;
+        if (DS.getTypeQualifiers() & TypeQual) {
+          if (!(Function.TypeQuals & TypeQual)) {
+            std::string Name(FixItName);
+            Name += " ";
+            Insertion = FixItHint::CreateInsertion(VS.getFirstLocation(), Name.c_str());
+            Function.TypeQuals |= TypeQual;
+            *QualifierLoc = SpecLoc.getRawEncoding();
+          }
+          Diag(SpecLoc, diag::err_declspec_after_virtspec)
+            << FixItName
+            << VirtSpecifiers::getSpecifierName(VS.getLastSpecifier())
+            << FixItHint::CreateRemoval(SpecLoc)
+            << Insertion;
+        }
+      };
+      DeclSpecCheck(DeclSpec::TQ_const, "const", DS.getConstSpecLoc(),
+                    &Function.ConstQualifierLoc);
+      DeclSpecCheck(DeclSpec::TQ_volatile, "volatile", DS.getVolatileSpecLoc(),
+                    &Function.VolatileQualifierLoc);
+      DeclSpecCheck(DeclSpec::TQ_restrict, "restrict", DS.getRestrictSpecLoc(),
+                    &Function.RestrictQualifierLoc);
+    }
+
+    // Parse ref-qualifiers.
+    bool RefQualifierIsLValueRef = true;
+    SourceLocation RefQualifierLoc;
+    if (ParseRefQualifier(RefQualifierIsLValueRef, RefQualifierLoc)) {
+      const char *Name = (RefQualifierIsLValueRef ? "& " : "&& ");
+      FixItHint Insertion = FixItHint::CreateInsertion(VS.getFirstLocation(), Name);
+      Function.RefQualifierIsLValueRef = RefQualifierIsLValueRef;
+      Function.RefQualifierLoc = RefQualifierLoc.getRawEncoding();
+
+      Diag(RefQualifierLoc, diag::err_declspec_after_virtspec)
+        << (RefQualifierIsLValueRef ? "&" : "&&")
+        << VirtSpecifiers::getSpecifierName(VS.getLastSpecifier())
+        << FixItHint::CreateRemoval(RefQualifierLoc)
+        << Insertion;
+      D.SetRangeEnd(RefQualifierLoc);
     }
   }
 }
@@ -2298,14 +2381,8 @@ void Parser::ParseCXXClassMemberDeclaration(AccessSpecifier AS,
   bool ExpectSemi = true;
 
   // Parse the first declarator.
-  ParseCXXMemberDeclaratorBeforeInitializer(DeclaratorInfo, VS, BitfieldSize,
-                                            LateParsedAttrs);
-
-  // If this has neither a name nor a bit width, something has gone seriously
-  // wrong. Skip until the semi-colon or }.
-  if (!DeclaratorInfo.hasName() && BitfieldSize.isUnset()) {
-    // If so, skip until the semi-colon or a }.
-    SkipUntil(tok::r_brace, StopAtSemi | StopBeforeMatch);
+  if (ParseCXXMemberDeclaratorBeforeInitializer(
+          DeclaratorInfo, VS, BitfieldSize, LateParsedAttrs)) {
     TryConsumeToken(tok::semi);
     return;
   }
@@ -2344,6 +2421,7 @@ void Parser::ParseCXXClassMemberDeclaration(AccessSpecifier AS,
           DefinitionKind = FDK_Deleted;
       }
     }
+    DeclaratorInfo.setFunctionDefinitionKind(DefinitionKind);
 
     // C++11 [dcl.attr.grammar] p4: If an attribute-specifier-seq appertains 
     // to a friend declaration, that declaration shall be a definition.
@@ -2354,7 +2432,7 @@ void Parser::ParseCXXClassMemberDeclaration(AccessSpecifier AS,
       ProhibitAttributes(FnAttrs);
     }
 
-    if (DefinitionKind) {
+    if (DefinitionKind != FDK_Declaration) {
       if (!DeclaratorInfo.isFunctionDeclarator()) {
         Diag(DeclaratorInfo.getIdentifierLoc(), diag::err_func_def_no_params);
         ConsumeBrace();
@@ -2376,7 +2454,7 @@ void Parser::ParseCXXClassMemberDeclaration(AccessSpecifier AS,
 
       Decl *FunDecl =
         ParseCXXInlineMethodDef(AS, AccessAttrs, DeclaratorInfo, TemplateInfo,
-                                VS, DefinitionKind, Init);
+                                VS, Init);
 
       if (FunDecl) {
         for (unsigned i = 0, ni = CommonLateParsedAttrs.size(); i < ni; ++i) {
@@ -2530,16 +2608,17 @@ void Parser::ParseCXXClassMemberDeclaration(AccessSpecifier AS,
     // Parse the next declarator.
     DeclaratorInfo.clear();
     VS.clear();
-    BitfieldSize = true;
-    Init = true;
+    BitfieldSize = ExprResult(/*Invalid=*/false);
+    Init = ExprResult(/*Invalid=*/false);
     HasInitializer = false;
     DeclaratorInfo.setCommaLoc(CommaLoc);
 
     // GNU attributes are allowed before the second and subsequent declarator.
     MaybeParseGNUAttributes(DeclaratorInfo);
 
-    ParseCXXMemberDeclaratorBeforeInitializer(DeclaratorInfo, VS, BitfieldSize,
-                                              LateParsedAttrs);
+    if (ParseCXXMemberDeclaratorBeforeInitializer(
+            DeclaratorInfo, VS, BitfieldSize, LateParsedAttrs))
+      break;
   }
 
   if (ExpectSemi &&
@@ -2617,6 +2696,55 @@ ExprResult Parser::ParseCXXMemberInitializer(Decl *D, bool IsFunction,
   return ParseInitializer();
 }
 
+void Parser::SkipCXXMemberSpecification(SourceLocation RecordLoc,
+                                        SourceLocation AttrFixitLoc,
+                                        unsigned TagType, Decl *TagDecl) {
+  // Skip the optional 'final' keyword.
+  if (getLangOpts().CPlusPlus && Tok.is(tok::identifier)) {
+    assert(isCXX11FinalKeyword() && "not a class definition");
+    ConsumeToken();
+
+    // Diagnose any C++11 attributes after 'final' keyword.
+    // We deliberately discard these attributes.
+    ParsedAttributesWithRange Attrs(AttrFactory);
+    CheckMisplacedCXX11Attribute(Attrs, AttrFixitLoc);
+
+    // This can only happen if we had malformed misplaced attributes;
+    // we only get called if there is a colon or left-brace after the
+    // attributes.
+    if (Tok.isNot(tok::colon) && Tok.isNot(tok::l_brace))
+      return;
+  }
+
+  // Skip the base clauses. This requires actually parsing them, because
+  // otherwise we can't be sure where they end (a left brace may appear
+  // within a template argument).
+  if (Tok.is(tok::colon)) {
+    // Enter the scope of the class so that we can correctly parse its bases.
+    ParseScope ClassScope(this, Scope::ClassScope|Scope::DeclScope);
+    ParsingClassDefinition ParsingDef(*this, TagDecl, /*NonNestedClass*/ true,
+                                      TagType == DeclSpec::TST_interface);
+    Actions.ActOnTagStartSkippedDefinition(getCurScope(), TagDecl);
+
+    // Parse the bases but don't attach them to the class.
+    ParseBaseClause(nullptr);
+
+    Actions.ActOnTagFinishSkippedDefinition();
+
+    if (!Tok.is(tok::l_brace)) {
+      Diag(PP.getLocForEndOfToken(PrevTokLocation),
+           diag::err_expected_lbrace_after_base_specifiers);
+      return;
+    }
+  }
+
+  // Skip the body.
+  assert(Tok.is(tok::l_brace));
+  BalancedDelimiterTracker T(*this, tok::l_brace);
+  T.consumeOpen();
+  T.skipToEnd();
+}
+
 /// ParseCXXMemberSpecification - Parse the class definition.
 ///
 ///       member-specification:
@@ -2911,6 +3039,10 @@ void Parser::ParseCXXMemberSpecification(SourceLocation RecordLoc,
     ParseLexedMemberInitializers(getCurrentClass());
     ParseLexedMethodDefs(getCurrentClass());
     PrevTokLocation = SavedPrevTokLocation;
+
+    // We've finished parsing everything, including default argument
+    // initializers.
+    Actions.ActOnFinishCXXMemberDefaultArgs(TagDecl);
   }
 
   if (TagDecl)
@@ -2965,9 +3097,11 @@ void Parser::DiagnoseUnexpectedNamespace(NamedDecl *D) {
 ///          mem-initializer ...[opt]
 ///          mem-initializer ...[opt] , mem-initializer-list
 void Parser::ParseConstructorInitializer(Decl *ConstructorDecl) {
-  assert(Tok.is(tok::colon) && "Constructor initializer always starts with ':'");
+  assert(Tok.is(tok::colon) &&
+         "Constructor initializer always starts with ':'");
 
-  // Poison the SEH identifiers so they are flagged as illegal in constructor initializers
+  // Poison the SEH identifiers so they are flagged as illegal in constructor
+  // initializers.
   PoisonSEHIdentifiersRAIIObject PoisonSEHIdentifiers(*this, true);
   SourceLocation ColonLoc = ConsumeToken();
 
@@ -3396,7 +3530,9 @@ IdentifierInfo *Parser::TryParseCXX11AttributeIdentifier(SourceLocation &Loc) {
     // Alternative tokens do not have identifier info, but their spelling
     // starts with an alphabetical character.
     SmallString<8> SpellingBuf;
-    StringRef Spelling = PP.getSpelling(Tok.getLocation(), SpellingBuf);
+    SourceLocation SpellingLoc =
+        PP.getSourceManager().getSpellingLoc(Tok.getLocation());
+    StringRef Spelling = PP.getSpelling(SpellingLoc, SpellingBuf);
     if (isLetter(Spelling[0])) {
       Loc = ConsumeToken();
       return &PP.getIdentifierTable().get(Spelling);
@@ -3474,7 +3610,6 @@ bool Parser::ParseCXX11AttributeArgs(IdentifierInfo *AttrName,
         // The attribute was allowed to have arguments, but none were provided
         // even though the attribute parsed successfully. This is an error.
         Diag(LParenLoc, diag::err_attribute_requires_arguments) << AttrName;
-        return false;
       } else if (!Attr->getMaxArgs()) {
         // The attribute parsed successfully, but was not allowed to have any
         // arguments. It doesn't matter whether any were provided -- the
@@ -3482,7 +3617,6 @@ bool Parser::ParseCXX11AttributeArgs(IdentifierInfo *AttrName,
         Diag(LParenLoc, diag::err_cxx11_attribute_forbids_arguments)
             << AttrName
             << FixItHint::CreateRemoval(SourceRange(LParenLoc, *EndLoc));
-        return false;
       }
     }
   }
diff --git a/lib/Parse/ParseExpr.cpp b/lib/Parse/ParseExpr.cpp
index d0d97de84c10..95a28a8d5a14 100644
--- a/lib/Parse/ParseExpr.cpp
+++ b/lib/Parse/ParseExpr.cpp
@@ -347,7 +347,11 @@ Parser::ParseRHSOfBinaryExpression(ExprResult LHS, prec::Level MinPrec) {
       RHS = ParseCastExpression(false);
 
     if (RHS.isInvalid()) {
+      // FIXME: Errors generated by the delayed typo correction should be
+      // printed before errors from parsing the RHS, not after.
       Actions.CorrectDelayedTyposInExpr(LHS);
+      if (TernaryMiddle.isUsable())
+        TernaryMiddle = Actions.CorrectDelayedTyposInExpr(TernaryMiddle);
       LHS = ExprError();
     }
 
@@ -380,7 +384,11 @@ Parser::ParseRHSOfBinaryExpression(ExprResult LHS, prec::Level MinPrec) {
       RHSIsInitList = false;
 
       if (RHS.isInvalid()) {
+        // FIXME: Errors generated by the delayed typo correction should be
+        // printed before errors from ParseRHSOfBinaryExpression, not after.
         Actions.CorrectDelayedTyposInExpr(LHS);
+        if (TernaryMiddle.isUsable())
+          TernaryMiddle = Actions.CorrectDelayedTyposInExpr(TernaryMiddle);
         LHS = ExprError();
       }
 
@@ -446,8 +454,8 @@ ExprResult Parser::ParseCastExpression(bool isUnaryExpression,
 namespace {
 class CastExpressionIdValidator : public CorrectionCandidateCallback {
  public:
-  CastExpressionIdValidator(bool AllowTypes, bool AllowNonTypes)
-      : AllowNonTypes(AllowNonTypes) {
+  CastExpressionIdValidator(Token Next, bool AllowTypes, bool AllowNonTypes)
+      : NextToken(Next), AllowNonTypes(AllowNonTypes) {
     WantTypeSpecifiers = WantFunctionLikeCasts = AllowTypes;
   }
 
@@ -458,11 +466,24 @@ class CastExpressionIdValidator : public CorrectionCandidateCallback {
 
     if (isa<TypeDecl>(ND))
       return WantTypeSpecifiers;
-    return AllowNonTypes &&
-           CorrectionCandidateCallback::ValidateCandidate(candidate);
+
+    if (!AllowNonTypes || !CorrectionCandidateCallback::ValidateCandidate(candidate))
+      return false;
+
+    if (!(NextToken.is(tok::equal) || NextToken.is(tok::arrow) ||
+          NextToken.is(tok::period)))
+      return true;
+
+    for (auto *C : candidate) {
+      NamedDecl *ND = C->getUnderlyingDecl();
+      if (isa<ValueDecl>(ND) && !isa<FunctionDecl>(ND))
+        return true;
+    }
+    return false;
   }
 
  private:
+  Token NextToken;
   bool AllowNonTypes;
 };
 }
@@ -908,14 +929,20 @@ ExprResult Parser::ParseCastExpression(bool isUnaryExpression,
     SourceLocation TemplateKWLoc;
     Token Replacement;
     auto Validator = llvm::make_unique<CastExpressionIdValidator>(
-        isTypeCast != NotTypeCast, isTypeCast != IsTypeCast);
+        Tok, isTypeCast != NotTypeCast, isTypeCast != IsTypeCast);
     Validator->IsAddressOfOperand = isAddressOfOperand;
-    Validator->WantRemainingKeywords = Tok.isNot(tok::r_paren);
+    if (Tok.is(tok::periodstar) || Tok.is(tok::arrowstar)) {
+      Validator->WantExpressionKeywords = false;
+      Validator->WantRemainingKeywords = false;
+    } else {
+      Validator->WantRemainingKeywords = Tok.isNot(tok::r_paren);
+    }
     Name.setIdentifier(&II, ILoc);
     Res = Actions.ActOnIdExpression(
         getCurScope(), ScopeSpec, TemplateKWLoc, Name, Tok.is(tok::l_paren),
         isAddressOfOperand, std::move(Validator),
-        /*IsInlineAsmIdentifier=*/false, &Replacement);
+        /*IsInlineAsmIdentifier=*/false,
+        Tok.is(tok::r_paren) ? nullptr : &Replacement);
     if (!Res.isInvalid() && !Res.get()) {
       UnconsumeToken(Replacement);
       return ParseCastExpression(isUnaryExpression, isAddressOfOperand,
@@ -1441,10 +1468,14 @@ Parser::ParsePostfixExpressionSuffix(ExprResult LHS) {
 
       if (OpKind == tok::l_paren || !LHS.isInvalid()) {
         if (Tok.isNot(tok::r_paren)) {
-          if (ParseExpressionList(ArgExprs, CommaLocs, &Sema::CodeCompleteCall,
-                                  LHS.get())) {
+          if (ParseExpressionList(ArgExprs, CommaLocs, [&] {
+                Actions.CodeCompleteCall(getCurScope(), LHS.get(), ArgExprs);
+             })) {
             (void)Actions.CorrectDelayedTyposInExpr(LHS);
             LHS = ExprError();
+          } else if (LHS.isInvalid()) {
+            for (auto &E : ArgExprs)
+              Actions.CorrectDelayedTyposInExpr(E);
           }
         }
       }
@@ -1453,7 +1484,19 @@ Parser::ParsePostfixExpressionSuffix(ExprResult LHS) {
       if (LHS.isInvalid()) {
         SkipUntil(tok::r_paren, StopAtSemi);
       } else if (Tok.isNot(tok::r_paren)) {
-        PT.consumeClose();
+        bool HadDelayedTypo = false;
+        if (Actions.CorrectDelayedTyposInExpr(LHS).get() != LHS.get())
+          HadDelayedTypo = true;
+        for (auto &E : ArgExprs)
+          if (Actions.CorrectDelayedTyposInExpr(E).get() != E)
+            HadDelayedTypo = true;
+        // If there were delayed typos in the LHS or ArgExprs, call SkipUntil
+        // instead of PT.consumeClose() to avoid emitting extra diagnostics for
+        // the unmatched l_paren.
+        if (HadDelayedTypo)
+          SkipUntil(tok::r_paren, StopAtSemi);
+        else
+          PT.consumeClose();
         LHS = ExprError();
       } else {
         assert((ArgExprs.size() == 0 || 
@@ -1510,14 +1553,14 @@ Parser::ParsePostfixExpressionSuffix(ExprResult LHS) {
         cutOffParsing();
         return ExprError();
       }
-      
+
       if (MayBePseudoDestructor && !LHS.isInvalid()) {
         LHS = ParseCXXPseudoDestructor(LHS.get(), OpLoc, OpKind, SS, 
                                        ObjectType);
         break;
       }
 
-      // Either the action has told is that this cannot be a
+      // Either the action has told us that this cannot be a
       // pseudo-destructor expression (based on the type of base
       // expression), or we didn't see a '~' in the right place. We
       // can still parse a destructor name here, but in that case it
@@ -1526,7 +1569,8 @@ Parser::ParsePostfixExpressionSuffix(ExprResult LHS) {
       // FIXME: Add support for explicit call of template constructor.
       SourceLocation TemplateKWLoc;
       UnqualifiedId Name;
-      if (getLangOpts().ObjC2 && OpKind == tok::period && Tok.is(tok::kw_class)) {
+      if (getLangOpts().ObjC2 && OpKind == tok::period &&
+          Tok.is(tok::kw_class)) {
         // Objective-C++:
         //   After a '.' in a member access expression, treat the keyword
         //   'class' as if it were an identifier.
@@ -1551,8 +1595,7 @@ Parser::ParsePostfixExpressionSuffix(ExprResult LHS) {
         LHS = Actions.ActOnMemberAccessExpr(getCurScope(), LHS.get(), OpLoc, 
                                             OpKind, SS, TemplateKWLoc, Name,
                                  CurParsedObjCImpl ? CurParsedObjCImpl->Dcl
-                                                   : nullptr,
-                                            Tok.is(tok::l_paren));
+                                                   : nullptr);
       break;
     }
     case tok::plusplus:    // postfix-expression: postfix-expression '++'
@@ -2088,6 +2131,17 @@ Parser::ParseParenExpression(ParenParseOption &ExprType, bool stopIfCastExpr,
     if (!getCurScope()->getFnParent() && !getCurScope()->getBlockParent()) {
       Result = ExprError(Diag(OpenLoc, diag::err_stmtexpr_file_scope));
     } else {
+      // Find the nearest non-record decl context. Variables declared in a
+      // statement expression behave as if they were declared in the enclosing
+      // function, block, or other code construct.
+      DeclContext *CodeDC = Actions.CurContext;
+      while (CodeDC->isRecord() || isa<EnumDecl>(CodeDC)) {
+        CodeDC = CodeDC->getParent();
+        assert(CodeDC && !CodeDC->isFileContext() &&
+               "statement expr not in code context");
+      }
+      Sema::ContextRAII SavedContext(Actions, CodeDC, /*NewThisContext=*/false);
+
       Actions.ActOnStartStmtExpr();
 
       StmtResult Stmt(ParseCompoundStatement(true));
@@ -2256,6 +2310,11 @@ Parser::ParseParenExpression(ParenParseOption &ExprType, bool stopIfCastExpr,
     InMessageExpressionRAIIObject InMessage(*this, false);
 
     Result = ParseExpression(MaybeTypeCast);
+    if (!getLangOpts().CPlusPlus && MaybeTypeCast && Result.isUsable()) {
+      // Correct typos in non-C++ code earlier so that implicit-cast-like
+      // expressions are parsed correctly.
+      Result = Actions.CorrectDelayedTyposInExpr(Result);
+    }
     ExprType = SimpleExpr;
 
     if (isFoldOperator(Tok.getKind()) && NextToken().is(tok::ellipsis))
@@ -2354,7 +2413,8 @@ ExprResult Parser::ParseGenericSelectionExpression() {
     // C11 6.5.1.1p3 "The controlling expression of a generic selection is
     // not evaluated."
     EnterExpressionEvaluationContext Unevaluated(Actions, Sema::Unevaluated);
-    ControllingExpr = ParseAssignmentExpression();
+    ControllingExpr =
+        Actions.CorrectDelayedTyposInExpr(ParseAssignmentExpression());
     if (ControllingExpr.isInvalid()) {
       SkipUntil(tok::r_paren, StopAtSemi);
       return ExprError();
@@ -2400,7 +2460,8 @@ ExprResult Parser::ParseGenericSelectionExpression() {
 
     // FIXME: These expressions should be parsed in a potentially potentially
     // evaluated context.
-    ExprResult ER(ParseAssignmentExpression());
+    ExprResult ER(
+        Actions.CorrectDelayedTyposInExpr(ParseAssignmentExpression()));
     if (ER.isInvalid()) {
       SkipUntil(tok::r_paren, StopAtSemi);
       return ExprError();
@@ -2493,17 +2554,14 @@ ExprResult Parser::ParseFoldExpression(ExprResult LHS,
 /// [C++0x]   assignment-expression
 /// [C++0x]   braced-init-list
 /// \endverbatim
-bool Parser::ParseExpressionList(SmallVectorImpl<Expr*> &Exprs,
+bool Parser::ParseExpressionList(SmallVectorImpl<Expr *> &Exprs,
                                  SmallVectorImpl<SourceLocation> &CommaLocs,
-                                 void (Sema::*Completer)(Scope *S,
-                                                         Expr *Data,
-                                                         ArrayRef<Expr *> Args),
-                                 Expr *Data) {
+                                 std::function<void()> Completer) {
   bool SawError = false;
   while (1) {
     if (Tok.is(tok::code_completion)) {
       if (Completer)
-        (Actions.*Completer)(getCurScope(), Data, Exprs);
+        Completer();
       else
         Actions.CodeCompleteOrdinaryName(getCurScope(), Sema::PCC_Expression);
       cutOffParsing();
diff --git a/lib/Parse/ParseExprCXX.cpp b/lib/Parse/ParseExprCXX.cpp
index 67496edf8c4b..ed9f75d7b4de 100644
--- a/lib/Parse/ParseExprCXX.cpp
+++ b/lib/Parse/ParseExprCXX.cpp
@@ -118,6 +118,7 @@ void Parser::CheckForLParenAfterColonColon() {
   // Eat the '('.
   ConsumeParen();
   Token RParen;
+  RParen.setLocation(SourceLocation());
   // Do we have a ')' ?
   NextTok = StarTok.is(tok::star) ? GetLookAheadToken(2) : GetLookAheadToken(1);
   if (NextTok.is(tok::r_paren)) {
@@ -194,6 +195,7 @@ bool Parser::ParseOptionalCXXScopeSpecifier(CXXScopeSpec &SS,
 
   if (Tok.is(tok::annot_cxxscope)) {
     assert(!LastII && "want last identifier but have already annotated scope");
+    assert(!MayBePseudoDestructor && "unexpected annot_cxxscope");
     Actions.RestoreNestedNameSpecifierAnnotation(Tok.getAnnotationValue(),
                                                  Tok.getAnnotationRange(),
                                                  SS);
@@ -208,6 +210,13 @@ bool Parser::ParseOptionalCXXScopeSpecifier(CXXScopeSpec &SS,
     SS = TemplateId->SS;
   }
 
+  // Has to happen before any "return false"s in this function.
+  bool CheckForDestructor = false;
+  if (MayBePseudoDestructor && *MayBePseudoDestructor) {
+    CheckForDestructor = true;
+    *MayBePseudoDestructor = false;
+  }
+
   if (LastII)
     *LastII = nullptr;
 
@@ -244,12 +253,6 @@ bool Parser::ParseOptionalCXXScopeSpecifier(CXXScopeSpec &SS,
     return Actions.ActOnSuperScopeSpecifier(SuperLoc, ConsumeToken(), SS);
   }
 
-  bool CheckForDestructor = false;
-  if (MayBePseudoDestructor && *MayBePseudoDestructor) {
-    CheckForDestructor = true;
-    *MayBePseudoDestructor = false;
-  }
-
   if (!HasScopeSpecifier &&
       (Tok.is(tok::kw_decltype) || Tok.is(tok::annot_decltype))) {
     DeclSpec DS(AttrFactory);
@@ -659,7 +662,8 @@ ExprResult Parser::ParseCXXIdExpression(bool isAddressOfOperand) {
   ParseOptionalCXXScopeSpecifier(SS, ParsedType(), /*EnteringContext=*/false);
 
   Token Replacement;
-  ExprResult Result = tryParseCXXIdExpression(SS, isAddressOfOperand, Replacement);
+  ExprResult Result =
+      tryParseCXXIdExpression(SS, isAddressOfOperand, Replacement);
   if (Result.isUnset()) {
     // If the ExprResult is valid but null, then typo correction suggested a
     // keyword replacement that needs to be reparsed.
@@ -1090,6 +1094,10 @@ ExprResult Parser::ParseLambdaExpressionAfterIntroducer(
     // compatible with GCC.
     MaybeParseGNUAttributes(Attr, &DeclEndLoc);
 
+    // MSVC-style attributes must be parsed before the mutable specifier to be
+    // compatible with MSVC.
+    MaybeParseMicrosoftDeclSpecs(Attr, &DeclEndLoc);
+
     // Parse 'mutable'[opt].
     SourceLocation MutableLoc;
     if (TryConsumeToken(tok::kw_mutable, MutableLoc))
@@ -1485,9 +1493,8 @@ Parser::ParseCXXPseudoDestructor(Expr *Base, SourceLocation OpLoc,
     ParseDecltypeSpecifier(DS);
     if (DS.getTypeSpecType() == TST_error)
       return ExprError();
-    return Actions.ActOnPseudoDestructorExpr(getCurScope(), Base, OpLoc, 
-                                             OpKind, TildeLoc, DS, 
-                                             Tok.is(tok::l_paren));
+    return Actions.ActOnPseudoDestructorExpr(getCurScope(), Base, OpLoc, OpKind,
+                                             TildeLoc, DS);
   }
 
   if (!Tok.is(tok::identifier)) {
@@ -1510,11 +1517,9 @@ Parser::ParseCXXPseudoDestructor(Expr *Base, SourceLocation OpLoc,
                                    /*AssumeTemplateName=*/true))
     return ExprError();
 
-  return Actions.ActOnPseudoDestructorExpr(getCurScope(), Base,
-                                           OpLoc, OpKind,
-                                           SS, FirstTypeName, CCLoc,
-                                           TildeLoc, SecondTypeName,
-                                           Tok.is(tok::l_paren));
+  return Actions.ActOnPseudoDestructorExpr(getCurScope(), Base, OpLoc, OpKind,
+                                           SS, FirstTypeName, CCLoc, TildeLoc,
+                                           SecondTypeName);
 }
 
 /// ParseCXXBoolLiteral - This handles the C++ Boolean literals.
@@ -1602,7 +1607,11 @@ Parser::ParseCXXTypeConstructExpression(const DeclSpec &DS) {
     CommaLocsTy CommaLocs;
 
     if (Tok.isNot(tok::r_paren)) {
-      if (ParseExpressionList(Exprs, CommaLocs)) {
+      if (ParseExpressionList(Exprs, CommaLocs, [&] {
+            Actions.CodeCompleteConstructor(getCurScope(),
+                                      TypeRep.get()->getCanonicalTypeInternal(),
+                                            DS.getLocEnd(), Exprs);
+         })) {
         SkipUntil(tok::r_paren, StopAtSemi);
         return ExprError();
       }
@@ -2509,14 +2518,23 @@ bool Parser::ParseUnqualifiedId(CXXScopeSpec &SS, bool EnteringContext,
     }
 
     // If the user wrote ~T::T, correct it to T::~T.
+    DeclaratorScopeObj DeclScopeObj(*this, SS);
     if (!TemplateSpecified && NextToken().is(tok::coloncolon)) {
+      // Don't let ParseOptionalCXXScopeSpecifier() "correct"
+      // `int A; struct { ~A::A(); };` to `int A; struct { ~A:A(); };`,
+      // it will confuse this recovery logic.
+      ColonProtectionRAIIObject ColonRAII(*this, false);
+
       if (SS.isSet()) {
         AnnotateScopeToken(SS, /*NewAnnotation*/true);
         SS.clear();
       }
       if (ParseOptionalCXXScopeSpecifier(SS, ObjectType, EnteringContext))
         return true;
-      if (Tok.isNot(tok::identifier) || NextToken().is(tok::coloncolon)) {
+      if (SS.isNotEmpty())
+        ObjectType = ParsedType();
+      if (Tok.isNot(tok::identifier) || NextToken().is(tok::coloncolon) ||
+          !SS.isSet()) {
         Diag(TildeLoc, diag::err_destructor_tilde_scope);
         return true;
       }
@@ -2525,6 +2543,10 @@ bool Parser::ParseUnqualifiedId(CXXScopeSpec &SS, bool EnteringContext,
       Diag(TildeLoc, diag::err_destructor_tilde_scope)
         << FixItHint::CreateRemoval(TildeLoc)
         << FixItHint::CreateInsertion(Tok.getLocation(), "~");
+
+      // Temporarily enter the scope for the rest of this function.
+      if (Actions.ShouldEnterDeclaratorScope(getCurScope(), SS))
+        DeclScopeObj.EnterDeclaratorScope();
     }
 
     // Parse the class-name (or template-name in a simple-template-id).
@@ -2668,7 +2690,14 @@ Parser::ParseCXXNewExpression(bool UseGlobal, SourceLocation Start) {
     ConstructorLParen = T.getOpenLocation();
     if (Tok.isNot(tok::r_paren)) {
       CommaLocsTy CommaLocs;
-      if (ParseExpressionList(ConstructorArgs, CommaLocs)) {
+      if (ParseExpressionList(ConstructorArgs, CommaLocs, [&] {
+            ParsedType TypeRep = Actions.ActOnTypeName(getCurScope(),
+                                                       DeclaratorInfo).get();
+            Actions.CodeCompleteConstructor(getCurScope(),
+                                      TypeRep.get()->getCanonicalTypeInternal(),
+                                            DeclaratorInfo.getLocEnd(),
+                                            ConstructorArgs);
+      })) {
         SkipUntil(tok::semi, StopAtSemi | StopBeforeMatch);
         return ExprError();
       }
diff --git a/lib/Parse/ParseInit.cpp b/lib/Parse/ParseInit.cpp
index 7fe9862d3992..42287d68b331 100644
--- a/lib/Parse/ParseInit.cpp
+++ b/lib/Parse/ParseInit.cpp
@@ -148,7 +148,7 @@ ExprResult Parser::ParseInitializerWithPotentialDesignator() {
 
     Diag(NameLoc, diag::ext_gnu_old_style_field_designator)
       << FixItHint::CreateReplacement(SourceRange(NameLoc, ColonLoc),
-                                      NewSyntax.str());
+                                      NewSyntax);
 
     Designation D;
     D.AddDesignator(Designator::getField(FieldName, SourceLocation(), NameLoc));
diff --git a/lib/Parse/ParseObjc.cpp b/lib/Parse/ParseObjc.cpp
index a597a1658cf6..691f53f05a8f 100644
--- a/lib/Parse/ParseObjc.cpp
+++ b/lib/Parse/ParseObjc.cpp
@@ -240,7 +240,7 @@ Decl *Parser::ParseObjCAtInterfaceDeclaration(SourceLocation AtLoc,
     SmallVector<Decl *, 8> ProtocolRefs;
     SmallVector<SourceLocation, 8> ProtocolLocs;
     if (Tok.is(tok::less) &&
-        ParseObjCProtocolReferences(ProtocolRefs, ProtocolLocs, true,
+        ParseObjCProtocolReferences(ProtocolRefs, ProtocolLocs, true, true,
                                     LAngleLoc, EndProtoLoc))
       return nullptr;
 
@@ -286,7 +286,7 @@ Decl *Parser::ParseObjCAtInterfaceDeclaration(SourceLocation AtLoc,
   SmallVector<SourceLocation, 8> ProtocolLocs;
   SourceLocation LAngleLoc, EndProtoLoc;
   if (Tok.is(tok::less) &&
-      ParseObjCProtocolReferences(ProtocolRefs, ProtocolLocs, true,
+      ParseObjCProtocolReferences(ProtocolRefs, ProtocolLocs, true, true,
                                   LAngleLoc, EndProtoLoc))
     return nullptr;
 
@@ -1151,7 +1151,7 @@ Decl *Parser::ParseObjCMethodDecl(SourceLocation mLoc,
 bool Parser::
 ParseObjCProtocolReferences(SmallVectorImpl<Decl *> &Protocols,
                             SmallVectorImpl<SourceLocation> &ProtocolLocs,
-                            bool WarnOnDeclarations,
+                            bool WarnOnDeclarations, bool ForObjCContainer,
                             SourceLocation &LAngleLoc, SourceLocation &EndLoc) {
   assert(Tok.is(tok::less) && "expected <");
 
@@ -1186,7 +1186,7 @@ ParseObjCProtocolReferences(SmallVectorImpl<Decl *> &Protocols,
     return true;
 
   // Convert the list of protocols identifiers into a list of protocol decls.
-  Actions.FindProtocolDeclaration(WarnOnDeclarations,
+  Actions.FindProtocolDeclaration(WarnOnDeclarations, ForObjCContainer,
                                   &ProtocolIdents[0], ProtocolIdents.size(),
                                   Protocols);
   return false;
@@ -1201,6 +1201,7 @@ bool Parser::ParseObjCProtocolQualifiers(DeclSpec &DS) {
   SmallVector<Decl *, 8> ProtocolDecl;
   SmallVector<SourceLocation, 8> ProtocolLocs;
   bool Result = ParseObjCProtocolReferences(ProtocolDecl, ProtocolLocs, false,
+                                            false,
                                             LAngleLoc, EndProtoLoc);
   DS.setProtocolQualifiers(ProtocolDecl.data(), ProtocolDecl.size(),
                            ProtocolLocs.data(), LAngleLoc);
@@ -1416,7 +1417,7 @@ Parser::ParseObjCAtProtocolDeclaration(SourceLocation AtLoc,
   SmallVector<Decl *, 8> ProtocolRefs;
   SmallVector<SourceLocation, 8> ProtocolLocs;
   if (Tok.is(tok::less) &&
-      ParseObjCProtocolReferences(ProtocolRefs, ProtocolLocs, false,
+      ParseObjCProtocolReferences(ProtocolRefs, ProtocolLocs, false, true,
                                   LAngleLoc, EndProtoLoc))
     return DeclGroupPtrTy();
 
diff --git a/lib/Parse/ParseOpenMP.cpp b/lib/Parse/ParseOpenMP.cpp
index 764619aae6f3..187289ee637d 100644
--- a/lib/Parse/ParseOpenMP.cpp
+++ b/lib/Parse/ParseOpenMP.cpp
@@ -223,6 +223,7 @@ Parser::ParseOpenMPDeclarativeOrExecutableDirective(bool StandAloneAllowed) {
     ParseScope OMPDirectiveScope(this, ScopeFlags);
     Actions.StartOpenMPDSABlock(DKind, DirName, Actions.getCurScope(), Loc);
 
+    Actions.StartOpenMPClauses();
     while (Tok.isNot(tok::annot_pragma_openmp_end)) {
       OpenMPClauseKind CKind =
           Tok.isAnnotation()
@@ -242,6 +243,7 @@ Parser::ParseOpenMPDeclarativeOrExecutableDirective(bool StandAloneAllowed) {
       if (Tok.is(tok::comma))
         ConsumeToken();
     }
+    Actions.EndOpenMPClauses();
     // End location of the directive.
     EndLoc = Tok.getLocation();
     // Consume final annot_pragma_openmp_end.
@@ -257,13 +259,8 @@ Parser::ParseOpenMPDeclarativeOrExecutableDirective(bool StandAloneAllowed) {
       // Parse statement
       AssociatedStmt = ParseStatement();
       Actions.ActOnFinishOfCompoundStmt();
-      if (!AssociatedStmt.isUsable()) {
-        Actions.ActOnCapturedRegionError();
-        CreateDirective = false;
-      } else {
-        AssociatedStmt = Actions.ActOnCapturedRegionEnd(AssociatedStmt.get());
-        CreateDirective = AssociatedStmt.isUsable();
-      }
+      AssociatedStmt = Actions.ActOnOpenMPRegionEnd(AssociatedStmt, Clauses);
+      CreateDirective = AssociatedStmt.isUsable();
     }
     if (CreateDirective)
       Directive = Actions.ActOnOpenMPExecutableDirective(
diff --git a/lib/Parse/ParsePragma.cpp b/lib/Parse/ParsePragma.cpp
index 473be5467eeb..a8641efb705c 100644
--- a/lib/Parse/ParsePragma.cpp
+++ b/lib/Parse/ParsePragma.cpp
@@ -198,9 +198,12 @@ void Parser::initializePragmaHandlers() {
     OpenMPHandler.reset(new PragmaNoOpenMPHandler());
   PP.AddPragmaHandler(OpenMPHandler.get());
 
-  if (getLangOpts().MicrosoftExt) {
+  if (getLangOpts().MicrosoftExt || getTargetInfo().getTriple().isPS4()) {
     MSCommentHandler.reset(new PragmaCommentHandler(Actions));
     PP.AddPragmaHandler(MSCommentHandler.get());
+  }
+
+  if (getLangOpts().MicrosoftExt) {
     MSDetectMismatchHandler.reset(new PragmaDetectMismatchHandler(Actions));
     PP.AddPragmaHandler(MSDetectMismatchHandler.get());
     MSPointersToMembers.reset(new PragmaMSPointersToMembers());
@@ -261,9 +264,12 @@ void Parser::resetPragmaHandlers() {
   PP.RemovePragmaHandler(OpenMPHandler.get());
   OpenMPHandler.reset();
 
-  if (getLangOpts().MicrosoftExt) {
+  if (getLangOpts().MicrosoftExt || getTargetInfo().getTriple().isPS4()) {
     PP.RemovePragmaHandler(MSCommentHandler.get());
     MSCommentHandler.reset();
+  }
+
+  if (getLangOpts().MicrosoftExt) {
     PP.RemovePragmaHandler(MSDetectMismatchHandler.get());
     MSDetectMismatchHandler.reset();
     PP.RemovePragmaHandler(MSPointersToMembers.get());
@@ -900,6 +906,7 @@ void PragmaGCCVisibilityHandler::HandlePragma(Preprocessor &PP,
       << "visibility";
     return;
   }
+  SourceLocation EndLoc = Tok.getLocation();
   PP.LexUnexpandedToken(Tok);
   if (Tok.isNot(tok::eod)) {
     PP.Diag(Tok.getLocation(), diag::warn_pragma_extra_tokens_at_eol)
@@ -911,6 +918,7 @@ void PragmaGCCVisibilityHandler::HandlePragma(Preprocessor &PP,
   Toks[0].startToken();
   Toks[0].setKind(tok::annot_pragma_vis);
   Toks[0].setLocation(VisLoc);
+  Toks[0].setAnnotationEndLoc(EndLoc);
   Toks[0].setAnnotationValue(
                           const_cast<void*>(static_cast<const void*>(VisType)));
   PP.EnterTokenStream(Toks, 1, /*DisableMacroExpansion=*/true,
@@ -1030,6 +1038,7 @@ void PragmaPackHandler::HandlePragma(Preprocessor &PP,
   Toks[0].startToken();
   Toks[0].setKind(tok::annot_pragma_pack);
   Toks[0].setLocation(PackLoc);
+  Toks[0].setAnnotationEndLoc(RParenLoc);
   Toks[0].setAnnotationValue(static_cast<void*>(Info));
   PP.EnterTokenStream(Toks, 1, /*DisableMacroExpansion=*/true,
                       /*OwnsTokens=*/false);
@@ -1048,6 +1057,7 @@ void PragmaMSStructHandler::HandlePragma(Preprocessor &PP,
     PP.Diag(Tok.getLocation(), diag::warn_pragma_ms_struct);
     return;
   }
+  SourceLocation EndLoc = Tok.getLocation();
   const IdentifierInfo *II = Tok.getIdentifierInfo();
   if (II->isStr("on")) {
     Kind = Sema::PMSST_ON;
@@ -1073,6 +1083,7 @@ void PragmaMSStructHandler::HandlePragma(Preprocessor &PP,
   Toks[0].startToken();
   Toks[0].setKind(tok::annot_pragma_msstruct);
   Toks[0].setLocation(MSStructTok.getLocation());
+  Toks[0].setAnnotationEndLoc(EndLoc);
   Toks[0].setAnnotationValue(reinterpret_cast<void*>(
                              static_cast<uintptr_t>(Kind)));
   PP.EnterTokenStream(Toks, 1, /*DisableMacroExpansion=*/true,
@@ -1128,6 +1139,7 @@ static void ParseAlignPragma(Preprocessor &PP, Token &FirstTok,
     return;
   }
 
+  SourceLocation EndLoc = Tok.getLocation();
   PP.Lex(Tok);
   if (Tok.isNot(tok::eod)) {
     PP.Diag(Tok.getLocation(), diag::warn_pragma_extra_tokens_at_eol)
@@ -1142,6 +1154,7 @@ static void ParseAlignPragma(Preprocessor &PP, Token &FirstTok,
   Toks[0].startToken();
   Toks[0].setKind(tok::annot_pragma_align);
   Toks[0].setLocation(FirstTok.getLocation());
+  Toks[0].setAnnotationEndLoc(EndLoc);
   Toks[0].setAnnotationValue(reinterpret_cast<void*>(
                              static_cast<uintptr_t>(Kind)));
   PP.EnterTokenStream(Toks, 1, /*DisableMacroExpansion=*/true,
@@ -1285,6 +1298,7 @@ void PragmaWeakHandler::HandlePragma(Preprocessor &PP,
     pragmaUnusedTok.startToken();
     pragmaUnusedTok.setKind(tok::annot_pragma_weakalias);
     pragmaUnusedTok.setLocation(WeakLoc);
+    pragmaUnusedTok.setAnnotationEndLoc(AliasName.getLocation());
     Toks[1] = WeakName;
     Toks[2] = AliasName;
     PP.EnterTokenStream(Toks, 3,
@@ -1297,6 +1311,7 @@ void PragmaWeakHandler::HandlePragma(Preprocessor &PP,
     pragmaUnusedTok.startToken();
     pragmaUnusedTok.setKind(tok::annot_pragma_weak);
     pragmaUnusedTok.setLocation(WeakLoc);
+    pragmaUnusedTok.setAnnotationEndLoc(WeakLoc);
     Toks[1] = WeakName;
     PP.EnterTokenStream(Toks, 2,
                         /*DisableMacroExpansion=*/true, /*OwnsTokens=*/false);
@@ -1342,6 +1357,7 @@ void PragmaRedefineExtnameHandler::HandlePragma(Preprocessor &PP,
   pragmaRedefTok.startToken();
   pragmaRedefTok.setKind(tok::annot_pragma_redefine_extname);
   pragmaRedefTok.setLocation(RedefLoc);
+  pragmaRedefTok.setAnnotationEndLoc(AliasName.getLocation());
   Toks[1] = RedefName;
   Toks[2] = AliasName;
   PP.EnterTokenStream(Toks, 3,
@@ -1364,6 +1380,7 @@ PragmaFPContractHandler::HandlePragma(Preprocessor &PP,
   Toks[0].startToken();
   Toks[0].setKind(tok::annot_pragma_fp_contract);
   Toks[0].setLocation(Tok.getLocation());
+  Toks[0].setAnnotationEndLoc(Tok.getLocation());
   Toks[0].setAnnotationValue(reinterpret_cast<void*>(
                              static_cast<uintptr_t>(OOS)));
   PP.EnterTokenStream(Toks, 1, /*DisableMacroExpansion=*/true,
@@ -1423,6 +1440,7 @@ PragmaOpenCLExtensionHandler::HandlePragma(Preprocessor &PP,
   Toks[0].setKind(tok::annot_pragma_opencl_extension);
   Toks[0].setLocation(NameLoc);
   Toks[0].setAnnotationValue(data.getOpaqueValue());
+  Toks[0].setAnnotationEndLoc(StateLoc);
   PP.EnterTokenStream(Toks, 1, /*DisableMacroExpansion=*/true,
                       /*OwnsTokens=*/false);
 
@@ -1471,7 +1489,7 @@ PragmaOpenMPHandler::HandlePragma(Preprocessor &PP,
   Token *Toks = new Token[Pragma.size()];
   std::copy(Pragma.begin(), Pragma.end(), Toks);
   PP.EnterTokenStream(Toks, Pragma.size(),
-                      /*DisableMacroExpansion=*/true, /*OwnsTokens=*/true);
+                      /*DisableMacroExpansion=*/false, /*OwnsTokens=*/true);
 }
 
 /// \brief Handle '#pragma pointers_to_members'
@@ -1554,6 +1572,7 @@ void PragmaMSPointersToMembers::HandlePragma(Preprocessor &PP,
     return;
   }
 
+  SourceLocation EndLoc = Tok.getLocation();
   PP.Lex(Tok);
   if (Tok.isNot(tok::eod)) {
     PP.Diag(Tok.getLocation(), diag::warn_pragma_extra_tokens_at_eol)
@@ -1565,6 +1584,7 @@ void PragmaMSPointersToMembers::HandlePragma(Preprocessor &PP,
   AnnotTok.startToken();
   AnnotTok.setKind(tok::annot_pragma_ms_pointers_to_members);
   AnnotTok.setLocation(PointersToMembersLoc);
+  AnnotTok.setAnnotationEndLoc(EndLoc);
   AnnotTok.setAnnotationValue(
       reinterpret_cast<void *>(static_cast<uintptr_t>(RepresentationMethod)));
   PP.EnterToken(AnnotTok);
@@ -1644,6 +1664,7 @@ void PragmaMSVtorDisp::HandlePragma(Preprocessor &PP,
     PP.Diag(VtorDispLoc, diag::warn_pragma_expected_rparen) << "vtordisp";
     return;
   }
+  SourceLocation EndLoc = Tok.getLocation();
   PP.Lex(Tok);
   if (Tok.isNot(tok::eod)) {
     PP.Diag(Tok.getLocation(), diag::warn_pragma_extra_tokens_at_eol)
@@ -1656,6 +1677,7 @@ void PragmaMSVtorDisp::HandlePragma(Preprocessor &PP,
   AnnotTok.startToken();
   AnnotTok.setKind(tok::annot_pragma_ms_vtordisp);
   AnnotTok.setLocation(VtorDispLoc);
+  AnnotTok.setAnnotationEndLoc(EndLoc);
   AnnotTok.setAnnotationValue(reinterpret_cast<void *>(
       static_cast<uintptr_t>((Kind << 16) | (Value & 0xFFFF))));
   PP.EnterToken(AnnotTok);
@@ -1672,10 +1694,13 @@ void PragmaMSPragma::HandlePragma(Preprocessor &PP,
   AnnotTok.startToken();
   AnnotTok.setKind(tok::annot_pragma_ms_pragma);
   AnnotTok.setLocation(Tok.getLocation());
+  AnnotTok.setAnnotationEndLoc(Tok.getLocation());
   SmallVector<Token, 8> TokenVector;
   // Suck up all of the tokens before the eod.
-  for (; Tok.isNot(tok::eod); PP.Lex(Tok))
+  for (; Tok.isNot(tok::eod); PP.Lex(Tok)) {
     TokenVector.push_back(Tok);
+    AnnotTok.setAnnotationEndLoc(Tok.getLocation());
+  }
   // Add a sentinal EoF token to the end of the list.
   TokenVector.push_back(EoF);
   // We must allocate this array with new because EnterTokenStream is going to
@@ -1786,6 +1811,14 @@ void PragmaCommentHandler::HandlePragma(Preprocessor &PP,
     return;
   }
 
+  // On PS4, issue a warning about any pragma comments other than
+  // #pragma comment lib.
+  if (PP.getTargetInfo().getTriple().isPS4() && Kind != Sema::PCK_Lib) {
+    PP.Diag(Tok.getLocation(), diag::warn_pragma_comment_ignored)
+      << II->getName();
+    return;
+  }
+
   // Read the optional string if present.
   PP.Lex(Tok);
   std::string ArgumentString;
@@ -1994,6 +2027,7 @@ void PragmaLoopHintHandler::HandlePragma(Preprocessor &PP,
     LoopHintTok.startToken();
     LoopHintTok.setKind(tok::annot_pragma_loop_hint);
     LoopHintTok.setLocation(PragmaName.getLocation());
+    LoopHintTok.setAnnotationEndLoc(PragmaName.getLocation());
     LoopHintTok.setAnnotationValue(static_cast<void *>(Info));
     TokenList.push_back(LoopHintTok);
   }
@@ -2076,6 +2110,7 @@ void PragmaUnrollHintHandler::HandlePragma(Preprocessor &PP,
   TokenArray[0].startToken();
   TokenArray[0].setKind(tok::annot_pragma_loop_hint);
   TokenArray[0].setLocation(PragmaName.getLocation());
+  TokenArray[0].setAnnotationEndLoc(PragmaName.getLocation());
   TokenArray[0].setAnnotationValue(static_cast<void *>(Info));
   PP.EnterTokenStream(TokenArray, 1, /*DisableMacroExpansion=*/false,
                       /*OwnsTokens=*/true);
diff --git a/lib/Parse/ParseStmt.cpp b/lib/Parse/ParseStmt.cpp
index 2a5f8406a50c..055bdeabf623 100644
--- a/lib/Parse/ParseStmt.cpp
+++ b/lib/Parse/ParseStmt.cpp
@@ -408,12 +408,6 @@ StmtResult Parser::ParseExprStatement() {
   return Actions.ActOnExprStmt(Expr);
 }
 
-StmtResult Parser::ParseSEHTryBlock() {
-  assert(Tok.is(tok::kw___try) && "Expected '__try'");
-  SourceLocation Loc = ConsumeToken();
-  return ParseSEHTryBlockCommon(Loc);
-}
-
 /// ParseSEHTryBlockCommon
 ///
 /// seh-try-block:
@@ -423,8 +417,11 @@ StmtResult Parser::ParseSEHTryBlock() {
 ///   seh-except-block
 ///   seh-finally-block
 ///
-StmtResult Parser::ParseSEHTryBlockCommon(SourceLocation TryLoc) {
-  if(Tok.isNot(tok::l_brace))
+StmtResult Parser::ParseSEHTryBlock() {
+  assert(Tok.is(tok::kw___try) && "Expected '__try'");
+  SourceLocation TryLoc = ConsumeToken();
+
+  if (Tok.isNot(tok::l_brace))
     return StmtError(Diag(Tok, diag::err_expected) << tok::l_brace);
 
   StmtResult TryBlock(ParseCompoundStatement(/*isStmtExpr=*/false,
@@ -441,7 +438,7 @@ StmtResult Parser::ParseSEHTryBlockCommon(SourceLocation TryLoc) {
     SourceLocation Loc = ConsumeToken();
     Handler = ParseSEHFinallyBlock(Loc);
   } else {
-    return StmtError(Diag(Tok,diag::err_seh_expected_handler));
+    return StmtError(Diag(Tok, diag::err_seh_expected_handler));
   }
 
   if(Handler.isInvalid())
@@ -466,14 +463,21 @@ StmtResult Parser::ParseSEHExceptBlock(SourceLocation ExceptLoc) {
   if (ExpectAndConsume(tok::l_paren))
     return StmtError();
 
-  ParseScope ExpectScope(this, Scope::DeclScope | Scope::ControlScope);
+  ParseScope ExpectScope(this, Scope::DeclScope | Scope::ControlScope |
+                                   Scope::SEHExceptScope);
 
   if (getLangOpts().Borland) {
     Ident__exception_info->setIsPoisoned(false);
     Ident___exception_info->setIsPoisoned(false);
     Ident_GetExceptionInfo->setIsPoisoned(false);
   }
-  ExprResult FilterExpr(ParseExpression());
+
+  ExprResult FilterExpr;
+  {
+    ParseScopeFlags FilterScope(this, getCurScope()->getFlags() |
+                                          Scope::SEHFilterScope);
+    FilterExpr = Actions.CorrectDelayedTyposInExpr(ParseExpression());
+  }
 
   if (getLangOpts().Borland) {
     Ident__exception_info->setIsPoisoned(true);
@@ -487,6 +491,9 @@ StmtResult Parser::ParseSEHExceptBlock(SourceLocation ExceptLoc) {
   if (ExpectAndConsume(tok::r_paren))
     return StmtError();
 
+  if (Tok.isNot(tok::l_brace))
+    return StmtError(Diag(Tok, diag::err_expected) << tok::l_brace);
+
   StmtResult Block(ParseCompoundStatement());
 
   if(Block.isInvalid())
@@ -500,16 +507,24 @@ StmtResult Parser::ParseSEHExceptBlock(SourceLocation ExceptLoc) {
 /// seh-finally-block:
 ///   '__finally' compound-statement
 ///
-StmtResult Parser::ParseSEHFinallyBlock(SourceLocation FinallyBlock) {
+StmtResult Parser::ParseSEHFinallyBlock(SourceLocation FinallyLoc) {
   PoisonIdentifierRAIIObject raii(Ident__abnormal_termination, false),
     raii2(Ident___abnormal_termination, false),
     raii3(Ident_AbnormalTermination, false);
 
+  if (Tok.isNot(tok::l_brace))
+    return StmtError(Diag(Tok, diag::err_expected) << tok::l_brace);
+
+  ParseScope FinallyScope(this, 0);
+  Actions.ActOnStartSEHFinallyBlock();
+
   StmtResult Block(ParseCompoundStatement());
-  if(Block.isInvalid())
+  if(Block.isInvalid()) {
+    Actions.ActOnAbortSEHFinallyBlock();
     return Block;
+  }
 
-  return Actions.ActOnSEHFinallyBlock(FinallyBlock,Block.get());
+  return Actions.ActOnFinishSEHFinallyBlock(FinallyLoc, Block.get());
 }
 
 /// Handle __leave
@@ -1253,7 +1268,7 @@ StmtResult Parser::ParseSwitchStatement(SourceLocation *TrailingElseLoc) {
   // We have incremented the mangling number for the SwitchScope and the
   // InnerScope, which is one too many.
   if (C99orCXX)
-    getCurScope()->decrementMSLocalManglingNumber();
+    getCurScope()->decrementMSManglingNumber();
 
   // Read the body statement.
   StmtResult Body(ParseStatement(TrailingElseLoc));
@@ -1674,6 +1689,12 @@ StmtResult Parser::ParseForStatement(SourceLocation *TrailingElseLoc) {
                                                      FirstPart.get(),
                                                      Collection.get(),
                                                      T.getCloseLocation());
+  } else {
+    // In OpenMP loop region loop control variable must be captured and be
+    // private. Perform analysis of first part (if any).
+    if (getLangOpts().OpenMP && FirstPart.isUsable()) {
+      Actions.ActOnOpenMPLoopInitialization(ForLoc, FirstPart.get());
+    }
   }
 
   // C99 6.8.5p5 - In C99, the body of the for statement is a scope, even if
@@ -1695,7 +1716,7 @@ StmtResult Parser::ParseForStatement(SourceLocation *TrailingElseLoc) {
   // It will only be incremented if the body contains other things that would
   // normally increment the mangling number (like a compound statement).
   if (C99orCXXorObjC)
-    getCurScope()->decrementMSLocalManglingNumber();
+    getCurScope()->decrementMSManglingNumber();
 
   // Read the body statement.
   StmtResult Body(ParseStatement(TrailingElseLoc));
diff --git a/lib/Parse/ParseStmtAsm.cpp b/lib/Parse/ParseStmtAsm.cpp
index 7bf4da6d1bd4..8ba9f15f0a40 100644
--- a/lib/Parse/ParseStmtAsm.cpp
+++ b/lib/Parse/ParseStmtAsm.cpp
@@ -530,7 +530,7 @@ StmtResult Parser::ParseMicrosoftAsmStatement(SourceLocation AsmLoc) {
       TheTarget->createMCAsmParser(*STI, *Parser, *MII, MCOptions));
 
   std::unique_ptr<llvm::MCInstPrinter> IP(
-      TheTarget->createMCInstPrinter(1, *MAI, *MII, *MRI, *STI));
+      TheTarget->createMCInstPrinter(llvm::Triple(TT), 1, *MAI, *MII, *MRI));
 
   // Change to the Intel dialect.
   Parser->setAssemblerDialect(1);
@@ -615,6 +615,7 @@ StmtResult Parser::ParseAsmStatement(bool &msAsm) {
     msAsm = true;
     return ParseMicrosoftAsmStatement(AsmLoc);
   }
+
   DeclSpec DS(AttrFactory);
   SourceLocation Loc = Tok.getLocation();
   ParseTypeQualifierListOpt(DS, AR_VendorAttributesParsed);
@@ -639,6 +640,15 @@ StmtResult Parser::ParseAsmStatement(bool &msAsm) {
   T.consumeOpen();
 
   ExprResult AsmString(ParseAsmStringLiteral());
+
+  // Check if GNU-style InlineAsm is disabled.
+  // Error on anything other than empty string.
+  if (!(getLangOpts().GNUAsm || AsmString.isInvalid())) {
+    const auto *SL = cast<StringLiteral>(AsmString.get());
+    if (!SL->getString().trim().empty())
+      Diag(Loc, diag::err_gnu_inline_asm_disabled);
+  }
+
   if (AsmString.isInvalid()) {
     // Consume up to and including the closing paren.
     T.skipToEnd();
diff --git a/lib/Parse/ParseTemplate.cpp b/lib/Parse/ParseTemplate.cpp
index 53de72cd3cc5..f1467fe553a0 100644
--- a/lib/Parse/ParseTemplate.cpp
+++ b/lib/Parse/ParseTemplate.cpp
@@ -14,6 +14,7 @@
 #include "clang/Parse/Parser.h"
 #include "RAIIObjectsForParser.h"
 #include "clang/AST/ASTConsumer.h"
+#include "clang/AST/ASTContext.h"
 #include "clang/AST/DeclTemplate.h"
 #include "clang/Parse/ParseDiagnostic.h"
 #include "clang/Sema/DeclSpec.h"
@@ -1301,7 +1302,8 @@ void Parser::ParseLateTemplatedFuncDef(LateParsedTemplate &LPT) {
   TemplateParameterDepthRAII CurTemplateDepthTracker(TemplateParameterDepth);
 
   // To restore the context after late parsing.
-  Sema::ContextRAII GlobalSavedContext(Actions, Actions.CurContext);
+  Sema::ContextRAII GlobalSavedContext(
+      Actions, Actions.Context.getTranslationUnitDecl());
 
   SmallVector<ParseScope*, 4> TemplateParamScopeStack;
 
diff --git a/lib/Parse/Parser.cpp b/lib/Parse/Parser.cpp
index 7ccd2092a2d7..dea7a6998a2d 100644
--- a/lib/Parse/Parser.cpp
+++ b/lib/Parse/Parser.cpp
@@ -38,6 +38,26 @@ public:
     return false;
   }
 };
+
+/// \brief RAIIObject to destroy the contents of a SmallVector of
+/// TemplateIdAnnotation pointers and clear the vector.
+class DestroyTemplateIdAnnotationsRAIIObj {
+  SmallVectorImpl<TemplateIdAnnotation *> &Container;
+
+public:
+  DestroyTemplateIdAnnotationsRAIIObj(
+      SmallVectorImpl<TemplateIdAnnotation *> &Container)
+      : Container(Container) {}
+
+  ~DestroyTemplateIdAnnotationsRAIIObj() {
+    for (SmallVectorImpl<TemplateIdAnnotation *>::iterator I =
+             Container.begin(),
+                                                           E = Container.end();
+         I != E; ++I)
+      (*I)->Destroy();
+    Container.clear();
+  }
+};
 } // end anonymous namespace
 
 IdentifierInfo *Parser::getSEHExceptKeyword() {
@@ -414,6 +434,15 @@ Parser::~Parser() {
 
   PP.clearCodeCompletionHandler();
 
+  if (getLangOpts().DelayedTemplateParsing &&
+      !PP.isIncrementalProcessingEnabled() && !TemplateIds.empty()) {
+    // If an ASTConsumer parsed delay-parsed templates in their
+    // HandleTranslationUnit() method, TemplateIds created there were not
+    // guarded by a DestroyTemplateIdAnnotationsRAIIObj object in
+    // ParseTopLevelDecl(). Destroy them here.
+    DestroyTemplateIdAnnotationsRAIIObj CleanupRAII(TemplateIds);
+  }
+
   assert(TemplateIds.empty() && "Still alive TemplateIdAnnotations around?");
 }
 
@@ -490,26 +519,6 @@ void Parser::Initialize() {
   ConsumeToken();
 }
 
-namespace {
-  /// \brief RAIIObject to destroy the contents of a SmallVector of
-  /// TemplateIdAnnotation pointers and clear the vector.
-  class DestroyTemplateIdAnnotationsRAIIObj {
-    SmallVectorImpl<TemplateIdAnnotation *> &Container;
-  public:
-    DestroyTemplateIdAnnotationsRAIIObj(SmallVectorImpl<TemplateIdAnnotation *>
-                                       &Container)
-      : Container(Container) {}
-
-    ~DestroyTemplateIdAnnotationsRAIIObj() {
-      for (SmallVectorImpl<TemplateIdAnnotation *>::iterator I =
-           Container.begin(), E = Container.end();
-           I != E; ++I)
-        (*I)->Destroy();
-      Container.clear();
-    }
-  };
-}
-
 void Parser::LateTemplateParserCleanupCallback(void *P) {
   // While this RAII helper doesn't bracket any actual work, the destructor will
   // clean up annotations that were created during ActOnEndOfTranslationUnit
@@ -541,8 +550,14 @@ bool Parser::ParseTopLevelDecl(DeclGroupPtrTy &Result) {
     return false;
 
   case tok::annot_module_begin:
+    Actions.ActOnModuleBegin(Tok.getLocation(), reinterpret_cast<Module *>(
+                                                    Tok.getAnnotationValue()));
+    ConsumeToken();
+    return false;
+
   case tok::annot_module_end:
-    // FIXME: Update visibility based on the submodule we're in.
+    Actions.ActOnModuleEnd(Tok.getLocation(), reinterpret_cast<Module *>(
+                                                  Tok.getAnnotationValue()));
     ConsumeToken();
     return false;
 
@@ -669,8 +684,18 @@ Parser::ParseExternalDeclaration(ParsedAttributesWithRange &attrs,
 
     SourceLocation StartLoc = Tok.getLocation();
     SourceLocation EndLoc;
+
     ExprResult Result(ParseSimpleAsm(&EndLoc));
 
+    // Check if GNU-style InlineAsm is disabled.
+    // Empty asm string is allowed because it will not introduce
+    // any assembly code.
+    if (!(getLangOpts().GNUAsm || Result.isInvalid())) {
+      const auto *SL = cast<StringLiteral>(Result.get());
+      if (!SL->getString().trim().empty())
+        Diag(StartLoc, diag::err_gnu_inline_asm_disabled);
+    }
+
     ExpectAndConsume(tok::semi, diag::err_expected_after,
                      "top-level asm block");
 
@@ -1048,7 +1073,6 @@ Decl *Parser::ParseFunctionDefinition(ParsingDeclarator &D,
 
   if (TryConsumeToken(tok::equal)) {
     assert(getLangOpts().CPlusPlus && "Only C++ function definitions have '='");
-    Actions.ActOnFinishFunctionBody(Res, nullptr, false);
 
     bool Delete = false;
     SourceLocation KWLoc;
@@ -1076,6 +1100,8 @@ Decl *Parser::ParseFunctionDefinition(ParsingDeclarator &D,
       SkipUntil(tok::semi);
     }
 
+    Stmt *GeneratedBody = Res ? Res->getBody() : nullptr;
+    Actions.ActOnFinishFunctionBody(Res, GeneratedBody, false);
     return Res;
   }
 
diff --git a/lib/Parse/RAIIObjectsForParser.h b/lib/Parse/RAIIObjectsForParser.h
index a0c9c1fb6c6f..36d87ebd8aca 100644
--- a/lib/Parse/RAIIObjectsForParser.h
+++ b/lib/Parse/RAIIObjectsForParser.h
@@ -58,6 +58,12 @@ namespace clang {
         Active = false;
       }
     }
+    SuppressAccessChecks(SuppressAccessChecks &&Other)
+      : S(Other.S), DiagnosticPool(std::move(Other.DiagnosticPool)),
+        State(Other.State), Active(Other.Active) {
+      Other.Active = false;
+    }
+    void operator=(SuppressAccessChecks &&Other) = delete;
 
     void done() {
       assert(Active && "trying to end an inactive suppression");
@@ -87,8 +93,8 @@ namespace clang {
     Sema::ParsingDeclState State;
     bool Popped;
 
-    ParsingDeclRAIIObject(const ParsingDeclRAIIObject &) LLVM_DELETED_FUNCTION;
-    void operator=(const ParsingDeclRAIIObject &) LLVM_DELETED_FUNCTION;
+    ParsingDeclRAIIObject(const ParsingDeclRAIIObject &) = delete;
+    void operator=(const ParsingDeclRAIIObject &) = delete;
 
   public:
     enum NoParent_t { NoParent };
@@ -244,8 +250,8 @@ namespace clang {
   /// the way they used to be.  This is used to handle __extension__ in the
   /// parser.
   class ExtensionRAIIObject {
-    ExtensionRAIIObject(const ExtensionRAIIObject &) LLVM_DELETED_FUNCTION;
-    void operator=(const ExtensionRAIIObject &) LLVM_DELETED_FUNCTION;
+    ExtensionRAIIObject(const ExtensionRAIIObject &) = delete;
+    void operator=(const ExtensionRAIIObject &) = delete;
 
     DiagnosticsEngine &Diags;
   public:
@@ -423,7 +429,13 @@ namespace clang {
       if (P.Tok.is(Close)) {
         LClose = (P.*Consumer)();
         return false;
-      } 
+      } else if (P.Tok.is(tok::semi) && P.NextToken().is(Close)) {
+        SourceLocation SemiLoc = P.ConsumeToken();
+        P.Diag(SemiLoc, diag::err_unexpected_semi)
+            << Close << FixItHint::CreateRemoval(SourceRange(SemiLoc, SemiLoc));
+        LClose = (P.*Consumer)();
+        return false;
+      }
       
       return diagnoseMissingClose();
     }
diff --git a/lib/Rewrite/RewriteRope.cpp b/lib/Rewrite/RewriteRope.cpp
index 1c82ee4a67ec..451ad07e4e84 100644
--- a/lib/Rewrite/RewriteRope.cpp
+++ b/lib/Rewrite/RewriteRope.cpp
@@ -89,9 +89,9 @@ namespace {
     bool IsLeaf;
 
     RopePieceBTreeNode(bool isLeaf) : Size(0), IsLeaf(isLeaf) {}
-    ~RopePieceBTreeNode() {}
-  public:
+    ~RopePieceBTreeNode() = default;
 
+  public:
     bool isLeaf() const { return IsLeaf; }
     unsigned size() const { return Size; }
 
diff --git a/lib/Rewrite/Rewriter.cpp b/lib/Rewrite/Rewriter.cpp
index 60cdcf703172..be09a363a61f 100644
--- a/lib/Rewrite/Rewriter.cpp
+++ b/lib/Rewrite/Rewriter.cpp
@@ -54,7 +54,7 @@ void RewriteBuffer::RemoveText(unsigned OrigOffset, unsigned Size,
   if (Size == 0) return;
 
   unsigned RealOffset = getMappedOffset(OrigOffset, true);
-  assert(RealOffset+Size < Buffer.size() && "Invalid location");
+  assert(RealOffset+Size <= Buffer.size() && "Invalid location");
 
   // Remove the dead characters.
   Buffer.erase(RealOffset, Size);
diff --git a/lib/Sema/AnalysisBasedWarnings.cpp b/lib/Sema/AnalysisBasedWarnings.cpp
index f666a9b46384..d697ecb21454 100644
--- a/lib/Sema/AnalysisBasedWarnings.cpp
+++ b/lib/Sema/AnalysisBasedWarnings.cpp
@@ -117,6 +117,7 @@ static void CheckUnreachable(Sema &S, AnalysisDeclContext &AC) {
   reachable_code::FindUnreachableCode(AC, S.getPreprocessor(), UC);
 }
 
+namespace {
 /// \brief Warn on logical operator errors in CFGBuilder
 class LogicalErrorHandler : public CFGCallback {
   Sema &S;
@@ -138,7 +139,7 @@ public:
     return false;
   }
 
-  void compareAlwaysTrue(const BinaryOperator *B, bool isAlwaysTrue) {
+  void compareAlwaysTrue(const BinaryOperator *B, bool isAlwaysTrue) override {
     if (HasMacroID(B))
       return;
 
@@ -147,7 +148,8 @@ public:
         << DiagRange << isAlwaysTrue;
   }
 
-  void compareBitwiseEquality(const BinaryOperator *B, bool isAlwaysTrue) {
+  void compareBitwiseEquality(const BinaryOperator *B,
+                              bool isAlwaysTrue) override {
     if (HasMacroID(B))
       return;
 
@@ -156,7 +158,7 @@ public:
         << DiagRange << isAlwaysTrue;
   }
 };
-
+} // namespace
 
 //===----------------------------------------------------------------------===//
 // Check for infinite self-recursion in functions
@@ -1332,9 +1334,7 @@ class UninitValsDiagReporter : public UninitVariablesHandler {
   
 public:
   UninitValsDiagReporter(Sema &S) : S(S), uses(nullptr) {}
-  ~UninitValsDiagReporter() { 
-    flushDiagnostics();
-  }
+  ~UninitValsDiagReporter() override { flushDiagnostics(); }
 
   MappedType &getUses(const VarDecl *vd) {
     if (!uses)
@@ -1438,7 +1438,7 @@ struct SortDiagBySourceLocation {
 //===----------------------------------------------------------------------===//
 namespace clang {
 namespace threadSafety {
-
+namespace {
 class ThreadSafetyReporter : public clang::threadSafety::ThreadSafetyHandler {
   Sema &S;
   DiagList Warnings;
@@ -1662,9 +1662,8 @@ class ThreadSafetyReporter : public clang::threadSafety::ThreadSafetyHandler {
     }
   }
 
-
-  virtual void handleNegativeNotHeld(StringRef Kind, Name LockName, Name Neg,
-                                     SourceLocation Loc) override {
+  void handleNegativeNotHeld(StringRef Kind, Name LockName, Name Neg,
+                             SourceLocation Loc) override {
     PartialDiagnosticAt Warning(Loc,
         S.PDiag(diag::warn_acquire_requires_negative_cap)
         << Kind << LockName << Neg);
@@ -1679,6 +1678,19 @@ class ThreadSafetyReporter : public clang::threadSafety::ThreadSafetyHandler {
     Warnings.push_back(DelayedDiag(Warning, getNotes()));
   }
 
+  void handleLockAcquiredBefore(StringRef Kind, Name L1Name, Name L2Name,
+                                SourceLocation Loc) override {
+    PartialDiagnosticAt Warning(Loc,
+      S.PDiag(diag::warn_acquired_before) << Kind << L1Name << L2Name);
+    Warnings.push_back(DelayedDiag(Warning, getNotes()));
+  }
+
+  void handleBeforeAfterCycle(Name L1Name, SourceLocation Loc) override {
+    PartialDiagnosticAt Warning(Loc,
+      S.PDiag(diag::warn_acquired_before_after_cycle) << L1Name);
+    Warnings.push_back(DelayedDiag(Warning, getNotes()));
+  }
+
   void enterFunction(const FunctionDecl* FD) override {
     CurrentFunction = FD;
   }
@@ -1687,9 +1699,9 @@ class ThreadSafetyReporter : public clang::threadSafety::ThreadSafetyHandler {
     CurrentFunction = 0;
   }
 };
-
-}
-}
+} // namespace
+} // namespace threadSafety
+} // namespace clang
 
 //===----------------------------------------------------------------------===//
 // -Wconsumed
@@ -1704,7 +1716,7 @@ class ConsumedWarningsHandler : public ConsumedWarningsHandlerBase {
   DiagList Warnings;
   
 public:
-  
+
   ConsumedWarningsHandler(Sema &S) : S(S) {}
 
   void emitDiagnostics() override {
@@ -1981,7 +1993,8 @@ AnalysisBasedWarnings::IssueWarnings(sema::AnalysisBasedWarnings::Policy P,
     if (!Diags.isIgnored(diag::warn_thread_safety_verbose, D->getLocStart()))
       Reporter.setVerbose(true);
 
-    threadSafety::runThreadSafetyAnalysis(AC, Reporter);
+    threadSafety::runThreadSafetyAnalysis(AC, Reporter,
+                                          &S.ThreadSafetyDeclCache);
     Reporter.emitDiagnostics();
   }
 
diff --git a/lib/Sema/CodeCompleteConsumer.cpp b/lib/Sema/CodeCompleteConsumer.cpp
index b2dc2d79616b..69ae4f01dc7b 100644
--- a/lib/Sema/CodeCompleteConsumer.cpp
+++ b/lib/Sema/CodeCompleteConsumer.cpp
@@ -251,19 +251,16 @@ const char *CodeCompletionString::getTypedText() const {
   return nullptr;
 }
 
-const char *CodeCompletionAllocator::CopyString(StringRef String) {
-  char *Mem = (char *)Allocate(String.size() + 1, 1);
-  std::copy(String.begin(), String.end(), Mem);
-  Mem[String.size()] = 0;
-  return Mem;
-}
-
-const char *CodeCompletionAllocator::CopyString(Twine String) {
+const char *CodeCompletionAllocator::CopyString(const Twine &String) {
+  SmallString<128> Data;
+  StringRef Ref = String.toStringRef(Data);
   // FIXME: It would be more efficient to teach Twine to tell us its size and
   // then add a routine there to fill in an allocated char* with the contents
   // of the string.
-  SmallString<128> Data;
-  return CopyString(String.toStringRef(Data));
+  char *Mem = (char *)Allocate(Ref.size() + 1, 1);
+  std::copy(Ref.begin(), Ref.end(), Mem);
+  Mem[Ref.size()] = 0;
+  return Mem;
 }
 
 StringRef CodeCompletionTUInfo::getParentName(const DeclContext *DC) {
@@ -483,6 +480,31 @@ PrintingCodeCompleteConsumer::ProcessCodeCompleteResults(Sema &SemaRef,
   }
 }
 
+// This function is used solely to preserve the former presentation of overloads
+// by "clang -cc1 -code-completion-at", since CodeCompletionString::getAsString
+// needs to be improved for printing the newer and more detailed overload
+// chunks.
+static std::string getOverloadAsString(const CodeCompletionString &CCS) {
+  std::string Result;
+  llvm::raw_string_ostream OS(Result);
+
+  for (auto &C : CCS) {
+    switch (C.Kind) {
+    case CodeCompletionString::CK_Informative:
+    case CodeCompletionString::CK_ResultType:
+      OS << "[#" << C.Text << "#]";
+      break;
+
+    case CodeCompletionString::CK_CurrentParameter:
+      OS << "<#" << C.Text << "#>";
+      break;
+
+    default: OS << C.Text; break;
+    }
+  }
+  return OS.str();
+}
+
 void 
 PrintingCodeCompleteConsumer::ProcessOverloadCandidates(Sema &SemaRef,
                                                         unsigned CurrentArg,
@@ -491,8 +513,9 @@ PrintingCodeCompleteConsumer::ProcessOverloadCandidates(Sema &SemaRef,
   for (unsigned I = 0; I != NumCandidates; ++I) {
     if (CodeCompletionString *CCS
           = Candidates[I].CreateSignatureString(CurrentArg, SemaRef,
-                                                getAllocator(), CCTUInfo)) {
-      OS << "OVERLOAD: " << CCS->getAsString() << "\n";
+                                                getAllocator(), CCTUInfo,
+                                                includeBriefComments())) {
+      OS << "OVERLOAD: " << getOverloadAsString(*CCS) << "\n";
     }
   }
 }
diff --git a/lib/Sema/DeclSpec.cpp b/lib/Sema/DeclSpec.cpp
index 349bb3258025..1e7fc750a9cf 100644
--- a/lib/Sema/DeclSpec.cpp
+++ b/lib/Sema/DeclSpec.cpp
@@ -983,11 +983,18 @@ void DeclSpec::Finish(DiagnosticsEngine &D, Preprocessor &PP, const PrintingPoli
                                  getSpecifierName((TST)TypeSpecType, Policy));
       }
 
-      // Only 'short' is valid with vector bool. (PIM 2.1)
-      if ((TypeSpecWidth != TSW_unspecified) && (TypeSpecWidth != TSW_short))
+      // Only 'short' and 'long long' are valid with vector bool. (PIM 2.1)
+      if ((TypeSpecWidth != TSW_unspecified) && (TypeSpecWidth != TSW_short) &&
+          (TypeSpecWidth != TSW_longlong))
         Diag(D, TSWLoc, diag::err_invalid_vector_bool_decl_spec)
           << getSpecifierName((TSW)TypeSpecWidth);
 
+      // vector bool long long requires VSX support.
+      if ((TypeSpecWidth == TSW_longlong) &&
+          (!PP.getTargetInfo().hasFeature("vsx")) &&
+          (!PP.getTargetInfo().hasFeature("power8-vector")))
+        Diag(D, TSTLoc, diag::err_invalid_vector_long_long_decl_spec);
+
       // Elements of vector bool are interpreted as unsigned. (PIM 2.1)
       if ((TypeSpecType == TST_char) || (TypeSpecType == TST_int) ||
           (TypeSpecWidth != TSW_unspecified))
@@ -1214,7 +1221,10 @@ void UnqualifiedId::setOperatorFunctionId(SourceLocation OperatorLoc,
 
 bool VirtSpecifiers::SetSpecifier(Specifier VS, SourceLocation Loc,
                                   const char *&PrevSpec) {
+  if (!FirstLocation.isValid())
+    FirstLocation = Loc;
   LastLocation = Loc;
+  LastSpecifier = VS;
   
   if (Specifiers & VS) {
     PrevSpec = getSpecifierName(VS);
diff --git a/lib/Sema/DelayedDiagnostic.cpp b/lib/Sema/DelayedDiagnostic.cpp
index 664a6b1a8998..ceea04f276c9 100644
--- a/lib/Sema/DelayedDiagnostic.cpp
+++ b/lib/Sema/DelayedDiagnostic.cpp
@@ -35,6 +35,8 @@ DelayedDiagnostic::makeAvailability(Sema::AvailabilityDiagnostic AD,
     case Sema::AD_Unavailable:
       DD.Kind = Unavailable;
       break;
+    case Sema::AD_Partial:
+      llvm_unreachable("AD_Partial diags should not be delayed");
   }
   DD.Triggered = false;
   DD.Loc = Loc;
diff --git a/lib/Sema/IdentifierResolver.cpp b/lib/Sema/IdentifierResolver.cpp
index 6586fb32787c..53263bac546f 100644
--- a/lib/Sema/IdentifierResolver.cpp
+++ b/lib/Sema/IdentifierResolver.cpp
@@ -98,7 +98,7 @@ bool IdentifierResolver::isDeclInScope(Decl *D, DeclContext *Ctx, Scope *S,
                                        bool AllowInlineNamespace) const {
   Ctx = Ctx->getRedeclContext();
 
-  if (Ctx->isFunctionOrMethod() || S->isFunctionPrototypeScope()) {
+  if (Ctx->isFunctionOrMethod() || (S && S->isFunctionPrototypeScope())) {
     // Ignore the scopes associated within transparent declaration contexts.
     while (S->getEntity() && S->getEntity()->isTransparentContext())
       S = S->getParent();
@@ -266,6 +266,11 @@ static DeclMatchKind compareDeclarations(NamedDecl *Existing, NamedDecl *New) {
 
   // If the declarations are redeclarations of each other, keep the newest one.
   if (Existing->getCanonicalDecl() == New->getCanonicalDecl()) {
+    // If we're adding an imported declaration, don't replace another imported
+    // declaration.
+    if (Existing->isFromASTFile() && New->isFromASTFile())
+      return DMK_Different;
+
     // If either of these is the most recent declaration, use it.
     Decl *MostRecent = Existing->getMostRecentDecl();
     if (Existing == MostRecent)
diff --git a/lib/Sema/JumpDiagnostics.cpp b/lib/Sema/JumpDiagnostics.cpp
index fd75c02bb1ed..aac28bedfa5d 100644
--- a/lib/Sema/JumpDiagnostics.cpp
+++ b/lib/Sema/JumpDiagnostics.cpp
@@ -338,6 +338,36 @@ void JumpScopeChecker::BuildScopeInformation(Stmt *S, unsigned &origParentScope)
     return;
   }
 
+  case Stmt::SEHTryStmtClass: {
+    SEHTryStmt *TS = cast<SEHTryStmt>(S);
+    unsigned newParentScope;
+    Scopes.push_back(GotoScope(ParentScope,
+                               diag::note_protected_by_seh_try,
+                               diag::note_exits_seh_try,
+                               TS->getSourceRange().getBegin()));
+    if (Stmt *TryBlock = TS->getTryBlock())
+      BuildScopeInformation(TryBlock, (newParentScope = Scopes.size()-1));
+
+    // Jump from __except or __finally into the __try are not allowed either.
+    if (SEHExceptStmt *Except = TS->getExceptHandler()) {
+      Scopes.push_back(GotoScope(ParentScope,
+                                 diag::note_protected_by_seh_except,
+                                 diag::note_exits_seh_except,
+                                 Except->getSourceRange().getBegin()));
+      BuildScopeInformation(Except->getBlock(), 
+                            (newParentScope = Scopes.size()-1));
+    } else if (SEHFinallyStmt *Finally = TS->getFinallyHandler()) {
+      Scopes.push_back(GotoScope(ParentScope,
+                                 diag::note_protected_by_seh_finally,
+                                 diag::note_exits_seh_finally,
+                                 Finally->getSourceRange().getBegin()));
+      BuildScopeInformation(Finally->getBlock(), 
+                            (newParentScope = Scopes.size()-1));
+    }
+
+    return;
+  }
+
   default:
     break;
   }
@@ -417,7 +447,8 @@ void JumpScopeChecker::BuildScopeInformation(Stmt *S, unsigned &origParentScope)
     unsigned newParentScope;
     // Disallow jumps into the protected statement of an @synchronized, but
     // allow jumps into the object expression it protects.
-    if (ObjCAtSynchronizedStmt *AS = dyn_cast<ObjCAtSynchronizedStmt>(SubStmt)){
+    if (ObjCAtSynchronizedStmt *AS =
+            dyn_cast<ObjCAtSynchronizedStmt>(SubStmt)) {
       // Recursively walk the AST for the @synchronized object expr, it is
       // evaluated in the normal scope.
       BuildScopeInformation(AS->getSynchExpr(), ParentScope);
@@ -434,14 +465,16 @@ void JumpScopeChecker::BuildScopeInformation(Stmt *S, unsigned &origParentScope)
     }
 
     // Disallow jumps into the protected statement of an @autoreleasepool.
-    if (ObjCAutoreleasePoolStmt *AS = dyn_cast<ObjCAutoreleasePoolStmt>(SubStmt)){
-      // Recursively walk the AST for the @autoreleasepool part, protected by a new
-      // scope.
+    if (ObjCAutoreleasePoolStmt *AS =
+            dyn_cast<ObjCAutoreleasePoolStmt>(SubStmt)) {
+      // Recursively walk the AST for the @autoreleasepool part, protected by a
+      // new scope.
       Scopes.push_back(GotoScope(ParentScope,
                                  diag::note_protected_by_objc_autoreleasepool,
                                  diag::note_exits_objc_autoreleasepool,
                                  AS->getAtLoc()));
-      BuildScopeInformation(AS->getSubStmt(), (newParentScope = Scopes.size()-1));
+      BuildScopeInformation(AS->getSubStmt(),
+                            (newParentScope = Scopes.size() - 1));
       continue;
     }
 
@@ -756,6 +789,18 @@ void JumpScopeChecker::CheckJump(Stmt *From, Stmt *To, SourceLocation DiagLoc,
   // Common case: exactly the same scope, which is fine.
   if (FromScope == ToScope) return;
 
+  // Warn on gotos out of __finally blocks.
+  if (isa<GotoStmt>(From) || isa<IndirectGotoStmt>(From)) {
+    // If FromScope > ToScope, FromScope is more nested and the jump goes to a
+    // less nested scope.  Check if it crosses a __finally along the way.
+    for (unsigned I = FromScope; I > ToScope; I = Scopes[I].ParentScope) {
+      if (Scopes[I].InDiag == diag::note_protected_by_seh_finally) {
+        S.Diag(From->getLocStart(), diag::warn_jump_out_of_seh_finally);
+        break;
+      }
+    }
+  }
+
   unsigned CommonScope = GetDeepestCommonScope(FromScope, ToScope);
 
   // It's okay to jump out from a nested scope.
diff --git a/lib/Sema/MultiplexExternalSemaSource.cpp b/lib/Sema/MultiplexExternalSemaSource.cpp
index 449ddf43114d..9ecb5a7fefbc 100644
--- a/lib/Sema/MultiplexExternalSemaSource.cpp
+++ b/lib/Sema/MultiplexExternalSemaSource.cpp
@@ -86,6 +86,14 @@ CXXBaseSpecifier *MultiplexExternalSemaSource::GetExternalCXXBaseSpecifiers(
   return nullptr;
 }
 
+CXXCtorInitializer **
+MultiplexExternalSemaSource::GetExternalCXXCtorInitializers(uint64_t Offset) {
+  for (auto *S : Sources)
+    if (auto *R = S->GetExternalCXXCtorInitializers(Offset))
+      return R;
+  return nullptr;
+}
+
 bool MultiplexExternalSemaSource::
 FindExternalVisibleDeclsByName(const DeclContext *DC, DeclarationName Name) {
   bool AnyDeclsFound = false;
@@ -204,7 +212,15 @@ void MultiplexExternalSemaSource::ReadUndefinedButUsed(
   for(size_t i = 0; i < Sources.size(); ++i)
     Sources[i]->ReadUndefinedButUsed(Undefined);
 }
-  
+
+void MultiplexExternalSemaSource::ReadMismatchingDeleteExpressions(
+    llvm::MapVector<FieldDecl *,
+                    llvm::SmallVector<std::pair<SourceLocation, bool>, 4>> &
+        Exprs) {
+  for (auto &Source : Sources)
+    Source->ReadMismatchingDeleteExpressions(Exprs);
+}
+
 bool MultiplexExternalSemaSource::LookupUnqualified(LookupResult &R, Scope *S){ 
   for(size_t i = 0; i < Sources.size(); ++i)
     Sources[i]->LookupUnqualified(R, S);
@@ -236,24 +252,12 @@ void MultiplexExternalSemaSource::ReadExtVectorDecls(
     Sources[i]->ReadExtVectorDecls(Decls);
 }
 
-void MultiplexExternalSemaSource::ReadDynamicClasses(
-                                       SmallVectorImpl<CXXRecordDecl*> &Decls) {
-  for(size_t i = 0; i < Sources.size(); ++i)
-    Sources[i]->ReadDynamicClasses(Decls);
-}
-
 void MultiplexExternalSemaSource::ReadUnusedLocalTypedefNameCandidates(
     llvm::SmallSetVector<const TypedefNameDecl *, 4> &Decls) {
   for(size_t i = 0; i < Sources.size(); ++i)
     Sources[i]->ReadUnusedLocalTypedefNameCandidates(Decls);
 }
 
-void MultiplexExternalSemaSource::ReadLocallyScopedExternCDecls(
-                                           SmallVectorImpl<NamedDecl*> &Decls) {
-  for(size_t i = 0; i < Sources.size(); ++i)
-    Sources[i]->ReadLocallyScopedExternCDecls(Decls);
-}
-
 void MultiplexExternalSemaSource::ReadReferencedSelectors(
                   SmallVectorImpl<std::pair<Selector, SourceLocation> > &Sels) {
   for(size_t i = 0; i < Sources.size(); ++i)
@@ -280,7 +284,7 @@ void MultiplexExternalSemaSource::ReadPendingInstantiations(
 }
 
 void MultiplexExternalSemaSource::ReadLateParsedTemplates(
-    llvm::DenseMap<const FunctionDecl *, LateParsedTemplate *> &LPTMap) {
+    llvm::MapVector<const FunctionDecl *, LateParsedTemplate *> &LPTMap) {
   for (size_t i = 0; i < Sources.size(); ++i)
     Sources[i]->ReadLateParsedTemplates(LPTMap);
 }
diff --git a/lib/Sema/Scope.cpp b/lib/Sema/Scope.cpp
index 6c7977882386..7c70048acfbf 100644
--- a/lib/Sema/Scope.cpp
+++ b/lib/Sema/Scope.cpp
@@ -38,7 +38,8 @@ void Scope::Init(Scope *parent, unsigned flags) {
     FnParent       = parent->FnParent;
     BlockParent    = parent->BlockParent;
     TemplateParamParent = parent->TemplateParamParent;
-    MSLocalManglingParent = parent->MSLocalManglingParent;
+    MSLastManglingParent = parent->MSLastManglingParent;
+    MSCurManglingNumber = getMSLastManglingNumber();
     if ((Flags & (FnScope | ClassScope | BlockScope | TemplateParamScope |
                   FunctionPrototypeScope | AtCatchScope | ObjCMethodScope)) ==
         0)
@@ -47,9 +48,10 @@ void Scope::Init(Scope *parent, unsigned flags) {
     Depth = 0;
     PrototypeDepth = 0;
     PrototypeIndex = 0;
-    MSLocalManglingParent = FnParent = BlockParent = nullptr;
+    MSLastManglingParent = FnParent = BlockParent = nullptr;
     TemplateParamParent = nullptr;
-    MSLocalManglingNumber = 1;
+    MSLastManglingNumber = 1;
+    MSCurManglingNumber = 1;
   }
 
   // If this scope is a function or contains breaks/continues, remember it.
@@ -57,8 +59,9 @@ void Scope::Init(Scope *parent, unsigned flags) {
   // The MS mangler uses the number of scopes that can hold declarations as
   // part of an external name.
   if (Flags & (ClassScope | FnScope)) {
-    MSLocalManglingNumber = getMSLocalManglingNumber();
-    MSLocalManglingParent = this;
+    MSLastManglingNumber = getMSLastManglingNumber();
+    MSLastManglingParent = this;
+    MSCurManglingNumber = 1;
   }
   if (flags & BreakScope)         BreakParent = this;
   if (flags & ContinueScope)      ContinueParent = this;
@@ -78,7 +81,7 @@ void Scope::Init(Scope *parent, unsigned flags) {
     else if ((flags & EnumScope))
       ; // Don't increment for enum scopes.
     else
-      incrementMSLocalManglingNumber();
+      incrementMSManglingNumber();
   }
 
   DeclsInScope.clear();
@@ -185,6 +188,9 @@ void Scope::dumpImpl(raw_ostream &OS) const {
     } else if (Flags & SEHTryScope) {
       OS << "SEHTryScope";
       Flags &= ~SEHTryScope;
+    } else if (Flags & SEHExceptScope) {
+      OS << "SEHExceptScope";
+      Flags &= ~SEHExceptScope;
     } else if (Flags & OpenMPDirectiveScope) {
       OS << "OpenMPDirectiveScope";
       Flags &= ~OpenMPDirectiveScope;
@@ -206,12 +212,13 @@ void Scope::dumpImpl(raw_ostream &OS) const {
     OS << "Parent: (clang::Scope*)" << Parent << '\n';
 
   OS << "Depth: " << Depth << '\n';
-  OS << "MSLocalManglingNumber: " << getMSLocalManglingNumber() << '\n';
+  OS << "MSLastManglingNumber: " << getMSLastManglingNumber() << '\n';
+  OS << "MSCurManglingNumber: " << getMSCurManglingNumber() << '\n';
   if (const DeclContext *DC = getEntity())
     OS << "Entity : (clang::DeclContext*)" << DC << '\n';
 
   if (NRVO.getInt())
-    OS << "NRVO not allowed";
+    OS << "NRVO not allowed\n";
   else if (NRVO.getPointer())
     OS << "NRVO candidate : (clang::VarDecl*)" << NRVO.getPointer() << '\n';
 }
diff --git a/lib/Sema/ScopeInfo.cpp b/lib/Sema/ScopeInfo.cpp
index 63ef3b2355fb..f80eadf18d56 100644
--- a/lib/Sema/ScopeInfo.cpp
+++ b/lib/Sema/ScopeInfo.cpp
@@ -33,6 +33,8 @@ void FunctionScopeInfo::Clear() {
   ObjCWarnForNoDesignatedInitChain = false;
   ObjCIsSecondaryInit = false;
   ObjCWarnForNoInitDelegation = false;
+  FirstCXXTryLoc = SourceLocation();
+  FirstSEHTryLoc = SourceLocation();
 
   SwitchStack.clear();
   Returns.clear();
diff --git a/lib/Sema/Sema.cpp b/lib/Sema/Sema.cpp
index b9aaf1616d68..50edc4261353 100644
--- a/lib/Sema/Sema.cpp
+++ b/lib/Sema/Sema.cpp
@@ -99,10 +99,11 @@ Sema::Sema(Preprocessor &pp, ASTContext &ctxt, ASTConsumer &consumer,
     GlobalNewDeleteDeclared(false),
     TUKind(TUKind),
     NumSFINAEErrors(0),
+    CachedFakeTopLevelModule(nullptr),
     AccessCheckingSFINAE(false), InNonInstantiationSFINAEContext(false),
     NonInstantiationEntries(0), ArgumentPackSubstitutionIndex(-1),
     CurrentInstantiationScope(nullptr), DisableTypoCorrection(false),
-    TyposCorrected(0), AnalysisWarnings(*this),
+    TyposCorrected(0), AnalysisWarnings(*this), ThreadSafetyDeclCache(nullptr),
     VarDataSharingAttributesStack(nullptr), CurScope(nullptr),
     Ident_super(nullptr), Ident___float128(nullptr)
 {
@@ -122,9 +123,7 @@ Sema::Sema(Preprocessor &pp, ASTContext &ctxt, ASTConsumer &consumer,
   PP.getDiagnostics().SetArgToStringFn(&FormatASTNodeDiagnosticArgument,
                                        &Context);
 
-  ExprEvalContexts.push_back(
-        ExpressionEvaluationContextRecord(PotentiallyEvaluated, 0,
-                                          false, nullptr, false));
+  ExprEvalContexts.emplace_back(PotentiallyEvaluated, 0, false, nullptr, false);
 
   FunctionScopes.push_back(new FunctionScopeInfo(Diags));
 
@@ -195,8 +194,9 @@ void Sema::Initialize() {
   }
 
   // Initialize Microsoft "predefined C++ types".
-  if (PP.getLangOpts().MSVCCompat && PP.getLangOpts().CPlusPlus) {
-    if (IdResolver.begin(&Context.Idents.get("type_info")) == IdResolver.end())
+  if (PP.getLangOpts().MSVCCompat) {
+    if (PP.getLangOpts().CPlusPlus &&
+        IdResolver.begin(&Context.Idents.get("type_info")) == IdResolver.end())
       PushOnScopeChains(Context.buildImplicitRecord("type_info", TTK_Class),
                         TUScope);
 
@@ -213,6 +213,29 @@ void Sema::Initialize() {
     addImplicitTypedef("image3d_t", Context.OCLImage3dTy);
     addImplicitTypedef("sampler_t", Context.OCLSamplerTy);
     addImplicitTypedef("event_t", Context.OCLEventTy);
+    if (getLangOpts().OpenCLVersion >= 200) {
+      addImplicitTypedef("atomic_int", Context.getAtomicType(Context.IntTy));
+      addImplicitTypedef("atomic_uint",
+                         Context.getAtomicType(Context.UnsignedIntTy));
+      addImplicitTypedef("atomic_long", Context.getAtomicType(Context.LongTy));
+      addImplicitTypedef("atomic_ulong",
+                         Context.getAtomicType(Context.UnsignedLongTy));
+      addImplicitTypedef("atomic_float",
+                         Context.getAtomicType(Context.FloatTy));
+      addImplicitTypedef("atomic_double",
+                         Context.getAtomicType(Context.DoubleTy));
+      // OpenCLC v2.0, s6.13.11.6 requires that atomic_flag is implemented as
+      // 32-bit integer and OpenCLC v2.0, s6.1.1 int is always 32-bit wide.
+      addImplicitTypedef("atomic_flag", Context.getAtomicType(Context.IntTy));
+      addImplicitTypedef("atomic_intptr_t",
+                         Context.getAtomicType(Context.getIntPtrType()));
+      addImplicitTypedef("atomic_uintptr_t",
+                         Context.getAtomicType(Context.getUIntPtrType()));
+      addImplicitTypedef("atomic_size_t",
+                         Context.getAtomicType(Context.getSizeType()));
+      addImplicitTypedef("atomic_ptrdiff_t",
+                         Context.getAtomicType(Context.getPointerDiffType()));
+    }
   }
 
   DeclarationName BuiltinVaList = &Context.Idents.get("__builtin_va_list");
@@ -243,6 +266,8 @@ Sema::~Sema() {
   if (isMultiplexExternalSource)
     delete ExternalSource;
 
+  threadSafety::threadSafetyCleanup(ThreadSafetyDeclCache);
+
   // Destroys data sharing attributes stack for OpenMP
   DestroyDataSharingAttributesStack();
 
@@ -337,18 +362,6 @@ ExprResult Sema::ImpCastExprToType(Expr *E, QualType Ty,
   if (ExprTy == TypeTy)
     return E;
 
-  // If this is a derived-to-base cast to a through a virtual base, we
-  // need a vtable.
-  if (Kind == CK_DerivedToBase &&
-      BasePathInvolvesVirtualBase(*BasePath)) {
-    QualType T = E->getType();
-    if (const PointerType *Pointer = T->getAs<PointerType>())
-      T = Pointer->getPointeeType();
-    if (const RecordType *RecordTy = T->getAs<RecordType>())
-      MarkVTableUsed(E->getLocStart(),
-                     cast<CXXRecordDecl>(RecordTy->getDecl()));
-  }
-
   if (ImplicitCastExpr *ImpCast = dyn_cast<ImplicitCastExpr>(E)) {
     if (ImpCast->getCastKind() == Kind && (!BasePath || BasePath->empty())) {
       ImpCast->setType(Ty);
@@ -512,14 +525,8 @@ void Sema::LoadExternalWeakUndeclaredIdentifiers() {
 
   SmallVector<std::pair<IdentifierInfo *, WeakInfo>, 4> WeakIDs;
   ExternalSource->ReadWeakUndeclaredIdentifiers(WeakIDs);
-  for (unsigned I = 0, N = WeakIDs.size(); I != N; ++I) {
-    llvm::DenseMap<IdentifierInfo*,WeakInfo>::iterator Pos
-      = WeakUndeclaredIdentifiers.find(WeakIDs[I].first);
-    if (Pos != WeakUndeclaredIdentifiers.end())
-      continue;
-
-    WeakUndeclaredIdentifiers.insert(WeakIDs[I]);
-  }
+  for (auto &WeakID : WeakIDs)
+    WeakUndeclaredIdentifiers.insert(WeakID);
 }
 
 
@@ -632,22 +639,6 @@ void Sema::ActOnEndOfTranslationUnit() {
   if (TUKind != TU_Prefix) {
     DiagnoseUseOfUnimplementedSelectors();
 
-    // If any dynamic classes have their key function defined within
-    // this translation unit, then those vtables are considered "used" and must
-    // be emitted.
-    for (DynamicClassesType::iterator I = DynamicClasses.begin(ExternalSource),
-                                      E = DynamicClasses.end();
-         I != E; ++I) {
-      assert(!(*I)->isDependentType() &&
-             "Should not see dependent types here!");
-      if (const CXXMethodDecl *KeyFunction =
-              Context.getCurrentKeyFunction(*I)) {
-        const FunctionDecl *Definition = nullptr;
-        if (KeyFunction->hasBody(Definition))
-          MarkVTableUsed(Definition->getLocation(), *I, true);
-      }
-    }
-
     // If DefinedUsedVTables ends up marking any virtual member functions it
     // might lead to more pending template instantiations, which we then need
     // to instantiate.
@@ -679,6 +670,8 @@ void Sema::ActOnEndOfTranslationUnit() {
 
   // All delayed member exception specs should be checked or we end up accepting
   // incompatible declarations.
+  // FIXME: This is wrong for TUKind == TU_Prefix. In that case, we need to
+  // write out the lists to the AST file (if any).
   assert(DelayedDefaultedMemberExceptionSpecs.empty());
   assert(DelayedExceptionSpecChecks.empty());
 
@@ -696,16 +689,13 @@ void Sema::ActOnEndOfTranslationUnit() {
   }
 
   // Check for #pragma weak identifiers that were never declared
-  // FIXME: This will cause diagnostics to be emitted in a non-determinstic
-  // order!  Iterating over a densemap like this is bad.
   LoadExternalWeakUndeclaredIdentifiers();
-  for (llvm::DenseMap<IdentifierInfo*,WeakInfo>::iterator
-       I = WeakUndeclaredIdentifiers.begin(),
-       E = WeakUndeclaredIdentifiers.end(); I != E; ++I) {
-    if (I->second.getUsed()) continue;
+  for (auto WeakID : WeakUndeclaredIdentifiers) {
+    if (WeakID.second.getUsed())
+      continue;
 
-    Diag(I->second.getLocation(), diag::warn_weak_identifier_undeclared)
-      << I->first;
+    Diag(WeakID.second.getLocation(), diag::warn_weak_identifier_undeclared)
+        << WeakID.first;
   }
 
   if (LangOpts.CPlusPlus11 &&
@@ -871,6 +861,17 @@ void Sema::ActOnEndOfTranslationUnit() {
     }
   }
 
+  if (!Diags.isIgnored(diag::warn_mismatched_delete_new, SourceLocation())) {
+    if (ExternalSource)
+      ExternalSource->ReadMismatchingDeleteExpressions(DeleteExprs);
+    for (const auto &DeletedFieldInfo : DeleteExprs) {
+      for (const auto &DeleteExprLoc : DeletedFieldInfo.second) {
+        AnalyzeDeleteExprMismatch(DeletedFieldInfo.first, DeleteExprLoc.first,
+                                  DeleteExprLoc.second);
+      }
+    }
+  }
+
   // Check we've noticed that we're no longer parsing the initializer for every
   // variable. If we miss cases, then at best we have a performance issue and
   // at worst a rejects-valid bug.
@@ -1230,6 +1231,9 @@ void ExternalSemaSource::ReadUndefinedButUsed(
                        llvm::DenseMap<NamedDecl *, SourceLocation> &Undefined) {
 }
 
+void ExternalSemaSource::ReadMismatchingDeleteExpressions(llvm::MapVector<
+    FieldDecl *, llvm::SmallVector<std::pair<SourceLocation, bool>, 4>> &) {}
+
 void PrettyDeclStackTraceEntry::print(raw_ostream &OS) const {
   SourceLocation Loc = this->Loc;
   if (!Loc.isValid() && TheDecl) Loc = TheDecl->getLocation();
@@ -1478,3 +1482,8 @@ CapturedRegionScopeInfo *Sema::getCurCapturedRegion() {
 
   return dyn_cast<CapturedRegionScopeInfo>(FunctionScopes.back());
 }
+
+const llvm::MapVector<FieldDecl *, Sema::DeleteLocs> &
+Sema::getMismatchingDeleteExpressions() const {
+  return DeleteExprs;
+}
diff --git a/lib/Sema/SemaAccess.cpp b/lib/Sema/SemaAccess.cpp
index 37240c23b691..0e973cc5ebaa 100644
--- a/lib/Sema/SemaAccess.cpp
+++ b/lib/Sema/SemaAccess.cpp
@@ -1462,7 +1462,7 @@ static Sema::AccessResult CheckAccess(Sema &S, SourceLocation Loc,
   case AR_inaccessible: return Sema::AR_inaccessible;
   case AR_dependent: return Sema::AR_dependent;
   }
-  llvm_unreachable("falling off end");
+  llvm_unreachable("invalid access result");
 }
 
 void Sema::HandleDelayedAccessCheck(DelayedDiagnostic &DD, Decl *D) {
diff --git a/lib/Sema/SemaAttr.cpp b/lib/Sema/SemaAttr.cpp
index 76297977ea0e..5a29bad29f4d 100644
--- a/lib/Sema/SemaAttr.cpp
+++ b/lib/Sema/SemaAttr.cpp
@@ -131,7 +131,7 @@ void Sema::AddAlignmentAttributesForRecord(RecordDecl *RD) {
 
 void Sema::AddMsStructLayoutForRecord(RecordDecl *RD) {
   if (MSStructPragmaOn)
-    RD->addAttr(MsStructAttr::CreateImplicit(Context));
+    RD->addAttr(MSStructAttr::CreateImplicit(Context));
 
   // FIXME: We should merge AddAlignmentAttributesForRecord with
   // AddMsStructLayoutForRecord into AddPragmaAttributesForRecord, which takes
@@ -422,6 +422,9 @@ void Sema::ActOnPragmaMSSeg(SourceLocation PragmaLocation,
   if (Action & PSK_Pop && Stack->Stack.empty())
     Diag(PragmaLocation, diag::warn_pragma_pop_failed) << PragmaName
         << "stack empty";
+  if (SegmentName &&
+      !checkSectionName(SegmentName->getLocStart(), SegmentName->getString()))
+    return;
   Stack->Act(PragmaLocation, Action, StackSlotLabel, SegmentName);
 }
 
diff --git a/lib/Sema/SemaCUDA.cpp b/lib/Sema/SemaCUDA.cpp
index 64222fbf8ac8..5973500826e3 100644
--- a/lib/Sema/SemaCUDA.cpp
+++ b/lib/Sema/SemaCUDA.cpp
@@ -62,6 +62,11 @@ Sema::CUDAFunctionTarget Sema::IdentifyCUDATarget(const FunctionDecl *D) {
 
 bool Sema::CheckCUDATarget(const FunctionDecl *Caller,
                            const FunctionDecl *Callee) {
+  // The CUDADisableTargetCallChecks short-circuits this check: we assume all
+  // cross-target calls are valid.
+  if (getLangOpts().CUDADisableTargetCallChecks)
+    return false;
+
   CUDAFunctionTarget CallerTarget = IdentifyCUDATarget(Caller),
                      CalleeTarget = IdentifyCUDATarget(Callee);
 
@@ -92,9 +97,21 @@ bool Sema::CheckCUDATarget(const FunctionDecl *Caller,
     if (Caller->isImplicit()) return false;
 
     bool InDeviceMode = getLangOpts().CUDAIsDevice;
-    if ((InDeviceMode && CalleeTarget != CFT_Device) ||
-        (!InDeviceMode && CalleeTarget != CFT_Host))
+    if (!InDeviceMode && CalleeTarget != CFT_Host)
+        return true;
+    if (InDeviceMode && CalleeTarget != CFT_Device) {
+      // Allow host device functions to call host functions if explicitly
+      // requested.
+      if (CalleeTarget == CFT_Host &&
+          getLangOpts().CUDAAllowHostCallsFromHostDevice) {
+        Diag(Caller->getLocation(),
+             diag::warn_host_calls_from_host_device)
+            << Callee->getNameAsString() << Caller->getNameAsString();
+        return false;
+      }
+
       return true;
+    }
   }
 
   return false;
diff --git a/lib/Sema/SemaCXXScopeSpec.cpp b/lib/Sema/SemaCXXScopeSpec.cpp
index 3e56e676a15c..9e146ed3a642 100644
--- a/lib/Sema/SemaCXXScopeSpec.cpp
+++ b/lib/Sema/SemaCXXScopeSpec.cpp
@@ -218,6 +218,7 @@ bool Sema::RequireCompleteDeclContext(CXXScopeSpec &SS,
   // Fixed enum types are complete, but they aren't valid as scopes
   // until we see a definition, so awkwardly pull out this special
   // case.
+  // FIXME: The definition might not be visible; complain if it is not.
   const EnumType *enumType = dyn_cast_or_null<EnumType>(tagType);
   if (!enumType || enumType->getDecl()->isCompleteDefinition())
     return false;
@@ -282,7 +283,11 @@ bool Sema::ActOnSuperScopeSpecifier(SourceLocation SuperLoc,
 
 /// \brief Determines whether the given declaration is an valid acceptable
 /// result for name lookup of a nested-name-specifier.
-bool Sema::isAcceptableNestedNameSpecifier(const NamedDecl *SD) {
+/// \param SD Declaration checked for nested-name-specifier.
+/// \param IsExtension If not null and the declaration is accepted as an
+/// extension, the pointed variable is assigned true.
+bool Sema::isAcceptableNestedNameSpecifier(const NamedDecl *SD,
+                                           bool *IsExtension) {
   if (!SD)
     return false;
 
@@ -298,14 +303,23 @@ bool Sema::isAcceptableNestedNameSpecifier(const NamedDecl *SD) {
   QualType T = Context.getTypeDeclType(cast<TypeDecl>(SD));
   if (T->isDependentType())
     return true;
-  else if (const TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(SD)) {
-    if (TD->getUnderlyingType()->isRecordType() ||
-        (Context.getLangOpts().CPlusPlus11 &&
-         TD->getUnderlyingType()->isEnumeralType()))
+  if (const TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(SD)) {
+    if (TD->getUnderlyingType()->isRecordType())
       return true;
-  } else if (isa<RecordDecl>(SD) ||
-             (Context.getLangOpts().CPlusPlus11 && isa<EnumDecl>(SD)))
+    if (TD->getUnderlyingType()->isEnumeralType()) {
+      if (Context.getLangOpts().CPlusPlus11)
+        return true;
+      if (IsExtension)
+        *IsExtension = true;
+    }
+  } else if (isa<RecordDecl>(SD)) {
     return true;
+  } else if (isa<EnumDecl>(SD)) {
+    if (Context.getLangOpts().CPlusPlus11)
+      return true;
+    if (IsExtension)
+      *IsExtension = true;
+  }
 
   return false;
 }
@@ -599,7 +613,13 @@ bool Sema::BuildCXXNestedNameSpecifier(Scope *S,
   }
 
   NamedDecl *SD = Found.getAsSingle<NamedDecl>();
-  if (isAcceptableNestedNameSpecifier(SD)) {
+  bool IsExtension = false;
+  bool AcceptSpec = isAcceptableNestedNameSpecifier(SD, &IsExtension);
+  if (!AcceptSpec && IsExtension) {
+    AcceptSpec = true;
+    Diag(IdentifierLoc, diag::ext_nested_name_spec_is_enum);
+  }
+  if (AcceptSpec) {
     if (!ObjectType.isNull() && !ObjectTypeSearchedInScope &&
         !getLangOpts().CPlusPlus11) {
       // C++03 [basic.lookup.classref]p4:
diff --git a/lib/Sema/SemaCast.cpp b/lib/Sema/SemaCast.cpp
index a4c2d9b51c26..091e77936c1e 100644
--- a/lib/Sema/SemaCast.cpp
+++ b/lib/Sema/SemaCast.cpp
@@ -389,6 +389,33 @@ static void diagnoseBadCast(Sema &S, unsigned msg, CastType castType,
 
   S.Diag(opRange.getBegin(), msg) << castType
     << src->getType() << destType << opRange << src->getSourceRange();
+
+  // Detect if both types are (ptr to) class, and note any incompleteness.
+  int DifferentPtrness = 0;
+  QualType From = destType;
+  if (auto Ptr = From->getAs<PointerType>()) {
+    From = Ptr->getPointeeType();
+    DifferentPtrness++;
+  }
+  QualType To = src->getType();
+  if (auto Ptr = To->getAs<PointerType>()) {
+    To = Ptr->getPointeeType();
+    DifferentPtrness--;
+  }
+  if (!DifferentPtrness) {
+    auto RecFrom = From->getAs<RecordType>();
+    auto RecTo = To->getAs<RecordType>();
+    if (RecFrom && RecTo) {
+      auto DeclFrom = RecFrom->getAsCXXRecordDecl();
+      if (!DeclFrom->isCompleteDefinition())
+        S.Diag(DeclFrom->getLocation(), diag::note_type_incomplete)
+          << DeclFrom->getDeclName();
+      auto DeclTo = RecTo->getAsCXXRecordDecl();
+      if (!DeclTo->isCompleteDefinition())
+        S.Diag(DeclTo->getLocation(), diag::note_type_incomplete)
+          << DeclTo->getDeclName();
+    }
+  }
 }
 
 /// UnwrapDissimilarPointerTypes - Like Sema::UnwrapSimilarPointerTypes,
@@ -665,12 +692,6 @@ void CastOperation::CheckDynamicCast() {
     }
 
     Kind = CK_DerivedToBase;
-
-    // If we are casting to or through a virtual base class, we need a
-    // vtable.
-    if (Self.BasePathInvolvesVirtualBase(BasePath))
-      Self.MarkVTableUsed(OpRange.getBegin(), 
-                          cast<CXXRecordDecl>(SrcRecord->getDecl()));
     return;
   }
 
@@ -682,8 +703,6 @@ void CastOperation::CheckDynamicCast() {
       << SrcPointee.getUnqualifiedType() << SrcExpr.get()->getSourceRange();
     SrcExpr = ExprError();
   }
-  Self.MarkVTableUsed(OpRange.getBegin(), 
-                      cast<CXXRecordDecl>(SrcRecord->getDecl()));
 
   // dynamic_cast is not available with -fno-rtti.
   // As an exception, dynamic_cast to void* is available because it doesn't
@@ -1087,6 +1106,14 @@ static TryCastResult TryStaticCast(Sema &Self, ExprResult &SrcExpr,
   if (!CStyle &&
       Self.CheckTollFreeBridgeStaticCast(DestType, SrcExpr.get(), Kind))
     return TC_Success;
+
+  // See if it looks like the user is trying to convert between
+  // related record types, and select a better diagnostic if so.
+  if (auto SrcPointer = SrcType->getAs<PointerType>())
+    if (auto DestPointer = DestType->getAs<PointerType>())
+      if (SrcPointer->getPointeeType()->getAs<RecordType>() &&
+          DestPointer->getPointeeType()->getAs<RecordType>())
+       msg = diag::err_bad_cxx_cast_unrelated_class;
   
   // We tried everything. Everything! Nothing works! :-(
   return TC_NotApplicable;
@@ -1790,8 +1817,8 @@ static TryCastResult TryReinterpretCast(Sema &Self, ExprResult &SrcExpr,
     //   can be explicitly converted to an rvalue of type "pointer to member
     //   of Y of type T2" if T1 and T2 are both function types or both object
     //   types.
-    if (DestMemPtr->getPointeeType()->isFunctionType() !=
-        SrcMemPtr->getPointeeType()->isFunctionType())
+    if (DestMemPtr->isMemberFunctionPointer() !=
+        SrcMemPtr->isMemberFunctionPointer())
       return TC_NotApplicable;
 
     // C++ 5.2.10p2: The reinterpret_cast operator shall not cast away
diff --git a/lib/Sema/SemaChecking.cpp b/lib/Sema/SemaChecking.cpp
index 8c3efdee9036..23a6fc3c4cd4 100644
--- a/lib/Sema/SemaChecking.cpp
+++ b/lib/Sema/SemaChecking.cpp
@@ -184,7 +184,7 @@ static bool SemaBuiltinCallWithStaticChain(Sema &S, CallExpr *BuiltinCall) {
     return true;
   }
 
-  QualType ReturnTy = CE->getCallReturnType();
+  QualType ReturnTy = CE->getCallReturnType(S.Context);
   QualType ArgTys[2] = { ReturnTy, ChainResult.get()->getType() };
   QualType BuiltinTy = S.Context.getFunctionType(
       ReturnTy, ArgTys, FunctionProtoType::ExtProtoInfo());
@@ -202,6 +202,28 @@ static bool SemaBuiltinCallWithStaticChain(Sema &S, CallExpr *BuiltinCall) {
   return false;
 }
 
+static bool SemaBuiltinSEHScopeCheck(Sema &SemaRef, CallExpr *TheCall,
+                                     Scope::ScopeFlags NeededScopeFlags,
+                                     unsigned DiagID) {
+  // Scopes aren't available during instantiation. Fortunately, builtin
+  // functions cannot be template args so they cannot be formed through template
+  // instantiation. Therefore checking once during the parse is sufficient.
+  if (!SemaRef.ActiveTemplateInstantiations.empty())
+    return false;
+
+  Scope *S = SemaRef.getCurScope();
+  while (S && !S->isSEHExceptScope())
+    S = S->getParent();
+  if (!S || !(S->getFlags() & NeededScopeFlags)) {
+    auto *DRE = cast<DeclRefExpr>(TheCall->getCallee()->IgnoreParenCasts());
+    SemaRef.Diag(TheCall->getExprLoc(), DiagID)
+        << DRE->getDecl()->getIdentifier();
+    return true;
+  }
+
+  return false;
+}
+
 ExprResult
 Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID,
                                CallExpr *TheCall) {
@@ -301,6 +323,11 @@ Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID,
     if (SemaBuiltinSetjmp(TheCall))
       return ExprError();
     break;
+  case Builtin::BI_setjmp:
+  case Builtin::BI_setjmpex:
+    if (checkArgCount(*this, TheCall, 1))
+      return true;
+    break;
 
   case Builtin::BI__builtin_classify_type:
     if (checkArgCount(*this, TheCall, 1)) return true;
@@ -465,6 +492,35 @@ Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID,
     if (SemaBuiltinCallWithStaticChain(*this, TheCall))
       return ExprError();
     break;
+
+  case Builtin::BI__exception_code:
+  case Builtin::BI_exception_code: {
+    if (SemaBuiltinSEHScopeCheck(*this, TheCall, Scope::SEHExceptScope,
+                                 diag::err_seh___except_block))
+      return ExprError();
+    break;
+  }
+  case Builtin::BI__exception_info:
+  case Builtin::BI_exception_info: {
+    if (SemaBuiltinSEHScopeCheck(*this, TheCall, Scope::SEHFilterScope,
+                                 diag::err_seh___except_filter))
+      return ExprError();
+    break;
+  }
+
+  case Builtin::BI__GetExceptionInfo:
+    if (checkArgCount(*this, TheCall, 1))
+      return ExprError();
+
+    if (CheckCXXThrowOperand(
+            TheCall->getLocStart(),
+            Context.getExceptionObjectType(FDecl->getParamDecl(0)->getType()),
+            TheCall))
+      return ExprError();
+
+    TheCall->setType(Context.VoidPtrTy);
+    break;
+
   }
 
   // Since the target specific builtins for each arch overlap, only check those
@@ -490,11 +546,21 @@ Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID,
         if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall))
           return ExprError();
         break;
+      case llvm::Triple::systemz:
+        if (CheckSystemZBuiltinFunctionCall(BuiltinID, TheCall))
+          return ExprError();
+        break;
       case llvm::Triple::x86:
       case llvm::Triple::x86_64:
         if (CheckX86BuiltinFunctionCall(BuiltinID, TheCall))
           return ExprError();
         break;
+      case llvm::Triple::ppc:
+      case llvm::Triple::ppc64:
+      case llvm::Triple::ppc64le:
+        if (CheckPPCBuiltinFunctionCall(BuiltinID, TheCall))
+          return ExprError();
+        break;
       default:
         break;
     }
@@ -557,7 +623,10 @@ static QualType getNeonEltType(NeonTypeFlags Flags, ASTContext &Context,
   case NeonTypeFlags::Poly16:
     return IsPolyUnsigned ? Context.UnsignedShortTy : Context.ShortTy;
   case NeonTypeFlags::Poly64:
-    return Context.UnsignedLongTy;
+    if (IsInt64Long)
+      return Context.UnsignedLongTy;
+    else
+      return Context.UnsignedLongLongTy;
   case NeonTypeFlags::Poly128:
     break;
   case NeonTypeFlags::Float16:
@@ -839,15 +908,161 @@ bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   return SemaBuiltinConstantArgRange(TheCall, i, l, u);
 }
 
+bool Sema::CheckPPCBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
+  unsigned i = 0, l = 0, u = 0;
+  bool Is64BitBltin = BuiltinID == PPC::BI__builtin_divde ||
+                      BuiltinID == PPC::BI__builtin_divdeu ||
+                      BuiltinID == PPC::BI__builtin_bpermd;
+  bool IsTarget64Bit = Context.getTargetInfo()
+                              .getTypeWidth(Context
+                                            .getTargetInfo()
+                                            .getIntPtrType()) == 64;
+  bool IsBltinExtDiv = BuiltinID == PPC::BI__builtin_divwe ||
+                       BuiltinID == PPC::BI__builtin_divweu ||
+                       BuiltinID == PPC::BI__builtin_divde ||
+                       BuiltinID == PPC::BI__builtin_divdeu;
+
+  if (Is64BitBltin && !IsTarget64Bit)
+      return Diag(TheCall->getLocStart(), diag::err_64_bit_builtin_32_bit_tgt)
+             << TheCall->getSourceRange();
+
+  if ((IsBltinExtDiv && !Context.getTargetInfo().hasFeature("extdiv")) ||
+      (BuiltinID == PPC::BI__builtin_bpermd &&
+       !Context.getTargetInfo().hasFeature("bpermd")))
+    return Diag(TheCall->getLocStart(), diag::err_ppc_builtin_only_on_pwr7)
+           << TheCall->getSourceRange();
+
+  switch (BuiltinID) {
+  default: return false;
+  case PPC::BI__builtin_altivec_crypto_vshasigmaw:
+  case PPC::BI__builtin_altivec_crypto_vshasigmad:
+    return SemaBuiltinConstantArgRange(TheCall, 1, 0, 1) ||
+           SemaBuiltinConstantArgRange(TheCall, 2, 0, 15);
+  case PPC::BI__builtin_tbegin:
+  case PPC::BI__builtin_tend: i = 0; l = 0; u = 1; break;
+  case PPC::BI__builtin_tsr: i = 0; l = 0; u = 7; break;
+  case PPC::BI__builtin_tabortwc:
+  case PPC::BI__builtin_tabortdc: i = 0; l = 0; u = 31; break;
+  case PPC::BI__builtin_tabortwci:
+  case PPC::BI__builtin_tabortdci:
+    return SemaBuiltinConstantArgRange(TheCall, 0, 0, 31) ||
+           SemaBuiltinConstantArgRange(TheCall, 2, 0, 31);
+  }
+  return SemaBuiltinConstantArgRange(TheCall, i, l, u);
+}
+
+bool Sema::CheckSystemZBuiltinFunctionCall(unsigned BuiltinID,
+                                           CallExpr *TheCall) {
+  if (BuiltinID == SystemZ::BI__builtin_tabort) {
+    Expr *Arg = TheCall->getArg(0);
+    llvm::APSInt AbortCode(32);
+    if (Arg->isIntegerConstantExpr(AbortCode, Context) &&
+        AbortCode.getSExtValue() >= 0 && AbortCode.getSExtValue() < 256)
+      return Diag(Arg->getLocStart(), diag::err_systemz_invalid_tabort_code)
+             << Arg->getSourceRange();
+  }
+
+  // For intrinsics which take an immediate value as part of the instruction,
+  // range check them here.
+  unsigned i = 0, l = 0, u = 0;
+  switch (BuiltinID) {
+  default: return false;
+  case SystemZ::BI__builtin_s390_lcbb: i = 1; l = 0; u = 15; break;
+  case SystemZ::BI__builtin_s390_verimb:
+  case SystemZ::BI__builtin_s390_verimh:
+  case SystemZ::BI__builtin_s390_verimf:
+  case SystemZ::BI__builtin_s390_verimg: i = 3; l = 0; u = 255; break;
+  case SystemZ::BI__builtin_s390_vfaeb:
+  case SystemZ::BI__builtin_s390_vfaeh:
+  case SystemZ::BI__builtin_s390_vfaef:
+  case SystemZ::BI__builtin_s390_vfaebs:
+  case SystemZ::BI__builtin_s390_vfaehs:
+  case SystemZ::BI__builtin_s390_vfaefs:
+  case SystemZ::BI__builtin_s390_vfaezb:
+  case SystemZ::BI__builtin_s390_vfaezh:
+  case SystemZ::BI__builtin_s390_vfaezf:
+  case SystemZ::BI__builtin_s390_vfaezbs:
+  case SystemZ::BI__builtin_s390_vfaezhs:
+  case SystemZ::BI__builtin_s390_vfaezfs: i = 2; l = 0; u = 15; break;
+  case SystemZ::BI__builtin_s390_vfidb:
+    return SemaBuiltinConstantArgRange(TheCall, 1, 0, 15) ||
+           SemaBuiltinConstantArgRange(TheCall, 2, 0, 15);
+  case SystemZ::BI__builtin_s390_vftcidb: i = 1; l = 0; u = 4095; break;
+  case SystemZ::BI__builtin_s390_vlbb: i = 1; l = 0; u = 15; break;
+  case SystemZ::BI__builtin_s390_vpdi: i = 2; l = 0; u = 15; break;
+  case SystemZ::BI__builtin_s390_vsldb: i = 2; l = 0; u = 15; break;
+  case SystemZ::BI__builtin_s390_vstrcb:
+  case SystemZ::BI__builtin_s390_vstrch:
+  case SystemZ::BI__builtin_s390_vstrcf:
+  case SystemZ::BI__builtin_s390_vstrczb:
+  case SystemZ::BI__builtin_s390_vstrczh:
+  case SystemZ::BI__builtin_s390_vstrczf:
+  case SystemZ::BI__builtin_s390_vstrcbs:
+  case SystemZ::BI__builtin_s390_vstrchs:
+  case SystemZ::BI__builtin_s390_vstrcfs:
+  case SystemZ::BI__builtin_s390_vstrczbs:
+  case SystemZ::BI__builtin_s390_vstrczhs:
+  case SystemZ::BI__builtin_s390_vstrczfs: i = 3; l = 0; u = 15; break;
+  }
+  return SemaBuiltinConstantArgRange(TheCall, i, l, u);
+}
+
 bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   unsigned i = 0, l = 0, u = 0;
   switch (BuiltinID) {
   default: return false;
   case X86::BI_mm_prefetch: i = 1; l = 0; u = 3; break;
+  case X86::BI__builtin_ia32_sha1rnds4: i = 2, l = 0; u = 3; break;
+  case X86::BI__builtin_ia32_vpermil2pd:
+  case X86::BI__builtin_ia32_vpermil2pd256:
+  case X86::BI__builtin_ia32_vpermil2ps:
+  case X86::BI__builtin_ia32_vpermil2ps256: i = 3, l = 0; u = 3; break;
+  case X86::BI__builtin_ia32_cmpb128_mask:
+  case X86::BI__builtin_ia32_cmpw128_mask:
+  case X86::BI__builtin_ia32_cmpd128_mask:
+  case X86::BI__builtin_ia32_cmpq128_mask:
+  case X86::BI__builtin_ia32_cmpb256_mask:
+  case X86::BI__builtin_ia32_cmpw256_mask:
+  case X86::BI__builtin_ia32_cmpd256_mask:
+  case X86::BI__builtin_ia32_cmpq256_mask:
+  case X86::BI__builtin_ia32_cmpb512_mask:
+  case X86::BI__builtin_ia32_cmpw512_mask:
+  case X86::BI__builtin_ia32_cmpd512_mask:
+  case X86::BI__builtin_ia32_cmpq512_mask:
+  case X86::BI__builtin_ia32_ucmpb128_mask:
+  case X86::BI__builtin_ia32_ucmpw128_mask:
+  case X86::BI__builtin_ia32_ucmpd128_mask:
+  case X86::BI__builtin_ia32_ucmpq128_mask:
+  case X86::BI__builtin_ia32_ucmpb256_mask:
+  case X86::BI__builtin_ia32_ucmpw256_mask:
+  case X86::BI__builtin_ia32_ucmpd256_mask:
+  case X86::BI__builtin_ia32_ucmpq256_mask:
+  case X86::BI__builtin_ia32_ucmpb512_mask:
+  case X86::BI__builtin_ia32_ucmpw512_mask:
+  case X86::BI__builtin_ia32_ucmpd512_mask:
+  case X86::BI__builtin_ia32_ucmpq512_mask: i = 2; l = 0; u = 7; break;
+  case X86::BI__builtin_ia32_roundps:
+  case X86::BI__builtin_ia32_roundpd:
+  case X86::BI__builtin_ia32_roundps256:
+  case X86::BI__builtin_ia32_roundpd256: i = 1, l = 0; u = 15; break;
+  case X86::BI__builtin_ia32_roundss:
+  case X86::BI__builtin_ia32_roundsd: i = 2, l = 0; u = 15; break;
   case X86::BI__builtin_ia32_cmpps:
   case X86::BI__builtin_ia32_cmpss:
   case X86::BI__builtin_ia32_cmppd:
-  case X86::BI__builtin_ia32_cmpsd: i = 2; l = 0; u = 31; break;
+  case X86::BI__builtin_ia32_cmpsd:
+  case X86::BI__builtin_ia32_cmpps256:
+  case X86::BI__builtin_ia32_cmppd256:
+  case X86::BI__builtin_ia32_cmpps512_mask:
+  case X86::BI__builtin_ia32_cmppd512_mask: i = 2; l = 0; u = 31; break;
+  case X86::BI__builtin_ia32_vpcomub:
+  case X86::BI__builtin_ia32_vpcomuw:
+  case X86::BI__builtin_ia32_vpcomud:
+  case X86::BI__builtin_ia32_vpcomuq:
+  case X86::BI__builtin_ia32_vpcomb:
+  case X86::BI__builtin_ia32_vpcomw:
+  case X86::BI__builtin_ia32_vpcomd:
+  case X86::BI__builtin_ia32_vpcomq: i = 2; l = 0; u = 7; break;
   }
   return SemaBuiltinConstantArgRange(TheCall, i, l, u);
 }
@@ -1119,11 +1334,14 @@ bool Sema::CheckObjCMethodCall(ObjCMethodDecl *Method, SourceLocation lbrac,
 
 bool Sema::CheckPointerCall(NamedDecl *NDecl, CallExpr *TheCall,
                             const FunctionProtoType *Proto) {
-  const VarDecl *V = dyn_cast<VarDecl>(NDecl);
-  if (!V)
+  QualType Ty;
+  if (const auto *V = dyn_cast<VarDecl>(NDecl))
+    Ty = V->getType();
+  else if (const auto *F = dyn_cast<FieldDecl>(NDecl))
+    Ty = F->getType();
+  else
     return false;
 
-  QualType Ty = V->getType();
   if (!Ty->isBlockPointerType() && !Ty->isFunctionPointerType())
     return false;
 
@@ -1220,9 +1438,10 @@ ExprResult Sema::SemaAtomicOpsOverloaded(ExprResult TheCallResult,
   //   M is C if C is an integer, and ptrdiff_t if C is a pointer, and
   //   the int parameters are for orderings.
 
-  assert(AtomicExpr::AO__c11_atomic_init == 0 &&
-         AtomicExpr::AO__c11_atomic_fetch_xor + 1 == AtomicExpr::AO__atomic_load
-         && "need to update code for modified C11 atomics");
+  static_assert(AtomicExpr::AO__c11_atomic_init == 0 &&
+                    AtomicExpr::AO__c11_atomic_fetch_xor + 1 ==
+                        AtomicExpr::AO__atomic_load,
+                "need to update code for modified C11 atomics");
   bool IsC11 = Op >= AtomicExpr::AO__c11_atomic_init &&
                Op <= AtomicExpr::AO__c11_atomic_fetch_xor;
   bool IsN = Op == AtomicExpr::AO__atomic_load_n ||
@@ -2039,7 +2258,7 @@ bool Sema::SemaBuiltinVAStartARM(CallExpr *Call) {
   if (checkBuiltinArgument(*this, Call, 0))
     return true;
 
-  static const struct {
+  const struct {
     unsigned ArgNo;
     QualType Type;
   } ArgumentTypes[] = {
@@ -2286,7 +2505,7 @@ bool Sema::SemaBuiltinAssume(CallExpr *TheCall) {
   if (Arg->isInstantiationDependent()) return false;
 
   if (Arg->HasSideEffects(Context))
-    return Diag(Arg->getLocStart(), diag::warn_assume_side_effects)
+    Diag(Arg->getLocStart(), diag::warn_assume_side_effects)
       << Arg->getSourceRange()
       << cast<FunctionDecl>(TheCall->getCalleeDecl())->getIdentifier();
 
@@ -2603,6 +2822,8 @@ Sema::FormatStringType Sema::GetFormatStringType(const FormatAttr *Format) {
   .Case("strftime", FST_Strftime)
   .Case("strfmon", FST_Strfmon)
   .Cases("kprintf", "cmn_err", "vcmn_err", "zcmn_err", FST_Kprintf)
+  .Case("freebsd_kprintf", FST_FreeBSDKPrintf)
+  .Case("os_trace", FST_OSTrace)
   .Default(FST_Unknown);
 }
 
@@ -3073,10 +3294,7 @@ void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall,
   if (InFunctionCall) {
     const Sema::SemaDiagnosticBuilder &D = S.Diag(Loc, PDiag);
     D << StringRange;
-    for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
-         I != E; ++I) {
-      D << *I;
-    }
+    D << FixIt;
   } else {
     S.Diag(IsStringLocation ? ArgumentExpr->getExprLoc() : Loc, PDiag)
       << ArgumentExpr->getSourceRange();
@@ -3086,10 +3304,7 @@ void CheckFormatHandler::EmitFormatDiagnostic(Sema &S, bool InFunctionCall,
              diag::note_format_string_defined);
 
     Note << StringRange;
-    for (ArrayRef<FixItHint>::iterator I = FixIt.begin(), E = FixIt.end();
-         I != E; ++I) {
-      Note << *I;
-    }
+    Note << FixIt;
   }
 }
 
@@ -3384,6 +3599,43 @@ CheckPrintfHandler::HandlePrintfSpecifier(const analyze_printf::PrintfSpecifier
     CoveredArgs.set(argIndex);
   }
 
+  // FreeBSD kernel extensions.
+  if (CS.getKind() == ConversionSpecifier::FreeBSDbArg ||
+      CS.getKind() == ConversionSpecifier::FreeBSDDArg) {
+    // We need at least two arguments.
+    if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex + 1))
+      return false;
+
+    // Claim the second argument.
+    CoveredArgs.set(argIndex + 1);
+
+    // Type check the first argument (int for %b, pointer for %D)
+    const Expr *Ex = getDataArg(argIndex);
+    const analyze_printf::ArgType &AT =
+      (CS.getKind() == ConversionSpecifier::FreeBSDbArg) ?
+        ArgType(S.Context.IntTy) : ArgType::CPointerTy;
+    if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType()))
+      EmitFormatDiagnostic(
+        S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
+        << AT.getRepresentativeTypeName(S.Context) << Ex->getType()
+        << false << Ex->getSourceRange(),
+        Ex->getLocStart(), /*IsStringLocation*/false,
+        getSpecifierRange(startSpecifier, specifierLen));
+
+    // Type check the second argument (char * for both %b and %D)
+    Ex = getDataArg(argIndex + 1);
+    const analyze_printf::ArgType &AT2 = ArgType::CStrTy;
+    if (AT2.isValid() && !AT2.matchesType(S.Context, Ex->getType()))
+      EmitFormatDiagnostic(
+        S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
+        << AT2.getRepresentativeTypeName(S.Context) << Ex->getType()
+        << false << Ex->getSourceRange(),
+        Ex->getLocStart(), /*IsStringLocation*/false,
+        getSpecifierRange(startSpecifier, specifierLen));
+
+     return true;
+  }
+
   // Check for using an Objective-C specific conversion specifier
   // in a non-ObjC literal.
   if (!ObjCContext && CS.isObjCArg()) {
@@ -3562,8 +3814,11 @@ CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
     ExprTy = TET->getUnderlyingExpr()->getType();
   }
 
-  if (AT.matchesType(S.Context, ExprTy))
+  analyze_printf::ArgType::MatchKind match = AT.matchesType(S.Context, ExprTy);
+
+  if (match == analyze_printf::ArgType::Match) {
     return true;
+  }
 
   // Look through argument promotions for our error message's reported type.
   // This includes the integral and floating promotions, but excludes array
@@ -3658,16 +3913,18 @@ CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
     CharSourceRange SpecRange = getSpecifierRange(StartSpecifier, SpecifierLen);
 
     if (IntendedTy == ExprTy && !ShouldNotPrintDirectly) {
+      unsigned diag = diag::warn_format_conversion_argument_type_mismatch;
+      if (match == analyze_format_string::ArgType::NoMatchPedantic) {
+        diag = diag::warn_format_conversion_argument_type_mismatch_pedantic;
+      }
       // In this case, the specifier is wrong and should be changed to match
       // the argument.
-      EmitFormatDiagnostic(
-        S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
-          << AT.getRepresentativeTypeName(S.Context) << IntendedTy << IsEnum
-          << E->getSourceRange(),
-        E->getLocStart(),
-        /*IsStringLocation*/false,
-        SpecRange,
-        FixItHint::CreateReplacement(SpecRange, os.str()));
+      EmitFormatDiagnostic(S.PDiag(diag)
+                               << AT.getRepresentativeTypeName(S.Context)
+                               << IntendedTy << IsEnum << E->getSourceRange(),
+                           E->getLocStart(),
+                           /*IsStringLocation*/ false, SpecRange,
+                           FixItHint::CreateReplacement(SpecRange, os.str()));
 
     } else {
       // The canonical type for formatting this value is different from the
@@ -3741,15 +3998,18 @@ CheckPrintfHandler::checkFormatExpr(const analyze_printf::PrintfSpecifier &FS,
     // arguments here.
     switch (S.isValidVarArgType(ExprTy)) {
     case Sema::VAK_Valid:
-    case Sema::VAK_ValidInCXX11:
+    case Sema::VAK_ValidInCXX11: {
+      unsigned diag = diag::warn_format_conversion_argument_type_mismatch;
+      if (match == analyze_printf::ArgType::NoMatchPedantic) {
+        diag = diag::warn_format_conversion_argument_type_mismatch_pedantic;
+      }
+
       EmitFormatDiagnostic(
-        S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
-          << AT.getRepresentativeTypeName(S.Context) << ExprTy << IsEnum
-          << CSR
-          << E->getSourceRange(),
-        E->getLocStart(), /*IsStringLocation*/false, CSR);
+          S.PDiag(diag) << AT.getRepresentativeTypeName(S.Context) << ExprTy
+                        << IsEnum << CSR << E->getSourceRange(),
+          E->getLocStart(), /*IsStringLocation*/ false, CSR);
       break;
-
+    }
     case Sema::VAK_Undefined:
     case Sema::VAK_MSVCUndefined:
       EmitFormatDiagnostic(
@@ -3881,13 +4141,13 @@ bool CheckScanfHandler::HandleScanfSpecifier(
                            FixItHint::CreateRemoval(R));
     }
   }
-  
+
   if (!FS.consumesDataArgument()) {
     // FIXME: Technically specifying a precision or field width here
     // makes no sense.  Worth issuing a warning at some point.
     return true;
   }
-  
+
   // Consume the argument.
   unsigned argIndex = FS.getArgIndex();
   if (argIndex < NumDataArgs) {
@@ -3896,7 +4156,7 @@ bool CheckScanfHandler::HandleScanfSpecifier(
       // function if we encounter some other error.
     CoveredArgs.set(argIndex);
   }
-  
+
   // Check the length modifier is valid with the given conversion specifier.
   if (!FS.hasValidLengthModifier(S.getASTContext().getTargetInfo()))
     HandleInvalidLengthModifier(FS, CS, startSpecifier, specifierLen,
@@ -3913,47 +4173,57 @@ bool CheckScanfHandler::HandleScanfSpecifier(
   // The remaining checks depend on the data arguments.
   if (HasVAListArg)
     return true;
-  
+
   if (!CheckNumArgs(FS, CS, startSpecifier, specifierLen, argIndex))
     return false;
-  
+
   // Check that the argument type matches the format specifier.
   const Expr *Ex = getDataArg(argIndex);
   if (!Ex)
     return true;
 
   const analyze_format_string::ArgType &AT = FS.getArgType(S.Context);
-  if (AT.isValid() && !AT.matchesType(S.Context, Ex->getType())) {
-    ScanfSpecifier fixedFS = FS;
-    bool success = fixedFS.fixType(Ex->getType(),
-                                   Ex->IgnoreImpCasts()->getType(),
-                                   S.getLangOpts(), S.Context);
-
-    if (success) {
-      // Get the fix string from the fixed format specifier.
-      SmallString<128> buf;
-      llvm::raw_svector_ostream os(buf);
-      fixedFS.toString(os);
 
-      EmitFormatDiagnostic(
-        S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
-          << AT.getRepresentativeTypeName(S.Context) << Ex->getType() << false
-          << Ex->getSourceRange(),
+  if (!AT.isValid()) {
+    return true;
+  }
+
+  analyze_format_string::ArgType::MatchKind match =
+      AT.matchesType(S.Context, Ex->getType());
+  if (match == analyze_format_string::ArgType::Match) {
+    return true;
+  }
+
+  ScanfSpecifier fixedFS = FS;
+  bool success = fixedFS.fixType(Ex->getType(), Ex->IgnoreImpCasts()->getType(),
+                                 S.getLangOpts(), S.Context);
+
+  unsigned diag = diag::warn_format_conversion_argument_type_mismatch;
+  if (match == analyze_format_string::ArgType::NoMatchPedantic) {
+    diag = diag::warn_format_conversion_argument_type_mismatch_pedantic;
+  }
+
+  if (success) {
+    // Get the fix string from the fixed format specifier.
+    SmallString<128> buf;
+    llvm::raw_svector_ostream os(buf);
+    fixedFS.toString(os);
+
+    EmitFormatDiagnostic(
+        S.PDiag(diag) << AT.getRepresentativeTypeName(S.Context)
+                      << Ex->getType() << false << Ex->getSourceRange(),
         Ex->getLocStart(),
-        /*IsStringLocation*/false,
+        /*IsStringLocation*/ false,
         getSpecifierRange(startSpecifier, specifierLen),
         FixItHint::CreateReplacement(
-          getSpecifierRange(startSpecifier, specifierLen),
-          os.str()));
-    } else {
-      EmitFormatDiagnostic(
-        S.PDiag(diag::warn_format_conversion_argument_type_mismatch)
-          << AT.getRepresentativeTypeName(S.Context) << Ex->getType() << false
-          << Ex->getSourceRange(),
-        Ex->getLocStart(),
-        /*IsStringLocation*/false,
-        getSpecifierRange(startSpecifier, specifierLen));
-    }
+            getSpecifierRange(startSpecifier, specifierLen), os.str()));
+  } else {
+    EmitFormatDiagnostic(S.PDiag(diag)
+                             << AT.getRepresentativeTypeName(S.Context)
+                             << Ex->getType() << false << Ex->getSourceRange(),
+                         Ex->getLocStart(),
+                         /*IsStringLocation*/ false,
+                         getSpecifierRange(startSpecifier, specifierLen));
   }
 
   return true;
@@ -4007,15 +4277,17 @@ void Sema::CheckFormatString(const StringLiteral *FExpr,
     return;
   }
   
-  if (Type == FST_Printf || Type == FST_NSString) {
+  if (Type == FST_Printf || Type == FST_NSString ||
+      Type == FST_FreeBSDKPrintf || Type == FST_OSTrace) {
     CheckPrintfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg,
-                         numDataArgs, (Type == FST_NSString),
+                         numDataArgs, (Type == FST_NSString || Type == FST_OSTrace),
                          Str, HasVAListArg, Args, format_idx,
                          inFunctionCall, CallType, CheckedVarArgs);
   
     if (!analyze_format_string::ParsePrintfString(H, Str, Str + StrLen,
                                                   getLangOpts(),
-                                                  Context.getTargetInfo()))
+                                                  Context.getTargetInfo(),
+                                                  Type == FST_FreeBSDKPrintf))
       H.DoneProcessing();
   } else if (Type == FST_Scanf) {
     CheckScanfHandler H(*this, FExpr, OrigFormatExpr, firstDataArg, numDataArgs,
@@ -4493,7 +4765,7 @@ static const CXXRecordDecl *getContainedDynamicClass(QualType T,
 
 /// \brief If E is a sizeof expression, returns its argument expression,
 /// otherwise returns NULL.
-static const Expr *getSizeOfExprArg(const Expr* E) {
+static const Expr *getSizeOfExprArg(const Expr *E) {
   if (const UnaryExprOrTypeTraitExpr *SizeOf =
       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
     if (SizeOf->getKind() == clang::UETT_SizeOf && !SizeOf->isArgumentType())
@@ -4503,7 +4775,7 @@ static const Expr *getSizeOfExprArg(const Expr* E) {
 }
 
 /// \brief If E is a sizeof expression, returns its argument type.
-static QualType getSizeOfArgType(const Expr* E) {
+static QualType getSizeOfArgType(const Expr *E) {
   if (const UnaryExprOrTypeTraitExpr *SizeOf =
       dyn_cast<UnaryExprOrTypeTraitExpr>(E))
     if (SizeOf->getKind() == clang::UETT_SizeOf)
@@ -4549,8 +4821,9 @@ void Sema::CheckMemaccessArguments(const CallExpr *Call,
     SourceRange ArgRange = Call->getArg(ArgIdx)->getSourceRange();
 
     QualType DestTy = Dest->getType();
+    QualType PointeeTy;
     if (const PointerType *DestPtrTy = DestTy->getAs<PointerType>()) {
-      QualType PointeeTy = DestPtrTy->getPointeeType();
+      PointeeTy = DestPtrTy->getPointeeType();
 
       // Never warn about void type pointers. This can be used to suppress
       // false positives.
@@ -4630,47 +4903,53 @@ void Sema::CheckMemaccessArguments(const CallExpr *Call,
           break;
         }
       }
+    } else if (DestTy->isArrayType()) {
+      PointeeTy = DestTy;
+    }
 
-      // Always complain about dynamic classes.
-      bool IsContained;
-      if (const CXXRecordDecl *ContainedRD =
-              getContainedDynamicClass(PointeeTy, IsContained)) {
-
-        unsigned OperationType = 0;
-        // "overwritten" if we're warning about the destination for any call
-        // but memcmp; otherwise a verb appropriate to the call.
-        if (ArgIdx != 0 || BId == Builtin::BImemcmp) {
-          if (BId == Builtin::BImemcpy)
-            OperationType = 1;
-          else if(BId == Builtin::BImemmove)
-            OperationType = 2;
-          else if (BId == Builtin::BImemcmp)
-            OperationType = 3;
-        }
-          
-        DiagRuntimeBehavior(
-          Dest->getExprLoc(), Dest,
-          PDiag(diag::warn_dyn_class_memaccess)
-            << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx)
-            << FnName << IsContained << ContainedRD << OperationType
-            << Call->getCallee()->getSourceRange());
-      } else if (PointeeTy.hasNonTrivialObjCLifetime() &&
-               BId != Builtin::BImemset)
-        DiagRuntimeBehavior(
-          Dest->getExprLoc(), Dest,
-          PDiag(diag::warn_arc_object_memaccess)
-            << ArgIdx << FnName << PointeeTy
-            << Call->getCallee()->getSourceRange());
-      else
-        continue;
+    if (PointeeTy == QualType())
+      continue;
 
+    // Always complain about dynamic classes.
+    bool IsContained;
+    if (const CXXRecordDecl *ContainedRD =
+            getContainedDynamicClass(PointeeTy, IsContained)) {
+
+      unsigned OperationType = 0;
+      // "overwritten" if we're warning about the destination for any call
+      // but memcmp; otherwise a verb appropriate to the call.
+      if (ArgIdx != 0 || BId == Builtin::BImemcmp) {
+        if (BId == Builtin::BImemcpy)
+          OperationType = 1;
+        else if(BId == Builtin::BImemmove)
+          OperationType = 2;
+        else if (BId == Builtin::BImemcmp)
+          OperationType = 3;
+      }
+        
       DiagRuntimeBehavior(
         Dest->getExprLoc(), Dest,
-        PDiag(diag::note_bad_memaccess_silence)
-          << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
-      break;
-    }
+        PDiag(diag::warn_dyn_class_memaccess)
+          << (BId == Builtin::BImemcmp ? ArgIdx + 2 : ArgIdx)
+          << FnName << IsContained << ContainedRD << OperationType
+          << Call->getCallee()->getSourceRange());
+    } else if (PointeeTy.hasNonTrivialObjCLifetime() &&
+             BId != Builtin::BImemset)
+      DiagRuntimeBehavior(
+        Dest->getExprLoc(), Dest,
+        PDiag(diag::warn_arc_object_memaccess)
+          << ArgIdx << FnName << PointeeTy
+          << Call->getCallee()->getSourceRange());
+    else
+      continue;
+
+    DiagRuntimeBehavior(
+      Dest->getExprLoc(), Dest,
+      PDiag(diag::note_bad_memaccess_silence)
+        << FixItHint::CreateInsertion(ArgRange.getBegin(), "(void*)"));
+    break;
   }
+
 }
 
 // A little helper routine: ignore addition and subtraction of integer literals.
@@ -5821,7 +6100,7 @@ static void DiagnoseOutOfRangeComparison(Sema &S, BinaryOperator *E,
   // TODO: Investigate using GetExprRange() to get tighter bounds
   // on the bit ranges.
   QualType OtherT = Other->getType();
-  if (const AtomicType *AT = dyn_cast<AtomicType>(OtherT))
+  if (const auto *AT = OtherT->getAs<AtomicType>())
     OtherT = AT->getValueType();
   IntRange OtherRange = IntRange::forValueOfType(S.Context, OtherT);
   unsigned OtherWidth = OtherRange.Width;
@@ -6665,8 +6944,11 @@ void AnalyzeImplicitConversions(Sema &S, Expr *OrigE, SourceLocation CC) {
       E = POE->getResultExpr();
   }
   
-  if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E))
-    return AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC);
+  if (const OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(E)) {
+    if (OVE->getSourceExpr())
+      AnalyzeImplicitConversions(S, OVE->getSourceExpr(), CC);
+    return;
+  }
   
   // Skip past explicit casts.
   if (isa<ExplicitCastExpr>(E)) {
@@ -6748,7 +7030,7 @@ static bool CheckForReference(Sema &SemaRef, const Expr *E,
     if (!M->getMemberDecl()->getType()->isReferenceType())
       return false;
   } else if (const CallExpr *Call = dyn_cast<CallExpr>(E)) {
-    if (!Call->getCallReturnType()->isReferenceType())
+    if (!Call->getCallReturnType(SemaRef.Context)->isReferenceType())
       return false;
     FD = Call->getDirectCallee();
   } else {
@@ -7493,6 +7775,35 @@ void Sema::CheckBitFieldInitialization(SourceLocation InitLoc,
   (void) AnalyzeBitFieldAssignment(*this, BitField, Init, InitLoc);
 }
 
+static void diagnoseArrayStarInParamType(Sema &S, QualType PType,
+                                         SourceLocation Loc) {
+  if (!PType->isVariablyModifiedType())
+    return;
+  if (const auto *PointerTy = dyn_cast<PointerType>(PType)) {
+    diagnoseArrayStarInParamType(S, PointerTy->getPointeeType(), Loc);
+    return;
+  }
+  if (const auto *ReferenceTy = dyn_cast<ReferenceType>(PType)) {
+    diagnoseArrayStarInParamType(S, ReferenceTy->getPointeeType(), Loc);
+    return;
+  }
+  if (const auto *ParenTy = dyn_cast<ParenType>(PType)) {
+    diagnoseArrayStarInParamType(S, ParenTy->getInnerType(), Loc);
+    return;
+  }
+
+  const ArrayType *AT = S.Context.getAsArrayType(PType);
+  if (!AT)
+    return;
+
+  if (AT->getSizeModifier() != ArrayType::Star) {
+    diagnoseArrayStarInParamType(S, AT->getElementType(), Loc);
+    return;
+  }
+
+  S.Diag(Loc, diag::err_array_star_in_function_definition);
+}
+
 /// CheckParmsForFunctionDef - Check that the parameters of the given
 /// function are appropriate for the definition of a function. This
 /// takes care of any checks that cannot be performed on the
@@ -7531,15 +7842,9 @@ bool Sema::CheckParmsForFunctionDef(ParmVarDecl *const *P,
     //   notation in their sequences of declarator specifiers to specify
     //   variable length array types.
     QualType PType = Param->getOriginalType();
-    while (const ArrayType *AT = Context.getAsArrayType(PType)) {
-      if (AT->getSizeModifier() == ArrayType::Star) {
-        // FIXME: This diagnostic should point the '[*]' if source-location
-        // information is added for it.
-        Diag(Param->getLocation(), diag::err_array_star_in_function_definition);
-        break;
-      }
-      PType= AT->getElementType();
-    }
+    // FIXME: This diagnostic should point the '[*]' if source-location
+    // information is added for it.
+    diagnoseArrayStarInParamType(*this, PType, Param->getLocation());
 
     // MSVC destroys objects passed by value in the callee.  Therefore a
     // function definition which takes such a parameter must be able to call the
@@ -8057,6 +8362,236 @@ static bool isSetterLikeSelector(Selector sel) {
   return !isLowercase(str.front());
 }
 
+static Optional<int> GetNSMutableArrayArgumentIndex(Sema &S,
+                                                    ObjCMessageExpr *Message) {
+  if (S.NSMutableArrayPointer.isNull()) {
+    IdentifierInfo *NSMutableArrayId =
+      S.NSAPIObj->getNSClassId(NSAPI::ClassId_NSMutableArray);
+    NamedDecl *IF = S.LookupSingleName(S.TUScope, NSMutableArrayId,
+                                       Message->getLocStart(),
+                                       Sema::LookupOrdinaryName);
+    ObjCInterfaceDecl *InterfaceDecl = dyn_cast_or_null<ObjCInterfaceDecl>(IF);
+    if (!InterfaceDecl) {
+      return None;
+    }
+    QualType NSMutableArrayObject =
+      S.Context.getObjCInterfaceType(InterfaceDecl);
+    S.NSMutableArrayPointer =
+      S.Context.getObjCObjectPointerType(NSMutableArrayObject);
+  }
+
+  if (S.NSMutableArrayPointer != Message->getReceiverType()) {
+    return None;
+  }
+
+  Selector Sel = Message->getSelector();
+
+  Optional<NSAPI::NSArrayMethodKind> MKOpt =
+    S.NSAPIObj->getNSArrayMethodKind(Sel);
+  if (!MKOpt) {
+    return None;
+  }
+
+  NSAPI::NSArrayMethodKind MK = *MKOpt;
+
+  switch (MK) {
+    case NSAPI::NSMutableArr_addObject:
+    case NSAPI::NSMutableArr_insertObjectAtIndex:
+    case NSAPI::NSMutableArr_setObjectAtIndexedSubscript:
+      return 0;
+    case NSAPI::NSMutableArr_replaceObjectAtIndex:
+      return 1;
+
+    default:
+      return None;
+  }
+
+  return None;
+}
+
+static
+Optional<int> GetNSMutableDictionaryArgumentIndex(Sema &S,
+                                                  ObjCMessageExpr *Message) {
+
+  if (S.NSMutableDictionaryPointer.isNull()) {
+    IdentifierInfo *NSMutableDictionaryId =
+      S.NSAPIObj->getNSClassId(NSAPI::ClassId_NSMutableDictionary);
+    NamedDecl *IF = S.LookupSingleName(S.TUScope, NSMutableDictionaryId,
+                                       Message->getLocStart(),
+                                       Sema::LookupOrdinaryName);
+    ObjCInterfaceDecl *InterfaceDecl = dyn_cast_or_null<ObjCInterfaceDecl>(IF);
+    if (!InterfaceDecl) {
+      return None;
+    }
+    QualType NSMutableDictionaryObject =
+      S.Context.getObjCInterfaceType(InterfaceDecl);
+    S.NSMutableDictionaryPointer =
+      S.Context.getObjCObjectPointerType(NSMutableDictionaryObject);
+  }
+
+  if (S.NSMutableDictionaryPointer != Message->getReceiverType()) {
+    return None;
+  }
+
+  Selector Sel = Message->getSelector();
+
+  Optional<NSAPI::NSDictionaryMethodKind> MKOpt =
+    S.NSAPIObj->getNSDictionaryMethodKind(Sel);
+  if (!MKOpt) {
+    return None;
+  }
+
+  NSAPI::NSDictionaryMethodKind MK = *MKOpt;
+
+  switch (MK) {
+    case NSAPI::NSMutableDict_setObjectForKey:
+    case NSAPI::NSMutableDict_setValueForKey:
+    case NSAPI::NSMutableDict_setObjectForKeyedSubscript:
+      return 0;
+
+    default:
+      return None;
+  }
+
+  return None;
+}
+
+static Optional<int> GetNSSetArgumentIndex(Sema &S, ObjCMessageExpr *Message) {
+
+  ObjCInterfaceDecl *InterfaceDecl;
+  if (S.NSMutableSetPointer.isNull()) {
+    IdentifierInfo *NSMutableSetId =
+      S.NSAPIObj->getNSClassId(NSAPI::ClassId_NSMutableSet);
+    NamedDecl *IF = S.LookupSingleName(S.TUScope, NSMutableSetId,
+                                       Message->getLocStart(),
+                                       Sema::LookupOrdinaryName);
+    InterfaceDecl = dyn_cast_or_null<ObjCInterfaceDecl>(IF);
+    if (InterfaceDecl) {
+      QualType NSMutableSetObject =
+        S.Context.getObjCInterfaceType(InterfaceDecl);
+      S.NSMutableSetPointer =
+        S.Context.getObjCObjectPointerType(NSMutableSetObject);
+    }
+  }
+
+  if (S.NSCountedSetPointer.isNull()) {
+    IdentifierInfo *NSCountedSetId =
+      S.NSAPIObj->getNSClassId(NSAPI::ClassId_NSCountedSet);
+    NamedDecl *IF = S.LookupSingleName(S.TUScope, NSCountedSetId,
+                                       Message->getLocStart(),
+                                       Sema::LookupOrdinaryName);
+    InterfaceDecl = dyn_cast_or_null<ObjCInterfaceDecl>(IF);
+    if (InterfaceDecl) {
+      QualType NSCountedSetObject =
+        S.Context.getObjCInterfaceType(InterfaceDecl);
+      S.NSCountedSetPointer =
+        S.Context.getObjCObjectPointerType(NSCountedSetObject);
+    }
+  }
+
+  if (S.NSMutableOrderedSetPointer.isNull()) {
+    IdentifierInfo *NSOrderedSetId =
+      S.NSAPIObj->getNSClassId(NSAPI::ClassId_NSMutableOrderedSet);
+    NamedDecl *IF = S.LookupSingleName(S.TUScope, NSOrderedSetId,
+                                       Message->getLocStart(),
+                                       Sema::LookupOrdinaryName);
+    InterfaceDecl = dyn_cast_or_null<ObjCInterfaceDecl>(IF);
+    if (InterfaceDecl) {
+      QualType NSOrderedSetObject =
+        S.Context.getObjCInterfaceType(InterfaceDecl);
+      S.NSMutableOrderedSetPointer =
+        S.Context.getObjCObjectPointerType(NSOrderedSetObject);
+    }
+  }
+
+  QualType ReceiverType = Message->getReceiverType();
+
+  bool IsMutableSet = !S.NSMutableSetPointer.isNull() &&
+    ReceiverType == S.NSMutableSetPointer;
+  bool IsMutableOrderedSet = !S.NSMutableOrderedSetPointer.isNull() &&
+    ReceiverType == S.NSMutableOrderedSetPointer;
+  bool IsCountedSet = !S.NSCountedSetPointer.isNull() &&
+    ReceiverType == S.NSCountedSetPointer;
+
+  if (!IsMutableSet && !IsMutableOrderedSet && !IsCountedSet) {
+    return None;
+  }
+
+  Selector Sel = Message->getSelector();
+
+  Optional<NSAPI::NSSetMethodKind> MKOpt = S.NSAPIObj->getNSSetMethodKind(Sel);
+  if (!MKOpt) {
+    return None;
+  }
+
+  NSAPI::NSSetMethodKind MK = *MKOpt;
+
+  switch (MK) {
+    case NSAPI::NSMutableSet_addObject:
+    case NSAPI::NSOrderedSet_setObjectAtIndex:
+    case NSAPI::NSOrderedSet_setObjectAtIndexedSubscript:
+    case NSAPI::NSOrderedSet_insertObjectAtIndex:
+      return 0;
+    case NSAPI::NSOrderedSet_replaceObjectAtIndexWithObject:
+      return 1;
+  }
+
+  return None;
+}
+
+void Sema::CheckObjCCircularContainer(ObjCMessageExpr *Message) {
+  if (!Message->isInstanceMessage()) {
+    return;
+  }
+
+  Optional<int> ArgOpt;
+
+  if (!(ArgOpt = GetNSMutableArrayArgumentIndex(*this, Message)) &&
+      !(ArgOpt = GetNSMutableDictionaryArgumentIndex(*this, Message)) &&
+      !(ArgOpt = GetNSSetArgumentIndex(*this, Message))) {
+    return;
+  }
+
+  int ArgIndex = *ArgOpt;
+
+  Expr *Receiver = Message->getInstanceReceiver()->IgnoreImpCasts();
+  if (OpaqueValueExpr *OE = dyn_cast<OpaqueValueExpr>(Receiver)) {
+    Receiver = OE->getSourceExpr()->IgnoreImpCasts();
+  }
+
+  Expr *Arg = Message->getArg(ArgIndex)->IgnoreImpCasts();
+  if (OpaqueValueExpr *OE = dyn_cast<OpaqueValueExpr>(Arg)) {
+    Arg = OE->getSourceExpr()->IgnoreImpCasts();
+  }
+
+  if (DeclRefExpr *ReceiverRE = dyn_cast<DeclRefExpr>(Receiver)) {
+    if (DeclRefExpr *ArgRE = dyn_cast<DeclRefExpr>(Arg)) {
+      if (ReceiverRE->getDecl() == ArgRE->getDecl()) {
+        ValueDecl *Decl = ReceiverRE->getDecl();
+        Diag(Message->getSourceRange().getBegin(),
+             diag::warn_objc_circular_container)
+          << Decl->getName();
+        Diag(Decl->getLocation(),
+             diag::note_objc_circular_container_declared_here)
+          << Decl->getName();
+      }
+    }
+  } else if (ObjCIvarRefExpr *IvarRE = dyn_cast<ObjCIvarRefExpr>(Receiver)) {
+    if (ObjCIvarRefExpr *IvarArgRE = dyn_cast<ObjCIvarRefExpr>(Arg)) {
+      if (IvarRE->getDecl() == IvarArgRE->getDecl()) {
+        ObjCIvarDecl *Decl = IvarRE->getDecl();
+        Diag(Message->getSourceRange().getBegin(),
+             diag::warn_objc_circular_container)
+          << Decl->getName();
+        Diag(Decl->getLocation(),
+             diag::note_objc_circular_container_declared_here)
+          << Decl->getName();
+      }
+    }
+  }
+
+}
+
 /// Check a message send to see if it's likely to cause a retain cycle.
 void Sema::checkRetainCycles(ObjCMessageExpr *msg) {
   // Only check instance methods whose selector looks like a setter.
@@ -8241,7 +8776,7 @@ bool ShouldDiagnoseEmptyStmtBody(const SourceManager &SourceMgr,
 
   // Get line numbers of statement and body.
   bool StmtLineInvalid;
-  unsigned StmtLine = SourceMgr.getSpellingLineNumber(StmtLoc,
+  unsigned StmtLine = SourceMgr.getPresumedLineNumber(StmtLoc,
                                                       &StmtLineInvalid);
   if (StmtLineInvalid)
     return false;
diff --git a/lib/Sema/SemaCodeComplete.cpp b/lib/Sema/SemaCodeComplete.cpp
index 48bdd2a7d82c..18d352ba829b 100644
--- a/lib/Sema/SemaCodeComplete.cpp
+++ b/lib/Sema/SemaCodeComplete.cpp
@@ -495,7 +495,6 @@ bool ResultBuilder::isInterestingDecl(const NamedDecl *ND,
   AsNestedNameSpecifier = false;
 
   ND = ND->getUnderlyingDecl();
-  unsigned IDNS = ND->getIdentifierNamespace();
 
   // Skip unnamed entities.
   if (!ND->getDeclName())
@@ -503,7 +502,7 @@ bool ResultBuilder::isInterestingDecl(const NamedDecl *ND,
   
   // Friend declarations and declarations introduced due to friends are never
   // added as results.
-  if (IDNS & (Decl::IDNS_OrdinaryFriend | Decl::IDNS_TagFriend))
+  if (ND->getFriendObjectKind() == Decl::FOK_Undeclared)
     return false;
   
   // Class template (partial) specializations are never added as results.
@@ -800,8 +799,8 @@ void ResultBuilder::MaybeAddConstructorResults(Result R) {
   DeclarationName ConstructorName
     = Context.DeclarationNames.getCXXConstructorName(
                                            Context.getCanonicalType(RecordTy));
-  DeclContext::lookup_const_result Ctors = Record->lookup(ConstructorName);
-  for (DeclContext::lookup_const_iterator I = Ctors.begin(),
+  DeclContext::lookup_result Ctors = Record->lookup(ConstructorName);
+  for (DeclContext::lookup_iterator I = Ctors.begin(),
                                           E = Ctors.end();
        I != E; ++I) {
     R.Declaration = *I;
@@ -2018,7 +2017,7 @@ static void AddOrdinaryNameResults(Sema::ParserCompletionContext CCC,
     if (SemaRef.getLangOpts().C11) {
       // _Alignof
       Builder.AddResultTypeChunk("size_t");
-      if (SemaRef.getASTContext().Idents.get("alignof").hasMacroDefinition())
+      if (SemaRef.PP.isMacroDefined("alignof"))
         Builder.AddTypedTextChunk("alignof");
       else
         Builder.AddTypedTextChunk("_Alignof");
@@ -2086,15 +2085,14 @@ static void AddResultTypeChunk(ASTContext &Context,
                                                     Result.getAllocator()));
 }
 
-static void MaybeAddSentinel(ASTContext &Context,
+static void MaybeAddSentinel(Preprocessor &PP,
                              const NamedDecl *FunctionOrMethod,
                              CodeCompletionBuilder &Result) {
   if (SentinelAttr *Sentinel = FunctionOrMethod->getAttr<SentinelAttr>())
     if (Sentinel->getSentinel() == 0) {
-      if (Context.getLangOpts().ObjC1 &&
-          Context.Idents.get("nil").hasMacroDefinition())
+      if (PP.getLangOpts().ObjC1 && PP.isMacroDefined("nil"))
         Result.AddTextChunk(", nil");
-      else if (Context.Idents.get("NULL").hasMacroDefinition())
+      else if (PP.isMacroDefined("NULL"))
         Result.AddTextChunk(", NULL");
       else
         Result.AddTextChunk(", (void*)0");
@@ -2118,8 +2116,7 @@ static std::string formatObjCParamQualifiers(unsigned ObjCQuals) {
   return Result;
 }
 
-static std::string FormatFunctionParameter(ASTContext &Context,
-                                           const PrintingPolicy &Policy,
+static std::string FormatFunctionParameter(const PrintingPolicy &Policy,
                                            const ParmVarDecl *Param,
                                            bool SuppressName = false,
                                            bool SuppressBlock = false) {
@@ -2218,7 +2215,7 @@ static std::string FormatFunctionParameter(ASTContext &Context,
     for (unsigned I = 0, N = Block.getNumParams(); I != N; ++I) {
       if (I)
         Params += ", ";
-      Params += FormatFunctionParameter(Context, Policy, Block.getParam(I),
+      Params += FormatFunctionParameter(Policy, Block.getParam(I),
                                         /*SuppressName=*/false,
                                         /*SuppressBlock=*/true);
 
@@ -2248,7 +2245,7 @@ static std::string FormatFunctionParameter(ASTContext &Context,
 }
 
 /// \brief Add function parameter chunks to the given code completion string.
-static void AddFunctionParameterChunks(ASTContext &Context,
+static void AddFunctionParameterChunks(Preprocessor &PP,
                                        const PrintingPolicy &Policy,
                                        const FunctionDecl *Function,
                                        CodeCompletionBuilder &Result,
@@ -2266,7 +2263,7 @@ static void AddFunctionParameterChunks(ASTContext &Context,
                                 Result.getCodeCompletionTUInfo());
       if (!FirstParameter)
         Opt.AddChunk(CodeCompletionString::CK_Comma);
-      AddFunctionParameterChunks(Context, Policy, Function, Opt, P, true);
+      AddFunctionParameterChunks(PP, Policy, Function, Opt, P, true);
       Result.AddOptionalChunk(Opt.TakeString());
       break;
     }
@@ -2279,9 +2276,8 @@ static void AddFunctionParameterChunks(ASTContext &Context,
     InOptional = false;
     
     // Format the placeholder string.
-    std::string PlaceholderStr = FormatFunctionParameter(Context, Policy, 
-                                                         Param);
-        
+    std::string PlaceholderStr = FormatFunctionParameter(Policy, Param);
+
     if (Function->isVariadic() && P == N - 1)
       PlaceholderStr += ", ...";
 
@@ -2296,7 +2292,7 @@ static void AddFunctionParameterChunks(ASTContext &Context,
       if (Proto->getNumParams() == 0)
         Result.AddPlaceholderChunk("...");
 
-      MaybeAddSentinel(Context, Function, Result);
+      MaybeAddSentinel(PP, Function, Result);
     }
 }
 
@@ -2309,7 +2305,11 @@ static void AddTemplateParameterChunks(ASTContext &Context,
                                        unsigned Start = 0,
                                        bool InDefaultArg = false) {
   bool FirstParameter = true;
-  
+
+  // Prefer to take the template parameter names from the first declaration of
+  // the template.
+  Template = cast<TemplateDecl>(Template->getCanonicalDecl());
+
   TemplateParameterList *Params = Template->getTemplateParameters();
   TemplateParameterList::iterator PEnd = Params->end();
   if (MaxParameters)
@@ -2572,11 +2572,7 @@ CodeCompletionResult::CreateCodeCompletionString(ASTContext &Ctx,
   }
   
   if (Kind == RK_Macro) {
-    const MacroDirective *MD = PP.getMacroDirectiveHistory(Macro);
-    assert(MD && "Not a macro?");
-    const MacroInfo *MI = MD->getMacroInfo();
-    assert((!MD->isDefined() || MI) && "missing MacroInfo for define");
-
+    const MacroInfo *MI = PP.getMacroInfo(Macro);
     Result.AddTypedTextChunk(
                             Result.getAllocator().CopyString(Macro->getName()));
 
@@ -2651,7 +2647,7 @@ CodeCompletionResult::CreateCodeCompletionString(ASTContext &Ctx,
                                    Ctx, Policy);
     AddTypedNameChunk(Ctx, Policy, ND, Result);
     Result.AddChunk(CodeCompletionString::CK_LeftParen);
-    AddFunctionParameterChunks(Ctx, Policy, Function, Result);
+    AddFunctionParameterChunks(PP, Policy, Function, Result);
     Result.AddChunk(CodeCompletionString::CK_RightParen);
     AddFunctionTypeQualsToCompletionString(Result, Function);
     return Result.TakeString();
@@ -2705,7 +2701,7 @@ CodeCompletionResult::CreateCodeCompletionString(ASTContext &Ctx,
     
     // Add the function parameters
     Result.AddChunk(CodeCompletionString::CK_LeftParen);
-    AddFunctionParameterChunks(Ctx, Policy, Function, Result);
+    AddFunctionParameterChunks(PP, Policy, Function, Result);
     Result.AddChunk(CodeCompletionString::CK_RightParen);
     AddFunctionTypeQualsToCompletionString(Result, Function);
     return Result.TakeString();
@@ -2766,7 +2762,7 @@ CodeCompletionResult::CreateCodeCompletionString(ASTContext &Ctx,
       std::string Arg;
       
       if ((*P)->getType()->isBlockPointerType() && !DeclaringEntity)
-        Arg = FormatFunctionParameter(Ctx, Policy, *P, true);
+        Arg = FormatFunctionParameter(Policy, *P, true);
       else {
         (*P)->getType().getAsStringInternal(Arg, Policy);
         Arg = "(" + formatObjCParamQualifiers((*P)->getObjCDeclQualifier()) 
@@ -2797,7 +2793,7 @@ CodeCompletionResult::CreateCodeCompletionString(ASTContext &Ctx,
           Result.AddPlaceholderChunk(", ...");
       }
       
-      MaybeAddSentinel(Ctx, Method, Result);
+      MaybeAddSentinel(PP, Method, Result);
     }
     
     return Result.TakeString();
@@ -2812,73 +2808,115 @@ CodeCompletionResult::CreateCodeCompletionString(ASTContext &Ctx,
   return Result.TakeString();
 }
 
+/// \brief Add function overload parameter chunks to the given code completion
+/// string.
+static void AddOverloadParameterChunks(ASTContext &Context,
+                                       const PrintingPolicy &Policy,
+                                       const FunctionDecl *Function,
+                                       const FunctionProtoType *Prototype,
+                                       CodeCompletionBuilder &Result,
+                                       unsigned CurrentArg,
+                                       unsigned Start = 0,
+                                       bool InOptional = false) {
+  bool FirstParameter = true;
+  unsigned NumParams = Function ? Function->getNumParams()
+                                : Prototype->getNumParams();
+
+  for (unsigned P = Start; P != NumParams; ++P) {
+    if (Function && Function->getParamDecl(P)->hasDefaultArg() && !InOptional) {
+      // When we see an optional default argument, put that argument and
+      // the remaining default arguments into a new, optional string.
+      CodeCompletionBuilder Opt(Result.getAllocator(),
+                                Result.getCodeCompletionTUInfo());
+      if (!FirstParameter)
+        Opt.AddChunk(CodeCompletionString::CK_Comma);
+      // Optional sections are nested.
+      AddOverloadParameterChunks(Context, Policy, Function, Prototype, Opt,
+                                 CurrentArg, P, /*InOptional=*/true);
+      Result.AddOptionalChunk(Opt.TakeString());
+      return;
+    }
+
+    if (FirstParameter)
+      FirstParameter = false;
+    else
+      Result.AddChunk(CodeCompletionString::CK_Comma);
+
+    InOptional = false;
+
+    // Format the placeholder string.
+    std::string Placeholder;
+    if (Function)
+      Placeholder = FormatFunctionParameter(Policy, Function->getParamDecl(P));
+    else
+      Placeholder = Prototype->getParamType(P).getAsString(Policy);
+
+    if (P == CurrentArg)
+      Result.AddCurrentParameterChunk(
+        Result.getAllocator().CopyString(Placeholder));
+    else
+      Result.AddPlaceholderChunk(Result.getAllocator().CopyString(Placeholder));
+  }
+
+  if (Prototype && Prototype->isVariadic()) {
+    CodeCompletionBuilder Opt(Result.getAllocator(),
+                              Result.getCodeCompletionTUInfo());
+    if (!FirstParameter)
+      Opt.AddChunk(CodeCompletionString::CK_Comma);
+
+    if (CurrentArg < NumParams)
+      Opt.AddPlaceholderChunk("...");
+    else
+      Opt.AddCurrentParameterChunk("...");
+
+    Result.AddOptionalChunk(Opt.TakeString());
+  }
+}
+
 CodeCompletionString *
 CodeCompleteConsumer::OverloadCandidate::CreateSignatureString(
-                                                          unsigned CurrentArg,
-                                                               Sema &S,
-                                     CodeCompletionAllocator &Allocator,
-                                     CodeCompletionTUInfo &CCTUInfo) const {
+                                             unsigned CurrentArg, Sema &S,
+                                             CodeCompletionAllocator &Allocator,
+                                             CodeCompletionTUInfo &CCTUInfo,
+                                             bool IncludeBriefComments) const {
   PrintingPolicy Policy = getCompletionPrintingPolicy(S);
 
   // FIXME: Set priority, availability appropriately.
   CodeCompletionBuilder Result(Allocator,CCTUInfo, 1, CXAvailability_Available);
   FunctionDecl *FDecl = getFunction();
-  AddResultTypeChunk(S.Context, Policy, FDecl, Result);
-  const FunctionProtoType *Proto 
+  const FunctionProtoType *Proto
     = dyn_cast<FunctionProtoType>(getFunctionType());
   if (!FDecl && !Proto) {
     // Function without a prototype. Just give the return type and a 
     // highlighted ellipsis.
     const FunctionType *FT = getFunctionType();
-    Result.AddTextChunk(GetCompletionTypeString(FT->getReturnType(), S.Context,
-                                                Policy, Result.getAllocator()));
+    Result.AddResultTypeChunk(Result.getAllocator().CopyString(
+      FT->getReturnType().getAsString(Policy)));
     Result.AddChunk(CodeCompletionString::CK_LeftParen);
     Result.AddChunk(CodeCompletionString::CK_CurrentParameter, "...");
     Result.AddChunk(CodeCompletionString::CK_RightParen);
     return Result.TakeString();
   }
-  
-  if (FDecl)
+
+  if (FDecl) {
+    if (IncludeBriefComments && CurrentArg < FDecl->getNumParams())
+      if (auto RC = S.getASTContext().getRawCommentForAnyRedecl(
+          FDecl->getParamDecl(CurrentArg)))
+        Result.addBriefComment(RC->getBriefText(S.getASTContext()));
+    AddResultTypeChunk(S.Context, Policy, FDecl, Result);
     Result.AddTextChunk(
-                    Result.getAllocator().CopyString(FDecl->getNameAsString()));
-  else
-    Result.AddTextChunk(Result.getAllocator().CopyString(
+      Result.getAllocator().CopyString(FDecl->getNameAsString()));
+  } else {
+    Result.AddResultTypeChunk(
+      Result.getAllocator().CopyString(
         Proto->getReturnType().getAsString(Policy)));
+  }
 
   Result.AddChunk(CodeCompletionString::CK_LeftParen);
-  unsigned NumParams = FDecl ? FDecl->getNumParams() : Proto->getNumParams();
-  for (unsigned I = 0; I != NumParams; ++I) {
-    if (I)
-      Result.AddChunk(CodeCompletionString::CK_Comma);
-    
-    std::string ArgString;
-    QualType ArgType;
-    
-    if (FDecl) {
-      ArgString = FDecl->getParamDecl(I)->getNameAsString();
-      ArgType = FDecl->getParamDecl(I)->getOriginalType();
-    } else {
-      ArgType = Proto->getParamType(I);
-    }
-    
-    ArgType.getAsStringInternal(ArgString, Policy);
-    
-    if (I == CurrentArg)
-      Result.AddChunk(CodeCompletionString::CK_CurrentParameter,
-                      Result.getAllocator().CopyString(ArgString));
-    else
-      Result.AddTextChunk(Result.getAllocator().CopyString(ArgString));
-  }
-  
-  if (Proto && Proto->isVariadic()) {
-    Result.AddChunk(CodeCompletionString::CK_Comma);
-    if (CurrentArg < NumParams)
-      Result.AddTextChunk("...");
-    else
-      Result.AddChunk(CodeCompletionString::CK_CurrentParameter, "...");
-  }
+  AddOverloadParameterChunks(S.getASTContext(), Policy, FDecl, Proto, Result,
+                             CurrentArg);
   Result.AddChunk(CodeCompletionString::CK_RightParen);
-  
+
   return Result.TakeString();
 }
 
@@ -2990,8 +3028,9 @@ static void AddMacroResults(Preprocessor &PP, ResultBuilder &Results,
   for (Preprocessor::macro_iterator M = PP.macro_begin(), 
                                  MEnd = PP.macro_end();
        M != MEnd; ++M) {
-    if (IncludeUndefined || M->first->hasMacroDefinition()) {
-      if (MacroInfo *MI = M->second->getMacroInfo())
+    auto MD = PP.getMacroDefinition(M->first);
+    if (IncludeUndefined || MD) {
+      if (MacroInfo *MI = MD.getMacroInfo())
         if (MI->isUsedForHeaderGuard())
           continue;
 
@@ -3816,7 +3855,74 @@ static bool anyNullArguments(ArrayRef<Expr *> Args) {
   return false;
 }
 
-void Sema::CodeCompleteCall(Scope *S, Expr *FnIn, ArrayRef<Expr *> Args) {
+typedef CodeCompleteConsumer::OverloadCandidate ResultCandidate;
+
+static void mergeCandidatesWithResults(Sema &SemaRef,
+                                      SmallVectorImpl<ResultCandidate> &Results,
+                                       OverloadCandidateSet &CandidateSet,
+                                       SourceLocation Loc) {
+  if (!CandidateSet.empty()) {
+    // Sort the overload candidate set by placing the best overloads first.
+    std::stable_sort(
+        CandidateSet.begin(), CandidateSet.end(),
+        [&](const OverloadCandidate &X, const OverloadCandidate &Y) {
+          return isBetterOverloadCandidate(SemaRef, X, Y, Loc);
+        });
+
+    // Add the remaining viable overload candidates as code-completion results.
+    for (auto &Candidate : CandidateSet)
+      if (Candidate.Viable)
+        Results.push_back(ResultCandidate(Candidate.Function));
+  }
+}
+
+/// \brief Get the type of the Nth parameter from a given set of overload
+/// candidates.
+static QualType getParamType(Sema &SemaRef,
+                             ArrayRef<ResultCandidate> Candidates,
+                             unsigned N) {
+
+  // Given the overloads 'Candidates' for a function call matching all arguments
+  // up to N, return the type of the Nth parameter if it is the same for all
+  // overload candidates.
+  QualType ParamType;
+  for (auto &Candidate : Candidates) {
+    if (auto FType = Candidate.getFunctionType())
+      if (auto Proto = dyn_cast<FunctionProtoType>(FType))
+        if (N < Proto->getNumParams()) {
+          if (ParamType.isNull())
+            ParamType = Proto->getParamType(N);
+          else if (!SemaRef.Context.hasSameUnqualifiedType(
+                        ParamType.getNonReferenceType(),
+                        Proto->getParamType(N).getNonReferenceType()))
+            // Otherwise return a default-constructed QualType.
+            return QualType();
+        }
+  }
+
+  return ParamType;
+}
+
+static void CodeCompleteOverloadResults(Sema &SemaRef, Scope *S,
+                                    MutableArrayRef<ResultCandidate> Candidates,
+                                        unsigned CurrentArg,
+                                 bool CompleteExpressionWithCurrentArg = true) {
+  QualType ParamType;
+  if (CompleteExpressionWithCurrentArg)
+    ParamType = getParamType(SemaRef, Candidates, CurrentArg);
+
+  if (ParamType.isNull())
+    SemaRef.CodeCompleteOrdinaryName(S, Sema::PCC_Expression);
+  else
+    SemaRef.CodeCompleteExpression(S, ParamType);
+
+  if (!Candidates.empty())
+    SemaRef.CodeCompleter->ProcessOverloadCandidates(SemaRef, CurrentArg,
+                                                     Candidates.data(),
+                                                     Candidates.size());
+}
+
+void Sema::CodeCompleteCall(Scope *S, Expr *Fn, ArrayRef<Expr *> Args) {
   if (!CodeCompleter)
     return;
 
@@ -3825,7 +3931,7 @@ void Sema::CodeCompleteCall(Scope *S, Expr *FnIn, ArrayRef<Expr *> Args) {
   // results. We may want to revisit this strategy in the future,
   // e.g., by merging the two kinds of results.
 
