diff --git a/clang/include/clang/Sema/SemaOpenACC.h b/clang/include/clang/Sema/SemaOpenACC.h
@@ -40,7 +40,12 @@ class SemaOpenACC : public SemaBase {
       OpenACCDefaultClauseKind DefaultClauseKind;
     };
 
-    std::variant<DefaultDetails> Details;
+    struct ConditionDetails {
+      Expr *ConditionExpr;
+    };
+
+    std::variant<std::monostate, DefaultDetails, ConditionDetails> Details =
+        std::monostate{};
 
   public:
     OpenACCParsedClause(OpenACCDirectiveKind DirKind,
@@ -63,6 +68,25 @@ class SemaOpenACC : public SemaBase {
       return std::get<DefaultDetails>(Details).DefaultClauseKind;
     }
 
+    const Expr *getConditionExpr() const {
+      return const_cast<OpenACCParsedClause *>(this)->getConditionExpr();
+    }
+
+    Expr *getConditionExpr() {
+      assert((ClauseKind == OpenACCClauseKind::If ||
+              (ClauseKind == OpenACCClauseKind::Self &&
+               DirKind != OpenACCDirectiveKind::Update)) &&
+             "Parsed clause kind does not have a condition expr");
+
+      // 'self' has an optional ConditionExpr, so be tolerant of that. This will
+      // assert in variant otherwise.
+      if (ClauseKind == OpenACCClauseKind::Self &&
+          std::holds_alternative<std::monostate>(Details))
+        return nullptr;
+
+      return std::get<ConditionDetails>(Details).ConditionExpr;
+    }
+
     void setLParenLoc(SourceLocation EndLoc) { LParenLoc = EndLoc; }
     void setEndLoc(SourceLocation EndLoc) { ClauseRange.setEnd(EndLoc); }
 
@@ -71,6 +95,20 @@ class SemaOpenACC : public SemaBase {
              "Parsed clause is not a default clause");
       Details = DefaultDetails{DefKind};
     }
+
+    void setConditionDetails(Expr *ConditionExpr) {
+      assert((ClauseKind == OpenACCClauseKind::If ||
+              (ClauseKind == OpenACCClauseKind::Self &&
+               DirKind != OpenACCDirectiveKind::Update)) &&
+             "Parsed clause kind does not have a condition expr");
+      // In C++ we can count on this being a 'bool', but in C this gets left as
+      // some sort of scalar that codegen will have to take care of converting.
+      assert((!ConditionExpr || ConditionExpr->isInstantiationDependent() ||
+              ConditionExpr->getType()->isScalarType()) &&
+             "Condition expression type not scalar/dependent");
+
+      Details = ConditionDetails{ConditionExpr};
+    }
   };
 
   SemaOpenACC(Sema &S);

diff --git a/clang/include/clang/Sema/SemaOpenMP.h b/clang/include/clang/Sema/SemaOpenMP.h
diff --git a/clang/include/clang/Serialization/ASTBitCodes.h b/clang/include/clang/Serialization/ASTBitCodes.h
@@ -698,6 +698,10 @@ enum ASTRecordTypes {
   /// Record code for an unterminated \#pragma clang assume_nonnull begin
   /// recorded in a preamble.
   PP_ASSUME_NONNULL_LOC = 67,
+
+  /// Record code for lexical and visible block for delayed namespace in
+  /// reduced BMI.
+  DELAYED_NAMESPACE_LEXICAL_VISIBLE_RECORD = 68,
 };
 
 /// Record types used within a source manager block.

diff --git a/clang/include/clang/Serialization/ASTReader.h b/clang/include/clang/Serialization/ASTReader.h
@@ -517,6 +517,20 @@ class ASTReader
   /// in the chain.
   DeclUpdateOffsetsMap DeclUpdateOffsets;
 
+  using DelayedNamespaceOffsetMapTy = llvm::DenseMap<
+      serialization::DeclID,
+      std::pair</*LexicalOffset*/ uint64_t, /*VisibleOffset*/ uint64_t>>;
+
+  /// Mapping from global declaration IDs to the lexical and visible block
+  /// offset for delayed namespace in reduced BMI.
+  ///
+  /// We can't use the existing DeclUpdate mechanism since the DeclUpdate
+  /// may only be applied in an outer most read. However, we need to know
+  /// whether or not a DeclContext has external storage during the recursive
+  /// reading. So we need to apply the offset immediately after we read the
+  /// namespace as if it is not delayed.
+  DelayedNamespaceOffsetMapTy DelayedNamespaceOffsetMap;
+
   struct PendingUpdateRecord {
     Decl *D;
     serialization::GlobalDeclID ID;
@@ -859,7 +873,7 @@ class ASTReader
 
   /// Our current depth in #pragma cuda force_host_device begin/end
   /// macros.
-  unsigned ForceCUDAHostDeviceDepth = 0;
+  unsigned ForceHostDeviceDepth = 0;
 
   /// The IDs of the declarations Sema stores directly.
   ///
@@ -2443,6 +2457,12 @@ class BitsUnpacker {
   uint32_t Value;
   uint32_t CurrentBitsIndex = ~0;
 };
+
+inline bool shouldSkipCheckingODR(const Decl *D) {
+  return D->getASTContext().getLangOpts().SkipODRCheckInGMF &&
+         D->isFromExplicitGlobalModule();
+}
+
 } // namespace clang
 
 #endif // LLVM_CLANG_SERIALIZATION_ASTREADER_H
diff --git a/clang/include/clang/Serialization/ASTWriter.h b/clang/include/clang/Serialization/ASTWriter.h
@@ -201,6 +201,16 @@ class ASTWriter : public ASTDeserializationListener,
   /// The declarations and types to emit.
   std::queue<DeclOrType> DeclTypesToEmit;
 
+  /// The delayed namespace to emit. Only meaningful for reduced BMI.
+  ///
+  /// In reduced BMI, we want to elide the unreachable declarations in
+  /// the global module fragment. However, in ASTWriterDecl, when we see
+  /// a namespace, all the declarations in the namespace would be emitted.
+  /// So the optimization become meaningless. To solve the issue, we
+  /// delay recording all the declarations until we emit all the declarations.
+  /// Then we can safely record the reached declarations only.
+  llvm::SmallVector<NamespaceDecl *, 16> DelayedNamespace;
+
   /// The first ID number we can use for our own declarations.
   serialization::DeclID FirstDeclID = serialization::NUM_PREDEF_DECL_IDS;
 
@@ -529,7 +539,8 @@ class ASTWriter : public ASTDeserializationListener,
   void WriteType(QualType T);
 
   bool isLookupResultExternal(StoredDeclsList &Result, DeclContext *DC);
-  bool isLookupResultEntirelyExternal(StoredDeclsList &Result, DeclContext *DC);
+  bool isLookupResultEntirelyExternalOrUnreachable(StoredDeclsList &Result,
+                                                   DeclContext *DC);
 
   void GenerateNameLookupTable(const DeclContext *DC,
                                llvm::SmallVectorImpl<char> &LookupTable);
@@ -704,6 +715,15 @@ class ASTWriter : public ASTDeserializationListener,
   /// declaration.
   serialization::DeclID getDeclID(const Decl *D);
 
+  /// Whether or not the declaration got emitted. If not, it wouldn't be
+  /// emitted.
+  ///
+  /// This may only be called after we've done the job to write the
+  /// declarations (marked by DoneWritingDeclsAndTypes).
+  ///
+  /// A declaration may only be omitted in reduced BMI.
+  bool wasDeclEmitted(const Decl *D) const;
+
   unsigned getAnonymousDeclarationNumber(const NamedDecl *D);
 
   /// Add a string to the given record.
@@ -798,6 +818,10 @@ class ASTWriter : public ASTDeserializationListener,
     return WritingModule && WritingModule->isNamedModule();
   }
 
+  bool isGeneratingReducedBMI() const { return GeneratingReducedBMI; }
+
+  bool getDoneWritingDeclsAndTypes() const { return DoneWritingDeclsAndTypes; }
+
 private:
   // ASTDeserializationListener implementation
   void ReaderInitialized(ASTReader *Reader) override;

diff --git a/clang/include/clang/Serialization/ModuleFileExtension.h b/clang/include/clang/Serialization/ModuleFileExtension.h
@@ -9,7 +9,6 @@
 #ifndef LLVM_CLANG_SERIALIZATION_MODULEFILEEXTENSION_H
 #define LLVM_CLANG_SERIALIZATION_MODULEFILEEXTENSION_H
 
-#include "llvm/ADT/IntrusiveRefCntPtr.h"
 #include "llvm/Support/ExtensibleRTTI.h"
 #include "llvm/Support/HashBuilder.h"
 #include "llvm/Support/MD5.h"

diff --git a/clang/include/clang/Serialization/PCHContainerOperations.h b/clang/include/clang/Serialization/PCHContainerOperations.h
@@ -12,7 +12,7 @@
 #include "clang/Basic/Module.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/StringMap.h"
-#include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/MemoryBufferRef.h"
 #include <memory>
 
 namespace llvm {

diff --git a/clang/include/clang/Tooling/DependencyScanning/DependencyScanningFilesystem.h b/clang/include/clang/Tooling/DependencyScanning/DependencyScanningFilesystem.h
@@ -142,6 +142,8 @@ class CachedFileSystemEntry {
   CachedFileContents *Contents;
 };
 
+using CachedRealPath = llvm::ErrorOr<std::string>;
+
 /// This class is a shared cache, that caches the 'stat' and 'open' calls to the
 /// underlying real file system, and the scanned preprocessor directives of
 /// files.
@@ -154,9 +156,11 @@ class DependencyScanningFilesystemSharedCache {
     /// The mutex that needs to be locked before mutation of any member.
     mutable std::mutex CacheLock;
 
-    /// Map from filenames to cached entries.
-    llvm::StringMap<const CachedFileSystemEntry *, llvm::BumpPtrAllocator>
-        EntriesByFilename;
+    /// Map from filenames to cached entries and real paths.
+    llvm::StringMap<
+        std::pair<const CachedFileSystemEntry *, const CachedRealPath *>,
+        llvm::BumpPtrAllocator>
+        CacheByFilename;
 
     /// Map from unique IDs to cached entries.
     llvm::DenseMap<llvm::sys::fs::UniqueID, const CachedFileSystemEntry *>
@@ -168,6 +172,9 @@ class DependencyScanningFilesystemSharedCache {
     /// The backing storage for cached contents.
     llvm::SpecificBumpPtrAllocator<CachedFileContents> ContentsStorage;
 
+    /// The backing storage for cached real paths.
+    llvm::SpecificBumpPtrAllocator<CachedRealPath> RealPathStorage;
+
     /// Returns entry associated with the filename or nullptr if none is found.
     const CachedFileSystemEntry *findEntryByFilename(StringRef Filename) const;
 
@@ -194,6 +201,17 @@ class DependencyScanningFilesystemSharedCache {
     const CachedFileSystemEntry &
     getOrInsertEntryForFilename(StringRef Filename,
                                 const CachedFileSystemEntry &Entry);
+
+    /// Returns the real path associated with the filename or nullptr if none is
+    /// found.
+    const CachedRealPath *findRealPathByFilename(StringRef Filename) const;
+
+    /// Returns the real path associated with the filename if there is some.
+    /// Otherwise, constructs new one with the given one, associates it with the
+    /// filename and returns the result.
+    const CachedRealPath &
+    getOrEmplaceRealPathForFilename(StringRef Filename,
+                                    llvm::ErrorOr<StringRef> RealPath);
   };
 
   DependencyScanningFilesystemSharedCache();
@@ -210,14 +228,17 @@ class DependencyScanningFilesystemSharedCache {
 /// This class is a local cache, that caches the 'stat' and 'open' calls to the
 /// underlying real file system.
 class DependencyScanningFilesystemLocalCache {
-  llvm::StringMap<const CachedFileSystemEntry *, llvm::BumpPtrAllocator> Cache;
+  llvm::StringMap<
+      std::pair<const CachedFileSystemEntry *, const CachedRealPath *>,
+      llvm::BumpPtrAllocator>
+      Cache;
 
 public:
   /// Returns entry associated with the filename or nullptr if none is found.
   const CachedFileSystemEntry *findEntryByFilename(StringRef Filename) const {
     assert(llvm::sys::path::is_absolute_gnu(Filename));
     auto It = Cache.find(Filename);
-    return It == Cache.end() ? nullptr : It->getValue();
+    return It == Cache.end() ? nullptr : It->getValue().first;
   }
 
   /// Associates the given entry with the filename and returns the given entry
@@ -226,9 +247,40 @@ class DependencyScanningFilesystemLocalCache {
   insertEntryForFilename(StringRef Filename,
                          const CachedFileSystemEntry &Entry) {
     assert(llvm::sys::path::is_absolute_gnu(Filename));
-    const auto *InsertedEntry = Cache.insert({Filename, &Entry}).first->second;
-    assert(InsertedEntry == &Entry && "entry already present");
-    return *InsertedEntry;
+    auto [It, Inserted] = Cache.insert({Filename, {&Entry, nullptr}});
+    auto &[CachedEntry, CachedRealPath] = It->getValue();
+    if (!Inserted) {
+      // The file is already present in the local cache. If we got here, it only
+      // contains the real path. Let's make sure the entry is populated too.
+      assert((!CachedEntry && CachedRealPath) && "entry already present");
+      CachedEntry = &Entry;
+    }
+    return *CachedEntry;
+  }
+
+  /// Returns real path associated with the filename or nullptr if none is
+  /// found.
+  const CachedRealPath *findRealPathByFilename(StringRef Filename) const {
+    assert(llvm::sys::path::is_absolute_gnu(Filename));
+    auto It = Cache.find(Filename);
+    return It == Cache.end() ? nullptr : It->getValue().second;
+  }
+
+  /// Associates the given real path with the filename and returns the given
+  /// entry pointer (for convenience).
+  const CachedRealPath &
+  insertRealPathForFilename(StringRef Filename,
+                            const CachedRealPath &RealPath) {
+    assert(llvm::sys::path::is_absolute_gnu(Filename));
+    auto [It, Inserted] = Cache.insert({Filename, {nullptr, &RealPath}});
+    auto &[CachedEntry, CachedRealPath] = It->getValue();
+    if (!Inserted) {
+      // The file is already present in the local cache. If we got here, it only
+      // contains the entry. Let's make sure the real path is populated too.
+      assert((!CachedRealPath && CachedEntry) && "real path already present");
+      CachedRealPath = &RealPath;
+    }
+    return *CachedRealPath;
   }
 };
 
@@ -296,6 +348,9 @@ class DependencyScanningWorkerFilesystem
   llvm::ErrorOr<std::unique_ptr<llvm::vfs::File>>
   openFileForRead(const Twine &Path) override;
 
+  std::error_code getRealPath(const Twine &Path,
+                              SmallVectorImpl<char> &Output) override;
+
   std::error_code setCurrentWorkingDirectory(const Twine &Path) override;
 
   /// Returns entry for the given filename.
@@ -310,6 +365,10 @@ class DependencyScanningWorkerFilesystem
   /// false if not (i.e. this entry is not a file or its scan fails).
   bool ensureDirectiveTokensArePopulated(EntryRef Entry);
 
