diff --git a/clang/include/clang/Sema/Sema.h b/clang/include/clang/Sema/Sema.h
@@ -184,6 +184,7 @@ class PseudoObjectExpr;
 class QualType;
 class SemaHLSL;
 class SemaOpenACC;
+class SemaSYCL;
 class StandardConversionSequence;
 class Stmt;
 class StringLiteral;
@@ -467,7 +468,6 @@ class Sema final : public SemaBase {
   // 37. Name Lookup for RISC-V Vector Intrinsic (SemaRISCVVectorLookup.cpp)
   // 38. CUDA (SemaCUDA.cpp)
   // 39. OpenMP Directives and Clauses (SemaOpenMP.cpp)
-  // 40. SYCL Constructs (SemaSYCL.cpp)
 
   /// \name Semantic Analysis
   /// Implementations are in Sema.cpp
@@ -974,6 +974,11 @@ class Sema final : public SemaBase {
     return *OpenACCPtr;
   }
 
+  SemaSYCL &SYCL() {
+    assert(SYCLPtr);
+    return *SYCLPtr;
+  }
+
 protected:
   friend class Parser;
   friend class InitializationSequence;
@@ -1006,6 +1011,7 @@ class Sema final : public SemaBase {
 
   std::unique_ptr<SemaHLSL> HLSLPtr;
   std::unique_ptr<SemaOpenACC> OpenACCPtr;
+  std::unique_ptr<SemaSYCL> SYCLPtr;
 
   ///@}
 
@@ -5439,8 +5445,7 @@ class Sema final : public SemaBase {
 
   ExprResult BuildDeclarationNameExpr(const CXXScopeSpec &SS, LookupResult &R,
                                       bool NeedsADL,
-                                      bool AcceptInvalidDecl = false,
-                                      bool NeedUnresolved = false);
+                                      bool AcceptInvalidDecl = false);
   ExprResult BuildDeclarationNameExpr(
       const CXXScopeSpec &SS, const DeclarationNameInfo &NameInfo, NamedDecl *D,
       NamedDecl *FoundD = nullptr,
@@ -5456,15 +5461,6 @@ class Sema final : public SemaBase {
   ExprResult ActOnPredefinedExpr(SourceLocation Loc, tok::TokenKind Kind);
   ExprResult ActOnIntegerConstant(SourceLocation Loc, uint64_t Val);
 
-  ExprResult BuildSYCLUniqueStableNameExpr(SourceLocation OpLoc,
-                                           SourceLocation LParen,
-                                           SourceLocation RParen,
-                                           TypeSourceInfo *TSI);
-  ExprResult ActOnSYCLUniqueStableNameExpr(SourceLocation OpLoc,
-                                           SourceLocation LParen,
-                                           SourceLocation RParen,
-                                           ParsedType ParsedTy);
-
   bool CheckLoopHintExpr(Expr *E, SourceLocation Loc);
 
   ExprResult ActOnNumericConstant(const Token &Tok, Scope *UDLScope = nullptr);
@@ -6592,10 +6588,7 @@ class Sema final : public SemaBase {
                             SourceLocation RParenLoc);
 
   //// ActOnCXXThis -  Parse 'this' pointer.
-  ExprResult ActOnCXXThis(SourceLocation Loc);
-
-  /// Check whether the type of 'this' is valid in the current context.
-  bool CheckCXXThisType(SourceLocation Loc, QualType Type);
+  ExprResult ActOnCXXThis(SourceLocation loc);
 
   /// Build a CXXThisExpr and mark it referenced in the current context.
   Expr *BuildCXXThisExpr(SourceLocation Loc, QualType Type, bool IsImplicit);
@@ -14520,44 +14513,6 @@ class Sema final : public SemaBase {
                         OpenMPDirectiveKind CancelRegion);
 
   ///@}
-
-  //
-  //
-  // -------------------------------------------------------------------------
-  //
-  //
-
-  /// \name SYCL Constructs
-  /// Implementations are in SemaSYCL.cpp
-  ///@{
-
-public:
-  /// Creates a SemaDiagnosticBuilder that emits the diagnostic if the current
-  /// context is "used as device code".
-  ///
-  /// - If CurLexicalContext is a kernel function or it is known that the
-  ///   function will be emitted for the device, emits the diagnostics
-  ///   immediately.
-  /// - If CurLexicalContext is a function and we are compiling
-  ///   for the device, but we don't know that this function will be codegen'ed
-  ///   for devive yet, creates a diagnostic which is emitted if and when we
-  ///   realize that the function will be codegen'ed.
-  ///
-  /// Example usage:
-  ///
-  /// Diagnose __float128 type usage only from SYCL device code if the current
-  /// target doesn't support it
-  /// if (!S.Context.getTargetInfo().hasFloat128Type() &&
-  ///     S.getLangOpts().SYCLIsDevice)
-  ///   SYCLDiagIfDeviceCode(Loc, diag::err_type_unsupported) << "__float128";
-  SemaDiagnosticBuilder SYCLDiagIfDeviceCode(SourceLocation Loc,
-                                             unsigned DiagID);
-
-  void deepTypeCheckForSYCLDevice(SourceLocation UsedAt,
-                                  llvm::DenseSet<QualType> Visited,
-                                  ValueDecl *DeclToCheck);
-
-  ///@}
 };
 
 DeductionFailureInfo

diff --git a/clang/include/clang/Sema/SemaOpenACC.h b/clang/include/clang/Sema/SemaOpenACC.h
@@ -19,6 +19,7 @@
 #include "clang/Basic/SourceLocation.h"
 #include "clang/Sema/Ownership.h"
 #include "clang/Sema/SemaBase.h"
+#include <variant>
 
 namespace clang {
 class OpenACCClause;
@@ -35,7 +36,11 @@ class SemaOpenACC : public SemaBase {
     SourceRange ClauseRange;
     SourceLocation LParenLoc;
 
-    // TODO OpenACC: Add variant here to store details of individual clauses.
+    struct DefaultDetails {
+      OpenACCDefaultClauseKind DefaultClauseKind;
+    };
+
+    std::variant<DefaultDetails> Details;
 
   public:
     OpenACCParsedClause(OpenACCDirectiveKind DirKind,
@@ -52,8 +57,20 @@ class SemaOpenACC : public SemaBase {
 
     SourceLocation getEndLoc() const { return ClauseRange.getEnd(); }
 
+    OpenACCDefaultClauseKind getDefaultClauseKind() const {
+      assert(ClauseKind == OpenACCClauseKind::Default &&
+             "Parsed clause is not a default clause");
+      return std::get<DefaultDetails>(Details).DefaultClauseKind;
+    }
+
     void setLParenLoc(SourceLocation EndLoc) { LParenLoc = EndLoc; }
     void setEndLoc(SourceLocation EndLoc) { ClauseRange.setEnd(EndLoc); }
+
+    void setDefaultDetails(OpenACCDefaultClauseKind DefKind) {
+      assert(ClauseKind == OpenACCClauseKind::Default &&
+             "Parsed clause is not a default clause");
+      Details = DefaultDetails{DefKind};
+    }
   };
 
   SemaOpenACC(Sema &S);

diff --git a/clang/include/clang/Sema/SemaSYCL.h b/clang/include/clang/Sema/SemaSYCL.h
@@ -0,0 +1,65 @@
+//===----- SemaSYCL.h ------- Semantic Analysis for SYCL constructs -------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+/// \file
+/// This file declares semantic analysis for SYCL constructs.
+///
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CLANG_SEMA_SEMASYCL_H
+#define LLVM_CLANG_SEMA_SEMASYCL_H
+
+#include "clang/AST/Decl.h"
+#include "clang/AST/Type.h"
+#include "clang/Basic/SourceLocation.h"
+#include "clang/Sema/Ownership.h"
+#include "clang/Sema/SemaBase.h"
+#include "llvm/ADT/DenseSet.h"
+
+namespace clang {
+
+class SemaSYCL : public SemaBase {
+public:
+  SemaSYCL(Sema &S);
+
+  /// Creates a SemaDiagnosticBuilder that emits the diagnostic if the current
+  /// context is "used as device code".
+  ///
+  /// - If CurLexicalContext is a kernel function or it is known that the
+  ///   function will be emitted for the device, emits the diagnostics
+  ///   immediately.
+  /// - If CurLexicalContext is a function and we are compiling
+  ///   for the device, but we don't know yet that this function will be
+  ///   codegen'ed for the devive, creates a diagnostic which is emitted if and
+  ///   when we realize that the function will be codegen'ed.
+  ///
+  /// Example usage:
+  ///
+  /// Diagnose __float128 type usage only from SYCL device code if the current
+  /// target doesn't support it
+  /// if (!S.Context.getTargetInfo().hasFloat128Type() &&
+  ///     S.getLangOpts().SYCLIsDevice)
+  ///   DiagIfDeviceCode(Loc, diag::err_type_unsupported) << "__float128";
+  SemaDiagnosticBuilder DiagIfDeviceCode(SourceLocation Loc, unsigned DiagID);
+
+  void deepTypeCheckForDevice(SourceLocation UsedAt,
+                              llvm::DenseSet<QualType> Visited,
+                              ValueDecl *DeclToCheck);
+
+  ExprResult BuildUniqueStableNameExpr(SourceLocation OpLoc,
+                                       SourceLocation LParen,
+                                       SourceLocation RParen,
+                                       TypeSourceInfo *TSI);
+  ExprResult ActOnUniqueStableNameExpr(SourceLocation OpLoc,
+                                       SourceLocation LParen,
+                                       SourceLocation RParen,
+                                       ParsedType ParsedTy);
+};
+
+} // namespace clang
+
+#endif // LLVM_CLANG_SEMA_SEMASYCL_H
diff --git a/clang/lib/AST/Interp/ByteCodeExprGen.cpp b/clang/lib/AST/Interp/ByteCodeExprGen.cpp
@@ -173,10 +173,18 @@ bool ByteCodeExprGen<Emitter>::VisitCastExpr(const CastExpr *CE) {
     return this->emitCastFloatingIntegral(*ToT, CE);
   }
 
-  case CK_NullToPointer:
+  case CK_NullToPointer: {
     if (DiscardResult)
       return true;
-    return this->emitNull(classifyPrim(CE->getType()), CE);
+
+    const Descriptor *Desc = nullptr;
+    const QualType PointeeType = CE->getType()->getPointeeType();
+    if (!PointeeType.isNull()) {
+      if (std::optional<PrimType> T = classify(PointeeType))
+        Desc = P.createDescriptor(SubExpr, *T);
+    }
+    return this->emitNull(classifyPrim(CE->getType()), Desc, CE);
+  }
 
   case CK_PointerToIntegral: {
     if (DiscardResult)
@@ -199,6 +207,41 @@ bool ByteCodeExprGen<Emitter>::VisitCastExpr(const CastExpr *CE) {
     return true;
   }
 
+  case CK_IntegralToPointer: {
+    QualType IntType = SubExpr->getType();
+    assert(IntType->isIntegralOrEnumerationType());
+    if (!this->visit(SubExpr))
+      return false;
+    // FIXME: I think the discard is wrong since the int->ptr cast might cause a
+    // diagnostic.
+    PrimType T = classifyPrim(IntType);
+    if (DiscardResult)
+      return this->emitPop(T, CE);
+
+    QualType PtrType = CE->getType();
+    assert(PtrType->isPointerType());
+
+    const Descriptor *Desc;
+    if (std::optional<PrimType> T = classify(PtrType->getPointeeType()))
+      Desc = P.createDescriptor(SubExpr, *T);
+    else if (PtrType->getPointeeType()->isVoidType())
+      Desc = nullptr;
+    else
+      Desc = P.createDescriptor(CE, PtrType->getPointeeType().getTypePtr(),
+                                Descriptor::InlineDescMD, true, false,
+                                /*IsMutable=*/false, nullptr);
+
+    if (!this->emitGetIntPtr(T, Desc, CE))
+      return false;
+
+    PrimType DestPtrT = classifyPrim(PtrType);
+    if (DestPtrT == PT_Ptr)
+      return true;
+
+    // In case we're converting the integer to a non-Pointer.
+    return this->emitDecayPtr(PT_Ptr, DestPtrT, CE);
+  }
+
   case CK_AtomicToNonAtomic:
   case CK_ConstructorConversion:
   case CK_FunctionToPointerDecay:
@@ -207,13 +250,31 @@ bool ByteCodeExprGen<Emitter>::VisitCastExpr(const CastExpr *CE) {
   case CK_UserDefinedConversion:
     return this->delegate(SubExpr);
 