-  Expr *Fn = (Expr *)FnIn;
+  // FIXME: Provide support for variadic template functions.
 
   // Ignore type-dependent call expressions entirely.
   if (!Fn || Fn->isTypeDependent() || anyNullArguments(Args) ||
@@ -3838,92 +3944,120 @@ void Sema::CodeCompleteCall(Scope *S, Expr *FnIn, ArrayRef<Expr *> Args) {
   SourceLocation Loc = Fn->getExprLoc();
   OverloadCandidateSet CandidateSet(Loc, OverloadCandidateSet::CSK_Normal);
 
-  // FIXME: What if we're calling something that isn't a function declaration?
-  // FIXME: What if we're calling a pseudo-destructor?
-  // FIXME: What if we're calling a member function?
-  
-  typedef CodeCompleteConsumer::OverloadCandidate ResultCandidate;
   SmallVector<ResultCandidate, 8> Results;
 
   Expr *NakedFn = Fn->IgnoreParenCasts();
-  if (UnresolvedLookupExpr *ULE = dyn_cast<UnresolvedLookupExpr>(NakedFn))
+  if (auto ULE = dyn_cast<UnresolvedLookupExpr>(NakedFn))
     AddOverloadedCallCandidates(ULE, Args, CandidateSet,
-                                /*PartialOverloading=*/ true);
-  else if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(NakedFn)) {
-    FunctionDecl *FDecl = dyn_cast<FunctionDecl>(DRE->getDecl());
-    if (FDecl) {
-      if (!getLangOpts().CPlusPlus || 
-          !FDecl->getType()->getAs<FunctionProtoType>())
-        Results.push_back(ResultCandidate(FDecl));
+                                /*PartialOverloading=*/true);
+  else if (auto UME = dyn_cast<UnresolvedMemberExpr>(NakedFn)) {
+    TemplateArgumentListInfo TemplateArgsBuffer, *TemplateArgs = nullptr;
+    if (UME->hasExplicitTemplateArgs()) {
+      UME->copyTemplateArgumentsInto(TemplateArgsBuffer);
+      TemplateArgs = &TemplateArgsBuffer;
+    }
+    SmallVector<Expr *, 12> ArgExprs(1, UME->getBase());
+    ArgExprs.append(Args.begin(), Args.end());
+    UnresolvedSet<8> Decls;
+    Decls.append(UME->decls_begin(), UME->decls_end());
+    AddFunctionCandidates(Decls, ArgExprs, CandidateSet, TemplateArgs,
+                          /*SuppressUsedConversions=*/false,
+                          /*PartialOverloading=*/true);
+  } else {
+    FunctionDecl *FD = nullptr;
+    if (auto MCE = dyn_cast<MemberExpr>(NakedFn))
+      FD = dyn_cast<FunctionDecl>(MCE->getMemberDecl());
+    else if (auto DRE = dyn_cast<DeclRefExpr>(NakedFn))
+      FD = dyn_cast<FunctionDecl>(DRE->getDecl());
+    if (FD) { // We check whether it's a resolved function declaration.
+      if (!getLangOpts().CPlusPlus ||
+          !FD->getType()->getAs<FunctionProtoType>())
+        Results.push_back(ResultCandidate(FD));
       else
-        // FIXME: access?
-        AddOverloadCandidate(FDecl, DeclAccessPair::make(FDecl, AS_none), Args,
-                             CandidateSet, false, /*PartialOverloading*/true);
+        AddOverloadCandidate(FD, DeclAccessPair::make(FD, FD->getAccess()),
+                             Args, CandidateSet,
+                             /*SuppressUsedConversions=*/false,
+                             /*PartialOverloading=*/true);
+
+    } else if (auto DC = NakedFn->getType()->getAsCXXRecordDecl()) {
+      // If expression's type is CXXRecordDecl, it may overload the function
+      // call operator, so we check if it does and add them as candidates.
+      // A complete type is needed to lookup for member function call operators.
+      if (!RequireCompleteType(Loc, NakedFn->getType(), 0)) {
+        DeclarationName OpName = Context.DeclarationNames
+                                 .getCXXOperatorName(OO_Call);
+        LookupResult R(*this, OpName, Loc, LookupOrdinaryName);
+        LookupQualifiedName(R, DC);
+        R.suppressDiagnostics();
+        SmallVector<Expr *, 12> ArgExprs(1, NakedFn);
+        ArgExprs.append(Args.begin(), Args.end());
+        AddFunctionCandidates(R.asUnresolvedSet(), ArgExprs, CandidateSet,
+                              /*ExplicitArgs=*/nullptr,
+                              /*SuppressUsedConversions=*/false,
+                              /*PartialOverloading=*/true);
+      }
+    } else {
+      // Lastly we check whether expression's type is function pointer or
+      // function.
+      QualType T = NakedFn->getType();
+      if (!T->getPointeeType().isNull())
+        T = T->getPointeeType();
+
+      if (auto FP = T->getAs<FunctionProtoType>()) {
+        if (!TooManyArguments(FP->getNumParams(), Args.size(),
+                             /*PartialOverloading=*/true) ||
+            FP->isVariadic())
+          Results.push_back(ResultCandidate(FP));
+      } else if (auto FT = T->getAs<FunctionType>())
+        // No prototype and declaration, it may be a K & R style function.
+        Results.push_back(ResultCandidate(FT));
     }
   }
-  
-  QualType ParamType;
-  
-  if (!CandidateSet.empty()) {
-    // Sort the overload candidate set by placing the best overloads first.
-    std::stable_sort(
-        CandidateSet.begin(), CandidateSet.end(),
-        [&](const OverloadCandidate &X, const OverloadCandidate &Y) {
-          return isBetterOverloadCandidate(*this, X, Y, Loc);
-        });
 
-    // Add the remaining viable overload candidates as code-completion reslults.
-    for (OverloadCandidateSet::iterator Cand = CandidateSet.begin(),
-                                     CandEnd = CandidateSet.end();
-         Cand != CandEnd; ++Cand) {
-      if (Cand->Viable)
-        Results.push_back(ResultCandidate(Cand->Function));
-    }
-
-    // From the viable candidates, try to determine the type of this parameter.
-    for (unsigned I = 0, N = Results.size(); I != N; ++I) {
-      if (const FunctionType *FType = Results[I].getFunctionType())
-        if (const FunctionProtoType *Proto = dyn_cast<FunctionProtoType>(FType))
-          if (Args.size() < Proto->getNumParams()) {
-            if (ParamType.isNull())
-              ParamType = Proto->getParamType(Args.size());
-            else if (!Context.hasSameUnqualifiedType(
-                          ParamType.getNonReferenceType(),
-                          Proto->getParamType(Args.size())
-                              .getNonReferenceType())) {
-              ParamType = QualType();
-              break;
-            }
-          }
-    }
-  } else {
-    // Try to determine the parameter type from the type of the expression
-    // being called.
-    QualType FunctionType = Fn->getType();
-    if (const PointerType *Ptr = FunctionType->getAs<PointerType>())
-      FunctionType = Ptr->getPointeeType();
-    else if (const BlockPointerType *BlockPtr
-                                    = FunctionType->getAs<BlockPointerType>())
-      FunctionType = BlockPtr->getPointeeType();
-    else if (const MemberPointerType *MemPtr
-                                    = FunctionType->getAs<MemberPointerType>())
-      FunctionType = MemPtr->getPointeeType();
-    
-    if (const FunctionProtoType *Proto
-                                  = FunctionType->getAs<FunctionProtoType>()) {
-      if (Args.size() < Proto->getNumParams())
-        ParamType = Proto->getParamType(Args.size());
+  mergeCandidatesWithResults(*this, Results, CandidateSet, Loc);
+  CodeCompleteOverloadResults(*this, S, Results, Args.size(),
+                              !CandidateSet.empty());
+}
+
+void Sema::CodeCompleteConstructor(Scope *S, QualType Type, SourceLocation Loc,
+                                   ArrayRef<Expr *> Args) {
+  if (!CodeCompleter)
+    return;
+
+  // A complete type is needed to lookup for constructors.
+  if (RequireCompleteType(Loc, Type, 0))
+    return;
+
+  CXXRecordDecl *RD = Type->getAsCXXRecordDecl();
+  if (!RD) {
+    CodeCompleteExpression(S, Type);
+    return;
+  }
+
+  // FIXME: Provide support for member initializers.
+  // FIXME: Provide support for variadic template constructors.
+
+  OverloadCandidateSet CandidateSet(Loc, OverloadCandidateSet::CSK_Normal);
+
+  for (auto C : LookupConstructors(RD)) {
+    if (auto FD = dyn_cast<FunctionDecl>(C)) {
+      AddOverloadCandidate(FD, DeclAccessPair::make(FD, C->getAccess()),
+                           Args, CandidateSet,
+                           /*SuppressUsedConversions=*/false,
+                           /*PartialOverloading=*/true);
+    } else if (auto FTD = dyn_cast<FunctionTemplateDecl>(C)) {
+      AddTemplateOverloadCandidate(FTD,
+                                   DeclAccessPair::make(FTD, C->getAccess()),
+                                   /*ExplicitTemplateArgs=*/nullptr,
+                                   Args, CandidateSet,
+                                   /*SuppressUsedConversions=*/false,
+                                   /*PartialOverloading=*/true);
     }
   }
 
-  if (ParamType.isNull())
-    CodeCompleteOrdinaryName(S, PCC_Expression);
-  else
-    CodeCompleteExpression(S, ParamType);
-  
-  if (!Results.empty())
-    CodeCompleter->ProcessOverloadCandidates(*this, Args.size(), Results.data(),
-                                             Results.size());
+  SmallVector<ResultCandidate, 8> Results;
+  mergeCandidatesWithResults(*this, Results, CandidateSet, Loc);
+  CodeCompleteOverloadResults(*this, S, Results, Args.size());
 }
 
 void Sema::CodeCompleteInitializer(Scope *S, Decl *D) {
@@ -4981,7 +5115,7 @@ void Sema::CodeCompleteObjCPassingType(Scope *S, ObjCDeclSpec &DS,
   // an action, e.g.,
   //   IBAction)<#selector#>:(id)sender
   if (DS.getObjCDeclQualifier() == 0 && !IsParameter &&
-      Context.Idents.get("IBAction").hasMacroDefinition()) {
+      PP.isMacroDefined("IBAction")) {
     CodeCompletionBuilder Builder(Results.getAllocator(),
                                   Results.getCodeCompletionTUInfo(),
                                   CCP_CodePattern, CXAvailability_Available);
diff --git a/lib/Sema/SemaDecl.cpp b/lib/Sema/SemaDecl.cpp
index c6af2edee512..89f4b3a0c1fb 100644
--- a/lib/Sema/SemaDecl.cpp
+++ b/lib/Sema/SemaDecl.cpp
@@ -128,45 +128,85 @@ bool Sema::isSimpleTypeSpecifier(tok::TokenKind Kind) const {
   return false;
 }
 
+namespace {
+enum class UnqualifiedTypeNameLookupResult {
+  NotFound,
+  FoundNonType,
+  FoundType
+};
+} // namespace
+
+/// \brief Tries to perform unqualified lookup of the type decls in bases for
+/// dependent class.
+/// \return \a NotFound if no any decls is found, \a FoundNotType if found not a
+/// type decl, \a FoundType if only type decls are found.
+static UnqualifiedTypeNameLookupResult
+lookupUnqualifiedTypeNameInBase(Sema &S, const IdentifierInfo &II,
+                                SourceLocation NameLoc,
+                                const CXXRecordDecl *RD) {
+  if (!RD->hasDefinition())
+    return UnqualifiedTypeNameLookupResult::NotFound;
+  // Look for type decls in base classes.
+  UnqualifiedTypeNameLookupResult FoundTypeDecl =
+      UnqualifiedTypeNameLookupResult::NotFound;
+  for (const auto &Base : RD->bases()) {
+    const CXXRecordDecl *BaseRD = nullptr;
+    if (auto *BaseTT = Base.getType()->getAs<TagType>())
+      BaseRD = BaseTT->getAsCXXRecordDecl();
+    else if (auto *TST = Base.getType()->getAs<TemplateSpecializationType>()) {
+      // Look for type decls in dependent base classes that have known primary
+      // templates.
+      if (!TST || !TST->isDependentType())
+        continue;
+      auto *TD = TST->getTemplateName().getAsTemplateDecl();
+      if (!TD)
+        continue;
+      auto *BasePrimaryTemplate =
+          dyn_cast_or_null<CXXRecordDecl>(TD->getTemplatedDecl());
+      if (!BasePrimaryTemplate)
+        continue;
+      BaseRD = BasePrimaryTemplate;
+    }
+    if (BaseRD) {
+      for (NamedDecl *ND : BaseRD->lookup(&II)) {
+        if (!isa<TypeDecl>(ND))
+          return UnqualifiedTypeNameLookupResult::FoundNonType;
+        FoundTypeDecl = UnqualifiedTypeNameLookupResult::FoundType;
+      }
+      if (FoundTypeDecl == UnqualifiedTypeNameLookupResult::NotFound) {
+        switch (lookupUnqualifiedTypeNameInBase(S, II, NameLoc, BaseRD)) {
+        case UnqualifiedTypeNameLookupResult::FoundNonType:
+          return UnqualifiedTypeNameLookupResult::FoundNonType;
+        case UnqualifiedTypeNameLookupResult::FoundType:
+          FoundTypeDecl = UnqualifiedTypeNameLookupResult::FoundType;
+          break;
+        case UnqualifiedTypeNameLookupResult::NotFound:
+          break;
+        }
+      }
+    }
+  }
+
+  return FoundTypeDecl;
+}
+
 static ParsedType recoverFromTypeInKnownDependentBase(Sema &S,
                                                       const IdentifierInfo &II,
                                                       SourceLocation NameLoc) {
-  // Find the first parent class template context, if any.
-  // FIXME: Perform the lookup in all enclosing class templates.
+  // Lookup in the parent class template context, if any.
   const CXXRecordDecl *RD = nullptr;
-  for (DeclContext *DC = S.CurContext; DC; DC = DC->getParent()) {
+  UnqualifiedTypeNameLookupResult FoundTypeDecl =
+      UnqualifiedTypeNameLookupResult::NotFound;
+  for (DeclContext *DC = S.CurContext;
+       DC && FoundTypeDecl == UnqualifiedTypeNameLookupResult::NotFound;
+       DC = DC->getParent()) {
+    // Look for type decls in dependent base classes that have known primary
+    // templates.
     RD = dyn_cast<CXXRecordDecl>(DC);
     if (RD && RD->getDescribedClassTemplate())
-      break;
+      FoundTypeDecl = lookupUnqualifiedTypeNameInBase(S, II, NameLoc, RD);
   }
-  if (!RD)
-    return ParsedType();
-
-  // Look for type decls in dependent base classes that have known primary
-  // templates.
-  bool FoundTypeDecl = false;
-  for (const auto &Base : RD->bases()) {
-    auto *TST = Base.getType()->getAs<TemplateSpecializationType>();
-    if (!TST || !TST->isDependentType())
-      continue;
-    auto *TD = TST->getTemplateName().getAsTemplateDecl();
-    if (!TD)
-      continue;
-    auto *BasePrimaryTemplate =
-        dyn_cast_or_null<CXXRecordDecl>(TD->getTemplatedDecl());
-    if (!BasePrimaryTemplate)
-      continue;
-    // FIXME: Allow lookup into non-dependent bases of dependent bases, possibly
-    // by calling or integrating with the main LookupQualifiedName mechanism.
-    for (NamedDecl *ND : BasePrimaryTemplate->lookup(&II)) {
-      if (FoundTypeDecl)
-        return ParsedType();
-      FoundTypeDecl = isa<TypeDecl>(ND);
-      if (!FoundTypeDecl)
-        return ParsedType();
-    }
-  }
-  if (!FoundTypeDecl)
+  if (FoundTypeDecl != UnqualifiedTypeNameLookupResult::FoundType)
     return ParsedType();
 
   // We found some types in dependent base classes.  Recover as if the user
@@ -1708,7 +1748,7 @@ NamedDecl *Sema::LazilyCreateBuiltin(IdentifierInfo *II, unsigned ID,
                                            Loc, Loc, II, R, /*TInfo=*/nullptr,
                                            SC_Extern,
                                            false,
-                                           /*hasPrototype=*/true);
+                                           R->isFunctionProtoType());
   New->setImplicit();
 
   // Create Decl objects for each parameter, adding them to the
@@ -1744,11 +1784,11 @@ NamedDecl *Sema::LazilyCreateBuiltin(IdentifierInfo *II, unsigned ID,
 /// should not consider because they are not permitted to conflict, e.g.,
 /// because they come from hidden sub-modules and do not refer to the same
 /// entity.
-static void filterNonConflictingPreviousDecls(ASTContext &context,
+static void filterNonConflictingPreviousDecls(Sema &S,
                                               NamedDecl *decl,
                                               LookupResult &previous){
   // This is only interesting when modules are enabled.
-  if (!context.getLangOpts().Modules)
+  if (!S.getLangOpts().Modules)
     return;
 
   // Empty sets are uninteresting.
@@ -1760,7 +1800,7 @@ static void filterNonConflictingPreviousDecls(ASTContext &context,
     NamedDecl *old = filter.next();
 
     // Non-hidden declarations are never ignored.
-    if (!old->isHidden())
+    if (S.isVisible(old))
       continue;
 
     if (!old->isExternallyVisible())
@@ -1774,11 +1814,11 @@ static void filterNonConflictingPreviousDecls(ASTContext &context,
 /// entity if their types are the same.
 /// FIXME: This is notionally doing the same thing as ASTReaderDecl's
 /// isSameEntity.
-static void filterNonConflictingPreviousTypedefDecls(ASTContext &Context,
+static void filterNonConflictingPreviousTypedefDecls(Sema &S,
                                                      TypedefNameDecl *Decl,
                                                      LookupResult &Previous) {
   // This is only interesting when modules are enabled.
-  if (!Context.getLangOpts().Modules)
+  if (!S.getLangOpts().Modules)
     return;
 
   // Empty sets are uninteresting.
@@ -1790,16 +1830,23 @@ static void filterNonConflictingPreviousTypedefDecls(ASTContext &Context,
     NamedDecl *Old = Filter.next();
 
     // Non-hidden declarations are never ignored.
-    if (!Old->isHidden())
+    if (S.isVisible(Old))
       continue;
 
     // Declarations of the same entity are not ignored, even if they have
     // different linkages.
-    if (auto *OldTD = dyn_cast<TypedefNameDecl>(Old))
-      if (Context.hasSameType(OldTD->getUnderlyingType(),
-                              Decl->getUnderlyingType()))
+    if (auto *OldTD = dyn_cast<TypedefNameDecl>(Old)) {
+      if (S.Context.hasSameType(OldTD->getUnderlyingType(),
+                                Decl->getUnderlyingType()))
         continue;
 
+      // If both declarations give a tag declaration a typedef name for linkage
+      // purposes, then they declare the same entity.
+      if (OldTD->getAnonDeclWithTypedefName(/*AnyRedecl*/true) &&
+          Decl->getAnonDeclWithTypedefName())
+        continue;
+    }
+
     if (!Old->isExternallyVisible())
       Filter.erase();
   }
@@ -1909,6 +1956,27 @@ void Sema::MergeTypedefNameDecl(TypedefNameDecl *New, LookupResult &OldDecls) {
   if (Old->isInvalidDecl())
     return New->setInvalidDecl();
 
+  if (auto *OldTD = dyn_cast<TypedefNameDecl>(Old)) {
+    auto *OldTag = OldTD->getAnonDeclWithTypedefName(/*AnyRedecl*/true);
+    auto *NewTag = New->getAnonDeclWithTypedefName();
+    NamedDecl *Hidden = nullptr;
+    if (getLangOpts().CPlusPlus && OldTag && NewTag &&
+        OldTag->getCanonicalDecl() != NewTag->getCanonicalDecl() &&
+        !hasVisibleDefinition(OldTag, &Hidden)) {
+      // There is a definition of this tag, but it is not visible. Use it
+      // instead of our tag.
+      New->setTypeForDecl(OldTD->getTypeForDecl());
+      if (OldTD->isModed())
+        New->setModedTypeSourceInfo(OldTD->getTypeSourceInfo(),
+                                    OldTD->getUnderlyingType());
+      else
+        New->setTypeSourceInfo(OldTD->getTypeSourceInfo());
+
+      // Make the old tag definition visible.
+      makeMergedDefinitionVisible(Hidden, NewTag->getLocation());
+    }
+  }
+
   // If the typedef types are not identical, reject them in all languages and
   // with any extensions enabled.
   if (isIncompatibleTypedef(Old, New))
@@ -1978,7 +2046,6 @@ void Sema::MergeTypedefNameDecl(TypedefNameDecl *New, LookupResult &OldDecls) {
   Diag(New->getLocation(), diag::ext_redefinition_of_typedef)
     << New->getDeclName();
   Diag(Old->getLocation(), diag::note_previous_definition);
-  return;
 }
 
 /// DeclhasAttr - returns true if decl Declaration already has the target
@@ -2643,7 +2710,7 @@ bool Sema::MergeFunctionDecl(FunctionDecl *New, NamedDecl *&OldD,
   // UndefinedButUsed.
   if (!Old->isInlined() && New->isInlined() &&
       !New->hasAttr<GNUInlineAttr>() &&
-      (getLangOpts().CPlusPlus || !getLangOpts().GNUInline) &&
+      !getLangOpts().GNUInline &&
       Old->isUsed(false) &&
       !Old->isDefined() && !New->isThisDeclarationADefinition())
     UndefinedButUsed.insert(std::make_pair(Old->getCanonicalDecl(),
@@ -2761,6 +2828,7 @@ bool Sema::MergeFunctionDecl(FunctionDecl *New, NamedDecl *&OldD,
             << New << New->getType();
         }
         Diag(OldLocation, PrevDiag) << Old << Old->getType();
+        return true;
 
       // Complain if this is an explicit declaration of a special
       // member that was initially declared implicitly.
@@ -3336,14 +3404,23 @@ void Sema::MergeVarDecl(VarDecl *New, LookupResult &Previous) {
   }
 
   // C++ doesn't have tentative definitions, so go right ahead and check here.
-  const VarDecl *Def;
+  VarDecl *Def;
   if (getLangOpts().CPlusPlus &&
       New->isThisDeclarationADefinition() == VarDecl::Definition &&
       (Def = Old->getDefinition())) {
-    Diag(New->getLocation(), diag::err_redefinition) << New;
-    Diag(Def->getLocation(), diag::note_previous_definition);
-    New->setInvalidDecl();
-    return;
+    NamedDecl *Hidden = nullptr;
+    if (!hasVisibleDefinition(Def, &Hidden) && 
+        (New->getDescribedVarTemplate() ||
+         New->getNumTemplateParameterLists() ||
+         New->getDeclContext()->isDependentContext())) {
+      // The previous definition is hidden, and multiple definitions are
+      // permitted (in separate TUs). Form another definition of it.
+    } else {
+      Diag(New->getLocation(), diag::err_redefinition) << New;
+      Diag(Def->getLocation(), diag::note_previous_definition);
+      New->setInvalidDecl();
+      return;
+    }
   }
 
   if (haveIncompatibleLanguageLinkages(Old, New)) {
@@ -3375,8 +3452,20 @@ Decl *Sema::ParsedFreeStandingDeclSpec(Scope *S, AccessSpecifier AS,
   return ParsedFreeStandingDeclSpec(S, AS, DS, MultiTemplateParamsArg());
 }
 
-static void HandleTagNumbering(Sema &S, const TagDecl *Tag, Scope *TagScope) {
-  if (!S.Context.getLangOpts().CPlusPlus)
+// The MS ABI changed between VS2013 and VS2015 with regard to numbers used to
+// disambiguate entities defined in different scopes.
+// While the VS2015 ABI fixes potential miscompiles, it is also breaks
+// compatibility.
+// We will pick our mangling number depending on which version of MSVC is being
+// targeted.
+static unsigned getMSManglingNumber(const LangOptions &LO, Scope *S) {
+  return LO.isCompatibleWithMSVC(LangOptions::MSVC2015)
+             ? S->getMSCurManglingNumber()
+             : S->getMSLastManglingNumber();
+}
+
+void Sema::handleTagNumbering(const TagDecl *Tag, Scope *TagScope) {
+  if (!Context.getLangOpts().CPlusPlus)
     return;
 
   if (isa<CXXRecordDecl>(Tag->getParent())) {
@@ -3385,21 +3474,62 @@ static void HandleTagNumbering(Sema &S, const TagDecl *Tag, Scope *TagScope) {
     if (!Tag->getName().empty() || Tag->getTypedefNameForAnonDecl())
       return;
     MangleNumberingContext &MCtx =
-        S.Context.getManglingNumberContext(Tag->getParent());
-    S.Context.setManglingNumber(
-        Tag, MCtx.getManglingNumber(Tag, TagScope->getMSLocalManglingNumber()));
+        Context.getManglingNumberContext(Tag->getParent());
+    Context.setManglingNumber(
+        Tag, MCtx.getManglingNumber(
+                 Tag, getMSManglingNumber(getLangOpts(), TagScope)));
     return;
   }
 
   // If this tag isn't a direct child of a class, number it if it is local.
   Decl *ManglingContextDecl;
-  if (MangleNumberingContext *MCtx =
-          S.getCurrentMangleNumberContext(Tag->getDeclContext(),
-                                          ManglingContextDecl)) {
-    S.Context.setManglingNumber(
-        Tag,
-        MCtx->getManglingNumber(Tag, TagScope->getMSLocalManglingNumber()));
+  if (MangleNumberingContext *MCtx = getCurrentMangleNumberContext(
+          Tag->getDeclContext(), ManglingContextDecl)) {
+    Context.setManglingNumber(
+        Tag, MCtx->getManglingNumber(
+                 Tag, getMSManglingNumber(getLangOpts(), TagScope)));
+  }
+}
+
+void Sema::setTagNameForLinkagePurposes(TagDecl *TagFromDeclSpec,
+                                        TypedefNameDecl *NewTD) {
+  // Do nothing if the tag is not anonymous or already has an
+  // associated typedef (from an earlier typedef in this decl group).
+  if (TagFromDeclSpec->getIdentifier())
+    return;
+  if (TagFromDeclSpec->getTypedefNameForAnonDecl())
+    return;
+
+  // A well-formed anonymous tag must always be a TUK_Definition.
+  assert(TagFromDeclSpec->isThisDeclarationADefinition());
+
+  // The type must match the tag exactly;  no qualifiers allowed.
+  if (!Context.hasSameType(NewTD->getUnderlyingType(),
+                           Context.getTagDeclType(TagFromDeclSpec)))
+    return;
+
+  // If we've already computed linkage for the anonymous tag, then
+  // adding a typedef name for the anonymous decl can change that
+  // linkage, which might be a serious problem.  Diagnose this as
+  // unsupported and ignore the typedef name.  TODO: we should
+  // pursue this as a language defect and establish a formal rule
+  // for how to handle it.
+  if (TagFromDeclSpec->hasLinkageBeenComputed()) {
+    Diag(NewTD->getLocation(), diag::err_typedef_changes_linkage);
+
+    SourceLocation tagLoc = TagFromDeclSpec->getInnerLocStart();
+    tagLoc = getLocForEndOfToken(tagLoc);
+
+    llvm::SmallString<40> textToInsert;
+    textToInsert += ' ';
+    textToInsert += NewTD->getIdentifier()->getName();
+    Diag(tagLoc, diag::note_typedef_changes_linkage)
+        << FixItHint::CreateInsertion(tagLoc, textToInsert);
+    return;
   }
+
+  // Otherwise, set this is the anon-decl typedef for the tag.
+  TagFromDeclSpec->setTypedefNameForAnonDecl(NewTD);
 }
 
 /// ParsedFreeStandingDeclSpec - This method is invoked when a declspec with
@@ -3431,7 +3561,7 @@ Decl *Sema::ParsedFreeStandingDeclSpec(Scope *S, AccessSpecifier AS,
   }
 
   if (Tag) {
-    HandleTagNumbering(*this, Tag, S);
+    handleTagNumbering(Tag, S);
     Tag->setFreeStanding();
     if (Tag->isInvalidDecl())
       return Tag;
@@ -3471,7 +3601,7 @@ Decl *Sema::ParsedFreeStandingDeclSpec(Scope *S, AccessSpecifier AS,
     return ActOnFriendTypeDecl(S, DS, TemplateParams);
   }
 
-  CXXScopeSpec &SS = DS.getTypeSpecScope();
+  const CXXScopeSpec &SS = DS.getTypeSpecScope();
   bool IsExplicitSpecialization =
     !TemplateParams.empty() && TemplateParams.back()->size() == 0;
   if (Tag && SS.isNotEmpty() && !Tag->isCompleteDefinition() &&
@@ -3498,7 +3628,8 @@ Decl *Sema::ParsedFreeStandingDeclSpec(Scope *S, AccessSpecifier AS,
         DS.getStorageClassSpec() != DeclSpec::SCS_typedef) {
       if (getLangOpts().CPlusPlus ||
           Record->getDeclContext()->isRecord())
-        return BuildAnonymousStructOrUnion(S, DS, AS, Record, Context.getPrintingPolicy());
+        return BuildAnonymousStructOrUnion(S, DS, AS, Record,
+                                           Context.getPrintingPolicy());
 
       DeclaresAnything = false;
     }
@@ -3726,8 +3857,7 @@ static bool InjectAnonymousStructOrUnionMembers(Sema &SemaRef, Scope *S,
         //   anonymous union is declared.
         unsigned OldChainingSize = Chaining.size();
         if (IndirectFieldDecl *IF = dyn_cast<IndirectFieldDecl>(VD))
-          for (auto *PI : IF->chain())
-            Chaining.push_back(PI);
+          Chaining.append(IF->chain_begin(), IF->chain_end());
         else
           Chaining.push_back(VD);
 
@@ -4054,10 +4184,11 @@ Decl *Sema::BuildAnonymousStructOrUnion(Scope *S, DeclSpec &DS,
   if (VarDecl *NewVD = dyn_cast<VarDecl>(Anon)) {
     if (getLangOpts().CPlusPlus && NewVD->isStaticLocal()) {
       Decl *ManglingContextDecl;
-      if (MangleNumberingContext *MCtx =
-              getCurrentMangleNumberContext(NewVD->getDeclContext(),
-                                            ManglingContextDecl)) {
-        Context.setManglingNumber(NewVD, MCtx->getManglingNumber(NewVD, S->getMSLocalManglingNumber()));
+      if (MangleNumberingContext *MCtx = getCurrentMangleNumberContext(
+              NewVD->getDeclContext(), ManglingContextDecl)) {
+        Context.setManglingNumber(
+            NewVD, MCtx->getManglingNumber(
+                       NewVD, getMSManglingNumber(getLangOpts(), S)));
         Context.setStaticLocalNumber(NewVD, MCtx->getStaticLocalNumber(NewVD));
       }
     }
@@ -4767,6 +4898,8 @@ static QualType TryToFixInvalidVariablyModifiedType(QualType T,
 
 static void
 FixInvalidVariablyModifiedTypeLoc(TypeLoc SrcTL, TypeLoc DstTL) {
+  SrcTL = SrcTL.getUnqualifiedLoc();
+  DstTL = DstTL.getUnqualifiedLoc();
   if (PointerTypeLoc SrcPTL = SrcTL.getAs<PointerTypeLoc>()) {
     PointerTypeLoc DstPTL = DstTL.castAs<PointerTypeLoc>();
     FixInvalidVariablyModifiedTypeLoc(SrcPTL.getPointeeLoc(),
@@ -4823,27 +4956,13 @@ Sema::RegisterLocallyScopedExternCDecl(NamedDecl *ND, Scope *S) {
     return;
 
   // Note that we have a locally-scoped external with this name.
-  // FIXME: There can be multiple such declarations if they are functions marked
-  // __attribute__((overloadable)) declared in function scope in C.
-  LocallyScopedExternCDecls[ND->getDeclName()] = ND;
+  Context.getExternCContextDecl()->makeDeclVisibleInContext(ND);
 }
 
 NamedDecl *Sema::findLocallyScopedExternCDecl(DeclarationName Name) {
-  if (ExternalSource) {
-    // Load locally-scoped external decls from the external source.
-    // FIXME: This is inefficient. Maybe add a DeclContext for extern "C" decls?
-    SmallVector<NamedDecl *, 4> Decls;
-    ExternalSource->ReadLocallyScopedExternCDecls(Decls);
-    for (unsigned I = 0, N = Decls.size(); I != N; ++I) {
-      llvm::DenseMap<DeclarationName, NamedDecl *>::iterator Pos
-        = LocallyScopedExternCDecls.find(Decls[I]->getDeclName());
-      if (Pos == LocallyScopedExternCDecls.end())
-        LocallyScopedExternCDecls[Decls[I]->getDeclName()] = Decls[I];
-    }
-  }
-
-  NamedDecl *D = LocallyScopedExternCDecls.lookup(Name);
-  return D ? D->getMostRecentDecl() : nullptr;
+  // FIXME: We can have multiple results via __attribute__((overloadable)).
+  auto Result = Context.getExternCContextDecl()->lookup(Name);
+  return Result.empty() ? nullptr : *Result.begin();
 }
 
 /// \brief Diagnose function specifiers on a declaration of an identifier that
@@ -4955,7 +5074,7 @@ Sema::ActOnTypedefNameDecl(Scope *S, DeclContext *DC, TypedefNameDecl *NewTD,
   // in an outer scope, it isn't the same thing.
   FilterLookupForScope(Previous, DC, S, /*ConsiderLinkage*/false,
                        /*AllowInlineNamespace*/false);
-  filterNonConflictingPreviousTypedefDecls(Context, NewTD, Previous);
+  filterNonConflictingPreviousTypedefDecls(*this, NewTD, Previous);
   if (!Previous.empty()) {
     Redeclaration = true;
     MergeTypedefNameDecl(NewTD, Previous);
@@ -5102,14 +5221,27 @@ static void checkAttributesAfterMerging(Sema &S, NamedDecl &ND) {
     if (ND.isExternallyVisible()) {
       S.Diag(Attr->getLocation(), diag::err_attribute_weakref_not_static);
       ND.dropAttr<WeakRefAttr>();
+      ND.dropAttr<AliasAttr>();
     }
   }
 
-  // 'selectany' only applies to externally visible varable declarations.
+  if (auto *VD = dyn_cast<VarDecl>(&ND)) {
+    if (VD->hasInit()) {
+      if (const auto *Attr = VD->getAttr<AliasAttr>()) {
+        assert(VD->isThisDeclarationADefinition() &&
+               !VD->isExternallyVisible() && "Broken AliasAttr handled late!");
+        S.Diag(Attr->getLocation(), diag::err_alias_is_definition) << VD;
+        VD->dropAttr<AliasAttr>();
+      }
+    }
+  }
+
+  // 'selectany' only applies to externally visible variable declarations.
   // It does not apply to functions.
   if (SelectAnyAttr *Attr = ND.getAttr<SelectAnyAttr>()) {
     if (isa<FunctionDecl>(ND) || !ND.isExternallyVisible()) {
-      S.Diag(Attr->getLocation(), diag::err_attribute_selectany_non_extern_data);
+      S.Diag(Attr->getLocation(),
+             diag::err_attribute_selectany_non_extern_data);
       ND.dropAttr<SelectAnyAttr>();
     }
   }
@@ -5565,8 +5697,9 @@ Sema::ActOnVariableDeclarator(Scope *S, Declarator &D, DeclContext *DC,
         }
       }
     } else {
-      assert(D.getName().getKind() != UnqualifiedId::IK_TemplateId &&
-             "should have a 'template<>' for this decl");
+      assert(
+          (Invalid || D.getName().getKind() != UnqualifiedId::IK_TemplateId) &&
+          "should have a 'template<>' for this decl");
     }
 
     if (IsVariableTemplateSpecialization) {
@@ -5627,6 +5760,7 @@ Sema::ActOnVariableDeclarator(Scope *S, Declarator &D, DeclContext *DC,
   if (IsLocalExternDecl)
     NewVD->setLocalExternDecl();
 
+  bool EmitTLSUnsupportedError = false;
   if (DeclSpec::TSCS TSCS = D.getDeclSpec().getThreadStorageClassSpec()) {
     // C++11 [dcl.stc]p4:
     //   When thread_local is applied to a variable of block scope the
@@ -5641,10 +5775,20 @@ Sema::ActOnVariableDeclarator(Scope *S, Declarator &D, DeclContext *DC,
       Diag(D.getDeclSpec().getThreadStorageClassSpecLoc(),
            diag::err_thread_non_global)
         << DeclSpec::getSpecifierName(TSCS);
-    else if (!Context.getTargetInfo().isTLSSupported())
-      Diag(D.getDeclSpec().getThreadStorageClassSpecLoc(),
-           diag::err_thread_unsupported);
-    else
+    else if (!Context.getTargetInfo().isTLSSupported()) {
+      if (getLangOpts().CUDA) {
+        // Postpone error emission until we've collected attributes required to
+        // figure out whether it's a host or device variable and whether the
+        // error should be ignored.
+        EmitTLSUnsupportedError = true;
+        // We still need to mark the variable as TLS so it shows up in AST with
+        // proper storage class for other tools to use even if we're not going
+        // to emit any code for it.
+        NewVD->setTSCSpec(TSCS);
+      } else
+        Diag(D.getDeclSpec().getThreadStorageClassSpecLoc(),
+             diag::err_thread_unsupported);
+    } else
       NewVD->setTSCSpec(TSCS);
   }
 
@@ -5693,6 +5837,9 @@ Sema::ActOnVariableDeclarator(Scope *S, Declarator &D, DeclContext *DC,
   ProcessDeclAttributes(S, NewVD, D);
 
   if (getLangOpts().CUDA) {
+    if (EmitTLSUnsupportedError && DeclAttrsMatchCUDAMode(getLangOpts(), NewVD))
+      Diag(D.getDeclSpec().getThreadStorageClassSpecLoc(),
+           diag::err_thread_unsupported);
     // CUDA B.2.5: "__shared__ and __constant__ variables have implied static
     // storage [duration]."
     if (SC == SC_None && S->getFnParent() != nullptr &&
@@ -5845,11 +5992,11 @@ Sema::ActOnVariableDeclarator(Scope *S, Declarator &D, DeclContext *DC,
 
   if (getLangOpts().CPlusPlus && NewVD->isStaticLocal()) {
     Decl *ManglingContextDecl;
-    if (MangleNumberingContext *MCtx =
-            getCurrentMangleNumberContext(NewVD->getDeclContext(),
-                                          ManglingContextDecl)) {
+    if (MangleNumberingContext *MCtx = getCurrentMangleNumberContext(
+            NewVD->getDeclContext(), ManglingContextDecl)) {
       Context.setManglingNumber(
-          NewVD, MCtx->getManglingNumber(NewVD, S->getMSLocalManglingNumber()));
+          NewVD, MCtx->getManglingNumber(
+                     NewVD, getMSManglingNumber(getLangOpts(), S)));
       Context.setStaticLocalNumber(NewVD, MCtx->getStaticLocalNumber(NewVD));
     }
   }
@@ -6243,7 +6390,7 @@ bool Sema::CheckVariableDeclaration(VarDecl *NewVD, LookupResult &Previous) {
     Previous.setShadowed();
 
   // Filter out any non-conflicting previous declarations.
-  filterNonConflictingPreviousDecls(Context, NewVD, Previous);
+  filterNonConflictingPreviousDecls(*this, NewVD, Previous);
 
   if (!Previous.empty()) {
     MergeVarDecl(NewVD, Previous);
@@ -6898,8 +7045,10 @@ static void checkIsValidOpenCLKernelParameter(
 
       // We have an error, now let's go back up through history and show where
       // the offending field came from
-      for (ArrayRef<const FieldDecl *>::const_iterator I = HistoryStack.begin() + 1,
-             E = HistoryStack.end(); I != E; ++I) {
+      for (ArrayRef<const FieldDecl *>::const_iterator
+               I = HistoryStack.begin() + 1,
+               E = HistoryStack.end();
+           I != E; ++I) {
         const FieldDecl *OuterField = *I;
         S.Diag(OuterField->getLocation(), diag::note_within_field_of_type)
           << OuterField->getType();
@@ -7010,12 +7159,12 @@ Sema::ActOnFunctionDeclarator(Scope *S, Declarator &D, DeclContext *DC,
 
         // Check that we can declare a template here.
         if (CheckTemplateDeclScope(S, TemplateParams))
-          return nullptr;
+          NewFD->setInvalidDecl();
 
         // A destructor cannot be a template.
         if (Name.getNameKind() == DeclarationName::CXXDestructorName) {
           Diag(NewFD->getLocation(), diag::err_destructor_template);
-          return nullptr;
+          NewFD->setInvalidDecl();
         }
         
         // If we're adding a template to a dependent context, we may need to 
@@ -7347,7 +7496,8 @@ Sema::ActOnFunctionDeclarator(Scope *S, Declarator &D, DeclContext *DC,
     NewFD->setInvalidDecl();
   }
 
-  if (D.isFunctionDefinition() && CodeSegStack.CurrentValue &&
+  // Apply an implicit SectionAttr if #pragma code_seg is active.
+  if (CodeSegStack.CurrentValue && D.isFunctionDefinition() &&
       !NewFD->hasAttr<SectionAttr>()) {
     NewFD->addAttr(
         SectionAttr::CreateImplicit(Context, SectionAttr::Declspec_allocate,
@@ -7363,23 +7513,10 @@ Sema::ActOnFunctionDeclarator(Scope *S, Declarator &D, DeclContext *DC,
   // Handle attributes.
   ProcessDeclAttributes(S, NewFD, D);
 
-  QualType RetType = NewFD->getReturnType();
-  const CXXRecordDecl *Ret = RetType->isRecordType() ?
-      RetType->getAsCXXRecordDecl() : RetType->getPointeeCXXRecordDecl();
-  if (!NewFD->isInvalidDecl() && !NewFD->hasAttr<WarnUnusedResultAttr>() &&
-      Ret && Ret->hasAttr<WarnUnusedResultAttr>()) {
-    const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(NewFD);
-    // Attach WarnUnusedResult to functions returning types with that attribute.
-    // Don't apply the attribute to that type's own non-static member functions
-    // (to avoid warning on things like assignment operators)
-    if (!MD || MD->getParent() != Ret)
-      NewFD->addAttr(WarnUnusedResultAttr::CreateImplicit(Context));
-  }
-
   if (getLangOpts().OpenCL) {
     // OpenCL v1.1 s6.5: Using an address space qualifier in a function return
     // type declaration will generate a compilation error.
-    unsigned AddressSpace = RetType.getAddressSpace();
+    unsigned AddressSpace = NewFD->getReturnType().getAddressSpace();
     if (AddressSpace == LangAS::opencl_local ||
         AddressSpace == LangAS::opencl_global ||
         AddressSpace == LangAS::opencl_constant) {
@@ -7402,7 +7539,7 @@ Sema::ActOnFunctionDeclarator(Scope *S, Declarator &D, DeclContext *DC,
       D.setRedeclaration(CheckFunctionDeclaration(S, NewFD, Previous,
                                                   isExplicitSpecialization));
     else if (!Previous.empty())
-      // Make graceful recovery from an invalid redeclaration.
+      // Recover gracefully from an invalid redeclaration.
       D.setRedeclaration(true);
     assert((NewFD->isInvalidDecl() || !D.isRedeclaration() ||
             Previous.getResultKind() != LookupResult::FoundOverloaded) &&
@@ -7543,6 +7680,9 @@ Sema::ActOnFunctionDeclarator(Scope *S, Declarator &D, DeclContext *DC,
       if (!NewFD->isInvalidDecl())
         D.setRedeclaration(CheckFunctionDeclaration(S, NewFD, Previous,
                                                     isExplicitSpecialization));
+      else if (!Previous.empty())
+        // Recover gracefully from an invalid redeclaration.
+        D.setRedeclaration(true);
     }
 
     assert((NewFD->isInvalidDecl() || !D.isRedeclaration() ||
@@ -7799,7 +7939,7 @@ bool Sema::CheckFunctionDeclaration(Scope *S, FunctionDecl *NewFD,
                                !Previous.isShadowed();
 
   // Filter out any non-conflicting previous declarations.
-  filterNonConflictingPreviousDecls(Context, NewFD, Previous);
+  filterNonConflictingPreviousDecls(*this, NewFD, Previous);
 
   bool Redeclaration = false;
   NamedDecl *OldDecl = nullptr;
@@ -7854,7 +7994,7 @@ bool Sema::CheckFunctionDeclaration(Scope *S, FunctionDecl *NewFD,
   // Check for a previous extern "C" declaration with this name.
   if (!Redeclaration &&
       checkForConflictWithNonVisibleExternC(*this, NewFD, Previous)) {
-    filterNonConflictingPreviousDecls(Context, NewFD, Previous);
+    filterNonConflictingPreviousDecls(*this, NewFD, Previous);
     if (!Previous.empty()) {
       // This is an extern "C" declaration with the same name as a previous
       // declaration, and thus redeclares that entity...
@@ -7952,28 +8092,8 @@ bool Sema::CheckFunctionDeclaration(Scope *S, FunctionDecl *NewFD,
       // This needs to happen first so that 'inline' propagates.
       NewFD->setPreviousDeclaration(cast<FunctionDecl>(OldDecl));
 
-      if (isa<CXXMethodDecl>(NewFD)) {
-        // A valid redeclaration of a C++ method must be out-of-line,
-        // but (unfortunately) it's not necessarily a definition
-        // because of templates, which means that the previous
-        // declaration is not necessarily from the class definition.
-
-        // For just setting the access, that doesn't matter.
-        CXXMethodDecl *oldMethod = cast<CXXMethodDecl>(OldDecl);
-        NewFD->setAccess(oldMethod->getAccess());
-
-        // Update the key-function state if necessary for this ABI.
-        if (NewFD->isInlined() &&
-            !Context.getTargetInfo().getCXXABI().canKeyFunctionBeInline()) {
-          // setNonKeyFunction needs to work with the original
-          // declaration from the class definition, and isVirtual() is
-          // just faster in that case, so map back to that now.
-          oldMethod = cast<CXXMethodDecl>(oldMethod->getFirstDecl());
-          if (oldMethod->isVirtual()) {
-            Context.setNonKeyFunction(oldMethod);
-          }
-        }
-      }
+      if (isa<CXXMethodDecl>(NewFD))
+        NewFD->setAccess(OldDecl->getAccess());
     }
   }
 
@@ -8153,6 +8273,12 @@ void Sema::CheckMain(FunctionDecl* FD, const DeclSpec& DS) {
 
   bool HasExtraParameters = (nparams > 3);
 
+  if (FTP->isVariadic()) {
+    Diag(FD->getLocation(), diag::ext_variadic_main);
+    // FIXME: if we had information about the location of the ellipsis, we
+    // could add a FixIt hint to remove it as a parameter.
+  }
+
   // Darwin passes an undocumented fourth argument of type char**.  If
   // other platforms start sprouting these, the logic below will start
   // getting shifty.
@@ -8543,8 +8669,13 @@ namespace {
         diag = diag::warn_uninit_self_reference_in_reference_init;
       } else if (cast<VarDecl>(OrigDecl)->isStaticLocal()) {
         diag = diag::warn_static_self_reference_in_init;
-      } else {
+      } else if (isa<TranslationUnitDecl>(OrigDecl->getDeclContext()) ||
+                 isa<NamespaceDecl>(OrigDecl->getDeclContext()) ||
+                 DRE->getDecl()->getType()->isRecordType()) {
         diag = diag::warn_uninit_self_reference_in_init;
+      } else {
+        // Local variables will be handled by the CFG analysis.
+        return;
       }
 
       S.DiagRuntimeBehavior(DRE->getLocStart(), DRE,
@@ -8586,7 +8717,7 @@ void Sema::AddInitializerToDecl(Decl *RealDecl, Expr *Init,
   // If there is no declaration, there was an error parsing it.  Just ignore
   // the initializer.
   if (!RealDecl || RealDecl->isInvalidDecl()) {
-    CorrectDelayedTyposInExpr(Init);
+    CorrectDelayedTyposInExpr(Init, dyn_cast_or_null<VarDecl>(RealDecl));
     return;
   }
 
@@ -8617,6 +8748,21 @@ void Sema::AddInitializerToDecl(Decl *RealDecl, Expr *Init,
 
   // C++11 [decl.spec.auto]p6. Deduce the type which 'auto' stands in for.
   if (TypeMayContainAuto && VDecl->getType()->isUndeducedType()) {
+    // Attempt typo correction early so that the type of the init expression can
+    // be deduced based on the chosen correction:if the original init contains a
+    // TypoExpr.
+    ExprResult Res = CorrectDelayedTyposInExpr(Init, VDecl);
+    if (!Res.isUsable()) {
+      RealDecl->setInvalidDecl();
+      return;
+    }
+
+    if (Res.get() != Init) {
+      Init = Res.get();
+      if (CXXDirectInit)
+        CXXDirectInit = dyn_cast<ParenListExpr>(Init);
+    }
+
     Expr *DeduceInit = Init;
     // Initializer could be a C++ direct-initializer. Deduction only works if it
     // contains exactly one expression.
@@ -8746,16 +8892,24 @@ void Sema::AddInitializerToDecl(Decl *RealDecl, Expr *Init,
       VDecl->setInvalidDecl();
   }
 
-  const VarDecl *Def;
+  VarDecl *Def;
   if ((Def = VDecl->getDefinition()) && Def != VDecl) {
-    Diag(VDecl->getLocation(), diag::err_redefinition)
-      << VDecl->getDeclName();
-    Diag(Def->getLocation(), diag::note_previous_definition);
-    VDecl->setInvalidDecl();
-    return;
+    NamedDecl *Hidden = nullptr;
+    if (!hasVisibleDefinition(Def, &Hidden) && 
+        (VDecl->getDescribedVarTemplate() ||
+         VDecl->getNumTemplateParameterLists() ||
+         VDecl->getDeclContext()->isDependentContext())) {
+      // The previous definition is hidden, and multiple definitions are
+      // permitted (in separate TUs). Form another definition of it.
+    } else {
+      Diag(VDecl->getLocation(), diag::err_redefinition)
+        << VDecl->getDeclName();
+      Diag(Def->getLocation(), diag::note_previous_definition);
+      VDecl->setInvalidDecl();
+      return;
+    }
   }
 
-  const VarDecl *PrevInit = nullptr;
   if (getLangOpts().CPlusPlus) {
     // C++ [class.static.data]p4
     //   If a static data member is of const integral or const
@@ -8769,10 +8923,12 @@ void Sema::AddInitializerToDecl(Decl *RealDecl, Expr *Init,
     // We already performed a redefinition check above, but for static
     // data members we also need to check whether there was an in-class
     // declaration with an initializer.
-    if (VDecl->isStaticDataMember() && VDecl->getAnyInitializer(PrevInit)) {
+    if (VDecl->isStaticDataMember() && VDecl->getCanonicalDecl()->hasInit()) {
       Diag(Init->getExprLoc(), diag::err_static_data_member_reinitialization)
           << VDecl->getDeclName();
-      Diag(PrevInit->getInit()->getExprLoc(), diag::note_previous_initializer) << 0;
+      Diag(VDecl->getCanonicalDecl()->getInit()->getExprLoc(),
+           diag::note_previous_initializer)
+          << 0;
       return;
     }  
 
@@ -8829,8 +8985,8 @@ void Sema::AddInitializerToDecl(Decl *RealDecl, Expr *Init,
 
     // Try to correct any TypoExprs in the initialization arguments.
     for (size_t Idx = 0; Idx < Args.size(); ++Idx) {
-      ExprResult Res =
-          CorrectDelayedTyposInExpr(Args[Idx], [this, Entity, Kind](Expr *E) {
+      ExprResult Res = CorrectDelayedTyposInExpr(
+          Args[Idx], VDecl, [this, Entity, Kind](Expr *E) {
             InitializationSequence Init(*this, Entity, Kind, MultiExprArg(E));
             return Init.Failed() ? ExprError() : E;
           });
@@ -9253,6 +9409,8 @@ void Sema::ActOnUninitializedDecl(Decl *RealDecl,
         Var->setInvalidDecl();
         return;
       }
+    } else {
+      return;
     }
 
     // The variable can not have an abstract class type.
@@ -9350,8 +9508,6 @@ void Sema::ActOnCXXForRangeDecl(Decl *D) {
   case SC_OpenCLWorkGroupLocal:
     llvm_unreachable("Unexpected storage class");
   }
-  if (VD->isConstexpr())
-    Error = 5;
   if (Error != -1) {
     Diag(VD->getOuterLocStart(), diag::err_for_range_storage_class)
       << VD->getDeclName() << Error;
@@ -9453,7 +9609,9 @@ void Sema::CheckCompleteVariableDeclaration(VarDecl *var) {
 
   }
 
-  if (var->isThisDeclarationADefinition() &&
+  // Apply section attributes and pragmas to global variables.
+  bool GlobalStorage = var->hasGlobalStorage();
+  if (GlobalStorage && var->isThisDeclarationADefinition() &&
       ActiveTemplateInstantiations.empty()) {
     PragmaStack<StringLiteral *> *Stack = nullptr;
     int SectionFlags = ASTContext::PSF_Implicit | ASTContext::PSF_Read;
@@ -9466,11 +9624,11 @@ void Sema::CheckCompleteVariableDeclaration(VarDecl *var) {
       Stack = &DataSegStack;
       SectionFlags |= ASTContext::PSF_Write;
     }
-    if (!var->hasAttr<SectionAttr>() && Stack->CurrentValue)
-      var->addAttr(
-          SectionAttr::CreateImplicit(Context, SectionAttr::Declspec_allocate,
-                                      Stack->CurrentValue->getString(),
-                                      Stack->CurrentPragmaLocation));
+    if (Stack->CurrentValue && !var->hasAttr<SectionAttr>()) {
+      var->addAttr(SectionAttr::CreateImplicit(
+          Context, SectionAttr::Declspec_allocate,
+          Stack->CurrentValue->getString(), Stack->CurrentPragmaLocation));
+    }
     if (const SectionAttr *SA = var->getAttr<SectionAttr>())
       if (UnifySection(SA->getName(), SectionFlags, var))
         var->dropAttr<SectionAttr>();
@@ -9513,7 +9671,7 @@ void Sema::CheckCompleteVariableDeclaration(VarDecl *var) {
   }
 
   Expr *Init = var->getInit();
-  bool IsGlobal = var->hasGlobalStorage() && !var->isStaticLocal();
+  bool IsGlobal = GlobalStorage && !var->isStaticLocal();
   QualType baseType = Context.getBaseElementType(type);
 
   if (!var->getDeclContext()->isDependentContext() &&
@@ -9625,18 +9783,6 @@ Sema::FinalizeDeclaration(Decl *ThisDecl) {
     }
   }
 
-  if (!VD->isInvalidDecl() &&
-      VD->isThisDeclarationADefinition() == VarDecl::TentativeDefinition) {
-    if (const VarDecl *Def = VD->getDefinition()) {
-      if (Def->hasAttr<AliasAttr>()) {
-        Diag(VD->getLocation(), diag::err_tentative_after_alias)
-            << VD->getDeclName();
-        Diag(Def->getLocation(), diag::note_previous_definition);
-        VD->setInvalidDecl();
-      }
-    }
-  }
-
   const DeclContext *DC = VD->getDeclContext();
   // If there's a #pragma GCC visibility in scope, and this isn't a class
   // member, set the visibility of this variable.
@@ -9699,7 +9845,7 @@ Sema::DeclGroupPtrTy Sema::FinalizeDeclaratorGroup(Scope *S, const DeclSpec &DS,
 
   if (DeclSpec::isDeclRep(DS.getTypeSpecType())) {
     if (TagDecl *Tag = dyn_cast_or_null<TagDecl>(DS.getRepAsDecl())) {
-      HandleTagNumbering(*this, Tag, S);
+      handleTagNumbering(Tag, S);
       if (!Tag->hasNameForLinkage() && !Tag->hasDeclaratorForAnonDecl())
         Tag->setDeclaratorForAnonDecl(FirstDeclaratorInGroup);
     }
@@ -9766,9 +9912,12 @@ void Sema::ActOnDocumentableDecl(Decl *D) {
 void Sema::ActOnDocumentableDecls(ArrayRef<Decl *> Group) {
   // Don't parse the comment if Doxygen diagnostics are ignored.
   if (Group.empty() || !Group[0])
-   return;
+    return;
 
-  if (Diags.isIgnored(diag::warn_doc_param_not_found, Group[0]->getLocation()))
+  if (Diags.isIgnored(diag::warn_doc_param_not_found,
+                      Group[0]->getLocation()) &&
+      Diags.isIgnored(diag::warn_unknown_comment_command_name,
+                      Group[0]->getLocation()))
     return;
 
   if (Group.size() >= 2) {
@@ -10058,7 +10207,7 @@ void Sema::ActOnFinishKNRParamDeclarations(Scope *S, Declarator &D,
             << "  int " << FTI.Params[i].Ident->getName() << ";\n";
         Diag(FTI.Params[i].IdentLoc, diag::ext_param_not_declared)
             << FTI.Params[i].Ident
-            << FixItHint::CreateInsertion(LocAfterDecls, Code.str());
+            << FixItHint::CreateInsertion(LocAfterDecls, Code);
 
         // Implicitly declare the argument as type 'int' for lack of a better
         // type.
@@ -10127,6 +10276,10 @@ static bool ShouldWarnAboutMissingPrototype(const FunctionDecl *FD,
   if (FD->hasAttr<OpenCLKernelAttr>())
     return false;
 
+  // Don't warn on explicitly deleted functions.
+  if (FD->isDeleted())
+    return false;
+
   bool MissingPrototype = true;
   for (const FunctionDecl *Prev = FD->getPreviousDecl();
        Prev; Prev = Prev->getPreviousDecl()) {
@@ -10157,6 +10310,18 @@ Sema::CheckForFunctionRedefinition(FunctionDecl *FD,
   if (canRedefineFunction(Definition, getLangOpts()))
     return;
 
+  // If we don't have a visible definition of the function, and it's inline or
+  // a template, it's OK to form another definition of it.
+  //
+  // FIXME: Should we skip the body of the function and use the old definition
+  // in this case? That may be necessary for functions that return local types
+  // through a deduced return type, or instantiate templates with local types.
+  if (!hasVisibleDefinition(Definition) &&
+      (Definition->isInlineSpecified() ||
+       Definition->getDescribedFunctionTemplate() ||
+       Definition->getNumTemplateParameterLists()))
+    return;
+
   if (getLangOpts().GNUMode && Definition->isInlineSpecified() &&
       Definition->getStorageClass() == SC_Extern)
     Diag(FD->getLocation(), diag::err_redefinition_extern_inline)
@@ -10269,30 +10434,6 @@ Decl *Sema::ActOnStartOfFunctionDef(Scope *FnBodyScope, Decl *D) {
                           diag::err_func_def_incomplete_result))
     FD->setInvalidDecl();
 
-  // GNU warning -Wmissing-prototypes:
-  //   Warn if a global function is defined without a previous
-  //   prototype declaration. This warning is issued even if the
-  //   definition itself provides a prototype. The aim is to detect
-  //   global functions that fail to be declared in header files.
-  const FunctionDecl *PossibleZeroParamPrototype = nullptr;
-  if (ShouldWarnAboutMissingPrototype(FD, PossibleZeroParamPrototype)) {
-    Diag(FD->getLocation(), diag::warn_missing_prototype) << FD;
-
-    if (PossibleZeroParamPrototype) {
-      // We found a declaration that is not a prototype,
-      // but that could be a zero-parameter prototype
-      if (TypeSourceInfo *TI =
-              PossibleZeroParamPrototype->getTypeSourceInfo()) {
-        TypeLoc TL = TI->getTypeLoc();
-        if (FunctionNoProtoTypeLoc FTL = TL.getAs<FunctionNoProtoTypeLoc>())
-          Diag(PossibleZeroParamPrototype->getLocation(),
-               diag::note_declaration_not_a_prototype)
-            << PossibleZeroParamPrototype
-            << FixItHint::CreateInsertion(FTL.getRParenLoc(), "void");
-      }
-    }
-  }
-
   if (FnBodyScope)
     PushDeclContext(FnBodyScope, FD);
 
@@ -10321,9 +10462,9 @@ Decl *Sema::ActOnStartOfFunctionDef(Scope *FnBodyScope, Decl *D) {
          I != E; ++I) {
       NamedDecl *D = *I;
 
-      // Some of these decls (like enums) may have been pinned to the translation unit
-      // for lack of a real context earlier. If so, remove from the translation unit
-      // and reattach to the current context.
+      // Some of these decls (like enums) may have been pinned to the
+      // translation unit for lack of a real context earlier. If so, remove
+      // from the translation unit and reattach to the current context.
       if (D->getLexicalDeclContext() == Context.getTranslationUnitDecl()) {
         // Is the decl actually in the context?
         for (const auto *DI : Context.getTranslationUnitDecl()->decls()) {
@@ -10468,6 +10609,23 @@ Decl *Sema::ActOnFinishFunctionBody(Decl *dcl, Stmt *Body,
         Context.adjustDeducedFunctionResultType(
             FD, SubstAutoType(ResultType.getType(), Context.VoidTy));
       }
+    } else if (getLangOpts().CPlusPlus11 && isLambdaCallOperator(FD)) {
+      auto *LSI = getCurLambda();
+      if (LSI->HasImplicitReturnType) {
+        deduceClosureReturnType(*LSI);
+
+        // C++11 [expr.prim.lambda]p4:
+        //   [...] if there are no return statements in the compound-statement
+        //   [the deduced type is] the type void
+        QualType RetType =
+            LSI->ReturnType.isNull() ? Context.VoidTy : LSI->ReturnType;
+
+        // Update the return type to the deduced type.
+        const FunctionProtoType *Proto =
+            FD->getType()->getAs<FunctionProtoType>();
+        FD->setType(Context.getFunctionType(RetType, Proto->getParamTypes(),
+                                            Proto->getExtProtoInfo()));
+      }
     }
 
     // The only way to be included in UndefinedButUsed is if there is an
@@ -10477,7 +10635,7 @@ Decl *Sema::ActOnFinishFunctionBody(Decl *dcl, Stmt *Body,
       if (!FD->isExternallyVisible())
         UndefinedButUsed.erase(FD);
       else if (FD->isInlined() &&
-               (LangOpts.CPlusPlus || !LangOpts.GNUInline) &&
+               !LangOpts.GNUInline &&
                (!FD->getPreviousDecl()->hasAttr<GNUInlineAttr>()))
         UndefinedButUsed.erase(FD);
     }
@@ -10494,7 +10652,7 @@ Decl *Sema::ActOnFinishFunctionBody(Decl *dcl, Stmt *Body,
 
     if (!FD->isInvalidDecl()) {
       // Don't diagnose unused parameters of defaulted or deleted functions.
-      if (Body)
+      if (!FD->isDeleted() && !FD->isDefaulted())
         DiagnoseUnusedParameters(FD->param_begin(), FD->param_end());
       DiagnoseSizeOfParametersAndReturnValue(FD->param_begin(), FD->param_end(),
                                              FD->getReturnType(), FD);
@@ -10512,7 +10670,55 @@ Decl *Sema::ActOnFinishFunctionBody(Decl *dcl, Stmt *Body,
           !FD->isDependentContext())
         computeNRVO(Body, getCurFunction());
     }
-    
+
+    // GNU warning -Wmissing-prototypes:
+    //   Warn if a global function is defined without a previous
+    //   prototype declaration. This warning is issued even if the
+    //   definition itself provides a prototype. The aim is to detect
+    //   global functions that fail to be declared in header files.
+    const FunctionDecl *PossibleZeroParamPrototype = nullptr;
+    if (ShouldWarnAboutMissingPrototype(FD, PossibleZeroParamPrototype)) {
+      Diag(FD->getLocation(), diag::warn_missing_prototype) << FD;
+
+      if (PossibleZeroParamPrototype) {
+        // We found a declaration that is not a prototype,
+        // but that could be a zero-parameter prototype
+        if (TypeSourceInfo *TI =
+                PossibleZeroParamPrototype->getTypeSourceInfo()) {
+          TypeLoc TL = TI->getTypeLoc();
+          if (FunctionNoProtoTypeLoc FTL = TL.getAs<FunctionNoProtoTypeLoc>())
+            Diag(PossibleZeroParamPrototype->getLocation(),
+                 diag::note_declaration_not_a_prototype)
+                << PossibleZeroParamPrototype
+                << FixItHint::CreateInsertion(FTL.getRParenLoc(), "void");
+        }
+      }
+    }
+
+    if (auto *MD = dyn_cast<CXXMethodDecl>(FD)) {
+      const CXXMethodDecl *KeyFunction;
+      if (MD->isOutOfLine() && (MD = MD->getCanonicalDecl()) &&
+          MD->isVirtual() &&
+          (KeyFunction = Context.getCurrentKeyFunction(MD->getParent())) &&
+          MD == KeyFunction->getCanonicalDecl()) {
+        // Update the key-function state if necessary for this ABI.
+        if (FD->isInlined() &&
+            !Context.getTargetInfo().getCXXABI().canKeyFunctionBeInline()) {
+          Context.setNonKeyFunction(MD);
+
+          // If the newly-chosen key function is already defined, then we
+          // need to mark the vtable as used retroactively.
+          KeyFunction = Context.getCurrentKeyFunction(MD->getParent());
+          const FunctionDecl *Definition;
+          if (KeyFunction && KeyFunction->isDefined(Definition))
+            MarkVTableUsed(Definition->getLocation(), MD->getParent(), true);
+        } else {
+          // We just defined they key function; mark the vtable as used.
+          MarkVTableUsed(FD->getLocation(), MD->getParent(), true);
+        }
+      }
+    }
+
     assert((FD == getCurFunctionDecl() || getCurLambda()->CallOperator == FD) &&
            "Function parsing confused");
   } else if (ObjCMethodDecl *MD = dyn_cast_or_null<ObjCMethodDecl>(dcl)) {
@@ -10561,7 +10767,8 @@ Decl *Sema::ActOnFinishFunctionBody(Decl *dcl, Stmt *Body,
     if (getCurFunction()->ObjCWarnForNoInitDelegation) {
       // Don't issue this warning for unavaialable inits.
       if (!MD->isUnavailable())
-        Diag(MD->getLocation(), diag::warn_objc_secondary_init_missing_init_call);
+        Diag(MD->getLocation(),
+             diag::warn_objc_secondary_init_missing_init_call);
       getCurFunction()->ObjCWarnForNoInitDelegation = false;
     }
   } else {
@@ -10573,7 +10780,7 @@ Decl *Sema::ActOnFinishFunctionBody(Decl *dcl, Stmt *Body,
          "handled in the block above.");
 
   // Verify and clean out per-function state.
-  if (Body) {
+  if (Body && (!FD || !FD->isDefaulted())) {
     // C++ constructors that have function-try-blocks can't have return
     // statements in the handlers of that block. (C++ [except.handle]p14)
     // Verify this.
@@ -10623,8 +10830,9 @@ Decl *Sema::ActOnFinishFunctionBody(Decl *dcl, Stmt *Body,
       }
     }
 
-    assert(ExprCleanupObjects.size() == ExprEvalContexts.back().NumCleanupObjects
-           && "Leftover temporaries in function");
+    assert(ExprCleanupObjects.size() ==
+               ExprEvalContexts.back().NumCleanupObjects &&
+           "Leftover temporaries in function");
     assert(!ExprNeedsCleanups && "Unaccounted cleanups in function");
     assert(MaybeODRUseExprs.empty() &&
            "Leftover expressions for odr-use checking");
@@ -10888,44 +11096,10 @@ TypedefDecl *Sema::ParseTypedefDecl(Scope *S, Declarator &D, QualType T,
   case TST_union:
   case TST_class: {
     TagDecl *tagFromDeclSpec = cast<TagDecl>(D.getDeclSpec().getRepAsDecl());
-
-    // Do nothing if the tag is not anonymous or already has an
-    // associated typedef (from an earlier typedef in this decl group).
-    if (tagFromDeclSpec->getIdentifier()) break;
-    if (tagFromDeclSpec->getTypedefNameForAnonDecl()) break;
-
-    // A well-formed anonymous tag must always be a TUK_Definition.
-    assert(tagFromDeclSpec->isThisDeclarationADefinition());
-
-    // The type must match the tag exactly;  no qualifiers allowed.
-    if (!Context.hasSameType(T, Context.getTagDeclType(tagFromDeclSpec)))
-      break;
-
-    // If we've already computed linkage for the anonymous tag, then
-    // adding a typedef name for the anonymous decl can change that
-    // linkage, which might be a serious problem.  Diagnose this as
-    // unsupported and ignore the typedef name.  TODO: we should
-    // pursue this as a language defect and establish a formal rule
-    // for how to handle it.
-    if (tagFromDeclSpec->hasLinkageBeenComputed()) {
-      Diag(D.getIdentifierLoc(), diag::err_typedef_changes_linkage);
-
-      SourceLocation tagLoc = D.getDeclSpec().getTypeSpecTypeLoc();
-      tagLoc = getLocForEndOfToken(tagLoc);
-
-      llvm::SmallString<40> textToInsert;
-      textToInsert += ' ';
-      textToInsert += D.getIdentifier()->getName();
-      Diag(tagLoc, diag::note_typedef_changes_linkage)
-        << FixItHint::CreateInsertion(tagLoc, textToInsert);
-      break;
-    }
-
-    // Otherwise, set this is the anon-decl typedef for the tag.
-    tagFromDeclSpec->setTypedefNameForAnonDecl(NewTD);
+    setTagNameForLinkagePurposes(tagFromDeclSpec, NewTD);
     break;
   }
-    
+
   default:
     break;
   }
@@ -11146,13 +11320,16 @@ static FixItHint createFriendTagNNSFixIt(Sema &SemaRef, NamedDecl *ND, Scope *S,
   return FixItHint::CreateInsertion(NameLoc, Insertion);
 }
 
-/// ActOnTag - This is invoked when we see 'struct foo' or 'struct {'.  In the
+/// \brief This is invoked when we see 'struct foo' or 'struct {'.  In the
 /// former case, Name will be non-null.  In the later case, Name will be null.
 /// TagSpec indicates what kind of tag this is. TUK indicates whether this is a
 /// reference/declaration/definition of a tag.
 ///
-/// IsTypeSpecifier is true if this is a type-specifier (or
+/// \param IsTypeSpecifier \c true if this is a type-specifier (or
 /// trailing-type-specifier) other than one in an alias-declaration.
+///
+/// \param SkipBody If non-null, will be set to indicate if the caller should
+/// skip the definition of this tag and treat it as if it were a declaration.
 Decl *Sema::ActOnTag(Scope *S, unsigned TagSpec, TagUseKind TUK,
                      SourceLocation KWLoc, CXXScopeSpec &SS,
                      IdentifierInfo *Name, SourceLocation NameLoc,
@@ -11163,7 +11340,7 @@ Decl *Sema::ActOnTag(Scope *S, unsigned TagSpec, TagUseKind TUK,
                      SourceLocation ScopedEnumKWLoc,
                      bool ScopedEnumUsesClassTag,
                      TypeResult UnderlyingType,
-                     bool IsTypeSpecifier) {
+                     bool IsTypeSpecifier, SkipBodyInfo *SkipBody) {
   // If this is not a definition, it must have a name.
   IdentifierInfo *OrigName = Name;
   assert((Name != nullptr || TUK == TUK_Definition) &&
@@ -11206,7 +11383,8 @@ Decl *Sema::ActOnTag(Scope *S, unsigned TagSpec, TagUseKind TUK,
                                                ModulePrivateLoc,
                                                /*FriendLoc*/SourceLocation(),
                                                TemplateParameterLists.size()-1,
-                                               TemplateParameterLists.data());
+                                               TemplateParameterLists.data(),
+                                               SkipBody);
         return Result.get();
       } else {
         // The "template<>" header is extraneous.
@@ -11472,6 +11650,10 @@ Decl *Sema::ActOnTag(Scope *S, unsigned TagSpec, TagUseKind TUK,
     }
   }
 
+  // If we have a known previous declaration to use, then use it.
+  if (Previous.empty() && SkipBody && SkipBody->Previous)
+    Previous.addDecl(SkipBody->Previous);
+
   if (!Previous.empty()) {
     NamedDecl *PrevDecl = Previous.getFoundDecl();
     NamedDecl *DirectPrevDecl =
@@ -11589,7 +11771,7 @@ Decl *Sema::ActOnTag(Scope *S, unsigned TagSpec, TagUseKind TUK,
 
           // Diagnose attempts to redefine a tag.
           if (TUK == TUK_Definition) {
-            if (TagDecl *Def = PrevTagDecl->getDefinition()) {
+            if (NamedDecl *Def = PrevTagDecl->getDefinition()) {
               // If we're defining a specialization and the previous definition
               // is from an implicit instantiation, don't emit an error
               // here; we'll catch this in the general case below.
@@ -11605,7 +11787,18 @@ Decl *Sema::ActOnTag(Scope *S, unsigned TagSpec, TagUseKind TUK,
                     TSK_ExplicitSpecialization;
               }
 
-              if (!IsExplicitSpecializationAfterInstantiation) {
+              NamedDecl *Hidden = nullptr;
+              if (SkipBody && getLangOpts().CPlusPlus &&
+                  !hasVisibleDefinition(Def, &Hidden)) {
+                // There is a definition of this tag, but it is not visible. We
+                // explicitly make use of C++'s one definition rule here, and
+                // assume that this definition is identical to the hidden one
+                // we already have. Make the existing definition visible and
+                // use it in place of this one.
+                SkipBody->ShouldSkip = true;
+                makeMergedDefinitionVisible(Hidden, KWLoc);
+                return Def;
+              } else if (!IsExplicitSpecializationAfterInstantiation) {
                 // A redeclaration in function prototype scope in C isn't
                 // visible elsewhere, so merely issue a warning.
                 if (!getLangOpts().CPlusPlus && S->containedInPrototypeScope())
@@ -12339,8 +12532,10 @@ FieldDecl *Sema::CheckFieldDecl(DeclarationName Name, QualType T,
     InvalidDecl = true;
 
   bool ZeroWidth = false;
+  if (InvalidDecl)
+    BitWidth = nullptr;
   // If this is declared as a bit-field, check the bit-field.
-  if (!InvalidDecl && BitWidth) {
+  if (BitWidth) {
     BitWidth = VerifyBitField(Loc, II, T, Record->isMsStruct(Context), BitWidth,
                               &ZeroWidth).get();
     if (!BitWidth) {
@@ -12354,7 +12549,8 @@ FieldDecl *Sema::CheckFieldDecl(DeclarationName Name, QualType T,
   if (!InvalidDecl && Mutable) {
     unsigned DiagID = 0;
     if (T->isReferenceType())
-      DiagID = diag::err_mutable_reference;
+      DiagID = getLangOpts().MSVCCompat ? diag::ext_mutable_reference
+                                        : diag::err_mutable_reference;
     else if (T.isConstQualified())
       DiagID = diag::err_mutable_const;
 
@@ -12363,8 +12559,10 @@ FieldDecl *Sema::CheckFieldDecl(DeclarationName Name, QualType T,
       if (D && D->getDeclSpec().getStorageClassSpecLoc().isValid())
         ErrLoc = D->getDeclSpec().getStorageClassSpecLoc();
       Diag(ErrLoc, DiagID);
-      Mutable = false;
-      InvalidDecl = true;
+      if (DiagID != diag::ext_mutable_reference) {
+        Mutable = false;
+        InvalidDecl = true;
+      }
     }
   }
 
@@ -13286,6 +13484,29 @@ EnumConstantDecl *Sema::CheckEnumConstant(EnumDecl *Enum,
                                   Val, EnumVal);
 }
 
+Sema::SkipBodyInfo Sema::shouldSkipAnonEnumBody(Scope *S, IdentifierInfo *II,
+                                                SourceLocation IILoc) {
+  if (!getLangOpts().Modules || !getLangOpts().CPlusPlus)
+    return SkipBodyInfo();
+
+  // We have an anonymous enum definition. Look up the first enumerator to
+  // determine if we should merge the definition with an existing one and
+  // skip the body.
+  NamedDecl *PrevDecl = LookupSingleName(S, II, IILoc, LookupOrdinaryName,
+                                         ForRedeclaration);
+  auto *PrevECD = dyn_cast_or_null<EnumConstantDecl>(PrevDecl);
+  NamedDecl *Hidden;
+  if (PrevECD &&
+      !hasVisibleDefinition(cast<NamedDecl>(PrevECD->getDeclContext()),
+                            &Hidden)) {
+    SkipBodyInfo Skip;
+    Skip.ShouldSkip = true;
+    Skip.Previous = Hidden;
+    return Skip;
+  }
+
+  return SkipBodyInfo();
+}
 
 Decl *Sema::ActOnEnumConstant(Scope *S, Decl *theEnumDecl, Decl *lastEnumConst,
                               SourceLocation IdLoc, IdentifierInfo *Id,
@@ -13528,6 +13749,49 @@ static void CheckForDuplicateEnumValues(Sema &S, ArrayRef<Decl *> Elements,
   }
 }
 
+bool
+Sema::IsValueInFlagEnum(const EnumDecl *ED, const llvm::APInt &Val,
+                        bool AllowMask) const {
+  FlagEnumAttr *FEAttr = ED->getAttr<FlagEnumAttr>();
+  assert(FEAttr && "looking for value in non-flag enum");
+
+  llvm::APInt FlagMask = ~FEAttr->getFlagBits();
+  unsigned Width = FlagMask.getBitWidth();
+
+  // We will try a zero-extended value for the regular check first.
+  llvm::APInt ExtVal = Val.zextOrSelf(Width);
+
+  // A value is in a flag enum if either its bits are a subset of the enum's
+  // flag bits (the first condition) or we are allowing masks and the same is
+  // true of its complement (the second condition). When masks are allowed, we
+  // allow the common idiom of ~(enum1 | enum2) to be a valid enum value.
+  //
+  // While it's true that any value could be used as a mask, the assumption is
+  // that a mask will have all of the insignificant bits set. Anything else is
+  // likely a logic error.
+  if (!(FlagMask & ExtVal))
+    return true;
+
+  if (AllowMask) {
+    // Try a one-extended value instead. This can happen if the enum is wider
+    // than the constant used, in C with extensions to allow for wider enums.
+    // The mask will still have the correct behaviour, so we give the user the
+    // benefit of the doubt.
+    //
+    // FIXME: This heuristic can cause weird results if the enum was extended
+    // to a larger type and is signed, because then bit-masks of smaller types
+    // that get extended will fall out of range (e.g. ~0x1u). We currently don't
+    // detect that case and will get a false positive for it. In most cases,
+    // though, it can be fixed by making it a signed type (e.g. ~0x1), so it may
+    // be fine just to accept this as a warning.
+    ExtVal |= llvm::APInt::getHighBitsSet(Width, Width - Val.getBitWidth());
+    if (!(FlagMask & ~ExtVal))
+      return true;
+  }
+
+  return false;
+}
+
 void Sema::ActOnEnumBody(SourceLocation EnumLoc, SourceLocation LBraceLoc,
                          SourceLocation RBraceLoc, Decl *EnumDeclX,
                          ArrayRef<Decl *> Elements,
@@ -13613,10 +13877,8 @@ void Sema::ActOnEnumBody(SourceLocation EnumLoc, SourceLocation LBraceLoc,
       BestPromotionType = Context.getPromotedIntegerType(BestType);
     else
       BestPromotionType = BestType;
-    // We don't need to set BestWidth, because BestType is going to be the type
-    // of the enumerators, but we do anyway because otherwise some compilers
-    // warn that it might be used uninitialized.
-    BestWidth = CharWidth;
+
+    BestWidth = Context.getIntWidth(BestType);
   }
   else if (NumNegativeBits) {
     // If there is a negative value, figure out the smallest integer type (of
@@ -13681,10 +13943,15 @@ void Sema::ActOnEnumBody(SourceLocation EnumLoc, SourceLocation LBraceLoc,
     }
   }
 
+  FlagEnumAttr *FEAttr = Enum->getAttr<FlagEnumAttr>();
+  if (FEAttr)
+    FEAttr->getFlagBits() = llvm::APInt(BestWidth, 0);
+
   // Loop over all of the enumerator constants, changing their types to match
-  // the type of the enum if needed.
-  for (unsigned i = 0, e = Elements.size(); i != e; ++i) {
-    EnumConstantDecl *ECD = cast_or_null<EnumConstantDecl>(Elements[i]);
+  // the type of the enum if needed. If we have a flag type, we also prepare the
+  // FlagBits cache.
+  for (auto *D : Elements) {
+    auto *ECD = cast_or_null<EnumConstantDecl>(D);
     if (!ECD) continue;  // Already issued a diagnostic.
 
     // Standard C says the enumerators have int type, but we allow, as an
@@ -13714,7 +13981,7 @@ void Sema::ActOnEnumBody(SourceLocation EnumLoc, SourceLocation LBraceLoc,
         // enum-specifier, each enumerator has the type of its
         // enumeration.
         ECD->setType(EnumType);
-      continue;
+      goto flagbits;
     } else {
       NewTy = BestType;
       NewWidth = BestWidth;
@@ -13741,8 +14008,32 @@ void Sema::ActOnEnumBody(SourceLocation EnumLoc, SourceLocation LBraceLoc,
       ECD->setType(EnumType);
     else
       ECD->setType(NewTy);
+
+flagbits:
+    // Check to see if we have a constant with exactly one bit set. Note that x
+    // & (x - 1) will be nonzero if and only if x has more than one bit set.
+    if (FEAttr) {
+      llvm::APInt ExtVal = InitVal.zextOrSelf(BestWidth);
+      if (ExtVal != 0 && !(ExtVal & (ExtVal - 1))) {
+        FEAttr->getFlagBits() |= ExtVal;
+      }
+    }
   }
 
+  if (FEAttr) {
+    for (Decl *D : Elements) {
+      EnumConstantDecl *ECD = cast_or_null<EnumConstantDecl>(D);
+      if (!ECD) continue;  // Already issued a diagnostic.
+
+      llvm::APSInt InitVal = ECD->getInitVal();
+      if (InitVal != 0 && !IsValueInFlagEnum(Enum, InitVal, true))
+        Diag(ECD->getLocation(), diag::warn_flag_enum_constant_out_of_range)
+          << ECD << Enum;
+    }
+  }
+
+
+
   Enum->completeDefinition(BestType, BestPromotionType,
                            NumPositiveBits, NumNegativeBits);
 
@@ -13804,6 +14095,8 @@ DeclResult Sema::ActOnModuleImport(SourceLocation AtLoc,
   if (!Mod)
     return true;
 
+  VisibleModules.setVisible(Mod, ImportLoc);
+
   checkModuleImportContext(*this, Mod, ImportLoc, CurContext);
 
   // FIXME: we should support importing a submodule within a different submodule
@@ -13839,9 +14132,46 @@ DeclResult Sema::ActOnModuleImport(SourceLocation AtLoc,
 void Sema::ActOnModuleInclude(SourceLocation DirectiveLoc, Module *Mod) {
   checkModuleImportContext(*this, Mod, DirectiveLoc, CurContext);
 
-  // FIXME: Should we synthesize an ImportDecl here?
-  getModuleLoader().makeModuleVisible(Mod, Module::AllVisible, DirectiveLoc,
-                                      /*Complain=*/true);
+  // Determine whether we're in the #include buffer for a module. The #includes
+  // in that buffer do not qualify as module imports; they're just an
+  // implementation detail of us building the module.
+  //
+  // FIXME: Should we even get ActOnModuleInclude calls for those?
+  bool IsInModuleIncludes =
+      TUKind == TU_Module &&
+      getSourceManager().isWrittenInMainFile(DirectiveLoc);
+
+  // If this module import was due to an inclusion directive, create an 
+  // implicit import declaration to capture it in the AST.
+  if (!IsInModuleIncludes) {
+    TranslationUnitDecl *TU = getASTContext().getTranslationUnitDecl();
+    ImportDecl *ImportD = ImportDecl::CreateImplicit(getASTContext(), TU,
+                                                     DirectiveLoc, Mod,
+                                                     DirectiveLoc);
+    TU->addDecl(ImportD);
+    Consumer.HandleImplicitImportDecl(ImportD);
+  }
+  
+  getModuleLoader().makeModuleVisible(Mod, Module::AllVisible, DirectiveLoc);
+  VisibleModules.setVisible(Mod, DirectiveLoc);
+}
+
+void Sema::ActOnModuleBegin(SourceLocation DirectiveLoc, Module *Mod) {
+  checkModuleImportContext(*this, Mod, DirectiveLoc, CurContext);
+
+  if (getLangOpts().ModulesLocalVisibility)
+    VisibleModulesStack.push_back(std::move(VisibleModules));
+  VisibleModules.setVisible(Mod, DirectiveLoc);
+}
+
+void Sema::ActOnModuleEnd(SourceLocation DirectiveLoc, Module *Mod) {
+  checkModuleImportContext(*this, Mod, DirectiveLoc, CurContext);
+
+  if (getLangOpts().ModulesLocalVisibility) {
+    VisibleModules = std::move(VisibleModulesStack.back());
+    VisibleModulesStack.pop_back();
+    VisibleModules.setVisible(Mod, DirectiveLoc);
+  }
 }
 
 void Sema::createImplicitModuleImportForErrorRecovery(SourceLocation Loc,
@@ -13858,8 +14188,8 @@ void Sema::createImplicitModuleImportForErrorRecovery(SourceLocation Loc,
   Consumer.HandleImplicitImportDecl(ImportD);
 
   // Make the module visible.
-  getModuleLoader().makeModuleVisible(Mod, Module::AllVisible, Loc,
-                                      /*Complain=*/false);
+  getModuleLoader().makeModuleVisible(Mod, Module::AllVisible, Loc);
+  VisibleModules.setVisible(Mod, Loc);
 }
 
 void Sema::ActOnPragmaRedefineExtname(IdentifierInfo* Name,
@@ -13917,7 +14247,10 @@ Decl *Sema::getObjCDeclContext() const {
 }
 
 AvailabilityResult Sema::getCurContextAvailability() const {
-  const Decl *D = cast<Decl>(getCurObjCLexicalContext());
+  const Decl *D = cast_or_null<Decl>(getCurObjCLexicalContext());
+  if (!D)
+    return AR_Available;
+
   // If we are within an Objective-C method, we should consult
   // both the availability of the method as well as the
   // enclosing class.  If the class is (say) deprecated,
diff --git a/lib/Sema/SemaDeclAttr.cpp b/lib/Sema/SemaDeclAttr.cpp
index dcab25e7fe72..31fe05592f00 100644
--- a/lib/Sema/SemaDeclAttr.cpp
+++ b/lib/Sema/SemaDeclAttr.cpp
@@ -51,6 +51,11 @@ namespace AttributeLangSupport {
 static bool isFunctionOrMethod(const Decl *D) {
   return (D->getFunctionType() != nullptr) || isa<ObjCMethodDecl>(D);
 }
+/// \brief Return true if the given decl has function type (function or
+/// function-typed variable) or an Objective-C method or a block.
+static bool isFunctionOrMethodOrBlock(const Decl *D) {
+  return isFunctionOrMethod(D) || isa<BlockDecl>(D);
+}
 
 /// Return true if the given decl has a declarator that should have
 /// been processed by Sema::GetTypeForDeclarator.
@@ -257,7 +262,7 @@ static bool checkFunctionOrMethodParameterIndex(Sema &S, const Decl *D,
                                                 unsigned AttrArgNum,
                                                 const Expr *IdxExpr,
                                                 uint64_t &Idx) {
-  assert(isFunctionOrMethod(D));
+  assert(isFunctionOrMethodOrBlock(D));
 
   // In C++ the implicit 'this' function parameter also counts.
   // Parameters are counted from one.
@@ -351,13 +356,13 @@ static bool isIntOrBool(Expr *Exp) {
 // Check to see if the type is a smart pointer of some kind.  We assume
 // it's a smart pointer if it defines both operator-> and operator*.
 static bool threadSafetyCheckIsSmartPointer(Sema &S, const RecordType* RT) {
-  DeclContextLookupConstResult Res1 = RT->getDecl()->lookup(
-    S.Context.DeclarationNames.getCXXOperatorName(OO_Star));
+  DeclContextLookupResult Res1 = RT->getDecl()->lookup(
+      S.Context.DeclarationNames.getCXXOperatorName(OO_Star));
   if (Res1.empty())
     return false;
 
-  DeclContextLookupConstResult Res2 = RT->getDecl()->lookup(
-    S.Context.DeclarationNames.getCXXOperatorName(OO_Arrow));
+  DeclContextLookupResult Res2 = RT->getDecl()->lookup(
+      S.Context.DeclarationNames.getCXXOperatorName(OO_Arrow));
   if (Res2.empty())
     return false;
 
@@ -1492,6 +1497,20 @@ static void handleAliasAttr(Sema &S, Decl *D, const AttributeList &Attr) {
     return;
   }
 
+  // Aliases should be on declarations, not definitions.
+  if (const auto *FD = dyn_cast<FunctionDecl>(D)) {
+    if (FD->isThisDeclarationADefinition()) {
+      S.Diag(Attr.getLoc(), diag::err_alias_is_definition) << FD;
+      return;
+    }
+  } else {
+    const auto *VD = cast<VarDecl>(D);
+    if (VD->isThisDeclarationADefinition() && VD->isExternallyVisible()) {
+      S.Diag(Attr.getLoc(), diag::err_alias_is_definition) << VD;
+      return;
+    }
+  }
+
   // FIXME: check if target symbol exists in current file
 
   D->addAttr(::new (S.Context) AliasAttr(Attr.getRange(), S.Context, Str,
@@ -1534,18 +1553,16 @@ static void handleTLSModelAttr(Sema &S, Decl *D,
                           Attr.getAttributeSpellingListIndex()));
 }
 
-static void handleMallocAttr(Sema &S, Decl *D, const AttributeList &Attr) {
-  if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) {
-    QualType RetTy = FD->getReturnType();
-    if (RetTy->isAnyPointerType() || RetTy->isBlockPointerType()) {
-      D->addAttr(::new (S.Context)
-                 MallocAttr(Attr.getRange(), S.Context,
-                            Attr.getAttributeSpellingListIndex()));
-      return;
-    }
+static void handleRestrictAttr(Sema &S, Decl *D, const AttributeList &Attr) {
+  QualType ResultType = getFunctionOrMethodResultType(D);
+  if (ResultType->isAnyPointerType() || ResultType->isBlockPointerType()) {
+    D->addAttr(::new (S.Context) RestrictAttr(
+        Attr.getRange(), S.Context, Attr.getAttributeSpellingListIndex()));
+    return;
   }
 
-  S.Diag(Attr.getLoc(), diag::warn_attribute_malloc_pointer_only);
+  S.Diag(Attr.getLoc(), diag::warn_attribute_return_pointers_only)
+      << Attr.getName() << getFunctionOrMethodResultSourceRange(D);
 }
 
 static void handleCommonAttr(Sema &S, Decl *D, const AttributeList &Attr) {
@@ -1589,7 +1606,7 @@ static void handleAnalyzerNoReturnAttr(Sema &S, Decl *D,
   
   // The checking path for 'noreturn' and 'analyzer_noreturn' are different
   // because 'analyzer_noreturn' does not impact the type.
-  if (!isFunctionOrMethod(D) && !isa<BlockDecl>(D)) {
+  if (!isFunctionOrMethodOrBlock(D)) {
     ValueDecl *VD = dyn_cast<ValueDecl>(D);
     if (!VD || (!VD->getType()->isBlockPointerType() &&
                 !VD->getType()->isFunctionPointerType())) {
@@ -2105,6 +2122,22 @@ static void handleObjCNSObject(Sema &S, Decl *D, const AttributeList &Attr) {
                               Attr.getAttributeSpellingListIndex()));
 }
 
+static void handleObjCIndependentClass(Sema &S, Decl *D, const AttributeList &Attr) {
+  if (TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(D)) {
+    QualType T = TD->getUnderlyingType();
+    if (!T->isObjCObjectPointerType()) {
+      S.Diag(TD->getLocation(), diag::warn_ptr_independentclass_attribute);
+      return;
+    }
+  } else {
+    S.Diag(D->getLocation(), diag::warn_independentclass_attribute);
+    return;
+  }
+  D->addAttr(::new (S.Context)
+             ObjCIndependentClassAttr(Attr.getRange(), S.Context,
+                              Attr.getAttributeSpellingListIndex()));
+}
+
 static void handleBlocksAttr(Sema &S, Decl *D, const AttributeList &Attr) {
   if (!Attr.isArgIdent(0)) {
     S.Diag(Attr.getLoc(), diag::err_attribute_argument_n_type)
@@ -2330,6 +2363,15 @@ SectionAttr *Sema::mergeSectionAttr(Decl *D, SourceRange Range,
                                      AttrSpellingListIndex);
 }
 
+bool Sema::checkSectionName(SourceLocation LiteralLoc, StringRef SecName) {
+  std::string Error = Context.getTargetInfo().isValidSectionSpecifier(SecName);
+  if (!Error.empty()) {
+    Diag(LiteralLoc, diag::err_attribute_section_invalid_for_target) << Error;
+    return false;
+  }
+  return true;
+}
+
 static void handleSectionAttr(Sema &S, Decl *D, const AttributeList &Attr) {
   // Make sure that there is a string literal as the sections's single
   // argument.
@@ -2338,6 +2380,9 @@ static void handleSectionAttr(Sema &S, Decl *D, const AttributeList &Attr) {
   if (!S.checkStringLiteralArgumentAttr(Attr, 0, Str, &LiteralLoc))
     return;
 
+  if (!S.checkSectionName(LiteralLoc, Str))
+    return;
+
   // If the target wants to validate the section specifier, make it happen.
   std::string Error = S.Context.getTargetInfo().isValidSectionSpecifier(Str);
   if (!Error.empty()) {
@@ -2481,6 +2526,8 @@ static FormatAttrKind getFormatAttrKind(StringRef Format) {
     .Cases("scanf", "printf", "printf0", "strfmon", SupportedFormat)
     .Cases("cmn_err", "vcmn_err", "zcmn_err", SupportedFormat)
     .Case("kprintf", SupportedFormat) // OpenBSD.
+    .Case("freebsd_kprintf", SupportedFormat) // FreeBSD.
+    .Case("os_trace", SupportedFormat)
 
     .Cases("gcc_diag", "gcc_cdiag", "gcc_cxxdiag", "gcc_tdiag", IgnoredFormat)
     .Default(InvalidFormat);
@@ -2837,6 +2884,16 @@ static void handleAlignedAttr(Sema &S, Decl *D, const AttributeList &Attr) {
   if (!Attr.isPackExpansion() && S.DiagnoseUnexpandedParameterPack(E))
     return;
 
+  if (E->isValueDependent()) {
+    if (const auto *TND = dyn_cast<TypedefNameDecl>(D)) {
+      if (!TND->getUnderlyingType()->isDependentType()) {
+        S.Diag(Attr.getLoc(), diag::err_alignment_dependent_typedef_name)
+            << E->getSourceRange();
+        return;
+      }
+    }
+  }
+
   S.AddAlignedAttr(Attr.getRange(), D, E, Attr.getAttributeSpellingListIndex(),
                    Attr.isPackExpansion());
 }
@@ -2939,12 +2996,15 @@ void Sema::AddAlignedAttr(SourceRange AttrRange, Decl *D, TypeSourceInfo *TS,
 void Sema::CheckAlignasUnderalignment(Decl *D) {
   assert(D->hasAttrs() && "no attributes on decl");
 
-  QualType Ty;
-  if (ValueDecl *VD = dyn_cast<ValueDecl>(D))
-    Ty = VD->getType();
-  else
-    Ty = Context.getTagDeclType(cast<TagDecl>(D));
-  if (Ty->isDependentType() || Ty->isIncompleteType())
+  QualType UnderlyingTy, DiagTy;
+  if (ValueDecl *VD = dyn_cast<ValueDecl>(D)) {
+    UnderlyingTy = DiagTy = VD->getType();
+  } else {
+    UnderlyingTy = DiagTy = Context.getTagDeclType(cast<TagDecl>(D));
+    if (EnumDecl *ED = dyn_cast<EnumDecl>(D))
+      UnderlyingTy = ED->getIntegerType();
+  }
+  if (DiagTy->isDependentType() || DiagTy->isIncompleteType())
     return;
 
   // C++11 [dcl.align]p5, C11 6.7.5/4:
@@ -2963,10 +3023,10 @@ void Sema::CheckAlignasUnderalignment(Decl *D) {
 
   if (AlignasAttr && Align) {
     CharUnits RequestedAlign = Context.toCharUnitsFromBits(Align);
-    CharUnits NaturalAlign = Context.getTypeAlignInChars(Ty);
+    CharUnits NaturalAlign = Context.getTypeAlignInChars(UnderlyingTy);
     if (NaturalAlign > RequestedAlign)
       Diag(AlignasAttr->getLocation(), diag::err_alignas_underaligned)
-        << Ty << (unsigned)NaturalAlign.getQuantity();
+        << DiagTy << (unsigned)NaturalAlign.getQuantity();
   }
 }
 
@@ -3314,11 +3374,6 @@ static void handleCallConvAttr(Sema &S, Decl *D, const AttributeList &Attr) {
                        Attr.getAttributeSpellingListIndex()));
     return;
   }
-  case AttributeList::AT_PnaclCall:
-    D->addAttr(::new (S.Context)
-               PnaclCallAttr(Attr.getRange(), S.Context,
-                             Attr.getAttributeSpellingListIndex()));
-    return;
   case AttributeList::AT_IntelOclBicc:
     D->addAttr(::new (S.Context)
                IntelOclBiccAttr(Attr.getRange(), S.Context,
@@ -3375,16 +3430,18 @@ bool Sema::CheckCallingConvAttr(const AttributeList &attr, CallingConv &CC,
     Diag(attr.getLoc(), diag::err_invalid_pcs);
     return true;
   }
-  case AttributeList::AT_PnaclCall: CC = CC_PnaclCall; break;
   case AttributeList::AT_IntelOclBicc: CC = CC_IntelOclBicc; break;
   default: llvm_unreachable("unexpected attribute kind");
   }
 
   const TargetInfo &TI = Context.getTargetInfo();
   TargetInfo::CallingConvCheckResult A = TI.checkCallingConvention(CC);
-  if (A == TargetInfo::CCCR_Warning) {
-    Diag(attr.getLoc(), diag::warn_cconv_ignored) << attr.getName();
+  if (A != TargetInfo::CCCR_OK) {
+    if (A == TargetInfo::CCCR_Warning)
+      Diag(attr.getLoc(), diag::warn_cconv_ignored) << attr.getName();
 
+    // This convention is not valid for the target. Use the default function or
+    // method calling convention.
     TargetInfo::CallingConvMethodType MT = TargetInfo::CCMT_Unknown;
     if (FD)
       MT = FD->isCXXInstanceMember() ? TargetInfo::CCMT_Member : 
@@ -3431,20 +3488,63 @@ bool Sema::CheckRegparmAttr(const AttributeList &Attr, unsigned &numParams) {
   return false;
 }
 
+// Checks whether an argument of launch_bounds attribute is acceptable
+// May output an error.
+static bool checkLaunchBoundsArgument(Sema &S, Expr *E,
+                                      const CUDALaunchBoundsAttr &Attr,
+                                      const unsigned Idx) {
+
+  if (S.DiagnoseUnexpandedParameterPack(E))
+    return false;
+
+  // Accept template arguments for now as they depend on something else.
+  // We'll get to check them when they eventually get instantiated.
+  if (E->isValueDependent())
+    return true;
+
+  llvm::APSInt I(64);
+  if (!E->isIntegerConstantExpr(I, S.Context)) {
+    S.Diag(E->getExprLoc(), diag::err_attribute_argument_n_type)
+        << &Attr << Idx << AANT_ArgumentIntegerConstant << E->getSourceRange();
+    return false;
+  }
+  // Make sure we can fit it in 32 bits.
+  if (!I.isIntN(32)) {
+    S.Diag(E->getExprLoc(), diag::err_ice_too_large) << I.toString(10, false)
+                                                     << 32 << /* Unsigned */ 1;
+    return false;
+  }
+  if (I < 0)
+    S.Diag(E->getExprLoc(), diag::warn_attribute_argument_n_negative)
+        << &Attr << Idx << E->getSourceRange();
+
+  return true;
+}
+
+void Sema::AddLaunchBoundsAttr(SourceRange AttrRange, Decl *D, Expr *MaxThreads,
+                               Expr *MinBlocks, unsigned SpellingListIndex) {
+  CUDALaunchBoundsAttr TmpAttr(AttrRange, Context, MaxThreads, MinBlocks,
+                               SpellingListIndex);
+
+  if (!checkLaunchBoundsArgument(*this, MaxThreads, TmpAttr, 0))
+    return;
+
+  if (MinBlocks && !checkLaunchBoundsArgument(*this, MinBlocks, TmpAttr, 1))
+    return;
+
+  D->addAttr(::new (Context) CUDALaunchBoundsAttr(
+      AttrRange, Context, MaxThreads, MinBlocks, SpellingListIndex));
+}
+
 static void handleLaunchBoundsAttr(Sema &S, Decl *D,
                                    const AttributeList &Attr) {
-  uint32_t MaxThreads, MinBlocks = 0;
-  if (!checkUInt32Argument(S, Attr, Attr.getArgAsExpr(0), MaxThreads, 1))
-    return;
-  if (Attr.getNumArgs() > 1 && !checkUInt32Argument(S, Attr,
-                                                    Attr.getArgAsExpr(1),
-                                                    MinBlocks, 2))
+  if (!checkAttributeAtLeastNumArgs(S, Attr, 1) ||
+      !checkAttributeAtMostNumArgs(S, Attr, 2))
     return;
 
-  D->addAttr(::new (S.Context)
-              CUDALaunchBoundsAttr(Attr.getRange(), S.Context,
-                                  MaxThreads, MinBlocks,
-                                  Attr.getAttributeSpellingListIndex()));
+  S.AddLaunchBoundsAttr(Attr.getRange(), D, Attr.getArgAsExpr(0),
+                        Attr.getNumArgs() > 1 ? Attr.getArgAsExpr(1) : nullptr,
+                        Attr.getAttributeSpellingListIndex());
 }
 
 static void handleArgumentWithTypeTagAttr(Sema &S, Decl *D,
@@ -3723,6 +3823,22 @@ static void handleObjCBridgeAttr(Sema &S, Scope *Sc, Decl *D,
     S.Diag(D->getLocStart(), diag::err_objc_attr_not_id) << Attr.getName() << 0;
     return;
   }
+
+  // Typedefs only allow objc_bridge(id) and have some additional checking.
+  if (auto TD = dyn_cast<TypedefNameDecl>(D)) {
+    if (!Parm->Ident->isStr("id")) {
+      S.Diag(Attr.getLoc(), diag::err_objc_attr_typedef_not_id)
+        << Attr.getName();
+      return;
+    }
+
+    // Only allow 'cv void *'.
+    QualType T = TD->getUnderlyingType();
+    if (!T->isVoidPointerType()) {
+      S.Diag(Attr.getLoc(), diag::err_objc_attr_typedef_not_void_pointer);
+      return;
+    }
+  }
   
   D->addAttr(::new (S.Context)
              ObjCBridgeAttr(Attr.getRange(), S.Context, Parm->Ident,
@@ -3769,6 +3885,10 @@ static void handleObjCDesignatedInitializer(Sema &S, Decl *D,
     IFace = CatDecl->getClassInterface();
   else
     IFace = cast<ObjCInterfaceDecl>(D->getDeclContext());
+
+  if (!IFace)
+    return;
+
   IFace->setHasDesignatedInitializers();
   D->addAttr(::new (S.Context)
                   ObjCDesignatedInitializerAttr(Attr.getRange(), S.Context,
@@ -4225,9 +4345,52 @@ static void handleDeprecatedAttr(Sema &S, Decl *D, const AttributeList &Attr) {
       return;
     }
   }
+
+  if (!S.getLangOpts().CPlusPlus14)
+    if (Attr.isCXX11Attribute() &&
+        !(Attr.hasScope() && Attr.getScopeName()->isStr("gnu")))
+      S.Diag(Attr.getLoc(), diag::ext_deprecated_attr_is_a_cxx14_extension);
+
   handleAttrWithMessage<DeprecatedAttr>(S, D, Attr);
 }
 
+static void handleNoSanitizeAttr(Sema &S, Decl *D, const AttributeList &Attr) {
+  if (!checkAttributeAtLeastNumArgs(S, Attr, 1))
+    return;
+
+  std::vector<std::string> Sanitizers;
+
+  for (unsigned I = 0, E = Attr.getNumArgs(); I != E; ++I) {
+    StringRef SanitizerName;
+    SourceLocation LiteralLoc;
+
+    if (!S.checkStringLiteralArgumentAttr(Attr, I, SanitizerName, &LiteralLoc))
+      return;
+
+    if (parseSanitizerValue(SanitizerName, /*AllowGroups=*/true) == 0)
+      S.Diag(LiteralLoc, diag::warn_unknown_sanitizer_ignored) << SanitizerName;
+
+    Sanitizers.push_back(SanitizerName);
+  }
+
+  D->addAttr(::new (S.Context) NoSanitizeAttr(
+      Attr.getRange(), S.Context, Sanitizers.data(), Sanitizers.size(),
+      Attr.getAttributeSpellingListIndex()));
+}
+
+static void handleNoSanitizeSpecificAttr(Sema &S, Decl *D,
+                                         const AttributeList &Attr) {
+  std::string SanitizerName =
+      llvm::StringSwitch<std::string>(Attr.getName()->getName())
+          .Case("no_address_safety_analysis", "address")
+          .Case("no_sanitize_address", "address")
+          .Case("no_sanitize_thread", "thread")
+          .Case("no_sanitize_memory", "memory");
+  D->addAttr(::new (S.Context)
+                 NoSanitizeAttr(Attr.getRange(), S.Context, &SanitizerName, 1,
+                                Attr.getAttributeSpellingListIndex()));
+}
+
 /// Handles semantic checking for features that are common to all attributes,
 /// such as checking whether a parameter was properly specified, or the correct
 /// number of arguments were passed, etc.
@@ -4395,6 +4558,9 @@ static void ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D,
   case AttributeList::AT_OptimizeNone:
     handleOptimizeNoneAttr(S, D, Attr);
     break;
+  case AttributeList::AT_FlagEnum:
+    handleSimpleAttribute<FlagEnumAttr>(S, D, Attr);
+    break;
   case AttributeList::AT_Flatten:
     handleSimpleAttribute<FlattenAttr>(S, D, Attr);
     break;
@@ -4419,8 +4585,8 @@ static void ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D,
   case AttributeList::AT_CUDALaunchBounds:
     handleLaunchBoundsAttr(S, D, Attr);
     break;
-  case AttributeList::AT_Malloc:
-    handleMallocAttr(S, D, Attr);
+  case AttributeList::AT_Restrict:
+    handleRestrictAttr(S, D, Attr);
     break;
   case AttributeList::AT_MayAlias:
     handleSimpleAttribute<MayAliasAttr>(S, D, Attr);
@@ -4608,6 +4774,9 @@ static void ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D,
   case AttributeList::AT_ObjCNSObject:
     handleObjCNSObject(S, D, Attr);
     break;
+  case AttributeList::AT_ObjCIndependentClass:
+    handleObjCIndependentClass(S, D, Attr);
+    break;
   case AttributeList::AT_Blocks:
     handleBlocksAttr(S, D, Attr);
     break;
@@ -4644,7 +4813,6 @@ static void ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D,
   case AttributeList::AT_MSABI:
   case AttributeList::AT_SysVABI:
   case AttributeList::AT_Pcs:
-  case AttributeList::AT_PnaclCall:
   case AttributeList::AT_IntelOclBicc:
     handleCallConvAttr(S, D, Attr);
     break;
@@ -4656,8 +4824,11 @@ static void ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D,
     break;
 
   // Microsoft attributes:
-  case AttributeList::AT_MsStruct:
-    handleSimpleAttribute<MsStructAttr>(S, D, Attr);
+  case AttributeList::AT_MSNoVTable:
+    handleSimpleAttribute<MSNoVTableAttr>(S, D, Attr);
+    break;
+  case AttributeList::AT_MSStruct:
+    handleSimpleAttribute<MSStructAttr>(S, D, Attr);
     break;
   case AttributeList::AT_Uuid:
     handleUuidAttr(S, D, Attr);
@@ -4688,18 +4859,15 @@ static void ProcessDeclAttribute(Sema &S, Scope *scope, Decl *D,
   case AttributeList::AT_ScopedLockable:
     handleSimpleAttribute<ScopedLockableAttr>(S, D, Attr);
     break;
-  case AttributeList::AT_NoSanitizeAddress:
-    handleSimpleAttribute<NoSanitizeAddressAttr>(S, D, Attr);
+  case AttributeList::AT_NoSanitize:
+    handleNoSanitizeAttr(S, D, Attr);
+    break;
+  case AttributeList::AT_NoSanitizeSpecific:
+    handleNoSanitizeSpecificAttr(S, D, Attr);
     break;
   case AttributeList::AT_NoThreadSafetyAnalysis:
     handleSimpleAttribute<NoThreadSafetyAnalysisAttr>(S, D, Attr);
     break;
-  case AttributeList::AT_NoSanitizeThread:
-    handleSimpleAttribute<NoSanitizeThreadAttr>(S, D, Attr);
-    break;
-  case AttributeList::AT_NoSanitizeMemory:
-    handleSimpleAttribute<NoSanitizeMemoryAttr>(S, D, Attr);
-    break;
   case AttributeList::AT_GuardedBy:
     handleGuardedByAttr(S, D, Attr);
     break;
@@ -4962,8 +5130,7 @@ void Sema::ProcessPragmaWeak(Scope *S, Decl *D) {
         ND = FD;
     if (ND) {
       if (IdentifierInfo *Id = ND->getIdentifier()) {
-        llvm::DenseMap<IdentifierInfo*,WeakInfo>::iterator I
-          = WeakUndeclaredIdentifiers.find(Id);
+        auto I = WeakUndeclaredIdentifiers.find(Id);
         if (I != WeakUndeclaredIdentifiers.end()) {
           WeakInfo W = I->second;
           DeclApplyPragmaWeak(S, ND, W);
@@ -5042,7 +5209,8 @@ static bool isDeclDeprecated(Decl *D) {
       return true;
     // A category implicitly has the availability of the interface.
     if (const ObjCCategoryDecl *CatD = dyn_cast<ObjCCategoryDecl>(D))
-      return CatD->getClassInterface()->isDeprecated();
+      if (const ObjCInterfaceDecl *Interface = CatD->getClassInterface())
+        return Interface->isDeprecated();
   } while ((D = cast_or_null<Decl>(D->getDeclContext())));
   return false;
 }
@@ -5053,12 +5221,13 @@ static bool isDeclUnavailable(Decl *D) {
       return true;
     // A category implicitly has the availability of the interface.
     if (const ObjCCategoryDecl *CatD = dyn_cast<ObjCCategoryDecl>(D))
-      return CatD->getClassInterface()->isUnavailable();
+      if (const ObjCInterfaceDecl *Interface = CatD->getClassInterface())
+        return Interface->isUnavailable();
   } while ((D = cast_or_null<Decl>(D->getDeclContext())));
   return false;
 }
 
-static void DoEmitAvailabilityWarning(Sema &S, DelayedDiagnostic::DDKind K,
+static void DoEmitAvailabilityWarning(Sema &S, Sema::AvailabilityDiagnostic K,
                                       Decl *Ctx, const NamedDecl *D,
                                       StringRef Message, SourceLocation Loc,
                                       const ObjCInterfaceDecl *UnknownObjCClass,
@@ -5075,8 +5244,8 @@ static void DoEmitAvailabilityWarning(Sema &S, DelayedDiagnostic::DDKind K,
 
   // Don't warn if our current context is deprecated or unavailable.
   switch (K) {
-  case DelayedDiagnostic::Deprecation:
-    if (isDeclDeprecated(Ctx))
+  case Sema::AD_Deprecation:
+    if (isDeclDeprecated(Ctx) || isDeclUnavailable(Ctx))
       return;
     diag = !ObjCPropertyAccess ? diag::warn_deprecated
                                : diag::warn_property_method_deprecated;
@@ -5086,7 +5255,7 @@ static void DoEmitAvailabilityWarning(Sema &S, DelayedDiagnostic::DDKind K,
     available_here_select_kind = /* deprecated */ 2;
     break;
 
-  case DelayedDiagnostic::Unavailable:
+  case Sema::AD_Unavailable:
     if (isDeclUnavailable(Ctx))
       return;
     diag = !ObjCPropertyAccess ? diag::err_unavailable
@@ -5097,8 +5266,13 @@ static void DoEmitAvailabilityWarning(Sema &S, DelayedDiagnostic::DDKind K,
     available_here_select_kind = /* unavailable */ 0;
     break;
 
-  default:
-    llvm_unreachable("Neither a deprecation or unavailable kind");
+  case Sema::AD_Partial:
+    diag = diag::warn_partial_availability;
+    diag_message = diag::warn_partial_message;
+    diag_fwdclass_message = diag::warn_partial_fwdclass_message;
+    property_note_select = /* partial */ 2;
+    available_here_select_kind = /* partial */ 3;
+    break;
   }
 
   if (!Message.empty()) {
@@ -5118,15 +5292,21 @@ static void DoEmitAvailabilityWarning(Sema &S, DelayedDiagnostic::DDKind K,
 
   S.Diag(D->getLocation(), diag::note_availability_specified_here)
       << D << available_here_select_kind;
+  if (K == Sema::AD_Partial)
+    S.Diag(Loc, diag::note_partial_availability_silence) << D;
 }
 
 static void handleDelayedAvailabilityCheck(Sema &S, DelayedDiagnostic &DD,
                                            Decl *Ctx) {
+  assert(DD.Kind == DelayedDiagnostic::Deprecation ||
+         DD.Kind == DelayedDiagnostic::Unavailable);
+  Sema::AvailabilityDiagnostic AD = DD.Kind == DelayedDiagnostic::Deprecation
+                                        ? Sema::AD_Deprecation
+                                        : Sema::AD_Unavailable;
   DD.Triggered = true;
-  DoEmitAvailabilityWarning(S, (DelayedDiagnostic::DDKind)DD.Kind, Ctx,
-                            DD.getDeprecationDecl(), DD.getDeprecationMessage(),
-                            DD.Loc, DD.getUnknownObjCClass(),
-                            DD.getObjCProperty(), false);
+  DoEmitAvailabilityWarning(
+      S, AD, Ctx, DD.getDeprecationDecl(), DD.getDeprecationMessage(), DD.Loc,
+      DD.getUnknownObjCClass(), DD.getObjCProperty(), false);
 }
 
 void Sema::PopParsingDeclaration(ParsingDeclState state, Decl *decl) {
@@ -5192,7 +5372,7 @@ void Sema::EmitAvailabilityWarning(AvailabilityDiagnostic AD,
                                    const ObjCPropertyDecl  *ObjCProperty,
                                    bool ObjCPropertyAccess) {
   // Delay if we're currently parsing a declaration.
-  if (DelayedDiagnostics.shouldDelayDiagnostics()) {
+  if (DelayedDiagnostics.shouldDelayDiagnostics() && AD != AD_Partial) {
     DelayedDiagnostics.add(DelayedDiagnostic::makeAvailability(
         AD, Loc, D, UnknownObjCClass, ObjCProperty, Message,
         ObjCPropertyAccess));
@@ -5200,16 +5380,6 @@ void Sema::EmitAvailabilityWarning(AvailabilityDiagnostic AD,
   }
 
   Decl *Ctx = cast<Decl>(getCurLexicalContext());
-  DelayedDiagnostic::DDKind K;
-  switch (AD) {
-    case AD_Deprecation:
-      K = DelayedDiagnostic::Deprecation;
-      break;
-    case AD_Unavailable:
-      K = DelayedDiagnostic::Unavailable;
-      break;
-  }
-
-  DoEmitAvailabilityWarning(*this, K, Ctx, D, Message, Loc,
-                            UnknownObjCClass, ObjCProperty, ObjCPropertyAccess);
+  DoEmitAvailabilityWarning(*this, AD, Ctx, D, Message, Loc, UnknownObjCClass,
+                            ObjCProperty, ObjCPropertyAccess);
 }
diff --git a/lib/Sema/SemaDeclCXX.cpp b/lib/Sema/SemaDeclCXX.cpp
index 510738ea8168..b1dfe0e95e2d 100644
--- a/lib/Sema/SemaDeclCXX.cpp
+++ b/lib/Sema/SemaDeclCXX.cpp
@@ -316,8 +316,17 @@ Sema::ActOnParamDefaultArgument(Decl *param, SourceLocation EqualLoc,
   if (DiagnoseUnexpandedParameterPack(DefaultArg, UPPC_DefaultArgument)) {
     Param->setInvalidDecl();
     return;
-  }    
-      
+  }
+
+  // C++11 [dcl.fct.default]p3
+  //   A default argument expression [...] shall not be specified for a
+  //   parameter pack.
+  if (Param->isParameterPack()) {
+    Diag(EqualLoc, diag::err_param_default_argument_on_parameter_pack)
+        << DefaultArg->getSourceRange();
+    return;
+  }
+
   // Check that the default argument is well-formed
   CheckDefaultArgumentVisitor DefaultArgChecker(DefaultArg, this);
   if (DefaultArgChecker.Visit(DefaultArg)) {
@@ -429,6 +438,45 @@ bool Sema::MergeCXXFunctionDecl(FunctionDecl *New, FunctionDecl *Old,
                                 Scope *S) {
   bool Invalid = false;
 
+  // The declaration context corresponding to the scope is the semantic
+  // parent, unless this is a local function declaration, in which case
+  // it is that surrounding function.
+  DeclContext *ScopeDC = New->isLocalExternDecl()
+                             ? New->getLexicalDeclContext()
+                             : New->getDeclContext();
+
+  // Find the previous declaration for the purpose of default arguments.
+  FunctionDecl *PrevForDefaultArgs = Old;
+  for (/**/; PrevForDefaultArgs;
+       // Don't bother looking back past the latest decl if this is a local
+       // extern declaration; nothing else could work.
+       PrevForDefaultArgs = New->isLocalExternDecl()
+                                ? nullptr
+                                : PrevForDefaultArgs->getPreviousDecl()) {
+    // Ignore hidden declarations.
+    if (!LookupResult::isVisible(*this, PrevForDefaultArgs))
+      continue;
+
+    if (S && !isDeclInScope(PrevForDefaultArgs, ScopeDC, S) &&
+        !New->isCXXClassMember()) {
+      // Ignore default arguments of old decl if they are not in
+      // the same scope and this is not an out-of-line definition of
+      // a member function.
+      continue;
+    }
+
+    if (PrevForDefaultArgs->isLocalExternDecl() != New->isLocalExternDecl()) {
+      // If only one of these is a local function declaration, then they are
+      // declared in different scopes, even though isDeclInScope may think
+      // they're in the same scope. (If both are local, the scope check is
+      // sufficent, and if neither is local, then they are in the same scope.)
+      continue;
+    }
+
+    // We found our guy.
+    break;
+  }
+
   // C++ [dcl.fct.default]p4:
   //   For non-template functions, default arguments can be added in
   //   later declarations of a function in the same
@@ -447,34 +495,17 @@ bool Sema::MergeCXXFunctionDecl(FunctionDecl *New, FunctionDecl *Old,
   //   in a member function definition that appears outside of the class
   //   definition are added to the set of default arguments provided by the
   //   member function declaration in the class definition.
-  for (unsigned p = 0, NumParams = Old->getNumParams(); p < NumParams; ++p) {
-    ParmVarDecl *OldParam = Old->getParamDecl(p);
+  for (unsigned p = 0, NumParams = PrevForDefaultArgs
+                                       ? PrevForDefaultArgs->getNumParams()
+                                       : 0;
+       p < NumParams; ++p) {
+    ParmVarDecl *OldParam = PrevForDefaultArgs->getParamDecl(p);
     ParmVarDecl *NewParam = New->getParamDecl(p);
 
-    bool OldParamHasDfl = OldParam->hasDefaultArg();
+    bool OldParamHasDfl = OldParam ? OldParam->hasDefaultArg() : false;
     bool NewParamHasDfl = NewParam->hasDefaultArg();
 
-    // The declaration context corresponding to the scope is the semantic
-    // parent, unless this is a local function declaration, in which case
-    // it is that surrounding function.
-    DeclContext *ScopeDC = New->isLocalExternDecl()
-                               ? New->getLexicalDeclContext()
-                               : New->getDeclContext();
-    if (S && !isDeclInScope(Old, ScopeDC, S) &&
-        !New->getDeclContext()->isRecord())
-      // Ignore default parameters of old decl if they are not in
-      // the same scope and this is not an out-of-line definition of
-      // a member function.
-      OldParamHasDfl = false;
-    if (New->isLocalExternDecl() != Old->isLocalExternDecl())
-      // If only one of these is a local function declaration, then they are
-      // declared in different scopes, even though isDeclInScope may think
-      // they're in the same scope. (If both are local, the scope check is
-      // sufficent, and if neither is local, then they are in the same scope.)
-      OldParamHasDfl = false;
-
     if (OldParamHasDfl && NewParamHasDfl) {
-
       unsigned DiagDefaultParamID =
         diag::err_param_default_argument_redefinition;
 
@@ -482,7 +513,7 @@ bool Sema::MergeCXXFunctionDecl(FunctionDecl *New, FunctionDecl *Old,
       // of template class. The new default parameter's value is ignored.
       Invalid = true;
       if (getLangOpts().MicrosoftExt) {
-        CXXMethodDecl* MD = dyn_cast<CXXMethodDecl>(New);
+        CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(New);
         if (MD && MD->getParent()->getDescribedClassTemplate()) {
           // Merge the old default argument into the new parameter.
           NewParam->setHasInheritedDefaultArg();
@@ -509,14 +540,12 @@ bool Sema::MergeCXXFunctionDecl(FunctionDecl *New, FunctionDecl *Old,
       
       // Look for the function declaration where the default argument was
       // actually written, which may be a declaration prior to Old.
-      for (FunctionDecl *Older = Old->getPreviousDecl();
-           Older; Older = Older->getPreviousDecl()) {
-        if (!Older->getParamDecl(p)->hasDefaultArg())
-          break;
-        
+      for (auto Older = PrevForDefaultArgs;
+           OldParam->hasInheritedDefaultArg(); /**/) {
+        Older = Older->getPreviousDecl();
         OldParam = Older->getParamDecl(p);
-      }        
-      
+      }
+
       Diag(OldParam->getLocation(), diag::note_previous_definition)
         << OldParam->getDefaultArgRange();
     } else if (OldParamHasDfl) {
@@ -524,7 +553,9 @@ bool Sema::MergeCXXFunctionDecl(FunctionDecl *New, FunctionDecl *Old,
       // It's important to use getInit() here;  getDefaultArg()
       // strips off any top-level ExprWithCleanups.
       NewParam->setHasInheritedDefaultArg();
-      if (OldParam->hasUninstantiatedDefaultArg())
+      if (OldParam->hasUnparsedDefaultArg())
+        NewParam->setUnparsedDefaultArg();
+      else if (OldParam->hasUninstantiatedDefaultArg())
         NewParam->setUninstantiatedDefaultArg(
                                       OldParam->getUninstantiatedDefaultArg());
       else
@@ -535,8 +566,9 @@ bool Sema::MergeCXXFunctionDecl(FunctionDecl *New, FunctionDecl *Old,
         Diag(NewParam->getLocation(),
              diag::err_param_default_argument_template_redecl)
           << NewParam->getDefaultArgRange();
-        Diag(Old->getLocation(), diag::note_template_prev_declaration)
-          << false;
+        Diag(PrevForDefaultArgs->getLocation(),
+             diag::note_template_prev_declaration)
+            << false;
       } else if (New->getTemplateSpecializationKind()
                    != TSK_ImplicitInstantiation &&
                  New->getTemplateSpecializationKind() != TSK_Undeclared) {
@@ -607,7 +639,8 @@ bool Sema::MergeCXXFunctionDecl(FunctionDecl *New, FunctionDecl *Old,
       << New << New->isConstexpr();
     Diag(Old->getLocation(), diag::note_previous_declaration);
     Invalid = true;
-  } else if (!Old->isInlined() && New->isInlined() && Old->isDefined(Def)) {
+  } else if (!Old->getMostRecentDecl()->isInlined() && New->isInlined() &&
+             Old->isDefined(Def)) {
     // C++11 [dcl.fcn.spec]p4:
     //   If the definition of a function appears in a translation unit before its
     //   first declaration as inline, the program is ill-formed.
@@ -689,16 +722,16 @@ void Sema::CheckCXXDefaultArguments(FunctionDecl *FD) {
       break;
   }
 
-  // C++ [dcl.fct.default]p4:
-  //   In a given function declaration, all parameters
-  //   subsequent to a parameter with a default argument shall
-  //   have default arguments supplied in this or previous
-  //   declarations. A default argument shall not be redefined
-  //   by a later declaration (not even to the same value).
+  // C++11 [dcl.fct.default]p4:
+  //   In a given function declaration, each parameter subsequent to a parameter
+  //   with a default argument shall have a default argument supplied in this or
+  //   a previous declaration or shall be a function parameter pack. A default
+  //   argument shall not be redefined by a later declaration (not even to the
+  //   same value).
   unsigned LastMissingDefaultArg = 0;
   for (; p < NumParams; ++p) {
     ParmVarDecl *Param = FD->getParamDecl(p);
-    if (!Param->hasDefaultArg()) {
+    if (!Param->hasDefaultArg() && !Param->isParameterPack()) {
       if (Param->isInvalidDecl())
         /* We already complained about this parameter. */;
       else if (Param->getIdentifier())
@@ -795,7 +828,8 @@ bool Sema::CheckConstexprFunctionDecl(const FunctionDecl *NewFD) {
     // - it shall not be virtual;
     const CXXMethodDecl *Method = dyn_cast<CXXMethodDecl>(NewFD);
     if (Method && Method->isVirtual()) {
-      Diag(NewFD->getLocation(), diag::err_constexpr_virtual);
+      Method = Method->getCanonicalDecl();
+      Diag(Method->getLocation(), diag::err_constexpr_virtual);
 
       // If it's not obvious why this function is virtual, find an overridden
       // function which uses the 'virtual' keyword.
@@ -1545,6 +1579,31 @@ Sema::ActOnBaseSpecifier(Decl *classdecl, SourceRange SpecifierRange,
   return true;
 }
 
+/// Use small set to collect indirect bases.  As this is only used
+/// locally, there's no need to abstract the small size parameter.
+typedef llvm::SmallPtrSet<QualType, 4> IndirectBaseSet;
+
+/// \brief Recursively add the bases of Type.  Don't add Type itself.
+static void
+NoteIndirectBases(ASTContext &Context, IndirectBaseSet &Set,
+                  const QualType &Type)
+{
+  // Even though the incoming type is a base, it might not be
+  // a class -- it could be a template parm, for instance.
+  if (auto Rec = Type->getAs<RecordType>()) {
+    auto Decl = Rec->getAsCXXRecordDecl();
+
+    // Iterate over its bases.
+    for (const auto &BaseSpec : Decl->bases()) {
+      QualType Base = Context.getCanonicalType(BaseSpec.getType())
+        .getUnqualifiedType();
+      if (Set.insert(Base).second)
+        // If we've not already seen it, recurse.
+        NoteIndirectBases(Context, Set, Base);
+    }
+  }
+}
+
 /// \brief Performs the actual work of attaching the given base class
 /// specifiers to a C++ class.
 bool Sema::AttachBaseSpecifiers(CXXRecordDecl *Class, CXXBaseSpecifier **Bases,
@@ -1558,6 +1617,10 @@ bool Sema::AttachBaseSpecifiers(CXXRecordDecl *Class, CXXBaseSpecifier **Bases,
   // class.
   std::map<QualType, CXXBaseSpecifier*, QualTypeOrdering> KnownBaseTypes;
 
+  // Used to track indirect bases so we can see if a direct base is
+  // ambiguous.
+  IndirectBaseSet IndirectBaseTypes;
+
   // Copy non-redundant base specifiers into permanent storage.
   unsigned NumGoodBases = 0;
   bool Invalid = false;
@@ -1585,6 +1648,11 @@ bool Sema::AttachBaseSpecifiers(CXXRecordDecl *Class, CXXBaseSpecifier **Bases,
       // Okay, add this new base class.
       KnownBase = Bases[idx];
       Bases[NumGoodBases++] = Bases[idx];
+
+      // Note this base's direct & indirect bases, if there could be ambiguity.
+      if (NumBases > 1)
+        NoteIndirectBases(Context, IndirectBaseTypes, NewBaseType);
+      
       if (const RecordType *Record = NewBaseType->getAs<RecordType>()) {
         const CXXRecordDecl *RD = cast<CXXRecordDecl>(Record->getDecl());
         if (Class->isInterface() &&
@@ -1605,11 +1673,33 @@ bool Sema::AttachBaseSpecifiers(CXXRecordDecl *Class, CXXBaseSpecifier **Bases,
 
   // Attach the remaining base class specifiers to the derived class.
   Class->setBases(Bases, NumGoodBases);
+  
+  for (unsigned idx = 0; idx < NumGoodBases; ++idx) {
+    // Check whether this direct base is inaccessible due to ambiguity.
+    QualType BaseType = Bases[idx]->getType();
+    CanQualType CanonicalBase = Context.getCanonicalType(BaseType)
+      .getUnqualifiedType();
+
+    if (IndirectBaseTypes.count(CanonicalBase)) {
+      CXXBasePaths Paths(/*FindAmbiguities=*/true, /*RecordPaths=*/true,
+                         /*DetectVirtual=*/true);
+      bool found
+        = Class->isDerivedFrom(CanonicalBase->getAsCXXRecordDecl(), Paths);
+      assert(found);
+      (void)found;
+
+      if (Paths.isAmbiguous(CanonicalBase))
+        Diag(Bases[idx]->getLocStart (), diag::warn_inaccessible_base_class)
+          << BaseType << getAmbiguousPathsDisplayString(Paths)
+          << Bases[idx]->getSourceRange();
+      else
+        assert(Bases[idx]->isVirtual());
+    }
 
-  // Delete the remaining (good) base class specifiers, since their
-  // data has been copied into the CXXRecordDecl.
-  for (unsigned idx = 0; idx < NumGoodBases; ++idx)
+    // Delete the base class specifier, since its data has been copied
+    // into the CXXRecordDecl.
     Context.Deallocate(Bases[idx]);
+  }
 
   return Invalid;
 }
@@ -1689,18 +1779,6 @@ void Sema::BuildBasePathArray(const CXXBasePaths &Paths,
     BasePathArray.push_back(const_cast<CXXBaseSpecifier*>(Path[I].Base));
 }
 
-/// \brief Determine whether the given base path includes a virtual
-/// base class.
-bool Sema::BasePathInvolvesVirtualBase(const CXXCastPath &BasePath) {
-  for (CXXCastPath::const_iterator B = BasePath.begin(), 
-                                BEnd = BasePath.end();
-       B != BEnd; ++B)
-    if ((*B)->isVirtual())
-      return true;
-
-  return false;
-}
-
 /// CheckDerivedToBaseConversion - Check whether the Derived-to-Base
 /// conversion (where Derived and Base are class types) is
 /// well-formed, meaning that the conversion is unambiguous (and
@@ -2159,7 +2237,8 @@ Sema::ActOnCXXMemberDeclarator(Scope *S, AccessSpecifier AS, Declarator &D,
       assert(Member && "HandleField never returns null");
     }
   } else {
-    assert(InitStyle == ICIS_NoInit || D.getDeclSpec().getStorageClassSpec() == DeclSpec::SCS_static);
+    assert(InitStyle == ICIS_NoInit ||
+           D.getDeclSpec().getStorageClassSpec() == DeclSpec::SCS_static);
 
     Member = HandleDeclarator(S, D, TemplateParameterLists);
     if (!Member)
@@ -3537,8 +3616,9 @@ BuildImplicitMemberInitializer(Sema &SemaRef, CXXConstructorDecl *Constructor,
       InitializationKind::CreateDirect(Loc, SourceLocation(), SourceLocation());
     
     Expr *CtorArgE = CtorArg.getAs<Expr>();
-    InitializationSequence InitSeq(SemaRef, Entities.back(), InitKind, CtorArgE);
-    
+    InitializationSequence InitSeq(SemaRef, Entities.back(), InitKind,
+                                   CtorArgE);
+
     ExprResult MemberInit
       = InitSeq.Perform(SemaRef, Entities.back(), InitKind, 
                         MultiExprArg(&CtorArgE, 1));
@@ -4659,15 +4739,15 @@ static void CheckAbstractClassUsage(AbstractUsageInfo &Info,
 }
 
 /// \brief Check class-level dllimport/dllexport attribute.
-static void checkDLLAttribute(Sema &S, CXXRecordDecl *Class) {
+void Sema::checkClassLevelDLLAttribute(CXXRecordDecl *Class) {
   Attr *ClassAttr = getDLLAttr(Class);
 
   // MSVC inherits DLL attributes to partial class template specializations.
-  if (S.Context.getTargetInfo().getCXXABI().isMicrosoft() && !ClassAttr) {
+  if (Context.getTargetInfo().getCXXABI().isMicrosoft() && !ClassAttr) {
     if (auto *Spec = dyn_cast<ClassTemplatePartialSpecializationDecl>(Class)) {
       if (Attr *TemplateAttr =
               getDLLAttr(Spec->getSpecializedTemplate()->getTemplatedDecl())) {
-        auto *A = cast<InheritableAttr>(TemplateAttr->clone(S.getASTContext()));
+        auto *A = cast<InheritableAttr>(TemplateAttr->clone(getASTContext()));
         A->setInherited(true);
         ClassAttr = A;
       }
@@ -4678,12 +4758,12 @@ static void checkDLLAttribute(Sema &S, CXXRecordDecl *Class) {
     return;
 
   if (!Class->isExternallyVisible()) {
-    S.Diag(Class->getLocation(), diag::err_attribute_dll_not_extern)
+    Diag(Class->getLocation(), diag::err_attribute_dll_not_extern)
         << Class << ClassAttr;
     return;
   }
 
-  if (S.Context.getTargetInfo().getCXXABI().isMicrosoft() &&
+  if (Context.getTargetInfo().getCXXABI().isMicrosoft() &&
       !ClassAttr->isInherited()) {
     // Diagnose dll attributes on members of class with dll attribute.
     for (Decl *Member : Class->decls()) {
@@ -4693,10 +4773,10 @@ static void checkDLLAttribute(Sema &S, CXXRecordDecl *Class) {
       if (!MemberAttr || MemberAttr->isInherited() || Member->isInvalidDecl())
         continue;
 
-      S.Diag(MemberAttr->getLocation(),
+      Diag(MemberAttr->getLocation(),
              diag::err_attribute_dll_member_of_dll_class)
           << MemberAttr << ClassAttr;
-      S.Diag(ClassAttr->getLocation(), diag::note_previous_attribute);
+      Diag(ClassAttr->getLocation(), diag::note_previous_attribute);
       Member->setInvalidDecl();
     }
   }
@@ -4709,15 +4789,20 @@ static void checkDLLAttribute(Sema &S, CXXRecordDecl *Class) {
   const bool ClassExported = ClassAttr->getKind() == attr::DLLExport;
   const bool ClassImported = !ClassExported;
 
+  TemplateSpecializationKind TSK = Class->getTemplateSpecializationKind();
+
+  // Don't dllexport explicit class template instantiation declarations.
+  if (ClassExported && TSK == TSK_ExplicitInstantiationDeclaration) {
+    Class->dropAttr<DLLExportAttr>();
+    return;
+  }
+
   // Force declaration of implicit members so they can inherit the attribute.
-  S.ForceDeclarationOfImplicitMembers(Class);
+  ForceDeclarationOfImplicitMembers(Class);
 
   // FIXME: MSVC's docs say all bases must be exportable, but this doesn't
   // seem to be true in practice?
 
