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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

diff --git a/clang/lib/Index/USRGeneration.cpp b/clang/lib/Index/USRGeneration.cpp
@@ -267,10 +267,13 @@ void USRGenerator::VisitFunctionDecl(const FunctionDecl *D) {
     Out << '>';
   }
 
+  QualType CanonicalType = D->getType().getCanonicalType();
   // Mangle in type information for the arguments.
-  for (auto *PD : D->parameters()) {
-    Out << '#';
-    VisitType(PD->getType());
+  if (const auto *FPT = CanonicalType->getAs<FunctionProtoType>()) {
+    for (QualType PT : FPT->param_types()) {
+      Out << '#';
+      VisitType(PT);
+    }
   }
   if (D->isVariadic())
     Out << '.';

diff --git a/clang/lib/Parse/ParseCXXInlineMethods.cpp b/clang/lib/Parse/ParseCXXInlineMethods.cpp
@@ -20,6 +20,49 @@
 
 using namespace clang;
 
+/// Parse the optional ("message") part of a deleted-function-body.
+StringLiteral *Parser::ParseCXXDeletedFunctionMessage() {
+  if (!Tok.is(tok::l_paren))
+    return nullptr;
+  StringLiteral *Message = nullptr;
+  BalancedDelimiterTracker BT{*this, tok::l_paren};
+  BT.consumeOpen();
+
+  if (isTokenStringLiteral()) {
+    ExprResult Res = ParseUnevaluatedStringLiteralExpression();
+    if (Res.isUsable()) {
+      Message = Res.getAs<StringLiteral>();
+      Diag(Message->getBeginLoc(), getLangOpts().CPlusPlus26
+                                       ? diag::warn_cxx23_delete_with_message
+                                       : diag::ext_delete_with_message)
+          << Message->getSourceRange();
+    }
+  } else {
+    Diag(Tok.getLocation(), diag::err_expected_string_literal)
+        << /*Source='in'*/ 0 << "'delete'";
+    SkipUntil(tok::r_paren, StopAtSemi | StopBeforeMatch);
+  }
+
+  BT.consumeClose();
+  return Message;
+}
+
+/// If we've encountered '= delete' in a context where it is ill-formed, such
+/// as in the declaration of a non-function, also skip the ("message") part if
+/// it is present to avoid issuing further diagnostics.
+void Parser::SkipDeletedFunctionBody() {
+  if (!Tok.is(tok::l_paren))
+    return;
+
+  BalancedDelimiterTracker BT{*this, tok::l_paren};
+  BT.consumeOpen();
+
+  // Just skip to the end of the current declaration.
+  SkipUntil(tok::r_paren, tok::comma, StopAtSemi | StopBeforeMatch);
+  if (Tok.is(tok::r_paren))
+    BT.consumeClose();
+}
+
 /// ParseCXXInlineMethodDef - We parsed and verified that the specified
 /// Declarator is a well formed C++ inline method definition. Now lex its body
 /// and store its tokens for parsing after the C++ class is complete.
@@ -70,7 +113,8 @@ NamedDecl *Parser::ParseCXXInlineMethodDef(
                       ? diag::warn_cxx98_compat_defaulted_deleted_function
                       : diag::ext_defaulted_deleted_function)
         << 1 /* deleted */;
-      Actions.SetDeclDeleted(FnD, KWLoc);
+      StringLiteral *Message = ParseCXXDeletedFunctionMessage();
+      Actions.SetDeclDeleted(FnD, KWLoc, Message);
       Delete = true;
       if (auto *DeclAsFunction = dyn_cast<FunctionDecl>(FnD)) {
         DeclAsFunction->setRangeEnd(KWEndLoc);

diff --git a/clang/lib/Parse/ParseDecl.cpp b/clang/lib/Parse/ParseDecl.cpp
@@ -26,7 +26,9 @@
 #include "clang/Sema/Lookup.h"
 #include "clang/Sema/ParsedTemplate.h"
 #include "clang/Sema/Scope.h"
+#include "clang/Sema/SemaCUDA.h"
 #include "clang/Sema/SemaDiagnostic.h"
+#include "clang/Sema/SemaOpenMP.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/ADT/SmallString.h"
 #include "llvm/ADT/StringSwitch.h"
@@ -2379,14 +2381,10 @@ Parser::DeclGroupPtrTy Parser::ParseDeclGroup(ParsingDeclSpec &DS,
       if (getLangOpts().CPlusPlus23) {
         auto &LastRecord = Actions.ExprEvalContexts.back();
         LastRecord.InLifetimeExtendingContext = true;
-
-        // Materialize non-`cv void` prvalue temporaries in discarded
-        // expressions. These materialized temporaries may be lifetime-extented.
-        LastRecord.InMaterializeTemporaryObjectContext = true;
       }
 
       if (getLangOpts().OpenMP)
-        Actions.startOpenMPCXXRangeFor();
+        Actions.OpenMP().startOpenMPCXXRangeFor();
       if (Tok.is(tok::l_brace))
         FRI->RangeExpr = ParseBraceInitializer();
       else
@@ -2664,7 +2662,8 @@ Decl *Parser::ParseDeclarationAfterDeclaratorAndAttributes(
     }
   }
 
-  Sema::CUDATargetContextRAII X(Actions, Sema::CTCK_InitGlobalVar, ThisDecl);
+  SemaCUDA::CUDATargetContextRAII X(Actions.CUDA(),
+                                    SemaCUDA::CTCK_InitGlobalVar, ThisDecl);
   switch (TheInitKind) {
   // Parse declarator '=' initializer.
   case InitKind::Equal: {
@@ -2676,6 +2675,7 @@ Decl *Parser::ParseDeclarationAfterDeclaratorAndAttributes(
           << 1 /* delete */;
       else
         Diag(ConsumeToken(), diag::err_deleted_non_function);
+      SkipDeletedFunctionBody();
     } else if (Tok.is(tok::kw_default)) {
       if (D.isFunctionDeclarator())
         Diag(ConsumeToken(), diag::err_default_delete_in_multiple_declaration)

diff --git a/clang/lib/Parse/ParseDeclCXX.cpp b/clang/lib/Parse/ParseDeclCXX.cpp
@@ -3397,6 +3397,7 @@ ExprResult Parser::ParseCXXMemberInitializer(Decl *D, bool IsFunction,
               << 1 /* delete */;
         else
           Diag(ConsumeToken(), diag::err_deleted_non_function);
+        SkipDeletedFunctionBody();
         return ExprError();
       }
     } else if (Tok.is(tok::kw_default)) {