-  case CK_BitCast:
+  case CK_BitCast: {
+    // Reject bitcasts to atomic types.
     if (CE->getType()->isAtomicType()) {
       if (!this->discard(SubExpr))
         return false;
       return this->emitInvalidCast(CastKind::Reinterpret, CE);
     }
-    return this->delegate(SubExpr);
+
+    if (DiscardResult)
+      return this->discard(SubExpr);
+
+    std::optional<PrimType> FromT = classify(SubExpr->getType());
+    std::optional<PrimType> ToT = classifyPrim(CE->getType());
+    if (!FromT || !ToT)
+      return false;
+
+    assert(isPtrType(*FromT));
+    assert(isPtrType(*ToT));
+    if (FromT == ToT)
+      return this->delegate(SubExpr);
+
+    if (!this->visit(SubExpr))
+      return false;
+    return this->emitDecayPtr(*FromT, *ToT, CE);
+  }
 
   case CK_IntegralToBoolean:
   case CK_IntegralCast: {
@@ -245,7 +306,7 @@ bool ByteCodeExprGen<Emitter>::VisitCastExpr(const CastExpr *CE) {
     if (!this->visit(SubExpr))
       return false;
 
-    if (!this->emitNull(PtrT, CE))
+    if (!this->emitNull(PtrT, nullptr, CE))
       return false;
 
     return this->emitNE(PtrT, CE);
@@ -455,7 +516,7 @@ bool ByteCodeExprGen<Emitter>::VisitBinaryOperator(const BinaryOperator *BO) {
 
   // Pointer arithmetic special case.
   if (BO->getOpcode() == BO_Add || BO->getOpcode() == BO_Sub) {
-    if (T == PT_Ptr || (LT == PT_Ptr && RT == PT_Ptr))
+    if (isPtrType(*T) || (isPtrType(*LT) && isPtrType(*RT)))
       return this->VisitPointerArithBinOp(BO);
   }
 
@@ -1033,6 +1094,34 @@ bool ByteCodeExprGen<Emitter>::VisitInitListExpr(const InitListExpr *E) {
     return true;
   }
 
+  if (const auto *VecT = E->getType()->getAs<VectorType>()) {
+    unsigned NumVecElements = VecT->getNumElements();
+    assert(NumVecElements >= E->getNumInits());
+
+    QualType ElemQT = VecT->getElementType();
+    PrimType ElemT = classifyPrim(ElemQT);
+
+    // All initializer elements.
+    unsigned InitIndex = 0;
+    for (const Expr *Init : E->inits()) {
+      if (!this->visit(Init))
+        return false;
+
+      if (!this->emitInitElem(ElemT, InitIndex, E))
+        return false;
+      ++InitIndex;
+    }
+
+    // Fill the rest with zeroes.
+    for (; InitIndex != NumVecElements; ++InitIndex) {
+      if (!this->visitZeroInitializer(ElemT, ElemQT, E))
+        return false;
+      if (!this->emitInitElem(ElemT, InitIndex, E))
+        return false;
+    }
+    return true;
+  }
+
   return false;
 }
 
@@ -2326,7 +2415,7 @@ bool ByteCodeExprGen<Emitter>::visitBool(const Expr *E) {
 
   // Convert pointers to bool.
   if (T == PT_Ptr || T == PT_FnPtr) {
-    if (!this->emitNull(*T, E))
+    if (!this->emitNull(*T, nullptr, E))
       return false;
     return this->emitNE(*T, E);
   }
@@ -2366,9 +2455,9 @@ bool ByteCodeExprGen<Emitter>::visitZeroInitializer(PrimType T, QualType QT,
   case PT_IntAPS:
     return this->emitZeroIntAPS(Ctx.getBitWidth(QT), E);
   case PT_Ptr:
-    return this->emitNullPtr(E);
+    return this->emitNullPtr(nullptr, E);
   case PT_FnPtr:
-    return this->emitNullFnPtr(E);
+    return this->emitNullFnPtr(nullptr, E);
   case PT_Float: {
     return this->emitConstFloat(APFloat::getZero(Ctx.getFloatSemantics(QT)), E);
   }
@@ -2952,7 +3041,7 @@ bool ByteCodeExprGen<Emitter>::VisitCXXNullPtrLiteralExpr(
   if (DiscardResult)
     return true;
 
-  return this->emitNullPtr(E);
+  return this->emitNullPtr(nullptr, E);
 }
 
 template <class Emitter>

diff --git a/clang/lib/AST/Interp/ByteCodeStmtGen.cpp b/clang/lib/AST/Interp/ByteCodeStmtGen.cpp
@@ -110,7 +110,7 @@ bool ByteCodeStmtGen<Emitter>::emitLambdaStaticInvokerBody(
   // one here, and we don't need one either because the lambda cannot have
   // any captures, as verified above. Emit a null pointer. This is then
   // special-cased when interpreting to not emit any misleading diagnostics.
-  if (!this->emitNullPtr(MD))
+  if (!this->emitNullPtr(nullptr, MD))
     return false;
 
   // Forward all arguments from the static invoker to the lambda call operator.

diff --git a/clang/lib/AST/Interp/Context.cpp b/clang/lib/AST/Interp/Context.cpp
@@ -120,7 +120,8 @@ std::optional<PrimType> Context::classify(QualType T) const {
   if (T->isBooleanType())
     return PT_Bool;
 
-  if (T->isAnyComplexType())
+  // We map these to primitive arrays.
+  if (T->isAnyComplexType() || T->isVectorType())
     return std::nullopt;
 
   if (T->isSignedIntegerOrEnumerationType()) {

diff --git a/clang/lib/AST/Interp/Descriptor.h b/clang/lib/AST/Interp/Descriptor.h
@@ -168,6 +168,7 @@ struct Descriptor final {
 
   const Decl *asDecl() const { return Source.dyn_cast<const Decl *>(); }
   const Expr *asExpr() const { return Source.dyn_cast<const Expr *>(); }
+  const DeclTy &getSource() const { return Source; }
 
   const ValueDecl *asValueDecl() const {
     return dyn_cast_if_present<ValueDecl>(asDecl());

diff --git a/clang/lib/AST/Interp/EvalEmitter.cpp b/clang/lib/AST/Interp/EvalEmitter.cpp
@@ -51,7 +51,8 @@ EvaluationResult EvalEmitter::interpretDecl(const VarDecl *VD,
   this->CheckFullyInitialized = CheckFullyInitialized;
   this->ConvertResultToRValue =
       VD->getAnyInitializer() &&
-      (VD->getAnyInitializer()->getType()->isAnyComplexType());
+      (VD->getAnyInitializer()->getType()->isAnyComplexType() ||
+       VD->getAnyInitializer()->getType()->isVectorType());
   EvalResult.setSource(VD);
 
   if (!this->visitDecl(VD) && EvalResult.empty())

diff --git a/clang/lib/AST/Interp/FunctionPointer.h b/clang/lib/AST/Interp/FunctionPointer.h
@@ -22,7 +22,11 @@ class FunctionPointer final {
   const Function *Func;
 
 public:
-  FunctionPointer() : Func(nullptr) {}
+  // FIXME: We might want to track the fact that the Function pointer
+  // has been created from an integer and is most likely garbage anyway.
+  FunctionPointer(uintptr_t IntVal = 0, const Descriptor *Desc = nullptr)
+      : Func(reinterpret_cast<const Function *>(IntVal)) {}
+
   FunctionPointer(const Function *Func) : Func(Func) { assert(Func); }
 
   const Function *getFunction() const { return Func; }
@@ -53,6 +57,10 @@ class FunctionPointer final {
     return toAPValue().getAsString(Ctx, Func->getDecl()->getType());
   }
 
+  uint64_t getIntegerRepresentation() const {
+    return static_cast<uint64_t>(reinterpret_cast<uintptr_t>(Func));
+  }
+
   ComparisonCategoryResult compare(const FunctionPointer &RHS) const {
     if (Func == RHS.Func)
       return ComparisonCategoryResult::Equal;

diff --git a/clang/lib/AST/Interp/Interp.cpp b/clang/lib/AST/Interp/Interp.cpp
@@ -282,6 +282,8 @@ bool CheckConstant(InterpState &S, CodePtr OpPC, const Descriptor *Desc) {
 }
 
 static bool CheckConstant(InterpState &S, CodePtr OpPC, const Pointer &Ptr) {
+  if (Ptr.isIntegralPointer())
+    return true;
   return CheckConstant(S, OpPC, Ptr.getDeclDesc());
 }
 
@@ -335,6 +337,9 @@ bool CheckConst(InterpState &S, CodePtr OpPC, const Pointer &Ptr) {
     return true;
   }
 
+  if (!Ptr.isBlockPointer())
+    return false;
+
   const QualType Ty = Ptr.getType();
   const SourceInfo &Loc = S.Current->getSource(OpPC);
   S.FFDiag(Loc, diag::note_constexpr_modify_const_type) << Ty;

diff --git a/clang/lib/AST/Interp/Interp.h b/clang/lib/AST/Interp/Interp.h
@@ -804,8 +804,7 @@ inline bool CmpHelperEQ<Pointer>(InterpState &S, CodePtr OpPC, CompareFn Fn) {
   for (const auto &P : {LHS, RHS}) {
     if (P.isZero())
       continue;
-    if (const ValueDecl *VD = P.getDeclDesc()->asValueDecl();
-        VD && VD->isWeak()) {
+    if (P.isWeak()) {
       const SourceInfo &Loc = S.Current->getSource(OpPC);
       S.FFDiag(Loc, diag::note_constexpr_pointer_weak_comparison)
           << P.toDiagnosticString(S.getCtx());
@@ -824,9 +823,9 @@ inline bool CmpHelperEQ<Pointer>(InterpState &S, CodePtr OpPC, CompareFn Fn) {
     // element in the same array are NOT equal. They have the same Base value,
     // but a different Offset. This is a pretty rare case, so we fix this here
     // by comparing pointers to the first elements.
-    if (!LHS.isDummy() && LHS.isArrayRoot())
+    if (!LHS.isZero() && !LHS.isDummy() && LHS.isArrayRoot())
       VL = LHS.atIndex(0).getByteOffset();
-    if (!RHS.isDummy() && RHS.isArrayRoot())
+    if (!RHS.isZero() && !RHS.isDummy() && RHS.isArrayRoot())
       VR = RHS.atIndex(0).getByteOffset();
 
     S.Stk.push<BoolT>(BoolT::from(Fn(Compare(VL, VR))));
@@ -1387,6 +1386,8 @@ bool Load(InterpState &S, CodePtr OpPC) {
   const Pointer &Ptr = S.Stk.peek<Pointer>();
   if (!CheckLoad(S, OpPC, Ptr))
     return false;
+  if (!Ptr.isBlockPointer())
+    return false;
   S.Stk.push<T>(Ptr.deref<T>());
   return true;
 }
@@ -1396,6 +1397,8 @@ bool LoadPop(InterpState &S, CodePtr OpPC) {
   const Pointer &Ptr = S.Stk.pop<Pointer>();
   if (!CheckLoad(S, OpPC, Ptr))
     return false;
+  if (!Ptr.isBlockPointer())
+    return false;
   S.Stk.push<T>(Ptr.deref<T>());
   return true;
 }
@@ -1534,8 +1537,12 @@ bool OffsetHelper(InterpState &S, CodePtr OpPC, const T &Offset,
     return true;
   }
 
-  if (!CheckNull(S, OpPC, Ptr, CSK_ArrayIndex))
-    return false;
+  if (!CheckNull(S, OpPC, Ptr, CSK_ArrayIndex)) {
+    // The CheckNull will have emitted a note already, but we only
+    // abort in C++, since this is fine in C.
+    if (S.getLangOpts().CPlusPlus)
+      return false;
+  }
 
   // Arrays of unknown bounds cannot have pointers into them.
   if (!CheckArray(S, OpPC, Ptr))
@@ -1561,23 +1568,25 @@ bool OffsetHelper(InterpState &S, CodePtr OpPC, const T &Offset,
     Invalid = true;
   };
 