-  TemplateSpecializationKind TSK =
-    Class->getTemplateSpecializationKind();
-
   for (Decl *Member : Class->decls()) {
     VarDecl *VD = dyn_cast<VarDecl>(Member);
     CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(Member);
@@ -4731,22 +4816,25 @@ static void checkDLLAttribute(Sema &S, CXXRecordDecl *Class) {
       if (MD->isDeleted())
         continue;
 
-      if (MD->isMoveAssignmentOperator() && ClassImported && MD->isInlined()) {
-        // Current MSVC versions don't export the move assignment operators, so
-        // don't attempt to import them if we have a definition.
-        continue;
-      }
+      if (MD->isInlined()) {
+        // MinGW does not import or export inline methods.
+        if (!Context.getTargetInfo().getCXXABI().isMicrosoft())
+          continue;
 
-      if (MD->isInlined() && ClassImported &&
-          !S.Context.getTargetInfo().getCXXABI().isMicrosoft()) {
-        // MinGW does not import inline functions.
-        continue;
+        // MSVC versions before 2015 don't export the move assignment operators,
+        // so don't attempt to import them if we have a definition.
+        if (ClassImported && MD->isMoveAssignmentOperator() &&
+            !getLangOpts().isCompatibleWithMSVC(LangOptions::MSVC2015))
+          continue;
       }
     }
 
+    if (!cast<NamedDecl>(Member)->isExternallyVisible())
+      continue;
+
     if (!getDLLAttr(Member)) {
       auto *NewAttr =
-          cast<InheritableAttr>(ClassAttr->clone(S.getASTContext()));
+          cast<InheritableAttr>(ClassAttr->clone(getASTContext()));
       NewAttr->setInherited(true);
       Member->addAttr(NewAttr);
     }
@@ -4761,7 +4849,7 @@ static void checkDLLAttribute(Sema &S, CXXRecordDecl *Class) {
         if (TSK == TSK_ImplicitInstantiation && !ClassAttr->isInherited())
           continue;
 
-        S.MarkFunctionReferenced(Class->getLocation(), MD);
+        MarkFunctionReferenced(Class->getLocation(), MD);
 
         // The function will be passed to the consumer when its definition is
         // encountered.
@@ -4772,11 +4860,17 @@ static void checkDLLAttribute(Sema &S, CXXRecordDecl *Class) {
         // defaulted methods, and the copy and move assignment operators. The
         // latter are exported even if they are trivial, because the address of
         // an operator can be taken and should compare equal accross libraries.
-        S.MarkFunctionReferenced(Class->getLocation(), MD);
+        DiagnosticErrorTrap Trap(Diags);
+        MarkFunctionReferenced(Class->getLocation(), MD);
+        if (Trap.hasErrorOccurred()) {
+          Diag(ClassAttr->getLocation(), diag::note_due_to_dllexported_class)
+              << Class->getName() << !getLangOpts().CPlusPlus11;
+          break;
+        }
 
         // There is no later point when we will see the definition of this
         // function, so pass it to the consumer now.
-        S.Consumer.HandleTopLevelDecl(DeclGroupRef(MD));
+        Consumer.HandleTopLevelDecl(DeclGroupRef(MD));
       }
     }
   }
@@ -4820,9 +4914,6 @@ void Sema::CheckCompletedCXXClass(CXXRecordDecl *Record) {
     }
   }
 
-  if (Record->isDynamicClass() && !Record->isDependentType())
-    DynamicClasses.push_back(Record);
-
   if (Record->getIdentifier()) {
     // C++ [class.mem]p13:
     //   If T is the name of a class, then each of the following shall have a 
@@ -4923,7 +5014,7 @@ void Sema::CheckCompletedCXXClass(CXXRecordDecl *Record) {
   //   have inheriting constructors.
   DeclareInheritingConstructors(Record);
 
-  checkDLLAttribute(*this, Record);
+  checkClassLevelDLLAttribute(Record);
 }
 
 /// Look up the special member function that would be called by a special
@@ -7318,7 +7409,7 @@ bool Sema::isStdInitializerList(QualType Ty, QualType *Element) {
     StdInitializerList = Template;
   }
 
-  if (Template != StdInitializerList)
+  if (Template->getCanonicalDecl() != StdInitializerList->getCanonicalDecl())
     return false;
 
   // This is an instance of std::initializer_list. Find the argument type.
@@ -8029,15 +8120,7 @@ NamedDecl *Sema::BuildUsingDeclaration(Scope *S, AccessSpecifier AS,
   if (RequireCompleteDeclContext(SS, LookupContext))
     return BuildInvalid();
 
-  // The normal rules do not apply to inheriting constructor declarations.
-  if (NameInfo.getName().getNameKind() == DeclarationName::CXXConstructorName) {
-    UsingDecl *UD = BuildValid();
-    CheckInheritingConstructorUsingDecl(UD);
-    return UD;
-  }
-
-  // Otherwise, look up the target name.
-
+  // Look up the target name.
   LookupResult R(*this, NameInfo, LookupOrdinaryName);
 
   // Unlike most lookups, we don't always want to hide tag
@@ -8056,8 +8139,12 @@ NamedDecl *Sema::BuildUsingDeclaration(Scope *S, AccessSpecifier AS,
 
   LookupQualifiedName(R, LookupContext);
 
-  // Try to correct typos if possible.
-  if (R.empty()) {
+  // Try to correct typos if possible. If constructor name lookup finds no
+  // results, that means the named class has no explicit constructors, and we
+  // suppressed declaring implicit ones (probably because it's dependent or
+  // invalid).
+  if (R.empty() &&
+      NameInfo.getName().getNameKind() != DeclarationName::CXXConstructorName) {
     if (TypoCorrection Corrected = CorrectTypo(
             R.getLookupNameInfo(), R.getLookupKind(), S, &SS,
             llvm::make_unique<UsingValidatorCCC>(
@@ -8087,16 +8174,12 @@ NamedDecl *Sema::BuildUsingDeclaration(Scope *S, AccessSpecifier AS,
         NameInfo.setName(Context.DeclarationNames.getCXXConstructorName(
             Context.getCanonicalType(Context.getRecordType(RD))));
         NameInfo.setNamedTypeInfo(nullptr);
-
-        // Build it and process it as an inheriting constructor.
-        UsingDecl *UD = BuildValid();
-        CheckInheritingConstructorUsingDecl(UD);
-        return UD;
+        for (auto *Ctor : LookupConstructors(RD))
+          R.addDecl(Ctor);
+      } else {
+        // FIXME: Pick up all the declarations if we found an overloaded function.
+        R.addDecl(ND);
       }
-
-      // FIXME: Pick up all the declarations if we found an overloaded function.
-      R.setLookupName(Corrected.getCorrection());
-      R.addDecl(ND);
     } else {
       Diag(IdentLoc, diag::err_no_member)
         << NameInfo.getName() << LookupContext << SS.getRange();
@@ -8136,6 +8219,18 @@ NamedDecl *Sema::BuildUsingDeclaration(Scope *S, AccessSpecifier AS,
   }
 
   UsingDecl *UD = BuildValid();
+
+  // The normal rules do not apply to inheriting constructor declarations.
+  if (NameInfo.getName().getNameKind() == DeclarationName::CXXConstructorName) {
+    // Suppress access diagnostics; the access check is instead performed at the
+    // point of use for an inheriting constructor.
+    R.suppressDiagnostics();
+    CheckInheritingConstructorUsingDecl(UD);
+    return UD;
+  }
+
+  // Otherwise, look up the target name.
+
   for (LookupResult::iterator I = R.begin(), E = R.end(); I != E; ++I) {
     UsingShadowDecl *PrevDecl = nullptr;
     if (!CheckUsingShadowDecl(UD, *I, Previous, PrevDecl))
@@ -8421,7 +8516,8 @@ Decl *Sema::ActOnAliasDeclaration(Scope *S,
                                   SourceLocation UsingLoc,
                                   UnqualifiedId &Name,
                                   AttributeList *AttrList,
-                                  TypeResult Type) {
+                                  TypeResult Type,
+                                  Decl *DeclFromDeclSpec) {
   // Skip up to the relevant declaration scope.
   while (S->getFlags() & Scope::TemplateParamScope)
     S = S->getParent();
@@ -8549,6 +8645,10 @@ Decl *Sema::ActOnAliasDeclaration(Scope *S,
 
     NewND = NewDecl;
   } else {
+    if (auto *TD = dyn_cast_or_null<TagDecl>(DeclFromDeclSpec)) {
+      setTagNameForLinkagePurposes(TD, NewTD);
+      handleTagNumbering(TD, S);
+    }
     ActOnTypedefNameDecl(S, CurContext, NewTD, Previous, Redeclaration);
     NewND = NewTD;
   }
@@ -9014,7 +9114,7 @@ private:
     ASTContext &Context = SemaRef.Context;
     DeclarationName Name = Context.DeclarationNames.getCXXConstructorName(
         Context.getCanonicalType(Context.getRecordType(Base)));
-    DeclContext::lookup_const_result Decls = Derived->lookup(Name);
+    DeclContext::lookup_result Decls = Derived->lookup(Name);
     return Decls.empty() ? Derived->getLocation() : Decls[0]->getLocation();
   }
 
@@ -9363,6 +9463,44 @@ void Sema::ActOnFinishCXXMemberDecls() {
   }
 }
 
+static void getDefaultArgExprsForConstructors(Sema &S, CXXRecordDecl *Class) {
+  // Don't do anything for template patterns.
+  if (Class->getDescribedClassTemplate())
+    return;
+
+  for (Decl *Member : Class->decls()) {
+    auto *CD = dyn_cast<CXXConstructorDecl>(Member);
+    if (!CD) {
+      // Recurse on nested classes.
+      if (auto *NestedRD = dyn_cast<CXXRecordDecl>(Member))
+        getDefaultArgExprsForConstructors(S, NestedRD);
+      continue;
+    } else if (!CD->isDefaultConstructor() || !CD->hasAttr<DLLExportAttr>()) {
+      continue;
+    }
+
+    for (unsigned I = 0, E = CD->getNumParams(); I != E; ++I) {
+      // Skip any default arguments that we've already instantiated.
+      if (S.Context.getDefaultArgExprForConstructor(CD, I))
+        continue;
+
+      Expr *DefaultArg = S.BuildCXXDefaultArgExpr(Class->getLocation(), CD,
+                                                  CD->getParamDecl(I)).get();
+      S.Context.addDefaultArgExprForConstructor(CD, I, DefaultArg);
+    }
+  }
+}
+
+void Sema::ActOnFinishCXXMemberDefaultArgs(Decl *D) {
+  auto *RD = dyn_cast<CXXRecordDecl>(D);
+
+  // Default constructors that are annotated with __declspec(dllexport) which
+  // have default arguments or don't use the standard calling convention are
+  // wrapped with a thunk called the default constructor closure.
+  if (RD && Context.getTargetInfo().getCXXABI().isMicrosoft())
+    getDefaultArgExprsForConstructors(*this, RD);
+}
+
 void Sema::AdjustDestructorExceptionSpec(CXXRecordDecl *ClassDecl,
                                          CXXDestructorDecl *Destructor) {
   assert(getLangOpts().CPlusPlus11 &&
@@ -9397,8 +9535,8 @@ namespace {
 //  copy/move operators. These classes serve as factory functions and help us
 //  avoid using the same Expr* in the AST twice.
 class ExprBuilder {
-  ExprBuilder(const ExprBuilder&) LLVM_DELETED_FUNCTION;
-  ExprBuilder &operator=(const ExprBuilder&) LLVM_DELETED_FUNCTION;
+  ExprBuilder(const ExprBuilder&) = delete;
+  ExprBuilder &operator=(const ExprBuilder&) = delete;
 
 protected:
   static Expr *assertNotNull(Expr *E) {
@@ -10101,7 +10239,9 @@ void Sema::DefineImplicitCopyAssignment(SourceLocation CurrentLocation,
   
   // Assign non-static members.
   for (auto *Field : ClassDecl->fields()) {
-    if (Field->isUnnamedBitfield())
+    // FIXME: We should form some kind of AST representation for the implied
+    // memcpy in a union copy operation.
+    if (Field->isUnnamedBitfield() || Field->getParent()->isUnion())
       continue;
 
     if (Field->isInvalidDecl()) {
@@ -10531,7 +10671,9 @@ void Sema::DefineImplicitMoveAssignment(SourceLocation CurrentLocation,
 
   // Assign non-static members.
   for (auto *Field : ClassDecl->fields()) {
-    if (Field->isUnnamedBitfield())
+    // FIXME: We should form some kind of AST representation for the implied
+    // memcpy in a union copy operation.
+    if (Field->isUnnamedBitfield() || Field->getParent()->isUnion())
       continue;
 
     if (Field->isInvalidDecl()) {
@@ -11881,7 +12023,7 @@ VarDecl *Sema::BuildExceptionDeclaration(Scope *S,
       //
       // We just pretend to initialize the object with itself, then make sure
       // it can be destroyed later.
-      QualType initType = ExDeclType;
+      QualType initType = Context.getExceptionObjectType(ExDeclType);
 
       InitializedEntity entity =
         InitializedEntity::InitializeVariable(ExDecl);
@@ -13122,7 +13264,8 @@ bool Sema::DefineUsedVTables() {
     DefinedAnything = true;
     MarkVirtualMembersReferenced(Loc, Class);
     CXXRecordDecl *Canonical = cast<CXXRecordDecl>(Class->getCanonicalDecl());
-    Consumer.HandleVTable(Class, VTablesUsed[Canonical]);
+    if (VTablesUsed[Canonical])
+      Consumer.HandleVTable(Class);
 
     // Optionally warn if we're emitting a weak vtable.
     if (Class->isExternallyVisible() &&
diff --git a/lib/Sema/SemaDeclObjC.cpp b/lib/Sema/SemaDeclObjC.cpp
index 7e3da941b339..38318791fd77 100644
--- a/lib/Sema/SemaDeclObjC.cpp
+++ b/lib/Sema/SemaDeclObjC.cpp
@@ -448,6 +448,19 @@ class ObjCInterfaceValidatorCCC : public CorrectionCandidateCallback {
 
 }
 
+static void diagnoseUseOfProtocols(Sema &TheSema,
+                                   ObjCContainerDecl *CD,
+                                   ObjCProtocolDecl *const *ProtoRefs,
+                                   unsigned NumProtoRefs,
+                                   const SourceLocation *ProtoLocs) {
+  assert(ProtoRefs);
+  // Diagnose availability in the context of the ObjC container.
+  Sema::ContextRAII SavedContext(TheSema, CD);
+  for (unsigned i = 0; i < NumProtoRefs; ++i) {
+    (void)TheSema.DiagnoseUseOfDecl(ProtoRefs[i], ProtoLocs[i]);
+  }
+}
+
 Decl *Sema::
 ActOnStartClassInterface(SourceLocation AtInterfaceLoc,
                          IdentifierInfo *ClassName, SourceLocation ClassLoc,
@@ -535,6 +548,8 @@ ActOnStartClassInterface(SourceLocation AtInterfaceLoc,
       ObjCInterfaceDecl *SuperClassDecl =
                                 dyn_cast_or_null<ObjCInterfaceDecl>(PrevDecl);
 
+      // Diagnose availability in the context of the @interface.
+      ContextRAII SavedContext(*this, IDecl);
       // Diagnose classes that inherit from deprecated classes.
       if (SuperClassDecl)
         (void)DiagnoseUseOfDecl(SuperClassDecl, SuperLoc);
@@ -591,6 +606,8 @@ ActOnStartClassInterface(SourceLocation AtInterfaceLoc,
 
   // Check then save referenced protocols.
   if (NumProtoRefs) {
+    diagnoseUseOfProtocols(*this, IDecl, (ObjCProtocolDecl*const*)ProtoRefs,
+                           NumProtoRefs, ProtoLocs);
     IDecl->setProtocolList((ObjCProtocolDecl*const*)ProtoRefs, NumProtoRefs,
                            ProtoLocs, Context);
     IDecl->setEndOfDefinitionLoc(EndProtoLoc);
@@ -617,8 +634,7 @@ void Sema::ActOnTypedefedProtocols(SmallVectorImpl<Decl *> &ProtocolRefs,
     QualType T = TDecl->getUnderlyingType();
     if (T->isObjCObjectType())
       if (const ObjCObjectType *OPT = T->getAs<ObjCObjectType>())
-        for (auto *I : OPT->quals())
-          ProtocolRefs.push_back(I);
+        ProtocolRefs.append(OPT->qual_begin(), OPT->qual_end());
   }
 }
 
@@ -752,6 +768,8 @@ Sema::ActOnStartProtocolInterface(SourceLocation AtProtoInterfaceLoc,
 
   if (!err && NumProtoRefs ) {
     /// Check then save referenced protocols.
+    diagnoseUseOfProtocols(*this, PDecl, (ObjCProtocolDecl*const*)ProtoRefs,
+                           NumProtoRefs, ProtoLocs);
     PDecl->setProtocolList((ObjCProtocolDecl*const*)ProtoRefs, NumProtoRefs,
                            ProtoLocs, Context);
   }
@@ -779,7 +797,7 @@ static bool NestedProtocolHasNoDefinition(ObjCProtocolDecl *PDecl,
 /// issues an error if they are not declared. It returns list of
 /// protocol declarations in its 'Protocols' argument.
 void
-Sema::FindProtocolDeclaration(bool WarnOnDeclarations,
+Sema::FindProtocolDeclaration(bool WarnOnDeclarations, bool ForObjCContainer,
                               const IdentifierLocPair *ProtocolId,
                               unsigned NumProtocols,
                               SmallVectorImpl<Decl *> &Protocols) {
@@ -805,8 +823,12 @@ Sema::FindProtocolDeclaration(bool WarnOnDeclarations,
     // If this is a forward protocol declaration, get its definition.
     if (!PDecl->isThisDeclarationADefinition() && PDecl->getDefinition())
       PDecl = PDecl->getDefinition();
-    
-    (void)DiagnoseUseOfDecl(PDecl, ProtocolId[i].second);
+
+    // For an objc container, delay protocol reference checking until after we
+    // can set the objc decl as the availability context, otherwise check now.
+    if (!ForObjCContainer) {
+      (void)DiagnoseUseOfDecl(PDecl, ProtocolId[i].second);
+    }
 
     // If this is a forward declaration and we are supposed to warn in this
     // case, do it.
@@ -935,7 +957,9 @@ ActOnStartCategoryInterface(SourceLocation AtInterfaceLoc,
   CurContext->addDecl(CDecl);
 
   if (NumProtoRefs) {
-    CDecl->setProtocolList((ObjCProtocolDecl*const*)ProtoRefs, NumProtoRefs, 
+    diagnoseUseOfProtocols(*this, CDecl, (ObjCProtocolDecl*const*)ProtoRefs,
+                           NumProtoRefs, ProtoLocs);
+    CDecl->setProtocolList((ObjCProtocolDecl*const*)ProtoRefs, NumProtoRefs,
                            ProtoLocs, Context);
     // Protocols in the class extension belong to the class.
     if (CDecl->IsClassExtension())
@@ -2241,8 +2265,14 @@ void Sema::addMethodToGlobalList(ObjCMethodList *List,
     if (getLangOpts().Modules && !getLangOpts().CurrentModule.empty())
       continue;
 
-    if (!MatchTwoMethodDeclarations(Method, List->getMethod()))
+    if (!MatchTwoMethodDeclarations(Method, List->getMethod())) {
+      // Even if two method types do not match, we would like to say
+      // there is more than one declaration so unavailability/deprecated
+      // warning is not too noisy.
+      if (!Method->isDefined())
+        List->setHasMoreThanOneDecl(true);
       continue;
+    }
 
     ObjCMethodDecl *PrevObjCMethod = List->getMethod();
 
@@ -2341,19 +2371,33 @@ bool Sema::CollectMultipleMethodsInGlobalPool(
   return Methods.size() > 1;
 }
 
-bool Sema::AreMultipleMethodsInGlobalPool(Selector Sel, bool instance) {
+bool Sema::AreMultipleMethodsInGlobalPool(Selector Sel, ObjCMethodDecl *BestMethod,
+                                          SourceRange R,
+                                          bool receiverIdOrClass) {
   GlobalMethodPool::iterator Pos = MethodPool.find(Sel);
   // Test for no method in the pool which should not trigger any warning by
   // caller.
   if (Pos == MethodPool.end())
     return true;
-  ObjCMethodList &MethList = instance ? Pos->second.first : Pos->second.second;
+  ObjCMethodList &MethList =
+    BestMethod->isInstanceMethod() ? Pos->second.first : Pos->second.second;
+  
+  // Diagnose finding more than one method in global pool
+  SmallVector<ObjCMethodDecl *, 4> Methods;
+  Methods.push_back(BestMethod);
+  for (ObjCMethodList *ML = &MethList; ML; ML = ML->getNext())
+    if (ObjCMethodDecl *M = ML->getMethod())
+      if (!M->isHidden() && M != BestMethod && !M->hasAttr<UnavailableAttr>())
+        Methods.push_back(M);
+  if (Methods.size() > 1)
+    DiagnoseMultipleMethodInGlobalPool(Methods, Sel, R, receiverIdOrClass);
+
   return MethList.hasMoreThanOneDecl();
 }
 
 ObjCMethodDecl *Sema::LookupMethodInGlobalPool(Selector Sel, SourceRange R,
                                                bool receiverIdOrClass,
-                                               bool warn, bool instance) {
+                                               bool instance) {
   if (ExternalSource)
     ReadMethodPool(Sel);
     
@@ -2365,31 +2409,23 @@ ObjCMethodDecl *Sema::LookupMethodInGlobalPool(Selector Sel, SourceRange R,
   ObjCMethodList &MethList = instance ? Pos->second.first : Pos->second.second;
   SmallVector<ObjCMethodDecl *, 4> Methods;
   for (ObjCMethodList *M = &MethList; M; M = M->getNext()) {
-    if (M->getMethod() && !M->getMethod()->isHidden()) {
-      // If we're not supposed to warn about mismatches, we're done.
-      if (!warn)
-        return M->getMethod();
-
-      Methods.push_back(M->getMethod());
-    }
+    if (M->getMethod() && !M->getMethod()->isHidden())
+      return M->getMethod();
   }
+  return nullptr;
+}
 
-  // If there aren't any visible methods, we're done.
-  // FIXME: Recover if there are any known-but-hidden methods?
-  if (Methods.empty())
-    return nullptr;
-
-  if (Methods.size() == 1)
-    return Methods[0];
-
+void Sema::DiagnoseMultipleMethodInGlobalPool(SmallVectorImpl<ObjCMethodDecl*> &Methods,
+                                              Selector Sel, SourceRange R,
+                                              bool receiverIdOrClass) {
   // We found multiple methods, so we may have to complain.
   bool issueDiagnostic = false, issueError = false;
 
   // We support a warning which complains about *any* difference in
   // method signature.
   bool strictSelectorMatch =
-      receiverIdOrClass && warn &&
-      !Diags.isIgnored(diag::warn_strict_multiple_method_decl, R.getBegin());
+  receiverIdOrClass &&
+  !Diags.isIgnored(diag::warn_strict_multiple_method_decl, R.getBegin());
   if (strictSelectorMatch) {
     for (unsigned I = 1, N = Methods.size(); I != N; ++I) {
       if (!MatchTwoMethodDeclarations(Methods[0], Methods[I], MMS_strict)) {
@@ -2414,7 +2450,7 @@ ObjCMethodDecl *Sema::LookupMethodInGlobalPool(Selector Sel, SourceRange R,
         break;
       }
     }
-
+  
   if (issueDiagnostic) {
     if (issueError)
       Diag(R.getBegin(), diag::err_arc_multiple_method_decl) << Sel << R;
@@ -2422,16 +2458,15 @@ ObjCMethodDecl *Sema::LookupMethodInGlobalPool(Selector Sel, SourceRange R,
       Diag(R.getBegin(), diag::warn_strict_multiple_method_decl) << Sel << R;
     else
       Diag(R.getBegin(), diag::warn_multiple_method_decl) << Sel << R;
-
+    
     Diag(Methods[0]->getLocStart(),
          issueError ? diag::note_possibility : diag::note_using)
-      << Methods[0]->getSourceRange();
+    << Methods[0]->getSourceRange();
     for (unsigned I = 1, N = Methods.size(); I != N; ++I) {
       Diag(Methods[I]->getLocStart(), diag::note_also_found)
-        << Methods[I]->getSourceRange();
-  }
+      << Methods[I]->getSourceRange();
+    }
   }
-  return Methods[0];
 }
 
 ObjCMethodDecl *Sema::LookupImplementedMethodInGlobalPool(Selector Sel) {
@@ -2442,12 +2477,16 @@ ObjCMethodDecl *Sema::LookupImplementedMethodInGlobalPool(Selector Sel) {
   GlobalMethods &Methods = Pos->second;
   for (const ObjCMethodList *Method = &Methods.first; Method;
        Method = Method->getNext())
-    if (Method->getMethod() && Method->getMethod()->isDefined())
+    if (Method->getMethod() &&
+        (Method->getMethod()->isDefined() ||
+         Method->getMethod()->isPropertyAccessor()))
       return Method->getMethod();
   
   for (const ObjCMethodList *Method = &Methods.second; Method;
        Method = Method->getNext())
-    if (Method->getMethod() && Method->getMethod()->isDefined())
+    if (Method->getMethod() &&
+        (Method->getMethod()->isDefined() ||
+         Method->getMethod()->isPropertyAccessor()))
       return Method->getMethod();
   return nullptr;
 }
@@ -2571,10 +2610,9 @@ Sema::ObjCContainerKind Sema::getObjCContainerKind() const {
     case Decl::ObjCProtocol:
       return Sema::OCK_Protocol;
     case Decl::ObjCCategory:
-      if (dyn_cast<ObjCCategoryDecl>(CurContext)->IsClassExtension())
+      if (cast<ObjCCategoryDecl>(CurContext)->IsClassExtension())
         return Sema::OCK_ClassExtension;
-      else
-        return Sema::OCK_Category;
+      return Sema::OCK_Category;
     case Decl::ObjCImplementation:
       return Sema::OCK_Implementation;
     case Decl::ObjCCategoryImpl:
@@ -3286,7 +3324,7 @@ Decl *Sema::ActOnMethodDeclaration(
       
     case OMF_alloc:
     case OMF_new:
-      InferRelatedResultType = ObjCMethod->isClassMethod();
+        InferRelatedResultType = ObjCMethod->isClassMethod();
       break;
         
     case OMF_init:
@@ -3297,7 +3335,8 @@ Decl *Sema::ActOnMethodDeclaration(
       break;
     }
     
-    if (InferRelatedResultType)
+    if (InferRelatedResultType &&
+        !ObjCMethod->getReturnType()->isObjCIndependentClassType())
       ObjCMethod->SetRelatedResultType();
   }
 
@@ -3487,12 +3526,11 @@ void Sema::DiagnoseUseOfUnimplementedSelectors() {
   if (ReferencedSelectors.empty() || 
       !Context.AnyObjCImplementation())
     return;
-  for (llvm::DenseMap<Selector, SourceLocation>::iterator S = 
-        ReferencedSelectors.begin(),
-       E = ReferencedSelectors.end(); S != E; ++S) {
-    Selector Sel = (*S).first;
+  for (auto &SelectorAndLocation : ReferencedSelectors) {
+    Selector Sel = SelectorAndLocation.first;
+    SourceLocation Loc = SelectorAndLocation.second;
     if (!LookupImplementedMethodInGlobalPool(Sel))
-      Diag((*S).second, diag::warn_unimplemented_selector) << Sel;
+      Diag(Loc, diag::warn_unimplemented_selector) << Sel;
   }
   return;
 }
diff --git a/lib/Sema/SemaExceptionSpec.cpp b/lib/Sema/SemaExceptionSpec.cpp
index 2387325be435..51d6acebada1 100644
--- a/lib/Sema/SemaExceptionSpec.cpp
+++ b/lib/Sema/SemaExceptionSpec.cpp
@@ -167,13 +167,13 @@ Sema::ResolveExceptionSpec(SourceLocation Loc, const FunctionProtoType *FPT) {
 void
 Sema::UpdateExceptionSpec(FunctionDecl *FD,
                           const FunctionProtoType::ExceptionSpecInfo &ESI) {
-  for (auto *Redecl : FD->redecls())
-    Context.adjustExceptionSpec(cast<FunctionDecl>(Redecl), ESI);
-
   // If we've fully resolved the exception specification, notify listeners.
   if (!isUnresolvedExceptionSpec(ESI.Type))
     if (auto *Listener = getASTMutationListener())
       Listener->ResolvedExceptionSpec(FD);
+
+  for (auto *Redecl : FD->redecls())
+    Context.adjustExceptionSpec(cast<FunctionDecl>(Redecl), ESI);
 }
 
 /// Determine whether a function has an implicitly-generated exception
@@ -437,7 +437,7 @@ bool Sema::CheckEquivalentExceptionSpec(const PartialDiagnostic &DiagID,
       OldNR != FunctionProtoType::NR_NoNoexcept &&
       NewNR != FunctionProtoType::NR_NoNoexcept) {
     Diag(NewLoc, DiagID);
-    if (NoteID.getDiagID() != 0)
+    if (NoteID.getDiagID() != 0 && OldLoc.isValid())
       Diag(OldLoc, NoteID);
     return true;
   }
@@ -518,7 +518,7 @@ bool Sema::CheckEquivalentExceptionSpec(const PartialDiagnostic &DiagID,
     }
 
     Diag(NewLoc, DiagID);
-    if (NoteID.getDiagID() != 0)
+    if (NoteID.getDiagID() != 0 && OldLoc.isValid())
       Diag(OldLoc, NoteID);
     return true;
   }
@@ -547,7 +547,7 @@ bool Sema::CheckEquivalentExceptionSpec(const PartialDiagnostic &DiagID,
     return false;
   }
   Diag(NewLoc, DiagID);
-  if (NoteID.getDiagID() != 0)
+  if (NoteID.getDiagID() != 0 && OldLoc.isValid())
     Diag(OldLoc, NoteID);
   return true;
 }
diff --git a/lib/Sema/SemaExpr.cpp b/lib/Sema/SemaExpr.cpp
index 8be11572b2e0..7ab269c3b6d5 100644
--- a/lib/Sema/SemaExpr.cpp
+++ b/lib/Sema/SemaExpr.cpp
@@ -76,25 +76,57 @@ bool Sema::CanUseDecl(NamedDecl *D) {
 static void DiagnoseUnusedOfDecl(Sema &S, NamedDecl *D, SourceLocation Loc) {
   // Warn if this is used but marked unused.
   if (D->hasAttr<UnusedAttr>()) {
-    const Decl *DC = cast<Decl>(S.getCurObjCLexicalContext());
-    if (!DC->hasAttr<UnusedAttr>())
+    const Decl *DC = cast_or_null<Decl>(S.getCurObjCLexicalContext());
+    if (DC && !DC->hasAttr<UnusedAttr>())
       S.Diag(Loc, diag::warn_used_but_marked_unused) << D->getDeclName();
   }
 }
 
-static AvailabilityResult DiagnoseAvailabilityOfDecl(Sema &S,
-                              NamedDecl *D, SourceLocation Loc,
-                              const ObjCInterfaceDecl *UnknownObjCClass,
-                              bool ObjCPropertyAccess) {
+static bool HasRedeclarationWithoutAvailabilityInCategory(const Decl *D) {
+  const auto *OMD = dyn_cast<ObjCMethodDecl>(D);
+  if (!OMD)
+    return false;
+  const ObjCInterfaceDecl *OID = OMD->getClassInterface();
+  if (!OID)
+    return false;
+
+  for (const ObjCCategoryDecl *Cat : OID->visible_categories())
+    if (ObjCMethodDecl *CatMeth =
+            Cat->getMethod(OMD->getSelector(), OMD->isInstanceMethod()))
+      if (!CatMeth->hasAttr<AvailabilityAttr>())
+        return true;
+  return false;
+}
+
+static AvailabilityResult
+DiagnoseAvailabilityOfDecl(Sema &S, NamedDecl *D, SourceLocation Loc,
+                           const ObjCInterfaceDecl *UnknownObjCClass,
+                           bool ObjCPropertyAccess) {
   // See if this declaration is unavailable or deprecated.
   std::string Message;
+  AvailabilityResult Result = D->getAvailability(&Message);
+
+  // For typedefs, if the typedef declaration appears available look
+  // to the underlying type to see if it is more restrictive.
+  while (const TypedefNameDecl *TD = dyn_cast<TypedefNameDecl>(D)) {
+    if (Result == AR_Available) {
+      if (const TagType *TT = TD->getUnderlyingType()->getAs<TagType>()) {
+        D = TT->getDecl();
+        Result = D->getAvailability(&Message);
+        continue;
+      }
+    }
+    break;
+  }
     
   // Forward class declarations get their attributes from their definition.
   if (ObjCInterfaceDecl *IDecl = dyn_cast<ObjCInterfaceDecl>(D)) {
-    if (IDecl->getDefinition())
+    if (IDecl->getDefinition()) {
       D = IDecl->getDefinition();
+      Result = D->getAvailability(&Message);
+    }
   }
-  AvailabilityResult Result = D->getAvailability(&Message);
+
   if (const EnumConstantDecl *ECD = dyn_cast<EnumConstantDecl>(D))
     if (Result == AR_Available) {
       const DeclContext *DC = ECD->getDeclContext();
@@ -103,7 +135,8 @@ static AvailabilityResult DiagnoseAvailabilityOfDecl(Sema &S,
     }
 
   const ObjCPropertyDecl *ObjCPDecl = nullptr;
-  if (Result == AR_Deprecated || Result == AR_Unavailable) {
+  if (Result == AR_Deprecated || Result == AR_Unavailable ||
+      AR_NotYetIntroduced) {
     if (const ObjCMethodDecl *MD = dyn_cast<ObjCMethodDecl>(D)) {
       if (const ObjCPropertyDecl *PD = MD->findPropertyDecl()) {
         AvailabilityResult PDeclResult = PD->getAvailability(nullptr);
@@ -115,7 +148,6 @@ static AvailabilityResult DiagnoseAvailabilityOfDecl(Sema &S,
   
   switch (Result) {
     case AR_Available:
-    case AR_NotYetIntroduced:
       break;
 
     case AR_Deprecated:
@@ -125,6 +157,34 @@ static AvailabilityResult DiagnoseAvailabilityOfDecl(Sema &S,
                                   ObjCPropertyAccess);
       break;
 
+    case AR_NotYetIntroduced: {
+      // Don't do this for enums, they can't be redeclared.
+      if (isa<EnumConstantDecl>(D) || isa<EnumDecl>(D))
+        break;
+ 
+      bool Warn = !D->getAttr<AvailabilityAttr>()->isInherited();
+      // Objective-C method declarations in categories are not modelled as
+      // redeclarations, so manually look for a redeclaration in a category
+      // if necessary.
+      if (Warn && HasRedeclarationWithoutAvailabilityInCategory(D))
+        Warn = false;
+      // In general, D will point to the most recent redeclaration. However,
+      // for `@class A;` decls, this isn't true -- manually go through the
+      // redecl chain in that case.
+      if (Warn && isa<ObjCInterfaceDecl>(D))
+        for (Decl *Redecl = D->getMostRecentDecl(); Redecl && Warn;
+             Redecl = Redecl->getPreviousDecl())
+          if (!Redecl->hasAttr<AvailabilityAttr>() ||
+              Redecl->getAttr<AvailabilityAttr>()->isInherited())
+            Warn = false;
+ 
+      if (Warn)
+        S.EmitAvailabilityWarning(Sema::AD_Partial, D, Message, Loc,
+                                  UnknownObjCClass, ObjCPDecl,
+                                  ObjCPropertyAccess);
+      break;
+    }
+
     case AR_Unavailable:
       if (S.getCurContextAvailability() != AR_Unavailable)
         S.EmitAvailabilityWarning(Sema::AD_Unavailable,
@@ -307,7 +367,8 @@ bool Sema::DiagnoseUseOfDecl(NamedDecl *D, SourceLocation Loc,
         DeduceReturnType(FD, Loc))
       return true;
   }
-  DiagnoseAvailabilityOfDecl(*this, D, Loc, UnknownObjCClass, ObjCPropertyAccess);
+  DiagnoseAvailabilityOfDecl(*this, D, Loc, UnknownObjCClass,
+                             ObjCPropertyAccess);
 
   DiagnoseUnusedOfDecl(*this, D, Loc);
 
@@ -405,12 +466,11 @@ void Sema::DiagnoseSentinelCalls(NamedDecl *D, SourceLocation Loc,
   SourceLocation MissingNilLoc
     = PP.getLocForEndOfToken(sentinelExpr->getLocEnd());
   std::string NullValue;
-  if (calleeType == CT_Method &&
-      PP.getIdentifierInfo("nil")->hasMacroDefinition())
+  if (calleeType == CT_Method && PP.isMacroDefined("nil"))
     NullValue = "nil";
   else if (getLangOpts().CPlusPlus11)
     NullValue = "nullptr";
-  else if (PP.getIdentifierInfo("NULL")->hasMacroDefinition())
+  else if (PP.isMacroDefined("NULL"))
     NullValue = "NULL";
   else
     NullValue = "(void*) 0";
@@ -3266,7 +3326,9 @@ ExprResult Sema::ActOnNumericConstant(const Token &Tok, Scope *UDLScope) {
     if (Ty == Context.DoubleTy) {
       if (getLangOpts().SinglePrecisionConstants) {
         Res = ImpCastExprToType(Res, Context.FloatTy, CK_FloatingCast).get();
-      } else if (getLangOpts().OpenCL && !getOpenCLOptions().cl_khr_fp64) {
+      } else if (getLangOpts().OpenCL &&
+                 !((getLangOpts().OpenCLVersion >= 120) ||
+                   getOpenCLOptions().cl_khr_fp64)) {
         Diag(Tok.getLocation(), diag::warn_double_const_requires_fp64);
         Res = ImpCastExprToType(Res, Context.FloatTy, CK_FloatingCast).get();
       }
@@ -3322,6 +3384,9 @@ ExprResult Sema::ActOnNumericConstant(const Token &Tok, Scope *UDLScope) {
           Diag(Tok.getLocation(), diag::err_int128_unsupported);
           Width = MaxWidth;
           Ty = Context.getIntMaxType();
+        } else if (Literal.MicrosoftInteger == 8 && !Literal.isUnsigned) {
+          Width = 8;
+          Ty = Context.CharTy;
         } else {
           Width = Literal.MicrosoftInteger;
           Ty = Context.getIntTypeForBitwidth(Width,
@@ -4558,6 +4623,83 @@ static bool checkArgsForPlaceholders(Sema &S, MultiExprArg args) {
   return hasInvalid;
 }
 
+/// If a builtin function has a pointer argument with no explicit address
+/// space, than it should be able to accept a pointer to any address
+/// space as input.  In order to do this, we need to replace the
+/// standard builtin declaration with one that uses the same address space
+/// as the call.
+///
+/// \returns nullptr If this builtin is not a candidate for a rewrite i.e.
+///                  it does not contain any pointer arguments without
+///                  an address space qualifer.  Otherwise the rewritten
+///                  FunctionDecl is returned.
+/// TODO: Handle pointer return types.
+static FunctionDecl *rewriteBuiltinFunctionDecl(Sema *Sema, ASTContext &Context,
+                                                const FunctionDecl *FDecl,
+                                                MultiExprArg ArgExprs) {
+
+  QualType DeclType = FDecl->getType();
+  const FunctionProtoType *FT = dyn_cast<FunctionProtoType>(DeclType);
+
+  if (!Context.BuiltinInfo.hasPtrArgsOrResult(FDecl->getBuiltinID()) ||
+      !FT || FT->isVariadic() || ArgExprs.size() != FT->getNumParams())
+    return nullptr;
+
+  bool NeedsNewDecl = false;
+  unsigned i = 0;
+  SmallVector<QualType, 8> OverloadParams;
+
+  for (QualType ParamType : FT->param_types()) {
+
+    // Convert array arguments to pointer to simplify type lookup.
+    Expr *Arg = Sema->DefaultFunctionArrayLvalueConversion(ArgExprs[i++]).get();
+    QualType ArgType = Arg->getType();
+    if (!ParamType->isPointerType() ||
+        ParamType.getQualifiers().hasAddressSpace() ||
+        !ArgType->isPointerType() ||
+        !ArgType->getPointeeType().getQualifiers().hasAddressSpace()) {
+      OverloadParams.push_back(ParamType);
+      continue;
+    }
+
+    NeedsNewDecl = true;
+    unsigned AS = ArgType->getPointeeType().getQualifiers().getAddressSpace();
+
+    QualType PointeeType = ParamType->getPointeeType();
+    PointeeType = Context.getAddrSpaceQualType(PointeeType, AS);
+    OverloadParams.push_back(Context.getPointerType(PointeeType));
+  }
+
+  if (!NeedsNewDecl)
+    return nullptr;
+
+  FunctionProtoType::ExtProtoInfo EPI;
+  QualType OverloadTy = Context.getFunctionType(FT->getReturnType(),
+                                                OverloadParams, EPI);
+  DeclContext *Parent = Context.getTranslationUnitDecl();
+  FunctionDecl *OverloadDecl = FunctionDecl::Create(Context, Parent,
+                                                    FDecl->getLocation(),
+                                                    FDecl->getLocation(),
+                                                    FDecl->getIdentifier(),
+                                                    OverloadTy,
+                                                    /*TInfo=*/nullptr,
+                                                    SC_Extern, false,
+                                                    /*hasPrototype=*/true);
+  SmallVector<ParmVarDecl*, 16> Params;
+  FT = cast<FunctionProtoType>(OverloadTy);
+  for (unsigned i = 0, e = FT->getNumParams(); i != e; ++i) {
+    QualType ParamType = FT->getParamType(i);
+    ParmVarDecl *Parm =
+        ParmVarDecl::Create(Context, OverloadDecl, SourceLocation(),
+                                SourceLocation(), nullptr, ParamType,
+                                /*TInfo=*/nullptr, SC_None, nullptr);
+    Parm->setScopeInfo(0, i);
+    Params.push_back(Parm);
+  }
+  OverloadDecl->setParams(Params);
+  return OverloadDecl;
+}
+
 /// ActOnCallExpr - Handle a call to Fn with the specified array of arguments.
 /// This provides the location of the left/right parens and a list of comma
 /// locations.
@@ -4661,10 +4803,24 @@ Sema::ActOnCallExpr(Scope *S, Expr *Fn, SourceLocation LParenLoc,
   if (UnaryOperator *UnOp = dyn_cast<UnaryOperator>(NakedFn))
     if (UnOp->getOpcode() == UO_AddrOf)
       NakedFn = UnOp->getSubExpr()->IgnoreParens();
-  
-  if (isa<DeclRefExpr>(NakedFn))
+
+  if (isa<DeclRefExpr>(NakedFn)) {
     NDecl = cast<DeclRefExpr>(NakedFn)->getDecl();
-  else if (isa<MemberExpr>(NakedFn))
+
+    FunctionDecl *FDecl = dyn_cast<FunctionDecl>(NDecl);
+    if (FDecl && FDecl->getBuiltinID()) {
+      // Rewrite the function decl for this builtin by replacing paramaters
+      // with no explicit address space with the address space of the arguments
+      // in ArgExprs.
+      if ((FDecl = rewriteBuiltinFunctionDecl(this, Context, FDecl, ArgExprs))) {
+        NDecl = FDecl;
+        Fn = DeclRefExpr::Create(Context, FDecl->getQualifierLoc(),
+                           SourceLocation(), FDecl, false,
+                           SourceLocation(), FDecl->getType(),
+                           Fn->getValueKind(), FDecl);
+      }
+    }
+  } else if (isa<MemberExpr>(NakedFn))
     NDecl = cast<MemberExpr>(NakedFn)->getMemberDecl();
 
   if (FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(NDecl)) {
@@ -5516,47 +5672,24 @@ bool Sema::DiagnoseConditionalForNull(Expr *LHSExpr, Expr *RHSExpr,
 }
 
 /// \brief Return false if the condition expression is valid, true otherwise.
-static bool checkCondition(Sema &S, Expr *Cond) {
+static bool checkCondition(Sema &S, Expr *Cond, SourceLocation QuestionLoc) {
   QualType CondTy = Cond->getType();
 
+  // OpenCL v1.1 s6.3.i says the condition cannot be a floating point type.
+  if (S.getLangOpts().OpenCL && CondTy->isFloatingType()) {
+    S.Diag(QuestionLoc, diag::err_typecheck_cond_expect_nonfloat)
+      << CondTy << Cond->getSourceRange();
+    return true;
+  }
+
   // C99 6.5.15p2
   if (CondTy->isScalarType()) return false;
 
-  // OpenCL v1.1 s6.3.i says the condition is allowed to be a vector or scalar.
-  if (S.getLangOpts().OpenCL && CondTy->isVectorType())
-    return false;
-
-  // Emit the proper error message.
-  S.Diag(Cond->getLocStart(), S.getLangOpts().OpenCL ?
-                              diag::err_typecheck_cond_expect_scalar :
-                              diag::err_typecheck_cond_expect_scalar_or_vector)
-    << CondTy;
+  S.Diag(QuestionLoc, diag::err_typecheck_cond_expect_scalar)
+    << CondTy << Cond->getSourceRange();
   return true;
 }
 
-/// \brief Return false if the two expressions can be converted to a vector,
-/// true otherwise
-static bool checkConditionalConvertScalarsToVectors(Sema &S, ExprResult &LHS,
-                                                    ExprResult &RHS,
-                                                    QualType CondTy) {
-  // Both operands should be of scalar type.
-  if (!LHS.get()->getType()->isScalarType()) {
-    S.Diag(LHS.get()->getLocStart(), diag::err_typecheck_cond_expect_scalar)
-      << CondTy;
-    return true;
-  }
-  if (!RHS.get()->getType()->isScalarType()) {
-    S.Diag(RHS.get()->getLocStart(), diag::err_typecheck_cond_expect_scalar)
-      << CondTy;
-    return true;
-  }
-
-  // Implicity convert these scalars to the type of the condition.
-  LHS = S.ImpCastExprToType(LHS.get(), CondTy, CK_IntegralCast);
-  RHS = S.ImpCastExprToType(RHS.get(), CondTy, CK_IntegralCast);
-  return false;
-}
-
 /// \brief Handle when one or both operands are void type.
 static QualType checkConditionalVoidType(Sema &S, ExprResult &LHS,
                                          ExprResult &RHS) {
@@ -5773,6 +5906,184 @@ static bool checkPointerIntegerMismatch(Sema &S, ExprResult &Int,
   return true;
 }
 
+/// \brief Simple conversion between integer and floating point types.
+///
+/// Used when handling the OpenCL conditional operator where the
+/// condition is a vector while the other operands are scalar.
+///
+/// OpenCL v1.1 s6.3.i and s6.11.6 together require that the scalar
+/// types are either integer or floating type. Between the two
+/// operands, the type with the higher rank is defined as the "result
+/// type". The other operand needs to be promoted to the same type. No
+/// other type promotion is allowed. We cannot use
+/// UsualArithmeticConversions() for this purpose, since it always
+/// promotes promotable types.
+static QualType OpenCLArithmeticConversions(Sema &S, ExprResult &LHS,
+                                            ExprResult &RHS,
+                                            SourceLocation QuestionLoc) {
+  LHS = S.DefaultFunctionArrayLvalueConversion(LHS.get());
+  if (LHS.isInvalid())
+    return QualType();
+  RHS = S.DefaultFunctionArrayLvalueConversion(RHS.get());
+  if (RHS.isInvalid())
+    return QualType();
+
+  // For conversion purposes, we ignore any qualifiers.
+  // For example, "const float" and "float" are equivalent.
+  QualType LHSType =
+    S.Context.getCanonicalType(LHS.get()->getType()).getUnqualifiedType();
+  QualType RHSType =
+    S.Context.getCanonicalType(RHS.get()->getType()).getUnqualifiedType();
+
+  if (!LHSType->isIntegerType() && !LHSType->isRealFloatingType()) {
+    S.Diag(QuestionLoc, diag::err_typecheck_cond_expect_int_float)
+      << LHSType << LHS.get()->getSourceRange();
+    return QualType();
+  }
+
+  if (!RHSType->isIntegerType() && !RHSType->isRealFloatingType()) {
+    S.Diag(QuestionLoc, diag::err_typecheck_cond_expect_int_float)
+      << RHSType << RHS.get()->getSourceRange();
+    return QualType();
+  }
+
+  // If both types are identical, no conversion is needed.
+  if (LHSType == RHSType)
+    return LHSType;
+
+  // Now handle "real" floating types (i.e. float, double, long double).
+  if (LHSType->isRealFloatingType() || RHSType->isRealFloatingType())
+    return handleFloatConversion(S, LHS, RHS, LHSType, RHSType,
+                                 /*IsCompAssign = */ false);
+
+  // Finally, we have two differing integer types.
+  return handleIntegerConversion<doIntegralCast, doIntegralCast>
+  (S, LHS, RHS, LHSType, RHSType, /*IsCompAssign = */ false);
+}
+
+/// \brief Convert scalar operands to a vector that matches the
+///        condition in length.
+///
+/// Used when handling the OpenCL conditional operator where the
+/// condition is a vector while the other operands are scalar.
+///
+/// We first compute the "result type" for the scalar operands
+/// according to OpenCL v1.1 s6.3.i. Both operands are then converted
+/// into a vector of that type where the length matches the condition
+/// vector type. s6.11.6 requires that the element types of the result
+/// and the condition must have the same number of bits.
+static QualType
+OpenCLConvertScalarsToVectors(Sema &S, ExprResult &LHS, ExprResult &RHS,
+                              QualType CondTy, SourceLocation QuestionLoc) {
+  QualType ResTy = OpenCLArithmeticConversions(S, LHS, RHS, QuestionLoc);
+  if (ResTy.isNull()) return QualType();
+
+  const VectorType *CV = CondTy->getAs<VectorType>();
+  assert(CV);
+
+  // Determine the vector result type
+  unsigned NumElements = CV->getNumElements();
+  QualType VectorTy = S.Context.getExtVectorType(ResTy, NumElements);
+
+  // Ensure that all types have the same number of bits
+  if (S.Context.getTypeSize(CV->getElementType())
+      != S.Context.getTypeSize(ResTy)) {
+    // Since VectorTy is created internally, it does not pretty print
+    // with an OpenCL name. Instead, we just print a description.
+    std::string EleTyName = ResTy.getUnqualifiedType().getAsString();
+    SmallString<64> Str;
+    llvm::raw_svector_ostream OS(Str);
+    OS << "(vector of " << NumElements << " '" << EleTyName << "' values)";
+    S.Diag(QuestionLoc, diag::err_conditional_vector_element_size)
+      << CondTy << OS.str();
+    return QualType();
+  }
+
+  // Convert operands to the vector result type
+  LHS = S.ImpCastExprToType(LHS.get(), VectorTy, CK_VectorSplat);
+  RHS = S.ImpCastExprToType(RHS.get(), VectorTy, CK_VectorSplat);
+
+  return VectorTy;
+}
+
+/// \brief Return false if this is a valid OpenCL condition vector
+static bool checkOpenCLConditionVector(Sema &S, Expr *Cond,
+                                       SourceLocation QuestionLoc) {
+  // OpenCL v1.1 s6.11.6 says the elements of the vector must be of
+  // integral type.
+  const VectorType *CondTy = Cond->getType()->getAs<VectorType>();
+  assert(CondTy);
+  QualType EleTy = CondTy->getElementType();
+  if (EleTy->isIntegerType()) return false;
+
+  S.Diag(QuestionLoc, diag::err_typecheck_cond_expect_nonfloat)
+    << Cond->getType() << Cond->getSourceRange();
+  return true;
+}
+
+/// \brief Return false if the vector condition type and the vector
+///        result type are compatible.
+///
+/// OpenCL v1.1 s6.11.6 requires that both vector types have the same
+/// number of elements, and their element types have the same number
+/// of bits.
+static bool checkVectorResult(Sema &S, QualType CondTy, QualType VecResTy,
+                              SourceLocation QuestionLoc) {
+  const VectorType *CV = CondTy->getAs<VectorType>();
+  const VectorType *RV = VecResTy->getAs<VectorType>();
+  assert(CV && RV);
+
+  if (CV->getNumElements() != RV->getNumElements()) {
+    S.Diag(QuestionLoc, diag::err_conditional_vector_size)
+      << CondTy << VecResTy;
+    return true;
+  }
+
+  QualType CVE = CV->getElementType();
+  QualType RVE = RV->getElementType();
+
+  if (S.Context.getTypeSize(CVE) != S.Context.getTypeSize(RVE)) {
+    S.Diag(QuestionLoc, diag::err_conditional_vector_element_size)
+      << CondTy << VecResTy;
+    return true;
+  }
+
+  return false;
+}
+
+/// \brief Return the resulting type for the conditional operator in
+///        OpenCL (aka "ternary selection operator", OpenCL v1.1
+///        s6.3.i) when the condition is a vector type.
+static QualType
+OpenCLCheckVectorConditional(Sema &S, ExprResult &Cond,
+                             ExprResult &LHS, ExprResult &RHS,
+                             SourceLocation QuestionLoc) {
+  Cond = S.DefaultFunctionArrayLvalueConversion(Cond.get()); 
+  if (Cond.isInvalid())
+    return QualType();
+  QualType CondTy = Cond.get()->getType();
+
+  if (checkOpenCLConditionVector(S, Cond.get(), QuestionLoc))
+    return QualType();
+
+  // If either operand is a vector then find the vector type of the
+  // result as specified in OpenCL v1.1 s6.3.i.
+  if (LHS.get()->getType()->isVectorType() ||
+      RHS.get()->getType()->isVectorType()) {
+    QualType VecResTy = S.CheckVectorOperands(LHS, RHS, QuestionLoc,
+                                              /*isCompAssign*/false);
+    if (VecResTy.isNull()) return QualType();
+    // The result type must match the condition type as specified in
+    // OpenCL v1.1 s6.11.6.
+    if (checkVectorResult(S, CondTy, VecResTy, QuestionLoc))
+      return QualType();
+    return VecResTy;
+  }
+
+  // Both operands are scalar.
+  return OpenCLConvertScalarsToVectors(S, LHS, RHS, CondTy, QuestionLoc);
+}
+
 /// Note that LHS is not null here, even if this is the gnu "x ?: y" extension.
 /// In that case, LHS = cond.
 /// C99 6.5.15
@@ -5796,11 +6107,16 @@ QualType Sema::CheckConditionalOperands(ExprResult &Cond, ExprResult &LHS,
   VK = VK_RValue;
   OK = OK_Ordinary;
 
+  // The OpenCL operator with a vector condition is sufficiently
+  // different to merit its own checker.
+  if (getLangOpts().OpenCL && Cond.get()->getType()->isVectorType())
+    return OpenCLCheckVectorConditional(*this, Cond, LHS, RHS, QuestionLoc);
+
   // First, check the condition.
   Cond = UsualUnaryConversions(Cond.get());
   if (Cond.isInvalid())
     return QualType();
-  if (checkCondition(*this, Cond.get()))
+  if (checkCondition(*this, Cond.get(), QuestionLoc))
     return QualType();
 
   // Now check the two expressions.
@@ -5812,17 +6128,9 @@ QualType Sema::CheckConditionalOperands(ExprResult &Cond, ExprResult &LHS,
   if (LHS.isInvalid() || RHS.isInvalid())
     return QualType();
 
-  QualType CondTy = Cond.get()->getType();
   QualType LHSTy = LHS.get()->getType();
   QualType RHSTy = RHS.get()->getType();
 
-  // If the condition is a vector, and both operands are scalar,
-  // attempt to implicity convert them to the vector type to act like the
-  // built in select. (OpenCL v1.1 s6.3.i)
-  if (getLangOpts().OpenCL && CondTy->isVectorType())
-    if (checkConditionalConvertScalarsToVectors(*this, LHS, RHS, CondTy))
-      return QualType();
-  
   // If both operands have arithmetic type, do the usual arithmetic conversions
   // to find a common type: C99 6.5.15p3,5.
   if (LHSTy->isArithmeticType() && RHSTy->isArithmeticType()) {
@@ -6116,6 +6424,8 @@ static bool ExprLooksBoolean(Expr *E) {
     return IsLogicOp(OP->getOpcode());
   if (UnaryOperator *OP = dyn_cast<UnaryOperator>(E))
     return OP->getOpcode() == UO_LNot;
+  if (E->getType()->isPointerType())
+    return true;
 
   return false;
 }
@@ -6218,6 +6528,8 @@ ExprResult Sema::ActOnConditionalOp(SourceLocation QuestionLoc,
   DiagnoseConditionalPrecedence(*this, QuestionLoc, Cond.get(), LHS.get(),
                                 RHS.get());
 
+  CheckBoolLikeConversion(Cond.get(), QuestionLoc);
+
   if (!commonExpr)
     return new (Context)
         ConditionalOperator(Cond.get(), QuestionLoc, LHS.get(), ColonLoc,
@@ -7178,9 +7490,12 @@ static void diagnoseArithmeticOnFunctionPointer(Sema &S, SourceLocation Loc,
 /// \returns True if pointer has incomplete type
 static bool checkArithmeticIncompletePointerType(Sema &S, SourceLocation Loc,
                                                  Expr *Operand) {
-  assert(Operand->getType()->isAnyPointerType() &&
-         !Operand->getType()->isDependentType());
-  QualType PointeeTy = Operand->getType()->getPointeeType();
+  QualType ResType = Operand->getType();
+  if (const AtomicType *ResAtomicType = ResType->getAs<AtomicType>())
+    ResType = ResAtomicType->getValueType();
+
+  assert(ResType->isAnyPointerType() && !ResType->isDependentType());
+  QualType PointeeTy = ResType->getPointeeType();
   return S.RequireCompleteType(Loc, PointeeTy,
                                diag::err_typecheck_arithmetic_incomplete_type,
                                PointeeTy, Operand->getSourceRange());
@@ -7196,9 +7511,13 @@ static bool checkArithmeticIncompletePointerType(Sema &S, SourceLocation Loc,
 /// \returns True when the operand is valid to use (even if as an extension).
 static bool checkArithmeticOpPointerOperand(Sema &S, SourceLocation Loc,
                                             Expr *Operand) {
-  if (!Operand->getType()->isAnyPointerType()) return true;
+  QualType ResType = Operand->getType();
+  if (const AtomicType *ResAtomicType = ResType->getAs<AtomicType>())
+    ResType = ResAtomicType->getValueType();
 
-  QualType PointeeTy = Operand->getType()->getPointeeType();
+  if (!ResType->isAnyPointerType()) return true;
+
+  QualType PointeeTy = ResType->getPointeeType();
   if (PointeeTy->isVoidType()) {
     diagnoseArithmeticOnVoidPointer(S, Loc, Operand);
     return !S.getLangOpts().CPlusPlus;
@@ -7558,7 +7877,7 @@ static void DiagnoseBadShiftValues(Sema& S, ExprResult &LHS, ExprResult &RHS,
   llvm::APSInt Right;
   // Check right/shifter operand
   if (RHS.get()->isValueDependent() ||
-      !RHS.get()->isIntegerConstantExpr(Right, S.Context))
+      !RHS.get()->EvaluateAsInt(Right, S.Context))
     return;
 
   if (Right.isNegative()) {
@@ -7605,7 +7924,7 @@ static void DiagnoseBadShiftValues(Sema& S, ExprResult &LHS, ExprResult &RHS,
   // turned off separately if needed.
   if (LeftBits == ResultBits - 1) {
     S.Diag(Loc, diag::warn_shift_result_sets_sign_bit)
-        << HexResult.str() << LHSType
+        << HexResult << LHSType
         << LHS.get()->getSourceRange() << RHS.get()->getSourceRange();
     return;
   }
@@ -7616,6 +7935,69 @@ static void DiagnoseBadShiftValues(Sema& S, ExprResult &LHS, ExprResult &RHS,
     << RHS.get()->getSourceRange();
 }
 
+/// \brief Return the resulting type when an OpenCL vector is shifted
+///        by a scalar or vector shift amount.
+static QualType checkOpenCLVectorShift(Sema &S,
+                                       ExprResult &LHS, ExprResult &RHS,
+                                       SourceLocation Loc, bool IsCompAssign) {
+  // OpenCL v1.1 s6.3.j says RHS can be a vector only if LHS is a vector.
+  if (!LHS.get()->getType()->isVectorType()) {
+    S.Diag(Loc, diag::err_shift_rhs_only_vector)
+      << RHS.get()->getType() << LHS.get()->getType()
+      << LHS.get()->getSourceRange() << RHS.get()->getSourceRange();
+    return QualType();
+  }
+
+  if (!IsCompAssign) {
+    LHS = S.UsualUnaryConversions(LHS.get());
+    if (LHS.isInvalid()) return QualType();
+  }
+
+  RHS = S.UsualUnaryConversions(RHS.get());
+  if (RHS.isInvalid()) return QualType();
+
+  QualType LHSType = LHS.get()->getType();
+  const VectorType *LHSVecTy = LHSType->getAs<VectorType>();
+  QualType LHSEleType = LHSVecTy->getElementType();
+
+  // Note that RHS might not be a vector.
+  QualType RHSType = RHS.get()->getType();
+  const VectorType *RHSVecTy = RHSType->getAs<VectorType>();
+  QualType RHSEleType = RHSVecTy ? RHSVecTy->getElementType() : RHSType;
+
+  // OpenCL v1.1 s6.3.j says that the operands need to be integers.
+  if (!LHSEleType->isIntegerType()) {
+    S.Diag(Loc, diag::err_typecheck_expect_int)
+      << LHS.get()->getType() << LHS.get()->getSourceRange();
+    return QualType();
+  }
+
+  if (!RHSEleType->isIntegerType()) {
+    S.Diag(Loc, diag::err_typecheck_expect_int)
+      << RHS.get()->getType() << RHS.get()->getSourceRange();
+    return QualType();
+  }
+
+  if (RHSVecTy) {
+    // OpenCL v1.1 s6.3.j says that for vector types, the operators
+    // are applied component-wise. So if RHS is a vector, then ensure
+    // that the number of elements is the same as LHS...
+    if (RHSVecTy->getNumElements() != LHSVecTy->getNumElements()) {
+      S.Diag(Loc, diag::err_typecheck_vector_lengths_not_equal)
+        << LHS.get()->getType() << RHS.get()->getType()
+        << LHS.get()->getSourceRange() << RHS.get()->getSourceRange();
+      return QualType();
+    }
+  } else {
+    // ...else expand RHS to match the number of elements in LHS.
+    QualType VecTy =
+      S.Context.getExtVectorType(RHSEleType, LHSVecTy->getNumElements());
+    RHS = S.ImpCastExprToType(RHS.get(), VecTy, CK_VectorSplat);
+  }
+
+  return LHSType;
+}
+
 // C99 6.5.7
 QualType Sema::CheckShiftOperands(ExprResult &LHS, ExprResult &RHS,
                                   SourceLocation Loc, unsigned Opc,
@@ -7624,8 +8006,11 @@ QualType Sema::CheckShiftOperands(ExprResult &LHS, ExprResult &RHS,
 
   // Vector shifts promote their scalar inputs to vector type.
   if (LHS.get()->getType()->isVectorType() ||
-      RHS.get()->getType()->isVectorType())
+      RHS.get()->getType()->isVectorType()) {
+    if (LangOpts.OpenCL)
+      return checkOpenCLVectorShift(*this, LHS, RHS, Loc, IsCompAssign);
     return CheckVectorOperands(LHS, RHS, Loc, IsCompAssign);
+  }
 
   // Shifts don't perform usual arithmetic conversions, they just do integer
   // promotions on each operand. C99 6.5.7p3
@@ -7807,8 +8192,7 @@ static bool hasIsEqualMethod(Sema &S, const Expr *LHS, const Expr *RHS) {
     if (Type->isObjCIdType()) {
       // For 'id', just check the global pool.
       Method = S.LookupInstanceMethodInGlobalPool(IsEqualSel, SourceRange(),
-                                                  /*receiverId=*/true,
-                                                  /*warn=*/false);
+                                                  /*receiverId=*/true);
     } else {
       // Check protocols.
       Method = S.LookupMethodInQualifiedType(IsEqualSel, Type,
@@ -8614,6 +8998,139 @@ static NonConstCaptureKind isReferenceToNonConstCapture(Sema &S, Expr *E) {
   return (isa<BlockDecl>(DC) ? NCCK_Block : NCCK_Lambda);
 }
 
+static bool IsTypeModifiable(QualType Ty, bool IsDereference) {
+  Ty = Ty.getNonReferenceType();
+  if (IsDereference && Ty->isPointerType())
+    Ty = Ty->getPointeeType();
+  return !Ty.isConstQualified();
+}
+
+/// Emit the "read-only variable not assignable" error and print notes to give
+/// more information about why the variable is not assignable, such as pointing
+/// to the declaration of a const variable, showing that a method is const, or
+/// that the function is returning a const reference.
+static void DiagnoseConstAssignment(Sema &S, const Expr *E,
+                                    SourceLocation Loc) {
+  // Update err_typecheck_assign_const and note_typecheck_assign_const
+  // when this enum is changed.
+  enum {
+    ConstFunction,
+    ConstVariable,
+    ConstMember,
+    ConstMethod,
+    ConstUnknown,  // Keep as last element
+  };
+
+  SourceRange ExprRange = E->getSourceRange();
+
+  // Only emit one error on the first const found.  All other consts will emit
+  // a note to the error.
+  bool DiagnosticEmitted = false;
+
+  // Track if the current expression is the result of a derefence, and if the
+  // next checked expression is the result of a derefence.
+  bool IsDereference = false;
+  bool NextIsDereference = false;
+
+  // Loop to process MemberExpr chains.
+  while (true) {
+    IsDereference = NextIsDereference;
+    NextIsDereference = false;
+
+    E = E->IgnoreParenImpCasts();
+    if (const MemberExpr *ME = dyn_cast<MemberExpr>(E)) {
+      NextIsDereference = ME->isArrow();
+      const ValueDecl *VD = ME->getMemberDecl();
+      if (const FieldDecl *Field = dyn_cast<FieldDecl>(VD)) {
+        // Mutable fields can be modified even if the class is const.
+        if (Field->isMutable()) {
+          assert(DiagnosticEmitted && "Expected diagnostic not emitted.");
+          break;
+        }
+
+        if (!IsTypeModifiable(Field->getType(), IsDereference)) {
+          if (!DiagnosticEmitted) {
+            S.Diag(Loc, diag::err_typecheck_assign_const)
+                << ExprRange << ConstMember << false /*static*/ << Field
+                << Field->getType();
+            DiagnosticEmitted = true;
+          }
+          S.Diag(VD->getLocation(), diag::note_typecheck_assign_const)
+              << ConstMember << false /*static*/ << Field << Field->getType()
+              << Field->getSourceRange();
+        }
+        E = ME->getBase();
+        continue;
+      } else if (const VarDecl *VDecl = dyn_cast<VarDecl>(VD)) {
+        if (VDecl->getType().isConstQualified()) {
+          if (!DiagnosticEmitted) {
+            S.Diag(Loc, diag::err_typecheck_assign_const)
+                << ExprRange << ConstMember << true /*static*/ << VDecl
+                << VDecl->getType();
+            DiagnosticEmitted = true;
+          }
+          S.Diag(VD->getLocation(), diag::note_typecheck_assign_const)
+              << ConstMember << true /*static*/ << VDecl << VDecl->getType()
+              << VDecl->getSourceRange();
+        }
+        // Static fields do not inherit constness from parents.
+        break;
+      }
+      break;
+    } // End MemberExpr
+    break;
+  }
+
+  if (const CallExpr *CE = dyn_cast<CallExpr>(E)) {
+    // Function calls
+    const FunctionDecl *FD = CE->getDirectCallee();
+    if (!IsTypeModifiable(FD->getReturnType(), IsDereference)) {
+      if (!DiagnosticEmitted) {
+        S.Diag(Loc, diag::err_typecheck_assign_const) << ExprRange
+                                                      << ConstFunction << FD;
+        DiagnosticEmitted = true;
+      }
+      S.Diag(FD->getReturnTypeSourceRange().getBegin(),
+             diag::note_typecheck_assign_const)
+          << ConstFunction << FD << FD->getReturnType()
+          << FD->getReturnTypeSourceRange();
+    }
+  } else if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) {
+    // Point to variable declaration.
+    if (const ValueDecl *VD = DRE->getDecl()) {
+      if (!IsTypeModifiable(VD->getType(), IsDereference)) {
+        if (!DiagnosticEmitted) {
+          S.Diag(Loc, diag::err_typecheck_assign_const)
+              << ExprRange << ConstVariable << VD << VD->getType();
+          DiagnosticEmitted = true;
+        }
+        S.Diag(VD->getLocation(), diag::note_typecheck_assign_const)
+            << ConstVariable << VD << VD->getType() << VD->getSourceRange();
+      }
+    }
+  } else if (isa<CXXThisExpr>(E)) {
+    if (const DeclContext *DC = S.getFunctionLevelDeclContext()) {
+      if (const CXXMethodDecl *MD = dyn_cast<CXXMethodDecl>(DC)) {
+        if (MD->isConst()) {
+          if (!DiagnosticEmitted) {
+            S.Diag(Loc, diag::err_typecheck_assign_const) << ExprRange
+                                                          << ConstMethod << MD;
+            DiagnosticEmitted = true;
+          }
+          S.Diag(MD->getLocation(), diag::note_typecheck_assign_const)
+              << ConstMethod << MD << MD->getSourceRange();
+        }
+      }
+    }
+  }
+
+  if (DiagnosticEmitted)
+    return;
+
+  // Can't determine a more specific message, so display the generic error.
+  S.Diag(Loc, diag::err_typecheck_assign_const) << ExprRange << ConstUnknown;
+}
+
 /// CheckForModifiableLvalue - Verify that E is a modifiable lvalue.  If not,
 /// emit an error and return true.  If so, return false.
 static bool CheckForModifiableLvalue(Expr *E, SourceLocation Loc, Sema &S) {
@@ -8630,8 +9147,6 @@ static bool CheckForModifiableLvalue(Expr *E, SourceLocation Loc, Sema &S) {
   bool NeedType = false;
   switch (IsLV) { // C99 6.5.16p2
   case Expr::MLV_ConstQualified:
-    DiagID = diag::err_typecheck_assign_const;
-
     // Use a specialized diagnostic when we're assigning to an object
     // from an enclosing function or block.
     if (NonConstCaptureKind NCCK = isReferenceToNonConstCapture(S, E)) {
@@ -8670,14 +9185,24 @@ static bool CheckForModifiableLvalue(Expr *E, SourceLocation Loc, Sema &S) {
           if (Loc != OrigLoc)
             Assign = SourceRange(OrigLoc, OrigLoc);
           S.Diag(Loc, DiagID) << E->getSourceRange() << Assign;
-          // We need to preserve the AST regardless, so migration tool 
+          // We need to preserve the AST regardless, so migration tool
           // can do its job.
           return false;
         }
       }
     }
 
+    // If none of the special cases above are triggered, then this is a
+    // simple const assignment.
+    if (DiagID == 0) {
+      DiagnoseConstAssignment(S, E, Loc);
+      return true;
+    }
+
     break;
+  case Expr::MLV_ConstAddrSpace:
+    DiagnoseConstAssignment(S, E, Loc);
+    return true;
   case Expr::MLV_ArrayType:
   case Expr::MLV_ArrayTemporary:
     DiagID = diag::err_typecheck_array_not_modifiable_lvalue;
@@ -9189,6 +9714,8 @@ QualType Sema::CheckAddressOfOperand(ExprResult &OrigOp, SourceLocation OpLoc) {
           !getLangOpts().CPlusPlus) {
         AddressOfError = AO_Register_Variable;
       }
+    } else if (isa<MSPropertyDecl>(dcl)) {
+      AddressOfError = AO_Property_Expansion;
     } else if (isa<FunctionTemplateDecl>(dcl)) {
       return Context.OverloadTy;
     } else if (isa<FieldDecl>(dcl) || isa<IndirectFieldDecl>(dcl)) {
@@ -11443,12 +11970,9 @@ void
 Sema::PushExpressionEvaluationContext(ExpressionEvaluationContext NewContext,
                                       Decl *LambdaContextDecl,
                                       bool IsDecltype) {
-  ExprEvalContexts.push_back(
-             ExpressionEvaluationContextRecord(NewContext,
-                                               ExprCleanupObjects.size(),
-                                               ExprNeedsCleanups,
-                                               LambdaContextDecl,
-                                               IsDecltype));
+  ExprEvalContexts.emplace_back(NewContext, ExprCleanupObjects.size(),
+                                ExprNeedsCleanups, LambdaContextDecl,
+                                IsDecltype);
   ExprNeedsCleanups = false;
   if (!MaybeODRUseExprs.empty())
     std::swap(MaybeODRUseExprs, ExprEvalContexts.back().SavedMaybeODRUseExprs);
@@ -11580,7 +12104,8 @@ void Sema::MarkFunctionReferenced(SourceLocation Loc, FunctionDecl *Func,
   // We (incorrectly) mark overload resolution as an unevaluated context, so we
   // can just check that here. Skip the rest of this function if we've already
   // marked the function as used.
-  if (Func->isUsed(false) || !IsPotentiallyEvaluatedContext(*this)) {
+  if (Func->isUsed(/*CheckUsedAttr=*/false) ||
+      !IsPotentiallyEvaluatedContext(*this)) {
     // C++11 [temp.inst]p3:
     //   Unless a function template specialization has been explicitly
     //   instantiated or explicitly specialized, the function template
@@ -11635,7 +12160,7 @@ void Sema::MarkFunctionReferenced(SourceLocation Loc, FunctionDecl *Func,
         return;
       DefineImplicitDestructor(Loc, Destructor);
     }
-    if (Destructor->isVirtual())
+    if (Destructor->isVirtual() && getLangOpts().AppleKext)
       MarkVTableUsed(Loc, Destructor->getParent());
   } else if (CXXMethodDecl *MethodDecl = dyn_cast<CXXMethodDecl>(Func)) {
     if (MethodDecl->isOverloadedOperator() &&
@@ -11655,7 +12180,7 @@ void Sema::MarkFunctionReferenced(SourceLocation Loc, FunctionDecl *Func,
         DefineImplicitLambdaToBlockPointerConversion(Loc, Conversion);
       else
         DefineImplicitLambdaToFunctionPointerConversion(Loc, Conversion);
-    } else if (MethodDecl->isVirtual())
+    } else if (MethodDecl->isVirtual() && getLangOpts().AppleKext)
       MarkVTableUsed(Loc, MethodDecl->getParent());
   }
 
@@ -11727,7 +12252,7 @@ void Sema::MarkFunctionReferenced(SourceLocation Loc, FunctionDecl *Func,
     if (mightHaveNonExternalLinkage(Func))
       UndefinedButUsed.insert(std::make_pair(Func->getCanonicalDecl(), Loc));
     else if (Func->getMostRecentDecl()->isInlined() &&
-             (LangOpts.CPlusPlus || !LangOpts.GNUInline) &&
+             !LangOpts.GNUInline &&
              !Func->getMostRecentDecl()->hasAttr<GNUInlineAttr>())
       UndefinedButUsed.insert(std::make_pair(Func->getCanonicalDecl(), Loc));
   }
@@ -12029,13 +12554,11 @@ static bool captureInCapturedRegion(CapturedRegionScopeInfo *RSI,
 }
 
 /// \brief Create a field within the lambda class for the variable
-///  being captured.  Handle Array captures.  
-static ExprResult addAsFieldToClosureType(Sema &S, 
-                                 LambdaScopeInfo *LSI,
-                                  VarDecl *Var, QualType FieldType, 
-                                  QualType DeclRefType,
-                                  SourceLocation Loc,
-                                  bool RefersToCapturedVariable) {
+/// being captured.
+static void addAsFieldToClosureType(Sema &S, LambdaScopeInfo *LSI, VarDecl *Var,
+                                    QualType FieldType, QualType DeclRefType,
+                                    SourceLocation Loc,
+                                    bool RefersToCapturedVariable) {
   CXXRecordDecl *Lambda = LSI->Lambda;
 
   // Build the non-static data member.
@@ -12046,111 +12569,8 @@ static ExprResult addAsFieldToClosureType(Sema &S,
   Field->setImplicit(true);
   Field->setAccess(AS_private);
   Lambda->addDecl(Field);
-
-  // C++11 [expr.prim.lambda]p21:
-  //   When the lambda-expression is evaluated, the entities that
-  //   are captured by copy are used to direct-initialize each
-  //   corresponding non-static data member of the resulting closure
-  //   object. (For array members, the array elements are
-  //   direct-initialized in increasing subscript order.) These
-  //   initializations are performed in the (unspecified) order in
-  //   which the non-static data members are declared.
-      
-  // Introduce a new evaluation context for the initialization, so
-  // that temporaries introduced as part of the capture are retained
-  // to be re-"exported" from the lambda expression itself.
-  EnterExpressionEvaluationContext scope(S, Sema::PotentiallyEvaluated);
-
-  // C++ [expr.prim.labda]p12:
-  //   An entity captured by a lambda-expression is odr-used (3.2) in
-  //   the scope containing the lambda-expression.
-  Expr *Ref = new (S.Context) DeclRefExpr(Var, RefersToCapturedVariable, 
-                                          DeclRefType, VK_LValue, Loc);
-  Var->setReferenced(true);
-  Var->markUsed(S.Context);
-
-  // When the field has array type, create index variables for each
-  // dimension of the array. We use these index variables to subscript
-  // the source array, and other clients (e.g., CodeGen) will perform
-  // the necessary iteration with these index variables.
-  SmallVector<VarDecl *, 4> IndexVariables;
-  QualType BaseType = FieldType;
-  QualType SizeType = S.Context.getSizeType();
-  LSI->ArrayIndexStarts.push_back(LSI->ArrayIndexVars.size());
-  while (const ConstantArrayType *Array
-                        = S.Context.getAsConstantArrayType(BaseType)) {
-    // Create the iteration variable for this array index.
-    IdentifierInfo *IterationVarName = nullptr;
-    {
-      SmallString<8> Str;
-      llvm::raw_svector_ostream OS(Str);
-      OS << "__i" << IndexVariables.size();
-      IterationVarName = &S.Context.Idents.get(OS.str());
-    }
-    VarDecl *IterationVar
-      = VarDecl::Create(S.Context, S.CurContext, Loc, Loc,
-                        IterationVarName, SizeType,
-                        S.Context.getTrivialTypeSourceInfo(SizeType, Loc),
-                        SC_None);
-    IndexVariables.push_back(IterationVar);
-    LSI->ArrayIndexVars.push_back(IterationVar);
-    
-    // Create a reference to the iteration variable.
-    ExprResult IterationVarRef
-      = S.BuildDeclRefExpr(IterationVar, SizeType, VK_LValue, Loc);
-    assert(!IterationVarRef.isInvalid() &&
-           "Reference to invented variable cannot fail!");
-    IterationVarRef = S.DefaultLvalueConversion(IterationVarRef.get());
-    assert(!IterationVarRef.isInvalid() &&
-           "Conversion of invented variable cannot fail!");
-    
-    // Subscript the array with this iteration variable.
-    ExprResult Subscript = S.CreateBuiltinArraySubscriptExpr(
-                             Ref, Loc, IterationVarRef.get(), Loc);
-    if (Subscript.isInvalid()) {
-      S.CleanupVarDeclMarking();
-      S.DiscardCleanupsInEvaluationContext();
-      return ExprError();
-    }
-
-    Ref = Subscript.get();
-    BaseType = Array->getElementType();
-  }
-
-  // Construct the entity that we will be initializing. For an array, this
-  // will be first element in the array, which may require several levels
-  // of array-subscript entities. 
-  SmallVector<InitializedEntity, 4> Entities;
-  Entities.reserve(1 + IndexVariables.size());
-  Entities.push_back(
-    InitializedEntity::InitializeLambdaCapture(Var->getIdentifier(), 
-        Field->getType(), Loc));
-  for (unsigned I = 0, N = IndexVariables.size(); I != N; ++I)
-    Entities.push_back(InitializedEntity::InitializeElement(S.Context,
-                                                            0,
-                                                            Entities.back()));
-
-  InitializationKind InitKind
-    = InitializationKind::CreateDirect(Loc, Loc, Loc);
-  InitializationSequence Init(S, Entities.back(), InitKind, Ref);
-  ExprResult Result(true);
-  if (!Init.Diagnose(S, Entities.back(), InitKind, Ref))
-    Result = Init.Perform(S, Entities.back(), InitKind, Ref);
-
-  // If this initialization requires any cleanups (e.g., due to a
-  // default argument to a copy constructor), note that for the
-  // lambda.
-  if (S.ExprNeedsCleanups)
-    LSI->ExprNeedsCleanups = true;
-
-  // Exit the expression evaluation context used for the capture.
-  S.CleanupVarDeclMarking();
-  S.DiscardCleanupsInEvaluationContext();
-  return Result;
 }
 
-
-
 /// \brief Capture the given variable in the lambda.
 static bool captureInLambda(LambdaScopeInfo *LSI,
                             VarDecl *Var, 
@@ -12228,14 +12648,9 @@ static bool captureInLambda(LambdaScopeInfo *LSI,
   }
 
   // Capture this variable in the lambda.
-  Expr *CopyExpr = nullptr;
-  if (BuildAndDiagnose) {
-    ExprResult Result = addAsFieldToClosureType(S, LSI, Var, 
-                                        CaptureType, DeclRefType, Loc,
-                                        RefersToCapturedVariable);
-    if (!Result.isInvalid())
-      CopyExpr = Result.get();
-  }
+  if (BuildAndDiagnose)
+    addAsFieldToClosureType(S, LSI, Var, CaptureType, DeclRefType, Loc,
+                            RefersToCapturedVariable);
     
   // Compute the type of a reference to this captured variable.
   if (ByRef)
@@ -12254,18 +12669,20 @@ static bool captureInLambda(LambdaScopeInfo *LSI,
   // Add the capture.
   if (BuildAndDiagnose)
     LSI->addCapture(Var, /*IsBlock=*/false, ByRef, RefersToCapturedVariable, 
-                    Loc, EllipsisLoc, CaptureType, CopyExpr);
+                    Loc, EllipsisLoc, CaptureType, /*CopyExpr=*/nullptr);
       
   return true;
 }
 
-bool Sema::tryCaptureVariable(VarDecl *Var, SourceLocation ExprLoc, 
-                              TryCaptureKind Kind, SourceLocation EllipsisLoc,
-                              bool BuildAndDiagnose, 
-                              QualType &CaptureType,
-                              QualType &DeclRefType,
-						                const unsigned *const FunctionScopeIndexToStopAt) {
-  bool Nested = Var->isInitCapture();
+bool Sema::tryCaptureVariable(
+    VarDecl *Var, SourceLocation ExprLoc, TryCaptureKind Kind,
+    SourceLocation EllipsisLoc, bool BuildAndDiagnose, QualType &CaptureType,
+    QualType &DeclRefType, const unsigned *const FunctionScopeIndexToStopAt) {
+  // An init-capture is notionally from the context surrounding its
+  // declaration, but its parent DC is the lambda class.
+  DeclContext *VarDC = Var->getDeclContext();
+  if (Var->isInitCapture())
+    VarDC = VarDC->getParent();
   
   DeclContext *DC = CurContext;
   const unsigned MaxFunctionScopesIndex = FunctionScopeIndexToStopAt 
@@ -12281,9 +12698,9 @@ bool Sema::tryCaptureVariable(VarDecl *Var, SourceLocation ExprLoc,
   }
 
   
-  // If the variable is declared in the current context (and is not an 
-  // init-capture), there is no need to capture it.
-  if (!Nested && Var->getDeclContext() == DC) return true;
+  // If the variable is declared in the current context, there is no need to
+  // capture it.
+  if (VarDC == DC) return true;
 
   // Capture global variables if it is required to use private copy of this
   // variable.
@@ -12301,6 +12718,7 @@ bool Sema::tryCaptureVariable(VarDecl *Var, SourceLocation ExprLoc,
   // the variable.
   CaptureType = Var->getType();
   DeclRefType = CaptureType.getNonReferenceType();
+  bool Nested = false;
   bool Explicit = (Kind != TryCapture_Implicit);
   unsigned FunctionScopesIndex = MaxFunctionScopesIndex;
   do {
@@ -12501,7 +12919,7 @@ bool Sema::tryCaptureVariable(VarDecl *Var, SourceLocation ExprLoc,
     FunctionScopesIndex--;
     DC = ParentDC;
     Explicit = false;
-  } while (!Var->getDeclContext()->Equals(DC));
+  } while (!VarDC->Equals(DC));
 
   // Walk back down the scope stack, (e.g. from outer lambda to inner lambda)
   // computing the type of the capture at each step, checking type-specific 
@@ -13740,7 +14158,17 @@ ExprResult Sema::CheckPlaceholderExpr(Expr *E) {
   // Bound member functions.
   case BuiltinType::BoundMember: {
     ExprResult result = E;
-    tryToRecoverWithCall(result, PDiag(diag::err_bound_member_function),
+    const Expr *BME = E->IgnoreParens();
+    PartialDiagnostic PD = PDiag(diag::err_bound_member_function);
+    // Try to give a nicer diagnostic if it is a bound member that we recognize.
+    if (isa<CXXPseudoDestructorExpr>(BME)) {
+      PD = PDiag(diag::err_dtor_expr_without_call) << /*pseudo-destructor*/ 1;
+    } else if (const auto *ME = dyn_cast<MemberExpr>(BME)) {
+      if (ME->getMemberNameInfo().getName().getNameKind() ==
+          DeclarationName::CXXDestructorName)
+        PD = PDiag(diag::err_dtor_expr_without_call) << /*destructor*/ 0;
+    }
+    tryToRecoverWithCall(result, PD,
                          /*complain*/ true);
     return result;
   }
diff --git a/lib/Sema/SemaExprCXX.cpp b/lib/Sema/SemaExprCXX.cpp
index a92b7ac472e7..7e305ffa8edc 100644
--- a/lib/Sema/SemaExprCXX.cpp
+++ b/lib/Sema/SemaExprCXX.cpp
@@ -113,6 +113,9 @@ ParsedType Sema::getDestructorName(SourceLocation TildeLoc,
   bool isDependent = false;
   bool LookInScope = false;
 
+  if (SS.isInvalid())
+    return ParsedType();
+
   // If we have an object type, it's because we are in a
   // pseudo-destructor-expression or a member access expression, and
   // we know what type we're looking for.
@@ -644,37 +647,98 @@ ExprResult Sema::BuildCXXThrow(SourceLocation OpLoc, Expr *Ex,
     Diag(OpLoc, diag::err_omp_simd_region_cannot_use_stmt) << "throw";
 
   if (Ex && !Ex->isTypeDependent()) {
-    ExprResult ExRes = CheckCXXThrowOperand(OpLoc, Ex, IsThrownVarInScope);
-    if (ExRes.isInvalid())
+    QualType ExceptionObjectTy = Context.getExceptionObjectType(Ex->getType());
+    if (CheckCXXThrowOperand(OpLoc, ExceptionObjectTy, Ex))
       return ExprError();
-    Ex = ExRes.get();
+
+    // Initialize the exception result.  This implicitly weeds out
+    // abstract types or types with inaccessible copy constructors.
+
+    // C++0x [class.copymove]p31:
+    //   When certain criteria are met, an implementation is allowed to omit the
+    //   copy/move construction of a class object [...]
+    //
+    //     - in a throw-expression, when the operand is the name of a
+    //       non-volatile automatic object (other than a function or
+    //       catch-clause
+    //       parameter) whose scope does not extend beyond the end of the
+    //       innermost enclosing try-block (if there is one), the copy/move
+    //       operation from the operand to the exception object (15.1) can be
+    //       omitted by constructing the automatic object directly into the
+    //       exception object
+    const VarDecl *NRVOVariable = nullptr;
+    if (IsThrownVarInScope)
+      NRVOVariable = getCopyElisionCandidate(QualType(), Ex, false);
+
+    InitializedEntity Entity = InitializedEntity::InitializeException(
+        OpLoc, ExceptionObjectTy,
+        /*NRVO=*/NRVOVariable != nullptr);
+    ExprResult Res = PerformMoveOrCopyInitialization(
+        Entity, NRVOVariable, QualType(), Ex, IsThrownVarInScope);
+    if (Res.isInvalid())
+      return ExprError();
+    Ex = Res.get();
   }
-  
+
   return new (Context)
       CXXThrowExpr(Ex, Context.VoidTy, OpLoc, IsThrownVarInScope);
 }
 
-/// CheckCXXThrowOperand - Validate the operand of a throw.
-ExprResult Sema::CheckCXXThrowOperand(SourceLocation ThrowLoc, Expr *E,
-                                      bool IsThrownVarInScope) {
-  // C++ [except.throw]p3:
-  //   A throw-expression initializes a temporary object, called the exception
-  //   object, the type of which is determined by removing any top-level
-  //   cv-qualifiers from the static type of the operand of throw and adjusting
-  //   the type from "array of T" or "function returning T" to "pointer to T"
-  //   or "pointer to function returning T", [...]
-  if (E->getType().hasQualifiers())
-    E = ImpCastExprToType(E, E->getType().getUnqualifiedType(), CK_NoOp,
-                          E->getValueKind()).get();
-
-  ExprResult Res = DefaultFunctionArrayConversion(E);
-  if (Res.isInvalid())
-    return ExprError();
-  E = Res.get();
+static void
+collectPublicBases(CXXRecordDecl *RD,
+                   llvm::DenseMap<CXXRecordDecl *, unsigned> &SubobjectsSeen,
+                   llvm::SmallPtrSetImpl<CXXRecordDecl *> &VBases,
+                   llvm::SetVector<CXXRecordDecl *> &PublicSubobjectsSeen,
+                   bool ParentIsPublic) {
+  for (const CXXBaseSpecifier &BS : RD->bases()) {
+    CXXRecordDecl *BaseDecl = BS.getType()->getAsCXXRecordDecl();
+    bool NewSubobject;
+    // Virtual bases constitute the same subobject.  Non-virtual bases are
+    // always distinct subobjects.
+    if (BS.isVirtual())
+      NewSubobject = VBases.insert(BaseDecl).second;
+    else
+      NewSubobject = true;
+
+    if (NewSubobject)
+      ++SubobjectsSeen[BaseDecl];
+
+    // Only add subobjects which have public access throughout the entire chain.
+    bool PublicPath = ParentIsPublic && BS.getAccessSpecifier() == AS_public;
+    if (PublicPath)
+      PublicSubobjectsSeen.insert(BaseDecl);
+
+    // Recurse on to each base subobject.
+    collectPublicBases(BaseDecl, SubobjectsSeen, VBases, PublicSubobjectsSeen,
+                       PublicPath);
+  }
+}
+
+static void getUnambiguousPublicSubobjects(
+    CXXRecordDecl *RD, llvm::SmallVectorImpl<CXXRecordDecl *> &Objects) {
+  llvm::DenseMap<CXXRecordDecl *, unsigned> SubobjectsSeen;
+  llvm::SmallSet<CXXRecordDecl *, 2> VBases;
+  llvm::SetVector<CXXRecordDecl *> PublicSubobjectsSeen;
+  SubobjectsSeen[RD] = 1;
+  PublicSubobjectsSeen.insert(RD);
+  collectPublicBases(RD, SubobjectsSeen, VBases, PublicSubobjectsSeen,
+                     /*ParentIsPublic=*/true);
+
+  for (CXXRecordDecl *PublicSubobject : PublicSubobjectsSeen) {
+    // Skip ambiguous objects.
+    if (SubobjectsSeen[PublicSubobject] > 1)
+      continue;
 
+    Objects.push_back(PublicSubobject);
+  }
+}
+
+/// CheckCXXThrowOperand - Validate the operand of a throw.
+bool Sema::CheckCXXThrowOperand(SourceLocation ThrowLoc,
+                                QualType ExceptionObjectTy, Expr *E) {
   //   If the type of the exception would be an incomplete type or a pointer
   //   to an incomplete type other than (cv) void the program is ill-formed.
-  QualType Ty = E->getType();
+  QualType Ty = ExceptionObjectTy;
   bool isPointer = false;
   if (const PointerType* Ptr = Ty->getAs<PointerType>()) {
     Ty = Ptr->getPointeeType();
@@ -682,49 +746,20 @@ ExprResult Sema::CheckCXXThrowOperand(SourceLocation ThrowLoc, Expr *E,
   }
   if (!isPointer || !Ty->isVoidType()) {
     if (RequireCompleteType(ThrowLoc, Ty,
-                            isPointer? diag::err_throw_incomplete_ptr
-                                     : diag::err_throw_incomplete,
+                            isPointer ? diag::err_throw_incomplete_ptr
+                                      : diag::err_throw_incomplete,
                             E->getSourceRange()))
-      return ExprError();
+      return true;
 
-    if (RequireNonAbstractType(ThrowLoc, E->getType(),
+    if (RequireNonAbstractType(ThrowLoc, ExceptionObjectTy,
                                diag::err_throw_abstract_type, E))
-      return ExprError();
+      return true;
   }
 
-  // Initialize the exception result.  This implicitly weeds out
-  // abstract types or types with inaccessible copy constructors.
-  
-  // C++0x [class.copymove]p31:
-  //   When certain criteria are met, an implementation is allowed to omit the 
-  //   copy/move construction of a class object [...]
-  //
-  //     - in a throw-expression, when the operand is the name of a 
-  //       non-volatile automatic object (other than a function or catch-clause 
-  //       parameter) whose scope does not extend beyond the end of the 
-  //       innermost enclosing try-block (if there is one), the copy/move 
-  //       operation from the operand to the exception object (15.1) can be 
-  //       omitted by constructing the automatic object directly into the 
-  //       exception object
-  const VarDecl *NRVOVariable = nullptr;
-  if (IsThrownVarInScope)
-    NRVOVariable = getCopyElisionCandidate(QualType(), E, false);
-
-  InitializedEntity Entity =
-      InitializedEntity::InitializeException(ThrowLoc, E->getType(),
-                                             /*NRVO=*/NRVOVariable != nullptr);
-  Res = PerformMoveOrCopyInitialization(Entity, NRVOVariable,
-                                        QualType(), E,
-                                        IsThrownVarInScope);
-  if (Res.isInvalid())
-    return ExprError();
-  E = Res.get();
-
   // If the exception has class type, we need additional handling.
-  const RecordType *RecordTy = Ty->getAs<RecordType>();
-  if (!RecordTy)
-    return E;
-  CXXRecordDecl *RD = cast<CXXRecordDecl>(RecordTy->getDecl());
+  CXXRecordDecl *RD = Ty->getAsCXXRecordDecl();
+  if (!RD)
+    return false;
 
   // If we are throwing a polymorphic class type or pointer thereof,
   // exception handling will make use of the vtable.
@@ -732,22 +767,69 @@ ExprResult Sema::CheckCXXThrowOperand(SourceLocation ThrowLoc, Expr *E,
 
   // If a pointer is thrown, the referenced object will not be destroyed.
   if (isPointer)
-    return E;
+    return false;
 
   // If the class has a destructor, we must be able to call it.
-  if (RD->hasIrrelevantDestructor())
-    return E;
+  if (!RD->hasIrrelevantDestructor()) {
+    if (CXXDestructorDecl *Destructor = LookupDestructor(RD)) {
+      MarkFunctionReferenced(E->getExprLoc(), Destructor);
+      CheckDestructorAccess(E->getExprLoc(), Destructor,
+                            PDiag(diag::err_access_dtor_exception) << Ty);
+      if (DiagnoseUseOfDecl(Destructor, E->getExprLoc()))
+        return true;
+    }
+  }
 
-  CXXDestructorDecl *Destructor = LookupDestructor(RD);
-  if (!Destructor)
-    return E;
+  // The MSVC ABI creates a list of all types which can catch the exception
+  // object.  This list also references the appropriate copy constructor to call
+  // if the object is caught by value and has a non-trivial copy constructor.
+  if (Context.getTargetInfo().getCXXABI().isMicrosoft()) {
+    // We are only interested in the public, unambiguous bases contained within
+    // the exception object.  Bases which are ambiguous or otherwise
+    // inaccessible are not catchable types.
+    llvm::SmallVector<CXXRecordDecl *, 2> UnambiguousPublicSubobjects;
+    getUnambiguousPublicSubobjects(RD, UnambiguousPublicSubobjects);
+
+    for (CXXRecordDecl *Subobject : UnambiguousPublicSubobjects) {
+      // Attempt to lookup the copy constructor.  Various pieces of machinery
+      // will spring into action, like template instantiation, which means this
+      // cannot be a simple walk of the class's decls.  Instead, we must perform
+      // lookup and overload resolution.
+      CXXConstructorDecl *CD = LookupCopyingConstructor(Subobject, 0);
+      if (!CD)
+        continue;
 
-  MarkFunctionReferenced(E->getExprLoc(), Destructor);
-  CheckDestructorAccess(E->getExprLoc(), Destructor,
-                        PDiag(diag::err_access_dtor_exception) << Ty);
-  if (DiagnoseUseOfDecl(Destructor, E->getExprLoc()))
-    return ExprError();
-  return E;
+      // Mark the constructor referenced as it is used by this throw expression.
+      MarkFunctionReferenced(E->getExprLoc(), CD);
+
+      // Skip this copy constructor if it is trivial, we don't need to record it
+      // in the catchable type data.
+      if (CD->isTrivial())
+        continue;
+
+      // The copy constructor is non-trivial, create a mapping from this class
+      // type to this constructor.
+      // N.B.  The selection of copy constructor is not sensitive to this
+      // particular throw-site.  Lookup will be performed at the catch-site to
+      // ensure that the copy constructor is, in fact, accessible (via
+      // friendship or any other means).
+      Context.addCopyConstructorForExceptionObject(Subobject, CD);
+
+      // We don't keep the instantiated default argument expressions around so
+      // we must rebuild them here.
+      for (unsigned I = 1, E = CD->getNumParams(); I != E; ++I) {
+        // Skip any default arguments that we've already instantiated.
+        if (Context.getDefaultArgExprForConstructor(CD, I))
+          continue;
+
+        Expr *DefaultArg =
+            BuildCXXDefaultArgExpr(ThrowLoc, CD, CD->getParamDecl(I)).get();
+        Context.addDefaultArgExprForConstructor(CD, I, DefaultArg);
+      }
+    }
+  }
+
+  return false;
 }
 
 QualType Sema::getCurrentThisType() {
@@ -982,18 +1064,21 @@ Sema::BuildCXXTypeConstructExpr(TypeSourceInfo *TInfo,
   Expr *Inner = Result.get();
   if (CXXBindTemporaryExpr *BTE = dyn_cast_or_null<CXXBindTemporaryExpr>(Inner))
     Inner = BTE->getSubExpr();
-  if (isa<InitListExpr>(Inner)) {
-    // If the list-initialization doesn't involve a constructor call, we'll get
-    // the initializer-list (with corrected type) back, but that's not what we
-    // want, since it will be treated as an initializer list in further
-    // processing. Explicitly insert a cast here.
+  if (!isa<CXXTemporaryObjectExpr>(Inner)) {
+    // If we created a CXXTemporaryObjectExpr, that node also represents the
+    // functional cast. Otherwise, create an explicit cast to represent
+    // the syntactic form of a functional-style cast that was used here.
+    //
+    // FIXME: Creating a CXXFunctionalCastExpr around a CXXConstructExpr
+    // would give a more consistent AST representation than using a
+    // CXXTemporaryObjectExpr. It's also weird that the functional cast
+    // is sometimes handled by initialization and sometimes not.
     QualType ResultType = Result.get()->getType();
     Result = CXXFunctionalCastExpr::Create(
         Context, ResultType, Expr::getValueKindForType(TInfo->getType()), TInfo,
         CK_NoOp, Result.get(), /*Path=*/nullptr, LParenLoc, RParenLoc);
   }
 
-  // FIXME: Improve AST representation?
   return Result;
 }
 
@@ -1333,8 +1418,9 @@ Sema::BuildCXXNew(SourceRange Range, bool UseGlobal,
                    << ConvTy->isEnumeralType() << ConvTy;
         }
 
-        virtual SemaDiagnosticBuilder diagnoseConversion(
-            Sema &S, SourceLocation Loc, QualType T, QualType ConvTy) override {
+        SemaDiagnosticBuilder diagnoseConversion(Sema &S, SourceLocation Loc,
+                                                 QualType T,
+                                                 QualType ConvTy) override {
           return S.Diag(Loc,
                         S.getLangOpts().CPlusPlus11
                           ? diag::warn_cxx98_compat_array_size_conversion
@@ -2048,7 +2134,7 @@ void Sema::DeclareGlobalNewDelete() {
 void Sema::DeclareGlobalAllocationFunction(DeclarationName Name,
                                            QualType Return,
                                            QualType Param1, QualType Param2,
-                                           bool AddMallocAttr) {
+                                           bool AddRestrictAttr) {
   DeclContext *GlobalCtx = Context.getTranslationUnitDecl();
   unsigned NumParams = Param2.isNull() ? 1 : 2;
 
@@ -2071,8 +2157,9 @@ void Sema::DeclareGlobalAllocationFunction(DeclarationName Name,
         // FIXME: Do we need to check for default arguments here?
         if (InitialParam1Type == Param1 &&
             (NumParams == 1 || InitialParam2Type == Param2)) {
-          if (AddMallocAttr && !Func->hasAttr<MallocAttr>())
-            Func->addAttr(MallocAttr::CreateImplicit(Context));
+          if (AddRestrictAttr && !Func->hasAttr<RestrictAttr>())
+            Func->addAttr(RestrictAttr::CreateImplicit(
+                Context, RestrictAttr::GNU_malloc));
           // Make the function visible to name lookup, even if we found it in
           // an unimported module. It either is an implicitly-declared global
           // allocation function, or is suppressing that function.
@@ -2110,9 +2197,14 @@ void Sema::DeclareGlobalAllocationFunction(DeclarationName Name,
                          SourceLocation(), Name,
                          FnType, /*TInfo=*/nullptr, SC_None, false, true);
   Alloc->setImplicit();
+  
+  // Implicit sized deallocation functions always have default visibility.
+  Alloc->addAttr(VisibilityAttr::CreateImplicit(Context,
+                                                VisibilityAttr::Default));
 
-  if (AddMallocAttr)
-    Alloc->addAttr(MallocAttr::CreateImplicit(Context));
+  if (AddRestrictAttr)
+    Alloc->addAttr(
+        RestrictAttr::CreateImplicit(Context, RestrictAttr::GNU_malloc));
 
   ParmVarDecl *ParamDecls[2];
   for (unsigned I = 0; I != NumParams; ++I) {
@@ -2247,6 +2339,260 @@ bool Sema::FindDeallocationFunction(SourceLocation StartLoc, CXXRecordDecl *RD,
   return false;
 }
 
+namespace {
+/// \brief Checks whether delete-expression, and new-expression used for
+///  initializing deletee have the same array form.
+class MismatchingNewDeleteDetector {
+public:
+  enum MismatchResult {
+    /// Indicates that there is no mismatch or a mismatch cannot be proven.
+    NoMismatch,
+    /// Indicates that variable is initialized with mismatching form of \a new.
+    VarInitMismatches,
+    /// Indicates that member is initialized with mismatching form of \a new.
+    MemberInitMismatches,
+    /// Indicates that 1 or more constructors' definitions could not been
+    /// analyzed, and they will be checked again at the end of translation unit.
+    AnalyzeLater
+  };
+
+  /// \param EndOfTU True, if this is the final analysis at the end of
+  /// translation unit. False, if this is the initial analysis at the point
+  /// delete-expression was encountered.
+  explicit MismatchingNewDeleteDetector(bool EndOfTU)
+      : IsArrayForm(false), Field(nullptr), EndOfTU(EndOfTU),
+        HasUndefinedConstructors(false) {}
+
+  /// \brief Checks whether pointee of a delete-expression is initialized with
+  /// matching form of new-expression.
+  ///
+  /// If return value is \c VarInitMismatches or \c MemberInitMismatches at the
+  /// point where delete-expression is encountered, then a warning will be
+  /// issued immediately. If return value is \c AnalyzeLater at the point where
+  /// delete-expression is seen, then member will be analyzed at the end of
+  /// translation unit. \c AnalyzeLater is returned iff at least one constructor
+  /// couldn't be analyzed. If at least one constructor initializes the member
+  /// with matching type of new, the return value is \c NoMismatch.
+  MismatchResult analyzeDeleteExpr(const CXXDeleteExpr *DE);
+  /// \brief Analyzes a class member.
+  /// \param Field Class member to analyze.
+  /// \param DeleteWasArrayForm Array form-ness of the delete-expression used
+  /// for deleting the \p Field.
+  MismatchResult analyzeField(FieldDecl *Field, bool DeleteWasArrayForm);
+  /// List of mismatching new-expressions used for initialization of the pointee
+  llvm::SmallVector<const CXXNewExpr *, 4> NewExprs;
+  /// Indicates whether delete-expression was in array form.
+  bool IsArrayForm;
+  FieldDecl *Field;
+
+private:
+  const bool EndOfTU;
+  /// \brief Indicates that there is at least one constructor without body.
+  bool HasUndefinedConstructors;
+  /// \brief Returns \c CXXNewExpr from given initialization expression.
+  /// \param E Expression used for initializing pointee in delete-expression.
+  /// E can be a single-element \c InitListExpr consisting of new-expression.
+  const CXXNewExpr *getNewExprFromInitListOrExpr(const Expr *E);
+  /// \brief Returns whether member is initialized with mismatching form of
+  /// \c new either by the member initializer or in-class initialization.
+  ///
+  /// If bodies of all constructors are not visible at the end of translation
+  /// unit or at least one constructor initializes member with the matching
+  /// form of \c new, mismatch cannot be proven, and this function will return
+  /// \c NoMismatch.
+  MismatchResult analyzeMemberExpr(const MemberExpr *ME);
+  /// \brief Returns whether variable is initialized with mismatching form of
+  /// \c new.
+  ///
+  /// If variable is initialized with matching form of \c new or variable is not
+  /// initialized with a \c new expression, this function will return true.
+  /// If variable is initialized with mismatching form of \c new, returns false.
+  /// \param D Variable to analyze.
+  bool hasMatchingVarInit(const DeclRefExpr *D);
+  /// \brief Checks whether the constructor initializes pointee with mismatching
+  /// form of \c new.
+  ///
+  /// Returns true, if member is initialized with matching form of \c new in
+  /// member initializer list. Returns false, if member is initialized with the
+  /// matching form of \c new in this constructor's initializer or given
+  /// constructor isn't defined at the point where delete-expression is seen, or
+  /// member isn't initialized by the constructor.
+  bool hasMatchingNewInCtor(const CXXConstructorDecl *CD);
+  /// \brief Checks whether member is initialized with matching form of
+  /// \c new in member initializer list.
+  bool hasMatchingNewInCtorInit(const CXXCtorInitializer *CI);
+  /// Checks whether member is initialized with mismatching form of \c new by
+  /// in-class initializer.
+  MismatchResult analyzeInClassInitializer();
+};
+}
+
+MismatchingNewDeleteDetector::MismatchResult
+MismatchingNewDeleteDetector::analyzeDeleteExpr(const CXXDeleteExpr *DE) {
+  NewExprs.clear();
+  assert(DE && "Expected delete-expression");
+  IsArrayForm = DE->isArrayForm();
+  const Expr *E = DE->getArgument()->IgnoreParenImpCasts();
+  if (const MemberExpr *ME = dyn_cast<const MemberExpr>(E)) {
+    return analyzeMemberExpr(ME);
+  } else if (const DeclRefExpr *D = dyn_cast<const DeclRefExpr>(E)) {
+    if (!hasMatchingVarInit(D))
+      return VarInitMismatches;
+  }
+  return NoMismatch;
+}
+
+const CXXNewExpr *
+MismatchingNewDeleteDetector::getNewExprFromInitListOrExpr(const Expr *E) {
+  assert(E != nullptr && "Expected a valid initializer expression");
+  E = E->IgnoreParenImpCasts();
+  if (const InitListExpr *ILE = dyn_cast<const InitListExpr>(E)) {
+    if (ILE->getNumInits() == 1)
+      E = dyn_cast<const CXXNewExpr>(ILE->getInit(0)->IgnoreParenImpCasts());
+  }
+
+  return dyn_cast_or_null<const CXXNewExpr>(E);
+}
+
+bool MismatchingNewDeleteDetector::hasMatchingNewInCtorInit(
+    const CXXCtorInitializer *CI) {
+  const CXXNewExpr *NE = nullptr;
+  if (Field == CI->getMember() &&
+      (NE = getNewExprFromInitListOrExpr(CI->getInit()))) {
+    if (NE->isArray() == IsArrayForm)
+      return true;
+    else
+      NewExprs.push_back(NE);
+  }
+  return false;
+}
+
+bool MismatchingNewDeleteDetector::hasMatchingNewInCtor(
+    const CXXConstructorDecl *CD) {
+  if (CD->isImplicit())
+    return false;
+  const FunctionDecl *Definition = CD;
+  if (!CD->isThisDeclarationADefinition() && !CD->isDefined(Definition)) {
+    HasUndefinedConstructors = true;
+    return EndOfTU;
+  }
+  for (const auto *CI : cast<const CXXConstructorDecl>(Definition)->inits()) {
+    if (hasMatchingNewInCtorInit(CI))
+      return true;
+  }
+  return false;
+}
+
+MismatchingNewDeleteDetector::MismatchResult
+MismatchingNewDeleteDetector::analyzeInClassInitializer() {
+  assert(Field != nullptr && "This should be called only for members");
+  if (const CXXNewExpr *NE =
+          getNewExprFromInitListOrExpr(Field->getInClassInitializer())) {
+    if (NE->isArray() != IsArrayForm) {
+      NewExprs.push_back(NE);
+      return MemberInitMismatches;
+    }
+  }
+  return NoMismatch;
+}
+
+MismatchingNewDeleteDetector::MismatchResult
+MismatchingNewDeleteDetector::analyzeField(FieldDecl *Field,
+                                           bool DeleteWasArrayForm) {
+  assert(Field != nullptr && "Analysis requires a valid class member.");
+  this->Field = Field;
+  IsArrayForm = DeleteWasArrayForm;
+  const CXXRecordDecl *RD = cast<const CXXRecordDecl>(Field->getParent());
+  for (const auto *CD : RD->ctors()) {
+    if (hasMatchingNewInCtor(CD))
+      return NoMismatch;
+  }
+  if (HasUndefinedConstructors)
+    return EndOfTU ? NoMismatch : AnalyzeLater;
+  if (!NewExprs.empty())
+    return MemberInitMismatches;
+  return Field->hasInClassInitializer() ? analyzeInClassInitializer()
+                                        : NoMismatch;
+}
+
+MismatchingNewDeleteDetector::MismatchResult
+MismatchingNewDeleteDetector::analyzeMemberExpr(const MemberExpr *ME) {
+  assert(ME != nullptr && "Expected a member expression");
+  if (FieldDecl *F = dyn_cast<FieldDecl>(ME->getMemberDecl()))
+    return analyzeField(F, IsArrayForm);
+  return NoMismatch;
+}
+
+bool MismatchingNewDeleteDetector::hasMatchingVarInit(const DeclRefExpr *D) {
+  const CXXNewExpr *NE = nullptr;
+  if (const VarDecl *VD = dyn_cast<const VarDecl>(D->getDecl())) {
+    if (VD->hasInit() && (NE = getNewExprFromInitListOrExpr(VD->getInit())) &&
+        NE->isArray() != IsArrayForm) {
+      NewExprs.push_back(NE);
+    }
+  }
+  return NewExprs.empty();
+}
+
+static void
+DiagnoseMismatchedNewDelete(Sema &SemaRef, SourceLocation DeleteLoc,
+                            const MismatchingNewDeleteDetector &Detector) {
+  SourceLocation EndOfDelete = SemaRef.getLocForEndOfToken(DeleteLoc);
+  FixItHint H;
+  if (!Detector.IsArrayForm)
+    H = FixItHint::CreateInsertion(EndOfDelete, "[]");
+  else {
+    SourceLocation RSquare = Lexer::findLocationAfterToken(
+        DeleteLoc, tok::l_square, SemaRef.getSourceManager(),
+        SemaRef.getLangOpts(), true);
+    if (RSquare.isValid())
+      H = FixItHint::CreateRemoval(SourceRange(EndOfDelete, RSquare));
+  }
+  SemaRef.Diag(DeleteLoc, diag::warn_mismatched_delete_new)
+      << Detector.IsArrayForm << H;
+
+  for (const auto *NE : Detector.NewExprs)
+    SemaRef.Diag(NE->getExprLoc(), diag::note_allocated_here)
+        << Detector.IsArrayForm;
+}
+
+void Sema::AnalyzeDeleteExprMismatch(const CXXDeleteExpr *DE) {
+  if (Diags.isIgnored(diag::warn_mismatched_delete_new, SourceLocation()))
+    return;
+  MismatchingNewDeleteDetector Detector(/*EndOfTU=*/false);
+  switch (Detector.analyzeDeleteExpr(DE)) {
+  case MismatchingNewDeleteDetector::VarInitMismatches:
+  case MismatchingNewDeleteDetector::MemberInitMismatches: {
+    DiagnoseMismatchedNewDelete(*this, DE->getLocStart(), Detector);
+    break;
+  }
+  case MismatchingNewDeleteDetector::AnalyzeLater: {
+    DeleteExprs[Detector.Field].push_back(
+        std::make_pair(DE->getLocStart(), DE->isArrayForm()));
+    break;
+  }
+  case MismatchingNewDeleteDetector::NoMismatch:
+    break;
+  }
+}
+
+void Sema::AnalyzeDeleteExprMismatch(FieldDecl *Field, SourceLocation DeleteLoc,
+                                     bool DeleteWasArrayForm) {
+  MismatchingNewDeleteDetector Detector(/*EndOfTU=*/true);
+  switch (Detector.analyzeField(Field, DeleteWasArrayForm)) {
+  case MismatchingNewDeleteDetector::VarInitMismatches:
+    llvm_unreachable("This analysis should have been done for class members.");
+  case MismatchingNewDeleteDetector::AnalyzeLater:
+    llvm_unreachable("Analysis cannot be postponed any point beyond end of "
+                     "translation unit.");
+  case MismatchingNewDeleteDetector::MemberInitMismatches:
+    DiagnoseMismatchedNewDelete(*this, DeleteLoc, Detector);
+    break;
+  case MismatchingNewDeleteDetector::NoMismatch:
+    break;
+  }
+}
+
 /// ActOnCXXDelete - Parsed a C++ 'delete' expression (C++ 5.3.5), as in:
 /// @code ::delete ptr; @endcode
 /// or
@@ -2362,12 +2708,6 @@ Sema::ActOnCXXDelete(SourceLocation StartLoc, bool UseGlobal,
       }
     }
 
-    // C++ [expr.delete]p2:
-    //   [Note: a pointer to a const type can be the operand of a
-    //   delete-expression; it is not necessary to cast away the constness
-    //   (5.2.11) of the pointer expression before it is used as the operand
-    //   of the delete-expression. ]
-
     if (Pointee->isArrayType() && !ArrayForm) {
       Diag(StartLoc, diag::warn_delete_array_type)
           << Type << Ex.get()->getSourceRange()
@@ -2442,7 +2782,7 @@ Sema::ActOnCXXDelete(SourceLocation StartLoc, bool UseGlobal,
           DeleteName);
 
     MarkFunctionReferenced(StartLoc, OperatorDelete);
-    
+
     // Check access and ambiguity of operator delete and destructor.
     if (PointeeRD) {
       if (CXXDestructorDecl *Dtor = LookupDestructor(PointeeRD)) {
@@ -2452,9 +2792,11 @@ Sema::ActOnCXXDelete(SourceLocation StartLoc, bool UseGlobal,
     }
   }
 
-  return new (Context) CXXDeleteExpr(
+  CXXDeleteExpr *Result = new (Context) CXXDeleteExpr(
       Context.VoidTy, UseGlobal, ArrayForm, ArrayFormAsWritten,
       UsualArrayDeleteWantsSize, OperatorDelete, Ex.get(), StartLoc);
+  AnalyzeDeleteExprMismatch(Result);
+  return Result;
 }
 
 /// \brief Check the use of the given variable as a C++ condition in an if,
@@ -2570,6 +2912,8 @@ static ExprResult BuildCXXCastArgument(Sema &S,
     S.CheckConstructorAccess(CastLoc, Constructor,
                              InitializedEntity::InitializeTemporary(Ty),
                              Constructor->getAccess());
+    if (S.DiagnoseUseOfDecl(Method, CastLoc))
+      return ExprError();
 
     ExprResult Result = S.BuildCXXConstructExpr(
         CastLoc, Ty, cast<CXXConstructorDecl>(Method),
@@ -2585,6 +2929,10 @@ static ExprResult BuildCXXCastArgument(Sema &S,
   case CK_UserDefinedConversion: {
     assert(!From->getType()->isPointerType() && "Arg can't have pointer type!");
 
+    S.CheckMemberOperatorAccess(CastLoc, From, /*arg*/ nullptr, FoundDecl);
+    if (S.DiagnoseUseOfDecl(Method, CastLoc))
+      return ExprError();
+
     // Create an implicit call expr that calls it.
     CXXConversionDecl *Conv = cast<CXXConversionDecl>(Method);
     ExprResult Result = S.BuildCXXMemberCallExpr(From, FoundDecl, Conv,
@@ -2596,8 +2944,6 @@ static ExprResult BuildCXXCastArgument(Sema &S,
                                       CK_UserDefinedConversion, Result.get(),
                                       nullptr, Result.get()->getValueKind());
 
-    S.CheckMemberOperatorAccess(CastLoc, From, /*arg*/ nullptr, FoundDecl);
-
     return S.MaybeBindToTemporary(Result.get());
   }
   }
@@ -2628,7 +2974,7 @@ Sema::PerformImplicitConversion(Expr *From, QualType ToType,
       FunctionDecl *FD = ICS.UserDefined.ConversionFunction;
       CastKind CastKind;
       QualType BeforeToType;
-      assert(FD && "FIXME: aggregate initialization from init list");
+      assert(FD && "no conversion function for user-defined conversion seq");
       if (const CXXConversionDecl *Conv = dyn_cast<CXXConversionDecl>(FD)) {
         CastKind = CK_UserDefinedConversion;
 
@@ -2989,8 +3335,18 @@ Sema::PerformImplicitConversion(Expr *From, QualType ToType,
     break;
 
   case ICK_Vector_Splat:
-    From = ImpCastExprToType(From, ToType, CK_VectorSplat, 
-                             VK_RValue, /*BasePath=*/nullptr, CCK).get();
+    // Vector splat from any arithmetic type to a vector.
+    // Cast to the element type.
+    {
+      QualType elType = ToType->getAs<ExtVectorType>()->getElementType();
+      if (elType != From->getType()) {
+        ExprResult E = From;
+        From = ImpCastExprToType(From, elType,
+                                 PrepareScalarCast(E, elType)).get();
+      }
+      From = ImpCastExprToType(From, ToType, CK_VectorSplat,
+                               VK_RValue, /*BasePath=*/nullptr, CCK).get();
+    }
     break;
 
   case ICK_Complex_Real:
@@ -3529,8 +3885,8 @@ static bool EvaluateUnaryTypeTrait(Sema &Self, TypeTrait UTT,
 
       bool FoundConstructor = false;
       unsigned FoundTQs;
-      DeclContext::lookup_const_result R = Self.LookupConstructors(RD);
-      for (DeclContext::lookup_const_iterator Con = R.begin(),
+      DeclContext::lookup_result R = Self.LookupConstructors(RD);
+      for (DeclContext::lookup_iterator Con = R.begin(),
            ConEnd = R.end(); Con != ConEnd; ++Con) {
         // A template constructor is never a copy constructor.
         // FIXME: However, it may actually be selected at the actual overload
@@ -3569,8 +3925,8 @@ static bool EvaluateUnaryTypeTrait(Sema &Self, TypeTrait UTT,
         return true;
 
       bool FoundConstructor = false;
-      DeclContext::lookup_const_result R = Self.LookupConstructors(RD);
-      for (DeclContext::lookup_const_iterator Con = R.begin(),
+      DeclContext::lookup_result R = Self.LookupConstructors(RD);
+      for (DeclContext::lookup_iterator Con = R.begin(),
            ConEnd = R.end(); Con != ConEnd; ++Con) {
         // FIXME: In C++0x, a constructor template can be a default constructor.
         if (isa<FunctionTemplateDecl>(*Con))
@@ -5139,7 +5495,7 @@ ExprResult Sema::ActOnDecltypeExpression(Expr *E) {
     if (Call == TopCall)
       continue;
 
-    if (CheckCallReturnType(Call->getCallReturnType(),
+    if (CheckCallReturnType(Call->getCallReturnType(Context),
                             Call->getLocStart(),
                             Call, Call->getDirectCallee()))
       return ExprError();
@@ -5202,10 +5558,11 @@ static void noteOperatorArrows(Sema &S,
   }
 }
 
-ExprResult
-Sema::ActOnStartCXXMemberReference(Scope *S, Expr *Base, SourceLocation OpLoc,
-                                   tok::TokenKind OpKind, ParsedType &ObjectType,
-                                   bool &MayBePseudoDestructor) {
+ExprResult Sema::ActOnStartCXXMemberReference(Scope *S, Expr *Base,
+                                              SourceLocation OpLoc,
+                                              tok::TokenKind OpKind,
+                                              ParsedType &ObjectType,
+                                              bool &MayBePseudoDestructor) {
   // Since this might be a postfix expression, get rid of ParenListExprs.
   ExprResult Result = MaybeConvertParenListExprToParenExpr(S, Base);
   if (Result.isInvalid()) return ExprError();
@@ -5339,20 +5696,6 @@ Sema::ActOnStartCXXMemberReference(Scope *S, Expr *Base, SourceLocation OpLoc,
   return Base;
 }
 
-ExprResult Sema::DiagnoseDtorReference(SourceLocation NameLoc,
-                                                   Expr *MemExpr) {
-  SourceLocation ExpectedLParenLoc = PP.getLocForEndOfToken(NameLoc);
-  Diag(MemExpr->getLocStart(), diag::err_dtor_expr_without_call)
-    << isa<CXXPseudoDestructorExpr>(MemExpr)
-    << FixItHint::CreateInsertion(ExpectedLParenLoc, "()");
-
-  return ActOnCallExpr(/*Scope*/ nullptr,
-                       MemExpr,
-                       /*LPLoc*/ ExpectedLParenLoc,
-                       None,
-                       /*RPLoc*/ ExpectedLParenLoc);
-}
-
 static bool CheckArrow(Sema& S, QualType& ObjectType, Expr *&Base, 
                    tok::TokenKind& OpKind, SourceLocation OpLoc) {
   if (Base->hasPlaceholderType()) {
@@ -5393,8 +5736,7 @@ ExprResult Sema::BuildPseudoDestructorExpr(Expr *Base,
                                            TypeSourceInfo *ScopeTypeInfo,
                                            SourceLocation CCLoc,
                                            SourceLocation TildeLoc,
-                                         PseudoDestructorTypeStorage Destructed,
-                                           bool HasTrailingLParen) {
+                                         PseudoDestructorTypeStorage Destructed) {
   TypeSourceInfo *DestructedTypeInfo = Destructed.getTypeSourceInfo();
 
   QualType ObjectType;
@@ -5482,10 +5824,7 @@ ExprResult Sema::BuildPseudoDestructorExpr(Expr *Base,
                                             TildeLoc,
                                             Destructed);
 
-  if (HasTrailingLParen)
-    return Result;
-
-  return DiagnoseDtorReference(Destructed.getLocation(), Result);
+  return Result;
 }
 
 ExprResult Sema::ActOnPseudoDestructorExpr(Scope *S, Expr *Base,
@@ -5495,8 +5834,7 @@ ExprResult Sema::ActOnPseudoDestructorExpr(Scope *S, Expr *Base,
                                            UnqualifiedId &FirstTypeName,
                                            SourceLocation CCLoc,
                                            SourceLocation TildeLoc,
-                                           UnqualifiedId &SecondTypeName,
-                                           bool HasTrailingLParen) {
+                                           UnqualifiedId &SecondTypeName) {
   assert((FirstTypeName.getKind() == UnqualifiedId::IK_TemplateId ||
           FirstTypeName.getKind() == UnqualifiedId::IK_Identifier) &&
          "Invalid first type name in pseudo-destructor");
@@ -5623,15 +5961,14 @@ ExprResult Sema::ActOnPseudoDestructorExpr(Scope *S, Expr *Base,
 
   return BuildPseudoDestructorExpr(Base, OpLoc, OpKind, SS,
                                    ScopeTypeInfo, CCLoc, TildeLoc,
-                                   Destructed, HasTrailingLParen);
+                                   Destructed);
 }
 
 ExprResult Sema::ActOnPseudoDestructorExpr(Scope *S, Expr *Base,
                                            SourceLocation OpLoc,
                                            tok::TokenKind OpKind,
                                            SourceLocation TildeLoc, 
-                                           const DeclSpec& DS,
-                                           bool HasTrailingLParen) {
+                                           const DeclSpec& DS) {
   QualType ObjectType;
   if (CheckArrow(*this, ObjectType, Base, OpKind, OpLoc))
     return ExprError();
@@ -5647,7 +5984,7 @@ ExprResult Sema::ActOnPseudoDestructorExpr(Scope *S, Expr *Base,
 
   return BuildPseudoDestructorExpr(Base, OpLoc, OpKind, CXXScopeSpec(),
                                    nullptr, SourceLocation(), TildeLoc,
-                                   Destructed, HasTrailingLParen);
+                                   Destructed);
 }
 
 ExprResult Sema::BuildCXXMemberCallExpr(Expr *E, NamedDecl *FoundDecl,
@@ -5685,10 +6022,9 @@ ExprResult Sema::BuildCXXMemberCallExpr(Expr *E, NamedDecl *FoundDecl,
   if (Exp.isInvalid())
     return true;
 
-  MemberExpr *ME =
-      new (Context) MemberExpr(Exp.get(), /*IsArrow=*/false, Method,
-                               SourceLocation(), Context.BoundMemberTy,
-                               VK_RValue, OK_Ordinary);
+  MemberExpr *ME = new (Context) MemberExpr(
+      Exp.get(), /*IsArrow=*/false, SourceLocation(), Method, SourceLocation(),
+      Context.BoundMemberTy, VK_RValue, OK_Ordinary);
   if (HadMultipleCandidates)
     ME->setHadMultipleCandidates(true);
   MarkMemberReferenced(ME);
@@ -6052,6 +6388,8 @@ public:
 class TransformTypos : public TreeTransform<TransformTypos> {
   typedef TreeTransform<TransformTypos> BaseTransform;
 
+  VarDecl *InitDecl; // A decl to avoid as a correction because it is in the
+                     // process of being initialized.
   llvm::function_ref<ExprResult(Expr *)> ExprFilter;
   llvm::SmallSetVector<TypoExpr *, 2> TypoExprs, AmbiguousTypoExprs;
   llvm::SmallDenseMap<TypoExpr *, ExprResult, 2> TransformCache;
@@ -6130,8 +6468,8 @@ class TransformTypos : public TreeTransform<TransformTypos> {
   }
 
 public:
-  TransformTypos(Sema &SemaRef, llvm::function_ref<ExprResult(Expr *)> Filter)
-      : BaseTransform(SemaRef), ExprFilter(Filter) {}
+  TransformTypos(Sema &SemaRef, VarDecl *InitDecl, llvm::function_ref<ExprResult(Expr *)> Filter)
+      : BaseTransform(SemaRef), InitDecl(InitDecl), ExprFilter(Filter) {}
 
   ExprResult RebuildCallExpr(Expr *Callee, SourceLocation LParenLoc,
                                    MultiExprArg Args,
@@ -6210,6 +6548,8 @@ public:
     // For the first TypoExpr and an uncached TypoExpr, find the next likely
     // typo correction and return it.
     while (TypoCorrection TC = State.Consumer->getNextCorrection()) {
+      if (InitDecl && TC.getCorrectionDecl() == InitDecl)
+        continue;
       ExprResult NE = State.RecoveryHandler ?
           State.RecoveryHandler(SemaRef, E, TC) :
           attemptRecovery(SemaRef, *State.Consumer, TC);
@@ -6234,8 +6574,9 @@ public:
 };
 }
 
-ExprResult Sema::CorrectDelayedTyposInExpr(
-    Expr *E, llvm::function_ref<ExprResult(Expr *)> Filter) {
+ExprResult
+Sema::CorrectDelayedTyposInExpr(Expr *E, VarDecl *InitDecl,
+                                llvm::function_ref<ExprResult(Expr *)> Filter) {
   // If the current evaluation context indicates there are uncorrected typos
   // and the current expression isn't guaranteed to not have typos, try to
   // resolve any TypoExpr nodes that might be in the expression.
@@ -6246,7 +6587,7 @@ ExprResult Sema::CorrectDelayedTyposInExpr(
     assert(TyposInContext < ~0U && "Recursive call of CorrectDelayedTyposInExpr");
     ExprEvalContexts.back().NumTypos = ~0U;
     auto TyposResolved = DelayedTypos.size();
-    auto Result = TransformTypos(*this, Filter).Transform(E);
+    auto Result = TransformTypos(*this, InitDecl, Filter).Transform(E);
     ExprEvalContexts.back().NumTypos = TyposInContext;
     TyposResolved -= DelayedTypos.size();
     if (Result.isInvalid() || Result.get() != E) {
diff --git a/lib/Sema/SemaExprMember.cpp b/lib/Sema/SemaExprMember.cpp
index af1cf9046113..e421349338b1 100644
--- a/lib/Sema/SemaExprMember.cpp
+++ b/lib/Sema/SemaExprMember.cpp
@@ -109,9 +109,8 @@ static IMAKind ClassifyImplicitMemberAccess(Sema &SemaRef,
     NamedDecl *D = *I;
 
     if (D->isCXXInstanceMember()) {
-      if (dyn_cast<FieldDecl>(D) || dyn_cast<MSPropertyDecl>(D)
-          || dyn_cast<IndirectFieldDecl>(D))
-        isField = true;
+      isField |= isa<FieldDecl>(D) || isa<MSPropertyDecl>(D) ||
+                 isa<IndirectFieldDecl>(D);
 
       CXXRecordDecl *R = cast<CXXRecordDecl>(D->getDeclContext());
       Classes.insert(R->getCanonicalDecl());
@@ -732,8 +731,8 @@ Sema::BuildMemberReferenceExpr(Expr *Base, QualType BaseType,
 
 static ExprResult
 BuildFieldReferenceExpr(Sema &S, Expr *BaseExpr, bool IsArrow,
-                        const CXXScopeSpec &SS, FieldDecl *Field,
-                        DeclAccessPair FoundDecl,
+                        SourceLocation OpLoc, const CXXScopeSpec &SS,
+                        FieldDecl *Field, DeclAccessPair FoundDecl,
                         const DeclarationNameInfo &MemberNameInfo);
 
 ExprResult
@@ -820,10 +819,10 @@ Sema::BuildAnonymousStructUnionMemberReference(const CXXScopeSpec &SS,
     
     // Make a nameInfo that properly uses the anonymous name.
     DeclarationNameInfo memberNameInfo(field->getDeclName(), loc);
-    
+
     result = BuildFieldReferenceExpr(*this, result, baseObjectIsPointer,
-                                     EmptySS, field, foundDecl,
-                                     memberNameInfo).get();
+                                     SourceLocation(), EmptySS, field,
+                                     foundDecl, memberNameInfo).get();
     if (!result)
       return ExprError();
 
@@ -841,9 +840,10 @@ Sema::BuildAnonymousStructUnionMemberReference(const CXXScopeSpec &SS,
     DeclAccessPair fakeFoundDecl =
         DeclAccessPair::make(field, field->getAccess());
 
-    result = BuildFieldReferenceExpr(*this, result, /*isarrow*/ false,
-                                     (FI == FEnd? SS : EmptySS), field,
-                                     fakeFoundDecl, memberNameInfo).get();
+    result =
+        BuildFieldReferenceExpr(*this, result, /*isarrow*/ false,
+                                SourceLocation(), (FI == FEnd ? SS : EmptySS),
+                                field, fakeFoundDecl, memberNameInfo).get();
   }
   
   return result;
@@ -863,18 +863,16 @@ BuildMSPropertyRefExpr(Sema &S, Expr *BaseExpr, bool IsArrow,
 }
 
 /// \brief Build a MemberExpr AST node.
-static MemberExpr *
-BuildMemberExpr(Sema &SemaRef, ASTContext &C, Expr *Base, bool isArrow,
-                const CXXScopeSpec &SS, SourceLocation TemplateKWLoc,
-                ValueDecl *Member, DeclAccessPair FoundDecl,
-                const DeclarationNameInfo &MemberNameInfo, QualType Ty,
-                ExprValueKind VK, ExprObjectKind OK,
-                const TemplateArgumentListInfo *TemplateArgs = nullptr) {
+static MemberExpr *BuildMemberExpr(
+    Sema &SemaRef, ASTContext &C, Expr *Base, bool isArrow,
+    SourceLocation OpLoc, const CXXScopeSpec &SS, SourceLocation TemplateKWLoc,
+    ValueDecl *Member, DeclAccessPair FoundDecl,
+    const DeclarationNameInfo &MemberNameInfo, QualType Ty, ExprValueKind VK,
+    ExprObjectKind OK, const TemplateArgumentListInfo *TemplateArgs = nullptr) {
   assert((!isArrow || Base->isRValue()) && "-> base must be a pointer rvalue");
-  MemberExpr *E =
-      MemberExpr::Create(C, Base, isArrow, SS.getWithLocInContext(C),
-                         TemplateKWLoc, Member, FoundDecl, MemberNameInfo,
-                         TemplateArgs, Ty, VK, OK);
+  MemberExpr *E = MemberExpr::Create(
+      C, Base, isArrow, OpLoc, SS.getWithLocInContext(C), TemplateKWLoc, Member,
+      FoundDecl, MemberNameInfo, TemplateArgs, Ty, VK, OK);
   SemaRef.MarkMemberReferenced(E);
   return E;
 }
@@ -969,8 +967,7 @@ Sema::BuildMemberReferenceExpr(Expr *BaseExpr, QualType BaseExprType,
           CXXScopeSpec TempSS(SS);
           RetryExpr = ActOnMemberAccessExpr(
               ExtraArgs->S, RetryExpr.get(), OpLoc, tok::arrow, TempSS,
-              TemplateKWLoc, ExtraArgs->Id, ExtraArgs->ObjCImpDecl,
-              ExtraArgs->HasTrailingLParen);
+              TemplateKWLoc, ExtraArgs->Id, ExtraArgs->ObjCImpDecl);
         }
         if (Trap.hasErrorOccurred())
           RetryExpr = ExprError();
@@ -1058,8 +1055,8 @@ Sema::BuildMemberReferenceExpr(Expr *BaseExpr, QualType BaseExprType,
     return ExprError();
 
   if (FieldDecl *FD = dyn_cast<FieldDecl>(MemberDecl))
-    return BuildFieldReferenceExpr(*this, BaseExpr, IsArrow,
-                                   SS, FD, FoundDecl, MemberNameInfo);
+    return BuildFieldReferenceExpr(*this, BaseExpr, IsArrow, OpLoc, SS, FD,
+                                   FoundDecl, MemberNameInfo);
 
   if (MSPropertyDecl *PD = dyn_cast<MSPropertyDecl>(MemberDecl))
     return BuildMSPropertyRefExpr(*this, BaseExpr, IsArrow, SS, PD,
@@ -1073,8 +1070,8 @@ Sema::BuildMemberReferenceExpr(Expr *BaseExpr, QualType BaseExprType,
                                                     OpLoc);
 
   if (VarDecl *Var = dyn_cast<VarDecl>(MemberDecl)) {
-    return BuildMemberExpr(*this, Context, BaseExpr, IsArrow, SS, TemplateKWLoc,
-                           Var, FoundDecl, MemberNameInfo,
+    return BuildMemberExpr(*this, Context, BaseExpr, IsArrow, OpLoc, SS,
+                           TemplateKWLoc, Var, FoundDecl, MemberNameInfo,
                            Var->getType().getNonReferenceType(), VK_LValue,
                            OK_Ordinary);
   }
@@ -1090,16 +1087,16 @@ Sema::BuildMemberReferenceExpr(Expr *BaseExpr, QualType BaseExprType,
       type = MemberFn->getType();
     }
 
-    return BuildMemberExpr(*this, Context, BaseExpr, IsArrow, SS, TemplateKWLoc,
-                           MemberFn, FoundDecl, MemberNameInfo, type, valueKind,
-                           OK_Ordinary);
+    return BuildMemberExpr(*this, Context, BaseExpr, IsArrow, OpLoc, SS,
+                           TemplateKWLoc, MemberFn, FoundDecl, MemberNameInfo,
+                           type, valueKind, OK_Ordinary);
   }
   assert(!isa<FunctionDecl>(MemberDecl) && "member function not C++ method?");
 
   if (EnumConstantDecl *Enum = dyn_cast<EnumConstantDecl>(MemberDecl)) {
-    return BuildMemberExpr(*this, Context, BaseExpr, IsArrow, SS, TemplateKWLoc,
-                           Enum, FoundDecl, MemberNameInfo, Enum->getType(),
-                           VK_RValue, OK_Ordinary);
+    return BuildMemberExpr(*this, Context, BaseExpr, IsArrow, OpLoc, SS,
+                           TemplateKWLoc, Enum, FoundDecl, MemberNameInfo,
+                           Enum->getType(), VK_RValue, OK_Ordinary);
   }
 
   // We found something that we didn't expect. Complain.
@@ -1521,7 +1518,15 @@ static ExprResult LookupMemberExpr(Sema &S, LookupResult &R,
   if (BaseType->isExtVectorType()) {
     // FIXME: this expr should store IsArrow.
     IdentifierInfo *Member = MemberName.getAsIdentifierInfo();
-    ExprValueKind VK = (IsArrow ? VK_LValue : BaseExpr.get()->getValueKind());
+    ExprValueKind VK;
+    if (IsArrow)
+      VK = VK_LValue;
+    else {
+      if (PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(BaseExpr.get()))
+        VK = POE->getSyntacticForm()->getValueKind();
+      else
+        VK = BaseExpr.get()->getValueKind();
+    }
     QualType ret = CheckExtVectorComponent(S, BaseType, VK, OpLoc,
                                            Member, MemberLoc);
     if (ret.isNull())
@@ -1591,9 +1596,6 @@ static ExprResult LookupMemberExpr(Sema &S, LookupResult &R,
 /// possibilities, including destructor and operator references.
 ///
 /// \param OpKind either tok::arrow or tok::period
-/// \param HasTrailingLParen whether the next token is '(', which
-///   is used to diagnose mis-uses of special members that can
-///   only be called
 /// \param ObjCImpDecl the current Objective-C \@implementation
 ///   decl; this is an ugly hack around the fact that Objective-C
 ///   \@implementations aren't properly put in the context chain
@@ -1603,24 +1605,10 @@ ExprResult Sema::ActOnMemberAccessExpr(Scope *S, Expr *Base,
                                        CXXScopeSpec &SS,
                                        SourceLocation TemplateKWLoc,
                                        UnqualifiedId &Id,
-                                       Decl *ObjCImpDecl,
-                                       bool HasTrailingLParen) {
+                                       Decl *ObjCImpDecl) {
   if (SS.isSet() && SS.isInvalid())
     return ExprError();
 
-  // The only way a reference to a destructor can be used is to
-  // immediately call it. If the next token is not a '(', produce
-  // a diagnostic and build the call now.
-  if (!HasTrailingLParen &&
-      Id.getKind() == UnqualifiedId::IK_DestructorName) {
-    ExprResult DtorAccess =
-        ActOnMemberAccessExpr(S, Base, OpLoc, OpKind, SS, TemplateKWLoc, Id,
-                              ObjCImpDecl, /*HasTrailingLParen*/true);
-    if (DtorAccess.isInvalid())
-      return DtorAccess;
-    return DiagnoseDtorReference(Id.getLocStart(), DtorAccess.get());
-  }
-
   // Warn about the explicit constructor calls Microsoft extension.
   if (getLangOpts().MicrosoftExt &&
       Id.getKind() == UnqualifiedId::IK_ConstructorName)
@@ -1653,8 +1641,7 @@ ExprResult Sema::ActOnMemberAccessExpr(Scope *S, Expr *Base,
                                     NameInfo, TemplateArgs);
   }
 
-  ActOnMemberAccessExtraArgs ExtraArgs = {S, Id, ObjCImpDecl,
-                                          HasTrailingLParen};
+  ActOnMemberAccessExtraArgs ExtraArgs = {S, Id, ObjCImpDecl};
   return BuildMemberReferenceExpr(Base, Base->getType(), OpLoc, IsArrow, SS,
                                   TemplateKWLoc, FirstQualifierInScope,
                                   NameInfo, TemplateArgs, &ExtraArgs);
@@ -1662,8 +1649,8 @@ ExprResult Sema::ActOnMemberAccessExpr(Scope *S, Expr *Base,
 
 static ExprResult
 BuildFieldReferenceExpr(Sema &S, Expr *BaseExpr, bool IsArrow,
-                        const CXXScopeSpec &SS, FieldDecl *Field,
-                        DeclAccessPair FoundDecl,
+                        SourceLocation OpLoc, const CXXScopeSpec &SS,
+                        FieldDecl *Field, DeclAccessPair FoundDecl,
                         const DeclarationNameInfo &MemberNameInfo) {
   // x.a is an l-value if 'a' has a reference type. Otherwise:
   // x.a is an l-value/x-value/pr-value if the base is (and note
@@ -1716,7 +1703,7 @@ BuildFieldReferenceExpr(Sema &S, Expr *BaseExpr, bool IsArrow,
                                   FoundDecl, Field);
   if (Base.isInvalid())
     return ExprError();
-  return BuildMemberExpr(S, S.Context, Base.get(), IsArrow, SS,
+  return BuildMemberExpr(S, S.Context, Base.get(), IsArrow, OpLoc, SS,
                          /*TemplateKWLoc=*/SourceLocation(), Field, FoundDecl,
                          MemberNameInfo, MemberType, VK, OK);
 }
diff --git a/lib/Sema/SemaExprObjC.cpp b/lib/Sema/SemaExprObjC.cpp
index 9c3b51c623d3..63b7485c8764 100644
--- a/lib/Sema/SemaExprObjC.cpp
+++ b/lib/Sema/SemaExprObjC.cpp
@@ -218,7 +218,9 @@ static ObjCMethodDecl *getNSNumberFactoryMethod(Sema &S, SourceLocation Loc,
       S.Diag(Loc, diag::err_undeclared_nsnumber);
       return nullptr;
     }
-    
+  }
+
+  if (S.NSNumberPointer.isNull()) {
     // generate the pointer to NSNumber type.
     QualType NSNumberObject = CX.getObjCInterfaceType(S.NSNumberDecl);
     S.NSNumberPointer = CX.getObjCObjectPointerType(NSNumberObject);
@@ -1041,7 +1043,7 @@ ExprResult Sema::ParseObjCSelectorExpression(Selector Sel,
                                              SourceLocation RParenLoc,
                                              bool WarnMultipleSelectors) {
   ObjCMethodDecl *Method = LookupInstanceMethodInGlobalPool(Sel,
-                             SourceRange(LParenLoc, RParenLoc), false, false);
+                             SourceRange(LParenLoc, RParenLoc));
   if (!Method)
     Method = LookupFactoryMethodInGlobalPool(Sel,
                                           SourceRange(LParenLoc, RParenLoc));
@@ -1059,15 +1061,11 @@ ExprResult Sema::ParseObjCSelectorExpression(Selector Sel,
   } else
     DiagnoseMismatchedSelectors(*this, AtLoc, Method, LParenLoc, RParenLoc,
                                 WarnMultipleSelectors);
-  
+
   if (Method &&
       Method->getImplementationControl() != ObjCMethodDecl::Optional &&
-      !getSourceManager().isInSystemHeader(Method->getLocation())) {
-    llvm::DenseMap<Selector, SourceLocation>::iterator Pos
-      = ReferencedSelectors.find(Sel);
-    if (Pos == ReferencedSelectors.end())
-      ReferencedSelectors.insert(std::make_pair(Sel, AtLoc));
-  }
+      !getSourceManager().isInSystemHeader(Method->getLocation()))
+    ReferencedSelectors.insert(std::make_pair(Sel, AtLoc));
 
   // In ARC, forbid the user from using @selector for 
   // retain/release/autorelease/dealloc/retainCount.
@@ -1507,64 +1505,6 @@ ObjCMethodDecl *Sema::LookupMethodInQualifiedType(Selector Sel,
   return nullptr;
 }
 
-static void DiagnoseARCUseOfWeakReceiver(Sema &S, Expr *Receiver) {
-  if (!Receiver)
-    return;
-  
-  if (OpaqueValueExpr *OVE = dyn_cast<OpaqueValueExpr>(Receiver))
-    Receiver = OVE->getSourceExpr();
-  
-  Expr *RExpr = Receiver->IgnoreParenImpCasts();
-  SourceLocation Loc = RExpr->getLocStart();
-  QualType T = RExpr->getType();
-  const ObjCPropertyDecl *PDecl = nullptr;
-  const ObjCMethodDecl *GDecl = nullptr;
-  if (PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(RExpr)) {
-    RExpr = POE->getSyntacticForm();
-    if (ObjCPropertyRefExpr *PRE = dyn_cast<ObjCPropertyRefExpr>(RExpr)) {
-      if (PRE->isImplicitProperty()) {
-        GDecl = PRE->getImplicitPropertyGetter();
-        if (GDecl) {
-          T = GDecl->getReturnType();
-        }
-      }
-      else {
-        PDecl = PRE->getExplicitProperty();
-        if (PDecl) {
-          T = PDecl->getType();
-        }
-      }
-    }
-  }
-  else if (ObjCMessageExpr *ME = dyn_cast<ObjCMessageExpr>(RExpr)) {
-    // See if receiver is a method which envokes a synthesized getter
-    // backing a 'weak' property.
-    ObjCMethodDecl *Method = ME->getMethodDecl();
-    if (Method && Method->getSelector().getNumArgs() == 0) {
-      PDecl = Method->findPropertyDecl();
-      if (PDecl)
-        T = PDecl->getType();
-    }
-  }
-  
-  if (T.getObjCLifetime() != Qualifiers::OCL_Weak) {
-    if (!PDecl)
-      return;
-    if (!(PDecl->getPropertyAttributes() & ObjCPropertyDecl::OBJC_PR_weak))
-      return;
-  }
-
-  S.Diag(Loc, diag::warn_receiver_is_weak)
-    << ((!PDecl && !GDecl) ? 0 : (PDecl ? 1 : 2));
-
-  if (PDecl)
-    S.Diag(PDecl->getLocation(), diag::note_property_declare);
-  else if (GDecl)
-    S.Diag(GDecl->getLocation(), diag::note_method_declared_at) << GDecl;
-
-  S.Diag(Loc, diag::note_arc_assign_to_strong);
-}
-
 /// HandleExprPropertyRefExpr - Handle foo.bar where foo is a pointer to an
 /// objective C interface.  This is a property reference expression.
 ExprResult Sema::
@@ -1751,29 +1691,30 @@ ActOnClassPropertyRefExpr(IdentifierInfo &receiverName,
       IsSuper = true;
 
       if (ObjCMethodDecl *CurMethod = tryCaptureObjCSelf(receiverNameLoc)) {
-        if (CurMethod->isInstanceMethod()) {
-          ObjCInterfaceDecl *Super =
-            CurMethod->getClassInterface()->getSuperClass();
-          if (!Super) {
-            // The current class does not have a superclass.
-            Diag(receiverNameLoc, diag::error_root_class_cannot_use_super)
-            << CurMethod->getClassInterface()->getIdentifier();
-            return ExprError();
+        if (ObjCInterfaceDecl *Class = CurMethod->getClassInterface()) {
+          if (CurMethod->isInstanceMethod()) {
+            ObjCInterfaceDecl *Super = Class->getSuperClass();
+            if (!Super) {
+              // The current class does not have a superclass.
+              Diag(receiverNameLoc, diag::error_root_class_cannot_use_super)
+              << Class->getIdentifier();
+              return ExprError();
+            }
+            QualType T = Context.getObjCInterfaceType(Super);
+            T = Context.getObjCObjectPointerType(T);
+
+            return HandleExprPropertyRefExpr(T->getAsObjCInterfacePointerType(),
+                                             /*BaseExpr*/nullptr,
+                                             SourceLocation()/*OpLoc*/,
+                                             &propertyName,
+                                             propertyNameLoc,
+                                             receiverNameLoc, T, true);
           }
-          QualType T = Context.getObjCInterfaceType(Super);
-          T = Context.getObjCObjectPointerType(T);
-
-          return HandleExprPropertyRefExpr(T->getAsObjCInterfacePointerType(),
-                                           /*BaseExpr*/nullptr,
-                                           SourceLocation()/*OpLoc*/, 
-                                           &propertyName,
-                                           propertyNameLoc,
-                                           receiverNameLoc, T, true);
-        }
 
-        // Otherwise, if this is a class method, try dispatching to our
-        // superclass.
-        IFace = CurMethod->getClassInterface()->getSuperClass();
+          // Otherwise, if this is a class method, try dispatching to our
+          // superclass.
+          IFace = Class->getSuperClass();
+        }
       }
     }
 
@@ -2452,8 +2393,11 @@ ExprResult Sema::BuildInstanceMessage(Expr *Receiver,
         if (ObjCMethodDecl *BestMethod =
               SelectBestMethod(Sel, ArgsIn, Method->isInstanceMethod()))
           Method = BestMethod;
-        if (!AreMultipleMethodsInGlobalPool(Sel, Method->isInstanceMethod()))
+        if (!AreMultipleMethodsInGlobalPool(Sel, Method,
+                                            SourceRange(LBracLoc, RBracLoc),
+                                            receiverIsId)) {
           DiagnoseUseOfDecl(Method, SelLoc);
+        }
       }
     } else if (ReceiverType->isObjCClassType() ||
                ReceiverType->isObjCQualifiedClassType()) {
@@ -2491,14 +2435,12 @@ ExprResult Sema::BuildInstanceMessage(Expr *Receiver,
           // If not messaging 'self', look for any factory method named 'Sel'.
           if (!Receiver || !isSelfExpr(Receiver)) {
             Method = LookupFactoryMethodInGlobalPool(Sel, 
-                                                SourceRange(LBracLoc, RBracLoc),
-                                                     true);
+                                                SourceRange(LBracLoc, RBracLoc));
             if (!Method) {
               // If no class (factory) method was found, check if an _instance_
               // method of the same name exists in the root class only.
               Method = LookupInstanceMethodInGlobalPool(Sel,
-                                               SourceRange(LBracLoc, RBracLoc),
-                                                        true);
+                                               SourceRange(LBracLoc, RBracLoc));
               if (Method)
                   if (const ObjCInterfaceDecl *ID =
                       dyn_cast<ObjCInterfaceDecl>(Method->getDeclContext())) {
@@ -2575,6 +2517,14 @@ ExprResult Sema::BuildInstanceMessage(Expr *Receiver,
             if (OCIType->qual_empty()) {
               Method = LookupInstanceMethodInGlobalPool(Sel,
                                               SourceRange(LBracLoc, RBracLoc));
+              if (Method) {
+                if (auto BestMethod =
+                      SelectBestMethod(Sel, ArgsIn, Method->isInstanceMethod()))
+                  Method = BestMethod;
+                AreMultipleMethodsInGlobalPool(Sel, Method,
+                                               SourceRange(LBracLoc, RBracLoc),
+                                               true);
+              }
               if (Method && !forwardClass)
                 Diag(SelLoc, diag::warn_maynot_respond)
                   << OCIType->getInterfaceDecl()->getIdentifier()
@@ -2757,15 +2707,6 @@ ExprResult Sema::BuildInstanceMessage(Expr *Receiver,
   }
 
   if (getLangOpts().ObjCAutoRefCount) {
-    // Do not warn about IBOutlet weak property receivers being set to null
-    // as this cannot asynchronously happen.
-    bool WarnWeakReceiver = true;
-    if (isImplicit && Method)
-      if (const ObjCPropertyDecl *PropertyDecl = Method->findPropertyDecl())
-        WarnWeakReceiver = !PropertyDecl->hasAttr<IBOutletAttr>();
-    if (WarnWeakReceiver)
-      DiagnoseARCUseOfWeakReceiver(*this, Receiver);
-    
     // In ARC, annotate delegate init calls.
     if (Result->getMethodFamily() == OMF_init &&
         (SuperLoc.isValid() || isSelfExpr(Receiver))) {
@@ -2796,7 +2737,9 @@ ExprResult Sema::BuildInstanceMessage(Expr *Receiver,
       }
     }
   }
-  
+
+  CheckObjCCircularContainer(Result);
+
   return MaybeBindToTemporary(Result);
 }
 
@@ -2805,8 +2748,7 @@ static void RemoveSelectorFromWarningCache(Sema &S, Expr* Arg) {
       dyn_cast<ObjCSelectorExpr>(Arg->IgnoreParenCasts())) {
     Selector Sel = OSE->getSelector();
     SourceLocation Loc = OSE->getAtLoc();
-    llvm::DenseMap<Selector, SourceLocation>::iterator Pos
-    = S.ReferencedSelectors.find(Sel);
+    auto Pos = S.ReferencedSelectors.find(Sel);
     if (Pos != S.ReferencedSelectors.end() && Pos->second == Loc)
       S.ReferencedSelectors.erase(Pos);
   }
@@ -3028,17 +2970,20 @@ namespace {
 
     /// Some declaration references are okay.
     ACCResult VisitDeclRefExpr(DeclRefExpr *e) {
-      // References to global constants from system headers are okay.
-      // These are things like 'kCFStringTransformToLatin'.  They are
-      // can also be assumed to be immune to retains.
       VarDecl *var = dyn_cast<VarDecl>(e->getDecl());
+      // References to global constants are okay.
       if (isAnyRetainable(TargetClass) &&
           isAnyRetainable(SourceClass) &&
           var &&
           var->getStorageClass() == SC_Extern &&
-          var->getType().isConstQualified() &&
-          Context.getSourceManager().isInSystemHeader(var->getLocation())) {
-        return ACC_bottom;
+          var->getType().isConstQualified()) {
+
+        // In system headers, they can also be assumed to be immune to retains.
+        // These are things like 'kCFStringTransformToLatin'.
+        if (Context.getSourceManager().isInSystemHeader(var->getLocation()))
+          return ACC_bottom;
+
+        return ACC_plusZero;
       }
 
       // Nothing else.
@@ -3421,7 +3366,7 @@ static bool CheckObjCBridgeNSCast(Sema &S, QualType castType, Expr *castExpr,
               ObjCInterfaceDecl *CastClass
                 = InterfacePointerType->getObjectType()->getInterface();
               if ((CastClass == ExprClass) ||
-                  (CastClass && ExprClass->isSuperClassOf(CastClass)))
+                  (CastClass && CastClass->isSuperClassOf(ExprClass)))
                 return true;
               if (warn)
                 S.Diag(castExpr->getLocStart(), diag::warn_objc_invalid_bridge)
@@ -3444,12 +3389,13 @@ static bool CheckObjCBridgeNSCast(Sema &S, QualType castType, Expr *castExpr,
               return false;
            }
           }
+        } else if (!castType->isObjCIdType()) {
+          S.Diag(castExpr->getLocStart(), diag::err_objc_cf_bridged_not_interface)
+            << castExpr->getType() << Parm;
+          S.Diag(TDNDecl->getLocStart(), diag::note_declared_at);
+          if (Target)
+            S.Diag(Target->getLocStart(), diag::note_declared_at);
         }
-        S.Diag(castExpr->getLocStart(), diag::err_objc_cf_bridged_not_interface)
-          << castExpr->getType() << Parm;
-        S.Diag(TDNDecl->getLocStart(), diag::note_declared_at);
-        if (Target)
-          S.Diag(Target->getLocStart(), diag::note_declared_at);
         return true;
       }
       return false;
@@ -3469,6 +3415,9 @@ static bool CheckObjCBridgeCFCast(Sema &S, QualType castType, Expr *castExpr,
     if (TB *ObjCBAttr = getObjCBridgeAttr<TB>(TD)) {
       if (IdentifierInfo *Parm = ObjCBAttr->getBridgedType()) {
         HadTheAttribute = true;
+        if (Parm->isStr("id"))
+          return true;
+
         NamedDecl *Target = nullptr;
         // Check for an existing type with this name.
         LookupResult R(S, DeclarationName(Parm), SourceLocation(),
diff --git a/lib/Sema/SemaFixItUtils.cpp b/lib/Sema/SemaFixItUtils.cpp
index 32b56bcddc63..2e327ecf231f 100644
--- a/lib/Sema/SemaFixItUtils.cpp
+++ b/lib/Sema/SemaFixItUtils.cpp
@@ -161,11 +161,8 @@ bool ConversionFixItGenerator::tryToFixConversion(const Expr *FullExpr,
 }
 
 static bool isMacroDefined(const Sema &S, SourceLocation Loc, StringRef Name) {
-  const IdentifierInfo *II = &S.getASTContext().Idents.get(Name);
-  if (!II->hadMacroDefinition()) return false;
-
-  MacroDirective *Macro = S.PP.getMacroDirectiveHistory(II);
-  return Macro && Macro->findDirectiveAtLoc(Loc, S.getSourceManager());
+  return (bool)S.PP.getMacroDefinitionAtLoc(&S.getASTContext().Idents.get(Name),
+                                            Loc);
 }
 
 static std::string getScalarZeroExpressionForType(
diff --git a/lib/Sema/SemaInit.cpp b/lib/Sema/SemaInit.cpp
index 569ef307474f..610e0a9a0389 100644
--- a/lib/Sema/SemaInit.cpp
+++ b/lib/Sema/SemaInit.cpp
@@ -149,9 +149,9 @@ static void updateStringLiteralType(Expr *E, QualType Ty) {
 static void CheckStringInit(Expr *Str, QualType &DeclT, const ArrayType *AT,
                             Sema &S) {
   // Get the length of the string as parsed.
-  uint64_t StrLength =
-    cast<ConstantArrayType>(Str->getType())->getSize().getZExtValue();
-
+  auto *ConstantArrayTy =
+      cast<ConstantArrayType>(Str->getType()->getAsArrayTypeUnsafe());
+  uint64_t StrLength = ConstantArrayTy->getSize().getZExtValue();
 
   if (const IncompleteArrayType *IAT = dyn_cast<IncompleteArrayType>(AT)) {
     // C99 6.7.8p14. We have an array of character type with unknown size
@@ -640,6 +640,9 @@ InitListChecker::InitListChecker(Sema &S, const InitializedEntity &Entity,
                                  InitListExpr *IL, QualType &T,
                                  bool VerifyOnly)
   : SemaRef(S), VerifyOnly(VerifyOnly) {
+  // FIXME: Check that IL isn't already the semantic form of some other
+  // InitListExpr. If it is, we'd create a broken AST.
+
   hadError = false;
 
   FullyStructuredList =
@@ -751,6 +754,68 @@ void InitListChecker::CheckImplicitInitList(const InitializedEntity &Entity,
   }
 }
 
+/// Warn that \p Entity was of scalar type and was initialized by a
+/// single-element braced initializer list.
+static void warnBracedScalarInit(Sema &S, const InitializedEntity &Entity,
+                                 SourceRange Braces) {
+  // Don't warn during template instantiation. If the initialization was
+  // non-dependent, we warned during the initial parse; otherwise, the
+  // type might not be scalar in some uses of the template.
+  if (!S.ActiveTemplateInstantiations.empty())
+    return;
+
+  unsigned DiagID = 0;
+
+  switch (Entity.getKind()) {
+  case InitializedEntity::EK_VectorElement:
+  case InitializedEntity::EK_ComplexElement:
+  case InitializedEntity::EK_ArrayElement:
+  case InitializedEntity::EK_Parameter:
+  case InitializedEntity::EK_Parameter_CF_Audited:
+  case InitializedEntity::EK_Result:
+    // Extra braces here are suspicious.
+    DiagID = diag::warn_braces_around_scalar_init;
+    break;
+
+  case InitializedEntity::EK_Member:
+    // Warn on aggregate initialization but not on ctor init list or
+    // default member initializer.
+    if (Entity.getParent())
+      DiagID = diag::warn_braces_around_scalar_init;
+    break;
+
+  case InitializedEntity::EK_Variable:
+  case InitializedEntity::EK_LambdaCapture:
+    // No warning, might be direct-list-initialization.
+    // FIXME: Should we warn for copy-list-initialization in these cases?
+    break;
+
+  case InitializedEntity::EK_New:
+  case InitializedEntity::EK_Temporary:
+  case InitializedEntity::EK_CompoundLiteralInit:
+    // No warning, braces are part of the syntax of the underlying construct.
+    break;
+
+  case InitializedEntity::EK_RelatedResult:
+    // No warning, we already warned when initializing the result.
+    break;
+
+  case InitializedEntity::EK_Exception:
+  case InitializedEntity::EK_Base:
+  case InitializedEntity::EK_Delegating:
+  case InitializedEntity::EK_BlockElement:
+    llvm_unreachable("unexpected braced scalar init");
+  }
+
+  if (DiagID) {
+    S.Diag(Braces.getBegin(), DiagID)
+      << Braces
+      << FixItHint::CreateRemoval(Braces.getBegin())
+      << FixItHint::CreateRemoval(Braces.getEnd());
+  }
+}
+
+
 /// Check whether the initializer \p IList (that was written with explicit
 /// braces) can be used to initialize an object of type \p T.
 ///
@@ -826,12 +891,9 @@ void InitListChecker::CheckExplicitInitList(const InitializedEntity &Entity,
     }
   }
 
-  if (!VerifyOnly && T->isScalarType() && IList->getNumInits() == 1 &&
-      !TopLevelObject)
-    SemaRef.Diag(IList->getLocStart(), diag::warn_braces_around_scalar_init)
-      << IList->getSourceRange()
-      << FixItHint::CreateRemoval(IList->getLocStart())
-      << FixItHint::CreateRemoval(IList->getLocEnd());
+  if (!VerifyOnly && T->isScalarType() &&
+      IList->getNumInits() == 1 && !isa<InitListExpr>(IList->getInit(0)))
+    warnBracedScalarInit(SemaRef, Entity, IList->getSourceRange());
 }
 
 void InitListChecker::CheckListElementTypes(const InitializedEntity &Entity,
@@ -904,7 +966,7 @@ void InitListChecker::CheckSubElementType(const InitializedEntity &Entity,
                               StructuredList, StructuredIndex);
 
   if (InitListExpr *SubInitList = dyn_cast<InitListExpr>(expr)) {
-    if (!ElemType->isRecordType() || ElemType->isAggregateType()) {
+    if (!SemaRef.getLangOpts().CPlusPlus) {
       InitListExpr *InnerStructuredList
         = getStructuredSubobjectInit(IList, Index, ElemType,
                                      StructuredList, StructuredIndex,
@@ -915,8 +977,6 @@ void InitListChecker::CheckSubElementType(const InitializedEntity &Entity,
       ++Index;
       return;
     }
-    assert(SemaRef.getLangOpts().CPlusPlus &&
-           "non-aggregate records are only possible in C++");
     // C++ initialization is handled later.
   } else if (isa<ImplicitValueInitExpr>(expr)) {
     // This happens during template instantiation when we see an InitListExpr
@@ -929,43 +989,24 @@ void InitListChecker::CheckSubElementType(const InitializedEntity &Entity,
     return;
   }
 
-  // FIXME: Need to handle atomic aggregate types with implicit init lists.
-  if (ElemType->isScalarType() || ElemType->isAtomicType())
-    return CheckScalarType(Entity, IList, ElemType, Index,
-                           StructuredList, StructuredIndex);
-
-  assert((ElemType->isRecordType() || ElemType->isVectorType() ||
-          ElemType->isArrayType()) && "Unexpected type");
-
-  if (const ArrayType *arrayType = SemaRef.Context.getAsArrayType(ElemType)) {
-    // arrayType can be incomplete if we're initializing a flexible
-    // array member.  There's nothing we can do with the completed
-    // type here, though.
-
-    if (IsStringInit(expr, arrayType, SemaRef.Context) == SIF_None) {
-      if (!VerifyOnly) {
-        CheckStringInit(expr, ElemType, arrayType, SemaRef);
-        UpdateStructuredListElement(StructuredList, StructuredIndex, expr);
-      }
-      ++Index;
-      return;
-    }
-
-    // Fall through for subaggregate initialization.
-
-  } else if (SemaRef.getLangOpts().CPlusPlus) {
-    // C++ [dcl.init.aggr]p12:
-    //   All implicit type conversions (clause 4) are considered when
-    //   initializing the aggregate member with an initializer from
-    //   an initializer-list. If the initializer can initialize a
-    //   member, the member is initialized. [...]
+  if (SemaRef.getLangOpts().CPlusPlus) {
+    // C++ [dcl.init.aggr]p2:
+    //   Each member is copy-initialized from the corresponding
+    //   initializer-clause.
 