diff --git a/clang/lib/Parse/ParseExpr.cpp b/clang/lib/Parse/ParseExpr.cpp
@@ -30,6 +30,8 @@
 #include "clang/Sema/EnterExpressionEvaluationContext.h"
 #include "clang/Sema/ParsedTemplate.h"
 #include "clang/Sema/Scope.h"
+#include "clang/Sema/SemaCUDA.h"
+#include "clang/Sema/SemaOpenMP.h"
 #include "clang/Sema/SemaSYCL.h"
 #include "clang/Sema/TypoCorrection.h"
 #include "llvm/ADT/SmallVector.h"
@@ -2074,7 +2076,7 @@ Parser::ParsePostfixExpressionSuffix(ExprResult LHS) {
           // replace this call to ActOnOpenACCArraySectionExpr in the future.
           // Eventually we'll genericize the OPenMPArraySectionExpr type as
           // well.
-          LHS = Actions.ActOnOMPArraySectionExpr(
+          LHS = Actions.OpenMP().ActOnOMPArraySectionExpr(
               LHS.get(), Loc, ArgExprs.empty() ? nullptr : ArgExprs[0],
               ColonLocFirst, ColonLocSecond, Length.get(), Stride.get(), RLoc);
         } else {
@@ -2129,10 +2131,8 @@ Parser::ParsePostfixExpressionSuffix(ExprResult LHS) {
         }
 
         if (!LHS.isInvalid()) {
-          ExprResult ECResult = Actions.ActOnCUDAExecConfigExpr(getCurScope(),
-                                    OpenLoc,
-                                    ExecConfigExprs,
-                                    CloseLoc);
+          ExprResult ECResult = Actions.CUDA().ActOnExecConfigExpr(
+              getCurScope(), OpenLoc, ExecConfigExprs, CloseLoc);
           if (ECResult.isInvalid())
             LHS = ExprError();
           else
@@ -3278,7 +3278,7 @@ Parser::ParseParenExpression(ParenParseOption &ExprType, bool stopIfCastExpr,
     if (ErrorFound) {
       Result = ExprError();
     } else if (!Result.isInvalid()) {
-      Result = Actions.ActOnOMPArrayShapingExpr(
+      Result = Actions.OpenMP().ActOnOMPArrayShapingExpr(
           Result.get(), OpenLoc, RParenLoc, OMPDimensions, OMPBracketsRanges);
     }
     return Result;

diff --git a/clang/lib/Parse/ParseOpenACC.cpp b/clang/lib/Parse/ParseOpenACC.cpp
@@ -835,19 +835,23 @@ Parser::OpenACCClauseParseResult Parser::ParseOpenACCClauseParams(
     case OpenACCClauseKind::Default: {
       Token DefKindTok = getCurToken();
 
-      if (expectIdentifierOrKeyword(*this))
-        break;
+      if (expectIdentifierOrKeyword(*this)) {
+        Parens.skipToEnd();
+        return OpenACCCanContinue();
+      }
 
       ConsumeToken();
 
       OpenACCDefaultClauseKind DefKind =
           getOpenACCDefaultClauseKind(DefKindTok);
 
-      if (DefKind == OpenACCDefaultClauseKind::Invalid)
+      if (DefKind == OpenACCDefaultClauseKind::Invalid) {
         Diag(DefKindTok, diag::err_acc_invalid_default_clause_kind);
-      else
-        ParsedClause.setDefaultDetails(DefKind);
+        Parens.skipToEnd();
+        return OpenACCCanContinue();
+      }
 
+      ParsedClause.setDefaultDetails(DefKind);
       break;
     }
     case OpenACCClauseKind::If: {
@@ -977,6 +981,8 @@ Parser::OpenACCClauseParseResult Parser::ParseOpenACCClauseParams(
       case OpenACCClauseKind::Self: {
         assert(DirKind != OpenACCDirectiveKind::Update);
         ExprResult CondExpr = ParseOpenACCConditionExpr();
+        ParsedClause.setConditionDetails(CondExpr.isUsable() ? CondExpr.get()
+                                                             : nullptr);
 
         if (CondExpr.isInvalid()) {
           Parens.skipToEnd();

diff --git a/clang/lib/Parse/ParseOpenMP.cpp b/clang/lib/Parse/ParseOpenMP.cpp
diff --git a/clang/lib/Parse/ParsePragma.cpp b/clang/lib/Parse/ParsePragma.cpp
@@ -21,6 +21,7 @@
 #include "clang/Parse/RAIIObjectsForParser.h"
 #include "clang/Sema/EnterExpressionEvaluationContext.h"
 #include "clang/Sema/Scope.h"
+#include "clang/Sema/SemaCUDA.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/StringSwitch.h"
 #include <optional>
@@ -3900,8 +3901,8 @@ void PragmaForceCUDAHostDeviceHandler::HandlePragma(
   }
 
   if (Info->isStr("begin"))
-    Actions.PushForceCUDAHostDevice();
-  else if (!Actions.PopForceCUDAHostDevice())
+    Actions.CUDA().PushForceHostDevice();
+  else if (!Actions.CUDA().PopForceHostDevice())
     PP.Diag(FirstTok.getLocation(),
             diag::err_pragma_cannot_end_force_cuda_host_device);
 

diff --git a/clang/lib/Parse/ParseStmt.cpp b/clang/lib/Parse/ParseStmt.cpp
@@ -22,6 +22,7 @@
 #include "clang/Sema/DeclSpec.h"
 #include "clang/Sema/EnterExpressionEvaluationContext.h"
 #include "clang/Sema/Scope.h"
+#include "clang/Sema/SemaOpenMP.h"
 #include "clang/Sema/TypoCorrection.h"
 #include "llvm/ADT/STLExtras.h"
 #include <optional>
@@ -2301,7 +2302,7 @@ StmtResult Parser::ParseForStatement(SourceLocation *TrailingElseLoc) {
     // In OpenMP loop region loop control variable must be captured and be
     // private. Perform analysis of first part (if any).
     if (getLangOpts().OpenMP && FirstPart.isUsable()) {
-      Actions.ActOnOpenMPLoopInitialization(ForLoc, FirstPart.get());
+      Actions.OpenMP().ActOnOpenMPLoopInitialization(ForLoc, FirstPart.get());
     }
   }
 

diff --git a/clang/lib/Parse/Parser.cpp b/clang/lib/Parse/Parser.cpp
@@ -1404,6 +1404,7 @@ Decl *Parser::ParseFunctionDefinition(ParsingDeclarator &D,
 
   // Parse function body eagerly if it is either '= delete;' or '= default;' as
   // ActOnStartOfFunctionDef needs to know whether the function is deleted.
+  StringLiteral *DeletedMessage = nullptr;
   Sema::FnBodyKind BodyKind = Sema::FnBodyKind::Other;
   SourceLocation KWLoc;
   if (TryConsumeToken(tok::equal)) {
@@ -1415,6 +1416,7 @@ Decl *Parser::ParseFunctionDefinition(ParsingDeclarator &D,
                       : diag::ext_defaulted_deleted_function)
           << 1 /* deleted */;
       BodyKind = Sema::FnBodyKind::Delete;
+      DeletedMessage = ParseCXXDeletedFunctionMessage();
     } else if (TryConsumeToken(tok::kw_default, KWLoc)) {
       Diag(KWLoc, getLangOpts().CPlusPlus11
                       ? diag::warn_cxx98_compat_defaulted_deleted_function
@@ -1473,7 +1475,7 @@ Decl *Parser::ParseFunctionDefinition(ParsingDeclarator &D,
   D.getMutableDeclSpec().abort();
 