-  T MaxOffset = T::from(MaxIndex - Index, Offset.bitWidth());
-  if constexpr (Op == ArithOp::Add) {
-    // If the new offset would be negative, bail out.
-    if (Offset.isNegative() && (Offset.isMin() || -Offset > Index))
-      DiagInvalidOffset();
-
-    // If the new offset would be out of bounds, bail out.
-    if (Offset.isPositive() && Offset > MaxOffset)
-      DiagInvalidOffset();
-  } else {
-    // If the new offset would be negative, bail out.
-    if (Offset.isPositive() && Index < Offset)
-      DiagInvalidOffset();
-
-    // If the new offset would be out of bounds, bail out.
-    if (Offset.isNegative() && (Offset.isMin() || -Offset > MaxOffset))
-      DiagInvalidOffset();
+  if (Ptr.isBlockPointer()) {
+    T MaxOffset = T::from(MaxIndex - Index, Offset.bitWidth());
+    if constexpr (Op == ArithOp::Add) {
+      // If the new offset would be negative, bail out.
+      if (Offset.isNegative() && (Offset.isMin() || -Offset > Index))
+        DiagInvalidOffset();
+
+      // If the new offset would be out of bounds, bail out.
+      if (Offset.isPositive() && Offset > MaxOffset)
+        DiagInvalidOffset();
+    } else {
+      // If the new offset would be negative, bail out.
+      if (Offset.isPositive() && Index < Offset)
+        DiagInvalidOffset();
+
+      // If the new offset would be out of bounds, bail out.
+      if (Offset.isNegative() && (Offset.isMin() || -Offset > MaxOffset))
+        DiagInvalidOffset();
+    }
   }
 
   if (Invalid && !Ptr.isDummy() && S.getLangOpts().CPlusPlus)
@@ -1661,6 +1670,11 @@ inline bool SubPtr(InterpState &S, CodePtr OpPC) {
   const Pointer &LHS = S.Stk.pop<Pointer>();
   const Pointer &RHS = S.Stk.pop<Pointer>();
 
+  if (RHS.isZero()) {
+    S.Stk.push<T>(T::from(LHS.getIndex()));
+    return true;
+  }
+
   if (!Pointer::hasSameBase(LHS, RHS) && S.getLangOpts().CPlusPlus) {
     // TODO: Diagnose.
     return false;
@@ -1839,8 +1853,9 @@ static inline bool ZeroIntAPS(InterpState &S, CodePtr OpPC, uint32_t BitWidth) {
 }
 
 template <PrimType Name, class T = typename PrimConv<Name>::T>
-inline bool Null(InterpState &S, CodePtr OpPC) {
-  S.Stk.push<T>();
+inline bool Null(InterpState &S, CodePtr OpPC, const Descriptor *Desc) {
+  // Note: Desc can be null.
+  S.Stk.push<T>(0, Desc);
   return true;
 }
 
@@ -2244,6 +2259,14 @@ inline bool GetFnPtr(InterpState &S, CodePtr OpPC, const Function *Func) {
   return true;
 }
 
+template <PrimType Name, class T = typename PrimConv<Name>::T>
+inline bool GetIntPtr(InterpState &S, CodePtr OpPC, const Descriptor *Desc) {
+  const T &IntVal = S.Stk.pop<T>();
+
+  S.Stk.push<Pointer>(static_cast<uint64_t>(IntVal), Desc);
+  return true;
+}
+
 /// Just emit a diagnostic. The expression that caused emission of this
 /// op is not valid in a constant context.
 inline bool Invalid(InterpState &S, CodePtr OpPC) {
@@ -2300,6 +2323,18 @@ inline bool CheckNonNullArg(InterpState &S, CodePtr OpPC) {
   return false;
 }
 
+/// OldPtr -> Integer -> NewPtr.
+template <PrimType TIn, PrimType TOut>
+inline bool DecayPtr(InterpState &S, CodePtr OpPC) {
+  static_assert(isPtrType(TIn) && isPtrType(TOut));
+  using FromT = typename PrimConv<TIn>::T;
+  using ToT = typename PrimConv<TOut>::T;
+
+  const FromT &OldPtr = S.Stk.pop<FromT>();
+  S.Stk.push<ToT>(ToT(OldPtr.getIntegerRepresentation(), nullptr));
+  return true;
+}
+
 //===----------------------------------------------------------------------===//
 // Read opcode arguments
 //===----------------------------------------------------------------------===//

diff --git a/clang/lib/AST/Interp/InterpBlock.cpp b/clang/lib/AST/Interp/InterpBlock.cpp
@@ -73,7 +73,7 @@ void Block::replacePointer(Pointer *Old, Pointer *New) {
   removePointer(Old);
   addPointer(New);
 
-  Old->Pointee = nullptr;
+  Old->PointeeStorage.BS.Pointee = nullptr;
 
 #ifndef NDEBUG
   assert(!hasPointer(Old));
@@ -104,7 +104,7 @@ DeadBlock::DeadBlock(DeadBlock *&Root, Block *Blk)
   // Transfer pointers.
   B.Pointers = Blk->Pointers;
   for (Pointer *P = Blk->Pointers; P; P = P->Next)
-    P->Pointee = &B;
+    P->PointeeStorage.BS.Pointee = &B;
 }
 
 void DeadBlock::free() {

diff --git a/clang/lib/AST/Interp/InterpBuiltin.cpp b/clang/lib/AST/Interp/InterpBuiltin.cpp
@@ -16,6 +16,16 @@
 namespace clang {
 namespace interp {
 
+static unsigned callArgSize(const InterpState &S, const CallExpr *C) {
+  unsigned O = 0;
+
+  for (const Expr *E : C->arguments()) {
+    O += align(primSize(*S.getContext().classify(E)));
+  }
+
+  return O;
+}
+
 template <typename T>
 static T getParam(const InterpFrame *Frame, unsigned Index) {
   assert(Frame->getFunction()->getNumParams() > Index);
@@ -816,18 +826,30 @@ static bool interp__builtin_carryop(InterpState &S, CodePtr OpPC,
 static bool interp__builtin_clz(InterpState &S, CodePtr OpPC,
                                 const InterpFrame *Frame, const Function *Func,
                                 const CallExpr *Call) {
+  unsigned CallSize = callArgSize(S, Call);
   unsigned BuiltinOp = Func->getBuiltinID();
   PrimType ValT = *S.getContext().classify(Call->getArg(0));
-  const APSInt &Val = peekToAPSInt(S.Stk, ValT);
+  const APSInt &Val = peekToAPSInt(S.Stk, ValT, CallSize);
 
   // When the argument is 0, the result of GCC builtins is undefined, whereas
   // for Microsoft intrinsics, the result is the bit-width of the argument.
   bool ZeroIsUndefined = BuiltinOp != Builtin::BI__lzcnt16 &&
                          BuiltinOp != Builtin::BI__lzcnt &&
                          BuiltinOp != Builtin::BI__lzcnt64;
 
-  if (ZeroIsUndefined && Val == 0)
-    return false;
+  if (Val == 0) {
+    if (Func->getBuiltinID() == Builtin::BI__builtin_clzg &&
+        Call->getNumArgs() == 2) {
+      // We have a fallback parameter.
+      PrimType FallbackT = *S.getContext().classify(Call->getArg(1));
+      const APSInt &Fallback = peekToAPSInt(S.Stk, FallbackT);
+      pushInteger(S, Fallback, Call->getType());
+      return true;
+    }
+
+    if (ZeroIsUndefined)
+      return false;
+  }
 
   pushInteger(S, Val.countl_zero(), Call->getType());
   return true;
@@ -836,11 +858,21 @@ static bool interp__builtin_clz(InterpState &S, CodePtr OpPC,
 static bool interp__builtin_ctz(InterpState &S, CodePtr OpPC,
                                 const InterpFrame *Frame, const Function *Func,
                                 const CallExpr *Call) {
+  unsigned CallSize = callArgSize(S, Call);
   PrimType ValT = *S.getContext().classify(Call->getArg(0));
-  const APSInt &Val = peekToAPSInt(S.Stk, ValT);
-
-  if (Val == 0)
+  const APSInt &Val = peekToAPSInt(S.Stk, ValT, CallSize);
+
+  if (Val == 0) {
+    if (Func->getBuiltinID() == Builtin::BI__builtin_ctzg &&
+        Call->getNumArgs() == 2) {
+      // We have a fallback parameter.
+      PrimType FallbackT = *S.getContext().classify(Call->getArg(1));
+      const APSInt &Fallback = peekToAPSInt(S.Stk, FallbackT);
+      pushInteger(S, Fallback, Call->getType());
+      return true;
+    }
     return false;
+  }
 
   pushInteger(S, Val.countr_zero(), Call->getType());
   return true;
@@ -1223,6 +1255,7 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const Function *F,
   case Builtin::BI__builtin_clzl:
   case Builtin::BI__builtin_clzll:
   case Builtin::BI__builtin_clzs:
+  case Builtin::BI__builtin_clzg:
   case Builtin::BI__lzcnt16: // Microsoft variants of count leading-zeroes
   case Builtin::BI__lzcnt:
   case Builtin::BI__lzcnt64:
@@ -1234,6 +1267,7 @@ bool InterpretBuiltin(InterpState &S, CodePtr OpPC, const Function *F,
   case Builtin::BI__builtin_ctzl:
   case Builtin::BI__builtin_ctzll:
   case Builtin::BI__builtin_ctzs:
+  case Builtin::BI__builtin_ctzg:
     if (!interp__builtin_ctz(S, OpPC, Frame, F, Call))
       return false;
     break;

diff --git a/clang/lib/AST/Interp/Opcodes.td b/clang/lib/AST/Interp/Opcodes.td
@@ -59,6 +59,7 @@ def ArgCastKind : ArgType { let Name = "CastKind"; }
 def ArgCallExpr : ArgType { let Name = "const CallExpr *"; }
 def ArgOffsetOfExpr : ArgType { let Name = "const OffsetOfExpr *"; }
 def ArgDeclRef : ArgType { let Name = "const DeclRefExpr *"; }
+def ArgDesc : ArgType { let Name = "const Descriptor *"; }
 def ArgCCI : ArgType { let Name = "const ComparisonCategoryInfo *"; }
 
 //===----------------------------------------------------------------------===//
@@ -272,6 +273,7 @@ def ZeroIntAPS : Opcode {
 // [] -> [Pointer]
 def Null : Opcode {
   let Types = [PtrTypeClass];
+  let Args = [ArgDesc];
   let HasGroup = 1;
 }
 
@@ -530,6 +532,11 @@ def GetFnPtr : Opcode {
   let Args = [ArgFunction];
 }
 
+def GetIntPtr : Opcode {
+  let Types = [AluTypeClass];
+  let Args = [ArgDesc];
+  let HasGroup = 1;
+}
 
 //===----------------------------------------------------------------------===//
 // Binary operators.
@@ -662,6 +669,11 @@ def CastPointerIntegral : Opcode {
   let HasGroup = 1;
 }
 
+def DecayPtr : Opcode {
+  let Types = [PtrTypeClass, PtrTypeClass];
+  let HasGroup = 1;
+}
+
 //===----------------------------------------------------------------------===//
 // Comparison opcodes.
 //===----------------------------------------------------------------------===//

diff --git a/clang/lib/AST/Interp/Pointer.cpp b/clang/lib/AST/Interp/Pointer.cpp
@@ -26,60 +26,95 @@ Pointer::Pointer(Block *Pointee)
 Pointer::Pointer(Block *Pointee, unsigned BaseAndOffset)
     : Pointer(Pointee, BaseAndOffset, BaseAndOffset) {}
 
-Pointer::Pointer(const Pointer &P) : Pointer(P.Pointee, P.Base, P.Offset) {}
+Pointer::Pointer(const Pointer &P)
+    : Offset(P.Offset), PointeeStorage(P.PointeeStorage),
+      StorageKind(P.StorageKind) {
 
-Pointer::Pointer(Pointer &&P)
-    : Pointee(P.Pointee), Base(P.Base), Offset(P.Offset) {
-  if (Pointee)
-    Pointee->replacePointer(&P, this);
+  if (isBlockPointer() && PointeeStorage.BS.Pointee)
+    PointeeStorage.BS.Pointee->addPointer(this);
 }
 
 Pointer::Pointer(Block *Pointee, unsigned Base, unsigned Offset)
-    : Pointee(Pointee), Base(Base), Offset(Offset) {
+    : Offset(Offset), StorageKind(Storage::Block) {
   assert((Base == RootPtrMark || Base % alignof(void *) == 0) && "wrong base");
+
+  PointeeStorage.BS = {Pointee, Base};
+
   if (Pointee)
     Pointee->addPointer(this);
 }
 
+Pointer::Pointer(Pointer &&P)
+    : Offset(P.Offset), PointeeStorage(P.PointeeStorage),
+      StorageKind(P.StorageKind) {
+
+  if (StorageKind == Storage::Block && PointeeStorage.BS.Pointee)
+    PointeeStorage.BS.Pointee->replacePointer(&P, this);
+}
+
 Pointer::~Pointer() {
-  if (Pointee) {
-    Pointee->removePointer(this);
-    Pointee->cleanup();
+  if (isIntegralPointer())
+    return;
+
+  if (PointeeStorage.BS.Pointee) {
+    PointeeStorage.BS.Pointee->removePointer(this);
+    PointeeStorage.BS.Pointee->cleanup();
   }
 }
 
 void Pointer::operator=(const Pointer &P) {
-  Block *Old = Pointee;
 