     // FIXME: Better EqualLoc?
     InitializationKind Kind =
       InitializationKind::CreateCopy(expr->getLocStart(), SourceLocation());
-    InitializationSequence Seq(SemaRef, Entity, Kind, expr);
+    InitializationSequence Seq(SemaRef, Entity, Kind, expr,
+                               /*TopLevelOfInitList*/ true);
 
-    if (Seq) {
+    // C++14 [dcl.init.aggr]p13:
+    //   If the assignment-expression can initialize a member, the member is
+    //   initialized. Otherwise [...] brace elision is assumed
+    //
+    // Brace elision is never performed if the element is not an
+    // assignment-expression.
+    if (Seq || isa<InitListExpr>(expr)) {
       if (!VerifyOnly) {
         ExprResult Result =
           Seq.Perform(SemaRef, Entity, Kind, expr);
@@ -974,13 +1015,38 @@ void InitListChecker::CheckSubElementType(const InitializedEntity &Entity,
 
         UpdateStructuredListElement(StructuredList, StructuredIndex,
                                     Result.getAs<Expr>());
-      }
+      } else if (!Seq)
+        hadError = true;
       ++Index;
       return;
     }
 
     // Fall through for subaggregate initialization
+  } else if (ElemType->isScalarType() || ElemType->isAtomicType()) {
+    // FIXME: Need to handle atomic aggregate types with implicit init lists.
+    return CheckScalarType(Entity, IList, ElemType, Index,
+                           StructuredList, StructuredIndex);
+  } else if (const ArrayType *arrayType =
+                 SemaRef.Context.getAsArrayType(ElemType)) {
+    // arrayType can be incomplete if we're initializing a flexible
+    // array member.  There's nothing we can do with the completed
+    // type here, though.
+
+    if (IsStringInit(expr, arrayType, SemaRef.Context) == SIF_None) {
+      if (!VerifyOnly) {
+        CheckStringInit(expr, ElemType, arrayType, SemaRef);
+        UpdateStructuredListElement(StructuredList, StructuredIndex, expr);
+      }
+      ++Index;
+      return;
+    }
+
+    // Fall through for subaggregate initialization.
+
   } else {
+    assert((ElemType->isRecordType() || ElemType->isVectorType()) &&
+           "Unexpected type");
+
     // C99 6.7.8p13:
     //
     //   The initializer for a structure or union object that has
@@ -990,10 +1056,8 @@ void InitListChecker::CheckSubElementType(const InitializedEntity &Entity,
     //   initial value of the object, including unnamed members, is
     //   that of the expression.
     ExprResult ExprRes = expr;
-    if ((ElemType->isRecordType() || ElemType->isVectorType()) &&
-        SemaRef.CheckSingleAssignmentConstraints(ElemType, ExprRes,
-                                                 !VerifyOnly)
-          != Sema::Incompatible) {
+    if (SemaRef.CheckSingleAssignmentConstraints(
+            ElemType, ExprRes, !VerifyOnly) != Sema::Incompatible) {
       if (ExprRes.isInvalid())
         hadError = true;
       else {
@@ -3037,6 +3101,28 @@ void InitializationSequence::SetOverloadFailure(FailureKind Failure,
 // Attempt initialization
 //===----------------------------------------------------------------------===//
 
+/// Tries to add a zero initializer. Returns true if that worked.
+static bool
+maybeRecoverWithZeroInitialization(Sema &S, InitializationSequence &Sequence,
+                                   const InitializedEntity &Entity) {
+  if (Entity.getKind() != InitializedEntity::EK_Variable)
+    return false;
+
+  VarDecl *VD = cast<VarDecl>(Entity.getDecl());
+  if (VD->getInit() || VD->getLocEnd().isMacroID())
+    return false;
+
+  QualType VariableTy = VD->getType().getCanonicalType();
+  SourceLocation Loc = S.getLocForEndOfToken(VD->getLocEnd());
+  std::string Init = S.getFixItZeroInitializerForType(VariableTy, Loc);
+  if (!Init.empty()) {
+    Sequence.AddZeroInitializationStep(Entity.getType());
+    Sequence.SetZeroInitializationFixit(Init, Loc);
+    return true;
+  }
+  return false;
+}
+
 static void MaybeProduceObjCObject(Sema &S,
                                    InitializationSequence &Sequence,
                                    const InitializedEntity &Entity) {
@@ -3107,15 +3193,13 @@ static OverloadingResult
 ResolveConstructorOverload(Sema &S, SourceLocation DeclLoc,
                            MultiExprArg Args,
                            OverloadCandidateSet &CandidateSet,
-                           ArrayRef<NamedDecl *> Ctors,
+                           DeclContext::lookup_result Ctors,
                            OverloadCandidateSet::iterator &Best,
                            bool CopyInitializing, bool AllowExplicit,
-                           bool OnlyListConstructors, bool InitListSyntax) {
+                           bool OnlyListConstructors, bool IsListInit) {
   CandidateSet.clear();
 
-  for (ArrayRef<NamedDecl *>::iterator
-         Con = Ctors.begin(), ConEnd = Ctors.end(); Con != ConEnd; ++Con) {
-    NamedDecl *D = *Con;
+  for (NamedDecl *D : Ctors) {
     DeclAccessPair FoundDecl = DeclAccessPair::make(D, D->getAccess());
     bool SuppressUserConversions = false;
 
@@ -3129,19 +3213,21 @@ ResolveConstructorOverload(Sema &S, SourceLocation DeclLoc,
       Constructor = cast<CXXConstructorDecl>(D);
 
       // C++11 [over.best.ics]p4:
-      //   However, when considering the argument of a constructor or
-      //   user-defined conversion function that is a candidate:
-      //    -- by 13.3.1.3 when invoked for the copying/moving of a temporary
-      //       in the second step of a class copy-initialization,
-      //    -- by 13.3.1.7 when passing the initializer list as a single
-      //       argument or when the initializer list has exactly one elementand
-      //       a conversion to some class X or reference to (possibly
-      //       cv-qualified) X is considered for the first parameter of a
-      //       constructor of X, or
-      //    -- by 13.3.1.4, 13.3.1.5, or 13.3.1.6 in all cases,
-      //   only standard conversion sequences and ellipsis conversion sequences
-      //   are considered.
-      if ((CopyInitializing || (InitListSyntax && Args.size() == 1)) &&
+      //   ... and the constructor or user-defined conversion function is a
+      //   candidate by
+      //   - 13.3.1.3, when the argument is the temporary in the second step
+      //     of a class copy-initialization, or
+      //   - 13.3.1.4, 13.3.1.5, or 13.3.1.6 (in all cases),
+      //   user-defined conversion sequences are not considered.
+      // FIXME: This breaks backward compatibility, e.g. PR12117. As a
+      //        temporary fix, let's re-instate the third bullet above until
+      //        there is a resolution in the standard, i.e.,
+      //   - 13.3.1.7 when the initializer list has exactly one element that is
+      //     itself an initializer list and a conversion to some class X or
+      //     reference to (possibly cv-qualified) X is considered for the first
+      //     parameter of a constructor of X.
+      if ((CopyInitializing ||
+           (IsListInit && Args.size() == 1 && isa<InitListExpr>(Args[0]))) &&
           Constructor->isCopyOrMoveConstructor())
         SuppressUserConversions = true;
     }
@@ -3178,16 +3264,19 @@ ResolveConstructorOverload(Sema &S, SourceLocation DeclLoc,
 /// \brief Attempt initialization by constructor (C++ [dcl.init]), which
 /// enumerates the constructors of the initialized entity and performs overload
 /// resolution to select the best.
-/// If InitListSyntax is true, this is list-initialization of a non-aggregate
-/// class type.
+/// \param IsListInit     Is this list-initialization?
+/// \param IsInitListCopy Is this non-list-initialization resulting from a
+///                       list-initialization from {x} where x is the same
+///                       type as the entity?
 static void TryConstructorInitialization(Sema &S,
                                          const InitializedEntity &Entity,
                                          const InitializationKind &Kind,
                                          MultiExprArg Args, QualType DestType,
                                          InitializationSequence &Sequence,
-                                         bool InitListSyntax = false) {
-  assert((!InitListSyntax || (Args.size() == 1 && isa<InitListExpr>(Args[0]))) &&
-         "InitListSyntax must come with a single initializer list argument.");
+                                         bool IsListInit = false,
+                                         bool IsInitListCopy = false) {
+  assert((!IsListInit || (Args.size() == 1 && isa<InitListExpr>(Args[0]))) &&
+         "IsListInit must come with a single initializer list argument.");
 
   // The type we're constructing needs to be complete.
   if (S.RequireCompleteType(Kind.getLocation(), DestType, 0)) {
@@ -3206,29 +3295,28 @@ static void TryConstructorInitialization(Sema &S,
 
   // Determine whether we are allowed to call explicit constructors or
   // explicit conversion operators.
-  bool AllowExplicit = Kind.AllowExplicit() || InitListSyntax;
+  bool AllowExplicit = Kind.AllowExplicit() || IsListInit;
   bool CopyInitialization = Kind.getKind() == InitializationKind::IK_Copy;
 
   //   - Otherwise, if T is a class type, constructors are considered. The
   //     applicable constructors are enumerated, and the best one is chosen
   //     through overload resolution.
-  DeclContext::lookup_result R = S.LookupConstructors(DestRecordDecl);
-  // The container holding the constructors can under certain conditions
-  // be changed while iterating (e.g. because of deserialization).
-  // To be safe we copy the lookup results to a new container.
-  SmallVector<NamedDecl*, 16> Ctors(R.begin(), R.end());
+  DeclContext::lookup_result Ctors = S.LookupConstructors(DestRecordDecl);
 
   OverloadingResult Result = OR_No_Viable_Function;
   OverloadCandidateSet::iterator Best;
   bool AsInitializerList = false;
 
-  // C++11 [over.match.list]p1:
-  //   When objects of non-aggregate type T are list-initialized, overload
-  //   resolution selects the constructor in two phases:
+  // C++11 [over.match.list]p1, per DR1467:
+  //   When objects of non-aggregate type T are list-initialized, such that
+  //   8.5.4 [dcl.init.list] specifies that overload resolution is performed
+  //   according to the rules in this section, overload resolution selects
+  //   the constructor in two phases:
+  //
   //   - Initially, the candidate functions are the initializer-list
   //     constructors of the class T and the argument list consists of the
   //     initializer list as a single argument.
-  if (InitListSyntax) {
+  if (IsListInit) {
     InitListExpr *ILE = cast<InitListExpr>(Args[0]);
     AsInitializerList = true;
 
@@ -3239,7 +3327,7 @@ static void TryConstructorInitialization(Sema &S,
                                           CandidateSet, Ctors, Best,
                                           CopyInitialization, AllowExplicit,
                                           /*OnlyListConstructor=*/true,
-                                          InitListSyntax);
+                                          IsListInit);
 
     // Time to unwrap the init list.
     Args = MultiExprArg(ILE->getInits(), ILE->getNumInits());
@@ -3256,10 +3344,10 @@ static void TryConstructorInitialization(Sema &S,
                                         CandidateSet, Ctors, Best,
                                         CopyInitialization, AllowExplicit,
                                         /*OnlyListConstructors=*/false,
-                                        InitListSyntax);
+                                        IsListInit);
   }
   if (Result) {
-    Sequence.SetOverloadFailure(InitListSyntax ?
+    Sequence.SetOverloadFailure(IsListInit ?
                       InitializationSequence::FK_ListConstructorOverloadFailed :
                       InitializationSequence::FK_ConstructorOverloadFailed,
                                 Result);
@@ -3273,7 +3361,8 @@ static void TryConstructorInitialization(Sema &S,
   if (Kind.getKind() == InitializationKind::IK_Default &&
       Entity.getType().isConstQualified() &&
       !cast<CXXConstructorDecl>(Best->Function)->isUserProvided()) {
-    Sequence.SetFailed(InitializationSequence::FK_DefaultInitOfConst);
+    if (!maybeRecoverWithZeroInitialization(S, Sequence, Entity))
+      Sequence.SetFailed(InitializationSequence::FK_DefaultInitOfConst);
     return;
   }
 
@@ -3281,7 +3370,7 @@ static void TryConstructorInitialization(Sema &S,
   //   In copy-list-initialization, if an explicit constructor is chosen, the
   //   initializer is ill-formed.
   CXXConstructorDecl *CtorDecl = cast<CXXConstructorDecl>(Best->Function);
-  if (InitListSyntax && !Kind.AllowExplicit() && CtorDecl->isExplicit()) {
+  if (IsListInit && !Kind.AllowExplicit() && CtorDecl->isExplicit()) {
     Sequence.SetFailed(InitializationSequence::FK_ExplicitConstructor);
     return;
   }
@@ -3289,10 +3378,9 @@ static void TryConstructorInitialization(Sema &S,
   // Add the constructor initialization step. Any cv-qualification conversion is
   // subsumed by the initialization.
   bool HadMultipleCandidates = (CandidateSet.size() > 1);
-  Sequence.AddConstructorInitializationStep(CtorDecl,
-                                            Best->FoundDecl.getAccess(),
-                                            DestType, HadMultipleCandidates,
-                                            InitListSyntax, AsInitializerList);
+  Sequence.AddConstructorInitializationStep(
+      CtorDecl, Best->FoundDecl.getAccess(), DestType, HadMultipleCandidates,
+      IsListInit | IsInitListCopy, AsInitializerList);
 }
 
 static bool
@@ -3350,6 +3438,11 @@ static void TryReferenceListInitialization(Sema &S,
     Sequence.SetFailed(InitializationSequence::FK_ReferenceBindingToInitList);
     return;
   }
+  // Can't reference initialize a compound literal.
+  if (Entity.getKind() == InitializedEntity::EK_CompoundLiteralInit) {
+    Sequence.SetFailed(InitializationSequence::FK_ReferenceBindingToInitList);
+    return;
+  }
 
   QualType DestType = Entity.getType();
   QualType cv1T1 = DestType->getAs<ReferenceType>()->getPointeeType();
@@ -3425,48 +3518,97 @@ static void TryListInitialization(Sema &S,
     TryReferenceListInitialization(S, Entity, Kind, InitList, Sequence);
     return;
   }
-  if (DestType->isRecordType()) {
-    if (S.RequireCompleteType(InitList->getLocStart(), DestType, 0)) {
-      Sequence.setIncompleteTypeFailure(DestType);
-      return;
-    }
 
-    // C++11 [dcl.init.list]p3:
-    //   - If T is an aggregate, aggregate initialization is performed.
-    if (!DestType->isAggregateType()) {
-      if (S.getLangOpts().CPlusPlus11) {
-        //   - Otherwise, if the initializer list has no elements and T is a
-        //     class type with a default constructor, the object is
-        //     value-initialized.
-        if (InitList->getNumInits() == 0) {
-          CXXRecordDecl *RD = DestType->getAsCXXRecordDecl();
-          if (RD->hasDefaultConstructor()) {
-            TryValueInitialization(S, Entity, Kind, Sequence, InitList);
-            return;
-          }
+  if (DestType->isRecordType() &&
+      S.RequireCompleteType(InitList->getLocStart(), DestType, 0)) {
+    Sequence.setIncompleteTypeFailure(DestType);
+    return;
+  }
+
+  // C++11 [dcl.init.list]p3, per DR1467:
+  // - If T is a class type and the initializer list has a single element of
+  //   type cv U, where U is T or a class derived from T, the object is
+  //   initialized from that element (by copy-initialization for
+  //   copy-list-initialization, or by direct-initialization for
+  //   direct-list-initialization).
+  // - Otherwise, if T is a character array and the initializer list has a
+  //   single element that is an appropriately-typed string literal
+  //   (8.5.2 [dcl.init.string]), initialization is performed as described
+  //   in that section.
+  // - Otherwise, if T is an aggregate, [...] (continue below).
+  if (S.getLangOpts().CPlusPlus11 && InitList->getNumInits() == 1) {
+    if (DestType->isRecordType()) {
+      QualType InitType = InitList->getInit(0)->getType();
+      if (S.Context.hasSameUnqualifiedType(InitType, DestType) ||
+          S.IsDerivedFrom(InitType, DestType)) {
+        Expr *InitAsExpr = InitList->getInit(0);
+        TryConstructorInitialization(S, Entity, Kind, InitAsExpr, DestType,
+                                     Sequence, /*InitListSyntax*/ false,
+                                     /*IsInitListCopy*/ true);
+        return;
+      }
+    }
+    if (const ArrayType *DestAT = S.Context.getAsArrayType(DestType)) {
+      Expr *SubInit[1] = {InitList->getInit(0)};
+      if (!isa<VariableArrayType>(DestAT) &&
+          IsStringInit(SubInit[0], DestAT, S.Context) == SIF_None) {
+        InitializationKind SubKind =
+            Kind.getKind() == InitializationKind::IK_DirectList
+                ? InitializationKind::CreateDirect(Kind.getLocation(),
+                                                   InitList->getLBraceLoc(),
+                                                   InitList->getRBraceLoc())
+                : Kind;
+        Sequence.InitializeFrom(S, Entity, SubKind, SubInit,
+                                /*TopLevelOfInitList*/ true);
+
+        // TryStringLiteralInitialization() (in InitializeFrom()) will fail if
+        // the element is not an appropriately-typed string literal, in which
+        // case we should proceed as in C++11 (below).
+        if (Sequence) {
+          Sequence.RewrapReferenceInitList(Entity.getType(), InitList);
+          return;
         }
+      }
+    }
+  }
 
-        //   - Otherwise, if T is a specialization of std::initializer_list<E>,
-        //     an initializer_list object constructed [...]
-        if (TryInitializerListConstruction(S, InitList, DestType, Sequence))
+  // C++11 [dcl.init.list]p3:
+  //   - If T is an aggregate, aggregate initialization is performed.
+  if (DestType->isRecordType() && !DestType->isAggregateType()) {
+    if (S.getLangOpts().CPlusPlus11) {
+      //   - Otherwise, if the initializer list has no elements and T is a
+      //     class type with a default constructor, the object is
+      //     value-initialized.
+      if (InitList->getNumInits() == 0) {
+        CXXRecordDecl *RD = DestType->getAsCXXRecordDecl();
+        if (RD->hasDefaultConstructor()) {
+          TryValueInitialization(S, Entity, Kind, Sequence, InitList);
           return;
+        }
+      }
 
-        //   - Otherwise, if T is a class type, constructors are considered.
-        Expr *InitListAsExpr = InitList;
-        TryConstructorInitialization(S, Entity, Kind, InitListAsExpr, DestType,
-                                     Sequence, /*InitListSyntax*/true);
-      } else
-        Sequence.SetFailed(
-            InitializationSequence::FK_InitListBadDestinationType);
-      return;
-    }
+      //   - Otherwise, if T is a specialization of std::initializer_list<E>,
+      //     an initializer_list object constructed [...]
+      if (TryInitializerListConstruction(S, InitList, DestType, Sequence))
+        return;
+
+      //   - Otherwise, if T is a class type, constructors are considered.
+      Expr *InitListAsExpr = InitList;
+      TryConstructorInitialization(S, Entity, Kind, InitListAsExpr, DestType,
+                                   Sequence, /*InitListSyntax*/ true);
+    } else
+      Sequence.SetFailed(InitializationSequence::FK_InitListBadDestinationType);
+    return;
   }
+
   if (S.getLangOpts().CPlusPlus && !DestType->isAggregateType() &&
       InitList->getNumInits() == 1 &&
       InitList->getInit(0)->getType()->isRecordType()) {
     //   - Otherwise, if the initializer list has a single element of type E
     //     [...references are handled above...], the object or reference is
-    //     initialized from that element; if a narrowing conversion is required
+    //     initialized from that element (by copy-initialization for
+    //     copy-list-initialization, or by direct-initialization for
+    //     direct-list-initialization); if a narrowing conversion is required
     //     to convert the element to T, the program is ill-formed.
     //
     // Per core-24034, this is direct-initialization if we were performing
@@ -3543,14 +3685,7 @@ static OverloadingResult TryRefInitWithConversionFunction(Sema &S,
     // to see if there is a suitable conversion.
     CXXRecordDecl *T1RecordDecl = cast<CXXRecordDecl>(T1RecordType->getDecl());
 
-    DeclContext::lookup_result R = S.LookupConstructors(T1RecordDecl);
-    // The container holding the constructors can under certain conditions
-    // be changed while iterating (e.g. because of deserialization).
-    // To be safe we copy the lookup results to a new container.
-    SmallVector<NamedDecl*, 16> Ctors(R.begin(), R.end());
-    for (SmallVectorImpl<NamedDecl *>::iterator
-           CI = Ctors.begin(), CE = Ctors.end(); CI != CE; ++CI) {
-      NamedDecl *D = *CI;
+    for (NamedDecl *D : S.LookupConstructors(T1RecordDecl)) {
       DeclAccessPair FoundDecl = DeclAccessPair::make(D, D->getAccess());
 
       // Find the constructor (which may be a template).
@@ -3586,11 +3721,8 @@ static OverloadingResult TryRefInitWithConversionFunction(Sema &S,
     // functions.
     CXXRecordDecl *T2RecordDecl = cast<CXXRecordDecl>(T2RecordType->getDecl());
 
-    std::pair<CXXRecordDecl::conversion_iterator,
-              CXXRecordDecl::conversion_iterator>
-      Conversions = T2RecordDecl->getVisibleConversionFunctions();
-    for (CXXRecordDecl::conversion_iterator
-           I = Conversions.first, E = Conversions.second; I != E; ++I) {
+    const auto &Conversions = T2RecordDecl->getVisibleConversionFunctions();
+    for (auto I = Conversions.begin(), E = Conversions.end(); I != E; ++I) {
       NamedDecl *D = *I;
       CXXRecordDecl *ActingDC = cast<CXXRecordDecl>(D->getDeclContext());
       if (isa<UsingShadowDecl>(D))
@@ -4127,7 +4259,8 @@ static void TryDefaultInitialization(Sema &S,
   //   a const-qualified type T, T shall be a class type with a user-provided
   //   default constructor.
   if (DestType.isConstQualified() && S.getLangOpts().CPlusPlus) {
-    Sequence.SetFailed(InitializationSequence::FK_DefaultInitOfConst);
+    if (!maybeRecoverWithZeroInitialization(S, Sequence, Entity))
+      Sequence.SetFailed(InitializationSequence::FK_DefaultInitOfConst);
     return;
   }
 
@@ -4218,11 +4351,9 @@ static void TryUserDefinedConversion(Sema &S,
       CXXRecordDecl *SourceRecordDecl
         = cast<CXXRecordDecl>(SourceRecordType->getDecl());
 
-      std::pair<CXXRecordDecl::conversion_iterator,
-                CXXRecordDecl::conversion_iterator>
-        Conversions = SourceRecordDecl->getVisibleConversionFunctions();
-      for (CXXRecordDecl::conversion_iterator
-             I = Conversions.first, E = Conversions.second; I != E; ++I) {
+      const auto &Conversions =
+          SourceRecordDecl->getVisibleConversionFunctions();
+      for (auto I = Conversions.begin(), E = Conversions.end(); I != E; ++I) {
         NamedDecl *D = *I;
         CXXRecordDecl *ActingDC = cast<CXXRecordDecl>(D->getDeclContext());
         if (isa<UsingShadowDecl>(D))
@@ -4408,8 +4539,7 @@ static void checkIndirectCopyRestoreSource(Sema &S, Expr *src) {
 
 /// \brief Determine whether we have compatible array types for the
 /// purposes of GNU by-copy array initialization.
-static bool hasCompatibleArrayTypes(ASTContext &Context,
-                                    const ArrayType *Dest, 
+static bool hasCompatibleArrayTypes(ASTContext &Context, const ArrayType *Dest,
                                     const ArrayType *Source) {
   // If the source and destination array types are equivalent, we're
   // done.
@@ -4668,7 +4798,7 @@ void InitializationSequence::InitializeFrom(Sema &S,
     return;
   }
 
-  // Determine whether we should consider writeback conversions for 
+  // Determine whether we should consider writeback conversions for
   // Objective-C ARC.
   bool allowObjCWritebackConversion = S.getLangOpts().ObjCAutoRefCount &&
          Entity.isParameterKind();
@@ -5022,6 +5152,8 @@ static ExprResult CopyObject(Sema &S,
                              const InitializedEntity &Entity,
                              ExprResult CurInit,
                              bool IsExtraneousCopy) {
+  if (CurInit.isInvalid())
+    return CurInit;
   // Determine which class type we're copying to.
   Expr *CurInitExpr = (Expr *)CurInit.get();
   CXXRecordDecl *Class = nullptr;
@@ -5636,6 +5768,115 @@ static void DiagnoseNarrowingInInitList(Sema &S,
                                         QualType EntityType,
                                         const Expr *PostInit);
 
+/// Provide warnings when std::move is used on construction.
+static void CheckMoveOnConstruction(Sema &S, const Expr *InitExpr,
+                                    bool IsReturnStmt) {
+  if (!InitExpr)
+    return;
+
+  QualType DestType = InitExpr->getType();
+  if (!DestType->isRecordType())
+    return;
+
+  unsigned DiagID = 0;
+  if (IsReturnStmt) {
+    const CXXConstructExpr *CCE =
+        dyn_cast<CXXConstructExpr>(InitExpr->IgnoreParens());
+    if (!CCE || CCE->getNumArgs() != 1)
+      return;
+
+    if (!CCE->getConstructor()->isCopyOrMoveConstructor())
+      return;
+
+    InitExpr = CCE->getArg(0)->IgnoreImpCasts();
+
+    // Remove implicit temporary and constructor nodes.
+    if (const MaterializeTemporaryExpr *MTE =
+            dyn_cast<MaterializeTemporaryExpr>(InitExpr)) {
+      InitExpr = MTE->GetTemporaryExpr()->IgnoreImpCasts();
+      while (const CXXConstructExpr *CCE =
+                 dyn_cast<CXXConstructExpr>(InitExpr)) {
+        if (isa<CXXTemporaryObjectExpr>(CCE))
+          return;
+        if (CCE->getNumArgs() == 0)
+          return;
+        if (CCE->getNumArgs() > 1 && !isa<CXXDefaultArgExpr>(CCE->getArg(1)))
+          return;
+        InitExpr = CCE->getArg(0);
+      }
+      InitExpr = InitExpr->IgnoreImpCasts();
+      DiagID = diag::warn_redundant_move_on_return;
+    }
+  }
+
+  // Find the std::move call and get the argument.
+  const CallExpr *CE = dyn_cast<CallExpr>(InitExpr->IgnoreParens());
+  if (!CE || CE->getNumArgs() != 1)
+    return;
+
+  const FunctionDecl *MoveFunction = CE->getDirectCallee();
+  if (!MoveFunction || !MoveFunction->isInStdNamespace() ||
+      !MoveFunction->getIdentifier() ||
+      !MoveFunction->getIdentifier()->isStr("move"))
+    return;
+
+  const Expr *Arg = CE->getArg(0)->IgnoreImplicit();
+
+  if (IsReturnStmt) {
+    const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(Arg->IgnoreParenImpCasts());
+    if (!DRE || DRE->refersToEnclosingVariableOrCapture())
+      return;
+
+    const VarDecl *VD = dyn_cast<VarDecl>(DRE->getDecl());
+    if (!VD || !VD->hasLocalStorage())
+      return;
+
+    if (!VD->getType()->isRecordType())
+      return;
+
+    if (DiagID == 0) {
+      DiagID = S.Context.hasSameUnqualifiedType(DestType, VD->getType())
+                   ? diag::warn_pessimizing_move_on_return
+                   : diag::warn_redundant_move_on_return;
+    }
+  } else {
+    DiagID = diag::warn_pessimizing_move_on_initialization;
+    const Expr *ArgStripped = Arg->IgnoreImplicit()->IgnoreParens();
+    if (!ArgStripped->isRValue() || !ArgStripped->getType()->isRecordType())
+      return;
+  }
+
+  S.Diag(CE->getLocStart(), DiagID);
+
+  // Get all the locations for a fix-it.  Don't emit the fix-it if any location
+  // is within a macro.
+  SourceLocation CallBegin = CE->getCallee()->getLocStart();
+  if (CallBegin.isMacroID())
+    return;
+  SourceLocation RParen = CE->getRParenLoc();
+  if (RParen.isMacroID())
+    return;
+  SourceLocation LParen;
+  SourceLocation ArgLoc = Arg->getLocStart();
+
+  // Special testing for the argument location.  Since the fix-it needs the
+  // location right before the argument, the argument location can be in a
+  // macro only if it is at the beginning of the macro.
+  while (ArgLoc.isMacroID() &&
+         S.getSourceManager().isAtStartOfImmediateMacroExpansion(ArgLoc)) {
+    ArgLoc = S.getSourceManager().getImmediateExpansionRange(ArgLoc).first;
+  }
+
+  if (LParen.isMacroID())
+    return;
+
+  LParen = ArgLoc.getLocWithOffset(-1);
+
+  S.Diag(CE->getLocStart(), diag::note_remove_move)
+      << FixItHint::CreateRemoval(SourceRange(CallBegin, LParen))
+      << FixItHint::CreateRemoval(SourceRange(RParen, RParen));
+}
+
 ExprResult
 InitializationSequence::Perform(Sema &S,
                                 const InitializedEntity &Entity,
@@ -5646,6 +5887,21 @@ InitializationSequence::Perform(Sema &S,
     Diagnose(S, Entity, Kind, Args);
     return ExprError();
   }
+  if (!ZeroInitializationFixit.empty()) {
+    unsigned DiagID = diag::err_default_init_const;
+    if (Decl *D = Entity.getDecl())
+      if (S.getLangOpts().MSVCCompat && D->hasAttr<SelectAnyAttr>())
+        DiagID = diag::ext_default_init_const;
+
+    // The initialization would have succeeded with this fixit. Since the fixit
+    // is on the error, we need to build a valid AST in this case, so this isn't
+    // handled in the Failed() branch above.
+    QualType DestType = Entity.getType();
+    S.Diag(Kind.getLocation(), DiagID)
+        << DestType << (bool)DestType->getAs<RecordType>()
+        << FixItHint::CreateInsertion(ZeroInitializationFixitLoc,
+                                      ZeroInitializationFixit);
+  }
 
   if (getKind() == DependentSequence) {
     // If the declaration is a non-dependent, incomplete array type
@@ -5820,15 +6076,6 @@ InitializationSequence::Perform(Sema &S,
                                          &BasePath, IgnoreBaseAccess))
         return ExprError();
 
-      if (S.BasePathInvolvesVirtualBase(BasePath)) {
-        QualType T = SourceType;
-        if (const PointerType *Pointer = T->getAs<PointerType>())
-          T = Pointer->getPointeeType();
-        if (const RecordType *RecordTy = T->getAs<RecordType>())
-          S.MarkVTableUsed(CurInit.get()->getLocStart(),
-                           cast<CXXRecordDecl>(RecordTy->getDecl()));
-      }
-
       ExprValueKind VK =
           Step->Kind == SK_CastDerivedToBaseLValue ?
               VK_LValue :
@@ -6359,6 +6606,12 @@ InitializationSequence::Perform(Sema &S,
                                   cast<FieldDecl>(Entity.getDecl()),
                                   CurInit.get());
 
+  // Check for std::move on construction.
+  if (const Expr *E = CurInit.get()) {
+    CheckMoveOnConstruction(S, E,
+                            Entity.getKind() == InitializedEntity::EK_Result);
+  }
+
   return CurInit;
 }
 
@@ -6455,26 +6708,6 @@ static void diagnoseListInit(Sema &S, const InitializedEntity &Entity,
          "Inconsistent init list check result.");
 }
 
-/// Prints a fixit for adding a null initializer for |Entity|. Call this only
-/// right after emitting a diagnostic.
-static void maybeEmitZeroInitializationFixit(Sema &S,
-                                             InitializationSequence &Sequence,
-                                             const InitializedEntity &Entity) {
-  if (Entity.getKind() != InitializedEntity::EK_Variable)
-    return;
-
-  VarDecl *VD = cast<VarDecl>(Entity.getDecl());
-  if (VD->getInit() || VD->getLocEnd().isMacroID())
-    return;
-
-  QualType VariableTy = VD->getType().getCanonicalType();
-  SourceLocation Loc = S.getLocForEndOfToken(VD->getLocEnd());
-  std::string Init = S.getFixItZeroInitializerForType(VariableTy, Loc);
-
-  S.Diag(Loc, diag::note_add_initializer)
-      << VD << FixItHint::CreateInsertion(Loc, Init);
-}
-
 bool InitializationSequence::Diagnose(Sema &S,
                                       const InitializedEntity &Entity,
                                       const InitializationKind &Kind,
@@ -6622,12 +6855,19 @@ bool InitializationSequence::Diagnose(Sema &S,
       << Args[0]->getSourceRange();
     break;
 
-  case FK_ReferenceInitDropsQualifiers:
+  case FK_ReferenceInitDropsQualifiers: {
+    QualType SourceType = Args[0]->getType();
+    QualType NonRefType = DestType.getNonReferenceType();
+    Qualifiers DroppedQualifiers =
+        SourceType.getQualifiers() - NonRefType.getQualifiers();
+
     S.Diag(Kind.getLocation(), diag::err_reference_bind_drops_quals)
-      << DestType.getNonReferenceType()
-      << Args[0]->getType()
+      << SourceType
+      << NonRefType
+      << DroppedQualifiers.getCVRQualifiers()
       << Args[0]->getSourceRange();
     break;
+  }
 
   case FK_ReferenceInitFailed:
     S.Diag(Kind.getLocation(), diag::err_reference_bind_failed)
@@ -6659,9 +6899,9 @@ bool InitializationSequence::Diagnose(Sema &S,
   case FK_TooManyInitsForScalar: {
     SourceRange R;
 
-    if (InitListExpr *InitList = dyn_cast<InitListExpr>(Args[0]))
-      R = SourceRange(InitList->getInit(0)->getLocEnd(),
-                      InitList->getLocEnd());
+    auto *InitList = dyn_cast<InitListExpr>(Args[0]);
+    if (InitList && InitList->getNumInits() == 1)
+      R = SourceRange(InitList->getInit(0)->getLocEnd(), InitList->getLocEnd());
     else
       R = SourceRange(Args.front()->getLocEnd(), Args.back()->getLocEnd());
 
@@ -6806,7 +7046,6 @@ bool InitializationSequence::Diagnose(Sema &S,
     } else {
       S.Diag(Kind.getLocation(), diag::err_default_init_const)
           << DestType << (bool)DestType->getAs<RecordType>();
-      maybeEmitZeroInitializationFixit(S, *this, Entity);
     }
     break;
 
@@ -7262,7 +7501,6 @@ Sema::PerformCopyInitialization(const InitializedEntity &Entity,
                                                            EqualLoc,
                                                            AllowExplicit);
   InitializationSequence Seq(*this, Entity, Kind, InitE, TopLevelOfInitList);
-  Init.get();
 
   ExprResult Result = Seq.Perform(*this, Entity, Kind, InitE);
 
diff --git a/lib/Sema/SemaLambda.cpp b/lib/Sema/SemaLambda.cpp
index 90a81f4ec452..8220641166b2 100644
--- a/lib/Sema/SemaLambda.cpp
+++ b/lib/Sema/SemaLambda.cpp
@@ -818,7 +818,6 @@ VarDecl *Sema::createLambdaInitCaptureVarDecl(SourceLocation Loc,
   NewVD->markUsed(Context);
   NewVD->setInit(Init);
   return NewVD;
-
 }
 
 FieldDecl *Sema::buildInitCaptureField(LambdaScopeInfo *LSI, VarDecl *Var) {
@@ -837,7 +836,8 @@ FieldDecl *Sema::buildInitCaptureField(LambdaScopeInfo *LSI, VarDecl *Var) {
 }
 
 void Sema::ActOnStartOfLambdaDefinition(LambdaIntroducer &Intro,
-                  Declarator &ParamInfo, Scope *CurScope) {
+                                        Declarator &ParamInfo,
+                                        Scope *CurScope) {
   // Determine if we're within a context where we know that the lambda will
   // be dependent, because there are template parameters in scope.
   bool KnownDependent = false;
@@ -930,12 +930,8 @@ void Sema::ActOnStartOfLambdaDefinition(LambdaIntroducer &Intro,
   PushDeclContext(CurScope, Method);
     
   // Build the lambda scope.
-  buildLambdaScope(LSI, Method,
-                       Intro.Range,
-                       Intro.Default, Intro.DefaultLoc,
-                       ExplicitParams,
-                       ExplicitResultType,
-                       !Method->isConst());
+  buildLambdaScope(LSI, Method, Intro.Range, Intro.Default, Intro.DefaultLoc,
+                   ExplicitParams, ExplicitResultType, !Method->isConst());
 
   // C++11 [expr.prim.lambda]p9:
   //   A lambda-expression whose smallest enclosing scope is a block scope is a
@@ -1137,7 +1133,7 @@ void Sema::ActOnStartOfLambdaDefinition(LambdaIntroducer &Intro,
 
 void Sema::ActOnLambdaError(SourceLocation StartLoc, Scope *CurScope,
                             bool IsInstantiation) {
-  LambdaScopeInfo *LSI = getCurLambda();
+  LambdaScopeInfo *LSI = cast<LambdaScopeInfo>(FunctionScopes.back());
 
   // Leave the expression-evaluation context.
   DiscardCleanupsInEvaluationContext();
@@ -1340,22 +1336,27 @@ static void addBlockPointerConversion(Sema &S,
                                       SourceRange IntroducerRange,
                                       CXXRecordDecl *Class,
                                       CXXMethodDecl *CallOperator) {
-  const FunctionProtoType *Proto
-    = CallOperator->getType()->getAs<FunctionProtoType>(); 
-  QualType BlockPtrTy;
-  {
-    FunctionProtoType::ExtProtoInfo ExtInfo = Proto->getExtProtoInfo();
-    ExtInfo.TypeQuals = 0;
-    QualType FunctionTy = S.Context.getFunctionType(
-        Proto->getReturnType(), Proto->getParamTypes(), ExtInfo);
-    BlockPtrTy = S.Context.getBlockPointerType(FunctionTy);
-  }
+  const FunctionProtoType *Proto =
+      CallOperator->getType()->getAs<FunctionProtoType>();
+
+  // The function type inside the block pointer type is the same as the call
+  // operator with some tweaks. The calling convention is the default free
+  // function convention, and the type qualifications are lost.
+  FunctionProtoType::ExtProtoInfo BlockEPI = Proto->getExtProtoInfo();
+  BlockEPI.ExtInfo =
+      BlockEPI.ExtInfo.withCallingConv(S.Context.getDefaultCallingConvention(
+          Proto->isVariadic(), /*IsCXXMethod=*/false));
+  BlockEPI.TypeQuals = 0;
+  QualType FunctionTy = S.Context.getFunctionType(
+      Proto->getReturnType(), Proto->getParamTypes(), BlockEPI);
+  QualType BlockPtrTy = S.Context.getBlockPointerType(FunctionTy);
+
+  FunctionProtoType::ExtProtoInfo ConversionEPI(
+      S.Context.getDefaultCallingConvention(
+          /*IsVariadic=*/false, /*IsCXXMethod=*/true));
+  ConversionEPI.TypeQuals = Qualifiers::Const;
+  QualType ConvTy = S.Context.getFunctionType(BlockPtrTy, None, ConversionEPI);
 
-  FunctionProtoType::ExtProtoInfo ExtInfo(S.Context.getDefaultCallingConvention(
-      /*IsVariadic=*/false, /*IsCXXMethod=*/true));
-  ExtInfo.TypeQuals = Qualifiers::Const;
-  QualType ConvTy = S.Context.getFunctionType(BlockPtrTy, None, ExtInfo);
-  
   SourceLocation Loc = IntroducerRange.getBegin();
   DeclarationName Name
     = S.Context.DeclarationNames.getCXXConversionFunctionName(
@@ -1374,15 +1375,131 @@ static void addBlockPointerConversion(Sema &S,
   Conversion->setImplicit(true);
   Class->addDecl(Conversion);
 }
+
+static ExprResult performLambdaVarCaptureInitialization(
+    Sema &S, LambdaScopeInfo::Capture &Capture,
+    FieldDecl *Field,
+    SmallVectorImpl<VarDecl *> &ArrayIndexVars,
+    SmallVectorImpl<unsigned> &ArrayIndexStarts) {
+  assert(Capture.isVariableCapture() && "not a variable capture");
+
+  auto *Var = Capture.getVariable();
+  SourceLocation Loc = Capture.getLocation();
+
+  // C++11 [expr.prim.lambda]p21:
+  //   When the lambda-expression is evaluated, the entities that
+  //   are captured by copy are used to direct-initialize each
+  //   corresponding non-static data member of the resulting closure
+  //   object. (For array members, the array elements are
+  //   direct-initialized in increasing subscript order.) These
+  //   initializations are performed in the (unspecified) order in
+  //   which the non-static data members are declared.
+      
+  // C++ [expr.prim.lambda]p12:
+  //   An entity captured by a lambda-expression is odr-used (3.2) in
+  //   the scope containing the lambda-expression.
+  ExprResult RefResult = S.BuildDeclarationNameExpr(
+      CXXScopeSpec(), DeclarationNameInfo(Var->getDeclName(), Loc), Var);
+  if (RefResult.isInvalid())
+    return ExprError();
+  Expr *Ref = RefResult.get();
+
+  QualType FieldType = Field->getType();
+
+  // When the variable has array type, create index variables for each
+  // dimension of the array. We use these index variables to subscript
+  // the source array, and other clients (e.g., CodeGen) will perform
+  // the necessary iteration with these index variables.
+  //
+  // FIXME: This is dumb. Add a proper AST representation for array
+  // copy-construction and use it here.
+  SmallVector<VarDecl *, 4> IndexVariables;
+  QualType BaseType = FieldType;
+  QualType SizeType = S.Context.getSizeType();
+  ArrayIndexStarts.push_back(ArrayIndexVars.size());
+  while (const ConstantArrayType *Array
+                        = S.Context.getAsConstantArrayType(BaseType)) {
+    // Create the iteration variable for this array index.
+    IdentifierInfo *IterationVarName = nullptr;
+    {
+      SmallString<8> Str;
+      llvm::raw_svector_ostream OS(Str);
+      OS << "__i" << IndexVariables.size();
+      IterationVarName = &S.Context.Idents.get(OS.str());
+    }
+    VarDecl *IterationVar = VarDecl::Create(
+        S.Context, S.CurContext, Loc, Loc, IterationVarName, SizeType,
+        S.Context.getTrivialTypeSourceInfo(SizeType, Loc), SC_None);
+    IterationVar->setImplicit();
+    IndexVariables.push_back(IterationVar);
+    ArrayIndexVars.push_back(IterationVar);
+    
+    // Create a reference to the iteration variable.
+    ExprResult IterationVarRef =
+        S.BuildDeclRefExpr(IterationVar, SizeType, VK_LValue, Loc);
+    assert(!IterationVarRef.isInvalid() &&
+           "Reference to invented variable cannot fail!");
+    IterationVarRef = S.DefaultLvalueConversion(IterationVarRef.get());
+    assert(!IterationVarRef.isInvalid() &&
+           "Conversion of invented variable cannot fail!");
+    
+    // Subscript the array with this iteration variable.
+    ExprResult Subscript =
+        S.CreateBuiltinArraySubscriptExpr(Ref, Loc, IterationVarRef.get(), Loc);
+    if (Subscript.isInvalid())
+      return ExprError();
+
+    Ref = Subscript.get();
+    BaseType = Array->getElementType();
+  }
+
+  // Construct the entity that we will be initializing. For an array, this
+  // will be first element in the array, which may require several levels
+  // of array-subscript entities. 
+  SmallVector<InitializedEntity, 4> Entities;
+  Entities.reserve(1 + IndexVariables.size());
+  Entities.push_back(InitializedEntity::InitializeLambdaCapture(
+      Var->getIdentifier(), FieldType, Loc));
+  for (unsigned I = 0, N = IndexVariables.size(); I != N; ++I)
+    Entities.push_back(
+        InitializedEntity::InitializeElement(S.Context, 0, Entities.back()));
+
+  InitializationKind InitKind = InitializationKind::CreateDirect(Loc, Loc, Loc);
+  InitializationSequence Init(S, Entities.back(), InitKind, Ref);
+  return Init.Perform(S, Entities.back(), InitKind, Ref);
+}
          
 ExprResult Sema::ActOnLambdaExpr(SourceLocation StartLoc, Stmt *Body, 
-                                 Scope *CurScope, 
-                                 bool IsInstantiation) {
+                                 Scope *CurScope) {
+  LambdaScopeInfo LSI = *cast<LambdaScopeInfo>(FunctionScopes.back());
+  ActOnFinishFunctionBody(LSI.CallOperator, Body);
+  return BuildLambdaExpr(StartLoc, Body->getLocEnd(), &LSI);
+}
+
+static LambdaCaptureDefault
+mapImplicitCaptureStyle(CapturingScopeInfo::ImplicitCaptureStyle ICS) {
+  switch (ICS) {
+  case CapturingScopeInfo::ImpCap_None:
+    return LCD_None;
+  case CapturingScopeInfo::ImpCap_LambdaByval:
+    return LCD_ByCopy;
+  case CapturingScopeInfo::ImpCap_CapturedRegion:
+  case CapturingScopeInfo::ImpCap_LambdaByref:
+    return LCD_ByRef;
+  case CapturingScopeInfo::ImpCap_Block:
+    llvm_unreachable("block capture in lambda");
+  }
+  llvm_unreachable("Unknown implicit capture style");
+}
+
+ExprResult Sema::BuildLambdaExpr(SourceLocation StartLoc, SourceLocation EndLoc,
+                                 LambdaScopeInfo *LSI) {
   // Collect information from the lambda scope.
   SmallVector<LambdaCapture, 4> Captures;
   SmallVector<Expr *, 4> CaptureInits;
-  LambdaCaptureDefault CaptureDefault;
-  SourceLocation CaptureDefaultLoc;
+  SourceLocation CaptureDefaultLoc = LSI->CaptureDefaultLoc;
+  LambdaCaptureDefault CaptureDefault =
+      mapImplicitCaptureStyle(LSI->ImpCaptureStyle);
   CXXRecordDecl *Class;
   CXXMethodDecl *CallOperator;
   SourceRange IntroducerRange;
@@ -1393,7 +1510,6 @@ ExprResult Sema::ActOnLambdaExpr(SourceLocation StartLoc, Stmt *Body,
   SmallVector<VarDecl *, 4> ArrayIndexVars;
   SmallVector<unsigned, 4> ArrayIndexStarts;
   {
-    LambdaScopeInfo *LSI = getCurLambda();
     CallOperator = LSI->CallOperator;
     Class = LSI->Lambda;
     IntroducerRange = LSI->IntroducerRange;
@@ -1401,11 +1517,21 @@ ExprResult Sema::ActOnLambdaExpr(SourceLocation StartLoc, Stmt *Body,
     ExplicitResultType = !LSI->HasImplicitReturnType;
     LambdaExprNeedsCleanups = LSI->ExprNeedsCleanups;
     ContainsUnexpandedParameterPack = LSI->ContainsUnexpandedParameterPack;
-    ArrayIndexVars.swap(LSI->ArrayIndexVars);
-    ArrayIndexStarts.swap(LSI->ArrayIndexStarts);
     
+    CallOperator->setLexicalDeclContext(Class);
+    Decl *TemplateOrNonTemplateCallOperatorDecl = 
+        CallOperator->getDescribedFunctionTemplate()  
+        ? CallOperator->getDescribedFunctionTemplate() 
+        : cast<Decl>(CallOperator);
+
+    TemplateOrNonTemplateCallOperatorDecl->setLexicalDeclContext(Class);
+    Class->addDecl(TemplateOrNonTemplateCallOperatorDecl);
+
+    PopExpressionEvaluationContext();
+
     // Translate captures.
-    for (unsigned I = 0, N = LSI->Captures.size(); I != N; ++I) {
+    auto CurField = Class->field_begin();
+    for (unsigned I = 0, N = LSI->Captures.size(); I != N; ++I, ++CurField) {
       LambdaScopeInfo::Capture From = LSI->Captures[I];
       assert(!From.isBlockCapture() && "Cannot capture __block variables");
       bool IsImplicit = I >= LSI->NumExplicitCaptures;
@@ -1417,83 +1543,33 @@ ExprResult Sema::ActOnLambdaExpr(SourceLocation StartLoc, Stmt *Body,
         CaptureInits.push_back(new (Context) CXXThisExpr(From.getLocation(),
                                                          getCurrentThisType(),
                                                          /*isImplicit=*/true));
+        ArrayIndexStarts.push_back(ArrayIndexVars.size());
         continue;
       }
       if (From.isVLATypeCapture()) {
         Captures.push_back(
             LambdaCapture(From.getLocation(), IsImplicit, LCK_VLAType));
         CaptureInits.push_back(nullptr);
+        ArrayIndexStarts.push_back(ArrayIndexVars.size());
         continue;
       }
 
       VarDecl *Var = From.getVariable();
-      LambdaCaptureKind Kind = From.isCopyCapture()? LCK_ByCopy : LCK_ByRef;
+      LambdaCaptureKind Kind = From.isCopyCapture() ? LCK_ByCopy : LCK_ByRef;
       Captures.push_back(LambdaCapture(From.getLocation(), IsImplicit, Kind,
                                        Var, From.getEllipsisLoc()));
-      CaptureInits.push_back(From.getInitExpr());
-    }
-
-    switch (LSI->ImpCaptureStyle) {
-    case CapturingScopeInfo::ImpCap_None:
-      CaptureDefault = LCD_None;
-      break;
-
-    case CapturingScopeInfo::ImpCap_LambdaByval:
-      CaptureDefault = LCD_ByCopy;
-      break;
-
-    case CapturingScopeInfo::ImpCap_CapturedRegion:
-    case CapturingScopeInfo::ImpCap_LambdaByref:
-      CaptureDefault = LCD_ByRef;
-      break;
-
-    case CapturingScopeInfo::ImpCap_Block:
-      llvm_unreachable("block capture in lambda");
-      break;
-    }
-    CaptureDefaultLoc = LSI->CaptureDefaultLoc;
-
-    // C++11 [expr.prim.lambda]p4:
-    //   If a lambda-expression does not include a
-    //   trailing-return-type, it is as if the trailing-return-type
-    //   denotes the following type:
-    //
-    // Skip for C++1y return type deduction semantics which uses
-    // different machinery.
-    // FIXME: Refactor and Merge the return type deduction machinery.
-    // FIXME: Assumes current resolution to core issue 975.
-    if (LSI->HasImplicitReturnType && !getLangOpts().CPlusPlus14) {
-      deduceClosureReturnType(*LSI);
-
-      //   - if there are no return statements in the
-      //     compound-statement, or all return statements return
-      //     either an expression of type void or no expression or
-      //     braced-init-list, the type void;
-      if (LSI->ReturnType.isNull()) {
-        LSI->ReturnType = Context.VoidTy;
+      Expr *Init = From.getInitExpr();
+      if (!Init) {
+        auto InitResult = performLambdaVarCaptureInitialization(
+            *this, From, *CurField, ArrayIndexVars, ArrayIndexStarts);
+        if (InitResult.isInvalid())
+          return ExprError();
+        Init = InitResult.get();
+      } else {
+        ArrayIndexStarts.push_back(ArrayIndexVars.size());
       }
-
-      // Create a function type with the inferred return type.
-      const FunctionProtoType *Proto
-        = CallOperator->getType()->getAs<FunctionProtoType>();
-      QualType FunctionTy = Context.getFunctionType(
-          LSI->ReturnType, Proto->getParamTypes(), Proto->getExtProtoInfo());
-      CallOperator->setType(FunctionTy);
+      CaptureInits.push_back(Init);
     }
-    // C++ [expr.prim.lambda]p7:
-    //   The lambda-expression's compound-statement yields the
-    //   function-body (8.4) of the function call operator [...].
-    ActOnFinishFunctionBody(CallOperator, Body, IsInstantiation);
-    CallOperator->setLexicalDeclContext(Class);
-    Decl *TemplateOrNonTemplateCallOperatorDecl = 
-        CallOperator->getDescribedFunctionTemplate()  
-        ? CallOperator->getDescribedFunctionTemplate() 
-        : cast<Decl>(CallOperator);
-
-    TemplateOrNonTemplateCallOperatorDecl->setLexicalDeclContext(Class);
-    Class->addDecl(TemplateOrNonTemplateCallOperatorDecl);
-
-    PopExpressionEvaluationContext();
 
     // C++11 [expr.prim.lambda]p6:
     //   The closure type for a lambda-expression with no lambda-capture
@@ -1529,7 +1605,7 @@ ExprResult Sema::ActOnLambdaExpr(SourceLocation StartLoc, Stmt *Body,
                                           Captures, 
                                           ExplicitParams, ExplicitResultType,
                                           CaptureInits, ArrayIndexVars, 
-                                          ArrayIndexStarts, Body->getLocEnd(),
+                                          ArrayIndexStarts, EndLoc,
                                           ContainsUnexpandedParameterPack);
 
   if (!CurContext->isDependentContext()) {
diff --git a/lib/Sema/SemaLookup.cpp b/lib/Sema/SemaLookup.cpp
index ad06872ab6fe..c745b136bede 100644
--- a/lib/Sema/SemaLookup.cpp
+++ b/lib/Sema/SemaLookup.cpp
@@ -13,6 +13,7 @@
 //===----------------------------------------------------------------------===//
 #include "clang/Sema/Lookup.h"
 #include "clang/AST/ASTContext.h"
+#include "clang/AST/ASTMutationListener.h"
 #include "clang/AST/CXXInheritance.h"
 #include "clang/AST/Decl.h"
 #include "clang/AST/DeclCXX.h"
@@ -23,7 +24,9 @@
 #include "clang/AST/ExprCXX.h"
 #include "clang/Basic/Builtins.h"
 #include "clang/Basic/LangOptions.h"
+#include "clang/Lex/HeaderSearch.h"
 #include "clang/Lex/ModuleLoader.h"
+#include "clang/Lex/Preprocessor.h"
 #include "clang/Sema/DeclSpec.h"
 #include "clang/Sema/ExternalSemaSource.h"
 #include "clang/Sema/Overload.h"
@@ -194,10 +197,11 @@ namespace {
     const_iterator begin() const { return list.begin(); }
     const_iterator end() const { return list.end(); }
 
-    std::pair<const_iterator,const_iterator>
+    llvm::iterator_range<const_iterator>
     getNamespacesFor(DeclContext *DC) const {
-      return std::equal_range(begin(), end(), DC->getPrimaryContext(),
-                              UnqualUsingEntry::Comparator());
+      return llvm::make_range(std::equal_range(begin(), end(),
+                                               DC->getPrimaryContext(),
+                                               UnqualUsingEntry::Comparator()));
     }
   };
 }
@@ -413,6 +417,10 @@ void LookupResult::resolveKind() {
     if (!Unique.insert(D).second) {
       // If it's not unique, pull something off the back (and
       // continue at this index).
+      // FIXME: This is wrong. We need to take the more recent declaration in
+      // order to get the right type, default arguments, etc. We also need to
+      // prefer visible declarations to hidden ones (for redeclaration lookup
+      // in modules builds).
       Decls[I] = Decls[--N];
       continue;
     }
@@ -670,8 +678,8 @@ static bool LookupDirect(Sema &S, LookupResult &R, const DeclContext *DC) {
     DeclareImplicitMemberFunctionsWithName(S, R.getLookupName(), DC);
 
   // Perform lookup into this declaration context.
-  DeclContext::lookup_const_result DR = DC->lookup(R.getLookupName());
-  for (DeclContext::lookup_const_iterator I = DR.begin(), E = DR.end(); I != E;
+  DeclContext::lookup_result DR = DC->lookup(R.getLookupName());
+  for (DeclContext::lookup_iterator I = DR.begin(), E = DR.end(); I != E;
        ++I) {
     NamedDecl *D = *I;
     if ((D = R.getAcceptableDecl(D))) {
@@ -765,11 +773,8 @@ CppNamespaceLookup(Sema &S, LookupResult &R, ASTContext &Context,
 
   // Perform direct name lookup into the namespaces nominated by the
   // using directives whose common ancestor is this namespace.
-  UnqualUsingDirectiveSet::const_iterator UI, UEnd;
-  std::tie(UI, UEnd) = UDirs.getNamespacesFor(NS);
-
-  for (; UI != UEnd; ++UI)
-    if (LookupDirect(S, R, UI->getNominatedNamespace()))
+  for (const UnqualUsingEntry &UUE : UDirs.getNamespacesFor(NS))
+    if (LookupDirect(S, R, UUE.getNominatedNamespace()))
       Found = true;
 
   R.resolveKind();
@@ -1167,8 +1172,59 @@ static Decl *getInstantiatedFrom(Decl *D, MemberSpecializationInfo *MSInfo) {
   return MSInfo->isExplicitSpecialization() ? D : MSInfo->getInstantiatedFrom();
 }
 
+Module *Sema::getOwningModule(Decl *Entity) {
+  // If it's imported, grab its owning module.
+  Module *M = Entity->getImportedOwningModule();
+  if (M || !isa<NamedDecl>(Entity) || !cast<NamedDecl>(Entity)->isHidden())
+    return M;
+  assert(!Entity->isFromASTFile() &&
+         "hidden entity from AST file has no owning module");
+
+  // It's local and hidden; grab or compute its owning module.
+  M = Entity->getLocalOwningModule();
+  if (M)
+    return M;
+
+  if (auto *Containing =
+          PP.getModuleContainingLocation(Entity->getLocation())) {
+    M = Containing;
+  } else if (Entity->isInvalidDecl() || Entity->getLocation().isInvalid()) {
+    // Don't bother tracking visibility for invalid declarations with broken
+    // locations.
+    cast<NamedDecl>(Entity)->setHidden(false);
+  } else {
+    // We need to assign a module to an entity that exists outside of any
+    // module, so that we can hide it from modules that we textually enter.
+    // Invent a fake module for all such entities.
+    if (!CachedFakeTopLevelModule) {
+      CachedFakeTopLevelModule =
+          PP.getHeaderSearchInfo().getModuleMap().findOrCreateModule(
+              "<top-level>", nullptr, false, false).first;
+
+      auto &SrcMgr = PP.getSourceManager();
+      SourceLocation StartLoc =
+          SrcMgr.getLocForStartOfFile(SrcMgr.getMainFileID());
+      auto &TopLevel =
+          VisibleModulesStack.empty() ? VisibleModules : VisibleModulesStack[0];
+      TopLevel.setVisible(CachedFakeTopLevelModule, StartLoc);
+    }
+
+    M = CachedFakeTopLevelModule;
+  }
+
+  if (M)
+    Entity->setLocalOwningModule(M);
+  return M;
+}
+
+void Sema::makeMergedDefinitionVisible(NamedDecl *ND, SourceLocation Loc) {
+  auto *M = PP.getModuleContainingLocation(Loc);
+  assert(M && "hidden definition not in any module");
+  Context.mergeDefinitionIntoModule(ND, M);
+}
+
 /// \brief Find the module in which the given declaration was defined.
-static Module *getDefiningModule(Decl *Entity) {
+static Module *getDefiningModule(Sema &S, Decl *Entity) {
   if (FunctionDecl *FD = dyn_cast<FunctionDecl>(Entity)) {
     // If this function was instantiated from a template, the defining module is
     // the module containing the pattern.
@@ -1190,15 +1246,16 @@ static Module *getDefiningModule(Decl *Entity) {
   // from a template.
   DeclContext *Context = Entity->getDeclContext();
   if (Context->isFileContext())
-    return Entity->getOwningModule();
-  return getDefiningModule(cast<Decl>(Context));
+    return S.getOwningModule(Entity);
+  return getDefiningModule(S, cast<Decl>(Context));
 }
 
 llvm::DenseSet<Module*> &Sema::getLookupModules() {
   unsigned N = ActiveTemplateInstantiations.size();
   for (unsigned I = ActiveTemplateInstantiationLookupModules.size();
        I != N; ++I) {
-    Module *M = getDefiningModule(ActiveTemplateInstantiations[I].Entity);
+    Module *M =
+        getDefiningModule(*this, ActiveTemplateInstantiations[I].Entity);
     if (M && !LookupModulesCache.insert(M).second)
       M = nullptr;
     ActiveTemplateInstantiationLookupModules.push_back(M);
@@ -1206,6 +1263,13 @@ llvm::DenseSet<Module*> &Sema::getLookupModules() {
   return LookupModulesCache;
 }
 
+bool Sema::hasVisibleMergedDefinition(NamedDecl *Def) {
+  for (Module *Merged : Context.getModulesWithMergedDefinition(Def))
+    if (isModuleVisible(Merged))
+      return true;
+  return false;
+}
+
 /// \brief Determine whether a declaration is visible to name lookup.
 ///
 /// This routine determines whether the declaration D is visible in the current
@@ -1215,10 +1279,48 @@ llvm::DenseSet<Module*> &Sema::getLookupModules() {
 /// path (by instantiating a template, you allow it to see the declarations that
 /// your module can see, including those later on in your module).
 bool LookupResult::isVisibleSlow(Sema &SemaRef, NamedDecl *D) {
-  assert(D->isHidden() && !SemaRef.ActiveTemplateInstantiations.empty() &&
-         "should not call this: not in slow case");
-  Module *DeclModule = D->getOwningModule();
-  assert(DeclModule && "hidden decl not from a module");
+  assert(D->isHidden() && "should not call this: not in slow case");
+  Module *DeclModule = SemaRef.getOwningModule(D);
+  if (!DeclModule) {
+    // getOwningModule() may have decided the declaration should not be hidden.
+    assert(!D->isHidden() && "hidden decl not from a module");
+    return true;
+  }
+
+  // If the owning module is visible, and the decl is not module private,
+  // then the decl is visible too. (Module private is ignored within the same
+  // top-level module.)
+  if (!D->isFromASTFile() || !D->isModulePrivate()) {
+    if (SemaRef.isModuleVisible(DeclModule))
+      return true;
+    // Also check merged definitions.
+    if (SemaRef.getLangOpts().ModulesLocalVisibility &&
+        SemaRef.hasVisibleMergedDefinition(D))
+      return true;
+  }
+
+  // If this declaration is not at namespace scope nor module-private,
+  // then it is visible if its lexical parent has a visible definition.
+  DeclContext *DC = D->getLexicalDeclContext();
+  if (!D->isModulePrivate() &&
+      DC && !DC->isFileContext() && !isa<LinkageSpecDecl>(DC)) {
+    // For a parameter, check whether our current template declaration's
+    // lexical context is visible, not whether there's some other visible
+    // definition of it, because parameters aren't "within" the definition.
+    if ((D->isTemplateParameter() || isa<ParmVarDecl>(D))
+            ? isVisible(SemaRef, cast<NamedDecl>(DC))
+            : SemaRef.hasVisibleDefinition(cast<NamedDecl>(DC))) {
+      if (SemaRef.ActiveTemplateInstantiations.empty() &&
+          // FIXME: Do something better in this case.
+          !SemaRef.getLangOpts().ModulesLocalVisibility) {
+        // Cache the fact that this declaration is implicitly visible because
+        // its parent has a visible definition.
+        D->setHidden(false);
+      }
+      return true;
+    }
+    return false;
+  }
 
   // Find the extra places where we need to look.
   llvm::DenseSet<Module*> &LookupModules = SemaRef.getLookupModules();
@@ -1243,6 +1345,10 @@ bool LookupResult::isVisibleSlow(Sema &SemaRef, NamedDecl *D) {
   return false;
 }
 
+bool Sema::isVisibleSlow(const NamedDecl *D) {
+  return LookupResult::isVisible(*this, const_cast<NamedDecl*>(D));
+}
+
 /// \brief Retrieve the visible declaration corresponding to D, if any.
 ///
 /// This routine determines whether the declaration D is visible in the current
@@ -1256,6 +1362,9 @@ static NamedDecl *findAcceptableDecl(Sema &SemaRef, NamedDecl *D) {
 
   for (auto RD : D->redecls()) {
     if (auto ND = dyn_cast<NamedDecl>(RD)) {
+      // FIXME: This is wrong in the case where the previous declaration is not
+      // visible in the same scope as D. This needs to be done much more
+      // carefully.
       if (LookupResult::isVisible(SemaRef, ND))
         return ND;
     }
@@ -3016,17 +3125,45 @@ static void LookupVisibleDecls(DeclContext *Ctx, LookupResult &Result,
   if (Visited.visitedContext(Ctx->getPrimaryContext()))
     return;
 
+  // Outside C++, lookup results for the TU live on identifiers.
+  if (isa<TranslationUnitDecl>(Ctx) &&
+      !Result.getSema().getLangOpts().CPlusPlus) {
+    auto &S = Result.getSema();
+    auto &Idents = S.Context.Idents;
+
+    // Ensure all external identifiers are in the identifier table.
+    if (IdentifierInfoLookup *External = Idents.getExternalIdentifierLookup()) {
+      std::unique_ptr<IdentifierIterator> Iter(External->getIdentifiers());
+      for (StringRef Name = Iter->Next(); !Name.empty(); Name = Iter->Next())
+        Idents.get(Name);
+    }
+
+    // Walk all lookup results in the TU for each identifier.
+    for (const auto &Ident : Idents) {
+      for (auto I = S.IdResolver.begin(Ident.getValue()),
+                E = S.IdResolver.end();
+           I != E; ++I) {
+        if (S.IdResolver.isDeclInScope(*I, Ctx)) {
+          if (NamedDecl *ND = Result.getAcceptableDecl(*I)) {
+            Consumer.FoundDecl(ND, Visited.checkHidden(ND), Ctx, InBaseClass);
+            Visited.add(ND);
+          }
+        }
+      }
+    }
+
+    return;
+  }
+
   if (CXXRecordDecl *Class = dyn_cast<CXXRecordDecl>(Ctx))
     Result.getSema().ForceDeclarationOfImplicitMembers(Class);
 
   // Enumerate all of the results in this context.
-  for (const auto &R : Ctx->lookups()) {
-    for (auto *I : R) {
-      if (NamedDecl *ND = dyn_cast<NamedDecl>(I)) {
-        if ((ND = Result.getAcceptableDecl(ND))) {
-          Consumer.FoundDecl(ND, Visited.checkHidden(ND), Ctx, InBaseClass);
-          Visited.add(ND);
-        }
+  for (DeclContextLookupResult R : Ctx->lookups()) {
+    for (auto *D : R) {
+      if (auto *ND = Result.getAcceptableDecl(D)) {
+        Consumer.FoundDecl(ND, Visited.checkHidden(ND), Ctx, InBaseClass);
+        Visited.add(ND);
       }
     }
   }
@@ -3209,10 +3346,8 @@ static void LookupVisibleDecls(Scope *S, LookupResult &Result,
   if (Entity) {
     // Lookup visible declarations in any namespaces found by using
     // directives.
-    UnqualUsingDirectiveSet::const_iterator UI, UEnd;
-    std::tie(UI, UEnd) = UDirs.getNamespacesFor(Entity);
-    for (; UI != UEnd; ++UI)
-      LookupVisibleDecls(const_cast<DeclContext *>(UI->getNominatedNamespace()),
+    for (const UnqualUsingEntry &UUE : UDirs.getNamespacesFor(Entity))
+      LookupVisibleDecls(const_cast<DeclContext *>(UUE.getNominatedNamespace()),
                          Result, /*QualifiedNameLookup=*/false,
                          /*InBaseClass=*/false, Consumer, Visited);
   }
@@ -3682,8 +3817,7 @@ void TypoCorrectionConsumer::performQualifiedLookups() {
 
 TypoCorrectionConsumer::NamespaceSpecifierSet::NamespaceSpecifierSet(
     ASTContext &Context, DeclContext *CurContext, CXXScopeSpec *CurScopeSpec)
-    : Context(Context), CurContextChain(buildContextChain(CurContext)),
-      isSorted(false) {
+    : Context(Context), CurContextChain(buildContextChain(CurContext)) {
   if (NestedNameSpecifier *NNS =
           CurScopeSpec ? CurScopeSpec->getScopeRep() : nullptr) {
     llvm::raw_string_ostream SpecifierOStream(CurNameSpecifier);
@@ -3702,7 +3836,6 @@ TypoCorrectionConsumer::NamespaceSpecifierSet::NamespaceSpecifierSet(
   }
 
   // Add the global context as a NestedNameSpecifier
-  Distances.insert(1);
   SpecifierInfo SI = {cast<DeclContext>(Context.getTranslationUnitDecl()),
                       NestedNameSpecifier::GlobalSpecifier(Context), 1};
   DistanceMap[1].push_back(SI);
@@ -3722,22 +3855,6 @@ auto TypoCorrectionConsumer::NamespaceSpecifierSet::buildContextChain(
   return Chain;
 }
 
-void TypoCorrectionConsumer::NamespaceSpecifierSet::sortNamespaces() {
-  SmallVector<unsigned, 4> sortedDistances;
-  sortedDistances.append(Distances.begin(), Distances.end());
-
-  if (sortedDistances.size() > 1)
-    std::sort(sortedDistances.begin(), sortedDistances.end());
-
-  Specifiers.clear();
-  for (auto D : sortedDistances) {
-    SpecifierInfoList &SpecList = DistanceMap[D];
-    Specifiers.append(SpecList.begin(), SpecList.end());
-  }
-
-  isSorted = true;
-}
-
 unsigned
 TypoCorrectionConsumer::NamespaceSpecifierSet::buildNestedNameSpecifier(
     DeclContextList &DeclChain, NestedNameSpecifier *&NNS) {
@@ -3818,8 +3935,6 @@ void TypoCorrectionConsumer::NamespaceSpecifierSet::addNameSpecifier(
         llvm::makeArrayRef(NewNameSpecifierIdentifiers));
   }
 
-  isSorted = false;
-  Distances.insert(NumSpecifiers);
   SpecifierInfo SI = {Ctx, NNS, NumSpecifiers};
   DistanceMap[NumSpecifiers].push_back(SI);
 }
@@ -4294,7 +4409,7 @@ TypoCorrection Sema::CorrectTypo(const DeclarationNameInfo &TypoName,
   // Record the failure's location if needed and return an empty correction. If
   // this was an unqualified lookup and we believe the callback object did not
   // filter out possible corrections, also cache the failure for the typo.
-  return FailedCorrection(Typo, TypoName.getLoc(), RecordFailure);
+  return FailedCorrection(Typo, TypoName.getLoc(), RecordFailure && !SecondBestTC);
 }
 
 /// \brief Try to "correct" a typo in the source code by finding
@@ -4347,9 +4462,7 @@ TypoExpr *Sema::CorrectTypoDelayed(
   TypoCorrection Empty;
   auto Consumer = makeTypoCorrectionConsumer(
       TypoName, LookupKind, S, SS, std::move(CCC), MemberContext,
-      EnteringContext, OPT,
-      /*SearchModules=*/(Mode == CTK_ErrorRecovery) && getLangOpts().Modules &&
-          getLangOpts().ModulesSearchAll);
+      EnteringContext, OPT, Mode == CTK_ErrorRecovery);
 
   if (!Consumer || Consumer->empty())
     return nullptr;
@@ -4491,18 +4604,18 @@ void Sema::diagnoseTypo(const TypoCorrection &Correction,
 
 /// Find which declaration we should import to provide the definition of
 /// the given declaration.
-static const NamedDecl *getDefinitionToImport(const NamedDecl *D) {
-  if (const VarDecl *VD = dyn_cast<VarDecl>(D))
+static NamedDecl *getDefinitionToImport(NamedDecl *D) {
+  if (VarDecl *VD = dyn_cast<VarDecl>(D))
     return VD->getDefinition();
   if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D))
-    return FD->isDefined(FD) ? FD : nullptr;
-  if (const TagDecl *TD = dyn_cast<TagDecl>(D))
+    return FD->isDefined(FD) ? const_cast<FunctionDecl*>(FD) : nullptr;
+  if (TagDecl *TD = dyn_cast<TagDecl>(D))
     return TD->getDefinition();
-  if (const ObjCInterfaceDecl *ID = dyn_cast<ObjCInterfaceDecl>(D))
+  if (ObjCInterfaceDecl *ID = dyn_cast<ObjCInterfaceDecl>(D))
     return ID->getDefinition();
-  if (const ObjCProtocolDecl *PD = dyn_cast<ObjCProtocolDecl>(D))
+  if (ObjCProtocolDecl *PD = dyn_cast<ObjCProtocolDecl>(D))
     return PD->getDefinition();
-  if (const TemplateDecl *TD = dyn_cast<TemplateDecl>(D))
+  if (TemplateDecl *TD = dyn_cast<TemplateDecl>(D))
     return getDefinitionToImport(TD->getTemplatedDecl());
   return nullptr;
 }
@@ -4535,10 +4648,10 @@ void Sema::diagnoseTypo(const TypoCorrection &Correction,
 
     // Suggest importing a module providing the definition of this entity, if
     // possible.
-    const NamedDecl *Def = getDefinitionToImport(Decl);
+    NamedDecl *Def = getDefinitionToImport(Decl);
     if (!Def)
       Def = Decl;
-    Module *Owner = Def->getOwningModule();
+    Module *Owner = getOwningModule(Def);
     assert(Owner && "definition of hidden declaration is not in a module");
 
     Diag(Correction.getCorrectionRange().getBegin(),
diff --git a/lib/Sema/SemaObjCProperty.cpp b/lib/Sema/SemaObjCProperty.cpp
index 72b6020351fa..5e7b4b8b859d 100644
--- a/lib/Sema/SemaObjCProperty.cpp
+++ b/lib/Sema/SemaObjCProperty.cpp
@@ -19,6 +19,7 @@
 #include "clang/AST/ExprObjC.h"
 #include "clang/Basic/SourceManager.h"
 #include "clang/Lex/Lexer.h"
+#include "clang/Lex/Preprocessor.h"
 #include "clang/Sema/Initialization.h"
 #include "llvm/ADT/DenseSet.h"
 #include "llvm/ADT/SmallString.h"
@@ -406,9 +407,11 @@ Sema::HandlePropertyInClassExtension(Scope *S,
     // this conversion is safe only because the wider type is for a 'readonly'
     // property in primary class and 'narrowed' type for a 'readwrite' property
     // in continuation class.
-    if (!isa<ObjCObjectPointerType>(PIDecl->getType()) ||
-        !isa<ObjCObjectPointerType>(PDecl->getType()) ||
-        (!isObjCPointerConversion(PDecl->getType(), PIDecl->getType(), 
+    QualType PrimaryClassPropertyT = Context.getCanonicalType(PIDecl->getType());
+    QualType ClassExtPropertyT = Context.getCanonicalType(PDecl->getType());
+    if (!isa<ObjCObjectPointerType>(PrimaryClassPropertyT) ||
+        !isa<ObjCObjectPointerType>(ClassExtPropertyT) ||
+        (!isObjCPointerConversion(ClassExtPropertyT, PrimaryClassPropertyT,
                                   ConvertedType, IncompatibleObjC))
         || IncompatibleObjC) {
       Diag(AtLoc, 
@@ -1854,6 +1857,39 @@ void Sema::DiagnoseOwningPropertyGetterSynthesis(const ObjCImplementationDecl *D
           Diag(PD->getLocation(), diag::err_cocoa_naming_owned_rule);
         else
           Diag(PD->getLocation(), diag::warn_cocoa_naming_owned_rule);
+
+        // Look for a getter explicitly declared alongside the property.
+        // If we find one, use its location for the note.
+        SourceLocation noteLoc = PD->getLocation();
+        SourceLocation fixItLoc;
+        for (auto *getterRedecl : method->redecls()) {
+          if (getterRedecl->isImplicit())
+            continue;
+          if (getterRedecl->getDeclContext() != PD->getDeclContext())
+            continue;
+          noteLoc = getterRedecl->getLocation();
+          fixItLoc = getterRedecl->getLocEnd();
+        }
+
+        Preprocessor &PP = getPreprocessor();
+        TokenValue tokens[] = {
+          tok::kw___attribute, tok::l_paren, tok::l_paren,
+          PP.getIdentifierInfo("objc_method_family"), tok::l_paren,
+          PP.getIdentifierInfo("none"), tok::r_paren,
+          tok::r_paren, tok::r_paren
+        };
+        StringRef spelling = "__attribute__((objc_method_family(none)))";
+        StringRef macroName = PP.getLastMacroWithSpelling(noteLoc, tokens);
+        if (!macroName.empty())
+          spelling = macroName;
+
+        auto noteDiag = Diag(noteLoc, diag::note_cocoa_naming_declare_family)
+            << method->getDeclName() << spelling;
+        if (fixItLoc.isValid()) {
+          SmallString<64> fixItText(" ");
+          fixItText += spelling;
+          noteDiag << FixItHint::CreateInsertion(fixItLoc, fixItText);
+        }
       }
     }
   }
diff --git a/lib/Sema/SemaOpenMP.cpp b/lib/Sema/SemaOpenMP.cpp
index d72942a2ffec..cfe8db3d658c 100644
--- a/lib/Sema/SemaOpenMP.cpp
+++ b/lib/Sema/SemaOpenMP.cpp
@@ -82,10 +82,12 @@ private:
   };
   typedef llvm::SmallDenseMap<VarDecl *, DSAInfo, 64> DeclSAMapTy;
   typedef llvm::SmallDenseMap<VarDecl *, DeclRefExpr *, 64> AlignedMapTy;
+  typedef llvm::DenseSet<VarDecl *> LoopControlVariablesSetTy;
 
   struct SharingMapTy {
     DeclSAMapTy SharingMap;
     AlignedMapTy AlignedMap;
+    LoopControlVariablesSetTy LCVSet;
     DefaultDataSharingAttributes DefaultAttr;
     SourceLocation DefaultAttrLoc;
     OpenMPDirectiveKind Directive;
@@ -93,22 +95,28 @@ private:
     Scope *CurScope;
     SourceLocation ConstructLoc;
     bool OrderedRegion;
+    unsigned CollapseNumber;
     SourceLocation InnerTeamsRegionLoc;
     SharingMapTy(OpenMPDirectiveKind DKind, DeclarationNameInfo Name,
                  Scope *CurScope, SourceLocation Loc)
-        : SharingMap(), AlignedMap(), DefaultAttr(DSA_unspecified),
+        : SharingMap(), AlignedMap(), LCVSet(), DefaultAttr(DSA_unspecified),
           Directive(DKind), DirectiveName(std::move(Name)), CurScope(CurScope),
-          ConstructLoc(Loc), OrderedRegion(false), InnerTeamsRegionLoc() {}
+          ConstructLoc(Loc), OrderedRegion(false), CollapseNumber(1),
+          InnerTeamsRegionLoc() {}
     SharingMapTy()
-        : SharingMap(), AlignedMap(), DefaultAttr(DSA_unspecified),
+        : SharingMap(), AlignedMap(), LCVSet(), DefaultAttr(DSA_unspecified),
           Directive(OMPD_unknown), DirectiveName(), CurScope(nullptr),
-          ConstructLoc(), OrderedRegion(false), InnerTeamsRegionLoc() {}
+          ConstructLoc(), OrderedRegion(false), CollapseNumber(1),
+          InnerTeamsRegionLoc() {}
   };
 
   typedef SmallVector<SharingMapTy, 64> StackTy;
 
   /// \brief Stack of used declaration and their data-sharing attributes.
   StackTy Stack;
+  /// \brief true, if check for DSA must be from parent directive, false, if
+  /// from current directive.
+  bool FromParent;
   Sema &SemaRef;
 
   typedef SmallVector<SharingMapTy, 8>::reverse_iterator reverse_iterator;
@@ -119,7 +127,10 @@ private:
   bool isOpenMPLocal(VarDecl *D, StackTy::reverse_iterator Iter);
 
 public:
-  explicit DSAStackTy(Sema &S) : Stack(1), SemaRef(S) {}
+  explicit DSAStackTy(Sema &S) : Stack(1), FromParent(false), SemaRef(S) {}
+
+  bool isFromParent() const { return FromParent; }
+  void setFromParent(bool Flag) { FromParent = Flag; }
 
   void push(OpenMPDirectiveKind DKind, const DeclarationNameInfo &DirName,
             Scope *CurScope, SourceLocation Loc) {
@@ -137,6 +148,12 @@ public:
   /// for diagnostics.
   DeclRefExpr *addUniqueAligned(VarDecl *D, DeclRefExpr *NewDE);
 
+  /// \brief Register specified variable as loop control variable.
+  void addLoopControlVariable(VarDecl *D);
+  /// \brief Check if the specified variable is a loop control variable for
+  /// current region.
+  bool isLoopControlVariable(VarDecl *D);
+
   /// \brief Adds explicit data sharing attribute to the specified declaration.
   void addDSA(VarDecl *D, DeclRefExpr *E, OpenMPClauseKind A);
 
@@ -209,6 +226,13 @@ public:
     return false;
   }
 
+  /// \brief Set collapse value for the region.
+  void setCollapseNumber(unsigned Val) { Stack.back().CollapseNumber = Val; }
+  /// \brief Return collapse value for region.
+  unsigned getCollapseNumber() const {
+    return Stack.back().CollapseNumber;
+  }
+
   /// \brief Marks current target region as one with closely nested teams
   /// region.
   void setParentTeamsRegionLoc(SourceLocation TeamsRegionLoc) {
@@ -238,6 +262,7 @@ bool isParallelOrTaskRegion(OpenMPDirectiveKind DKind) {
 
 DSAStackTy::DSAVarData DSAStackTy::getDSA(StackTy::reverse_iterator Iter,
                                           VarDecl *D) {
+  D = D->getCanonicalDecl();
   DSAVarData DVar;
   if (Iter == std::prev(Stack.rend())) {
     // OpenMP [2.9.1.1, Data-sharing Attribute Rules for Variables Referenced
@@ -308,8 +333,7 @@ DSAStackTy::DSAVarData DSAStackTy::getDSA(StackTy::reverse_iterator Iter,
     //  bound to the current team is shared.
     if (DVar.DKind == OMPD_task) {
       DSAVarData DVarTemp;
-      for (StackTy::reverse_iterator I = std::next(Iter),
-                                     EE = std::prev(Stack.rend());
+      for (StackTy::reverse_iterator I = std::next(Iter), EE = Stack.rend();
            I != EE; ++I) {
         // OpenMP [2.9.1.1, Data-sharing Attribute Rules for Variables
         // Referenced
@@ -343,6 +367,7 @@ DSAStackTy::DSAVarData DSAStackTy::getDSA(StackTy::reverse_iterator Iter,
 
 DeclRefExpr *DSAStackTy::addUniqueAligned(VarDecl *D, DeclRefExpr *NewDE) {
   assert(Stack.size() > 1 && "Data sharing attributes stack is empty");
+  D = D->getCanonicalDecl();
   auto It = Stack.back().AlignedMap.find(D);
   if (It == Stack.back().AlignedMap.end()) {
     assert(NewDE && "Unexpected nullptr expr to be added into aligned map");
@@ -355,7 +380,20 @@ DeclRefExpr *DSAStackTy::addUniqueAligned(VarDecl *D, DeclRefExpr *NewDE) {
   return nullptr;
 }
 
+void DSAStackTy::addLoopControlVariable(VarDecl *D) {
+  assert(Stack.size() > 1 && "Data-sharing attributes stack is empty");
+  D = D->getCanonicalDecl();
+  Stack.back().LCVSet.insert(D);
+}
+
+bool DSAStackTy::isLoopControlVariable(VarDecl *D) {
+  assert(Stack.size() > 1 && "Data-sharing attributes stack is empty");
+  D = D->getCanonicalDecl();
+  return Stack.back().LCVSet.count(D) > 0;
+}
+
 void DSAStackTy::addDSA(VarDecl *D, DeclRefExpr *E, OpenMPClauseKind A) {
+  D = D->getCanonicalDecl();
   if (A == OMPC_threadprivate) {
     Stack[0].SharingMap[D].Attributes = A;
     Stack[0].SharingMap[D].RefExpr = E;
@@ -367,6 +405,7 @@ void DSAStackTy::addDSA(VarDecl *D, DeclRefExpr *E, OpenMPClauseKind A) {
 }
 
 bool DSAStackTy::isOpenMPLocal(VarDecl *D, StackTy::reverse_iterator Iter) {
+  D = D->getCanonicalDecl();
   if (Stack.size() > 2) {
     reverse_iterator I = Iter, E = std::prev(Stack.rend());
     Scope *TopScope = nullptr;
@@ -385,16 +424,41 @@ bool DSAStackTy::isOpenMPLocal(VarDecl *D, StackTy::reverse_iterator Iter) {
   return false;
 }
 
+/// \brief Build a variable declaration for OpenMP loop iteration variable.
+static VarDecl *buildVarDecl(Sema &SemaRef, SourceLocation Loc, QualType Type,
+                             StringRef Name) {
+  DeclContext *DC = SemaRef.CurContext;
+  IdentifierInfo *II = &SemaRef.PP.getIdentifierTable().get(Name);
+  TypeSourceInfo *TInfo = SemaRef.Context.getTrivialTypeSourceInfo(Type, Loc);
+  VarDecl *Decl =
+      VarDecl::Create(SemaRef.Context, DC, Loc, Loc, II, Type, TInfo, SC_None);
+  Decl->setImplicit();
+  return Decl;
+}
+
+static DeclRefExpr *buildDeclRefExpr(Sema &S, VarDecl *D, QualType Ty,
+                                     SourceLocation Loc,
+                                     bool RefersToCapture = false) {
+  D->setReferenced();
+  D->markUsed(S.Context);
+  return DeclRefExpr::Create(S.getASTContext(), NestedNameSpecifierLoc(),
+                             SourceLocation(), D, RefersToCapture, Loc, Ty,
+                             VK_LValue);
+}
+
 DSAStackTy::DSAVarData DSAStackTy::getTopDSA(VarDecl *D, bool FromParent) {
+  D = D->getCanonicalDecl();
   DSAVarData DVar;
 
   // OpenMP [2.9.1.1, Data-sharing Attribute Rules for Variables Referenced
   // in a Construct, C/C++, predetermined, p.1]
   //  Variables appearing in threadprivate directives are threadprivate.
   if (D->getTLSKind() != VarDecl::TLS_None ||
-      D->getStorageClass() == SC_Register) {
-    DVar.CKind = OMPC_threadprivate;
-    return DVar;
+      (D->getStorageClass() == SC_Register && D->hasAttr<AsmLabelAttr>() &&
+       !D->isLocalVarDecl())) {
+    addDSA(D, buildDeclRefExpr(SemaRef, D, D->getType().getNonReferenceType(),
+                               D->getLocation()),
+           OMPC_threadprivate);
   }
   if (Stack[0].SharingMap.count(D)) {
     DVar.RefExpr = Stack[0].SharingMap[D].RefExpr;
@@ -421,29 +485,28 @@ DSAStackTy::DSAVarData DSAStackTy::getTopDSA(VarDecl *D, bool FromParent) {
       DVar.CKind = OMPC_private;
       return DVar;
     }
-  }
 
-  // OpenMP [2.9.1.1, Data-sharing Attribute Rules for Variables Referenced
-  // in a Construct, C/C++, predetermined, p.4]
-  //  Static data members are shared.
-  if (D->isStaticDataMember()) {
-    // Variables with const-qualified type having no mutable member may be
-    // listed in a firstprivate clause, even if they are static data members.
-    DSAVarData DVarTemp = hasDSA(D, MatchesAnyClause(OMPC_firstprivate),
-                                 MatchesAlways(), FromParent);
-    if (DVarTemp.CKind == OMPC_firstprivate && DVarTemp.RefExpr)
-      return DVar;
+    // OpenMP [2.9.1.1, Data-sharing Attribute Rules for Variables Referenced
+    // in a Construct, C/C++, predetermined, p.4]
+    //  Static data members are shared.
+    // OpenMP [2.9.1.1, Data-sharing Attribute Rules for Variables Referenced
+    // in a Construct, C/C++, predetermined, p.7]
+    //  Variables with static storage duration that are declared in a scope
+    //  inside the construct are shared.
+    if (D->isStaticDataMember() || D->isStaticLocal()) {
+      DSAVarData DVarTemp =
+          hasDSA(D, isOpenMPPrivate, MatchesAlways(), FromParent);
+      if (DVarTemp.CKind != OMPC_unknown && DVarTemp.RefExpr)
+        return DVar;
 
-    DVar.CKind = OMPC_shared;
-    return DVar;
+      DVar.CKind = OMPC_shared;
+      return DVar;
+    }
   }
 
   QualType Type = D->getType().getNonReferenceType().getCanonicalType();
   bool IsConstant = Type.isConstant(SemaRef.getASTContext());
-  while (Type->isArrayType()) {
-    QualType ElemType = cast<ArrayType>(Type.getTypePtr())->getElementType();
-    Type = ElemType.getNonReferenceType().getCanonicalType();
-  }
+  Type = SemaRef.getASTContext().getBaseElementType(Type);
   // OpenMP [2.9.1.1, Data-sharing Attribute Rules for Variables Referenced
   // in a Construct, C/C++, predetermined, p.6]
   //  Variables with const qualified type having no mutable member are
@@ -463,15 +526,6 @@ DSAStackTy::DSAVarData DSAStackTy::getTopDSA(VarDecl *D, bool FromParent) {
     return DVar;
   }
 
-  // OpenMP [2.9.1.1, Data-sharing Attribute Rules for Variables Referenced
-  // in a Construct, C/C++, predetermined, p.7]
-  //  Variables with static storage duration that are declared in a scope
-  //  inside the construct are shared.
-  if (D->isStaticLocal()) {
-    DVar.CKind = OMPC_shared;
-    return DVar;
-  }
-
   // Explicitly specified attributes and local variables with predetermined
   // attributes.
   auto I = std::prev(StartI);
@@ -485,6 +539,7 @@ DSAStackTy::DSAVarData DSAStackTy::getTopDSA(VarDecl *D, bool FromParent) {
 }
 
 DSAStackTy::DSAVarData DSAStackTy::getImplicitDSA(VarDecl *D, bool FromParent) {
+  D = D->getCanonicalDecl();
   auto StartI = Stack.rbegin();
   auto EndI = std::prev(Stack.rend());
   if (FromParent && StartI != EndI) {
@@ -497,6 +552,7 @@ template <class ClausesPredicate, class DirectivesPredicate>
 DSAStackTy::DSAVarData DSAStackTy::hasDSA(VarDecl *D, ClausesPredicate CPred,
                                           DirectivesPredicate DPred,
                                           bool FromParent) {
+  D = D->getCanonicalDecl();
   auto StartI = std::next(Stack.rbegin());
   auto EndI = std::prev(Stack.rend());
   if (FromParent && StartI != EndI) {
@@ -516,6 +572,7 @@ template <class ClausesPredicate, class DirectivesPredicate>
 DSAStackTy::DSAVarData
 DSAStackTy::hasInnermostDSA(VarDecl *D, ClausesPredicate CPred,
                             DirectivesPredicate DPred, bool FromParent) {
+  D = D->getCanonicalDecl();
   auto StartI = std::next(Stack.rbegin());
   auto EndI = std::prev(Stack.rend());
   if (FromParent && StartI != EndI) {
@@ -554,12 +611,17 @@ void Sema::InitDataSharingAttributesStack() {
 
 bool Sema::IsOpenMPCapturedVar(VarDecl *VD) {
   assert(LangOpts.OpenMP && "OpenMP is not allowed");
+  VD = VD->getCanonicalDecl();
   if (DSAStack->getCurrentDirective() != OMPD_unknown) {
-    auto DVarPrivate = DSAStack->getTopDSA(VD, /*FromParent=*/false);
+    if (DSAStack->isLoopControlVariable(VD) ||
+        (VD->hasLocalStorage() &&
+         isParallelOrTaskRegion(DSAStack->getCurrentDirective())))
+      return true;
+    auto DVarPrivate = DSAStack->getTopDSA(VD, DSAStack->isFromParent());
     if (DVarPrivate.CKind != OMPC_unknown && isOpenMPPrivate(DVarPrivate.CKind))
       return true;
     DVarPrivate = DSAStack->hasDSA(VD, isOpenMPPrivate, MatchesAlways(),
-                                   /*FromParent=*/false);
+                                   DSAStack->isFromParent());
     return DVarPrivate.CKind != OMPC_unknown;
   }
   return false;
@@ -574,6 +636,14 @@ void Sema::StartOpenMPDSABlock(OpenMPDirectiveKind DKind,
   PushExpressionEvaluationContext(PotentiallyEvaluated);
 }
 
+void Sema::StartOpenMPClauses() {
+  DSAStack->setFromParent(/*Flag=*/true);
+}
+
+void Sema::EndOpenMPClauses() {
+  DSAStack->setFromParent(/*Flag=*/false);
+}
+
 void Sema::EndOpenMPDSABlock(Stmt *CurDirective) {
   // OpenMP [2.14.3.5, Restrictions, C/C++, p.1]
   //  A variable of class type (or array thereof) that appears in a lastprivate
@@ -581,46 +651,41 @@ void Sema::EndOpenMPDSABlock(Stmt *CurDirective) {
   //  class type, unless the list item is also specified in a firstprivate
   //  clause.
   if (auto D = dyn_cast_or_null<OMPExecutableDirective>(CurDirective)) {
-    for (auto C : D->clauses()) {
-      if (auto Clause = dyn_cast<OMPLastprivateClause>(C)) {
-        for (auto VarRef : Clause->varlists()) {
-          if (VarRef->isValueDependent() || VarRef->isTypeDependent())
+    for (auto *C : D->clauses()) {
+      if (auto *Clause = dyn_cast<OMPLastprivateClause>(C)) {
+        SmallVector<Expr *, 8> PrivateCopies;
+        for (auto *DE : Clause->varlists()) {
+          if (DE->isValueDependent() || DE->isTypeDependent()) {
+            PrivateCopies.push_back(nullptr);
             continue;
-          auto VD = cast<VarDecl>(cast<DeclRefExpr>(VarRef)->getDecl());
+          }
+          auto *VD = cast<VarDecl>(cast<DeclRefExpr>(DE)->getDecl());
+          QualType Type = VD->getType();
           auto DVar = DSAStack->getTopDSA(VD, false);
           if (DVar.CKind == OMPC_lastprivate) {
-            SourceLocation ELoc = VarRef->getExprLoc();
-            auto Type = VarRef->getType();
-            if (Type->isArrayType())
-              Type = QualType(Type->getArrayElementTypeNoTypeQual(), 0);
-            CXXRecordDecl *RD =
-                getLangOpts().CPlusPlus ? Type->getAsCXXRecordDecl() : nullptr;
-            // FIXME This code must be replaced by actual constructing of the
-            // lastprivate variable.
-            if (RD) {
-              CXXConstructorDecl *CD = LookupDefaultConstructor(RD);
-              PartialDiagnostic PD =
-                  PartialDiagnostic(PartialDiagnostic::NullDiagnostic());
-              if (!CD ||
-                  CheckConstructorAccess(
-                      ELoc, CD, InitializedEntity::InitializeTemporary(Type),
-                      CD->getAccess(), PD) == AR_inaccessible ||
-                  CD->isDeleted()) {
-                Diag(ELoc, diag::err_omp_required_method)
-                    << getOpenMPClauseName(OMPC_lastprivate) << 0;
-                bool IsDecl = VD->isThisDeclarationADefinition(Context) ==
-                              VarDecl::DeclarationOnly;
-                Diag(VD->getLocation(), IsDecl ? diag::note_previous_decl
-                                               : diag::note_defined_here)
-                    << VD;
-                Diag(RD->getLocation(), diag::note_previous_decl) << RD;
-                continue;
-              }
-              MarkFunctionReferenced(ELoc, CD);
-              DiagnoseUseOfDecl(CD, ELoc);
-            }
+            // Generate helper private variable and initialize it with the
+            // default value. The address of the original variable is replaced
+            // by the address of the new private variable in CodeGen. This new
+            // variable is not added to IdResolver, so the code in the OpenMP
+            // region uses original variable for proper diagnostics.
+            auto *VDPrivate =
+                buildVarDecl(*this, DE->getExprLoc(), Type.getUnqualifiedType(),
+                             VD->getName());
+            ActOnUninitializedDecl(VDPrivate, /*TypeMayContainAuto=*/false);
+            if (VDPrivate->isInvalidDecl())
+              continue;
+            PrivateCopies.push_back(buildDeclRefExpr(
+                *this, VDPrivate, DE->getType(), DE->getExprLoc()));
+          } else {
+            // The variable is also a firstprivate, so initialization sequence
+            // for private copy is generated already.
+            PrivateCopies.push_back(nullptr);
           }
         }
+        // Set initializers to private copies if no errors were found.
+        if (PrivateCopies.size() == Clause->varlist_size()) {
+          Clause->setPrivateCopies(PrivateCopies);
+        }
       }
     }
   }
@@ -630,6 +695,10 @@ void Sema::EndOpenMPDSABlock(Stmt *CurDirective) {
   PopExpressionEvaluationContext();
 }
 
+static bool FinishOpenMPLinearClause(OMPLinearClause &Clause, DeclRefExpr *IV,
+                                     Expr *NumIterations, Sema &SemaRef,
+                                     Scope *S);
+
 namespace {
 
 class VarDeclFilterCCC : public CorrectionCandidateCallback {
@@ -763,14 +832,14 @@ ExprResult Sema::ActOnOpenMPIdExpression(Scope *CurScope,
   // OpenMP [2.9.2, Restrictions, C/C++, p.2-6]
   //   A threadprivate directive must lexically precede all references to any
   //   of the variables in its list.
-  if (VD->isUsed()) {
+  if (VD->isUsed() && !DSAStack->isThreadPrivate(VD)) {
     Diag(Id.getLoc(), diag::err_omp_var_used)
         << getOpenMPDirectiveName(OMPD_threadprivate) << VD;
     return ExprError();
   }
 
   QualType ExprType = VD->getType().getNonReferenceType();
-  ExprResult DE = BuildDeclRefExpr(VD, ExprType, VK_LValue, Id.getLoc());
+  ExprResult DE = buildDeclRefExpr(*this, VD, ExprType, Id.getLoc());
   return DE;
 }
 
@@ -821,6 +890,13 @@ Sema::CheckOMPThreadPrivateDecl(SourceLocation Loc, ArrayRef<Expr *> VarList) {
     VarDecl *VD = cast<VarDecl>(DE->getDecl());
     SourceLocation ILoc = DE->getExprLoc();
 
+    QualType QType = VD->getType();
+    if (QType->isDependentType() || QType->isInstantiationDependentType()) {
+      // It will be analyzed later.
+      Vars.push_back(DE);
+      continue;
+    }
+
     // OpenMP [2.9.2, Restrictions, C/C++, p.10]
     //   A threadprivate variable must not have an incomplete type.
     if (RequireCompleteType(ILoc, VD->getType(),
@@ -843,7 +919,8 @@ Sema::CheckOMPThreadPrivateDecl(SourceLocation Loc, ArrayRef<Expr *> VarList) {
 
     // Check if this is a TLS variable.
     if (VD->getTLSKind() != VarDecl::TLS_None ||
-        VD->getStorageClass() == SC_Register) {
+        (VD->getStorageClass() == SC_Register && VD->hasAttr<AsmLabelAttr>() &&
+         !VD->isLocalVarDecl())) {
       Diag(ILoc, diag::err_omp_var_thread_local)
           << VD << ((VD->getTLSKind() != VarDecl::TLS_None) ? 0 : 1);
       bool IsDecl =
@@ -1121,7 +1198,11 @@ void Sema::ActOnOpenMPRegionStart(OpenMPDirectiveKind DKind, Scope *CurScope) {
     break;
   }
   case OMPD_parallel_sections: {
+    QualType KmpInt32Ty = Context.getIntTypeForBitwidth(32, 1);
+    QualType KmpInt32PtrTy = Context.getPointerType(KmpInt32Ty);
     Sema::CapturedParamNameType Params[] = {
+        std::make_pair(".global_tid.", KmpInt32PtrTy),
+        std::make_pair(".bound_tid.", KmpInt32PtrTy),
         std::make_pair(StringRef(), QualType()) // __context with shared vars
     };
     ActOnCapturedRegionStart(DSAStack->getConstructLoc(), CurScope, CR_OpenMP,
@@ -1129,11 +1210,28 @@ void Sema::ActOnOpenMPRegionStart(OpenMPDirectiveKind DKind, Scope *CurScope) {
     break;
   }
   case OMPD_task: {
+    QualType KmpInt32Ty = Context.getIntTypeForBitwidth(32, 1);
+    QualType Args[] = {Context.VoidPtrTy.withConst().withRestrict()};
+    FunctionProtoType::ExtProtoInfo EPI;
+    EPI.Variadic = true;
+    QualType CopyFnType = Context.getFunctionType(Context.VoidTy, Args, EPI);
     Sema::CapturedParamNameType Params[] = {
+        std::make_pair(".global_tid.", KmpInt32Ty),
+        std::make_pair(".part_id.", KmpInt32Ty),
+        std::make_pair(".privates.",
+                       Context.VoidPtrTy.withConst().withRestrict()),
+        std::make_pair(
+            ".copy_fn.",
+            Context.getPointerType(CopyFnType).withConst().withRestrict()),
         std::make_pair(StringRef(), QualType()) // __context with shared vars
     };
     ActOnCapturedRegionStart(DSAStack->getConstructLoc(), CurScope, CR_OpenMP,
                              Params);
+    // Mark this captured region as inlined, because we don't use outlined
+    // function directly.
+    getCurCapturedRegion()->TheCapturedDecl->addAttr(
+        AlwaysInlineAttr::CreateImplicit(
+            Context, AlwaysInlineAttr::Keyword_forceinline, SourceRange()));
     break;
   }
   case OMPD_ordered: {
@@ -1183,6 +1281,36 @@ void Sema::ActOnOpenMPRegionStart(OpenMPDirectiveKind DKind, Scope *CurScope) {
   }
 }
 
+StmtResult Sema::ActOnOpenMPRegionEnd(StmtResult S,
+                                      ArrayRef<OMPClause *> Clauses) {
+  if (!S.isUsable()) {
+    ActOnCapturedRegionError();
+    return StmtError();
+  }
+  // This is required for proper codegen.
+  for (auto *Clause : Clauses) {
+    if (isOpenMPPrivate(Clause->getClauseKind()) ||
+        Clause->getClauseKind() == OMPC_copyprivate) {
+      // Mark all variables in private list clauses as used in inner region.
+      for (auto *VarRef : Clause->children()) {
+        if (auto *E = cast_or_null<Expr>(VarRef)) {
+          MarkDeclarationsReferencedInExpr(E);
+        }
+      }
+    } else if (isParallelOrTaskRegion(DSAStack->getCurrentDirective()) &&
+               Clause->getClauseKind() == OMPC_schedule) {
+      // Mark all variables in private list clauses as used in inner region.
+      // Required for proper codegen of combined directives.
+      // TODO: add processing for other clauses.
+      if (auto *E = cast_or_null<Expr>(
+              cast<OMPScheduleClause>(Clause)->getHelperChunkSize())) {
+          MarkDeclarationsReferencedInExpr(E);
+        }
+    }
+  }
+  return ActOnCapturedRegionEnd(S.get());
+}
+
 static bool CheckNestingOfRegions(Sema &SemaRef, DSAStackTy *Stack,
                                   OpenMPDirectiveKind CurrentRegion,
                                   const DeclarationNameInfo &CurrentName,
@@ -1914,7 +2042,7 @@ public:
         TestIsStrictOp(false), SubtractStep(false) {}
   /// \brief Check init-expr for canonical loop form and save loop counter
   /// variable - #Var and its initialization value - #LB.
-  bool CheckInit(Stmt *S);
+  bool CheckInit(Stmt *S, bool EmitDiags = true);
   /// \brief Check test-expr for canonical form, save upper-bound (#UB), flags
   /// for less/greater and for strict/non-strict comparison.
   bool CheckCond(Expr *S);
@@ -1935,6 +2063,8 @@ public:
   bool ShouldSubtractStep() const { return SubtractStep; }
   /// \brief Build the expression to calculate the number of iterations.
   Expr *BuildNumIterations(Scope *S, const bool LimitedType) const;
+  /// \brief Build the precondition expression for the loops.
+  Expr *BuildPreCond(Scope *S, Expr *Cond) const;
   /// \brief Build reference expression to the counter be used for codegen.
   Expr *BuildCounterVar() const;
   /// \brief Build initization of the counter be used for codegen.
@@ -2052,7 +2182,7 @@ bool OpenMPIterationSpaceChecker::SetStep(Expr *NewStep, bool Subtract) {
   return false;
 }
 
-bool OpenMPIterationSpaceChecker::CheckInit(Stmt *S) {
+bool OpenMPIterationSpaceChecker::CheckInit(Stmt *S, bool EmitDiags) {
   // Check init-expr for canonical loop form and save loop counter
   // variable - #Var and its initialization value - #LB.
   // OpenMP [2.6] Canonical loop form. init-expr may be one of the following:
@@ -2062,7 +2192,9 @@ bool OpenMPIterationSpaceChecker::CheckInit(Stmt *S) {
   //   pointer-type var = lb
   //
   if (!S) {
-    SemaRef.Diag(DefaultLoc, diag::err_omp_loop_not_canonical_init);
+    if (EmitDiags) {
+      SemaRef.Diag(DefaultLoc, diag::err_omp_loop_not_canonical_init);
+    }
     return true;
   }
   InitSrcRange = S->getSourceRange();
@@ -2078,7 +2210,7 @@ bool OpenMPIterationSpaceChecker::CheckInit(Stmt *S) {
       if (auto Var = dyn_cast_or_null<VarDecl>(DS->getSingleDecl())) {
         if (Var->hasInit()) {
           // Accept non-canonical init form here but emit ext. warning.
-          if (Var->getInitStyle() != VarDecl::CInit)
+          if (Var->getInitStyle() != VarDecl::CInit && EmitDiags)
             SemaRef.Diag(S->getLocStart(),
                          diag::ext_omp_loop_not_canonical_init)
                 << S->getSourceRange();
@@ -2092,8 +2224,10 @@ bool OpenMPIterationSpaceChecker::CheckInit(Stmt *S) {
         return SetVarAndLB(dyn_cast<VarDecl>(DRE->getDecl()), DRE,
                            CE->getArg(1));
 
-  SemaRef.Diag(S->getLocStart(), diag::err_omp_loop_not_canonical_init)
-      << S->getSourceRange();
+  if (EmitDiags) {
+    SemaRef.Diag(S->getLocStart(), diag::err_omp_loop_not_canonical_init)
+        << S->getSourceRange();
+  }
   return true;
 }
 
@@ -2338,11 +2472,22 @@ OpenMPIterationSpaceChecker::BuildNumIterations(Scope *S,
   return Diff.get();
 }
 
+Expr *OpenMPIterationSpaceChecker::BuildPreCond(Scope *S, Expr *Cond) const {
+  // Try to build LB <op> UB, where <op> is <, >, <=, or >=.
+  bool Suppress = SemaRef.getDiagnostics().getSuppressAllDiagnostics();
+  SemaRef.getDiagnostics().setSuppressAllDiagnostics(/*Val=*/true);
+  auto CondExpr = SemaRef.BuildBinOp(
+      S, DefaultLoc, TestIsLessOp ? (TestIsStrictOp ? BO_LT : BO_LE)
+                                  : (TestIsStrictOp ? BO_GT : BO_GE),
+      LB, UB);
+  SemaRef.getDiagnostics().setSuppressAllDiagnostics(Suppress);
+  // Otherwise use original loop conditon and evaluate it in runtime.
+  return CondExpr.isUsable() ? CondExpr.get() : Cond;
+}
+
 /// \brief Build reference expression to the counter be used for codegen.
 Expr *OpenMPIterationSpaceChecker::BuildCounterVar() const {
-  return DeclRefExpr::Create(SemaRef.Context, NestedNameSpecifierLoc(),
-                             GetIncrementSrcRange().getBegin(), Var, false,
-                             DefaultLoc, Var->getType(), VK_LValue);
+  return buildDeclRefExpr(SemaRef, Var, Var->getType(), DefaultLoc);
 }
 
 /// \brief Build initization of the counter be used for codegen.
@@ -2353,6 +2498,8 @@ Expr *OpenMPIterationSpaceChecker::BuildCounterStep() const { return Step; }
 
 /// \brief Iteration space of a single for loop.
 struct LoopIterationSpace {
+  /// \brief Condition of the loop.
+  Expr *PreCond;
   /// \brief This expression calculates the number of iterations in the loop.
   /// It is always possible to calculate it before starting the loop.
   Expr *NumIterations;
@@ -2375,6 +2522,20 @@ struct LoopIterationSpace {
 
 } // namespace
 
+void Sema::ActOnOpenMPLoopInitialization(SourceLocation ForLoc, Stmt *Init) {
+  assert(getLangOpts().OpenMP && "OpenMP is not active.");
+  assert(Init && "Expected loop in canonical form.");
+  unsigned CollapseIteration = DSAStack->getCollapseNumber();
+  if (CollapseIteration > 0 &&
+      isOpenMPLoopDirective(DSAStack->getCurrentDirective())) {
+    OpenMPIterationSpaceChecker ISC(*this, ForLoc);
+    if (!ISC.CheckInit(Init, /*EmitDiags=*/false)) {
+      DSAStack->addLoopControlVariable(ISC.GetLoopVar());
+    }
+    DSAStack->setCollapseNumber(CollapseIteration - 1);
+  }
+}
+
 /// \brief Called on a for stmt to check and extract its iteration space
 /// for further processing (such as collapsing).
 static bool CheckOpenMPIterationSpace(
@@ -2453,32 +2614,27 @@ static bool CheckOpenMPIterationSpace(
           ? ((NestedLoopCount == 1) ? OMPC_linear : OMPC_lastprivate)
           : OMPC_private;
   if (((isOpenMPSimdDirective(DKind) && DVar.CKind != OMPC_unknown &&
-        DVar.CKind != PredeterminedCKind) ||
+        DVar.CKind != OMPC_threadprivate && DVar.CKind != PredeterminedCKind) ||
        (isOpenMPWorksharingDirective(DKind) && !isOpenMPSimdDirective(DKind) &&
         DVar.CKind != OMPC_unknown && DVar.CKind != OMPC_private &&
-        DVar.CKind != OMPC_lastprivate)) &&
-      (DVar.CKind != OMPC_private || DVar.RefExpr != nullptr)) {
+        DVar.CKind != OMPC_lastprivate && DVar.CKind != OMPC_threadprivate)) &&
+      ((DVar.CKind != OMPC_private && DVar.CKind != OMPC_threadprivate) ||
+       DVar.RefExpr != nullptr)) {
     SemaRef.Diag(Init->getLocStart(), diag::err_omp_loop_var_dsa)
         << getOpenMPClauseName(DVar.CKind) << getOpenMPDirectiveName(DKind)
         << getOpenMPClauseName(PredeterminedCKind);
-    ReportOriginalDSA(SemaRef, &DSA, Var, DVar, true);
+    if (DVar.RefExpr == nullptr)
+      DVar.CKind = PredeterminedCKind;
+    ReportOriginalDSA(SemaRef, &DSA, Var, DVar, /*IsLoopIterVar=*/true);
     HasErrors = true;
   } else if (LoopVarRefExpr != nullptr) {
     // Make the loop iteration variable private (for worksharing constructs),
     // linear (for simd directives with the only one associated loop) or
     // lastprivate (for simd directives with several collapsed loops).
-    // FIXME: the next check and error message must be removed once the
-    // capturing of global variables in loops is fixed.
     if (DVar.CKind == OMPC_unknown)
       DVar = DSA.hasDSA(Var, isOpenMPPrivate, MatchesAlways(),
                         /*FromParent=*/false);
-    if (!Var->hasLocalStorage() && DVar.CKind == OMPC_unknown) {
-      SemaRef.Diag(Init->getLocStart(), diag::err_omp_global_loop_var_dsa)
-          << getOpenMPClauseName(PredeterminedCKind)
-          << getOpenMPDirectiveName(DKind);
-      HasErrors = true;
-    } else
-      DSA.addDSA(Var, LoopVarRefExpr, PredeterminedCKind);
+    DSA.addDSA(Var, LoopVarRefExpr, PredeterminedCKind);
   }
 
   assert(isOpenMPLoopDirective(DKind) && "DSA for non-loop vars");
@@ -2493,6 +2649,7 @@ static bool CheckOpenMPIterationSpace(
     return HasErrors;
 
   // Build the loop's iteration space representation.
+  ResultIterSpace.PreCond = ISC.BuildPreCond(DSA.getCurScope(), For->getCond());
   ResultIterSpace.NumIterations = ISC.BuildNumIterations(
       DSA.getCurScope(), /* LimitedType */ isOpenMPWorksharingDirective(DKind));
   ResultIterSpace.CounterVar = ISC.BuildCounterVar();
@@ -2503,7 +2660,8 @@ static bool CheckOpenMPIterationSpace(
   ResultIterSpace.IncSrcRange = ISC.GetIncrementSrcRange();
   ResultIterSpace.Subtract = ISC.ShouldSubtractStep();
 
-  HasErrors |= (ResultIterSpace.NumIterations == nullptr ||
+  HasErrors |= (ResultIterSpace.PreCond == nullptr ||
+                ResultIterSpace.NumIterations == nullptr ||
                 ResultIterSpace.CounterVar == nullptr ||
                 ResultIterSpace.CounterInit == nullptr ||
                 ResultIterSpace.CounterStep == nullptr);
@@ -2511,18 +2669,6 @@ static bool CheckOpenMPIterationSpace(
   return HasErrors;
 }
 
-/// \brief Build a variable declaration for OpenMP loop iteration variable.
-static VarDecl *BuildVarDecl(Sema &SemaRef, SourceLocation Loc, QualType Type,
-                             StringRef Name) {
-  DeclContext *DC = SemaRef.CurContext;
-  IdentifierInfo *II = &SemaRef.PP.getIdentifierTable().get(Name);
-  TypeSourceInfo *TInfo = SemaRef.Context.getTrivialTypeSourceInfo(Type, Loc);
-  VarDecl *Decl =
-      VarDecl::Create(SemaRef.Context, DC, Loc, Loc, II, Type, TInfo, SC_None);
-  Decl->setImplicit();
-  return Decl;
-}
-
 /// \brief Build 'VarRef = Start + Iter * Step'.
 static ExprResult BuildCounterUpdate(Sema &SemaRef, Scope *S,
                                      SourceLocation Loc, ExprResult VarRef,
@@ -2648,6 +2794,9 @@ CheckOpenMPLoop(OpenMPDirectiveKind DKind, Expr *NestedLoopCountExpr,
 
   // Last iteration number is (I1 * I2 * ... In) - 1, where I1, I2 ... In are
   // the iteration counts of the collapsed for loops.
+  // Precondition tests if there is at least one iteration (all conditions are
+  // true).
+  auto PreCond = ExprResult(IterSpaces[0].PreCond);
   auto N0 = IterSpaces[0].NumIterations;
   ExprResult LastIteration32 = WidenIterationCount(32 /* Bits */, N0, SemaRef);
   ExprResult LastIteration64 = WidenIterationCount(64 /* Bits */, N0, SemaRef);
@@ -2660,6 +2809,10 @@ CheckOpenMPLoop(OpenMPDirectiveKind DKind, Expr *NestedLoopCountExpr,
 
   Scope *CurScope = DSA.getCurScope();
   for (unsigned Cnt = 1; Cnt < NestedLoopCount; ++Cnt) {
+    if (PreCond.isUsable()) {
+      PreCond = SemaRef.BuildBinOp(CurScope, SourceLocation(), BO_LAnd,
+                                   PreCond.get(), IterSpaces[Cnt].PreCond);
+    }
     auto N = IterSpaces[Cnt].NumIterations;
     AllCountsNeedLessThan32Bits &= C.getTypeSize(N->getType()) < 32;
     if (LastIteration32.isUsable())
@@ -2703,10 +2856,10 @@ CheckOpenMPLoop(OpenMPDirectiveKind DKind, Expr *NestedLoopCountExpr,
   if (!IsConstant) {
     SourceLocation SaveLoc;
     VarDecl *SaveVar =
-        BuildVarDecl(SemaRef, SaveLoc, LastIteration.get()->getType(),
+        buildVarDecl(SemaRef, SaveLoc, LastIteration.get()->getType(),
                      ".omp.last.iteration");
-    ExprResult SaveRef = SemaRef.BuildDeclRefExpr(
-        SaveVar, LastIteration.get()->getType(), VK_LValue, SaveLoc);
+    ExprResult SaveRef = buildDeclRefExpr(
+        SemaRef, SaveVar, LastIteration.get()->getType(), SaveLoc);
     CalcLastIteration = SemaRef.BuildBinOp(CurScope, SaveLoc, BO_Assign,
                                            SaveRef.get(), LastIteration.get());
     LastIteration = SaveRef;
@@ -2721,25 +2874,20 @@ CheckOpenMPLoop(OpenMPDirectiveKind DKind, Expr *NestedLoopCountExpr,
 
   SourceLocation InitLoc = IterSpaces[0].InitSrcRange.getBegin();
 
-  // Precondition tests if there is at least one iteration (LastIteration > 0).
-  ExprResult PreCond = SemaRef.BuildBinOp(
-      CurScope, InitLoc, BO_GT, LastIteration.get(),
-      SemaRef.ActOnIntegerConstant(SourceLocation(), 0).get());
-
   QualType VType = LastIteration.get()->getType();
   // Build variables passed into runtime, nesessary for worksharing directives.
   ExprResult LB, UB, IL, ST, EUB;
   if (isOpenMPWorksharingDirective(DKind)) {
     // Lower bound variable, initialized with zero.
-    VarDecl *LBDecl = BuildVarDecl(SemaRef, InitLoc, VType, ".omp.lb");
-    LB = SemaRef.BuildDeclRefExpr(LBDecl, VType, VK_LValue, InitLoc);
+    VarDecl *LBDecl = buildVarDecl(SemaRef, InitLoc, VType, ".omp.lb");
+    LB = buildDeclRefExpr(SemaRef, LBDecl, VType, InitLoc);
     SemaRef.AddInitializerToDecl(
         LBDecl, SemaRef.ActOnIntegerConstant(InitLoc, 0).get(),
         /*DirectInit*/ false, /*TypeMayContainAuto*/ false);
 
     // Upper bound variable, initialized with last iteration number.
-    VarDecl *UBDecl = BuildVarDecl(SemaRef, InitLoc, VType, ".omp.ub");
-    UB = SemaRef.BuildDeclRefExpr(UBDecl, VType, VK_LValue, InitLoc);
+    VarDecl *UBDecl = buildVarDecl(SemaRef, InitLoc, VType, ".omp.ub");
+    UB = buildDeclRefExpr(SemaRef, UBDecl, VType, InitLoc);
     SemaRef.AddInitializerToDecl(UBDecl, LastIteration.get(),
                                  /*DirectInit*/ false,
                                  /*TypeMayContainAuto*/ false);
@@ -2747,15 +2895,15 @@ CheckOpenMPLoop(OpenMPDirectiveKind DKind, Expr *NestedLoopCountExpr,
     // A 32-bit variable-flag where runtime returns 1 for the last iteration.
     // This will be used to implement clause 'lastprivate'.
     QualType Int32Ty = SemaRef.Context.getIntTypeForBitwidth(32, true);
-    VarDecl *ILDecl = BuildVarDecl(SemaRef, InitLoc, Int32Ty, ".omp.is_last");
-    IL = SemaRef.BuildDeclRefExpr(ILDecl, Int32Ty, VK_LValue, InitLoc);
+    VarDecl *ILDecl = buildVarDecl(SemaRef, InitLoc, Int32Ty, ".omp.is_last");
+    IL = buildDeclRefExpr(SemaRef, ILDecl, Int32Ty, InitLoc);
     SemaRef.AddInitializerToDecl(
         ILDecl, SemaRef.ActOnIntegerConstant(InitLoc, 0).get(),
         /*DirectInit*/ false, /*TypeMayContainAuto*/ false);
 
     // Stride variable returned by runtime (we initialize it to 1 by default).
-    VarDecl *STDecl = BuildVarDecl(SemaRef, InitLoc, VType, ".omp.stride");
-    ST = SemaRef.BuildDeclRefExpr(STDecl, VType, VK_LValue, InitLoc);
+    VarDecl *STDecl = buildVarDecl(SemaRef, InitLoc, VType, ".omp.stride");
+    ST = buildDeclRefExpr(SemaRef, STDecl, VType, InitLoc);
     SemaRef.AddInitializerToDecl(
         STDecl, SemaRef.ActOnIntegerConstant(InitLoc, 1).get(),
         /*DirectInit*/ false, /*TypeMayContainAuto*/ false);
@@ -2775,8 +2923,8 @@ CheckOpenMPLoop(OpenMPDirectiveKind DKind, Expr *NestedLoopCountExpr,
   ExprResult IV;
   ExprResult Init;
   {
-    VarDecl *IVDecl = BuildVarDecl(SemaRef, InitLoc, VType, ".omp.iv");
-    IV = SemaRef.BuildDeclRefExpr(IVDecl, VType, VK_LValue, InitLoc);
+    VarDecl *IVDecl = buildVarDecl(SemaRef, InitLoc, VType, ".omp.iv");
+    IV = buildDeclRefExpr(SemaRef, IVDecl, VType, InitLoc);
     Expr *RHS = isOpenMPWorksharingDirective(DKind)
                     ? LB.get()
                     : SemaRef.ActOnIntegerConstant(SourceLocation(), 0).get();
@@ -2864,9 +3012,13 @@ CheckOpenMPLoop(OpenMPDirectiveKind DKind, Expr *NestedLoopCountExpr,
         break;
       }
 
-      // Build update: IS.CounterVar = IS.Start + Iter * IS.Step
+      // Build update: IS.CounterVar(Private) = IS.Start + Iter * IS.Step
+      auto *CounterVar = buildDeclRefExpr(
+          SemaRef, cast<VarDecl>(cast<DeclRefExpr>(IS.CounterVar)->getDecl()),
+          IS.CounterVar->getType(), IS.CounterVar->getExprLoc(),
+          /*RefersToCapture=*/true);
       ExprResult Update =
-          BuildCounterUpdate(SemaRef, CurScope, UpdLoc, IS.CounterVar,
+          BuildCounterUpdate(SemaRef, CurScope, UpdLoc, CounterVar,
                              IS.CounterInit, Iter, IS.CounterStep, IS.Subtract);
       if (!Update.isUsable()) {
         HasErrors = true;
@@ -2875,7 +3027,7 @@ CheckOpenMPLoop(OpenMPDirectiveKind DKind, Expr *NestedLoopCountExpr,
 
       // Build final: IS.CounterVar = IS.Start + IS.NumIters * IS.Step
       ExprResult Final = BuildCounterUpdate(
-          SemaRef, CurScope, UpdLoc, IS.CounterVar, IS.CounterInit,
+          SemaRef, CurScope, UpdLoc, CounterVar, IS.CounterInit,
           IS.NumIterations, IS.CounterStep, IS.Subtract);
       if (!Final.isUsable()) {
         HasErrors = true;
@@ -2915,6 +3067,7 @@ CheckOpenMPLoop(OpenMPDirectiveKind DKind, Expr *NestedLoopCountExpr,
   // Save results
   Built.IterationVarRef = IV.get();
   Built.LastIteration = LastIteration.get();
+  Built.NumIterations = NumIterations.get();
   Built.CalcLastIteration = CalcLastIteration.get();
   Built.PreCond = PreCond.get();
   Built.Cond = Cond.get();
@@ -2933,11 +3086,11 @@ CheckOpenMPLoop(OpenMPDirectiveKind DKind, Expr *NestedLoopCountExpr,
 }
 
 static Expr *GetCollapseNumberExpr(ArrayRef<OMPClause *> Clauses) {
-  auto CollapseFilter = [](const OMPClause *C) -> bool {
+  auto &&CollapseFilter = [](const OMPClause *C) -> bool {
     return C->getClauseKind() == OMPC_collapse;
   };
   OMPExecutableDirective::filtered_clause_iterator<decltype(CollapseFilter)> I(
-      Clauses, CollapseFilter);
+      Clauses, std::move(CollapseFilter));
   if (I)
     return cast<OMPCollapseClause>(*I)->getNumForLoops();
   return nullptr;
@@ -2958,6 +3111,16 @@ StmtResult Sema::ActOnOpenMPSimdDirective(
   assert((CurContext->isDependentContext() || B.builtAll()) &&
          "omp simd loop exprs were not built");
 
+  if (!CurContext->isDependentContext()) {
+    // Finalize the clauses that need pre-built expressions for CodeGen.
+    for (auto C : Clauses) {
+      if (auto LC = dyn_cast<OMPLinearClause>(C))
+        if (FinishOpenMPLinearClause(*LC, cast<DeclRefExpr>(B.IterationVarRef),
+                                     B.NumIterations, *this, CurScope))
+          return StmtError();
+    }
+  }
+
   getCurFunction()->setHasBranchProtectedScope();
   return OMPSimdDirective::Create(Context, StartLoc, EndLoc, NestedLoopCount,
                                   Clauses, AStmt, B);
@@ -3055,6 +3218,23 @@ StmtResult Sema::ActOnOpenMPSingleDirective(ArrayRef<OMPClause *> Clauses,
 
   getCurFunction()->setHasBranchProtectedScope();
 
+  // OpenMP [2.7.3, single Construct, Restrictions]
+  // The copyprivate clause must not be used with the nowait clause.
+  OMPClause *Nowait = nullptr;
+  OMPClause *Copyprivate = nullptr;
+  for (auto *Clause : Clauses) {
+    if (Clause->getClauseKind() == OMPC_nowait)
+      Nowait = Clause;
+    else if (Clause->getClauseKind() == OMPC_copyprivate)
+      Copyprivate = Clause;
+    if (Copyprivate && Nowait) {
+      Diag(Copyprivate->getLocStart(),
+           diag::err_omp_single_copyprivate_with_nowait);
+      Diag(Nowait->getLocStart(), diag::note_omp_nowait_clause_here);
+      return StmtError();
+    }
+  }
+
   return OMPSingleDirective::Create(Context, StartLoc, EndLoc, Clauses, AStmt);
 }
 
@@ -3219,6 +3399,246 @@ StmtResult Sema::ActOnOpenMPOrderedDirective(Stmt *AStmt,
   return OMPOrderedDirective::Create(Context, StartLoc, EndLoc, AStmt);
 }
 
+namespace {
+/// \brief Helper class for checking expression in 'omp atomic [update]'
+/// construct.
+class OpenMPAtomicUpdateChecker {
+  /// \brief Error results for atomic update expressions.
+  enum ExprAnalysisErrorCode {
+    /// \brief A statement is not an expression statement.
+    NotAnExpression,
+    /// \brief Expression is not builtin binary or unary operation.
+    NotABinaryOrUnaryExpression,
+    /// \brief Unary operation is not post-/pre- increment/decrement operation.
+    NotAnUnaryIncDecExpression,
+    /// \brief An expression is not of scalar type.
+    NotAScalarType,
+    /// \brief A binary operation is not an assignment operation.
+    NotAnAssignmentOp,
+    /// \brief RHS part of the binary operation is not a binary expression.
+    NotABinaryExpression,
+    /// \brief RHS part is not additive/multiplicative/shift/biwise binary
+    /// expression.
+    NotABinaryOperator,
+    /// \brief RHS binary operation does not have reference to the updated LHS
+    /// part.
+    NotAnUpdateExpression,
+    /// \brief No errors is found.
+    NoError
+  };
+  /// \brief Reference to Sema.
+  Sema &SemaRef;
+  /// \brief A location for note diagnostics (when error is found).
+  SourceLocation NoteLoc;
+  /// \brief 'x' lvalue part of the source atomic expression.
+  Expr *X;
+  /// \brief 'expr' rvalue part of the source atomic expression.
+  Expr *E;
+  /// \brief Helper expression of the form
+  /// 'OpaqueValueExpr(x) binop OpaqueValueExpr(expr)' or
+  /// 'OpaqueValueExpr(expr) binop OpaqueValueExpr(x)'.
+  Expr *UpdateExpr;
+  /// \brief Is 'x' a LHS in a RHS part of full update expression. It is
+  /// important for non-associative operations.
+  bool IsXLHSInRHSPart;
+  BinaryOperatorKind Op;
+  SourceLocation OpLoc;
+  /// \brief true if the source expression is a postfix unary operation, false
+  /// if it is a prefix unary operation.
+  bool IsPostfixUpdate;
+
+public:
+  OpenMPAtomicUpdateChecker(Sema &SemaRef)
+      : SemaRef(SemaRef), X(nullptr), E(nullptr), UpdateExpr(nullptr),
+        IsXLHSInRHSPart(false), Op(BO_PtrMemD), IsPostfixUpdate(false) {}
+  /// \brief Check specified statement that it is suitable for 'atomic update'
+  /// constructs and extract 'x', 'expr' and Operation from the original
+  /// expression. If DiagId and NoteId == 0, then only check is performed
+  /// without error notification.
+  /// \param DiagId Diagnostic which should be emitted if error is found.
+  /// \param NoteId Diagnostic note for the main error message.
+  /// \return true if statement is not an update expression, false otherwise.
+  bool checkStatement(Stmt *S, unsigned DiagId = 0, unsigned NoteId = 0);
+  /// \brief Return the 'x' lvalue part of the source atomic expression.
+  Expr *getX() const { return X; }
+  /// \brief Return the 'expr' rvalue part of the source atomic expression.
+  Expr *getExpr() const { return E; }
+  /// \brief Return the update expression used in calculation of the updated
+  /// value. Always has form 'OpaqueValueExpr(x) binop OpaqueValueExpr(expr)' or
+  /// 'OpaqueValueExpr(expr) binop OpaqueValueExpr(x)'.
+  Expr *getUpdateExpr() const { return UpdateExpr; }
+  /// \brief Return true if 'x' is LHS in RHS part of full update expression,
+  /// false otherwise.
+  bool isXLHSInRHSPart() const { return IsXLHSInRHSPart; }
+
+  /// \brief true if the source expression is a postfix unary operation, false
+  /// if it is a prefix unary operation.
+  bool isPostfixUpdate() const { return IsPostfixUpdate; }
+
+private:
+  bool checkBinaryOperation(BinaryOperator *AtomicBinOp, unsigned DiagId = 0,
+                            unsigned NoteId = 0);
+};
+} // namespace
+
+bool OpenMPAtomicUpdateChecker::checkBinaryOperation(
+    BinaryOperator *AtomicBinOp, unsigned DiagId, unsigned NoteId) {
+  ExprAnalysisErrorCode ErrorFound = NoError;
+  SourceLocation ErrorLoc, NoteLoc;
+  SourceRange ErrorRange, NoteRange;
+  // Allowed constructs are:
+  //  x = x binop expr;
+  //  x = expr binop x;
+  if (AtomicBinOp->getOpcode() == BO_Assign) {
+    X = AtomicBinOp->getLHS();
+    if (auto *AtomicInnerBinOp = dyn_cast<BinaryOperator>(
+            AtomicBinOp->getRHS()->IgnoreParenImpCasts())) {
+      if (AtomicInnerBinOp->isMultiplicativeOp() ||
+          AtomicInnerBinOp->isAdditiveOp() || AtomicInnerBinOp->isShiftOp() ||
+          AtomicInnerBinOp->isBitwiseOp()) {
+        Op = AtomicInnerBinOp->getOpcode();
+        OpLoc = AtomicInnerBinOp->getOperatorLoc();
+        auto *LHS = AtomicInnerBinOp->getLHS();
+        auto *RHS = AtomicInnerBinOp->getRHS();
+        llvm::FoldingSetNodeID XId, LHSId, RHSId;
+        X->IgnoreParenImpCasts()->Profile(XId, SemaRef.getASTContext(),
+                                          /*Canonical=*/true);
+        LHS->IgnoreParenImpCasts()->Profile(LHSId, SemaRef.getASTContext(),
+                                            /*Canonical=*/true);
+        RHS->IgnoreParenImpCasts()->Profile(RHSId, SemaRef.getASTContext(),
+                                            /*Canonical=*/true);
+        if (XId == LHSId) {
+          E = RHS;
+          IsXLHSInRHSPart = true;
+        } else if (XId == RHSId) {
+          E = LHS;
+          IsXLHSInRHSPart = false;
+        } else {
+          ErrorLoc = AtomicInnerBinOp->getExprLoc();
+          ErrorRange = AtomicInnerBinOp->getSourceRange();
+          NoteLoc = X->getExprLoc();
+          NoteRange = X->getSourceRange();
+          ErrorFound = NotAnUpdateExpression;
+        }
+      } else {
+        ErrorLoc = AtomicInnerBinOp->getExprLoc();
+        ErrorRange = AtomicInnerBinOp->getSourceRange();
+        NoteLoc = AtomicInnerBinOp->getOperatorLoc();
+        NoteRange = SourceRange(NoteLoc, NoteLoc);
+        ErrorFound = NotABinaryOperator;
+      }
+    } else {
+      NoteLoc = ErrorLoc = AtomicBinOp->getRHS()->getExprLoc();
+      NoteRange = ErrorRange = AtomicBinOp->getRHS()->getSourceRange();
+      ErrorFound = NotABinaryExpression;
+    }
+  } else {
+    ErrorLoc = AtomicBinOp->getExprLoc();
+    ErrorRange = AtomicBinOp->getSourceRange();
+    NoteLoc = AtomicBinOp->getOperatorLoc();
+    NoteRange = SourceRange(NoteLoc, NoteLoc);
+    ErrorFound = NotAnAssignmentOp;
+  }
+  if (ErrorFound != NoError && DiagId != 0 && NoteId != 0) {
+    SemaRef.Diag(ErrorLoc, DiagId) << ErrorRange;
+    SemaRef.Diag(NoteLoc, NoteId) << ErrorFound << NoteRange;
+    return true;
+  } else if (SemaRef.CurContext->isDependentContext())
+    E = X = UpdateExpr = nullptr;
+  return ErrorFound != NoError;
+}
+
+bool OpenMPAtomicUpdateChecker::checkStatement(Stmt *S, unsigned DiagId,
+                                               unsigned NoteId) {
+  ExprAnalysisErrorCode ErrorFound = NoError;
+  SourceLocation ErrorLoc, NoteLoc;
+  SourceRange ErrorRange, NoteRange;
+  // Allowed constructs are:
+  //  x++;
+  //  x--;
+  //  ++x;
+  //  --x;
+  //  x binop= expr;
+  //  x = x binop expr;
+  //  x = expr binop x;
+  if (auto *AtomicBody = dyn_cast<Expr>(S)) {
+    AtomicBody = AtomicBody->IgnoreParenImpCasts();
+    if (AtomicBody->getType()->isScalarType() ||
+        AtomicBody->isInstantiationDependent()) {
+      if (auto *AtomicCompAssignOp = dyn_cast<CompoundAssignOperator>(
+              AtomicBody->IgnoreParenImpCasts())) {
+        // Check for Compound Assignment Operation
+        Op = BinaryOperator::getOpForCompoundAssignment(
+            AtomicCompAssignOp->getOpcode());
+        OpLoc = AtomicCompAssignOp->getOperatorLoc();
+        E = AtomicCompAssignOp->getRHS();
+        X = AtomicCompAssignOp->getLHS();
+        IsXLHSInRHSPart = true;
+      } else if (auto *AtomicBinOp = dyn_cast<BinaryOperator>(
+                     AtomicBody->IgnoreParenImpCasts())) {
+        // Check for Binary Operation
+        if(checkBinaryOperation(AtomicBinOp, DiagId, NoteId))
+          return true;
+      } else if (auto *AtomicUnaryOp =
+                 dyn_cast<UnaryOperator>(AtomicBody->IgnoreParenImpCasts())) {
+        // Check for Unary Operation
+        if (AtomicUnaryOp->isIncrementDecrementOp()) {
+          IsPostfixUpdate = AtomicUnaryOp->isPostfix();
+          Op = AtomicUnaryOp->isIncrementOp() ? BO_Add : BO_Sub;
+          OpLoc = AtomicUnaryOp->getOperatorLoc();
+          X = AtomicUnaryOp->getSubExpr();
+          E = SemaRef.ActOnIntegerConstant(OpLoc, /*uint64_t Val=*/1).get();
+          IsXLHSInRHSPart = true;
+        } else {
+          ErrorFound = NotAnUnaryIncDecExpression;
+          ErrorLoc = AtomicUnaryOp->getExprLoc();
+          ErrorRange = AtomicUnaryOp->getSourceRange();
+          NoteLoc = AtomicUnaryOp->getOperatorLoc();
+          NoteRange = SourceRange(NoteLoc, NoteLoc);
+        }
+      } else {
+        ErrorFound = NotABinaryOrUnaryExpression;
+        NoteLoc = ErrorLoc = AtomicBody->getExprLoc();
+        NoteRange = ErrorRange = AtomicBody->getSourceRange();
+      }
+    } else {
+      ErrorFound = NotAScalarType;
+      NoteLoc = ErrorLoc = AtomicBody->getLocStart();
+      NoteRange = ErrorRange = SourceRange(NoteLoc, NoteLoc);
+    }
+  } else {
+    ErrorFound = NotAnExpression;
+    NoteLoc = ErrorLoc = S->getLocStart();
+    NoteRange = ErrorRange = SourceRange(NoteLoc, NoteLoc);
+  }
+  if (ErrorFound != NoError && DiagId != 0 && NoteId != 0) {
+    SemaRef.Diag(ErrorLoc, DiagId) << ErrorRange;
+    SemaRef.Diag(NoteLoc, NoteId) << ErrorFound << NoteRange;
+    return true;
+  } else if (SemaRef.CurContext->isDependentContext())
+    E = X = UpdateExpr = nullptr;
+  if (ErrorFound == NoError && E && X) {
+    // Build an update expression of form 'OpaqueValueExpr(x) binop
+    // OpaqueValueExpr(expr)' or 'OpaqueValueExpr(expr) binop
+    // OpaqueValueExpr(x)' and then cast it to the type of the 'x' expression.
+    auto *OVEX = new (SemaRef.getASTContext())
+        OpaqueValueExpr(X->getExprLoc(), X->getType(), VK_RValue);
+    auto *OVEExpr = new (SemaRef.getASTContext())
+        OpaqueValueExpr(E->getExprLoc(), E->getType(), VK_RValue);
+    auto Update =
+        SemaRef.CreateBuiltinBinOp(OpLoc, Op, IsXLHSInRHSPart ? OVEX : OVEExpr,
+                                   IsXLHSInRHSPart ? OVEExpr : OVEX);
+    if (Update.isInvalid())
+      return true;
+    Update = SemaRef.PerformImplicitConversion(Update.get(), X->getType(),
+                                               Sema::AA_Casting);
+    if (Update.isInvalid())
+      return true;
+    UpdateExpr = Update.get();
+  }
+  return ErrorFound != NoError;
+}
+
 StmtResult Sema::ActOnOpenMPAtomicDirective(ArrayRef<OMPClause *> Clauses,
                                             Stmt *AStmt,
                                             SourceLocation StartLoc,
@@ -3230,7 +3650,6 @@ StmtResult Sema::ActOnOpenMPAtomicDirective(ArrayRef<OMPClause *> Clauses,
   // top and a single exit at the bottom.
   // The point of exit cannot be a branch out of the structured block.
   // longjmp() and throw() must not violate the entry/exit criteria.
-  // TODO further analysis of associated statements and clauses.
   OpenMPClauseKind AtomicKind = OMPC_unknown;
   SourceLocation AtomicKindLoc;
   for (auto *C : Clauses) {
@@ -3250,9 +3669,15 @@ StmtResult Sema::ActOnOpenMPAtomicDirective(ArrayRef<OMPClause *> Clauses,
   }
 
   auto Body = CS->getCapturedStmt();
+  if (auto *EWC = dyn_cast<ExprWithCleanups>(Body))
+    Body = EWC->getSubExpr();
+
   Expr *X = nullptr;
   Expr *V = nullptr;
   Expr *E = nullptr;
+  Expr *UE = nullptr;
+  bool IsXLHSInRHSPart = false;
+  bool IsPostfixUpdate = false;
   // OpenMP [2.12.6, atomic Construct]
   // In the next expressions:
   // * x and v (as applicable) are both l-value expressions with scalar type.
@@ -3275,14 +3700,14 @@ StmtResult Sema::ActOnOpenMPAtomicDirective(ArrayRef<OMPClause *> Clauses,
   // expr or subexpressions of expr.
   // * For forms that allow multiple occurrences of x, the number of times
   // that x is evaluated is unspecified.
-  enum {
-    NotAnExpression,
-    NotAnAssignmentOp,
-    NotAScalarType,
-    NotAnLValue,
-    NoError
-  } ErrorFound = NoError;
   if (AtomicKind == OMPC_read) {
+    enum {
+      NotAnExpression,
+      NotAnAssignmentOp,
+      NotAScalarType,
+      NotAnLValue,
+      NoError
+    } ErrorFound = NoError;
     SourceLocation ErrorLoc, NoteLoc;
     SourceRange ErrorRange, NoteRange;
     // If clause is read:
@@ -3338,6 +3763,13 @@ StmtResult Sema::ActOnOpenMPAtomicDirective(ArrayRef<OMPClause *> Clauses,
     } else if (CurContext->isDependentContext())
       V = X = nullptr;
   } else if (AtomicKind == OMPC_write) {
+    enum {
+      NotAnExpression,
+      NotAnAssignmentOp,
+      NotAScalarType,
+      NotAnLValue,
+      NoError
+    } ErrorFound = NoError;
     SourceLocation ErrorLoc, NoteLoc;
     SourceRange ErrorRange, NoteRange;
     // If clause is write:
@@ -3346,8 +3778,8 @@ StmtResult Sema::ActOnOpenMPAtomicDirective(ArrayRef<OMPClause *> Clauses,
       auto AtomicBinOp =
           dyn_cast<BinaryOperator>(AtomicBody->IgnoreParenImpCasts());
       if (AtomicBinOp && AtomicBinOp->getOpcode() == BO_Assign) {
-        X = AtomicBinOp->getLHS()->IgnoreParenImpCasts();
-        E = AtomicBinOp->getRHS()->IgnoreParenImpCasts();
+        X = AtomicBinOp->getLHS();
+        E = AtomicBinOp->getRHS();
         if ((X->isInstantiationDependent() || X->getType()->isScalarType()) &&
             (E->isInstantiationDependent() || E->getType()->isScalarType())) {
           if (!X->isLValue()) {
@@ -3392,28 +3824,240 @@ StmtResult Sema::ActOnOpenMPAtomicDirective(ArrayRef<OMPClause *> Clauses,
     } else if (CurContext->isDependentContext())
       E = X = nullptr;
   } else if (AtomicKind == OMPC_update || AtomicKind == OMPC_unknown) {
-    if (!isa<Expr>(Body)) {
-      Diag(Body->getLocStart(),
-           diag::err_omp_atomic_update_not_expression_statement)
-          << (AtomicKind == OMPC_update);
+    // If clause is update:
+    //  x++;
+    //  x--;
+    //  ++x;
+    //  --x;
+    //  x binop= expr;
+    //  x = x binop expr;
+    //  x = expr binop x;
+    OpenMPAtomicUpdateChecker Checker(*this);
+    if (Checker.checkStatement(
+            Body, (AtomicKind == OMPC_update)
+                      ? diag::err_omp_atomic_update_not_expression_statement
+                      : diag::err_omp_atomic_not_expression_statement,
+            diag::note_omp_atomic_update))
       return StmtError();
+    if (!CurContext->isDependentContext()) {
+      E = Checker.getExpr();
+      X = Checker.getX();
+      UE = Checker.getUpdateExpr();
+      IsXLHSInRHSPart = Checker.isXLHSInRHSPart();
     }
   } else if (AtomicKind == OMPC_capture) {
-    if (isa<Expr>(Body) && !isa<BinaryOperator>(Body)) {
-      Diag(Body->getLocStart(),
-           diag::err_omp_atomic_capture_not_expression_statement);
-      return StmtError();
-    } else if (!isa<Expr>(Body) && !isa<CompoundStmt>(Body)) {
-      Diag(Body->getLocStart(),
-           diag::err_omp_atomic_capture_not_compound_statement);
-      return StmtError();
+    enum {
+      NotAnAssignmentOp,
+      NotACompoundStatement,
+      NotTwoSubstatements,
+      NotASpecificExpression,
+      NoError
+    } ErrorFound = NoError;
+    SourceLocation ErrorLoc, NoteLoc;
+    SourceRange ErrorRange, NoteRange;
+    if (auto *AtomicBody = dyn_cast<Expr>(Body)) {
+      // If clause is a capture:
+      //  v = x++;
+      //  v = x--;
+      //  v = ++x;
+      //  v = --x;
+      //  v = x binop= expr;
+      //  v = x = x binop expr;
+      //  v = x = expr binop x;
+      auto *AtomicBinOp =
+          dyn_cast<BinaryOperator>(AtomicBody->IgnoreParenImpCasts());
+      if (AtomicBinOp && AtomicBinOp->getOpcode() == BO_Assign) {
+        V = AtomicBinOp->getLHS();
+        Body = AtomicBinOp->getRHS()->IgnoreParenImpCasts();
+        OpenMPAtomicUpdateChecker Checker(*this);
+        if (Checker.checkStatement(
+                Body, diag::err_omp_atomic_capture_not_expression_statement,
+                diag::note_omp_atomic_update))
+          return StmtError();
+        E = Checker.getExpr();
+        X = Checker.getX();
+        UE = Checker.getUpdateExpr();
+        IsXLHSInRHSPart = Checker.isXLHSInRHSPart();
+        IsPostfixUpdate = Checker.isPostfixUpdate();
+      } else {
+        ErrorLoc = AtomicBody->getExprLoc();
+        ErrorRange = AtomicBody->getSourceRange();
+        NoteLoc = AtomicBinOp ? AtomicBinOp->getOperatorLoc()
+                              : AtomicBody->getExprLoc();
+        NoteRange = AtomicBinOp ? AtomicBinOp->getSourceRange()
+                                : AtomicBody->getSourceRange();
+        ErrorFound = NotAnAssignmentOp;
+      }
+      if (ErrorFound != NoError) {
+        Diag(ErrorLoc, diag::err_omp_atomic_capture_not_expression_statement)
+            << ErrorRange;
+        Diag(NoteLoc, diag::note_omp_atomic_capture) << ErrorFound << NoteRange;
+        return StmtError();
+      } else if (CurContext->isDependentContext()) {
+        UE = V = E = X = nullptr;
+      }
+    } else {
+      // If clause is a capture:
+      //  { v = x; x = expr; }
+      //  { v = x; x++; }
+      //  { v = x; x--; }
+      //  { v = x; ++x; }
+      //  { v = x; --x; }
+      //  { v = x; x binop= expr; }
+      //  { v = x; x = x binop expr; }
+      //  { v = x; x = expr binop x; }
+      //  { x++; v = x; }
+      //  { x--; v = x; }
+      //  { ++x; v = x; }
+      //  { --x; v = x; }
+      //  { x binop= expr; v = x; }
+      //  { x = x binop expr; v = x; }
+      //  { x = expr binop x; v = x; }
+      if (auto *CS = dyn_cast<CompoundStmt>(Body)) {
+        // Check that this is { expr1; expr2; }
+        if (CS->size() == 2) {
+          auto *First = CS->body_front();
+          auto *Second = CS->body_back();
+          if (auto *EWC = dyn_cast<ExprWithCleanups>(First))
+            First = EWC->getSubExpr()->IgnoreParenImpCasts();
+          if (auto *EWC = dyn_cast<ExprWithCleanups>(Second))
+            Second = EWC->getSubExpr()->IgnoreParenImpCasts();
+          // Need to find what subexpression is 'v' and what is 'x'.
+          OpenMPAtomicUpdateChecker Checker(*this);
+          bool IsUpdateExprFound = !Checker.checkStatement(Second);
+          BinaryOperator *BinOp = nullptr;
+          if (IsUpdateExprFound) {
+            BinOp = dyn_cast<BinaryOperator>(First);
+            IsUpdateExprFound = BinOp && BinOp->getOpcode() == BO_Assign;
+          }
+          if (IsUpdateExprFound && !CurContext->isDependentContext()) {
+            //  { v = x; x++; }
+            //  { v = x; x--; }
+            //  { v = x; ++x; }
+            //  { v = x; --x; }
+            //  { v = x; x binop= expr; }
+            //  { v = x; x = x binop expr; }
+            //  { v = x; x = expr binop x; }
+            // Check that the first expression has form v = x.
+            auto *PossibleX = BinOp->getRHS()->IgnoreParenImpCasts();
+            llvm::FoldingSetNodeID XId, PossibleXId;
+            Checker.getX()->Profile(XId, Context, /*Canonical=*/true);
+            PossibleX->Profile(PossibleXId, Context, /*Canonical=*/true);
+            IsUpdateExprFound = XId == PossibleXId;
+            if (IsUpdateExprFound) {
+              V = BinOp->getLHS();
+              X = Checker.getX();
+              E = Checker.getExpr();
+              UE = Checker.getUpdateExpr();
+              IsXLHSInRHSPart = Checker.isXLHSInRHSPart();
+              IsPostfixUpdate = true;
+            }
+          }
+          if (!IsUpdateExprFound) {
+            IsUpdateExprFound = !Checker.checkStatement(First);
+            BinOp = nullptr;
+            if (IsUpdateExprFound) {
+              BinOp = dyn_cast<BinaryOperator>(Second);
+              IsUpdateExprFound = BinOp && BinOp->getOpcode() == BO_Assign;
+            }
+            if (IsUpdateExprFound && !CurContext->isDependentContext()) {
+              //  { x++; v = x; }
+              //  { x--; v = x; }
+              //  { ++x; v = x; }
+              //  { --x; v = x; }
+              //  { x binop= expr; v = x; }
+              //  { x = x binop expr; v = x; }
+              //  { x = expr binop x; v = x; }
+              // Check that the second expression has form v = x.
+              auto *PossibleX = BinOp->getRHS()->IgnoreParenImpCasts();
+              llvm::FoldingSetNodeID XId, PossibleXId;
+              Checker.getX()->Profile(XId, Context, /*Canonical=*/true);
+              PossibleX->Profile(PossibleXId, Context, /*Canonical=*/true);
+              IsUpdateExprFound = XId == PossibleXId;
+              if (IsUpdateExprFound) {
+                V = BinOp->getLHS();
+                X = Checker.getX();
+                E = Checker.getExpr();
+                UE = Checker.getUpdateExpr();
+                IsXLHSInRHSPart = Checker.isXLHSInRHSPart();
+                IsPostfixUpdate = false;
+              }
+            }
+          }
+          if (!IsUpdateExprFound) {
+            //  { v = x; x = expr; }
+            auto *FirstBinOp = dyn_cast<BinaryOperator>(First);
+            if (!FirstBinOp || FirstBinOp->getOpcode() != BO_Assign) {
+              ErrorFound = NotAnAssignmentOp;
+              NoteLoc = ErrorLoc = FirstBinOp ? FirstBinOp->getOperatorLoc()
+                                              : First->getLocStart();
+              NoteRange = ErrorRange = FirstBinOp
+                                           ? FirstBinOp->getSourceRange()
+                                           : SourceRange(ErrorLoc, ErrorLoc);
+            } else {
+              auto *SecondBinOp = dyn_cast<BinaryOperator>(Second);
+              if (!SecondBinOp || SecondBinOp->getOpcode() != BO_Assign) {
+                ErrorFound = NotAnAssignmentOp;
+                NoteLoc = ErrorLoc = SecondBinOp ? SecondBinOp->getOperatorLoc()
+                                                 : Second->getLocStart();
+                NoteRange = ErrorRange = SecondBinOp
+                                             ? SecondBinOp->getSourceRange()
+                                             : SourceRange(ErrorLoc, ErrorLoc);
+              } else {
+                auto *PossibleXRHSInFirst =
+                    FirstBinOp->getRHS()->IgnoreParenImpCasts();
+                auto *PossibleXLHSInSecond =
+                    SecondBinOp->getLHS()->IgnoreParenImpCasts();
+                llvm::FoldingSetNodeID X1Id, X2Id;
+                PossibleXRHSInFirst->Profile(X1Id, Context, /*Canonical=*/true);
+                PossibleXLHSInSecond->Profile(X2Id, Context,
+                                              /*Canonical=*/true);
+                IsUpdateExprFound = X1Id == X2Id;
+                if (IsUpdateExprFound) {
+                  V = FirstBinOp->getLHS();
+                  X = SecondBinOp->getLHS();
+                  E = SecondBinOp->getRHS();
+                  UE = nullptr;
+                  IsXLHSInRHSPart = false;
+                  IsPostfixUpdate = true;
+                } else {
+                  ErrorFound = NotASpecificExpression;
+                  ErrorLoc = FirstBinOp->getExprLoc();
+                  ErrorRange = FirstBinOp->getSourceRange();
+                  NoteLoc = SecondBinOp->getLHS()->getExprLoc();
+                  NoteRange = SecondBinOp->getRHS()->getSourceRange();
+                }
+              }
+            }
+          }
+        } else {
+          NoteLoc = ErrorLoc = Body->getLocStart();
+          NoteRange = ErrorRange =
+              SourceRange(Body->getLocStart(), Body->getLocStart());
+          ErrorFound = NotTwoSubstatements;
+        }
+      } else {
+        NoteLoc = ErrorLoc = Body->getLocStart();
+        NoteRange = ErrorRange =
+            SourceRange(Body->getLocStart(), Body->getLocStart());
+        ErrorFound = NotACompoundStatement;
+      }
+      if (ErrorFound != NoError) {
+        Diag(ErrorLoc, diag::err_omp_atomic_capture_not_compound_statement)
+            << ErrorRange;
+        Diag(NoteLoc, diag::note_omp_atomic_capture) << ErrorFound << NoteRange;
+        return StmtError();
+      } else if (CurContext->isDependentContext()) {
+        UE = V = E = X = nullptr;
+      }
     }
   }
 
   getCurFunction()->setHasBranchProtectedScope();
 
   return OMPAtomicDirective::Create(Context, StartLoc, EndLoc, Clauses, AStmt,
-                                    X, V, E);
+                                    X, V, E, UE, IsXLHSInRHSPart,
+                                    IsPostfixUpdate);
 }
 
 StmtResult Sema::ActOnOpenMPTargetDirective(ArrayRef<OMPClause *> Clauses,
@@ -3655,6 +4299,9 @@ ExprResult Sema::VerifyPositiveIntegerConstantInClause(Expr *E,
         << E->getSourceRange();
     return ExprError();
   }
+  if (CKind == OMPC_collapse) {
+    DSAStack->setCollapseNumber(Result.getExtValue());
+  }
   return ICE;
 }
 
@@ -3882,6 +4529,7 @@ OMPClause *Sema::ActOnOpenMPScheduleClause(
     return nullptr;
   }
   Expr *ValExpr = ChunkSize;
+  Expr *HelperValExpr = nullptr;
   if (ChunkSize) {
     if (!ChunkSize->isValueDependent() && !ChunkSize->isTypeDependent() &&
         !ChunkSize->isInstantiationDependent() &&
@@ -3898,17 +4546,25 @@ OMPClause *Sema::ActOnOpenMPScheduleClause(
       //  chunk_size must be a loop invariant integer expression with a positive
       //  value.
       llvm::APSInt Result;
-      if (ValExpr->isIntegerConstantExpr(Result, Context) &&
-          Result.isSigned() && !Result.isStrictlyPositive()) {
-        Diag(ChunkSizeLoc, diag::err_omp_negative_expression_in_clause)
-            << "schedule" << ChunkSize->getSourceRange();
-        return nullptr;
+      if (ValExpr->isIntegerConstantExpr(Result, Context)) {
+        if (Result.isSigned() && !Result.isStrictlyPositive()) {
+          Diag(ChunkSizeLoc, diag::err_omp_negative_expression_in_clause)
+              << "schedule" << ChunkSize->getSourceRange();
+          return nullptr;
+        }
+      } else if (isParallelOrTaskRegion(DSAStack->getCurrentDirective())) {
+        auto *ImpVar = buildVarDecl(*this, ChunkSize->getExprLoc(),
+                                    ChunkSize->getType(), ".chunk.");
+        auto *ImpVarRef = buildDeclRefExpr(*this, ImpVar, ChunkSize->getType(),
+                                           ChunkSize->getExprLoc(),
+                                           /*RefersToCapture=*/true);
+        HelperValExpr = ImpVarRef;
       }
     }
   }
 
   return new (Context) OMPScheduleClause(StartLoc, LParenLoc, KindLoc, CommaLoc,
-                                         EndLoc, Kind, ValExpr);
+                                         EndLoc, Kind, ValExpr, HelperValExpr);
 }
 
 OMPClause *Sema::ActOnOpenMPClause(OpenMPClauseKind Kind,
@@ -4133,14 +4789,6 @@ OMPClause *Sema::ActOnOpenMPPrivateClause(ArrayRef<Expr *> VarList,
       continue;
     }
 
-    // OpenMP [2.9.3.3, Restrictions, C/C++, p.1]
-    //  A variable of class type (or array thereof) that appears in a private
-    //  clause requires an accessible, unambiguous default constructor for the
-    //  class type.
-    while (Type->isArrayType()) {
-      Type = cast<ArrayType>(Type.getTypePtr())->getElementType();
-    }
-
     // OpenMP [2.9.1.1, Data-sharing Attribute Rules for Variables Referenced
     // in a Construct]
     //  Variables with the predetermined data-sharing attributes may not be
@@ -4156,25 +4804,36 @@ OMPClause *Sema::ActOnOpenMPPrivateClause(ArrayRef<Expr *> VarList,
       continue;
     }
 
+    // Variably modified types are not supported for tasks.
+    if (!Type->isAnyPointerType() && Type->isVariablyModifiedType() &&
+        DSAStack->getCurrentDirective() == OMPD_task) {
+      Diag(ELoc, diag::err_omp_variably_modified_type_not_supported)
+          << getOpenMPClauseName(OMPC_private) << Type
+          << getOpenMPDirectiveName(DSAStack->getCurrentDirective());
+      bool IsDecl =
+          VD->isThisDeclarationADefinition(Context) == VarDecl::DeclarationOnly;
+      Diag(VD->getLocation(),
+           IsDecl ? diag::note_previous_decl : diag::note_defined_here)
+          << VD;
+      continue;
+    }
+
+    // OpenMP [2.9.3.3, Restrictions, C/C++, p.1]
+    //  A variable of class type (or array thereof) that appears in a private
+    //  clause requires an accessible, unambiguous default constructor for the
+    //  class type.
     // Generate helper private variable and initialize it with the default
     // value. The address of the original variable is replaced by the address of
     // the new private variable in CodeGen. This new variable is not added to
     // IdResolver, so the code in the OpenMP region uses original variable for
     // proper diagnostics.
-    auto VDPrivate =
-        VarDecl::Create(Context, CurContext, DE->getLocStart(),
-                        DE->getExprLoc(), VD->getIdentifier(), VD->getType(),
-                        VD->getTypeSourceInfo(), /*S*/ SC_Auto);
-    ActOnUninitializedDecl(VDPrivate, /*TypeMayContainAuto*/ false);
+    Type = Type.getUnqualifiedType();
+    auto VDPrivate = buildVarDecl(*this, DE->getExprLoc(), Type, VD->getName());
+    ActOnUninitializedDecl(VDPrivate, /*TypeMayContainAuto=*/false);
     if (VDPrivate->isInvalidDecl())
       continue;
-    CurContext->addDecl(VDPrivate);
-    auto VDPrivateRefExpr =
-        DeclRefExpr::Create(Context, /*QualifierLoc*/ NestedNameSpecifierLoc(),
-                            /*TemplateKWLoc*/ SourceLocation(), VDPrivate,
-                            /*RefersToEnclosingVariableOrCapture*/ false,
-                            /*NameLoc*/ SourceLocation(), DE->getType(),
-                            /*VK*/ VK_LValue);
+    auto VDPrivateRefExpr = buildDeclRefExpr(
+        *this, VDPrivate, DE->getType().getUnqualifiedType(), DE->getExprLoc());
 
     DSAStack->addDSA(VD, DE, OMPC_private);
     Vars.push_back(DE);
@@ -4285,14 +4944,12 @@ OMPClause *Sema::ActOnOpenMPFirstprivateClause(ArrayRef<Expr *> VarList,
     //  A variable of class type (or array thereof) that appears in a private
     //  clause requires an accessible, unambiguous copy constructor for the
     //  class type.
-    Type = Context.getBaseElementType(Type);
+    auto ElemType = Context.getBaseElementType(Type).getNonReferenceType();
 
     // If an implicit firstprivate variable found it was checked already.
     if (!IsImplicitClause) {
       DSAStackTy::DSAVarData DVar = DSAStack->getTopDSA(VD, false);
-      Type = Type.getNonReferenceType().getCanonicalType();
-      bool IsConstant = Type.isConstant(Context);
-      Type = Context.getBaseElementType(Type);
+      bool IsConstant = ElemType.isConstant(Context);
       // OpenMP [2.4.13, Data-sharing Attribute Clauses]
       //  A list item that specifies a given variable may not appear in more
       // than one clause on the same directive, except that a variable may be
@@ -4375,10 +5032,22 @@ OMPClause *Sema::ActOnOpenMPFirstprivateClause(ArrayRef<Expr *> VarList,
       }
     }
 
+    // Variably modified types are not supported for tasks.
+    if (!Type->isAnyPointerType() && Type->isVariablyModifiedType() &&
+        DSAStack->getCurrentDirective() == OMPD_task) {
+      Diag(ELoc, diag::err_omp_variably_modified_type_not_supported)
+          << getOpenMPClauseName(OMPC_firstprivate) << Type
+          << getOpenMPDirectiveName(DSAStack->getCurrentDirective());
+      bool IsDecl =
+          VD->isThisDeclarationADefinition(Context) == VarDecl::DeclarationOnly;
+      Diag(VD->getLocation(),
+           IsDecl ? diag::note_previous_decl : diag::note_defined_here)
+          << VD;
+      continue;
+    }
+
     Type = Type.getUnqualifiedType();
-    auto VDPrivate = VarDecl::Create(Context, CurContext, DE->getLocStart(),
-                                     ELoc, VD->getIdentifier(), VD->getType(),
-                                     VD->getTypeSourceInfo(), /*S*/ SC_Auto);
+    auto VDPrivate = buildVarDecl(*this, ELoc, Type, VD->getName());
     // Generate helper private variable and initialize it with the value of the
     // original variable. The address of the original variable is replaced by
     // the address of the new private variable in the CodeGen. This new variable
@@ -4387,19 +5056,16 @@ OMPClause *Sema::ActOnOpenMPFirstprivateClause(ArrayRef<Expr *> VarList,
     Expr *VDInitRefExpr = nullptr;
     // For arrays generate initializer for single element and replace it by the
     // original array element in CodeGen.
-    if (DE->getType()->isArrayType()) {
-      auto VDInit = VarDecl::Create(Context, CurContext, DE->getLocStart(),
-                                    ELoc, VD->getIdentifier(), Type,
-                                    VD->getTypeSourceInfo(), /*S*/ SC_Auto);
-      CurContext->addHiddenDecl(VDInit);
-      VDInitRefExpr = DeclRefExpr::Create(
-          Context, /*QualifierLoc*/ NestedNameSpecifierLoc(),
-          /*TemplateKWLoc*/ SourceLocation(), VDInit,
-          /*RefersToEnclosingVariableOrCapture*/ true, ELoc, Type,
-          /*VK*/ VK_LValue);
-      VDInit->setIsUsed();
+    if (Type->isArrayType()) {
+      auto VDInit =
+          buildVarDecl(*this, DE->getExprLoc(), ElemType, VD->getName());
+      VDInitRefExpr = buildDeclRefExpr(*this, VDInit, ElemType, ELoc);
       auto Init = DefaultLvalueConversion(VDInitRefExpr).get();
-      InitializedEntity Entity = InitializedEntity::InitializeVariable(VDInit);
+      ElemType = ElemType.getUnqualifiedType();
+      auto *VDInitTemp = buildVarDecl(*this, DE->getLocStart(), ElemType,
+                                      ".firstprivate.temp");
+      InitializedEntity Entity =
+          InitializedEntity::InitializeVariable(VDInitTemp);
       InitializationKind Kind = InitializationKind::CreateCopy(ELoc, ELoc);
 
       InitializationSequence InitSeq(*this, Entity, Kind, Init);
@@ -4409,15 +5075,13 @@ OMPClause *Sema::ActOnOpenMPFirstprivateClause(ArrayRef<Expr *> VarList,
       else
         VDPrivate->setInit(Result.getAs<Expr>());
     } else {
-      AddInitializerToDecl(
-          VDPrivate,
-          DefaultLvalueConversion(
-              DeclRefExpr::Create(Context, NestedNameSpecifierLoc(),
-                                  SourceLocation(), DE->getDecl(),
-                                  /*RefersToEnclosingVariableOrCapture=*/true,
-                                  DE->getExprLoc(), DE->getType(),
-                                  /*VK=*/VK_LValue)).get(),
-          /*DirectInit=*/false, /*TypeMayContainAuto=*/false);
+      auto *VDInit =
+          buildVarDecl(*this, DE->getLocStart(), Type, ".firstprivate.temp");
+      VDInitRefExpr =
+          buildDeclRefExpr(*this, VDInit, DE->getType(), DE->getExprLoc());
+      AddInitializerToDecl(VDPrivate,
+                           DefaultLvalueConversion(VDInitRefExpr).get(),
+                           /*DirectInit=*/false, /*TypeMayContainAuto=*/false);
     }
     if (VDPrivate->isInvalidDecl()) {
       if (IsImplicitClause) {
@@ -4427,12 +5091,8 @@ OMPClause *Sema::ActOnOpenMPFirstprivateClause(ArrayRef<Expr *> VarList,
       continue;
     }
     CurContext->addDecl(VDPrivate);
-    auto VDPrivateRefExpr =
-        DeclRefExpr::Create(Context, /*QualifierLoc*/ NestedNameSpecifierLoc(),
-                            /*TemplateKWLoc*/ SourceLocation(), VDPrivate,
-                            /*RefersToEnclosingVariableOrCapture*/ false,
-                            DE->getLocStart(), DE->getType(),
-                            /*VK*/ VK_LValue);
+    auto VDPrivateRefExpr = buildDeclRefExpr(
+        *this, VDPrivate, DE->getType().getUnqualifiedType(), DE->getExprLoc());
     DSAStack->addDSA(VD, DE, OMPC_firstprivate);
     Vars.push_back(DE);
     PrivateCopies.push_back(VDPrivateRefExpr);
@@ -4451,11 +5111,17 @@ OMPClause *Sema::ActOnOpenMPLastprivateClause(ArrayRef<Expr *> VarList,
                                               SourceLocation LParenLoc,
                                               SourceLocation EndLoc) {
   SmallVector<Expr *, 8> Vars;
+  SmallVector<Expr *, 8> SrcExprs;
+  SmallVector<Expr *, 8> DstExprs;
+  SmallVector<Expr *, 8> AssignmentOps;
   for (auto &RefExpr : VarList) {
     assert(RefExpr && "NULL expr in OpenMP lastprivate clause.");
     if (isa<DependentScopeDeclRefExpr>(RefExpr)) {
       // It will be analyzed later.
       Vars.push_back(RefExpr);
+      SrcExprs.push_back(nullptr);
+      DstExprs.push_back(nullptr);
+      AssignmentOps.push_back(nullptr);
       continue;
     }
 
@@ -4477,6 +5143,9 @@ OMPClause *Sema::ActOnOpenMPLastprivateClause(ArrayRef<Expr *> VarList,
     if (Type->isDependentType() || Type->isInstantiationDependentType()) {
       // It will be analyzed later.
       Vars.push_back(DE);
+      SrcExprs.push_back(nullptr);
+      DstExprs.push_back(nullptr);
+      AssignmentOps.push_back(nullptr);
       continue;
     }
 