+  /// Check whether \p Path exists. By default checks cached result of \c
+  /// status(), and falls back on FS if unable to do so.
+  bool exists(const Twine &Path) override;
+
 private:
   /// For a filename that's not yet associated with any entry in the caches,
   /// uses the underlying filesystem to either look up the entry based in the
@@ -402,6 +461,10 @@ class DependencyScanningWorkerFilesystem
   llvm::ErrorOr<std::string> WorkingDirForCacheLookup;
 
   void updateWorkingDirForCacheLookup();
+
+  llvm::ErrorOr<StringRef>
+  tryGetFilenameForLookup(StringRef OriginalFilename,
+                          llvm::SmallVectorImpl<char> &PathBuf) const;
 };
 
 } // end namespace dependencies

diff --git a/clang/include/clang/Tooling/DependencyScanning/ModuleDepCollector.h b/clang/include/clang/Tooling/DependencyScanning/ModuleDepCollector.h
@@ -308,6 +308,11 @@ class ModuleDepCollector final : public DependencyCollector {
                                 ModuleDeps &Deps);
 };
 
+/// Resets codegen options that don't affect modules/PCH.
+void resetBenignCodeGenOptions(frontend::ActionKind ProgramAction,
+                               const LangOptions &LangOpts,
+                               CodeGenOptions &CGOpts);
+
 } // end namespace dependencies
 } // end namespace tooling
 } // end namespace clang

diff --git a/clang/lib/APINotes/APINotesReader.cpp b/clang/lib/APINotes/APINotesReader.cpp
diff --git a/clang/lib/AST/ASTImporter.cpp b/clang/lib/AST/ASTImporter.cpp
@@ -3947,6 +3947,14 @@ ExpectedDecl ASTNodeImporter::VisitFunctionDecl(FunctionDecl *D) {
     // decl and its redeclarations may be required.
   }
 
+  StringLiteral *Msg = D->getDeletedMessage();
+  if (Msg) {
+    auto Imported = import(Msg);
+    if (!Imported)
+      return Imported.takeError();
+    Msg = *Imported;
+  }
+
   ToFunction->setQualifierInfo(ToQualifierLoc);
   ToFunction->setAccess(D->getAccess());
   ToFunction->setLexicalDeclContext(LexicalDC);
@@ -3961,6 +3969,11 @@ ExpectedDecl ASTNodeImporter::VisitFunctionDecl(FunctionDecl *D) {
   ToFunction->setRangeEnd(ToEndLoc);
   ToFunction->setDefaultLoc(ToDefaultLoc);
 
+  if (Msg)
+    ToFunction->setDefaultedOrDeletedInfo(
+        FunctionDecl::DefaultedOrDeletedFunctionInfo::Create(
+            Importer.getToContext(), {}, Msg));
+
   // Set the parameters.
   for (auto *Param : Parameters) {
     Param->setOwningFunction(ToFunction);

diff --git a/clang/lib/AST/Decl.cpp b/clang/lib/AST/Decl.cpp
@@ -3058,7 +3058,7 @@ FunctionDecl::FunctionDecl(Kind DK, ASTContext &C, DeclContext *DC,
   FunctionDeclBits.IsTrivialForCall = false;
   FunctionDeclBits.IsDefaulted = false;
   FunctionDeclBits.IsExplicitlyDefaulted = false;
-  FunctionDeclBits.HasDefaultedFunctionInfo = false;
+  FunctionDeclBits.HasDefaultedOrDeletedInfo = false;
   FunctionDeclBits.IsIneligibleOrNotSelected = false;
   FunctionDeclBits.HasImplicitReturnZero = false;
   FunctionDeclBits.IsLateTemplateParsed = false;
@@ -3092,30 +3092,65 @@ bool FunctionDecl::isVariadic() const {
   return false;
 }
 
-FunctionDecl::DefaultedFunctionInfo *
-FunctionDecl::DefaultedFunctionInfo::Create(ASTContext &Context,
-                                            ArrayRef<DeclAccessPair> Lookups) {
-  DefaultedFunctionInfo *Info = new (Context.Allocate(
-      totalSizeToAlloc<DeclAccessPair>(Lookups.size()),
-      std::max(alignof(DefaultedFunctionInfo), alignof(DeclAccessPair))))
-      DefaultedFunctionInfo;
+FunctionDecl::DefaultedOrDeletedFunctionInfo *
+FunctionDecl::DefaultedOrDeletedFunctionInfo::Create(
+    ASTContext &Context, ArrayRef<DeclAccessPair> Lookups,
+    StringLiteral *DeletedMessage) {
+  static constexpr size_t Alignment =
+      std::max({alignof(DefaultedOrDeletedFunctionInfo),
+                alignof(DeclAccessPair), alignof(StringLiteral *)});
+  size_t Size = totalSizeToAlloc<DeclAccessPair, StringLiteral *>(
+      Lookups.size(), DeletedMessage != nullptr);
+
+  DefaultedOrDeletedFunctionInfo *Info =
+      new (Context.Allocate(Size, Alignment)) DefaultedOrDeletedFunctionInfo;
   Info->NumLookups = Lookups.size();
+  Info->HasDeletedMessage = DeletedMessage != nullptr;
+
   std::uninitialized_copy(Lookups.begin(), Lookups.end(),
                           Info->getTrailingObjects<DeclAccessPair>());
+  if (DeletedMessage)
+    *Info->getTrailingObjects<StringLiteral *>() = DeletedMessage;
   return Info;
 }
 
-void FunctionDecl::setDefaultedFunctionInfo(DefaultedFunctionInfo *Info) {
-  assert(!FunctionDeclBits.HasDefaultedFunctionInfo && "already have this");
+void FunctionDecl::setDefaultedOrDeletedInfo(
+    DefaultedOrDeletedFunctionInfo *Info) {
+  assert(!FunctionDeclBits.HasDefaultedOrDeletedInfo && "already have this");
   assert(!Body && "can't replace function body with defaulted function info");
 
-  FunctionDeclBits.HasDefaultedFunctionInfo = true;
-  DefaultedInfo = Info;
+  FunctionDeclBits.HasDefaultedOrDeletedInfo = true;
+  DefaultedOrDeletedInfo = Info;
+}
+
+void FunctionDecl::setDeletedAsWritten(bool D, StringLiteral *Message) {
+  FunctionDeclBits.IsDeleted = D;
+
+  if (Message) {
+    assert(isDeletedAsWritten() && "Function must be deleted");
+    if (FunctionDeclBits.HasDefaultedOrDeletedInfo)
+      DefaultedOrDeletedInfo->setDeletedMessage(Message);
+    else
+      setDefaultedOrDeletedInfo(DefaultedOrDeletedFunctionInfo::Create(
+          getASTContext(), /*Lookups=*/{}, Message));
+  }
+}
+
+void FunctionDecl::DefaultedOrDeletedFunctionInfo::setDeletedMessage(
+    StringLiteral *Message) {
+  // We should never get here with the DefaultedOrDeletedInfo populated, but
+  // no space allocated for the deleted message, since that would require
+  // recreating this, but setDefaultedOrDeletedInfo() disallows overwriting
+  // an already existing DefaultedOrDeletedFunctionInfo.
+  assert(HasDeletedMessage &&
+         "No space to store a delete message in this DefaultedOrDeletedInfo");
+  *getTrailingObjects<StringLiteral *>() = Message;
 }
 
-FunctionDecl::DefaultedFunctionInfo *
-FunctionDecl::getDefaultedFunctionInfo() const {
-  return FunctionDeclBits.HasDefaultedFunctionInfo ? DefaultedInfo : nullptr;
+FunctionDecl::DefaultedOrDeletedFunctionInfo *
+FunctionDecl::getDefalutedOrDeletedInfo() const {
+  return FunctionDeclBits.HasDefaultedOrDeletedInfo ? DefaultedOrDeletedInfo
+                                                    : nullptr;
 }
 
 bool FunctionDecl::hasBody(const FunctionDecl *&Definition) const {
@@ -3202,7 +3237,7 @@ Stmt *FunctionDecl::getBody(const FunctionDecl *&Definition) const {
   if (!hasBody(Definition))
     return nullptr;
 
-  assert(!Definition->FunctionDeclBits.HasDefaultedFunctionInfo &&
+  assert(!Definition->FunctionDeclBits.HasDefaultedOrDeletedInfo &&
          "definition should not have a body");
   if (Definition->Body)
     return Definition->Body.get(getASTContext().getExternalSource());
@@ -3211,7 +3246,7 @@ Stmt *FunctionDecl::getBody(const FunctionDecl *&Definition) const {
 }
 
 void FunctionDecl::setBody(Stmt *B) {
-  FunctionDeclBits.HasDefaultedFunctionInfo = false;
+  FunctionDeclBits.HasDefaultedOrDeletedInfo = false;
   Body = LazyDeclStmtPtr(B);
   if (B)
     EndRangeLoc = B->getEndLoc();
@@ -4499,7 +4534,7 @@ unsigned FunctionDecl::getODRHash() {
   }
 
   class ODRHash Hash;
-  Hash.AddFunctionDecl(this, /*SkipBody=*/shouldSkipCheckingODR());
+  Hash.AddFunctionDecl(this);
   setHasODRHash(true);
   ODRHash = Hash.CalculateHash();
   return ODRHash;

diff --git a/clang/lib/AST/DeclBase.cpp b/clang/lib/AST/DeclBase.cpp
@@ -1106,11 +1106,6 @@ bool Decl::isFromExplicitGlobalModule() const {
   return getOwningModule() && getOwningModule()->isExplicitGlobalModule();
 }
 
-bool Decl::shouldSkipCheckingODR() const {
-  return getASTContext().getLangOpts().SkipODRCheckInGMF &&
-         isFromExplicitGlobalModule();
-}
-
 static Decl::Kind getKind(const Decl *D) { return D->getKind(); }
 static Decl::Kind getKind(const DeclContext *DC) { return DC->getDeclKind(); }
 

diff --git a/clang/lib/AST/DeclPrinter.cpp b/clang/lib/AST/DeclPrinter.cpp
@@ -822,9 +822,14 @@ void DeclPrinter::VisitFunctionDecl(FunctionDecl *D) {
 
   if (D->isPureVirtual())
     Out << " = 0";
-  else if (D->isDeletedAsWritten())
+  else if (D->isDeletedAsWritten()) {
     Out << " = delete";
-  else if (D->isExplicitlyDefaulted())
+    if (const StringLiteral *M = D->getDeletedMessage()) {
+      Out << "(";
+      M->outputString(Out);
+      Out << ")";
+    }
+  } else if (D->isExplicitlyDefaulted())
     Out << " = default";
   else if (D->doesThisDeclarationHaveABody()) {
     if (!Policy.TerseOutput) {

diff --git a/clang/lib/AST/Interp/ByteCodeExprGen.cpp b/clang/lib/AST/Interp/ByteCodeExprGen.cpp
@@ -262,7 +262,7 @@ bool ByteCodeExprGen<Emitter>::VisitCastExpr(const CastExpr *CE) {
       return this->discard(SubExpr);
 
     std::optional<PrimType> FromT = classify(SubExpr->getType());
-    std::optional<PrimType> ToT = classifyPrim(CE->getType());
+    std::optional<PrimType> ToT = classify(CE->getType());
     if (!FromT || !ToT)
       return false;
 
@@ -398,6 +398,35 @@ bool ByteCodeExprGen<Emitter>::VisitCastExpr(const CastExpr *CE) {
     return true;
   }
 
+  case CK_VectorSplat: {
+    assert(!classify(CE->getType()));
+    assert(classify(SubExpr->getType()));
+    assert(CE->getType()->isVectorType());
+
+    if (DiscardResult)
+      return this->discard(SubExpr);
+
+    assert(Initializing); // FIXME: Not always correct.
+    const auto *VT = CE->getType()->getAs<VectorType>();
+    PrimType ElemT = classifyPrim(SubExpr);
+    unsigned ElemOffset = allocateLocalPrimitive(
+        SubExpr, ElemT, /*IsConst=*/true, /*IsExtended=*/false);
+
+    if (!this->visit(SubExpr))
+      return false;
+    if (!this->emitSetLocal(ElemT, ElemOffset, CE))
+      return false;
+
+    for (unsigned I = 0; I != VT->getNumElements(); ++I) {
+      if (!this->emitGetLocal(ElemT, ElemOffset, CE))
+        return false;
+      if (!this->emitInitElem(ElemT, I, CE))
+        return false;
+    }
+
+    return true;
+  }
+
   case CK_ToVoid:
     return discard(SubExpr);
 
@@ -1251,17 +1280,46 @@ bool ByteCodeExprGen<Emitter>::VisitUnaryExprOrTypeTraitExpr(
     return this->emitConst(Size.getQuantity(), E);
   }
 
+  if (Kind == UETT_VectorElements) {
+    if (const auto *VT = E->getTypeOfArgument()->getAs<VectorType>())
+      return this->emitConst(VT->getNumElements(), E);
+
+    // FIXME: Apparently we need to catch the fact that a sizeless vector type
+    // has been passed and diagnose that (at run time).
+    assert(E->getTypeOfArgument()->isSizelessVectorType());
+  }
+
   return false;
 }
 
 template <class Emitter>
 bool ByteCodeExprGen<Emitter>::VisitMemberExpr(const MemberExpr *E) {
   // 'Base.Member'
   const Expr *Base = E->getBase();
+  const ValueDecl *Member = E->getMemberDecl();
 
   if (DiscardResult)
     return this->discard(Base);
 
+  // MemberExprs are almost always lvalues, in which case we don't need to
+  // do the load. But sometimes they aren't.
+  const auto maybeLoadValue = [&]() -> bool {
+    if (E->isGLValue())
+      return true;
+    if (std::optional<PrimType> T = classify(E))
+      return this->emitLoadPop(*T, E);
+    return false;
+  };
+
+  if (const auto *VD = dyn_cast<VarDecl>(Member)) {
+    // I am almost confident in saying that a var decl must be static
+    // and therefore registered as a global variable. But this will probably
+    // turn out to be wrong some time in the future, as always.
+    if (auto GlobalIndex = P.getGlobal(VD))
+      return this->emitGetPtrGlobal(*GlobalIndex, E) && maybeLoadValue();
+    return false;
+  }
+
   if (Initializing) {
     if (!this->delegate(Base))
       return false;
@@ -1271,16 +1329,14 @@ bool ByteCodeExprGen<Emitter>::VisitMemberExpr(const MemberExpr *E) {
   }
 
   // Base above gives us a pointer on the stack.
-  // TODO: Implement non-FieldDecl members.
-  const ValueDecl *Member = E->getMemberDecl();
   if (const auto *FD = dyn_cast<FieldDecl>(Member)) {
     const RecordDecl *RD = FD->getParent();
     const Record *R = getRecord(RD);
     const Record::Field *F = R->getField(FD);
     // Leave a pointer to the field on the stack.
     if (F->Decl->getType()->isReferenceType())
-      return this->emitGetFieldPop(PT_Ptr, F->Offset, E);
-    return this->emitGetPtrField(F->Offset, E);
+      return this->emitGetFieldPop(PT_Ptr, F->Offset, E) && maybeLoadValue();
+    return this->emitGetPtrField(F->Offset, E) && maybeLoadValue();
   }
 
   return false;
@@ -1615,7 +1671,7 @@ bool ByteCodeExprGen<Emitter>::VisitCompoundAssignOperator(
     return false;
   if (!this->emitLoad(*LT, E))
     return false;
-  if (*LT != *LHSComputationT) {
+  if (LT != LHSComputationT) {
     if (!this->emitCast(*LT, *LHSComputationT, E))
       return false;
   }
@@ -1671,7 +1727,7 @@ bool ByteCodeExprGen<Emitter>::VisitCompoundAssignOperator(
   }
 