-  if (Pointee)
-    Pointee->removePointer(this);
+  if (!this->isIntegralPointer() || !P.isBlockPointer())
+    assert(P.StorageKind == StorageKind);
 
-  Offset = P.Offset;
-  Base = P.Base;
+  bool WasBlockPointer = isBlockPointer();
+  StorageKind = P.StorageKind;
+  if (StorageKind == Storage::Block) {
+    Block *Old = PointeeStorage.BS.Pointee;
+    if (WasBlockPointer && PointeeStorage.BS.Pointee)
+      PointeeStorage.BS.Pointee->removePointer(this);
 
-  Pointee = P.Pointee;
-  if (Pointee)
-    Pointee->addPointer(this);
+    Offset = P.Offset;
+    PointeeStorage.BS = P.PointeeStorage.BS;
+
+    if (PointeeStorage.BS.Pointee)
+      PointeeStorage.BS.Pointee->addPointer(this);
+
+    if (WasBlockPointer && Old)
+      Old->cleanup();
 
-  if (Old)
-    Old->cleanup();
+  } else if (StorageKind == Storage::Int) {
+    PointeeStorage.Int = P.PointeeStorage.Int;
+  } else {
+    assert(false && "Unhandled storage kind");
+  }
 }
 
 void Pointer::operator=(Pointer &&P) {
-  Block *Old = Pointee;
+  if (!this->isIntegralPointer() || !P.isBlockPointer())
+    assert(P.StorageKind == StorageKind);
 
-  if (Pointee)
-    Pointee->removePointer(this);
+  bool WasBlockPointer = isBlockPointer();
+  StorageKind = P.StorageKind;
+  if (StorageKind == Storage::Block) {
+    Block *Old = PointeeStorage.BS.Pointee;
+    if (WasBlockPointer && PointeeStorage.BS.Pointee)
+      PointeeStorage.BS.Pointee->removePointer(this);
 
-  Offset = P.Offset;
-  Base = P.Base;
+    Offset = P.Offset;
+    PointeeStorage.BS = P.PointeeStorage.BS;
 
-  Pointee = P.Pointee;
-  if (Pointee)
-    Pointee->replacePointer(&P, this);
+    if (PointeeStorage.BS.Pointee)
+      PointeeStorage.BS.Pointee->addPointer(this);
+
+    if (WasBlockPointer && Old)
+      Old->cleanup();
 
-  if (Old)
-    Old->cleanup();
+  } else if (StorageKind == Storage::Int) {
+    PointeeStorage.Int = P.PointeeStorage.Int;
+  } else {
+    assert(false && "Unhandled storage kind");
+  }
 }
 
 APValue Pointer::toAPValue() const {
@@ -88,6 +123,11 @@ APValue Pointer::toAPValue() const {
   if (isZero())
     return APValue(static_cast<const Expr *>(nullptr), CharUnits::Zero(), Path,
                    /*IsOnePastEnd=*/false, /*IsNullPtr=*/true);
+  if (isIntegralPointer())
+    return APValue(static_cast<const Expr *>(nullptr),
+                   CharUnits::fromQuantity(asIntPointer().Value + this->Offset),
+                   Path,
+                   /*IsOnePastEnd=*/false, /*IsNullPtr=*/false);
 
   // Build the lvalue base from the block.
   const Descriptor *Desc = getDeclDesc();
@@ -137,19 +177,52 @@ APValue Pointer::toAPValue() const {
   return APValue(Base, Offset, Path, IsOnePastEnd, /*IsNullPtr=*/false);
 }
 
+void Pointer::print(llvm::raw_ostream &OS) const {
+  OS << PointeeStorage.BS.Pointee << " (";
+  if (isBlockPointer()) {
+    OS << "Block) {";
+
+    if (PointeeStorage.BS.Base == RootPtrMark)
+      OS << "rootptr, ";
+    else
+      OS << PointeeStorage.BS.Base << ", ";
+
+    if (Offset == PastEndMark)
+      OS << "pastend, ";
+    else
+      OS << Offset << ", ";
+
+    if (isBlockPointer() && PointeeStorage.BS.Pointee)
+      OS << PointeeStorage.BS.Pointee->getSize();
+    else
+      OS << "nullptr";
+  } else {
+    OS << "Int) {";
+    OS << PointeeStorage.Int.Value << ", " << PointeeStorage.Int.Desc;
+  }
+  OS << "}";
+}
+
 std::string Pointer::toDiagnosticString(const ASTContext &Ctx) const {
-  if (!Pointee)
+  if (isZero())
     return "nullptr";
 
+  if (isIntegralPointer())
+    return (Twine("&(") + Twine(asIntPointer().Value + Offset) + ")").str();
+
   return toAPValue().getAsString(Ctx, getType());
 }
 
 bool Pointer::isInitialized() const {
-  assert(Pointee && "Cannot check if null pointer was initialized");
+  if (isIntegralPointer())
+    return true;
+
+  assert(PointeeStorage.BS.Pointee &&
+         "Cannot check if null pointer was initialized");
   const Descriptor *Desc = getFieldDesc();
   assert(Desc);
   if (Desc->isPrimitiveArray()) {
-    if (isStatic() && Base == 0)
+    if (isStatic() && PointeeStorage.BS.Base == 0)
       return true;
 
     InitMapPtr &IM = getInitMap();
@@ -164,17 +237,20 @@ bool Pointer::isInitialized() const {
   }
 
   // Field has its bit in an inline descriptor.
-  return Base == 0 || getInlineDesc()->IsInitialized;
+  return PointeeStorage.BS.Base == 0 || getInlineDesc()->IsInitialized;
 }
 
 void Pointer::initialize() const {
-  assert(Pointee && "Cannot initialize null pointer");
+  if (isIntegralPointer())
+    return;
+
+  assert(PointeeStorage.BS.Pointee && "Cannot initialize null pointer");
   const Descriptor *Desc = getFieldDesc();
 
   assert(Desc);
   if (Desc->isPrimitiveArray()) {
     // Primitive global arrays don't have an initmap.
-    if (isStatic() && Base == 0)
+    if (isStatic() && PointeeStorage.BS.Base == 0)
       return;
 
     // Nothing to do for these.
@@ -200,13 +276,15 @@ void Pointer::initialize() const {
   }
 
   // Field has its bit in an inline descriptor.
-  assert(Base != 0 && "Only composite fields can be initialised");
+  assert(PointeeStorage.BS.Base != 0 &&
+         "Only composite fields can be initialised");
   getInlineDesc()->IsInitialized = true;
 }
 
 void Pointer::activate() const {
   // Field has its bit in an inline descriptor.
-  assert(Base != 0 && "Only composite fields can be initialised");
+  assert(PointeeStorage.BS.Base != 0 &&
+         "Only composite fields can be initialised");
   getInlineDesc()->IsActive = true;
 }
 
@@ -215,11 +293,23 @@ void Pointer::deactivate() const {
 }
 
 bool Pointer::hasSameBase(const Pointer &A, const Pointer &B) {
-  return A.Pointee == B.Pointee;
+  // Two null pointers always have the same base.
+  if (A.isZero() && B.isZero())
+    return true;
+
+  if (A.isIntegralPointer() && B.isIntegralPointer())
+    return true;
+
+  if (A.isIntegralPointer() || B.isIntegralPointer())
+    return A.getSource() == B.getSource();
+
+  return A.asBlockPointer().Pointee == B.asBlockPointer().Pointee;
 }
 
 bool Pointer::hasSameArray(const Pointer &A, const Pointer &B) {
-  return hasSameBase(A, B) && A.Base == B.Base && A.getFieldDesc()->IsArray;
+  return hasSameBase(A, B) &&
+         A.PointeeStorage.BS.Base == B.PointeeStorage.BS.Base &&
+         A.getFieldDesc()->IsArray;
 }
 
 std::optional<APValue> Pointer::toRValue(const Context &Ctx) const {
@@ -342,6 +432,25 @@ std::optional<APValue> Pointer::toRValue(const Context &Ctx) const {
       return false;
     }
 
+    // Vector types.
+    if (const auto *VT = Ty->getAs<VectorType>()) {
+      assert(Ptr.getFieldDesc()->isPrimitiveArray());
+      QualType ElemTy = VT->getElementType();
+      PrimType ElemT = *Ctx.classify(ElemTy);
+
+      SmallVector<APValue> Values;
+      Values.reserve(VT->getNumElements());
+      for (unsigned I = 0; I != VT->getNumElements(); ++I) {
+        TYPE_SWITCH(ElemT, {
+          Values.push_back(Ptr.atIndex(I).deref<T>().toAPValue());
+        });
+      }
+
+      assert(Values.size() == VT->getNumElements());
+      R = APValue(Values.data(), Values.size());
+      return true;
+    }
+
     llvm_unreachable("invalid value to return");
   };
 

diff --git a/clang/lib/AST/Interp/Pointer.h b/clang/lib/AST/Interp/Pointer.h
diff --git a/clang/lib/AST/Interp/PrimType.h b/clang/lib/AST/Interp/PrimType.h
@@ -46,6 +46,10 @@ enum PrimType : unsigned {
   PT_FnPtr,
 };
 
+inline constexpr bool isPtrType(PrimType T) {
+  return T == PT_Ptr || T == PT_FnPtr;
+}
+
 enum class CastKind : uint8_t {
   Reinterpret,
   Atomic,

diff --git a/clang/lib/AST/Interp/Program.cpp b/clang/lib/AST/Interp/Program.cpp
@@ -411,5 +411,12 @@ Descriptor *Program::createDescriptor(const DeclTy &D, const Type *Ty,
                               IsMutable);
   }
 
+  // Same with vector types.
+  if (const auto *VT = Ty->getAs<VectorType>()) {
+    PrimType ElemTy = *Ctx.classify(VT->getElementType());
+    return allocateDescriptor(D, ElemTy, MDSize, VT->getNumElements(), IsConst,
+                              IsTemporary, IsMutable);
+  }
+
   return nullptr;
 }
diff --git a/clang/lib/AST/OpenACCClause.cpp b/clang/lib/AST/OpenACCClause.cpp
@@ -15,3 +15,22 @@
 #include "clang/AST/ASTContext.h"
 
 using namespace clang;
+
+OpenACCDefaultClause *OpenACCDefaultClause::Create(const ASTContext &C,
+                                                   OpenACCDefaultClauseKind K,
+                                                   SourceLocation BeginLoc,
+                                                   SourceLocation LParenLoc,
+                                                   SourceLocation EndLoc) {
+  void *Mem =
+      C.Allocate(sizeof(OpenACCDefaultClause), alignof(OpenACCDefaultClause));
+
+  return new (Mem) OpenACCDefaultClause(K, BeginLoc, LParenLoc, EndLoc);
+}
+
+//===----------------------------------------------------------------------===//
+//  OpenACC clauses printing methods
+//===----------------------------------------------------------------------===//
+void OpenACCClausePrinter::VisitOpenACCDefaultClause(
+    const OpenACCDefaultClause &C) {
+  OS << "default(" << C.getDefaultClauseKind() << ")";
+}
diff --git a/clang/lib/AST/ParentMapContext.cpp b/clang/lib/AST/ParentMapContext.cpp
@@ -61,7 +61,26 @@ class ParentMapContext::ParentMap {
   template <typename, typename...> friend struct ::MatchParents;
 
   /// Contains parents of a node.
-  using ParentVector = llvm::SmallVector<DynTypedNode, 2>;
+  class ParentVector {
+  public:
+    ParentVector() = default;
+    explicit ParentVector(size_t N, const DynTypedNode &Value) {
+      Items.reserve(N);
+      for (; N > 0; --N)
+        push_back(Value);
+    }
+    bool contains(const DynTypedNode &Value) {
+      return Seen.contains(Value);
+    }
+    void push_back(const DynTypedNode &Value) {
+      if (!Value.getMemoizationData() || Seen.insert(Value).second)
+        Items.push_back(Value);
+    }
+    llvm::ArrayRef<DynTypedNode> view() const { return Items; }
+  private:
+    llvm::SmallVector<DynTypedNode, 2> Items;
+    llvm::SmallDenseSet<DynTypedNode, 2> Seen;
+  };
 