@@ -4521,6 +5190,7 @@ OMPClause *Sema::ActOnOpenMPLastprivateClause(ArrayRef<Expr *> VarList,
     // lastprivate clause on a worksharing construct if any of the corresponding
     // worksharing regions ever binds to any of the corresponding parallel
     // regions.
+    DSAStackTy::DSAVarData TopDVar = DVar;
     if (isOpenMPWorksharingDirective(CurrDir) &&
         !isOpenMPParallelDirective(CurrDir)) {
       DVar = DSAStack->getImplicitDSA(VD, true);
@@ -4540,65 +5210,39 @@ OMPClause *Sema::ActOnOpenMPLastprivateClause(ArrayRef<Expr *> VarList,
     //  A variable of class type (or array thereof) that appears in a
     //  lastprivate clause requires an accessible, unambiguous copy assignment
     //  operator for the class type.
-    while (Type.getNonReferenceType()->isArrayType())
-      Type = cast<ArrayType>(Type.getNonReferenceType().getTypePtr())
-                 ->getElementType();
-    CXXRecordDecl *RD = getLangOpts().CPlusPlus
-                            ? Type.getNonReferenceType()->getAsCXXRecordDecl()
-                            : nullptr;
-    // FIXME This code must be replaced by actual copying and destructing of the
-    // lastprivate variable.
-    if (RD) {
-      CXXMethodDecl *MD = LookupCopyingAssignment(RD, 0, false, 0);
-      DeclAccessPair FoundDecl = DeclAccessPair::make(MD, MD->getAccess());
-      if (MD) {
-        if (CheckMemberAccess(ELoc, RD, FoundDecl) == AR_inaccessible ||
-            MD->isDeleted()) {
-          Diag(ELoc, diag::err_omp_required_method)
-              << getOpenMPClauseName(OMPC_lastprivate) << 2;
-          bool IsDecl = VD->isThisDeclarationADefinition(Context) ==
-                        VarDecl::DeclarationOnly;
-          Diag(VD->getLocation(),
-               IsDecl ? diag::note_previous_decl : diag::note_defined_here)
-              << VD;
-          Diag(RD->getLocation(), diag::note_previous_decl) << RD;
-          continue;
-        }
-        MarkFunctionReferenced(ELoc, MD);
-        DiagnoseUseOfDecl(MD, ELoc);
-      }
-
-      CXXDestructorDecl *DD = RD->getDestructor();
-      if (DD) {
-        PartialDiagnostic PD =
-            PartialDiagnostic(PartialDiagnostic::NullDiagnostic());
-        if (CheckDestructorAccess(ELoc, DD, PD) == AR_inaccessible ||
-            DD->isDeleted()) {
-          Diag(ELoc, diag::err_omp_required_method)
-              << getOpenMPClauseName(OMPC_lastprivate) << 4;
-          bool IsDecl = VD->isThisDeclarationADefinition(Context) ==
-                        VarDecl::DeclarationOnly;
-          Diag(VD->getLocation(),
-               IsDecl ? diag::note_previous_decl : diag::note_defined_here)
-              << VD;
-          Diag(RD->getLocation(), diag::note_previous_decl) << RD;
-          continue;
-        }
-        MarkFunctionReferenced(ELoc, DD);
-        DiagnoseUseOfDecl(DD, ELoc);
-      }
-    }
+    Type = Context.getBaseElementType(Type).getNonReferenceType();
+    auto *SrcVD = buildVarDecl(*this, DE->getLocStart(),
+                               Type.getUnqualifiedType(), ".lastprivate.src");
+    auto *PseudoSrcExpr = buildDeclRefExpr(
+        *this, SrcVD, Type.getUnqualifiedType(), DE->getExprLoc());
+    auto *DstVD =
+        buildVarDecl(*this, DE->getLocStart(), Type, ".lastprivate.dst");
+    auto *PseudoDstExpr =
+        buildDeclRefExpr(*this, DstVD, Type, DE->getExprLoc());
+    // For arrays generate assignment operation for single element and replace
+    // it by the original array element in CodeGen.
+    auto AssignmentOp = BuildBinOp(/*S=*/nullptr, DE->getExprLoc(), BO_Assign,
+                                   PseudoDstExpr, PseudoSrcExpr);
+    if (AssignmentOp.isInvalid())
+      continue;
+    AssignmentOp = ActOnFinishFullExpr(AssignmentOp.get(), DE->getExprLoc(),
+                                       /*DiscardedValue=*/true);
+    if (AssignmentOp.isInvalid())
+      continue;
 
-    if (DVar.CKind != OMPC_firstprivate)
+    if (TopDVar.CKind != OMPC_firstprivate)
       DSAStack->addDSA(VD, DE, OMPC_lastprivate);
     Vars.push_back(DE);
+    SrcExprs.push_back(PseudoSrcExpr);
+    DstExprs.push_back(PseudoDstExpr);
+    AssignmentOps.push_back(AssignmentOp.get());
   }
 
   if (Vars.empty())
     return nullptr;
 
   return OMPLastprivateClause::Create(Context, StartLoc, LParenLoc, EndLoc,
-                                      Vars);
+                                      Vars, SrcExprs, DstExprs, AssignmentOps);
 }
 
 OMPClause *Sema::ActOnOpenMPSharedClause(ArrayRef<Expr *> VarList,
@@ -4718,19 +5362,19 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
   switch (OOK) {
   case OO_Plus:
   case OO_Minus:
-    BOK = BO_AddAssign;
+    BOK = BO_Add;
     break;
   case OO_Star:
-    BOK = BO_MulAssign;
+    BOK = BO_Mul;
     break;
   case OO_Amp:
-    BOK = BO_AndAssign;
+    BOK = BO_And;
     break;
   case OO_Pipe:
-    BOK = BO_OrAssign;
+    BOK = BO_Or;
     break;
   case OO_Caret:
-    BOK = BO_XorAssign;
+    BOK = BO_Xor;
     break;
   case OO_AmpAmp:
     BOK = BO_LAnd;
@@ -4738,7 +5382,44 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
   case OO_PipePipe:
     BOK = BO_LOr;
     break;
-  default:
+  case OO_New:
+  case OO_Delete:
+  case OO_Array_New:
+  case OO_Array_Delete:
+  case OO_Slash:
+  case OO_Percent:
+  case OO_Tilde:
+  case OO_Exclaim:
+  case OO_Equal:
+  case OO_Less:
+  case OO_Greater:
+  case OO_LessEqual:
+  case OO_GreaterEqual:
+  case OO_PlusEqual:
+  case OO_MinusEqual:
+  case OO_StarEqual:
+  case OO_SlashEqual:
+  case OO_PercentEqual:
+  case OO_CaretEqual:
+  case OO_AmpEqual:
+  case OO_PipeEqual:
+  case OO_LessLess:
+  case OO_GreaterGreater:
+  case OO_LessLessEqual:
+  case OO_GreaterGreaterEqual:
+  case OO_EqualEqual:
+  case OO_ExclaimEqual:
+  case OO_PlusPlus:
+  case OO_MinusMinus:
+  case OO_Comma:
+  case OO_ArrowStar:
+  case OO_Arrow:
+  case OO_Call:
+  case OO_Subscript:
+  case OO_Conditional:
+  case NUM_OVERLOADED_OPERATORS:
+    llvm_unreachable("Unexpected reduction identifier");
+  case OO_None:
     if (auto II = DN.getAsIdentifierInfo()) {
       if (II->isStr("max"))
         BOK = BO_GT;
@@ -4760,11 +5441,17 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
   }
 
   SmallVector<Expr *, 8> Vars;
+  SmallVector<Expr *, 8> LHSs;
+  SmallVector<Expr *, 8> RHSs;
+  SmallVector<Expr *, 8> ReductionOps;
   for (auto RefExpr : VarList) {
     assert(RefExpr && "nullptr expr in OpenMP reduction clause.");
     if (isa<DependentScopeDeclRefExpr>(RefExpr)) {
       // It will be analyzed later.
       Vars.push_back(RefExpr);
+      LHSs.push_back(nullptr);
+      RHSs.push_back(nullptr);
+      ReductionOps.push_back(nullptr);
       continue;
     }
 
@@ -4773,6 +5460,9 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
         RefExpr->containsUnexpandedParameterPack()) {
       // It will be analyzed later.
       Vars.push_back(RefExpr);
+      LHSs.push_back(nullptr);
+      RHSs.push_back(nullptr);
+      ReductionOps.push_back(nullptr);
       continue;
     }
 
@@ -4865,23 +5555,6 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
           << VD;
       continue;
     }
-    bool Suppress = getDiagnostics().getSuppressAllDiagnostics();
-    getDiagnostics().setSuppressAllDiagnostics(true);
-    ExprResult ReductionOp =
-        BuildBinOp(DSAStack->getCurScope(), ReductionId.getLocStart(), BOK,
-                   RefExpr, RefExpr);
-    getDiagnostics().setSuppressAllDiagnostics(Suppress);
-    if (ReductionOp.isInvalid()) {
-      Diag(ELoc, diag::err_omp_reduction_id_not_compatible) << Type
-                                                            << ReductionIdRange;
-      bool IsDecl =
-          VD->isThisDeclarationADefinition(Context) == VarDecl::DeclarationOnly;
-      Diag(VD->getLocation(),
-           IsDecl ? diag::note_previous_decl : diag::note_defined_here)
-          << VD;
-      continue;
-    }
-
     // OpenMP [2.14.1.1, Data-sharing Attribute Rules for Variables Referenced
     // in a Construct]
     //  Variables with the predetermined data-sharing attributes may not be
@@ -4924,55 +5597,160 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
         continue;
       }
     }
-
-    CXXRecordDecl *RD = getLangOpts().CPlusPlus
-                            ? Type.getNonReferenceType()->getAsCXXRecordDecl()
-                            : nullptr;
-    // FIXME This code must be replaced by actual constructing/destructing of
-    // the reduction variable.
-    if (RD) {
-      CXXConstructorDecl *CD = LookupDefaultConstructor(RD);
-      PartialDiagnostic PD =
-          PartialDiagnostic(PartialDiagnostic::NullDiagnostic());
-      if (!CD ||
-          CheckConstructorAccess(ELoc, CD,
-                                 InitializedEntity::InitializeTemporary(Type),
-                                 CD->getAccess(), PD) == AR_inaccessible ||
-          CD->isDeleted()) {
-        Diag(ELoc, diag::err_omp_required_method)
-            << getOpenMPClauseName(OMPC_reduction) << 0;
-        bool IsDecl = VD->isThisDeclarationADefinition(Context) ==
-                      VarDecl::DeclarationOnly;
-        Diag(VD->getLocation(),
-             IsDecl ? diag::note_previous_decl : diag::note_defined_here)
-            << VD;
-        Diag(RD->getLocation(), diag::note_previous_decl) << RD;
-        continue;
+    Type = Type.getNonLValueExprType(Context).getUnqualifiedType();
+    auto *LHSVD = buildVarDecl(*this, ELoc, Type, ".reduction.lhs");
+    auto *RHSVD = buildVarDecl(*this, ELoc, Type, VD->getName());
+    // Add initializer for private variable.
+    Expr *Init = nullptr;
+    switch (BOK) {
+    case BO_Add:
+    case BO_Xor:
+    case BO_Or:
+    case BO_LOr:
+      // '+', '-', '^', '|', '||' reduction ops - initializer is '0'.
+      if (Type->isScalarType() || Type->isAnyComplexType()) {
+        Init = ActOnIntegerConstant(ELoc, /*Val=*/0).get();
       }
-      MarkFunctionReferenced(ELoc, CD);
-      DiagnoseUseOfDecl(CD, ELoc);
-
-      CXXDestructorDecl *DD = RD->getDestructor();
-      if (DD) {
-        if (CheckDestructorAccess(ELoc, DD, PD) == AR_inaccessible ||
-            DD->isDeleted()) {
-          Diag(ELoc, diag::err_omp_required_method)
-              << getOpenMPClauseName(OMPC_reduction) << 4;
-          bool IsDecl = VD->isThisDeclarationADefinition(Context) ==
-                        VarDecl::DeclarationOnly;
-          Diag(VD->getLocation(),
-               IsDecl ? diag::note_previous_decl : diag::note_defined_here)
-              << VD;
-          Diag(RD->getLocation(), diag::note_previous_decl) << RD;
-          continue;
+      break;
+    case BO_Mul:
+    case BO_LAnd:
+      if (Type->isScalarType() || Type->isAnyComplexType()) {
+        // '*' and '&&' reduction ops - initializer is '1'.
+        Init = ActOnIntegerConstant(ELoc, /*Val=*/1).get();
+      }
+      break;
+    case BO_And: {
+      // '&' reduction op - initializer is '~0'.
+      QualType OrigType = Type;
+      if (auto *ComplexTy = OrigType->getAs<ComplexType>()) {
+        Type = ComplexTy->getElementType();
+      }
+      if (Type->isRealFloatingType()) {
+        llvm::APFloat InitValue =
+            llvm::APFloat::getAllOnesValue(Context.getTypeSize(Type),
+                                           /*isIEEE=*/true);
+        Init = FloatingLiteral::Create(Context, InitValue, /*isexact=*/true,
+                                       Type, ELoc);
+      } else if (Type->isScalarType()) {
+        auto Size = Context.getTypeSize(Type);
+        QualType IntTy = Context.getIntTypeForBitwidth(Size, /*Signed=*/0);
+        llvm::APInt InitValue = llvm::APInt::getAllOnesValue(Size);
+        Init = IntegerLiteral::Create(Context, InitValue, IntTy, ELoc);
+      }
+      if (Init && OrigType->isAnyComplexType()) {
+        // Init = 0xFFFF + 0xFFFFi;
+        auto *Im = new (Context) ImaginaryLiteral(Init, OrigType);
+        Init = CreateBuiltinBinOp(ELoc, BO_Add, Init, Im).get();
+      }
+      Type = OrigType;
+      break;
+    }
+    case BO_LT:
+    case BO_GT: {
+      // 'min' reduction op - initializer is 'Largest representable number in
+      // the reduction list item type'.
+      // 'max' reduction op - initializer is 'Least representable number in
+      // the reduction list item type'.
+      if (Type->isIntegerType() || Type->isPointerType()) {
+        bool IsSigned = Type->hasSignedIntegerRepresentation();
+        auto Size = Context.getTypeSize(Type);
+        QualType IntTy =
+            Context.getIntTypeForBitwidth(Size, /*Signed=*/IsSigned);
+        llvm::APInt InitValue =
+            (BOK != BO_LT)
+                ? IsSigned ? llvm::APInt::getSignedMinValue(Size)
+                           : llvm::APInt::getMinValue(Size)
+                : IsSigned ? llvm::APInt::getSignedMaxValue(Size)
+                           : llvm::APInt::getMaxValue(Size);
+        Init = IntegerLiteral::Create(Context, InitValue, IntTy, ELoc);
+        if (Type->isPointerType()) {
+          // Cast to pointer type.
+          auto CastExpr = BuildCStyleCastExpr(
+              SourceLocation(), Context.getTrivialTypeSourceInfo(Type, ELoc),
+              SourceLocation(), Init);
+          if (CastExpr.isInvalid())
+            continue;
+          Init = CastExpr.get();
         }
-        MarkFunctionReferenced(ELoc, DD);
-        DiagnoseUseOfDecl(DD, ELoc);
+      } else if (Type->isRealFloatingType()) {
+        llvm::APFloat InitValue = llvm::APFloat::getLargest(
+            Context.getFloatTypeSemantics(Type), BOK != BO_LT);
+        Init = FloatingLiteral::Create(Context, InitValue, /*isexact=*/true,
+                                       Type, ELoc);
+      }
+      break;
+    }
+    case BO_PtrMemD:
+    case BO_PtrMemI:
+    case BO_MulAssign:
+    case BO_Div:
+    case BO_Rem:
+    case BO_Sub:
+    case BO_Shl:
+    case BO_Shr:
+    case BO_LE:
+    case BO_GE:
+    case BO_EQ:
+    case BO_NE:
+    case BO_AndAssign:
+    case BO_XorAssign:
+    case BO_OrAssign:
+    case BO_Assign:
+    case BO_AddAssign:
+    case BO_SubAssign:
+    case BO_DivAssign:
+    case BO_RemAssign:
+    case BO_ShlAssign:
+    case BO_ShrAssign:
+    case BO_Comma:
+      llvm_unreachable("Unexpected reduction operation");
+    }
+    if (Init) {
+      AddInitializerToDecl(RHSVD, Init, /*DirectInit=*/false,
+                           /*TypeMayContainAuto=*/false);
+    } else {
+      ActOnUninitializedDecl(RHSVD, /*TypeMayContainAuto=*/false);
+    }
+    if (!RHSVD->hasInit()) {
+      Diag(ELoc, diag::err_omp_reduction_id_not_compatible) << Type
+                                                            << ReductionIdRange;
+      bool IsDecl =
+          VD->isThisDeclarationADefinition(Context) == VarDecl::DeclarationOnly;
+      Diag(VD->getLocation(),
+           IsDecl ? diag::note_previous_decl : diag::note_defined_here)
+          << VD;
+      continue;
+    }
+    auto *LHSDRE = buildDeclRefExpr(*this, LHSVD, Type, ELoc);
+    auto *RHSDRE = buildDeclRefExpr(*this, RHSVD, Type, ELoc);
+    ExprResult ReductionOp =
+        BuildBinOp(DSAStack->getCurScope(), ReductionId.getLocStart(), BOK,
+                   LHSDRE, RHSDRE);
+    if (ReductionOp.isUsable()) {
+      if (BOK != BO_LT && BOK != BO_GT) {
+        ReductionOp =
+            BuildBinOp(DSAStack->getCurScope(), ReductionId.getLocStart(),
+                       BO_Assign, LHSDRE, ReductionOp.get());
+      } else {
+        auto *ConditionalOp = new (Context) ConditionalOperator(
+            ReductionOp.get(), SourceLocation(), LHSDRE, SourceLocation(),
+            RHSDRE, Type, VK_LValue, OK_Ordinary);
+        ReductionOp =
+            BuildBinOp(DSAStack->getCurScope(), ReductionId.getLocStart(),
+                       BO_Assign, LHSDRE, ConditionalOp);
+      }
+      if (ReductionOp.isUsable()) {
+        ReductionOp = ActOnFinishFullExpr(ReductionOp.get());
       }
     }
+    if (ReductionOp.isInvalid())
+      continue;
 
     DSAStack->addDSA(VD, DE, OMPC_reduction);
     Vars.push_back(DE);
+    LHSs.push_back(LHSDRE);
+    RHSs.push_back(RHSDRE);
+    ReductionOps.push_back(ReductionOp.get());
   }
 
   if (Vars.empty())
@@ -4980,7 +5758,8 @@ OMPClause *Sema::ActOnOpenMPReductionClause(
 
   return OMPReductionClause::Create(
       Context, StartLoc, LParenLoc, ColonLoc, EndLoc, Vars,
-      ReductionIdScopeSpec.getWithLocInContext(Context), ReductionId);
+      ReductionIdScopeSpec.getWithLocInContext(Context), ReductionId, LHSs,
+      RHSs, ReductionOps);
 }
 
 OMPClause *Sema::ActOnOpenMPLinearClause(ArrayRef<Expr *> VarList, Expr *Step,
@@ -4989,11 +5768,13 @@ OMPClause *Sema::ActOnOpenMPLinearClause(ArrayRef<Expr *> VarList, Expr *Step,
                                          SourceLocation ColonLoc,
                                          SourceLocation EndLoc) {
   SmallVector<Expr *, 8> Vars;
+  SmallVector<Expr *, 8> Inits;
   for (auto &RefExpr : VarList) {
     assert(RefExpr && "NULL expr in OpenMP linear clause.");
     if (isa<DependentScopeDeclRefExpr>(RefExpr)) {
       // It will be analyzed later.
       Vars.push_back(RefExpr);
+      Inits.push_back(nullptr);
       continue;
     }
 
@@ -5035,6 +5816,7 @@ OMPClause *Sema::ActOnOpenMPLinearClause(ArrayRef<Expr *> VarList, Expr *Step,
     if (QType->isDependentType() || QType->isInstantiationDependentType()) {
       // It will be analyzed later.
       Vars.push_back(DE);
+      Inits.push_back(nullptr);
       continue;
     }
 
@@ -5080,14 +5862,22 @@ OMPClause *Sema::ActOnOpenMPLinearClause(ArrayRef<Expr *> VarList, Expr *Step,
       continue;
     }
 
+    // Build var to save initial value.
+    VarDecl *Init = buildVarDecl(*this, ELoc, QType, ".linear.start");
+    AddInitializerToDecl(Init, DefaultLvalueConversion(DE).get(),
+                         /*DirectInit*/ false, /*TypeMayContainAuto*/ false);
+    auto InitRef = buildDeclRefExpr(
+        *this, Init, DE->getType().getUnqualifiedType(), DE->getExprLoc());
     DSAStack->addDSA(VD, DE, OMPC_linear);
     Vars.push_back(DE);
+    Inits.push_back(InitRef);
   }
 
   if (Vars.empty())
     return nullptr;
 
   Expr *StepExpr = Step;
+  Expr *CalcStepExpr = nullptr;
   if (Step && !Step->isValueDependent() && !Step->isTypeDependent() &&
       !Step->isInstantiationDependent() &&
       !Step->containsUnexpandedParameterPack()) {
@@ -5097,17 +5887,82 @@ OMPClause *Sema::ActOnOpenMPLinearClause(ArrayRef<Expr *> VarList, Expr *Step,
       return nullptr;
     StepExpr = Val.get();
 
+    // Build var to save the step value.
+    VarDecl *SaveVar =
+        buildVarDecl(*this, StepLoc, StepExpr->getType(), ".linear.step");
+    ExprResult SaveRef =
+        buildDeclRefExpr(*this, SaveVar, StepExpr->getType(), StepLoc);
+    ExprResult CalcStep =
+        BuildBinOp(CurScope, StepLoc, BO_Assign, SaveRef.get(), StepExpr);
+
     // Warn about zero linear step (it would be probably better specified as
     // making corresponding variables 'const').
     llvm::APSInt Result;
-    if (StepExpr->isIntegerConstantExpr(Result, Context) &&
-        !Result.isNegative() && !Result.isStrictlyPositive())
+    bool IsConstant = StepExpr->isIntegerConstantExpr(Result, Context);
+    if (IsConstant && !Result.isNegative() && !Result.isStrictlyPositive())
       Diag(StepLoc, diag::warn_omp_linear_step_zero) << Vars[0]
                                                      << (Vars.size() > 1);
+    if (!IsConstant && CalcStep.isUsable()) {
+      // Calculate the step beforehand instead of doing this on each iteration.
+      // (This is not used if the number of iterations may be kfold-ed).
+      CalcStepExpr = CalcStep.get();
+    }
   }
 
   return OMPLinearClause::Create(Context, StartLoc, LParenLoc, ColonLoc, EndLoc,
-                                 Vars, StepExpr);
+                                 Vars, Inits, StepExpr, CalcStepExpr);
+}
+
+static bool FinishOpenMPLinearClause(OMPLinearClause &Clause, DeclRefExpr *IV,
+                                     Expr *NumIterations, Sema &SemaRef,
+                                     Scope *S) {
+  // Walk the vars and build update/final expressions for the CodeGen.
+  SmallVector<Expr *, 8> Updates;
+  SmallVector<Expr *, 8> Finals;
+  Expr *Step = Clause.getStep();
+  Expr *CalcStep = Clause.getCalcStep();
+  // OpenMP [2.14.3.7, linear clause]
+  // If linear-step is not specified it is assumed to be 1.
+  if (Step == nullptr)
+    Step = SemaRef.ActOnIntegerConstant(SourceLocation(), 1).get();
+  else if (CalcStep)
+    Step = cast<BinaryOperator>(CalcStep)->getLHS();
+  bool HasErrors = false;
+  auto CurInit = Clause.inits().begin();
+  for (auto &RefExpr : Clause.varlists()) {
+    Expr *InitExpr = *CurInit;
+
+    // Build privatized reference to the current linear var.
+    auto DE = cast<DeclRefExpr>(RefExpr);
+    auto PrivateRef =
+        buildDeclRefExpr(SemaRef, cast<VarDecl>(DE->getDecl()),
+                         DE->getType().getUnqualifiedType(), DE->getExprLoc(),
+                         /*RefersToCapture=*/true);
+
+    // Build update: Var = InitExpr + IV * Step
+    ExprResult Update =
+        BuildCounterUpdate(SemaRef, S, RefExpr->getExprLoc(), PrivateRef,
+                           InitExpr, IV, Step, /* Subtract */ false);
+    Update = SemaRef.ActOnFinishFullExpr(Update.get());
+
+    // Build final: Var = InitExpr + NumIterations * Step
+    ExprResult Final =
+        BuildCounterUpdate(SemaRef, S, RefExpr->getExprLoc(), PrivateRef,
+                           InitExpr, NumIterations, Step, /* Subtract */ false);
+    Final = SemaRef.ActOnFinishFullExpr(Final.get());
+    if (!Update.isUsable() || !Final.isUsable()) {
+      Updates.push_back(nullptr);
+      Finals.push_back(nullptr);
+      HasErrors = true;
+    } else {
+      Updates.push_back(Update.get());
+      Finals.push_back(Final.get());
+    }
+    ++CurInit;
+  }
+  Clause.setUpdates(Updates);
+  Clause.setFinals(Finals);
+  return HasErrors;
 }
 
 OMPClause *Sema::ActOnOpenMPAlignedClause(
@@ -5137,10 +5992,8 @@ OMPClause *Sema::ActOnOpenMPAlignedClause(
     // OpenMP  [2.8.1, simd construct, Restrictions]
     // The type of list items appearing in the aligned clause must be
     // array, pointer, reference to array, or reference to pointer.
-    QualType QType = DE->getType()
-                         .getNonReferenceType()
-                         .getUnqualifiedType()
-                         .getCanonicalType();
+    QualType QType = VD->getType();
+    QType = QType.getNonReferenceType().getUnqualifiedType().getCanonicalType();
     const Type *Ty = QType.getTypePtrOrNull();
     if (!Ty || (!Ty->isDependentType() && !Ty->isArrayType() &&
                 !Ty->isPointerType())) {
@@ -5190,11 +6043,17 @@ OMPClause *Sema::ActOnOpenMPCopyinClause(ArrayRef<Expr *> VarList,
                                          SourceLocation LParenLoc,
                                          SourceLocation EndLoc) {
   SmallVector<Expr *, 8> Vars;
+  SmallVector<Expr *, 8> SrcExprs;
+  SmallVector<Expr *, 8> DstExprs;
+  SmallVector<Expr *, 8> AssignmentOps;
   for (auto &RefExpr : VarList) {
     assert(RefExpr && "NULL expr in OpenMP copyin clause.");
     if (isa<DependentScopeDeclRefExpr>(RefExpr)) {
       // It will be analyzed later.
       Vars.push_back(RefExpr);
+      SrcExprs.push_back(nullptr);
+      DstExprs.push_back(nullptr);
+      AssignmentOps.push_back(nullptr);
       continue;
     }
 
@@ -5216,6 +6075,9 @@ OMPClause *Sema::ActOnOpenMPCopyinClause(ArrayRef<Expr *> VarList,
     if (Type->isDependentType() || Type->isInstantiationDependentType()) {
       // It will be analyzed later.
       Vars.push_back(DE);
+      SrcExprs.push_back(nullptr);
+      DstExprs.push_back(nullptr);
+      AssignmentOps.push_back(nullptr);
       continue;
     }
 
@@ -5232,40 +6094,38 @@ OMPClause *Sema::ActOnOpenMPCopyinClause(ArrayRef<Expr *> VarList,
     //  A variable of class type (or array thereof) that appears in a
     //  copyin clause requires an accessible, unambiguous copy assignment
     //  operator for the class type.
-    Type = Context.getBaseElementType(Type);
-    CXXRecordDecl *RD =
-        getLangOpts().CPlusPlus ? Type->getAsCXXRecordDecl() : nullptr;
-    // FIXME This code must be replaced by actual assignment of the
-    // threadprivate variable.
-    if (RD) {
-      CXXMethodDecl *MD = LookupCopyingAssignment(RD, 0, false, 0);
-      DeclAccessPair FoundDecl = DeclAccessPair::make(MD, MD->getAccess());
-      if (MD) {
-        if (CheckMemberAccess(ELoc, RD, FoundDecl) == AR_inaccessible ||
-            MD->isDeleted()) {
-          Diag(ELoc, diag::err_omp_required_method)
-              << getOpenMPClauseName(OMPC_copyin) << 2;
-          bool IsDecl = VD->isThisDeclarationADefinition(Context) ==
-                        VarDecl::DeclarationOnly;
-          Diag(VD->getLocation(),
-               IsDecl ? diag::note_previous_decl : diag::note_defined_here)
-              << VD;
-          Diag(RD->getLocation(), diag::note_previous_decl) << RD;
-          continue;
-        }
-        MarkFunctionReferenced(ELoc, MD);
-        DiagnoseUseOfDecl(MD, ELoc);
-      }
-    }
+    auto ElemType = Context.getBaseElementType(Type).getNonReferenceType();
+    auto *SrcVD = buildVarDecl(*this, DE->getLocStart(),
+                               ElemType.getUnqualifiedType(), ".copyin.src");
+    auto *PseudoSrcExpr = buildDeclRefExpr(
+        *this, SrcVD, ElemType.getUnqualifiedType(), DE->getExprLoc());
+    auto *DstVD =
+        buildVarDecl(*this, DE->getLocStart(), ElemType, ".copyin.dst");
+    auto *PseudoDstExpr =
+        buildDeclRefExpr(*this, DstVD, ElemType, DE->getExprLoc());
+    // For arrays generate assignment operation for single element and replace
+    // it by the original array element in CodeGen.
+    auto AssignmentOp = BuildBinOp(/*S=*/nullptr, DE->getExprLoc(), BO_Assign,
+                                   PseudoDstExpr, PseudoSrcExpr);
+    if (AssignmentOp.isInvalid())
+      continue;
+    AssignmentOp = ActOnFinishFullExpr(AssignmentOp.get(), DE->getExprLoc(),
+                                       /*DiscardedValue=*/true);
+    if (AssignmentOp.isInvalid())
+      continue;
 
     DSAStack->addDSA(VD, DE, OMPC_copyin);
     Vars.push_back(DE);
+    SrcExprs.push_back(PseudoSrcExpr);
+    DstExprs.push_back(PseudoDstExpr);
+    AssignmentOps.push_back(AssignmentOp.get());
   }
 
   if (Vars.empty())
     return nullptr;
 
-  return OMPCopyinClause::Create(Context, StartLoc, LParenLoc, EndLoc, Vars);
+  return OMPCopyinClause::Create(Context, StartLoc, LParenLoc, EndLoc, Vars,
+                                 SrcExprs, DstExprs, AssignmentOps);
 }
 
 OMPClause *Sema::ActOnOpenMPCopyprivateClause(ArrayRef<Expr *> VarList,
@@ -5273,11 +6133,17 @@ OMPClause *Sema::ActOnOpenMPCopyprivateClause(ArrayRef<Expr *> VarList,
                                               SourceLocation LParenLoc,
                                               SourceLocation EndLoc) {
   SmallVector<Expr *, 8> Vars;
+  SmallVector<Expr *, 8> SrcExprs;
+  SmallVector<Expr *, 8> DstExprs;
+  SmallVector<Expr *, 8> AssignmentOps;
   for (auto &RefExpr : VarList) {
     assert(RefExpr && "NULL expr in OpenMP copyprivate clause.");
     if (isa<DependentScopeDeclRefExpr>(RefExpr)) {
       // It will be analyzed later.
       Vars.push_back(RefExpr);
+      SrcExprs.push_back(nullptr);
+      DstExprs.push_back(nullptr);
+      AssignmentOps.push_back(nullptr);
       continue;
     }
 
@@ -5299,6 +6165,9 @@ OMPClause *Sema::ActOnOpenMPCopyprivateClause(ArrayRef<Expr *> VarList,
     if (Type->isDependentType() || Type->isInstantiationDependentType()) {
       // It will be analyzed later.
       Vars.push_back(DE);
+      SrcExprs.push_back(nullptr);
+      DstExprs.push_back(nullptr);
+      AssignmentOps.push_back(nullptr);
       continue;
     }
 
@@ -5307,8 +6176,8 @@ OMPClause *Sema::ActOnOpenMPCopyprivateClause(ArrayRef<Expr *> VarList,
     //  private or firstprivate clause on the single construct.
     if (!DSAStack->isThreadPrivate(VD)) {
       auto DVar = DSAStack->getTopDSA(VD, false);
-      if (DVar.CKind != OMPC_copyprivate && DVar.CKind != OMPC_unknown &&
-          !(DVar.CKind == OMPC_private && !DVar.RefExpr)) {
+      if (DVar.CKind != OMPC_unknown && DVar.CKind != OMPC_copyprivate &&
+          DVar.RefExpr) {
         Diag(ELoc, diag::err_omp_wrong_dsa)
             << getOpenMPClauseName(DVar.CKind)
             << getOpenMPClauseName(OMPC_copyprivate);
@@ -5331,45 +6200,54 @@ OMPClause *Sema::ActOnOpenMPCopyprivateClause(ArrayRef<Expr *> VarList,
       }
     }
 
+    // Variably modified types are not supported.
+    if (!Type->isAnyPointerType() && Type->isVariablyModifiedType()) {
+      Diag(ELoc, diag::err_omp_variably_modified_type_not_supported)
+          << getOpenMPClauseName(OMPC_copyprivate) << Type
+          << getOpenMPDirectiveName(DSAStack->getCurrentDirective());
+      bool IsDecl =
+          VD->isThisDeclarationADefinition(Context) == VarDecl::DeclarationOnly;
+      Diag(VD->getLocation(),
+           IsDecl ? diag::note_previous_decl : diag::note_defined_here)
+          << VD;
+      continue;
+    }
+
     // OpenMP [2.14.4.1, Restrictions, C/C++, p.2]
     //  A variable of class type (or array thereof) that appears in a
     //  copyin clause requires an accessible, unambiguous copy assignment
     //  operator for the class type.
-    Type = Context.getBaseElementType(Type);
-    CXXRecordDecl *RD =
-        getLangOpts().CPlusPlus ? Type->getAsCXXRecordDecl() : nullptr;
-    // FIXME This code must be replaced by actual assignment of the
-    // threadprivate variable.
-    if (RD) {
-      CXXMethodDecl *MD = LookupCopyingAssignment(RD, 0, false, 0);
-      DeclAccessPair FoundDecl = DeclAccessPair::make(MD, MD->getAccess());
-      if (MD) {
-        if (CheckMemberAccess(ELoc, RD, FoundDecl) == AR_inaccessible ||
-            MD->isDeleted()) {
-          Diag(ELoc, diag::err_omp_required_method)
-              << getOpenMPClauseName(OMPC_copyprivate) << 2;
-          bool IsDecl = VD->isThisDeclarationADefinition(Context) ==
-                        VarDecl::DeclarationOnly;
-          Diag(VD->getLocation(),
-               IsDecl ? diag::note_previous_decl : diag::note_defined_here)
-              << VD;
-          Diag(RD->getLocation(), diag::note_previous_decl) << RD;
-          continue;
-        }
-        MarkFunctionReferenced(ELoc, MD);
-        DiagnoseUseOfDecl(MD, ELoc);
-      }
-    }
+    Type = Context.getBaseElementType(Type).getUnqualifiedType();
+    auto *SrcVD =
+        buildVarDecl(*this, DE->getLocStart(), Type, ".copyprivate.src");
+    auto *PseudoSrcExpr =
+        buildDeclRefExpr(*this, SrcVD, Type, DE->getExprLoc());
+    auto *DstVD =
+        buildVarDecl(*this, DE->getLocStart(), Type, ".copyprivate.dst");
+    auto *PseudoDstExpr =
+        buildDeclRefExpr(*this, DstVD, Type, DE->getExprLoc());
+    auto AssignmentOp = BuildBinOp(/*S=*/nullptr, DE->getExprLoc(), BO_Assign,
+                                   PseudoDstExpr, PseudoSrcExpr);
+    if (AssignmentOp.isInvalid())
+      continue;
+    AssignmentOp = ActOnFinishFullExpr(AssignmentOp.get(), DE->getExprLoc(),
+                                       /*DiscardedValue=*/true);
+    if (AssignmentOp.isInvalid())
+      continue;
 
     // No need to mark vars as copyprivate, they are already threadprivate or
     // implicitly private.
     Vars.push_back(DE);
+    SrcExprs.push_back(PseudoSrcExpr);
+    DstExprs.push_back(PseudoDstExpr);
+    AssignmentOps.push_back(AssignmentOp.get());
   }
 
   if (Vars.empty())
     return nullptr;
 
-  return OMPCopyprivateClause::Create(Context, StartLoc, LParenLoc, EndLoc, Vars);
+  return OMPCopyprivateClause::Create(Context, StartLoc, LParenLoc, EndLoc,
+                                      Vars, SrcExprs, DstExprs, AssignmentOps);
 }
 
 OMPClause *Sema::ActOnOpenMPFlushClause(ArrayRef<Expr *> VarList,
diff --git a/lib/Sema/SemaOverload.cpp b/lib/Sema/SemaOverload.cpp
index 9195ee59075d..9d87a103ac66 100644
--- a/lib/Sema/SemaOverload.cpp
+++ b/lib/Sema/SemaOverload.cpp
@@ -286,6 +286,16 @@ StandardConversionSequence::getNarrowingKind(ASTContext &Ctx,
   QualType FromType = getToType(0);
   QualType ToType = getToType(1);
   switch (Second) {
+  // 'bool' is an integral type; dispatch to the right place to handle it.
+  case ICK_Boolean_Conversion:
+    if (FromType->isRealFloatingType())
+      goto FloatingIntegralConversion;
+    if (FromType->isIntegralOrUnscopedEnumerationType())
+      goto IntegralConversion;
+    // Boolean conversions can be from pointers and pointers to members
+    // [conv.bool], and those aren't considered narrowing conversions.
+    return NK_Not_Narrowing;
+
   // -- from a floating-point type to an integer type, or
   //
   // -- from an integer type or unscoped enumeration type to a floating-point
@@ -293,6 +303,7 @@ StandardConversionSequence::getNarrowingKind(ASTContext &Ctx,
   //    value after conversion will fit into the target type and will produce
   //    the original value when converted back to the original type, or
   case ICK_Floating_Integral:
+  FloatingIntegralConversion:
     if (FromType->isRealFloatingType() && ToType->isIntegralType(Ctx)) {
       return NK_Type_Narrowing;
     } else if (FromType->isIntegralType(Ctx) && ToType->isRealFloatingType()) {
@@ -357,13 +368,8 @@ StandardConversionSequence::getNarrowingKind(ASTContext &Ctx,
   //    the source is a constant expression and the actual value after
   //    conversion will fit into the target type and will produce the original
   //    value when converted back to the original type.
-  case ICK_Boolean_Conversion:  // Bools are integers too.
-    if (!FromType->isIntegralOrUnscopedEnumerationType()) {
-      // Boolean conversions can be from pointers and pointers to members
-      // [conv.bool], and those aren't considered narrowing conversions.
-      return NK_Not_Narrowing;
-    }  // Otherwise, fall through to the integral case.
-  case ICK_Integral_Conversion: {
+  case ICK_Integral_Conversion:
+  IntegralConversion: {
     assert(FromType->isIntegralOrUnscopedEnumerationType());
     assert(ToType->isIntegralOrUnscopedEnumerationType());
     const bool FromSigned = FromType->isSignedIntegerOrEnumerationType();
@@ -1098,11 +1104,11 @@ TryUserDefinedConversion(Sema &S, Expr *From, QualType ToType,
   // Attempt user-defined conversion.
   OverloadCandidateSet Conversions(From->getExprLoc(),
                                    OverloadCandidateSet::CSK_Normal);
-  OverloadingResult UserDefResult
-    = IsUserDefinedConversion(S, From, ToType, ICS.UserDefined, Conversions,
-                              AllowExplicit, AllowObjCConversionOnExplicit);
-
-  if (UserDefResult == OR_Success) {
+  switch (IsUserDefinedConversion(S, From, ToType, ICS.UserDefined,
+                                  Conversions, AllowExplicit,
+                                  AllowObjCConversionOnExplicit)) {
+  case OR_Success:
+  case OR_Deleted:
     ICS.setUserDefined();
     ICS.UserDefined.Before.setAsIdentityConversion();
     // C++ [over.ics.user]p4:
@@ -1131,7 +1137,9 @@ TryUserDefinedConversion(Sema &S, Expr *From, QualType ToType,
           ICS.Standard.Second = ICK_Derived_To_Base;
       }
     }
-  } else if (UserDefResult == OR_Ambiguous && !SuppressUserConversions) {
+    break;
+
+  case OR_Ambiguous:
     ICS.setAmbiguous();
     ICS.Ambiguous.setFromType(From->getType());
     ICS.Ambiguous.setToType(ToType);
@@ -1139,8 +1147,12 @@ TryUserDefinedConversion(Sema &S, Expr *From, QualType ToType,
          Cand != Conversions.end(); ++Cand)
       if (Cand->Viable)
         ICS.Ambiguous.addConversion(Cand->Function);
-  } else {
+    break;
+
+    // Fall through.
+  case OR_No_Viable_Function:
     ICS.setBad(BadConversionSequence::no_conversion, From, ToType);
+    break;
   }
 
   return ICS;
@@ -1740,18 +1752,20 @@ bool Sema::IsIntegralPromotion(Expr *From, QualType FromType, QualType ToType) {
       return false;
 
     // We can perform an integral promotion to the underlying type of the enum,
-    // even if that's not the promoted type.
+    // even if that's not the promoted type. Note that the check for promoting
+    // the underlying type is based on the type alone, and does not consider
+    // the bitfield-ness of the actual source expression.
     if (FromEnumType->getDecl()->isFixed()) {
       QualType Underlying = FromEnumType->getDecl()->getIntegerType();
       return Context.hasSameUnqualifiedType(Underlying, ToType) ||
-             IsIntegralPromotion(From, Underlying, ToType);
+             IsIntegralPromotion(nullptr, Underlying, ToType);
     }
 
     // We have already pre-calculated the promotion type, so this is trivial.
     if (ToType->isIntegerType() &&
         !RequireCompleteType(From->getLocStart(), FromType, 0))
-      return Context.hasSameUnqualifiedType(ToType,
-                                FromEnumType->getDecl()->getPromotionType());
+      return Context.hasSameUnqualifiedType(
+          ToType, FromEnumType->getDecl()->getPromotionType());
   }
 
   // C++0x [conv.prom]p2:
@@ -1799,13 +1813,12 @@ bool Sema::IsIntegralPromotion(Expr *From, QualType FromType, QualType ToType) {
   // other value of that type for promotion purposes (C++ 4.5p3).
   // FIXME: We should delay checking of bit-fields until we actually perform the
   // conversion.
-  using llvm::APSInt;
-  if (From)
+  if (From) {
     if (FieldDecl *MemberDecl = From->getSourceBitField()) {
-      APSInt BitWidth;
+      llvm::APSInt BitWidth;
       if (FromType->isIntegralType(Context) &&
           MemberDecl->getBitWidth()->isIntegerConstantExpr(BitWidth, Context)) {
-        APSInt ToSize(BitWidth.getBitWidth(), BitWidth.isUnsigned());
+        llvm::APSInt ToSize(BitWidth.getBitWidth(), BitWidth.isUnsigned());
         ToSize = Context.getTypeSize(ToType);
 
         // Are we promoting to an int from a bitfield that fits in an int?
@@ -1823,6 +1836,7 @@ bool Sema::IsIntegralPromotion(Expr *From, QualType FromType, QualType ToType) {
         return false;
       }
     }
+  }
 
   // An rvalue of type bool can be converted to an rvalue of type int,
   // with false becoming zero and true becoming one (C++ 4.5p4).
@@ -2940,7 +2954,10 @@ IsInitializerListConstructorConversion(Sema &S, Expr *From, QualType ToType,
   bool HadMultipleCandidates = (CandidateSet.size() > 1);
 
   OverloadCandidateSet::iterator Best;
-  switch (CandidateSet.BestViableFunction(S, From->getLocStart(), Best, true)) {
+  switch (auto Result = 
+            CandidateSet.BestViableFunction(S, From->getLocStart(), 
+                                            Best, true)) {
+  case OR_Deleted:
   case OR_Success: {
     // Record the standard conversion we used and the conversion function.
     CXXConstructorDecl *Constructor = cast<CXXConstructorDecl>(Best->Function);
@@ -2953,13 +2970,11 @@ IsInitializerListConstructorConversion(Sema &S, Expr *From, QualType ToType,
     User.After.setAsIdentityConversion();
     User.After.setFromType(ThisType->getAs<PointerType>()->getPointeeType());
     User.After.setAllToTypes(ToType);
-    return OR_Success;
+    return Result;
   }
 
   case OR_No_Viable_Function:
     return OR_No_Viable_Function;
-  case OR_Deleted:
-    return OR_Deleted;
   case OR_Ambiguous:
     return OR_Ambiguous;
   }
@@ -3093,11 +3108,8 @@ IsUserDefinedConversion(Sema &S, Expr *From, QualType ToType,
     if (CXXRecordDecl *FromRecordDecl
          = dyn_cast<CXXRecordDecl>(FromRecordType->getDecl())) {
       // Add all of the conversion functions as candidates.
-      std::pair<CXXRecordDecl::conversion_iterator,
-                CXXRecordDecl::conversion_iterator>
-        Conversions = FromRecordDecl->getVisibleConversionFunctions();
-      for (CXXRecordDecl::conversion_iterator
-             I = Conversions.first, E = Conversions.second; I != E; ++I) {
+      const auto &Conversions = FromRecordDecl->getVisibleConversionFunctions();
+      for (auto I = Conversions.begin(), E = Conversions.end(); I != E; ++I) {
         DeclAccessPair FoundDecl = I.getPair();
         NamedDecl *D = FoundDecl.getDecl();
         CXXRecordDecl *ActingContext = cast<CXXRecordDecl>(D->getDeclContext());
@@ -3129,8 +3141,10 @@ IsUserDefinedConversion(Sema &S, Expr *From, QualType ToType,
   bool HadMultipleCandidates = (CandidateSet.size() > 1);
 
   OverloadCandidateSet::iterator Best;
-  switch (CandidateSet.BestViableFunction(S, From->getLocStart(), Best, true)) {
+  switch (auto Result = CandidateSet.BestViableFunction(S, From->getLocStart(),
+                                                        Best, true)) {
   case OR_Success:
+  case OR_Deleted:
     // Record the standard conversion we used and the conversion function.
     if (CXXConstructorDecl *Constructor
           = dyn_cast<CXXConstructorDecl>(Best->Function)) {
@@ -3158,7 +3172,7 @@ IsUserDefinedConversion(Sema &S, Expr *From, QualType ToType,
       User.After.setAsIdentityConversion();
       User.After.setFromType(ThisType->getAs<PointerType>()->getPointeeType());
       User.After.setAllToTypes(ToType);
-      return OR_Success;
+      return Result;
     }
     if (CXXConversionDecl *Conversion
                  = dyn_cast<CXXConversionDecl>(Best->Function)) {
@@ -3184,15 +3198,12 @@ IsUserDefinedConversion(Sema &S, Expr *From, QualType ToType,
       //   user-defined conversion sequence (see 13.3.3 and
       //   13.3.3.1).
       User.After = Best->FinalConversion;
-      return OR_Success;
+      return Result;
     }
     llvm_unreachable("Not a constructor or conversion function?");
 
   case OR_No_Viable_Function:
     return OR_No_Viable_Function;
-  case OR_Deleted:
-    // No conversion here! We're done.
-    return OR_Deleted;
 
   case OR_Ambiguous:
     return OR_Ambiguous;
@@ -3329,7 +3340,26 @@ CompareImplicitConversionSequences(Sema &S,
   // Two implicit conversion sequences of the same form are
   // indistinguishable conversion sequences unless one of the
   // following rules apply: (C++ 13.3.3.2p3):
+  
+  // List-initialization sequence L1 is a better conversion sequence than
+  // list-initialization sequence L2 if:
+  // - L1 converts to std::initializer_list<X> for some X and L2 does not, or,
+  //   if not that,
+  // - L1 converts to type "array of N1 T", L2 converts to type "array of N2 T",
+  //   and N1 is smaller than N2.,
+  // even if one of the other rules in this paragraph would otherwise apply.
+  if (!ICS1.isBad()) {
+    if (ICS1.isStdInitializerListElement() &&
+        !ICS2.isStdInitializerListElement())
+      return ImplicitConversionSequence::Better;
+    if (!ICS1.isStdInitializerListElement() &&
+        ICS2.isStdInitializerListElement())
+      return ImplicitConversionSequence::Worse;
+  }
+
   if (ICS1.isStandard())
+    // Standard conversion sequence S1 is a better conversion sequence than
+    // standard conversion sequence S2 if [...]
     Result = CompareStandardConversionSequences(S,
                                                 ICS1.Standard, ICS2.Standard);
   else if (ICS1.isUserDefined()) {
@@ -3350,19 +3380,6 @@ CompareImplicitConversionSequences(Sema &S,
                                           ICS2.UserDefined.ConversionFunction);
   }
 
-  // List-initialization sequence L1 is a better conversion sequence than
-  // list-initialization sequence L2 if L1 converts to std::initializer_list<X>
-  // for some X and L2 does not.
-  if (Result == ImplicitConversionSequence::Indistinguishable &&
-      !ICS1.isBad()) {
-    if (ICS1.isStdInitializerListElement() &&
-        !ICS2.isStdInitializerListElement())
-      return ImplicitConversionSequence::Better;
-    if (!ICS1.isStdInitializerListElement() &&
-        ICS2.isStdInitializerListElement())
-      return ImplicitConversionSequence::Worse;
-  }
-
   return Result;
 }
 
@@ -4034,11 +4051,8 @@ FindConversionForRefInit(Sema &S, ImplicitConversionSequence &ICS,
     = dyn_cast<CXXRecordDecl>(T2->getAs<RecordType>()->getDecl());
 
   OverloadCandidateSet CandidateSet(DeclLoc, OverloadCandidateSet::CSK_Normal);
-  std::pair<CXXRecordDecl::conversion_iterator,
-            CXXRecordDecl::conversion_iterator>
-    Conversions = T2RecordDecl->getVisibleConversionFunctions();
-  for (CXXRecordDecl::conversion_iterator
-         I = Conversions.first, E = Conversions.second; I != E; ++I) {
+  const auto &Conversions = T2RecordDecl->getVisibleConversionFunctions();
+  for (auto I = Conversions.begin(), E = Conversions.end(); I != E; ++I) {
     NamedDecl *D = *I;
     CXXRecordDecl *ActingDC = cast<CXXRecordDecl>(D->getDeclContext());
     if (isa<UsingShadowDecl>(D))
@@ -4254,16 +4268,6 @@ TryReferenceInit(Sema &S, Expr *Init, QualType DeclType,
   //     -- Otherwise, the reference shall be an lvalue reference to a
   //        non-volatile const type (i.e., cv1 shall be const), or the reference
   //        shall be an rvalue reference.
-  //
-  // We actually handle one oddity of C++ [over.ics.ref] at this
-  // point, which is that, due to p2 (which short-circuits reference
-  // binding by only attempting a simple conversion for non-direct
-  // bindings) and p3's strange wording, we allow a const volatile
-  // reference to bind to an rvalue. Hence the check for the presence
-  // of "const" rather than checking for "const" being the only
-  // qualifier.
-  // This is also the point where rvalue references and lvalue inits no longer
-  // go together.
   if (!isRValRef && (!T1.isConstQualified() || T1.isVolatileQualified()))
     return ICS;
 
@@ -4456,11 +4460,57 @@ TryListConversion(Sema &S, InitListExpr *From, QualType ToType,
   if (S.RequireCompleteType(From->getLocStart(), ToType, 0))
     return Result;
 
+  // Per DR1467:
+  //   If the parameter type is a class X and the initializer list has a single
+  //   element of type cv U, where U is X or a class derived from X, the
+  //   implicit conversion sequence is the one required to convert the element
+  //   to the parameter type.
+  //
+  //   Otherwise, if the parameter type is a character array [... ]
+  //   and the initializer list has a single element that is an
+  //   appropriately-typed string literal (8.5.2 [dcl.init.string]), the
+  //   implicit conversion sequence is the identity conversion.
+  if (From->getNumInits() == 1) {
+    if (ToType->isRecordType()) {
+      QualType InitType = From->getInit(0)->getType();
+      if (S.Context.hasSameUnqualifiedType(InitType, ToType) ||
+          S.IsDerivedFrom(InitType, ToType))
+        return TryCopyInitialization(S, From->getInit(0), ToType,
+                                     SuppressUserConversions,
+                                     InOverloadResolution,
+                                     AllowObjCWritebackConversion);
+    }
+    // FIXME: Check the other conditions here: array of character type,
+    // initializer is a string literal.
+    if (ToType->isArrayType()) {
+      InitializedEntity Entity =
+        InitializedEntity::InitializeParameter(S.Context, ToType,
+                                               /*Consumed=*/false);
+      if (S.CanPerformCopyInitialization(Entity, From)) {
+        Result.setStandard();
+        Result.Standard.setAsIdentityConversion();
+        Result.Standard.setFromType(ToType);
+        Result.Standard.setAllToTypes(ToType);
+        return Result;
+      }
+    }
+  }
+
+  // C++14 [over.ics.list]p2: Otherwise, if the parameter type [...] (below).
   // C++11 [over.ics.list]p2:
   //   If the parameter type is std::initializer_list<X> or "array of X" and
   //   all the elements can be implicitly converted to X, the implicit
   //   conversion sequence is the worst conversion necessary to convert an
   //   element of the list to X.
+  //
+  // C++14 [over.ics.list]p3:
+  //   Otherwise, if the parameter type is "array of N X", if the initializer
+  //   list has exactly N elements or if it has fewer than N elements and X is
+  //   default-constructible, and if all the elements of the initializer list
+  //   can be implicitly converted to X, the implicit conversion sequence is
+  //   the worst conversion necessary to convert an element of the list to X.
+  //
+  // FIXME: We're missing a lot of these checks.
   bool toStdInitializerList = false;
   QualType X;
   if (ToType->isArrayType())
@@ -4499,6 +4549,7 @@ TryListConversion(Sema &S, InitListExpr *From, QualType ToType,
     return Result;
   }
 
+  // C++14 [over.ics.list]p4:
   // C++11 [over.ics.list]p3:
   //   Otherwise, if the parameter is a non-aggregate class X and overload
   //   resolution chooses a single best constructor [...] the implicit
@@ -4514,6 +4565,7 @@ TryListConversion(Sema &S, InitListExpr *From, QualType ToType,
                                     /*AllowObjCConversionOnExplicit=*/false);
   }
 
+  // C++14 [over.ics.list]p5:
   // C++11 [over.ics.list]p4:
   //   Otherwise, if the parameter has an aggregate type which can be
   //   initialized from the initializer list [...] the implicit conversion
@@ -4540,6 +4592,7 @@ TryListConversion(Sema &S, InitListExpr *From, QualType ToType,
     return Result;
   }
 
+  // C++14 [over.ics.list]p6:
   // C++11 [over.ics.list]p5:
   //   Otherwise, if the parameter is a reference, see 13.3.3.1.4.
   if (ToType->isReferenceType()) {
@@ -4608,14 +4661,15 @@ TryListConversion(Sema &S, InitListExpr *From, QualType ToType,
     return Result;
   }
 
+  // C++14 [over.ics.list]p7:
   // C++11 [over.ics.list]p6:
   //   Otherwise, if the parameter type is not a class:
   if (!ToType->isRecordType()) {
-    //    - if the initializer list has one element, the implicit conversion
-    //      sequence is the one required to convert the element to the
-    //      parameter type.
+    //    - if the initializer list has one element that is not itself an
+    //      initializer list, the implicit conversion sequence is the one
+    //      required to convert the element to the parameter type.
     unsigned NumInits = From->getNumInits();
-    if (NumInits == 1)
+    if (NumInits == 1 && !isa<InitListExpr>(From->getInit(0)))
       Result = TryCopyInitialization(S, From->getInit(0), ToType,
                                      SuppressUserConversions,
                                      InOverloadResolution,
@@ -4631,6 +4685,7 @@ TryListConversion(Sema &S, InitListExpr *From, QualType ToType,
     return Result;
   }
 
+  // C++14 [over.ics.list]p8:
   // C++11 [over.ics.list]p7:
   //   In all cases other than those enumerated above, no conversion is possible
   return Result;
@@ -5347,21 +5402,18 @@ ExprResult Sema::PerformContextualImplicitConversion(
   UnresolvedSet<4>
       ViableConversions; // These are *potentially* viable in C++1y.
   UnresolvedSet<4> ExplicitConversions;
-  std::pair<CXXRecordDecl::conversion_iterator,
-            CXXRecordDecl::conversion_iterator> Conversions =
+  const auto &Conversions =
       cast<CXXRecordDecl>(RecordTy->getDecl())->getVisibleConversionFunctions();
 
   bool HadMultipleCandidates =
-      (std::distance(Conversions.first, Conversions.second) > 1);
+      (std::distance(Conversions.begin(), Conversions.end()) > 1);
 
   // To check that there is only one target type, in C++1y:
   QualType ToType;
   bool HasUniqueTargetType = true;
 
   // Collect explicit or viable (potentially in C++1y) conversions.
-  for (CXXRecordDecl::conversion_iterator I = Conversions.first,
-                                          E = Conversions.second;
-       I != E; ++I) {
+  for (auto I = Conversions.begin(), E = Conversions.end(); I != E; ++I) {
     NamedDecl *D = (*I)->getUnderlyingDecl();
     CXXConversionDecl *Conversion;
     FunctionTemplateDecl *ConvTemplate = dyn_cast<FunctionTemplateDecl>(D);
@@ -5554,7 +5606,8 @@ Sema::AddOverloadCandidate(FunctionDecl *Function,
       // is irrelevant.
       AddMethodCandidate(Method, FoundDecl, Method->getParent(),
                          QualType(), Expr::Classification::makeSimpleLValue(),
-                         Args, CandidateSet, SuppressUserConversions);
+                         Args, CandidateSet, SuppressUserConversions,
+                         PartialOverloading);
       return;
     }
     // We treat a constructor like a non-member function, since its object
@@ -5615,7 +5668,7 @@ Sema::AddOverloadCandidate(FunctionDecl *Function,
   // (C++ 13.3.2p2): A candidate function having fewer than m
   // parameters is viable only if it has an ellipsis in its parameter
   // list (8.3.5).
-  if ((Args.size() + (PartialOverloading && Args.size())) > NumParams &&
+  if (TooManyArguments(NumParams, Args.size(), PartialOverloading) &&
       !Proto->isVariadic()) {
     Candidate.Viable = false;
     Candidate.FailureKind = ovl_fail_too_many_arguments;
@@ -5850,8 +5903,9 @@ EnableIfAttr *Sema::CheckEnableIf(FunctionDecl *Function, ArrayRef<Expr *> Args,
 void Sema::AddFunctionCandidates(const UnresolvedSetImpl &Fns,
                                  ArrayRef<Expr *> Args,
                                  OverloadCandidateSet& CandidateSet,
+                                 TemplateArgumentListInfo *ExplicitTemplateArgs,
                                  bool SuppressUserConversions,
-                               TemplateArgumentListInfo *ExplicitTemplateArgs) {
+                                 bool PartialOverloading) {
   for (UnresolvedSetIterator F = Fns.begin(), E = Fns.end(); F != E; ++F) {
     NamedDecl *D = F.getDecl()->getUnderlyingDecl();
     if (FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) {
@@ -5860,10 +5914,10 @@ void Sema::AddFunctionCandidates(const UnresolvedSetImpl &Fns,
                            cast<CXXMethodDecl>(FD)->getParent(),
                            Args[0]->getType(), Args[0]->Classify(Context),
                            Args.slice(1), CandidateSet,
-                           SuppressUserConversions);
+                           SuppressUserConversions, PartialOverloading);
       else
         AddOverloadCandidate(FD, F.getPair(), Args, CandidateSet,
-                             SuppressUserConversions);
+                             SuppressUserConversions, PartialOverloading);
     } else {
       FunctionTemplateDecl *FunTmpl = cast<FunctionTemplateDecl>(D);
       if (isa<CXXMethodDecl>(FunTmpl->getTemplatedDecl()) &&
@@ -5873,11 +5927,13 @@ void Sema::AddFunctionCandidates(const UnresolvedSetImpl &Fns,
                                    ExplicitTemplateArgs,
                                    Args[0]->getType(),
                                    Args[0]->Classify(Context), Args.slice(1),
-                                   CandidateSet, SuppressUserConversions);
+                                   CandidateSet, SuppressUserConversions,
+                                   PartialOverloading);
       else
         AddTemplateOverloadCandidate(FunTmpl, F.getPair(),
                                      ExplicitTemplateArgs, Args,
-                                     CandidateSet, SuppressUserConversions);
+                                     CandidateSet, SuppressUserConversions,
+                                     PartialOverloading);
     }
   }
 }
@@ -5925,7 +5981,8 @@ Sema::AddMethodCandidate(CXXMethodDecl *Method, DeclAccessPair FoundDecl,
                          Expr::Classification ObjectClassification,
                          ArrayRef<Expr *> Args,
                          OverloadCandidateSet &CandidateSet,
-                         bool SuppressUserConversions) {
+                         bool SuppressUserConversions,
+                         bool PartialOverloading) {
   const FunctionProtoType *Proto
     = dyn_cast<FunctionProtoType>(Method->getType()->getAs<FunctionType>());
   assert(Proto && "Methods without a prototype cannot be overloaded");
@@ -5958,7 +6015,8 @@ Sema::AddMethodCandidate(CXXMethodDecl *Method, DeclAccessPair FoundDecl,
   // (C++ 13.3.2p2): A candidate function having fewer than m
   // parameters is viable only if it has an ellipsis in its parameter
   // list (8.3.5).
-  if (Args.size() > NumParams && !Proto->isVariadic()) {
+  if (TooManyArguments(NumParams, Args.size(), PartialOverloading) &&
+      !Proto->isVariadic()) {
     Candidate.Viable = false;
     Candidate.FailureKind = ovl_fail_too_many_arguments;
     return;
@@ -5970,7 +6028,7 @@ Sema::AddMethodCandidate(CXXMethodDecl *Method, DeclAccessPair FoundDecl,
   // parameter list is truncated on the right, so that there are
   // exactly m parameters.
   unsigned MinRequiredArgs = Method->getMinRequiredArguments();
-  if (Args.size() < MinRequiredArgs) {
+  if (Args.size() < MinRequiredArgs && !PartialOverloading) {
     // Not enough arguments.
     Candidate.Viable = false;
     Candidate.FailureKind = ovl_fail_too_few_arguments;
@@ -6052,7 +6110,8 @@ Sema::AddMethodTemplateCandidate(FunctionTemplateDecl *MethodTmpl,
                                  Expr::Classification ObjectClassification,
                                  ArrayRef<Expr *> Args,
                                  OverloadCandidateSet& CandidateSet,
-                                 bool SuppressUserConversions) {
+                                 bool SuppressUserConversions,
+                                 bool PartialOverloading) {
   if (!CandidateSet.isNewCandidate(MethodTmpl))
     return;
 
@@ -6069,7 +6128,7 @@ Sema::AddMethodTemplateCandidate(FunctionTemplateDecl *MethodTmpl,
   FunctionDecl *Specialization = nullptr;
   if (TemplateDeductionResult Result
       = DeduceTemplateArguments(MethodTmpl, ExplicitTemplateArgs, Args,
-                                Specialization, Info)) {
+                                Specialization, Info, PartialOverloading)) {
     OverloadCandidate &Candidate = CandidateSet.addCandidate();
     Candidate.FoundDecl = FoundDecl;
     Candidate.Function = MethodTmpl->getTemplatedDecl();
@@ -6090,7 +6149,7 @@ Sema::AddMethodTemplateCandidate(FunctionTemplateDecl *MethodTmpl,
          "Specialization is not a member function?");
   AddMethodCandidate(cast<CXXMethodDecl>(Specialization), FoundDecl,
                      ActingContext, ObjectType, ObjectClassification, Args,
-                     CandidateSet, SuppressUserConversions);
+                     CandidateSet, SuppressUserConversions, PartialOverloading);
 }
 
 /// \brief Add a C++ function template specialization as a candidate
@@ -6102,7 +6161,8 @@ Sema::AddTemplateOverloadCandidate(FunctionTemplateDecl *FunctionTemplate,
                                  TemplateArgumentListInfo *ExplicitTemplateArgs,
                                    ArrayRef<Expr *> Args,
                                    OverloadCandidateSet& CandidateSet,
-                                   bool SuppressUserConversions) {
+                                   bool SuppressUserConversions,
+                                   bool PartialOverloading) {
   if (!CandidateSet.isNewCandidate(FunctionTemplate))
     return;
 
@@ -6119,7 +6179,7 @@ Sema::AddTemplateOverloadCandidate(FunctionTemplateDecl *FunctionTemplate,
   FunctionDecl *Specialization = nullptr;
   if (TemplateDeductionResult Result
         = DeduceTemplateArguments(FunctionTemplate, ExplicitTemplateArgs, Args,
-                                  Specialization, Info)) {
+                                  Specialization, Info, PartialOverloading)) {
     OverloadCandidate &Candidate = CandidateSet.addCandidate();
     Candidate.FoundDecl = FoundDecl;
     Candidate.Function = FunctionTemplate->getTemplatedDecl();
@@ -6137,7 +6197,7 @@ Sema::AddTemplateOverloadCandidate(FunctionTemplateDecl *FunctionTemplate,
   // deduction as a candidate.
   assert(Specialization && "Missing function template specialization?");
   AddOverloadCandidate(Specialization, FoundDecl, Args, CandidateSet,
-                       SuppressUserConversions);
+                       SuppressUserConversions, PartialOverloading);
 }
 
 /// Determine whether this is an allowable conversion from the result
@@ -6884,12 +6944,7 @@ BuiltinCandidateTypeSet::AddTypesConvertedFrom(QualType Ty,
       return;
 
     CXXRecordDecl *ClassDecl = cast<CXXRecordDecl>(TyRec->getDecl());
-    std::pair<CXXRecordDecl::conversion_iterator,
-              CXXRecordDecl::conversion_iterator>
-      Conversions = ClassDecl->getVisibleConversionFunctions();
-    for (CXXRecordDecl::conversion_iterator
-           I = Conversions.first, E = Conversions.second; I != E; ++I) {
-      NamedDecl *D = I.getDecl();
+    for (NamedDecl *D : ClassDecl->getVisibleConversionFunctions()) {
       if (isa<UsingShadowDecl>(D))
         D = cast<UsingShadowDecl>(D)->getTargetDecl();
 
@@ -6953,13 +7008,7 @@ static  Qualifiers CollectVRQualifiers(ASTContext &Context, Expr* ArgExpr) {
     if (!ClassDecl->hasDefinition())
       return VRQuals;
 
-    std::pair<CXXRecordDecl::conversion_iterator,
-              CXXRecordDecl::conversion_iterator>
-      Conversions = ClassDecl->getVisibleConversionFunctions();
-
-    for (CXXRecordDecl::conversion_iterator
-           I = Conversions.first, E = Conversions.second; I != E; ++I) {
-      NamedDecl *D = I.getDecl();
+    for (NamedDecl *D : ClassDecl->getVisibleConversionFunctions()) {
       if (isa<UsingShadowDecl>(D))
         D = cast<UsingShadowDecl>(D)->getTargetDecl();
       if (CXXConversionDecl *Conv = dyn_cast<CXXConversionDecl>(D)) {
@@ -8078,7 +8127,7 @@ void Sema::AddBuiltinOperatorCandidates(OverloadedOperatorKind Op,
   bool HasArithmeticOrEnumeralCandidateType = false;
   SmallVector<BuiltinCandidateTypeSet, 2> CandidateTypes;
   for (unsigned ArgIdx = 0, N = Args.size(); ArgIdx != N; ++ArgIdx) {
-    CandidateTypes.push_back(BuiltinCandidateTypeSet(*this));
+    CandidateTypes.emplace_back(*this);
     CandidateTypes[ArgIdx].AddTypesConvertedFrom(Args[ArgIdx]->getType(),
                                                  OpLoc,
                                                  true,
@@ -8291,7 +8340,7 @@ Sema::AddArgumentDependentLookupCandidates(DeclarationName Name,
     } else
       AddTemplateOverloadCandidate(cast<FunctionTemplateDecl>(*I),
                                    FoundDecl, ExplicitTemplateArgs,
-                                   Args, CandidateSet);
+                                   Args, CandidateSet, PartialOverloading);
   }
 }
 
@@ -9451,10 +9500,7 @@ struct CompareOverloadCandidatesForDisplay {
         numLFixes = (numLFixes == 0) ? UINT_MAX : numLFixes;
         numRFixes = (numRFixes == 0) ? UINT_MAX : numRFixes;
         if (numLFixes != numRFixes) {
-          if (numLFixes < numRFixes)
-            return true;
-          else
-            return false;
+          return numLFixes < numRFixes;
         }
 
         // If there's any ordering between the defined conversions...
@@ -10363,7 +10409,8 @@ static void AddOverloadedCallCandidate(Sema &S,
       assert(!KnownValid && "Explicit template arguments?");
       return;
     }
-    S.AddOverloadCandidate(Func, FoundDecl, Args, CandidateSet, false,
+    S.AddOverloadCandidate(Func, FoundDecl, Args, CandidateSet,
+                           /*SuppressUsedConversions=*/false,
                            PartialOverloading);
     return;
   }
@@ -10371,7 +10418,9 @@ static void AddOverloadedCallCandidate(Sema &S,
   if (FunctionTemplateDecl *FuncTemplate
       = dyn_cast<FunctionTemplateDecl>(Callee)) {
     S.AddTemplateOverloadCandidate(FuncTemplate, FoundDecl,
-                                   ExplicitTemplateArgs, Args, CandidateSet);
+                                   ExplicitTemplateArgs, Args, CandidateSet,
+                                   /*SuppressUsedConversions=*/false,
+                                   PartialOverloading);
     return;
   }
 
@@ -10896,7 +10945,7 @@ Sema::CreateOverloadedUnaryOp(SourceLocation OpLoc, unsigned OpcIn,
   OverloadCandidateSet CandidateSet(OpLoc, OverloadCandidateSet::CSK_Operator);
 
   // Add the candidates from the given function set.
-  AddFunctionCandidates(Fns, ArgsArray, CandidateSet, false);
+  AddFunctionCandidates(Fns, ArgsArray, CandidateSet);
 
   // Add operator candidates that are member functions.
   AddMemberOperatorCandidates(Op, OpLoc, ArgsArray, CandidateSet);
@@ -11105,7 +11154,7 @@ Sema::CreateOverloadedBinOp(SourceLocation OpLoc,
   OverloadCandidateSet CandidateSet(OpLoc, OverloadCandidateSet::CSK_Operator);
 
   // Add the candidates from the given function set.
-  AddFunctionCandidates(Fns, Args, CandidateSet, false);
+  AddFunctionCandidates(Fns, Args, CandidateSet);
 
   // Add operator candidates that are member functions.
   AddMemberOperatorCandidates(Op, OpLoc, Args, CandidateSet);
@@ -11546,6 +11595,10 @@ Sema::BuildCallToMemberFunction(Scope *S, Expr *MemExprE,
     return MaybeBindToTemporary(call);
   }
 
+  if (isa<CXXPseudoDestructorExpr>(NakedMemExpr))
+    return new (Context)
+        CallExpr(Context, MemExprE, Args, Context.VoidTy, VK_RValue, RParenLoc);
+
   UnbridgedCastsSet UnbridgedCasts;
   if (checkArgPlaceholdersForOverload(*this, Args, UnbridgedCasts))
     return ExprError();
@@ -11805,11 +11858,9 @@ Sema::BuildCallToObjectOfClassType(Scope *S, Expr *Obj,
   //   functions for each conversion function declared in an
   //   accessible base class provided the function is not hidden
   //   within T by another intervening declaration.
-  std::pair<CXXRecordDecl::conversion_iterator,
-            CXXRecordDecl::conversion_iterator> Conversions
-    = cast<CXXRecordDecl>(Record->getDecl())->getVisibleConversionFunctions();
-  for (CXXRecordDecl::conversion_iterator
-         I = Conversions.first, E = Conversions.second; I != E; ++I) {
+  const auto &Conversions =
+      cast<CXXRecordDecl>(Record->getDecl())->getVisibleConversionFunctions();
+  for (auto I = Conversions.begin(), E = Conversions.end(); I != E; ++I) {
     NamedDecl *D = *I;
     CXXRecordDecl *ActingContext = cast<CXXRecordDecl>(D->getDeclContext());
     if (isa<UsingShadowDecl>(D))
@@ -12154,8 +12205,8 @@ ExprResult Sema::BuildLiteralOperatorCall(LookupResult &R,
 
   OverloadCandidateSet CandidateSet(UDSuffixLoc,
                                     OverloadCandidateSet::CSK_Normal);
-  AddFunctionCandidates(R.asUnresolvedSet(), Args, CandidateSet, true,
-                        TemplateArgs);
+  AddFunctionCandidates(R.asUnresolvedSet(), Args, CandidateSet, TemplateArgs,
+                        /*SuppressUserConversions=*/true);
 
   bool HadMultipleCandidates = (CandidateSet.size() > 1);
 
@@ -12440,15 +12491,11 @@ Expr *Sema::FixOverloadedFunctionReference(Expr *E, DeclAccessPair Found,
       type = Context.BoundMemberTy;
     }
 
-    MemberExpr *ME = MemberExpr::Create(Context, Base,
-                                        MemExpr->isArrow(),
-                                        MemExpr->getQualifierLoc(),
-                                        MemExpr->getTemplateKeywordLoc(),
-                                        Fn,
-                                        Found,
-                                        MemExpr->getMemberNameInfo(),
-                                        TemplateArgs,
-                                        type, valueKind, OK_Ordinary);
+    MemberExpr *ME = MemberExpr::Create(
+        Context, Base, MemExpr->isArrow(), MemExpr->getOperatorLoc(),
+        MemExpr->getQualifierLoc(), MemExpr->getTemplateKeywordLoc(), Fn, Found,
+        MemExpr->getMemberNameInfo(), TemplateArgs, type, valueKind,
+        OK_Ordinary);
     ME->setHadMultipleCandidates(true);
     MarkMemberReferenced(ME);
     return ME;
diff --git a/lib/Sema/SemaPseudoObject.cpp b/lib/Sema/SemaPseudoObject.cpp
index 5e92d5d07c51..3e465af9632c 100644
--- a/lib/Sema/SemaPseudoObject.cpp
+++ b/lib/Sema/SemaPseudoObject.cpp
@@ -1051,17 +1051,13 @@ Sema::ObjCSubscriptKind
   
   // Look for a conversion to an integral, enumeration type, or
   // objective-C pointer type.
-  std::pair<CXXRecordDecl::conversion_iterator,
-            CXXRecordDecl::conversion_iterator> Conversions
-    = cast<CXXRecordDecl>(RecordTy->getDecl())->getVisibleConversionFunctions();
-  
   int NoIntegrals=0, NoObjCIdPointers=0;
   SmallVector<CXXConversionDecl *, 4> ConversionDecls;
-    
-  for (CXXRecordDecl::conversion_iterator
-         I = Conversions.first, E = Conversions.second; I != E; ++I) {
-    if (CXXConversionDecl *Conversion
-        = dyn_cast<CXXConversionDecl>((*I)->getUnderlyingDecl())) {
+
+  for (NamedDecl *D : cast<CXXRecordDecl>(RecordTy->getDecl())
+                          ->getVisibleConversionFunctions()) {
+    if (CXXConversionDecl *Conversion =
+            dyn_cast<CXXConversionDecl>(D->getUnderlyingDecl())) {
       QualType CT = Conversion->getConversionType().getNonReferenceType();
       if (CT->isIntegralOrEnumerationType()) {
         ++NoIntegrals;
@@ -1196,7 +1192,7 @@ bool ObjCSubscriptOpBuilder::findAtIndexGetter() {
     AtIndexGetter = 
       S.LookupInstanceMethodInGlobalPool(AtIndexGetterSelector, 
                                          RefExpr->getSourceRange(), 
-                                         true, false);
+                                         true);
   }
   
   if (AtIndexGetter) {
@@ -1318,7 +1314,7 @@ bool ObjCSubscriptOpBuilder::findAtIndexSetter() {
     AtIndexSetter = 
       S.LookupInstanceMethodInGlobalPool(AtIndexSetterSelector, 
                                          RefExpr->getSourceRange(), 
-                                         true, false);
+                                         true);
   }
   
   bool err = false;
@@ -1446,7 +1442,7 @@ ExprResult MSPropertyOpBuilder::buildGet() {
   ExprResult GetterExpr = S.ActOnMemberAccessExpr(
     S.getCurScope(), RefExpr->getBaseExpr(), SourceLocation(),
     RefExpr->isArrow() ? tok::arrow : tok::period, SS, SourceLocation(),
-    GetterName, nullptr, true);
+    GetterName, nullptr);
   if (GetterExpr.isInvalid()) {
     S.Diag(RefExpr->getMemberLoc(),
            diag::error_cannot_find_suitable_accessor) << 0 /* getter */
@@ -1476,7 +1472,7 @@ ExprResult MSPropertyOpBuilder::buildSet(Expr *op, SourceLocation sl,
   ExprResult SetterExpr = S.ActOnMemberAccessExpr(
     S.getCurScope(), RefExpr->getBaseExpr(), SourceLocation(),
     RefExpr->isArrow() ? tok::arrow : tok::period, SS, SourceLocation(),
-    SetterName, nullptr, true);
+    SetterName, nullptr);
   if (SetterExpr.isInvalid()) {
     S.Diag(RefExpr->getMemberLoc(),
            diag::error_cannot_find_suitable_accessor) << 1 /* setter */
diff --git a/lib/Sema/SemaStmt.cpp b/lib/Sema/SemaStmt.cpp
index 0d1da4540eb3..5c72529caa8a 100644
--- a/lib/Sema/SemaStmt.cpp
+++ b/lib/Sema/SemaStmt.cpp
@@ -15,6 +15,7 @@
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/ASTDiagnostic.h"
 #include "clang/AST/CharUnits.h"
+#include "clang/AST/CXXInheritance.h"
 #include "clang/AST/DeclObjC.h"
 #include "clang/AST/EvaluatedExprVisitor.h"
 #include "clang/AST/ExprCXX.h"
@@ -23,12 +24,14 @@
 #include "clang/AST/StmtCXX.h"
 #include "clang/AST/StmtObjC.h"
 #include "clang/AST/TypeLoc.h"
+#include "clang/AST/TypeOrdering.h"
 #include "clang/Lex/Preprocessor.h"
 #include "clang/Sema/Initialization.h"
 #include "clang/Sema/Lookup.h"
 #include "clang/Sema/Scope.h"
 #include "clang/Sema/ScopeInfo.h"
 #include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallString.h"
@@ -236,7 +239,9 @@ void Sema::DiagnoseUnusedExprResult(const Stmt *S) {
     // is written in a macro body, only warn if it has the warn_unused_result
     // attribute.
     if (const Decl *FD = CE->getCalleeDecl()) {
-      if (FD->hasAttr<WarnUnusedResultAttr>()) {
+      const FunctionDecl *Func = dyn_cast<FunctionDecl>(FD);
+      if (Func ? Func->hasUnusedResultAttr()
+               : FD->hasAttr<WarnUnusedResultAttr>()) {
         Diag(Loc, diag::warn_unused_result) << R1 << R2;
         return;
       }
@@ -265,10 +270,6 @@ void Sema::DiagnoseUnusedExprResult(const Stmt *S) {
         Diag(Loc, diag::warn_unused_result) << R1 << R2;
         return;
       }
-      if (MD->isPropertyAccessor()) {
-        Diag(Loc, diag::warn_unused_property_expr);
-        return;
-      }
     }
   } else if (const PseudoObjectExpr *POE = dyn_cast<PseudoObjectExpr>(E)) {
     const Expr *Source = POE->getSyntacticForm();
@@ -687,26 +688,39 @@ static void checkCaseValue(Sema &S, SourceLocation Loc, const llvm::APSInt &Val,
   }
 }
 
+typedef SmallVector<std::pair<llvm::APSInt, EnumConstantDecl*>, 64> EnumValsTy;
+
 /// Returns true if we should emit a diagnostic about this case expression not
 /// being a part of the enum used in the switch controlling expression.
-static bool ShouldDiagnoseSwitchCaseNotInEnum(const ASTContext &Ctx,
+static bool ShouldDiagnoseSwitchCaseNotInEnum(const Sema &S,
                                               const EnumDecl *ED,
-                                              const Expr *CaseExpr) {
-  // Don't warn if the 'case' expression refers to a static const variable of
-  // the enum type.
-  CaseExpr = CaseExpr->IgnoreParenImpCasts();
-  if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(CaseExpr)) {
+                                              const Expr *CaseExpr,
+                                              EnumValsTy::iterator &EI,
+                                              EnumValsTy::iterator &EIEnd,
+                                              const llvm::APSInt &Val) {
+  bool FlagType = ED->hasAttr<FlagEnumAttr>();
+
+  if (const DeclRefExpr *DRE =
+          dyn_cast<DeclRefExpr>(CaseExpr->IgnoreParenImpCasts())) {
     if (const VarDecl *VD = dyn_cast<VarDecl>(DRE->getDecl())) {
-      if (!VD->hasGlobalStorage())
-        return true;
       QualType VarType = VD->getType();
-      if (!VarType.isConstQualified())
-        return true;
-      QualType EnumType = Ctx.getTypeDeclType(ED);
-      if (Ctx.hasSameUnqualifiedType(EnumType, VarType))
+      QualType EnumType = S.Context.getTypeDeclType(ED);
+      if (VD->hasGlobalStorage() && VarType.isConstQualified() &&
+          S.Context.hasSameUnqualifiedType(EnumType, VarType))
         return false;
     }
   }
+
+  if (FlagType) {
+    return !S.IsValueInFlagEnum(ED, Val, false);
+  } else {
+    while (EI != EIEnd && EI->first < Val)
+      EI++;
+
+    if (EI != EIEnd && EI->first == Val)
+      return false;
+  }
+
   return true;
 }
 
@@ -897,12 +911,12 @@ Sema::ActOnFinishSwitchStmt(SourceLocation SwitchLoc, Stmt *Switch,
           if (PrevString == CurrString)
             Diag(CaseVals[i].second->getLHS()->getLocStart(),
                  diag::err_duplicate_case) <<
-                 (PrevString.empty() ? CaseValStr.str() : PrevString);
+                 (PrevString.empty() ? StringRef(CaseValStr) : PrevString);
           else
             Diag(CaseVals[i].second->getLHS()->getLocStart(),
                  diag::err_duplicate_case_differing_expr) <<
-                 (PrevString.empty() ? CaseValStr.str() : PrevString) <<
-                 (CurrString.empty() ? CaseValStr.str() : CurrString) <<
+                 (PrevString.empty() ? StringRef(CaseValStr) : PrevString) <<
+                 (CurrString.empty() ? StringRef(CaseValStr) : CurrString) <<
                  CaseValStr;
 
           Diag(CaseVals[i-1].second->getLHS()->getLocStart(),
@@ -1046,8 +1060,6 @@ Sema::ActOnFinishSwitchStmt(SourceLocation SwitchLoc, Stmt *Switch,
     // If switch has default case, then ignore it.
     if (!CaseListIsErroneous  && !HasConstantCond && ET) {
       const EnumDecl *ED = ET->getDecl();
-      typedef SmallVector<std::pair<llvm::APSInt, EnumConstantDecl*>, 64>
-        EnumValsTy;
       EnumValsTy EnumVals;
 
       // Gather all enum values, set their type and sort them,
@@ -1058,57 +1070,48 @@ Sema::ActOnFinishSwitchStmt(SourceLocation SwitchLoc, Stmt *Switch,
         EnumVals.push_back(std::make_pair(Val, EDI));
       }
       std::stable_sort(EnumVals.begin(), EnumVals.end(), CmpEnumVals);
-      EnumValsTy::iterator EIend =
+      auto EI = EnumVals.begin(), EIEnd =
         std::unique(EnumVals.begin(), EnumVals.end(), EqEnumVals);
 
       // See which case values aren't in enum.
-      EnumValsTy::const_iterator EI = EnumVals.begin();
       for (CaseValsTy::const_iterator CI = CaseVals.begin();
-           CI != CaseVals.end(); CI++) {
-        while (EI != EIend && EI->first < CI->first)
-          EI++;
-        if (EI == EIend || EI->first > CI->first) {
-          Expr *CaseExpr = CI->second->getLHS();
-          if (ShouldDiagnoseSwitchCaseNotInEnum(Context, ED, CaseExpr))
-            Diag(CaseExpr->getExprLoc(), diag::warn_not_in_enum)
-              << CondTypeBeforePromotion;
-        }
+          CI != CaseVals.end(); CI++) {
+        Expr *CaseExpr = CI->second->getLHS();
+        if (ShouldDiagnoseSwitchCaseNotInEnum(*this, ED, CaseExpr, EI, EIEnd,
+                                              CI->first))
+          Diag(CaseExpr->getExprLoc(), diag::warn_not_in_enum)
+            << CondTypeBeforePromotion;
       }
+
       // See which of case ranges aren't in enum
       EI = EnumVals.begin();
       for (CaseRangesTy::const_iterator RI = CaseRanges.begin();
-           RI != CaseRanges.end() && EI != EIend; RI++) {
-        while (EI != EIend && EI->first < RI->first)
-          EI++;
-
-        if (EI == EIend || EI->first != RI->first) {
-          Expr *CaseExpr = RI->second->getLHS();
-          if (ShouldDiagnoseSwitchCaseNotInEnum(Context, ED, CaseExpr))
-            Diag(CaseExpr->getExprLoc(), diag::warn_not_in_enum)
-              << CondTypeBeforePromotion;
-        }
+          RI != CaseRanges.end(); RI++) {
+        Expr *CaseExpr = RI->second->getLHS();
+        if (ShouldDiagnoseSwitchCaseNotInEnum(*this, ED, CaseExpr, EI, EIEnd,
+                                              RI->first))
+          Diag(CaseExpr->getExprLoc(), diag::warn_not_in_enum)
+            << CondTypeBeforePromotion;
 
         llvm::APSInt Hi =
           RI->second->getRHS()->EvaluateKnownConstInt(Context);
         AdjustAPSInt(Hi, CondWidth, CondIsSigned);
-        while (EI != EIend && EI->first < Hi)
-          EI++;
-        if (EI == EIend || EI->first != Hi) {
-          Expr *CaseExpr = RI->second->getRHS();
-          if (ShouldDiagnoseSwitchCaseNotInEnum(Context, ED, CaseExpr))
-            Diag(CaseExpr->getExprLoc(), diag::warn_not_in_enum)
-              << CondTypeBeforePromotion;
-        }
+
+        CaseExpr = RI->second->getRHS();
+        if (ShouldDiagnoseSwitchCaseNotInEnum(*this, ED, CaseExpr, EI, EIEnd,
+                                              Hi))
+          Diag(CaseExpr->getExprLoc(), diag::warn_not_in_enum)
+            << CondTypeBeforePromotion;
       }
 
       // Check which enum vals aren't in switch
-      CaseValsTy::const_iterator CI = CaseVals.begin();
-      CaseRangesTy::const_iterator RI = CaseRanges.begin();
+      auto CI = CaseVals.begin();
+      auto RI = CaseRanges.begin();
       bool hasCasesNotInSwitch = false;
 
       SmallVector<DeclarationName,8> UnhandledNames;
 
-      for (EI = EnumVals.begin(); EI != EIend; EI++){
+      for (EI = EnumVals.begin(); EI != EIEnd; EI++){
         // Drop unneeded case values
         while (CI != CaseVals.end() && CI->first < EI->first)
           CI++;
@@ -1135,29 +1138,15 @@ Sema::ActOnFinishSwitchStmt(SourceLocation SwitchLoc, Stmt *Switch,
         Diag(TheDefaultStmt->getDefaultLoc(), diag::warn_unreachable_default);
 
       // Produce a nice diagnostic if multiple values aren't handled.
-      switch (UnhandledNames.size()) {
-      case 0: break;
-      case 1:
-        Diag(CondExpr->getExprLoc(), TheDefaultStmt
-          ? diag::warn_def_missing_case1 : diag::warn_missing_case1)
-          << UnhandledNames[0];
-        break;
-      case 2:
-        Diag(CondExpr->getExprLoc(), TheDefaultStmt
-          ? diag::warn_def_missing_case2 : diag::warn_missing_case2)
-          << UnhandledNames[0] << UnhandledNames[1];
-        break;
-      case 3:
-        Diag(CondExpr->getExprLoc(), TheDefaultStmt
-          ? diag::warn_def_missing_case3 : diag::warn_missing_case3)
-          << UnhandledNames[0] << UnhandledNames[1] << UnhandledNames[2];
-        break;
-      default:
-        Diag(CondExpr->getExprLoc(), TheDefaultStmt
-          ? diag::warn_def_missing_cases : diag::warn_missing_cases)
-          << (unsigned)UnhandledNames.size()
-          << UnhandledNames[0] << UnhandledNames[1] << UnhandledNames[2];
-        break;
+      if (!UnhandledNames.empty()) {
+        DiagnosticBuilder DB = Diag(CondExpr->getExprLoc(),
+                                    TheDefaultStmt ? diag::warn_def_missing_case
+                                                   : diag::warn_missing_case)
+                               << (int)UnhandledNames.size();
+
+        for (size_t I = 0, E = std::min(UnhandledNames.size(), (size_t)3);
+             I != E; ++I)
+          DB << UnhandledNames[I];
       }
 
       if (!hasCasesNotInSwitch)
@@ -1195,30 +1184,37 @@ Sema::DiagnoseAssignmentEnum(QualType DstType, QualType SrcType,
         llvm::APSInt RhsVal = SrcExpr->EvaluateKnownConstInt(Context);
         AdjustAPSInt(RhsVal, DstWidth, DstIsSigned);
         const EnumDecl *ED = ET->getDecl();
-        typedef SmallVector<std::pair<llvm::APSInt, EnumConstantDecl *>, 64>
-            EnumValsTy;
-        EnumValsTy EnumVals;
-
-        // Gather all enum values, set their type and sort them,
-        // allowing easier comparison with rhs constant.
-        for (auto *EDI : ED->enumerators()) {
-          llvm::APSInt Val = EDI->getInitVal();
-          AdjustAPSInt(Val, DstWidth, DstIsSigned);
-          EnumVals.push_back(std::make_pair(Val, EDI));
-        }
-        if (EnumVals.empty())
-          return;
-        std::stable_sort(EnumVals.begin(), EnumVals.end(), CmpEnumVals);
-        EnumValsTy::iterator EIend =
-            std::unique(EnumVals.begin(), EnumVals.end(), EqEnumVals);
-
-        // See which values aren't in the enum.
-        EnumValsTy::const_iterator EI = EnumVals.begin();
-        while (EI != EIend && EI->first < RhsVal)
-          EI++;
-        if (EI == EIend || EI->first != RhsVal) {
-          Diag(SrcExpr->getExprLoc(), diag::warn_not_in_enum_assignment)
+
+        if (ED->hasAttr<FlagEnumAttr>()) {
+          if (!IsValueInFlagEnum(ED, RhsVal, true))
+            Diag(SrcExpr->getExprLoc(), diag::warn_not_in_enum_assignment)
               << DstType.getUnqualifiedType();
+        } else {
+          typedef SmallVector<std::pair<llvm::APSInt, EnumConstantDecl *>, 64>
+              EnumValsTy;
+          EnumValsTy EnumVals;
+
+          // Gather all enum values, set their type and sort them,
+          // allowing easier comparison with rhs constant.
+          for (auto *EDI : ED->enumerators()) {
+            llvm::APSInt Val = EDI->getInitVal();
+            AdjustAPSInt(Val, DstWidth, DstIsSigned);
+            EnumVals.push_back(std::make_pair(Val, EDI));
+          }
+          if (EnumVals.empty())
+            return;
+          std::stable_sort(EnumVals.begin(), EnumVals.end(), CmpEnumVals);
+          EnumValsTy::iterator EIend =
+              std::unique(EnumVals.begin(), EnumVals.end(), EqEnumVals);
+
+          // See which values aren't in the enum.
+          EnumValsTy::const_iterator EI = EnumVals.begin();
+          while (EI != EIend && EI->first < RhsVal)
+            EI++;
+          if (EI == EIend || EI->first != RhsVal) {
+            Diag(SrcExpr->getExprLoc(), diag::warn_not_in_enum_assignment)
+                << DstType.getUnqualifiedType();
+          }
         }
       }
     }
@@ -1832,6 +1828,15 @@ Sema::ActOnObjCForCollectionStmt(SourceLocation ForLoc,
 /// \return true if an error occurs.
 static bool FinishForRangeVarDecl(Sema &SemaRef, VarDecl *Decl, Expr *Init,
                                   SourceLocation Loc, int DiagID) {
+  if (Decl->getType()->isUndeducedType()) {
+    ExprResult Res = SemaRef.CorrectDelayedTyposInExpr(Init);
+    if (!Res.isUsable()) {
+      Decl->setInvalidDecl();
+      return true;
+    }
+    Init = Res.get();
+  }
+
   // Deduce the type for the iterator variable now rather than leaving it to
   // AddInitializerToDecl, so we can produce a more suitable diagnostic.
   QualType InitType;
@@ -2368,6 +2373,156 @@ StmtResult Sema::FinishObjCForCollectionStmt(Stmt *S, Stmt *B) {
   return S;
 }
 
+// Warn when the loop variable is a const reference that creates a copy.
+// Suggest using the non-reference type for copies.  If a copy can be prevented
+// suggest the const reference type that would do so.
+// For instance, given "for (const &Foo : Range)", suggest
+// "for (const Foo : Range)" to denote a copy is made for the loop.  If
+// possible, also suggest "for (const &Bar : Range)" if this type prevents
+// the copy altogether.
+static void DiagnoseForRangeReferenceVariableCopies(Sema &SemaRef,
+                                                    const VarDecl *VD,
+                                                    QualType RangeInitType) {
+  const Expr *InitExpr = VD->getInit();
+  if (!InitExpr)
+    return;
+
+  QualType VariableType = VD->getType();
+
+  const MaterializeTemporaryExpr *MTE =
+      dyn_cast<MaterializeTemporaryExpr>(InitExpr);
+
+  // No copy made.
+  if (!MTE)
+    return;
+
+  const Expr *E = MTE->GetTemporaryExpr()->IgnoreImpCasts();
+
+  // Searching for either UnaryOperator for dereference of a pointer or
+  // CXXOperatorCallExpr for handling iterators.
+  while (!isa<CXXOperatorCallExpr>(E) && !isa<UnaryOperator>(E)) {
+    if (const CXXConstructExpr *CCE = dyn_cast<CXXConstructExpr>(E)) {
+      E = CCE->getArg(0);
+    } else if (const CXXMemberCallExpr *Call = dyn_cast<CXXMemberCallExpr>(E)) {
+      const MemberExpr *ME = cast<MemberExpr>(Call->getCallee());
+      E = ME->getBase();
+    } else {
+      const MaterializeTemporaryExpr *MTE = cast<MaterializeTemporaryExpr>(E);
+      E = MTE->GetTemporaryExpr();
+    }
+    E = E->IgnoreImpCasts();
+  }
+
+  bool ReturnsReference = false;
+  if (isa<UnaryOperator>(E)) {
+    ReturnsReference = true;
+  } else {
+    const CXXOperatorCallExpr *Call = cast<CXXOperatorCallExpr>(E);
+    const FunctionDecl *FD = Call->getDirectCallee();
+    QualType ReturnType = FD->getReturnType();
+    ReturnsReference = ReturnType->isReferenceType();
+  }
+
+  if (ReturnsReference) {
+    // Loop variable creates a temporary.  Suggest either to go with
+    // non-reference loop variable to indiciate a copy is made, or
+    // the correct time to bind a const reference.
+    SemaRef.Diag(VD->getLocation(), diag::warn_for_range_const_reference_copy)
+        << VD << VariableType << E->getType();
+    QualType NonReferenceType = VariableType.getNonReferenceType();
+    NonReferenceType.removeLocalConst();
+    QualType NewReferenceType =
+        SemaRef.Context.getLValueReferenceType(E->getType().withConst());
+    SemaRef.Diag(VD->getLocStart(), diag::note_use_type_or_non_reference)
+        << NonReferenceType << NewReferenceType << VD->getSourceRange();
+  } else {
+    // The range always returns a copy, so a temporary is always created.
+    // Suggest removing the reference from the loop variable.
+    SemaRef.Diag(VD->getLocation(), diag::warn_for_range_variable_always_copy)
+        << VD << RangeInitType;
+    QualType NonReferenceType = VariableType.getNonReferenceType();
+    NonReferenceType.removeLocalConst();
+    SemaRef.Diag(VD->getLocStart(), diag::note_use_non_reference_type)
+        << NonReferenceType << VD->getSourceRange();
+  }
+}
+
+// Warns when the loop variable can be changed to a reference type to
+// prevent a copy.  For instance, if given "for (const Foo x : Range)" suggest
+// "for (const Foo &x : Range)" if this form does not make a copy.
+static void DiagnoseForRangeConstVariableCopies(Sema &SemaRef,
+                                                const VarDecl *VD) {
+  const Expr *InitExpr = VD->getInit();
+  if (!InitExpr)
+    return;
+
+  QualType VariableType = VD->getType();
+
+  if (const CXXConstructExpr *CE = dyn_cast<CXXConstructExpr>(InitExpr)) {
+    if (!CE->getConstructor()->isCopyConstructor())
+      return;
+  } else if (const CastExpr *CE = dyn_cast<CastExpr>(InitExpr)) {
+    if (CE->getCastKind() != CK_LValueToRValue)
+      return;
+  } else {
+    return;
+  }
+
+  // TODO: Determine a maximum size that a POD type can be before a diagnostic
+  // should be emitted.  Also, only ignore POD types with trivial copy
+  // constructors.
+  if (VariableType.isPODType(SemaRef.Context))
+    return;
+
+  // Suggest changing from a const variable to a const reference variable
+  // if doing so will prevent a copy.
+  SemaRef.Diag(VD->getLocation(), diag::warn_for_range_copy)
+      << VD << VariableType << InitExpr->getType();
+  SemaRef.Diag(VD->getLocStart(), diag::note_use_reference_type)
+      << SemaRef.Context.getLValueReferenceType(VariableType)
+      << VD->getSourceRange();
+}
+
+/// DiagnoseForRangeVariableCopies - Diagnose three cases and fixes for them.
+/// 1) for (const foo &x : foos) where foos only returns a copy.  Suggest
+///    using "const foo x" to show that a copy is made
+/// 2) for (const bar &x : foos) where bar is a temporary intialized by bar.
+///    Suggest either "const bar x" to keep the copying or "const foo& x" to
+///    prevent the copy.
+/// 3) for (const foo x : foos) where x is constructed from a reference foo.
+///    Suggest "const foo &x" to prevent the copy.
+static void DiagnoseForRangeVariableCopies(Sema &SemaRef,
+                                           const CXXForRangeStmt *ForStmt) {
+  if (SemaRef.Diags.isIgnored(diag::warn_for_range_const_reference_copy,
+                              ForStmt->getLocStart()) &&
+      SemaRef.Diags.isIgnored(diag::warn_for_range_variable_always_copy,
+                              ForStmt->getLocStart()) &&
+      SemaRef.Diags.isIgnored(diag::warn_for_range_copy,
+                              ForStmt->getLocStart())) {
+    return;
+  }
+
+  const VarDecl *VD = ForStmt->getLoopVariable();
+  if (!VD)
+    return;
+
+  QualType VariableType = VD->getType();
+
+  if (VariableType->isIncompleteType())
+    return;
+
+  const Expr *InitExpr = VD->getInit();
+  if (!InitExpr)
+    return;
+
+  if (VariableType->isReferenceType()) {
+    DiagnoseForRangeReferenceVariableCopies(SemaRef, VD,
+                                            ForStmt->getRangeInit()->getType());
+  } else if (VariableType.isConstQualified()) {
+    DiagnoseForRangeConstVariableCopies(SemaRef, VD);
+  }
+}
+
 /// FinishCXXForRangeStmt - Attach the body to a C++0x for-range statement.
 /// This is a separate step from ActOnCXXForRangeStmt because analysis of the
 /// body cannot be performed until after the type of the range variable is
@@ -2385,6 +2540,8 @@ StmtResult Sema::FinishCXXForRangeStmt(Stmt *S, Stmt *B) {
   DiagnoseEmptyStmtBody(ForStmt->getRParenLoc(), B,
                         diag::warn_empty_range_based_for_body);
 
+  DiagnoseForRangeVariableCopies(*this, ForStmt);
+
   return S;
 }
 
@@ -2423,6 +2580,14 @@ Sema::ActOnIndirectGotoStmt(SourceLocation GotoLoc, SourceLocation StarLoc,
   return new (Context) IndirectGotoStmt(GotoLoc, StarLoc, E);
 }
 
+static void CheckJumpOutOfSEHFinally(Sema &S, SourceLocation Loc,
+                                     const Scope &DestScope) {
+  if (!S.CurrentSEHFinally.empty() &&
+      DestScope.Contains(*S.CurrentSEHFinally.back())) {
+    S.Diag(Loc, diag::warn_jump_out_of_seh_finally);
+  }
+}
+
 StmtResult
 Sema::ActOnContinueStmt(SourceLocation ContinueLoc, Scope *CurScope) {
   Scope *S = CurScope->getContinueParent();
@@ -2430,6 +2595,7 @@ Sema::ActOnContinueStmt(SourceLocation ContinueLoc, Scope *CurScope) {
     // C99 6.8.6.2p1: A break shall appear only in or as a loop body.
     return StmtError(Diag(ContinueLoc, diag::err_continue_not_in_loop));
   }
+  CheckJumpOutOfSEHFinally(*this, ContinueLoc, *S);
 
   return new (Context) ContinueStmt(ContinueLoc);
 }
@@ -2444,6 +2610,7 @@ Sema::ActOnBreakStmt(SourceLocation BreakLoc, Scope *CurScope) {
   if (S->isOpenMPLoopScope())
     return StmtError(Diag(BreakLoc, diag::err_omp_loop_cannot_use_stmt)
                      << "break");
+  CheckJumpOutOfSEHFinally(*this, BreakLoc, *S);
 
   return new (Context) BreakStmt(BreakLoc);
 }
@@ -2903,6 +3070,8 @@ Sema::ActOnReturnStmt(SourceLocation ReturnLoc, Expr *RetValExp,
     CurScope->setNoNRVO();
   }
 
+  CheckJumpOutOfSEHFinally(*this, ReturnLoc, *CurScope->getFnParent());
+
   return R;
 }
 
@@ -3179,7 +3348,7 @@ Sema::ActOnObjCAtThrowStmt(SourceLocation AtLoc, Expr *Throw,
     Diag(AtLoc, diag::err_objc_exceptions_disabled) << "@throw";
 
   if (!Throw) {
-    // @throw without an expression designates a rethrow (which much occur
+    // @throw without an expression designates a rethrow (which must occur
     // in the context of an @catch clause).
     Scope *AtCatchParent = CurScope;
     while (AtCatchParent && !AtCatchParent->isAtCatchScope())
@@ -3251,35 +3420,112 @@ Sema::ActOnObjCAutoreleasePoolStmt(SourceLocation AtLoc, Stmt *Body) {
 }
 
 namespace {
+class CatchHandlerType {
+  QualType QT;
+  unsigned IsPointer : 1;
 
-class TypeWithHandler {
-  QualType t;
-  CXXCatchStmt *stmt;
-public:
-  TypeWithHandler(const QualType &type, CXXCatchStmt *statement)
-  : t(type), stmt(statement) {}
+  // This is a special constructor to be used only with DenseMapInfo's
+  // getEmptyKey() and getTombstoneKey() functions.
+  friend struct llvm::DenseMapInfo<CatchHandlerType>;
+  enum Unique { ForDenseMap };
+  CatchHandlerType(QualType QT, Unique) : QT(QT), IsPointer(false) {}
 
-  // An arbitrary order is fine as long as it places identical
-  // types next to each other.
-  bool operator<(const TypeWithHandler &y) const {
-    if (t.getAsOpaquePtr() < y.t.getAsOpaquePtr())
-      return true;
-    if (t.getAsOpaquePtr() > y.t.getAsOpaquePtr())
+public:
+  /// Used when creating a CatchHandlerType from a handler type; will determine
+  /// whether the type is a pointer or reference and will strip off the the top
+  /// level pointer and cv-qualifiers.
+  CatchHandlerType(QualType Q) : QT(Q), IsPointer(false) {
+    if (QT->isPointerType())
+      IsPointer = true;
+
+    if (IsPointer || QT->isReferenceType())
+      QT = QT->getPointeeType();
+    QT = QT.getUnqualifiedType();
+  }
+
+  /// Used when creating a CatchHandlerType from a base class type; pretends the
+  /// type passed in had the pointer qualifier, does not need to get an
+  /// unqualified type.
+  CatchHandlerType(QualType QT, bool IsPointer)
+      : QT(QT), IsPointer(IsPointer) {}
+
+  QualType underlying() const { return QT; }
+  bool isPointer() const { return IsPointer; }
+
+  friend bool operator==(const CatchHandlerType &LHS,
+                         const CatchHandlerType &RHS) {
+    // If the pointer qualification does not match, we can return early.
+    if (LHS.IsPointer != RHS.IsPointer)
       return false;
-    else
-      return getTypeSpecStartLoc() < y.getTypeSpecStartLoc();
+    // Otherwise, check the underlying type without cv-qualifiers.
+    return LHS.QT == RHS.QT;
+  }
+};
+} // namespace
+
+namespace llvm {
+template <> struct DenseMapInfo<CatchHandlerType> {
+  static CatchHandlerType getEmptyKey() {
+    return CatchHandlerType(DenseMapInfo<QualType>::getEmptyKey(),
+                       CatchHandlerType::ForDenseMap);
   }
 
-  bool operator==(const TypeWithHandler& other) const {
-    return t == other.t;
+  static CatchHandlerType getTombstoneKey() {
+    return CatchHandlerType(DenseMapInfo<QualType>::getTombstoneKey(),
+                       CatchHandlerType::ForDenseMap);
   }
 
-  CXXCatchStmt *getCatchStmt() const { return stmt; }
-  SourceLocation getTypeSpecStartLoc() const {
-    return stmt->getExceptionDecl()->getTypeSpecStartLoc();
+  static unsigned getHashValue(const CatchHandlerType &Base) {
+    return DenseMapInfo<QualType>::getHashValue(Base.underlying());
   }
+
+  static bool isEqual(const CatchHandlerType &LHS,
+                      const CatchHandlerType &RHS) {
+    return LHS == RHS;
+  }
+};
+
+// It's OK to treat CatchHandlerType as a POD type.
+template <> struct isPodLike<CatchHandlerType> {
+  static const bool value = true;
 };
+}
+
+namespace {
+class CatchTypePublicBases {
+  ASTContext &Ctx;
+  const llvm::DenseMap<CatchHandlerType, CXXCatchStmt *> &TypesToCheck;
+  const bool CheckAgainstPointer;
+
+  CXXCatchStmt *FoundHandler;
+  CanQualType FoundHandlerType;
 
+public:
+  CatchTypePublicBases(
+      ASTContext &Ctx,
+      const llvm::DenseMap<CatchHandlerType, CXXCatchStmt *> &T, bool C)
+      : Ctx(Ctx), TypesToCheck(T), CheckAgainstPointer(C),
+        FoundHandler(nullptr) {}
+
+  CXXCatchStmt *getFoundHandler() const { return FoundHandler; }
+  CanQualType getFoundHandlerType() const { return FoundHandlerType; }
+
+  static bool FindPublicBasesOfType(const CXXBaseSpecifier *S, CXXBasePath &,
+                                    void *User) {
+    auto &PBOT = *reinterpret_cast<CatchTypePublicBases *>(User);
+    if (S->getAccessSpecifier() == AccessSpecifier::AS_public) {
+      CatchHandlerType Check(S->getType(), PBOT.CheckAgainstPointer);
+      auto M = PBOT.TypesToCheck;
+      auto I = M.find(Check);
+      if (I != M.end()) {
+        PBOT.FoundHandler = I->second;
+        PBOT.FoundHandlerType = PBOT.Ctx.getCanonicalType(S->getType());
+        return true;
+      }
+    }
+    return false;
+  }
+};
 }
 
 /// ActOnCXXTryBlock - Takes a try compound-statement and a number of
@@ -3289,74 +3535,120 @@ StmtResult Sema::ActOnCXXTryBlock(SourceLocation TryLoc, Stmt *TryBlock,
   // Don't report an error if 'try' is used in system headers.
   if (!getLangOpts().CXXExceptions &&
       !getSourceManager().isInSystemHeader(TryLoc))
-      Diag(TryLoc, diag::err_exceptions_disabled) << "try";
+    Diag(TryLoc, diag::err_exceptions_disabled) << "try";
 
   if (getCurScope() && getCurScope()->isOpenMPSimdDirectiveScope())
     Diag(TryLoc, diag::err_omp_simd_region_cannot_use_stmt) << "try";
 
+  sema::FunctionScopeInfo *FSI = getCurFunction();
+
+  // C++ try is incompatible with SEH __try.
+  if (!getLangOpts().Borland && FSI->FirstSEHTryLoc.isValid()) {
+    Diag(TryLoc, diag::err_mixing_cxx_try_seh_try);
+    Diag(FSI->FirstSEHTryLoc, diag::note_conflicting_try_here) << "'__try'";
+  }
+
   const unsigned NumHandlers = Handlers.size();
-  assert(NumHandlers > 0 &&
+  assert(!Handlers.empty() &&
          "The parser shouldn't call this if there are no handlers.");
 
-  SmallVector<TypeWithHandler, 8> TypesWithHandlers;
-
+  llvm::DenseMap<CatchHandlerType, CXXCatchStmt *> HandledTypes;
   for (unsigned i = 0; i < NumHandlers; ++i) {
-    CXXCatchStmt *Handler = cast<CXXCatchStmt>(Handlers[i]);
-    if (!Handler->getExceptionDecl()) {
-      if (i < NumHandlers - 1)
-        return StmtError(Diag(Handler->getLocStart(),
-                              diag::err_early_catch_all));
+    CXXCatchStmt *H = cast<CXXCatchStmt>(Handlers[i]);
 
+    // Diagnose when the handler is a catch-all handler, but it isn't the last
+    // handler for the try block. [except.handle]p5. Also, skip exception
+    // declarations that are invalid, since we can't usefully report on them.
+    if (!H->getExceptionDecl()) {
+      if (i < NumHandlers - 1)
+        return StmtError(Diag(H->getLocStart(), diag::err_early_catch_all));
+      continue;
+    } else if (H->getExceptionDecl()->isInvalidDecl())
       continue;
-    }
-
-    const QualType CaughtType = Handler->getCaughtType();
-    const QualType CanonicalCaughtType = Context.getCanonicalType(CaughtType);
-    TypesWithHandlers.push_back(TypeWithHandler(CanonicalCaughtType, Handler));
-  }
-
-  // Detect handlers for the same type as an earlier one.
-  if (NumHandlers > 1) {
-    llvm::array_pod_sort(TypesWithHandlers.begin(), TypesWithHandlers.end());
-
-    TypeWithHandler prev = TypesWithHandlers[0];
-    for (unsigned i = 1; i < TypesWithHandlers.size(); ++i) {
-      TypeWithHandler curr = TypesWithHandlers[i];
 
-      if (curr == prev) {
-        Diag(curr.getTypeSpecStartLoc(),
-             diag::warn_exception_caught_by_earlier_handler)
-          << curr.getCatchStmt()->getCaughtType().getAsString();
-        Diag(prev.getTypeSpecStartLoc(),
-             diag::note_previous_exception_handler)
-          << prev.getCatchStmt()->getCaughtType().getAsString();
+    // Walk the type hierarchy to diagnose when this type has already been
+    // handled (duplication), or cannot be handled (derivation inversion). We
+    // ignore top-level cv-qualifiers, per [except.handle]p3
+    CatchHandlerType HandlerCHT =
+        (QualType)Context.getCanonicalType(H->getCaughtType());
+
+    // We can ignore whether the type is a reference or a pointer; we need the
+    // underlying declaration type in order to get at the underlying record
+    // decl, if there is one.
+    QualType Underlying = HandlerCHT.underlying();
+    if (auto *RD = Underlying->getAsCXXRecordDecl()) {
+      if (!RD->hasDefinition())
+        continue;
+      // Check that none of the public, unambiguous base classes are in the
+      // map ([except.handle]p1). Give the base classes the same pointer
+      // qualification as the original type we are basing off of. This allows
+      // comparison against the handler type using the same top-level pointer
+      // as the original type.
+      CXXBasePaths Paths;
+      Paths.setOrigin(RD);
+      CatchTypePublicBases CTPB(Context, HandledTypes, HandlerCHT.isPointer());
+      if (RD->lookupInBases(CatchTypePublicBases::FindPublicBasesOfType, &CTPB,
+                            Paths)) {
+        const CXXCatchStmt *Problem = CTPB.getFoundHandler();
+        if (!Paths.isAmbiguous(CTPB.getFoundHandlerType())) {
+          Diag(H->getExceptionDecl()->getTypeSpecStartLoc(),
+               diag::warn_exception_caught_by_earlier_handler)
+              << H->getCaughtType();
+          Diag(Problem->getExceptionDecl()->getTypeSpecStartLoc(),
+                diag::note_previous_exception_handler)
+              << Problem->getCaughtType();
+        }
       }
+    }
 
-      prev = curr;
+    // Add the type the list of ones we have handled; diagnose if we've already
+    // handled it.
+    auto R = HandledTypes.insert(std::make_pair(H->getCaughtType(), H));
+    if (!R.second) {
+      const CXXCatchStmt *Problem = R.first->second;
+      Diag(H->getExceptionDecl()->getTypeSpecStartLoc(),
+           diag::warn_exception_caught_by_earlier_handler)
+          << H->getCaughtType();
+      Diag(Problem->getExceptionDecl()->getTypeSpecStartLoc(),
+           diag::note_previous_exception_handler)
+          << Problem->getCaughtType();
     }
   }
 
-  getCurFunction()->setHasBranchProtectedScope();
-
-  // FIXME: We should detect handlers that cannot catch anything because an
-  // earlier handler catches a superclass. Need to find a method that is not
-  // quadratic for this.
-  // Neither of these are explicitly forbidden, but every compiler detects them
-  // and warns.
+  FSI->setHasCXXTry(TryLoc);
 
   return CXXTryStmt::Create(Context, TryLoc, TryBlock, Handlers);
 }
 
-StmtResult
-Sema::ActOnSEHTryBlock(bool IsCXXTry,
-                       SourceLocation TryLoc,
-                       Stmt *TryBlock,
-                       Stmt *Handler) {
+StmtResult Sema::ActOnSEHTryBlock(bool IsCXXTry, SourceLocation TryLoc,
+                                  Stmt *TryBlock, Stmt *Handler) {
   assert(TryBlock && Handler);
 
-  getCurFunction()->setHasBranchProtectedScope();
+  sema::FunctionScopeInfo *FSI = getCurFunction();
+
+  // SEH __try is incompatible with C++ try. Borland appears to support this,
+  // however.
+  if (!getLangOpts().Borland) {
+    if (FSI->FirstCXXTryLoc.isValid()) {
+      Diag(TryLoc, diag::err_mixing_cxx_try_seh_try);
+      Diag(FSI->FirstCXXTryLoc, diag::note_conflicting_try_here) << "'try'";
+    }
+  }
+
+  FSI->setHasSEHTry(TryLoc);
 
-  return SEHTryStmt::Create(Context,IsCXXTry,TryLoc,TryBlock,Handler);
+  // Reject __try in Obj-C methods, blocks, and captured decls, since we don't
+  // track if they use SEH.
+  DeclContext *DC = CurContext;
+  while (DC && !DC->isFunctionOrMethod())
+    DC = DC->getParent();
+  FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(DC);
+  if (FD)
+    FD->setUsesSEHTry(true);
+  else
+    Diag(TryLoc, diag::err_seh_try_outside_functions);
+
+  return SEHTryStmt::Create(Context, IsCXXTry, TryLoc, TryBlock, Handler);
 }
 
 StmtResult
@@ -3374,11 +3666,18 @@ Sema::ActOnSEHExceptBlock(SourceLocation Loc,
   return SEHExceptStmt::Create(Context,Loc,FilterExpr,Block);
 }
 
-StmtResult
-Sema::ActOnSEHFinallyBlock(SourceLocation Loc,
-                           Stmt *Block) {
+void Sema::ActOnStartSEHFinallyBlock() {
+  CurrentSEHFinally.push_back(CurScope);
+}
+
+void Sema::ActOnAbortSEHFinallyBlock() {
+  CurrentSEHFinally.pop_back();
+}
+
+StmtResult Sema::ActOnFinishSEHFinallyBlock(SourceLocation Loc, Stmt *Block) {
   assert(Block);
-  return SEHFinallyStmt::Create(Context,Loc,Block);
+  CurrentSEHFinally.pop_back();
+  return SEHFinallyStmt::Create(Context, Loc, Block);
 }
 
 StmtResult
@@ -3388,6 +3687,7 @@ Sema::ActOnSEHLeaveStmt(SourceLocation Loc, Scope *CurScope) {
     SEHTryParent = SEHTryParent->getParent();
   if (!SEHTryParent)
     return StmtError(Diag(Loc, diag::err_ms___leave_not_in___try));
+  CheckJumpOutOfSEHFinally(*this, Loc, *SEHTryParent);
 
   return new (Context) SEHLeaveStmt(Loc);
 }
diff --git a/lib/Sema/SemaStmtAsm.cpp b/lib/Sema/SemaStmtAsm.cpp
index 0d32581e8daa..9f48616ea5bf 100644
--- a/lib/Sema/SemaStmtAsm.cpp
+++ b/lib/Sema/SemaStmtAsm.cpp
@@ -12,6 +12,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "clang/Sema/SemaInternal.h"
+#include "clang/AST/ExprCXX.h"
 #include "clang/AST/RecordLayout.h"
 #include "clang/AST/TypeLoc.h"
 #include "clang/Basic/TargetInfo.h"
@@ -86,6 +87,11 @@ static bool CheckNakedParmReference(Expr *E, Sema &S) {
   WorkList.push_back(E);
   while (WorkList.size()) {
     Expr *E = WorkList.pop_back_val();
+    if (isa<CXXThisExpr>(E)) {
+      S.Diag(E->getLocStart(), diag::err_asm_naked_this_ref);
+      S.Diag(Func->getAttr<NakedAttr>()->getLocation(), diag::note_attribute);
+      return true;
+    }
     if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) {
       if (isa<ParmVarDecl>(DRE->getDecl())) {
         S.Diag(DRE->getLocStart(), diag::err_asm_naked_parm_ref);
@@ -118,6 +124,9 @@ StmtResult Sema::ActOnGCCAsmStmt(SourceLocation AsmLoc, bool IsSimple,
   // The parser verifies that there is a string literal here.
   assert(AsmString->isAscii());
 
+  bool ValidateConstraints =
+      DeclAttrsMatchCUDAMode(getLangOpts(), getCurFunctionDecl());
+
   for (unsigned i = 0; i != NumOutputs; i++) {
     StringLiteral *Literal = Constraints[i];
     assert(Literal->isAscii());
@@ -127,7 +136,8 @@ StmtResult Sema::ActOnGCCAsmStmt(SourceLocation AsmLoc, bool IsSimple,
       OutputName = Names[i]->getName();
 
     TargetInfo::ConstraintInfo Info(Literal->getString(), OutputName);
-    if (!Context.getTargetInfo().validateOutputConstraint(Info))
+    if (ValidateConstraints &&
+        !Context.getTargetInfo().validateOutputConstraint(Info))
       return StmtError(Diag(Literal->getLocStart(),
                             diag::err_asm_invalid_output_constraint)
                        << Info.getConstraintStr());
@@ -201,8 +211,9 @@ StmtResult Sema::ActOnGCCAsmStmt(SourceLocation AsmLoc, bool IsSimple,
       InputName = Names[i]->getName();
 
     TargetInfo::ConstraintInfo Info(Literal->getString(), InputName);
-    if (!Context.getTargetInfo().validateInputConstraint(OutputConstraintInfos.data(),
-                                                NumOutputs, Info)) {
+    if (ValidateConstraints &&
+        !Context.getTargetInfo().validateInputConstraint(
+            OutputConstraintInfos.data(), NumOutputs, Info)) {
       return StmtError(Diag(Literal->getLocStart(),
                             diag::err_asm_invalid_input_constraint)
                        << Info.getConstraintStr());
@@ -307,32 +318,22 @@ StmtResult Sema::ActOnGCCAsmStmt(SourceLocation AsmLoc, bool IsSimple,
     if (!Piece.isOperand()) continue;
 
     // Look for the correct constraint index.
-    unsigned Idx = 0;
-    unsigned ConstraintIdx = 0;
-    for (unsigned i = 0, e = NS->getNumOutputs(); i != e; ++i, ++ConstraintIdx) {
-      TargetInfo::ConstraintInfo &Info = OutputConstraintInfos[i];
-      if (Idx == Piece.getOperandNo())
-        break;
-      ++Idx;
-
-      if (Info.isReadWrite()) {
-        if (Idx == Piece.getOperandNo())
-          break;
-        ++Idx;
-      }
-    }
+    unsigned ConstraintIdx = Piece.getOperandNo();
+    unsigned NumOperands = NS->getNumOutputs() + NS->getNumInputs();
 
-    for (unsigned i = 0, e = NS->getNumInputs(); i != e; ++i, ++ConstraintIdx) {
-      TargetInfo::ConstraintInfo &Info = InputConstraintInfos[i];
-      if (Idx == Piece.getOperandNo())
-        break;
-      ++Idx;
+    // Look for the (ConstraintIdx - NumOperands + 1)th constraint with
+    // modifier '+'.
+    if (ConstraintIdx >= NumOperands) {
+      unsigned I = 0, E = NS->getNumOutputs();
 
-      if (Info.isReadWrite()) {
-        if (Idx == Piece.getOperandNo())
+      for (unsigned Cnt = ConstraintIdx - NumOperands; I != E; ++I)
+        if (OutputConstraintInfos[I].isReadWrite() && Cnt-- == 0) {
+          ConstraintIdx = I;
           break;
-        ++Idx;
-      }
+        }
+
+      assert(I != E && "Invalid operand number should have been caught in "
+                       " AnalyzeAsmString");
     }
 
     // Now that we have the right indexes go ahead and check.
diff --git a/lib/Sema/SemaTemplate.cpp b/lib/Sema/SemaTemplate.cpp
index 1e7176222b57..37eeee2f8867 100644
--- a/lib/Sema/SemaTemplate.cpp
+++ b/lib/Sema/SemaTemplate.cpp
@@ -836,7 +836,8 @@ Sema::CheckClassTemplate(Scope *S, unsigned TagSpec, TagUseKind TUK,
                          AccessSpecifier AS, SourceLocation ModulePrivateLoc,
                          SourceLocation FriendLoc,
                          unsigned NumOuterTemplateParamLists,
-                         TemplateParameterList** OuterTemplateParamLists) {
+                         TemplateParameterList** OuterTemplateParamLists,
+                         SkipBodyInfo *SkipBody) {
   assert(TemplateParams && TemplateParams->size() > 0 &&
          "No template parameters");
   assert(TUK != TUK_Reference && "Can only declare or define class templates");
@@ -993,6 +994,19 @@ Sema::CheckClassTemplate(Scope *S, unsigned TagSpec, TagUseKind TUK,
     // Check for redefinition of this class template.
     if (TUK == TUK_Definition) {
       if (TagDecl *Def = PrevRecordDecl->getDefinition()) {
+        // If we have a prior definition that is not visible, treat this as
+        // simply making that previous definition visible.
+        NamedDecl *Hidden = nullptr;
+        if (SkipBody && !hasVisibleDefinition(Def, &Hidden)) {
+          SkipBody->ShouldSkip = true;
+          auto *Tmpl = cast<CXXRecordDecl>(Hidden)->getDescribedClassTemplate();
+          assert(Tmpl && "original definition of a class template is not a "
+                         "class template?");
+          makeMergedDefinitionVisible(Hidden, KWLoc);
+          makeMergedDefinitionVisible(Tmpl, KWLoc);
+          return Def;
+        }
+
         Diag(NameLoc, diag::err_redefinition) << Name;
         Diag(Def->getLocation(), diag::note_previous_definition);
         // FIXME: Would it make sense to try to "forget" the previous
@@ -1296,6 +1310,9 @@ bool Sema::CheckTemplateParameterList(TemplateParameterList *NewParams,
       // Merge default arguments for template type parameters.
       TemplateTypeParmDecl *OldTypeParm
           = OldParams? cast<TemplateTypeParmDecl>(*OldParam) : nullptr;
+      // FIXME: There might be a visible declaration of this template parameter.
+      if (OldTypeParm && !LookupResult::isVisible(*this, OldTypeParm))
+        OldTypeParm = nullptr;
 
       if (NewTypeParm->isParameterPack()) {
         assert(!NewTypeParm->hasDefaultArgument() &&
@@ -1341,6 +1358,8 @@ bool Sema::CheckTemplateParameterList(TemplateParameterList *NewParams,
       // Merge default arguments for non-type template parameters
       NonTypeTemplateParmDecl *OldNonTypeParm
         = OldParams? cast<NonTypeTemplateParmDecl>(*OldParam) : nullptr;
+      if (OldNonTypeParm && !LookupResult::isVisible(*this, OldNonTypeParm))
+        OldNonTypeParm = nullptr;
       if (NewNonTypeParm->isParameterPack()) {
         assert(!NewNonTypeParm->hasDefaultArgument() &&
                "Parameter packs can't have a default argument!");
@@ -1388,6 +1407,8 @@ bool Sema::CheckTemplateParameterList(TemplateParameterList *NewParams,
       // Merge default arguments for template template parameters
       TemplateTemplateParmDecl *OldTemplateParm
         = OldParams? cast<TemplateTemplateParmDecl>(*OldParam) : nullptr;
+      if (OldTemplateParm && !LookupResult::isVisible(*this, OldTemplateParm))
+        OldTemplateParm = nullptr;
       if (NewTemplateParm->isParameterPack()) {
         assert(!NewTemplateParm->hasDefaultArgument() &&
                "Parameter packs can't have a default argument!");
@@ -1798,7 +1819,7 @@ TemplateParameterList *Sema::MatchTemplateParametersToScopeSpecifier(
       }
     } else if (const TemplateSpecializationType *TST
                                      = T->getAs<TemplateSpecializationType>()) {
-      if (TemplateDecl *Template = TST->getTemplateName().getAsTemplateDecl()) {        
+      if (TemplateDecl *Template = TST->getTemplateName().getAsTemplateDecl()) {
         ExpectedTemplateParams = Template->getTemplateParameters();
         NeedNonemptyTemplateHeader = true;        
       }
@@ -5835,11 +5856,13 @@ static bool CheckTemplateSpecializationScope(Sema &S,
     if (isa<TranslationUnitDecl>(SpecializedContext))
       S.Diag(Loc, diag::err_template_spec_redecl_global_scope)
         << EntityKind << Specialized;
-    else if (isa<NamespaceDecl>(SpecializedContext))
-      S.Diag(Loc, diag::err_template_spec_redecl_out_of_scope)
-        << EntityKind << Specialized
-        << cast<NamedDecl>(SpecializedContext);
-    else
+    else if (isa<NamespaceDecl>(SpecializedContext)) {
+      int Diag = diag::err_template_spec_redecl_out_of_scope;
+      if (S.getLangOpts().MicrosoftExt)
+        Diag = diag::ext_ms_template_spec_redecl_out_of_scope;
+      S.Diag(Loc, Diag) << EntityKind << Specialized
+                        << cast<NamedDecl>(SpecializedContext);
+    } else
       llvm_unreachable("unexpected namespace context for specialization");
 
     S.Diag(Specialized->getLocation(), diag::note_specialized_entity);
@@ -6036,7 +6059,9 @@ Sema::ActOnClassTemplateSpecialization(Scope *S, unsigned TagSpec,
                                        SourceLocation ModulePrivateLoc,
                                        TemplateIdAnnotation &TemplateId,
                                        AttributeList *Attr,
-                               MultiTemplateParamsArg TemplateParameterLists) {
+                                       MultiTemplateParamsArg
+                                           TemplateParameterLists,
+                                       SkipBodyInfo *SkipBody) {
   assert(TUK != TUK_Reference && "References are not specializations");
 
   CXXScopeSpec &SS = TemplateId.SS;
@@ -6347,7 +6372,14 @@ Sema::ActOnClassTemplateSpecialization(Scope *S, unsigned TagSpec,
 
   // Check that this isn't a redefinition of this specialization.
   if (TUK == TUK_Definition) {
-    if (RecordDecl *Def = Specialization->getDefinition()) {
+    RecordDecl *Def = Specialization->getDefinition();
+    NamedDecl *Hidden = nullptr;
+    if (Def && SkipBody && !hasVisibleDefinition(Def, &Hidden)) {
+      SkipBody->ShouldSkip = true;
+      makeMergedDefinitionVisible(Hidden, KWLoc);
+      // From here on out, treat this as just a redeclaration.
+      TUK = TUK_Declaration;
+    } else if (Def) {
       SourceRange Range(TemplateNameLoc, RAngleLoc);
       Diag(TemplateNameLoc, diag::err_redefinition)
         << Context.getTypeDeclType(Specialization) << Range;
@@ -7177,9 +7209,27 @@ Sema::ActOnExplicitInstantiation(Scope *S,
   //   There are two forms of explicit instantiation: an explicit instantiation
   //   definition and an explicit instantiation declaration. An explicit
   //   instantiation declaration begins with the extern keyword. [...]
-  TemplateSpecializationKind TSK
-    = ExternLoc.isInvalid()? TSK_ExplicitInstantiationDefinition
-                           : TSK_ExplicitInstantiationDeclaration;
+  TemplateSpecializationKind TSK = ExternLoc.isInvalid()
+                                       ? TSK_ExplicitInstantiationDefinition
+                                       : TSK_ExplicitInstantiationDeclaration;
+
+  if (TSK == TSK_ExplicitInstantiationDeclaration) {
+    // Check for dllexport class template instantiation declarations.
+    for (AttributeList *A = Attr; A; A = A->getNext()) {
+      if (A->getKind() == AttributeList::AT_DLLExport) {
+        Diag(ExternLoc,
+             diag::warn_attribute_dllexport_explicit_instantiation_decl);
+        Diag(A->getLoc(), diag::note_attribute);
+        break;
+      }
+    }
+
+    if (auto *A = ClassTemplate->getTemplatedDecl()->getAttr<DLLExportAttr>()) {
+      Diag(ExternLoc,
+           diag::warn_attribute_dllexport_explicit_instantiation_decl);
+      Diag(A->getLocation(), diag::note_attribute);
+    }
+  }
 
   // Translate the parser's template argument list in our AST format.
   TemplateArgumentListInfo TemplateArgs(LAngleLoc, RAngleLoc);
@@ -7315,10 +7365,19 @@ Sema::ActOnExplicitInstantiation(Scope *S,
     // Fix a TSK_ExplicitInstantiationDeclaration followed by a
     // TSK_ExplicitInstantiationDefinition
     if (Old_TSK == TSK_ExplicitInstantiationDeclaration &&
-        TSK == TSK_ExplicitInstantiationDefinition)
+        TSK == TSK_ExplicitInstantiationDefinition) {
       // FIXME: Need to notify the ASTMutationListener that we did this.
       Def->setTemplateSpecializationKind(TSK);
 
+      if (!getDLLAttr(Def) && getDLLAttr(Specialization)) {
+        auto *A = cast<InheritableAttr>(
+            getDLLAttr(Specialization)->clone(getASTContext()));
+        A->setInherited(true);
+        Def->addAttr(A);
+        checkClassLevelDLLAttribute(Def);
+      }
+    }
+
     InstantiateClassTemplateSpecializationMembers(TemplateNameLoc, Def, TSK);
   }
 
@@ -8285,7 +8344,7 @@ void Sema::MarkAsLateParsedTemplate(FunctionDecl *FD, Decl *FnD,
   // Take tokens to avoid allocations
   LPT->Toks.swap(Toks);
   LPT->D = FnD;
-  LateParsedTemplateMap[FD] = LPT;
+  LateParsedTemplateMap.insert(std::make_pair(FD, LPT));
 
   FD->setLateTemplateParsed(true);
 }
diff --git a/lib/Sema/SemaTemplateDeduction.cpp b/lib/Sema/SemaTemplateDeduction.cpp
index dd2a4d26979e..6f676ad7c55d 100644
--- a/lib/Sema/SemaTemplateDeduction.cpp
+++ b/lib/Sema/SemaTemplateDeduction.cpp
@@ -91,30 +91,6 @@ DeduceTemplateArguments(Sema &S,
                         TemplateDeductionInfo &Info,
                         SmallVectorImpl<DeducedTemplateArgument> &Deduced);
 
-/// \brief Whether template argument deduction for two reference parameters
-/// resulted in the argument type, parameter type, or neither type being more
-/// qualified than the other.
-enum DeductionQualifierComparison {
-  NeitherMoreQualified = 0,
-  ParamMoreQualified,
-  ArgMoreQualified
-};
-
-/// \brief Stores the result of comparing two reference parameters while
-/// performing template argument deduction for partial ordering of function
-/// templates.
-struct RefParamPartialOrderingComparison {
-  /// \brief Whether the parameter type is an rvalue reference type.
-  bool ParamIsRvalueRef;
-  /// \brief Whether the argument type is an rvalue reference type.
-  bool ArgIsRvalueRef;
-
-  /// \brief Whether the parameter or argument (or neither) is more qualified.
-  DeductionQualifierComparison Qualifiers;
-};
-
-
-
 static Sema::TemplateDeductionResult
 DeduceTemplateArgumentsByTypeMatch(Sema &S,
                                    TemplateParameterList *TemplateParams,
@@ -124,9 +100,7 @@ DeduceTemplateArgumentsByTypeMatch(Sema &S,
                                    SmallVectorImpl<DeducedTemplateArgument> &
                                                       Deduced,
                                    unsigned TDF,
-                                   bool PartialOrdering = false,
-                            SmallVectorImpl<RefParamPartialOrderingComparison> *
-                                                 RefParamComparisons = nullptr);
+                                   bool PartialOrdering = false);
 
 static Sema::TemplateDeductionResult
 DeduceTemplateArguments(Sema &S,
@@ -602,6 +576,7 @@ struct clang::DeducedPack {
   DeducedPack *Outer;
 };
 
+namespace {
 /// A scope in which we're performing pack deduction.
 class PackDeductionScope {
 public:
@@ -756,6 +731,7 @@ private:
 
   SmallVector<DeducedPack, 2> Packs;
 };
+} // namespace
 
 /// \brief Deduce the template arguments by comparing the list of parameter
 /// types to the list of argument types, as in the parameter-type-lists of
@@ -784,9 +760,6 @@ private:
 /// deduction for during partial ordering for a call
 /// (C++0x [temp.deduct.partial]).
 ///
-/// \param RefParamComparisons If we're performing template argument deduction
-/// in the context of partial ordering, the set of qualifier comparisons.
-///
 /// \returns the result of template argument deduction so far. Note that a
 /// "success" result means that template argument deduction has not yet failed,
 /// but it may still fail, later, for other reasons.
@@ -798,9 +771,7 @@ DeduceTemplateArguments(Sema &S,
                         TemplateDeductionInfo &Info,
                         SmallVectorImpl<DeducedTemplateArgument> &Deduced,
                         unsigned TDF,
-                        bool PartialOrdering = false,
-                        SmallVectorImpl<RefParamPartialOrderingComparison> *
-                                                RefParamComparisons = nullptr) {
+                        bool PartialOrdering = false) {
   // Fast-path check to see if we have too many/too few arguments.
   if (NumParams != NumArgs &&
       !(NumParams && isa<PackExpansionType>(Params[NumParams - 1])) &&
@@ -836,8 +807,7 @@ DeduceTemplateArguments(Sema &S,
             = DeduceTemplateArgumentsByTypeMatch(S, TemplateParams,
                                                  Params[ParamIdx], Args[ArgIdx],
                                                  Info, Deduced, TDF,
-                                                 PartialOrdering,
-                                                 RefParamComparisons))
+                                                 PartialOrdering))
         return Result;
 
       ++ArgIdx;
@@ -869,8 +839,7 @@ DeduceTemplateArguments(Sema &S,
       if (Sema::TemplateDeductionResult Result
             = DeduceTemplateArgumentsByTypeMatch(S, TemplateParams, Pattern,
                                                  Args[ArgIdx], Info, Deduced,
-                                                 TDF, PartialOrdering,
-                                                 RefParamComparisons))
+                                                 TDF, PartialOrdering))
         return Result;
 
       PackScope.nextPackElement();
@@ -967,9 +936,6 @@ bool Sema::isSameOrCompatibleFunctionType(CanQualType Param,
 /// \param PartialOrdering Whether we're performing template argument deduction
 /// in the context of partial ordering (C++0x [temp.deduct.partial]).
 ///
-/// \param RefParamComparisons If we're performing template argument deduction
-/// in the context of partial ordering, the set of qualifier comparisons.
-///
 /// \returns the result of template argument deduction so far. Note that a
 /// "success" result means that template argument deduction has not yet failed,
 /// but it may still fail, later, for other reasons.
@@ -980,9 +946,7 @@ DeduceTemplateArgumentsByTypeMatch(Sema &S,
                                    TemplateDeductionInfo &Info,
                             SmallVectorImpl<DeducedTemplateArgument> &Deduced,
                                    unsigned TDF,
-                                   bool PartialOrdering,
-                            SmallVectorImpl<RefParamPartialOrderingComparison> *
-                                                          RefParamComparisons) {
+                                   bool PartialOrdering) {
   // We only want to look at the canonical types, since typedefs and
   // sugar are not part of template argument deduction.
   QualType Param = S.Context.getCanonicalType(ParamIn);
@@ -995,7 +959,7 @@ DeduceTemplateArgumentsByTypeMatch(Sema &S,
     Arg = ArgExpansion->getPattern();
 
   if (PartialOrdering) {
-    // C++0x [temp.deduct.partial]p5:
+    // C++11 [temp.deduct.partial]p5:
     //   Before the partial ordering is done, certain transformations are
     //   performed on the types used for partial ordering:
     //     - If P is a reference type, P is replaced by the type referred to.
@@ -1008,42 +972,42 @@ DeduceTemplateArgumentsByTypeMatch(Sema &S,
     if (ArgRef)
       Arg = ArgRef->getPointeeType();
 
-    if (RefParamComparisons && ParamRef && ArgRef) {
-      // C++0x [temp.deduct.partial]p6:
-      //   If both P and A were reference types (before being replaced with the
-      //   type referred to above), determine which of the two types (if any) is
-      //   more cv-qualified than the other; otherwise the types are considered
-      //   to be equally cv-qualified for partial ordering purposes. The result
-      //   of this determination will be used below.
+    if (ParamRef && ArgRef && S.Context.hasSameUnqualifiedType(Param, Arg)) {
+      // C++11 [temp.deduct.partial]p9:
+      //   If, for a given type, deduction succeeds in both directions (i.e.,
+      //   the types are identical after the transformations above) and both
+      //   P and A were reference types [...]:
+      //     - if [one type] was an lvalue reference and [the other type] was
+      //       not, [the other type] is not considered to be at least as
+      //       specialized as [the first type]
+      //     - if [one type] is more cv-qualified than [the other type],
+      //       [the other type] is not considered to be at least as specialized
+      //       as [the first type]
+      // Objective-C ARC adds:
+      //     - [one type] has non-trivial lifetime, [the other type] has
+      //       __unsafe_unretained lifetime, and the types are otherwise
+      //       identical
       //
-      // We save this information for later, using it only when deduction
-      // succeeds in both directions.
-      RefParamPartialOrderingComparison Comparison;
-      Comparison.ParamIsRvalueRef = ParamRef->getAs<RValueReferenceType>();
-      Comparison.ArgIsRvalueRef = ArgRef->getAs<RValueReferenceType>();
-      Comparison.Qualifiers = NeitherMoreQualified;
-      
+      // A is "considered to be at least as specialized" as P iff deduction
+      // succeeds, so we model this as a deduction failure. Note that
+      // [the first type] is P and [the other type] is A here; the standard
+      // gets this backwards.
       Qualifiers ParamQuals = Param.getQualifiers();
       Qualifiers ArgQuals = Arg.getQualifiers();
-      if (ParamQuals.isStrictSupersetOf(ArgQuals))
-        Comparison.Qualifiers = ParamMoreQualified;
-      else if (ArgQuals.isStrictSupersetOf(ParamQuals))
-        Comparison.Qualifiers = ArgMoreQualified;
-      else if (ArgQuals.getObjCLifetime() != ParamQuals.getObjCLifetime() &&
-               ArgQuals.withoutObjCLifetime()
-                 == ParamQuals.withoutObjCLifetime()) {
-        // Prefer binding to non-__unsafe_autoretained parameters.
-        if (ArgQuals.getObjCLifetime() == Qualifiers::OCL_ExplicitNone &&
-            ParamQuals.getObjCLifetime())
-          Comparison.Qualifiers = ParamMoreQualified;
-        else if (ParamQuals.getObjCLifetime() == Qualifiers::OCL_ExplicitNone &&
-                 ArgQuals.getObjCLifetime())
-          Comparison.Qualifiers = ArgMoreQualified;
+      if ((ParamRef->isLValueReferenceType() &&
+           !ArgRef->isLValueReferenceType()) ||
+          ParamQuals.isStrictSupersetOf(ArgQuals) ||
+          (ParamQuals.hasNonTrivialObjCLifetime() &&
+           ArgQuals.getObjCLifetime() == Qualifiers::OCL_ExplicitNone &&
+           ParamQuals.withoutObjCLifetime() ==
+               ArgQuals.withoutObjCLifetime())) {
+        Info.FirstArg = TemplateArgument(ParamIn);
+        Info.SecondArg = TemplateArgument(ArgIn);
+        return Sema::TDK_NonDeducedMismatch;
       }
-      RefParamComparisons->push_back(Comparison);
     }
 
-    // C++0x [temp.deduct.partial]p7:
+    // C++11 [temp.deduct.partial]p7:
     //   Remove any top-level cv-qualifiers:
     //     - If P is a cv-qualified type, P is replaced by the cv-unqualified
     //       version of P.
@@ -2784,7 +2748,8 @@ Sema::FinishTemplateArgumentDeduction(FunctionTemplateDecl *FunctionTemplate,
                                       unsigned NumExplicitlySpecified,
                                       FunctionDecl *&Specialization,
                                       TemplateDeductionInfo &Info,
-        SmallVectorImpl<OriginalCallArg> const *OriginalCallArgs) {
+        SmallVectorImpl<OriginalCallArg> const *OriginalCallArgs,
+                                      bool PartialOverloading) {
   TemplateParameterList *TemplateParams
     = FunctionTemplate->getTemplateParameters();
 
@@ -2911,6 +2876,8 @@ Sema::FinishTemplateArgumentDeduction(FunctionTemplateDecl *FunctionTemplate,
                          const_cast<NamedDecl *>(TemplateParams->getParam(I)));
       Info.reset(TemplateArgumentList::CreateCopy(Context, Builder.data(),
                                                   Builder.size()));
+      if (PartialOverloading) break;
+
       return HasDefaultArg ? TDK_SubstitutionFailure : TDK_Incomplete;
     }
 
@@ -3136,34 +3103,16 @@ static bool AdjustFunctionParmAndArgTypesForDeduction(Sema &S,
   //   are ignored for type deduction.
   if (ParamType.hasQualifiers())
     ParamType = ParamType.getUnqualifiedType();
-  const ReferenceType *ParamRefType = ParamType->getAs<ReferenceType>();
-  if (ParamRefType) {
-    QualType PointeeType = ParamRefType->getPointeeType();
 
-    // If the argument has incomplete array type, try to complete its type.
-    if (ArgType->isIncompleteArrayType() && !S.RequireCompleteExprType(Arg, 0))
-      ArgType = Arg->getType();
-
-    //   [C++0x] If P is an rvalue reference to a cv-unqualified
-    //   template parameter and the argument is an lvalue, the type
-    //   "lvalue reference to A" is used in place of A for type
-    //   deduction.
-    if (isa<RValueReferenceType>(ParamType)) {
-      if (!PointeeType.getQualifiers() &&
-          isa<TemplateTypeParmType>(PointeeType) &&
-          Arg->Classify(S.Context).isLValue() &&
-          Arg->getType() != S.Context.OverloadTy &&
-          Arg->getType() != S.Context.BoundMemberTy)
-        ArgType = S.Context.getLValueReferenceType(ArgType);
-    }
-
-    //   [...] If P is a reference type, the type referred to by P is used
-    //   for type deduction.
-    ParamType = PointeeType;
-  }
+  //   [...] If P is a reference type, the type referred to by P is
+  //   used for type deduction.
+  const ReferenceType *ParamRefType = ParamType->getAs<ReferenceType>();
+  if (ParamRefType)
+    ParamType = ParamRefType->getPointeeType();
 
-  // Overload sets usually make this parameter an undeduced
-  // context, but there are sometimes special circumstances.
+  // Overload sets usually make this parameter an undeduced context,
+  // but there are sometimes special circumstances.  Typically
+  // involving a template-id-expr.
   if (ArgType == S.Context.OverloadTy) {
     ArgType = ResolveOverloadForDeduction(S, TemplateParams,
                                           Arg, ParamType,
@@ -3173,12 +3122,17 @@ static bool AdjustFunctionParmAndArgTypesForDeduction(Sema &S,
   }
 
   if (ParamRefType) {
+    // If the argument has incomplete array type, try to complete its type.
+    if (ArgType->isIncompleteArrayType() && !S.RequireCompleteExprType(Arg, 0))
+      ArgType = Arg->getType();
+
     // C++0x [temp.deduct.call]p3:
-    //   [...] If P is of the form T&&, where T is a template parameter, and
-    //   the argument is an lvalue, the type A& is used in place of A for
-    //   type deduction.
+    //   If P is an rvalue reference to a cv-unqualified template
+    //   parameter and the argument is an lvalue, the type "lvalue
+    //   reference to A" is used in place of A for type deduction.
     if (ParamRefType->isRValueReferenceType() &&
-        ParamRefType->getAs<TemplateTypeParmType>() &&
+        !ParamType.getQualifiers() &&
+        isa<TemplateTypeParmType>(ParamType) &&
         Arg->isLValue())
       ArgType = S.Context.getLValueReferenceType(ArgType);
   } else {
@@ -3298,26 +3252,28 @@ DeduceTemplateArgumentByListElement(Sema &S,
 Sema::TemplateDeductionResult Sema::DeduceTemplateArguments(
     FunctionTemplateDecl *FunctionTemplate,
     TemplateArgumentListInfo *ExplicitTemplateArgs, ArrayRef<Expr *> Args,
-    FunctionDecl *&Specialization, TemplateDeductionInfo &Info) {
+    FunctionDecl *&Specialization, TemplateDeductionInfo &Info,
+    bool PartialOverloading) {
   if (FunctionTemplate->isInvalidDecl())
     return TDK_Invalid;
 
   FunctionDecl *Function = FunctionTemplate->getTemplatedDecl();
+  unsigned NumParams = Function->getNumParams();
 
   // C++ [temp.deduct.call]p1:
   //   Template argument deduction is done by comparing each function template
   //   parameter type (call it P) with the type of the corresponding argument
   //   of the call (call it A) as described below.
   unsigned CheckArgs = Args.size();
-  if (Args.size() < Function->getMinRequiredArguments())
+  if (Args.size() < Function->getMinRequiredArguments() && !PartialOverloading)
     return TDK_TooFewArguments;
-  else if (Args.size() > Function->getNumParams()) {
+  else if (TooManyArguments(NumParams, Args.size(), PartialOverloading)) {
     const FunctionProtoType *Proto
       = Function->getType()->getAs<FunctionProtoType>();
     if (Proto->isTemplateVariadic())
       /* Do nothing */;
     else if (Proto->isVariadic())
-      CheckArgs = Function->getNumParams();
+      CheckArgs = NumParams;
     else
       return TDK_TooManyArguments;
   }
@@ -3344,7 +3300,7 @@ Sema::TemplateDeductionResult Sema::DeduceTemplateArguments(
     NumExplicitlySpecified = Deduced.size();
   } else {
     // Just fill in the parameter types from the function declaration.
-    for (unsigned I = 0, N = Function->getNumParams(); I != N; ++I)
+    for (unsigned I = 0; I != NumParams; ++I)
       ParamTypes.push_back(Function->getParamDecl(I)->getType());
   }
 
@@ -3352,8 +3308,8 @@ Sema::TemplateDeductionResult Sema::DeduceTemplateArguments(
   Deduced.resize(TemplateParams->size());
   unsigned ArgIdx = 0;
   SmallVector<OriginalCallArg, 4> OriginalCallArgs;
-  for (unsigned ParamIdx = 0, NumParams = ParamTypes.size();
-       ParamIdx != NumParams; ++ParamIdx) {
+  for (unsigned ParamIdx = 0, NumParamTypes = ParamTypes.size();
+       ParamIdx != NumParamTypes; ++ParamIdx) {
     QualType OrigParamType = ParamTypes[ParamIdx];
     QualType ParamType = OrigParamType;
     
@@ -3422,7 +3378,7 @@ Sema::TemplateDeductionResult Sema::DeduceTemplateArguments(
     //   the function parameter pack. For a function parameter pack that does
     //   not occur at the end of the parameter-declaration-list, the type of
     //   the parameter pack is a non-deduced context.
-    if (ParamIdx + 1 < NumParams)
+    if (ParamIdx + 1 < NumParamTypes)
       break;
 
     QualType ParamPattern = ParamExpansion->getPattern();
@@ -3492,7 +3448,8 @@ Sema::TemplateDeductionResult Sema::DeduceTemplateArguments(
 
   return FinishTemplateArgumentDeduction(FunctionTemplate, Deduced,
                                          NumExplicitlySpecified, Specialization,
-                                         Info, &OriginalCallArgs);
+                                         Info, &OriginalCallArgs,
+                                         PartialOverloading);
 }
 
 QualType Sema::adjustCCAndNoReturn(QualType ArgFunctionType,
@@ -3699,8 +3656,10 @@ SpecializeCorrespondingLambdaCallOperatorAndInvoker(
   FunctionTemplateDecl *InvokerTemplate = LambdaClass->
                   getLambdaStaticInvoker()->getDescribedFunctionTemplate();
 
-  Sema::TemplateDeductionResult LLVM_ATTRIBUTE_UNUSED Result
-    = S.FinishTemplateArgumentDeduction(InvokerTemplate, DeducedArguments, 0, 
+#ifndef NDEBUG
+  Sema::TemplateDeductionResult LLVM_ATTRIBUTE_UNUSED Result =
+#endif
+    S.FinishTemplateArgumentDeduction(InvokerTemplate, DeducedArguments, 0, 
           InvokerSpecialized, TDInfo);
   assert(Result == Sema::TDK_Success && 
     "If the call operator succeeded so should the invoker!");
@@ -4153,8 +4112,7 @@ static bool isAtLeastAsSpecializedAs(Sema &S,
                                      FunctionTemplateDecl *FT1,
                                      FunctionTemplateDecl *FT2,
                                      TemplatePartialOrderingContext TPOC,
-                                     unsigned NumCallArguments1,
-    SmallVectorImpl<RefParamPartialOrderingComparison> *RefParamComparisons) {
+                                     unsigned NumCallArguments1) {
   FunctionDecl *FD1 = FT1->getTemplatedDecl();
   FunctionDecl *FD2 = FT2->getTemplatedDecl();
   const FunctionProtoType *Proto1 = FD1->getType()->getAs<FunctionProtoType>();
@@ -4219,8 +4177,7 @@ static bool isAtLeastAsSpecializedAs(Sema &S,
       Args2.resize(NumComparedArguments);
     if (DeduceTemplateArguments(S, TemplateParams, Args2.data(), Args2.size(),
                                 Args1.data(), Args1.size(), Info, Deduced,
-                                TDF_None, /*PartialOrdering=*/true,
-                                RefParamComparisons))
+                                TDF_None, /*PartialOrdering=*/true))
       return false;
 
     break;
@@ -4232,7 +4189,7 @@ static bool isAtLeastAsSpecializedAs(Sema &S,
     if (DeduceTemplateArgumentsByTypeMatch(
             S, TemplateParams, Proto2->getReturnType(), Proto1->getReturnType(),
             Info, Deduced, TDF_None,
-            /*PartialOrdering=*/true, RefParamComparisons))
+            /*PartialOrdering=*/true))
       return false;
     break;
 
@@ -4242,8 +4199,7 @@ static bool isAtLeastAsSpecializedAs(Sema &S,
     if (DeduceTemplateArgumentsByTypeMatch(S, TemplateParams,
                                            FD2->getType(), FD1->getType(),
                                            Info, Deduced, TDF_None,
-                                           /*PartialOrdering=*/true,
-                                           RefParamComparisons))
+                                           /*PartialOrdering=*/true))
       return false;
     break;
   }
@@ -4342,83 +4298,17 @@ Sema::getMoreSpecializedTemplate(FunctionTemplateDecl *FT1,
                                  TemplatePartialOrderingContext TPOC,
                                  unsigned NumCallArguments1,
                                  unsigned NumCallArguments2) {
-  SmallVector<RefParamPartialOrderingComparison, 4> RefParamComparisons;
   bool Better1 = isAtLeastAsSpecializedAs(*this, Loc, FT1, FT2, TPOC,
-                                          NumCallArguments1, nullptr);
+                                          NumCallArguments1);
   bool Better2 = isAtLeastAsSpecializedAs(*this, Loc, FT2, FT1, TPOC,
-                                          NumCallArguments2,
-                                          &RefParamComparisons);
+                                          NumCallArguments2);
 
   if (Better1 != Better2) // We have a clear winner
-    return Better1? FT1 : FT2;
+    return Better1 ? FT1 : FT2;
 
   if (!Better1 && !Better2) // Neither is better than the other
     return nullptr;
 
-  // C++0x [temp.deduct.partial]p10:
-  //   If for each type being considered a given template is at least as
-  //   specialized for all types and more specialized for some set of types and
-  //   the other template is not more specialized for any types or is not at
-  //   least as specialized for any types, then the given template is more
-  //   specialized than the other template. Otherwise, neither template is more
-  //   specialized than the other.
-  Better1 = false;
-  Better2 = false;
-  for (unsigned I = 0, N = RefParamComparisons.size(); I != N; ++I) {
-    // C++0x [temp.deduct.partial]p9:
-    //   If, for a given type, deduction succeeds in both directions (i.e., the
-    //   types are identical after the transformations above) and both P and A
-    //   were reference types (before being replaced with the type referred to
-    //   above):
-
-    //     -- if the type from the argument template was an lvalue reference
-    //        and the type from the parameter template was not, the argument
-    //        type is considered to be more specialized than the other;
-    //        otherwise,
-    if (!RefParamComparisons[I].ArgIsRvalueRef &&
-        RefParamComparisons[I].ParamIsRvalueRef) {
-      Better2 = true;
-      if (Better1)
-        return nullptr;
-      continue;
-    } else if (!RefParamComparisons[I].ParamIsRvalueRef &&
-               RefParamComparisons[I].ArgIsRvalueRef) {
-      Better1 = true;
-      if (Better2)
-        return nullptr;
-      continue;
-    }
-
-    //     -- if the type from the argument template is more cv-qualified than
-    //        the type from the parameter template (as described above), the
-    //        argument type is considered to be more specialized than the
-    //        other; otherwise,
-    switch (RefParamComparisons[I].Qualifiers) {
-    case NeitherMoreQualified:
-      break;
-
-    case ParamMoreQualified:
-      Better1 = true;
-      if (Better2)
-        return nullptr;
-      continue;
-
-    case ArgMoreQualified:
-      Better2 = true;
-      if (Better1)
-        return nullptr;
-      continue;
-    }
-
-    //     -- neither type is more specialized than the other.
-  }
-
-  assert(!(Better1 && Better2) && "Should have broken out in the loop above");
-  if (Better1)
-    return FT1;
-  else if (Better2)
-    return FT2;
-
   // FIXME: This mimics what GCC implements, but doesn't match up with the
   // proposed resolution for core issue 692. This area needs to be sorted out,
   // but for now we attempt to maintain compatibility.
@@ -4591,8 +4481,7 @@ Sema::getMoreSpecializedPartialSpecialization(
   bool Better1 = !DeduceTemplateArgumentsByTypeMatch(*this,
                                             PS2->getTemplateParameters(),
                                             PT2, PT1, Info, Deduced, TDF_None,
-                                            /*PartialOrdering=*/true,
-                                            /*RefParamComparisons=*/nullptr);
+                                            /*PartialOrdering=*/true);
   if (Better1) {
     SmallVector<TemplateArgument, 4> DeducedArgs(Deduced.begin(),Deduced.end());
     InstantiatingTemplate Inst(*this, Loc, PS2, DeducedArgs, Info);
@@ -4605,8 +4494,7 @@ Sema::getMoreSpecializedPartialSpecialization(
   Deduced.resize(PS1->getTemplateParameters()->size());
   bool Better2 = !DeduceTemplateArgumentsByTypeMatch(
       *this, PS1->getTemplateParameters(), PT1, PT2, Info, Deduced, TDF_None,
-      /*PartialOrdering=*/true,
-      /*RefParamComparisons=*/nullptr);
+      /*PartialOrdering=*/true);
   if (Better2) {
     SmallVector<TemplateArgument, 4> DeducedArgs(Deduced.begin(),
                                                  Deduced.end());
@@ -4649,8 +4537,7 @@ Sema::getMoreSpecializedPartialSpecialization(
   Deduced.resize(PS2->getTemplateParameters()->size());
   bool Better1 = !DeduceTemplateArgumentsByTypeMatch(
       *this, PS2->getTemplateParameters(), PT2, PT1, Info, Deduced, TDF_None,
-      /*PartialOrdering=*/true,
-      /*RefParamComparisons=*/nullptr);
+      /*PartialOrdering=*/true);
   if (Better1) {
     SmallVector<TemplateArgument, 4> DeducedArgs(Deduced.begin(),
                                                  Deduced.end());
@@ -4666,8 +4553,7 @@ Sema::getMoreSpecializedPartialSpecialization(
   bool Better2 = !DeduceTemplateArgumentsByTypeMatch(*this,
                                             PS1->getTemplateParameters(),
                                             PT1, PT2, Info, Deduced, TDF_None,
-                                            /*PartialOrdering=*/true,
-                                            /*RefParamComparisons=*/nullptr);
+                                            /*PartialOrdering=*/true);
   if (Better2) {
     SmallVector<TemplateArgument, 4> DeducedArgs(Deduced.begin(),Deduced.end());
     InstantiatingTemplate Inst(*this, Loc, PS1, DeducedArgs, Info);
diff --git a/lib/Sema/SemaTemplateInstantiate.cpp b/lib/Sema/SemaTemplateInstantiate.cpp
index 6ac7175cf30f..82ff7c0ca4f5 100644
--- a/lib/Sema/SemaTemplateInstantiate.cpp
+++ b/lib/Sema/SemaTemplateInstantiate.cpp
@@ -200,13 +200,19 @@ bool Sema::ActiveTemplateInstantiation::isInstantiationRecord() const {
   llvm_unreachable("Invalid InstantiationKind!");
 }
 
-void Sema::InstantiatingTemplate::Initialize(
-    ActiveTemplateInstantiation::InstantiationKind Kind,
+Sema::InstantiatingTemplate::InstantiatingTemplate(
+    Sema &SemaRef, ActiveTemplateInstantiation::InstantiationKind Kind,
     SourceLocation PointOfInstantiation, SourceRange InstantiationRange,
     Decl *Entity, NamedDecl *Template, ArrayRef<TemplateArgument> TemplateArgs,
-    sema::TemplateDeductionInfo *DeductionInfo) {
-  SavedInNonInstantiationSFINAEContext =
-      SemaRef.InNonInstantiationSFINAEContext;
+    sema::TemplateDeductionInfo *DeductionInfo)
+    : SemaRef(SemaRef), SavedInNonInstantiationSFINAEContext(
+                            SemaRef.InNonInstantiationSFINAEContext) {
+  // Don't allow further instantiation if a fatal error has occcured.  Any
+  // diagnostics we might have raised will not be visible.
+  if (SemaRef.Diags.hasFatalErrorOccurred()) {
+    Invalid = true;
+    return;
+  }
   Invalid = CheckInstantiationDepth(PointOfInstantiation, InstantiationRange);
   if (!Invalid) {
     ActiveTemplateInstantiation Inst;
@@ -225,124 +231,98 @@ void Sema::InstantiatingTemplate::Initialize(
   }
 }
 