   // And now cast from LHSComputationT to ResultT.
-  if (*ResultT != *LHSComputationT) {
+  if (ResultT != LHSComputationT) {
     if (!this->emitCast(*LHSComputationT, *ResultT, E))
       return false;
   }

diff --git a/clang/lib/AST/Interp/ByteCodeExprGen.h b/clang/lib/AST/Interp/ByteCodeExprGen.h
@@ -148,13 +148,20 @@ class ByteCodeExprGen : public ConstStmtVisitor<ByteCodeExprGen<Emitter>, bool>,
     return Ctx.classify(Ty);
   }
 
-  /// Classifies a known primitive type
+  /// Classifies a known primitive type.
   PrimType classifyPrim(QualType Ty) const {
     if (auto T = classify(Ty)) {
       return *T;
     }
     llvm_unreachable("not a primitive type");
   }
+  /// Classifies a known primitive expression.
+  PrimType classifyPrim(const Expr *E) const {
+    if (auto T = classify(E))
+      return *T;
+    llvm_unreachable("not a primitive type");
+  }
+
   /// Evaluates an expression and places the result on the stack. If the
   /// expression is of composite type, a local variable will be created
   /// and a pointer to said variable will be placed on the stack.

diff --git a/clang/lib/AST/Interp/Disasm.cpp b/clang/lib/AST/Interp/Disasm.cpp
@@ -140,7 +140,7 @@ LLVM_DUMP_METHOD void Program::dump(llvm::raw_ostream &OS) const {
     const Descriptor *Desc = G->block()->getDescriptor();
     Pointer GP = getPtrGlobal(GI);
 
-    OS << GI << ": " << (void *)G->block() << " ";
+    OS << GI << ": " << (const void *)G->block() << " ";
     {
       ColorScope SC(OS, true,
                     GP.isInitialized()
@@ -264,3 +264,19 @@ LLVM_DUMP_METHOD void Record::dump(llvm::raw_ostream &OS, unsigned Indentation,
     ++I;
   }
 }
+
+LLVM_DUMP_METHOD void Block::dump(llvm::raw_ostream &OS) const {
+  {
+    ColorScope SC(OS, true, {llvm::raw_ostream::BRIGHT_BLUE, true});
+    OS << "Block " << (const void *)this << "\n";
+  }
+  unsigned NPointers = 0;
+  for (const Pointer *P = Pointers; P; P = P->Next) {
+    ++NPointers;
+  }
+  OS << "  Pointers: " << NPointers << "\n";
+  OS << "  Dead: " << IsDead << "\n";
+  OS << "  Static: " << IsStatic << "\n";
+  OS << "  Extern: " << IsExtern << "\n";
+  OS << "  Initialized: " << IsInitialized << "\n";
+}
diff --git a/clang/lib/AST/Interp/FunctionPointer.h b/clang/lib/AST/Interp/FunctionPointer.h
@@ -32,6 +32,13 @@ class FunctionPointer final {
 
   const Function *getFunction() const { return Func; }
   bool isZero() const { return !Func; }
+  bool isValid() const { return Valid; }
+  bool isWeak() const {
+    if (!Func || !Valid)
+      return false;
+
+    return Func->getDecl()->isWeak();
+  }
 
   APValue toAPValue() const {
     if (!Func)

diff --git a/clang/lib/AST/Interp/Interp.h b/clang/lib/AST/Interp/Interp.h
@@ -758,7 +758,7 @@ inline bool CmpHelperEQ<FunctionPointer>(InterpState &S, CodePtr OpPC,
 
   // We cannot compare against weak declarations at compile time.
   for (const auto &FP : {LHS, RHS}) {
-    if (!FP.isZero() && FP.getFunction()->getDecl()->isWeak()) {
+    if (FP.isWeak()) {
       const SourceInfo &Loc = S.Current->getSource(OpPC);
       S.FFDiag(Loc, diag::note_constexpr_pointer_weak_comparison)
           << FP.toDiagnosticString(S.getCtx());
@@ -2236,6 +2236,10 @@ inline bool CallPtr(InterpState &S, CodePtr OpPC, uint32_t ArgSize,
         << const_cast<Expr *>(E) << E->getSourceRange();
     return false;
   }
+
+  if (!FuncPtr.isValid())
+    return false;
+
   assert(F);
 
   // Check argument nullability state.

diff --git a/clang/lib/AST/Interp/InterpBlock.h b/clang/lib/AST/Interp/InterpBlock.h
@@ -118,6 +118,9 @@ class Block final {
     IsInitialized = false;
   }
 
+  void dump() const { dump(llvm::errs()); }
+  void dump(llvm::raw_ostream &OS) const;
+
 protected:
   friend class Pointer;
   friend class DeadBlock;

diff --git a/clang/lib/AST/Interp/InterpBuiltin.cpp b/clang/lib/AST/Interp/InterpBuiltin.cpp
@@ -977,6 +977,117 @@ static bool interp__builtin_complex(InterpState &S, CodePtr OpPC,
   return true;
 }
 
+/// __builtin_is_aligned()
+/// __builtin_align_up()
+/// __builtin_align_down()
+/// The first parameter is either an integer or a pointer.
+/// The second parameter is the requested alignment as an integer.
+static bool interp__builtin_is_aligned_up_down(InterpState &S, CodePtr OpPC,
+                                               const InterpFrame *Frame,
+                                               const Function *Func,
+                                               const CallExpr *Call) {
+  unsigned BuiltinOp = Func->getBuiltinID();
+  unsigned CallSize = callArgSize(S, Call);
+
+  PrimType AlignmentT = *S.Ctx.classify(Call->getArg(1));
+  const APSInt &Alignment = peekToAPSInt(S.Stk, AlignmentT);
+
+  if (Alignment < 0 || !Alignment.isPowerOf2()) {
+    S.FFDiag(Call, diag::note_constexpr_invalid_alignment) << Alignment;
+    return false;
+  }
+  unsigned SrcWidth = S.getCtx().getIntWidth(Call->getArg(0)->getType());
+  APSInt MaxValue(APInt::getOneBitSet(SrcWidth, SrcWidth - 1));
+  if (APSInt::compareValues(Alignment, MaxValue) > 0) {
+    S.FFDiag(Call, diag::note_constexpr_alignment_too_big)
+        << MaxValue << Call->getArg(0)->getType() << Alignment;
+    return false;
+  }
+
+  // The first parameter is either an integer or a pointer (but not a function
+  // pointer).
+  PrimType FirstArgT = *S.Ctx.classify(Call->getArg(0));
+
+  if (isIntegralType(FirstArgT)) {
+    const APSInt &Src = peekToAPSInt(S.Stk, FirstArgT, CallSize);
+    APSInt Align = Alignment.extOrTrunc(Src.getBitWidth());
+    if (BuiltinOp == Builtin::BI__builtin_align_up) {
+      APSInt AlignedVal =
+          APSInt((Src + (Align - 1)) & ~(Align - 1), Src.isUnsigned());
+      pushInteger(S, AlignedVal, Call->getType());
+    } else if (BuiltinOp == Builtin::BI__builtin_align_down) {
+      APSInt AlignedVal = APSInt(Src & ~(Align - 1), Src.isUnsigned());
+      pushInteger(S, AlignedVal, Call->getType());
+    } else {
+      assert(*S.Ctx.classify(Call->getType()) == PT_Bool);
+      S.Stk.push<Boolean>((Src & (Align - 1)) == 0);
+    }
+    return true;
+  }
+
+  assert(FirstArgT == PT_Ptr);
+  const Pointer &Ptr = S.Stk.peek<Pointer>(CallSize);
+
+  unsigned PtrOffset = Ptr.getByteOffset();
+  PtrOffset = Ptr.getIndex();
+  CharUnits BaseAlignment =
+      S.getCtx().getDeclAlign(Ptr.getDeclDesc()->asValueDecl());
+  CharUnits PtrAlign =
+      BaseAlignment.alignmentAtOffset(CharUnits::fromQuantity(PtrOffset));
+
+  if (BuiltinOp == Builtin::BI__builtin_is_aligned) {
+    if (PtrAlign.getQuantity() >= Alignment) {
+      S.Stk.push<Boolean>(true);
+      return true;
+    }
+    // If the alignment is not known to be sufficient, some cases could still
+    // be aligned at run time. However, if the requested alignment is less or
+    // equal to the base alignment and the offset is not aligned, we know that
+    // the run-time value can never be aligned.
+    if (BaseAlignment.getQuantity() >= Alignment &&
+        PtrAlign.getQuantity() < Alignment) {
+      S.Stk.push<Boolean>(false);
+      return true;
+    }
+
+    S.FFDiag(Call->getArg(0), diag::note_constexpr_alignment_compute)
+        << Alignment;
+    return false;
+  }
+
+  assert(BuiltinOp == Builtin::BI__builtin_align_down ||
+         BuiltinOp == Builtin::BI__builtin_align_up);
+
+  // For align_up/align_down, we can return the same value if the alignment
+  // is known to be greater or equal to the requested value.
+  if (PtrAlign.getQuantity() >= Alignment) {
+    S.Stk.push<Pointer>(Ptr);
+    return true;
+  }
+
+  // The alignment could be greater than the minimum at run-time, so we cannot
+  // infer much about the resulting pointer value. One case is possible:
+  // For `_Alignas(32) char buf[N]; __builtin_align_down(&buf[idx], 32)` we
+  // can infer the correct index if the requested alignment is smaller than
+  // the base alignment so we can perform the computation on the offset.
+  if (BaseAlignment.getQuantity() >= Alignment) {
+    assert(Alignment.getBitWidth() <= 64 &&
+           "Cannot handle > 64-bit address-space");
+    uint64_t Alignment64 = Alignment.getZExtValue();
+    CharUnits NewOffset =
+        CharUnits::fromQuantity(BuiltinOp == Builtin::BI__builtin_align_down
+                                    ? llvm::alignDown(PtrOffset, Alignment64)
+                                    : llvm::alignTo(PtrOffset, Alignment64));
+
+    S.Stk.push<Pointer>(Ptr.atIndex(NewOffset.getQuantity()));
+    return true;
+  }
+
+  // Otherwise, we cannot constant-evaluate the result.
+  S.FFDiag(Call->getArg(0), diag::note_constexpr_alignment_adjust) << Alignment;
+  return false;
+}
+
 bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const Function *F,
                       const CallExpr *Call) {
   const InterpFrame *Frame = S.Current;
@@ -1291,6 +1402,13 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const Function *F,
       return false;
     break;
 
+  case Builtin::BI__builtin_is_aligned:
+  case Builtin::BI__builtin_align_up:
+  case Builtin::BI__builtin_align_down:
+    if (!interp__builtin_is_aligned_up_down(S, OpPC, Frame, F, Call))
+      return false;
+    break;
+
   default:
     S.FFDiag(S.Current->getLocation(OpPC),
              diag::note_invalid_subexpr_in_const_expr)

diff --git a/clang/lib/AST/Interp/InterpFrame.cpp b/clang/lib/AST/Interp/InterpFrame.cpp
@@ -152,6 +152,13 @@ void print(llvm::raw_ostream &OS, const Pointer &P, ASTContext &Ctx,
 }
 
 void InterpFrame::describe(llvm::raw_ostream &OS) const {
+  // We create frames for builtin functions as well, but we can't reliably
+  // diagnose them. The 'in call to' diagnostics for them add no value to the
+  // user _and_ it doesn't generally work since the argument types don't always
+  // match the function prototype. Just ignore them.
+  if (const auto *F = getFunction(); F && F->isBuiltin())
+    return;
+
   const FunctionDecl *F = getCallee();
   if (const auto *M = dyn_cast<CXXMethodDecl>(F);
       M && M->isInstance() && !isa<CXXConstructorDecl>(F)) {

diff --git a/clang/lib/AST/Interp/InterpState.h b/clang/lib/AST/Interp/InterpState.h
@@ -89,7 +89,11 @@ class InterpState final : public State, public SourceMapper {
 
   /// Delegates source mapping to the mapper.
   SourceInfo getSource(const Function *F, CodePtr PC) const override {
-    return M ? M->getSource(F, PC) : F->getSource(PC);
+    if (M)
+      return M->getSource(F, PC);
+
+    assert(F && "Function cannot be null");
+    return F->getSource(PC);
   }
 
   Context &getContext() const { return Ctx; }

diff --git a/clang/lib/AST/Interp/Pointer.h b/clang/lib/AST/Interp/Pointer.h
@@ -241,13 +241,10 @@ class Pointer {
 
   /// Checks if the pointer is null.
   bool isZero() const {
-    if (Offset != 0)
-      return false;
-
     if (isBlockPointer())
       return asBlockPointer().Pointee == nullptr;
     assert(isIntegralPointer());
-    return asIntPointer().Value == 0;
+    return asIntPointer().Value == 0 && Offset == 0;
   }
   /// Checks if the pointer is live.
   bool isLive() const {

diff --git a/clang/lib/AST/Interp/State.cpp b/clang/lib/AST/Interp/State.cpp
@@ -155,7 +155,8 @@ void State::addCallStack(unsigned Limit) {
     SmallString<128> Buffer;
     llvm::raw_svector_ostream Out(Buffer);
     F->describe(Out);
-    addDiag(CallRange.getBegin(), diag::note_constexpr_call_here)
-        << Out.str() << CallRange;
+    if (!Buffer.empty())
+      addDiag(CallRange.getBegin(), diag::note_constexpr_call_here)
+          << Out.str() << CallRange;
   }
 }
diff --git a/clang/lib/AST/JSONNodeDumper.cpp b/clang/lib/AST/JSONNodeDumper.cpp
@@ -975,6 +975,9 @@ void JSONNodeDumper::VisitFunctionDecl(const FunctionDecl *FD) {
   if (FD->isDefaulted())
     JOS.attribute("explicitlyDefaulted",
                   FD->isDeleted() ? "deleted" : "default");
+
+  if (StringLiteral *Msg = FD->getDeletedMessage())
+    JOS.attribute("deletedMessage", Msg->getString());
 }
 
 void JSONNodeDumper::VisitEnumDecl(const EnumDecl *ED) {
@@ -1576,6 +1579,14 @@ void JSONNodeDumper::VisitMaterializeTemporaryExpr(
   attributeOnlyIfTrue("boundToLValueRef", MTE->isBoundToLvalueReference());
 }
 
+void JSONNodeDumper::VisitCXXDefaultArgExpr(const CXXDefaultArgExpr *Node) {
+  attributeOnlyIfTrue("hasRewrittenInit", Node->hasRewrittenInit());
+}
+
+void JSONNodeDumper::VisitCXXDefaultInitExpr(const CXXDefaultInitExpr *Node) {
+  attributeOnlyIfTrue("hasRewrittenInit", Node->hasRewrittenInit());
+}
+
 void JSONNodeDumper::VisitCXXDependentScopeMemberExpr(
     const CXXDependentScopeMemberExpr *DSME) {
   JOS.attribute("isArrow", DSME->isArrow());

diff --git a/clang/lib/AST/ODRHash.cpp b/clang/lib/AST/ODRHash.cpp
@@ -696,6 +696,12 @@ void ODRHash::AddFunctionDecl(const FunctionDecl *Function,
   AddBoolean(Function->isDeletedAsWritten());
   AddBoolean(Function->isExplicitlyDefaulted());
 