   if (BodyKind != Sema::FnBodyKind::Other) {
-    Actions.SetFunctionBodyKind(Res, KWLoc, BodyKind);
+    Actions.SetFunctionBodyKind(Res, KWLoc, BodyKind, DeletedMessage);
     Stmt *GeneratedBody = Res ? Res->getBody() : nullptr;
     Actions.ActOnFinishFunctionBody(Res, GeneratedBody, false);
     return Res;

diff --git a/clang/lib/Sema/Sema.cpp b/clang/lib/Sema/Sema.cpp
@@ -41,10 +41,12 @@
 #include "clang/Sema/RISCVIntrinsicManager.h"
 #include "clang/Sema/Scope.h"
 #include "clang/Sema/ScopeInfo.h"
+#include "clang/Sema/SemaCUDA.h"
 #include "clang/Sema/SemaConsumer.h"
 #include "clang/Sema/SemaHLSL.h"
 #include "clang/Sema/SemaInternal.h"
 #include "clang/Sema/SemaOpenACC.h"
+#include "clang/Sema/SemaOpenMP.h"
 #include "clang/Sema/SemaSYCL.h"
 #include "clang/Sema/TemplateDeduction.h"
 #include "clang/Sema/TemplateInstCallback.h"
@@ -199,8 +201,10 @@ Sema::Sema(Preprocessor &pp, ASTContext &ctxt, ASTConsumer &consumer,
       LateTemplateParser(nullptr), LateTemplateParserCleanup(nullptr),
       OpaqueParser(nullptr), CurContext(nullptr), ExternalSource(nullptr),
       CurScope(nullptr), Ident_super(nullptr),
+      CUDAPtr(std::make_unique<SemaCUDA>(*this)),
       HLSLPtr(std::make_unique<SemaHLSL>(*this)),
       OpenACCPtr(std::make_unique<SemaOpenACC>(*this)),
+      OpenMPPtr(std::make_unique<SemaOpenMP>(*this)),
       SYCLPtr(std::make_unique<SemaSYCL>(*this)),
       MSPointerToMemberRepresentationMethod(
           LangOpts.getMSPointerToMemberRepresentationMethod()),
@@ -224,8 +228,7 @@ Sema::Sema(Preprocessor &pp, ASTContext &ctxt, ASTConsumer &consumer,
       StringWithUTF8StringMethod(nullptr),
       ValueWithBytesObjCTypeMethod(nullptr), NSArrayDecl(nullptr),
       ArrayWithObjectsMethod(nullptr), NSDictionaryDecl(nullptr),
-      DictionaryWithObjectsMethod(nullptr), CodeCompleter(CodeCompleter),
-      VarDataSharingAttributesStack(nullptr) {
+      DictionaryWithObjectsMethod(nullptr), CodeCompleter(CodeCompleter) {
   assert(pp.TUKind == TUKind);
   TUScope = nullptr;
 
@@ -250,7 +253,7 @@ Sema::Sema(Preprocessor &pp, ASTContext &ctxt, ASTConsumer &consumer,
       nullptr, ExpressionEvaluationContextRecord::EK_Other);
 
   // Initialization of data sharing attributes stack for OpenMP
-  InitDataSharingAttributesStack();
+  OpenMP().InitDataSharingAttributesStack();
 
   std::unique_ptr<sema::SemaPPCallbacks> Callbacks =
       std::make_unique<sema::SemaPPCallbacks>();
@@ -499,7 +502,7 @@ Sema::~Sema() {
   threadSafety::threadSafetyCleanup(ThreadSafetyDeclCache);
 
   // Destroys data sharing attributes stack for OpenMP
-  DestroyDataSharingAttributesStack();
+  OpenMP().DestroyDataSharingAttributesStack();
 
   // Detach from the PP callback handler which outlives Sema since it's owned
   // by the preprocessor.
@@ -1157,7 +1160,7 @@ void Sema::ActOnEndOfTranslationUnit() {
 
   DiagnoseUnterminatedPragmaAlignPack();
   DiagnoseUnterminatedPragmaAttribute();
-  DiagnoseUnterminatedOpenMPDeclareTarget();
+  OpenMP().DiagnoseUnterminatedOpenMPDeclareTarget();
 
   // All delayed member exception specs should be checked or we end up accepting
   // incompatible declarations.
@@ -1635,15 +1638,15 @@ bool Sema::hasUncompilableErrorOccurred() const {
 // Print notes showing how we can reach FD starting from an a priori
 // known-callable function.
 static void emitCallStackNotes(Sema &S, const FunctionDecl *FD) {
-  auto FnIt = S.DeviceKnownEmittedFns.find(FD);
-  while (FnIt != S.DeviceKnownEmittedFns.end()) {
+  auto FnIt = S.CUDA().DeviceKnownEmittedFns.find(FD);
+  while (FnIt != S.CUDA().DeviceKnownEmittedFns.end()) {
     // Respect error limit.
     if (S.Diags.hasFatalErrorOccurred())
       return;
     DiagnosticBuilder Builder(
         S.Diags.Report(FnIt->second.Loc, diag::note_called_by));
     Builder << FnIt->second.FD;
-    FnIt = S.DeviceKnownEmittedFns.find(FnIt->second.FD);
+    FnIt = S.CUDA().DeviceKnownEmittedFns.find(FnIt->second.FD);
   }
 }
 
@@ -1745,9 +1748,9 @@ class DeferredDiagnosticsEmitter
     // Finalize analysis of OpenMP-specific constructs.
     if (Caller && S.LangOpts.OpenMP && UsePath.size() == 1 &&
         (ShouldEmitRootNode || InOMPDeviceContext))
-      S.finalizeOpenMPDelayedAnalysis(Caller, FD, Loc);
+      S.OpenMP().finalizeOpenMPDelayedAnalysis(Caller, FD, Loc);
     if (Caller)
-      S.DeviceKnownEmittedFns[FD] = {Caller, Loc};
+      S.CUDA().DeviceKnownEmittedFns[FD] = {Caller, Loc};
     // Always emit deferred diagnostics for the direct users. This does not
     // lead to explosion of diagnostics since each user is visited at most
     // twice.
@@ -1836,8 +1839,8 @@ void Sema::emitDeferredDiags() {
 // which other not-known-emitted functions.
 //
 // When we see something which is illegal if the current function is emitted
-// (usually by way of CUDADiagIfDeviceCode, CUDADiagIfHostCode, or
-// CheckCUDACall), we first check if the current function is known-emitted.  If
+// (usually by way of DiagIfDeviceCode, DiagIfHostCode, or
+// CheckCall), we first check if the current function is known-emitted.  If
 // so, we immediately output the diagnostic.
 //
 // Otherwise, we "defer" the diagnostic.  It sits in Sema::DeviceDeferredDiags
@@ -1897,11 +1900,11 @@ Sema::targetDiag(SourceLocation Loc, unsigned DiagID, const FunctionDecl *FD) {
   FD = FD ? FD : getCurFunctionDecl();
   if (LangOpts.OpenMP)
     return LangOpts.OpenMPIsTargetDevice
-               ? diagIfOpenMPDeviceCode(Loc, DiagID, FD)
-               : diagIfOpenMPHostCode(Loc, DiagID, FD);
+               ? OpenMP().diagIfOpenMPDeviceCode(Loc, DiagID, FD)
+               : OpenMP().diagIfOpenMPHostCode(Loc, DiagID, FD);
   if (getLangOpts().CUDA)
-    return getLangOpts().CUDAIsDevice ? CUDADiagIfDeviceCode(Loc, DiagID)
-                                      : CUDADiagIfHostCode(Loc, DiagID);
+    return getLangOpts().CUDAIsDevice ? CUDA().DiagIfDeviceCode(Loc, DiagID)
+                                      : CUDA().DiagIfHostCode(Loc, DiagID);
 