-Sema::InstantiatingTemplate::
-InstantiatingTemplate(Sema &SemaRef, SourceLocation PointOfInstantiation,
-                      Decl *Entity,
-                      SourceRange InstantiationRange)
-  : SemaRef(SemaRef)
-{
-  Initialize(ActiveTemplateInstantiation::TemplateInstantiation,
-             PointOfInstantiation, InstantiationRange, Entity);
-}
-
-Sema::InstantiatingTemplate::
-InstantiatingTemplate(Sema &SemaRef, SourceLocation PointOfInstantiation,
-                      FunctionDecl *Entity, ExceptionSpecification,
-                      SourceRange InstantiationRange)
-  : SemaRef(SemaRef)
-{
-  Initialize(ActiveTemplateInstantiation::ExceptionSpecInstantiation,
-             PointOfInstantiation, InstantiationRange, Entity);
-}
-
-Sema::InstantiatingTemplate::
-InstantiatingTemplate(Sema &SemaRef, SourceLocation PointOfInstantiation,
-                      TemplateDecl *Template,
-                      ArrayRef<TemplateArgument> TemplateArgs,
-                      SourceRange InstantiationRange)
-  : SemaRef(SemaRef)
-{
-  Initialize(ActiveTemplateInstantiation::DefaultTemplateArgumentInstantiation,
-             PointOfInstantiation, InstantiationRange,
-             Template, nullptr, TemplateArgs);
-}
-
-Sema::InstantiatingTemplate::
-InstantiatingTemplate(Sema &SemaRef, SourceLocation PointOfInstantiation,
-                      FunctionTemplateDecl *FunctionTemplate,
-                      ArrayRef<TemplateArgument> TemplateArgs,
-                      ActiveTemplateInstantiation::InstantiationKind Kind,
-                      sema::TemplateDeductionInfo &DeductionInfo,
-                      SourceRange InstantiationRange)
-  : SemaRef(SemaRef)
-{
-  Initialize(Kind, PointOfInstantiation, InstantiationRange,
-             FunctionTemplate, nullptr, TemplateArgs, &DeductionInfo);
-}
-
-Sema::InstantiatingTemplate::
-InstantiatingTemplate(Sema &SemaRef, SourceLocation PointOfInstantiation,
-                      ClassTemplatePartialSpecializationDecl *PartialSpec,
-                      ArrayRef<TemplateArgument> TemplateArgs,
-                      sema::TemplateDeductionInfo &DeductionInfo,
-                      SourceRange InstantiationRange)
-  : SemaRef(SemaRef)
-{
-  Initialize(ActiveTemplateInstantiation::DeducedTemplateArgumentSubstitution,
-             PointOfInstantiation, InstantiationRange,
-             PartialSpec, nullptr, TemplateArgs, &DeductionInfo);
-}
+Sema::InstantiatingTemplate::InstantiatingTemplate(
+    Sema &SemaRef, SourceLocation PointOfInstantiation, Decl *Entity,
+    SourceRange InstantiationRange)
+    : InstantiatingTemplate(SemaRef,
+                            ActiveTemplateInstantiation::TemplateInstantiation,
+                            PointOfInstantiation, InstantiationRange, Entity) {}
+
+Sema::InstantiatingTemplate::InstantiatingTemplate(
+    Sema &SemaRef, SourceLocation PointOfInstantiation, FunctionDecl *Entity,
+    ExceptionSpecification, SourceRange InstantiationRange)
+    : InstantiatingTemplate(
+          SemaRef, ActiveTemplateInstantiation::ExceptionSpecInstantiation,
+          PointOfInstantiation, InstantiationRange, Entity) {}
+
+Sema::InstantiatingTemplate::InstantiatingTemplate(
+    Sema &SemaRef, SourceLocation PointOfInstantiation, TemplateDecl *Template,
+    ArrayRef<TemplateArgument> TemplateArgs, SourceRange InstantiationRange)
+    : InstantiatingTemplate(
+          SemaRef,
+          ActiveTemplateInstantiation::DefaultTemplateArgumentInstantiation,
+          PointOfInstantiation, InstantiationRange, Template, nullptr,
+          TemplateArgs) {}
+
+Sema::InstantiatingTemplate::InstantiatingTemplate(
+    Sema &SemaRef, SourceLocation PointOfInstantiation,
+    FunctionTemplateDecl *FunctionTemplate,
+    ArrayRef<TemplateArgument> TemplateArgs,
+    ActiveTemplateInstantiation::InstantiationKind Kind,
+    sema::TemplateDeductionInfo &DeductionInfo, SourceRange InstantiationRange)
+    : InstantiatingTemplate(SemaRef, Kind, PointOfInstantiation,
+                            InstantiationRange, FunctionTemplate, nullptr,
+                            TemplateArgs, &DeductionInfo) {}
+
+Sema::InstantiatingTemplate::InstantiatingTemplate(
+    Sema &SemaRef, SourceLocation PointOfInstantiation,
+    ClassTemplatePartialSpecializationDecl *PartialSpec,
+    ArrayRef<TemplateArgument> TemplateArgs,
+    sema::TemplateDeductionInfo &DeductionInfo, SourceRange InstantiationRange)
+    : InstantiatingTemplate(
+          SemaRef,
+          ActiveTemplateInstantiation::DeducedTemplateArgumentSubstitution,
+          PointOfInstantiation, InstantiationRange, PartialSpec, nullptr,
+          TemplateArgs, &DeductionInfo) {}
 
 Sema::InstantiatingTemplate::InstantiatingTemplate(
     Sema &SemaRef, SourceLocation PointOfInstantiation,
     VarTemplatePartialSpecializationDecl *PartialSpec,
     ArrayRef<TemplateArgument> TemplateArgs,
     sema::TemplateDeductionInfo &DeductionInfo, SourceRange InstantiationRange)
-  : SemaRef(SemaRef)
-{
-  Initialize(ActiveTemplateInstantiation::DeducedTemplateArgumentSubstitution,
-             PointOfInstantiation, InstantiationRange,
-             PartialSpec, nullptr, TemplateArgs, &DeductionInfo);
-}
-
-Sema::InstantiatingTemplate::
-InstantiatingTemplate(Sema &SemaRef, SourceLocation PointOfInstantiation,
-                      ParmVarDecl *Param,
-                      ArrayRef<TemplateArgument> TemplateArgs,
-                      SourceRange InstantiationRange)
-  : SemaRef(SemaRef)
-{
-  Initialize(ActiveTemplateInstantiation::DefaultFunctionArgumentInstantiation,
-             PointOfInstantiation, InstantiationRange,
-             Param, nullptr, TemplateArgs);
-}
-
-
-Sema::InstantiatingTemplate::
-InstantiatingTemplate(Sema &SemaRef, SourceLocation PointOfInstantiation,
-                      NamedDecl *Template, NonTypeTemplateParmDecl *Param,
-                      ArrayRef<TemplateArgument> TemplateArgs,
-                      SourceRange InstantiationRange)
-  : SemaRef(SemaRef)
-{
-  Initialize(ActiveTemplateInstantiation::PriorTemplateArgumentSubstitution,
-             PointOfInstantiation, InstantiationRange,
-             Param, Template, TemplateArgs);
-}
-
-Sema::InstantiatingTemplate::
-InstantiatingTemplate(Sema &SemaRef, SourceLocation PointOfInstantiation,
-                      NamedDecl *Template, TemplateTemplateParmDecl *Param,
-                      ArrayRef<TemplateArgument> TemplateArgs,
-                      SourceRange InstantiationRange)
-  : SemaRef(SemaRef)
-{
-  Initialize(ActiveTemplateInstantiation::PriorTemplateArgumentSubstitution,
-             PointOfInstantiation, InstantiationRange,
-             Param, Template, TemplateArgs);
-}
-
-Sema::InstantiatingTemplate::
-InstantiatingTemplate(Sema &SemaRef, SourceLocation PointOfInstantiation,
-                      TemplateDecl *Template, NamedDecl *Param,
-                      ArrayRef<TemplateArgument> TemplateArgs,
-                      SourceRange InstantiationRange)
-  : SemaRef(SemaRef)
-{
-  Initialize(ActiveTemplateInstantiation::DefaultTemplateArgumentChecking,
-             PointOfInstantiation, InstantiationRange,
-             Param, Template, TemplateArgs);
-}
+    : InstantiatingTemplate(
+          SemaRef,
+          ActiveTemplateInstantiation::DeducedTemplateArgumentSubstitution,
+          PointOfInstantiation, InstantiationRange, PartialSpec, nullptr,
+          TemplateArgs, &DeductionInfo) {}
+
+Sema::InstantiatingTemplate::InstantiatingTemplate(
+    Sema &SemaRef, SourceLocation PointOfInstantiation, ParmVarDecl *Param,
+    ArrayRef<TemplateArgument> TemplateArgs, SourceRange InstantiationRange)
+    : InstantiatingTemplate(
+          SemaRef,
+          ActiveTemplateInstantiation::DefaultFunctionArgumentInstantiation,
+          PointOfInstantiation, InstantiationRange, Param, nullptr,
+          TemplateArgs) {}
+
+Sema::InstantiatingTemplate::InstantiatingTemplate(
+    Sema &SemaRef, SourceLocation PointOfInstantiation, NamedDecl *Template,
+    NonTypeTemplateParmDecl *Param, ArrayRef<TemplateArgument> TemplateArgs,
+    SourceRange InstantiationRange)
+    : InstantiatingTemplate(
+          SemaRef,
+          ActiveTemplateInstantiation::PriorTemplateArgumentSubstitution,
+          PointOfInstantiation, InstantiationRange, Param, Template,
+          TemplateArgs) {}
+
+Sema::InstantiatingTemplate::InstantiatingTemplate(
+    Sema &SemaRef, SourceLocation PointOfInstantiation, NamedDecl *Template,
+    TemplateTemplateParmDecl *Param, ArrayRef<TemplateArgument> TemplateArgs,
+    SourceRange InstantiationRange)
+    : InstantiatingTemplate(
+          SemaRef,
+          ActiveTemplateInstantiation::PriorTemplateArgumentSubstitution,
+          PointOfInstantiation, InstantiationRange, Param, Template,
+          TemplateArgs) {}
+
+Sema::InstantiatingTemplate::InstantiatingTemplate(
+    Sema &SemaRef, SourceLocation PointOfInstantiation, TemplateDecl *Template,
+    NamedDecl *Param, ArrayRef<TemplateArgument> TemplateArgs,
+    SourceRange InstantiationRange)
+    : InstantiatingTemplate(
+          SemaRef, ActiveTemplateInstantiation::DefaultTemplateArgumentChecking,
+          PointOfInstantiation, InstantiationRange, Param, Template,
+          TemplateArgs) {}
 
 void Sema::InstantiatingTemplate::Clear() {
   if (!Invalid) {
@@ -734,6 +714,20 @@ namespace {
     }
 
     void transformedLocalDecl(Decl *Old, Decl *New) {
+      // If we've instantiated the call operator of a lambda or the call
+      // operator template of a generic lambda, update the "instantiation of"
+      // information.
+      auto *NewMD = dyn_cast<CXXMethodDecl>(New);
+      if (NewMD && isLambdaCallOperator(NewMD)) {
+        auto *OldMD = dyn_cast<CXXMethodDecl>(Old);
+        if (auto *NewTD = NewMD->getDescribedFunctionTemplate())
+          NewTD->setInstantiatedFromMemberTemplate(
+              OldMD->getDescribedFunctionTemplate());
+        else
+          NewMD->setInstantiationOfMemberFunction(OldMD,
+                                                  TSK_ImplicitInstantiation);
+      }
+      
       SemaRef.CurrentInstantiationScope->InstantiatedLocal(Old, New);
     }
     
@@ -836,28 +830,6 @@ namespace {
       return TreeTransform<TemplateInstantiator>::TransformLambdaExpr(E);
     }
 
-    ExprResult TransformLambdaScope(LambdaExpr *E,
-        CXXMethodDecl *NewCallOperator, 
-        ArrayRef<InitCaptureInfoTy> InitCaptureExprsAndTypes) {
-      CXXMethodDecl *const OldCallOperator = E->getCallOperator();   
-      // In the generic lambda case, we set the NewTemplate to be considered
-      // an "instantiation" of the OldTemplate.
-      if (FunctionTemplateDecl *const NewCallOperatorTemplate = 
-            NewCallOperator->getDescribedFunctionTemplate()) {
-        
-        FunctionTemplateDecl *const OldCallOperatorTemplate = 
-                              OldCallOperator->getDescribedFunctionTemplate();
-        NewCallOperatorTemplate->setInstantiatedFromMemberTemplate(
-                                                     OldCallOperatorTemplate);
-      } else 
-        // For a non-generic lambda we set the NewCallOperator to 
-        // be an instantiation of the OldCallOperator.
-        NewCallOperator->setInstantiationOfMemberFunction(OldCallOperator,
-                                                    TSK_ImplicitInstantiation);
-      
-      return inherited::TransformLambdaScope(E, NewCallOperator, 
-          InitCaptureExprsAndTypes);
-    }
     TemplateParameterList *TransformTemplateParameterList(
                               TemplateParameterList *OrigTPL)  {
       if (!OrigTPL || !OrigTPL->size()) return OrigTPL;
@@ -1767,7 +1739,7 @@ Sema::SubstBaseSpecifiers(CXXRecordDecl *Instantiation,
                           const MultiLevelTemplateArgumentList &TemplateArgs) {
   bool Invalid = false;
   SmallVector<CXXBaseSpecifier*, 4> InstantiatedBases;
-  for (const auto Base : Pattern->bases()) {
+  for (const auto &Base : Pattern->bases()) {
     if (!Base.getType()->isDependentType()) {
       if (const CXXRecordDecl *RD = Base.getType()->getAsCXXRecordDecl()) {
         if (RD->isInvalidDecl())
@@ -2063,6 +2035,10 @@ Sema::InstantiateClass(SourceLocation PointOfInstantiation,
               SourceLocation(), SourceLocation(), nullptr);
   CheckCompletedCXXClass(Instantiation);
 
+  // Default arguments are parsed, if not instantiated. We can go instantiate
+  // default arg exprs for default constructors if necessary now.
+  ActOnFinishCXXMemberDefaultArgs(Instantiation);
+
   // Instantiate late parsed attributes, and attach them to their decls.
   // See Sema::InstantiateAttrs
   for (LateInstantiatedAttrVec::iterator I = LateAttrs.begin(),
@@ -2257,7 +2233,7 @@ bool Sema::InstantiateInClassInitializer(
   EnterExpressionEvaluationContext EvalContext(*this,
                                                Sema::PotentiallyEvaluated);
 
-  LocalInstantiationScope Scope(*this);
+  LocalInstantiationScope Scope(*this, true);
 
   // Instantiate the initializer.
   ActOnStartCXXInClassMemberInitializer();
@@ -2812,6 +2788,16 @@ LocalInstantiationScope::findInstantiationOf(const Decl *D) {
       isa<TemplateTemplateParmDecl>(D))
     return nullptr;
 
+  // Local types referenced prior to definition may require instantiation.
+  if (const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(D))
+    if (RD->isLocalClass())
+      return nullptr;
+
+  // Enumeration types referenced prior to definition may appear as a result of
+  // error recovery.
+  if (isa<EnumDecl>(D))
+    return nullptr;
+
   // If we didn't find the decl, then we either have a sema bug, or we have a
   // forward reference to a label declaration.  Return null to indicate that
   // we have an uninstantiated label.
diff --git a/lib/Sema/SemaTemplateInstantiateDecl.cpp b/lib/Sema/SemaTemplateInstantiateDecl.cpp
index 40e86175b2cd..5c994f86bd2c 100644
--- a/lib/Sema/SemaTemplateInstantiateDecl.cpp
+++ b/lib/Sema/SemaTemplateInstantiateDecl.cpp
@@ -202,6 +202,31 @@ static void instantiateDependentEnableIfAttr(
   New->addAttr(EIA);
 }
 
+// Constructs and adds to New a new instance of CUDALaunchBoundsAttr using
+// template A as the base and arguments from TemplateArgs.
+static void instantiateDependentCUDALaunchBoundsAttr(
+    Sema &S, const MultiLevelTemplateArgumentList &TemplateArgs,
+    const CUDALaunchBoundsAttr &Attr, Decl *New) {
+  // The alignment expression is a constant expression.
+  EnterExpressionEvaluationContext Unevaluated(S, Sema::ConstantEvaluated);
+
+  ExprResult Result = S.SubstExpr(Attr.getMaxThreads(), TemplateArgs);
+  if (Result.isInvalid())
+    return;
+  Expr *MaxThreads = Result.getAs<Expr>();
+
+  Expr *MinBlocks = nullptr;
+  if (Attr.getMinBlocks()) {
+    Result = S.SubstExpr(Attr.getMinBlocks(), TemplateArgs);
+    if (Result.isInvalid())
+      return;
+    MinBlocks = Result.getAs<Expr>();
+  }
+
+  S.AddLaunchBoundsAttr(Attr.getLocation(), New, MaxThreads, MinBlocks,
+                        Attr.getSpellingListIndex());
+}
+
 void Sema::InstantiateAttrs(const MultiLevelTemplateArgumentList &TemplateArgs,
                             const Decl *Tmpl, Decl *New,
                             LateInstantiatedAttrVec *LateAttrs,
@@ -233,6 +258,13 @@ void Sema::InstantiateAttrs(const MultiLevelTemplateArgumentList &TemplateArgs,
       continue;
     }
 
+    if (const CUDALaunchBoundsAttr *CUDALaunchBounds =
+            dyn_cast<CUDALaunchBoundsAttr>(TmplAttr)) {
+      instantiateDependentCUDALaunchBoundsAttr(*this, TemplateArgs,
+                                               *CUDALaunchBounds, New);
+      continue;
+    }
+
     // Existing DLL attribute on the instantiation takes precedence.
     if (TmplAttr->getKind() == attr::DLLExport ||
         TmplAttr->getKind() == attr::DLLImport) {
@@ -289,6 +321,11 @@ TemplateDeclInstantiator::VisitTranslationUnitDecl(TranslationUnitDecl *D) {
 }
 
 Decl *
+TemplateDeclInstantiator::VisitExternCContextDecl(ExternCContextDecl *D) {
+  llvm_unreachable("extern \"C\" context cannot be instantiated");
+}
+
+Decl *
 TemplateDeclInstantiator::VisitLabelDecl(LabelDecl *D) {
   LabelDecl *Inst = LabelDecl::Create(SemaRef.Context, Owner, D->getLocation(),
                                       D->getIdentifier());
@@ -1265,11 +1302,19 @@ Decl *TemplateDeclInstantiator::VisitCXXRecordDecl(CXXRecordDecl *D) {
   // DR1484 clarifies that the members of a local class are instantiated as part
   // of the instantiation of their enclosing entity.
   if (D->isCompleteDefinition() && D->isLocalClass()) {
+    Sema::SavePendingLocalImplicitInstantiationsRAII
+        SavedPendingLocalImplicitInstantiations(SemaRef);
+
     SemaRef.InstantiateClass(D->getLocation(), Record, D, TemplateArgs,
                              TSK_ImplicitInstantiation,
                              /*Complain=*/true);
+
     SemaRef.InstantiateClassMembers(D->getLocation(), Record, TemplateArgs,
                                     TSK_ImplicitInstantiation);
+
+    // This class may have local implicit instantiations that need to be
+    // performed within this scope.
+    SemaRef.PerformPendingInstantiations(/*LocalOnly=*/true);
   }
 
   SemaRef.DiagnoseUnusedNestedTypedefs(Record);
@@ -3313,12 +3358,8 @@ void Sema::InstantiateFunctionDefinition(SourceLocation PointOfInstantiation,
   // it marks vtables used in late parsed templates as used.
   SavePendingLocalImplicitInstantiationsRAII
       SavedPendingLocalImplicitInstantiations(*this);
-  std::unique_ptr<SavePendingInstantiationsAndVTableUsesRAII>
-      SavePendingInstantiationsAndVTableUses;
-  if (Recursive) {
-    SavePendingInstantiationsAndVTableUses.reset(
-        new SavePendingInstantiationsAndVTableUsesRAII(*this));
-  }
+  SavePendingInstantiationsAndVTableUsesRAII
+      SavePendingInstantiationsAndVTableUses(*this, /*Enabled=*/Recursive);
 
   // Call the LateTemplateParser callback if there is a need to late parse
   // a templated function definition.
@@ -3463,8 +3504,8 @@ void Sema::InstantiateFunctionDefinition(SourceLocation PointOfInstantiation,
     // instantiation of this template.
     PerformPendingInstantiations();
 
-    // Restore PendingInstantiations and VTableUses.
-    SavePendingInstantiationsAndVTableUses.reset();
+    // PendingInstantiations and VTableUses are restored through
+    // SavePendingInstantiationsAndVTableUses's destructor.
   }
 }
 
@@ -3780,12 +3821,8 @@ void Sema::InstantiateVariableDefinition(SourceLocation PointOfInstantiation,
       // If we're performing recursive template instantiation, create our own
       // queue of pending implicit instantiations that we will instantiate
       // later, while we're still within our own instantiation context.
-      std::unique_ptr<SavePendingInstantiationsAndVTableUsesRAII>
-          SavePendingInstantiationsAndVTableUses;
-      if (Recursive) {
-        SavePendingInstantiationsAndVTableUses.reset(
-            new SavePendingInstantiationsAndVTableUsesRAII(*this));
-      }
+      SavePendingInstantiationsAndVTableUsesRAII
+          SavePendingInstantiationsAndVTableUses(*this, /*Enabled=*/Recursive);
 
       LocalInstantiationScope Local(*this);
 
@@ -3812,8 +3849,8 @@ void Sema::InstantiateVariableDefinition(SourceLocation PointOfInstantiation,
         // instantiation of this template.
         PerformPendingInstantiations();
 
-        // Restore PendingInstantiations and VTableUses.
-        SavePendingInstantiationsAndVTableUses.reset();
+        // PendingInstantiations and VTableUses are restored through
+        // SavePendingInstantiationsAndVTableUses's destructor.
       }
     }
 
@@ -3899,12 +3936,8 @@ void Sema::InstantiateVariableDefinition(SourceLocation PointOfInstantiation,
   // while we're still within our own instantiation context.
   SavePendingLocalImplicitInstantiationsRAII
       SavedPendingLocalImplicitInstantiations(*this);
-  std::unique_ptr<SavePendingInstantiationsAndVTableUsesRAII>
-      SavePendingInstantiationsAndVTableUses;
-  if (Recursive) {
-    SavePendingInstantiationsAndVTableUses.reset(
-        new SavePendingInstantiationsAndVTableUsesRAII(*this));
-  }
+  SavePendingInstantiationsAndVTableUsesRAII
+      SavePendingInstantiationsAndVTableUses(*this, /*Enabled=*/Recursive);
 
   // Enter the scope of this instantiation. We don't use
   // PushDeclContext because we don't have a scope.
@@ -3970,8 +4003,8 @@ void Sema::InstantiateVariableDefinition(SourceLocation PointOfInstantiation,
     // instantiation of this template.
     PerformPendingInstantiations();
 
-    // Restore PendingInstantiations and VTableUses.
-    SavePendingInstantiationsAndVTableUses.reset();
+    // PendingInstantiations and VTableUses are restored through
+    // SavePendingInstantiationsAndVTableUses's destructor.
   }
 }
 
@@ -4409,6 +4442,30 @@ NamedDecl *Sema::FindInstantiatedDecl(SourceLocation Loc, NamedDecl *D,
     if (D->isInvalidDecl())
       return nullptr;
 
+    // Normally this function only searches for already instantiated declaration
+    // however we have to make an exclusion for local types used before
+    // definition as in the code:
+    //
+    //   template<typename T> void f1() {
+    //     void g1(struct x1);
+    //     struct x1 {};
+    //   }
+    //
+    // In this case instantiation of the type of 'g1' requires definition of
+    // 'x1', which is defined later. Error recovery may produce an enum used
+    // before definition. In these cases we need to instantiate relevant
+    // declarations here.
+    bool NeedInstantiate = false;
+    if (CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(D))
+      NeedInstantiate = RD->isLocalClass();
+    else
+      NeedInstantiate = isa<EnumDecl>(D);
+    if (NeedInstantiate) {
+      Decl *Inst = SubstDecl(D, CurContext, TemplateArgs);
+      CurrentInstantiationScope->InstantiatedLocal(D, Inst);
+      return cast<TypeDecl>(Inst);
+    }
+
     // If we didn't find the decl, then we must have a label decl that hasn't
     // been found yet.  Lazily instantiate it and return it now.
     assert(isa<LabelDecl>(D));
diff --git a/lib/Sema/SemaTemplateVariadic.cpp b/lib/Sema/SemaTemplateVariadic.cpp
index e4fab71d995b..fd3ba3549b4c 100644
--- a/lib/Sema/SemaTemplateVariadic.cpp
+++ b/lib/Sema/SemaTemplateVariadic.cpp
@@ -251,15 +251,10 @@ Sema::DiagnoseUnexpandedParameterPacks(SourceLocation Loc,
       Locations.push_back(Unexpanded[I].second);
   }
 
-  DiagnosticBuilder DB
-    = Names.size() == 0? Diag(Loc, diag::err_unexpanded_parameter_pack_0)
-                           << (int)UPPC
-    : Names.size() == 1? Diag(Loc, diag::err_unexpanded_parameter_pack_1)
-                           << (int)UPPC << Names[0]
-    : Names.size() == 2? Diag(Loc, diag::err_unexpanded_parameter_pack_2)
-                           << (int)UPPC << Names[0] << Names[1]
-    : Diag(Loc, diag::err_unexpanded_parameter_pack_3_or_more)
-        << (int)UPPC << Names[0] << Names[1];
+  DiagnosticBuilder DB = Diag(Loc, diag::err_unexpanded_parameter_pack)
+                         << (int)UPPC << (int)Names.size();
+  for (size_t I = 0, E = std::min(Names.size(), (size_t)2); I != E; ++I)
+    DB << Names[I];
 
   for (unsigned I = 0, N = Locations.size(); I != N; ++I)
     DB << SourceRange(Locations[I]);
diff --git a/lib/Sema/SemaType.cpp b/lib/Sema/SemaType.cpp
index 0f96a1cbce35..57a4689c6d72 100644
--- a/lib/Sema/SemaType.cpp
+++ b/lib/Sema/SemaType.cpp
@@ -112,7 +112,6 @@ static void diagnoseBadTypeAttribute(Sema &S, const AttributeList &attr,
     case AttributeList::AT_SysVABI: \
     case AttributeList::AT_Regparm: \
     case AttributeList::AT_Pcs: \
-    case AttributeList::AT_PnaclCall: \
     case AttributeList::AT_IntelOclBicc
 
 // Microsoft-specific type qualifiers.
@@ -689,6 +688,31 @@ static void maybeSynthesizeBlockSignature(TypeProcessingState &state,
   state.setCurrentChunkIndex(declarator.getNumTypeObjects());
 }
 
+static void diagnoseAndRemoveTypeQualifiers(Sema &S, const DeclSpec &DS,
+                                            unsigned &TypeQuals,
+                                            QualType TypeSoFar,
+                                            unsigned RemoveTQs,
+                                            unsigned DiagID) {
+  // If this occurs outside a template instantiation, warn the user about
+  // it; they probably didn't mean to specify a redundant qualifier.
+  typedef std::pair<DeclSpec::TQ, SourceLocation> QualLoc;
+  for (QualLoc Qual : {QualLoc(DeclSpec::TQ_const, DS.getConstSpecLoc()),
+                       QualLoc(DeclSpec::TQ_volatile, DS.getVolatileSpecLoc()),
+                       QualLoc(DeclSpec::TQ_atomic, DS.getAtomicSpecLoc())}) {
+    if (!(RemoveTQs & Qual.first))
+      continue;
+
+    if (S.ActiveTemplateInstantiations.empty()) {
+      if (TypeQuals & Qual.first)
+        S.Diag(Qual.second, DiagID)
+          << DeclSpec::getSpecifierName(Qual.first) << TypeSoFar
+          << FixItHint::CreateRemoval(Qual.second);
+    }
+
+    TypeQuals &= ~Qual.first;
+  }
+}
+
 /// \brief Convert the specified declspec to the appropriate type
 /// object.
 /// \param state Specifies the declarator containing the declaration specifier
@@ -869,8 +893,11 @@ static QualType ConvertDeclSpecToType(TypeProcessingState &state) {
     else
       Result = Context.DoubleTy;
 
-    if (S.getLangOpts().OpenCL && !S.getOpenCLOptions().cl_khr_fp64) {
-      S.Diag(DS.getTypeSpecTypeLoc(), diag::err_double_requires_fp64);
+    if (S.getLangOpts().OpenCL &&
+        !((S.getLangOpts().OpenCLVersion >= 120) ||
+          S.getOpenCLOptions().cl_khr_fp64)) {
+      S.Diag(DS.getTypeSpecTypeLoc(), diag::err_type_requires_extension)
+          << Result << "cl_khr_fp64";
       declarator.setInvalidType(true);
     }
     break;
@@ -946,6 +973,30 @@ static QualType ConvertDeclSpecToType(TypeProcessingState &state) {
           << DS.getSourceRange();
         declarator.setInvalidType(true);
       }
+    } else if (S.getLangOpts().OpenCL) {
+      if (const AtomicType *AT = Result->getAs<AtomicType>()) {
+        const BuiltinType *BT = AT->getValueType()->getAs<BuiltinType>();
+        bool NoExtTypes = BT && (BT->getKind() == BuiltinType::Int ||
+                                 BT->getKind() == BuiltinType::UInt ||
+                                 BT->getKind() == BuiltinType::Float);
+        if (!S.getOpenCLOptions().cl_khr_int64_base_atomics && !NoExtTypes) {
+          S.Diag(DS.getTypeSpecTypeLoc(), diag::err_type_requires_extension)
+              << Result << "cl_khr_int64_base_atomics";
+          declarator.setInvalidType(true);
+        }
+        if (!S.getOpenCLOptions().cl_khr_int64_extended_atomics &&
+            !NoExtTypes) {
+          S.Diag(DS.getTypeSpecTypeLoc(), diag::err_type_requires_extension)
+              << Result << "cl_khr_int64_extended_atomics";
+          declarator.setInvalidType(true);
+        }
+        if (!S.getOpenCLOptions().cl_khr_fp64 && BT &&
+            BT->getKind() == BuiltinType::Double) {
+          S.Diag(DS.getTypeSpecTypeLoc(), diag::err_type_requires_extension)
+              << Result << "cl_khr_fp64";
+          declarator.setInvalidType(true);
+        }
+      }
     }
 
     // TypeQuals handled by caller.
@@ -1091,24 +1142,22 @@ static QualType ConvertDeclSpecToType(TypeProcessingState &state) {
 
   // Apply const/volatile/restrict qualifiers to T.
   if (unsigned TypeQuals = DS.getTypeQualifiers()) {
-
-    // Warn about CV qualifiers on functions: C99 6.7.3p8: "If the specification
-    // of a function type includes any type qualifiers, the behavior is
-    // undefined."
-    if (Result->isFunctionType() && TypeQuals) {
-      if (TypeQuals & DeclSpec::TQ_const)
-        S.Diag(DS.getConstSpecLoc(), diag::warn_typecheck_function_qualifiers)
-          << Result << DS.getSourceRange();
-      else if (TypeQuals & DeclSpec::TQ_volatile)
-        S.Diag(DS.getVolatileSpecLoc(),
-               diag::warn_typecheck_function_qualifiers)
-            << Result << DS.getSourceRange();
-      else {
-        assert((TypeQuals & (DeclSpec::TQ_restrict | DeclSpec::TQ_atomic)) &&
-               "Has CVRA quals but not C, V, R, or A?");
-        // No diagnostic; we'll diagnose 'restrict' or '_Atomic' applied to a
-        // function type later, in BuildQualifiedType.
-      }
+    // Warn about CV qualifiers on function types.
+    // C99 6.7.3p8:
+    //   If the specification of a function type includes any type qualifiers,
+    //   the behavior is undefined.
+    // C++11 [dcl.fct]p7:
+    //   The effect of a cv-qualifier-seq in a function declarator is not the
+    //   same as adding cv-qualification on top of the function type. In the
+    //   latter case, the cv-qualifiers are ignored.
+    if (TypeQuals && Result->isFunctionType()) {
+      diagnoseAndRemoveTypeQualifiers(
+          S, DS, TypeQuals, Result, DeclSpec::TQ_const | DeclSpec::TQ_volatile,
+          S.getLangOpts().CPlusPlus
+              ? diag::warn_typecheck_function_qualifiers_ignored
+              : diag::warn_typecheck_function_qualifiers_unspecified);
+      // No diagnostic for 'restrict' or '_Atomic' applied to a
+      // function type; we'll diagnose those later, in BuildQualifiedType.
     }
 
     // C++11 [dcl.ref]p1:
@@ -1119,25 +1168,11 @@ static QualType ConvertDeclSpecToType(TypeProcessingState &state) {
     // There don't appear to be any other contexts in which a cv-qualified
     // reference type could be formed, so the 'ill-formed' clause here appears
     // to never happen.
-    if (DS.getTypeSpecType() == DeclSpec::TST_typename &&
-        TypeQuals && Result->isReferenceType()) {
-      // If this occurs outside a template instantiation, warn the user about
-      // it; they probably didn't mean to specify a redundant qualifier.
-      typedef std::pair<DeclSpec::TQ, SourceLocation> QualLoc;
-      QualLoc Quals[] = {
-        QualLoc(DeclSpec::TQ_const, DS.getConstSpecLoc()),
-        QualLoc(DeclSpec::TQ_volatile, DS.getVolatileSpecLoc()),
-        QualLoc(DeclSpec::TQ_atomic, DS.getAtomicSpecLoc())
-      };
-      for (unsigned I = 0, N = llvm::array_lengthof(Quals); I != N; ++I) {
-        if (S.ActiveTemplateInstantiations.empty()) {
-          if (TypeQuals & Quals[I].first)
-            S.Diag(Quals[I].second, diag::warn_typecheck_reference_qualifiers)
-              << DeclSpec::getSpecifierName(Quals[I].first) << Result
-              << FixItHint::CreateRemoval(Quals[I].second);
-        }
-        TypeQuals &= ~Quals[I].first;
-      }
+    if (TypeQuals && Result->isReferenceType()) {
+      diagnoseAndRemoveTypeQualifiers(
+          S, DS, TypeQuals, Result,
+          DeclSpec::TQ_const | DeclSpec::TQ_volatile | DeclSpec::TQ_atomic,
+          diag::warn_typecheck_reference_qualifiers);
     }
 
     // C90 6.5.3 constraints: "The same type qualifier shall not appear more
@@ -1809,13 +1844,7 @@ QualType Sema::BuildMemberPointerType(QualType T, QualType Class,
   // exception specification.
   if (CheckDistantExceptionSpec(T)) {
     Diag(Loc, diag::err_distant_exception_spec);
-
-    // FIXME: If we're doing this as part of template instantiation,
-    // we should return immediately.
-
-    // Build the type anyway, but use the canonical type so that the
-    // exception specifiers are stripped off.
-    T = Context.getCanonicalType(T);
+    return QualType();
   }
 
   // C++ 8.3.3p3: A pointer to member shall not point to ... a member
@@ -2463,6 +2492,10 @@ getCCForDeclaratorChunk(Sema &S, Declarator &D,
       // in a member pointer.
       IsCXXInstanceMethod =
           D.getTypeObject(I).Kind == DeclaratorChunk::MemberPointer;
+    } else if (D.getContext() == Declarator::LambdaExprContext) {
+      // This can only be a call operator for a lambda, which is an instance
+      // method.
+      IsCXXInstanceMethod = true;
     } else {
       // We're the innermost decl chunk, so must be a function declarator.
       assert(D.isFunctionDeclarator());
@@ -2476,8 +2509,24 @@ getCCForDeclaratorChunk(Sema &S, Declarator &D,
     }
   }
 
-  return S.Context.getDefaultCallingConvention(FTI.isVariadic,
-                                               IsCXXInstanceMethod);
+  CallingConv CC = S.Context.getDefaultCallingConvention(FTI.isVariadic,
+                                                         IsCXXInstanceMethod);
+
+  // Attribute AT_OpenCLKernel affects the calling convention only on
+  // the SPIR target, hence it cannot be treated as a calling
+  // convention attribute. This is the simplest place to infer
+  // "spir_kernel" for OpenCL kernels on SPIR.
+  if (CC == CC_SpirFunction) {
+    for (const AttributeList *Attr = D.getDeclSpec().getAttributes().getList();
+         Attr; Attr = Attr->getNext()) {
+      if (Attr->getKind() == AttributeList::AT_OpenCLKernel) {
+        CC = CC_SpirKernel;
+        break;
+      }
+    }
+  }
+
+  return CC;
 }
 
 static TypeSourceInfo *GetFullTypeForDeclarator(TypeProcessingState &state,
@@ -2792,8 +2841,16 @@ static TypeSourceInfo *GetFullTypeForDeclarator(TypeProcessingState &state,
       // class type in C++.
       if ((T.getCVRQualifiers() || T->isAtomicType()) &&
           !(S.getLangOpts().CPlusPlus &&
-            (T->isDependentType() || T->isRecordType())))
-        diagnoseRedundantReturnTypeQualifiers(S, T, D, chunkIndex);
+            (T->isDependentType() || T->isRecordType()))) {
+	if (T->isVoidType() && !S.getLangOpts().CPlusPlus &&
+	    D.getFunctionDefinitionKind() == FDK_Definition) {
+	  // [6.9.1/3] qualified void return is invalid on a C
+	  // function definition.  Apparently ok on declarations and
+	  // in C++ though (!)
+	  S.Diag(DeclType.Loc, diag::err_func_returning_qualified_void) << T;
+	} else
+	  diagnoseRedundantReturnTypeQualifiers(S, T, D, chunkIndex);
+      }
 
       // Objective-C ARC ownership qualifiers are ignored on the function
       // return type (by type canonicalization). Complain if this attribute
@@ -3424,8 +3481,6 @@ static AttributeList::Kind getAttrListKind(AttributedType::Kind kind) {
   case AttributedType::attr_pcs:
   case AttributedType::attr_pcs_vfp:
     return AttributeList::AT_Pcs;
-  case AttributedType::attr_pnaclcall:
-    return AttributeList::AT_PnaclCall;
   case AttributedType::attr_inteloclbicc:
     return AttributeList::AT_IntelOclBicc;
   case AttributedType::attr_ms_abi:
@@ -4454,8 +4509,6 @@ static AttributedType::Kind getCCTypeAttrKind(AttributeList &Attr) {
         .Case("aapcs", AttributedType::attr_pcs)
         .Case("aapcs-vfp", AttributedType::attr_pcs_vfp);
   }
-  case AttributeList::AT_PnaclCall:
-    return AttributedType::attr_pnaclcall;
   case AttributeList::AT_IntelOclBicc:
     return AttributedType::attr_inteloclbicc;
   case AttributeList::AT_MSABI:
@@ -5086,14 +5139,18 @@ bool Sema::RequireCompleteType(SourceLocation Loc, QualType T,
 /// \param D The definition of the entity.
 /// \param Suggested Filled in with the declaration that should be made visible
 ///        in order to provide a definition of this entity.
-static bool hasVisibleDefinition(Sema &S, NamedDecl *D, NamedDecl **Suggested) {
+bool Sema::hasVisibleDefinition(NamedDecl *D, NamedDecl **Suggested) {
   // Easy case: if we don't have modules, all declarations are visible.
-  if (!S.getLangOpts().Modules)
+  if (!getLangOpts().Modules)
     return true;
 
   // If this definition was instantiated from a template, map back to the
   // pattern from which it was instantiated.
-  if (auto *RD = dyn_cast<CXXRecordDecl>(D)) {
+  if (isa<TagDecl>(D) && cast<TagDecl>(D)->isBeingDefined()) {
+    // We're in the middle of defining it; this definition should be treated
+    // as visible.
+    return true;
+  } else if (auto *RD = dyn_cast<CXXRecordDecl>(D)) {
     if (auto *Pattern = RD->getTemplateInstantiationPattern())
       RD = Pattern;
     D = RD->getDefinition();
@@ -5104,7 +5161,7 @@ static bool hasVisibleDefinition(Sema &S, NamedDecl *D, NamedDecl **Suggested) {
       // If the enum has a fixed underlying type, any declaration of it will do.
       *Suggested = nullptr;
       for (auto *Redecl : ED->redecls()) {
-        if (LookupResult::isVisible(S, Redecl))
+        if (LookupResult::isVisible(*this, Redecl))
           return true;
         if (Redecl->isThisDeclarationADefinition() ||
             (Redecl->isCanonicalDecl() && !*Suggested))
@@ -5119,7 +5176,7 @@ static bool hasVisibleDefinition(Sema &S, NamedDecl *D, NamedDecl **Suggested) {
   // FIXME: If we merged any other decl into D, and that declaration is visible,
   // then we should consider a definition to be visible.
   *Suggested = D;
-  return LookupResult::isVisible(S, D);
+  return LookupResult::isVisible(*this, D);
 }
 
 /// Locks in the inheritance model for the given class and all of its bases.
@@ -5170,13 +5227,13 @@ bool Sema::RequireCompleteTypeImpl(SourceLocation Loc, QualType T,
     // If we know about the definition but it is not visible, complain.
     NamedDecl *SuggestedDef = nullptr;
     if (!Diagnoser.Suppressed && Def &&
-        !hasVisibleDefinition(*this, Def, &SuggestedDef)) {
+        !hasVisibleDefinition(Def, &SuggestedDef)) {
       // Suppress this error outside of a SFINAE context if we've already
       // emitted the error once for this type. There's no usefulness in
       // repeating the diagnostic.
       // FIXME: Add a Fix-It that imports the corresponding module or includes
       // the header.
-      Module *Owner = SuggestedDef->getOwningModule();
+      Module *Owner = getOwningModule(SuggestedDef);
       Diag(Loc, diag::err_module_private_definition)
         << T << Owner->getFullModuleName();
       Diag(SuggestedDef->getLocation(), diag::note_previous_definition);
diff --git a/lib/Sema/TreeTransform.h b/lib/Sema/TreeTransform.h
index 36abbb624af7..f5249fdeb017 100644
--- a/lib/Sema/TreeTransform.h
+++ b/lib/Sema/TreeTransform.h
@@ -619,11 +619,6 @@ public:
 
   StmtResult TransformCompoundStmt(CompoundStmt *S, bool IsStmtExpr);
   ExprResult TransformCXXNamedCastExpr(CXXNamedCastExpr *E);
-  
-  typedef std::pair<ExprResult, QualType> InitCaptureInfoTy;
-  /// \brief Transform the captures and body of a lambda expression.
-  ExprResult TransformLambdaScope(LambdaExpr *E, CXXMethodDecl *CallOperator, 
-       ArrayRef<InitCaptureInfoTy> InitCaptureExprsAndTypes);
 
   TemplateParameterList *TransformTemplateParameterList(
         TemplateParameterList *TPL) {
@@ -1702,7 +1697,7 @@ public:
   }
 
   StmtResult RebuildSEHFinallyStmt(SourceLocation Loc, Stmt *Block) {
-    return getSema().ActOnSEHFinallyBlock(Loc, Block);
+    return SEHFinallyStmt::Create(getSema().getASTContext(), Loc, Block);
   }
 
   /// \brief Build a new predefined expression.
@@ -1870,11 +1865,9 @@ public:
         return ExprError();
       Base = BaseResult.get();
       ExprValueKind VK = isArrow ? VK_LValue : Base->getValueKind();
-      MemberExpr *ME =
-        new (getSema().Context) MemberExpr(Base, isArrow,
-                                           Member, MemberNameInfo,
-                                           cast<FieldDecl>(Member)->getType(),
-                                           VK, OK_Ordinary);
+      MemberExpr *ME = new (getSema().Context)
+          MemberExpr(Base, isArrow, OpLoc, Member, MemberNameInfo,
+                     cast<FieldDecl>(Member)->getType(), VK, OK_Ordinary);
       return ME;
     }
 
@@ -2625,6 +2618,31 @@ public:
                                         RBracLoc, Args);
   }
 
+  /// \brief Build a new Objective-C instance/class message to 'super'.
+  ExprResult RebuildObjCMessageExpr(SourceLocation SuperLoc,
+                                    Selector Sel,
+                                    ArrayRef<SourceLocation> SelectorLocs,
+                                    ObjCMethodDecl *Method,
+                                    SourceLocation LBracLoc,
+                                    MultiExprArg Args,
+                                    SourceLocation RBracLoc) {
+    ObjCInterfaceDecl *Class = Method->getClassInterface();
+    QualType ReceiverTy = SemaRef.Context.getObjCInterfaceType(Class);
+    
+    return Method->isInstanceMethod() ? SemaRef.BuildInstanceMessage(nullptr,
+                                          ReceiverTy,
+                                          SuperLoc,
+                                          Sel, Method, LBracLoc, SelectorLocs,
+                                          RBracLoc, Args)
+                                      : SemaRef.BuildClassMessage(nullptr,
+                                          ReceiverTy,
+                                          SuperLoc,
+                                          Sel, Method, LBracLoc, SelectorLocs,
+                                          RBracLoc, Args);
+
+      
+  }
+
   /// \brief Build a new Objective-C ivar reference expression.
   ///
   /// By default, performs semantic analysis to build the new expression.
@@ -6647,7 +6665,16 @@ StmtResult TreeTransform<Derived>::TransformOMPExecutableDirective(
     if (!D->getAssociatedStmt()) {
       return StmtError();
     }
-    AssociatedStmt = getDerived().TransformStmt(D->getAssociatedStmt());
+    getDerived().getSema().ActOnOpenMPRegionStart(D->getDirectiveKind(),
+                                                  /*CurScope=*/nullptr);
+    StmtResult Body;
+    {
+      Sema::CompoundScopeRAII CompoundScope(getSema());
+      Body = getDerived().TransformStmt(
+          cast<CapturedStmt>(D->getAssociatedStmt())->getCapturedStmt());
+    }
+    AssociatedStmt =
+        getDerived().getSema().ActOnOpenMPRegionEnd(Body, TClauses);
     if (AssociatedStmt.isInvalid()) {
       return StmtError();
     }
@@ -7834,6 +7861,9 @@ TreeTransform<Derived>::TransformExtVectorElementExpr(ExtVectorElementExpr *E) {
 template<typename Derived>
 ExprResult
 TreeTransform<Derived>::TransformInitListExpr(InitListExpr *E) {
+  if (InitListExpr *Syntactic = E->getSyntacticForm())
+    E = Syntactic;
+
   bool InitChanged = false;
 
   SmallVector<Expr*, 4> Inits;
@@ -7841,8 +7871,12 @@ TreeTransform<Derived>::TransformInitListExpr(InitListExpr *E) {
                                   Inits, &InitChanged))
     return ExprError();
 
-  if (!getDerived().AlwaysRebuild() && !InitChanged)
-    return E;
+  if (!getDerived().AlwaysRebuild() && !InitChanged) {
+    // FIXME: Attempt to reuse the existing syntactic form of the InitListExpr
+    // in some cases. We can't reuse it in general, because the syntactic and
+    // semantic forms are linked, and we can't know that semantic form will
+    // match even if the syntactic form does.
+  }
 
   return getDerived().RebuildInitList(E->getLBraceLoc(), Inits,
                                       E->getRBraceLoc(), E->getType());
@@ -9092,13 +9126,14 @@ ExprResult
 TreeTransform<Derived>::TransformLambdaExpr(LambdaExpr *E) {
   // Transform any init-capture expressions before entering the scope of the
   // lambda body, because they are not semantically within that scope.
+  typedef std::pair<ExprResult, QualType> InitCaptureInfoTy;
   SmallVector<InitCaptureInfoTy, 8> InitCaptureExprsAndTypes;
   InitCaptureExprsAndTypes.resize(E->explicit_capture_end() -
-      E->explicit_capture_begin());
+                                  E->explicit_capture_begin());
   for (LambdaExpr::capture_iterator C = E->capture_begin(),
                                     CEnd = E->capture_end();
        C != CEnd; ++C) {
-    if (!C->isInitCapture())
+    if (!E->isInitCapture(C))
       continue;
     EnterExpressionEvaluationContext EEEC(getSema(),
                                           Sema::PotentiallyEvaluated);
@@ -9120,12 +9155,9 @@ TreeTransform<Derived>::TransformLambdaExpr(LambdaExpr *E) {
         std::make_pair(NewExprInitResult, NewInitCaptureType);
   }
 
-  LambdaScopeInfo *LSI = getSema().PushLambdaScope();
-  Sema::FunctionScopeRAII FuncScopeCleanup(getSema());
-
   // Transform the template parameters, and add them to the current
   // instantiation scope. The null case is handled correctly.
-  LSI->GLTemplateParameterList = getDerived().TransformTemplateParameterList(
+  auto TPL = getDerived().TransformTemplateParameterList(
       E->getTemplateParameterList());
 
   // Transform the type of the original lambda's call operator.
@@ -9153,6 +9185,10 @@ TreeTransform<Derived>::TransformLambdaExpr(LambdaExpr *E) {
                                                         NewCallOpType);
   }
 
+  LambdaScopeInfo *LSI = getSema().PushLambdaScope();
+  Sema::FunctionScopeRAII FuncScopeCleanup(getSema());
+  LSI->GLTemplateParameterList = TPL;
+
   // Create the local class that will describe the lambda.
   CXXRecordDecl *Class
     = getSema().createLambdaClosureType(E->getIntroducerRange(),
@@ -9169,34 +9205,22 @@ TreeTransform<Derived>::TransformLambdaExpr(LambdaExpr *E) {
   LSI->CallOperator = NewCallOperator;
 
   getDerived().transformAttrs(E->getCallOperator(), NewCallOperator);
-
-  // TransformLambdaScope will manage the function scope, so we can disable the
-  // cleanup.
-  FuncScopeCleanup.disable();
-
-  return getDerived().TransformLambdaScope(E, NewCallOperator, 
-      InitCaptureExprsAndTypes);
-}
-
-template<typename Derived>
-ExprResult
-TreeTransform<Derived>::TransformLambdaScope(LambdaExpr *E,
-    CXXMethodDecl *CallOperator, 
-    ArrayRef<InitCaptureInfoTy> InitCaptureExprsAndTypes) {
-  bool Invalid = false;
+  getDerived().transformedLocalDecl(E->getCallOperator(), NewCallOperator);
 
   // Introduce the context of the call operator.
-  Sema::ContextRAII SavedContext(getSema(), CallOperator,
+  Sema::ContextRAII SavedContext(getSema(), NewCallOperator,
                                  /*NewThisContext*/false);
 
-  LambdaScopeInfo *const LSI = getSema().getCurLambda();
   // Enter the scope of the lambda.
-  getSema().buildLambdaScope(LSI, CallOperator, E->getIntroducerRange(),
-                                 E->getCaptureDefault(),
-                                 E->getCaptureDefaultLoc(),
-                                 E->hasExplicitParameters(),
-                                 E->hasExplicitResultType(),
-                                 E->isMutable());
+  getSema().buildLambdaScope(LSI, NewCallOperator,
+                             E->getIntroducerRange(),
+                             E->getCaptureDefault(),
+                             E->getCaptureDefaultLoc(),
+                             E->hasExplicitParameters(),
+                             E->hasExplicitResultType(),
+                             E->isMutable());
+
+  bool Invalid = false;
 
   // Transform captures.
   bool FinishedExplicitCaptures = false;
@@ -9221,8 +9245,7 @@ TreeTransform<Derived>::TransformLambdaScope(LambdaExpr *E,
       continue;
 
     // Rebuild init-captures, including the implied field declaration.
-    if (C->isInitCapture()) {
-      
+    if (E->isInitCapture(C)) {
       InitCaptureInfoTy InitExprTypePair = 
           InitCaptureExprsAndTypes[C - E->capture_begin()];
       ExprResult Init = InitExprTypePair.first;
@@ -9309,28 +9332,34 @@ TreeTransform<Derived>::TransformLambdaScope(LambdaExpr *E,
   if (!FinishedExplicitCaptures)
     getSema().finishLambdaExplicitCaptures(LSI);
 
-
   // Enter a new evaluation context to insulate the lambda from any
   // cleanups from the enclosing full-expression.
   getSema().PushExpressionEvaluationContext(Sema::PotentiallyEvaluated);
 
-  if (Invalid) {
-    getSema().ActOnLambdaError(E->getLocStart(), /*CurScope=*/nullptr,
-                               /*IsInstantiation=*/true);
-    return ExprError();
-  }
-
   // Instantiate the body of the lambda expression.
-  StmtResult Body = getDerived().TransformStmt(E->getBody());
+  StmtResult Body =
+      Invalid ? StmtError() : getDerived().TransformStmt(E->getBody());
+
+  // ActOnLambda* will pop the function scope for us.
+  FuncScopeCleanup.disable();
+
   if (Body.isInvalid()) {
+    SavedContext.pop();
     getSema().ActOnLambdaError(E->getLocStart(), /*CurScope=*/nullptr,
                                /*IsInstantiation=*/true);
     return ExprError();
   }
 
-  return getSema().ActOnLambdaExpr(E->getLocStart(), Body.get(),
-                                   /*CurScope=*/nullptr,
-                                   /*IsInstantiation=*/true);
+  // Copy the LSI before ActOnFinishFunctionBody removes it.
+  // FIXME: This is dumb. Store the lambda information somewhere that outlives
+  // the call operator.
+  auto LSICopy = *LSI;
+  getSema().ActOnFinishFunctionBody(NewCallOperator, Body.get(),
+                                    /*IsInstantiation*/ true);
+  SavedContext.pop();
+
+  return getSema().BuildLambdaExpr(E->getLocStart(), Body.get()->getLocEnd(),
+                                   &LSICopy);
 }
 
 template<typename Derived>
@@ -10042,6 +10071,19 @@ TreeTransform<Derived>::TransformObjCMessageExpr(ObjCMessageExpr *E) {
                                                Args,
                                                E->getRightLoc());
   }
+  else if (E->getReceiverKind() == ObjCMessageExpr::SuperClass ||
+           E->getReceiverKind() == ObjCMessageExpr::SuperInstance) {
+    // Build a new class message send to 'super'.
+    SmallVector<SourceLocation, 16> SelLocs;
+    E->getSelectorLocs(SelLocs);
+    return getDerived().RebuildObjCMessageExpr(E->getSuperLoc(),
+                                               E->getSelector(),
+                                               SelLocs,
+                                               E->getMethodDecl(),
+                                               E->getLeftLoc(),
+                                               Args,
+                                               E->getRightLoc());
+  }
 
   // Instance message: transform the receiver
   assert(E->getReceiverKind() == ObjCMessageExpr::Instance &&
@@ -10720,11 +10762,9 @@ TreeTransform<Derived>::RebuildCXXPseudoDestructorExpr(Expr *Base,
        !BaseType->getAs<PointerType>()->getPointeeType()
                                               ->template getAs<RecordType>())){
     // This pseudo-destructor expression is still a pseudo-destructor.
-    return SemaRef.BuildPseudoDestructorExpr(Base, OperatorLoc,
-                                             isArrow? tok::arrow : tok::period,
-                                             SS, ScopeType, CCLoc, TildeLoc,
-                                             Destroyed,
-                                             /*FIXME?*/true);
+    return SemaRef.BuildPseudoDestructorExpr(
+        Base, OperatorLoc, isArrow ? tok::arrow : tok::period, SS, ScopeType,
+        CCLoc, TildeLoc, Destroyed);
   }
 
   TypeSourceInfo *DestroyedType = Destroyed.getTypeSourceInfo();
diff --git a/lib/Serialization/ASTCommon.cpp b/lib/Serialization/ASTCommon.cpp
index 13393225b60e..85c574c2021a 100644
--- a/lib/Serialization/ASTCommon.cpp
+++ b/lib/Serialization/ASTCommon.cpp
@@ -94,6 +94,7 @@ serialization::getDefinitiveDeclContext(const DeclContext *DC) {
   switch (DC->getDeclKind()) {
   // These entities may have multiple definitions.
   case Decl::TranslationUnit:
+  case Decl::ExternCContext:
   case Decl::Namespace:
   case Decl::LinkageSpec:
     return nullptr;
@@ -149,7 +150,11 @@ serialization::getDefinitiveDeclContext(const DeclContext *DC) {
 
 bool serialization::isRedeclarableDeclKind(unsigned Kind) {
   switch (static_cast<Decl::Kind>(Kind)) {
-  case Decl::TranslationUnit: // Special case of a "merged" declaration.
+  case Decl::TranslationUnit:
+  case Decl::ExternCContext:
+    // Special case of a "merged" declaration.
+    return true;
+
   case Decl::Namespace:
   case Decl::NamespaceAlias:
   case Decl::Typedef:
@@ -223,6 +228,24 @@ bool serialization::isRedeclarableDeclKind(unsigned Kind) {
 }
 
 bool serialization::needsAnonymousDeclarationNumber(const NamedDecl *D) {
+  // Friend declarations in dependent contexts aren't anonymous in the usual
+  // sense, but they cannot be found by name lookup in their semantic context
+  // (or indeed in any context), so we treat them as anonymous.
+  //
+  // This doesn't apply to friend tag decls; Sema makes those available to name
+  // lookup in the surrounding context.
+  if (D->getFriendObjectKind() &&
+      D->getLexicalDeclContext()->isDependentContext() && !isa<TagDecl>(D)) {
+    // For function templates and class templates, the template is numbered and
+    // not its pattern.
+    if (auto *FD = dyn_cast<FunctionDecl>(D))
+      return !FD->getDescribedFunctionTemplate();
+    if (auto *RD = dyn_cast<CXXRecordDecl>(D))
+      return !RD->getDescribedClassTemplate();
+    return true;
+  }
+
+  // Otherwise, we only care about anonymous class members.
   if (D->getDeclName() || !isa<CXXRecordDecl>(D->getLexicalDeclContext()))
     return false;
   return isa<TagDecl>(D) || isa<FieldDecl>(D);
diff --git a/lib/Serialization/ASTCommon.h b/lib/Serialization/ASTCommon.h
index 38a0ff5fbd4e..79d1817159c6 100644
--- a/lib/Serialization/ASTCommon.h
+++ b/lib/Serialization/ASTCommon.h
@@ -15,6 +15,7 @@
 #define LLVM_CLANG_LIB_SERIALIZATION_ASTCOMMON_H
 
 #include "clang/AST/ASTContext.h"
+#include "clang/AST/DeclFriend.h"
 #include "clang/Serialization/ASTBitCodes.h"
 
 namespace clang {
@@ -28,12 +29,14 @@ enum DeclUpdateKind {
   UPD_CXX_ADDED_FUNCTION_DEFINITION,
   UPD_CXX_INSTANTIATED_STATIC_DATA_MEMBER,
   UPD_CXX_INSTANTIATED_CLASS_DEFINITION,
+  UPD_CXX_RESOLVED_DTOR_DELETE,
   UPD_CXX_RESOLVED_EXCEPTION_SPEC,
   UPD_CXX_DEDUCED_RETURN_TYPE,
   UPD_DECL_MARKED_USED,
   UPD_MANGLING_NUMBER,
   UPD_STATIC_LOCAL_NUMBER,
-  UPD_DECL_MARKED_OPENMP_THREADPRIVATE
+  UPD_DECL_MARKED_OPENMP_THREADPRIVATE,
+  UPD_DECL_EXPORTED
 };
 
 TypeIdx TypeIdxFromBuiltin(const BuiltinType *BT);
@@ -85,6 +88,24 @@ bool isRedeclarableDeclKind(unsigned Kind);
 /// declaration number.
 bool needsAnonymousDeclarationNumber(const NamedDecl *D);
 
+/// \brief Visit each declaration within \c DC that needs an anonymous
+/// declaration number and call \p Visit with the declaration and its number.
+template<typename Fn> void numberAnonymousDeclsWithin(const DeclContext *DC,
+                                                      Fn Visit) {
+  unsigned Index = 0;
+  for (Decl *LexicalD : DC->decls()) {
+    // For a friend decl, we care about the declaration within it, if any.
+    if (auto *FD = dyn_cast<FriendDecl>(LexicalD))
+      LexicalD = FD->getFriendDecl();
+
+    auto *ND = dyn_cast_or_null<NamedDecl>(LexicalD);
+    if (!ND || !needsAnonymousDeclarationNumber(ND))
+      continue;
+
+    Visit(ND, Index++);
+  }
+}
+
 } // namespace serialization
 
 } // namespace clang
diff --git a/lib/Serialization/ASTReader.cpp b/lib/Serialization/ASTReader.cpp
index 416164ebb7d2..5e3a82795049 100644
--- a/lib/Serialization/ASTReader.cpp
+++ b/lib/Serialization/ASTReader.cpp
@@ -1,4 +1,4 @@
-//===--- ASTReader.cpp - AST File Reader ----------------------------------===//
+//===-- ASTReader.cpp - AST File Reader ----------------------------------===//
 //
 //                     The LLVM Compiler Infrastructure
 //
@@ -89,11 +89,13 @@ ChainedASTReaderListener::ReadLanguageOptions(const LangOptions &LangOpts,
          Second->ReadLanguageOptions(LangOpts, Complain,
                                      AllowCompatibleDifferences);
 }
-bool
-ChainedASTReaderListener::ReadTargetOptions(const TargetOptions &TargetOpts,
-                                            bool Complain) {
-  return First->ReadTargetOptions(TargetOpts, Complain) ||
-         Second->ReadTargetOptions(TargetOpts, Complain);
+bool ChainedASTReaderListener::ReadTargetOptions(
+    const TargetOptions &TargetOpts, bool Complain,
+    bool AllowCompatibleDifferences) {
+  return First->ReadTargetOptions(TargetOpts, Complain,
+                                  AllowCompatibleDifferences) ||
+         Second->ReadTargetOptions(TargetOpts, Complain,
+                                   AllowCompatibleDifferences);
 }
 bool ChainedASTReaderListener::ReadDiagnosticOptions(
     IntrusiveRefCntPtr<DiagnosticOptions> DiagOpts, bool Complain) {
@@ -108,9 +110,12 @@ ChainedASTReaderListener::ReadFileSystemOptions(const FileSystemOptions &FSOpts,
 }
 
 bool ChainedASTReaderListener::ReadHeaderSearchOptions(
-    const HeaderSearchOptions &HSOpts, bool Complain) {
-  return First->ReadHeaderSearchOptions(HSOpts, Complain) ||
-         Second->ReadHeaderSearchOptions(HSOpts, Complain);
+    const HeaderSearchOptions &HSOpts, StringRef SpecificModuleCachePath,
+    bool Complain) {
+  return First->ReadHeaderSearchOptions(HSOpts, SpecificModuleCachePath,
+                                        Complain) ||
+         Second->ReadHeaderSearchOptions(HSOpts, SpecificModuleCachePath,
+                                         Complain);
 }
 bool ChainedASTReaderListener::ReadPreprocessorOptions(
     const PreprocessorOptions &PPOpts, bool Complain,
@@ -229,7 +234,8 @@ static bool checkLanguageOptions(const LangOptions &LangOpts,
 /// \returns true if the target options mis-match, false otherwise.
 static bool checkTargetOptions(const TargetOptions &TargetOpts,
                                const TargetOptions &ExistingTargetOpts,
-                               DiagnosticsEngine *Diags) {
+                               DiagnosticsEngine *Diags,
+                               bool AllowCompatibleDifferences = true) {
 #define CHECK_TARGET_OPT(Field, Name)                             \
   if (TargetOpts.Field != ExistingTargetOpts.Field) {             \
     if (Diags)                                                    \
@@ -238,9 +244,16 @@ static bool checkTargetOptions(const TargetOptions &TargetOpts,
     return true;                                                  \
   }
 
+  // The triple and ABI must match exactly.
   CHECK_TARGET_OPT(Triple, "target");
-  CHECK_TARGET_OPT(CPU, "target CPU");
   CHECK_TARGET_OPT(ABI, "target ABI");
+
+  // We can tolerate different CPUs in many cases, notably when one CPU
+  // supports a strict superset of another. When allowing compatible
+  // differences skip this check.
+  if (!AllowCompatibleDifferences)
+    CHECK_TARGET_OPT(CPU, "target CPU");
+
 #undef CHECK_TARGET_OPT
 
   // Compare feature sets.
@@ -252,43 +265,31 @@ static bool checkTargetOptions(const TargetOptions &TargetOpts,
   std::sort(ExistingFeatures.begin(), ExistingFeatures.end());
   std::sort(ReadFeatures.begin(), ReadFeatures.end());
 
-  unsigned ExistingIdx = 0, ExistingN = ExistingFeatures.size();
-  unsigned ReadIdx = 0, ReadN = ReadFeatures.size();
-  while (ExistingIdx < ExistingN && ReadIdx < ReadN) {
-    if (ExistingFeatures[ExistingIdx] == ReadFeatures[ReadIdx]) {
-      ++ExistingIdx;
-      ++ReadIdx;
-      continue;
-    }
-
-    if (ReadFeatures[ReadIdx] < ExistingFeatures[ExistingIdx]) {
-      if (Diags)
-        Diags->Report(diag::err_pch_targetopt_feature_mismatch)
-          << false << ReadFeatures[ReadIdx];
-      return true;
-    }
-
-    if (Diags)
-      Diags->Report(diag::err_pch_targetopt_feature_mismatch)
-        << true << ExistingFeatures[ExistingIdx];
-    return true;
-  }
+  // We compute the set difference in both directions explicitly so that we can
+  // diagnose the differences differently.
+  SmallVector<StringRef, 4> UnmatchedExistingFeatures, UnmatchedReadFeatures;
+  std::set_difference(
+      ExistingFeatures.begin(), ExistingFeatures.end(), ReadFeatures.begin(),
+      ReadFeatures.end(), std::back_inserter(UnmatchedExistingFeatures));
+  std::set_difference(ReadFeatures.begin(), ReadFeatures.end(),
+                      ExistingFeatures.begin(), ExistingFeatures.end(),
+                      std::back_inserter(UnmatchedReadFeatures));
+
+  // If we are allowing compatible differences and the read feature set is
+  // a strict subset of the existing feature set, there is nothing to diagnose.
+  if (AllowCompatibleDifferences && UnmatchedReadFeatures.empty())
+    return false;
 
-  if (ExistingIdx < ExistingN) {
-    if (Diags)
+  if (Diags) {
+    for (StringRef Feature : UnmatchedReadFeatures)
       Diags->Report(diag::err_pch_targetopt_feature_mismatch)
-        << true << ExistingFeatures[ExistingIdx];
-    return true;
-  }
-
-  if (ReadIdx < ReadN) {
-    if (Diags)
+          << /* is-existing-feature */ false << Feature;
+    for (StringRef Feature : UnmatchedExistingFeatures)
       Diags->Report(diag::err_pch_targetopt_feature_mismatch)
-        << false << ReadFeatures[ReadIdx];
-    return true;
+          << /* is-existing-feature */ true << Feature;
   }
 
-  return false;
+  return !UnmatchedReadFeatures.empty() || !UnmatchedExistingFeatures.empty();
 }
 
 bool
@@ -302,10 +303,12 @@ PCHValidator::ReadLanguageOptions(const LangOptions &LangOpts,
 }
 
 bool PCHValidator::ReadTargetOptions(const TargetOptions &TargetOpts,
-                                     bool Complain) {
+                                     bool Complain,
+                                     bool AllowCompatibleDifferences) {
   const TargetOptions &ExistingTargetOpts = PP.getTargetInfo().getTargetOpts();
   return checkTargetOptions(TargetOpts, ExistingTargetOpts,
-                            Complain? &Reader.Diags : nullptr);
+                            Complain ? &Reader.Diags : nullptr,
+                            AllowCompatibleDifferences);
 }
 
 namespace {
@@ -464,7 +467,7 @@ collectMacroDefinitions(const PreprocessorOptions &PPOpts,
     Macros[MacroName] = std::make_pair(MacroBody, false);
   }
 }
-         
+
 /// \brief Check the preprocessor options deserialized from the control block
 /// against the preprocessor options in an existing preprocessor.
 ///
@@ -591,6 +594,36 @@ bool PCHValidator::ReadPreprocessorOptions(const PreprocessorOptions &PPOpts,
                                   PP.getLangOpts());
 }
 
+/// Check the header search options deserialized from the control block
+/// against the header search options in an existing preprocessor.
+///
+/// \param Diags If non-null, produce diagnostics for any mismatches incurred.
+static bool checkHeaderSearchOptions(const HeaderSearchOptions &HSOpts,
+                                     StringRef SpecificModuleCachePath,
+                                     StringRef ExistingModuleCachePath,
+                                     DiagnosticsEngine *Diags,
+                                     const LangOptions &LangOpts) {
+  if (LangOpts.Modules) {
+    if (SpecificModuleCachePath != ExistingModuleCachePath) {
+      if (Diags)
+        Diags->Report(diag::err_pch_modulecache_mismatch)
+          << SpecificModuleCachePath << ExistingModuleCachePath;
+      return true;
+    }
+  }
+
+  return false;
+}
+
+bool PCHValidator::ReadHeaderSearchOptions(const HeaderSearchOptions &HSOpts,
+                                           StringRef SpecificModuleCachePath,
+                                           bool Complain) {
+  return checkHeaderSearchOptions(HSOpts, SpecificModuleCachePath,
+                                  PP.getHeaderSearchInfo().getModuleCachePath(),
+                                  Complain ? &Reader.Diags : nullptr,
+                                  PP.getLangOpts());
+}
+
 void PCHValidator::ReadCounter(const ModuleFile &M, unsigned Value) {
   PP.setCounterValue(Value);
 }
@@ -744,8 +777,6 @@ IdentifierInfo *ASTIdentifierLookupTrait::ReadData(const internal_key_type& k,
   Bits >>= 1;
   bool ExtensionToken = Bits & 0x01;
   Bits >>= 1;
-  bool hasSubmoduleMacros = Bits & 0x01;
-  Bits >>= 1;
   bool hadMacroDefinition = Bits & 0x01;
   Bits >>= 1;
 
@@ -787,49 +818,8 @@ IdentifierInfo *ASTIdentifierLookupTrait::ReadData(const internal_key_type& k,
     uint32_t MacroDirectivesOffset =
         endian::readNext<uint32_t, little, unaligned>(d);
     DataLen -= 4;
-    SmallVector<uint32_t, 8> LocalMacroIDs;
-    if (hasSubmoduleMacros) {
-      while (true) {
-        uint32_t LocalMacroID =
-            endian::readNext<uint32_t, little, unaligned>(d);
-        DataLen -= 4;
-        if (LocalMacroID == 0xdeadbeef) break;
-        LocalMacroIDs.push_back(LocalMacroID);
-      }
-    }
-
-    if (F.Kind == MK_ImplicitModule || F.Kind == MK_ExplicitModule) {
-      // Macro definitions are stored from newest to oldest, so reverse them
-      // before registering them.
-      llvm::SmallVector<unsigned, 8> MacroSizes;
-      for (SmallVectorImpl<uint32_t>::iterator
-             I = LocalMacroIDs.begin(), E = LocalMacroIDs.end(); I != E; /**/) {
-        unsigned Size = 1;
-
-        static const uint32_t HasOverridesFlag = 0x80000000U;
-        if (I + 1 != E && (I[1] & HasOverridesFlag))
-          Size += 1 + (I[1] & ~HasOverridesFlag);
 
-        MacroSizes.push_back(Size);
-        I += Size;
-      }
-
-      SmallVectorImpl<uint32_t>::iterator I = LocalMacroIDs.end();
-      for (SmallVectorImpl<unsigned>::reverse_iterator SI = MacroSizes.rbegin(),
-                                                       SE = MacroSizes.rend();
-           SI != SE; ++SI) {
-        I -= *SI;
-
-        uint32_t LocalMacroID = *I;
-        ArrayRef<uint32_t> Overrides;
-        if (*SI != 1)
-          Overrides = llvm::makeArrayRef(&I[2], *SI - 2);
-        Reader.addPendingMacroFromModule(II, &F, LocalMacroID, Overrides);
-      }
-      assert(I == LocalMacroIDs.begin());
-    } else {
-      Reader.addPendingMacroFromPCH(II, &F, MacroDirectivesOffset);
-    }
+    Reader.addPendingMacro(II, &F, MacroDirectivesOffset);
   }
 
   Reader.SetIdentifierInfo(ID, II);
@@ -1393,6 +1383,7 @@ MacroInfo *ASTReader::ReadMacroRecord(ModuleFile &F, uint64_t Offset) {
     PreprocessorRecordTypes RecType =
       (PreprocessorRecordTypes)Stream.readRecord(Entry.ID, Record);
     switch (RecType) {
+    case PP_MODULE_MACRO:
     case PP_MACRO_DIRECTIVE_HISTORY:
       return Macro;
 
@@ -1441,10 +1432,10 @@ MacroInfo *ASTReader::ReadMacroRecord(ModuleFile &F, uint64_t Offset) {
         PreprocessedEntityID
             GlobalID = getGlobalPreprocessedEntityID(F, Record[NextIndex]);
         PreprocessingRecord &PPRec = *PP.getPreprocessingRecord();
-        PreprocessingRecord::PPEntityID
-          PPID = PPRec.getPPEntityID(GlobalID-1, /*isLoaded=*/true);
-        MacroDefinition *PPDef =
-          cast_or_null<MacroDefinition>(PPRec.getPreprocessedEntity(PPID));
+        PreprocessingRecord::PPEntityID PPID =
+            PPRec.getPPEntityID(GlobalID - 1, /*isLoaded=*/true);
+        MacroDefinitionRecord *PPDef = cast_or_null<MacroDefinitionRecord>(
+            PPRec.getPreprocessedEntity(PPID));
         if (PPDef)
           PPRec.RegisterMacroDefinition(Macro, PPDef);
       }
@@ -1586,24 +1577,9 @@ HeaderFileInfoTrait::ReadData(internal_key_ref key, const unsigned char *d,
   return HFI;
 }
 
-void
-ASTReader::addPendingMacroFromModule(IdentifierInfo *II, ModuleFile *M,
-                                     GlobalMacroID GMacID,
-                                     ArrayRef<SubmoduleID> Overrides) {
-  assert(NumCurrentElementsDeserializing > 0 &&"Missing deserialization guard");
-  SubmoduleID *OverrideData = nullptr;
-  if (!Overrides.empty()) {
-    OverrideData = new (Context) SubmoduleID[Overrides.size() + 1];
-    OverrideData[0] = Overrides.size();
-    for (unsigned I = 0; I != Overrides.size(); ++I)
-      OverrideData[I + 1] = getGlobalSubmoduleID(*M, Overrides[I]);
-  }
-  PendingMacroIDs[II].push_back(PendingMacroInfo(M, GMacID, OverrideData));
-}
-
-void ASTReader::addPendingMacroFromPCH(IdentifierInfo *II,
-                                       ModuleFile *M,
-                                       uint64_t MacroDirectivesOffset) {
+void ASTReader::addPendingMacro(IdentifierInfo *II,
+                                ModuleFile *M,
+                                uint64_t MacroDirectivesOffset) {
   assert(NumCurrentElementsDeserializing > 0 &&"Missing deserialization guard");
   PendingMacroIDs[II].push_back(PendingMacroInfo(M, MacroDirectivesOffset));
 }
@@ -1747,110 +1723,81 @@ void ASTReader::markIdentifierUpToDate(IdentifierInfo *II) {
     IdentifierGeneration[II] = getGeneration();
 }
 
-struct ASTReader::ModuleMacroInfo {
-  SubmoduleID SubModID;
-  MacroInfo *MI;
-  SubmoduleID *Overrides;
-  // FIXME: Remove this.
-  ModuleFile *F;
-
-  bool isDefine() const { return MI; }
-
-  SubmoduleID getSubmoduleID() const { return SubModID; }
-
-  ArrayRef<SubmoduleID> getOverriddenSubmodules() const {
-    if (!Overrides)
-      return None;
-    return llvm::makeArrayRef(Overrides + 1, *Overrides);
-  }
-
-  MacroDirective *import(Preprocessor &PP, SourceLocation ImportLoc) const {
-    if (!MI)
-      return PP.AllocateUndefMacroDirective(ImportLoc, SubModID,
-                                            getOverriddenSubmodules());
-    return PP.AllocateDefMacroDirective(MI, ImportLoc, SubModID,
-                                        getOverriddenSubmodules());
-  }
-};
-
-ASTReader::ModuleMacroInfo *
-ASTReader::getModuleMacro(const PendingMacroInfo &PMInfo) {
-  ModuleMacroInfo Info;
-
-  uint32_t ID = PMInfo.ModuleMacroData.MacID;
-  if (ID & 1) {
-    // Macro undefinition.
-    Info.SubModID = getGlobalSubmoduleID(*PMInfo.M, ID >> 1);
-    Info.MI = nullptr;
-  } else {
-    // Macro definition.
-    GlobalMacroID GMacID = getGlobalMacroID(*PMInfo.M, ID >> 1);
-    assert(GMacID);
-
-    // If this macro has already been loaded, don't do so again.
-    // FIXME: This is highly dubious. Multiple macro definitions can have the
-    // same MacroInfo (and hence the same GMacID) due to #pragma push_macro etc.
-    if (MacrosLoaded[GMacID - NUM_PREDEF_MACRO_IDS])
-      return nullptr;
+void ASTReader::resolvePendingMacro(IdentifierInfo *II,
+                                    const PendingMacroInfo &PMInfo) {
+  ModuleFile &M = *PMInfo.M;
 
-    Info.MI = getMacro(GMacID);
-    Info.SubModID = Info.MI->getOwningModuleID();
-  }
-  Info.Overrides = PMInfo.ModuleMacroData.Overrides;
-  Info.F = PMInfo.M;
+  BitstreamCursor &Cursor = M.MacroCursor;
+  SavedStreamPosition SavedPosition(Cursor);
+  Cursor.JumpToBit(PMInfo.MacroDirectivesOffset);
 
-  return new (Context) ModuleMacroInfo(Info);
-}
+  struct ModuleMacroRecord {
+    SubmoduleID SubModID;
+    MacroInfo *MI;
+    SmallVector<SubmoduleID, 8> Overrides;
+  };
+  llvm::SmallVector<ModuleMacroRecord, 8> ModuleMacros;
 
-void ASTReader::resolvePendingMacro(IdentifierInfo *II,
-                                    const PendingMacroInfo &PMInfo) {
-  assert(II);
+  // We expect to see a sequence of PP_MODULE_MACRO records listing exported
+  // macros, followed by a PP_MACRO_DIRECTIVE_HISTORY record with the complete
+  // macro histroy.
+  RecordData Record;
+  while (true) {
+    llvm::BitstreamEntry Entry =
+        Cursor.advance(BitstreamCursor::AF_DontPopBlockAtEnd);
+    if (Entry.Kind != llvm::BitstreamEntry::Record) {
+      Error("malformed block record in AST file");
+      return;
+    }
 
-  if (PMInfo.M->Kind != MK_ImplicitModule &&
-      PMInfo.M->Kind != MK_ExplicitModule) {
-    installPCHMacroDirectives(II, *PMInfo.M,
-                              PMInfo.PCHMacroData.MacroDirectivesOffset);
-    return;
-  }
+    Record.clear();
+    switch ((PreprocessorRecordTypes)Cursor.readRecord(Entry.ID, Record)) {
+    case PP_MACRO_DIRECTIVE_HISTORY:
+      break;
 
-  // Module Macro.
+    case PP_MODULE_MACRO: {
+      ModuleMacros.push_back(ModuleMacroRecord());
+      auto &Info = ModuleMacros.back();
+      Info.SubModID = getGlobalSubmoduleID(M, Record[0]);
+      Info.MI = getMacro(getGlobalMacroID(M, Record[1]));
+      for (int I = 2, N = Record.size(); I != N; ++I)
+        Info.Overrides.push_back(getGlobalSubmoduleID(M, Record[I]));
+      continue;
+    }
 
-  ModuleMacroInfo *MMI = getModuleMacro(PMInfo);
-  if (!MMI)
-    return;
+    default:
+      Error("malformed block record in AST file");
+      return;
+    }
 
-  Module *Owner = getSubmodule(MMI->getSubmoduleID());
-  if (Owner && Owner->NameVisibility == Module::Hidden) {
-    // Macros in the owning module are hidden. Just remember this macro to
-    // install if we make this module visible.
-    HiddenNamesMap[Owner].HiddenMacros.insert(std::make_pair(II, MMI));
-  } else {
-    installImportedMacro(II, MMI, Owner);
+    // We found the macro directive history; that's the last record
+    // for this macro.
+    break;
   }
-}
-
-void ASTReader::installPCHMacroDirectives(IdentifierInfo *II,
-                                          ModuleFile &M, uint64_t Offset) {
-  assert(M.Kind != MK_ImplicitModule && M.Kind != MK_ExplicitModule);
 
-  BitstreamCursor &Cursor = M.MacroCursor;
-  SavedStreamPosition SavedPosition(Cursor);
-  Cursor.JumpToBit(Offset);
+  // Module macros are listed in reverse dependency order.
+  {
+    std::reverse(ModuleMacros.begin(), ModuleMacros.end());
+    llvm::SmallVector<ModuleMacro*, 8> Overrides;
+    for (auto &MMR : ModuleMacros) {
+      Overrides.clear();
+      for (unsigned ModID : MMR.Overrides) {
+        Module *Mod = getSubmodule(ModID);
+        auto *Macro = PP.getModuleMacro(Mod, II);
+        assert(Macro && "missing definition for overridden macro");
+        Overrides.push_back(Macro);
+      }
 
-  llvm::BitstreamEntry Entry =
-      Cursor.advance(BitstreamCursor::AF_DontPopBlockAtEnd);
-  if (Entry.Kind != llvm::BitstreamEntry::Record) {
-    Error("malformed block record in AST file");
-    return;
+      bool Inserted = false;
+      Module *Owner = getSubmodule(MMR.SubModID);
+      PP.addModuleMacro(Owner, II, MMR.MI, Overrides, Inserted);
+    }
   }
 
-  RecordData Record;
-  PreprocessorRecordTypes RecType =
-    (PreprocessorRecordTypes)Cursor.readRecord(Entry.ID, Record);
-  if (RecType != PP_MACRO_DIRECTIVE_HISTORY) {
-    Error("malformed block record in AST file");
+  // Don't read the directive history for a module; we don't have anywhere
+  // to put it.
+  if (M.Kind == MK_ImplicitModule || M.Kind == MK_ExplicitModule)
     return;
-  }
 
   // Deserialize the macro directives history in reverse source-order.
   MacroDirective *Latest = nullptr, *Earliest = nullptr;
@@ -1861,31 +1808,12 @@ void ASTReader::installPCHMacroDirectives(IdentifierInfo *II,
     MacroDirective::Kind K = (MacroDirective::Kind)Record[Idx++];
     switch (K) {
     case MacroDirective::MD_Define: {
-      GlobalMacroID GMacID = getGlobalMacroID(M, Record[Idx++]);
-      MacroInfo *MI = getMacro(GMacID);
-      SubmoduleID ImportedFrom = Record[Idx++];
-      bool IsAmbiguous = Record[Idx++];
-      llvm::SmallVector<unsigned, 4> Overrides;
-      if (ImportedFrom) {
-        Overrides.insert(Overrides.end(),
-                         &Record[Idx] + 1, &Record[Idx] + 1 + Record[Idx]);
-        Idx += Overrides.size() + 1;
-      }
-      DefMacroDirective *DefMD =
-          PP.AllocateDefMacroDirective(MI, Loc, ImportedFrom, Overrides);
-      DefMD->setAmbiguous(IsAmbiguous);
-      MD = DefMD;
+      MacroInfo *MI = getMacro(getGlobalMacroID(M, Record[Idx++]));
+      MD = PP.AllocateDefMacroDirective(MI, Loc);
       break;
     }
     case MacroDirective::MD_Undefine: {
-      SubmoduleID ImportedFrom = Record[Idx++];
-      llvm::SmallVector<unsigned, 4> Overrides;
-      if (ImportedFrom) {
-        Overrides.insert(Overrides.end(),
-                         &Record[Idx] + 1, &Record[Idx] + 1 + Record[Idx]);
-        Idx += Overrides.size() + 1;
-      }
-      MD = PP.AllocateUndefMacroDirective(Loc, ImportedFrom, Overrides);
+      MD = PP.AllocateUndefMacroDirective(Loc);
       break;
     }
     case MacroDirective::MD_Visibility:
@@ -1901,177 +1829,8 @@ void ASTReader::installPCHMacroDirectives(IdentifierInfo *II,
     Earliest = MD;
   }
 
-  PP.setLoadedMacroDirective(II, Latest);
-}
-
-/// \brief For the given macro definitions, check if they are both in system
-/// modules.
-static bool areDefinedInSystemModules(MacroInfo *PrevMI, MacroInfo *NewMI,
-                                      Module *NewOwner, ASTReader &Reader) {
-  assert(PrevMI && NewMI);
-  Module *PrevOwner = nullptr;
-  if (SubmoduleID PrevModID = PrevMI->getOwningModuleID())
-    PrevOwner = Reader.getSubmodule(PrevModID);
-  SourceManager &SrcMgr = Reader.getSourceManager();
-  bool PrevInSystem
-    = PrevOwner? PrevOwner->IsSystem
-               : SrcMgr.isInSystemHeader(PrevMI->getDefinitionLoc());
-  bool NewInSystem
-    = NewOwner? NewOwner->IsSystem
-              : SrcMgr.isInSystemHeader(NewMI->getDefinitionLoc());
-  if (PrevOwner && PrevOwner == NewOwner)
-    return false;
-  return PrevInSystem && NewInSystem;
-}
-
-void ASTReader::removeOverriddenMacros(IdentifierInfo *II,
-                                       SourceLocation ImportLoc,
-                                       AmbiguousMacros &Ambig,
-                                       ArrayRef<SubmoduleID> Overrides) {
-  for (unsigned OI = 0, ON = Overrides.size(); OI != ON; ++OI) {
-    SubmoduleID OwnerID = Overrides[OI];
-
-    // If this macro is not yet visible, remove it from the hidden names list.
-    // It won't be there if we're in the middle of making the owner visible.
-    Module *Owner = getSubmodule(OwnerID);
-    auto HiddenIt = HiddenNamesMap.find(Owner);
-    if (HiddenIt != HiddenNamesMap.end()) {
-      HiddenNames &Hidden = HiddenIt->second;
-      HiddenMacrosMap::iterator HI = Hidden.HiddenMacros.find(II);
-      if (HI != Hidden.HiddenMacros.end()) {
-        // Register the macro now so we don't lose it when we re-export.
-        PP.appendMacroDirective(II, HI->second->import(PP, ImportLoc));
-
-        auto SubOverrides = HI->second->getOverriddenSubmodules();
-        Hidden.HiddenMacros.erase(HI);
-        removeOverriddenMacros(II, ImportLoc, Ambig, SubOverrides);
-      }
-    }
-
-    // If this macro is already in our list of conflicts, remove it from there.
-    Ambig.erase(
-        std::remove_if(Ambig.begin(), Ambig.end(), [&](DefMacroDirective *MD) {
-          return MD->getInfo()->getOwningModuleID() == OwnerID;
-        }),
-        Ambig.end());
-  }
-}
-
-ASTReader::AmbiguousMacros *
-ASTReader::removeOverriddenMacros(IdentifierInfo *II,
-                                  SourceLocation ImportLoc,
-                                  ArrayRef<SubmoduleID> Overrides) {
-  MacroDirective *Prev = PP.getMacroDirective(II);
-  if (!Prev && Overrides.empty())
-    return nullptr;
-
-  DefMacroDirective *PrevDef = Prev ? Prev->getDefinition().getDirective()
-                                    : nullptr;
-  if (PrevDef && PrevDef->isAmbiguous()) {
-    // We had a prior ambiguity. Check whether we resolve it (or make it worse).
-    AmbiguousMacros &Ambig = AmbiguousMacroDefs[II];
-    Ambig.push_back(PrevDef);
-
-    removeOverriddenMacros(II, ImportLoc, Ambig, Overrides);
-
-    if (!Ambig.empty())
-      return &Ambig;
-
-    AmbiguousMacroDefs.erase(II);
-  } else {
-    // There's no ambiguity yet. Maybe we're introducing one.
-    AmbiguousMacros Ambig;
-    if (PrevDef)
-      Ambig.push_back(PrevDef);
-
-    removeOverriddenMacros(II, ImportLoc, Ambig, Overrides);
-
-    if (!Ambig.empty()) {
-      AmbiguousMacros &Result = AmbiguousMacroDefs[II];
-      std::swap(Result, Ambig);
-      return &Result;
-    }
-  }
-
-  // We ended up with no ambiguity.
-  return nullptr;
-}
-
-void ASTReader::installImportedMacro(IdentifierInfo *II, ModuleMacroInfo *MMI,
-                                     Module *Owner) {
-  assert(II && Owner);
-
-  SourceLocation ImportLoc = Owner->MacroVisibilityLoc;
-  if (ImportLoc.isInvalid()) {
-    // FIXME: If we made macros from this module visible but didn't provide a
-    // source location for the import, we don't have a location for the macro.
-    // Use the location at which the containing module file was first imported
-    // for now.
-    ImportLoc = MMI->F->DirectImportLoc;
-    assert(ImportLoc.isValid() && "no import location for a visible macro?");
-  }
-
-  AmbiguousMacros *Prev =
-      removeOverriddenMacros(II, ImportLoc, MMI->getOverriddenSubmodules());
-
-  // Create a synthetic macro definition corresponding to the import (or null
-  // if this was an undefinition of the macro).
-  MacroDirective *Imported = MMI->import(PP, ImportLoc);
-  DefMacroDirective *MD = dyn_cast<DefMacroDirective>(Imported);
-
-  // If there's no ambiguity, just install the macro.
-  if (!Prev) {
-    PP.appendMacroDirective(II, Imported);
-    return;
-  }
-  assert(!Prev->empty());
-
-  if (!MD) {
-    // We imported a #undef that didn't remove all prior definitions. The most
-    // recent prior definition remains, and we install it in the place of the
-    // imported directive, as if by a local #pragma pop_macro.
-    MacroInfo *NewMI = Prev->back()->getInfo();
-    Prev->pop_back();
-    MD = PP.AllocateDefMacroDirective(NewMI, ImportLoc);
-
-    // Install our #undef first so that we don't lose track of it. We'll replace
-    // this with whichever macro definition ends up winning.
-    PP.appendMacroDirective(II, Imported);
-  }
-
-  // We're introducing a macro definition that creates or adds to an ambiguity.
-  // We can resolve that ambiguity if this macro is token-for-token identical to
-  // all of the existing definitions.
-  MacroInfo *NewMI = MD->getInfo();
-  assert(NewMI && "macro definition with no MacroInfo?");
-  while (!Prev->empty()) {
-    MacroInfo *PrevMI = Prev->back()->getInfo();
-    assert(PrevMI && "macro definition with no MacroInfo?");
-
-    // Before marking the macros as ambiguous, check if this is a case where
-    // both macros are in system headers. If so, we trust that the system
-    // did not get it wrong. This also handles cases where Clang's own
-    // headers have a different spelling of certain system macros:
-    //   #define LONG_MAX __LONG_MAX__ (clang's limits.h)
-    //   #define LONG_MAX 0x7fffffffffffffffL (system's limits.h)
-    //
-    // FIXME: Remove the defined-in-system-headers check. clang's limits.h
-    // overrides the system limits.h's macros, so there's no conflict here.
-    if (NewMI != PrevMI &&
-        !PrevMI->isIdenticalTo(*NewMI, PP, /*Syntactically=*/true) &&
-        !areDefinedInSystemModules(PrevMI, NewMI, Owner, *this))
-      break;
-
-    // The previous definition is the same as this one (or both are defined in
-    // system modules so we can assume they're equivalent); we don't need to
-    // track it any more.
-    Prev->pop_back();
-  }
-
-  if (!Prev->empty())
-    MD->setAmbiguous(true);
-
-  PP.appendMacroDirective(II, MD);
+  if (Latest)
+    PP.setLoadedMacroDirective(II, Latest);
 }
 
 ASTReader::InputFileInfo
@@ -2426,6 +2185,9 @@ ASTReader::ReadControlBlock(ModuleFile &F,
       break;
     }
 
+    case KNOWN_MODULE_FILES:
+      break;
+
     case LANGUAGE_OPTIONS: {
       bool Complain = (ClientLoadCapabilities & ARR_ConfigurationMismatch) == 0;
       // FIXME: The &F == *ModuleMgr.begin() check is wrong for modules.
@@ -2440,7 +2202,8 @@ ASTReader::ReadControlBlock(ModuleFile &F,
     case TARGET_OPTIONS: {
       bool Complain = (ClientLoadCapabilities & ARR_ConfigurationMismatch)==0;
       if (Listener && &F == *ModuleMgr.begin() &&
-          ParseTargetOptions(Record, Complain, *Listener) &&
+          ParseTargetOptions(Record, Complain, *Listener,
+                             AllowCompatibleConfigurationMismatch) &&
           !DisableValidation && !AllowConfigurationMismatch)
         return ConfigurationMismatch;
       break;
@@ -2543,7 +2306,7 @@ ASTReader::ReadControlBlock(ModuleFile &F,
     case INPUT_FILE_OFFSETS:
       NumInputs = Record[0];
       NumUserInputs = Record[1];
-      F.InputFileOffsets = (const uint32_t *)Blob.data();
+      F.InputFileOffsets = (const uint64_t *)Blob.data();
       F.InputFilesLoaded.resize(NumInputs);
       break;
     }
@@ -2795,6 +2558,8 @@ ASTReader::ReadASTBlock(ModuleFile &F, unsigned ClientLoadCapabilities) {
     }
 
     case EAGERLY_DESERIALIZED_DECLS:
+      // FIXME: Skip reading this record if our ASTConsumer doesn't care
+      // about "interesting" decls (for instance, if we're building a module).
       for (unsigned I = 0, N = Record.size(); I != N; ++I)
         EagerlyDeserializedDecls.push_back(getGlobalDeclID(F, Record[I]));
       break;
@@ -2859,11 +2624,6 @@ ASTReader::ReadASTBlock(ModuleFile &F, unsigned ClientLoadCapabilities) {
       }
       break;
 
-    case LOCALLY_SCOPED_EXTERN_C_DECLS:
-      for (unsigned I = 0, N = Record.size(); I != N; ++I)
-        LocallyScopedExternCDecls.push_back(getGlobalDeclID(F, Record[I]));
-      break;
-
     case SELECTOR_OFFSETS: {
       F.SelectorOffsets = (const uint32_t *)Blob.data();
       F.LocalNumSelectors = Record[0];
@@ -3066,11 +2826,6 @@ ASTReader::ReadASTBlock(ModuleFile &F, unsigned ClientLoadCapabilities) {
       }
       break;
 
-    case DYNAMIC_CLASSES:
-      for (unsigned I = 0, N = Record.size(); I != N; ++I)
-        DynamicClasses.push_back(getGlobalDeclID(F, Record[I]));
-      break;
-
     case PENDING_IMPLICIT_INSTANTIATIONS:
       if (PendingInstantiations.size() % 2 != 0) {
         Error("Invalid existing PendingInstantiations");
@@ -3172,16 +2927,26 @@ ASTReader::ReadASTBlock(ModuleFile &F, unsigned ClientLoadCapabilities) {
     case OBJC_CATEGORIES:
       F.ObjCCategories.swap(Record);
       break;
-        
+
     case CXX_BASE_SPECIFIER_OFFSETS: {
       if (F.LocalNumCXXBaseSpecifiers != 0) {
         Error("duplicate CXX_BASE_SPECIFIER_OFFSETS record in AST file");
         return Failure;
       }
-      
+
       F.LocalNumCXXBaseSpecifiers = Record[0];
       F.CXXBaseSpecifiersOffsets = (const uint32_t *)Blob.data();
-      NumCXXBaseSpecifiersLoaded += F.LocalNumCXXBaseSpecifiers;
+      break;
+    }
+
+    case CXX_CTOR_INITIALIZERS_OFFSETS: {
+      if (F.LocalNumCXXCtorInitializers != 0) {
+        Error("duplicate CXX_CTOR_INITIALIZERS_OFFSETS record in AST file");
+        return Failure;
+      }
+
+      F.LocalNumCXXCtorInitializers = Record[0];
+      F.CXXCtorInitializersOffsets = (const uint32_t *)Blob.data();
       break;
     }
 
@@ -3256,6 +3021,18 @@ ASTReader::ReadASTBlock(ModuleFile &F, unsigned ClientLoadCapabilities) {
             ReadSourceLocation(F, Record, I).getRawEncoding());
       }
       break;
+    case DELETE_EXPRS_TO_ANALYZE:
+      for (unsigned I = 0, N = Record.size(); I != N;) {
+        DelayedDeleteExprs.push_back(getGlobalDeclID(F, Record[I++]));
+        const uint64_t Count = Record[I++];
+        DelayedDeleteExprs.push_back(Count);
+        for (uint64_t C = 0; C < Count; ++C) {
+          DelayedDeleteExprs.push_back(ReadSourceLocation(F, Record, I).getRawEncoding());
+          bool IsArrayForm = Record[I++] == 1;
+          DelayedDeleteExprs.push_back(IsArrayForm);
+        }
+      }
+      break;
 
     case IMPORTED_MODULES: {
       if (F.Kind != MK_ImplicitModule && F.Kind != MK_ExplicitModule) {
@@ -3288,16 +3065,6 @@ ASTReader::ReadASTBlock(ModuleFile &F, unsigned ClientLoadCapabilities) {
       break;
     }
         
-    case MERGED_DECLARATIONS: {
-      for (unsigned Idx = 0; Idx < Record.size(); /* increment in loop */) {
-        GlobalDeclID CanonID = getGlobalDeclID(F, Record[Idx++]);
-        SmallVectorImpl<GlobalDeclID> &Decls = StoredMergedDecls[CanonID];
-        for (unsigned N = Record[Idx++]; N > 0; --N)
-          Decls.push_back(getGlobalDeclID(F, Record[Idx++]));
-      }
-      break;
-    }
-
     case MACRO_OFFSET: {
       if (F.LocalNumMacros != 0) {
         Error("duplicate MACRO_OFFSET record in AST file");
@@ -3322,11 +3089,6 @@ ASTReader::ReadASTBlock(ModuleFile &F, unsigned ClientLoadCapabilities) {
       break;
     }
 
-    case MACRO_TABLE: {
-      // FIXME: Not used yet.
-      break;
-    }
-
     case LATE_PARSED_TEMPLATE: {
       LateParsedTemplates.append(Record.begin(), Record.end());
       break;
@@ -3472,10 +3234,9 @@ static void moveMethodToBackOfGlobalList(Sema &S, ObjCMethodDecl *Method) {
   }
 }
 
-void ASTReader::makeNamesVisible(const HiddenNames &Names, Module *Owner,
-                                 bool FromFinalization) {
-  // FIXME: Only do this if Owner->NameVisibility == AllVisible.
-  for (Decl *D : Names.HiddenDecls) {
+void ASTReader::makeNamesVisible(const HiddenNames &Names, Module *Owner) {
+  assert(Owner->NameVisibility != Module::Hidden && "nothing to make visible?");
+  for (Decl *D : Names) {
     bool wasHidden = D->Hidden;
     D->Hidden = false;
 
@@ -3485,22 +3246,11 @@ void ASTReader::makeNamesVisible(const HiddenNames &Names, Module *Owner,
       }
     }
   }
-
-  assert((FromFinalization || Owner->NameVisibility >= Module::MacrosVisible) &&
-         "nothing to make visible?");
-  for (const auto &Macro : Names.HiddenMacros) {
-    if (FromFinalization)
-      PP.appendMacroDirective(Macro.first,
-                              Macro.second->import(PP, SourceLocation()));
-    else
-      installImportedMacro(Macro.first, Macro.second, Owner);
-  }
 }
 
 void ASTReader::makeModuleVisible(Module *Mod,
                                   Module::NameVisibilityKind NameVisibility,
-                                  SourceLocation ImportLoc,
-                                  bool Complain) {
+                                  SourceLocation ImportLoc) {
   llvm::SmallPtrSet<Module *, 4> Visited;
   SmallVector<Module *, 4> Stack;
   Stack.push_back(Mod);
@@ -3519,9 +3269,6 @@ void ASTReader::makeModuleVisible(Module *Mod,
     }
 
     // Update the module's name visibility.
-    if (NameVisibility >= Module::MacrosVisible &&
-        Mod->NameVisibility < Module::MacrosVisible)
-      Mod->MacroVisibilityLoc = ImportLoc;
     Mod->NameVisibility = NameVisibility;
 
     // If we've already deserialized any names from this module,
@@ -3530,8 +3277,7 @@ void ASTReader::makeModuleVisible(Module *Mod,
     if (Hidden != HiddenNamesMap.end()) {
       auto HiddenNames = std::move(*Hidden);
       HiddenNamesMap.erase(Hidden);
-      makeNamesVisible(HiddenNames.second, HiddenNames.first,
-                       /*FromFinalization*/false);
+      makeNamesVisible(HiddenNames.second, HiddenNames.first);
       assert(HiddenNamesMap.find(Mod) == HiddenNamesMap.end() &&
              "making names visible added hidden names");
     }
@@ -3545,20 +3291,6 @@ void ASTReader::makeModuleVisible(Module *Mod,
       if (Visited.insert(Exported).second)
         Stack.push_back(Exported);
     }
-
-    // Detect any conflicts.
-    if (Complain) {
-      assert(ImportLoc.isValid() && "Missing import location");
-      for (unsigned I = 0, N = Mod->Conflicts.size(); I != N; ++I) {
-        if (Mod->Conflicts[I].Other->NameVisibility >= NameVisibility) {
-          Diag(ImportLoc, diag::warn_module_conflict)
-            << Mod->getFullModuleName()
-            << Mod->Conflicts[I].Other->getFullModuleName()
-            << Mod->Conflicts[I].Message;
-          // FIXME: Need note where the other module was imported.
-        }
-      }
-    }
   }
 }
 
@@ -3717,7 +3449,7 @@ ASTReader::ASTReadResult ASTReader::ReadAST(const std::string &FileName,
 
     case UnresolvedModuleRef::Import:
       if (ResolvedMod)
-        Unresolved.Mod->Imports.push_back(ResolvedMod);
+        Unresolved.Mod->Imports.insert(ResolvedMod);
       continue;
 
     case UnresolvedModuleRef::Export:
@@ -3785,6 +3517,14 @@ ASTReader::ASTReadResult ASTReader::ReadAST(const std::string &FileName,
 
 static ASTFileSignature readASTFileSignature(llvm::BitstreamReader &StreamFile);
 
+/// \brief Whether \p Stream starts with the AST/PCH file magic number 'CPCH'.
+static bool startsWithASTFileMagic(BitstreamCursor &Stream) {
+  return Stream.Read(8) == 'C' &&
+         Stream.Read(8) == 'P' &&
+         Stream.Read(8) == 'C' &&
+         Stream.Read(8) == 'H';
+}
+
 ASTReader::ASTReadResult
 ASTReader::ReadASTCore(StringRef FileName,
                        ModuleKind Type,
@@ -3854,10 +3594,7 @@ ASTReader::ReadASTCore(StringRef FileName,
   F.SizeInBits = F.Buffer->getBufferSize() * 8;
   
   // Sniff for the signature.
-  if (Stream.Read(8) != 'C' ||
-      Stream.Read(8) != 'P' ||
-      Stream.Read(8) != 'C' ||
-      Stream.Read(8) != 'H') {
+  if (!startsWithASTFileMagic(Stream)) {
     Diag(diag::err_not_a_pch_file) << FileName;
     return Failure;
   }
@@ -3924,7 +3661,7 @@ ASTReader::ReadASTCore(StringRef FileName,
   return Success;
 }
 
-void ASTReader::InitializeContext() {  
+void ASTReader::InitializeContext() {
   // If there's a listener, notify them that we "read" the translation unit.
   if (DeserializationListener)
     DeserializationListener->DeclRead(PREDEF_DECL_TRANSLATION_UNIT_ID, 
@@ -4047,24 +3784,19 @@ void ASTReader::InitializeContext() {
   }
 
   // Re-export any modules that were imported by a non-module AST file.
-  // FIXME: This does not make macro-only imports visible again. It also doesn't
-  // make #includes mapped to module imports visible.
+  // FIXME: This does not make macro-only imports visible again.
   for (auto &Import : ImportedModules) {
-    if (Module *Imported = getSubmodule(Import.ID))
+    if (Module *Imported = getSubmodule(Import.ID)) {
       makeModuleVisible(Imported, Module::AllVisible,
-                        /*ImportLoc=*/Import.ImportLoc,
-                        /*Complain=*/false);
+                        /*ImportLoc=*/Import.ImportLoc);
+      PP.makeModuleVisible(Imported, Import.ImportLoc);
+    }
   }
   ImportedModules.clear();
 }
 
 void ASTReader::finalizeForWriting() {
-  while (!HiddenNamesMap.empty()) {
-    auto HiddenNames = std::move(*HiddenNamesMap.begin());
-    HiddenNamesMap.erase(HiddenNamesMap.begin());
-    makeNamesVisible(HiddenNames.second, HiddenNames.first,
-                     /*FromFinalization*/true);
-  }
+  // Nothing to do for now.
 }
 
 /// \brief Given a cursor at the start of an AST file, scan ahead and drop the
@@ -4097,14 +3829,12 @@ static bool SkipCursorToBlock(BitstreamCursor &Cursor, unsigned BlockID) {
   }
 }
 
+/// \brief Reads and return the signature record from \p StreamFile's control
+/// block, or else returns 0.
 static ASTFileSignature readASTFileSignature(llvm::BitstreamReader &StreamFile){
   BitstreamCursor Stream(StreamFile);
-  if (Stream.Read(8) != 'C' ||
-      Stream.Read(8) != 'P' ||
-      Stream.Read(8) != 'C' ||
-      Stream.Read(8) != 'H') {
+  if (!startsWithASTFileMagic(Stream))
     return 0;
-  }
 
   // Scan for the CONTROL_BLOCK_ID block.
   if (SkipCursorToBlock(Stream, CONTROL_BLOCK_ID))
@@ -4146,10 +3876,7 @@ std::string ASTReader::getOriginalSourceFile(const std::string &ASTFileName,
   BitstreamCursor Stream(StreamFile);
 
   // Sniff for the signature.
-  if (Stream.Read(8) != 'C' ||
-      Stream.Read(8) != 'P' ||
-      Stream.Read(8) != 'C' ||
-      Stream.Read(8) != 'H') {
+  if (!startsWithASTFileMagic(Stream)) {
     Diags.Report(diag::err_fe_not_a_pch_file) << ASTFileName;
     return std::string();
   }
@@ -4184,16 +3911,19 @@ namespace {
     const LangOptions &ExistingLangOpts;
     const TargetOptions &ExistingTargetOpts;
     const PreprocessorOptions &ExistingPPOpts;
+    std::string ExistingModuleCachePath;
     FileManager &FileMgr;
-    
+
   public:
     SimplePCHValidator(const LangOptions &ExistingLangOpts,
                        const TargetOptions &ExistingTargetOpts,
                        const PreprocessorOptions &ExistingPPOpts,
+                       StringRef ExistingModuleCachePath,
                        FileManager &FileMgr)
       : ExistingLangOpts(ExistingLangOpts),
         ExistingTargetOpts(ExistingTargetOpts),
         ExistingPPOpts(ExistingPPOpts),
+        ExistingModuleCachePath(ExistingModuleCachePath),
         FileMgr(FileMgr)
     {
     }
@@ -4203,9 +3933,17 @@ namespace {
       return checkLanguageOptions(ExistingLangOpts, LangOpts, nullptr,
                                   AllowCompatibleDifferences);
     }
-    bool ReadTargetOptions(const TargetOptions &TargetOpts,
-                           bool Complain) override {
-      return checkTargetOptions(ExistingTargetOpts, TargetOpts, nullptr);
+    bool ReadTargetOptions(const TargetOptions &TargetOpts, bool Complain,
+                           bool AllowCompatibleDifferences) override {
+      return checkTargetOptions(ExistingTargetOpts, TargetOpts, nullptr,
+                                AllowCompatibleDifferences);
+    }
+    bool ReadHeaderSearchOptions(const HeaderSearchOptions &HSOpts,
+                                 StringRef SpecificModuleCachePath,
+                                 bool Complain) override {
+      return checkHeaderSearchOptions(HSOpts, SpecificModuleCachePath,
+                                      ExistingModuleCachePath,
+                                      nullptr, ExistingLangOpts);
     }
     bool ReadPreprocessorOptions(const PreprocessorOptions &PPOpts,
                                  bool Complain,
@@ -4220,6 +3958,8 @@ bool ASTReader::readASTFileControlBlock(StringRef Filename,
                                         FileManager &FileMgr,
                                         ASTReaderListener &Listener) {
   // Open the AST file.
+  // FIXME: This allows use of the VFS; we do not allow use of the
+  // VFS when actually loading a module.
   auto Buffer = FileMgr.getBufferForFile(Filename);
   if (!Buffer) {
     return true;
@@ -4232,12 +3972,8 @@ bool ASTReader::readASTFileControlBlock(StringRef Filename,
   BitstreamCursor Stream(StreamFile);
 
   // Sniff for the signature.
-  if (Stream.Read(8) != 'C' ||
-      Stream.Read(8) != 'P' ||
-      Stream.Read(8) != 'C' ||
-      Stream.Read(8) != 'H') {
+  if (!startsWithASTFileMagic(Stream))
     return true;
-  }
 
   // Scan for the CONTROL_BLOCK_ID block.
   if (SkipCursorToBlock(Stream, CONTROL_BLOCK_ID))
@@ -4310,7 +4046,8 @@ bool ASTReader::readASTFileControlBlock(StringRef Filename,
       break;
 
     case TARGET_OPTIONS:
-      if (ParseTargetOptions(Record, false, Listener))
+      if (ParseTargetOptions(Record, false, Listener,
+                             /*AllowCompatibleConfigurationMismatch*/ false))
         return true;
       break;
 
@@ -4343,7 +4080,7 @@ bool ASTReader::readASTFileControlBlock(StringRef Filename,
 
       unsigned NumInputFiles = Record[0];
       unsigned NumUserFiles = Record[1];
-      const uint32_t *InputFileOffs = (const uint32_t *)Blob.data();
+      const uint64_t *InputFileOffs = (const uint64_t *)Blob.data();
       for (unsigned I = 0; I != NumInputFiles; ++I) {
         // Go find this input file.
         bool isSystemFile = I >= NumUserFiles;
@@ -4389,6 +4126,20 @@ bool ASTReader::readASTFileControlBlock(StringRef Filename,
       break;
     }
 
+    case KNOWN_MODULE_FILES: {
+      // Known-but-not-technically-used module files are treated as imports.
+      if (!NeedsImports)
+        break;
+
+      unsigned Idx = 0, N = Record.size();
+      while (Idx < N) {
+        std::string Filename = ReadString(Record, Idx);
+        ResolveImportedPath(Filename, ModuleDir);
+        Listener.visitImport(Filename);
+      }
+      break;
+    }
+
     default:
       // No other validation to perform.
       break;
@@ -4401,8 +4152,10 @@ bool ASTReader::isAcceptableASTFile(StringRef Filename,
                                     FileManager &FileMgr,
                                     const LangOptions &LangOpts,
                                     const TargetOptions &TargetOpts,
-                                    const PreprocessorOptions &PPOpts) {
-  SimplePCHValidator validator(LangOpts, TargetOpts, PPOpts, FileMgr);
+                                    const PreprocessorOptions &PPOpts,
+                                    std::string ExistingModuleCachePath) {
+  SimplePCHValidator validator(LangOpts, TargetOpts, PPOpts,
+                               ExistingModuleCachePath, FileMgr);
   return !readASTFileControlBlock(Filename, FileMgr, validator);
 }
 
@@ -4529,13 +4282,19 @@ ASTReader::ReadSubmoduleBlock(ModuleFile &F, unsigned ClientLoadCapabilities) {
     }
         
     case SUBMODULE_UMBRELLA_HEADER: {
-      if (const FileEntry *Umbrella = PP.getFileManager().getFile(Blob)) {
+      std::string Filename = Blob;
+      ResolveImportedPath(F, Filename);
+      if (auto *Umbrella = PP.getFileManager().getFile(Filename)) {
         if (!CurrentModule->getUmbrellaHeader())
-          ModMap.setUmbrellaHeader(CurrentModule, Umbrella);
-        else if (CurrentModule->getUmbrellaHeader() != Umbrella) {
-          if ((ClientLoadCapabilities & ARR_OutOfDate) == 0)
-            Error("mismatched umbrella headers in submodule");
-          return OutOfDate;
+          ModMap.setUmbrellaHeader(CurrentModule, Umbrella, Blob);
+        else if (CurrentModule->getUmbrellaHeader().Entry != Umbrella) {
+          // This can be a spurious difference caused by changing the VFS to
+          // point to a different copy of the file, and it is too late to
+          // to rebuild safely.
+          // FIXME: If we wrote the virtual paths instead of the 'real' paths,
+          // after input file validation only real problems would remain and we
+          // could just error. For now, assume it's okay.
+          break;
         }
       }
       break;
@@ -4561,11 +4320,12 @@ ASTReader::ReadSubmoduleBlock(ModuleFile &F, unsigned ClientLoadCapabilities) {
     }
 
     case SUBMODULE_UMBRELLA_DIR: {
-      if (const DirectoryEntry *Umbrella
-                                  = PP.getFileManager().getDirectory(Blob)) {
+      std::string Dirname = Blob;
+      ResolveImportedPath(F, Dirname);
+      if (auto *Umbrella = PP.getFileManager().getDirectory(Dirname)) {
         if (!CurrentModule->getUmbrellaDir())
-          ModMap.setUmbrellaDir(CurrentModule, Umbrella);
-        else if (CurrentModule->getUmbrellaDir() != Umbrella) {
+          ModMap.setUmbrellaDir(CurrentModule, Umbrella, Blob);
+        else if (CurrentModule->getUmbrellaDir().Entry != Umbrella) {
           if ((ClientLoadCapabilities & ARR_OutOfDate) == 0)
             Error("mismatched umbrella directories in submodule");
           return OutOfDate;
@@ -4696,9 +4456,9 @@ bool ASTReader::ParseLanguageOptions(const RecordData &Record,
                                       AllowCompatibleDifferences);
 }
 
-bool ASTReader::ParseTargetOptions(const RecordData &Record,
-                                   bool Complain,
-                                   ASTReaderListener &Listener) {
+bool ASTReader::ParseTargetOptions(const RecordData &Record, bool Complain,
+                                   ASTReaderListener &Listener,
+                                   bool AllowCompatibleDifferences) {
   unsigned Idx = 0;
   TargetOptions TargetOpts;
   TargetOpts.Triple = ReadString(Record, Idx);
@@ -4711,7 +4471,8 @@ bool ASTReader::ParseTargetOptions(const RecordData &Record,
     TargetOpts.Features.push_back(ReadString(Record, Idx));
   }
 
-  return Listener.ReadTargetOptions(TargetOpts, Complain);
+  return Listener.ReadTargetOptions(TargetOpts, Complain,
+                                    AllowCompatibleDifferences);
 }
 
 bool ASTReader::ParseDiagnosticOptions(const RecordData &Record, bool Complain,
@@ -4773,8 +4534,10 @@ bool ASTReader::ParseHeaderSearchOptions(const RecordData &Record,
   HSOpts.UseStandardSystemIncludes = Record[Idx++];
   HSOpts.UseStandardCXXIncludes = Record[Idx++];
   HSOpts.UseLibcxx = Record[Idx++];
+  std::string SpecificModuleCachePath = ReadString(Record, Idx);
 
-  return Listener.ReadHeaderSearchOptions(HSOpts, Complain);
+  return Listener.ReadHeaderSearchOptions(HSOpts, SpecificModuleCachePath,
+                                          Complain);
 }
 
 bool ASTReader::ParsePreprocessorOptions(const RecordData &Record,
@@ -4823,21 +4586,22 @@ ASTReader::getModulePreprocessedEntity(unsigned GlobalIndex) {
   return std::make_pair(M, LocalIndex);
 }
 
-std::pair<PreprocessingRecord::iterator, PreprocessingRecord::iterator>
+llvm::iterator_range<PreprocessingRecord::iterator>
 ASTReader::getModulePreprocessedEntities(ModuleFile &Mod) const {
   if (PreprocessingRecord *PPRec = PP.getPreprocessingRecord())
     return PPRec->getIteratorsForLoadedRange(Mod.BasePreprocessedEntityID,
                                              Mod.NumPreprocessedEntities);
 
-  return std::make_pair(PreprocessingRecord::iterator(),
-                        PreprocessingRecord::iterator());
+  return llvm::make_range(PreprocessingRecord::iterator(),
+                          PreprocessingRecord::iterator());
 }
 
-std::pair<ASTReader::ModuleDeclIterator, ASTReader::ModuleDeclIterator>
+llvm::iterator_range<ASTReader::ModuleDeclIterator>
 ASTReader::getModuleFileLevelDecls(ModuleFile &Mod) {
-  return std::make_pair(ModuleDeclIterator(this, &Mod, Mod.FileSortedDecls),
-                        ModuleDeclIterator(this, &Mod,
-                                 Mod.FileSortedDecls + Mod.NumFileSortedDecls));
+  return llvm::make_range(
+      ModuleDeclIterator(this, &Mod, Mod.FileSortedDecls),
+      ModuleDeclIterator(this, &Mod,
+                         Mod.FileSortedDecls + Mod.NumFileSortedDecls));
 }
 
 PreprocessedEntity *ASTReader::ReadPreprocessedEntity(unsigned Index) {
@@ -4873,13 +4637,14 @@ PreprocessedEntity *ASTReader::ReadPreprocessedEntity(unsigned Index) {
   case PPD_MACRO_EXPANSION: {
     bool isBuiltin = Record[0];
     IdentifierInfo *Name = nullptr;
-    MacroDefinition *Def = nullptr;
+    MacroDefinitionRecord *Def = nullptr;
     if (isBuiltin)
       Name = getLocalIdentifier(M, Record[1]);
     else {
-      PreprocessedEntityID
-          GlobalID = getGlobalPreprocessedEntityID(M, Record[1]);
-      Def =cast<MacroDefinition>(PPRec.getLoadedPreprocessedEntity(GlobalID-1));
+      PreprocessedEntityID GlobalID =
+          getGlobalPreprocessedEntityID(M, Record[1]);
+      Def = cast<MacroDefinitionRecord>(
+          PPRec.getLoadedPreprocessedEntity(GlobalID - 1));
     }
 
     MacroExpansion *ME;
@@ -4895,8 +4660,7 @@ PreprocessedEntity *ASTReader::ReadPreprocessedEntity(unsigned Index) {
     // Decode the identifier info and then check again; if the macro is
     // still defined and associated with the identifier,
     IdentifierInfo *II = getLocalIdentifier(M, Record[0]);
-    MacroDefinition *MD
-      = new (PPRec) MacroDefinition(II, Range);
+    MacroDefinitionRecord *MD = new (PPRec) MacroDefinitionRecord(II, Range);
 
     if (DeserializationListener)
       DeserializationListener->MacroDefinitionRead(PPID, MD);
@@ -6081,6 +5845,12 @@ Decl *ASTReader::GetExternalDecl(uint32_t ID) {
   return GetDecl(ID);
 }
 
+template<typename TemplateSpecializationDecl>
+static void completeRedeclChainForTemplateSpecialization(Decl *D) {
+  if (auto *TSD = dyn_cast<TemplateSpecializationDecl>(D))
+    TSD->getSpecializedTemplate()->LoadLazySpecializations();
+}
+
 void ASTReader::CompleteRedeclChain(const Decl *D) {
   if (NumCurrentElementsDeserializing) {
     // We arrange to not care about the complete redeclaration chain while we're
@@ -6114,6 +5884,47 @@ void ASTReader::CompleteRedeclChain(const Decl *D) {
       D->getDeclContext()->decls_begin();
     }
   }
+
+  if (auto *CTSD = dyn_cast<ClassTemplateSpecializationDecl>(D))
+    CTSD->getSpecializedTemplate()->LoadLazySpecializations();
+  if (auto *VTSD = dyn_cast<VarTemplateSpecializationDecl>(D))
+    VTSD->getSpecializedTemplate()->LoadLazySpecializations();
+  if (auto *FD = dyn_cast<FunctionDecl>(D)) {
+    if (auto *Template = FD->getPrimaryTemplate())
+      Template->LoadLazySpecializations();
+  }
+}
+
+uint64_t ASTReader::ReadCXXCtorInitializersRef(ModuleFile &M,
+                                               const RecordData &Record,
+                                               unsigned &Idx) {
+  if (Idx >= Record.size() || Record[Idx] > M.LocalNumCXXCtorInitializers) {
+    Error("malformed AST file: missing C++ ctor initializers");
+    return 0;
+  }
+
+  unsigned LocalID = Record[Idx++];
+  return getGlobalBitOffset(M, M.CXXCtorInitializersOffsets[LocalID - 1]);
+}
+
+CXXCtorInitializer **
+ASTReader::GetExternalCXXCtorInitializers(uint64_t Offset) {
+  RecordLocation Loc = getLocalBitOffset(Offset);
+  BitstreamCursor &Cursor = Loc.F->DeclsCursor;
+  SavedStreamPosition SavedPosition(Cursor);
+  Cursor.JumpToBit(Loc.Offset);
+  ReadingKindTracker ReadingKind(Read_Decl, *this);
+
+  RecordData Record;
+  unsigned Code = Cursor.ReadCode();
+  unsigned RecCode = Cursor.readRecord(Code, Record);
+  if (RecCode != DECL_CXX_CTOR_INITIALIZERS) {
+    Error("malformed AST file: missing C++ ctor initializers");
+    return nullptr;
+  }
+
+  unsigned Idx = 0;
+  return ReadCXXCtorInitializers(*Loc.F, Record, Idx);
 }
 
 uint64_t ASTReader::readCXXBaseSpecifiers(ModuleFile &M,
@@ -6165,6 +5976,10 @@ ASTReader::getGlobalDeclID(ModuleFile &F, LocalDeclID LocalID) const {
 
 bool ASTReader::isDeclIDFromModule(serialization::GlobalDeclID ID,
                                    ModuleFile &M) const {
+  // Predefined decls aren't from any module.
+  if (ID < NUM_PREDEF_DECL_IDS)
+    return false;
+
   GlobalDeclMapType::const_iterator I = GlobalDeclMap.find(ID);
   assert(I != GlobalDeclMap.end() && "Corrupted global declaration map");
   return &M == I->second;
@@ -6197,39 +6012,55 @@ SourceLocation ASTReader::getSourceLocationForDeclID(GlobalDeclID ID) {
   return ReadSourceLocation(*Rec.F, RawLocation);
 }
 
-Decl *ASTReader::GetExistingDecl(DeclID ID) {
-  if (ID < NUM_PREDEF_DECL_IDS) {
-    switch ((PredefinedDeclIDs)ID) {
-    case PREDEF_DECL_NULL_ID:
-      return nullptr;
+static Decl *getPredefinedDecl(ASTContext &Context, PredefinedDeclIDs ID) {
+  switch (ID) {
+  case PREDEF_DECL_NULL_ID:
+    return nullptr;
+
+  case PREDEF_DECL_TRANSLATION_UNIT_ID:
+    return Context.getTranslationUnitDecl();
 
-    case PREDEF_DECL_TRANSLATION_UNIT_ID:
-      return Context.getTranslationUnitDecl();
+  case PREDEF_DECL_OBJC_ID_ID:
+    return Context.getObjCIdDecl();
 
-    case PREDEF_DECL_OBJC_ID_ID:
-      return Context.getObjCIdDecl();
+  case PREDEF_DECL_OBJC_SEL_ID:
+    return Context.getObjCSelDecl();
 
-    case PREDEF_DECL_OBJC_SEL_ID:
-      return Context.getObjCSelDecl();
+  case PREDEF_DECL_OBJC_CLASS_ID:
+    return Context.getObjCClassDecl();
 
-    case PREDEF_DECL_OBJC_CLASS_ID:
-      return Context.getObjCClassDecl();
+  case PREDEF_DECL_OBJC_PROTOCOL_ID:
+    return Context.getObjCProtocolDecl();
 
-    case PREDEF_DECL_OBJC_PROTOCOL_ID:
-      return Context.getObjCProtocolDecl();
+  case PREDEF_DECL_INT_128_ID:
+    return Context.getInt128Decl();
 
-    case PREDEF_DECL_INT_128_ID:
-      return Context.getInt128Decl();
+  case PREDEF_DECL_UNSIGNED_INT_128_ID:
+    return Context.getUInt128Decl();
 
-    case PREDEF_DECL_UNSIGNED_INT_128_ID:
-      return Context.getUInt128Decl();
+  case PREDEF_DECL_OBJC_INSTANCETYPE_ID:
+    return Context.getObjCInstanceTypeDecl();
 
-    case PREDEF_DECL_OBJC_INSTANCETYPE_ID:
-      return Context.getObjCInstanceTypeDecl();
+  case PREDEF_DECL_BUILTIN_VA_LIST_ID:
+    return Context.getBuiltinVaListDecl();
 
-    case PREDEF_DECL_BUILTIN_VA_LIST_ID:
-      return Context.getBuiltinVaListDecl();
+  case PREDEF_DECL_EXTERN_C_CONTEXT_ID:
+    return Context.getExternCContextDecl();
+  }
+  llvm_unreachable("PredefinedDeclIDs unknown enum value");
+}
+
+Decl *ASTReader::GetExistingDecl(DeclID ID) {
+  if (ID < NUM_PREDEF_DECL_IDS) {
+    Decl *D = getPredefinedDecl(Context, (PredefinedDeclIDs)ID);
+    if (D) {
+      // Track that we have merged the declaration with ID \p ID into the
+      // pre-existing predefined declaration \p D.
+      auto &Merged = MergedDecls[D->getCanonicalDecl()];
+      if (Merged.empty())
+        Merged.push_back(ID);
     }
+    return D;
   }
 
   unsigned Index = ID - NUM_PREDEF_DECL_IDS;
@@ -6326,10 +6157,7 @@ namespace {
         PredefsVisited[I] = false;
     }
 
-    static bool visit(ModuleFile &M, bool Preorder, void *UserData) {
-      if (Preorder)
-        return false;
-
+    static bool visitPostorder(ModuleFile &M, void *UserData) {
       FindExternalLexicalDeclsVisitor *This
         = static_cast<FindExternalLexicalDeclsVisitor *>(UserData);
 
@@ -6371,7 +6199,8 @@ ExternalLoadResult ASTReader::FindExternalLexicalDecls(const DeclContext *DC,
   // There might be lexical decls in multiple modules, for the TU at
   // least. Walk all of the modules in the order they were loaded.
   FindExternalLexicalDeclsVisitor Visitor(*this, DC, isKindWeWant, Decls);
-  ModuleMgr.visitDepthFirst(&FindExternalLexicalDeclsVisitor::visit, &Visitor);
+  ModuleMgr.visitDepthFirst(
+      nullptr, &FindExternalLexicalDeclsVisitor::visitPostorder, &Visitor);
   ++NumLexicalDeclContextsRead;
   return ELR_Success;
 }
@@ -6460,13 +6289,16 @@ namespace {
     ArrayRef<const DeclContext *> Contexts;
     DeclarationName Name;
     SmallVectorImpl<NamedDecl *> &Decls;
+    llvm::SmallPtrSetImpl<NamedDecl *> &DeclSet;
 
   public:
     DeclContextNameLookupVisitor(ASTReader &Reader,
                                  ArrayRef<const DeclContext *> Contexts,
                                  DeclarationName Name,
-                                 SmallVectorImpl<NamedDecl *> &Decls)
-      : Reader(Reader), Contexts(Contexts), Name(Name), Decls(Decls) { }
+                                 SmallVectorImpl<NamedDecl *> &Decls,
+                                 llvm::SmallPtrSetImpl<NamedDecl *> &DeclSet)
+      : Reader(Reader), Contexts(Contexts), Name(Name), Decls(Decls),
+        DeclSet(DeclSet) { }
 
     static bool visit(ModuleFile &M, void *UserData) {
       DeclContextNameLookupVisitor *This
@@ -6515,7 +6347,8 @@ namespace {
 
         // Record this declaration.
         FoundAnything = true;
-        This->Decls.push_back(ND);
+        if (This->DeclSet.insert(ND).second)
+          This->Decls.push_back(ND);
       }
 
       return FoundAnything;
@@ -6555,6 +6388,7 @@ ASTReader::FindExternalVisibleDeclsByName(const DeclContext *DC,
   Deserializing LookupResults(this);
 
   SmallVector<NamedDecl *, 64> Decls;
+  llvm::SmallPtrSet<NamedDecl*, 64> DeclSet;
 
   // Compute the declaration contexts we need to look into. Multiple such
   // declaration contexts occur when two declaration contexts from disjoint
@@ -6572,7 +6406,7 @@ ASTReader::FindExternalVisibleDeclsByName(const DeclContext *DC,
   }
 
   auto LookUpInContexts = [&](ArrayRef<const DeclContext*> Contexts) {
-    DeclContextNameLookupVisitor Visitor(*this, Contexts, Name, Decls);
+    DeclContextNameLookupVisitor Visitor(*this, Contexts, Name, Decls, DeclSet);
 
     // If we can definitively determine which module file to look into,
     // only look there. Otherwise, look in all module files.
@@ -6592,19 +6426,14 @@ ASTReader::FindExternalVisibleDeclsByName(const DeclContext *DC,
   // individually, because finding an entity in one of them doesn't imply that
   // we can't find a different entity in another one.
   if (isa<CXXRecordDecl>(DC)) {
-    auto Kind = Name.getNameKind();
-    if (Kind == DeclarationName::CXXConstructorName ||
-        Kind == DeclarationName::CXXDestructorName ||
-        (Kind == DeclarationName::CXXOperatorName &&
-         Name.getCXXOverloadedOperator() == OO_Equal)) {
-      auto Merged = MergedLookups.find(DC);
-      if (Merged != MergedLookups.end()) {
-        for (unsigned I = 0; I != Merged->second.size(); ++I) {
-          LookUpInContexts(Merged->second[I]);
-          // We might have just added some more merged lookups. If so, our
-          // iterator is now invalid, so grab a fresh one before continuing.
-          Merged = MergedLookups.find(DC);
-        }
+    auto Merged = MergedLookups.find(DC);
+    if (Merged != MergedLookups.end()) {
+      for (unsigned I = 0; I != Merged->second.size(); ++I) {
+        const DeclContext *Context = Merged->second[I];
+        LookUpInContexts(Context);
+        // We might have just added some more merged lookups. If so, our
+        // iterator is now invalid, so grab a fresh one before continuing.
+        Merged = MergedLookups.find(DC);
       }
     }
   }
@@ -6621,6 +6450,7 @@ namespace {
     ASTReader &Reader;
     SmallVectorImpl<const DeclContext *> &Contexts;
     DeclsMap &Decls;
+    llvm::SmallPtrSet<NamedDecl *, 256> DeclSet;
     bool VisitAll;
 
   public:
@@ -6665,7 +6495,8 @@ namespace {
 
           // Record this declaration.
           FoundAnything = true;
-          This->Decls[ND->getDeclName()].push_back(ND);
+          if (This->DeclSet.insert(ND).second)
+            This->Decls[ND->getDeclName()].push_back(ND);
         }
       }
 
@@ -6731,6 +6562,12 @@ void ASTReader::PassInterestingDeclsToConsumer() {
   SaveAndRestore<bool> GuardPassingDeclsToConsumer(PassingDeclsToConsumer,
                                                    true);
 
+  // Ensure that we've loaded all potentially-interesting declarations
+  // that need to be eagerly loaded.
+  for (auto ID : EagerlyDeserializedDecls)
+    GetDecl(ID);
+  EagerlyDeserializedDecls.clear();
+
   while (!InterestingDecls.empty()) {
     Decl *D = InterestingDecls.front();
     InterestingDecls.pop_front();
@@ -6749,17 +6586,11 @@ void ASTReader::PassInterestingDeclToConsumer(Decl *D) {
 void ASTReader::StartTranslationUnit(ASTConsumer *Consumer) {
   this->Consumer = Consumer;
 
-  if (!Consumer)
-    return;
+  if (Consumer)
+    PassInterestingDeclsToConsumer();
 
-  for (unsigned I = 0, N = EagerlyDeserializedDecls.size(); I != N; ++I) {
-    // Force deserialization of this decl, which will cause it to be queued for
-    // passing to the consumer.
-    GetDecl(EagerlyDeserializedDecls[I]);
-  }
-  EagerlyDeserializedDecls.clear();
-
-  PassInterestingDeclsToConsumer();
+  if (DeserializationListener)
+    DeserializationListener->ReaderInitialized(this);
 }
 
 void ASTReader::PrintStats() {
@@ -7195,6 +7026,21 @@ void ASTReader::ReadUndefinedButUsed(
   }
 }
 
+void ASTReader::ReadMismatchingDeleteExpressions(llvm::MapVector<
+    FieldDecl *, llvm::SmallVector<std::pair<SourceLocation, bool>, 4>> &
+                                                     Exprs) {
+  for (unsigned Idx = 0, N = DelayedDeleteExprs.size(); Idx != N;) {
+    FieldDecl *FD = cast<FieldDecl>(GetDecl(DelayedDeleteExprs[Idx++]));
+    uint64_t Count = DelayedDeleteExprs[Idx++];
+    for (uint64_t C = 0; C < Count; ++C) {
+      SourceLocation DeleteLoc =
+          SourceLocation::getFromRawEncoding(DelayedDeleteExprs[Idx++]);
+      const bool IsArrayForm = DelayedDeleteExprs[Idx++];
+      Exprs[FD].push_back(std::make_pair(DeleteLoc, IsArrayForm));
+    }
+  }
+}
+
 void ASTReader::ReadTentativeDefinitions(
                   SmallVectorImpl<VarDecl *> &TentativeDefs) {
   for (unsigned I = 0, N = TentativeDefinitions.size(); I != N; ++I) {
@@ -7237,16 +7083,6 @@ void ASTReader::ReadExtVectorDecls(SmallVectorImpl<TypedefNameDecl *> &Decls) {
   ExtVectorDecls.clear();
 }
 
-void ASTReader::ReadDynamicClasses(SmallVectorImpl<CXXRecordDecl *> &Decls) {
-  for (unsigned I = 0, N = DynamicClasses.size(); I != N; ++I) {
-    CXXRecordDecl *D
-      = dyn_cast_or_null<CXXRecordDecl>(GetDecl(DynamicClasses[I]));
-    if (D)
-      Decls.push_back(D);
-  }
-  DynamicClasses.clear();
-}
-
 void ASTReader::ReadUnusedLocalTypedefNameCandidates(
     llvm::SmallSetVector<const TypedefNameDecl *, 4> &Decls) {
   for (unsigned I = 0, N = UnusedLocalTypedefNameCandidates.size(); I != N;
@@ -7259,17 +7095,6 @@ void ASTReader::ReadUnusedLocalTypedefNameCandidates(
   UnusedLocalTypedefNameCandidates.clear();
 }
 
-void 
-ASTReader::ReadLocallyScopedExternCDecls(SmallVectorImpl<NamedDecl *> &Decls) {
-  for (unsigned I = 0, N = LocallyScopedExternCDecls.size(); I != N; ++I) {
-    NamedDecl *D
-      = dyn_cast_or_null<NamedDecl>(GetDecl(LocallyScopedExternCDecls[I]));
-    if (D)
-      Decls.push_back(D);
-  }
-  LocallyScopedExternCDecls.clear();
-}
-
 void ASTReader::ReadReferencedSelectors(
        SmallVectorImpl<std::pair<Selector, SourceLocation> > &Sels) {
   if (ReferencedSelectorsData.empty())
@@ -7333,7 +7158,7 @@ void ASTReader::ReadPendingInstantiations(
 }
 
 void ASTReader::ReadLateParsedTemplates(
-    llvm::DenseMap<const FunctionDecl *, LateParsedTemplate *> &LPTMap) {
+    llvm::MapVector<const FunctionDecl *, LateParsedTemplate *> &LPTMap) {
   for (unsigned Idx = 0, N = LateParsedTemplates.size(); Idx < N;
        /* In loop */) {
     FunctionDecl *FD = cast<FunctionDecl>(GetDecl(LateParsedTemplates[Idx++]));
@@ -7349,7 +7174,7 @@ void ASTReader::ReadLateParsedTemplates(
     for (unsigned T = 0; T < TokN; ++T)
       LT->Toks.push_back(ReadToken(*F, LateParsedTemplates, Idx));
 
-    LPTMap[FD] = LT;
+    LPTMap.insert(std::make_pair(FD, LT));
   }
 
   LateParsedTemplates.clear();
@@ -7840,92 +7665,89 @@ ASTReader::ReadCXXBaseSpecifier(ModuleFile &F,
   return Result;
 }
 
-std::pair<CXXCtorInitializer **, unsigned>
+CXXCtorInitializer **
 ASTReader::ReadCXXCtorInitializers(ModuleFile &F, const RecordData &Record,
                                    unsigned &Idx) {
-  CXXCtorInitializer **CtorInitializers = nullptr;
   unsigned NumInitializers = Record[Idx++];
-  if (NumInitializers) {
-    CtorInitializers
-        = new (Context) CXXCtorInitializer*[NumInitializers];
-    for (unsigned i=0; i != NumInitializers; ++i) {
-      TypeSourceInfo *TInfo = nullptr;
-      bool IsBaseVirtual = false;
-      FieldDecl *Member = nullptr;
-      IndirectFieldDecl *IndirectMember = nullptr;
-
-      CtorInitializerType Type = (CtorInitializerType)Record[Idx++];
-      switch (Type) {
-      case CTOR_INITIALIZER_BASE:
-        TInfo = GetTypeSourceInfo(F, Record, Idx);
-        IsBaseVirtual = Record[Idx++];
-        break;
-          
-      case CTOR_INITIALIZER_DELEGATING:
-        TInfo = GetTypeSourceInfo(F, Record, Idx);
-        break;
+  assert(NumInitializers && "wrote ctor initializers but have no inits");
+  auto **CtorInitializers = new (Context) CXXCtorInitializer*[NumInitializers];
+  for (unsigned i = 0; i != NumInitializers; ++i) {
+    TypeSourceInfo *TInfo = nullptr;
+    bool IsBaseVirtual = false;
+    FieldDecl *Member = nullptr;
+    IndirectFieldDecl *IndirectMember = nullptr;
 
-       case CTOR_INITIALIZER_MEMBER:
-        Member = ReadDeclAs<FieldDecl>(F, Record, Idx);
-        break;
+    CtorInitializerType Type = (CtorInitializerType)Record[Idx++];
+    switch (Type) {
+    case CTOR_INITIALIZER_BASE:
+      TInfo = GetTypeSourceInfo(F, Record, Idx);
+      IsBaseVirtual = Record[Idx++];
+      break;
 
-       case CTOR_INITIALIZER_INDIRECT_MEMBER:
-        IndirectMember = ReadDeclAs<IndirectFieldDecl>(F, Record, Idx);
-        break;
-      }
+    case CTOR_INITIALIZER_DELEGATING:
+      TInfo = GetTypeSourceInfo(F, Record, Idx);
+      break;
 
-      SourceLocation MemberOrEllipsisLoc = ReadSourceLocation(F, Record, Idx);
-      Expr *Init = ReadExpr(F);
-      SourceLocation LParenLoc = ReadSourceLocation(F, Record, Idx);
-      SourceLocation RParenLoc = ReadSourceLocation(F, Record, Idx);
-      bool IsWritten = Record[Idx++];
-      unsigned SourceOrderOrNumArrayIndices;
-      SmallVector<VarDecl *, 8> Indices;
-      if (IsWritten) {
-        SourceOrderOrNumArrayIndices = Record[Idx++];
-      } else {
-        SourceOrderOrNumArrayIndices = Record[Idx++];
-        Indices.reserve(SourceOrderOrNumArrayIndices);
-        for (unsigned i=0; i != SourceOrderOrNumArrayIndices; ++i)
-          Indices.push_back(ReadDeclAs<VarDecl>(F, Record, Idx));
-      }
+     case CTOR_INITIALIZER_MEMBER:
+      Member = ReadDeclAs<FieldDecl>(F, Record, Idx);
+      break;
 
-      CXXCtorInitializer *BOMInit;
-      if (Type == CTOR_INITIALIZER_BASE) {
-        BOMInit = new (Context) CXXCtorInitializer(Context, TInfo, IsBaseVirtual,
-                                             LParenLoc, Init, RParenLoc,
-                                             MemberOrEllipsisLoc);
-      } else if (Type == CTOR_INITIALIZER_DELEGATING) {
-        BOMInit = new (Context) CXXCtorInitializer(Context, TInfo, LParenLoc,
-                                                   Init, RParenLoc);
-      } else if (IsWritten) {
-        if (Member)
-          BOMInit = new (Context) CXXCtorInitializer(Context, Member, MemberOrEllipsisLoc,
-                                               LParenLoc, Init, RParenLoc);
-        else 
-          BOMInit = new (Context) CXXCtorInitializer(Context, IndirectMember,
-                                               MemberOrEllipsisLoc, LParenLoc,
-                                               Init, RParenLoc);
+     case CTOR_INITIALIZER_INDIRECT_MEMBER:
+      IndirectMember = ReadDeclAs<IndirectFieldDecl>(F, Record, Idx);
+      break;
+    }
+
+    SourceLocation MemberOrEllipsisLoc = ReadSourceLocation(F, Record, Idx);
+    Expr *Init = ReadExpr(F);
+    SourceLocation LParenLoc = ReadSourceLocation(F, Record, Idx);
+    SourceLocation RParenLoc = ReadSourceLocation(F, Record, Idx);
+    bool IsWritten = Record[Idx++];
+    unsigned SourceOrderOrNumArrayIndices;
+    SmallVector<VarDecl *, 8> Indices;
+    if (IsWritten) {
+      SourceOrderOrNumArrayIndices = Record[Idx++];
+    } else {
+      SourceOrderOrNumArrayIndices = Record[Idx++];
+      Indices.reserve(SourceOrderOrNumArrayIndices);
+      for (unsigned i=0; i != SourceOrderOrNumArrayIndices; ++i)
+        Indices.push_back(ReadDeclAs<VarDecl>(F, Record, Idx));
+    }
+
+    CXXCtorInitializer *BOMInit;
+    if (Type == CTOR_INITIALIZER_BASE) {
+      BOMInit = new (Context)
+          CXXCtorInitializer(Context, TInfo, IsBaseVirtual, LParenLoc, Init,
+                             RParenLoc, MemberOrEllipsisLoc);
+    } else if (Type == CTOR_INITIALIZER_DELEGATING) {
+      BOMInit = new (Context)
+          CXXCtorInitializer(Context, TInfo, LParenLoc, Init, RParenLoc);
+    } else if (IsWritten) {
+      if (Member)
+        BOMInit = new (Context) CXXCtorInitializer(
+            Context, Member, MemberOrEllipsisLoc, LParenLoc, Init, RParenLoc);
+      else
+        BOMInit = new (Context)
+            CXXCtorInitializer(Context, IndirectMember, MemberOrEllipsisLoc,
+                               LParenLoc, Init, RParenLoc);
+    } else {
+      if (IndirectMember) {
+        assert(Indices.empty() && "Indirect field improperly initialized");
+        BOMInit = new (Context)
+            CXXCtorInitializer(Context, IndirectMember, MemberOrEllipsisLoc,
+                               LParenLoc, Init, RParenLoc);
       } else {
-        if (IndirectMember) {
-          assert(Indices.empty() && "Indirect field improperly initialized");
-          BOMInit = new (Context) CXXCtorInitializer(Context, IndirectMember,
-                                                     MemberOrEllipsisLoc, LParenLoc,
-                                                     Init, RParenLoc);
-        } else {
-          BOMInit = CXXCtorInitializer::Create(Context, Member, MemberOrEllipsisLoc,
-                                               LParenLoc, Init, RParenLoc,
-                                               Indices.data(), Indices.size());
-        }
+        BOMInit = CXXCtorInitializer::Create(
+            Context, Member, MemberOrEllipsisLoc, LParenLoc, Init, RParenLoc,
+            Indices.data(), Indices.size());
       }
-
-      if (IsWritten)
-        BOMInit->setSourceOrder(SourceOrderOrNumArrayIndices);
-      CtorInitializers[i] = BOMInit;
     }
+
+    if (IsWritten)
+      BOMInit->setSourceOrder(SourceOrderOrNumArrayIndices);
+    CtorInitializers[i] = BOMInit;
   }
 
-  return std::make_pair(CtorInitializers, NumInitializers);
+  return CtorInitializers;
 }
 
 NestedNameSpecifier *
@@ -8203,7 +8025,7 @@ void ASTReader::getInputFiles(ModuleFile &F,
 
 std::string ASTReader::getOwningModuleNameForDiagnostic(const Decl *D) {
   // If we know the owning module, use it.
-  if (Module *M = D->getOwningModule())
+  if (Module *M = D->getImportedOwningModule())
     return M->getFullModuleName();
 
   // Otherwise, use the name of the top-level module the decl is within.
@@ -8243,9 +8065,10 @@ void ASTReader::finishPendingActions() {
 
     // Load pending declaration chains.
     for (unsigned I = 0; I != PendingDeclChains.size(); ++I) {
-      loadPendingDeclChain(PendingDeclChains[I]);
       PendingDeclChainsKnown.erase(PendingDeclChains[I]);
+      loadPendingDeclChain(PendingDeclChains[I]);
     }
+    assert(PendingDeclChainsKnown.empty());
     PendingDeclChains.clear();
 
     // Make the most recent of the top-level declarations visible.
@@ -8298,7 +8121,12 @@ void ASTReader::finishPendingActions() {
       loadDeclUpdateRecords(Update.first, Update.second);
     }
   }
-  
+
+  // At this point, all update records for loaded decls are in place, so any
+  // fake class definitions should have become real.
+  assert(PendingFakeDefinitionData.empty() &&
+         "faked up a class definition but never saw the real one");
+
   // If we deserialized any C++ or Objective-C class definitions, any
   // Objective-C protocol definitions, or any redeclarable templates, make sure
   // that all redeclarations point to the definitions. Note that this can only 
@@ -8309,10 +8137,12 @@ void ASTReader::finishPendingActions() {
         // Make sure that the TagType points at the definition.
         const_cast<TagType*>(TagT)->decl = TD;
       }
-      
+
       if (auto RD = dyn_cast<CXXRecordDecl>(D)) {
-        for (auto R : RD->redecls()) {
-          assert((R == D) == R->isThisDeclarationADefinition() &&
+        for (auto *R = getMostRecentExistingDecl(RD); R;
+             R = R->getPreviousDecl()) {
+          assert((R == D) ==
+                     cast<CXXRecordDecl>(R)->isThisDeclarationADefinition() &&
                  "declaration thinks it's the definition but it isn't");
           cast<CXXRecordDecl>(R)->DefinitionData = RD->DefinitionData;
         }
@@ -8320,34 +8150,36 @@ void ASTReader::finishPendingActions() {
 
       continue;
     }
-    
+
     if (auto ID = dyn_cast<ObjCInterfaceDecl>(D)) {
       // Make sure that the ObjCInterfaceType points at the definition.
       const_cast<ObjCInterfaceType *>(cast<ObjCInterfaceType>(ID->TypeForDecl))
         ->Decl = ID;
-      
-      for (auto R : ID->redecls())
-        R->Data = ID->Data;
-      
+
+      for (auto *R = getMostRecentExistingDecl(ID); R; R = R->getPreviousDecl())
+        cast<ObjCInterfaceDecl>(R)->Data = ID->Data;
+
       continue;
     }
-    
+
     if (auto PD = dyn_cast<ObjCProtocolDecl>(D)) {
-      for (auto R : PD->redecls())
-        R->Data = PD->Data;
-      
+      for (auto *R = getMostRecentExistingDecl(PD); R; R = R->getPreviousDecl())
+        cast<ObjCProtocolDecl>(R)->Data = PD->Data;
+
       continue;
     }
-    
+
     auto RTD = cast<RedeclarableTemplateDecl>(D)->getCanonicalDecl();
-    for (auto R : RTD->redecls())
-      R->Common = RTD->Common;
+    for (auto *R = getMostRecentExistingDecl(RTD); R; R = R->getPreviousDecl())
+      cast<RedeclarableTemplateDecl>(R)->Common = RTD->Common;
   }
   PendingDefinitions.clear();
 
   // Load the bodies of any functions or methods we've encountered. We do
   // this now (delayed) so that we can be sure that the declaration chains
   // have been fully wired up.
+  // FIXME: There seems to be no point in delaying this, it does not depend
+  // on the redecl chains having been wired up.
   for (PendingBodiesMap::iterator PB = PendingBodies.begin(),
                                PBEnd = PendingBodies.end();
        PB != PBEnd; ++PB) {
@@ -8364,6 +8196,11 @@ void ASTReader::finishPendingActions() {
       MD->setLazyBody(PB->second);
   }
   PendingBodies.clear();
+
+  // Do some cleanup.
+  for (auto *ND : PendingMergedDefinitionsToDeduplicate)
+    getContext().deduplicateMergedDefinitonsFor(ND);
+  PendingMergedDefinitionsToDeduplicate.clear();
 }
 
 void ASTReader::diagnoseOdrViolations() {
@@ -8440,6 +8277,10 @@ void ASTReader::diagnoseOdrViolations() {
       // completed. We only really need to mark FieldDecls as invalid here.
       if (!isa<TagDecl>(D))
         D->setInvalidDecl();
+      
+      // Ensure we don't accidentally recursively enter deserialization while
+      // we're producing our diagnostic.
+      Deserializing RecursionGuard(this);
 
       std::string CanonDefModule =
           getOwningModuleNameForDiagnostic(cast<Decl>(CanonDef));
@@ -8461,6 +8302,13 @@ void ASTReader::diagnoseOdrViolations() {
     }
   }
 
+  if (OdrMergeFailures.empty())
+    return;
+
+  // Ensure we don't accidentally recursively enter deserialization while
+  // we're producing our diagnostics.
+  Deserializing RecursionGuard(this);
+
   // Issue any pending ODR-failure diagnostics.
   for (auto &Merge : OdrMergeFailures) {
     // If we've already pointed out a specific problem with this class, don't
@@ -8514,6 +8362,17 @@ void ASTReader::FinishedDeserializing() {
   --NumCurrentElementsDeserializing;
 
   if (NumCurrentElementsDeserializing == 0) {
+    // Propagate exception specification updates along redeclaration chains.
+    while (!PendingExceptionSpecUpdates.empty()) {
+      auto Updates = std::move(PendingExceptionSpecUpdates);
+      PendingExceptionSpecUpdates.clear();
+      for (auto Update : Updates) {
+        auto *FPT = Update.second->getType()->castAs<FunctionProtoType>();
+        SemaObj->UpdateExceptionSpec(Update.second,
+                                     FPT->getExtProtoInfo().ExceptionSpec);
+      }
+    }
+
     diagnoseOdrViolations();
 
     // We are not in recursive loading, so it's safe to pass the "interesting"
@@ -8524,7 +8383,18 @@ void ASTReader::FinishedDeserializing() {
 }
 
 void ASTReader::pushExternalDeclIntoScope(NamedDecl *D, DeclarationName Name) {
-  D = D->getMostRecentDecl();
+  if (IdentifierInfo *II = Name.getAsIdentifierInfo()) {
+    // Remove any fake results before adding any real ones.
+    auto It = PendingFakeLookupResults.find(II);
+    if (It != PendingFakeLookupResults.end()) {
+      for (auto *ND : PendingFakeLookupResults[II])
+        SemaObj->IdResolver.RemoveDecl(ND);
+      // FIXME: this works around module+PCH performance issue.
+      // Rather than erase the result from the map, which is O(n), just clear
+      // the vector of NamedDecls.
+      It->second.clear();
+    }
+  }
 
   if (SemaObj->IdResolver.tryAddTopLevelDecl(D, Name) && SemaObj->TUScope) {
     SemaObj->TUScope->AddDecl(D);
@@ -8562,8 +8432,7 @@ ASTReader::ASTReader(Preprocessor &PP, ASTContext &Context, StringRef isysroot,
       NumLexicalDeclContextsRead(0), TotalLexicalDeclContexts(0),
       NumVisibleDeclContextsRead(0), TotalVisibleDeclContexts(0),
       TotalModulesSizeInBits(0), NumCurrentElementsDeserializing(0),
-      PassingDeclsToConsumer(false), NumCXXBaseSpecifiersLoaded(0),
-      ReadingKind(Read_None) {
+      PassingDeclsToConsumer(false), ReadingKind(Read_None) {
   SourceMgr.setExternalSLocEntrySource(this);
 }
 
diff --git a/lib/Serialization/ASTReaderDecl.cpp b/lib/Serialization/ASTReaderDecl.cpp
index a783183d2ef6..02273ed229a5 100644
--- a/lib/Serialization/ASTReaderDecl.cpp
+++ b/lib/Serialization/ASTReaderDecl.cpp
@@ -66,7 +66,12 @@ namespace clang {
     serialization::DeclID ReadDeclID(const RecordData &R, unsigned &I) {
       return Reader.ReadDeclID(F, R, I);
     }
-    
+
+    void ReadDeclIDList(SmallVectorImpl<DeclID> &IDs) {
+      for (unsigned I = 0, Size = Record[Idx++]; I != Size; ++I)
+        IDs.push_back(ReadDeclID(Record, Idx));
+    }
+
     Decl *ReadDecl(const RecordData &R, unsigned &I) {
       return Reader.ReadDecl(F, R, I);
     }
@@ -103,11 +108,11 @@ namespace clang {
       return Reader.getSubmodule(readSubmoduleID(R, I));
     }
 
-    void ReadCXXRecordDefinition(CXXRecordDecl *D);
+    void ReadCXXRecordDefinition(CXXRecordDecl *D, bool Update);
     void ReadCXXDefinitionData(struct CXXRecordDecl::DefinitionData &Data,
                                const RecordData &R, unsigned &I);
     void MergeDefinitionData(CXXRecordDecl *D,
-                             struct CXXRecordDecl::DefinitionData &NewDD);
+                             struct CXXRecordDecl::DefinitionData &&NewDD);
 
     static NamedDecl *getAnonymousDeclForMerging(ASTReader &Reader,
                                                  DeclContext *DC,
@@ -118,43 +123,42 @@ namespace clang {
     /// \brief RAII class used to capture the first ID within a redeclaration
     /// chain and to introduce it into the list of pending redeclaration chains
     /// on destruction.
-    ///
-    /// The caller can choose not to introduce this ID into the list of pending
-    /// redeclaration chains by calling \c suppress().
     class RedeclarableResult {
       ASTReader &Reader;
       GlobalDeclID FirstID;
+      Decl *MergeWith;
       mutable bool Owning;
       Decl::Kind DeclKind;
-      
-      void operator=(RedeclarableResult &) LLVM_DELETED_FUNCTION;
-      
+
+      void operator=(RedeclarableResult &) = delete;
+
     public:
       RedeclarableResult(ASTReader &Reader, GlobalDeclID FirstID,
-                         Decl::Kind DeclKind)
-        : Reader(Reader), FirstID(FirstID), Owning(true), DeclKind(DeclKind) { }
-
-      RedeclarableResult(const RedeclarableResult &Other)
-        : Reader(Other.Reader), FirstID(Other.FirstID), Owning(Other.Owning) ,
-          DeclKind(Other.DeclKind)
-      { 
+                         Decl *MergeWith, Decl::Kind DeclKind)
+        : Reader(Reader), FirstID(FirstID), MergeWith(MergeWith),
+          Owning(true), DeclKind(DeclKind) {}
+
+      RedeclarableResult(RedeclarableResult &&Other)
+        : Reader(Other.Reader), FirstID(Other.FirstID),
+          MergeWith(Other.MergeWith), Owning(Other.Owning),
+          DeclKind(Other.DeclKind) {
         Other.Owning = false;
       }
 
       ~RedeclarableResult() {
-        if (FirstID && Owning && isRedeclarableDeclKind(DeclKind) &&
-            Reader.PendingDeclChainsKnown.insert(FirstID).second)
-          Reader.PendingDeclChains.push_back(FirstID);
+        if (FirstID && Owning && isRedeclarableDeclKind(DeclKind)) {
+          auto Canon = Reader.GetDecl(FirstID)->getCanonicalDecl();
+          if (Reader.PendingDeclChainsKnown.insert(Canon).second)
+            Reader.PendingDeclChains.push_back(Canon);
+        }
       }
-      
+
       /// \brief Retrieve the first ID.
       GlobalDeclID getFirstID() const { return FirstID; }
-      
-      /// \brief Do not introduce this declaration ID into the set of pending
-      /// declaration chains.
-      void suppress() {
-        Owning = false;
-      }
+
+      /// \brief Get a known declaration that this should be merged with, if
+      /// any.
+      Decl *getKnownMergeTarget() const { return MergeWith; }
     };
 
     /// \brief Class used to capture the result of searching for an existing
@@ -171,7 +175,7 @@ namespace clang {
       unsigned AnonymousDeclNumber;
       IdentifierInfo *TypedefNameForLinkage;
 
-      void operator=(FindExistingResult&) LLVM_DELETED_FUNCTION;
+      void operator=(FindExistingResult&) = delete;
 
     public:
       FindExistingResult(ASTReader &Reader)
@@ -205,6 +209,8 @@ namespace clang {
       operator T*() const { return dyn_cast_or_null<T>(Existing); }
     };
 
+    static DeclContext *getPrimaryContextForMerging(ASTReader &Reader,
+                                                    DeclContext *DC);
     FindExistingResult findExisting(NamedDecl *D);
 
   public:
@@ -216,10 +222,17 @@ namespace clang {
           TypedefNameForLinkage(nullptr), HasPendingBody(false) {}
 
     template <typename DeclT>
+    static Decl *getMostRecentDeclImpl(Redeclarable<DeclT> *D);
+    static Decl *getMostRecentDeclImpl(...);
+    static Decl *getMostRecentDecl(Decl *D);
+
+    template <typename DeclT>
     static void attachPreviousDeclImpl(ASTReader &Reader,
-                                       Redeclarable<DeclT> *D, Decl *Previous);
+                                       Redeclarable<DeclT> *D, Decl *Previous,
+                                       Decl *Canon);
     static void attachPreviousDeclImpl(ASTReader &Reader, ...);
-    static void attachPreviousDecl(ASTReader &Reader, Decl *D, Decl *Previous);
+    static void attachPreviousDecl(ASTReader &Reader, Decl *D, Decl *Previous,
+                                   Decl *Canon);
 
     template <typename DeclT>
     static void attachLatestDeclImpl(Redeclarable<DeclT> *D, Decl *Latest);
@@ -390,9 +403,14 @@ void ASTDeclReader::Visit(Decl *D) {
     // FunctionDecl's body was written last after all other Stmts/Exprs.
     // We only read it if FD doesn't already have a body (e.g., from another
     // module).
-    // FIXME: Also consider = default and = delete.
     // FIXME: Can we diagnose ODR violations somehow?
     if (Record[Idx++]) {
+      if (auto *CD = dyn_cast<CXXConstructorDecl>(FD)) {
+        CD->NumCtorInitializers = Record[Idx++];
+        if (CD->NumCtorInitializers)
+          CD->CtorInitializers =
+              Reader.ReadCXXCtorInitializersRef(F, Record, Idx);
+      }
       Reader.PendingBodies[FD] = GetCurrentCursorOffset();
       HasPendingBody = true;
     }
@@ -440,24 +458,28 @@ void ASTDeclReader::VisitDecl(Decl *D) {
   D->FromASTFile = true;
   D->setModulePrivate(Record[Idx++]);
   D->Hidden = D->isModulePrivate();
-  
+
   // Determine whether this declaration is part of a (sub)module. If so, it
   // may not yet be visible.
   if (unsigned SubmoduleID = readSubmoduleID(Record, Idx)) {
     // Store the owning submodule ID in the declaration.
     D->setOwningModuleID(SubmoduleID);
-    
-    // Module-private declarations are never visible, so there is no work to do.
-    if (!D->isModulePrivate()) {
-      if (Module *Owner = Reader.getSubmodule(SubmoduleID)) {
-        if (Owner->NameVisibility != Module::AllVisible) {
-          // The owning module is not visible. Mark this declaration as hidden.
-          D->Hidden = true;
-          
-          // Note that this declaration was hidden because its owning module is 
-          // not yet visible.
-          Reader.HiddenNamesMap[Owner].HiddenDecls.push_back(D);
-        }
+
+    if (D->Hidden) {
+      // Module-private declarations are never visible, so there is no work to do.
+    } else if (Reader.getContext().getLangOpts().ModulesLocalVisibility) {
+      // If local visibility is being tracked, this declaration will become
+      // hidden and visible as the owning module does. Inform Sema that this
+      // declaration might not be visible.
+      D->Hidden = true;
+    } else if (Module *Owner = Reader.getSubmodule(SubmoduleID)) {
+      if (Owner->NameVisibility != Module::AllVisible) {
+        // The owning module is not visible. Mark this declaration as hidden.
+        D->Hidden = true;
+        
+        // Note that this declaration was hidden because its owning module is 
+        // not yet visible.
+        Reader.HiddenNamesMap[Owner].push_back(D);
       }
     }
   }
@@ -470,8 +492,7 @@ void ASTDeclReader::VisitTranslationUnitDecl(TranslationUnitDecl *TU) {
 void ASTDeclReader::VisitNamedDecl(NamedDecl *ND) {
   VisitDecl(ND);
   ND->setDeclName(Reader.ReadDeclarationName(F, Record, Idx));
-  if (needsAnonymousDeclarationNumber(ND))
-    AnonymousDeclNumber = Record[Idx++];
+  AnonymousDeclNumber = Record[Idx++];
 }
 
 void ASTDeclReader::VisitTypeDecl(TypeDecl *TD) {
@@ -981,8 +1002,9 @@ void ASTDeclReader::VisitObjCImplementationDecl(ObjCImplementationDecl *D) {
   D->setIvarRBraceLoc(ReadSourceLocation(Record, Idx));
   D->setHasNonZeroConstructors(Record[Idx++]);
   D->setHasDestructors(Record[Idx++]);
-  std::tie(D->IvarInitializers, D->NumIvarInitializers) =
-      Reader.ReadCXXCtorInitializers(F, Record, Idx);
+  D->NumIvarInitializers = Record[Idx++];
+  if (D->NumIvarInitializers)
+    D->IvarInitializers = Reader.ReadCXXCtorInitializersRef(F, Record, Idx);
 }
 
 
@@ -1041,13 +1063,15 @@ ASTDeclReader::RedeclarableResult ASTDeclReader::VisitVarDeclImpl(VarDecl *VD) {
   VD->VarDeclBits.SClass = (StorageClass)Record[Idx++];
   VD->VarDeclBits.TSCSpec = Record[Idx++];
   VD->VarDeclBits.InitStyle = Record[Idx++];
-  VD->VarDeclBits.ExceptionVar = Record[Idx++];
-  VD->VarDeclBits.NRVOVariable = Record[Idx++];
-  VD->VarDeclBits.CXXForRangeDecl = Record[Idx++];
-  VD->VarDeclBits.ARCPseudoStrong = Record[Idx++];
-  VD->VarDeclBits.IsConstexpr = Record[Idx++];
-  VD->VarDeclBits.IsInitCapture = Record[Idx++];
-  VD->VarDeclBits.PreviousDeclInSameBlockScope = Record[Idx++];
+  if (!isa<ParmVarDecl>(VD)) {
+    VD->NonParmVarDeclBits.ExceptionVar = Record[Idx++];
+    VD->NonParmVarDeclBits.NRVOVariable = Record[Idx++];
+    VD->NonParmVarDeclBits.CXXForRangeDecl = Record[Idx++];
+    VD->NonParmVarDeclBits.ARCPseudoStrong = Record[Idx++];
+    VD->NonParmVarDeclBits.IsConstexpr = Record[Idx++];
+    VD->NonParmVarDeclBits.IsInitCapture = Record[Idx++];
+    VD->NonParmVarDeclBits.PreviousDeclInSameBlockScope = Record[Idx++];
+  }
   Linkage VarLinkage = Linkage(Record[Idx++]);
   VD->setCachedLinkage(VarLinkage);
 
@@ -1189,13 +1213,13 @@ void ASTDeclReader::VisitNamespaceDecl(NamespaceDecl *D) {
   D->LocStart = ReadSourceLocation(Record, Idx);
   D->RBraceLoc = ReadSourceLocation(Record, Idx);
 