   /// Maps from a node to its parents. This is used for nodes that have
   /// pointer identity only, which are more common and we can save space by
@@ -99,7 +118,7 @@ class ParentMapContext::ParentMap {
       return llvm::ArrayRef<DynTypedNode>();
     }
     if (const auto *V = I->second.template dyn_cast<ParentVector *>()) {
-      return llvm::ArrayRef(*V);
+      return V->view();
     }
     return getSingleDynTypedNodeFromParentMap(I->second);
   }
@@ -252,7 +271,7 @@ class ParentMapContext::ParentMap {
       const auto *S = It->second.dyn_cast<const Stmt *>();
       if (!S) {
         if (auto *Vec = It->second.dyn_cast<ParentVector *>())
-          return llvm::ArrayRef(*Vec);
+          return Vec->view();
         return getSingleDynTypedNodeFromParentMap(It->second);
       }
       const auto *P = dyn_cast<Expr>(S);

diff --git a/clang/lib/AST/StmtOpenMP.cpp b/clang/lib/AST/StmtOpenMP.cpp
@@ -2431,7 +2431,7 @@ OMPTeamsGenericLoopDirective::CreateEmpty(const ASTContext &C,
 OMPTargetTeamsGenericLoopDirective *OMPTargetTeamsGenericLoopDirective::Create(
     const ASTContext &C, SourceLocation StartLoc, SourceLocation EndLoc,
     unsigned CollapsedNum, ArrayRef<OMPClause *> Clauses, Stmt *AssociatedStmt,
-    const HelperExprs &Exprs) {
+    const HelperExprs &Exprs, bool CanBeParallelFor) {
   auto *Dir = createDirective<OMPTargetTeamsGenericLoopDirective>(
       C, Clauses, AssociatedStmt,
       numLoopChildren(CollapsedNum, OMPD_target_teams_loop), StartLoc, EndLoc,
@@ -2473,6 +2473,7 @@ OMPTargetTeamsGenericLoopDirective *OMPTargetTeamsGenericLoopDirective::Create(
   Dir->setCombinedNextUpperBound(Exprs.DistCombinedFields.NUB);
   Dir->setCombinedDistCond(Exprs.DistCombinedFields.DistCond);
   Dir->setCombinedParForInDistCond(Exprs.DistCombinedFields.ParForInDistCond);
+  Dir->setCanBeParallelFor(CanBeParallelFor);
   return Dir;
 }
 

diff --git a/clang/lib/AST/StmtProfile.cpp b/clang/lib/AST/StmtProfile.cpp
@@ -2456,7 +2456,12 @@ class OpenACCClauseProfiler
       Visit(Clause);
     }
   }
+  void VisitOpenACCDefaultClause(const OpenACCDefaultClause &Clause);
 };
+
+/// Nothing to do here, there are no sub-statements.
+void OpenACCClauseProfiler::VisitOpenACCDefaultClause(
+    const OpenACCDefaultClause &Clause) {}
 } // namespace
 
 void StmtProfiler::VisitOpenACCComputeConstruct(

diff --git a/clang/lib/AST/TextNodeDumper.cpp b/clang/lib/AST/TextNodeDumper.cpp
@@ -390,6 +390,17 @@ void TextNodeDumper::Visit(const OpenACCClause *C) {
   {
     ColorScope Color(OS, ShowColors, AttrColor);
     OS << C->getClauseKind();
+
+    // Handle clauses with parens for types that have no children, likely
+    // because there is no sub expression.
+    switch (C->getClauseKind()) {
+    case OpenACCClauseKind::Default:
+      OS << '(' << cast<OpenACCDefaultClause>(C)->getDefaultClauseKind() << ')';
+      break;
+    default:
+      // Nothing to do here.
+      break;
+    }
   }
   dumpPointer(C);
   dumpSourceRange(SourceRange(C->getBeginLoc(), C->getEndLoc()));

diff --git a/clang/lib/AST/TypePrinter.cpp b/clang/lib/AST/TypePrinter.cpp
@@ -1213,10 +1213,13 @@ void TypePrinter::printDecltypeBefore(const DecltypeType *T, raw_ostream &OS) {
 
 void TypePrinter::printPackIndexingBefore(const PackIndexingType *T,
                                           raw_ostream &OS) {
-  if (T->hasSelectedType())
+  if (T->hasSelectedType()) {
     OS << T->getSelectedType();
-  else
-    OS << T->getPattern() << "...[" << T->getIndexExpr() << "]";
+  } else {
+    OS << T->getPattern() << "...[";
+    T->getIndexExpr()->printPretty(OS, nullptr, Policy);
+    OS << "]";
+  }
   spaceBeforePlaceHolder(OS);
 }
 

diff --git a/clang/lib/Basic/Targets/AArch64.cpp b/clang/lib/Basic/Targets/AArch64.cpp
@@ -1543,10 +1543,13 @@ WindowsARM64TargetInfo::getBuiltinVaListKind() const {
 TargetInfo::CallingConvCheckResult
 WindowsARM64TargetInfo::checkCallingConvention(CallingConv CC) const {
   switch (CC) {
+  case CC_X86VectorCall:
+    if (getTriple().isWindowsArm64EC())
+      return CCCR_OK;
+    return CCCR_Ignore;
   case CC_X86StdCall:
   case CC_X86ThisCall:
   case CC_X86FastCall:
-  case CC_X86VectorCall:
     return CCCR_Ignore;
   case CC_C:
   case CC_OpenCLKernel:

diff --git a/clang/lib/CodeGen/CGCall.cpp b/clang/lib/CodeGen/CGCall.cpp
@@ -5591,6 +5591,12 @@ RValue CodeGenFunction::EmitCall(const CGFunctionInfo &CallInfo,
                              /*AttrOnCallSite=*/true,
                              /*IsThunk=*/false);
 
+  if (CallingConv == llvm::CallingConv::X86_VectorCall &&
+      getTarget().getTriple().isWindowsArm64EC()) {
+    CGM.Error(Loc, "__vectorcall calling convention is not currently "
+                   "supported");
+  }
+
   if (const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(CurFuncDecl)) {
     if (FD->hasAttr<StrictFPAttr>())
       // All calls within a strictfp function are marked strictfp

diff --git a/clang/lib/CodeGen/CGCleanup.cpp b/clang/lib/CodeGen/CGCleanup.cpp
@@ -667,7 +667,8 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
 
   // - whether there's a fallthrough
   llvm::BasicBlock *FallthroughSource = Builder.GetInsertBlock();
-  bool HasFallthrough = (FallthroughSource != nullptr && IsActive);
+  bool HasFallthrough =
+      FallthroughSource != nullptr && (IsActive || HasExistingBranches);
 
   // Branch-through fall-throughs leave the insertion point set to the
   // end of the last cleanup, which points to the current scope.  The
@@ -692,7 +693,11 @@ 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 && !IsActive) {
+  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);
     llvm::BasicBlock *prebranchDest;
 
     // If the prebranch is semantically branching through the next
@@ -765,6 +770,7 @@ 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,
@@ -781,6 +787,7 @@ void CodeGenFunction::PopCleanupBlock(bool FallthroughIsBranchThrough) {
       }
 
       destroyOptimisticNormalEntry(*this, Scope);
+      Scope.MarkEmitted();
       EHStack.popCleanup();
 
       EmitCleanup(*this, Fn, cleanupFlags, NormalActiveFlag);
@@ -916,6 +923,7 @@ 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
@@ -16,8 +16,11 @@
 #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;
@@ -266,6 +269,51 @@ 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.
@@ -298,7 +346,7 @@ class alignas(8) EHCleanupScope : public EHScope {
                  EHScopeStack::stable_iterator enclosingEH)
       : EHScope(EHScope::Cleanup, enclosingEH),
         EnclosingNormal(enclosingNormal), NormalBlock(nullptr),
-        ActiveFlag(Address::invalid()), ExtInfo(nullptr),
+        ActiveFlag(Address::invalid()), ExtInfo(nullptr), AuxAllocas(nullptr),
         FixupDepth(fixupDepth) {
     CleanupBits.IsNormalCleanup = isNormal;
     CleanupBits.IsEHCleanup = isEH;
@@ -312,8 +360,15 @@ 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
@@ -19,6 +19,7 @@
 #include "CodeGenFunction.h"
 #include "CodeGenModule.h"
 #include "ConstantEmitter.h"
+#include "EHScopeStack.h"
 #include "PatternInit.h"
 #include "TargetInfo.h"
 #include "clang/AST/ASTContext.h"
@@ -2201,6 +2202,24 @@ 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);
 }
@@ -2217,16 +2236,19 @@ 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.
@@ -2241,13 +2263,12 @@ void CodeGenFunction::pushLifetimeExtendedDestroy(CleanupKind cleanupKind,
   Address ActiveFlag = createCleanupActiveFlag();
   SavedType SavedAddr = saveValueInCond(addr);
 
-  if (cleanupKind & EHCleanup) {
-    EHStack.pushCleanup<ConditionalCleanupType>(
-        static_cast<CleanupKind>(cleanupKind & ~NormalCleanup), SavedAddr, type,
-        destroyer, useEHCleanupForArray);
-    initFullExprCleanupWithFlag(ActiveFlag);
-  }
+  pushCleanupAndDeferDeactivation<ConditionalCleanupType>(
+      cleanupKind, 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);
@@ -2442,9 +2463,9 @@ namespace {
   };
 } // end anonymous namespace
 
-/// pushIrregularPartialArrayCleanup - Push an EH cleanup to destroy
-/// already-constructed elements of the given array.  The cleanup
-/// may be popped with DeactivateCleanupBlock or PopCleanupBlock.
+/// pushIrregularPartialArrayCleanup - Push a NormalAndEHCleanup 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
@@ -2453,10 +2474,9 @@ void CodeGenFunction::pushIrregularPartialArrayCleanup(llvm::Value *arrayBegin,
                                                        QualType elementType,
                                                        CharUnits elementAlign,
                                                        Destroyer *destroyer) {
-  pushFullExprCleanup<IrregularPartialArrayDestroy>(EHCleanup,
-                                                    arrayBegin, arrayEndPointer,
-                                                    elementType, elementAlign,
-                                                    destroyer);
+  pushFullExprCleanup<IrregularPartialArrayDestroy>(
+      NormalAndEHCleanup, 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
@@ -115,10 +115,16 @@ 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)
-    return Builder.CreateAlloca(Ty, ArraySize, Name);
-  return new llvm::AllocaInst(Ty, CGM.getDataLayout().getAllocaAddrSpace(),
-                              ArraySize, Name, AllocaInsertPt);
+    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;
 }
 
 /// CreateDefaultAlignTempAlloca - This creates an alloca with the

diff --git a/clang/lib/CodeGen/CGExprAgg.cpp b/clang/lib/CodeGen/CGExprAgg.cpp
@@ -15,6 +15,7 @@
 #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"
@@ -24,6 +25,7 @@
 #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;
@@ -558,24 +560,27 @@ 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();
-  EHScopeStack::stable_iterator cleanup;
-  llvm::Instruction *cleanupDominator = nullptr;
-  if (CGF.needsEHCleanup(dtorKind)) {
+  CodeGenFunction::CleanupDeactivationScope deactivation(CGF);
+
+  if (dtorKind) {
+    CodeGenFunction::AllocaTrackerRAII allocaTracker(CGF);
     // 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");
-    cleanupDominator = Builder.CreateStore(begin, endOfInit);
+    Builder.CreateStore(begin, endOfInit);
     CGF.pushIrregularPartialArrayCleanup(begin, endOfInit, elementType,
                                          elementAlign,
                                          CGF.getDestroyer(dtorKind));
-    cleanup = CGF.EHStack.stable_begin();
+    cast<EHCleanupScope>(*CGF.EHStack.find(CGF.EHStack.stable_begin()))
+        .AddAuxAllocas(allocaTracker.Take());
 
-  // Otherwise, remember that we didn't need a cleanup.
-  } else {
-    dtorKind = QualType::DK_none;
+    CGF.DeferredDeactivationCleanupStack.push_back(
+        {CGF.EHStack.stable_begin(), dominatingIP});
   }
 
   llvm::Value *one = llvm::ConstantInt::get(CGF.SizeTy, 1);
@@ -671,9 +676,6 @@ 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);
 }
 
 //===----------------------------------------------------------------------===//
@@ -1374,9 +1376,8 @@ 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.
-  SmallVector<EHScopeStack::stable_iterator, 16> Cleanups;
-  llvm::Instruction *CleanupDominator = nullptr;
+  // initializers throws an exception or contains branch out of the expressions.
+  CodeGenFunction::CleanupDeactivationScope scope(CGF);
 