   if (getLangOpts().SYCLIsDevice)
     return SYCL().DiagIfDeviceCode(Loc, DiagID);
@@ -2129,7 +2132,7 @@ void Sema::PushFunctionScope() {
     FunctionScopes.push_back(new FunctionScopeInfo(getDiagnostics()));
   }
   if (LangOpts.OpenMP)
-    pushOpenMPFunctionRegion();
+    OpenMP().pushOpenMPFunctionRegion();
 }
 
 void Sema::PushBlockScope(Scope *BlockScope, BlockDecl *Block) {
@@ -2249,7 +2252,7 @@ Sema::PopFunctionScopeInfo(const AnalysisBasedWarnings::Policy *WP,
                                PoppedFunctionScopeDeleter(this));
 
   if (LangOpts.OpenMP)
-    popOpenMPFunctionRegion(Scope.get());
+    OpenMP().popOpenMPFunctionRegion(Scope.get());
 
   // Issue any analysis-based warnings.
   if (WP && D)
@@ -2685,7 +2688,9 @@ void Sema::PushCapturedRegionScope(Scope *S, CapturedDecl *CD, RecordDecl *RD,
                                    unsigned OpenMPCaptureLevel) {
   auto *CSI = new CapturedRegionScopeInfo(
       getDiagnostics(), S, CD, RD, CD->getContextParam(), K,
-      (getLangOpts().OpenMP && K == CR_OpenMP) ? getOpenMPNestingLevel() : 0,
+      (getLangOpts().OpenMP && K == CR_OpenMP)
+          ? OpenMP().getOpenMPNestingLevel()
+          : 0,
       OpenMPCaptureLevel);
   CSI->ReturnType = Context.VoidTy;
   FunctionScopes.push_back(CSI);

diff --git a/clang/lib/Sema/SemaBase.cpp b/clang/lib/Sema/SemaBase.cpp
@@ -1,5 +1,6 @@
 #include "clang/Sema/SemaBase.h"
 #include "clang/Sema/Sema.h"
+#include "clang/Sema/SemaCUDA.h"
 
 namespace clang {
 
@@ -70,8 +71,8 @@ Sema::SemaDiagnosticBuilder SemaBase::Diag(SourceLocation Loc, unsigned DiagID,
   }
 
   SemaDiagnosticBuilder DB = getLangOpts().CUDAIsDevice
-                                 ? SemaRef.CUDADiagIfDeviceCode(Loc, DiagID)
-                                 : SemaRef.CUDADiagIfHostCode(Loc, DiagID);
+                                 ? SemaRef.CUDA().DiagIfDeviceCode(Loc, DiagID)
+                                 : SemaRef.CUDA().DiagIfHostCode(Loc, DiagID);
   SetIsLastErrorImmediate(DB.isImmediate());
   return DB;
 }

diff --git a/clang/lib/Sema/SemaCUDA.cpp b/clang/lib/Sema/SemaCUDA.cpp
diff --git a/clang/lib/Sema/SemaCast.cpp b/clang/lib/Sema/SemaCast.cpp
@@ -498,10 +498,22 @@ static bool tryDiagnoseOverloadedCast(Sema &S, CastType CT,
     howManyCandidates = OCD_AmbiguousCandidates;
     break;
 
-  case OR_Deleted:
-    msg = diag::err_ovl_deleted_conversion_in_cast;
-    howManyCandidates = OCD_ViableCandidates;
-    break;
+  case OR_Deleted: {
+    OverloadCandidateSet::iterator Best;
+    [[maybe_unused]] OverloadingResult Res =
+        candidates.BestViableFunction(S, range.getBegin(), Best);
+    assert(Res == OR_Deleted && "Inconsistent overload resolution");
+
+    StringLiteral *Msg = Best->Function->getDeletedMessage();
+    candidates.NoteCandidates(
+        PartialDiagnosticAt(range.getBegin(),
+                            S.PDiag(diag::err_ovl_deleted_conversion_in_cast)
+                                << CT << srcType << destType << (Msg != nullptr)
+                                << (Msg ? Msg->getString() : StringRef())
+                                << range << src->getSourceRange()),
+        S, OCD_ViableCandidates, src);
+    return true;
+  }
   }
 
   candidates.NoteCandidates(

diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
@@ -19721,6 +19721,27 @@ bool Sema::IsLayoutCompatible(QualType T1, QualType T2) const {
   return isLayoutCompatible(getASTContext(), T1, T2);
 }
 
+//===-------------- Pointer interconvertibility ----------------------------//
+
+bool Sema::IsPointerInterconvertibleBaseOf(const TypeSourceInfo *Base,
+                                           const TypeSourceInfo *Derived) {
+  QualType BaseT = Base->getType()->getCanonicalTypeUnqualified();
+  QualType DerivedT = Derived->getType()->getCanonicalTypeUnqualified();
+
+  if (BaseT->isStructureOrClassType() && DerivedT->isStructureOrClassType() &&
+      getASTContext().hasSameType(BaseT, DerivedT))
+    return true;
+
+  if (!IsDerivedFrom(Derived->getTypeLoc().getBeginLoc(), DerivedT, BaseT))
+    return false;
+
+  // Per [basic.compound]/4.3, containing object has to be standard-layout.
+  if (DerivedT->getAsCXXRecordDecl()->isStandardLayout())
+    return true;
+
+  return false;
+}
+
 //===--- CHECK: pointer_with_type_tag attribute: datatypes should match ----//
 
 /// Given a type tag expression find the type tag itself.

diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
@@ -45,8 +45,10 @@
 #include "clang/Sema/ParsedTemplate.h"
 #include "clang/Sema/Scope.h"
 #include "clang/Sema/ScopeInfo.h"
+#include "clang/Sema/SemaCUDA.h"
 #include "clang/Sema/SemaHLSL.h"
 #include "clang/Sema/SemaInternal.h"
+#include "clang/Sema/SemaOpenMP.h"
 #include "clang/Sema/Template.h"
 #include "llvm/ADT/STLForwardCompat.h"
 #include "llvm/ADT/SmallString.h"
@@ -6167,11 +6169,12 @@ Decl *Sema::ActOnDeclarator(Scope *S, Declarator &D) {
   // Check if we are in an `omp begin/end declare variant` scope. Handle this
   // declaration only if the `bind_to_declaration` extension is set.
   SmallVector<FunctionDecl *, 4> Bases;
-  if (LangOpts.OpenMP && isInOpenMPDeclareVariantScope())
-    if (getOMPTraitInfoForSurroundingScope()->isExtensionActive(llvm::omp::TraitProperty::
-              implementation_extension_bind_to_declaration))
-    ActOnStartOfFunctionDefinitionInOpenMPDeclareVariantScope(
-        S, D, MultiTemplateParamsArg(), Bases);
+  if (LangOpts.OpenMP && OpenMP().isInOpenMPDeclareVariantScope())
+    if (OpenMP().getOMPTraitInfoForSurroundingScope()->isExtensionActive(
+            llvm::omp::TraitProperty::
+                implementation_extension_bind_to_declaration))
+      OpenMP().ActOnStartOfFunctionDefinitionInOpenMPDeclareVariantScope(
+          S, D, MultiTemplateParamsArg(), Bases);
 