+  // Defer loading the anonymous namespace until we've finished merging
+  // this namespace; loading it might load a later declaration of the
+  // same namespace, and we have an invariant that older declarations
+  // get merged before newer ones try to merge.
+  GlobalDeclID AnonNamespace = 0;
   if (Redecl.getFirstID() == ThisDeclID) {
-    // Each module has its own anonymous namespace, which is disjoint from
-    // any other module's anonymous namespaces, so don't attach the anonymous
-    // namespace at all.
-    NamespaceDecl *Anon = ReadDeclAs<NamespaceDecl>(Record, Idx);
-    if (F.Kind != MK_ImplicitModule && F.Kind != MK_ExplicitModule)
-      D->setAnonymousNamespace(Anon);
+    AnonNamespace = ReadDeclID(Record, Idx);
   } else {
     // Link this namespace back to the first declaration, which has already
     // been deserialized.
@@ -1203,6 +1227,15 @@ void ASTDeclReader::VisitNamespaceDecl(NamespaceDecl *D) {
   }
 
   mergeRedeclarable(D, Redecl);
+
+  if (AnonNamespace) {
+    // Each module has its own anonymous namespace, which is disjoint from
+    // any other module's anonymous namespaces, so don't attach the anonymous
+    // namespace at all.
+    NamespaceDecl *Anon = cast<NamespaceDecl>(Reader.GetDecl(AnonNamespace));
+    if (F.Kind != MK_ImplicitModule && F.Kind != MK_ExplicitModule)
+      D->setAnonymousNamespace(Anon);
+  }
 }
 
 void ASTDeclReader::VisitNamespaceAliasDecl(NamespaceAliasDecl *D) {
@@ -1353,7 +1386,7 @@ void ASTDeclReader::ReadCXXDefinitionData(
 }
 
 void ASTDeclReader::MergeDefinitionData(
-    CXXRecordDecl *D, struct CXXRecordDecl::DefinitionData &MergeDD) {
+    CXXRecordDecl *D, struct CXXRecordDecl::DefinitionData &&MergeDD) {
   assert(D->DefinitionData.getNotUpdated() &&
          "merging class definition into non-definition");
   auto &DD = *D->DefinitionData.getNotUpdated();
@@ -1364,14 +1397,45 @@ void ASTDeclReader::MergeDefinitionData(
   // FIXME: We only need to do this if the merged definition declares members
   // that this definition did not declare, or if it defines members that this
   // definition did not define.
-  if (MergeDD.DeclaredSpecialMembers && DD.Definition != MergeDD.Definition) {
+  if (DD.Definition != MergeDD.Definition) {
     Reader.MergedLookups[DD.Definition].push_back(MergeDD.Definition);
     DD.Definition->setHasExternalVisibleStorage();
+
+    if (DD.Definition->isHidden()) {
+      // If MergeDD is visible or becomes visible, make the definition visible.
+      if (!MergeDD.Definition->isHidden())
+        DD.Definition->Hidden = false;
+      else if (Reader.getContext().getLangOpts().ModulesLocalVisibility) {
+        Reader.getContext().mergeDefinitionIntoModule(
+            DD.Definition, MergeDD.Definition->getImportedOwningModule(),
+            /*NotifyListeners*/ false);
+        Reader.PendingMergedDefinitionsToDeduplicate.insert(DD.Definition);
+      } else {
+        auto SubmoduleID = MergeDD.Definition->getOwningModuleID();
+        assert(SubmoduleID && "hidden definition in no module");
+        Reader.HiddenNamesMap[Reader.getSubmodule(SubmoduleID)].push_back(
+            DD.Definition);
+      }
+    }
+  }
+
+  auto PFDI = Reader.PendingFakeDefinitionData.find(&DD);
+  if (PFDI != Reader.PendingFakeDefinitionData.end() &&
+      PFDI->second == ASTReader::PendingFakeDefinitionKind::Fake) {
+    // We faked up this definition data because we found a class for which we'd
+    // not yet loaded the definition. Replace it with the real thing now.
+    assert(!DD.IsLambda && !MergeDD.IsLambda && "faked up lambda definition?");
+    PFDI->second = ASTReader::PendingFakeDefinitionKind::FakeLoaded;
+
+    // Don't change which declaration is the definition; that is required
+    // to be invariant once we select it.
+    auto *Def = DD.Definition;
+    DD = std::move(MergeDD);
+    DD.Definition = Def;
+    return;
   }
 
   // FIXME: Move this out into a .def file?
-  // FIXME: Issue a diagnostic on a mismatched MATCH_FIELD, rather than
-  // asserting; this can happen in the case of an ODR violation.
   bool DetectedOdrViolation = false;
 #define OR_FIELD(Field) DD.Field |= MergeDD.Field;
 #define MATCH_FIELD(Field) \
@@ -1442,7 +1506,7 @@ void ASTDeclReader::MergeDefinitionData(
     Reader.PendingOdrMergeFailures[DD.Definition].push_back(MergeDD.Definition);
 }
 
-void ASTDeclReader::ReadCXXRecordDefinition(CXXRecordDecl *D) {
+void ASTDeclReader::ReadCXXRecordDefinition(CXXRecordDecl *D, bool Update) {
   struct CXXRecordDecl::DefinitionData *DD;
   ASTContext &C = Reader.getContext();
 
@@ -1457,19 +1521,29 @@ void ASTDeclReader::ReadCXXRecordDefinition(CXXRecordDecl *D) {
 
   ReadCXXDefinitionData(*DD, Record, Idx);
 
-  // If we're reading an update record, we might already have a definition for
-  // this record. If so, just merge into it.
-  if (D->DefinitionData.getNotUpdated()) {
-    MergeDefinitionData(D, *DD);
+  // We might already have a definition for this record. This can happen either
+  // because we're reading an update record, or because we've already done some
+  // merging. Either way, just merge into it.
+  CXXRecordDecl *Canon = D->getCanonicalDecl();
+  if (auto *CanonDD = Canon->DefinitionData.getNotUpdated()) {
+    if (CanonDD->Definition != DD->Definition)
+      Reader.MergedDeclContexts.insert(
+          std::make_pair(DD->Definition, CanonDD->Definition));
+    MergeDefinitionData(Canon, std::move(*DD));
+    D->DefinitionData = Canon->DefinitionData;
     return;
   }
 
   // Propagate the DefinitionData pointer to the canonical declaration, so
   // that all other deserialized declarations will see it.
-  CXXRecordDecl *Canon = D->getCanonicalDecl();
   if (Canon == D) {
     D->DefinitionData = DD;
     D->IsCompleteDefinition = true;
+
+    // If this is an update record, we can have redeclarations already. Make a
+    // note that we need to propagate the DefinitionData pointer onto them.
+    if (Update)
+      Reader.PendingDefinitions.insert(D);
   } else if (auto *CanonDD = Canon->DefinitionData.getNotUpdated()) {
     // We have already deserialized a definition of this record. This
     // definition is no longer really a definition. Note that the pre-existing
@@ -1478,7 +1552,7 @@ void ASTDeclReader::ReadCXXRecordDefinition(CXXRecordDecl *D) {
         std::make_pair(D, CanonDD->Definition));
     D->DefinitionData = Canon->DefinitionData;
     D->IsCompleteDefinition = false;
-    MergeDefinitionData(D, *DD);
+    MergeDefinitionData(D, std::move(*DD));
   } else {
     Canon->DefinitionData = DD;
     D->DefinitionData = Canon->DefinitionData;
@@ -1535,7 +1609,7 @@ ASTDeclReader::VisitCXXRecordDeclImpl(CXXRecordDecl *D) {
 
   bool WasDefinition = Record[Idx++];
   if (WasDefinition)
-    ReadCXXRecordDefinition(D);
+    ReadCXXRecordDefinition(D, /*Update*/false);
   else
     // Propagate DefinitionData pointer from the canonical declaration.
     D->DefinitionData = D->getCanonicalDecl()->DefinitionData;
@@ -1576,17 +1650,20 @@ void ASTDeclReader::VisitCXXConstructorDecl(CXXConstructorDecl *D) {
   VisitCXXMethodDecl(D);
 
   if (auto *CD = ReadDeclAs<CXXConstructorDecl>(Record, Idx))
-    D->setInheritedConstructor(CD);
+    if (D->isCanonicalDecl())
+      D->setInheritedConstructor(CD->getCanonicalDecl());
   D->IsExplicitSpecified = Record[Idx++];
-  // FIXME: We should defer loading this until we need the constructor's body.
-  std::tie(D->CtorInitializers, D->NumCtorInitializers) =
-      Reader.ReadCXXCtorInitializers(F, Record, Idx);
 }
 
 void ASTDeclReader::VisitCXXDestructorDecl(CXXDestructorDecl *D) {
   VisitCXXMethodDecl(D);
 
-  D->OperatorDelete = ReadDeclAs<FunctionDecl>(Record, Idx);
+  if (auto *OperatorDelete = ReadDeclAs<FunctionDecl>(Record, Idx)) {
+    auto *Canon = cast<CXXDestructorDecl>(D->getCanonicalDecl());
+    // FIXME: Check consistency if we have an old and new operator delete.
+    if (!Canon->OperatorDelete)
+      Canon->OperatorDelete = OperatorDelete;
+  }
 }
 
 void ASTDeclReader::VisitCXXConversionDecl(CXXConversionDecl *D) {
@@ -1690,36 +1767,34 @@ ASTDeclReader::VisitRedeclarableTemplateDecl(RedeclarableTemplateDecl *D) {
   return Redecl;
 }
 
+static DeclID *newDeclIDList(ASTContext &Context, DeclID *Old,
+                             SmallVectorImpl<DeclID> &IDs) {
+  assert(!IDs.empty() && "no IDs to add to list");
+  if (Old) {
+    IDs.insert(IDs.end(), Old + 1, Old + 1 + Old[0]);
+    std::sort(IDs.begin(), IDs.end());
+    IDs.erase(std::unique(IDs.begin(), IDs.end()), IDs.end());
+  }
+
+  auto *Result = new (Context) DeclID[1 + IDs.size()];
+  *Result = IDs.size();
+  std::copy(IDs.begin(), IDs.end(), Result + 1);
+  return Result;
+}
+
 void ASTDeclReader::VisitClassTemplateDecl(ClassTemplateDecl *D) {
   RedeclarableResult Redecl = VisitRedeclarableTemplateDecl(D);
 
   if (ThisDeclID == Redecl.getFirstID()) {
     // This ClassTemplateDecl owns a CommonPtr; read it to keep track of all of
     // the specializations.
-    SmallVector<serialization::DeclID, 2> SpecIDs;
-    SpecIDs.push_back(0);
-    
-    // Specializations.
-    unsigned Size = Record[Idx++];
-    SpecIDs[0] += Size;
-    for (unsigned I = 0; I != Size; ++I)
-      SpecIDs.push_back(ReadDeclID(Record, Idx));
-
-    // Partial specializations.
-    Size = Record[Idx++];
-    SpecIDs[0] += Size;
-    for (unsigned I = 0; I != Size; ++I)
-      SpecIDs.push_back(ReadDeclID(Record, Idx));
-
-    ClassTemplateDecl::Common *CommonPtr = D->getCommonPtr();
-    if (SpecIDs[0]) {
-      typedef serialization::DeclID DeclID;
-      
-      // FIXME: Append specializations!
-      CommonPtr->LazySpecializations
-        = new (Reader.getContext()) DeclID [SpecIDs.size()];
-      memcpy(CommonPtr->LazySpecializations, SpecIDs.data(), 
-             SpecIDs.size() * sizeof(DeclID));
+    SmallVector<serialization::DeclID, 32> SpecIDs;
+    ReadDeclIDList(SpecIDs);
+
+    if (!SpecIDs.empty()) {
+      auto *CommonPtr = D->getCommonPtr();
+      CommonPtr->LazySpecializations = newDeclIDList(
+          Reader.getContext(), CommonPtr->LazySpecializations, SpecIDs);
     }
   }
 
@@ -1741,30 +1816,13 @@ void ASTDeclReader::VisitVarTemplateDecl(VarTemplateDecl *D) {
   if (ThisDeclID == Redecl.getFirstID()) {
     // This VarTemplateDecl owns a CommonPtr; read it to keep track of all of
     // the specializations.
-    SmallVector<serialization::DeclID, 2> SpecIDs;
-    SpecIDs.push_back(0);
-
-    // Specializations.
-    unsigned Size = Record[Idx++];
-    SpecIDs[0] += Size;
-    for (unsigned I = 0; I != Size; ++I)
-      SpecIDs.push_back(ReadDeclID(Record, Idx));
-
-    // Partial specializations.
-    Size = Record[Idx++];
-    SpecIDs[0] += Size;
-    for (unsigned I = 0; I != Size; ++I)
-      SpecIDs.push_back(ReadDeclID(Record, Idx));
-
-    VarTemplateDecl::Common *CommonPtr = D->getCommonPtr();
-    if (SpecIDs[0]) {
-      typedef serialization::DeclID DeclID;
+    SmallVector<serialization::DeclID, 32> SpecIDs;
+    ReadDeclIDList(SpecIDs);
 
-      // FIXME: Append specializations!
-      CommonPtr->LazySpecializations =
-          new (Reader.getContext()) DeclID[SpecIDs.size()];
-      memcpy(CommonPtr->LazySpecializations, SpecIDs.data(),
-             SpecIDs.size() * sizeof(DeclID));
+    if (!SpecIDs.empty()) {
+      auto *CommonPtr = D->getCommonPtr();
+      CommonPtr->LazySpecializations = newDeclIDList(
+          Reader.getContext(), CommonPtr->LazySpecializations, SpecIDs);
     }
   }
 }
@@ -1823,7 +1881,7 @@ ASTDeclReader::VisitClassTemplateSpecializationDeclImpl(
         // definition.
         if (auto *DDD = D->DefinitionData.getNotUpdated()) {
           if (auto *CanonDD = CanonSpec->DefinitionData.getNotUpdated()) {
-            MergeDefinitionData(CanonSpec, *DDD);
+            MergeDefinitionData(CanonSpec, std::move(*DDD));
             Reader.PendingDefinitions.erase(D);
             Reader.MergedDeclContexts.insert(
                 std::make_pair(D, CanonDD->Definition));
@@ -1876,17 +1934,13 @@ void ASTDeclReader::VisitFunctionTemplateDecl(FunctionTemplateDecl *D) {
 
   if (ThisDeclID == Redecl.getFirstID()) {
     // This FunctionTemplateDecl owns a CommonPtr; read it.
+    SmallVector<serialization::DeclID, 32> SpecIDs;
+    ReadDeclIDList(SpecIDs);
 
-    // Read the function specialization declaration IDs. The specializations
-    // themselves will be loaded if they're needed.
-    if (unsigned NumSpecs = Record[Idx++]) {
-      // FIXME: Append specializations!
-      FunctionTemplateDecl::Common *CommonPtr = D->getCommonPtr();
-      CommonPtr->LazySpecializations = new (Reader.getContext())
-          serialization::DeclID[NumSpecs + 1];
-      CommonPtr->LazySpecializations[0] = NumSpecs;
-      for (unsigned I = 0; I != NumSpecs; ++I)
-        CommonPtr->LazySpecializations[I + 1] = ReadDeclID(Record, Idx);
+    if (!SpecIDs.empty()) {
+      auto *CommonPtr = D->getCommonPtr();
+      CommonPtr->LazySpecializations = newDeclIDList(
+          Reader.getContext(), CommonPtr->LazySpecializations, SpecIDs);
     }
   }
 }
@@ -2049,15 +2103,25 @@ ASTDeclReader::VisitDeclContext(DeclContext *DC) {
 }
 
 template <typename T>
-ASTDeclReader::RedeclarableResult 
+ASTDeclReader::RedeclarableResult
 ASTDeclReader::VisitRedeclarable(Redeclarable<T> *D) {
   DeclID FirstDeclID = ReadDeclID(Record, Idx);
-  
+  Decl *MergeWith = nullptr;
+
   // 0 indicates that this declaration was the only declaration of its entity,
   // and is used for space optimization.
   if (FirstDeclID == 0)
     FirstDeclID = ThisDeclID;
-  
+  else if (unsigned N = Record[Idx++]) {
+    // We have some declarations that must be before us in our redeclaration
+    // chain. Read them now, and remember that we ought to merge with one of
+    // them.
+    // FIXME: Provide a known merge target to the second and subsequent such
+    // declaration.
+    for (unsigned I = 0; I != N; ++I)
+      MergeWith = ReadDecl(Record, Idx/*, MergeWith*/);
+  }
+
   T *FirstDecl = cast_or_null<T>(Reader.GetDecl(FirstDeclID));
   if (FirstDecl != D) {
     // We delay loading of the redeclaration chain to avoid deeply nested calls.
@@ -2065,6 +2129,7 @@ ASTDeclReader::VisitRedeclarable(Redeclarable<T> *D) {
     // which is the one that matters and mark the real previous DeclID to be
     // loaded & attached later on.
     D->RedeclLink = Redeclarable<T>::PreviousDeclLink(FirstDecl);
+    D->First = FirstDecl->getCanonicalDecl();
   }    
   
   // Note that this declaration has been deserialized.
@@ -2072,7 +2137,7 @@ ASTDeclReader::VisitRedeclarable(Redeclarable<T> *D) {
                              
   // The result structure takes care to note that we need to load the 
   // other declaration chains for this ID.
-  return RedeclarableResult(Reader, FirstDeclID,
+  return RedeclarableResult(Reader, FirstDeclID, MergeWith,
                             static_cast<T *>(D)->getKind());
 }
 
@@ -2083,23 +2148,19 @@ void ASTDeclReader::mergeRedeclarable(Redeclarable<T> *DBase,
                                       RedeclarableResult &Redecl,
                                       DeclID TemplatePatternID) {
   T *D = static_cast<T*>(DBase);
-  T *DCanon = D->getCanonicalDecl();
-  if (D != DCanon &&
-      // IDs < NUM_PREDEF_DECL_IDS are not loaded from an AST file.
-      Redecl.getFirstID() >= NUM_PREDEF_DECL_IDS &&
-      (!Reader.getContext().getLangOpts().Modules ||
-       Reader.getOwningModuleFile(DCanon) == Reader.getOwningModuleFile(D))) {
-    // All redeclarations between this declaration and its originally-canonical
-    // declaration get pulled in when we load DCanon; we don't need to
-    // perform any more merging now.
-    Redecl.suppress();
-  }
 
   // If modules are not available, there is no reason to perform this merge.
   if (!Reader.getContext().getLangOpts().Modules)
     return;
 
-  if (FindExistingResult ExistingRes = findExisting(D))
+  // If we're not the canonical declaration, we don't need to merge.
+  if (!DBase->isFirstDecl())
+    return;
+
+  if (auto *Existing = Redecl.getKnownMergeTarget())
+    // We already know of an existing declaration we should merge with.
+    mergeRedeclarable(D, cast<T>(Existing), Redecl, TemplatePatternID);
+  else if (FindExistingResult ExistingRes = findExisting(D))
     if (T *Existing = ExistingRes)
       mergeRedeclarable(D, Existing, Redecl, TemplatePatternID);
 }
@@ -2120,7 +2181,8 @@ void ASTDeclReader::mergeTemplatePattern(RedeclarableTemplateDecl *D,
   auto *DPattern = D->getTemplatedDecl();
   auto *ExistingPattern = Existing->getTemplatedDecl();
   RedeclarableResult Result(Reader, DPattern->getCanonicalDecl()->getGlobalID(),
-                            DPattern->getKind());
+                            /*MergeWith*/ExistingPattern, DPattern->getKind());
+
   if (auto *DClass = dyn_cast<CXXRecordDecl>(DPattern)) {
     // Merge with any existing definition.
     // FIXME: This is duplicated in several places. Refactor.
@@ -2128,13 +2190,14 @@ void ASTDeclReader::mergeTemplatePattern(RedeclarableTemplateDecl *D,
         cast<CXXRecordDecl>(ExistingPattern)->getCanonicalDecl();
     if (auto *DDD = DClass->DefinitionData.getNotUpdated()) {
       if (auto *ExistingDD = ExistingClass->DefinitionData.getNotUpdated()) {
-        MergeDefinitionData(ExistingClass, *DDD);
+        MergeDefinitionData(ExistingClass, std::move(*DDD));
         Reader.PendingDefinitions.erase(DClass);
         Reader.MergedDeclContexts.insert(
             std::make_pair(DClass, ExistingDD->Definition));
         DClass->IsCompleteDefinition = false;
       } else {
         ExistingClass->DefinitionData = DClass->DefinitionData;
+        Reader.PendingDefinitions.insert(DClass);
       }
     }
     DClass->DefinitionData = ExistingClass->DefinitionData;
@@ -2163,14 +2226,18 @@ void ASTDeclReader::mergeRedeclarable(Redeclarable<T> *DBase, T *Existing,
   T *ExistingCanon = Existing->getCanonicalDecl();
   T *DCanon = D->getCanonicalDecl();
   if (ExistingCanon != DCanon) {
-    assert(DCanon->getGlobalID() == Redecl.getFirstID());
+    assert(DCanon->getGlobalID() == Redecl.getFirstID() &&
+           "already merged this declaration");
 
     // Have our redeclaration link point back at the canonical declaration
     // of the existing declaration, so that this declaration has the
     // appropriate canonical declaration.
     D->RedeclLink = Redeclarable<T>::PreviousDeclLink(ExistingCanon);
+    D->First = ExistingCanon;
 
     // When we merge a namespace, update its pointer to the first namespace.
+    // We cannot have loaded any redeclarations of this declaration yet, so
+    // there's nothing else that needs to be updated.
     if (auto *Namespace = dyn_cast<NamespaceDecl>(D))
       Namespace->AnonOrFirstNamespaceAndInline.setPointer(
           assert_cast<NamespaceDecl*>(ExistingCanon));
@@ -2181,14 +2248,11 @@ void ASTDeclReader::mergeRedeclarable(Redeclarable<T> *DBase, T *Existing,
           DTemplate, assert_cast<RedeclarableTemplateDecl*>(ExistingCanon),
           TemplatePatternID);
 
-    // If this declaration was the canonical declaration, make a note of
-    // that. We accept the linear algorithm here because the number of
-    // unique canonical declarations of an entity should always be tiny.
+    // If this declaration was the canonical declaration, make a note of that.
     if (DCanon == D) {
-      SmallVectorImpl<DeclID> &Merged = Reader.MergedDecls[ExistingCanon];
-      if (std::find(Merged.begin(), Merged.end(), Redecl.getFirstID())
-            == Merged.end())
-        Merged.push_back(Redecl.getFirstID());
+      Reader.MergedDecls[ExistingCanon].push_back(Redecl.getFirstID());
+      if (Reader.PendingDeclChainsKnown.insert(ExistingCanon).second)
+        Reader.PendingDeclChains.push_back(ExistingCanon);
     }
   }
 }
@@ -2529,42 +2593,71 @@ static bool isSameEntity(NamedDecl *X, NamedDecl *Y) {
 
 /// Find the context in which we should search for previous declarations when
 /// looking for declarations to merge.
-static DeclContext *getPrimaryContextForMerging(DeclContext *DC) {
+DeclContext *ASTDeclReader::getPrimaryContextForMerging(ASTReader &Reader,
+                                                        DeclContext *DC) {
   if (NamespaceDecl *ND = dyn_cast<NamespaceDecl>(DC))
     return ND->getOriginalNamespace();
 
-  // There is one tricky case here: if DC is a class with no definition, then
-  // we're merging a declaration whose definition is added by an update record,
-  // but we've not yet loaded that update record. In this case, we use the
-  // canonical declaration for merging until we get a real definition.
-  // FIXME: When we add a definition, we may need to move the partial lookup
-  // information from the canonical declaration onto the chosen definition.
-  if (CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(DC))
-    return RD->getPrimaryContext();
+  if (CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(DC)) {
+    // Try to dig out the definition.
+    auto *DD = RD->DefinitionData.getNotUpdated();
+    if (!DD)
+      DD = RD->getCanonicalDecl()->DefinitionData.getNotUpdated();
+
+    // If there's no definition yet, then DC's definition is added by an update
+    // record, but we've not yet loaded that update record. In this case, we
+    // commit to DC being the canonical definition now, and will fix this when
+    // we load the update record.
+    if (!DD) {
+      DD = new (Reader.Context) struct CXXRecordDecl::DefinitionData(RD);
+      RD->IsCompleteDefinition = true;
+      RD->DefinitionData = DD;
+      RD->getCanonicalDecl()->DefinitionData = DD;
+
+      // Track that we did this horrible thing so that we can fix it later.
+      Reader.PendingFakeDefinitionData.insert(
+          std::make_pair(DD, ASTReader::PendingFakeDefinitionKind::Fake));
+    }
+
+    return DD->Definition;
+  }
 
   if (EnumDecl *ED = dyn_cast<EnumDecl>(DC))
     return ED->getASTContext().getLangOpts().CPlusPlus? ED->getDefinition()
                                                       : nullptr;
 
+  // We can see the TU here only if we have no Sema object. In that case,
+  // there's no TU scope to look in, so using the DC alone is sufficient.
+  if (auto *TU = dyn_cast<TranslationUnitDecl>(DC))
+    return TU;
+
   return nullptr;
 }
 
 ASTDeclReader::FindExistingResult::~FindExistingResult() {
+  // Record that we had a typedef name for linkage whether or not we merge
+  // with that declaration.
+  if (TypedefNameForLinkage) {
+    DeclContext *DC = New->getDeclContext()->getRedeclContext();
+    Reader.ImportedTypedefNamesForLinkage.insert(
+        std::make_pair(std::make_pair(DC, TypedefNameForLinkage), New));
+    return;
+  }
+
   if (!AddResult || Existing)
     return;
 
   DeclarationName Name = New->getDeclName();
   DeclContext *DC = New->getDeclContext()->getRedeclContext();
-  if (TypedefNameForLinkage) {
-    Reader.ImportedTypedefNamesForLinkage.insert(
-        std::make_pair(std::make_pair(DC, TypedefNameForLinkage), New));
-  } else if (!Name) {
-    assert(needsAnonymousDeclarationNumber(New));
+  if (needsAnonymousDeclarationNumber(New)) {
     setAnonymousDeclForMerging(Reader, New->getLexicalDeclContext(),
                                AnonymousDeclNumber, New);
-  } else if (DC->isTranslationUnit() && Reader.SemaObj) {
-    Reader.SemaObj->IdResolver.tryAddTopLevelDecl(New, Name);
-  } else if (DeclContext *MergeDC = getPrimaryContextForMerging(DC)) {
+  } else if (DC->isTranslationUnit() && Reader.SemaObj &&
+             !Reader.getContext().getLangOpts().CPlusPlus) {
+    if (Reader.SemaObj->IdResolver.tryAddTopLevelDecl(New, Name))
+      Reader.PendingFakeLookupResults[Name.getAsIdentifierInfo()]
+            .push_back(New);
+  } else if (DeclContext *MergeDC = getPrimaryContextForMerging(Reader, DC)) {
     // Add the declaration to its redeclaration context so later merging
     // lookups will find it.
     MergeDC->makeDeclVisibleInContextImpl(New, /*Internal*/true);
@@ -2582,12 +2675,11 @@ static NamedDecl *getDeclForMerging(NamedDecl *Found,
   // If we found a typedef declaration that gives a name to some other
   // declaration, then we want that inner declaration. Declarations from
   // AST files are handled via ImportedTypedefNamesForLinkage.
-  if (Found->isFromASTFile()) return 0;
-  if (auto *TND = dyn_cast<TypedefNameDecl>(Found)) {
-    if (auto *TT = TND->getTypeSourceInfo()->getType()->getAs<TagType>())
-      if (TT->getDecl()->getTypedefNameForAnonDecl() == TND)
-        return TT->getDecl();
-  }
+  if (Found->isFromASTFile())
+    return 0;
+
+  if (auto *TND = dyn_cast<TypedefNameDecl>(Found))
+    return TND->getAnonDeclWithTypedefName();
 
   return 0;
 }
@@ -2608,17 +2700,12 @@ NamedDecl *ASTDeclReader::getAnonymousDeclForMerging(ASTReader &Reader,
   // If this is the first time, but we have parsed a declaration of the context,
   // build the anonymous declaration list from the parsed declaration.
   if (!cast<Decl>(DC)->isFromASTFile()) {
-    unsigned Index = 0;
-    for (Decl *LexicalD : DC->decls()) {
-      auto *ND = dyn_cast<NamedDecl>(LexicalD);
-      if (!ND || !needsAnonymousDeclarationNumber(ND))
-        continue;
-      if (Previous.size() == Index)
+    numberAnonymousDeclsWithin(DC, [&](NamedDecl *ND, unsigned Number) {
+      if (Previous.size() == Number)
         Previous.push_back(cast<NamedDecl>(ND->getCanonicalDecl()));
       else
-        Previous[Index] = cast<NamedDecl>(ND->getCanonicalDecl());
-      ++Index;
-    }
+        Previous[Number] = cast<NamedDecl>(ND->getCanonicalDecl());
+    });
   }
 
   return Index < Previous.size() ? Previous[Index] : nullptr;
@@ -2646,8 +2733,6 @@ ASTDeclReader::FindExistingResult ASTDeclReader::findExisting(NamedDecl *D) {
     // unmergeable contexts.
     FindExistingResult Result(Reader, D, /*Existing=*/nullptr,
                               AnonymousDeclNumber, TypedefNameForLinkage);
-    // FIXME: We may still need to pull in the redeclaration chain; there can
-    // be redeclarations via 'decltype'.
     Result.suppress();
     return Result;
   }
@@ -2667,16 +2752,16 @@ ASTDeclReader::FindExistingResult ASTDeclReader::findExisting(NamedDecl *D) {
     // was not imported.
   }
 
-  if (!Name) {
+  if (needsAnonymousDeclarationNumber(D)) {
     // This is an anonymous declaration that we may need to merge. Look it up
     // in its context by number.
-    assert(needsAnonymousDeclarationNumber(D));
     if (auto *Existing = getAnonymousDeclForMerging(
             Reader, D->getLexicalDeclContext(), AnonymousDeclNumber))
       if (isSameEntity(Existing, D))
         return FindExistingResult(Reader, D, Existing, AnonymousDeclNumber,
                                   TypedefNameForLinkage);
-  } else if (DC->isTranslationUnit() && Reader.SemaObj) {
+  } else if (DC->isTranslationUnit() && Reader.SemaObj &&
+             !Reader.getContext().getLangOpts().CPlusPlus) {
     IdentifierResolver &IdResolver = Reader.SemaObj->IdResolver;
 
     // Temporarily consider the identifier to be up-to-date. We don't want to
@@ -2710,7 +2795,7 @@ ASTDeclReader::FindExistingResult ASTDeclReader::findExisting(NamedDecl *D) {
           return FindExistingResult(Reader, D, Existing, AnonymousDeclNumber,
                                     TypedefNameForLinkage);
     }
-  } else if (DeclContext *MergeDC = getPrimaryContextForMerging(DC)) {
+  } else if (DeclContext *MergeDC = getPrimaryContextForMerging(Reader, DC)) {
     DeclContext::lookup_result R = MergeDC->noload_lookup(Name);
     for (DeclContext::lookup_iterator I = R.begin(), E = R.end(); I != E; ++I) {
       if (NamedDecl *Existing = getDeclForMerging(*I, TypedefNameForLinkage))
@@ -2738,20 +2823,47 @@ ASTDeclReader::FindExistingResult ASTDeclReader::findExisting(NamedDecl *D) {
 }
 
 template<typename DeclT>
+Decl *ASTDeclReader::getMostRecentDeclImpl(Redeclarable<DeclT> *D) {
+  return D->RedeclLink.getLatestNotUpdated();
+}
+Decl *ASTDeclReader::getMostRecentDeclImpl(...) {
+  llvm_unreachable("getMostRecentDecl on non-redeclarable declaration");
+}
+
+Decl *ASTDeclReader::getMostRecentDecl(Decl *D) {
+  assert(D);
+
+  switch (D->getKind()) {
+#define ABSTRACT_DECL(TYPE)
+#define DECL(TYPE, BASE)                               \
+  case Decl::TYPE:                                     \
+    return getMostRecentDeclImpl(cast<TYPE##Decl>(D));
+#include "clang/AST/DeclNodes.inc"
+  }
+  llvm_unreachable("unknown decl kind");
+}
+
+Decl *ASTReader::getMostRecentExistingDecl(Decl *D) {
+  return ASTDeclReader::getMostRecentDecl(D->getCanonicalDecl());
+}
+
+template<typename DeclT>
 void ASTDeclReader::attachPreviousDeclImpl(ASTReader &Reader,
                                            Redeclarable<DeclT> *D,
-                                           Decl *Previous) {
+                                           Decl *Previous, Decl *Canon) {
   D->RedeclLink.setPrevious(cast<DeclT>(Previous));
+  D->First = cast<DeclT>(Previous)->First;
 }
 namespace clang {
 template<>
 void ASTDeclReader::attachPreviousDeclImpl(ASTReader &Reader,
                                            Redeclarable<FunctionDecl> *D,
-                                           Decl *Previous) {
+                                           Decl *Previous, Decl *Canon) {
   FunctionDecl *FD = static_cast<FunctionDecl*>(D);
   FunctionDecl *PrevFD = cast<FunctionDecl>(Previous);
 
   FD->RedeclLink.setPrevious(PrevFD);
+  FD->First = PrevFD->First;
 
   // If the previous declaration is an inline function declaration, then this
   // declaration is too.
@@ -2774,16 +2886,17 @@ void ASTDeclReader::attachPreviousDeclImpl(ASTReader &Reader,
     FD->IsInline = true;
   }
 
-  // If this declaration has an unresolved exception specification but the
-  // previous declaration had a resolved one, resolve the exception
-  // specification now.
+  // If we need to propagate an exception specification along the redecl
+  // chain, make a note of that so that we can do so later.
   auto *FPT = FD->getType()->getAs<FunctionProtoType>();
   auto *PrevFPT = PrevFD->getType()->getAs<FunctionProtoType>();
-  if (FPT && PrevFPT &&
-      isUnresolvedExceptionSpec(FPT->getExceptionSpecType()) &&
-      !isUnresolvedExceptionSpec(PrevFPT->getExceptionSpecType())) {
-    Reader.Context.adjustExceptionSpec(
-        FD, PrevFPT->getExtProtoInfo().ExceptionSpec);
+  if (FPT && PrevFPT) {
+    bool IsUnresolved = isUnresolvedExceptionSpec(FPT->getExceptionSpecType());
+    bool WasUnresolved =
+        isUnresolvedExceptionSpec(PrevFPT->getExceptionSpecType());
+    if (IsUnresolved != WasUnresolved)
+      Reader.PendingExceptionSpecUpdates.insert(
+          std::make_pair(Canon, IsUnresolved ? PrevFD : FD));
   }
 }
 }
@@ -2792,14 +2905,14 @@ void ASTDeclReader::attachPreviousDeclImpl(ASTReader &Reader, ...) {
 }
 
 void ASTDeclReader::attachPreviousDecl(ASTReader &Reader, Decl *D,
-                                       Decl *Previous) {
+                                       Decl *Previous, Decl *Canon) {
   assert(D && Previous);
 
   switch (D->getKind()) {
 #define ABSTRACT_DECL(TYPE)
-#define DECL(TYPE, BASE)                                           \
-  case Decl::TYPE:                                                 \
-    attachPreviousDeclImpl(Reader, cast<TYPE##Decl>(D), Previous); \
+#define DECL(TYPE, BASE)                                                  \
+  case Decl::TYPE:                                                        \
+    attachPreviousDeclImpl(Reader, cast<TYPE##Decl>(D), Previous, Canon); \
     break;
 #include "clang/AST/DeclNodes.inc"
   }
@@ -2859,29 +2972,6 @@ void ASTReader::markIncompleteDeclChain(Decl *D) {
   }
 }
 
-ASTReader::MergedDeclsMap::iterator
-ASTReader::combineStoredMergedDecls(Decl *Canon, GlobalDeclID CanonID) {
-  // If we don't have any stored merged declarations, just look in the
-  // merged declarations set.
-  StoredMergedDeclsMap::iterator StoredPos = StoredMergedDecls.find(CanonID);
-  if (StoredPos == StoredMergedDecls.end())
-    return MergedDecls.find(Canon);
-
-  // Append the stored merged declarations to the merged declarations set.
-  MergedDeclsMap::iterator Pos = MergedDecls.find(Canon);
-  if (Pos == MergedDecls.end())
-    Pos = MergedDecls.insert(std::make_pair(Canon,
-                                            SmallVector<DeclID, 2>())).first;
-  Pos->second.append(StoredPos->second.begin(), StoredPos->second.end());
-  StoredMergedDecls.erase(StoredPos);
-
-  // Sort and uniquify the set of merged declarations.
-  llvm::array_pod_sort(Pos->second.begin(), Pos->second.end());
-  Pos->second.erase(std::unique(Pos->second.begin(), Pos->second.end()),
-                    Pos->second.end());
-  return Pos;
-}
-
 /// \brief Read the declaration at the given offset from the AST file.
 Decl *ASTReader::ReadDeclRecord(DeclID ID) {
   unsigned Index = ID - NUM_PREDEF_DECL_IDS;
@@ -3086,6 +3176,9 @@ Decl *ASTReader::ReadDeclRecord(DeclID ID) {
   case DECL_CXX_BASE_SPECIFIERS:
     Error("attempt to read a C++ base-specifier record as a declaration");
     return nullptr;
+  case DECL_CXX_CTOR_INITIALIZERS:
+    Error("attempt to read a C++ ctor initializer record as a declaration");
+    return nullptr;
   case DECL_IMPORT:
     // Note: last entry of the ImportDecl record is the number of stored source 
     // locations.
@@ -3199,47 +3292,55 @@ void ASTReader::loadDeclUpdateRecords(serialization::DeclID ID, Decl *D) {
 }
 
 namespace {
-  /// \brief Module visitor class that finds all of the redeclarations of a 
-  /// 
+  /// \brief Module visitor class that finds all of the redeclarations of a
+  /// redeclarable declaration.
   class RedeclChainVisitor {
     ASTReader &Reader;
     SmallVectorImpl<DeclID> &SearchDecls;
     llvm::SmallPtrSetImpl<Decl *> &Deserialized;
     GlobalDeclID CanonID;
     SmallVector<Decl *, 4> Chain;
-    
+
   public:
     RedeclChainVisitor(ASTReader &Reader, SmallVectorImpl<DeclID> &SearchDecls,
                        llvm::SmallPtrSetImpl<Decl *> &Deserialized,
                        GlobalDeclID CanonID)
       : Reader(Reader), SearchDecls(SearchDecls), Deserialized(Deserialized),
-        CanonID(CanonID) { 
-      for (unsigned I = 0, N = SearchDecls.size(); I != N; ++I)
-        addToChain(Reader.GetDecl(SearchDecls[I]));
+        CanonID(CanonID) {
+      // Ensure that the canonical ID goes at the start of the chain.
+      addToChain(Reader.GetDecl(CanonID));
     }
-    
-    static bool visit(ModuleFile &M, bool Preorder, void *UserData) {
-      if (Preorder)
-        return false;
-      
-      return static_cast<RedeclChainVisitor *>(UserData)->visit(M);
+
+    static ModuleManager::DFSPreorderControl
+    visitPreorder(ModuleFile &M, void *UserData) {
+      return static_cast<RedeclChainVisitor *>(UserData)->visitPreorder(M);
     }
-    
+
+    static bool visitPostorder(ModuleFile &M, void *UserData) {
+      return static_cast<RedeclChainVisitor *>(UserData)->visitPostorder(M);
+    }
+
     void addToChain(Decl *D) {
       if (!D)
         return;
-      
+
       if (Deserialized.erase(D))
         Chain.push_back(D);
     }
-    
+
     void searchForID(ModuleFile &M, GlobalDeclID GlobalID) {
       // Map global ID of the first declaration down to the local ID
       // used in this module file.
       DeclID ID = Reader.mapGlobalIDToModuleFileGlobalID(M, GlobalID);
       if (!ID)
         return;
-      
+
+      // If the search decl was from this module, add it to the chain before any
+      // of its redeclarations in this module or users of it, and after any from
+      // imported modules.
+      if (CanonID != GlobalID && Reader.isDeclIDFromModule(GlobalID, M))
+        addToChain(Reader.GetDecl(GlobalID));
+
       // Perform a binary search to find the local redeclarations for this
       // declaration (if any).
       const LocalRedeclarationsInfo Compare = { ID, 0 };
@@ -3269,8 +3370,36 @@ namespace {
       for (unsigned I = 0; I != N; ++I)
         addToChain(Reader.GetLocalDecl(M, M.RedeclarationChains[Offset++]));
     }
-    
-    bool visit(ModuleFile &M) {
+
+    bool needsToVisitImports(ModuleFile &M, GlobalDeclID GlobalID) {
+      DeclID ID = Reader.mapGlobalIDToModuleFileGlobalID(M, GlobalID);
+      if (!ID)
+        return false;
+
+      const LocalRedeclarationsInfo Compare = {ID, 0};
+      const LocalRedeclarationsInfo *Result = std::lower_bound(
+          M.RedeclarationsMap,
+          M.RedeclarationsMap + M.LocalNumRedeclarationsInMap, Compare);
+      if (Result == M.RedeclarationsMap + M.LocalNumRedeclarationsInMap ||
+          Result->FirstID != ID) {
+        return true;
+      }
+      unsigned Offset = Result->Offset;
+      unsigned N = M.RedeclarationChains[Offset];
+      // We don't need to visit a module or any of its imports if we've already
+      // deserialized the redecls from this module.
+      return N != 0;
+    }
+
+    ModuleManager::DFSPreorderControl visitPreorder(ModuleFile &M) {
+      for (unsigned I = 0, N = SearchDecls.size(); I != N; ++I) {
+        if (needsToVisitImports(M, SearchDecls[I]))
+          return ModuleManager::Continue;
+      }
+      return ModuleManager::SkipImports;
+    }
+
+    bool visitPostorder(ModuleFile &M) {
       // Visit each of the declarations.
       for (unsigned I = 0, N = SearchDecls.size(); I != N; ++I)
         searchForID(M, SearchDecls[I]);
@@ -3285,41 +3414,46 @@ namespace {
   };
 }
 
-void ASTReader::loadPendingDeclChain(serialization::GlobalDeclID ID) {
-  Decl *D = GetDecl(ID);  
-  Decl *CanonDecl = D->getCanonicalDecl();
-  
+void ASTReader::loadPendingDeclChain(Decl *CanonDecl) {
+  // The decl might have been merged into something else after being added to
+  // our list. If it was, just skip it.
+  if (!CanonDecl->isCanonicalDecl())
+    return;
+
   // Determine the set of declaration IDs we'll be searching for.
-  SmallVector<DeclID, 1> SearchDecls;
-  GlobalDeclID CanonID = 0;
-  if (D == CanonDecl) {
-    SearchDecls.push_back(ID); // Always first.
-    CanonID = ID;
-  }
-  MergedDeclsMap::iterator MergedPos = combineStoredMergedDecls(CanonDecl, ID);
+  SmallVector<DeclID, 16> SearchDecls;
+  GlobalDeclID CanonID = CanonDecl->getGlobalID();
+  if (CanonID)
+    SearchDecls.push_back(CanonDecl->getGlobalID()); // Always first.
+  MergedDeclsMap::iterator MergedPos = MergedDecls.find(CanonDecl);
   if (MergedPos != MergedDecls.end())
     SearchDecls.append(MergedPos->second.begin(), MergedPos->second.end());
-  
+
   // Build up the list of redeclarations.
   RedeclChainVisitor Visitor(*this, SearchDecls, RedeclsDeserialized, CanonID);
-  ModuleMgr.visitDepthFirst(&RedeclChainVisitor::visit, &Visitor);
-  
+  ModuleMgr.visitDepthFirst(&RedeclChainVisitor::visitPreorder,
+                            &RedeclChainVisitor::visitPostorder, &Visitor);
+
   // Retrieve the chains.
   ArrayRef<Decl *> Chain = Visitor.getChain();
-  if (Chain.empty())
+  if (Chain.empty() || (Chain.size() == 1 && Chain[0] == CanonDecl))
     return;
-    
+
   // Hook up the chains.
-  Decl *MostRecent = CanonDecl->getMostRecentDecl();
+  //
+  // FIXME: We have three different dispatches on decl kind here; maybe
+  // we should instead generate one loop per kind and dispatch up-front?
+  Decl *MostRecent = ASTDeclReader::getMostRecentDecl(CanonDecl);
+  if (!MostRecent)
+    MostRecent = CanonDecl;
   for (unsigned I = 0, N = Chain.size(); I != N; ++I) {
     if (Chain[I] == CanonDecl)
       continue;
 
-    ASTDeclReader::attachPreviousDecl(*this, Chain[I], MostRecent);
+    ASTDeclReader::attachPreviousDecl(*this, Chain[I], MostRecent, CanonDecl);
     MostRecent = Chain[I];
   }
-  
-  ASTDeclReader::attachLatestDecl(CanonDecl, MostRecent);  
+  ASTDeclReader::attachLatestDecl(CanonDecl, MostRecent);
 }
 
 namespace {
@@ -3513,13 +3647,25 @@ void ASTDeclReader::UpdateDecl(Decl *D, ModuleFile &ModuleFile,
   while (Idx < Record.size()) {
     switch ((DeclUpdateKind)Record[Idx++]) {
     case UPD_CXX_ADDED_IMPLICIT_MEMBER: {
+      auto *RD = cast<CXXRecordDecl>(D);
       // FIXME: If we also have an update record for instantiating the
       // definition of D, we need that to happen before we get here.
       Decl *MD = Reader.ReadDecl(ModuleFile, Record, Idx);
       assert(MD && "couldn't read decl from update record");
       // FIXME: We should call addHiddenDecl instead, to add the member
       // to its DeclContext.
-      cast<CXXRecordDecl>(D)->addedMember(MD);
+      RD->addedMember(MD);
+
+      // If we've added a new special member to a class definition that is not
+      // the canonical definition, then we need special member lookups in the
+      // canonical definition to also look into our class.
+      auto *DD = RD->DefinitionData.getNotUpdated();
+      if (DD && DD->Definition != RD) {
+        auto &Merged = Reader.MergedLookups[DD->Definition];
+        // FIXME: Avoid the linear-time scan here.
+        if (std::find(Merged.begin(), Merged.end(), RD) == Merged.end())
+          Merged.push_back(RD);
+      }
       break;
     }
 
@@ -3567,13 +3713,12 @@ void ASTDeclReader::UpdateDecl(Decl *D, ModuleFile &ModuleFile,
         });
       }
       FD->setInnerLocStart(Reader.ReadSourceLocation(ModuleFile, Record, Idx));
-      if (auto *CD = dyn_cast<CXXConstructorDecl>(FD))
-        std::tie(CD->CtorInitializers, CD->NumCtorInitializers) =
-            Reader.ReadCXXCtorInitializers(ModuleFile, Record, Idx);
-      if (auto *DD = dyn_cast<CXXDestructorDecl>(FD))
-        // FIXME: Check consistency.
-        DD->setOperatorDelete(Reader.ReadDeclAs<FunctionDecl>(ModuleFile,
-                                                              Record, Idx));
+      if (auto *CD = dyn_cast<CXXConstructorDecl>(FD)) {
+        CD->NumCtorInitializers = Record[Idx++];
+        if (CD->NumCtorInitializers)
+          CD->CtorInitializers =
+              Reader.ReadCXXCtorInitializersRef(F, Record, Idx);
+      }
       // Store the offset of the body so we can lazily load it later.
       Reader.PendingBodies[FD] = GetCurrentCursorOffset();
       HasPendingBody = true;
@@ -3583,16 +3728,20 @@ void ASTDeclReader::UpdateDecl(Decl *D, ModuleFile &ModuleFile,
 
     case UPD_CXX_INSTANTIATED_CLASS_DEFINITION: {
       auto *RD = cast<CXXRecordDecl>(D);
-      bool HadDefinition = RD->getDefinition();
-      ReadCXXRecordDefinition(RD);
+      auto *OldDD = RD->DefinitionData.getNotUpdated();
+      bool HadRealDefinition =
+          OldDD && (OldDD->Definition != RD ||
+                    !Reader.PendingFakeDefinitionData.count(OldDD));
+      ReadCXXRecordDefinition(RD, /*Update*/true);
+
       // Visible update is handled separately.
       uint64_t LexicalOffset = Record[Idx++];
-      if (!HadDefinition && LexicalOffset) {
+      if (!HadRealDefinition && LexicalOffset) {
         RD->setHasExternalLexicalStorage(true);
         Reader.ReadDeclContextStorage(ModuleFile, ModuleFile.DeclsCursor,
-                                          std::make_pair(LexicalOffset, 0),
-                                          ModuleFile.DeclContextInfos[RD]);
-        Reader.PendingDefinitions.insert(RD);
+                                      std::make_pair(LexicalOffset, 0),
+                                      ModuleFile.DeclContextInfos[RD]);
+        Reader.PendingFakeDefinitionData.erase(OldDD);
       }
 
       auto TSK = (TemplateSpecializationKind)Record[Idx++];
@@ -3636,24 +3785,36 @@ void ASTDeclReader::UpdateDecl(Decl *D, ModuleFile &ModuleFile,
       break;
     }
 
+    case UPD_CXX_RESOLVED_DTOR_DELETE: {
+      // Set the 'operator delete' directly to avoid emitting another update
+      // record.
+      auto *Del = Reader.ReadDeclAs<FunctionDecl>(ModuleFile, Record, Idx);
+      auto *First = cast<CXXDestructorDecl>(D->getCanonicalDecl());
+      // FIXME: Check consistency if we have an old and new operator delete.
+      if (!First->OperatorDelete)
+        First->OperatorDelete = Del;
+      break;
+    }
+
     case UPD_CXX_RESOLVED_EXCEPTION_SPEC: {
-      // FIXME: This doesn't send the right notifications if there are
-      // ASTMutationListeners other than an ASTWriter.
       FunctionProtoType::ExceptionSpecInfo ESI;
       SmallVector<QualType, 8> ExceptionStorage;
       Reader.readExceptionSpec(ModuleFile, ExceptionStorage, ESI, Record, Idx);
-      for (auto *Redecl : merged_redecls(D)) {
-        auto *FD = cast<FunctionDecl>(Redecl);
-        auto *FPT = FD->getType()->castAs<FunctionProtoType>();
-        if (!isUnresolvedExceptionSpec(FPT->getExceptionSpecType())) {
-          // AST invariant: if any exception spec in the redecl chain is
-          // resolved, all are resolved. We don't need to go any further.
-          // FIXME: If the exception spec is resolved, check that it matches.
-          break;
-        }
+
+      // Update this declaration's exception specification, if needed.
+      auto *FD = cast<FunctionDecl>(D);
+      auto *FPT = FD->getType()->castAs<FunctionProtoType>();
+      // FIXME: If the exception specification is already present, check that it
+      // matches.
+      if (isUnresolvedExceptionSpec(FPT->getExceptionSpecType())) {
         FD->setType(Reader.Context.getFunctionType(
             FPT->getReturnType(), FPT->getParamTypes(),
             FPT->getExtProtoInfo().withExceptionSpec(ESI)));
+
+        // When we get to the end of deserializing, see if there are other decls
+        // that we need to propagate this exception specification onto.
+        Reader.PendingExceptionSpecUpdates.insert(
+            std::make_pair(FD->getCanonicalDecl(), FD));
       }
       break;
     }
@@ -3685,10 +3846,30 @@ void ASTDeclReader::UpdateDecl(Decl *D, ModuleFile &ModuleFile,
     case UPD_STATIC_LOCAL_NUMBER:
       Reader.Context.setStaticLocalNumber(cast<VarDecl>(D), Record[Idx++]);
       break;
+
     case UPD_DECL_MARKED_OPENMP_THREADPRIVATE:
       D->addAttr(OMPThreadPrivateDeclAttr::CreateImplicit(
           Reader.Context, ReadSourceRange(Record, Idx)));
       break;
+
+    case UPD_DECL_EXPORTED:
+      unsigned SubmoduleID = readSubmoduleID(Record, Idx);
+      Module *Owner = SubmoduleID ? Reader.getSubmodule(SubmoduleID) : nullptr;
+      if (Reader.getContext().getLangOpts().ModulesLocalVisibility) {
+        // FIXME: This doesn't send the right notifications if there are
+        // ASTMutationListeners other than an ASTWriter.
+        Reader.getContext().mergeDefinitionIntoModule(cast<NamedDecl>(D), Owner,
+                                                      /*NotifyListeners*/false);
+        Reader.PendingMergedDefinitionsToDeduplicate.insert(cast<NamedDecl>(D));
+      } else if (Owner && Owner->NameVisibility != Module::AllVisible) {
+        // If Owner is made visible at some later point, make this declaration
+        // visible too.
+        Reader.HiddenNamesMap[Owner].push_back(D);
+      } else {
+        // The declaration is now visible.
+        D->Hidden = false;
+      }
+      break;
     }
   }
 }
diff --git a/lib/Serialization/ASTReaderStmt.cpp b/lib/Serialization/ASTReaderStmt.cpp
index 4ef2e73062a5..d1ecd4672488 100644
--- a/lib/Serialization/ASTReaderStmt.cpp
+++ b/lib/Serialization/ASTReaderStmt.cpp
@@ -486,7 +486,7 @@ void ASTStmtReader::VisitStringLiteral(StringLiteral *E) {
 
   // Read string data
   SmallString<16> Str(&Record[Idx], &Record[Idx] + Len);
-  E->setString(Reader.getContext(), Str.str(), kind, isPascal);
+  E->setString(Reader.getContext(), Str, kind, isPascal);
   Idx += Len;
 
   // Read source locations
@@ -1826,6 +1826,7 @@ void OMPClauseReader::VisitOMPScheduleClause(OMPScheduleClause *C) {
   C->setScheduleKind(
        static_cast<OpenMPScheduleClauseKind>(Record[Idx++]));
   C->setChunkSize(Reader->Reader.ReadSubExpr());
+  C->setHelperChunkSize(Reader->Reader.ReadSubExpr());
   C->setLParenLoc(Reader->ReadSourceLocation(Record, Idx));
   C->setScheduleKindLoc(Reader->ReadSourceLocation(Record, Idx));
   C->setCommaLoc(Reader->ReadSourceLocation(Record, Idx));
@@ -1889,6 +1890,22 @@ void OMPClauseReader::VisitOMPLastprivateClause(OMPLastprivateClause *C) {
   for (unsigned i = 0; i != NumVars; ++i)
     Vars.push_back(Reader->Reader.ReadSubExpr());
   C->setVarRefs(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setPrivateCopies(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setSourceExprs(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setDestinationExprs(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setAssignmentOps(Vars);
 }
 
 void OMPClauseReader::VisitOMPSharedClause(OMPSharedClause *C) {
@@ -1917,6 +1934,18 @@ void OMPClauseReader::VisitOMPReductionClause(OMPReductionClause *C) {
   for (unsigned i = 0; i != NumVars; ++i)
     Vars.push_back(Reader->Reader.ReadSubExpr());
   C->setVarRefs(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setLHSExprs(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setRHSExprs(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setReductionOps(Vars);
 }
 
 void OMPClauseReader::VisitOMPLinearClause(OMPLinearClause *C) {
@@ -1928,7 +1957,20 @@ void OMPClauseReader::VisitOMPLinearClause(OMPLinearClause *C) {
   for (unsigned i = 0; i != NumVars; ++i)
     Vars.push_back(Reader->Reader.ReadSubExpr());
   C->setVarRefs(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setInits(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setUpdates(Vars);
+  Vars.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Vars.push_back(Reader->Reader.ReadSubExpr());
+  C->setFinals(Vars);
   C->setStep(Reader->Reader.ReadSubExpr());
+  C->setCalcStep(Reader->Reader.ReadSubExpr());
 }
 
 void OMPClauseReader::VisitOMPAlignedClause(OMPAlignedClause *C) {
@@ -1946,21 +1988,45 @@ void OMPClauseReader::VisitOMPAlignedClause(OMPAlignedClause *C) {
 void OMPClauseReader::VisitOMPCopyinClause(OMPCopyinClause *C) {
   C->setLParenLoc(Reader->ReadSourceLocation(Record, Idx));
   unsigned NumVars = C->varlist_size();
-  SmallVector<Expr *, 16> Vars;
-  Vars.reserve(NumVars);
+  SmallVector<Expr *, 16> Exprs;
+  Exprs.reserve(NumVars);
   for (unsigned i = 0; i != NumVars; ++i)
-    Vars.push_back(Reader->Reader.ReadSubExpr());
-  C->setVarRefs(Vars);
+    Exprs.push_back(Reader->Reader.ReadSubExpr());
+  C->setVarRefs(Exprs);
+  Exprs.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Exprs.push_back(Reader->Reader.ReadSubExpr());
+  C->setSourceExprs(Exprs);
+  Exprs.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Exprs.push_back(Reader->Reader.ReadSubExpr());
+  C->setDestinationExprs(Exprs);
+  Exprs.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Exprs.push_back(Reader->Reader.ReadSubExpr());
+  C->setAssignmentOps(Exprs);
 }
 
 void OMPClauseReader::VisitOMPCopyprivateClause(OMPCopyprivateClause *C) {
   C->setLParenLoc(Reader->ReadSourceLocation(Record, Idx));
   unsigned NumVars = C->varlist_size();
-  SmallVector<Expr *, 16> Vars;
-  Vars.reserve(NumVars);
+  SmallVector<Expr *, 16> Exprs;
+  Exprs.reserve(NumVars);
   for (unsigned i = 0; i != NumVars; ++i)
-    Vars.push_back(Reader->Reader.ReadSubExpr());
-  C->setVarRefs(Vars);
+    Exprs.push_back(Reader->Reader.ReadSubExpr());
+  C->setVarRefs(Exprs);
+  Exprs.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Exprs.push_back(Reader->Reader.ReadSubExpr());
+  C->setSourceExprs(Exprs);
+  Exprs.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Exprs.push_back(Reader->Reader.ReadSubExpr());
+  C->setDestinationExprs(Exprs);
+  Exprs.clear();
+  for (unsigned i = 0; i != NumVars; ++i)
+    Exprs.push_back(Reader->Reader.ReadSubExpr());
+  C->setAssignmentOps(Exprs);
 }
 
 void OMPClauseReader::VisitOMPFlushClause(OMPFlushClause *C) {
@@ -2135,6 +2201,9 @@ void ASTStmtReader::VisitOMPAtomicDirective(OMPAtomicDirective *D) {
   D->setX(Reader.ReadSubExpr());
   D->setV(Reader.ReadSubExpr());
   D->setExpr(Reader.ReadSubExpr());
+  D->setUpdateExpr(Reader.ReadSubExpr());
+  D->IsXLHSInRHSPart = Record[Idx++] != 0;
+  D->IsPostfixUpdate = Record[Idx++] != 0;
 }
 
 void ASTStmtReader::VisitOMPTargetDirective(OMPTargetDirective *D) {
@@ -2423,11 +2492,12 @@ Stmt *ASTReader::ReadStmtFromStream(ModuleFile &F) {
       SourceLocation MemberLoc = ReadSourceLocation(F, Record, Idx);
       DeclarationNameInfo MemberNameInfo(MemberD->getDeclName(), MemberLoc);
       bool IsArrow = Record[Idx++];
+      SourceLocation OperatorLoc = ReadSourceLocation(F, Record, Idx);
 
-      S = MemberExpr::Create(Context, Base, IsArrow, QualifierLoc,
+      S = MemberExpr::Create(Context, Base, IsArrow, OperatorLoc, QualifierLoc,
                              TemplateKWLoc, MemberD, FoundDecl, MemberNameInfo,
-                             HasTemplateKWAndArgsInfo ? &ArgInfo : nullptr,
-                             T, VK, OK);
+                             HasTemplateKWAndArgsInfo ? &ArgInfo : nullptr, T,
+                             VK, OK);
       ReadDeclarationNameLoc(F, cast<MemberExpr>(S)->MemberDNLoc,
                              MemberD->getDeclName(), Record, Idx);
       if (HadMultipleCandidates)
diff --git a/lib/Serialization/ASTWriter.cpp b/lib/Serialization/ASTWriter.cpp
index 6c60d45503de..bf74c8471c3d 100644
--- a/lib/Serialization/ASTWriter.cpp
+++ b/lib/Serialization/ASTWriter.cpp
@@ -60,14 +60,14 @@ using namespace clang;
 using namespace clang::serialization;
 
 template <typename T, typename Allocator>
-static StringRef data(const std::vector<T, Allocator> &v) {
+static StringRef bytes(const std::vector<T, Allocator> &v) {
   if (v.empty()) return StringRef();
   return StringRef(reinterpret_cast<const char*>(&v[0]),
                          sizeof(T) * v.size());
 }
 
 template <typename T>
-static StringRef data(const SmallVectorImpl<T> &v) {
+static StringRef bytes(const SmallVectorImpl<T> &v) {
   return StringRef(reinterpret_cast<const char*>(v.data()),
                          sizeof(T) * v.size());
 }
@@ -867,6 +867,7 @@ void ASTWriter::WriteBlockInfoBlock() {
   RECORD(MODULE_NAME);
   RECORD(MODULE_MAP_FILE);
   RECORD(IMPORTS);
+  RECORD(KNOWN_MODULE_FILES);
   RECORD(LANGUAGE_OPTIONS);
   RECORD(TARGET_OPTIONS);
   RECORD(ORIGINAL_FILE);
@@ -892,7 +893,6 @@ void ASTWriter::WriteBlockInfoBlock() {
   RECORD(STATISTICS);
   RECORD(TENTATIVE_DEFINITIONS);
   RECORD(UNUSED_FILESCOPED_DECLS);
-  RECORD(LOCALLY_SCOPED_EXTERN_C_DECLS);
   RECORD(SELECTOR_OFFSETS);
   RECORD(METHOD_POOL);
   RECORD(PP_COUNTER_VALUE);
@@ -924,11 +924,9 @@ void ASTWriter::WriteBlockInfoBlock() {
   RECORD(OBJC_CATEGORIES_MAP);
   RECORD(FILE_SORTED_DECLS);
   RECORD(IMPORTED_MODULES);
-  RECORD(MERGED_DECLARATIONS);
   RECORD(LOCAL_REDECLARATIONS);
   RECORD(OBJC_CATEGORIES);
   RECORD(MACRO_OFFSET);
-  RECORD(MACRO_TABLE);
   RECORD(LATE_PARSED_TEMPLATE);
   RECORD(OPTIMIZE_PRAGMA_OPTIONS);
 
@@ -941,10 +939,12 @@ void ASTWriter::WriteBlockInfoBlock() {
 
   // Preprocessor Block.
   BLOCK(PREPROCESSOR_BLOCK);
-  RECORD(PP_MACRO_OBJECT_LIKE);
+  RECORD(PP_MACRO_DIRECTIVE_HISTORY);
   RECORD(PP_MACRO_FUNCTION_LIKE);
+  RECORD(PP_MACRO_OBJECT_LIKE);
+  RECORD(PP_MODULE_MACRO);
   RECORD(PP_TOKEN);
-  
+
   // Decls and Types block.
   BLOCK(DECLTYPES_BLOCK);
   RECORD(TYPE_EXT_QUAL);
@@ -1062,7 +1062,8 @@ void ASTWriter::WriteBlockInfoBlock() {
 /// to an absolute path and removing nested './'s.
 ///
 /// \return \c true if the path was changed.
-bool cleanPathForOutput(FileManager &FileMgr, SmallVectorImpl<char> &Path) {
+static bool cleanPathForOutput(FileManager &FileMgr,
+                               SmallVectorImpl<char> &Path) {
   bool Changed = false;
 
   if (!llvm::sys::path::is_absolute(StringRef(Path.data(), Path.size()))) {
@@ -1159,12 +1160,17 @@ void ASTWriter::WriteControlBlock(Preprocessor &PP, ASTContext &Context,
   Stream.EmitRecordWithBlob(MetadataAbbrevCode, Record,
                             getClangFullRepositoryVersion());
 
-  // Signature
-  Record.clear();
-  Record.push_back(getSignature());
-  Stream.EmitRecord(SIGNATURE, Record);
-
   if (WritingModule) {
+    // For implicit modules we output a signature that we can use to ensure
+    // duplicate module builds don't collide in the cache as their output order
+    // is non-deterministic.
+    // FIXME: Remove this when output is deterministic.
+    if (Context.getLangOpts().ImplicitModules) {
+      Record.clear();
+      Record.push_back(getSignature());
+      Stream.EmitRecord(SIGNATURE, Record);
+    }
+
     // Module name
     BitCodeAbbrev *Abbrev = new BitCodeAbbrev();
     Abbrev->Add(BitCodeAbbrevOp(MODULE_NAME));
@@ -1222,20 +1228,28 @@ void ASTWriter::WriteControlBlock(Preprocessor &PP, ASTContext &Context,
     serialization::ModuleManager &Mgr = Chain->getModuleManager();
     Record.clear();
 
-    for (ModuleManager::ModuleIterator M = Mgr.begin(), MEnd = Mgr.end();
-         M != MEnd; ++M) {
+    for (auto *M : Mgr) {
       // Skip modules that weren't directly imported.
-      if (!(*M)->isDirectlyImported())
+      if (!M->isDirectlyImported())
         continue;
 
-      Record.push_back((unsigned)(*M)->Kind); // FIXME: Stable encoding
-      AddSourceLocation((*M)->ImportLoc, Record);
-      Record.push_back((*M)->File->getSize());
-      Record.push_back((*M)->File->getModificationTime());
-      Record.push_back((*M)->Signature);
-      AddPath((*M)->FileName, Record);
+      Record.push_back((unsigned)M->Kind); // FIXME: Stable encoding
+      AddSourceLocation(M->ImportLoc, Record);
+      Record.push_back(M->File->getSize());
+      Record.push_back(M->File->getModificationTime());
+      Record.push_back(M->Signature);
+      AddPath(M->FileName, Record);
     }
     Stream.EmitRecord(IMPORTS, Record);
+
+    // Also emit a list of known module files that were not imported,
+    // but are made available by this module.
+    // FIXME: Should we also include a signature here?
+    Record.clear();
+    for (auto *E : Mgr.getAdditionalKnownModuleFiles())
+      AddPath(E->getName(), Record);
+    if (!Record.empty())
+      Stream.EmitRecord(KNOWN_MODULE_FILES, Record);
   }
 
   // Language options.
@@ -1341,6 +1355,8 @@ void ASTWriter::WriteControlBlock(Preprocessor &PP, ASTContext &Context,
   Record.push_back(HSOpts.UseStandardSystemIncludes);
   Record.push_back(HSOpts.UseStandardCXXIncludes);
   Record.push_back(HSOpts.UseLibcxx);
+  // Write out the specific module cache path that contains the module files.
+  AddString(PP.getHeaderSearchInfo().getModuleCachePath(), Record);
   Stream.EmitRecord(HEADER_SEARCH_OPTIONS, Record);
 
   // Preprocessor options.
@@ -1467,7 +1483,7 @@ void ASTWriter::WriteInputFiles(SourceManager &SourceMgr,
 
   unsigned UserFilesNum = 0;
   // Write out all of the input files.
-  std::vector<uint32_t> InputFileOffsets;
+  std::vector<uint64_t> InputFileOffsets;
   for (std::deque<InputFileEntry>::iterator
          I = SortedFiles.begin(), E = SortedFiles.end(); I != E; ++I) {
     const InputFileEntry &Entry = *I;
@@ -1514,7 +1530,7 @@ void ASTWriter::WriteInputFiles(SourceManager &SourceMgr,
   Record.push_back(INPUT_FILE_OFFSETS);
   Record.push_back(InputFileOffsets.size());
   Record.push_back(UserFilesNum);
-  Stream.EmitRecordWithBlob(OffsetsAbbrevCode, Record, data(InputFileOffsets));
+  Stream.EmitRecordWithBlob(OffsetsAbbrevCode, Record, bytes(InputFileOffsets));
 }
 
 //===----------------------------------------------------------------------===//
@@ -1756,7 +1772,7 @@ void ASTWriter::WriteHeaderSearch(const HeaderSearch &HS) {
   Record.push_back(NumHeaderSearchEntries);
   Record.push_back(TableData.size());
   TableData.append(GeneratorTrait.strings_begin(),GeneratorTrait.strings_end());
-  Stream.EmitRecordWithBlob(TableAbbrev, Record, TableData.str());
+  Stream.EmitRecordWithBlob(TableAbbrev, Record, TableData);
   
   // Free all of the strings we had to duplicate.
   for (unsigned I = 0, N = SavedStrings.size(); I != N; ++I)
@@ -1909,7 +1925,7 @@ void ASTWriter::WriteSourceManagerBlock(SourceManager &SourceMgr,
   Record.push_back(SOURCE_LOCATION_OFFSETS);
   Record.push_back(SLocEntryOffsets.size());
   Record.push_back(SourceMgr.getNextLocalOffset() - 1); // skip dummy
-  Stream.EmitRecordWithBlob(SLocOffsetsAbbrev, Record, data(SLocEntryOffsets));
+  Stream.EmitRecordWithBlob(SLocOffsetsAbbrev, Record, bytes(SLocEntryOffsets));
 
   // Write the source location entry preloads array, telling the AST
   // reader which source locations entries it should load eagerly.
@@ -1956,52 +1972,6 @@ void ASTWriter::WriteSourceManagerBlock(SourceManager &SourceMgr,
 // Preprocessor Serialization
 //===----------------------------------------------------------------------===//
 
-namespace {
-class ASTMacroTableTrait {
-public:
-  typedef IdentID key_type;
-  typedef key_type key_type_ref;
-
-  struct Data {
-    uint32_t MacroDirectivesOffset;
-  };
-
-  typedef Data data_type;
-  typedef const data_type &data_type_ref;
-  typedef unsigned hash_value_type;
-  typedef unsigned offset_type;
-
-  static hash_value_type ComputeHash(IdentID IdID) {
-    return llvm::hash_value(IdID);
-  }
-
-  std::pair<unsigned,unsigned>
-  static EmitKeyDataLength(raw_ostream& Out,
-                           key_type_ref Key, data_type_ref Data) {
-    unsigned KeyLen = 4; // IdentID.
-    unsigned DataLen = 4; // MacroDirectivesOffset.
-    return std::make_pair(KeyLen, DataLen);
-  }
-
-  static void EmitKey(raw_ostream& Out, key_type_ref Key, unsigned KeyLen) {
-    using namespace llvm::support;
-    endian::Writer<little>(Out).write<uint32_t>(Key);
-  }
-
-  static void EmitData(raw_ostream& Out, key_type_ref Key, data_type_ref Data,
-                       unsigned) {
-    using namespace llvm::support;
-    endian::Writer<little>(Out).write<uint32_t>(Data.MacroDirectivesOffset);
-  }
-};
-} // end anonymous namespace
-
-static int compareMacroDirectives(
-    const std::pair<const IdentifierInfo *, MacroDirective *> *X,
-    const std::pair<const IdentifierInfo *, MacroDirective *> *Y) {
-  return X->first->getName().compare(Y->first->getName());
-}
-
 static bool shouldIgnoreMacro(MacroDirective *MD, bool IsModule,
                               const Preprocessor &PP) {
   if (MacroInfo *MI = MD->getMacroInfo())
@@ -2009,10 +1979,6 @@ static bool shouldIgnoreMacro(MacroDirective *MD, bool IsModule,
       return true;
 
   if (IsModule) {
-    // Re-export any imported directives.
-    if (MD->isImported())
-      return false;
-
     SourceLocation Loc = MD->getLocation();
     if (Loc.isInvalid())
       return true;
@@ -2032,6 +1998,7 @@ void ASTWriter::WritePreprocessor(const Preprocessor &PP, bool IsModule) {
     WritePreprocessorDetail(*PPRec);
 
   RecordData Record;
+  RecordData ModuleMacroRecord;
 
   // If the preprocessor __COUNTER__ value has been bumped, remember it.
   if (PP.getCounterValue() != 0) {
@@ -2052,75 +2019,75 @@ void ASTWriter::WritePreprocessor(const Preprocessor &PP, bool IsModule) {
   // Loop over all the macro directives that are live at the end of the file,
   // emitting each to the PP section.
 
-  // Construct the list of macro directives that need to be serialized.
-  SmallVector<std::pair<const IdentifierInfo *, MacroDirective *>, 2>
-    MacroDirectives;
-  for (Preprocessor::macro_iterator
-         I = PP.macro_begin(/*IncludeExternalMacros=*/false),
-         E = PP.macro_end(/*IncludeExternalMacros=*/false);
-       I != E; ++I) {
-    MacroDirectives.push_back(std::make_pair(I->first, I->second));
-  }
-
+  // Construct the list of identifiers with macro directives that need to be
+  // serialized.
+  SmallVector<const IdentifierInfo *, 128> MacroIdentifiers;
+  for (auto &Id : PP.getIdentifierTable())
+    if (Id.second->hadMacroDefinition() &&
+        (!Id.second->isFromAST() ||
+         Id.second->hasChangedSinceDeserialization()))
+      MacroIdentifiers.push_back(Id.second);
   // Sort the set of macro definitions that need to be serialized by the
   // name of the macro, to provide a stable ordering.
-  llvm::array_pod_sort(MacroDirectives.begin(), MacroDirectives.end(),
-                       &compareMacroDirectives);
-
-  llvm::OnDiskChainedHashTableGenerator<ASTMacroTableTrait> Generator;
+  std::sort(MacroIdentifiers.begin(), MacroIdentifiers.end(),
+            llvm::less_ptr<IdentifierInfo>());
 
   // Emit the macro directives as a list and associate the offset with the
   // identifier they belong to.
-  for (unsigned I = 0, N = MacroDirectives.size(); I != N; ++I) {
-    const IdentifierInfo *Name = MacroDirectives[I].first;
-    uint64_t MacroDirectiveOffset = Stream.GetCurrentBitNo();
-    MacroDirective *MD = MacroDirectives[I].second;
-
-    // If the macro or identifier need no updates, don't write the macro history
-    // for this one.
-    // FIXME: Chain the macro history instead of re-writing it.
-    if (MD->isFromPCH() &&
-        Name->isFromAST() && !Name->hasChangedSinceDeserialization())
-      continue;
+  for (const IdentifierInfo *Name : MacroIdentifiers) {
+    MacroDirective *MD = PP.getLocalMacroDirectiveHistory(Name);
+    auto StartOffset = Stream.GetCurrentBitNo();
 
     // Emit the macro directives in reverse source order.
     for (; MD; MD = MD->getPrevious()) {
+      // Once we hit an ignored macro, we're done: the rest of the chain
+      // will all be ignored macros.
       if (shouldIgnoreMacro(MD, IsModule, PP))
-        continue;
+        break;
 
       AddSourceLocation(MD->getLocation(), Record);
       Record.push_back(MD->getKind());
       if (auto *DefMD = dyn_cast<DefMacroDirective>(MD)) {
-        MacroID InfoID = getMacroRef(DefMD->getInfo(), Name);
-        Record.push_back(InfoID);
-        Record.push_back(DefMD->getOwningModuleID());
-        Record.push_back(DefMD->isAmbiguous());
-      } else if (auto *UndefMD = dyn_cast<UndefMacroDirective>(MD)) {
-        Record.push_back(UndefMD->getOwningModuleID());
-      } else {
-        auto *VisMD = cast<VisibilityMacroDirective>(MD);
+        Record.push_back(getMacroRef(DefMD->getInfo(), Name));
+      } else if (auto *VisMD = dyn_cast<VisibilityMacroDirective>(MD)) {
         Record.push_back(VisMD->isPublic());
       }
+    }
 
-      if (MD->isImported()) {
-        auto Overrides = MD->getOverriddenModules();
-        Record.push_back(Overrides.size());
-        for (auto Override : Overrides)
-          Record.push_back(Override);
+    // Write out any exported module macros.
+    bool EmittedModuleMacros = false;
+    if (IsModule) {
+      auto Leafs = PP.getLeafModuleMacros(Name);
+      SmallVector<ModuleMacro*, 8> Worklist(Leafs.begin(), Leafs.end());
+      llvm::DenseMap<ModuleMacro*, unsigned> Visits;
+      while (!Worklist.empty()) {
+        auto *Macro = Worklist.pop_back_val();
+
+        // Emit a record indicating this submodule exports this macro.
+        ModuleMacroRecord.push_back(
+            getSubmoduleID(Macro->getOwningModule()));
+        ModuleMacroRecord.push_back(getMacroRef(Macro->getMacroInfo(), Name));
+        for (auto *M : Macro->overrides())
+          ModuleMacroRecord.push_back(getSubmoduleID(M->getOwningModule()));
+
+        Stream.EmitRecord(PP_MODULE_MACRO, ModuleMacroRecord);
+        ModuleMacroRecord.clear();
+
+        // Enqueue overridden macros once we've visited all their ancestors.
+        for (auto *M : Macro->overrides())
+          if (++Visits[M] == M->getNumOverridingMacros())
+            Worklist.push_back(M);
+
+        EmittedModuleMacros = true;
       }
     }
-    if (Record.empty())
+
+    if (Record.empty() && !EmittedModuleMacros)
       continue;
 
+    IdentMacroDirectivesOffsetMap[Name] = StartOffset;
     Stream.EmitRecord(PP_MACRO_DIRECTIVE_HISTORY, Record);
     Record.clear();
-
-    IdentMacroDirectivesOffsetMap[Name] = MacroDirectiveOffset;
-
-    IdentID NameID = getIdentifierRef(Name);
-    ASTMacroTableTrait::Data data;
-    data.MacroDirectivesOffset = MacroDirectiveOffset;
-    Generator.insert(NameID, data);
   }
 
   /// \brief Offsets of each of the macros into the bitstream, indexed by
@@ -2196,33 +2163,9 @@ void ASTWriter::WritePreprocessor(const Preprocessor &PP, bool IsModule) {
 
   Stream.ExitBlock();
 
-  // Create the on-disk hash table in a buffer.
-  SmallString<4096> MacroTable;
-  uint32_t BucketOffset;
-  {
-    using namespace llvm::support;
-    llvm::raw_svector_ostream Out(MacroTable);
-    // Make sure that no bucket is at offset 0
-    endian::Writer<little>(Out).write<uint32_t>(0);
-    BucketOffset = Generator.Emit(Out);
-  }
-
-  // Write the macro table
-  using namespace llvm;
-  BitCodeAbbrev *Abbrev = new BitCodeAbbrev();
-  Abbrev->Add(BitCodeAbbrevOp(MACRO_TABLE));
-  Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 32));
-  Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Blob));
-  unsigned MacroTableAbbrev = Stream.EmitAbbrev(Abbrev);
-
-  Record.push_back(MACRO_TABLE);
-  Record.push_back(BucketOffset);
-  Stream.EmitRecordWithBlob(MacroTableAbbrev, Record, MacroTable.str());
-  Record.clear();
-
   // Write the offsets table for macro IDs.
   using namespace llvm;
-  Abbrev = new BitCodeAbbrev();
+  auto *Abbrev = new BitCodeAbbrev();
   Abbrev->Add(BitCodeAbbrevOp(MACRO_OFFSET));
   Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 32)); // # of macros
   Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 32)); // first ID
@@ -2234,7 +2177,7 @@ void ASTWriter::WritePreprocessor(const Preprocessor &PP, bool IsModule) {
   Record.push_back(MacroOffsets.size());
   Record.push_back(FirstMacroID - NUM_PREDEF_MACRO_IDS);
   Stream.EmitRecordWithBlob(MacroOffsetAbbrev, Record,
-                            data(MacroOffsets));
+                            bytes(MacroOffsets));
 }
 
 void ASTWriter::WritePreprocessorDetail(PreprocessingRecord &PPRec) {
@@ -2274,13 +2217,13 @@ void ASTWriter::WritePreprocessorDetail(PreprocessingRecord &PPRec) {
        (void)++E, ++NumPreprocessingRecords, ++NextPreprocessorEntityID) {
     Record.clear();
 
-    PreprocessedEntityOffsets.push_back(PPEntityOffset((*E)->getSourceRange(),
-                                                     Stream.GetCurrentBitNo()));
+    PreprocessedEntityOffsets.push_back(
+        PPEntityOffset((*E)->getSourceRange(), Stream.GetCurrentBitNo()));
 
-    if (MacroDefinition *MD = dyn_cast<MacroDefinition>(*E)) {
+    if (MacroDefinitionRecord *MD = dyn_cast<MacroDefinitionRecord>(*E)) {
       // Record this macro definition's ID.
       MacroDefinitions[MD] = NextPreprocessorEntityID;
-      
+
       AddIdentifierRef(MD->getName(), Record);
       Stream.EmitRecord(PPD_MACRO_DEFINITION, Record);
       continue;
@@ -2332,7 +2275,7 @@ void ASTWriter::WritePreprocessorDetail(PreprocessingRecord &PPRec) {
     Record.push_back(PPD_ENTITIES_OFFSETS);
     Record.push_back(FirstPreprocessorEntityID - NUM_PREDEF_PP_ENTITY_IDS);
     Stream.EmitRecordWithBlob(PPEOffsetAbbrev, Record,
-                              data(PreprocessedEntityOffsets));
+                              bytes(PreprocessedEntityOffsets));
   }
 }
 
@@ -2369,19 +2312,6 @@ static unsigned getNumberOfModules(Module *Mod) {
 }
 
 void ASTWriter::WriteSubmodules(Module *WritingModule) {
-  // Determine the dependencies of our module and each of it's submodules.
-  // FIXME: This feels like it belongs somewhere else, but there are no
-  // other consumers of this information.
-  SourceManager &SrcMgr = PP->getSourceManager();
-  ModuleMap &ModMap = PP->getHeaderSearchInfo().getModuleMap();
-  for (const auto *I : Context->local_imports()) {
-    if (Module *ImportedFrom
-          = ModMap.inferModuleFromLocation(FullSourceLoc(I->getLocation(), 
-                                                         SrcMgr))) {
-      ImportedFrom->Imports.push_back(I->getImportedModule());
-    }
-  }
-  
   // Enter the submodule description block.
   Stream.EnterSubblock(SUBMODULE_BLOCK_ID, /*bits for abbreviations*/5);
   
@@ -2509,16 +2439,16 @@ void ASTWriter::WriteSubmodules(Module *WritingModule) {
     }
 
     // Emit the umbrella header, if there is one.
-    if (const FileEntry *UmbrellaHeader = Mod->getUmbrellaHeader()) {
+    if (auto UmbrellaHeader = Mod->getUmbrellaHeader()) {
       Record.clear();
       Record.push_back(SUBMODULE_UMBRELLA_HEADER);
-      Stream.EmitRecordWithBlob(UmbrellaAbbrev, Record, 
-                                UmbrellaHeader->getName());
-    } else if (const DirectoryEntry *UmbrellaDir = Mod->getUmbrellaDir()) {
+      Stream.EmitRecordWithBlob(UmbrellaAbbrev, Record,
+                                UmbrellaHeader.NameAsWritten);
+    } else if (auto UmbrellaDir = Mod->getUmbrellaDir()) {
       Record.clear();
       Record.push_back(SUBMODULE_UMBRELLA_DIR);
       Stream.EmitRecordWithBlob(UmbrellaDirAbbrev, Record, 
-                                UmbrellaDir->getName());      
+                                UmbrellaDir.NameAsWritten);
     }
 
     // Emit the headers.
@@ -2566,8 +2496,7 @@ void ASTWriter::WriteSubmodules(Module *WritingModule) {
       Record.clear();
       for (unsigned I = 0, N = Mod->Exports.size(); I != N; ++I) {
         if (Module *Exported = Mod->Exports[I].getPointer()) {
-          unsigned ExportedID = SubmoduleIDs[Exported];
-          assert(ExportedID > 0 && "Unknown submodule ID?");
+          unsigned ExportedID = getSubmoduleID(Exported);
           Record.push_back(ExportedID);
         } else {
           Record.push_back(0);
@@ -2618,9 +2547,14 @@ void ASTWriter::WriteSubmodules(Module *WritingModule) {
   }
   
   Stream.ExitBlock();
-  
-  assert((NextSubmoduleID - FirstSubmoduleID
-            == getNumberOfModules(WritingModule)) && "Wrong # of submodules");
+
+  // FIXME: This can easily happen, if we have a reference to a submodule that
+  // did not result in us loading a module file for that submodule. For
+  // instance, a cross-top-level-module 'conflict' declaration will hit this.
+  assert((NextSubmoduleID - FirstSubmoduleID ==
+          getNumberOfModules(WritingModule)) &&
+         "Wrong # of submodules; found a reference to a non-local, "
+         "non-imported submodule?");
 }
 
 serialization::SubmoduleID 
@@ -2684,6 +2618,29 @@ void ASTWriter::WritePragmaDiagnosticMappings(const DiagnosticsEngine &Diag,
     Stream.EmitRecord(DIAG_PRAGMA_MAPPINGS, Record);
 }
 
+void ASTWriter::WriteCXXCtorInitializersOffsets() {
+  if (CXXCtorInitializersOffsets.empty())
+    return;
+
+  RecordData Record;
+
+  // Create a blob abbreviation for the C++ ctor initializer offsets.
+  using namespace llvm;
+
+  BitCodeAbbrev *Abbrev = new BitCodeAbbrev();
+  Abbrev->Add(BitCodeAbbrevOp(CXX_CTOR_INITIALIZERS_OFFSETS));
+  Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Fixed, 32)); // size
+  Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Blob));
+  unsigned CtorInitializersOffsetAbbrev = Stream.EmitAbbrev(Abbrev);
+
+  // Write the base specifier offsets table.
+  Record.clear();
+  Record.push_back(CXX_CTOR_INITIALIZERS_OFFSETS);
+  Record.push_back(CXXCtorInitializersOffsets.size());
+  Stream.EmitRecordWithBlob(CtorInitializersOffsetAbbrev, Record,
+                            bytes(CXXCtorInitializersOffsets));
+}
+
 void ASTWriter::WriteCXXBaseSpecifiersOffsets() {
   if (CXXBaseSpecifiersOffsets.empty())
     return;
@@ -2704,7 +2661,7 @@ void ASTWriter::WriteCXXBaseSpecifiersOffsets() {
   Record.push_back(CXX_BASE_SPECIFIER_OFFSETS);
   Record.push_back(CXXBaseSpecifiersOffsets.size());
   Stream.EmitRecordWithBlob(BaseSpecifierOffsetAbbrev, Record,
-                            data(CXXBaseSpecifiersOffsets));
+                            bytes(CXXBaseSpecifiersOffsets));
 }
 
 //===----------------------------------------------------------------------===//
@@ -2780,7 +2737,7 @@ uint64_t ASTWriter::WriteDeclContextLexicalBlock(ASTContext &Context,
     Decls.push_back(std::make_pair(D->getKind(), GetDeclRef(D)));
 
   ++NumLexicalDeclContexts;
-  Stream.EmitRecordWithBlob(DeclContextLexicalAbbrev, Record, data(Decls));
+  Stream.EmitRecordWithBlob(DeclContextLexicalAbbrev, Record, bytes(Decls));
   return Offset;
 }
 
@@ -2799,7 +2756,7 @@ void ASTWriter::WriteTypeDeclOffsets() {
   Record.push_back(TYPE_OFFSET);
   Record.push_back(TypeOffsets.size());
   Record.push_back(FirstTypeID - NUM_PREDEF_TYPE_IDS);
-  Stream.EmitRecordWithBlob(TypeOffsetAbbrev, Record, data(TypeOffsets));
+  Stream.EmitRecordWithBlob(TypeOffsetAbbrev, Record, bytes(TypeOffsets));
 
   // Write the declaration offsets array
   Abbrev = new BitCodeAbbrev();
@@ -2812,22 +2769,25 @@ void ASTWriter::WriteTypeDeclOffsets() {
   Record.push_back(DECL_OFFSET);
   Record.push_back(DeclOffsets.size());
   Record.push_back(FirstDeclID - NUM_PREDEF_DECL_IDS);
-  Stream.EmitRecordWithBlob(DeclOffsetAbbrev, Record, data(DeclOffsets));
+  Stream.EmitRecordWithBlob(DeclOffsetAbbrev, Record, bytes(DeclOffsets));
 }
 
 void ASTWriter::WriteFileDeclIDsMap() {
   using namespace llvm;
   RecordData Record;
 
+  SmallVector<std::pair<FileID, DeclIDInFileInfo *>, 64> SortedFileDeclIDs(
+      FileDeclIDs.begin(), FileDeclIDs.end());
+  std::sort(SortedFileDeclIDs.begin(), SortedFileDeclIDs.end(),
+            llvm::less_first());
+
   // Join the vectors of DeclIDs from all files.
-  SmallVector<DeclID, 256> FileSortedIDs;
-  for (FileDeclIDsTy::iterator
-         FI = FileDeclIDs.begin(), FE = FileDeclIDs.end(); FI != FE; ++FI) {
-    DeclIDInFileInfo &Info = *FI->second;
-    Info.FirstDeclIndex = FileSortedIDs.size();
-    for (LocDeclIDsTy::iterator
-           DI = Info.DeclIDs.begin(), DE = Info.DeclIDs.end(); DI != DE; ++DI)
-      FileSortedIDs.push_back(DI->second);
+  SmallVector<DeclID, 256> FileGroupedDeclIDs;
+  for (auto &FileDeclEntry : SortedFileDeclIDs) {
+    DeclIDInFileInfo &Info = *FileDeclEntry.second;
+    Info.FirstDeclIndex = FileGroupedDeclIDs.size();
+    for (auto &LocDeclEntry : Info.DeclIDs)
+      FileGroupedDeclIDs.push_back(LocDeclEntry.second);
   }
 
   BitCodeAbbrev *Abbrev = new BitCodeAbbrev();
@@ -2836,8 +2796,8 @@ void ASTWriter::WriteFileDeclIDsMap() {
   Abbrev->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::Blob));
   unsigned AbbrevCode = Stream.EmitAbbrev(Abbrev);
   Record.push_back(FILE_SORTED_DECLS);
-  Record.push_back(FileSortedIDs.size());
-  Stream.EmitRecordWithBlob(AbbrevCode, Record, data(FileSortedIDs));
+  Record.push_back(FileGroupedDeclIDs.size());
+  Stream.EmitRecordWithBlob(AbbrevCode, Record, bytes(FileGroupedDeclIDs));
 }
 
 void ASTWriter::WriteComments() {
@@ -2988,13 +2948,12 @@ void ASTWriter::WriteSelectors(Sema &SemaRef) {
     // Create the on-disk hash table representation. We walk through every
     // selector we've seen and look it up in the method pool.
     SelectorOffsets.resize(NextSelectorID - FirstSelectorID);
-    for (llvm::DenseMap<Selector, SelectorID>::iterator
-             I = SelectorIDs.begin(), E = SelectorIDs.end();
-         I != E; ++I) {
-      Selector S = I->first;
+    for (auto &SelectorAndID : SelectorIDs) {
+      Selector S = SelectorAndID.first;
+      SelectorID ID = SelectorAndID.second;
       Sema::GlobalMethodPool::iterator F = SemaRef.MethodPool.find(S);
       ASTMethodPoolTrait::data_type Data = {
-        I->second,
+        ID,
         ObjCMethodList(),
         ObjCMethodList()
       };
@@ -3004,7 +2963,7 @@ void ASTWriter::WriteSelectors(Sema &SemaRef) {
       }
       // Only write this selector if it's not in an existing AST or something
       // changed.
-      if (Chain && I->second < FirstSelectorID) {
+      if (Chain && ID < FirstSelectorID) {
         // Selector already exists. Did it change?
         bool changed = false;
         for (ObjCMethodList *M = &Data.Instance;
@@ -3051,7 +3010,7 @@ void ASTWriter::WriteSelectors(Sema &SemaRef) {
     Record.push_back(METHOD_POOL);
     Record.push_back(BucketOffset);
     Record.push_back(NumTableEntries);
-    Stream.EmitRecordWithBlob(MethodPoolAbbrev, Record, MethodPool.str());
+    Stream.EmitRecordWithBlob(MethodPoolAbbrev, Record, MethodPool);
 
     // Create a blob abbreviation for the selector table offsets.
     Abbrev = new BitCodeAbbrev();
@@ -3067,7 +3026,7 @@ void ASTWriter::WriteSelectors(Sema &SemaRef) {
     Record.push_back(SelectorOffsets.size());
     Record.push_back(FirstSelectorID - NUM_PREDEF_SELECTOR_IDS);
     Stream.EmitRecordWithBlob(SelectorOffsetAbbrev, Record,
-                              data(SelectorOffsets));
+                              bytes(SelectorOffsets));
   }
 }
 
@@ -3082,11 +3041,9 @@ void ASTWriter::WriteReferencedSelectorsPool(Sema &SemaRef) {
   // Note: this writes out all references even for a dependent AST. But it is
   // very tricky to fix, and given that @selector shouldn't really appear in
   // headers, probably not worth it. It's not a correctness issue.
-  for (DenseMap<Selector, SourceLocation>::iterator S =
-       SemaRef.ReferencedSelectors.begin(),
-       E = SemaRef.ReferencedSelectors.end(); S != E; ++S) {
-    Selector Sel = (*S).first;
-    SourceLocation Loc = (*S).second;
+  for (auto &SelectorAndLocation : SemaRef.ReferencedSelectors) {
+    Selector Sel = SelectorAndLocation.first;
+    SourceLocation Loc = SelectorAndLocation.second;
     AddSelectorRef(Sel, Record);
     AddSourceLocation(Loc, Record);
   }
@@ -3097,174 +3054,59 @@ void ASTWriter::WriteReferencedSelectorsPool(Sema &SemaRef) {
 // Identifier Table Serialization
 //===----------------------------------------------------------------------===//
 
+/// Determine the declaration that should be put into the name lookup table to
+/// represent the given declaration in this module. This is usually D itself,
+/// but if D was imported and merged into a local declaration, we want the most
+/// recent local declaration instead. The chosen declaration will be the most
+/// recent declaration in any module that imports this one.
+static NamedDecl *getDeclForLocalLookup(const LangOptions &LangOpts,
+                                        NamedDecl *D) {
+  if (!LangOpts.Modules || !D->isFromASTFile())
+    return D;
+
+  if (Decl *Redecl = D->getPreviousDecl()) {
+    // For Redeclarable decls, a prior declaration might be local.
+    for (; Redecl; Redecl = Redecl->getPreviousDecl()) {
+      if (!Redecl->isFromASTFile())
+        return cast<NamedDecl>(Redecl);
+      // If we find a decl from a (chained-)PCH stop since we won't find a
+      // local one.
+      if (D->getOwningModuleID() == 0)
+        break;
+    }
+  } else if (Decl *First = D->getCanonicalDecl()) {
+    // For Mergeable decls, the first decl might be local.
+    if (!First->isFromASTFile())
+      return cast<NamedDecl>(First);
+  }
+
+  // All declarations are imported. Our most recent declaration will also be
+  // the most recent one in anyone who imports us.
+  return D;
+}
+
 namespace {
 class ASTIdentifierTableTrait {
   ASTWriter &Writer;
   Preprocessor &PP;
   IdentifierResolver &IdResolver;
-  bool IsModule;
   
-  /// \brief Determines whether this is an "interesting" identifier
-  /// that needs a full IdentifierInfo structure written into the hash
-  /// table.
-  bool isInterestingIdentifier(IdentifierInfo *II, MacroDirective *&Macro) {
-    if (II->isPoisoned() ||
+  /// \brief Determines whether this is an "interesting" identifier that needs a
+  /// full IdentifierInfo structure written into the hash table. Notably, this
+  /// doesn't check whether the name has macros defined; use PublicMacroIterator
+  /// to check that.
+  bool isInterestingIdentifier(IdentifierInfo *II, uint64_t MacroOffset) {
+    if (MacroOffset ||
+        II->isPoisoned() ||
         II->isExtensionToken() ||
         II->getObjCOrBuiltinID() ||
         II->hasRevertedTokenIDToIdentifier() ||
         II->getFETokenInfo<void>())
       return true;
 
-    return hadMacroDefinition(II, Macro);
-  }
-
-  bool hadMacroDefinition(IdentifierInfo *II, MacroDirective *&Macro) {
-    if (!II->hadMacroDefinition())
-      return false;
-
-    if (Macro || (Macro = PP.getMacroDirectiveHistory(II))) {
-      if (!IsModule)
-        return !shouldIgnoreMacro(Macro, IsModule, PP);
-
-      MacroState State;
-      if (getFirstPublicSubmoduleMacro(Macro, State))
-        return true;
-    }
-
     return false;
   }
 
-  enum class SubmoduleMacroState {
-    /// We've seen nothing about this macro.
-    None,
-    /// We've seen a public visibility directive.
-    Public,
-    /// We've either exported a macro for this module or found that the
-    /// module's definition of this macro is private.
-    Done
-  };
-  typedef llvm::DenseMap<SubmoduleID, SubmoduleMacroState> MacroState;
-
-  MacroDirective *
-  getFirstPublicSubmoduleMacro(MacroDirective *MD, MacroState &State) {
-    if (MacroDirective *NextMD = getPublicSubmoduleMacro(MD, State))
-      return NextMD;
-    return nullptr;
-  }
-
-  MacroDirective *
-  getNextPublicSubmoduleMacro(MacroDirective *MD, MacroState &State) {
-    if (MacroDirective *NextMD =
-            getPublicSubmoduleMacro(MD->getPrevious(), State))
-      return NextMD;
-    return nullptr;
-  }
-
-  /// \brief Traverses the macro directives history and returns the next
-  /// public macro definition or undefinition that has not been found so far.
-  ///
-  /// A macro that is defined in submodule A and undefined in submodule B
-  /// will still be considered as defined/exported from submodule A.
-  MacroDirective *getPublicSubmoduleMacro(MacroDirective *MD,
-                                          MacroState &State) {
-    if (!MD)
-      return nullptr;
-
-    Optional<bool> IsPublic;
-    for (; MD; MD = MD->getPrevious()) {
-      // Once we hit an ignored macro, we're done: the rest of the chain
-      // will all be ignored macros.
-      if (shouldIgnoreMacro(MD, IsModule, PP))
-        break;
-
-      // If this macro was imported, re-export it.
-      if (MD->isImported())
-        return MD;
-
-      SubmoduleID ModID = getSubmoduleID(MD);
-      auto &S = State[ModID];
-      assert(ModID && "found macro in no submodule");
-
-      if (S == SubmoduleMacroState::Done)
-        continue;
-
-      if (auto *VisMD = dyn_cast<VisibilityMacroDirective>(MD)) {
-        // The latest visibility directive for a name in a submodule affects all
-        // the directives that come before it.
-        if (S == SubmoduleMacroState::None)
-          S = VisMD->isPublic() ? SubmoduleMacroState::Public
-                                : SubmoduleMacroState::Done;
-      } else {
-        S = SubmoduleMacroState::Done;
-        return MD;
-      }
-    }
-
-    return nullptr;
-  }
-
-  ArrayRef<SubmoduleID>
-  getOverriddenSubmodules(MacroDirective *MD,
-                          SmallVectorImpl<SubmoduleID> &ScratchSpace) {
-    assert(!isa<VisibilityMacroDirective>(MD) &&
-           "only #define and #undef can override");
-    if (MD->isImported())
-      return MD->getOverriddenModules();
-
-    ScratchSpace.clear();
-    SubmoduleID ModID = getSubmoduleID(MD);
-    for (MD = MD->getPrevious(); MD; MD = MD->getPrevious()) {
-      if (shouldIgnoreMacro(MD, IsModule, PP))
-        break;
-
-      // If this is a definition from a submodule import, that submodule's
-      // definition is overridden by the definition or undefinition that we
-      // started with.
-      if (MD->isImported()) {
-        if (auto *DefMD = dyn_cast<DefMacroDirective>(MD)) {
-          SubmoduleID DefModuleID = DefMD->getInfo()->getOwningModuleID();
-          assert(DefModuleID && "imported macro has no owning module");
-          ScratchSpace.push_back(DefModuleID);
-        } else if (auto *UndefMD = dyn_cast<UndefMacroDirective>(MD)) {
-          // If we override a #undef, we override anything that #undef overrides.
-          // We don't need to override it, since an active #undef doesn't affect
-          // the meaning of a macro.
-          auto Overrides = UndefMD->getOverriddenModules();
-          ScratchSpace.insert(ScratchSpace.end(),
-                              Overrides.begin(), Overrides.end());
-        }
-      }
-
-      // Stop once we leave the original macro's submodule.
-      //
-      // Either this submodule #included another submodule of the same
-      // module or it just happened to be built after the other module.
-      // In the former case, we override the submodule's macro.
-      //
-      // FIXME: In the latter case, we shouldn't do so, but we can't tell
-      // these cases apart.
-      //
-      // FIXME: We can leave this submodule and re-enter it if it #includes a
-      // header within a different submodule of the same module. In such cases
-      // the overrides list will be incomplete.
-      SubmoduleID DirectiveModuleID = getSubmoduleID(MD);
-      if (DirectiveModuleID != ModID) {
-        if (DirectiveModuleID && !MD->isImported())
-          ScratchSpace.push_back(DirectiveModuleID);
-        break;
-      }
-    }
-
-    std::sort(ScratchSpace.begin(), ScratchSpace.end());
-    ScratchSpace.erase(std::unique(ScratchSpace.begin(), ScratchSpace.end()),
-                       ScratchSpace.end());
-    return ScratchSpace;
-  }
-
-  SubmoduleID getSubmoduleID(MacroDirective *MD) {
-    return Writer.inferSubmoduleIDFromLocation(MD->getLocation());
-  }
-
 public:
   typedef IdentifierInfo* key_type;
   typedef key_type  key_type_ref;
@@ -3275,9 +3117,9 @@ public:
   typedef unsigned hash_value_type;
   typedef unsigned offset_type;
 
-  ASTIdentifierTableTrait(ASTWriter &Writer, Preprocessor &PP, 
-                          IdentifierResolver &IdResolver, bool IsModule)
-    : Writer(Writer), PP(PP), IdResolver(IdResolver), IsModule(IsModule) { }
+  ASTIdentifierTableTrait(ASTWriter &Writer, Preprocessor &PP,
+                          IdentifierResolver &IdResolver)
+      : Writer(Writer), PP(PP), IdResolver(IdResolver) {}
 
   static hash_value_type ComputeHash(const IdentifierInfo* II) {
     return llvm::HashString(II->getName());
@@ -3287,25 +3129,12 @@ public:
   EmitKeyDataLength(raw_ostream& Out, IdentifierInfo* II, IdentID ID) {
     unsigned KeyLen = II->getLength() + 1;
     unsigned DataLen = 4; // 4 bytes for the persistent ID << 1
-    MacroDirective *Macro = nullptr;
-    if (isInterestingIdentifier(II, Macro)) {
+    auto MacroOffset = Writer.getMacroDirectivesOffset(II);
+    if (isInterestingIdentifier(II, MacroOffset)) {
       DataLen += 2; // 2 bytes for builtin ID
       DataLen += 2; // 2 bytes for flags
-      if (hadMacroDefinition(II, Macro)) {
+      if (MacroOffset)
         DataLen += 4; // MacroDirectives offset.
-        if (IsModule) {
-          MacroState State;
-          SmallVector<SubmoduleID, 16> Scratch;
-          for (MacroDirective *MD = getFirstPublicSubmoduleMacro(Macro, State);
-               MD; MD = getNextPublicSubmoduleMacro(MD, State)) {
-            DataLen += 4; // MacroInfo ID or ModuleID.
-            if (unsigned NumOverrides =
-                    getOverriddenSubmodules(MD, Scratch).size())
-              DataLen += 4 * (1 + NumOverrides);
-          }
-          DataLen += 4; // 0 terminator.
-        }
-      }
 
       for (IdentifierResolver::iterator D = IdResolver.begin(II),
                                      DEnd = IdResolver.end();
@@ -3315,6 +3144,7 @@ public:
     using namespace llvm::support;
     endian::Writer<little> LE(Out);
 
+    assert((uint16_t)DataLen == DataLen && (uint16_t)KeyLen == KeyLen);
     LE.write<uint16_t>(DataLen);
     // We emit the key length after the data length so that every
     // string is preceded by a 16-bit length. This matches the PTH
@@ -3331,25 +3161,13 @@ public:
     Out.write(II->getNameStart(), KeyLen);
   }
 
-  static void emitMacroOverrides(raw_ostream &Out,
-                                 ArrayRef<SubmoduleID> Overridden) {
-    if (!Overridden.empty()) {
-      using namespace llvm::support;
-      endian::Writer<little> LE(Out);
-      LE.write<uint32_t>(Overridden.size() | 0x80000000U);
-      for (unsigned I = 0, N = Overridden.size(); I != N; ++I) {
-        assert(Overridden[I] && "zero module ID for override");
-        LE.write<uint32_t>(Overridden[I]);
-      }
-    }
-  }
-
   void EmitData(raw_ostream& Out, IdentifierInfo* II,
                 IdentID ID, unsigned) {
     using namespace llvm::support;
     endian::Writer<little> LE(Out);
-    MacroDirective *Macro = nullptr;
-    if (!isInterestingIdentifier(II, Macro)) {
+
+    auto MacroOffset = Writer.getMacroDirectivesOffset(II);
+    if (!isInterestingIdentifier(II, MacroOffset)) {
       LE.write<uint32_t>(ID << 1);
       return;
     }
@@ -3359,77 +3177,29 @@ public:
     assert((Bits & 0xffff) == Bits && "ObjCOrBuiltinID too big for ASTReader.");
     LE.write<uint16_t>(Bits);
     Bits = 0;
-    bool HadMacroDefinition = hadMacroDefinition(II, Macro);
+    bool HadMacroDefinition = MacroOffset != 0;
     Bits = (Bits << 1) | unsigned(HadMacroDefinition);
-    Bits = (Bits << 1) | unsigned(IsModule);
     Bits = (Bits << 1) | unsigned(II->isExtensionToken());
     Bits = (Bits << 1) | unsigned(II->isPoisoned());
     Bits = (Bits << 1) | unsigned(II->hasRevertedTokenIDToIdentifier());
     Bits = (Bits << 1) | unsigned(II->isCPlusPlusOperatorKeyword());
     LE.write<uint16_t>(Bits);
 
-    if (HadMacroDefinition) {
-      LE.write<uint32_t>(Writer.getMacroDirectivesOffset(II));
-      if (IsModule) {
-        // Write the IDs of macros coming from different submodules.
-        MacroState State;
-        SmallVector<SubmoduleID, 16> Scratch;
-        for (MacroDirective *MD = getFirstPublicSubmoduleMacro(Macro, State);
-             MD; MD = getNextPublicSubmoduleMacro(MD, State)) {
-          if (DefMacroDirective *DefMD = dyn_cast<DefMacroDirective>(MD)) {
-            // FIXME: If this macro directive was created by #pragma pop_macros,
-            // or if it was created implicitly by resolving conflicting macros,
-            // it may be for a different submodule from the one in the MacroInfo
-            // object. If so, we should write out its owning ModuleID.
-            MacroID InfoID = Writer.getMacroID(DefMD->getInfo());
-            assert(InfoID);
-            LE.write<uint32_t>(InfoID << 1);
-          } else {
-            auto *UndefMD = cast<UndefMacroDirective>(MD);
-            SubmoduleID Mod = UndefMD->isImported()
-                                  ? UndefMD->getOwningModuleID()
-                                  : getSubmoduleID(UndefMD);
-            LE.write<uint32_t>((Mod << 1) | 1);
-          }
-          emitMacroOverrides(Out, getOverriddenSubmodules(MD, Scratch));
-        }
-        LE.write<uint32_t>(0xdeadbeef);
-      }
-    }
+    if (HadMacroDefinition)
+      LE.write<uint32_t>(MacroOffset);
 
     // Emit the declaration IDs in reverse order, because the
     // IdentifierResolver provides the declarations as they would be
     // visible (e.g., the function "stat" would come before the struct
-    // "stat"), but the ASTReader adds declarations to the end of the list 
-    // (so we need to see the struct "status" before the function "status").
+    // "stat"), but the ASTReader adds declarations to the end of the list
+    // (so we need to see the struct "stat" before the function "stat").
     // Only emit declarations that aren't from a chained PCH, though.
-    SmallVector<Decl *, 16> Decls(IdResolver.begin(II),
-                                  IdResolver.end());
-    for (SmallVectorImpl<Decl *>::reverse_iterator D = Decls.rbegin(),
-                                                DEnd = Decls.rend();
+    SmallVector<NamedDecl *, 16> Decls(IdResolver.begin(II), IdResolver.end());
+    for (SmallVectorImpl<NamedDecl *>::reverse_iterator D = Decls.rbegin(),
+                                                        DEnd = Decls.rend();
          D != DEnd; ++D)
-      LE.write<uint32_t>(Writer.getDeclID(getMostRecentLocalDecl(*D)));
-  }
-
-  /// \brief Returns the most recent local decl or the given decl if there are
-  /// no local ones. The given decl is assumed to be the most recent one.
-  Decl *getMostRecentLocalDecl(Decl *Orig) {
-    // The only way a "from AST file" decl would be more recent from a local one
-    // is if it came from a module.
-    if (!PP.getLangOpts().Modules)
-      return Orig;
-
-    // Look for a local in the decl chain.
-    for (Decl *D = Orig; D; D = D->getPreviousDecl()) {
-      if (!D->isFromASTFile())
-        return D;
-      // If we come up a decl from a (chained-)PCH stop since we won't find a
-      // local one.
-      if (D->getOwningModuleID() == 0)
-        break;
-    }
-
-    return Orig;
+      LE.write<uint32_t>(
+          Writer.getDeclID(getDeclForLocalLookup(PP.getLangOpts(), *D)));
   }
 };
 } // end anonymous namespace
@@ -3448,29 +3218,33 @@ void ASTWriter::WriteIdentifierTable(Preprocessor &PP,
   // strings.
   {
     llvm::OnDiskChainedHashTableGenerator<ASTIdentifierTableTrait> Generator;
-    ASTIdentifierTableTrait Trait(*this, PP, IdResolver, IsModule);
+    ASTIdentifierTableTrait Trait(*this, PP, IdResolver);
 
     // Look for any identifiers that were named while processing the
     // headers, but are otherwise not needed. We add these to the hash
     // table to enable checking of the predefines buffer in the case
     // where the user adds new macro definitions when building the AST
     // file.
+    SmallVector<const IdentifierInfo *, 128> IIs;
     for (IdentifierTable::iterator ID = PP.getIdentifierTable().begin(),
                                 IDEnd = PP.getIdentifierTable().end();
          ID != IDEnd; ++ID)
-      getIdentifierRef(ID->second);
+      IIs.push_back(ID->second);
+    // Sort the identifiers lexicographically before getting them references so
+    // that their order is stable.
+    std::sort(IIs.begin(), IIs.end(), llvm::less_ptr<IdentifierInfo>());
+    for (const IdentifierInfo *II : IIs)
+      getIdentifierRef(II);
 
     // Create the on-disk hash table representation. We only store offsets
     // for identifiers that appear here for the first time.
     IdentifierOffsets.resize(NextIdentID - FirstIdentID);
-    for (llvm::DenseMap<const IdentifierInfo *, IdentID>::iterator
-           ID = IdentifierIDs.begin(), IDEnd = IdentifierIDs.end();
-         ID != IDEnd; ++ID) {
-      assert(ID->first && "NULL identifier in identifier table");
-      if (!Chain || !ID->first->isFromAST() || 
-          ID->first->hasChangedSinceDeserialization())
-        Generator.insert(const_cast<IdentifierInfo *>(ID->first), ID->second,
-                         Trait);
+    for (auto IdentIDPair : IdentifierIDs) {
+      IdentifierInfo *II = const_cast<IdentifierInfo *>(IdentIDPair.first);
+      IdentID ID = IdentIDPair.second;
+      assert(II && "NULL identifier in identifier table");
+      if (!Chain || !II->isFromAST() || II->hasChangedSinceDeserialization())
+        Generator.insert(II, ID, Trait);
     }
 
     // Create the on-disk hash table in a buffer.
@@ -3478,7 +3252,6 @@ void ASTWriter::WriteIdentifierTable(Preprocessor &PP,
     uint32_t BucketOffset;
     {
       using namespace llvm::support;
-      ASTIdentifierTableTrait Trait(*this, PP, IdResolver, IsModule);
       llvm::raw_svector_ostream Out(IdentifierTable);
       // Make sure that no bucket is at offset 0
       endian::Writer<little>(Out).write<uint32_t>(0);
@@ -3496,7 +3269,7 @@ void ASTWriter::WriteIdentifierTable(Preprocessor &PP,
     RecordData Record;
     Record.push_back(IDENTIFIER_TABLE);
     Record.push_back(BucketOffset);
-    Stream.EmitRecordWithBlob(IDTableAbbrev, Record, IdentifierTable.str());
+    Stream.EmitRecordWithBlob(IDTableAbbrev, Record, IdentifierTable);
   }
 
   // Write the offsets table for identifier IDs.
@@ -3517,38 +3290,13 @@ void ASTWriter::WriteIdentifierTable(Preprocessor &PP,
   Record.push_back(IdentifierOffsets.size());
   Record.push_back(FirstIdentID - NUM_PREDEF_IDENT_IDS);
   Stream.EmitRecordWithBlob(IdentifierOffsetAbbrev, Record,
-                            data(IdentifierOffsets));
+                            bytes(IdentifierOffsets));
 }
 
 //===----------------------------------------------------------------------===//
 // DeclContext's Name Lookup Table Serialization
 //===----------------------------------------------------------------------===//
 
-/// Determine the declaration that should be put into the name lookup table to
-/// represent the given declaration in this module. This is usually D itself,
-/// but if D was imported and merged into a local declaration, we want the most
-/// recent local declaration instead. The chosen declaration will be the most
-/// recent declaration in any module that imports this one.
-static NamedDecl *getDeclForLocalLookup(NamedDecl *D) {
-  if (!D->isFromASTFile())
-    return D;
-
-  if (Decl *Redecl = D->getPreviousDecl()) {
-    // For Redeclarable decls, a prior declaration might be local.
-    for (; Redecl; Redecl = Redecl->getPreviousDecl())
-      if (!Redecl->isFromASTFile())
-        return cast<NamedDecl>(Redecl);
-  } else if (Decl *First = D->getCanonicalDecl()) {
-    // For Mergeable decls, the first decl might be local.
-    if (!First->isFromASTFile())
-      return cast<NamedDecl>(First);
-  }
-
-  // All declarations are imported. Our most recent declaration will also be
-  // the most recent one in anyone who imports us.
-  return D;
-}
-
 namespace {
 // Trait used for the on-disk hash table used in the method pool.
 class ASTDeclContextNameLookupTrait {
@@ -3666,119 +3414,198 @@ public:
     LE.write<uint16_t>(Lookup.size());
     for (DeclContext::lookup_iterator I = Lookup.begin(), E = Lookup.end();
          I != E; ++I)
-      LE.write<uint32_t>(Writer.GetDeclRef(getDeclForLocalLookup(*I)));
+      LE.write<uint32_t>(
+          Writer.GetDeclRef(getDeclForLocalLookup(Writer.getLangOpts(), *I)));
 
     assert(Out.tell() - Start == DataLen && "Data length is wrong");
   }
 };
 } // end anonymous namespace
 
-template<typename Visitor>
-static void visitLocalLookupResults(const DeclContext *ConstDC,
-                                    bool NeedToReconcileExternalVisibleStorage,
-                                    Visitor AddLookupResult) {
+bool ASTWriter::isLookupResultExternal(StoredDeclsList &Result,
+                                       DeclContext *DC) {
+  return Result.hasExternalDecls() && DC->NeedToReconcileExternalVisibleStorage;
+}
+
+bool ASTWriter::isLookupResultEntirelyExternal(StoredDeclsList &Result,
+                                               DeclContext *DC) {
+  for (auto *D : Result.getLookupResult())
+    if (!getDeclForLocalLookup(getLangOpts(), D)->isFromASTFile())
+      return false;
+
+  return true;
+}
+
+uint32_t
+ASTWriter::GenerateNameLookupTable(const DeclContext *ConstDC,
+                                   llvm::SmallVectorImpl<char> &LookupTable) {
+  assert(!ConstDC->HasLazyLocalLexicalLookups &&
+         !ConstDC->HasLazyExternalLexicalLookups &&
+         "must call buildLookups first");
+
   // FIXME: We need to build the lookups table, which is logically const.
   DeclContext *DC = const_cast<DeclContext*>(ConstDC);
   assert(DC == DC->getPrimaryContext() && "only primary DC has lookup table");
 
-  SmallVector<DeclarationName, 16> ExternalNames;
+  // Create the on-disk hash table representation.
+  llvm::OnDiskChainedHashTableGenerator<ASTDeclContextNameLookupTrait>
+      Generator;
+  ASTDeclContextNameLookupTrait Trait(*this);
+
+  // The first step is to collect the declaration names which we need to
+  // serialize into the name lookup table, and to collect them in a stable
+  // order.
+  SmallVector<DeclarationName, 16> Names;
+
+  // We also build up small sets of the constructor and conversion function
+  // names which are visible.
+  llvm::SmallSet<DeclarationName, 8> ConstructorNameSet, ConversionNameSet;
+
   for (auto &Lookup : *DC->buildLookup()) {
-    if (Lookup.second.hasExternalDecls() ||
-        NeedToReconcileExternalVisibleStorage) {
-      // We don't know for sure what declarations are found by this name,
-      // because the external source might have a different set from the set
-      // that are in the lookup map, and we can't update it now without
-      // risking invalidating our lookup iterator. So add it to a queue to
-      // deal with later.
-      ExternalNames.push_back(Lookup.first);
+    auto &Name = Lookup.first;
+    auto &Result = Lookup.second;
+
+    // If there are no local declarations in our lookup result, we don't
+    // need to write an entry for the name at all unless we're rewriting
+    // the decl context. If we can't write out a lookup set without
+    // performing more deserialization, just skip this entry.
+    if (isLookupResultExternal(Result, DC) && !isRewritten(cast<Decl>(DC)) &&
+        isLookupResultEntirelyExternal(Result, DC))
       continue;
-    }
 
-    AddLookupResult(Lookup.first, Lookup.second.getLookupResult());
-  }
+    // We also skip empty results. If any of the results could be external and
+    // the currently available results are empty, then all of the results are
+    // external and we skip it above. So the only way we get here with an empty
+    // results is when no results could have been external *and* we have
+    // external results.
+    //
+    // FIXME: While we might want to start emitting on-disk entries for negative
+    // lookups into a decl context as an optimization, today we *have* to skip
+    // them because there are names with empty lookup results in decl contexts
+    // which we can't emit in any stable ordering: we lookup constructors and
+    // conversion functions in the enclosing namespace scope creating empty
+    // results for them. This in almost certainly a bug in Clang's name lookup,
+    // but that is likely to be hard or impossible to fix and so we tolerate it
+    // here by omitting lookups with empty results.
+    if (Lookup.second.getLookupResult().empty())
+      continue;
 
-  // Add the names we needed to defer. Note, this shouldn't add any new decls
-  // to the list we need to serialize: any new declarations we find here should
-  // be imported from an external source.
-  // FIXME: What if the external source isn't an ASTReader?
-  for (const auto &Name : ExternalNames)
-    AddLookupResult(Name, DC->lookup(Name));
-}
-
-void ASTWriter::AddUpdatedDeclContext(const DeclContext *DC) {
-  if (UpdatedDeclContexts.insert(DC).second && WritingAST) {
-    // Ensure we emit all the visible declarations.
-    visitLocalLookupResults(DC, DC->NeedToReconcileExternalVisibleStorage,
-                            [&](DeclarationName Name,
-                                DeclContext::lookup_const_result Result) {
-      for (auto *Decl : Result)
-        GetDeclRef(getDeclForLocalLookup(Decl));
-    });
+    switch (Lookup.first.getNameKind()) {
+    default:
+      Names.push_back(Lookup.first);
+      break;
+
+    case DeclarationName::CXXConstructorName:
+      assert(isa<CXXRecordDecl>(DC) &&
+             "Cannot have a constructor name outside of a class!");
+      ConstructorNameSet.insert(Name);
+      break;
+
+    case DeclarationName::CXXConversionFunctionName:
+      assert(isa<CXXRecordDecl>(DC) &&
+             "Cannot have a conversion function name outside of a class!");
+      ConversionNameSet.insert(Name);
+      break;
+    }
   }
-}
 
-uint32_t
-ASTWriter::GenerateNameLookupTable(const DeclContext *DC,
-                                   llvm::SmallVectorImpl<char> &LookupTable) {
-  assert(!DC->LookupPtr.getInt() && "must call buildLookups first");
+  // Sort the names into a stable order.
+  std::sort(Names.begin(), Names.end());
+
+  if (auto *D = dyn_cast<CXXRecordDecl>(DC)) {
+    // We need to establish an ordering of constructor and conversion function
+    // names, and they don't have an intrinsic ordering.
+
+    // First we try the easy case by forming the current context's constructor
+    // name and adding that name first. This is a very useful optimization to
+    // avoid walking the lexical declarations in many cases, and it also
+    // handles the only case where a constructor name can come from some other
+    // lexical context -- when that name is an implicit constructor merged from
+    // another declaration in the redecl chain. Any non-implicit constructor or
+    // conversion function which doesn't occur in all the lexical contexts
+    // would be an ODR violation.
+    auto ImplicitCtorName = Context->DeclarationNames.getCXXConstructorName(
+        Context->getCanonicalType(Context->getRecordType(D)));
+    if (ConstructorNameSet.erase(ImplicitCtorName))
+      Names.push_back(ImplicitCtorName);
+
+    // If we still have constructors or conversion functions, we walk all the
+    // names in the decl and add the constructors and conversion functions
+    // which are visible in the order they lexically occur within the context.
+    if (!ConstructorNameSet.empty() || !ConversionNameSet.empty())
+      for (Decl *ChildD : cast<CXXRecordDecl>(DC)->decls())
+        if (auto *ChildND = dyn_cast<NamedDecl>(ChildD)) {
+          auto Name = ChildND->getDeclName();
+          switch (Name.getNameKind()) {
+          default:
+            continue;
+
+          case DeclarationName::CXXConstructorName:
+            if (ConstructorNameSet.erase(Name))
+              Names.push_back(Name);
+            break;
+
+          case DeclarationName::CXXConversionFunctionName:
+            if (ConversionNameSet.erase(Name))
+              Names.push_back(Name);
+            break;
+          }
 
-  llvm::OnDiskChainedHashTableGenerator<ASTDeclContextNameLookupTrait>
-      Generator;
-  ASTDeclContextNameLookupTrait Trait(*this);
+          if (ConstructorNameSet.empty() && ConversionNameSet.empty())
+            break;
+        }
 
-  // Create the on-disk hash table representation.
-  DeclarationName ConstructorName;
-  DeclarationName ConversionName;
+    assert(ConstructorNameSet.empty() && "Failed to find all of the visible "
+                                         "constructors by walking all the "
+                                         "lexical members of the context.");
+    assert(ConversionNameSet.empty() && "Failed to find all of the visible "
+                                        "conversion functions by walking all "
+                                        "the lexical members of the context.");
+  }
+
+  // Next we need to do a lookup with each name into this decl context to fully
+  // populate any results from external sources. We don't actually use the
+  // results of these lookups because we only want to use the results after all
+  // results have been loaded and the pointers into them will be stable.
+  for (auto &Name : Names)
+    DC->lookup(Name);
+
+  // Now we need to insert the results for each name into the hash table. For
+  // constructor names and conversion function names, we actually need to merge
+  // all of the results for them into one list of results each and insert
+  // those.
   SmallVector<NamedDecl *, 8> ConstructorDecls;
-  SmallVector<NamedDecl *, 4> ConversionDecls;
+  SmallVector<NamedDecl *, 8> ConversionDecls;
 
-  visitLocalLookupResults(DC, DC->NeedToReconcileExternalVisibleStorage,
-                          [&](DeclarationName Name,
-                              DeclContext::lookup_result Result) {
-    if (Result.empty())
-      return;
+  // Now loop over the names, either inserting them or appending for the two
+  // special cases.
+  for (auto &Name : Names) {
+    DeclContext::lookup_result Result = DC->noload_lookup(Name);
 
-    // Different DeclarationName values of certain kinds are mapped to
-    // identical serialized keys, because we don't want to use type
-    // identifiers in the keys (since type ids are local to the module).
     switch (Name.getNameKind()) {
+    default:
+      Generator.insert(Name, Result, Trait);
+      break;
+
     case DeclarationName::CXXConstructorName:
-      // There may be different CXXConstructorName DeclarationName values
-      // in a DeclContext because a UsingDecl that inherits constructors
-      // has the DeclarationName of the inherited constructors.
-      if (!ConstructorName)
-        ConstructorName = Name;
       ConstructorDecls.append(Result.begin(), Result.end());
-      return;
+      break;
 
     case DeclarationName::CXXConversionFunctionName:
-      if (!ConversionName)
-        ConversionName = Name;
       ConversionDecls.append(Result.begin(), Result.end());
-      return;
-
-    default:
       break;
     }
-
-    Generator.insert(Name, Result, Trait);
-  });
-
-  // Add the constructors.
-  if (!ConstructorDecls.empty()) {
-    Generator.insert(ConstructorName,
-                     DeclContext::lookup_result(ConstructorDecls.begin(),
-                                                ConstructorDecls.end()),
-                     Trait);
   }
 
-  // Add the conversion functions.
-  if (!ConversionDecls.empty()) {
-    Generator.insert(ConversionName,
-                     DeclContext::lookup_result(ConversionDecls.begin(),
-                                                ConversionDecls.end()),
-                     Trait);
-  }
+  // Handle our two special cases if we ended up having any. We arbitrarily use
+  // the first declaration's name here because the name itself isn't part of
+  // the key, only the kind of name is used.
+  if (!ConstructorDecls.empty())
+    Generator.insert(ConstructorDecls.front()->getDeclName(),
+                     DeclContext::lookup_result(ConstructorDecls), Trait);
+  if (!ConversionDecls.empty())
+    Generator.insert(ConversionDecls.front()->getDeclName(),
+                     DeclContext::lookup_result(ConversionDecls), Trait);
 
   // Create the on-disk hash table in a buffer.
   llvm::raw_svector_ostream Out(LookupTable);
@@ -3798,9 +3625,8 @@ uint64_t ASTWriter::WriteDeclContextVisibleBlock(ASTContext &Context,
   if (DC->getPrimaryContext() != DC)
     return 0;
 
-  // Since there is no name lookup into functions or methods, don't bother to
-  // build a visible-declarations table for these entities.
-  if (DC->isFunctionOrMethod())
+  // Skip contexts which don't support name lookup.
+  if (!DC->isLookupContext())
     return 0;
 
   // If not in C++, we perform name lookup for the translation unit via the
@@ -3827,7 +3653,7 @@ uint64_t ASTWriter::WriteDeclContextVisibleBlock(ASTContext &Context,
   Record.push_back(DECL_CONTEXT_VISIBLE);
   Record.push_back(BucketOffset);
   Stream.EmitRecordWithBlob(DeclContextVisibleLookupAbbrev, Record,
-                            LookupTable.str());
+                            LookupTable);
   ++NumVisibleDeclContexts;
   return Offset;
 }
@@ -3852,7 +3678,7 @@ void ASTWriter::WriteDeclContextVisibleUpdate(const DeclContext *DC) {
   Record.push_back(UPDATE_VISIBLE);
   Record.push_back(getDeclID(cast<Decl>(DC)));
   Record.push_back(BucketOffset);
-  Stream.EmitRecordWithBlob(UpdateVisibleAbbrev, Record, LookupTable.str());
+  Stream.EmitRecordWithBlob(UpdateVisibleAbbrev, Record, LookupTable);
 }
 
 /// \brief Write an FP_PRAGMA_OPTIONS block for the given FPOptions.
@@ -3902,18 +3728,6 @@ void ASTWriter::WriteRedeclarations() {
       }
     }
 
-    if (!First->isFromASTFile() && Chain) {
-      Decl *FirstFromAST = MostRecent;
-      for (Decl *Prev = MostRecent; Prev; Prev = Prev->getPreviousDecl()) {
-        if (Prev->isFromASTFile())
-          FirstFromAST = Prev;
-      }
-
-      // FIXME: Do we need to do this for the first declaration from each
-      // redeclaration chain that was merged into this one?
-      Chain->MergedDecls[FirstFromAST].push_back(getDeclID(First));
-    }
-
     LocalRedeclChains[Offset] = Size;
     
     // Reverse the set of local redeclarations, so that we store them in
@@ -4008,25 +3822,6 @@ void ASTWriter::WriteObjCCategories() {
   Stream.EmitRecord(OBJC_CATEGORIES, Categories);
 }
 
-void ASTWriter::WriteMergedDecls() {
-  if (!Chain || Chain->MergedDecls.empty())
-    return;
-  
-  RecordData Record;
-  for (ASTReader::MergedDeclsMap::iterator I = Chain->MergedDecls.begin(),
-                                        IEnd = Chain->MergedDecls.end();
-       I != IEnd; ++I) {
-    DeclID CanonID = I->first->isFromASTFile()? I->first->getGlobalID()
-                                              : GetDeclRef(I->first);
-    assert(CanonID && "Merged declaration not known?");
-    
-    Record.push_back(CanonID);
-    Record.push_back(I->second.size());
-    Record.append(I->second.begin(), I->second.end());
-  }
-  Stream.EmitRecord(MERGED_DECLARATIONS, Record);
-}
-
 void ASTWriter::WriteLateParsedTemplates(Sema &SemaRef) {
   Sema::LateParsedTemplateMapT &LPTMap = SemaRef.LateParsedTemplateMap;
 
@@ -4034,11 +3829,10 @@ void ASTWriter::WriteLateParsedTemplates(Sema &SemaRef) {
     return;
 
   RecordData Record;
-  for (Sema::LateParsedTemplateMapT::iterator It = LPTMap.begin(),
-                                              ItEnd = LPTMap.end();
-       It != ItEnd; ++It) {
-    LateParsedTemplate *LPT = It->second;
-    AddDeclRef(It->first, Record);
+  for (auto LPTMapEntry : LPTMap) {
+    const FunctionDecl *FD = LPTMapEntry.first;
+    LateParsedTemplate *LPT = LPTMapEntry.second;
+    AddDeclRef(FD, Record);
     AddDeclRef(LPT->D, Record);
     Record.push_back(LPT->Toks.size());
 
@@ -4175,7 +3969,8 @@ ASTWriter::ASTWriter(llvm::BitstreamWriter &Stream)
       FirstSelectorID(NUM_PREDEF_SELECTOR_IDS), NextSelectorID(FirstSelectorID),
       CollectedStmts(&StmtsToEmit), NumStatements(0), NumMacros(0),
       NumLexicalDeclContexts(0), NumVisibleDeclContexts(0),
-      NextCXXBaseSpecifiersID(1), TypeExtQualAbbrev(0),
+      NextCXXBaseSpecifiersID(1), NextCXXCtorInitializersID(1),
+      TypeExtQualAbbrev(0),
       TypeFunctionProtoAbbrev(0), DeclParmVarAbbrev(0),
       DeclContextLexicalAbbrev(0), DeclContextVisibleLookupAbbrev(0),
       UpdateVisibleAbbrev(0), DeclRecordAbbrev(0), DeclTypedefAbbrev(0),
@@ -4188,6 +3983,11 @@ ASTWriter::~ASTWriter() {
   llvm::DeleteContainerSeconds(FileDeclIDs);
 }
 
+const LangOptions &ASTWriter::getLangOpts() const {
+  assert(WritingAST && "can't determine lang opts when not writing AST");
+  return Context->getLangOpts();
+}
+
 void ASTWriter::WriteAST(Sema &SemaRef,
                          const std::string &OutputFile,
                          Module *WritingModule, StringRef isysroot,
@@ -4258,47 +4058,8 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
     DeclIDs[Context.ObjCInstanceTypeDecl] = PREDEF_DECL_OBJC_INSTANCETYPE_ID;
   if (Context.BuiltinVaListDecl)
     DeclIDs[Context.getBuiltinVaListDecl()] = PREDEF_DECL_BUILTIN_VA_LIST_ID;
-
-  if (!Chain) {
-    // Make sure that we emit IdentifierInfos (and any attached
-    // declarations) for builtins. We don't need to do this when we're
-    // emitting chained PCH files, because all of the builtins will be
-    // in the original PCH file.
-    // FIXME: Modules won't like this at all.
-    IdentifierTable &Table = PP.getIdentifierTable();
-    SmallVector<const char *, 32> BuiltinNames;
-    if (!Context.getLangOpts().NoBuiltin) {
-      Context.BuiltinInfo.GetBuiltinNames(BuiltinNames);
-    }
-    for (unsigned I = 0, N = BuiltinNames.size(); I != N; ++I)
-      getIdentifierRef(&Table.get(BuiltinNames[I]));
-  }
-
-  // If there are any out-of-date identifiers, bring them up to date.
-  if (ExternalPreprocessorSource *ExtSource = PP.getExternalSource()) {
-    // Find out-of-date identifiers.
-    SmallVector<IdentifierInfo *, 4> OutOfDate;
-    for (IdentifierTable::iterator ID = PP.getIdentifierTable().begin(),
-                                IDEnd = PP.getIdentifierTable().end();
-         ID != IDEnd; ++ID) {
-      if (ID->second->isOutOfDate())
-        OutOfDate.push_back(ID->second);
-    }
-
-    // Update the out-of-date identifiers.
-    for (unsigned I = 0, N = OutOfDate.size(); I != N; ++I) {
-      ExtSource->updateOutOfDateIdentifier(*OutOfDate[I]);
-    }
-  }
-
-  // If we saw any DeclContext updates before we started writing the AST file,
-  // make sure all visible decls in those DeclContexts are written out.
-  if (!UpdatedDeclContexts.empty()) {
-    auto OldUpdatedDeclContexts = std::move(UpdatedDeclContexts);
-    UpdatedDeclContexts.clear();
-    for (auto *DC : OldUpdatedDeclContexts)
-      AddUpdatedDeclContext(DC);
-  }
+  if (Context.ExternCContext)
+    DeclIDs[Context.ExternCContext] = PREDEF_DECL_EXTERN_C_CONTEXT_ID;
 
   // Build a record containing all of the tentative definitions in this file, in
   // TentativeDefinitions order.  Generally, this record will be empty for
@@ -4322,31 +4083,15 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
   // entire table, since later PCH files in a PCH chain are only interested in
   // the results at the end of the chain.
   RecordData WeakUndeclaredIdentifiers;
-  if (!SemaRef.WeakUndeclaredIdentifiers.empty()) {
-    for (llvm::DenseMap<IdentifierInfo*,WeakInfo>::iterator
-         I = SemaRef.WeakUndeclaredIdentifiers.begin(),
-         E = SemaRef.WeakUndeclaredIdentifiers.end(); I != E; ++I) {
-      AddIdentifierRef(I->first, WeakUndeclaredIdentifiers);
-      AddIdentifierRef(I->second.getAlias(), WeakUndeclaredIdentifiers);
-      AddSourceLocation(I->second.getLocation(), WeakUndeclaredIdentifiers);
-      WeakUndeclaredIdentifiers.push_back(I->second.getUsed());
-    }
+  for (auto &WeakUndeclaredIdentifier : SemaRef.WeakUndeclaredIdentifiers) {
+    IdentifierInfo *II = WeakUndeclaredIdentifier.first;
+    WeakInfo &WI = WeakUndeclaredIdentifier.second;
+    AddIdentifierRef(II, WeakUndeclaredIdentifiers);
+    AddIdentifierRef(WI.getAlias(), WeakUndeclaredIdentifiers);
+    AddSourceLocation(WI.getLocation(), WeakUndeclaredIdentifiers);
+    WeakUndeclaredIdentifiers.push_back(WI.getUsed());
   }
 
-  // Build a record containing all of the locally-scoped extern "C"
-  // declarations in this header file. Generally, this record will be
-  // empty.
-  RecordData LocallyScopedExternCDecls;
-  // FIXME: This is filling in the AST file in densemap order which is
-  // nondeterminstic!
-  for (llvm::DenseMap<DeclarationName, NamedDecl *>::iterator
-         TD = SemaRef.LocallyScopedExternCDecls.begin(),
-         TDEnd = SemaRef.LocallyScopedExternCDecls.end();
-       TD != TDEnd; ++TD) {
-    if (!TD->second->isFromASTFile())
-      AddDeclRef(TD->second, LocallyScopedExternCDecls);
-  }
-  
   // Build a record containing all of the ext_vector declarations.
   RecordData ExtVectorDecls;
   AddLazyVectorDecls(*this, SemaRef.ExtVectorDecls, ExtVectorDecls);
@@ -4366,10 +4111,6 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
   for (const TypedefNameDecl *TD : SemaRef.UnusedLocalTypedefNameCandidates)
     AddDeclRef(TD, UnusedLocalTypedefNameCandidates);
 
-  // Build a record containing all of dynamic classes declarations.
-  RecordData DynamicClasses;
-  AddLazyVectorDecls(*this, SemaRef.DynamicClasses, DynamicClasses);
-
   // Build a record containing all of pending implicit instantiations.
   RecordData PendingInstantiations;
   for (std::deque<Sema::PendingImplicitInstantiation>::iterator
@@ -4414,6 +4155,20 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
     AddSourceLocation(I->second, UndefinedButUsed);
   }
 
+  // Build a record containing all delete-expressions that we would like to
+  // analyze later in AST.
+  RecordData DeleteExprsToAnalyze;
+
+  for (const auto &DeleteExprsInfo :
+       SemaRef.getMismatchingDeleteExpressions()) {
+    AddDeclRef(DeleteExprsInfo.first, DeleteExprsToAnalyze);
+    DeleteExprsToAnalyze.push_back(DeleteExprsInfo.second.size());
+    for (const auto &DeleteLoc : DeleteExprsInfo.second) {
+      AddSourceLocation(DeleteLoc.first, DeleteExprsToAnalyze);
+      DeleteExprsToAnalyze.push_back(DeleteLoc.second);
+    }
+  }
+
   // Write the control block
   WriteControlBlock(PP, Context, isysroot, OutputFile);
 
@@ -4443,7 +4198,7 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
   Record.clear();
   Record.push_back(TU_UPDATE_LEXICAL);
   Stream.EmitRecordWithBlob(TuUpdateLexicalAbbrev, Record,
-                            data(NewGlobalDecls));
+                            bytes(NewGlobalDecls));
   
   // And a visible updates block for the translation unit.
   Abv = new llvm::BitCodeAbbrev();
@@ -4453,6 +4208,10 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
   Abv->Add(llvm::BitCodeAbbrevOp(llvm::BitCodeAbbrevOp::Blob));
   UpdateVisibleAbbrev = Stream.EmitAbbrev(Abv);
   WriteDeclContextVisibleUpdate(TU);
+
+  // If we have any extern "C" names, write out a visible update for them.
+  if (Context.ExternCContext)
+    WriteDeclContextVisibleUpdate(Context.ExternCContext);
   
   // If the translation unit has an anonymous namespace, and we don't already
   // have an update block for it, write it as an update block.
@@ -4485,16 +4244,21 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
 
   // Make sure all decls associated with an identifier are registered for
   // serialization.
+  llvm::SmallVector<const IdentifierInfo*, 256> IIs;
   for (IdentifierTable::iterator ID = PP.getIdentifierTable().begin(),
                               IDEnd = PP.getIdentifierTable().end();
        ID != IDEnd; ++ID) {
     const IdentifierInfo *II = ID->second;
-    if (!Chain || !II->isFromAST() || II->hasChangedSinceDeserialization()) {
-      for (IdentifierResolver::iterator D = SemaRef.IdResolver.begin(II),
-                                     DEnd = SemaRef.IdResolver.end();
-           D != DEnd; ++D) {
-        GetDeclRef(*D);
-      }
+    if (!Chain || !II->isFromAST() || II->hasChangedSinceDeserialization())
+      IIs.push_back(II);
+  }
+  // Sort the identifiers to visit based on their name.
+  std::sort(IIs.begin(), IIs.end(), llvm::less_ptr<IdentifierInfo>());
+  for (const IdentifierInfo *II : IIs) {
+    for (IdentifierResolver::iterator D = SemaRef.IdResolver.begin(II),
+                                   DEnd = SemaRef.IdResolver.end();
+         D != DEnd; ++D) {
+      GetDeclRef(*D);
     }
   }
 
@@ -4601,6 +4365,7 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
   if (!DeclUpdatesOffsetsRecord.empty())
     Stream.EmitRecord(DECL_UPDATE_OFFSETS, DeclUpdatesOffsetsRecord);
   WriteCXXBaseSpecifiersOffsets();
+  WriteCXXCtorInitializersOffsets();
   WriteFileDeclIDsMap();
   WriteSourceManagerBlock(Context.getSourceManager(), PP);
 
@@ -4637,11 +4402,6 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
     Stream.EmitRecord(WEAK_UNDECLARED_IDENTIFIERS,
                       WeakUndeclaredIdentifiers);
 
-  // Write the record containing locally-scoped extern "C" definitions.
-  if (!LocallyScopedExternCDecls.empty())
-    Stream.EmitRecord(LOCALLY_SCOPED_EXTERN_C_DECLS,
-                      LocallyScopedExternCDecls);
-
   // Write the record containing ext_vector type names.
   if (!ExtVectorDecls.empty())
     Stream.EmitRecord(EXT_VECTOR_DECLS, ExtVectorDecls);
@@ -4650,10 +4410,6 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
   if (!VTableUses.empty())
     Stream.EmitRecord(VTABLE_USES, VTableUses);
 
-  // Write the record containing dynamic classes declarations.
-  if (!DynamicClasses.empty())
-    Stream.EmitRecord(DYNAMIC_CLASSES, DynamicClasses);
-
   // Write the record containing potentially unused local typedefs.
   if (!UnusedLocalTypedefNameCandidates.empty())
     Stream.EmitRecord(UNUSED_LOCAL_TYPEDEF_NAME_CANDIDATES,
@@ -4682,7 +4438,10 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
   // Write the undefined internal functions and variables, and inline functions.
   if (!UndefinedButUsed.empty())
     Stream.EmitRecord(UNDEFINED_BUT_USED, UndefinedButUsed);
-  
+
+  if (!DeleteExprsToAnalyze.empty())
+    Stream.EmitRecord(DELETE_EXPRS_TO_ANALYZE, DeleteExprsToAnalyze);
+
   // Write the visible updates to DeclContexts.
   for (auto *DC : UpdatedDeclContexts)
     WriteDeclContextVisibleUpdate(DC);
@@ -4720,7 +4479,7 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
         // FIXME: If the module has macros imported then later has declarations
         // imported, this location won't be the right one as a location for the
         // declaration imports.
-        AddSourceLocation(Import.M->MacroVisibilityLoc, ImportedModules);
+        AddSourceLocation(PP.getModuleImportLoc(Import.M), ImportedModules);
       }
 
       Stream.EmitRecord(IMPORTED_MODULES, ImportedModules);
@@ -4729,7 +4488,6 @@ void ASTWriter::WriteASTCore(Sema &SemaRef,
 
   WriteDeclReplacementsBlock();
   WriteRedeclarations();
-  WriteMergedDecls();
   WriteObjCCategories();
   WriteLateParsedTemplates(SemaRef);
   if(!WritingModule)
@@ -4784,7 +4542,7 @@ void ASTWriter::WriteDeclUpdatesBlocks(RecordDataImpl &OffsetsRecord) {
 
       case UPD_CXX_INSTANTIATED_CLASS_DEFINITION: {
         auto *RD = cast<CXXRecordDecl>(D);
-        AddUpdatedDeclContext(RD->getPrimaryContext());
+        UpdatedDeclContexts.insert(RD->getPrimaryContext());
         AddCXXDefinitionData(RD, Record);
         Record.push_back(WriteDeclContextLexicalBlock(
             *Context, const_cast<CXXRecordDecl *>(RD)));
@@ -4828,6 +4586,10 @@ void ASTWriter::WriteDeclUpdatesBlocks(RecordDataImpl &OffsetsRecord) {
         break;
       }
 
+      case UPD_CXX_RESOLVED_DTOR_DELETE:
+        AddDeclRef(Update.getDecl(), Record);
+        break;
+
       case UPD_CXX_RESOLVED_EXCEPTION_SPEC:
         addExceptionSpec(
             *this,
@@ -4846,10 +4608,15 @@ void ASTWriter::WriteDeclUpdatesBlocks(RecordDataImpl &OffsetsRecord) {
       case UPD_STATIC_LOCAL_NUMBER:
         Record.push_back(Update.getNumber());
         break;
+
       case UPD_DECL_MARKED_OPENMP_THREADPRIVATE:
         AddSourceRange(D->getAttr<OMPThreadPrivateDeclAttr>()->getRange(),
                        Record);
         break;
+
+      case UPD_DECL_EXPORTED:
+        Record.push_back(getSubmoduleID(Update.getModule()));
+        break;
       }
     }
 
@@ -4859,8 +4626,6 @@ void ASTWriter::WriteDeclUpdatesBlocks(RecordDataImpl &OffsetsRecord) {
       Record.push_back(Def->isInlined());
       AddSourceLocation(Def->getInnerLocStart(), Record);
       AddFunctionDefinition(Def, Record);
-      if (auto *DD = dyn_cast<CXXDestructorDecl>(Def))
-        Record.push_back(GetDeclRef(DD->getOperatorDelete()));
     }
 
     OffsetsRecord.push_back(GetDeclRef(D));
@@ -4868,11 +4633,7 @@ void ASTWriter::WriteDeclUpdatesBlocks(RecordDataImpl &OffsetsRecord) {
 
     Stream.EmitRecord(DECL_UPDATES, Record);
 
-    // Flush any statements that were written as part of this update record.
-    FlushStmts();
-
-    // Flush C++ base specifiers, if there are any.
-    FlushCXXBaseSpecifiers();
+    FlushPendingAfterDecl();
   }
 }
 
@@ -4953,8 +4714,7 @@ MacroID ASTWriter::getMacroID(MacroInfo *MI) {
 }
 
 uint64_t ASTWriter::getMacroDirectivesOffset(const IdentifierInfo *Name) {
-  assert(IdentMacroDirectivesOffsetMap[Name] && "not set!");
-  return IdentMacroDirectivesOffsetMap[Name];
+  return IdentMacroDirectivesOffsetMap.lookup(Name);
 }
 
 void ASTWriter::AddSelectorRef(const Selector SelRef, RecordDataImpl &Record) {
@@ -4984,8 +4744,16 @@ void ASTWriter::AddCXXTemporary(const CXXTemporary *Temp, RecordDataImpl &Record
   AddDeclRef(Temp->getDestructor(), Record);
 }
 
+void ASTWriter::AddCXXCtorInitializersRef(ArrayRef<CXXCtorInitializer *> Inits,
+                                          RecordDataImpl &Record) {
+  assert(!Inits.empty() && "Empty ctor initializer sets are not recorded");
+  CXXCtorInitializersToWrite.push_back(
+      QueuedCXXCtorInitializers(NextCXXCtorInitializersID, Inits));
+  Record.push_back(NextCXXCtorInitializersID++);
+}
+
 void ASTWriter::AddCXXBaseSpecifiersRef(CXXBaseSpecifier const *Bases,
-                                      CXXBaseSpecifier const *BasesEnd,
+                                        CXXBaseSpecifier const *BasesEnd,
                                         RecordDataImpl &Record) {
   assert(Bases != BasesEnd && "Empty base-specifier sets are not recorded");
   CXXBaseSpecifiersToWrite.push_back(
@@ -5060,46 +4828,40 @@ void ASTWriter::AddTypeRef(QualType T, RecordDataImpl &Record) {
   Record.push_back(GetOrCreateTypeID(T));
 }
 
-TypeID ASTWriter::GetOrCreateTypeID( QualType T) {
-  assert(Context);
-  return MakeTypeID(*Context, T,
-              std::bind1st(std::mem_fun(&ASTWriter::GetOrCreateTypeIdx), this));
-}
-
-TypeID ASTWriter::getTypeID(QualType T) const {
+TypeID ASTWriter::GetOrCreateTypeID(QualType T) {
   assert(Context);
-  return MakeTypeID(*Context, T,
-              std::bind1st(std::mem_fun(&ASTWriter::getTypeIdx), this));
-}
+  return MakeTypeID(*Context, T, [&](QualType T) -> TypeIdx {
+    if (T.isNull())
+      return TypeIdx();
+    assert(!T.getLocalFastQualifiers());
 
-TypeIdx ASTWriter::GetOrCreateTypeIdx(QualType T) {
-  if (T.isNull())
-    return TypeIdx();
-  assert(!T.getLocalFastQualifiers());
+    TypeIdx &Idx = TypeIdxs[T];
+    if (Idx.getIndex() == 0) {
+      if (DoneWritingDeclsAndTypes) {
+        assert(0 && "New type seen after serializing all the types to emit!");
+        return TypeIdx();
+      }
 
-  TypeIdx &Idx = TypeIdxs[T];
-  if (Idx.getIndex() == 0) {
-    if (DoneWritingDeclsAndTypes) {
-      assert(0 && "New type seen after serializing all the types to emit!");
-      return TypeIdx();
+      // We haven't seen this type before. Assign it a new ID and put it
+      // into the queue of types to emit.
+      Idx = TypeIdx(NextTypeID++);
+      DeclTypesToEmit.push(T);
     }
-
-    // We haven't seen this type before. Assign it a new ID and put it
-    // into the queue of types to emit.
-    Idx = TypeIdx(NextTypeID++);
-    DeclTypesToEmit.push(T);
-  }
-  return Idx;
+    return Idx;
+  });
 }
 
-TypeIdx ASTWriter::getTypeIdx(QualType T) const {
-  if (T.isNull())
-    return TypeIdx();
-  assert(!T.getLocalFastQualifiers());
+TypeID ASTWriter::getTypeID(QualType T) const {
+  assert(Context);
+  return MakeTypeID(*Context, T, [&](QualType T) -> TypeIdx {
+    if (T.isNull())
+      return TypeIdx();
+    assert(!T.getLocalFastQualifiers());
 
-  TypeIdxMap::const_iterator I = TypeIdxs.find(T);
-  assert(I != TypeIdxs.end() && "Type not emitted!");
-  return I->second;
+    TypeIdxMap::const_iterator I = TypeIdxs.find(T);
+    assert(I != TypeIdxs.end() && "Type not emitted!");
+    return I->second;
+  });
 }
 
 void ASTWriter::AddDeclRef(const Decl *D, RecordDataImpl &Record) {
@@ -5234,13 +4996,10 @@ unsigned ASTWriter::getAnonymousDeclarationNumber(const NamedDecl *D) {
   // already done so.
   auto It = AnonymousDeclarationNumbers.find(D);
   if (It == AnonymousDeclarationNumbers.end()) {
-    unsigned Index = 0;
-    for (Decl *LexicalD : D->getLexicalDeclContext()->decls()) {
-      auto *ND = dyn_cast<NamedDecl>(LexicalD);
-      if (!ND || !needsAnonymousDeclarationNumber(ND))
-        continue;
-      AnonymousDeclarationNumbers[ND] = Index++;
-    }
+    auto *DC = D->getLexicalDeclContext();
+    numberAnonymousDeclsWithin(DC, [&](const NamedDecl *ND, unsigned Number) {
+      AnonymousDeclarationNumbers[ND] = Number;
+    });
 
     It = AnonymousDeclarationNumbers.find(D);
     assert(It != AnonymousDeclarationNumbers.end() &&
@@ -5555,7 +5314,8 @@ void ASTWriter::AddCXXBaseSpecifier(const CXXBaseSpecifier &Base,
 
 void ASTWriter::FlushCXXBaseSpecifiers() {
   RecordData Record;
-  for (unsigned I = 0, N = CXXBaseSpecifiersToWrite.size(); I != N; ++I) {
+  unsigned N = CXXBaseSpecifiersToWrite.size();
+  for (unsigned I = 0; I != N; ++I) {
     Record.clear();
     
     // Record the offset of this base-specifier set.
@@ -5579,6 +5339,8 @@ void ASTWriter::FlushCXXBaseSpecifiers() {
     FlushStmts();
   }
 
+  assert(N == CXXBaseSpecifiersToWrite.size() &&
+         "added more base specifiers while writing base specifiers");
   CXXBaseSpecifiersToWrite.clear();
 }
 
@@ -5620,6 +5382,36 @@ void ASTWriter::AddCXXCtorInitializers(
   }
 }
 
+void ASTWriter::FlushCXXCtorInitializers() {
+  RecordData Record;
+
+  unsigned N = CXXCtorInitializersToWrite.size();
+  (void)N; // Silence unused warning in non-assert builds.
+  for (auto &Init : CXXCtorInitializersToWrite) {
+    Record.clear();
+
+    // Record the offset of this mem-initializer list.
+    unsigned Index = Init.ID - 1;
+    if (Index == CXXCtorInitializersOffsets.size())
+      CXXCtorInitializersOffsets.push_back(Stream.GetCurrentBitNo());
+    else {
+      if (Index > CXXCtorInitializersOffsets.size())
+        CXXCtorInitializersOffsets.resize(Index + 1);
+      CXXCtorInitializersOffsets[Index] = Stream.GetCurrentBitNo();
+    }
+
+    AddCXXCtorInitializers(Init.Inits.data(), Init.Inits.size(), Record);
+    Stream.EmitRecord(serialization::DECL_CXX_CTOR_INITIALIZERS, Record);
+
+    // Flush any expressions that were written as part of the initializers.
+    FlushStmts();
+  }
+
+  assert(N == CXXCtorInitializersToWrite.size() &&
+         "added more ctor initializers while writing ctor initializers");
+  CXXCtorInitializersToWrite.clear();
+}
+
 void ASTWriter::AddCXXDefinitionData(const CXXRecordDecl *D, RecordDataImpl &Record) {
   auto &Data = D->data();
   Record.push_back(Data.IsLambda);
@@ -5725,6 +5517,8 @@ void ASTWriter::ReaderInitialized(ASTReader *Reader) {
 
   Chain = Reader;
 
+  // Note, this will get called multiple times, once one the reader starts up
+  // and again each time it's done reading a PCH or module.
   FirstDeclID = NUM_PREDEF_DECL_IDS + Chain->getTotalNumDecls();
   FirstTypeID = NUM_PREDEF_TYPE_IDS + Chain->getTotalNumTypes();
   FirstIdentID = NUM_PREDEF_IDENT_IDS + Chain->getTotalNumIdentifiers();
@@ -5772,7 +5566,7 @@ void ASTWriter::SelectorRead(SelectorID ID, Selector S) {
 }
 
 void ASTWriter::MacroDefinitionRead(serialization::PreprocessedEntityID ID,
-                                    MacroDefinition *MD) {
+                                    MacroDefinitionRecord *MD) {
   assert(MacroDefinitions.find(MD) == MacroDefinitions.end());
   MacroDefinitions[MD] = ID;
 }
@@ -5809,7 +5603,7 @@ void ASTWriter::AddedVisibleDecl(const DeclContext *DC, const Decl *D) {
 
   assert(!getDefinitiveDeclContext(DC) && "DeclContext not definitive!");
   assert(!WritingAST && "Already writing the AST!");
-  AddUpdatedDeclContext(DC);
+  UpdatedDeclContexts.insert(DC);
   UpdatingVisibleDecls.push_back(D);
 }
 
@@ -5863,21 +5657,36 @@ void ASTWriter::AddedCXXTemplateSpecialization(const FunctionTemplateDecl *TD,
 }
 
 void ASTWriter::ResolvedExceptionSpec(const FunctionDecl *FD) {
-  assert(!WritingAST && "Already writing the AST!");
-  FD = FD->getCanonicalDecl();
-  if (!FD->isFromASTFile())
-    return; // Not a function declared in PCH and defined outside.
-
-  DeclUpdates[FD].push_back(UPD_CXX_RESOLVED_EXCEPTION_SPEC);
+  assert(!DoneWritingDeclsAndTypes && "Already done writing updates!");
+  if (!Chain) return;
+  Chain->forEachFormerlyCanonicalImportedDecl(FD, [&](const Decl *D) {
+    // If we don't already know the exception specification for this redecl
+    // chain, add an update record for it.
+    if (isUnresolvedExceptionSpec(cast<FunctionDecl>(D)
+                                      ->getType()
+                                      ->castAs<FunctionProtoType>()
+                                      ->getExceptionSpecType()))
+      DeclUpdates[D].push_back(UPD_CXX_RESOLVED_EXCEPTION_SPEC);
+  });
 }
 
 void ASTWriter::DeducedReturnType(const FunctionDecl *FD, QualType ReturnType) {
   assert(!WritingAST && "Already writing the AST!");
-  FD = FD->getCanonicalDecl();
-  if (!FD->isFromASTFile())
-    return; // Not a function declared in PCH and defined outside.
+  if (!Chain) return;
+  Chain->forEachFormerlyCanonicalImportedDecl(FD, [&](const Decl *D) {
+    DeclUpdates[D].push_back(
+        DeclUpdate(UPD_CXX_DEDUCED_RETURN_TYPE, ReturnType));
+  });
+}
 
-  DeclUpdates[FD].push_back(DeclUpdate(UPD_CXX_DEDUCED_RETURN_TYPE, ReturnType));
+void ASTWriter::ResolvedOperatorDelete(const CXXDestructorDecl *DD,
+                                       const FunctionDecl *Delete) {
+  assert(!WritingAST && "Already writing the AST!");
+  assert(Delete && "Not given an operator delete");
+  if (!Chain) return;
+  Chain->forEachFormerlyCanonicalImportedDecl(DD, [&](const Decl *D) {
+    DeclUpdates[D].push_back(DeclUpdate(UPD_CXX_RESOLVED_DTOR_DELETE, Delete));
+  });
 }
 
 void ASTWriter::CompletedImplicitDefinition(const FunctionDecl *D) {
@@ -5894,8 +5703,7 @@ void ASTWriter::FunctionDefinitionInstantiated(const FunctionDecl *D) {
   if (!D->isFromASTFile())
     return;
 
-  DeclUpdates[D].push_back(
-      DeclUpdate(UPD_CXX_ADDED_FUNCTION_DEFINITION));
+  DeclUpdates[D].push_back(DeclUpdate(UPD_CXX_ADDED_FUNCTION_DEFINITION));
 }
 
 void ASTWriter::StaticDataMemberInstantiated(const VarDecl *D) {
@@ -5951,3 +5759,12 @@ void ASTWriter::DeclarationMarkedOpenMPThreadPrivate(const Decl *D) {
 
   DeclUpdates[D].push_back(DeclUpdate(UPD_DECL_MARKED_OPENMP_THREADPRIVATE));
 }
+
+void ASTWriter::RedefinedHiddenDefinition(const NamedDecl *D, Module *M) {
+  assert(!WritingAST && "Already writing the AST!");
+  assert(D->isHidden() && "expected a hidden declaration");
+  if (!D->isFromASTFile())
+    return;
+
+  DeclUpdates[D].push_back(DeclUpdate(UPD_DECL_EXPORTED, M));
+}
diff --git a/lib/Serialization/ASTWriterDecl.cpp b/lib/Serialization/ASTWriterDecl.cpp
index c97c2d8e48d2..0fa4f936a35a 100644
--- a/lib/Serialization/ASTWriterDecl.cpp
+++ b/lib/Serialization/ASTWriterDecl.cpp
@@ -133,11 +133,66 @@ namespace clang {
 
     void AddFunctionDefinition(const FunctionDecl *FD) {
       assert(FD->doesThisDeclarationHaveABody());
-      if (auto *CD = dyn_cast<CXXConstructorDecl>(FD))
-        Writer.AddCXXCtorInitializers(CD->CtorInitializers,
-                                      CD->NumCtorInitializers, Record);
+      if (auto *CD = dyn_cast<CXXConstructorDecl>(FD)) {
+        Record.push_back(CD->NumCtorInitializers);
+        if (CD->NumCtorInitializers)
+          Writer.AddCXXCtorInitializersRef(
+              llvm::makeArrayRef(CD->init_begin(), CD->init_end()), Record);
+      }
       Writer.AddStmt(FD->getBody());
     }
+
+    /// Get the specialization decl from an entry in the specialization list.
+    template <typename EntryType>
+    typename RedeclarableTemplateDecl::SpecEntryTraits<EntryType>::DeclType *
+    getSpecializationDecl(EntryType &T) {
+      return RedeclarableTemplateDecl::SpecEntryTraits<EntryType>::getDecl(&T);
+    }
+
+    /// Get the list of partial specializations from a template's common ptr.
+    template<typename T>
+    decltype(T::PartialSpecializations) &getPartialSpecializations(T *Common) {
+      return Common->PartialSpecializations;
+    }
+    ArrayRef<Decl> getPartialSpecializations(FunctionTemplateDecl::Common *) {
+      return None;
+    }
+
+    template<typename Decl>
+    void AddTemplateSpecializations(Decl *D) {
+      auto *Common = D->getCommonPtr();
+
+      // If we have any lazy specializations, and the external AST source is
+      // our chained AST reader, we can just write out the DeclIDs. Otherwise,
+      // we need to resolve them to actual declarations.
+      if (Writer.Chain != Writer.Context->getExternalSource() &&
+          Common->LazySpecializations) {
+        D->LoadLazySpecializations();
+        assert(!Common->LazySpecializations);
+      }
+
+      auto &Specializations = Common->Specializations;
+      auto &&PartialSpecializations = getPartialSpecializations(Common);
+      ArrayRef<DeclID> LazySpecializations;
+      if (auto *LS = Common->LazySpecializations)
+        LazySpecializations = ArrayRef<DeclID>(LS + 1, LS + 1 + LS[0]);
+
+      Record.push_back(Specializations.size() +
+                       PartialSpecializations.size() +
+                       LazySpecializations.size());
+      for (auto &Entry : Specializations) {
+        auto *D = getSpecializationDecl(Entry);
+        assert(D->isCanonicalDecl() && "non-canonical decl in set");
+        Writer.AddDeclRef(D, Record);
+      }
+      for (auto &Entry : PartialSpecializations) {
+        auto *D = getSpecializationDecl(Entry);
+        assert(D->isCanonicalDecl() && "non-canonical decl in set");
+        Writer.AddDeclRef(D, Record);
+      }
+      for (DeclID ID : LazySpecializations)
+        Record.push_back(ID);
+    }
   };
 }
 
@@ -157,7 +212,7 @@ void ASTDeclWriter::Visit(Decl *D) {
   if (FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) {
     Record.push_back(FD->doesThisDeclarationHaveABody());
     if (FD->doesThisDeclarationHaveABody())
-      Writer.AddStmt(FD->getBody());
+      AddFunctionDefinition(FD);
   }
 }
 
@@ -188,7 +243,7 @@ void ASTDeclWriter::VisitDecl(Decl *D) {
     while (auto *NS = dyn_cast<NamespaceDecl>(DC->getRedeclContext())) {
       if (!NS->isFromASTFile())
         break;
-      Writer.AddUpdatedDeclContext(NS->getPrimaryContext());
+      Writer.UpdatedDeclContexts.insert(NS->getPrimaryContext());
       if (!NS->isInlineNamespace())
         break;
       DC = NS->getParent();
@@ -203,8 +258,9 @@ void ASTDeclWriter::VisitTranslationUnitDecl(TranslationUnitDecl *D) {
 void ASTDeclWriter::VisitNamedDecl(NamedDecl *D) {
   VisitDecl(D);
   Writer.AddDeclarationName(D->getDeclName(), Record);
-  if (needsAnonymousDeclarationNumber(D))
-    Record.push_back(Writer.getAnonymousDeclarationNumber(D));
+  Record.push_back(needsAnonymousDeclarationNumber(D)
+                       ? Writer.getAnonymousDeclarationNumber(D)
+                       : 0);
 }
 
 void ASTDeclWriter::VisitTypeDecl(TypeDecl *D) {
@@ -658,8 +714,10 @@ void ASTDeclWriter::VisitObjCImplementationDecl(ObjCImplementationDecl *D) {
   Writer.AddSourceLocation(D->getIvarRBraceLoc(), Record);
   Record.push_back(D->hasNonZeroConstructors());
   Record.push_back(D->hasDestructors());
-  Writer.AddCXXCtorInitializers(D->IvarInitializers, D->NumIvarInitializers,
-                                Record);
+  Record.push_back(D->NumIvarInitializers);
+  if (D->NumIvarInitializers)
+    Writer.AddCXXCtorInitializersRef(
+        llvm::makeArrayRef(D->init_begin(), D->init_end()), Record);
   Code = serialization::DECL_OBJC_IMPLEMENTATION;
 }
 