   CXXRecordDecl::field_iterator CurField = E->getLambdaClass()->field_begin();
   for (LambdaExpr::const_capture_init_iterator i = E->capture_init_begin(),
@@ -1395,28 +1396,12 @@ AggExprEmitter::VisitLambdaExpr(LambdaExpr *E) {
     if (QualType::DestructionKind DtorKind =
             CurField->getType().isDestructedType()) {
       assert(LV.isSimple());
-      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());
-      }
+      if (DtorKind)
+        CGF.pushDestroyAndDeferDeactivation(
+            NormalAndEHCleanup, LV.getAddress(CGF), CurField->getType(),
+            CGF.getDestroyer(DtorKind), false);
     }
   }
-
-  // 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) {
@@ -1705,14 +1690,7 @@ 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;
-  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());
-  };
+  CodeGenFunction::CleanupDeactivationScope DeactivateCleanups(CGF);
 
   unsigned curInitIndex = 0;
 
@@ -1735,10 +1713,8 @@ void AggExprEmitter::VisitCXXParenListOrInitListExpr(
       CGF.EmitAggExpr(InitExprs[curInitIndex++], AggSlot);
 
       if (QualType::DestructionKind dtorKind =
-              Base.getType().isDestructedType()) {
-        CGF.pushDestroy(dtorKind, V, Base.getType());
-        addCleanup(CGF.EHStack.stable_begin());
-      }
+              Base.getType().isDestructedType())
+        CGF.pushDestroyAndDeferDeactivation(dtorKind, V, Base.getType());
     }
   }
 
@@ -1813,10 +1789,10 @@ void AggExprEmitter::VisitCXXParenListOrInitListExpr(
     if (QualType::DestructionKind dtorKind
           = field->getType().isDestructedType()) {
       assert(LV.isSimple());
-      if (CGF.needsEHCleanup(dtorKind)) {
-        CGF.pushDestroy(EHCleanup, LV.getAddress(CGF), field->getType(),
-                        CGF.getDestroyer(dtorKind), false);
-        addCleanup(CGF.EHStack.stable_begin());
+      if (dtorKind) {
+        CGF.pushDestroyAndDeferDeactivation(
+            NormalAndEHCleanup, LV.getAddress(CGF), field->getType(),
+            CGF.getDestroyer(dtorKind), false);
         pushedCleanup = true;
       }
     }
@@ -1829,17 +1805,6 @@ 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();
-  EHScopeStack::stable_iterator Cleanup;
-  llvm::Instruction *CleanupDominator = nullptr;
+  CleanupDeactivationScope deactivation(*this);
+  bool pushedCleanup = false;
 
   CharUnits ElementSize = getContext().getTypeSizeInChars(ElementType);
   CharUnits ElementAlign =
@@ -1105,19 +1105,24 @@ void CodeGenFunction::EmitNewArrayInitializer(
     }
 
     // Enter a partial-destruction Cleanup if necessary.
-    if (needsEHCleanup(DtorKind)) {
+    if (DtorKind) {
+      AllocaTrackerRAII AllocaTracker(*this);
       // 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));
-      Cleanup = EHStack.stable_begin();
+      cast<EHCleanupScope>(*EHStack.find(EHStack.stable_begin()))
+          .AddAuxAllocas(AllocaTracker.Take());
+      DeferredDeactivationCleanupStack.push_back(
+          {EHStack.stable_begin(), DominatingIP});
+      pushedCleanup = true;
     }
 
     CharUnits StartAlign = CurPtr.getAlignment();
@@ -1164,9 +1169,6 @@ 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;
   }
 
@@ -1281,24 +1283,22 @@ void CodeGenFunction::EmitNewArrayInitializer(
     Builder.CreateStore(CurPtr.emitRawPointer(*this), EndOfInit);
 
   // Enter a partial-destruction Cleanup if necessary.
-  if (!CleanupDominator && needsEHCleanup(DtorKind)) {
-    llvm::Value *BeginPtrRaw = BeginPtr.emitRawPointer(*this);
-    llvm::Value *CurPtrRaw = CurPtr.emitRawPointer(*this);
-    pushRegularPartialArrayCleanup(BeginPtrRaw, CurPtrRaw, ElementType,
+  if (!pushedCleanup && needsEHCleanup(DtorKind)) {
+    llvm::Instruction *DominatingIP =
+        Builder.CreateFlagLoad(llvm::ConstantInt::getNullValue(Int8PtrTy));
+    pushRegularPartialArrayCleanup(BeginPtr.emitRawPointer(*this),
+                                   CurPtr.emitRawPointer(*this), ElementType,
                                    ElementAlign, getDestroyer(DtorKind));
-    Cleanup = EHStack.stable_begin();
-    CleanupDominator = Builder.CreateUnreachable();
+    DeferredDeactivationCleanupStack.push_back(
+        {EHStack.stable_begin(), DominatingIP});
   }
 
   // Emit the initializer into this element.
   StoreAnyExprIntoOneUnit(*this, Init, Init->getType(), CurPtr,
                           AggValueSlot::DoesNotOverlap);
 
   // Leave the Cleanup if we entered one.
-  if (CleanupDominator) {
-    DeactivateCleanupBlock(Cleanup, CleanupDominator);
-    CleanupDominator->eraseFromParent();
-  }
+  deactivation.ForceDeactivate();
 
   // Advance to the next element by adjusting the pointer type as necessary.
   llvm::Value *NextPtr = Builder.CreateConstInBoundsGEP1_32(

diff --git a/clang/lib/CodeGen/CGOpenMPRuntime.cpp b/clang/lib/CodeGen/CGOpenMPRuntime.cpp
@@ -2656,11 +2656,12 @@ void CGOpenMPRuntime::emitForStaticFinish(CodeGenFunction &CGF,
   // Call __kmpc_for_static_fini(ident_t *loc, kmp_int32 tid);
   llvm::Value *Args[] = {
       emitUpdateLocation(CGF, Loc,
-                         isOpenMPDistributeDirective(DKind)
+                         isOpenMPDistributeDirective(DKind) ||
+                                 (DKind == OMPD_target_teams_loop)
                              ? OMP_IDENT_WORK_DISTRIBUTE
-                             : isOpenMPLoopDirective(DKind)
-                                   ? OMP_IDENT_WORK_LOOP
-                                   : OMP_IDENT_WORK_SECTIONS),
+                         : isOpenMPLoopDirective(DKind)
+                             ? OMP_IDENT_WORK_LOOP
+                             : OMP_IDENT_WORK_SECTIONS),
       getThreadID(CGF, Loc)};
   auto DL = ApplyDebugLocation::CreateDefaultArtificial(CGF, Loc);
   if (isOpenMPDistributeDirective(DKind) &&
@@ -8885,7 +8886,8 @@ getNestedDistributeDirective(ASTContext &Ctx, const OMPExecutableDirective &D) {
     OpenMPDirectiveKind DKind = NestedDir->getDirectiveKind();
     switch (D.getDirectiveKind()) {
     case OMPD_target:
-      // For now, just treat 'target teams loop' as if it's distributed.
+      // For now, treat 'target' with nested 'teams loop' as if it's
+      // distributed (target teams distribute).
       if (isOpenMPDistributeDirective(DKind) || DKind == OMPD_teams_loop)
         return NestedDir;
       if (DKind == OMPD_teams) {
@@ -9369,7 +9371,8 @@ llvm::Value *CGOpenMPRuntime::emitTargetNumIterationsCall(
         SizeEmitter) {
   OpenMPDirectiveKind Kind = D.getDirectiveKind();
   const OMPExecutableDirective *TD = &D;
-  // Get nested teams distribute kind directive, if any.
+  // Get nested teams distribute kind directive, if any. For now, treat
+  // 'target_teams_loop' as if it's really a target_teams_distribute.
   if ((!isOpenMPDistributeDirective(Kind) || !isOpenMPTeamsDirective(Kind)) &&
       Kind != OMPD_target_teams_loop)
     TD = getNestedDistributeDirective(CGM.getContext(), D);

diff --git a/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp b/clang/lib/CodeGen/CGOpenMPRuntimeGPU.cpp
@@ -646,7 +646,6 @@ static bool supportsSPMDExecutionMode(ASTContext &Ctx,
   case OMPD_target:
   case OMPD_target_teams:
     return hasNestedSPMDDirective(Ctx, D);
-  case OMPD_target_teams_loop:
   case OMPD_target_parallel_loop:
   case OMPD_target_parallel:
   case OMPD_target_parallel_for:
@@ -658,6 +657,12 @@ static bool supportsSPMDExecutionMode(ASTContext &Ctx,
     return true;
   case OMPD_target_teams_distribute:
     return false;
+  case OMPD_target_teams_loop:
+    // Whether this is true or not depends on how the directive will
+    // eventually be emitted.
+    if (auto *TTLD = dyn_cast<OMPTargetTeamsGenericLoopDirective>(&D))
+      return TTLD->canBeParallelFor();
+    return false;
   case OMPD_parallel:
   case OMPD_for:
   case OMPD_parallel_for:

diff --git a/clang/lib/CodeGen/CGStmtOpenMP.cpp b/clang/lib/CodeGen/CGStmtOpenMP.cpp
@@ -24,6 +24,7 @@
 #include "clang/AST/StmtVisitor.h"
 #include "clang/Basic/OpenMPKinds.h"
 #include "clang/Basic/PrettyStackTrace.h"
+#include "clang/Basic/SourceManager.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/BinaryFormat/Dwarf.h"
 #include "llvm/Frontend/OpenMP/OMPConstants.h"
@@ -34,11 +35,14 @@
 #include "llvm/IR/IntrinsicInst.h"
 #include "llvm/IR/Metadata.h"
 #include "llvm/Support/AtomicOrdering.h"
+#include "llvm/Support/Debug.h"
 #include <optional>
 using namespace clang;
 using namespace CodeGen;
 using namespace llvm::omp;
 
+#define TTL_CODEGEN_TYPE "target-teams-loop-codegen"
+
 static const VarDecl *getBaseDecl(const Expr *Ref);
 
 namespace {
@@ -1432,9 +1436,12 @@ void CodeGenFunction::EmitOMPReductionClauseFinal(
           *this, D.getBeginLoc(),
           isOpenMPWorksharingDirective(D.getDirectiveKind()));
     }
+    bool TeamsLoopCanBeParallel = false;
+    if (auto *TTLD = dyn_cast<OMPTargetTeamsGenericLoopDirective>(&D))
+      TeamsLoopCanBeParallel = TTLD->canBeParallelFor();
     bool WithNowait = D.getSingleClause<OMPNowaitClause>() ||
                       isOpenMPParallelDirective(D.getDirectiveKind()) ||
-                      ReductionKind == OMPD_simd;
+                      TeamsLoopCanBeParallel || ReductionKind == OMPD_simd;
     bool SimpleReduction = ReductionKind == OMPD_simd;
     // Emit nowait reduction if nowait clause is present or directive is a
     // parallel directive (it always has implicit barrier).
@@ -7928,11 +7935,9 @@ void CodeGenFunction::EmitOMPParallelGenericLoopDirective(
 void CodeGenFunction::EmitOMPTeamsGenericLoopDirective(
     const OMPTeamsGenericLoopDirective &S) {
   // To be consistent with current behavior of 'target teams loop', emit
-  // 'teams loop' as if its constituent constructs are 'distribute,
-  // 'parallel, and 'for'.
+  // 'teams loop' as if its constituent constructs are 'teams' and 'distribute'.
   auto &&CodeGenDistribute = [&S](CodeGenFunction &CGF, PrePostActionTy &) {
-    CGF.EmitOMPDistributeLoop(S, emitInnerParallelForWhenCombined,
-                              S.getDistInc());
+    CGF.EmitOMPDistributeLoop(S, emitOMPLoopBodyWithStopPoint, S.getInc());
   };
 
   // Emit teams region as a standalone region.
@@ -7946,15 +7951,33 @@ void CodeGenFunction::EmitOMPTeamsGenericLoopDirective(
                                                     CodeGenDistribute);
     CGF.EmitOMPReductionClauseFinal(S, /*ReductionKind=*/OMPD_teams);
   };
-  emitCommonOMPTeamsDirective(*this, S, OMPD_distribute_parallel_for, CodeGen);
+  emitCommonOMPTeamsDirective(*this, S, OMPD_distribute, CodeGen);
   emitPostUpdateForReductionClause(*this, S,
                                    [](CodeGenFunction &) { return nullptr; });
 }
 