   Decl *Dcl = HandleDeclarator(S, D, MultiTemplateParamsArg());
 
@@ -6180,7 +6183,8 @@ Decl *Sema::ActOnDeclarator(Scope *S, Declarator &D) {
     Dcl->setTopLevelDeclInObjCContainer();
 
   if (!Bases.empty())
-    ActOnFinishedFunctionDefinitionInOpenMPDeclareVariantScope(Dcl, Bases);
+    OpenMP().ActOnFinishedFunctionDefinitionInOpenMPDeclareVariantScope(Dcl,
+                                                                        Bases);
 
   return Dcl;
 }
@@ -6567,8 +6571,8 @@ NamedDecl *Sema::HandleDeclarator(Scope *S, Declarator &D,
   if (New->getDeclName() && AddToScope)
     PushOnScopeChains(New, S);
 
-  if (isInOpenMPDeclareTargetContext())
-    checkDeclIsAllowedInOpenMPTarget(nullptr, New);
+  if (OpenMP().isInOpenMPDeclareTargetContext())
+    OpenMP().checkDeclIsAllowedInOpenMPTarget(nullptr, New);
 
   return New;
 }
@@ -10595,12 +10599,12 @@ Sema::ActOnFunctionDeclarator(Scope *S, Declarator &D, DeclContext *DC,
 
     // We do not add HD attributes to specializations here because
     // they may have different constexpr-ness compared to their
-    // templates and, after maybeAddCUDAHostDeviceAttrs() is applied,
+    // templates and, after maybeAddHostDeviceAttrs() is applied,
     // may end up with different effective targets. Instead, a
     // specialization inherits its target attributes from its template
     // in the CheckFunctionTemplateSpecialization() call below.
     if (getLangOpts().CUDA && !isFunctionTemplateSpecialization)
-      maybeAddCUDAHostDeviceAttrs(NewFD, Previous);
+      CUDA().maybeAddHostDeviceAttrs(NewFD, Previous);
 
     // Handle explict specializations of function templates
     // and friend function declarations with an explicit
@@ -10898,12 +10902,12 @@ Sema::ActOnFunctionDeclarator(Scope *S, Declarator &D, DeclContext *DC,
 
   if (getLangOpts().CUDA) {
     IdentifierInfo *II = NewFD->getIdentifier();
-    if (II && II->isStr(getCudaConfigureFuncName()) &&
+    if (II && II->isStr(CUDA().getConfigureFuncName()) &&
         !NewFD->isInvalidDecl() &&
         NewFD->getDeclContext()->getRedeclContext()->isTranslationUnit()) {
       if (!R->castAs<FunctionType>()->getReturnType()->isScalarType())
         Diag(NewFD->getLocation(), diag::err_config_scalar_return)
-            << getCudaConfigureFuncName();
+            << CUDA().getConfigureFuncName();
       Context.setcudaConfigureCallDecl(NewFD);
     }
 
@@ -12267,7 +12271,7 @@ bool Sema::CheckFunctionDeclaration(Scope *S, FunctionDecl *NewFD,
   }
 
   if (LangOpts.OpenMP)
-    ActOnFinishedFunctionDefinitionInOpenMPAssumeScope(NewFD);
+    OpenMP().ActOnFinishedFunctionDefinitionInOpenMPAssumeScope(NewFD);
 
   // Semantic checking for this function declaration (in isolation).
 
@@ -12398,7 +12402,7 @@ bool Sema::CheckFunctionDeclaration(Scope *S, FunctionDecl *NewFD,
     }
 
     if (!Redeclaration && LangOpts.CUDA)
-      checkCUDATargetOverload(NewFD, Previous);
+      CUDA().checkTargetOverload(NewFD, Previous);
   }
 
   // Check if the function definition uses any AArch64 SME features without
@@ -12667,7 +12671,7 @@ void Sema::CheckMSVCRTEntryPoint(FunctionDecl *FD) {
   }
 }
 
-bool Sema::CheckForConstantInitializer(Expr *Init, QualType DclT) {
+bool Sema::CheckForConstantInitializer(Expr *Init, unsigned DiagID) {
   // FIXME: Need strict checking.  In C89, we need to check for
   // any assignment, increment, decrement, function-calls, or
   // commas outside of a sizeof.  In C99, it's the same list,
@@ -12685,8 +12689,7 @@ bool Sema::CheckForConstantInitializer(Expr *Init, QualType DclT) {
   const Expr *Culprit;
   if (Init->isConstantInitializer(Context, false, &Culprit))
     return false;
-  Diag(Culprit->getExprLoc(), diag::err_init_element_not_constant)
-    << Culprit->getSourceRange();
+  Diag(Culprit->getExprLoc(), DiagID) << Culprit->getSourceRange();
   return true;
 }
 
@@ -13804,29 +13807,24 @@ void Sema::AddInitializerToDecl(Decl *RealDecl, Expr *Init, bool DirectInit) {
     // OpenCL v1.2 s6.5.3: __constant locals must be constant-initialized.
     // This is true even in C++ for OpenCL.
     } else if (VDecl->getType().getAddressSpace() == LangAS::opencl_constant) {
-      CheckForConstantInitializer(Init, DclT);
+      CheckForConstantInitializer(Init);
 
-    // Otherwise, C++ does not restrict the initializer.
+      // Otherwise, C++ does not restrict the initializer.
     } else if (getLangOpts().CPlusPlus) {
       // do nothing
 
     // C99 6.7.8p4: All the expressions in an initializer for an object that has
     // static storage duration shall be constant expressions or string literals.
     } else if (VDecl->getStorageClass() == SC_Static) {
-      CheckForConstantInitializer(Init, DclT);
+      CheckForConstantInitializer(Init);
 
-    // C89 is stricter than C99 for aggregate initializers.
-    // C89 6.5.7p3: All the expressions [...] in an initializer list
-    // for an object that has aggregate or union type shall be
-    // constant expressions.
+      // C89 is stricter than C99 for aggregate initializers.
+      // C89 6.5.7p3: All the expressions [...] in an initializer list
+      // for an object that has aggregate or union type shall be
+      // constant expressions.
     } else if (!getLangOpts().C99 && VDecl->getType()->isAggregateType() &&
                isa<InitListExpr>(Init)) {
-      const Expr *Culprit;
-      if (!Init->isConstantInitializer(Context, false, &Culprit)) {
-        Diag(Culprit->getExprLoc(),
-             diag::ext_aggregate_init_not_constant)
-          << Culprit->getSourceRange();
-      }
+      CheckForConstantInitializer(Init, diag::ext_aggregate_init_not_constant);
     }
 
     if (auto *E = dyn_cast<ExprWithCleanups>(Init))
@@ -13959,7 +13957,7 @@ void Sema::AddInitializerToDecl(Decl *RealDecl, Expr *Init, bool DirectInit) {
     // Avoid duplicate diagnostics for constexpr variables.
     if (!getLangOpts().CPlusPlus && !VDecl->isInvalidDecl() &&
         !VDecl->isConstexpr())
-      CheckForConstantInitializer(Init, DclT);
+      CheckForConstantInitializer(Init);
   }
 