+  StringLiteral *DeletedMessage = Function->getDeletedMessage();
+  AddBoolean(DeletedMessage);
+
+  if (DeletedMessage)
+    ID.AddString(DeletedMessage->getBytes());
+
   AddDecl(Function);
 
   AddQualType(Function->getReturnType());

diff --git a/clang/lib/AST/OpenACCClause.cpp b/clang/lib/AST/OpenACCClause.cpp
@@ -13,6 +13,7 @@
 
 #include "clang/AST/OpenACCClause.h"
 #include "clang/AST/ASTContext.h"
+#include "clang/AST/Expr.h"
 
 using namespace clang;
 
@@ -27,10 +28,73 @@ OpenACCDefaultClause *OpenACCDefaultClause::Create(const ASTContext &C,
   return new (Mem) OpenACCDefaultClause(K, BeginLoc, LParenLoc, EndLoc);
 }
 
+OpenACCIfClause *OpenACCIfClause::Create(const ASTContext &C,
+                                         SourceLocation BeginLoc,
+                                         SourceLocation LParenLoc,
+                                         Expr *ConditionExpr,
+                                         SourceLocation EndLoc) {
+  void *Mem = C.Allocate(sizeof(OpenACCIfClause), alignof(OpenACCIfClause));
+  return new (Mem) OpenACCIfClause(BeginLoc, LParenLoc, ConditionExpr, EndLoc);
+}
+
+OpenACCIfClause::OpenACCIfClause(SourceLocation BeginLoc,
+                                 SourceLocation LParenLoc, Expr *ConditionExpr,
+                                 SourceLocation EndLoc)
+    : OpenACCClauseWithCondition(OpenACCClauseKind::If, BeginLoc, LParenLoc,
+                                 ConditionExpr, EndLoc) {
+  assert(ConditionExpr && "if clause requires condition expr");
+  assert((ConditionExpr->isInstantiationDependent() ||
+          ConditionExpr->getType()->isScalarType()) &&
+         "Condition expression type not scalar/dependent");
+}
+
+OpenACCSelfClause *OpenACCSelfClause::Create(const ASTContext &C,
+                                             SourceLocation BeginLoc,
+                                             SourceLocation LParenLoc,
+                                             Expr *ConditionExpr,
+                                             SourceLocation EndLoc) {
+  void *Mem = C.Allocate(sizeof(OpenACCIfClause), alignof(OpenACCIfClause));
+  return new (Mem)
+      OpenACCSelfClause(BeginLoc, LParenLoc, ConditionExpr, EndLoc);
+}
+
+OpenACCSelfClause::OpenACCSelfClause(SourceLocation BeginLoc,
+                                     SourceLocation LParenLoc,
+                                     Expr *ConditionExpr, SourceLocation EndLoc)
+    : OpenACCClauseWithCondition(OpenACCClauseKind::Self, BeginLoc, LParenLoc,
+                                 ConditionExpr, EndLoc) {
+  assert((!ConditionExpr || ConditionExpr->isInstantiationDependent() ||
+          ConditionExpr->getType()->isScalarType()) &&
+         "Condition expression type not scalar/dependent");
+}
+
+OpenACCClause::child_range OpenACCClause::children() {
+  switch (getClauseKind()) {
+  default:
+    assert(false && "Clause children function not implemented");
+    break;
+#define VISIT_CLAUSE(CLAUSE_NAME)                                              \
+  case OpenACCClauseKind::CLAUSE_NAME:                                         \
+    return cast<OpenACC##CLAUSE_NAME##Clause>(this)->children();
+
+#include "clang/Basic/OpenACCClauses.def"
+  }
+  return child_range(child_iterator(), child_iterator());
+}
+
 //===----------------------------------------------------------------------===//
 //  OpenACC clauses printing methods
 //===----------------------------------------------------------------------===//
-void OpenACCClausePrinter::VisitOpenACCDefaultClause(
-    const OpenACCDefaultClause &C) {
+void OpenACCClausePrinter::VisitDefaultClause(const OpenACCDefaultClause &C) {
   OS << "default(" << C.getDefaultClauseKind() << ")";
 }
+
+void OpenACCClausePrinter::VisitIfClause(const OpenACCIfClause &C) {
+  OS << "if(" << C.getConditionExpr() << ")";
+}
+
+void OpenACCClausePrinter::VisitSelfClause(const OpenACCSelfClause &C) {
+  OS << "self";
+  if (const Expr *CondExpr = C.getConditionExpr())
+    OS << "(" << CondExpr << ")";
+}
diff --git a/clang/lib/AST/StmtProfile.cpp b/clang/lib/AST/StmtProfile.cpp
@@ -2071,13 +2071,31 @@ StmtProfiler::VisitLambdaExpr(const LambdaExpr *S) {
   }
 
   CXXRecordDecl *Lambda = S->getLambdaClass();
-  ID.AddInteger(Lambda->getODRHash());
-
   for (const auto &Capture : Lambda->captures()) {
     ID.AddInteger(Capture.getCaptureKind());
     if (Capture.capturesVariable())
       VisitDecl(Capture.getCapturedVar());
   }
+
+  // Profiling the body of the lambda may be dangerous during deserialization.
+  // So we'd like only to profile the signature here.
+  ODRHash Hasher;
+  // FIXME: We can't get the operator call easily by
+  // `CXXRecordDecl::getLambdaCallOperator()` if we're in deserialization.
+  // So we have to do something raw here.
+  for (auto *SubDecl : Lambda->decls()) {
+    FunctionDecl *Call = nullptr;
+    if (auto *FTD = dyn_cast<FunctionTemplateDecl>(SubDecl))
+      Call = FTD->getTemplatedDecl();
+    else if (auto *FD = dyn_cast<FunctionDecl>(SubDecl))
+      Call = FD;
+
+    if (!Call)
+      continue;
+
+    Hasher.AddFunctionDecl(Call, /*SkipBody=*/true);
+  }
+  ID.AddInteger(Hasher.CalculateHash());
 }
 
 void
@@ -2445,9 +2463,10 @@ void StmtProfiler::VisitTemplateArgument(const TemplateArgument &Arg) {
 namespace {
 class OpenACCClauseProfiler
     : public OpenACCClauseVisitor<OpenACCClauseProfiler> {
+  StmtProfiler &Profiler;
 
 public:
-  OpenACCClauseProfiler() = default;
+  OpenACCClauseProfiler(StmtProfiler &P) : Profiler(P) {}
 
   void VisitOpenACCClauseList(ArrayRef<const OpenACCClause *> Clauses) {
     for (const OpenACCClause *Clause : Clauses) {
@@ -2456,20 +2475,35 @@ class OpenACCClauseProfiler
       Visit(Clause);
     }
   }
-  void VisitOpenACCDefaultClause(const OpenACCDefaultClause &Clause);
+
+#define VISIT_CLAUSE(CLAUSE_NAME)                                              \
+  void Visit##CLAUSE_NAME##Clause(const OpenACC##CLAUSE_NAME##Clause &Clause);
+
+#include "clang/Basic/OpenACCClauses.def"
 };
 
 /// Nothing to do here, there are no sub-statements.
-void OpenACCClauseProfiler::VisitOpenACCDefaultClause(
+void OpenACCClauseProfiler::VisitDefaultClause(
     const OpenACCDefaultClause &Clause) {}
+
+void OpenACCClauseProfiler::VisitIfClause(const OpenACCIfClause &Clause) {
+  assert(Clause.hasConditionExpr() &&
+         "if clause requires a valid condition expr");
+  Profiler.VisitStmt(Clause.getConditionExpr());
+}
+
+void OpenACCClauseProfiler::VisitSelfClause(const OpenACCSelfClause &Clause) {
+  if (Clause.hasConditionExpr())
+    Profiler.VisitStmt(Clause.getConditionExpr());
+}
 } // namespace
 
 void StmtProfiler::VisitOpenACCComputeConstruct(
     const OpenACCComputeConstruct *S) {
   // VisitStmt handles children, so the AssociatedStmt is handled.
   VisitStmt(S);
 
-  OpenACCClauseProfiler P;
+  OpenACCClauseProfiler P{*this};
   P.VisitOpenACCClauseList(S->clauses());
 }
 

diff --git a/clang/lib/AST/TextNodeDumper.cpp b/clang/lib/AST/TextNodeDumper.cpp
@@ -397,6 +397,12 @@ void TextNodeDumper::Visit(const OpenACCClause *C) {
     case OpenACCClauseKind::Default:
       OS << '(' << cast<OpenACCDefaultClause>(C)->getDefaultClauseKind() << ')';
       break;
+    case OpenACCClauseKind::If:
+    case OpenACCClauseKind::Self:
+      // The condition expression will be printed as a part of the 'children',
+      // but print 'clause' here so it is clear what is happening from the dump.
+      OS << " clause";
+      break;
     default:
       // Nothing to do here.
       break;
@@ -1450,23 +1456,13 @@ void TextNodeDumper::VisitExpressionTraitExpr(const ExpressionTraitExpr *Node) {
 }
 
 void TextNodeDumper::VisitCXXDefaultArgExpr(const CXXDefaultArgExpr *Node) {
-  if (Node->hasRewrittenInit()) {
+  if (Node->hasRewrittenInit())
     OS << " has rewritten init";
-    AddChild([=] {
-      ColorScope Color(OS, ShowColors, StmtColor);
-      Visit(Node->getExpr());
-    });
-  }
 }
 
 void TextNodeDumper::VisitCXXDefaultInitExpr(const CXXDefaultInitExpr *Node) {
-  if (Node->hasRewrittenInit()) {
+  if (Node->hasRewrittenInit())
     OS << " has rewritten init";
-    AddChild([=] {
-      ColorScope Color(OS, ShowColors, StmtColor);
-      Visit(Node->getExpr());
-    });
-  }
 }
 
 void TextNodeDumper::VisitMaterializeTemporaryExpr(
@@ -1966,6 +1962,9 @@ void TextNodeDumper::VisitFunctionDecl(const FunctionDecl *D) {
   if (D->isTrivial())
     OS << " trivial";
 
+  if (const StringLiteral *M = D->getDeletedMessage())
+    AddChild("delete message", [=] { Visit(M); });
+
   if (D->isIneligibleOrNotSelected())
     OS << (isa<CXXDestructorDecl>(D) ? " not_selected" : " ineligible");
 

diff --git a/clang/lib/Analysis/FlowSensitive/DataflowEnvironment.cpp b/clang/lib/Analysis/FlowSensitive/DataflowEnvironment.cpp
@@ -508,6 +508,11 @@ class ResultObjectVisitor : public RecursiveASTVisitor<ResultObjectVisitor> {
         isa<CXXStdInitializerListExpr>(E)) {
       return;
     }
+    if (auto *Op = dyn_cast<BinaryOperator>(E);
+        Op && Op->getOpcode() == BO_Cmp) {
+      // Builtin `<=>` returns a `std::strong_ordering` object.
+      return;
+    }
 
     if (auto *InitList = dyn_cast<InitListExpr>(E)) {
       if (!InitList->isSemanticForm())

diff --git a/clang/lib/Basic/Cuda.cpp b/clang/lib/Basic/Cuda.cpp
@@ -86,7 +86,7 @@ static const CudaArchToStringMap arch_names[] = {
     // clang-format off
     {CudaArch::UNUSED, "", ""},
     SM2(20, "compute_20"), SM2(21, "compute_20"), // Fermi
-    SM(30), SM(32), SM(35), SM(37),  // Kepler
+    SM(30), {CudaArch::SM_32_, "sm_32", "compute_32"}, SM(35), SM(37),  // Kepler
     SM(50), SM(52), SM(53),          // Maxwell
     SM(60), SM(61), SM(62),          // Pascal
     SM(70), SM(72),                  // Volta
@@ -186,7 +186,7 @@ CudaVersion MinVersionForCudaArch(CudaArch A) {
   case CudaArch::SM_20:
   case CudaArch::SM_21:
   case CudaArch::SM_30:
-  case CudaArch::SM_32:
+  case CudaArch::SM_32_:
   case CudaArch::SM_35:
   case CudaArch::SM_37:
   case CudaArch::SM_50:
@@ -231,7 +231,7 @@ CudaVersion MaxVersionForCudaArch(CudaArch A) {
   case CudaArch::SM_21:
     return CudaVersion::CUDA_80;
   case CudaArch::SM_30:
-  case CudaArch::SM_32:
+  case CudaArch::SM_32_:
     return CudaVersion::CUDA_102;
   case CudaArch::SM_35:
   case CudaArch::SM_37:

diff --git a/clang/lib/Basic/Module.cpp b/clang/lib/Basic/Module.cpp
@@ -305,6 +305,10 @@ bool Module::directlyUses(const Module *Requested) {
   if (Requested->fullModuleNameIs({"_Builtin_stddef", "max_align_t"}) ||
       Requested->fullModuleNameIs({"_Builtin_stddef_wint_t"}))
     return true;
+  // Darwin is allowed is to use our builtin 'ptrauth.h' and its accompanying
+  // module.
+  if (!Requested->Parent && Requested->Name == "ptrauth")
+    return true;
 
   if (NoUndeclaredIncludes)
     UndeclaredUses.insert(Requested);

diff --git a/clang/lib/Basic/Targets/NVPTX.cpp b/clang/lib/Basic/Targets/NVPTX.cpp
@@ -239,7 +239,7 @@ void NVPTXTargetInfo::getTargetDefines(const LangOptions &Opts,
         return "210";
       case CudaArch::SM_30:
         return "300";
-      case CudaArch::SM_32:
+      case CudaArch::SM_32_:
         return "320";
       case CudaArch::SM_35:
         return "350";

diff --git a/clang/lib/Basic/Targets/SPIR.h b/clang/lib/Basic/Targets/SPIR.h
@@ -259,7 +259,7 @@ class LLVM_LIBRARY_VISIBILITY SPIR32TargetInfo : public SPIRTargetInfo {
     SizeType = TargetInfo::UnsignedInt;
     PtrDiffType = IntPtrType = TargetInfo::SignedInt;
     resetDataLayout("e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-"
-                    "v96:128-v192:256-v256:256-v512:512-v1024:1024");
+                    "v96:128-v192:256-v256:256-v512:512-v1024:1024-G1");
   }
 
   void getTargetDefines(const LangOptions &Opts,
@@ -276,7 +276,7 @@ class LLVM_LIBRARY_VISIBILITY SPIR64TargetInfo : public SPIRTargetInfo {
     SizeType = TargetInfo::UnsignedLong;
     PtrDiffType = IntPtrType = TargetInfo::SignedLong;
     resetDataLayout("e-i64:64-v16:16-v24:32-v32:32-v48:64-"
-                    "v96:128-v192:256-v256:256-v512:512-v1024:1024");
+                    "v96:128-v192:256-v256:256-v512:512-v1024:1024-G1");
   }
 
   void getTargetDefines(const LangOptions &Opts,
@@ -336,7 +336,7 @@ class LLVM_LIBRARY_VISIBILITY SPIRV32TargetInfo : public BaseSPIRVTargetInfo {
     SizeType = TargetInfo::UnsignedInt;
     PtrDiffType = IntPtrType = TargetInfo::SignedInt;
     resetDataLayout("e-p:32:32-i64:64-v16:16-v24:32-v32:32-v48:64-"
-                    "v96:128-v192:256-v256:256-v512:512-v1024:1024");
+                    "v96:128-v192:256-v256:256-v512:512-v1024:1024-G1");
   }
 
   void getTargetDefines(const LangOptions &Opts,
@@ -357,7 +357,7 @@ class LLVM_LIBRARY_VISIBILITY SPIRV64TargetInfo : public BaseSPIRVTargetInfo {
     SizeType = TargetInfo::UnsignedLong;
     PtrDiffType = IntPtrType = TargetInfo::SignedLong;
     resetDataLayout("e-i64:64-v16:16-v24:32-v32:32-v48:64-"
-                    "v96:128-v192:256-v256:256-v512:512-v1024:1024");
+                    "v96:128-v192:256-v256:256-v512:512-v1024:1024-G1");
   }
 
   void getTargetDefines(const LangOptions &Opts,

diff --git a/clang/lib/CIR/CMakeLists.txt b/clang/lib/CIR/CMakeLists.txt
diff --git a/clang/lib/CMakeLists.txt b/clang/lib/CMakeLists.txt
@@ -31,3 +31,7 @@ if(CLANG_INCLUDE_TESTS)
 endif()
 add_subdirectory(Interpreter)
 add_subdirectory(Support)
+
+if(CLANG_ENABLE_CIR)
+  add_subdirectory(CIR)
+endif()
diff --git a/clang/lib/CodeGen/CGAtomic.cpp b/clang/lib/CodeGen/CGAtomic.cpp
@@ -1806,7 +1806,11 @@ void AtomicInfo::EmitAtomicUpdateOp(
                                              /*NumReservedValues=*/2);
   PHI->addIncoming(OldVal, CurBB);
   Address NewAtomicAddr = CreateTempAlloca();
-  Address NewAtomicIntAddr = castToAtomicIntPointer(NewAtomicAddr);
+  Address NewAtomicIntAddr =
+      shouldCastToInt(NewAtomicAddr.getElementType(), /*CmpXchg=*/true)
+          ? castToAtomicIntPointer(NewAtomicAddr)
+          : NewAtomicAddr;
+
   if ((LVal.isBitField() && BFI.Size != ValueSizeInBits) ||
       requiresMemSetZero(getAtomicAddress().getElementType())) {
     CGF.Builder.CreateStore(PHI, NewAtomicIntAddr);