-static void
-emitTargetTeamsGenericLoopRegion(CodeGenFunction &CGF,
-                                 const OMPTargetTeamsGenericLoopDirective &S,
-                                 PrePostActionTy &Action) {
+static void emitTargetTeamsLoopCodegenStatus(CodeGenFunction &CGF,
+                                             std::string StatusMsg,
+                                             const OMPExecutableDirective &D) {
+#ifndef NDEBUG
+  bool IsDevice = CGF.CGM.getLangOpts().OpenMPIsTargetDevice;
+  if (IsDevice)
+    StatusMsg += ": DEVICE";
+  else
+    StatusMsg += ": HOST";
+  SourceLocation L = D.getBeginLoc();
+  auto &SM = CGF.getContext().getSourceManager();
+  PresumedLoc PLoc = SM.getPresumedLoc(L);
+  const char *FileName = PLoc.isValid() ? PLoc.getFilename() : nullptr;
+  unsigned LineNo =
+      PLoc.isValid() ? PLoc.getLine() : SM.getExpansionLineNumber(L);
+  llvm::dbgs() << StatusMsg << ": " << FileName << ": " << LineNo << "\n";
+#endif
+}
+
+static void emitTargetTeamsGenericLoopRegionAsParallel(
+    CodeGenFunction &CGF, PrePostActionTy &Action,
+    const OMPTargetTeamsGenericLoopDirective &S) {
   Action.Enter(CGF);
   // Emit 'teams loop' as if its constituent constructs are 'distribute,
   // 'parallel, and 'for'.
@@ -7974,19 +7997,50 @@ emitTargetTeamsGenericLoopRegion(CodeGenFunction &CGF,
         CGF, OMPD_distribute, CodeGenDistribute, /*HasCancel=*/false);
     CGF.EmitOMPReductionClauseFinal(S, /*ReductionKind=*/OMPD_teams);
   };
-
+  DEBUG_WITH_TYPE(TTL_CODEGEN_TYPE,
+                  emitTargetTeamsLoopCodegenStatus(
+                      CGF, TTL_CODEGEN_TYPE " as parallel for", S));
   emitCommonOMPTeamsDirective(CGF, S, OMPD_distribute_parallel_for,
                               CodeGenTeams);
   emitPostUpdateForReductionClause(CGF, S,
                                    [](CodeGenFunction &) { return nullptr; });
 }
 
-/// Emit combined directive 'target teams loop' as if its constituent
-/// constructs are 'target', 'teams', 'distribute', 'parallel', and 'for'.
+static void emitTargetTeamsGenericLoopRegionAsDistribute(
+    CodeGenFunction &CGF, PrePostActionTy &Action,
+    const OMPTargetTeamsGenericLoopDirective &S) {
+  Action.Enter(CGF);
+  // Emit 'teams loop' as if its constituent construct is 'distribute'.
+  auto &&CodeGenDistribute = [&S](CodeGenFunction &CGF, PrePostActionTy &) {
+    CGF.EmitOMPDistributeLoop(S, emitOMPLoopBodyWithStopPoint, S.getInc());
+  };
+
+  // Emit teams region as a standalone region.
+  auto &&CodeGen = [&S, &CodeGenDistribute](CodeGenFunction &CGF,
+                                            PrePostActionTy &Action) {
+    Action.Enter(CGF);
+    CodeGenFunction::OMPPrivateScope PrivateScope(CGF);
+    CGF.EmitOMPReductionClauseInit(S, PrivateScope);
+    (void)PrivateScope.Privatize();
+    CGF.CGM.getOpenMPRuntime().emitInlinedDirective(
+        CGF, OMPD_distribute, CodeGenDistribute, /*HasCancel=*/false);
+    CGF.EmitOMPReductionClauseFinal(S, /*ReductionKind=*/OMPD_teams);
+  };
+  DEBUG_WITH_TYPE(TTL_CODEGEN_TYPE,
+                  emitTargetTeamsLoopCodegenStatus(
+                      CGF, TTL_CODEGEN_TYPE " as distribute", S));
+  emitCommonOMPTeamsDirective(CGF, S, OMPD_distribute, CodeGen);
+  emitPostUpdateForReductionClause(CGF, S,
+                                   [](CodeGenFunction &) { return nullptr; });
+}
+
 void CodeGenFunction::EmitOMPTargetTeamsGenericLoopDirective(
     const OMPTargetTeamsGenericLoopDirective &S) {
   auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
-    emitTargetTeamsGenericLoopRegion(CGF, S, Action);
+    if (S.canBeParallelFor())
+      emitTargetTeamsGenericLoopRegionAsParallel(CGF, Action, S);
+    else
+      emitTargetTeamsGenericLoopRegionAsDistribute(CGF, Action, S);
   };
   emitCommonOMPTargetDirective(*this, S, CodeGen);
 }
@@ -7996,7 +8050,10 @@ void CodeGenFunction::EmitOMPTargetTeamsGenericLoopDeviceFunction(
     const OMPTargetTeamsGenericLoopDirective &S) {
   // Emit SPMD target parallel loop region as a standalone region.
   auto &&CodeGen = [&S](CodeGenFunction &CGF, PrePostActionTy &Action) {
-    emitTargetTeamsGenericLoopRegion(CGF, S, Action);
+    if (S.canBeParallelFor())
+      emitTargetTeamsGenericLoopRegionAsParallel(CGF, Action, S);
+    else
+      emitTargetTeamsGenericLoopRegionAsDistribute(CGF, Action, S);
   };
   llvm::Function *Fn;
   llvm::Constant *Addr;

diff --git a/clang/lib/CodeGen/CodeGenFunction.cpp b/clang/lib/CodeGen/CodeGenFunction.cpp
@@ -91,6 +91,8 @@ 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);
@@ -346,6 +348,10 @@ 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
@@ -39,6 +39,7 @@
 #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"
@@ -670,6 +671,51 @@ 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;
@@ -875,6 +921,19 @@ 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() {
@@ -926,6 +985,7 @@ class CodeGenFunction : public CodeGenTypeCache {
   class RunCleanupsScope {
     EHScopeStack::stable_iterator CleanupStackDepth, OldCleanupScopeDepth;
     size_t LifetimeExtendedCleanupStackSize;
+    CleanupDeactivationScope DeactivateCleanups;
     bool OldDidCallStackSave;
   protected:
     bool PerformCleanup;
@@ -940,8 +1000,7 @@ class CodeGenFunction : public CodeGenTypeCache {
   public:
     /// Enter a new cleanup scope.
     explicit RunCleanupsScope(CodeGenFunction &CGF)
-      : PerformCleanup(true), CGF(CGF)
-    {
+        : DeactivateCleanups(CGF), PerformCleanup(true), CGF(CGF) {
       CleanupStackDepth = CGF.EHStack.stable_begin();
       LifetimeExtendedCleanupStackSize =
           CGF.LifetimeExtendedCleanupStack.size();
@@ -971,6 +1030,7 @@ 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;
@@ -2160,6 +2220,11 @@ 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);
@@ -2698,6 +2763,33 @@ 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
@@ -2087,6 +2087,14 @@ void CodeGenModule::SetLLVMFunctionAttributes(GlobalDecl GD,
   llvm::AttributeList PAL;
   ConstructAttributeList(F->getName(), Info, GD, PAL, CallingConv,
                          /*AttrOnCallSite=*/false, IsThunk);
+  if (CallingConv == llvm::CallingConv::X86_VectorCall &&
+      getTarget().getTriple().isWindowsArm64EC()) {
+    SourceLocation Loc;
+    if (const Decl *D = GD.getDecl())
+      Loc = D->getLocation();
+
+    Error(Loc, "__vectorcall calling convention is not currently supported");
+  }
   F->setAttributes(PAL);
   F->setCallingConv(static_cast<llvm::CallingConv::ID>(CallingConv));
 }

diff --git a/clang/lib/CodeGen/Targets/X86.cpp b/clang/lib/CodeGen/Targets/X86.cpp
@@ -2106,8 +2106,11 @@ void X86_64ABIInfo::classify(QualType Ty, uint64_t OffsetBase, Class &Lo,
         postMerge(Size, Lo, Hi);
         return;
       }
+
+      bool IsInMemory =
+          Offset % getContext().getTypeAlign(i->getType().getCanonicalType());
       // Note, skip this test for bit-fields, see below.
-      if (!BitField && Offset % getContext().getTypeAlign(i->getType())) {
+      if (!BitField && IsInMemory) {
         Lo = Memory;
         postMerge(Size, Lo, Hi);
         return;

diff --git a/clang/lib/Driver/ToolChains/AMDGPU.cpp b/clang/lib/Driver/ToolChains/AMDGPU.cpp
@@ -670,6 +670,10 @@ void amdgpu::getAMDGPUTargetFeatures(const Driver &D,
                    options::OPT_mno_wavefrontsize64, false))
     Features.push_back("+wavefrontsize64");
 
+  if (Args.hasFlag(options::OPT_mamdgpu_precise_memory_op,
+                   options::OPT_mno_amdgpu_precise_memory_op, false))
+    Features.push_back("+precise-memory");
+
   handleTargetFeaturesGroup(D, Triple, Args, Features,
                             options::OPT_m_amdgpu_Features_Group);
 }

diff --git a/clang/lib/Driver/ToolChains/Flang.cpp b/clang/lib/Driver/ToolChains/Flang.cpp
@@ -264,7 +264,7 @@ static void addVSDefines(const ToolChain &TC, const ArgList &Args,
   CmdArgs.push_back(Args.MakeArgString("-D_MSC_FULL_VER=" + Twine(ver)));
   CmdArgs.push_back(Args.MakeArgString("-D_WIN32"));
 
-  llvm::Triple triple = TC.getTriple();
+  const llvm::Triple &triple = TC.getTriple();
   if (triple.isAArch64()) {
     CmdArgs.push_back("-D_M_ARM64=1");
   } else if (triple.isX86() && triple.isArch32Bit()) {
@@ -589,7 +589,7 @@ static void addFloatingPointOptions(const Driver &D, const ArgList &Args,
 
   if (!HonorINFs && !HonorNaNs && AssociativeMath && ReciprocalMath &&
       ApproxFunc && !SignedZeros &&
-      (FPContract == "fast" || FPContract == "")) {
+      (FPContract == "fast" || FPContract.empty())) {
     CmdArgs.push_back("-ffast-math");
     return;
   }
@@ -679,7 +679,10 @@ void Flang::ConstructJob(Compilation &C, const JobAction &JA,
   CmdArgs.push_back(Args.MakeArgString(TripleStr));
 
   if (isa<PreprocessJobAction>(JA)) {
-      CmdArgs.push_back("-E");
+    CmdArgs.push_back("-E");
+    if (Args.getLastArg(options::OPT_dM)) {
+      CmdArgs.push_back("-dM");
+    }
   } else if (isa<CompileJobAction>(JA) || isa<BackendJobAction>(JA)) {
     if (JA.getType() == types::TY_Nothing) {
       CmdArgs.push_back("-fsyntax-only");

diff --git a/clang/lib/Format/TokenAnnotator.cpp b/clang/lib/Format/TokenAnnotator.cpp
@@ -2354,7 +2354,8 @@ class AnnotatingParser {
     // Line.MightBeFunctionDecl can only be true after the parentheses of a
     // function declaration have been found. In this case, 'Current' is a
     // trailing token of this declaration and thus cannot be a name.
-    if (Current.is(Keywords.kw_instanceof)) {
+    if ((Style.isJavaScript() || Style.Language == FormatStyle::LK_Java) &&
+        Current.is(Keywords.kw_instanceof)) {
       Current.setType(TT_BinaryOperator);
     } else if (isStartOfName(Current) &&
                (!Line.MightBeFunctionDecl || Current.NestingLevel != 0)) {

diff --git a/clang/lib/Format/UnwrappedLineParser.cpp b/clang/lib/Format/UnwrappedLineParser.cpp
@@ -538,16 +538,6 @@ void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) {
         if (Style.Language == FormatStyle::LK_Proto) {
           ProbablyBracedList = NextTok->isOneOf(tok::comma, tok::r_square);
         } else {
-          // Skip NextTok over preprocessor lines, otherwise we may not
-          // properly diagnose the block as a braced intializer
-          // if the comma separator appears after the pp directive.
-          while (NextTok->is(tok::hash)) {
-            ScopedMacroState MacroState(*Line, Tokens, NextTok);
-            do {
-              NextTok = Tokens->getNextToken();
-            } while (NextTok->isNot(tok::eof));
-          }
-
           // Using OriginalColumn to distinguish between ObjC methods and
           // binary operators is a bit hacky.
           bool NextIsObjCMethod = NextTok->isOneOf(tok::plus, tok::minus) &&
@@ -606,6 +596,16 @@ void UnwrappedLineParser::calculateBraceTypes(bool ExpectClassBody) {
             NextTok = Tokens->getNextToken();
             ProbablyBracedList = NextTok->isNot(tok::l_square);
           }
+
+          // Cpp macro definition body that is a nonempty braced list or block:
+          if (IsCpp && Line->InMacroBody && PrevTok != FormatTok &&
+              !FormatTok->Previous && NextTok->is(tok::eof) &&
+              // A statement can end with only `;` (simple statement), a block
+              // closing brace (compound statement), or `:` (label statement).
+              // If PrevTok is a block opening brace, Tok ends an empty block.
+              !PrevTok->isOneOf(tok::semi, BK_Block, tok::colon)) {
+            ProbablyBracedList = true;
+          }
         }
         if (ProbablyBracedList) {
           Tok->setBlockKind(BK_BracedInit);

diff --git a/clang/lib/Headers/intrin.h b/clang/lib/Headers/intrin.h
@@ -26,7 +26,7 @@
 #include <armintr.h>
 #endif
 