   QualType InitType = Init->getType();
@@ -14415,7 +14413,7 @@ StmtResult Sema::ActOnCXXForRangeIdentifier(Scope *S, SourceLocation IdentLoc,
 void Sema::CheckCompleteVariableDeclaration(VarDecl *var) {
   if (var->isInvalidDecl()) return;
 
-  MaybeAddCUDAConstantAttr(var);
+  CUDA().MaybeAddConstantAttr(var);
 
   if (getLangOpts().OpenCL) {
     // OpenCL v2.0 s6.12.5 - Every block variable declaration must have an
@@ -14829,7 +14827,7 @@ void Sema::FinalizeDeclaration(Decl *ThisDecl) {
   // variables whether they are local or not. CUDA also allows
   // constant initializers for __constant__ and __device__ variables.
   if (getLangOpts().CUDA)
-    checkAllowedCUDAInitializer(VD);
+    CUDA().checkAllowedInitializer(VD);
 
   // Grab the dllimport or dllexport attribute off of the VarDecl.
   const InheritableAttr *DLLAttr = getDLLAttr(VD);
@@ -14955,7 +14953,7 @@ Sema::DeclGroupPtrTy Sema::FinalizeDeclaratorGroup(Scope *S, const DeclSpec &DS,
       if (auto *VD = dyn_cast<VarDecl>(D);
           LangOpts.OpenMP && VD && VD->hasAttr<OMPDeclareTargetDeclAttr>() &&
           VD->hasGlobalStorage())
-        ActOnOpenMPDeclareTargetInitializer(D);
+        OpenMP().ActOnOpenMPDeclareTargetInitializer(D);
       // For declarators, there are some additional syntactic-ish checks we need
       // to perform.
       if (auto *DD = dyn_cast<DeclaratorDecl>(D)) {
@@ -15494,16 +15492,17 @@ Sema::ActOnStartOfFunctionDef(Scope *FnBodyScope, Declarator &D,
   // specialization function under the OpenMP context defined as part of the
   // `omp begin declare variant`.
   SmallVector<FunctionDecl *, 4> Bases;
-  if (LangOpts.OpenMP && isInOpenMPDeclareVariantScope())
-    ActOnStartOfFunctionDefinitionInOpenMPDeclareVariantScope(
+  if (LangOpts.OpenMP && OpenMP().isInOpenMPDeclareVariantScope())
+    OpenMP().ActOnStartOfFunctionDefinitionInOpenMPDeclareVariantScope(
         ParentScope, D, TemplateParameterLists, Bases);
 
   D.setFunctionDefinitionKind(FunctionDefinitionKind::Definition);
   Decl *DP = HandleDeclarator(ParentScope, D, TemplateParameterLists);
   Decl *Dcl = ActOnStartOfFunctionDef(FnBodyScope, DP, SkipBody, BodyKind);
 
   if (!Bases.empty())
-    ActOnFinishedFunctionDefinitionInOpenMPDeclareVariantScope(Dcl, Bases);
+    OpenMP().ActOnFinishedFunctionDefinitionInOpenMPDeclareVariantScope(Dcl,
+                                                                        Bases);
 
   return Dcl;
 }
@@ -16082,7 +16081,17 @@ Decl *Sema::ActOnFinishFunctionBody(Decl *dcl, Stmt *Body,
     // This is meant to pop the context added in ActOnStartOfFunctionDef().
     ExitFunctionBodyRAII ExitRAII(*this, isLambdaCallOperator(FD));
     if (FD) {
-      FD->setBody(Body);
+      // If this is called by Parser::ParseFunctionDefinition() after marking
+      // the declaration as deleted, and if the deleted-function-body contains
+      // a message (C++26), then a DefaultedOrDeletedInfo will have already been
+      // added to store that message; do not overwrite it in that case.
+      //
+      // Since this would always set the body to 'nullptr' in that case anyway,
+      // which is already done when the function decl is initially created,
+      // always skipping this irrespective of whether there is a delete message
+      // should not be a problem.
+      if (!FD->isDeletedAsWritten())
+        FD->setBody(Body);
       FD->setWillHaveBody(false);
       CheckImmediateEscalatingFunctionDefinition(FD, FSI);
 
@@ -20640,7 +20649,7 @@ Sema::FunctionEmissionStatus Sema::getEmissionStatus(const FunctionDecl *FD,
         return FunctionEmissionStatus::OMPDiscarded;
     // If we have an explicit value for the device type, or we are in a target
     // declare context, we need to emit all extern and used symbols.
-    if (isInOpenMPDeclareTargetContext() || DevTy)
+    if (OpenMP().isInOpenMPDeclareTargetContext() || DevTy)
       if (IsEmittedForExternalSymbol())
         return FunctionEmissionStatus::Emitted;
     // Device mode only emits what it must, if it wasn't tagged yet and needed,
@@ -20666,7 +20675,7 @@ Sema::FunctionEmissionStatus Sema::getEmissionStatus(const FunctionDecl *FD,
     // when compiling for host, device and global functions are never emitted.
     // (Technically, we do emit a host-side stub for global functions, but this
     // doesn't count for our purposes here.)
-    CUDAFunctionTarget T = IdentifyCUDATarget(FD);
+    CUDAFunctionTarget T = CUDA().IdentifyTarget(FD);
     if (LangOpts.CUDAIsDevice && T == CUDAFunctionTarget::Host)
       return FunctionEmissionStatus::CUDADiscarded;
     if (!LangOpts.CUDAIsDevice &&
@@ -20691,5 +20700,5 @@ bool Sema::shouldIgnoreInHostDeviceCheck(FunctionDecl *Callee) {
   // for host, only HD functions actually called from the host get marked as
   // known-emitted.
   return LangOpts.CUDA && !LangOpts.CUDAIsDevice &&
-         IdentifyCUDATarget(Callee) == CUDAFunctionTarget::Global;
+         CUDA().IdentifyTarget(Callee) == CUDAFunctionTarget::Global;
 }
diff --git a/clang/lib/Sema/SemaDeclAttr.cpp b/clang/lib/Sema/SemaDeclAttr.cpp
@@ -39,6 +39,7 @@
 #include "clang/Sema/ParsedAttr.h"
 #include "clang/Sema/Scope.h"
 #include "clang/Sema/ScopeInfo.h"
+#include "clang/Sema/SemaCUDA.h"
 #include "clang/Sema/SemaHLSL.h"
 #include "clang/Sema/SemaInternal.h"
 #include "llvm/ADT/STLExtras.h"
@@ -5099,8 +5100,8 @@ static void handleSharedAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
     return;
   }
   if (S.getLangOpts().CUDA && VD->hasLocalStorage() &&
-      S.CUDADiagIfHostCode(AL.getLoc(), diag::err_cuda_host_shared)
-          << llvm::to_underlying(S.CurrentCUDATarget()))
+      S.CUDA().DiagIfHostCode(AL.getLoc(), diag::err_cuda_host_shared)
+          << llvm::to_underlying(S.CUDA().CurrentTarget()))
     return;
   D->addAttr(::new (S.Context) CUDASharedAttr(S.Context, AL));
 }
@@ -5189,8 +5190,9 @@ static void handleCallConvAttr(Sema &S, Decl *D, const ParsedAttr &AL) {
   // Diagnostic is emitted elsewhere: here we store the (valid) AL
   // in the Decl node for syntactic reasoning, e.g., pretty-printing.
   CallingConv CC;
-  if (S.CheckCallingConvAttr(AL, CC, /*FD*/ nullptr,
-                             S.IdentifyCUDATarget(dyn_cast<FunctionDecl>(D))))
+  if (S.CheckCallingConvAttr(
+          AL, CC, /*FD*/ nullptr,
+          S.CUDA().IdentifyTarget(dyn_cast<FunctionDecl>(D))))
     return;
 