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -18194,7 +18194,8 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
     Value *Op0 = EmitScalarExpr(E->getArg(0));
     return Builder.CreateIntrinsic(
         /*ReturnType=*/llvm::Type::getInt1Ty(getLLVMContext()),
-        Intrinsic::dx_any, ArrayRef<Value *>{Op0}, nullptr, "dx.any");
+        CGM.getHLSLRuntime().getAnyIntrinsic(), ArrayRef<Value *>{Op0}, nullptr,
+        "hlsl.any");
   }
   case Builtin::BI__builtin_hlsl_elementwise_clamp: {
     Value *OpX = EmitScalarExpr(E->getArg(0));
@@ -18303,9 +18304,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
     Value *Op0 = EmitScalarExpr(E->getArg(0));
     if (!E->getArg(0)->getType()->hasFloatingRepresentation())
       llvm_unreachable("rcp operand must have a float representation");
-    return Builder.CreateIntrinsic(
-        /*ReturnType=*/Op0->getType(), Intrinsic::dx_rcp,
-        ArrayRef<Value *>{Op0}, nullptr, "dx.rcp");
+    llvm::Type *Ty = Op0->getType();
+    llvm::Type *EltTy = Ty->getScalarType();
+    Constant *One =
+        Ty->isVectorTy()
+            ? ConstantVector::getSplat(
+                  ElementCount::getFixed(
+                      dyn_cast<FixedVectorType>(Ty)->getNumElements()),
+                  ConstantFP::get(EltTy, 1.0))
+            : ConstantFP::get(EltTy, 1.0);
+    return Builder.CreateFDiv(One, Op0, "hlsl.rcp");
   }
   case Builtin::BI__builtin_hlsl_elementwise_rsqrt: {
     Value *Op0 = EmitScalarExpr(E->getArg(0));

diff --git a/clang/lib/CodeGen/CGCleanup.cpp b/clang/lib/CodeGen/CGCleanup.cpp
@@ -667,8 +667,7 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
 
   // - whether there's a fallthrough
   llvm::BasicBlock *FallthroughSource = Builder.GetInsertBlock();
-  bool HasFallthrough =
-      FallthroughSource != nullptr && (IsActive || HasExistingBranches);
+  bool HasFallthrough = (FallthroughSource != nullptr && IsActive);
 
   // Branch-through fall-throughs leave the insertion point set to the
   // end of the last cleanup, which points to the current scope.  The
@@ -693,11 +692,7 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
 
   // If we have a prebranched fallthrough into an inactive normal
   // cleanup, rewrite it so that it leads to the appropriate place.
-  if (Scope.isNormalCleanup() && HasPrebranchedFallthrough &&
-      !RequiresNormalCleanup) {
-    // FIXME: Come up with a program which would need forwarding prebranched
-    // fallthrough and add tests. Otherwise delete this and assert against it.
-    assert(!IsActive);
+  if (Scope.isNormalCleanup() && HasPrebranchedFallthrough && !IsActive) {
     llvm::BasicBlock *prebranchDest;
 
     // If the prebranch is semantically branching through the next
@@ -770,7 +765,6 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
         EmitSehCppScopeEnd();
     }
     destroyOptimisticNormalEntry(*this, Scope);
-    Scope.MarkEmitted();
     EHStack.popCleanup();
   } else {
     // If we have a fallthrough and no other need for the cleanup,
@@ -787,7 +781,6 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
       }
 
       destroyOptimisticNormalEntry(*this, Scope);
-      Scope.MarkEmitted();
       EHStack.popCleanup();
 
       EmitCleanup(*this, Fn, cleanupFlags, NormalActiveFlag);
@@ -923,7 +916,6 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
       }
 
       // IV.  Pop the cleanup and emit it.
-      Scope.MarkEmitted();
       EHStack.popCleanup();
       assert(EHStack.hasNormalCleanups() == HasEnclosingCleanups);
 

diff --git a/clang/lib/CodeGen/CGCleanup.h b/clang/lib/CodeGen/CGCleanup.h
@@ -16,11 +16,8 @@
 #include "EHScopeStack.h"
 
 #include "Address.h"
-#include "llvm/ADT/STLExtras.h"
-#include "llvm/ADT/SetVector.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallVector.h"
-#include "llvm/IR/Instruction.h"
 
 namespace llvm {
 class BasicBlock;
@@ -269,51 +266,6 @@ class alignas(8) EHCleanupScope : public EHScope {
   };
   mutable struct ExtInfo *ExtInfo;
 
-  /// Erases auxillary allocas and their usages for an unused cleanup.
-  /// Cleanups should mark these allocas as 'used' if the cleanup is
-  /// emitted, otherwise these instructions would be erased.
-  struct AuxillaryAllocas {
-    SmallVector<llvm::Instruction *, 1> AuxAllocas;
-    bool used = false;
-
-    // Records a potentially unused instruction to be erased later.
-    void Add(llvm::AllocaInst *Alloca) { AuxAllocas.push_back(Alloca); }
-
-    // Mark all recorded instructions as used. These will not be erased later.
-    void MarkUsed() {
-      used = true;
-      AuxAllocas.clear();
-    }
-
-    ~AuxillaryAllocas() {
-      if (used)
-        return;
-      llvm::SetVector<llvm::Instruction *> Uses;
-      for (auto *Inst : llvm::reverse(AuxAllocas))
-        CollectUses(Inst, Uses);
-      // Delete uses in the reverse order of insertion.
-      for (auto *I : llvm::reverse(Uses))
-        I->eraseFromParent();
-    }
-
-  private:
-    void CollectUses(llvm::Instruction *I,
-                     llvm::SetVector<llvm::Instruction *> &Uses) {
-      if (!I || !Uses.insert(I))
-        return;
-      for (auto *User : I->users())
-        CollectUses(cast<llvm::Instruction>(User), Uses);
-    }
-  };
-  mutable struct AuxillaryAllocas *AuxAllocas;
-
-  AuxillaryAllocas &getAuxillaryAllocas() {
-    if (!AuxAllocas) {
-      AuxAllocas = new struct AuxillaryAllocas();
-    }
-    return *AuxAllocas;
-  }
-
   /// 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.
@@ -346,7 +298,7 @@ class alignas(8) EHCleanupScope : public EHScope {
                  EHScopeStack::stable_iterator enclosingEH)
       : EHScope(EHScope::Cleanup, enclosingEH),
         EnclosingNormal(enclosingNormal), NormalBlock(nullptr),
-        ActiveFlag(Address::invalid()), ExtInfo(nullptr), AuxAllocas(nullptr),
+        ActiveFlag(Address::invalid()), ExtInfo(nullptr),
         FixupDepth(fixupDepth) {
     CleanupBits.IsNormalCleanup = isNormal;
     CleanupBits.IsEHCleanup = isEH;
@@ -360,15 +312,8 @@ class alignas(8) EHCleanupScope : public EHScope {
   }
 
   void Destroy() {
-    if (AuxAllocas)
-      delete AuxAllocas;
     delete ExtInfo;
   }
-  void AddAuxAllocas(llvm::SmallVector<llvm::AllocaInst *> Allocas) {
-    for (auto *Alloca : Allocas)
-      getAuxillaryAllocas().Add(Alloca);
-  }
-  void MarkEmitted() { getAuxillaryAllocas().MarkUsed(); }
   // Objects of EHCleanupScope are not destructed. Use Destroy().
   ~EHCleanupScope() = delete;
 

diff --git a/clang/lib/CodeGen/CGDecl.cpp b/clang/lib/CodeGen/CGDecl.cpp
@@ -19,7 +19,6 @@
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
 #include "ConstantEmitter.h"
-#include "EHScopeStack.h"
 #include "PatternInit.h"
 #include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
@@ -2202,24 +2201,6 @@ void CodeGenFunction::pushDestroy(CleanupKind cleanupKind, Address addr,
                                      destroyer, useEHCleanupForArray);
 }
 
-// Pushes a destroy and defers its deactivation until its
-// CleanupDeactivationScope is exited.
-void CodeGenFunction::pushDestroyAndDeferDeactivation(
-    QualType::DestructionKind dtorKind, Address addr, QualType type) {
-  assert(dtorKind && "cannot push destructor for trivial type");
-
-  CleanupKind cleanupKind = getCleanupKind(dtorKind);
-  pushDestroyAndDeferDeactivation(
-      cleanupKind, addr, type, getDestroyer(dtorKind), cleanupKind & EHCleanup);
-}
-
-void CodeGenFunction::pushDestroyAndDeferDeactivation(
-    CleanupKind cleanupKind, Address addr, QualType type, Destroyer *destroyer,
-    bool useEHCleanupForArray) {
-  pushCleanupAndDeferDeactivation<DestroyObject>(
-      cleanupKind, addr, type, destroyer, useEHCleanupForArray);
-}
-
 void CodeGenFunction::pushStackRestore(CleanupKind Kind, Address SPMem) {
   EHStack.pushCleanup<CallStackRestore>(Kind, SPMem);
 }
@@ -2236,19 +2217,16 @@ void CodeGenFunction::pushLifetimeExtendedDestroy(CleanupKind cleanupKind,
   // If we're not in a conditional branch, we don't need to bother generating a
   // conditional cleanup.
   if (!isInConditionalBranch()) {
+    // Push an EH-only cleanup for the object now.
     // FIXME: When popping normal cleanups, we need to keep this EH cleanup
     // around in case a temporary's destructor throws an exception.
+    if (cleanupKind & EHCleanup)
+      EHStack.pushCleanup<DestroyObject>(
+          static_cast<CleanupKind>(cleanupKind & ~NormalCleanup), addr, type,
+          destroyer, useEHCleanupForArray);
 
-    // Add the cleanup to the EHStack. After the full-expr, this would be
-    // deactivated before being popped from the stack.
-    pushDestroyAndDeferDeactivation(cleanupKind, addr, type, destroyer,
-                                    useEHCleanupForArray);
-
-    // Since this is lifetime-extended, push it once again to the EHStack after
-    // the full expression.
     return pushCleanupAfterFullExprWithActiveFlag<DestroyObject>(
-        cleanupKind, Address::invalid(), addr, type, destroyer,
-        useEHCleanupForArray);
+        cleanupKind, Address::invalid(), addr, type, destroyer, useEHCleanupForArray);
   }
 
   // Otherwise, we should only destroy the object if it's been initialized.
@@ -2263,12 +2241,13 @@ void CodeGenFunction::pushLifetimeExtendedDestroy(CleanupKind cleanupKind,
   Address ActiveFlag = createCleanupActiveFlag();
   SavedType SavedAddr = saveValueInCond(addr);
 
-  pushCleanupAndDeferDeactivation<ConditionalCleanupType>(
-      cleanupKind, SavedAddr, type, destroyer, useEHCleanupForArray);
-  initFullExprCleanupWithFlag(ActiveFlag);
+  if (cleanupKind & EHCleanup) {
+    EHStack.pushCleanup<ConditionalCleanupType>(
+        static_cast<CleanupKind>(cleanupKind & ~NormalCleanup), SavedAddr, type,
+        destroyer, useEHCleanupForArray);
+    initFullExprCleanupWithFlag(ActiveFlag);
+  }
 
-  // Since this is lifetime-extended, push it once again to the EHStack after
-  // the full expression.
   pushCleanupAfterFullExprWithActiveFlag<ConditionalCleanupType>(
       cleanupKind, ActiveFlag, SavedAddr, type, destroyer,
       useEHCleanupForArray);
@@ -2463,9 +2442,9 @@ namespace {
   };
 } // end anonymous namespace
 
-/// pushIrregularPartialArrayCleanup - Push a NormalAndEHCleanup to
-/// destroy already-constructed elements of the given array.  The cleanup may be
-/// popped with DeactivateCleanupBlock or PopCleanupBlock.
+/// pushIrregularPartialArrayCleanup - Push an EH cleanup to destroy
+/// already-constructed elements of the given array.  The cleanup
+/// may be popped with DeactivateCleanupBlock or PopCleanupBlock.
 ///
 /// \param elementType - the immediate element type of the array;
 ///   possibly still an array type
@@ -2474,9 +2453,10 @@ void CodeGenFunction::pushIrregularPartialArrayCleanup(llvm::Value *arrayBegin,
                                                        QualType elementType,
                                                        CharUnits elementAlign,
                                                        Destroyer *destroyer) {
-  pushFullExprCleanup<IrregularPartialArrayDestroy>(
-      NormalAndEHCleanup, arrayBegin, arrayEndPointer, elementType,
-      elementAlign, destroyer);
+  pushFullExprCleanup<IrregularPartialArrayDestroy>(EHCleanup,
+                                                    arrayBegin, arrayEndPointer,
+                                                    elementType, elementAlign,
+                                                    destroyer);
 }
 