@@ -732,13 +790,15 @@ void ASTDeclWriter::VisitVarDecl(VarDecl *D) {
   Record.push_back(D->getStorageClass());
   Record.push_back(D->getTSCSpec());
   Record.push_back(D->getInitStyle());
-  Record.push_back(D->isExceptionVariable());
-  Record.push_back(D->isNRVOVariable());
-  Record.push_back(D->isCXXForRangeDecl());
-  Record.push_back(D->isARCPseudoStrong());
-  Record.push_back(D->isConstexpr());
-  Record.push_back(D->isInitCapture());
-  Record.push_back(D->isPreviousDeclInSameBlockScope());
+  if (!isa<ParmVarDecl>(D)) {
+    Record.push_back(D->isExceptionVariable());
+    Record.push_back(D->isNRVOVariable());
+    Record.push_back(D->isCXXForRangeDecl());
+    Record.push_back(D->isARCPseudoStrong());
+    Record.push_back(D->isConstexpr());
+    Record.push_back(D->isInitCapture());
+    Record.push_back(D->isPreviousDeclInSameBlockScope());
+  }
   Record.push_back(D->getLinkageInternal());
 
   if (D->getInit()) {
@@ -920,17 +980,34 @@ void ASTDeclWriter::VisitNamespaceDecl(NamespaceDecl *D) {
   if (Writer.hasChain() && !D->isOriginalNamespace() &&
       D->getOriginalNamespace()->isFromASTFile()) {
     NamespaceDecl *NS = D->getOriginalNamespace();
-    Writer.AddUpdatedDeclContext(NS);
-
-    // Make sure all visible decls are written. They will be recorded later.
-    if (StoredDeclsMap *Map = NS->buildLookup()) {
-      for (StoredDeclsMap::iterator D = Map->begin(), DEnd = Map->end();
-           D != DEnd; ++D) {
-        DeclContext::lookup_result R = D->second.getLookupResult();
-        for (DeclContext::lookup_iterator I = R.begin(), E = R.end(); I != E;
-             ++I)
-          Writer.GetDeclRef(*I);
+    Writer.UpdatedDeclContexts.insert(NS);
+
+    // Make sure all visible decls are written. They will be recorded later. We
+    // do this using a side data structure so we can sort the names into
+    // a deterministic order.
+    StoredDeclsMap *Map = NS->buildLookup();
+    SmallVector<std::pair<DeclarationName, DeclContext::lookup_result>, 16>
+        LookupResults;
+    LookupResults.reserve(Map->size());
+    for (auto &Entry : *Map)
+      LookupResults.push_back(
+          std::make_pair(Entry.first, Entry.second.getLookupResult()));
+
+    std::sort(LookupResults.begin(), LookupResults.end(), llvm::less_first());
+    for (auto &NameAndResult : LookupResults) {
+      DeclarationName Name = NameAndResult.first;
+      DeclContext::lookup_result Result = NameAndResult.second;
+      if (Name.getNameKind() == DeclarationName::CXXConstructorName ||
+          Name.getNameKind() == DeclarationName::CXXConversionFunctionName) {
+        // We have to work around a name lookup bug here where negative lookup
+        // results for these names get cached in namespace lookup tables.
+        assert(Result.empty() && "Cannot have a constructor or conversion "
+                                 "function name in a namespace!");
+        continue;
       }
+
+      for (NamedDecl *ND : Result)
+        Writer.GetDeclRef(ND);
     }
   }
 
@@ -1067,8 +1144,6 @@ void ASTDeclWriter::VisitCXXConstructorDecl(CXXConstructorDecl *D) {
 
   Writer.AddDeclRef(D->getInheritedConstructor(), Record);
   Record.push_back(D->IsExplicitSpecified);
-  Writer.AddCXXCtorInitializers(D->CtorInitializers, D->NumCtorInitializers,
-                                Record);
 
   Code = serialization::DECL_CXX_CONSTRUCTOR;
 }
@@ -1076,7 +1151,7 @@ void ASTDeclWriter::VisitCXXConstructorDecl(CXXConstructorDecl *D) {
 void ASTDeclWriter::VisitCXXDestructorDecl(CXXDestructorDecl *D) {
   VisitCXXMethodDecl(D);
 
-  Writer.AddDeclRef(D->OperatorDelete, Record);
+  Writer.AddDeclRef(D->getOperatorDelete(), Record);
 
   Code = serialization::DECL_CXX_DESTRUCTOR;
 }
@@ -1171,24 +1246,8 @@ void ASTDeclWriter::VisitRedeclarableTemplateDecl(RedeclarableTemplateDecl *D) {
 void ASTDeclWriter::VisitClassTemplateDecl(ClassTemplateDecl *D) {
   VisitRedeclarableTemplateDecl(D);
 
-  if (D->isFirstDecl()) {
-    typedef llvm::FoldingSetVector<ClassTemplateSpecializationDecl> CTSDSetTy;
-    CTSDSetTy &CTSDSet = D->getSpecializations();
-    Record.push_back(CTSDSet.size());
-    for (CTSDSetTy::iterator I=CTSDSet.begin(), E = CTSDSet.end(); I!=E; ++I) {
-      assert(I->isCanonicalDecl() && "Expected only canonical decls in set");
-      Writer.AddDeclRef(&*I, Record);
-    }
-
-    typedef llvm::FoldingSetVector<ClassTemplatePartialSpecializationDecl>
-      CTPSDSetTy;
-    CTPSDSetTy &CTPSDSet = D->getPartialSpecializations();
-    Record.push_back(CTPSDSet.size());
-    for (CTPSDSetTy::iterator I=CTPSDSet.begin(), E=CTPSDSet.end(); I!=E; ++I) {
-      assert(I->isCanonicalDecl() && "Expected only canonical decls in set");
-      Writer.AddDeclRef(&*I, Record); 
-    }
-  }
+  if (D->isFirstDecl())
+    AddTemplateSpecializations(D);
   Code = serialization::DECL_CLASS_TEMPLATE;
 }
 
@@ -1246,26 +1305,8 @@ void ASTDeclWriter::VisitClassTemplatePartialSpecializationDecl(
 void ASTDeclWriter::VisitVarTemplateDecl(VarTemplateDecl *D) {
   VisitRedeclarableTemplateDecl(D);
 
-  if (D->isFirstDecl()) {
-    typedef llvm::FoldingSetVector<VarTemplateSpecializationDecl> VTSDSetTy;
-    VTSDSetTy &VTSDSet = D->getSpecializations();
-    Record.push_back(VTSDSet.size());
-    for (VTSDSetTy::iterator I = VTSDSet.begin(), E = VTSDSet.end(); I != E;
-         ++I) {
-      assert(I->isCanonicalDecl() && "Expected only canonical decls in set");
-      Writer.AddDeclRef(&*I, Record);
-    }
-
-    typedef llvm::FoldingSetVector<VarTemplatePartialSpecializationDecl>
-    VTPSDSetTy;
-    VTPSDSetTy &VTPSDSet = D->getPartialSpecializations();
-    Record.push_back(VTPSDSet.size());
-    for (VTPSDSetTy::iterator I = VTPSDSet.begin(), E = VTPSDSet.end(); I != E;
-         ++I) {
-      assert(I->isCanonicalDecl() && "Expected only canonical decls in set");
-      Writer.AddDeclRef(&*I, Record);
-    }
-  }
+  if (D->isFirstDecl())
+    AddTemplateSpecializations(D);
   Code = serialization::DECL_VAR_TEMPLATE;
 }
 
@@ -1330,19 +1371,8 @@ void ASTDeclWriter::VisitClassScopeFunctionSpecializationDecl(
 void ASTDeclWriter::VisitFunctionTemplateDecl(FunctionTemplateDecl *D) {
   VisitRedeclarableTemplateDecl(D);
 
-  if (D->isFirstDecl()) {
-    // This FunctionTemplateDecl owns the CommonPtr; write it.
-
-    // Write the function specialization declarations.
-    Record.push_back(D->getSpecializations().size());
-    for (llvm::FoldingSetVector<FunctionTemplateSpecializationInfo>::iterator
-           I = D->getSpecializations().begin(),
-           E = D->getSpecializations().end()   ; I != E; ++I) {
-      assert(I->Function->isCanonicalDecl() &&
-             "Expected only canonical decls in set");
-      Writer.AddDeclRef(I->Function, Record);
-    }
-  }
+  if (D->isFirstDecl())
+    AddTemplateSpecializations(D);
   Code = serialization::DECL_FUNCTION_TEMPLATE;
 }
 
@@ -1448,25 +1478,59 @@ void ASTDeclWriter::VisitDeclContext(DeclContext *DC, uint64_t LexicalOffset,
 template <typename T>
 void ASTDeclWriter::VisitRedeclarable(Redeclarable<T> *D) {
   T *First = D->getFirstDecl();
-  if (First->getMostRecentDecl() != First) {
+  T *MostRecent = First->getMostRecentDecl();
+  if (MostRecent != First) {
     assert(isRedeclarableDeclKind(static_cast<T *>(D)->getKind()) &&
            "Not considered redeclarable?");
-    
+
     // There is more than one declaration of this entity, so we will need to
     // write a redeclaration chain.
     Writer.AddDeclRef(First, Record);
     Writer.Redeclarations.insert(First);
 
+    auto *Previous = D->getPreviousDecl();
+
+    // In a modules build, we can have imported declarations after a local
+    // canonical declaration. If this is the first local declaration, emit
+    // a list of all such imported declarations so that we can ensure they
+    // are loaded before we are. This allows us to rebuild the redecl chain
+    // in the right order on reload (all declarations imported by a module
+    // should be before all declarations provided by that module).
+    bool EmitImportedMergedCanonicalDecls = false;
+    if (Context.getLangOpts().Modules && Writer.Chain) {
+      auto *PreviousLocal = Previous;
+      while (PreviousLocal && PreviousLocal->isFromASTFile())
+        PreviousLocal = PreviousLocal->getPreviousDecl();
+      if (!PreviousLocal)
+        EmitImportedMergedCanonicalDecls = true;
+    }
+    if (EmitImportedMergedCanonicalDecls) {
+      llvm::SmallMapVector<ModuleFile*, Decl*, 16> FirstInModule;
+      for (auto *Redecl = MostRecent; Redecl;
+           Redecl = Redecl->getPreviousDecl())
+        if (Redecl->isFromASTFile())
+          FirstInModule[Writer.Chain->getOwningModuleFile(Redecl)] = Redecl;
+      // FIXME: If FirstInModule has entries for modules A and B, and B imports
+      // A (directly or indirectly), we don't need to write the entry for A.
+      Record.push_back(FirstInModule.size());
+      for (auto I = FirstInModule.rbegin(), E = FirstInModule.rend();
+           I != E; ++I)
+        Writer.AddDeclRef(I->second, Record);
+    } else
+      Record.push_back(0);
+
     // Make sure that we serialize both the previous and the most-recent 
     // declarations, which (transitively) ensures that all declarations in the
     // chain get serialized.
-    (void)Writer.GetDeclRef(D->getPreviousDecl());
-    (void)Writer.GetDeclRef(First->getMostRecentDecl());
+    //
+    // FIXME: This is not correct; when we reach an imported declaration we
+    // won't emit its previous declaration.
+    (void)Writer.GetDeclRef(Previous);
+    (void)Writer.GetDeclRef(MostRecent);
   } else {
     // We use the sentinel value 0 to indicate an only declaration.
     Record.push_back(0);
   }
-  
 }
 
 void ASTDeclWriter::VisitOMPThreadPrivateDecl(OMPThreadPrivateDecl *D) {
@@ -1504,6 +1568,7 @@ void ASTWriter::WriteDeclAbbrevs() {
   // NamedDecl
   Abv->Add(BitCodeAbbrevOp(0));                       // NameKind = Identifier
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Name
+  Abv->Add(BitCodeAbbrevOp(0));                       // AnonDeclNumber
   // ValueDecl
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Type
   // DeclaratorDecl
@@ -1536,6 +1601,7 @@ void ASTWriter::WriteDeclAbbrevs() {
   // NamedDecl
   Abv->Add(BitCodeAbbrevOp(0));                       // NameKind = Identifier
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Name
+  Abv->Add(BitCodeAbbrevOp(0));                       // AnonDeclNumber
   // ValueDecl
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Type
   // DeclaratorDecl
@@ -1573,6 +1639,7 @@ void ASTWriter::WriteDeclAbbrevs() {
   // NamedDecl
   Abv->Add(BitCodeAbbrevOp(0));                       // NameKind = Identifier
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Name
+  Abv->Add(BitCodeAbbrevOp(0));                       // AnonDeclNumber
   // TypeDecl
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Source Location
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Type Ref
@@ -1620,6 +1687,7 @@ void ASTWriter::WriteDeclAbbrevs() {
   // NamedDecl
   Abv->Add(BitCodeAbbrevOp(0));                       // NameKind = Identifier
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Name
+  Abv->Add(BitCodeAbbrevOp(0));                       // AnonDeclNumber
   // TypeDecl
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Source Location
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Type Ref
@@ -1662,6 +1730,7 @@ void ASTWriter::WriteDeclAbbrevs() {
   // NamedDecl
   Abv->Add(BitCodeAbbrevOp(0));                       // NameKind = Identifier
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Name
+  Abv->Add(BitCodeAbbrevOp(0));                       // AnonDeclNumber
   // ValueDecl
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Type
   // DeclaratorDecl
@@ -1671,13 +1740,6 @@ void ASTWriter::WriteDeclAbbrevs() {
   Abv->Add(BitCodeAbbrevOp(0));                       // StorageClass
   Abv->Add(BitCodeAbbrevOp(0));                       // getTSCSpec
   Abv->Add(BitCodeAbbrevOp(0));                       // hasCXXDirectInitializer
-  Abv->Add(BitCodeAbbrevOp(0));                       // isExceptionVariable
-  Abv->Add(BitCodeAbbrevOp(0));                       // isNRVOVariable
-  Abv->Add(BitCodeAbbrevOp(0));                       // isCXXForRangeDecl
-  Abv->Add(BitCodeAbbrevOp(0));                       // isARCPseudoStrong
-  Abv->Add(BitCodeAbbrevOp(0));                       // isConstexpr
-  Abv->Add(BitCodeAbbrevOp(0));                       // isInitCapture
-  Abv->Add(BitCodeAbbrevOp(0));                       // isPrevDeclInSameScope
   Abv->Add(BitCodeAbbrevOp(0));                       // Linkage
   Abv->Add(BitCodeAbbrevOp(0));                       // HasInit
   Abv->Add(BitCodeAbbrevOp(0));                   // HasMemberSpecializationInfo
@@ -1715,6 +1777,7 @@ void ASTWriter::WriteDeclAbbrevs() {
   // NamedDecl
   Abv->Add(BitCodeAbbrevOp(0));                       // NameKind = Identifier
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Name
+  Abv->Add(BitCodeAbbrevOp(0));                       // AnonDeclNumber
   // TypeDecl
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Source Location
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Type Ref
@@ -1743,6 +1806,7 @@ void ASTWriter::WriteDeclAbbrevs() {
   // NamedDecl
   Abv->Add(BitCodeAbbrevOp(0));                       // NameKind = Identifier
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Name
+  Abv->Add(BitCodeAbbrevOp(0));                       // AnonDeclNumber
   // ValueDecl
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6)); // Type
   // DeclaratorDecl
@@ -1788,6 +1852,7 @@ void ASTWriter::WriteDeclAbbrevs() {
   // NamedDecl
   Abv->Add(BitCodeAbbrevOp(DeclarationName::Identifier)); // NameKind
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6));   // Identifier
+  Abv->Add(BitCodeAbbrevOp(0));                         // AnonDeclNumber
   // ValueDecl
   Abv->Add(BitCodeAbbrevOp(BitCodeAbbrevOp::VBR, 6));   // Type
   // DeclaratorDecl
@@ -1948,9 +2013,10 @@ void ASTWriter::WriteDecl(ASTContext &Context, Decl *D) {
 
   // Determine the ID for this declaration.
   serialization::DeclID ID;
-  if (D->isFromASTFile())
+  if (D->isFromASTFile()) {
+    assert(isRewritten(D) && "should not be emitting imported decl");
     ID = getDeclID(D);
-  else {
+  } else {
     serialization::DeclID &IDR = DeclIDs[D];
     if (IDR == 0)
       IDR = NextDeclID++;
@@ -2015,12 +2081,10 @@ void ASTWriter::WriteDecl(ASTContext &Context, Decl *D) {
                             D->getDeclKindName() + "'");
   Stream.EmitRecord(W.Code, Record, W.AbbrevToUse);
 
-  // Flush any expressions that were written as part of this declaration.
-  FlushStmts();
-  
-  // Flush C++ base specifiers, if there are any.
-  FlushCXXBaseSpecifiers();
-  
+  // Flush any expressions, base specifiers, and ctor initializers that
+  // were written as part of this declaration.
+  FlushPendingAfterDecl();
+
   // Note declarations that should be deserialized eagerly so that we can add
   // them to a record in the AST file later.
   if (isRequiredDecl(D, Context))
diff --git a/lib/Serialization/ASTWriterStmt.cpp b/lib/Serialization/ASTWriterStmt.cpp
index e980ce783f5e..ec822f02c67f 100644
--- a/lib/Serialization/ASTWriterStmt.cpp
+++ b/lib/Serialization/ASTWriterStmt.cpp
@@ -553,6 +553,7 @@ void ASTStmtWriter::VisitMemberExpr(MemberExpr *E) {
   Writer.AddDeclRef(E->getMemberDecl(), Record);
   Writer.AddSourceLocation(E->getMemberLoc(), Record);
   Record.push_back(E->isArrow());
+  Writer.AddSourceLocation(E->getOperatorLoc(), Record);
   Writer.AddDeclarationNameLoc(E->MemberDNLoc,
                                E->getMemberDecl()->getDeclName(), Record);
   Code = serialization::EXPR_MEMBER;
@@ -1744,6 +1745,7 @@ void OMPClauseWriter::VisitOMPProcBindClause(OMPProcBindClause *C) {
 void OMPClauseWriter::VisitOMPScheduleClause(OMPScheduleClause *C) {
   Record.push_back(C->getScheduleKind());
   Writer->Writer.AddStmt(C->getChunkSize());
+  Writer->Writer.AddStmt(C->getHelperChunkSize());
   Writer->Writer.AddSourceLocation(C->getLParenLoc(), Record);
   Writer->Writer.AddSourceLocation(C->getScheduleKindLoc(), Record);
   Writer->Writer.AddSourceLocation(C->getCommaLoc(), Record);
@@ -1797,6 +1799,14 @@ void OMPClauseWriter::VisitOMPLastprivateClause(OMPLastprivateClause *C) {
   Writer->Writer.AddSourceLocation(C->getLParenLoc(), Record);
   for (auto *VE : C->varlists())
     Writer->Writer.AddStmt(VE);
+  for (auto *E : C->private_copies())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->source_exprs())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->destination_exprs())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->assignment_ops())
+    Writer->Writer.AddStmt(E);
 }
 
 void OMPClauseWriter::VisitOMPSharedClause(OMPSharedClause *C) {
@@ -1814,15 +1824,32 @@ void OMPClauseWriter::VisitOMPReductionClause(OMPReductionClause *C) {
   Writer->Writer.AddDeclarationNameInfo(C->getNameInfo(), Record);
   for (auto *VE : C->varlists())
     Writer->Writer.AddStmt(VE);
+  for (auto *E : C->lhs_exprs())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->rhs_exprs())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->reduction_ops())
+    Writer->Writer.AddStmt(E);
 }
 
 void OMPClauseWriter::VisitOMPLinearClause(OMPLinearClause *C) {
   Record.push_back(C->varlist_size());
   Writer->Writer.AddSourceLocation(C->getLParenLoc(), Record);
   Writer->Writer.AddSourceLocation(C->getColonLoc(), Record);
-  for (auto *VE : C->varlists())
+  for (auto *VE : C->varlists()) {
+    Writer->Writer.AddStmt(VE);
+  }
+  for (auto *VE : C->inits()) {
     Writer->Writer.AddStmt(VE);
+  }
+  for (auto *VE : C->updates()) {
+    Writer->Writer.AddStmt(VE);
+  }
+  for (auto *VE : C->finals()) {
+    Writer->Writer.AddStmt(VE);
+  }
   Writer->Writer.AddStmt(C->getStep());
+  Writer->Writer.AddStmt(C->getCalcStep());
 }
 
 void OMPClauseWriter::VisitOMPAlignedClause(OMPAlignedClause *C) {
@@ -1839,6 +1866,12 @@ void OMPClauseWriter::VisitOMPCopyinClause(OMPCopyinClause *C) {
   Writer->Writer.AddSourceLocation(C->getLParenLoc(), Record);
   for (auto *VE : C->varlists())
     Writer->Writer.AddStmt(VE);
+  for (auto *E : C->source_exprs())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->destination_exprs())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->assignment_ops())
+    Writer->Writer.AddStmt(E);
 }
 
 void OMPClauseWriter::VisitOMPCopyprivateClause(OMPCopyprivateClause *C) {
@@ -1846,6 +1879,12 @@ void OMPClauseWriter::VisitOMPCopyprivateClause(OMPCopyprivateClause *C) {
   Writer->Writer.AddSourceLocation(C->getLParenLoc(), Record);
   for (auto *VE : C->varlists())
     Writer->Writer.AddStmt(VE);
+  for (auto *E : C->source_exprs())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->destination_exprs())
+    Writer->Writer.AddStmt(E);
+  for (auto *E : C->assignment_ops())
+    Writer->Writer.AddStmt(E);
 }
 
 void OMPClauseWriter::VisitOMPFlushClause(OMPFlushClause *C) {
@@ -1990,6 +2029,9 @@ void ASTStmtWriter::VisitOMPAtomicDirective(OMPAtomicDirective *D) {
   Writer.AddStmt(D->getX());
   Writer.AddStmt(D->getV());
   Writer.AddStmt(D->getExpr());
+  Writer.AddStmt(D->getUpdateExpr());
+  Record.push_back(D->isXLHSInRHSPart() ? 1 : 0);
+  Record.push_back(D->isPostfixUpdate() ? 1 : 0);
   Code = serialization::STMT_OMP_ATOMIC_DIRECTIVE;
 }
 
diff --git a/lib/Serialization/GlobalModuleIndex.cpp b/lib/Serialization/GlobalModuleIndex.cpp
index 479138804547..1b52b441134f 100644
--- a/lib/Serialization/GlobalModuleIndex.cpp
+++ b/lib/Serialization/GlobalModuleIndex.cpp
@@ -757,7 +757,7 @@ void GlobalModuleIndexBuilder::writeIndex(llvm::BitstreamWriter &Stream) {
     Record.clear();
     Record.push_back(IDENTIFIER_INDEX);
     Record.push_back(BucketOffset);
-    Stream.EmitRecordWithBlob(IDTableAbbrev, Record, IdentifierTable.str());
+    Stream.EmitRecordWithBlob(IDTableAbbrev, Record, IdentifierTable);
   }
 
   Stream.ExitBlock();
@@ -841,12 +841,12 @@ GlobalModuleIndex::writeIndex(FileManager &FileMgr, StringRef Path) {
     return EC_IOError;
 
   // Remove the old index file. It isn't relevant any more.
-  llvm::sys::fs::remove(IndexPath.str());
+  llvm::sys::fs::remove(IndexPath);
 
   // Rename the newly-written index file to the proper name.
-  if (llvm::sys::fs::rename(IndexTmpPath.str(), IndexPath.str())) {
+  if (llvm::sys::fs::rename(IndexTmpPath, IndexPath)) {
     // Rename failed; just remove the 
-    llvm::sys::fs::remove(IndexTmpPath.str());
+    llvm::sys::fs::remove(IndexTmpPath);
     return EC_IOError;
   }
 
diff --git a/lib/Serialization/Module.cpp b/lib/Serialization/Module.cpp
index 6c48a41187c2..3b237d5529c6 100644
--- a/lib/Serialization/Module.cpp
+++ b/lib/Serialization/Module.cpp
@@ -38,6 +38,7 @@ ModuleFile::ModuleFile(ModuleKind Kind, unsigned Generation)
     SelectorLookupTableData(nullptr), SelectorLookupTable(nullptr),
     LocalNumDecls(0), DeclOffsets(nullptr), BaseDeclID(0),
     LocalNumCXXBaseSpecifiers(0), CXXBaseSpecifiersOffsets(nullptr),
+    LocalNumCXXCtorInitializers(0), CXXCtorInitializersOffsets(nullptr),
     FileSortedDecls(nullptr), NumFileSortedDecls(0),
     RedeclarationsMap(nullptr), LocalNumRedeclarationsInMap(0),
     ObjCCategoriesMap(nullptr), LocalNumObjCCategoriesInMap(0),
diff --git a/lib/Serialization/ModuleManager.cpp b/lib/Serialization/ModuleManager.cpp
index ac98ca0b8720..30d9c89a1244 100644
--- a/lib/Serialization/ModuleManager.cpp
+++ b/lib/Serialization/ModuleManager.cpp
@@ -58,8 +58,7 @@ ModuleManager::addModule(StringRef FileName, ModuleKind Type,
                          unsigned Generation,
                          off_t ExpectedSize, time_t ExpectedModTime,
                          ASTFileSignature ExpectedSignature,
-                         std::function<ASTFileSignature(llvm::BitstreamReader &)>
-                             ReadSignature,
+                         ASTFileSignatureReader ReadSignature,
                          ModuleFile *&Module,
                          std::string &ErrorStr) {
   Module = nullptr;
@@ -95,6 +94,8 @@ ModuleManager::addModule(StringRef FileName, ModuleKind Type,
     New->File = Entry;
     New->ImportLoc = ImportLoc;
     Chain.push_back(New);
+    if (!ImportedBy)
+      Roots.push_back(New);
     NewModule = true;
     ModuleEntry = New;
 
@@ -156,7 +157,12 @@ ModuleManager::addModule(StringRef FileName, ModuleKind Type,
         // invalidate the file cache for Entry, and that is not safe if this
         // module is *itself* up to date, but has an out-of-date importer.
         Modules.erase(Entry);
+        assert(Chain.back() == ModuleEntry);
         Chain.pop_back();
+        if (Roots.back() == ModuleEntry)
+          Roots.pop_back();
+        else
+          assert(ImportedBy);
         delete ModuleEntry;
       }
       return OutOfDate;
@@ -187,12 +193,15 @@ void ModuleManager::removeModules(
   // Collect the set of module file pointers that we'll be removing.
   llvm::SmallPtrSet<ModuleFile *, 4> victimSet(first, last);
 
+  auto IsVictim = [&](ModuleFile *MF) {
+    return victimSet.count(MF);
+  };
   // Remove any references to the now-destroyed modules.
   for (unsigned i = 0, n = Chain.size(); i != n; ++i) {
-    Chain[i]->ImportedBy.remove_if([&](ModuleFile *MF) {
-      return victimSet.count(MF);
-    });
+    Chain[i]->ImportedBy.remove_if(IsVictim);
   }
+  Roots.erase(std::remove_if(Roots.begin(), Roots.end(), IsVictim),
+              Roots.end());
 
   // Delete the modules and erase them from the various structures.
   for (ModuleIterator victim = first; victim != last; ++victim) {
@@ -227,6 +236,15 @@ ModuleManager::addInMemoryBuffer(StringRef FileName,
   InMemoryBuffers[Entry] = std::move(Buffer);
 }
 
+bool ModuleManager::addKnownModuleFile(StringRef FileName) {
+  const FileEntry *File;
+  if (lookupModuleFile(FileName, 0, 0, File))
+    return true;
+  if (!Modules.count(File))
+    AdditionalKnownModuleFiles.insert(File);
+  return false;
+}
+
 ModuleManager::VisitState *ModuleManager::allocateVisitState() {
   // Fast path: if we have a cached state, use it.
   if (FirstVisitState) {
@@ -263,6 +281,8 @@ void ModuleManager::setGlobalIndex(GlobalModuleIndex *Index) {
 }
 
 void ModuleManager::moduleFileAccepted(ModuleFile *MF) {
+  AdditionalKnownModuleFiles.remove(MF->File);
+
   if (!GlobalIndex || GlobalIndex->loadedModuleFile(MF))
     return;
 
@@ -388,16 +408,38 @@ ModuleManager::visit(bool (*Visitor)(ModuleFile &M, void *UserData),
   returnVisitState(State);
 }
 
+static void markVisitedDepthFirst(ModuleFile &M,
+                                  SmallVectorImpl<bool> &Visited) {
+  for (llvm::SetVector<ModuleFile *>::iterator IM = M.Imports.begin(),
+                                               IMEnd = M.Imports.end();
+       IM != IMEnd; ++IM) {
+    if (Visited[(*IM)->Index])
+      continue;
+    Visited[(*IM)->Index] = true;
+    if (!M.DirectlyImported)
+      markVisitedDepthFirst(**IM, Visited);
+  }
+}
+
 /// \brief Perform a depth-first visit of the current module.
-static bool visitDepthFirst(ModuleFile &M, 
-                            bool (*Visitor)(ModuleFile &M, bool Preorder, 
-                                            void *UserData), 
-                            void *UserData,
-                            SmallVectorImpl<bool> &Visited) {
-  // Preorder visitation
-  if (Visitor(M, /*Preorder=*/true, UserData))
-    return true;
-  
+static bool visitDepthFirst(
+    ModuleFile &M,
+    ModuleManager::DFSPreorderControl (*PreorderVisitor)(ModuleFile &M,
+                                                         void *UserData),
+    bool (*PostorderVisitor)(ModuleFile &M, void *UserData), void *UserData,
+    SmallVectorImpl<bool> &Visited) {
+  if (PreorderVisitor) {
+    switch (PreorderVisitor(M, UserData)) {
+    case ModuleManager::Abort:
+      return true;
+    case ModuleManager::SkipImports:
+      markVisitedDepthFirst(M, Visited);
+      return false;
+    case ModuleManager::Continue:
+      break;
+    }
+  }
+
   // Visit children
   for (llvm::SetVector<ModuleFile *>::iterator IM = M.Imports.begin(),
                                             IMEnd = M.Imports.end();
@@ -406,24 +448,27 @@ static bool visitDepthFirst(ModuleFile &M,
       continue;
     Visited[(*IM)->Index] = true;
 
-    if (visitDepthFirst(**IM, Visitor, UserData, Visited))
+    if (visitDepthFirst(**IM, PreorderVisitor, PostorderVisitor, UserData, Visited))
       return true;
   }  
   
-  // Postorder visitation
-  return Visitor(M, /*Preorder=*/false, UserData);
+  if (PostorderVisitor)
+    return PostorderVisitor(M, UserData);
+
+  return false;
 }
 
-void ModuleManager::visitDepthFirst(bool (*Visitor)(ModuleFile &M, bool Preorder, 
-                                                    void *UserData), 
-                                    void *UserData) {
+void ModuleManager::visitDepthFirst(
+    ModuleManager::DFSPreorderControl (*PreorderVisitor)(ModuleFile &M,
+                                                         void *UserData),
+    bool (*PostorderVisitor)(ModuleFile &M, void *UserData), void *UserData) {
   SmallVector<bool, 16> Visited(size(), false);
-  for (unsigned I = 0, N = Chain.size(); I != N; ++I) {
-    if (Visited[Chain[I]->Index])
+  for (unsigned I = 0, N = Roots.size(); I != N; ++I) {
+    if (Visited[Roots[I]->Index])
       continue;
-    Visited[Chain[I]->Index] = true;
+    Visited[Roots[I]->Index] = true;
 
-    if (::visitDepthFirst(*Chain[I], Visitor, UserData, Visited))
+    if (::visitDepthFirst(*Roots[I], PreorderVisitor, PostorderVisitor, UserData, Visited))
       return;
   }
 }
diff --git a/lib/StaticAnalyzer/Checkers/CStringChecker.cpp b/lib/StaticAnalyzer/Checkers/CStringChecker.cpp
index e91a7e16802e..0f5741bf9e68 100644
--- a/lib/StaticAnalyzer/Checkers/CStringChecker.cpp
+++ b/lib/StaticAnalyzer/Checkers/CStringChecker.cpp
@@ -1922,10 +1922,6 @@ bool CStringChecker::evalCall(const CallExpr *CE, CheckerContext &C) const {
   if (!evalFunction)
     return false;
 
-  // Make sure each function sets its own description.
-  // (But don't bother in a release build.)
-  assert(!(CurrentFunctionDescription = nullptr));
-
   // Check and evaluate the call.
   (this->*evalFunction)(C, CE);
 
diff --git a/lib/StaticAnalyzer/Checkers/CheckSecuritySyntaxOnly.cpp b/lib/StaticAnalyzer/Checkers/CheckSecuritySyntaxOnly.cpp
index 45768b2cdf76..0beb917833d6 100644
--- a/lib/StaticAnalyzer/Checkers/CheckSecuritySyntaxOnly.cpp
+++ b/lib/StaticAnalyzer/Checkers/CheckSecuritySyntaxOnly.cpp
@@ -28,6 +28,7 @@ using namespace ento;
 static bool isArc4RandomAvailable(const ASTContext &Ctx) {
   const llvm::Triple &T = Ctx.getTargetInfo().getTriple();
   return T.getVendor() == llvm::Triple::Apple ||
+         T.getOS() == llvm::Triple::CloudABI ||
          T.getOS() == llvm::Triple::FreeBSD ||
          T.getOS() == llvm::Triple::NetBSD ||
          T.getOS() == llvm::Triple::OpenBSD ||
diff --git a/lib/StaticAnalyzer/Checkers/Checkers.td b/lib/StaticAnalyzer/Checkers/Checkers.td
index ba5b4fa3c66c..d1d6ac277ffe 100644
--- a/lib/StaticAnalyzer/Checkers/Checkers.td
+++ b/lib/StaticAnalyzer/Checkers/Checkers.td
@@ -295,7 +295,7 @@ def UnixAPIChecker : Checker<"API">,
   HelpText<"Check calls to various UNIX/Posix functions">,
   DescFile<"UnixAPIChecker.cpp">;
 
-def MallocPessimistic : Checker<"Malloc">,
+def MallocChecker: Checker<"Malloc">,
   HelpText<"Check for memory leaks, double free, and use-after-free problems. Traces memory managed by malloc()/free().">,
   DescFile<"MallocChecker.cpp">;
   
@@ -315,10 +315,6 @@ def ChrootChecker : Checker<"Chroot">,
   HelpText<"Check improper use of chroot">,
   DescFile<"ChrootChecker.cpp">;
 
-def MallocOptimistic : Checker<"MallocWithAnnotations">,
-  HelpText<"Check for memory leaks, double free, and use-after-free problems. Traces memory managed by malloc()/free(). Assumes that all user-defined functions which might free a pointer are annotated.">,
-  DescFile<"MallocChecker.cpp">;
-
 def PthreadLockChecker : Checker<"PthreadLock">,
   HelpText<"Simple lock -> unlock checker">,
   DescFile<"PthreadLockChecker.cpp">;
diff --git a/lib/StaticAnalyzer/Checkers/DeadStoresChecker.cpp b/lib/StaticAnalyzer/Checkers/DeadStoresChecker.cpp
index c1ea76735c41..f4be5b3e82f4 100644
--- a/lib/StaticAnalyzer/Checkers/DeadStoresChecker.cpp
+++ b/lib/StaticAnalyzer/Checkers/DeadStoresChecker.cpp
@@ -142,7 +142,7 @@ public:
       : cfg(cfg), Ctx(ctx), BR(br), Checker(checker), AC(ac), Parents(parents),
         Escaped(escaped), currentBlock(nullptr) {}
 
-  virtual ~DeadStoreObs() {}
+  ~DeadStoreObs() override {}
 
   bool isLive(const LiveVariables::LivenessValues &Live, const VarDecl *D) {
     if (Live.isLive(D))
diff --git a/lib/StaticAnalyzer/Checkers/DereferenceChecker.cpp b/lib/StaticAnalyzer/Checkers/DereferenceChecker.cpp
index 4ee02230649d..2e442c7d875c 100644
--- a/lib/StaticAnalyzer/Checkers/DereferenceChecker.cpp
+++ b/lib/StaticAnalyzer/Checkers/DereferenceChecker.cpp
@@ -162,7 +162,7 @@ void DereferenceChecker::reportBug(ProgramStateRef State, const Stmt *S,
   os.flush();
   BugReport *report =
     new BugReport(*BT_null,
-                  buf.empty() ? BT_null->getDescription() : buf.str(),
+                  buf.empty() ? BT_null->getDescription() : StringRef(buf),
                   N);
 
   bugreporter::trackNullOrUndefValue(N, bugreporter::getDerefExpr(S), *report);
diff --git a/lib/StaticAnalyzer/Checkers/InterCheckerAPI.h b/lib/StaticAnalyzer/Checkers/InterCheckerAPI.h
index b7549fd17da9..d38d63cd05ce 100644
--- a/lib/StaticAnalyzer/Checkers/InterCheckerAPI.h
+++ b/lib/StaticAnalyzer/Checkers/InterCheckerAPI.h
@@ -13,6 +13,8 @@
 #ifndef LLVM_CLANG_LIB_STATICANALYZER_CHECKERS_INTERCHECKERAPI_H
 #define LLVM_CLANG_LIB_STATICANALYZER_CHECKERS_INTERCHECKERAPI_H
 namespace clang {
+class CheckerManager;
+
 namespace ento {
 
 /// Register the checker which evaluates CString API calls.
diff --git a/lib/StaticAnalyzer/Checkers/IvarInvalidationChecker.cpp b/lib/StaticAnalyzer/Checkers/IvarInvalidationChecker.cpp
index 1926600b6d6e..02c12095b5a5 100644
--- a/lib/StaticAnalyzer/Checkers/IvarInvalidationChecker.cpp
+++ b/lib/StaticAnalyzer/Checkers/IvarInvalidationChecker.cpp
@@ -336,7 +336,7 @@ const ObjCIvarDecl *IvarInvalidationCheckerImpl::findPropertyBackingIvar(
       llvm::raw_svector_ostream os(PropNameWithUnderscore);
       os << '_' << PropName;
     }
-    if (IvarName == PropNameWithUnderscore.str())
+    if (IvarName == PropNameWithUnderscore)
       return Iv;
   }
 
diff --git a/lib/StaticAnalyzer/Checkers/MacOSKeychainAPIChecker.cpp b/lib/StaticAnalyzer/Checkers/MacOSKeychainAPIChecker.cpp
index 13ea4d3ed090..52e29368cea3 100644
--- a/lib/StaticAnalyzer/Checkers/MacOSKeychainAPIChecker.cpp
+++ b/lib/StaticAnalyzer/Checkers/MacOSKeychainAPIChecker.cpp
@@ -137,7 +137,7 @@ private:
 
   public:
     SecKeychainBugVisitor(SymbolRef S) : Sym(S) {}
-    virtual ~SecKeychainBugVisitor() {}
+    ~SecKeychainBugVisitor() override {}
 
     void Profile(llvm::FoldingSetNodeID &ID) const override {
       static int X = 0;
@@ -292,7 +292,11 @@ void MacOSKeychainAPIChecker::checkPreStmt(const CallExpr *CE,
   // If it is a call to an allocator function, it could be a double allocation.
   idx = getTrackedFunctionIndex(funName, true);
   if (idx != InvalidIdx) {
-    const Expr *ArgExpr = CE->getArg(FunctionsToTrack[idx].Param);
+    unsigned paramIdx = FunctionsToTrack[idx].Param;
+    if (CE->getNumArgs() <= paramIdx)
+      return;
+
+    const Expr *ArgExpr = CE->getArg(paramIdx);
     if (SymbolRef V = getAsPointeeSymbol(ArgExpr, C))
       if (const AllocationState *AS = State->get<AllocatedData>(V)) {
         if (!definitelyReturnedError(AS->Region, State, C.getSValBuilder())) {
@@ -325,8 +329,12 @@ void MacOSKeychainAPIChecker::checkPreStmt(const CallExpr *CE,
   if (idx == InvalidIdx)
     return;
 
+  unsigned paramIdx = FunctionsToTrack[idx].Param;
+  if (CE->getNumArgs() <= paramIdx)
+    return;
+
   // Check the argument to the deallocator.
-  const Expr *ArgExpr = CE->getArg(FunctionsToTrack[idx].Param);
+  const Expr *ArgExpr = CE->getArg(paramIdx);
   SVal ArgSVal = State->getSVal(ArgExpr, C.getLocationContext());
 
   // Undef is reported by another checker.
@@ -499,9 +507,11 @@ MacOSKeychainAPIChecker::getAllocationNode(const ExplodedNode *N,
   while (N) {
     if (!N->getState()->get<AllocatedData>(Sym))
       break;
-    // Allocation node, is the last node in the current context in which the
-    // symbol was tracked.
-    if (N->getLocationContext() == LeakContext)
+    // Allocation node, is the last node in the current or parent context in
+    // which the symbol was tracked.
+    const LocationContext *NContext = N->getLocationContext();
+    if (NContext == LeakContext ||
+        NContext->isParentOf(LeakContext))
       AllocNode = N;
     N = N->pred_empty() ? nullptr : *(N->pred_begin());
   }
diff --git a/lib/StaticAnalyzer/Checkers/MallocChecker.cpp b/lib/StaticAnalyzer/Checkers/MallocChecker.cpp
index aee5a43048b9..0cf009407349 100644
--- a/lib/StaticAnalyzer/Checkers/MallocChecker.cpp
+++ b/lib/StaticAnalyzer/Checkers/MallocChecker.cpp
@@ -43,12 +43,15 @@ enum AllocationFamily {
   AF_Malloc,
   AF_CXXNew,
   AF_CXXNewArray,
-  AF_IfNameIndex
+  AF_IfNameIndex,
+  AF_Alloca
 };
 
 class RefState {
   enum Kind { // Reference to allocated memory.
               Allocated,
+              // Reference to zero-allocated memory.
+              AllocatedOfSizeZero,
               // Reference to released/freed memory.
               Released,
               // The responsibility for freeing resources has transferred from
@@ -61,8 +64,8 @@ class RefState {
   };
 
   const Stmt *S;
-  unsigned K : 2; // Kind enum, but stored as a bitfield.
-  unsigned Family : 30; // Rest of 32-bit word, currently just an allocation 
+  unsigned K : 3; // Kind enum, but stored as a bitfield.
+  unsigned Family : 29; // Rest of 32-bit word, currently just an allocation 
                         // family.
 
   RefState(Kind k, const Stmt *s, unsigned family) 
@@ -71,6 +74,7 @@ class RefState {
   }
 public:
   bool isAllocated() const { return K == Allocated; }
+  bool isAllocatedOfSizeZero() const { return K == AllocatedOfSizeZero; }
   bool isReleased() const { return K == Released; }
   bool isRelinquished() const { return K == Relinquished; }
   bool isEscaped() const { return K == Escaped; }
@@ -86,6 +90,10 @@ public:
   static RefState getAllocated(unsigned family, const Stmt *s) {
     return RefState(Allocated, s, family);
   }
+  static RefState getAllocatedOfSizeZero(const RefState *RS) {
+    return RefState(AllocatedOfSizeZero, RS->getStmt(),
+                    RS->getAllocationFamily());
+  }
   static RefState getReleased(unsigned family, const Stmt *s) { 
     return RefState(Released, s, family);
   }
@@ -106,6 +114,7 @@ public:
     switch (static_cast<Kind>(K)) {
 #define CASE(ID) case ID: OS << #ID; break;
     CASE(Allocated)
+    CASE(AllocatedOfSizeZero)
     CASE(Released)
     CASE(Relinquished)
     CASE(Escaped)
@@ -160,16 +169,16 @@ class MallocChecker : public Checker<check::DeadSymbols,
 {
 public:
   MallocChecker()
-      : II_malloc(nullptr), II_free(nullptr), II_realloc(nullptr),
-        II_calloc(nullptr), II_valloc(nullptr), II_reallocf(nullptr),
-        II_strndup(nullptr), II_strdup(nullptr), II_kmalloc(nullptr),
-        II_if_nameindex(nullptr), II_if_freenameindex(nullptr) {}
+      : II_alloca(nullptr), II_malloc(nullptr), II_free(nullptr), 
+        II_realloc(nullptr), II_calloc(nullptr), II_valloc(nullptr),
+        II_reallocf(nullptr), II_strndup(nullptr), II_strdup(nullptr), 
+        II_kmalloc(nullptr), II_if_nameindex(nullptr),
+        II_if_freenameindex(nullptr) {}
 
   /// In pessimistic mode, the checker assumes that it does not know which
   /// functions might free the memory.
   enum CheckKind {
-    CK_MallocPessimistic,
-    CK_MallocOptimistic,
+    CK_MallocChecker,
     CK_NewDeleteChecker,
     CK_NewDeleteLeaksChecker,
     CK_MismatchedDeallocatorChecker,
@@ -182,6 +191,8 @@ public:
     MOK_Any
   };
 
+  DefaultBool IsOptimistic;
+
   DefaultBool ChecksEnabled[CK_NumCheckKinds];
   CheckName CheckNames[CK_NumCheckKinds];
 
@@ -216,11 +227,14 @@ private:
   mutable std::unique_ptr<BugType> BT_Leak[CK_NumCheckKinds];
   mutable std::unique_ptr<BugType> BT_UseFree[CK_NumCheckKinds];
   mutable std::unique_ptr<BugType> BT_BadFree[CK_NumCheckKinds];
+  mutable std::unique_ptr<BugType> BT_FreeAlloca[CK_NumCheckKinds];
   mutable std::unique_ptr<BugType> BT_MismatchedDealloc;
   mutable std::unique_ptr<BugType> BT_OffsetFree[CK_NumCheckKinds];
-  mutable IdentifierInfo *II_malloc, *II_free, *II_realloc, *II_calloc,
-                         *II_valloc, *II_reallocf, *II_strndup, *II_strdup,
-                         *II_kmalloc, *II_if_nameindex, *II_if_freenameindex;
+  mutable std::unique_ptr<BugType> BT_UseZerroAllocated[CK_NumCheckKinds];
+  mutable IdentifierInfo *II_alloca, *II_malloc, *II_free, *II_realloc,
+                         *II_calloc, *II_valloc, *II_reallocf, *II_strndup,
+                         *II_strdup, *II_kmalloc, *II_if_nameindex,
+                         *II_if_freenameindex;
   mutable Optional<uint64_t> KernelZeroFlagVal;
 
   void initIdentifierInfo(ASTContext &C) const;
@@ -252,22 +266,24 @@ private:
                       MemoryOperationKind MemKind) const;
   bool isStandardNewDelete(const FunctionDecl *FD, ASTContext &C) const;
   ///@}
+
+  /// \brief Perform a zero-allocation check.
+  ProgramStateRef ProcessZeroAllocation(CheckerContext &C, const Expr *E,
+                                        const unsigned AllocationSizeArg,
+                                        ProgramStateRef State) const;
+
   ProgramStateRef MallocMemReturnsAttr(CheckerContext &C,
                                        const CallExpr *CE,
-                                       const OwnershipAttr* Att) const;
+                                       const OwnershipAttr* Att,
+                                       ProgramStateRef State) const;
   static ProgramStateRef MallocMemAux(CheckerContext &C, const CallExpr *CE,
-                                     const Expr *SizeEx, SVal Init,
-                                     ProgramStateRef State,
-                                     AllocationFamily Family = AF_Malloc) {
-    return MallocMemAux(C, CE,
-                        State->getSVal(SizeEx, C.getLocationContext()),
-                        Init, State, Family);
-  }
-
+                                      const Expr *SizeEx, SVal Init,
+                                      ProgramStateRef State,
+                                      AllocationFamily Family = AF_Malloc);
   static ProgramStateRef MallocMemAux(CheckerContext &C, const CallExpr *CE,
-                                     SVal SizeEx, SVal Init,
-                                     ProgramStateRef State,
-                                     AllocationFamily Family = AF_Malloc);
+                                      SVal SizeEx, SVal Init,
+                                      ProgramStateRef State,
+                                      AllocationFamily Family = AF_Malloc);
 
   // Check if this malloc() for special flags. At present that means M_ZERO or
   // __GFP_ZERO (in which case, treat it like calloc).
@@ -281,7 +297,8 @@ private:
                        AllocationFamily Family = AF_Malloc);
 
   ProgramStateRef FreeMemAttr(CheckerContext &C, const CallExpr *CE,
-                              const OwnershipAttr* Att) const;
+                              const OwnershipAttr* Att,
+                              ProgramStateRef State) const;
   ProgramStateRef FreeMemAux(CheckerContext &C, const CallExpr *CE,
                              ProgramStateRef state, unsigned Num,
                              bool Hold,
@@ -295,14 +312,19 @@ private:
                              bool ReturnsNullOnFailure = false) const;
 
   ProgramStateRef ReallocMem(CheckerContext &C, const CallExpr *CE,
-                             bool FreesMemOnFailure) const;
-  static ProgramStateRef CallocMem(CheckerContext &C, const CallExpr *CE);
+                             bool FreesMemOnFailure, 
+                             ProgramStateRef State) const;
+  static ProgramStateRef CallocMem(CheckerContext &C, const CallExpr *CE,
+                                   ProgramStateRef State);
   
   ///\brief Check if the memory associated with this symbol was released.
   bool isReleased(SymbolRef Sym, CheckerContext &C) const;
 
   bool checkUseAfterFree(SymbolRef Sym, CheckerContext &C, const Stmt *S) const;
 
+  void checkUseZeroAllocated(SymbolRef Sym, CheckerContext &C, 
+                             const Stmt *S) const;
+
   bool checkDoubleDelete(SymbolRef Sym, CheckerContext &C) const;
 
   /// Check if the function is known free memory, or if it is
@@ -330,15 +352,20 @@ private:
   /// Tells if a given family/call/symbol is tracked by the current checker.
   /// Sets CheckKind to the kind of the checker responsible for this
   /// family/call/symbol.
-  Optional<CheckKind> getCheckIfTracked(AllocationFamily Family) const;
+  Optional<CheckKind> getCheckIfTracked(AllocationFamily Family,
+                                        bool IsALeakCheck = false) const;
   Optional<CheckKind> getCheckIfTracked(CheckerContext &C,
-                                        const Stmt *AllocDeallocStmt) const;
-  Optional<CheckKind> getCheckIfTracked(CheckerContext &C, SymbolRef Sym) const;
+                                        const Stmt *AllocDeallocStmt,
+                                        bool IsALeakCheck = false) const;
+  Optional<CheckKind> getCheckIfTracked(CheckerContext &C, SymbolRef Sym, 
+                                        bool IsALeakCheck = false) const;
   ///@}
   static bool SummarizeValue(raw_ostream &os, SVal V);
   static bool SummarizeRegion(raw_ostream &os, const MemRegion *MR);
   void ReportBadFree(CheckerContext &C, SVal ArgVal, SourceRange Range, 
                      const Expr *DeallocExpr) const;
+  void ReportFreeAlloca(CheckerContext &C, SVal ArgVal,
+                        SourceRange Range) const;
   void ReportMismatchedDealloc(CheckerContext &C, SourceRange Range,
                                const Expr *DeallocExpr, const RefState *RS,
                                SymbolRef Sym, bool OwnershipTransferred) const;
@@ -352,6 +379,9 @@ private:
 
   void ReportDoubleDelete(CheckerContext &C, SymbolRef Sym) const;
 
+  void ReportUseZeroAllocated(CheckerContext &C, SourceRange Range,
+                              SymbolRef Sym) const;
+
   /// Find the location of the allocation for Sym on the path leading to the
   /// exploded node N.
   LeakInfo getAllocationSite(const ExplodedNode *N, SymbolRef Sym,
@@ -384,7 +414,7 @@ private:
     MallocBugVisitor(SymbolRef S, bool isLeak = false)
        : Sym(S), Mode(Normal), FailedReallocSymbol(nullptr), IsLeak(isLeak) {}
 
-    virtual ~MallocBugVisitor() {}
+    ~MallocBugVisitor() override {}
 
     void Profile(llvm::FoldingSetNodeID &ID) const override {
       static int X = 0;
@@ -396,7 +426,9 @@ private:
                             const Stmt *Stmt) {
       // Did not track -> allocated. Other state (released) -> allocated.
       return (Stmt && (isa<CallExpr>(Stmt) || isa<CXXNewExpr>(Stmt)) &&
-              (S && S->isAllocated()) && (!SPrev || !SPrev->isAllocated()));
+              (S && (S->isAllocated() || S->isAllocatedOfSizeZero())) && 
+              (!SPrev || !(SPrev->isAllocated() || 
+                           SPrev->isAllocatedOfSizeZero())));
     }
 
     inline bool isReleased(const RefState *S, const RefState *SPrev,
@@ -422,7 +454,9 @@ private:
       // check. If we have to handle more cases here, it might be cleaner just
       // to track this extra bit in the state itself.
       return ((!Stmt || !isa<CallExpr>(Stmt)) &&
-              (S && S->isAllocated()) && (SPrev && !SPrev->isAllocated()));
+              (S && (S->isAllocated() || S->isAllocatedOfSizeZero())) &&
+              (SPrev && !(SPrev->isAllocated() ||
+                          SPrev->isAllocatedOfSizeZero())));
     }
 
     PathDiagnosticPiece *VisitNode(const ExplodedNode *N,
@@ -497,6 +531,7 @@ public:
 void MallocChecker::initIdentifierInfo(ASTContext &Ctx) const {
   if (II_malloc)
     return;
+  II_alloca = &Ctx.Idents.get("alloca");
   II_malloc = &Ctx.Idents.get("malloc");
   II_free = &Ctx.Idents.get("free");
   II_realloc = &Ctx.Idents.get("realloc");
@@ -517,6 +552,9 @@ bool MallocChecker::isMemFunction(const FunctionDecl *FD, ASTContext &C) const {
   if (isCMemFunction(FD, C, AF_IfNameIndex, MemoryOperationKind::MOK_Any))
     return true;
 
+  if (isCMemFunction(FD, C, AF_Alloca, MemoryOperationKind::MOK_Any))
+    return true;
+
   if (isStandardNewDelete(FD, C))
     return true;
 
@@ -560,12 +598,17 @@ bool MallocChecker::isCMemFunction(const FunctionDecl *FD,
       if (FunI == II_if_nameindex)
         return true;
     }
+
+    if (Family == AF_Alloca && CheckAlloc) {
+      if (FunI == II_alloca)
+        return true;
+    }
   }
 
   if (Family != AF_Malloc)
     return false;
 
-  if (ChecksEnabled[CK_MallocOptimistic] && FD->hasAttrs()) {
+  if (IsOptimistic && FD->hasAttrs()) {
     for (const auto *I : FD->specific_attrs<OwnershipAttr>()) {
       OwnershipAttr::OwnershipKind OwnKind = I->getOwnKind();
       if(OwnKind == OwnershipAttr::Takes || OwnKind == OwnershipAttr::Holds) {
@@ -712,6 +755,8 @@ void MallocChecker::checkPostStmt(const CallExpr *CE, CheckerContext &C) const {
         return;
       if (CE->getNumArgs() < 3) {
         State = MallocMemAux(C, CE, CE->getArg(0), UndefinedVal(), State);
+        if (CE->getNumArgs() == 1)
+          State = ProcessZeroAllocation(C, CE, 0, State);
       } else if (CE->getNumArgs() == 3) {
         llvm::Optional<ProgramStateRef> MaybeState =
           performKernelMalloc(CE, C, State);
@@ -731,31 +776,43 @@ void MallocChecker::checkPostStmt(const CallExpr *CE, CheckerContext &C) const {
       if (CE->getNumArgs() < 1)
         return;
       State = MallocMemAux(C, CE, CE->getArg(0), UndefinedVal(), State);
+      State = ProcessZeroAllocation(C, CE, 0, State);
     } else if (FunI == II_realloc) {
-      State = ReallocMem(C, CE, false);
+      State = ReallocMem(C, CE, false, State);
+      State = ProcessZeroAllocation(C, CE, 1, State);
     } else if (FunI == II_reallocf) {
-      State = ReallocMem(C, CE, true);
+      State = ReallocMem(C, CE, true, State);
+      State = ProcessZeroAllocation(C, CE, 1, State);
     } else if (FunI == II_calloc) {
-      State = CallocMem(C, CE);
+      State = CallocMem(C, CE, State);
+      State = ProcessZeroAllocation(C, CE, 0, State);
+      State = ProcessZeroAllocation(C, CE, 1, State);
     } else if (FunI == II_free) {
       State = FreeMemAux(C, CE, State, 0, false, ReleasedAllocatedMemory);
     } else if (FunI == II_strdup) {
       State = MallocUpdateRefState(C, CE, State);
     } else if (FunI == II_strndup) {
       State = MallocUpdateRefState(C, CE, State);
-    }
-    else if (isStandardNewDelete(FD, C.getASTContext())) {
+    } else if (FunI == II_alloca) {
+      State = MallocMemAux(C, CE, CE->getArg(0), UndefinedVal(), State,
+                           AF_Alloca);
+      State = ProcessZeroAllocation(C, CE, 0, State);
+    } else if (isStandardNewDelete(FD, C.getASTContext())) {
       // Process direct calls to operator new/new[]/delete/delete[] functions
       // as distinct from new/new[]/delete/delete[] expressions that are 
       // processed by the checkPostStmt callbacks for CXXNewExpr and 
       // CXXDeleteExpr.
       OverloadedOperatorKind K = FD->getOverloadedOperator();
-      if (K == OO_New)
+      if (K == OO_New) {
         State = MallocMemAux(C, CE, CE->getArg(0), UndefinedVal(), State,
                              AF_CXXNew);
-      else if (K == OO_Array_New)
+        State = ProcessZeroAllocation(C, CE, 0, State);
+      }
+      else if (K == OO_Array_New) {
         State = MallocMemAux(C, CE, CE->getArg(0), UndefinedVal(), State,
                              AF_CXXNewArray);
+        State = ProcessZeroAllocation(C, CE, 0, State);
+      }
       else if (K == OO_Delete || K == OO_Array_Delete)
         State = FreeMemAux(C, CE, State, 0, false, ReleasedAllocatedMemory);
       else
@@ -770,19 +827,18 @@ void MallocChecker::checkPostStmt(const CallExpr *CE, CheckerContext &C) const {
     }
   }
 
-  if (ChecksEnabled[CK_MallocOptimistic] ||
-      ChecksEnabled[CK_MismatchedDeallocatorChecker]) {
+  if (IsOptimistic || ChecksEnabled[CK_MismatchedDeallocatorChecker]) {
     // Check all the attributes, if there are any.
     // There can be multiple of these attributes.
     if (FD->hasAttrs())
       for (const auto *I : FD->specific_attrs<OwnershipAttr>()) {
         switch (I->getOwnKind()) {
         case OwnershipAttr::Returns:
-          State = MallocMemReturnsAttr(C, CE, I);
+          State = MallocMemReturnsAttr(C, CE, I, State);
           break;
         case OwnershipAttr::Takes:
         case OwnershipAttr::Holds:
-          State = FreeMemAttr(C, CE, I);
+          State = FreeMemAttr(C, CE, I, State);
           break;
         }
       }
@@ -790,6 +846,68 @@ void MallocChecker::checkPostStmt(const CallExpr *CE, CheckerContext &C) const {
   C.addTransition(State);
 }
 
+// Performs a 0-sized allocations check.
+ProgramStateRef MallocChecker::ProcessZeroAllocation(CheckerContext &C,
+                                               const Expr *E,
+                                               const unsigned AllocationSizeArg,
+                                               ProgramStateRef State) const {
+  if (!State)
+    return nullptr;
+
+  const Expr *Arg = nullptr;
+
+  if (const CallExpr *CE = dyn_cast<CallExpr>(E)) {
+    Arg = CE->getArg(AllocationSizeArg);
+  }
+  else if (const CXXNewExpr *NE = dyn_cast<CXXNewExpr>(E)) {
+    if (NE->isArray())
+      Arg = NE->getArraySize();
+    else
+      return State;
+  }
+  else
+    llvm_unreachable("not a CallExpr or CXXNewExpr");
+
+  assert(Arg);
+
+  Optional<DefinedSVal> DefArgVal = 
+      State->getSVal(Arg, C.getLocationContext()).getAs<DefinedSVal>();
+
+  if (!DefArgVal)
+    return State;
+
+  // Check if the allocation size is 0.
+  ProgramStateRef TrueState, FalseState;
+  SValBuilder &SvalBuilder = C.getSValBuilder();
+  DefinedSVal Zero =
+      SvalBuilder.makeZeroVal(Arg->getType()).castAs<DefinedSVal>();
+
+  std::tie(TrueState, FalseState) = 
+      State->assume(SvalBuilder.evalEQ(State, *DefArgVal, Zero));
+
+  if (TrueState && !FalseState) {
+    SVal retVal = State->getSVal(E, C.getLocationContext());
+    SymbolRef Sym = retVal.getAsLocSymbol();
+    if (!Sym)
+      return State;
+
+    const RefState *RS = State->get<RegionState>(Sym);
+    if (!RS)
+      return State; // TODO: change to assert(RS); after realloc() will 
+                    // guarantee have a RegionState attached.
+
+    if (!RS->isAllocated())
+      return State;
+
+    return TrueState->set<RegionState>(Sym,
+                                       RefState::getAllocatedOfSizeZero(RS));
+  }
+
+  // Assume the value is non-zero going forward.
+  assert(FalseState);
+  return FalseState;
+}
+
 static QualType getDeepPointeeType(QualType T) {
   QualType Result = T, PointeeType = T->getPointeeType();
   while (!PointeeType.isNull()) {
@@ -849,6 +967,7 @@ void MallocChecker::checkPostStmt(const CXXNewExpr *NE,
   // existing binding.
   State = MallocUpdateRefState(C, NE, State, NE->isArray() ? AF_CXXNewArray 
                                                            : AF_CXXNew);
+  State = ProcessZeroAllocation(C, NE, 0, State);
   C.addTransition(State);
 }
 
@@ -919,15 +1038,31 @@ void MallocChecker::checkPostObjCMessage(const ObjCMethodCall &Call,
 
 ProgramStateRef
 MallocChecker::MallocMemReturnsAttr(CheckerContext &C, const CallExpr *CE,
-                                    const OwnershipAttr *Att) const {
+                                    const OwnershipAttr *Att, 
+                                    ProgramStateRef State) const {
+  if (!State)
+    return nullptr;
+
   if (Att->getModule() != II_malloc)
     return nullptr;
 
   OwnershipAttr::args_iterator I = Att->args_begin(), E = Att->args_end();
   if (I != E) {
-    return MallocMemAux(C, CE, CE->getArg(*I), UndefinedVal(), C.getState());
+    return MallocMemAux(C, CE, CE->getArg(*I), UndefinedVal(), State);
   }
-  return MallocMemAux(C, CE, UnknownVal(), UndefinedVal(), C.getState());
+  return MallocMemAux(C, CE, UnknownVal(), UndefinedVal(), State);
+}
+
+ProgramStateRef MallocChecker::MallocMemAux(CheckerContext &C,
+                                            const CallExpr *CE,
+                                            const Expr *SizeEx, SVal Init,
+                                            ProgramStateRef State,
+                                            AllocationFamily Family) {
+  if (!State)
+    return nullptr;
+
+  return MallocMemAux(C, CE, State->getSVal(SizeEx, C.getLocationContext()),
+                      Init, State, Family);
 }
 
 ProgramStateRef MallocChecker::MallocMemAux(CheckerContext &C,
@@ -935,6 +1070,8 @@ ProgramStateRef MallocChecker::MallocMemAux(CheckerContext &C,
                                            SVal Size, SVal Init,
                                            ProgramStateRef State,
                                            AllocationFamily Family) {
+  if (!State)
+    return nullptr;
 
   // We expect the malloc functions to return a pointer.
   if (!Loc::isLocType(CE->getType()))
@@ -976,6 +1113,9 @@ ProgramStateRef MallocChecker::MallocUpdateRefState(CheckerContext &C,
                                                     const Expr *E,
                                                     ProgramStateRef State,
                                                     AllocationFamily Family) {
+  if (!State)
+    return nullptr;
+
   // Get the return value.
   SVal retVal = State->getSVal(E, C.getLocationContext());
 
@@ -992,11 +1132,14 @@ ProgramStateRef MallocChecker::MallocUpdateRefState(CheckerContext &C,
 
 ProgramStateRef MallocChecker::FreeMemAttr(CheckerContext &C,
                                            const CallExpr *CE,
-                                           const OwnershipAttr *Att) const {
+                                           const OwnershipAttr *Att, 
+                                           ProgramStateRef State) const {
+  if (!State)
+    return nullptr;
+
   if (Att->getModule() != II_malloc)
     return nullptr;
 
-  ProgramStateRef State = C.getState();
   bool ReleasedAllocated = false;
 
   for (const auto &Arg : Att->args()) {
@@ -1011,15 +1154,18 @@ ProgramStateRef MallocChecker::FreeMemAttr(CheckerContext &C,
 
 ProgramStateRef MallocChecker::FreeMemAux(CheckerContext &C,
                                           const CallExpr *CE,
-                                          ProgramStateRef state,
+                                          ProgramStateRef State,
                                           unsigned Num,
                                           bool Hold,
                                           bool &ReleasedAllocated,
                                           bool ReturnsNullOnFailure) const {
+  if (!State)
+    return nullptr;
+
   if (CE->getNumArgs() < (Num + 1))
     return nullptr;
 
-  return FreeMemAux(C, CE->getArg(Num), CE, state, Hold,
+  return FreeMemAux(C, CE->getArg(Num), CE, State, Hold,
                     ReleasedAllocated, ReturnsNullOnFailure);
 }
 
@@ -1065,6 +1211,9 @@ AllocationFamily MallocChecker::getAllocationFamily(CheckerContext &C,
     if (isCMemFunction(FD, Ctx, AF_IfNameIndex, MemoryOperationKind::MOK_Any))
       return AF_IfNameIndex;
 
+    if (isCMemFunction(FD, Ctx, AF_Alloca, MemoryOperationKind::MOK_Any))
+      return AF_Alloca;
+
     return AF_None;
   }
 
@@ -1129,6 +1278,7 @@ void MallocChecker::printExpectedAllocName(raw_ostream &os, CheckerContext &C,
     case AF_CXXNew: os << "'new'"; return;
     case AF_CXXNewArray: os << "'new[]'"; return;
     case AF_IfNameIndex: os << "'if_nameindex()'"; return;
+    case AF_Alloca:
     case AF_None: llvm_unreachable("not a deallocation expression");
   }
 }
@@ -1140,7 +1290,8 @@ void MallocChecker::printExpectedDeallocName(raw_ostream &os,
     case AF_CXXNew: os << "'delete'"; return;
     case AF_CXXNewArray: os << "'delete[]'"; return;
     case AF_IfNameIndex: os << "'if_freenameindex()'"; return;
-    case AF_None: llvm_unreachable("suspicious AF_None argument");
+    case AF_Alloca:
+    case AF_None: llvm_unreachable("suspicious argument");
   }
 }
 
@@ -1152,6 +1303,9 @@ ProgramStateRef MallocChecker::FreeMemAux(CheckerContext &C,
                                           bool &ReleasedAllocated,
                                           bool ReturnsNullOnFailure) const {
 
+  if (!State)
+    return nullptr;
+
   SVal ArgVal = State->getSVal(ArgExpr, C.getLocationContext());
   if (!ArgVal.getAs<DefinedOrUnknownSVal>())
     return nullptr;
@@ -1191,8 +1345,8 @@ ProgramStateRef MallocChecker::FreeMemAux(CheckerContext &C,
   
   const MemSpaceRegion *MS = R->getMemorySpace();
   
-  // Parameters, locals, statics, globals, and memory returned by alloca() 
-  // shouldn't be freed.
+  // Parameters, locals, statics, globals, and memory returned by 
+  // __builtin_alloca() shouldn't be freed.
   if (!(isa<UnknownSpaceRegion>(MS) || isa<HeapSpaceRegion>(MS))) {
     // FIXME: at the time this code was written, malloc() regions were
     // represented by conjured symbols, which are all in UnknownSpaceRegion.
@@ -1201,8 +1355,12 @@ ProgramStateRef MallocChecker::FreeMemAux(CheckerContext &C,
     // Of course, free() can work on memory allocated outside the current
     // function, so UnknownSpaceRegion is always a possibility.
     // False negatives are better than false positives.
-    
-    ReportBadFree(C, ArgVal, ArgExpr->getSourceRange(), ParentExpr);
+
+    if (isa<AllocaRegion>(R))
+      ReportFreeAlloca(C, ArgVal, ArgExpr->getSourceRange());
+    else
+      ReportBadFree(C, ArgVal, ArgExpr->getSourceRange(), ParentExpr);
+
     return nullptr;
   }
 
@@ -1218,6 +1376,12 @@ ProgramStateRef MallocChecker::FreeMemAux(CheckerContext &C,
 
   if (RsBase) {
 
+    // Memory returned by alloca() shouldn't be freed.
+    if (RsBase->getAllocationFamily() == AF_Alloca) {
+      ReportFreeAlloca(C, ArgVal, ArgExpr->getSourceRange());
+      return nullptr;
+    }
+
     // Check for double free first.
     if ((RsBase->isReleased() || RsBase->isRelinquished()) &&
         !didPreviousFreeFail(State, SymBase, PreviousRetStatusSymbol)) {
@@ -1227,7 +1391,8 @@ ProgramStateRef MallocChecker::FreeMemAux(CheckerContext &C,
 
     // If the pointer is allocated or escaped, but we are now trying to free it,
     // check that the call to free is proper.
-    } else if (RsBase->isAllocated() || RsBase->isEscaped()) {
+    } else if (RsBase->isAllocated() || RsBase->isAllocatedOfSizeZero() || 
+               RsBase->isEscaped()) {
 
       // Check if an expected deallocation function matches the real one.
       bool DeallocMatchesAlloc =
@@ -1252,7 +1417,8 @@ ProgramStateRef MallocChecker::FreeMemAux(CheckerContext &C,
     }
   }
 
-  ReleasedAllocated = (RsBase != nullptr) && RsBase->isAllocated();
+  ReleasedAllocated = (RsBase != nullptr) && (RsBase->isAllocated() || 
+                                              RsBase->isAllocatedOfSizeZero());
 
   // Clean out the info on previous call to free return info.
   State = State->remove<FreeReturnValue>(SymBase);
@@ -1281,21 +1447,26 @@ ProgramStateRef MallocChecker::FreeMemAux(CheckerContext &C,
 }
 
 Optional<MallocChecker::CheckKind>
-MallocChecker::getCheckIfTracked(AllocationFamily Family) const {
+MallocChecker::getCheckIfTracked(AllocationFamily Family,
+                                 bool IsALeakCheck) const {
   switch (Family) {
   case AF_Malloc:
+  case AF_Alloca:
   case AF_IfNameIndex: {
-    if (ChecksEnabled[CK_MallocOptimistic]) {
-      return CK_MallocOptimistic;
-    } else if (ChecksEnabled[CK_MallocPessimistic]) {
-      return CK_MallocPessimistic;
-    }
+    if (ChecksEnabled[CK_MallocChecker])
+      return CK_MallocChecker;
+
     return Optional<MallocChecker::CheckKind>();
   }
   case AF_CXXNew:
   case AF_CXXNewArray: {
-    if (ChecksEnabled[CK_NewDeleteChecker]) {
-      return CK_NewDeleteChecker;
+    if (IsALeakCheck) {
+      if (ChecksEnabled[CK_NewDeleteLeaksChecker])
+        return CK_NewDeleteLeaksChecker;
+    } 
+    else {
+      if (ChecksEnabled[CK_NewDeleteChecker])
+        return CK_NewDeleteChecker;
     }
     return Optional<MallocChecker::CheckKind>();
   }
@@ -1308,16 +1479,18 @@ MallocChecker::getCheckIfTracked(AllocationFamily Family) const {
 
 Optional<MallocChecker::CheckKind>
 MallocChecker::getCheckIfTracked(CheckerContext &C,
-                                 const Stmt *AllocDeallocStmt) const {
-  return getCheckIfTracked(getAllocationFamily(C, AllocDeallocStmt));
+                                 const Stmt *AllocDeallocStmt,
+                                 bool IsALeakCheck) const {
+  return getCheckIfTracked(getAllocationFamily(C, AllocDeallocStmt),
+                           IsALeakCheck);
 }
 
 Optional<MallocChecker::CheckKind>
-MallocChecker::getCheckIfTracked(CheckerContext &C, SymbolRef Sym) const {
-
+MallocChecker::getCheckIfTracked(CheckerContext &C, SymbolRef Sym,
+                                 bool IsALeakCheck) const {
   const RefState *RS = C.getState()->get<RegionState>(Sym);
   assert(RS);
-  return getCheckIfTracked(RS->getAllocationFamily());
+  return getCheckIfTracked(RS->getAllocationFamily(), IsALeakCheck);
 }
 
 bool MallocChecker::SummarizeValue(raw_ostream &os, SVal V) {
@@ -1411,8 +1584,7 @@ void MallocChecker::ReportBadFree(CheckerContext &C, SVal ArgVal,
                                   SourceRange Range, 
                                   const Expr *DeallocExpr) const {
 
-  if (!ChecksEnabled[CK_MallocOptimistic] &&
-      !ChecksEnabled[CK_MallocPessimistic] &&
+  if (!ChecksEnabled[CK_MallocChecker] &&
       !ChecksEnabled[CK_NewDeleteChecker])
     return;
 
@@ -1433,23 +1605,19 @@ void MallocChecker::ReportBadFree(CheckerContext &C, SVal ArgVal,
     while (const ElementRegion *ER = dyn_cast_or_null<ElementRegion>(MR))
       MR = ER->getSuperRegion();
 
-    if (MR && isa<AllocaRegion>(MR))
-      os << "Memory allocated by alloca() should not be deallocated";
-    else {
-      os << "Argument to ";
-      if (!printAllocDeallocName(os, C, DeallocExpr))
-        os << "deallocator";
-
-      os << " is ";
-      bool Summarized = MR ? SummarizeRegion(os, MR) 
-                           : SummarizeValue(os, ArgVal);
-      if (Summarized)
-        os << ", which is not memory allocated by ";
-      else
-        os << "not memory allocated by ";
+    os << "Argument to ";
+    if (!printAllocDeallocName(os, C, DeallocExpr))
+      os << "deallocator";
 
-      printExpectedAllocName(os, C, DeallocExpr);
-    }
+    os << " is ";
+    bool Summarized = MR ? SummarizeRegion(os, MR) 
+                         : SummarizeValue(os, ArgVal);
+    if (Summarized)
+      os << ", which is not memory allocated by ";
+    else
+      os << "not memory allocated by ";
+
+    printExpectedAllocName(os, C, DeallocExpr);
 
     BugReport *R = new BugReport(*BT_BadFree[*CheckKind], os.str(), N);
     R->markInteresting(MR);
@@ -1458,6 +1626,31 @@ void MallocChecker::ReportBadFree(CheckerContext &C, SVal ArgVal,
   }
 }
 
+void MallocChecker::ReportFreeAlloca(CheckerContext &C, SVal ArgVal, 
+                                     SourceRange Range) const {
+
+  Optional<MallocChecker::CheckKind> CheckKind;
+
+  if (ChecksEnabled[CK_MallocChecker])
+    CheckKind = CK_MallocChecker;
+  else if (ChecksEnabled[CK_MismatchedDeallocatorChecker])
+    CheckKind = CK_MismatchedDeallocatorChecker;
+  else
+    return;
+
+  if (ExplodedNode *N = C.generateSink()) {
+    if (!BT_FreeAlloca[*CheckKind])
+      BT_FreeAlloca[*CheckKind].reset(
+          new BugType(CheckNames[*CheckKind], "Free alloca()", "Memory Error"));
+
+    BugReport *R = new BugReport(*BT_FreeAlloca[*CheckKind], 
+                 "Memory allocated by alloca() should not be deallocated", N);
+    R->markInteresting(ArgVal.getAsRegion());
+    R->addRange(Range);
+    C.emitReport(R);
+  }
+}
+
 void MallocChecker::ReportMismatchedDealloc(CheckerContext &C, 
                                             SourceRange Range,
                                             const Expr *DeallocExpr, 
@@ -1517,8 +1710,8 @@ void MallocChecker::ReportOffsetFree(CheckerContext &C, SVal ArgVal,
                                      SourceRange Range, const Expr *DeallocExpr,
                                      const Expr *AllocExpr) const {
 
-  if (!ChecksEnabled[CK_MallocOptimistic] &&
-      !ChecksEnabled[CK_MallocPessimistic] &&
+
+  if (!ChecksEnabled[CK_MallocChecker] &&
       !ChecksEnabled[CK_NewDeleteChecker])
     return;
 
@@ -1573,8 +1766,7 @@ void MallocChecker::ReportOffsetFree(CheckerContext &C, SVal ArgVal,
 void MallocChecker::ReportUseAfterFree(CheckerContext &C, SourceRange Range,
                                        SymbolRef Sym) const {
 
-  if (!ChecksEnabled[CK_MallocOptimistic] &&
-      !ChecksEnabled[CK_MallocPessimistic] &&
+  if (!ChecksEnabled[CK_MallocChecker] &&
       !ChecksEnabled[CK_NewDeleteChecker])
     return;
 
@@ -1601,8 +1793,7 @@ void MallocChecker::ReportDoubleFree(CheckerContext &C, SourceRange Range,
                                      bool Released, SymbolRef Sym, 
                                      SymbolRef PrevSym) const {
 
-  if (!ChecksEnabled[CK_MallocOptimistic] &&
-      !ChecksEnabled[CK_MallocPessimistic] &&
+  if (!ChecksEnabled[CK_MallocChecker] &&
       !ChecksEnabled[CK_NewDeleteChecker])
     return;
 
@@ -1637,7 +1828,6 @@ void MallocChecker::ReportDoubleDelete(CheckerContext &C, SymbolRef Sym) const {
   Optional<MallocChecker::CheckKind> CheckKind = getCheckIfTracked(C, Sym);
   if (!CheckKind.hasValue())
     return;
-  assert(*CheckKind == CK_NewDeleteChecker && "invalid check kind");
 
   if (ExplodedNode *N = C.generateSink()) {
     if (!BT_DoubleDelete)
@@ -1653,16 +1843,49 @@ void MallocChecker::ReportDoubleDelete(CheckerContext &C, SymbolRef Sym) const {
   }
 }
 
+void MallocChecker::ReportUseZeroAllocated(CheckerContext &C,
+                                           SourceRange Range,
+                                           SymbolRef Sym) const {
+
+  if (!ChecksEnabled[CK_MallocChecker] &&
+      !ChecksEnabled[CK_NewDeleteChecker])
+    return;
+
+  Optional<MallocChecker::CheckKind> CheckKind = getCheckIfTracked(C, Sym);
+
+  if (!CheckKind.hasValue())
+    return;
+
+  if (ExplodedNode *N = C.generateSink()) {
+    if (!BT_UseZerroAllocated[*CheckKind])
+      BT_UseZerroAllocated[*CheckKind].reset(new BugType(
+          CheckNames[*CheckKind], "Use of zero allocated", "Memory Error"));
+
+    BugReport *R = new BugReport(*BT_UseZerroAllocated[*CheckKind],
+                                 "Use of zero-allocated memory", N);
+
+    R->addRange(Range);
+    if (Sym) {
+      R->markInteresting(Sym);
+      R->addVisitor(llvm::make_unique<MallocBugVisitor>(Sym));
+    }
+    C.emitReport(R);
+  }
+}
+
 ProgramStateRef MallocChecker::ReallocMem(CheckerContext &C,
                                           const CallExpr *CE,
-                                          bool FreesOnFail) const {
+                                          bool FreesOnFail,
+                                          ProgramStateRef State) const {
+  if (!State)
+    return nullptr;
+
   if (CE->getNumArgs() < 2)
     return nullptr;
 
-  ProgramStateRef state = C.getState();
   const Expr *arg0Expr = CE->getArg(0);
   const LocationContext *LCtx = C.getLocationContext();
-  SVal Arg0Val = state->getSVal(arg0Expr, LCtx);
+  SVal Arg0Val = State->getSVal(arg0Expr, LCtx);
   if (!Arg0Val.getAs<DefinedOrUnknownSVal>())
     return nullptr;
   DefinedOrUnknownSVal arg0Val = Arg0Val.castAs<DefinedOrUnknownSVal>();
@@ -1670,7 +1893,7 @@ ProgramStateRef MallocChecker::ReallocMem(CheckerContext &C,
   SValBuilder &svalBuilder = C.getSValBuilder();
 
   DefinedOrUnknownSVal PtrEQ =
-    svalBuilder.evalEQ(state, arg0Val, svalBuilder.makeNull());
+    svalBuilder.evalEQ(State, arg0Val, svalBuilder.makeNull());
 
   // Get the size argument. If there is no size arg then give up.
   const Expr *Arg1 = CE->getArg(1);
@@ -1678,20 +1901,20 @@ ProgramStateRef MallocChecker::ReallocMem(CheckerContext &C,
     return nullptr;
 
   // Get the value of the size argument.
-  SVal Arg1ValG = state->getSVal(Arg1, LCtx);
+  SVal Arg1ValG = State->getSVal(Arg1, LCtx);
   if (!Arg1ValG.getAs<DefinedOrUnknownSVal>())
     return nullptr;
   DefinedOrUnknownSVal Arg1Val = Arg1ValG.castAs<DefinedOrUnknownSVal>();
 
   // Compare the size argument to 0.
   DefinedOrUnknownSVal SizeZero =
-    svalBuilder.evalEQ(state, Arg1Val,
+    svalBuilder.evalEQ(State, Arg1Val,
                        svalBuilder.makeIntValWithPtrWidth(0, false));
 
   ProgramStateRef StatePtrIsNull, StatePtrNotNull;
-  std::tie(StatePtrIsNull, StatePtrNotNull) = state->assume(PtrEQ);
+  std::tie(StatePtrIsNull, StatePtrNotNull) = State->assume(PtrEQ);
   ProgramStateRef StateSizeIsZero, StateSizeNotZero;
-  std::tie(StateSizeIsZero, StateSizeNotZero) = state->assume(SizeZero);
+  std::tie(StateSizeIsZero, StateSizeNotZero) = State->assume(SizeZero);
   // We only assume exceptional states if they are definitely true; if the
   // state is under-constrained, assume regular realloc behavior.
   bool PrtIsNull = StatePtrIsNull && !StatePtrNotNull;
@@ -1711,7 +1934,7 @@ ProgramStateRef MallocChecker::ReallocMem(CheckerContext &C,
   // Get the from and to pointer symbols as in toPtr = realloc(fromPtr, size).
   assert(!PrtIsNull);
   SymbolRef FromPtr = arg0Val.getAsSymbol();
-  SVal RetVal = state->getSVal(CE, LCtx);
+  SVal RetVal = State->getSVal(CE, LCtx);
   SymbolRef ToPtr = RetVal.getAsSymbol();
   if (!FromPtr || !ToPtr)
     return nullptr;
@@ -1731,7 +1954,7 @@ ProgramStateRef MallocChecker::ReallocMem(CheckerContext &C,
 
   // Default behavior.
   if (ProgramStateRef stateFree =
-        FreeMemAux(C, CE, state, 0, false, ReleasedAllocated)) {
+        FreeMemAux(C, CE, State, 0, false, ReleasedAllocated)) {
 
     ProgramStateRef stateRealloc = MallocMemAux(C, CE, CE->getArg(1),
                                                 UnknownVal(), stateFree);
@@ -1755,20 +1978,23 @@ ProgramStateRef MallocChecker::ReallocMem(CheckerContext &C,
   return nullptr;
 }
 
-ProgramStateRef MallocChecker::CallocMem(CheckerContext &C, const CallExpr *CE){
+ProgramStateRef MallocChecker::CallocMem(CheckerContext &C, const CallExpr *CE, 
+                                         ProgramStateRef State) {
+  if (!State)
+    return nullptr;
+
   if (CE->getNumArgs() < 2)
     return nullptr;
 
-  ProgramStateRef state = C.getState();
   SValBuilder &svalBuilder = C.getSValBuilder();
   const LocationContext *LCtx = C.getLocationContext();
-  SVal count = state->getSVal(CE->getArg(0), LCtx);
-  SVal elementSize = state->getSVal(CE->getArg(1), LCtx);
-  SVal TotalSize = svalBuilder.evalBinOp(state, BO_Mul, count, elementSize,
+  SVal count = State->getSVal(CE->getArg(0), LCtx);
+  SVal elementSize = State->getSVal(CE->getArg(1), LCtx);
+  SVal TotalSize = svalBuilder.evalBinOp(State, BO_Mul, count, elementSize,
                                         svalBuilder.getContext().getSizeType());  
   SVal zeroVal = svalBuilder.makeZeroVal(svalBuilder.getContext().CharTy);
 
-  return MallocMemAux(C, CE, TotalSize, zeroVal, state);
+  return MallocMemAux(C, CE, TotalSize, zeroVal, State);
 }
 
 LeakInfo
@@ -1801,9 +2027,11 @@ MallocChecker::getAllocationSite(const ExplodedNode *N, SymbolRef Sym,
         }
       }
 
-    // Allocation node, is the last node in the current context in which the
-    // symbol was tracked.
-    if (N->getLocationContext() == LeakContext)
+    // Allocation node, is the last node in the current or parent context in
+    // which the symbol was tracked.
+    const LocationContext *NContext = N->getLocationContext();
+    if (NContext == LeakContext ||
+        NContext->isParentOf(LeakContext))
       AllocNode = N;
     N = N->pred_empty() ? nullptr : *(N->pred_begin());
   }
@@ -1814,23 +2042,22 @@ MallocChecker::getAllocationSite(const ExplodedNode *N, SymbolRef Sym,
 void MallocChecker::reportLeak(SymbolRef Sym, ExplodedNode *N,
                                CheckerContext &C) const {
 
-  if (!ChecksEnabled[CK_MallocOptimistic] &&
-      !ChecksEnabled[CK_MallocPessimistic] &&
+  if (!ChecksEnabled[CK_MallocChecker] &&
       !ChecksEnabled[CK_NewDeleteLeaksChecker])
     return;
 
   const RefState *RS = C.getState()->get<RegionState>(Sym);
   assert(RS && "cannot leak an untracked symbol");
   AllocationFamily Family = RS->getAllocationFamily();
-  Optional<MallocChecker::CheckKind> CheckKind = getCheckIfTracked(Family);
-  if (!CheckKind.hasValue())
+
+  if (Family == AF_Alloca)
     return;
 
-  // Special case for new and new[]; these are controlled by a separate checker
-  // flag so that they can be selectively disabled.
-  if (Family == AF_CXXNew || Family == AF_CXXNewArray)
-    if (!ChecksEnabled[CK_NewDeleteLeaksChecker])
-      return;
+  Optional<MallocChecker::CheckKind>
+      CheckKind = getCheckIfTracked(Family, true);
+
+  if (!CheckKind.hasValue())
+    return;
 
   assert(N);
   if (!BT_Leak[*CheckKind]) {
@@ -1893,7 +2120,7 @@ void MallocChecker::checkDeadSymbols(SymbolReaper &SymReaper,
   SmallVector<SymbolRef, 2> Errors;
   for (RegionStateTy::iterator I = RS.begin(), E = RS.end(); I != E; ++I) {
     if (SymReaper.isDead(I->first)) {
-      if (I->second.isAllocated())
+      if (I->second.isAllocated() || I->second.isAllocatedOfSizeZero())
         Errors.push_back(I->first);
       // Remove the dead symbol from the map.
       RS = F.remove(RS, I->first);
@@ -1949,8 +2176,7 @@ void MallocChecker::checkPreCall(const CallEvent &Call,
       return;
 
     ASTContext &Ctx = C.getASTContext();
-    if ((ChecksEnabled[CK_MallocOptimistic] ||
-         ChecksEnabled[CK_MallocPessimistic]) &&
+    if (ChecksEnabled[CK_MallocChecker] &&
         (isCMemFunction(FD, Ctx, AF_Malloc, MemoryOperationKind::MOK_Free) ||
          isCMemFunction(FD, Ctx, AF_IfNameIndex,
                         MemoryOperationKind::MOK_Free)))
@@ -2062,6 +2288,15 @@ bool MallocChecker::checkUseAfterFree(SymbolRef Sym, CheckerContext &C,
   return false;
 }
 
+void MallocChecker::checkUseZeroAllocated(SymbolRef Sym, CheckerContext &C,
+                                          const Stmt *S) const {
+  assert(Sym);
+  const RefState *RS = C.getState()->get<RegionState>(Sym);
+
+  if (RS && RS->isAllocatedOfSizeZero())
+    ReportUseZeroAllocated(C, RS->getStmt()->getSourceRange(), Sym);
+}
+
 bool MallocChecker::checkDoubleDelete(SymbolRef Sym, CheckerContext &C) const {
 
   if (isReleased(Sym, C)) {
@@ -2075,8 +2310,10 @@ bool MallocChecker::checkDoubleDelete(SymbolRef Sym, CheckerContext &C) const {
 void MallocChecker::checkLocation(SVal l, bool isLoad, const Stmt *S,
                                   CheckerContext &C) const {
   SymbolRef Sym = l.getLocSymbolInBase();
-  if (Sym)
+  if (Sym) {
     checkUseAfterFree(Sym, C, S);
+    checkUseZeroAllocated(Sym, C, S);
+  }
 }
 
 // If a symbolic region is assumed to NULL (or another constant), stop tracking
@@ -2320,7 +2557,8 @@ ProgramStateRef MallocChecker::checkPointerEscapeAux(ProgramStateRef State,
       continue;
     
     if (const RefState *RS = State->get<RegionState>(sym)) {
-      if (RS->isAllocated() && CheckRefState(RS)) {
+      if ((RS->isAllocated() || RS->isAllocatedOfSizeZero()) &&
+          CheckRefState(RS)) {
         State = State->remove<RegionState>(sym);
         State = State->set<RegionState>(sym, RefState::getEscaped(RS));
       }
@@ -2443,6 +2681,8 @@ void MallocChecker::printState(raw_ostream &Out, ProgramStateRef State,
       const RefState *RefS = State->get<RegionState>(I.getKey());
       AllocationFamily Family = RefS->getAllocationFamily();
       Optional<MallocChecker::CheckKind> CheckKind = getCheckIfTracked(Family);
+      if (!CheckKind.hasValue())
+         CheckKind = getCheckIfTracked(Family, true);
 
       I.getKey()->dumpToStream(Out);
       Out << " : ";
@@ -2457,6 +2697,8 @@ void MallocChecker::printState(raw_ostream &Out, ProgramStateRef State,
 void ento::registerNewDeleteLeaksChecker(CheckerManager &mgr) {
   registerCStringCheckerBasic(mgr);
   MallocChecker *checker = mgr.registerChecker<MallocChecker>();
+  checker->IsOptimistic = mgr.getAnalyzerOptions().getBooleanOption(
+      "Optimistic", false, checker);
   checker->ChecksEnabled[MallocChecker::CK_NewDeleteLeaksChecker] = true;
   checker->CheckNames[MallocChecker::CK_NewDeleteLeaksChecker] =
       mgr.getCurrentCheckName();
@@ -2470,11 +2712,12 @@ void ento::registerNewDeleteLeaksChecker(CheckerManager &mgr) {
   void ento::register##name(CheckerManager &mgr) {                             \
     registerCStringCheckerBasic(mgr);                                          \
     MallocChecker *checker = mgr.registerChecker<MallocChecker>();             \
+    checker->IsOptimistic = mgr.getAnalyzerOptions().getBooleanOption(         \
+        "Optimistic", false, checker);                                         \
     checker->ChecksEnabled[MallocChecker::CK_##name] = true;                   \
     checker->CheckNames[MallocChecker::CK_##name] = mgr.getCurrentCheckName(); \
   }
 
-REGISTER_CHECKER(MallocPessimistic)
-REGISTER_CHECKER(MallocOptimistic)
+REGISTER_CHECKER(MallocChecker)
 REGISTER_CHECKER(NewDeleteChecker)
 REGISTER_CHECKER(MismatchedDeallocatorChecker)
diff --git a/lib/StaticAnalyzer/Checkers/MallocOverflowSecurityChecker.cpp b/lib/StaticAnalyzer/Checkers/MallocOverflowSecurityChecker.cpp
index f38ce77dc6b6..e91347999dc1 100644
--- a/lib/StaticAnalyzer/Checkers/MallocOverflowSecurityChecker.cpp
+++ b/lib/StaticAnalyzer/Checkers/MallocOverflowSecurityChecker.cpp
@@ -142,13 +142,13 @@ private:
           }
         }
       }
-      else if (isa<MemberExpr>(E)) {
+      else if (const auto *ME = dyn_cast<MemberExpr>(E)) {
         // No points-to analysis, just look at the member
-        const Decl * EmeMD = dyn_cast<MemberExpr>(E)->getMemberDecl();
+        const Decl *EmeMD = ME->getMemberDecl();
         while (i != e) {
           --i;
-          if (isa<MemberExpr>(i->variable)) {
-            if (dyn_cast<MemberExpr>(i->variable)->getMemberDecl() == EmeMD)
+          if (const auto *ME_i = dyn_cast<MemberExpr>(i->variable)) {
+            if (ME_i->getMemberDecl() == EmeMD)
               i = toScanFor.erase (i);
           }
         }
diff --git a/lib/StaticAnalyzer/Checkers/RetainCountChecker.cpp b/lib/StaticAnalyzer/Checkers/RetainCountChecker.cpp
index 9a460ba50ddd..58c27d49acce 100644
--- a/lib/StaticAnalyzer/Checkers/RetainCountChecker.cpp
+++ b/lib/StaticAnalyzer/Checkers/RetainCountChecker.cpp
@@ -94,6 +94,16 @@ public:
     ErrorReturnedNotOwned
   };
 
+  /// Tracks how an object referenced by an ivar has been used.
+  ///
+  /// This accounts for us not knowing if an arbitrary ivar is supposed to be
+  /// stored at +0 or +1.
+  enum class IvarAccessHistory {
+    None,
+    AccessedDirectly,
+    ReleasedAfterDirectAccess
+  };
+
 private:
   /// The number of outstanding retains.
   unsigned Cnt;
@@ -121,14 +131,16 @@ private:
   /// This setting should not be propagated to state derived from this state.
   /// Once we start deriving new states, it would be inconsistent to override
   /// them.
-  unsigned IsOverridable : 1;
+  unsigned RawIvarAccessHistory : 2;
 
   RefVal(Kind k, RetEffect::ObjKind o, unsigned cnt, unsigned acnt, QualType t,
-         bool Overridable = false)
+         IvarAccessHistory IvarAccess)
     : Cnt(cnt), ACnt(acnt), T(t), RawKind(static_cast<unsigned>(k)),
-      RawObjectKind(static_cast<unsigned>(o)), IsOverridable(Overridable) {
+      RawObjectKind(static_cast<unsigned>(o)),
+      RawIvarAccessHistory(static_cast<unsigned>(IvarAccess)) {
     assert(getKind() == k && "not enough bits for the kind");
     assert(getObjKind() == o && "not enough bits for the object kind");
+    assert(getIvarAccessHistory() == IvarAccess && "not enough bits");
   }
 
 public:
@@ -144,20 +156,24 @@ public:
   void clearCounts() {
     Cnt = 0;
     ACnt = 0;
-    IsOverridable = false;
   }
   void setCount(unsigned i) {
     Cnt = i;
-    IsOverridable = false;
   }
   void setAutoreleaseCount(unsigned i) {
     ACnt = i;
-    IsOverridable = false;
   }
 
   QualType getType() const { return T; }
 
-  bool isOverridable() const { return IsOverridable; }
+  /// Returns what the analyzer knows about direct accesses to a particular
+  /// instance variable.
+  ///
+  /// If the object with this refcount wasn't originally from an Objective-C
+  /// ivar region, this should always return IvarAccessHistory::None.
+  IvarAccessHistory getIvarAccessHistory() const {
+    return static_cast<IvarAccessHistory>(RawIvarAccessHistory);
+  }
 
   bool isOwned() const {
     return getKind() == Owned;
@@ -181,7 +197,7 @@ public:
   /// Most commonly, this is an owned object with a retain count of +1.
   static RefVal makeOwned(RetEffect::ObjKind o, QualType t,
                           unsigned Count = 1) {
-    return RefVal(Owned, o, Count, 0, t);
+    return RefVal(Owned, o, Count, 0, t, IvarAccessHistory::None);
   }
 
   /// Create a state for an object whose lifetime is not the responsibility of
@@ -190,47 +206,49 @@ public:
   /// Most commonly, this is an unowned object with a retain count of +0.
   static RefVal makeNotOwned(RetEffect::ObjKind o, QualType t,
                              unsigned Count = 0) {
-    return RefVal(NotOwned, o, Count, 0, t);
-  }
-
-  /// Create an "overridable" state for an unowned object at +0.
-  ///
-  /// An overridable state is one that provides a good approximation of the
-  /// reference counting state now, but which may be discarded later if the
-  /// checker sees the object being used in new ways.
-  static RefVal makeOverridableNotOwned(RetEffect::ObjKind o, QualType t) {
-    return RefVal(NotOwned, o, 0, 0, t, /*Overridable=*/true);
+    return RefVal(NotOwned, o, Count, 0, t, IvarAccessHistory::None);
   }
 
   RefVal operator-(size_t i) const {
     return RefVal(getKind(), getObjKind(), getCount() - i,
-                  getAutoreleaseCount(), getType());
+                  getAutoreleaseCount(), getType(), getIvarAccessHistory());
   }
 
   RefVal operator+(size_t i) const {
     return RefVal(getKind(), getObjKind(), getCount() + i,
-                  getAutoreleaseCount(), getType());
+                  getAutoreleaseCount(), getType(), getIvarAccessHistory());
   }
 
   RefVal operator^(Kind k) const {
     return RefVal(k, getObjKind(), getCount(), getAutoreleaseCount(),
-                  getType());
+                  getType(), getIvarAccessHistory());
   }
 
   RefVal autorelease() const {
     return RefVal(getKind(), getObjKind(), getCount(), getAutoreleaseCount()+1,
-                  getType());
+                  getType(), getIvarAccessHistory());
+  }
+
+  RefVal withIvarAccess() const {
+    assert(getIvarAccessHistory() == IvarAccessHistory::None);
+    return RefVal(getKind(), getObjKind(), getCount(), getAutoreleaseCount(),
+                  getType(), IvarAccessHistory::AccessedDirectly);
+  }
+  RefVal releaseViaIvar() const {
+    assert(getIvarAccessHistory() == IvarAccessHistory::AccessedDirectly);
+    return RefVal(getKind(), getObjKind(), getCount(), getAutoreleaseCount(),
+                  getType(), IvarAccessHistory::ReleasedAfterDirectAccess);
   }
 
   // Comparison, profiling, and pretty-printing.
 
   bool hasSameState(const RefVal &X) const {
-    return getKind() == X.getKind() && Cnt == X.Cnt && ACnt == X.ACnt;
+    return getKind() == X.getKind() && Cnt == X.Cnt && ACnt == X.ACnt &&
+           getIvarAccessHistory() == X.getIvarAccessHistory();
   }
 
   bool operator==(const RefVal& X) const {
-    return T == X.T && hasSameState(X) && getObjKind() == X.getObjKind() &&
-           IsOverridable == X.IsOverridable;
+    return T == X.T && hasSameState(X) && getObjKind() == X.getObjKind();
   }
   
   void Profile(llvm::FoldingSetNodeID& ID) const {
@@ -239,7 +257,7 @@ public:
     ID.AddInteger(Cnt);
     ID.AddInteger(ACnt);
     ID.AddInteger(RawObjectKind);
-    ID.AddBoolean(IsOverridable);
+    ID.AddInteger(RawIvarAccessHistory);
   }
 
   void print(raw_ostream &Out) const;
@@ -249,9 +267,6 @@ void RefVal::print(raw_ostream &Out) const {
   if (!T.isNull())
     Out << "Tracked " << T.getAsString() << '/';
 
-  if (isOverridable())
-    Out << "(overridable) ";
-
   switch (getKind()) {
     default: llvm_unreachable("Invalid RefVal kind");
     case Owned: {
@@ -323,8 +338,18 @@ void RefVal::print(raw_ostream &Out) const {
       break;
   }
 
+  switch (getIvarAccessHistory()) {
+  case IvarAccessHistory::None:
+    break;
+  case IvarAccessHistory::AccessedDirectly:
+    Out << " [direct ivar access]";
+    break;
+  case IvarAccessHistory::ReleasedAfterDirectAccess:
+    Out << " [released after direct ivar access]";
+  }
+
   if (ACnt) {
-    Out << " [ARC +" << ACnt << ']';
+    Out << " [autorelease -" << ACnt << ']';
   }
 }
 } //end anonymous namespace
@@ -1763,12 +1788,11 @@ namespace {
       addGCModeDescription(LOpts, GCEnabled);
     }
 
-    std::pair<ranges_iterator, ranges_iterator> getRanges() override {
+    llvm::iterator_range<ranges_iterator> getRanges() override {
       const CFRefBug& BugTy = static_cast<CFRefBug&>(getBugType());
       if (!BugTy.isLeak())
         return BugReport::getRanges();
-      else
-        return std::make_pair(ranges_iterator(), ranges_iterator());
+      return llvm::make_range(ranges_iterator(), ranges_iterator());
     }
   };
 
@@ -1829,6 +1853,16 @@ static bool isNumericLiteralExpression(const Expr *E) {
          isa<CXXBoolLiteralExpr>(E);
 }
 
+/// Returns true if this stack frame is for an Objective-C method that is a
+/// property getter or setter whose body has been synthesized by the analyzer.
+static bool isSynthesizedAccessor(const StackFrameContext *SFC) {
+  auto Method = dyn_cast_or_null<ObjCMethodDecl>(SFC->getDecl());
+  if (!Method || !Method->isPropertyAccessor())
+    return false;
+
+  return SFC->getAnalysisDeclContext()->isBodyAutosynthesized();
+}
+
 PathDiagnosticPiece *CFRefReportVisitor::VisitNode(const ExplodedNode *N,
                                                    const ExplodedNode *PrevN,
                                                    BugReporterContext &BRC,
@@ -1859,6 +1893,11 @@ PathDiagnosticPiece *CFRefReportVisitor::VisitNode(const ExplodedNode *N,
   if (!PrevT) {
     const Stmt *S = N->getLocation().castAs<StmtPoint>().getStmt();
 
+    if (isa<ObjCIvarRefExpr>(S) &&
+        isSynthesizedAccessor(LCtx->getCurrentStackFrame())) {
+      S = LCtx->getCurrentStackFrame()->getCallSite();
+    }
+
     if (isa<ObjCArrayLiteral>(S)) {
       os << "NSArray literal is an object with a +0 retain count";
     }
@@ -1883,6 +1922,9 @@ PathDiagnosticPiece *CFRefReportVisitor::VisitNode(const ExplodedNode *N,
         os << "oxed expression produces an object with a +0 retain count";
       }
     }
+    else if (isa<ObjCIvarRefExpr>(S)) {
+      os << "Object loaded from instance variable";
+    }
     else {      
       if (const CallExpr *CE = dyn_cast<CallExpr>(S)) {
         // Get the name of the callee (if it is available).
@@ -2034,7 +2076,6 @@ PathDiagnosticPiece *CFRefReportVisitor::VisitNode(const ExplodedNode *N,
       switch (CurrV.getKind()) {
         case RefVal::Owned:
         case RefVal::NotOwned:
-
           if (PrevV.getCount() == CurrV.getCount()) {
             // Did an autorelease message get sent?
             if (PrevV.getAutoreleaseCount() == CurrV.getAutoreleaseCount())
@@ -2062,6 +2103,11 @@ PathDiagnosticPiece *CFRefReportVisitor::VisitNode(const ExplodedNode *N,
           break;
 
         case RefVal::Released:
+          if (CurrV.getIvarAccessHistory() ==
+                RefVal::IvarAccessHistory::ReleasedAfterDirectAccess &&
+              CurrV.getIvarAccessHistory() != PrevV.getIvarAccessHistory()) {
+            os << "Strong instance variable relinquished. ";
+          }
           os << "Object released.";
           break;
 
@@ -2143,7 +2189,7 @@ static AllocationInfo
 GetAllocationSite(ProgramStateManager& StateMgr, const ExplodedNode *N,
                   SymbolRef Sym) {
   const ExplodedNode *AllocationNode = N;
-  const ExplodedNode *AllocationNodeInCurrentContext = N;
+  const ExplodedNode *AllocationNodeInCurrentOrParentContext = N;
   const MemRegion *FirstBinding = nullptr;
   const LocationContext *LeakContext = N->getLocationContext();
 
@@ -2173,10 +2219,15 @@ GetAllocationSite(ProgramStateManager& StateMgr, const ExplodedNode *N,
     // AllocationNode is the last node in which the symbol was tracked.
     AllocationNode = N;
 
-    // AllocationNodeInCurrentContext, is the last node in the current context
-    // in which the symbol was tracked.
-    if (NContext == LeakContext)
-      AllocationNodeInCurrentContext = N;
+    // AllocationNodeInCurrentContext, is the last node in the current or
+    // parent context in which the symbol was tracked.
+    //
+    // Note that the allocation site might be in the parent conext. For example,
+    // the case where an allocation happens in a block that captures a reference
+    // to it and that reference is overwritten/dropped by another call to
+    // the block.
+    if (NContext == LeakContext || NContext->isParentOf(LeakContext))
+      AllocationNodeInCurrentOrParentContext = N;
 
     // Find the last init that was called on the given symbol and store the
     // init method's location context.
@@ -2214,7 +2265,7 @@ GetAllocationSite(ProgramStateManager& StateMgr, const ExplodedNode *N,
     FirstBinding = nullptr;
   }
 
-  return AllocationInfo(AllocationNodeInCurrentContext,
+  return AllocationInfo(AllocationNodeInCurrentOrParentContext,
                         FirstBinding,
                         InterestingMethodContext);
 }
@@ -2345,20 +2396,8 @@ CFRefLeakReport::CFRefLeakReport(CFRefBug &D, const LangOptions &LOpts,
   ProgramPoint P = AllocNode->getLocation();
   if (Optional<CallExitEnd> Exit = P.getAs<CallExitEnd>())
     AllocStmt = Exit->getCalleeContext()->getCallSite();
-  else {
-    // We are going to get a BlockEdge when the leak and allocation happen in
-    // different, non-nested frames (contexts). For example, the case where an
-    // allocation happens in a block that captures a reference to it and
-    // that reference is overwritten/dropped by another call to the block.
-    if (Optional<BlockEdge> Edge = P.getAs<BlockEdge>()) {
-      if (Optional<CFGStmt> St = Edge->getDst()->front().getAs<CFGStmt>()) {
-        AllocStmt = St->getStmt();
-      }
-    }
-    else {
-      AllocStmt = P.castAs<PostStmt>().getStmt();
-    }
-  }
+  else
+    AllocStmt = P.castAs<PostStmt>().getStmt();
   assert(AllocStmt && "Cannot find allocation statement");
 
   PathDiagnosticLocation AllocLocation =
@@ -2436,9 +2475,7 @@ public:
     : ShouldResetSummaryLog(false),
       IncludeAllocationLine(shouldIncludeAllocationSiteInLeakDiagnostics(AO)) {}
 
-  virtual ~RetainCountChecker() {
-    DeleteContainerSeconds(DeadSymbolTags);
-  }
+  ~RetainCountChecker() override { DeleteContainerSeconds(DeadSymbolTags); }
 
   void checkEndAnalysis(ExplodedGraph &G, BugReporter &BR,
                         ExprEngine &Eng) const {
@@ -2774,17 +2811,64 @@ void RetainCountChecker::checkPostStmt(const ObjCBoxedExpr *Ex,
   C.addTransition(State);
 }
 
+static bool wasLoadedFromIvar(SymbolRef Sym) {
+  if (auto DerivedVal = dyn_cast<SymbolDerived>(Sym))
+    return isa<ObjCIvarRegion>(DerivedVal->getRegion());
+  if (auto RegionVal = dyn_cast<SymbolRegionValue>(Sym))
+    return isa<ObjCIvarRegion>(RegionVal->getRegion());
+  return false;
+}
+
 void RetainCountChecker::checkPostStmt(const ObjCIvarRefExpr *IRE,
                                        CheckerContext &C) const {
+  Optional<Loc> IVarLoc = C.getSVal(IRE).getAs<Loc>();
+  if (!IVarLoc)
+    return;
+
   ProgramStateRef State = C.getState();
-  // If an instance variable was previously accessed through a property,
-  // it may have a synthesized refcount of +0. Override right now that we're
-  // doing direct access.
-  if (Optional<Loc> IVarLoc = C.getSVal(IRE).getAs<Loc>())
-    if (SymbolRef Sym = State->getSVal(*IVarLoc).getAsSymbol())
-      if (const RefVal *RV = getRefBinding(State, Sym))
-        if (RV->isOverridable())
-          State = removeRefBinding(State, Sym);
+  SymbolRef Sym = State->getSVal(*IVarLoc).getAsSymbol();
+  if (!Sym || !wasLoadedFromIvar(Sym))
+    return;
+
+  // Accessing an ivar directly is unusual. If we've done that, be more
+  // forgiving about what the surrounding code is allowed to do.
+
+  QualType Ty = Sym->getType();
+  RetEffect::ObjKind Kind;
+  if (Ty->isObjCRetainableType())
+    Kind = RetEffect::ObjC;
+  else if (coreFoundation::isCFObjectRef(Ty))
+    Kind = RetEffect::CF;
+  else
+    return;
+
+  // If the value is already known to be nil, don't bother tracking it.
+  ConstraintManager &CMgr = State->getConstraintManager();
+  if (CMgr.isNull(State, Sym).isConstrainedTrue())
+    return;
+
+  if (const RefVal *RV = getRefBinding(State, Sym)) {
+    // If we've seen this symbol before, or we're only seeing it now because
+    // of something the analyzer has synthesized, don't do anything.
+    if (RV->getIvarAccessHistory() != RefVal::IvarAccessHistory::None ||
+        isSynthesizedAccessor(C.getStackFrame())) {
+      return;
+    }
+
+    // Note that this value has been loaded from an ivar.
+    C.addTransition(setRefBinding(State, Sym, RV->withIvarAccess()));
+    return;
+  }
+
+  RefVal PlusZero = RefVal::makeNotOwned(Kind, Ty);
+
+  // In a synthesized accessor, the effective retain count is +0.
+  if (isSynthesizedAccessor(C.getStackFrame())) {
+    C.addTransition(setRefBinding(State, Sym, PlusZero));
+    return;
+  }
+
+  State = setRefBinding(State, Sym, PlusZero.withIvarAccess());
   C.addTransition(State);
 }
 
@@ -2828,16 +2912,6 @@ static QualType GetReturnType(const Expr *RetE, ASTContext &Ctx) {
   return RetTy;
 }
 
-static bool wasSynthesizedProperty(const ObjCMethodCall *Call,
-                                   ExplodedNode *N) {
-  if (!Call || !Call->getDecl()->isPropertyAccessor())
-    return false;
-
-  CallExitEnd PP = N->getLocation().castAs<CallExitEnd>();
-  const StackFrameContext *Frame = PP.getCalleeContext();
-  return Frame->getAnalysisDeclContext()->isBodyAutosynthesized();
-}
-
 // We don't always get the exact modeling of the function with regards to the
 // retain count checker even when the function is inlined. For example, we need
 // to stop tracking the symbols which were marked with StopTrackingHard.
@@ -2872,19 +2946,6 @@ void RetainCountChecker::processSummaryOfInlined(const RetainSummary &Summ,
     SymbolRef Sym = CallOrMsg.getReturnValue().getAsSymbol();
     if (Sym)
       state = removeRefBinding(state, Sym);
-  } else if (RE.getKind() == RetEffect::NotOwnedSymbol) {
-    if (wasSynthesizedProperty(MsgInvocation, C.getPredecessor())) {
-      // Believe the summary if we synthesized the body of a property getter
-      // and the return value is currently untracked. If the corresponding
-      // instance variable is later accessed directly, however, we're going to
-      // want to override this state, so that the owning object can perform
-      // reference counting operations on its own ivars.
-      SymbolRef Sym = CallOrMsg.getReturnValue().getAsSymbol();
-      if (Sym && !getRefBinding(state, Sym))
-        state = setRefBinding(state, Sym,
-                              RefVal::makeOverridableNotOwned(RE.getObjKind(),
-                                                              Sym->getType()));
-    }
   }
   
   C.addTransition(state);
@@ -3125,11 +3186,16 @@ RetainCountChecker::updateSymbol(ProgramStateRef state, SymbolRef sym,
 
         case RefVal::Owned:
           assert(V.getCount() > 0);
-          if (V.getCount() == 1)
-            V = V ^ (E == DecRefBridgedTransferred ? RefVal::NotOwned
-                                                   : RefVal::Released);
-          else if (E == DecRefAndStopTrackingHard)
+          if (V.getCount() == 1) {
+            if (E == DecRefBridgedTransferred ||
+                V.getIvarAccessHistory() ==
+                  RefVal::IvarAccessHistory::AccessedDirectly)
+              V = V ^ RefVal::NotOwned;
+            else
+              V = V ^ RefVal::Released;
+          } else if (E == DecRefAndStopTrackingHard) {
             return removeRefBinding(state, sym);
+          }
 
           V = V - 1;
           break;
@@ -3139,6 +3205,13 @@ RetainCountChecker::updateSymbol(ProgramStateRef state, SymbolRef sym,
             if (E == DecRefAndStopTrackingHard)
               return removeRefBinding(state, sym);
             V = V - 1;
+          } else if (V.getIvarAccessHistory() ==
+                       RefVal::IvarAccessHistory::AccessedDirectly) {
+            // Assume that the instance variable was holding on the object at
+            // +1, and we just didn't know.
+            if (E == DecRefAndStopTrackingHard)
+              return removeRefBinding(state, sym);
+            V = V.releaseViaIvar() ^ RefVal::Released;
           } else {
             V = V ^ RefVal::ErrorReleaseNotOwned;
             hasErr = V.getKind();
@@ -3162,6 +3235,16 @@ void RetainCountChecker::processNonLeakError(ProgramStateRef St,
                                              RefVal::Kind ErrorKind,
                                              SymbolRef Sym,
                                              CheckerContext &C) const {
+  // HACK: Ignore retain-count issues on values accessed through ivars,
+  // because of cases like this:
+  //   [_contentView retain];
+  //   [_contentView removeFromSuperview];
+  //   [self addSubview:_contentView]; // invalidates 'self'
+  //   [_contentView release];
+  if (const RefVal *RV = getRefBinding(St, Sym))
+    if (RV->getIvarAccessHistory() != RefVal::IvarAccessHistory::None)
+      return;
+
   ExplodedNode *N = C.generateSink(St);
   if (!N)
     return;
@@ -3229,7 +3312,7 @@ bool RetainCountChecker::evalCall(const CallExpr *CE, CheckerContext &C) const {
   // See if it's one of the specific functions we know how to eval.
   bool canEval = false;
 
-  QualType ResultTy = CE->getCallReturnType();
+  QualType ResultTy = CE->getCallReturnType(C.getASTContext());
   if (ResultTy->isObjCIdType()) {
     // Handle: id NSMakeCollectable(CFTypeRef)
     canEval = II->isStr("NSMakeCollectable");
@@ -3388,6 +3471,15 @@ void RetainCountChecker::checkReturnWithRetEffect(const ReturnStmt *S,
                                                   RetEffect RE, RefVal X,
                                                   SymbolRef Sym,
                                               ProgramStateRef state) const {
+  // HACK: Ignore retain-count issues on values accessed through ivars,
+  // because of cases like this:
+  //   [_contentView retain];
+  //   [_contentView removeFromSuperview];
+  //   [self addSubview:_contentView]; // invalidates 'self'
+  //   [_contentView release];
+  if (X.getIvarAccessHistory() != RefVal::IvarAccessHistory::None)
+    return;
+
   // Any leaks or other errors?
   if (X.isReturnedOwned() && X.getCount() == 0) {
     if (RE.getKind() != RetEffect::NoRet) {
@@ -3428,22 +3520,31 @@ void RetainCountChecker::checkReturnWithRetEffect(const ReturnStmt *S,
     }
   } else if (X.isReturnedNotOwned()) {
     if (RE.isOwned()) {
-      // Trying to return a not owned object to a caller expecting an
-      // owned object.
-      state = setRefBinding(state, Sym, X ^ RefVal::ErrorReturnedNotOwned);
-
-      static CheckerProgramPointTag ReturnNotOwnedTag(this, 
-                                                      "ReturnNotOwnedForOwned");
-      ExplodedNode *N = C.addTransition(state, Pred, &ReturnNotOwnedTag);
-      if (N) {
-        if (!returnNotOwnedForOwned)
-          returnNotOwnedForOwned.reset(new ReturnedNotOwnedForOwned(this));
-
-        CFRefReport *report =
-            new CFRefReport(*returnNotOwnedForOwned,
-                            C.getASTContext().getLangOpts(), 
-                            C.isObjCGCEnabled(), SummaryLog, N, Sym);
-        C.emitReport(report);
+      if (X.getIvarAccessHistory() ==
+            RefVal::IvarAccessHistory::AccessedDirectly) {
+        // Assume the method was trying to transfer a +1 reference from a
+        // strong ivar to the caller.
+        state = setRefBinding(state, Sym,
+                              X.releaseViaIvar() ^ RefVal::ReturnedOwned);
+      } else {
+        // Trying to return a not owned object to a caller expecting an
+        // owned object.
+        state = setRefBinding(state, Sym, X ^ RefVal::ErrorReturnedNotOwned);
+
+        static CheckerProgramPointTag
+            ReturnNotOwnedTag(this, "ReturnNotOwnedForOwned");
+
+        ExplodedNode *N = C.addTransition(state, Pred, &ReturnNotOwnedTag);
+        if (N) {
+          if (!returnNotOwnedForOwned)
+            returnNotOwnedForOwned.reset(new ReturnedNotOwnedForOwned(this));
+
+          CFRefReport *report =
+              new CFRefReport(*returnNotOwnedForOwned,
+                              C.getASTContext().getLangOpts(), 
+                              C.isObjCGCEnabled(), SummaryLog, N, Sym);
+          C.emitReport(report);
+        }
       }
     }
   }
@@ -3594,6 +3695,14 @@ RetainCountChecker::handleAutoreleaseCounts(ProgramStateRef state,
   if (V.getKind() == RefVal::ReturnedOwned)
     ++Cnt;
 
+  // If we would over-release here, but we know the value came from an ivar,
+  // assume it was a strong ivar that's just been relinquished.
+  if (ACnt > Cnt &&
+      V.getIvarAccessHistory() == RefVal::IvarAccessHistory::AccessedDirectly) {
+    V = V.releaseViaIvar();
+    --ACnt;
+  }
+
   if (ACnt <= Cnt) {
     if (ACnt == Cnt) {
       V.clearCounts();
@@ -3608,6 +3717,15 @@ RetainCountChecker::handleAutoreleaseCounts(ProgramStateRef state,
     return setRefBinding(state, Sym, V);
   }
 
+  // HACK: Ignore retain-count issues on values accessed through ivars,
+  // because of cases like this:
+  //   [_contentView retain];
+  //   [_contentView removeFromSuperview];
+  //   [self addSubview:_contentView]; // invalidates 'self'
+  //   [_contentView release];
+  if (V.getIvarAccessHistory() != RefVal::IvarAccessHistory::None)
+    return state;
+
   // Woah!  More autorelease counts then retain counts left.
   // Emit hard error.
   V = V ^ RefVal::ErrorOverAutorelease;
@@ -3641,11 +3759,22 @@ ProgramStateRef
 RetainCountChecker::handleSymbolDeath(ProgramStateRef state,
                                       SymbolRef sid, RefVal V,
                                     SmallVectorImpl<SymbolRef> &Leaked) const {
-  bool hasLeak = false;
-  if (V.isOwned())
+  bool hasLeak;
+
+  // HACK: Ignore retain-count issues on values accessed through ivars,
+  // because of cases like this:
+  //   [_contentView retain];
+  //   [_contentView removeFromSuperview];
+  //   [self addSubview:_contentView]; // invalidates 'self'
+  //   [_contentView release];
+  if (V.getIvarAccessHistory() != RefVal::IvarAccessHistory::None)
+    hasLeak = false;
+  else if (V.isOwned())
     hasLeak = true;
   else if (V.isNotOwned() || V.isReturnedOwned())
     hasLeak = (V.getCount() > 0);
+  else
+    hasLeak = false;
 
   if (!hasLeak)
     return removeRefBinding(state, sid);
diff --git a/lib/StaticAnalyzer/Core/AnalyzerOptions.cpp b/lib/StaticAnalyzer/Core/AnalyzerOptions.cpp
index d717e3fe86d2..1696bcfac9c1 100644
--- a/lib/StaticAnalyzer/Core/AnalyzerOptions.cpp
+++ b/lib/StaticAnalyzer/Core/AnalyzerOptions.cpp
@@ -13,12 +13,14 @@
 //===----------------------------------------------------------------------===//
 
 #include "clang/StaticAnalyzer/Core/AnalyzerOptions.h"
+#include "clang/StaticAnalyzer/Core/Checker.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/StringSwitch.h"
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/raw_ostream.h"
 
 using namespace clang;
+using namespace ento;
 using namespace llvm;
 
 AnalyzerOptions::UserModeKind AnalyzerOptions::getUserMode() {
@@ -100,12 +102,37 @@ AnalyzerOptions::mayInlineCXXMemberFunction(CXXInlineableMemberKind K) {
 
 static StringRef toString(bool b) { return b ? "true" : "false"; }
 
-bool AnalyzerOptions::getBooleanOption(StringRef Name, bool DefaultVal) {
+StringRef AnalyzerOptions::getCheckerOption(StringRef CheckerName,
+                                            StringRef OptionName,
+                                            StringRef Default,
+                                            bool SearchInParents) {
+  // Search for a package option if the option for the checker is not specified
+  // and search in parents is enabled.
+  ConfigTable::const_iterator E = Config.end();
+  do {
+    ConfigTable::const_iterator I =
+        Config.find((Twine(CheckerName) + ":" + OptionName).str());
+    if (I != E)
+      return StringRef(I->getValue());
+    size_t Pos = CheckerName.rfind('.');
+    if (Pos == StringRef::npos)
+      return Default;
+    CheckerName = CheckerName.substr(0, Pos);
+  } while (!CheckerName.empty() && SearchInParents);
+  return Default;
+}
+
+bool AnalyzerOptions::getBooleanOption(StringRef Name, bool DefaultVal,
+                                       const CheckerBase *C,
+                                       bool SearchInParents) {
   // FIXME: We should emit a warning here if the value is something other than
   // "true", "false", or the empty string (meaning the default value),
   // but the AnalyzerOptions doesn't have access to a diagnostic engine.
+  StringRef Default = toString(DefaultVal);
   StringRef V =
-      Config.insert(std::make_pair(Name, toString(DefaultVal))).first->second;
+      C ? getCheckerOption(C->getTagDescription(), Name, Default,
+                           SearchInParents)
+        : StringRef(Config.insert(std::make_pair(Name, Default)).first->second);
   return llvm::StringSwitch<bool>(V)
       .Case("true", true)
       .Case("false", false)
@@ -113,9 +140,10 @@ bool AnalyzerOptions::getBooleanOption(StringRef Name, bool DefaultVal) {
 }
 
 bool AnalyzerOptions::getBooleanOption(Optional<bool> &V, StringRef Name,
-                                       bool DefaultVal) {
+                                       bool DefaultVal, const CheckerBase *C,
+                                       bool SearchInParents) {
   if (!V.hasValue())
-    V = getBooleanOption(Name, DefaultVal);
+    V = getBooleanOption(Name, DefaultVal, C, SearchInParents);
   return V.getValue();
 }
 
@@ -199,19 +227,35 @@ bool AnalyzerOptions::shouldWriteStableReportFilename() {
                           /* Default = */ false);
 }
 
-int AnalyzerOptions::getOptionAsInteger(StringRef Name, int DefaultVal) {
+int AnalyzerOptions::getOptionAsInteger(StringRef Name, int DefaultVal,
+                                        const CheckerBase *C,
+                                        bool SearchInParents) {
   SmallString<10> StrBuf;
   llvm::raw_svector_ostream OS(StrBuf);
   OS << DefaultVal;
 
-  StringRef V = Config.insert(std::make_pair(Name, OS.str())).first->second;
+  StringRef V = C ? getCheckerOption(C->getTagDescription(), Name, OS.str(),
+                                     SearchInParents)
+                  : StringRef(Config.insert(std::make_pair(Name, OS.str()))
+                                  .first->second);
+
   int Res = DefaultVal;
   bool b = V.getAsInteger(10, Res);
   assert(!b && "analyzer-config option should be numeric");
-  (void) b;
+  (void)b;
   return Res;
 }
 
+StringRef AnalyzerOptions::getOptionAsString(StringRef Name,
+                                             StringRef DefaultVal,
+                                             const CheckerBase *C,
+                                             bool SearchInParents) {
+  return C ? getCheckerOption(C->getTagDescription(), Name, DefaultVal,
+                              SearchInParents)
+           : StringRef(
+                 Config.insert(std::make_pair(Name, DefaultVal)).first->second);
+}
+
 unsigned AnalyzerOptions::getAlwaysInlineSize() {
   if (!AlwaysInlineSize.hasValue())
     AlwaysInlineSize = getOptionAsInteger("ipa-always-inline-size", 3);
@@ -281,4 +325,3 @@ bool AnalyzerOptions::shouldPrunePaths() {
 bool AnalyzerOptions::shouldConditionalizeStaticInitializers() {
   return getBooleanOption("cfg-conditional-static-initializers", true);
 }
-
diff --git a/lib/StaticAnalyzer/Core/BugReporter.cpp b/lib/StaticAnalyzer/Core/BugReporter.cpp
index dff81e383ea6..97e97ef8c4d6 100644
--- a/lib/StaticAnalyzer/Core/BugReporter.cpp
+++ b/lib/StaticAnalyzer/Core/BugReporter.cpp
@@ -2702,22 +2702,22 @@ const Stmt *BugReport::getStmt() const {
   return S;
 }
 
-std::pair<BugReport::ranges_iterator, BugReport::ranges_iterator>
-BugReport::getRanges() {
-    // If no custom ranges, add the range of the statement corresponding to
-    // the error node.
-    if (Ranges.empty()) {
-      if (const Expr *E = dyn_cast_or_null<Expr>(getStmt()))
-        addRange(E->getSourceRange());
-      else
-        return std::make_pair(ranges_iterator(), ranges_iterator());
-    }
+llvm::iterator_range<BugReport::ranges_iterator> BugReport::getRanges() {
+  // If no custom ranges, add the range of the statement corresponding to
+  // the error node.
+  if (Ranges.empty()) {
+    if (const Expr *E = dyn_cast_or_null<Expr>(getStmt()))
+      addRange(E->getSourceRange());
+    else
+      return llvm::make_range(ranges_iterator(), ranges_iterator());
+  }
 
-    // User-specified absence of range info.
-    if (Ranges.size() == 1 && !Ranges.begin()->isValid())
-      return std::make_pair(ranges_iterator(), ranges_iterator());
+  // User-specified absence of range info.
+  if (Ranges.size() == 1 && !Ranges.begin()->isValid())
+    return llvm::make_range(ranges_iterator(), ranges_iterator());
 
-    return std::make_pair(Ranges.begin(), Ranges.end());
+  return llvm::iterator_range<BugReport::ranges_iterator>(Ranges.begin(),
+                                                          Ranges.end());
 }
 
 PathDiagnosticLocation BugReport::getLocation(const SourceManager &SM) const {
@@ -2763,9 +2763,7 @@ void BugReporter::FlushReports() {
   // warnings and new BugTypes.
   // FIXME: Only NSErrorChecker needs BugType's FlushReports.
   // Turn NSErrorChecker into a proper checker and remove this.
-  SmallVector<const BugType*, 16> bugTypes;
-  for (BugTypesTy::iterator I=BugTypes.begin(), E=BugTypes.end(); I!=E; ++I)
-    bugTypes.push_back(*I);
+  SmallVector<const BugType *, 16> bugTypes(BugTypes.begin(), BugTypes.end());
   for (SmallVectorImpl<const BugType *>::iterator
          I = bugTypes.begin(), E = bugTypes.end(); I != E; ++I)
     const_cast<BugType*>(*I)->FlushReports(*this);
@@ -3055,8 +3053,7 @@ static void CompactPathDiagnostic(PathPieces &path, const SourceManager& SM) {
   // Now take the pieces and construct a new PathDiagnostic.
   path.clear();
 
-  for (PiecesTy::iterator I=Pieces.begin(), E=Pieces.end(); I!=E; ++I)
-    path.push_back(*I);
+  path.insert(path.end(), Pieces.begin(), Pieces.end());
 }
 
 bool GRBugReporter::generatePathDiagnostic(PathDiagnostic& PD,
@@ -3434,10 +3431,8 @@ void BugReporter::FlushReport(BugReport *exampleReport,
     PathDiagnosticLocation L = exampleReport->getLocation(getSourceManager());
     auto piece = llvm::make_unique<PathDiagnosticEventPiece>(
         L, exampleReport->getDescription());
-    BugReport::ranges_iterator Beg, End;
-    std::tie(Beg, End) = exampleReport->getRanges();
-    for ( ; Beg != End; ++Beg)
-      piece->addRange(*Beg);
+    for (const SourceRange &Range : exampleReport->getRanges())
+      piece->addRange(Range);
     D->setEndOfPath(std::move(piece));
   }
 
diff --git a/lib/StaticAnalyzer/Core/BugReporterVisitors.cpp b/lib/StaticAnalyzer/Core/BugReporterVisitors.cpp
index 2d56bd088497..b906cc9e7624 100644
--- a/lib/StaticAnalyzer/Core/BugReporterVisitors.cpp
+++ b/lib/StaticAnalyzer/Core/BugReporterVisitors.cpp
@@ -111,15 +111,14 @@ std::unique_ptr<PathDiagnosticPiece> BugReporterVisitor::getDefaultEndPath(
   PathDiagnosticLocation L =
     PathDiagnosticLocation::createEndOfPath(EndPathNode,BRC.getSourceManager());
 
-  BugReport::ranges_iterator Beg, End;
-  std::tie(Beg, End) = BR.getRanges();
+  const auto &Ranges = BR.getRanges();
 
   // Only add the statement itself as a range if we didn't specify any
   // special ranges for this report.
-  auto P = llvm::make_unique<PathDiagnosticEventPiece>(L, BR.getDescription(),
-                                                       Beg == End);
-  for (; Beg != End; ++Beg)
-    P->addRange(*Beg);
+  auto P = llvm::make_unique<PathDiagnosticEventPiece>(
+      L, BR.getDescription(), Ranges.begin() == Ranges.end());
+  for (const SourceRange &Range : Ranges)
+    P->addRange(Range);
 
   return std::move(P);
 }
diff --git a/lib/StaticAnalyzer/Core/Checker.cpp b/lib/StaticAnalyzer/Core/Checker.cpp
index 1a3965acaf42..22352111d14a 100644
--- a/lib/StaticAnalyzer/Core/Checker.cpp
+++ b/lib/StaticAnalyzer/Core/Checker.cpp
@@ -36,11 +36,3 @@ raw_ostream& clang::ento::operator<<(raw_ostream &Out,
   Out << Checker.getCheckName().getName();
   return Out;
 }
-
-void Checker<check::_VoidCheck, check::_VoidCheck, check::_VoidCheck,
-             check::_VoidCheck, check::_VoidCheck, check::_VoidCheck,
-             check::_VoidCheck, check::_VoidCheck, check::_VoidCheck,
-             check::_VoidCheck, check::_VoidCheck, check::_VoidCheck,
-             check::_VoidCheck, check::_VoidCheck, check::_VoidCheck,
-             check::_VoidCheck, check::_VoidCheck, check::_VoidCheck
-             >::anchor() { }
diff --git a/lib/StaticAnalyzer/Core/ExprEngine.cpp b/lib/StaticAnalyzer/Core/ExprEngine.cpp
index 4699df8819bd..8b7f18fdf6ac 100644
--- a/lib/StaticAnalyzer/Core/ExprEngine.cpp
+++ b/lib/StaticAnalyzer/Core/ExprEngine.cpp
@@ -1901,6 +1901,9 @@ void ExprEngine::VisitLvalArraySubscriptExpr(const ArraySubscriptExpr *A,
   getCheckerManager().runCheckersForPreStmt(checkerPreStmt, Pred, A, *this);
 
   StmtNodeBuilder Bldr(checkerPreStmt, Dst, *currBldrCtx);
+  assert(A->isGLValue() ||
+          (!AMgr.getLangOpts().CPlusPlus &&
+           A->getType().isCForbiddenLValueType()));
 
   for (ExplodedNodeSet::iterator it = checkerPreStmt.begin(),
                                  ei = checkerPreStmt.end(); it != ei; ++it) {
@@ -1909,7 +1912,6 @@ void ExprEngine::VisitLvalArraySubscriptExpr(const ArraySubscriptExpr *A,
     SVal V = state->getLValue(A->getType(),
                               state->getSVal(Idx, LCtx),
                               state->getSVal(Base, LCtx));
-    assert(A->isGLValue());
     Bldr.generateNode(A, *it, state->BindExpr(A, LCtx, V), nullptr,
                       ProgramPoint::PostLValueKind);
   }
@@ -2646,17 +2648,6 @@ struct DOTGraphTraits<ExplodedNode*> :
 } // end llvm namespace
 #endif
 
-#ifndef NDEBUG
-template <typename ITERATOR>
-ExplodedNode *GetGraphNode(ITERATOR I) { return *I; }
-
-template <> ExplodedNode*
-GetGraphNode<llvm::DenseMap<ExplodedNode*, Expr*>::iterator>
-  (llvm::DenseMap<ExplodedNode*, Expr*>::iterator I) {
-  return I->first;
-}
-#endif
-
 void ExprEngine::ViewGraph(bool trim) {
 #ifndef NDEBUG
   if (trim) {
diff --git a/lib/StaticAnalyzer/Core/ExprEngineC.cpp b/lib/StaticAnalyzer/Core/ExprEngineC.cpp
index ffda52709dc2..1777ea97a402 100644
--- a/lib/StaticAnalyzer/Core/ExprEngineC.cpp
+++ b/lib/StaticAnalyzer/Core/ExprEngineC.cpp
@@ -31,7 +31,7 @@ void ExprEngine::VisitBinaryOperator(const BinaryOperator* B,
   ExplodedNodeSet Tmp2;
   getCheckerManager().runCheckersForPreStmt(CheckedSet, Pred, B, *this);
     
-  // With both the LHS and RHS evaluated, process the operation itself.    
+  // With both the LHS and RHS evaluated, process the operation itself.
   for (ExplodedNodeSet::iterator it=CheckedSet.begin(), ei=CheckedSet.end();
          it != ei; ++it) {
       
diff --git a/lib/StaticAnalyzer/Core/HTMLDiagnostics.cpp b/lib/StaticAnalyzer/Core/HTMLDiagnostics.cpp
index 88b5464d44be..3c1a3b4111d1 100644
--- a/lib/StaticAnalyzer/Core/HTMLDiagnostics.cpp
+++ b/lib/StaticAnalyzer/Core/HTMLDiagnostics.cpp
@@ -45,7 +45,7 @@ class HTMLDiagnostics : public PathDiagnosticConsumer {
 public:
   HTMLDiagnostics(AnalyzerOptions &AnalyzerOpts, const std::string& prefix, const Preprocessor &pp);
 
-  virtual ~HTMLDiagnostics() { FlushDiagnostics(nullptr); }
+  ~HTMLDiagnostics() override { FlushDiagnostics(nullptr); }
 
   void FlushDiagnosticsImpl(std::vector<const PathDiagnostic *> &Diags,
                             FilesMade *filesMade) override;
@@ -282,7 +282,7 @@ void HTMLDiagnostics::ReportDiag(const PathDiagnostic& D,
       llvm::sys::path::append(Model, Directory, "report-%%%%%%.html");
 
       if (std::error_code EC =
-          llvm::sys::fs::createUniqueFile(Model.str(), FD, ResultPath)) {
+          llvm::sys::fs::createUniqueFile(Model, FD, ResultPath)) {
           llvm::errs() << "warning: could not create file in '" << Directory
                        << "': " << EC.message() << '\n';
           return;
@@ -302,12 +302,12 @@ void HTMLDiagnostics::ReportDiag(const PathDiagnostic& D,
                    << "-" << i << ".html";
           llvm::sys::path::append(Model, Directory,
                                   filename.str());
-          EC = llvm::sys::fs::openFileForWrite(Model.str(),
+          EC = llvm::sys::fs::openFileForWrite(Model,
                                                FD,
                                                llvm::sys::fs::F_RW |
                                                llvm::sys::fs::F_Excl);
           if (EC && EC != std::errc::file_exists) {
-              llvm::errs() << "warning: could not create file '" << Model.str()
+              llvm::errs() << "warning: could not create file '" << Model
                            << "': " << EC.message() << '\n';
               return;
           }
diff --git a/lib/StaticAnalyzer/Core/MemRegion.cpp b/lib/StaticAnalyzer/Core/MemRegion.cpp
index 76cead623298..1fa675433b56 100644
--- a/lib/StaticAnalyzer/Core/MemRegion.cpp
+++ b/lib/StaticAnalyzer/Core/MemRegion.cpp
@@ -1372,10 +1372,11 @@ void BlockDataRegion::LazyInitializeReferencedVars() {
     return;
 
   AnalysisDeclContext *AC = getCodeRegion()->getAnalysisDeclContext();
-  AnalysisDeclContext::referenced_decls_iterator I, E;
-  std::tie(I, E) = AC->getReferencedBlockVars(BC->getDecl());
+  const auto &ReferencedBlockVars = AC->getReferencedBlockVars(BC->getDecl());
+  auto NumBlockVars =
+      std::distance(ReferencedBlockVars.begin(), ReferencedBlockVars.end());
 
-  if (I == E) {
+  if (NumBlockVars == 0) {
     ReferencedVars = (void*) 0x1;
     return;
   }
@@ -1386,14 +1387,14 @@ void BlockDataRegion::LazyInitializeReferencedVars() {
 
   typedef BumpVector<const MemRegion*> VarVec;
   VarVec *BV = (VarVec*) A.Allocate<VarVec>();
-  new (BV) VarVec(BC, E - I);
+  new (BV) VarVec(BC, NumBlockVars);
   VarVec *BVOriginal = (VarVec*) A.Allocate<VarVec>();
-  new (BVOriginal) VarVec(BC, E - I);
+  new (BVOriginal) VarVec(BC, NumBlockVars);
 
-  for ( ; I != E; ++I) {
+  for (const VarDecl *VD : ReferencedBlockVars) {
     const VarRegion *VR = nullptr;
     const VarRegion *OriginalVR = nullptr;
-    std::tie(VR, OriginalVR) = getCaptureRegions(*I);
+    std::tie(VR, OriginalVR) = getCaptureRegions(VD);
     assert(VR);
     assert(OriginalVR);
     BV->push_back(VR, BC);
diff --git a/lib/StaticAnalyzer/Core/PathDiagnostic.cpp b/lib/StaticAnalyzer/Core/PathDiagnostic.cpp
index b971fff642d4..c4900313cad4 100644
--- a/lib/StaticAnalyzer/Core/PathDiagnostic.cpp
+++ b/lib/StaticAnalyzer/Core/PathDiagnostic.cpp
@@ -432,11 +432,15 @@ void PathDiagnosticConsumer::FlushDiagnostics(
 
   // Sort the diagnostics so that they are always emitted in a deterministic
   // order.
-  if (!BatchDiags.empty())
-    std::sort(BatchDiags.begin(), BatchDiags.end(),
-              [](const PathDiagnostic *X, const PathDiagnostic *Y) {
-      return X != Y && compare(*X, *Y);
-    });
+  int (*Comp)(const PathDiagnostic *const *, const PathDiagnostic *const *) =
+      [](const PathDiagnostic *const *X, const PathDiagnostic *const *Y) {
+        assert(*X != *Y && "PathDiagnostics not uniqued!");
+        if (compare(**X, **Y))
+          return -1;
+        assert(compare(**Y, **X) && "Not a total order!");
+        return 1;
+      };
+  array_pod_sort(BatchDiags.begin(), BatchDiags.end(), Comp);
 
   FlushDiagnosticsImpl(BatchDiags, Files);
 
@@ -452,7 +456,7 @@ void PathDiagnosticConsumer::FlushDiagnostics(
 }
 
 PathDiagnosticConsumer::FilesMade::~FilesMade() {
-  for (PDFileEntry &Entry : *this)
+  for (PDFileEntry &Entry : Set)
     Entry.~PDFileEntry();
 }
 
@@ -462,11 +466,11 @@ void PathDiagnosticConsumer::FilesMade::addDiagnostic(const PathDiagnostic &PD,
   llvm::FoldingSetNodeID NodeID;
   NodeID.Add(PD);
   void *InsertPos;
-  PDFileEntry *Entry = FindNodeOrInsertPos(NodeID, InsertPos);
+  PDFileEntry *Entry = Set.FindNodeOrInsertPos(NodeID, InsertPos);
   if (!Entry) {
     Entry = Alloc.Allocate<PDFileEntry>();
     Entry = new (Entry) PDFileEntry(NodeID);
-    InsertNode(Entry, InsertPos);
+    Set.InsertNode(Entry, InsertPos);
   }
   
   // Allocate persistent storage for the file name.
@@ -483,7 +487,7 @@ PathDiagnosticConsumer::FilesMade::getFiles(const PathDiagnostic &PD) {
   llvm::FoldingSetNodeID NodeID;
   NodeID.Add(PD);
   void *InsertPos;
-  PDFileEntry *Entry = FindNodeOrInsertPos(NodeID, InsertPos);
+  PDFileEntry *Entry = Set.FindNodeOrInsertPos(NodeID, InsertPos);
   if (!Entry)
     return nullptr;
   return &Entry->files;
diff --git a/lib/StaticAnalyzer/Core/PlistDiagnostics.cpp b/lib/StaticAnalyzer/Core/PlistDiagnostics.cpp
index a2c66f881482..e0aff589e053 100644
--- a/lib/StaticAnalyzer/Core/PlistDiagnostics.cpp
+++ b/lib/StaticAnalyzer/Core/PlistDiagnostics.cpp
@@ -37,7 +37,7 @@ namespace {
                      const LangOptions &LangOpts,
                      bool supportsMultipleFiles);
 
-    virtual ~PlistDiagnostics() {}
+    ~PlistDiagnostics() override {}
 
     void FlushDiagnosticsImpl(std::vector<const PathDiagnostic *> &Diags,
                               FilesMade *filesMade) override;
@@ -106,13 +106,14 @@ static void ReportControlFlow(raw_ostream &o,
     // by forcing to use only the beginning of the range.  This simplifies the layout
     // logic for clients.
     Indent(o, indent) << "<key>start</key>\n";
-    SourceLocation StartEdge = I->getStart().asRange().getBegin();
-    EmitRange(o, SM, LangOpts, CharSourceRange::getTokenRange(StartEdge), FM,
+    SourceRange StartEdge(
+        SM.getExpansionLoc(I->getStart().asRange().getBegin()));
+    EmitRange(o, SM, Lexer::getAsCharRange(StartEdge, SM, LangOpts), FM,
               indent + 1);
 
     Indent(o, indent) << "<key>end</key>\n";
-    SourceLocation EndEdge = I->getEnd().asRange().getBegin();
-    EmitRange(o, SM, LangOpts, CharSourceRange::getTokenRange(EndEdge), FM,
+    SourceRange EndEdge(SM.getExpansionLoc(I->getEnd().asRange().getBegin()));
+    EmitRange(o, SM, Lexer::getAsCharRange(EndEdge, SM, LangOpts), FM,
               indent + 1);
 
     --indent;
@@ -154,7 +155,7 @@ static void ReportEvent(raw_ostream &o, const PathDiagnosticPiece& P,
   FullSourceLoc L = P.getLocation().asLocation();
 
   Indent(o, indent) << "<key>location</key>\n";
-  EmitLocation(o, SM, LangOpts, L, FM, indent);
+  EmitLocation(o, SM, L, FM, indent);
 
   // Output the ranges (if any).
   ArrayRef<SourceRange> Ranges = P.getRanges();
@@ -163,11 +164,10 @@ static void ReportEvent(raw_ostream &o, const PathDiagnosticPiece& P,
     Indent(o, indent) << "<key>ranges</key>\n";
     Indent(o, indent) << "<array>\n";
     ++indent;
-    for (ArrayRef<SourceRange>::iterator I = Ranges.begin(), E = Ranges.end();
-         I != E; ++I) {
-      EmitRange(o, SM, LangOpts, CharSourceRange::getTokenRange(*I), FM,
-                indent + 1);
-    }
+    for (auto &R : Ranges)
+      EmitRange(o, SM,
+                Lexer::getAsCharRange(SM.getExpansionRange(R), SM, LangOpts),
+                FM, indent + 1);
     --indent;
     Indent(o, indent) << "</array>\n";
   }
@@ -387,7 +387,9 @@ void PlistDiagnostics::FlushDiagnosticsImpl(
     EmitString(o, D->getCategory()) << '\n';
     o << "   <key>type</key>";
     EmitString(o, D->getBugType()) << '\n';
-    
+    o << "   <key>check_name</key>";
+    EmitString(o, D->getCheckName()) << '\n';
+ 
     // Output information about the semantic context where
     // the issue occurred.
     if (const Decl *DeclWithIssue = D->getDeclWithIssue()) {
@@ -453,7 +455,7 @@ void PlistDiagnostics::FlushDiagnosticsImpl(
 
     // Output the location of the bug.
     o << "  <key>location</key>\n";
-    EmitLocation(o, *SM, LangOpts, D->getLocation().asLocation(), FM, 2);
+    EmitLocation(o, *SM, D->getLocation().asLocation(), FM, 2);
 
     // Output the diagnostic to the sub-diagnostic client, if any.
     if (!filesMade->empty()) {
diff --git a/lib/StaticAnalyzer/Core/RegionStore.cpp b/lib/StaticAnalyzer/Core/RegionStore.cpp
index 45056226c9e0..6d41fc2146fe 100644
--- a/lib/StaticAnalyzer/Core/RegionStore.cpp
+++ b/lib/StaticAnalyzer/Core/RegionStore.cpp
@@ -767,7 +767,7 @@ static inline bool isUnionField(const FieldRegion *FR) {
 
 typedef SmallVector<const FieldDecl *, 8> FieldVector;
 
-void getSymbolicOffsetFields(BindingKey K, FieldVector &Fields) {
+static void getSymbolicOffsetFields(BindingKey K, FieldVector &Fields) {
   assert(K.hasSymbolicOffset() && "Not implemented for concrete offset keys");
 
   const MemRegion *Base = K.getConcreteOffsetRegion();
diff --git a/lib/StaticAnalyzer/Core/SimpleConstraintManager.h b/lib/StaticAnalyzer/Core/SimpleConstraintManager.h
index a72d1d4f9218..135cd4ef8649 100644
--- a/lib/StaticAnalyzer/Core/SimpleConstraintManager.h
+++ b/lib/StaticAnalyzer/Core/SimpleConstraintManager.h
@@ -27,7 +27,7 @@ class SimpleConstraintManager : public ConstraintManager {
 public:
   SimpleConstraintManager(SubEngine *subengine, SValBuilder &SB)
     : SU(subengine), SVB(SB) {}
-  virtual ~SimpleConstraintManager();
+  ~SimpleConstraintManager() override;
 
   //===------------------------------------------------------------------===//
   // Common implementation for the interface provided by ConstraintManager.
diff --git a/lib/StaticAnalyzer/Core/SimpleSValBuilder.cpp b/lib/StaticAnalyzer/Core/SimpleSValBuilder.cpp
index df9e4d6f9199..b3cab87c8080 100644
--- a/lib/StaticAnalyzer/Core/SimpleSValBuilder.cpp
+++ b/lib/StaticAnalyzer/Core/SimpleSValBuilder.cpp
@@ -29,7 +29,7 @@ public:
   SimpleSValBuilder(llvm::BumpPtrAllocator &alloc, ASTContext &context,
                     ProgramStateManager &stateMgr)
                     : SValBuilder(alloc, context, stateMgr) {}
-  virtual ~SimpleSValBuilder() {}
+  ~SimpleSValBuilder() override {}
 
   SVal evalMinus(NonLoc val) override;
   SVal evalComplement(NonLoc val) override;
diff --git a/lib/StaticAnalyzer/Frontend/AnalysisConsumer.cpp b/lib/StaticAnalyzer/Frontend/AnalysisConsumer.cpp
index 183ef358df63..fbeffb8aac80 100644
--- a/lib/StaticAnalyzer/Frontend/AnalysisConsumer.cpp
+++ b/lib/StaticAnalyzer/Frontend/AnalysisConsumer.cpp
@@ -92,7 +92,7 @@ class ClangDiagPathDiagConsumer : public PathDiagnosticConsumer {
 public:
   ClangDiagPathDiagConsumer(DiagnosticsEngine &Diag)
     : Diag(Diag), IncludePath(false) {}
-  virtual ~ClangDiagPathDiagConsumer() {}
+  ~ClangDiagPathDiagConsumer() override {}
   StringRef getName() const override { return "ClangDiags"; }
 
   bool supportsLogicalOpControlFlow() const override { return true; }
@@ -199,7 +199,7 @@ public:
     }
   }
 
-  ~AnalysisConsumer() {
+  ~AnalysisConsumer() override {
     if (Opts->PrintStats)
       delete TUTotalTimer;
   }
@@ -373,8 +373,7 @@ public:
     return true;
   }
 
-  virtual void
-  AddDiagnosticConsumer(PathDiagnosticConsumer *Consumer) override {
+  void AddDiagnosticConsumer(PathDiagnosticConsumer *Consumer) override {
     PathConsumers.push_back(Consumer);
   }
 
@@ -590,7 +589,7 @@ AnalysisConsumer::getModeForDecl(Decl *D, AnalysisMode Mode) {
   // - System headers: don't run any checks.
   SourceManager &SM = Ctx->getSourceManager();
   SourceLocation SL = SM.getExpansionLoc(D->getLocation());
-  if (!Opts->AnalyzeAll && !SM.isInMainFile(SL)) {
+  if (!Opts->AnalyzeAll && !SM.isWrittenInMainFile(SL)) {
     if (SL.isInvalid() || SM.isInSystemHeader(SL))
       return AM_None;
     return Mode & ~AM_Path;
@@ -724,7 +723,7 @@ class UbigraphViz : public ExplodedNode::Auditor {
 public:
   UbigraphViz(std::unique_ptr<raw_ostream> Out, StringRef Filename);
 
-  ~UbigraphViz();
+  ~UbigraphViz() override;
 
   void AddEdge(ExplodedNode *Src, ExplodedNode *Dst) override;
 };
@@ -735,7 +734,7 @@ static std::unique_ptr<ExplodedNode::Auditor> CreateUbiViz() {
   SmallString<128> P;
   int FD;
   llvm::sys::fs::createTemporaryFile("llvm_ubi", "", FD, P);
-  llvm::errs() << "Writing '" << P.str() << "'.\n";
+  llvm::errs() << "Writing '" << P << "'.\n";
 
   auto Stream = llvm::make_unique<llvm::raw_fd_ostream>(FD, true);
 
diff --git a/lib/StaticAnalyzer/Frontend/CheckerRegistration.cpp b/lib/StaticAnalyzer/Frontend/CheckerRegistration.cpp
index 36565cb6e2ca..b3ff79750b49 100644
--- a/lib/StaticAnalyzer/Frontend/CheckerRegistration.cpp
+++ b/lib/StaticAnalyzer/Frontend/CheckerRegistration.cpp
@@ -125,7 +125,7 @@ ento::createCheckerManager(AnalyzerOptions &opts, const LangOptions &langOpts,
 
   }
 
-  return std::move(checkerMgr);
+  return checkerMgr;
 }
 
 void ento::printCheckerHelp(raw_ostream &out, ArrayRef<std::string> plugins) {
diff --git a/lib/StaticAnalyzer/Frontend/ModelInjector.h b/lib/StaticAnalyzer/Frontend/ModelInjector.h
index fd24e32f3a3d..e23bf8abf384 100644
--- a/lib/StaticAnalyzer/Frontend/ModelInjector.h
+++ b/lib/StaticAnalyzer/Frontend/ModelInjector.h
@@ -43,8 +43,8 @@ namespace ento {
 class ModelInjector : public CodeInjector {
 public:
   ModelInjector(CompilerInstance &CI);
-  Stmt *getBody(const FunctionDecl *D);
-  Stmt *getBody(const ObjCMethodDecl *D);
+  Stmt *getBody(const FunctionDecl *D) override;
+  Stmt *getBody(const ObjCMethodDecl *D) override;
 
 private:
   /// \brief Synthesize a body for a declaration
diff --git a/lib/Tooling/CommonOptionsParser.cpp b/lib/Tooling/CommonOptionsParser.cpp
index 91c74a4c3359..adae1781f450 100644
--- a/lib/Tooling/CommonOptionsParser.cpp
+++ b/lib/Tooling/CommonOptionsParser.cpp
@@ -54,6 +54,7 @@ const char *const CommonOptionsParser::HelpMessage =
     "\tsuffix of a path in the compile command database.\n"
     "\n";
 
+namespace {
 class ArgumentsAdjustingCompilations : public CompilationDatabase {
 public:
   ArgumentsAdjustingCompilations(
@@ -89,6 +90,7 @@ private:
     return Commands;
   }
 };
+} // namespace
 
 CommonOptionsParser::CommonOptionsParser(int &argc, const char **argv,
                                          cl::OptionCategory &Category,
@@ -112,15 +114,7 @@ CommonOptionsParser::CommonOptionsParser(int &argc, const char **argv,
       cl::desc("Additional argument to prepend to the compiler command line"),
       cl::cat(Category));
 
-  // Hide unrelated options.
-  StringMap<cl::Option*> Options;
-  cl::getRegisteredOptions(Options);
-  for (StringMap<cl::Option *>::iterator I = Options.begin(), E = Options.end();
-       I != E; ++I) {
-    if (I->second->Category != &Category && I->first() != "help" &&
-        I->first() != "version")
-      I->second->setHiddenFlag(cl::ReallyHidden);
-  }
+  cl::HideUnrelatedOptions(Category);
 
   Compilations.reset(FixedCompilationDatabase::loadFromCommandLine(argc,
                                                                    argv));
diff --git a/lib/Tooling/CompilationDatabase.cpp b/lib/Tooling/CompilationDatabase.cpp
index 7613988c96d9..2514f027f472 100644
--- a/lib/Tooling/CompilationDatabase.cpp
+++ b/lib/Tooling/CompilationDatabase.cpp
@@ -29,9 +29,8 @@
 #include "llvm/Support/Path.h"
 #include <sstream>
 #include <system_error>
-
-namespace clang {
-namespace tooling {
+using namespace clang;
+using namespace tooling;
 
 CompilationDatabase::~CompilationDatabase() {}
 
@@ -109,6 +108,7 @@ CompilationDatabase::autoDetectFromDirectory(StringRef SourceDir,
 
 CompilationDatabasePlugin::~CompilationDatabasePlugin() {}
 
+namespace {
 // Helper for recursively searching through a chain of actions and collecting
 // all inputs, direct and indirect, of compile jobs.
 struct CompileJobAnalyzer {
@@ -156,8 +156,8 @@ public:
   // recording for our own purposes.
   UnusedInputDiagConsumer(DiagnosticConsumer *Other) : Other(Other) {}
 
-  virtual void HandleDiagnostic(DiagnosticsEngine::Level DiagLevel,
-                                const Diagnostic &Info) override {
+  void HandleDiagnostic(DiagnosticsEngine::Level DiagLevel,
+                        const Diagnostic &Info) override {
     if (Info.getID() == clang::diag::warn_drv_input_file_unused) {
       // Arg 1 for this diagnostic is the option that didn't get used.
       UnusedInputs.push_back(Info.getArgStdStr(0));
@@ -183,6 +183,7 @@ struct MatchesAny {
 private:
   ArrayRef<std::string> Arr;
 };
+} // namespace
 
 /// \brief Strips any positional args and possible argv[0] from a command-line
 /// provided by the user to construct a FixedCompilationDatabase.
@@ -282,11 +283,9 @@ static bool stripPositionalArgs(std::vector<const char *> Args,
   return true;
 }
 
-FixedCompilationDatabase *
-FixedCompilationDatabase::loadFromCommandLine(int &Argc,
-                                              const char **Argv,
-                                              Twine Directory) {
-  const char **DoubleDash = std::find(Argv, Argv + Argc, StringRef("--"));
+FixedCompilationDatabase *FixedCompilationDatabase::loadFromCommandLine(
+    int &Argc, const char *const *Argv, Twine Directory) {
+  const char *const *DoubleDash = std::find(Argv, Argv + Argc, StringRef("--"));
   if (DoubleDash == Argv + Argc)
     return nullptr;
   std::vector<const char *> CommandLine(DoubleDash + 1, Argv + Argc);
@@ -324,6 +323,9 @@ FixedCompilationDatabase::getAllCompileCommands() const {
   return std::vector<CompileCommand>();
 }
 
+namespace clang {
+namespace tooling {
+
 // This anchor is used to force the linker to link in the generated object file
 // and thus register the JSONCompilationDatabasePlugin.
 extern volatile int JSONAnchorSource;
diff --git a/lib/Tooling/Core/Replacement.cpp b/lib/Tooling/Core/Replacement.cpp
index 525f7dfa91c6..b9fc92bb6c71 100644
--- a/lib/Tooling/Core/Replacement.cpp
+++ b/lib/Tooling/Core/Replacement.cpp
@@ -77,11 +77,11 @@ bool Replacement::apply(Rewriter &Rewrite) const {
 }
 
 std::string Replacement::toString() const {
-  std::string result;
-  llvm::raw_string_ostream stream(result);
-  stream << FilePath << ": " << ReplacementRange.getOffset() << ":+"
+  std::string Result;
+  llvm::raw_string_ostream Stream(Result);
+  Stream << FilePath << ": " << ReplacementRange.getOffset() << ":+"
          << ReplacementRange.getLength() << ":\"" << ReplacementText << "\"";
-  return result;
+  return Stream.str();
 }
 
 bool operator<(const Replacement &LHS, const Replacement &RHS) {
diff --git a/lib/Tooling/FileMatchTrie.cpp b/lib/Tooling/FileMatchTrie.cpp
index dc9999ef965a..86ed036e2dbe 100644
--- a/lib/Tooling/FileMatchTrie.cpp
+++ b/lib/Tooling/FileMatchTrie.cpp
@@ -17,18 +17,20 @@
 #include "llvm/Support/Path.h"
 #include "llvm/Support/raw_ostream.h"
 #include <sstream>
+using namespace clang;
+using namespace tooling;
 
-namespace clang {
-namespace tooling {
-
+namespace {
 /// \brief Default \c PathComparator using \c llvm::sys::fs::equivalent().
 struct DefaultPathComparator : public PathComparator {
-  virtual ~DefaultPathComparator() {}
   bool equivalent(StringRef FileA, StringRef FileB) const override {
     return FileA == FileB || llvm::sys::fs::equivalent(FileA, FileB);
   }
 };
+}
 
+namespace clang {
+namespace tooling {
 /// \brief A node of the \c FileMatchTrie.
 ///
 /// Each node has storage for up to one path and a map mapping a path segment to
@@ -156,6 +158,8 @@ private:
   // The children of this node stored in a map based on the next path segment.
   llvm::StringMap<FileMatchTrieNode> Children;
 };
+} // end namespace tooling
+} // end namespace clang
 
 FileMatchTrie::FileMatchTrie()
   : Root(new FileMatchTrieNode), Comparator(new DefaultPathComparator()) {}
@@ -183,6 +187,3 @@ StringRef FileMatchTrie::findEquivalent(StringRef FileName,
     Error << "Path is ambiguous";
   return Result;
 }
-
-} // end namespace tooling
-} // end namespace clang
diff --git a/lib/Tooling/JSONCompilationDatabase.cpp b/lib/Tooling/JSONCompilationDatabase.cpp
index 3b5f7e28d02d..7dc211e934b2 100644
--- a/lib/Tooling/JSONCompilationDatabase.cpp
+++ b/lib/Tooling/JSONCompilationDatabase.cpp
@@ -176,7 +176,7 @@ JSONCompilationDatabase::getCompileCommands(StringRef FilePath) const {
 
   std::string Error;
   llvm::raw_string_ostream ES(Error);
-  StringRef Match = MatchTrie.findEquivalent(NativeFilePath.str(), ES);
+  StringRef Match = MatchTrie.findEquivalent(NativeFilePath, ES);
   if (Match.empty())
     return std::vector<CompileCommand>();
   llvm::StringMap< std::vector<CompileCommandRef> >::const_iterator
@@ -307,13 +307,13 @@ bool JSONCompilationDatabase::parse(std::string &ErrorMessage) {
       SmallString<128> AbsolutePath(
           Directory->getValue(DirectoryStorage));
       llvm::sys::path::append(AbsolutePath, FileName);
-      llvm::sys::path::native(AbsolutePath.str(), NativeFilePath);
+      llvm::sys::path::native(AbsolutePath, NativeFilePath);
     } else {
       llvm::sys::path::native(FileName, NativeFilePath);
     }
     IndexByFile[NativeFilePath].push_back(
         CompileCommandRef(Directory, Command));
-    MatchTrie.insert(NativeFilePath.str());
+    MatchTrie.insert(NativeFilePath);
   }
   return true;
 }
diff --git a/lib/Tooling/Tooling.cpp b/lib/Tooling/Tooling.cpp
index 60371fb5e0e6..e100003cee87 100644
--- a/lib/Tooling/Tooling.cpp
+++ b/lib/Tooling/Tooling.cpp
@@ -90,7 +90,7 @@ static const llvm::opt::ArgStringList *getCC1Arguments(
 }
 
 /// \brief Returns a clang build invocation initialized from the CC1 flags.
-static clang::CompilerInvocation *newInvocation(
+clang::CompilerInvocation *newInvocation(
     clang::DiagnosticsEngine *Diagnostics,
     const llvm::opt::ArgStringList &CC1Args) {
   assert(!CC1Args.empty() && "Must at least contain the program name!");