   if (!isa<ObjCMethodDecl>(D)) {
@@ -5495,7 +5497,7 @@ bool Sema::CheckCallingConvAttr(const ParsedAttr &Attrs, CallingConv &CC,
   if (LangOpts.CUDA) {
     auto *Aux = Context.getAuxTargetInfo();
     assert(FD || CFT != CUDAFunctionTarget::InvalidTarget);
-    auto CudaTarget = FD ? IdentifyCUDATarget(FD) : CFT;
+    auto CudaTarget = FD ? CUDA().IdentifyTarget(FD) : CFT;
     bool CheckHost = false, CheckDevice = false;
     switch (CudaTarget) {
     case CUDAFunctionTarget::HostDevice:

diff --git a/clang/lib/Sema/SemaDeclCXX.cpp b/clang/lib/Sema/SemaDeclCXX.cpp
@@ -42,7 +42,9 @@
 #include "clang/Sema/ParsedTemplate.h"
 #include "clang/Sema/Scope.h"
 #include "clang/Sema/ScopeInfo.h"
+#include "clang/Sema/SemaCUDA.h"
 #include "clang/Sema/SemaInternal.h"
+#include "clang/Sema/SemaOpenMP.h"
 #include "clang/Sema/Template.h"
 #include "llvm/ADT/ArrayRef.h"
 #include "llvm/ADT/STLExtras.h"
@@ -961,8 +963,8 @@ Sema::ActOnDecompositionDeclarator(Scope *S, Declarator &D,
     CurContext->addHiddenDecl(New);
   }
 
-  if (isInOpenMPDeclareTargetContext())
-    checkDeclIsAllowedInOpenMPTarget(nullptr, New);
+  if (OpenMP().isInOpenMPDeclareTargetContext())
+    OpenMP().checkDeclIsAllowedInOpenMPTarget(nullptr, New);
 
   return New;
 }
@@ -7981,7 +7983,7 @@ class DefaultedComparisonVisitor {
   DefaultedComparisonVisitor(Sema &S, CXXRecordDecl *RD, FunctionDecl *FD,
                              DefaultedComparisonKind DCK)
       : S(S), RD(RD), FD(FD), DCK(DCK) {
-    if (auto *Info = FD->getDefaultedFunctionInfo()) {
+    if (auto *Info = FD->getDefalutedOrDeletedInfo()) {
       // FIXME: Change CreateOverloadedBinOp to take an ArrayRef instead of an
       // UnresolvedSet to avoid this copy.
       Fns.assign(Info->getUnqualifiedLookups().begin(),
@@ -8849,8 +8851,9 @@ bool Sema::CheckExplicitlyDefaultedComparison(Scope *S, FunctionDecl *FD,
     UnresolvedSet<32> Operators;
     lookupOperatorsForDefaultedComparison(*this, S, Operators,
                                           FD->getOverloadedOperator());
-    FD->setDefaultedFunctionInfo(FunctionDecl::DefaultedFunctionInfo::Create(
-        Context, Operators.pairs()));
+    FD->setDefaultedOrDeletedInfo(
+        FunctionDecl::DefaultedOrDeletedFunctionInfo::Create(
+            Context, Operators.pairs()));
   }
 
   // C++2a [class.compare.default]p1:
@@ -9876,15 +9879,15 @@ bool Sema::ShouldDeleteSpecialMember(CXXMethodDecl *MD,
     // failed.
     // For inherited constructors (non-null ICI), CSM may be passed so that MD
     // is treated as certain special member, which may not reflect what special
-    // member MD really is. However inferCUDATargetForImplicitSpecialMember
+    // member MD really is. However inferTargetForImplicitSpecialMember
     // expects CSM to match MD, therefore recalculate CSM.
     assert(ICI || CSM == getSpecialMember(MD));
     auto RealCSM = CSM;
     if (ICI)
       RealCSM = getSpecialMember(MD);
 
-    return inferCUDATargetForImplicitSpecialMember(RD, RealCSM, MD,
-                                                   SMI.ConstArg, Diagnose);
+    return CUDA().inferTargetForImplicitSpecialMember(RD, RealCSM, MD,
+                                                      SMI.ConstArg, Diagnose);
   }
 
   return false;
@@ -14055,7 +14058,7 @@ CXXConstructorDecl *Sema::DeclareImplicitDefaultConstructor(
   setupImplicitSpecialMemberType(DefaultCon, Context.VoidTy, std::nullopt);
 
   if (getLangOpts().CUDA)
-    inferCUDATargetForImplicitSpecialMember(
+    CUDA().inferTargetForImplicitSpecialMember(
         ClassDecl, CXXSpecialMemberKind::DefaultConstructor, DefaultCon,
         /* ConstRHS */ false,
         /* Diagnose */ false);
@@ -14341,7 +14344,7 @@ CXXDestructorDecl *Sema::DeclareImplicitDestructor(CXXRecordDecl *ClassDecl) {
   setupImplicitSpecialMemberType(Destructor, Context.VoidTy, std::nullopt);
 
   if (getLangOpts().CUDA)
-    inferCUDATargetForImplicitSpecialMember(
+    CUDA().inferTargetForImplicitSpecialMember(
         ClassDecl, CXXSpecialMemberKind::Destructor, Destructor,
         /* ConstRHS */ false,
         /* Diagnose */ false);
@@ -14983,7 +14986,7 @@ CXXMethodDecl *Sema::DeclareImplicitCopyAssignment(CXXRecordDecl *ClassDecl) {
   setupImplicitSpecialMemberType(CopyAssignment, RetType, ArgType);
 