-#if defined(__aarch64__)
+#if defined(__aarch64__) || defined(__arm64ec__)
 #include <arm64intr.h>
 #endif
 

diff --git a/clang/lib/Parse/ParseExpr.cpp b/clang/lib/Parse/ParseExpr.cpp
@@ -30,6 +30,7 @@
 #include "clang/Sema/EnterExpressionEvaluationContext.h"
 #include "clang/Sema/ParsedTemplate.h"
 #include "clang/Sema/Scope.h"
+#include "clang/Sema/SemaSYCL.h"
 #include "clang/Sema/TypoCorrection.h"
 #include "llvm/ADT/SmallVector.h"
 #include <optional>
@@ -2490,8 +2491,8 @@ ExprResult Parser::ParseSYCLUniqueStableNameExpression() {
   if (T.consumeClose())
     return ExprError();
 
-  return Actions.ActOnSYCLUniqueStableNameExpr(OpLoc, T.getOpenLocation(),
-                                               T.getCloseLocation(), Ty.get());
+  return Actions.SYCL().ActOnUniqueStableNameExpr(
+      OpLoc, T.getOpenLocation(), T.getCloseLocation(), Ty.get());
 }
 
 /// Parse a sizeof or alignof expression.

diff --git a/clang/lib/Parse/ParseOpenACC.cpp b/clang/lib/Parse/ParseOpenACC.cpp
@@ -831,16 +831,19 @@ Parser::OpenACCClauseParseResult Parser::ParseOpenACCClauseParams(
 
       ConsumeToken();
 
-      if (getOpenACCDefaultClauseKind(DefKindTok) ==
-          OpenACCDefaultClauseKind::Invalid)
+      OpenACCDefaultClauseKind DefKind =
+          getOpenACCDefaultClauseKind(DefKindTok);
+
+      if (DefKind == OpenACCDefaultClauseKind::Invalid)
         Diag(DefKindTok, diag::err_acc_invalid_default_clause_kind);
+      else
+        ParsedClause.setDefaultDetails(DefKind);
 
       break;
     }
     case OpenACCClauseKind::If: {
       ExprResult CondExpr = ParseOpenACCConditionalExpr(*this);
-      // An invalid expression can be just about anything, so just give up on
-      // this clause list.
+
       if (CondExpr.isInvalid()) {
         Parens.skipToEnd();
         return OpenACCCanContinue();
@@ -962,8 +965,7 @@ Parser::OpenACCClauseParseResult Parser::ParseOpenACCClauseParams(
       case OpenACCClauseKind::Self: {
         assert(DirKind != OpenACCDirectiveKind::Update);
         ExprResult CondExpr = ParseOpenACCConditionalExpr(*this);
-        // An invalid expression can be just about anything, so just give up on
-        // this clause list.
+
         if (CondExpr.isInvalid()) {
           Parens.skipToEnd();
           return OpenACCCanContinue();

diff --git a/clang/lib/Sema/Sema.cpp b/clang/lib/Sema/Sema.cpp
@@ -45,6 +45,7 @@
 #include "clang/Sema/SemaHLSL.h"
 #include "clang/Sema/SemaInternal.h"
 #include "clang/Sema/SemaOpenACC.h"
+#include "clang/Sema/SemaSYCL.h"
 #include "clang/Sema/TemplateDeduction.h"
 #include "clang/Sema/TemplateInstCallback.h"
 #include "clang/Sema/TypoCorrection.h"
@@ -201,6 +202,7 @@ Sema::Sema(Preprocessor &pp, ASTContext &ctxt, ASTConsumer &consumer,
       CurScope(nullptr), Ident_super(nullptr),
       HLSLPtr(std::make_unique<SemaHLSL>(*this)),
       OpenACCPtr(std::make_unique<SemaOpenACC>(*this)),
+      SYCLPtr(std::make_unique<SemaSYCL>(*this)),
       MSPointerToMemberRepresentationMethod(
           LangOpts.getMSPointerToMemberRepresentationMethod()),
       MSStructPragmaOn(false), VtorDispStack(LangOpts.getVtorDispMode()),
@@ -1903,7 +1905,7 @@ Sema::targetDiag(SourceLocation Loc, unsigned DiagID, const FunctionDecl *FD) {
                                       : CUDADiagIfHostCode(Loc, DiagID);
 
   if (getLangOpts().SYCLIsDevice)
-    return SYCLDiagIfDeviceCode(Loc, DiagID);
+    return SYCL().DiagIfDeviceCode(Loc, DiagID);
 
   return SemaDiagnosticBuilder(SemaDiagnosticBuilder::K_Immediate, Loc, DiagID,
                                FD, *this);
@@ -1919,7 +1921,7 @@ void Sema::checkTypeSupport(QualType Ty, SourceLocation Loc, ValueDecl *D) {
   // constant byte size like zero length arrays. So, do a deep check for SYCL.
   if (D && LangOpts.SYCLIsDevice) {
     llvm::DenseSet<QualType> Visited;
-    deepTypeCheckForSYCLDevice(Loc, Visited, D);
+    SYCL().deepTypeCheckForDevice(Loc, Visited, D);
   }
 
   Decl *C = cast<Decl>(getCurLexicalContext());

diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
@@ -7938,6 +7938,7 @@ void Sema::checkCall(NamedDecl *FDecl, const FunctionProtoType *Proto,
     // For variadic functions, we may have more args than parameters.
     // For some K&R functions, we may have less args than parameters.
     const auto N = std::min<unsigned>(Proto->getNumParams(), Args.size());
+    bool AnyScalableArgsOrRet = Proto->getReturnType()->isSizelessVectorType();
     for (unsigned ArgIdx = 0; ArgIdx < N; ++ArgIdx) {
       // Args[ArgIdx] can be null in malformed code.
       if (const Expr *Arg = Args[ArgIdx]) {
@@ -7951,6 +7952,8 @@ void Sema::checkCall(NamedDecl *FDecl, const FunctionProtoType *Proto,
           checkAIXMemberAlignment((Arg->getExprLoc()), Arg);
 
         QualType ParamTy = Proto->getParamType(ArgIdx);
+        if (ParamTy->isSizelessVectorType())
+          AnyScalableArgsOrRet = true;
         QualType ArgTy = Arg->getType();
         CheckArgAlignment(Arg->getExprLoc(), FDecl, std::to_string(ArgIdx + 1),
                           ArgTy, ParamTy);
@@ -7971,6 +7974,23 @@ void Sema::checkCall(NamedDecl *FDecl, const FunctionProtoType *Proto,
       }
     }
 
+    // If the call requires a streaming-mode change and has scalable vector
+    // arguments or return values, then warn the user that the streaming and
+    // non-streaming vector lengths may be different.
+    const auto *CallerFD = dyn_cast<FunctionDecl>(CurContext);
+    if (CallerFD && (!FD || !FD->getBuiltinID()) && AnyScalableArgsOrRet) {
+      bool IsCalleeStreaming =
+          ExtInfo.AArch64SMEAttributes & FunctionType::SME_PStateSMEnabledMask;
+      bool IsCalleeStreamingCompatible =
+          ExtInfo.AArch64SMEAttributes &
+          FunctionType::SME_PStateSMCompatibleMask;
+      ArmStreamingType CallerFnType = getArmStreamingFnType(CallerFD);
+      if (!IsCalleeStreamingCompatible &&
+          (CallerFnType == ArmStreamingCompatible ||
+           ((CallerFnType == ArmStreaming) ^ IsCalleeStreaming)))
+        Diag(Loc, diag::warn_sme_streaming_pass_return_vl_to_non_streaming);
+    }
+
     FunctionType::ArmStateValue CalleeArmZAState =
         FunctionType::getArmZAState(ExtInfo.AArch64SMEAttributes);
     FunctionType::ArmStateValue CalleeArmZT0State =
@@ -7979,7 +7999,7 @@ void Sema::checkCall(NamedDecl *FDecl, const FunctionProtoType *Proto,
         CalleeArmZT0State != FunctionType::ARM_None) {
       bool CallerHasZAState = false;
       bool CallerHasZT0State = false;
-      if (const auto *CallerFD = dyn_cast<FunctionDecl>(CurContext)) {
+      if (CallerFD) {
         auto *Attr = CallerFD->getAttr<ArmNewAttr>();
         if (Attr && Attr->isNewZA())
           CallerHasZAState = true;

diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
@@ -12395,12 +12395,22 @@ bool Sema::CheckFunctionDeclaration(Scope *S, FunctionDecl *NewFD,
   }
 
   // Check if the function definition uses any AArch64 SME features without
-  // having the '+sme' feature enabled.
+  // having the '+sme' feature enabled and warn user if sme locally streaming
+  // function returns or uses arguments with VL-based types.
   if (DeclIsDefn) {
     const auto *Attr = NewFD->getAttr<ArmNewAttr>();
     bool UsesSM = NewFD->hasAttr<ArmLocallyStreamingAttr>();
     bool UsesZA = Attr && Attr->isNewZA();
     bool UsesZT0 = Attr && Attr->isNewZT0();
+
+    if (NewFD->hasAttr<ArmLocallyStreamingAttr>()) {
+      if (NewFD->getReturnType()->isSizelessVectorType() ||
+          llvm::any_of(NewFD->parameters(), [](ParmVarDecl *P) {
+            return P->getOriginalType()->isSizelessVectorType();
+          }))
+        Diag(NewFD->getLocation(),
+             diag::warn_sme_locally_streaming_has_vl_args_returns);
+    }
     if (const auto *FPT = NewFD->getType()->getAs<FunctionProtoType>()) {
       FunctionProtoType::ExtProtoInfo EPI = FPT->getExtProtoInfo();
       UsesSM |=

diff --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
@@ -3442,11 +3442,10 @@ static bool ShouldLookupResultBeMultiVersionOverload(const LookupResult &R) {
 
 ExprResult Sema::BuildDeclarationNameExpr(const CXXScopeSpec &SS,
                                           LookupResult &R, bool NeedsADL,
-                                          bool AcceptInvalidDecl,
-                                          bool NeedUnresolved) {
+                                          bool AcceptInvalidDecl) {
   // If this is a single, fully-resolved result and we don't need ADL,
   // just build an ordinary singleton decl ref.
-  if (!NeedUnresolved && !NeedsADL && R.isSingleResult() &&
+  if (!NeedsADL && R.isSingleResult() &&
       !R.getAsSingle<FunctionTemplateDecl>() &&
       !ShouldLookupResultBeMultiVersionOverload(R))
     return BuildDeclarationNameExpr(SS, R.getLookupNameInfo(), R.getFoundDecl(),
@@ -3795,28 +3794,6 @@ ExprResult Sema::BuildPredefinedExpr(SourceLocation Loc,
                                 SL);
 }
 
-ExprResult Sema::BuildSYCLUniqueStableNameExpr(SourceLocation OpLoc,
-                                               SourceLocation LParen,
-                                               SourceLocation RParen,
-                                               TypeSourceInfo *TSI) {
-  return SYCLUniqueStableNameExpr::Create(Context, OpLoc, LParen, RParen, TSI);
-}
-
-ExprResult Sema::ActOnSYCLUniqueStableNameExpr(SourceLocation OpLoc,
-                                               SourceLocation LParen,
-                                               SourceLocation RParen,
-                                               ParsedType ParsedTy) {
-  TypeSourceInfo *TSI = nullptr;
-  QualType Ty = GetTypeFromParser(ParsedTy, &TSI);
-
-  if (Ty.isNull())
-    return ExprError();
-  if (!TSI)
-    TSI = Context.getTrivialTypeSourceInfo(Ty, LParen);
-
-  return BuildSYCLUniqueStableNameExpr(OpLoc, LParen, RParen, TSI);
-}
-
 ExprResult Sema::ActOnPredefinedExpr(SourceLocation Loc, tok::TokenKind Kind) {
   return BuildPredefinedExpr(Loc, getPredefinedExprKind(Kind));
 }