 /// pushRegularPartialArrayCleanup - Push an EH cleanup to destroy

diff --git a/clang/lib/CodeGen/CGExpr.cpp b/clang/lib/CodeGen/CGExpr.cpp
@@ -115,16 +115,10 @@ RawAddress CodeGenFunction::CreateTempAlloca(llvm::Type *Ty, CharUnits Align,
 llvm::AllocaInst *CodeGenFunction::CreateTempAlloca(llvm::Type *Ty,
                                                     const Twine &Name,
                                                     llvm::Value *ArraySize) {
-  llvm::AllocaInst *Alloca;
   if (ArraySize)
-    Alloca = Builder.CreateAlloca(Ty, ArraySize, Name);
-  else
-    Alloca = new llvm::AllocaInst(Ty, CGM.getDataLayout().getAllocaAddrSpace(),
-                                  ArraySize, Name, AllocaInsertPt);
-  if (Allocas) {
-    Allocas->Add(Alloca);
-  }
-  return Alloca;
+    return Builder.CreateAlloca(Ty, ArraySize, Name);
+  return new llvm::AllocaInst(Ty, CGM.getDataLayout().getAllocaAddrSpace(),
+                              ArraySize, Name, AllocaInsertPt);
 }
 
 /// CreateDefaultAlignTempAlloca - This creates an alloca with the

diff --git a/clang/lib/CodeGen/CGExprAgg.cpp b/clang/lib/CodeGen/CGExprAgg.cpp
@@ -15,7 +15,6 @@
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
 #include "ConstantEmitter.h"
-#include "EHScopeStack.h"
 #include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
 #include "clang/AST/Attr.h"
@@ -25,7 +24,6 @@
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/Function.h"
 #include "llvm/IR/GlobalVariable.h"
-#include "llvm/IR/Instruction.h"
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Intrinsics.h"
 using namespace clang;
@@ -560,27 +558,24 @@ void AggExprEmitter::EmitArrayInit(Address DestPtr, llvm::ArrayType *AType,
   // For that, we'll need an EH cleanup.
   QualType::DestructionKind dtorKind = elementType.isDestructedType();
   Address endOfInit = Address::invalid();
-  CodeGenFunction::CleanupDeactivationScope deactivation(CGF);
-
-  if (dtorKind) {
-    CodeGenFunction::AllocaTrackerRAII allocaTracker(CGF);
+  EHScopeStack::stable_iterator cleanup;
+  llvm::Instruction *cleanupDominator = nullptr;
+  if (CGF.needsEHCleanup(dtorKind)) {
     // In principle we could tell the cleanup where we are more
     // directly, but the control flow can get so varied here that it
     // would actually be quite complex.  Therefore we go through an
     // alloca.
-    llvm::Instruction *dominatingIP =
-        Builder.CreateFlagLoad(llvm::ConstantInt::getNullValue(CGF.Int8PtrTy));
     endOfInit = CGF.CreateTempAlloca(begin->getType(), CGF.getPointerAlign(),
                                      "arrayinit.endOfInit");
-    Builder.CreateStore(begin, endOfInit);
+    cleanupDominator = Builder.CreateStore(begin, endOfInit);
     CGF.pushIrregularPartialArrayCleanup(begin, endOfInit, elementType,
                                          elementAlign,
                                          CGF.getDestroyer(dtorKind));
-    cast<EHCleanupScope>(*CGF.EHStack.find(CGF.EHStack.stable_begin()))
-        .AddAuxAllocas(allocaTracker.Take());
+    cleanup = CGF.EHStack.stable_begin();
 
-    CGF.DeferredDeactivationCleanupStack.push_back(
-        {CGF.EHStack.stable_begin(), dominatingIP});
+  // Otherwise, remember that we didn't need a cleanup.
+  } else {
+    dtorKind = QualType::DK_none;
   }
 
   llvm::Value *one = llvm::ConstantInt::get(CGF.SizeTy, 1);
@@ -676,6 +671,9 @@ void AggExprEmitter::EmitArrayInit(Address DestPtr, llvm::ArrayType *AType,
 
     CGF.EmitBlock(endBB);
   }
+
+  // Leave the partial-array cleanup if we entered one.
+  if (dtorKind) CGF.DeactivateCleanupBlock(cleanup, cleanupDominator);
 }
 
 //===----------------------------------------------------------------------===//
@@ -1376,8 +1374,9 @@ AggExprEmitter::VisitLambdaExpr(LambdaExpr *E) {
   LValue SlotLV = CGF.MakeAddrLValue(Slot.getAddress(), E->getType());
 
   // We'll need to enter cleanup scopes in case any of the element
-  // initializers throws an exception or contains branch out of the expressions.
-  CodeGenFunction::CleanupDeactivationScope scope(CGF);
+  // initializers throws an exception.
+  SmallVector<EHScopeStack::stable_iterator, 16> Cleanups;
+  llvm::Instruction *CleanupDominator = nullptr;
 
   CXXRecordDecl::field_iterator CurField = E->getLambdaClass()->field_begin();
   for (LambdaExpr::const_capture_init_iterator i = E->capture_init_begin(),
@@ -1396,12 +1395,28 @@ AggExprEmitter::VisitLambdaExpr(LambdaExpr *E) {
     if (QualType::DestructionKind DtorKind =
             CurField->getType().isDestructedType()) {
       assert(LV.isSimple());
-      if (DtorKind)
-        CGF.pushDestroyAndDeferDeactivation(
-            NormalAndEHCleanup, LV.getAddress(CGF), CurField->getType(),
-            CGF.getDestroyer(DtorKind), false);
+      if (CGF.needsEHCleanup(DtorKind)) {
+        if (!CleanupDominator)
+          CleanupDominator = CGF.Builder.CreateAlignedLoad(
+              CGF.Int8Ty,
+              llvm::Constant::getNullValue(CGF.Int8PtrTy),
+              CharUnits::One()); // placeholder
+
+        CGF.pushDestroy(EHCleanup, LV.getAddress(CGF), CurField->getType(),
+                        CGF.getDestroyer(DtorKind), false);
+        Cleanups.push_back(CGF.EHStack.stable_begin());
+      }
     }
   }
+
+  // Deactivate all the partial cleanups in reverse order, which
+  // generally means popping them.
+  for (unsigned i = Cleanups.size(); i != 0; --i)
+    CGF.DeactivateCleanupBlock(Cleanups[i-1], CleanupDominator);
+
+  // Destroy the placeholder if we made one.
+  if (CleanupDominator)
+    CleanupDominator->eraseFromParent();
 }
 
 void AggExprEmitter::VisitExprWithCleanups(ExprWithCleanups *E) {
@@ -1690,7 +1705,14 @@ void AggExprEmitter::VisitCXXParenListOrInitListExpr(
   // We'll need to enter cleanup scopes in case any of the element
   // initializers throws an exception.
   SmallVector<EHScopeStack::stable_iterator, 16> cleanups;
-  CodeGenFunction::CleanupDeactivationScope DeactivateCleanups(CGF);
+  llvm::Instruction *cleanupDominator = nullptr;
+  auto addCleanup = [&](const EHScopeStack::stable_iterator &cleanup) {
+    cleanups.push_back(cleanup);
+    if (!cleanupDominator) // create placeholder once needed
+      cleanupDominator = CGF.Builder.CreateAlignedLoad(
+          CGF.Int8Ty, llvm::Constant::getNullValue(CGF.Int8PtrTy),
+          CharUnits::One());
+  };
 
   unsigned curInitIndex = 0;
 
@@ -1713,8 +1735,10 @@ void AggExprEmitter::VisitCXXParenListOrInitListExpr(
       CGF.EmitAggExpr(InitExprs[curInitIndex++], AggSlot);
 
       if (QualType::DestructionKind dtorKind =
-              Base.getType().isDestructedType())
-        CGF.pushDestroyAndDeferDeactivation(dtorKind, V, Base.getType());
+              Base.getType().isDestructedType()) {
+        CGF.pushDestroy(dtorKind, V, Base.getType());
+        addCleanup(CGF.EHStack.stable_begin());
+      }
     }
   }
 
@@ -1789,10 +1813,10 @@ void AggExprEmitter::VisitCXXParenListOrInitListExpr(
     if (QualType::DestructionKind dtorKind
           = field->getType().isDestructedType()) {
       assert(LV.isSimple());
-      if (dtorKind) {
-        CGF.pushDestroyAndDeferDeactivation(
-            NormalAndEHCleanup, LV.getAddress(CGF), field->getType(),
-            CGF.getDestroyer(dtorKind), false);
+      if (CGF.needsEHCleanup(dtorKind)) {
+        CGF.pushDestroy(EHCleanup, LV.getAddress(CGF), field->getType(),
+                        CGF.getDestroyer(dtorKind), false);
+        addCleanup(CGF.EHStack.stable_begin());
         pushedCleanup = true;
       }
     }
@@ -1805,6 +1829,17 @@ void AggExprEmitter::VisitCXXParenListOrInitListExpr(
         if (GEP->use_empty())
           GEP->eraseFromParent();
   }
+
+  // Deactivate all the partial cleanups in reverse order, which
+  // generally means popping them.
+  assert((cleanupDominator || cleanups.empty()) &&
+         "Missing cleanupDominator before deactivating cleanup blocks");
+  for (unsigned i = cleanups.size(); i != 0; --i)
+    CGF.DeactivateCleanupBlock(cleanups[i-1], cleanupDominator);
+
+  // Destroy the placeholder if we made one.
+  if (cleanupDominator)
+    cleanupDominator->eraseFromParent();
 }
 
 void AggExprEmitter::VisitArrayInitLoopExpr(const ArrayInitLoopExpr *E,

diff --git a/clang/lib/CodeGen/CGExprCXX.cpp b/clang/lib/CodeGen/CGExprCXX.cpp
@@ -1008,8 +1008,8 @@ void CodeGenFunction::EmitNewArrayInitializer(
   const Expr *Init = E->getInitializer();
   Address EndOfInit = Address::invalid();
   QualType::DestructionKind DtorKind = ElementType.isDestructedType();
-  CleanupDeactivationScope deactivation(*this);
-  bool pushedCleanup = false;
+  EHScopeStack::stable_iterator Cleanup;
+  llvm::Instruction *CleanupDominator = nullptr;
 
   CharUnits ElementSize = getContext().getTypeSizeInChars(ElementType);
   CharUnits ElementAlign =
@@ -1105,24 +1105,19 @@ void CodeGenFunction::EmitNewArrayInitializer(
     }
 
     // Enter a partial-destruction Cleanup if necessary.
-    if (DtorKind) {
-      AllocaTrackerRAII AllocaTracker(*this);
+    if (needsEHCleanup(DtorKind)) {
       // In principle we could tell the Cleanup where we are more
       // directly, but the control flow can get so varied here that it
       // would actually be quite complex.  Therefore we go through an
       // alloca.
-      llvm::Instruction *DominatingIP =
-          Builder.CreateFlagLoad(llvm::ConstantInt::getNullValue(Int8PtrTy));
       EndOfInit = CreateTempAlloca(BeginPtr.getType(), getPointerAlign(),
                                    "array.init.end");
+      CleanupDominator =
+          Builder.CreateStore(BeginPtr.emitRawPointer(*this), EndOfInit);
       pushIrregularPartialArrayCleanup(BeginPtr.emitRawPointer(*this),
                                        EndOfInit, ElementType, ElementAlign,
                                        getDestroyer(DtorKind));
-      cast<EHCleanupScope>(*EHStack.find(EHStack.stable_begin()))
-          .AddAuxAllocas(AllocaTracker.Take());
-      DeferredDeactivationCleanupStack.push_back(
-          {EHStack.stable_begin(), DominatingIP});
-      pushedCleanup = true;
+      Cleanup = EHStack.stable_begin();
     }
 
     CharUnits StartAlign = CurPtr.getAlignment();
@@ -1169,6 +1164,9 @@ void CodeGenFunction::EmitNewArrayInitializer(
   // initialization.
   llvm::ConstantInt *ConstNum = dyn_cast<llvm::ConstantInt>(NumElements);
   if (ConstNum && ConstNum->getZExtValue() <= InitListElements) {
+    // If there was a Cleanup, deactivate it.
+    if (CleanupDominator)
+      DeactivateCleanupBlock(Cleanup, CleanupDominator);
     return;
   }
 
@@ -1283,22 +1281,24 @@ void CodeGenFunction::EmitNewArrayInitializer(
     Builder.CreateStore(CurPtr.emitRawPointer(*this), EndOfInit);
 
   // Enter a partial-destruction Cleanup if necessary.
-  if (!pushedCleanup && needsEHCleanup(DtorKind)) {
-    llvm::Instruction *DominatingIP =
-        Builder.CreateFlagLoad(llvm::ConstantInt::getNullValue(Int8PtrTy));
-    pushRegularPartialArrayCleanup(BeginPtr.emitRawPointer(*this),
-                                   CurPtr.emitRawPointer(*this), ElementType,
+  if (!CleanupDominator && needsEHCleanup(DtorKind)) {
+    llvm::Value *BeginPtrRaw = BeginPtr.emitRawPointer(*this);
+    llvm::Value *CurPtrRaw = CurPtr.emitRawPointer(*this);
+    pushRegularPartialArrayCleanup(BeginPtrRaw, CurPtrRaw, ElementType,
                                    ElementAlign, getDestroyer(DtorKind));
-    DeferredDeactivationCleanupStack.push_back(
-        {EHStack.stable_begin(), DominatingIP});
+    Cleanup = EHStack.stable_begin();
+    CleanupDominator = Builder.CreateUnreachable();
   }
 
   // Emit the initializer into this element.
   StoreAnyExprIntoOneUnit(*this, Init, Init->getType(), CurPtr,
                           AggValueSlot::DoesNotOverlap);
 
   // Leave the Cleanup if we entered one.
-  deactivation.ForceDeactivate();
+  if (CleanupDominator) {
+    DeactivateCleanupBlock(Cleanup, CleanupDominator);
+    CleanupDominator->eraseFromParent();
+  }
 
   // Advance to the next element by adjusting the pointer type as necessary.
   llvm::Value *NextPtr = Builder.CreateConstInBoundsGEP1_32(

diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -73,6 +73,7 @@ class CGHLSLRuntime {
   //===----------------------------------------------------------------------===//
 
   GENERATE_HLSL_INTRINSIC_FUNCTION(All, all)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Any, any)
   GENERATE_HLSL_INTRINSIC_FUNCTION(ThreadId, thread_id)
 
   //===----------------------------------------------------------------------===//

diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -3466,7 +3466,7 @@ void CGOpenMPRuntimeGPU::processRequiresDirective(
       case CudaArch::SM_20:
       case CudaArch::SM_21:
       case CudaArch::SM_30:
-      case CudaArch::SM_32:
+      case CudaArch::SM_32_:
       case CudaArch::SM_35:
       case CudaArch::SM_37:
       case CudaArch::SM_50:

diff --git a/clang/lib/CodeGen/CGRecordLayoutBuilder.cpp b/clang/lib/CodeGen/CGRecordLayoutBuilder.cpp
@@ -41,10 +41,11 @@ namespace {
 ///   contains enough information to determine where the runs break.  Microsoft
 ///   and Itanium follow different rules and use different codepaths.
 /// * It is desired that, when possible, bitfields use the appropriate iN type
-///   when lowered to llvm types.  For example unsigned x : 24 gets lowered to
+///   when lowered to llvm types. For example unsigned x : 24 gets lowered to
 ///   i24.  This isn't always possible because i24 has storage size of 32 bit
-///   and if it is possible to use that extra byte of padding we must use
-///   [i8 x 3] instead of i24.  The function clipTailPadding does this.
+///   and if it is possible to use that extra byte of padding we must use [i8 x
+///   3] instead of i24. This is computed when accumulating bitfields in
+///   accumulateBitfields.
 ///   C++ examples that require clipping:
 ///   struct { int a : 24; char b; }; // a must be clipped, b goes at offset 3
 ///   struct A { int a : 24; ~A(); }; // a must be clipped because:
@@ -62,11 +63,7 @@ namespace {
 ///   that the tail padding is not used in the complete class.) However,
 ///   because LLVM reads from the complete type it can generate incorrect code
 ///   if we do not clip the tail padding off of the bitfield in the complete
-///   layout.  This introduces a somewhat awkward extra unnecessary clip stage.
-///   The location of the clip is stored internally as a sentinel of type
-///   SCISSOR.  If LLVM were updated to read base types (which it probably
-///   should because locations of things such as VBases are bogus in the llvm
-///   type anyway) then we could eliminate the SCISSOR.
+///   layout.
 /// * Itanium allows nearly empty primary virtual bases.  These bases don't get
 ///   get their own storage because they're laid out as part of another base
 ///   or at the beginning of the structure.  Determining if a VBase actually
@@ -200,9 +197,7 @@ struct CGRecordLowering {
                      const CXXRecordDecl *Query) const;
   void calculateZeroInit();
   CharUnits calculateTailClippingOffset(bool isNonVirtualBaseType) const;
-  /// Lowers bitfield storage types to I8 arrays for bitfields with tail
-  /// padding that is or can potentially be used.
-  void clipTailPadding();
+  void checkBitfieldClipping() const;
   /// Determines if we need a packed llvm struct.
   void determinePacked(bool NVBaseType);
   /// Inserts padding everywhere it's needed.
@@ -305,7 +300,7 @@ void CGRecordLowering::lower(bool NVBaseType) {
   }
   llvm::stable_sort(Members);
   Members.push_back(StorageInfo(Size, getIntNType(8)));
-  clipTailPadding();
+  checkBitfieldClipping();
   determinePacked(NVBaseType);
   insertPadding();
   Members.pop_back();
@@ -531,6 +526,7 @@ CGRecordLowering::accumulateBitFields(bool isNonVirtualBaseType,
   // available padding characters.
   RecordDecl::field_iterator BestEnd = Begin;
   CharUnits BestEndOffset;
+  bool BestClipped; // Whether the representation must be in a byte array.
 
   for (;;) {
     // AtAlignedBoundary is true iff Field is the (potential) start of a new
@@ -593,10 +589,9 @@ CGRecordLowering::accumulateBitFields(bool isNonVirtualBaseType,
         // this is the best seen so far.
         BestEnd = Field;
         BestEndOffset = BeginOffset + AccessSize;
-        if (Types.getCodeGenOpts().FineGrainedBitfieldAccesses)
-          // Fine-grained access, so no merging of spans.
-          InstallBest = true;
-        else if (!BitSizeSinceBegin)
+        // Assume clipped until proven not below.
+        BestClipped = true;
+        if (!BitSizeSinceBegin)
           // A zero-sized initial span -- this will install nothing and reset
           // for another.
           InstallBest = true;
@@ -624,6 +619,12 @@ CGRecordLowering::accumulateBitFields(bool isNonVirtualBaseType,
             // The access unit is not at a naturally aligned offset within the
             // structure.
             InstallBest = true;
+
+          if (InstallBest && BestEnd == Field)
+            // We're installing the first span, whose clipping was presumed
+            // above. Compute it correctly.
+            if (getSize(Type) == AccessSize)
+              BestClipped = false;
         }
 
         if (!InstallBest) {
@@ -656,11 +657,15 @@ CGRecordLowering::accumulateBitFields(bool isNonVirtualBaseType,
             // access unit.
             BestEndOffset = BeginOffset + TypeSize;
             BestEnd = Field;
+            BestClipped = false;
           }
 
           if (Barrier)
             // The next field is a barrier that we cannot merge across.
             InstallBest = true;
+          else if (Types.getCodeGenOpts().FineGrainedBitfieldAccesses)
+            // Fine-grained access, so no merging of spans.
+            InstallBest = true;
           else
             // Otherwise, we're not installing. Update the bit size
             // of the current span to go all the way to LimitOffset, which is
@@ -679,7 +684,17 @@ CGRecordLowering::accumulateBitFields(bool isNonVirtualBaseType,
         // Add the storage member for the access unit to the record. The
         // bitfields get the offset of their storage but come afterward and
         // remain there after a stable sort.
-        llvm::Type *Type = getIntNType(Context.toBits(AccessSize));
+        llvm::Type *Type;
+        if (BestClipped) {
+          assert(getSize(getIntNType(Context.toBits(AccessSize))) >
+                     AccessSize &&
+                 "Clipped access need not be clipped");
+          Type = getByteArrayType(AccessSize);
+        } else {
+          Type = getIntNType(Context.toBits(AccessSize));
+          assert(getSize(Type) == AccessSize &&
+                 "Unclipped access must be clipped");
+        }
         Members.push_back(StorageInfo(BeginOffset, Type));
         for (; Begin != BestEnd; ++Begin)
           if (!Begin->isZeroLengthBitField(Context))
@@ -934,32 +949,21 @@ void CGRecordLowering::calculateZeroInit() {
   }
 }
 
-void CGRecordLowering::clipTailPadding() {
-  std::vector<MemberInfo>::iterator Prior = Members.begin();
-  CharUnits Tail = getSize(Prior->Data);
-  for (std::vector<MemberInfo>::iterator Member = Prior + 1,
-                                         MemberEnd = Members.end();
-       Member != MemberEnd; ++Member) {
+// Verify accumulateBitfields computed the correct storage representations.
+void CGRecordLowering::checkBitfieldClipping() const {
+#ifndef NDEBUG
+  auto Tail = CharUnits::Zero();
+  for (const auto &M : Members) {
     // Only members with data and the scissor can cut into tail padding.
-    if (!Member->Data && Member->Kind != MemberInfo::Scissor)
+    if (!M.Data && M.Kind != MemberInfo::Scissor)
       continue;
-    if (Member->Offset < Tail) {
-      assert(Prior->Kind == MemberInfo::Field &&
-             "Only storage fields have tail padding!");
-      if (!Prior->FD || Prior->FD->isBitField())
-        Prior->Data = getByteArrayType(bitsToCharUnits(llvm::alignTo(
-            cast<llvm::IntegerType>(Prior->Data)->getIntegerBitWidth(), 8)));
-      else {
-        assert(Prior->FD->hasAttr<NoUniqueAddressAttr>() &&
-               "should not have reused this field's tail padding");
-        Prior->Data = getByteArrayType(
-            Context.getTypeInfoDataSizeInChars(Prior->FD->getType()).Width);
-      }
-    }
-    if (Member->Data)
-      Prior = Member;
-    Tail = Prior->Offset + getSize(Prior->Data);
+
+    assert(M.Offset >= Tail && "Bitfield access unit is not clipped");
+    Tail = M.Offset;
+    if (M.Data)
+      Tail += getSize(M.Data);
   }
+#endif
 }
 
 void CGRecordLowering::determinePacked(bool NVBaseType) {

diff --git a/clang/lib/CodeGen/CodeGenAction.cpp b/clang/lib/CodeGen/CodeGenAction.cpp
@@ -25,8 +25,11 @@
 #include "clang/CodeGen/ModuleBuilder.h"
 #include "clang/Driver/DriverDiagnostic.h"
 #include "clang/Frontend/CompilerInstance.h"
+#include "clang/Frontend/FrontendActions.h"
 #include "clang/Frontend/FrontendDiagnostic.h"
+#include "clang/Frontend/MultiplexConsumer.h"
 #include "clang/Lex/Preprocessor.h"
+#include "clang/Serialization/ASTWriter.h"
 #include "llvm/ADT/Hashing.h"
 #include "llvm/Bitcode/BitcodeReader.h"
 #include "llvm/CodeGen/MachineOptimizationRemarkEmitter.h"
@@ -1003,6 +1006,12 @@ CodeGenerator *CodeGenAction::getCodeGenerator() const {
   return BEConsumer->getCodeGenerator();
 }
 
+bool CodeGenAction::BeginSourceFileAction(CompilerInstance &CI) {
+  if (CI.getFrontendOpts().GenReducedBMI)
+    CI.getLangOpts().setCompilingModule(LangOptions::CMK_ModuleInterface);
+  return true;
+}
+
 static std::unique_ptr<raw_pwrite_stream>
 GetOutputStream(CompilerInstance &CI, StringRef InFile, BackendAction Action) {
   switch (Action) {
@@ -1061,6 +1070,16 @@ CodeGenAction::CreateASTConsumer(CompilerInstance &CI, StringRef InFile) {
     CI.getPreprocessor().addPPCallbacks(std::move(Callbacks));
   }
 
+  if (CI.getFrontendOpts().GenReducedBMI &&
+      !CI.getFrontendOpts().ModuleOutputPath.empty()) {
+    std::vector<std::unique_ptr<ASTConsumer>> Consumers(2);
+    Consumers[0] = std::make_unique<ReducedBMIGenerator>(
+        CI.getPreprocessor(), CI.getModuleCache(),
+        CI.getFrontendOpts().ModuleOutputPath);
+    Consumers[1] = std::move(Result);
+    return std::make_unique<MultiplexConsumer>(std::move(Consumers));
+  }
+
   return std::move(Result);
 }
 

diff --git a/clang/lib/CodeGen/CodeGenFunction.cpp b/clang/lib/CodeGen/CodeGenFunction.cpp
@@ -91,8 +91,6 @@ CodeGenFunction::CodeGenFunction(CodeGenModule &cgm, bool suppressNewContext)
 
 CodeGenFunction::~CodeGenFunction() {
   assert(LifetimeExtendedCleanupStack.empty() && "failed to emit a cleanup");
-  assert(DeferredDeactivationCleanupStack.empty() &&
-         "missed to deactivate a cleanup");
 
   if (getLangOpts().OpenMP && CurFn)
     CGM.getOpenMPRuntime().functionFinished(*this);
@@ -348,10 +346,6 @@ static void EmitIfUsed(CodeGenFunction &CGF, llvm::BasicBlock *BB) {
 void CodeGenFunction::FinishFunction(SourceLocation EndLoc) {
   assert(BreakContinueStack.empty() &&
          "mismatched push/pop in break/continue stack!");
-  assert(LifetimeExtendedCleanupStack.empty() &&
-         "mismatched push/pop of cleanups in EHStack!");
-  assert(DeferredDeactivationCleanupStack.empty() &&
-         "mismatched activate/deactivate of cleanups!");
 
   bool OnlySimpleReturnStmts = NumSimpleReturnExprs > 0
     && NumSimpleReturnExprs == NumReturnExprs

diff --git a/clang/lib/CodeGen/CodeGenFunction.h b/clang/lib/CodeGen/CodeGenFunction.h
@@ -39,7 +39,6 @@
 #include "llvm/ADT/MapVector.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/Frontend/OpenMP/OMPIRBuilder.h"
-#include "llvm/IR/Instructions.h"
 #include "llvm/IR/ValueHandle.h"
 #include "llvm/Support/Debug.h"
 #include "llvm/Transforms/Utils/SanitizerStats.h"
@@ -671,51 +670,6 @@ class CodeGenFunction : public CodeGenTypeCache {
 
   EHScopeStack EHStack;
   llvm::SmallVector<char, 256> LifetimeExtendedCleanupStack;
-
-  // A stack of cleanups which were added to EHStack but have to be deactivated
-  // later before being popped or emitted. These are usually deactivated on
-  // exiting a `CleanupDeactivationScope` scope. For instance, after a
-  // full-expr.
-  //
-  // These are specially useful for correctly emitting cleanups while
-  // encountering branches out of expression (through stmt-expr or coroutine
-  // suspensions).
-  struct DeferredDeactivateCleanup {
-    EHScopeStack::stable_iterator Cleanup;
-    llvm::Instruction *DominatingIP;
-  };
-  llvm::SmallVector<DeferredDeactivateCleanup> DeferredDeactivationCleanupStack;
-
-  // Enters a new scope for capturing cleanups which are deferred to be
-  // deactivated, all of which will be deactivated once the scope is exited.
-  struct CleanupDeactivationScope {
-    CodeGenFunction &CGF;
-    size_t OldDeactivateCleanupStackSize;
-    bool Deactivated;
-    CleanupDeactivationScope(CodeGenFunction &CGF)
-        : CGF(CGF), OldDeactivateCleanupStackSize(
-                        CGF.DeferredDeactivationCleanupStack.size()),
-          Deactivated(false) {}
-
-    void ForceDeactivate() {
-      assert(!Deactivated && "Deactivating already deactivated scope");
-      auto &Stack = CGF.DeferredDeactivationCleanupStack;
-      for (size_t I = Stack.size(); I > OldDeactivateCleanupStackSize; I--) {
-        CGF.DeactivateCleanupBlock(Stack[I - 1].Cleanup,
-                                   Stack[I - 1].DominatingIP);
-        Stack[I - 1].DominatingIP->eraseFromParent();
-      }
-      Stack.resize(OldDeactivateCleanupStackSize);
-      Deactivated = true;
-    }
-
-    ~CleanupDeactivationScope() {
-      if (Deactivated)
-        return;
-      ForceDeactivate();
-    }
-  };
-
   llvm::SmallVector<const JumpDest *, 2> SEHTryEpilogueStack;
 
   llvm::Instruction *CurrentFuncletPad = nullptr;
@@ -921,19 +875,6 @@ class CodeGenFunction : public CodeGenTypeCache {
       new (Buffer + sizeof(Header) + sizeof(T)) RawAddress(ActiveFlag);
   }
 
-  // Push a cleanup onto EHStack and deactivate it later. It is usually
-  // deactivated when exiting a `CleanupDeactivationScope` (for example: after a
-  // full expression).
-  template <class T, class... As>
-  void pushCleanupAndDeferDeactivation(CleanupKind Kind, As... A) {
-    // Placeholder dominating IP for this cleanup.
-    llvm::Instruction *DominatingIP =
-        Builder.CreateFlagLoad(llvm::Constant::getNullValue(Int8PtrTy));
-    EHStack.pushCleanup<T>(Kind, A...);
-    DeferredDeactivationCleanupStack.push_back(
-        {EHStack.stable_begin(), DominatingIP});
-  }
-
   /// Set up the last cleanup that was pushed as a conditional
   /// full-expression cleanup.
   void initFullExprCleanup() {
@@ -985,7 +926,6 @@ class CodeGenFunction : public CodeGenTypeCache {
   class RunCleanupsScope {
     EHScopeStack::stable_iterator CleanupStackDepth, OldCleanupScopeDepth;
     size_t LifetimeExtendedCleanupStackSize;
-    CleanupDeactivationScope DeactivateCleanups;
     bool OldDidCallStackSave;
   protected:
     bool PerformCleanup;
@@ -1000,7 +940,8 @@ class CodeGenFunction : public CodeGenTypeCache {
   public:
     /// Enter a new cleanup scope.
     explicit RunCleanupsScope(CodeGenFunction &CGF)
-        : DeactivateCleanups(CGF), PerformCleanup(true), CGF(CGF) {
+      : PerformCleanup(true), CGF(CGF)
+    {
       CleanupStackDepth = CGF.EHStack.stable_begin();
       LifetimeExtendedCleanupStackSize =
           CGF.LifetimeExtendedCleanupStack.size();
@@ -1030,7 +971,6 @@ class CodeGenFunction : public CodeGenTypeCache {
     void ForceCleanup(std::initializer_list<llvm::Value**> ValuesToReload = {}) {
       assert(PerformCleanup && "Already forced cleanup");
       CGF.DidCallStackSave = OldDidCallStackSave;
-      DeactivateCleanups.ForceDeactivate();
       CGF.PopCleanupBlocks(CleanupStackDepth, LifetimeExtendedCleanupStackSize,
                            ValuesToReload);
       PerformCleanup = false;
@@ -2220,11 +2160,6 @@ class CodeGenFunction : public CodeGenTypeCache {
                      Address addr, QualType type);
   void pushDestroy(CleanupKind kind, Address addr, QualType type,
                    Destroyer *destroyer, bool useEHCleanupForArray);
-  void pushDestroyAndDeferDeactivation(QualType::DestructionKind dtorKind,
-                                       Address addr, QualType type);
-  void pushDestroyAndDeferDeactivation(CleanupKind cleanupKind, Address addr,
-                                       QualType type, Destroyer *destroyer,
-                                       bool useEHCleanupForArray);
   void pushLifetimeExtendedDestroy(CleanupKind kind, Address addr,
                                    QualType type, Destroyer *destroyer,
                                    bool useEHCleanupForArray);
@@ -2763,33 +2698,6 @@ class CodeGenFunction : public CodeGenTypeCache {
                             TBAAAccessInfo *TBAAInfo = nullptr);
   LValue EmitLoadOfPointerLValue(Address Ptr, const PointerType *PtrTy);
 
-private:
-  struct AllocaTracker {
-    void Add(llvm::AllocaInst *I) { Allocas.push_back(I); }
-    llvm::SmallVector<llvm::AllocaInst *> Take() { return std::move(Allocas); }
-
-  private:
-    llvm::SmallVector<llvm::AllocaInst *> Allocas;
-  };
-  AllocaTracker *Allocas = nullptr;
-
-public:
-  // Captures all the allocas created during the scope of its RAII object.
-  struct AllocaTrackerRAII {
-    AllocaTrackerRAII(CodeGenFunction &CGF)
-        : CGF(CGF), OldTracker(CGF.Allocas) {
-      CGF.Allocas = &Tracker;
-    }
-    ~AllocaTrackerRAII() { CGF.Allocas = OldTracker; }
-
-    llvm::SmallVector<llvm::AllocaInst *> Take() { return Tracker.Take(); }
-
-  private:
-    CodeGenFunction &CGF;
-    AllocaTracker *OldTracker;
-    AllocaTracker Tracker;
-  };
-
   /// CreateTempAlloca - This creates an alloca and inserts it into the entry
   /// block if \p ArraySize is nullptr, otherwise inserts it at the current
   /// insertion point of the builder. The caller is responsible for setting an

diff --git a/clang/lib/CodeGen/CodeGenModule.cpp b/clang/lib/CodeGen/CodeGenModule.cpp
@@ -3952,8 +3952,20 @@ bool CodeGenModule::shouldEmitFunction(GlobalDecl GD) {
   // behavior may break ABI compatibility of the current unit.
   if (const Module *M = F->getOwningModule();
       M && M->getTopLevelModule()->isNamedModule() &&
-      getContext().getCurrentNamedModule() != M->getTopLevelModule())
-    return false;
+      getContext().getCurrentNamedModule() != M->getTopLevelModule()) {
+    // There are practices to mark template member function as always-inline
+    // and mark the template as extern explicit instantiation but not give
+    // the definition for member function. So we have to emit the function
+    // from explicitly instantiation with always-inline.
+    //
+    // See https://github.com/llvm/llvm-project/issues/86893 for details.
+    //
+    // TODO: Maybe it is better to give it a warning if we call a non-inline
+    // function from other module units which is marked as always-inline.
+    if (!F->isTemplateInstantiation() || !F->hasAttr<AlwaysInlineAttr>()) {
+      return false;
+    }
+  }
 
   if (F->hasAttr<NoInlineAttr>())
     return false;

diff --git a/clang/lib/Driver/Driver.cpp b/clang/lib/Driver/Driver.cpp
@@ -4756,6 +4756,14 @@ Action *Driver::ConstructPhaseAction(
     if (Args.hasArg(options::OPT_extract_api))
       return C.MakeAction<ExtractAPIJobAction>(Input, types::TY_API_INFO);
 
+    // With 'fexperimental-modules-reduced-bmi', we don't want to run the
+    // precompile phase unless the user specified '--precompile'. In the case
+    // the '--precompile' flag is enabled, we will try to emit the reduced BMI
+    // as a by product in GenerateModuleInterfaceAction.
+    if (Args.hasArg(options::OPT_modules_reduced_bmi) &&
+        !Args.getLastArg(options::OPT__precompile))
+      return Input;
+
     types::ID OutputTy = getPrecompiledType(Input->getType());
     assert(OutputTy != types::TY_INVALID &&
            "Cannot precompile this input type!");
@@ -5916,8 +5924,10 @@ const char *Driver::GetNamedOutputPath(Compilation &C, const JobAction &JA,
   // If we're emitting a module output with the specified option
   // `-fmodule-output`.
   if (!AtTopLevel && isa<PrecompileJobAction>(JA) &&
-      JA.getType() == types::TY_ModuleFile && SpecifiedModuleOutput)
+      JA.getType() == types::TY_ModuleFile && SpecifiedModuleOutput) {
+    assert(!C.getArgs().hasArg(options::OPT_modules_reduced_bmi));
     return GetModuleOutputPath(C, JA, BaseInput);
+  }
 
   // Output to a temporary file?
   if ((!AtTopLevel && !isSaveTempsEnabled() &&

diff --git a/clang/lib/Driver/ToolChain.cpp b/clang/lib/Driver/ToolChain.cpp
@@ -796,7 +796,13 @@ ToolChain::getTargetSubDirPath(StringRef BaseDir) const {
 std::optional<std::string> ToolChain::getRuntimePath() const {
   SmallString<128> P(D.ResourceDir);
   llvm::sys::path::append(P, "lib");
-  return getTargetSubDirPath(P);
+  if (auto Ret = getTargetSubDirPath(P))
+    return Ret;
+  // Darwin does not use per-target runtime directory.
+  if (Triple.isOSDarwin())
+    return {};
+  llvm::sys::path::append(P, Triple.str());
+  return std::string(P);
 }
 
 std::optional<std::string> ToolChain::getStdlibPath() const {

diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -346,11 +346,14 @@ static bool addExceptionArgs(const ArgList &Args, types::ID InputType,
   bool EH = Args.hasFlag(options::OPT_fexceptions, options::OPT_fno_exceptions,
                          false);
 
-  bool EHa = Args.hasFlag(options::OPT_fasync_exceptions,
-                          options::OPT_fno_async_exceptions, false);
-  if (EHa) {
-    CmdArgs.push_back("-fasync-exceptions");
-    EH = true;
+  // Async exceptions are Windows MSVC only.
+  if (Triple.isWindowsMSVCEnvironment()) {
+    bool EHa = Args.hasFlag(options::OPT_fasync_exceptions,
+                            options::OPT_fno_async_exceptions, false);
+    if (EHa) {
+      CmdArgs.push_back("-fasync-exceptions");
+      EH = true;
+    }
   }
 
   // Obj-C exceptions are enabled by default, regardless of -fexceptions. This
@@ -4045,6 +4048,24 @@ static bool RenderModulesOptions(Compilation &C, const Driver &D,
   // module fragment.
   CmdArgs.push_back("-fskip-odr-check-in-gmf");
 
+  if (Args.hasArg(options::OPT_modules_reduced_bmi) &&
+      (Input.getType() == driver::types::TY_CXXModule ||
+       Input.getType() == driver::types::TY_PP_CXXModule)) {
+    CmdArgs.push_back("-fexperimental-modules-reduced-bmi");
+
+    if (Args.hasArg(options::OPT_fmodule_output_EQ))
+      Args.AddLastArg(CmdArgs, options::OPT_fmodule_output_EQ);
+    else
+      CmdArgs.push_back(Args.MakeArgString(
+          "-fmodule-output=" +
+          getCXX20NamedModuleOutputPath(Args, Input.getBaseInput())));
+  }
+
+  // Noop if we see '-fexperimental-modules-reduced-bmi' with other translation
+  // units than module units. This is more user friendly to allow end uers to
+  // enable this feature without asking for help from build systems.
+  Args.ClaimAllArgs(options::OPT_modules_reduced_bmi);
+
   // We need to include the case the input file is a module file here.
   // Since the default compilation model for C++ module interface unit will
   // create temporary module file and compile the temporary module file
@@ -8084,7 +8105,8 @@ struct EHFlags {
 ///      The 'a' modifier is unimplemented and fundamentally hard in LLVM IR.
 /// - c: Assume that extern "C" functions are implicitly nounwind.
 /// The default is /EHs-c-, meaning cleanups are disabled.
-static EHFlags parseClangCLEHFlags(const Driver &D, const ArgList &Args) {
+static EHFlags parseClangCLEHFlags(const Driver &D, const ArgList &Args,
+                                   bool isWindowsMSVC) {
   EHFlags EH;
 
   std::vector<std::string> EHArgs =
@@ -8094,8 +8116,15 @@ static EHFlags parseClangCLEHFlags(const Driver &D, const ArgList &Args) {
       switch (EHVal[I]) {
       case 'a':
         EH.Asynch = maybeConsumeDash(EHVal, I);
-        if (EH.Asynch)
+        if (EH.Asynch) {
+          // Async exceptions are Windows MSVC only.
+          if (!isWindowsMSVC) {
+            EH.Asynch = false;
+            D.Diag(clang::diag::warn_drv_unused_argument) << "/EHa" << EHVal;
+            continue;
+          }
           EH.Synch = false;
+        }
         continue;
       case 'c':
         EH.NoUnwindC = maybeConsumeDash(EHVal, I);
@@ -8159,7 +8188,8 @@ void Clang::AddClangCLArgs(const ArgList &Args, types::ID InputType,
 
   const Driver &D = getToolChain().getDriver();
 
-  EHFlags EH = parseClangCLEHFlags(D, Args);
+  bool IsWindowsMSVC = getToolChain().getTriple().isWindowsMSVCEnvironment();
+  EHFlags EH = parseClangCLEHFlags(D, Args, IsWindowsMSVC);
   if (!isNVPTX && (EH.Synch || EH.Asynch)) {
     if (types::isCXX(InputType))
       CmdArgs.push_back("-fcxx-exceptions");

diff --git a/clang/lib/Driver/ToolChains/CommonArgs.cpp b/clang/lib/Driver/ToolChains/CommonArgs.cpp
@@ -114,6 +114,7 @@ static bool useFramePointerForTargetByDefault(const llvm::opt::ArgList &Args,
   case llvm::Triple::csky:
   case llvm::Triple::loongarch32:
   case llvm::Triple::loongarch64:
+  case llvm::Triple::m68k:
     return !clang::driver::tools::areOptimizationsEnabled(Args);
   default:
     break;

diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -786,6 +786,10 @@ void Flang::ConstructJob(Compilation &C, const JobAction &JA,
     }
   }
 
+  // Pass the path to compiler resource files.
+  CmdArgs.push_back("-resource-dir");
+  CmdArgs.push_back(D.ResourceDir.c_str());
+
   // Offloading related options
   addOffloadOptions(C, Inputs, JA, Args, CmdArgs);
 

diff --git a/clang/lib/Driver/ToolChains/Linux.cpp b/clang/lib/Driver/ToolChains/Linux.cpp
@@ -244,8 +244,9 @@ Linux::Linux(const Driver &D, const llvm::Triple &Triple, const ArgList &Args)
       // Android ARM uses max-page-size=4096 to reduce VMA usage.
       ExtraOpts.push_back("-z");
       ExtraOpts.push_back("max-page-size=4096");
-    } else if (Triple.isAArch64()) {
+    } else if (Triple.isAArch64() || Triple.getArch() == llvm::Triple::x86_64) {
       // Android AArch64 uses max-page-size=16384 to support 4k/16k page sizes.
+      // Android emulates a 16k page size for app testing on x86_64 machines.
       ExtraOpts.push_back("-z");
       ExtraOpts.push_back("max-page-size=16384");
     }

diff --git a/clang/lib/Format/ContinuationIndenter.cpp b/clang/lib/Format/ContinuationIndenter.cpp
@@ -684,7 +684,13 @@ void ContinuationIndenter::addTokenOnCurrentLine(LineState &State, bool DryRun,
         // arguments to function calls. We do this by ensuring that either all
         // arguments (including any lambdas) go on the same line as the function
         // call, or we break before the first argument.
-        auto PrevNonComment = Current.getPreviousNonComment();
+        const auto *Prev = Current.Previous;
+        if (!Prev)
+          return false;
+        // For example, `/*Newline=*/false`.
+        if (Prev->is(TT_BlockComment) && Current.SpacesRequiredBefore == 0)
+          return false;
+        const auto *PrevNonComment = Current.getPreviousNonComment();
         if (!PrevNonComment || PrevNonComment->isNot(tok::l_paren))
           return false;
         if (Current.isOneOf(tok::comment, tok::l_paren, TT_LambdaLSquare))

diff --git a/clang/lib/Format/Format.cpp b/clang/lib/Format/Format.cpp
@@ -3891,7 +3891,11 @@ static FormatStyle::LanguageKind getLanguageByFileName(StringRef FileName) {
       FileName.ends_with_insensitive(".protodevel")) {
     return FormatStyle::LK_Proto;
   }
-  if (FileName.ends_with_insensitive(".textpb") ||
+  // txtpb is the canonical extension, and textproto is the legacy canonical
+  // extension
+  // https://protobuf.dev/reference/protobuf/textformat-spec/#text-format-files
+  if (FileName.ends_with_insensitive(".txtpb") ||
+      FileName.ends_with_insensitive(".textpb") ||
       FileName.ends_with_insensitive(".pb.txt") ||
       FileName.ends_with_insensitive(".textproto") ||
       FileName.ends_with_insensitive(".asciipb")) {

diff --git a/clang/lib/Frontend/FrontendActions.cpp b/clang/lib/Frontend/FrontendActions.cpp
@@ -281,6 +281,13 @@ GenerateModuleInterfaceAction::CreateASTConsumer(CompilerInstance &CI,
   if (Consumers.empty())
     return nullptr;
 
+  if (CI.getFrontendOpts().GenReducedBMI &&
+      !CI.getFrontendOpts().ModuleOutputPath.empty()) {
+    Consumers.push_back(std::make_unique<ReducedBMIGenerator>(
+        CI.getPreprocessor(), CI.getModuleCache(),
+        CI.getFrontendOpts().ModuleOutputPath));
+  }
+
   return std::make_unique<MultiplexConsumer>(std::move(Consumers));
 }
 

diff --git a/clang/lib/Frontend/InitPreprocessor.cpp b/clang/lib/Frontend/InitPreprocessor.cpp
@@ -747,6 +747,9 @@ static void InitializeCPlusPlusFeatureTestMacros(const LangOptions &LangOpts,
   Builder.defineMacro("__cpp_named_character_escapes", "202207L");
   Builder.defineMacro("__cpp_placeholder_variables", "202306L");
 
+  // C++26 features supported in earlier language modes.
+  Builder.defineMacro("__cpp_deleted_function", "202403L");
+
   if (LangOpts.Char8)
     Builder.defineMacro("__cpp_char8_t", "202207L");
   Builder.defineMacro("__cpp_impl_destroying_delete", "201806L");