   if (getLangOpts().CUDA)
-    inferCUDATargetForImplicitSpecialMember(
+    CUDA().inferTargetForImplicitSpecialMember(
         ClassDecl, CXXSpecialMemberKind::CopyAssignment, CopyAssignment,
         /* ConstRHS */ Const,
         /* Diagnose */ false);
@@ -15335,7 +15338,7 @@ CXXMethodDecl *Sema::DeclareImplicitMoveAssignment(CXXRecordDecl *ClassDecl) {
   setupImplicitSpecialMemberType(MoveAssignment, RetType, ArgType);
 
   if (getLangOpts().CUDA)
-    inferCUDATargetForImplicitSpecialMember(
+    CUDA().inferTargetForImplicitSpecialMember(
         ClassDecl, CXXSpecialMemberKind::MoveAssignment, MoveAssignment,
         /* ConstRHS */ false,
         /* Diagnose */ false);
@@ -15733,7 +15736,7 @@ CXXConstructorDecl *Sema::DeclareImplicitCopyConstructor(
   setupImplicitSpecialMemberType(CopyConstructor, Context.VoidTy, ArgType);
 
   if (getLangOpts().CUDA)
-    inferCUDATargetForImplicitSpecialMember(
+    CUDA().inferTargetForImplicitSpecialMember(
         ClassDecl, CXXSpecialMemberKind::CopyConstructor, CopyConstructor,
         /* ConstRHS */ Const,
         /* Diagnose */ false);
@@ -15878,7 +15881,7 @@ CXXConstructorDecl *Sema::DeclareImplicitMoveConstructor(
   setupImplicitSpecialMemberType(MoveConstructor, Context.VoidTy, ArgType);
 
   if (getLangOpts().CUDA)
-    inferCUDATargetForImplicitSpecialMember(
+    CUDA().inferTargetForImplicitSpecialMember(
         ClassDecl, CXXSpecialMemberKind::MoveConstructor, MoveConstructor,
         /* ConstRHS */ false,
         /* Diagnose */ false);
@@ -16184,7 +16187,7 @@ ExprResult Sema::BuildCXXConstructExpr(
              DeclInitType->getBaseElementTypeUnsafe()->getAsCXXRecordDecl()) &&
          "given constructor for wrong type");
   MarkFunctionReferenced(ConstructLoc, Constructor);
-  if (getLangOpts().CUDA && !CheckCUDACall(ConstructLoc, Constructor))
+  if (getLangOpts().CUDA && !CUDA().CheckCall(ConstructLoc, Constructor))
     return ExprError();
 
   return CheckForImmediateInvocation(
@@ -18158,7 +18161,8 @@ NamedDecl *Sema::ActOnFriendFunctionDecl(Scope *S, Declarator &D,
   return ND;
 }
 
-void Sema::SetDeclDeleted(Decl *Dcl, SourceLocation DelLoc) {
+void Sema::SetDeclDeleted(Decl *Dcl, SourceLocation DelLoc,
+                          StringLiteral *Message) {
   AdjustDeclIfTemplate(Dcl);
 
   FunctionDecl *Fn = dyn_cast_or_null<FunctionDecl>(Dcl);
@@ -18207,7 +18211,7 @@ void Sema::SetDeclDeleted(Decl *Dcl, SourceLocation DelLoc) {
   // C++11 [dcl.fct.def.delete]p4:
   //  A deleted function is implicitly inline.
   Fn->setImplicitlyInline();
-  Fn->setDeletedAsWritten();
+  Fn->setDeletedAsWritten(true, Message);
 }
 
 void Sema::SetDeclDefaulted(Decl *Dcl, SourceLocation DefaultLoc) {
@@ -18320,11 +18324,11 @@ void Sema::DiagnoseReturnInConstructorExceptionHandler(CXXTryStmt *TryBlock) {
   }
 }
 
-void Sema::SetFunctionBodyKind(Decl *D, SourceLocation Loc,
-                               FnBodyKind BodyKind) {
+void Sema::SetFunctionBodyKind(Decl *D, SourceLocation Loc, FnBodyKind BodyKind,
+                               StringLiteral *DeletedMessage) {
   switch (BodyKind) {
   case FnBodyKind::Delete:
-    SetDeclDeleted(D, Loc);
+    SetDeclDeleted(D, Loc, DeletedMessage);
     break;
   case FnBodyKind::Default:
     SetDeclDefaulted(D, Loc);
@@ -18651,8 +18655,8 @@ void Sema::MarkVTableUsed(SourceLocation Loc, CXXRecordDecl *Class,
   // Do not mark as used if compiling for the device outside of the target
   // region.
   if (TUKind != TU_Prefix && LangOpts.OpenMP && LangOpts.OpenMPIsTargetDevice &&
-      !isInOpenMPDeclareTargetContext() &&
-      !isInOpenMPTargetExecutionDirective()) {
+      !OpenMP().isInOpenMPDeclareTargetContext() &&
+      !OpenMP().isInOpenMPTargetExecutionDirective()) {
     if (!DefinitionRequired)
       MarkVirtualMembersReferenced(Loc, Class);
     return;

diff --git a/clang/lib/Sema/SemaExpr.cpp b/clang/lib/Sema/SemaExpr.cpp
diff --git a/clang/lib/Sema/SemaExprCXX.cpp b/clang/lib/Sema/SemaExprCXX.cpp
@@ -38,6 +38,7 @@
 #include "clang/Sema/ParsedTemplate.h"
 #include "clang/Sema/Scope.h"
 #include "clang/Sema/ScopeInfo.h"
+#include "clang/Sema/SemaCUDA.h"
 #include "clang/Sema/SemaInternal.h"
 #include "clang/Sema/SemaLambda.h"
 #include "clang/Sema/Template.h"
@@ -884,8 +885,8 @@ ExprResult Sema::BuildCXXThrow(SourceLocation OpLoc, Expr *Ex,
 
   // Exceptions aren't allowed in CUDA device code.
   if (getLangOpts().CUDA)
-    CUDADiagIfDeviceCode(OpLoc, diag::err_cuda_device_exceptions)
-        << "throw" << llvm::to_underlying(CurrentCUDATarget());
+    CUDA().DiagIfDeviceCode(OpLoc, diag::err_cuda_device_exceptions)
+        << "throw" << llvm::to_underlying(CUDA().CurrentTarget());
 
   if (getCurScope() && getCurScope()->isOpenMPSimdDirectiveScope())
     Diag(OpLoc, diag::err_omp_simd_region_cannot_use_stmt) << "throw";
@@ -1415,42 +1416,26 @@ bool Sema::CheckCXXThisCapture(SourceLocation Loc, const bool Explicit,
 }
 
 ExprResult Sema::ActOnCXXThis(SourceLocation Loc) {
-  // C++20 [expr.prim.this]p1:
-  //   The keyword this names a pointer to the object for which an
-  //   implicit object member function is invoked or a non-static
-  //   data member's initializer is evaluated.
+  /// C++ 9.3.2: In the body of a non-static member function, the keyword this
+  /// is a non-lvalue expression whose value is the address of the object for
+  /// which the function is called.
   QualType ThisTy = getCurrentThisType();
 
-  if (CheckCXXThisType(Loc, ThisTy))
-    return ExprError();
+  if (ThisTy.isNull()) {
+    DeclContext *DC = getFunctionLevelDeclContext();
 
-  return BuildCXXThisExpr(Loc, ThisTy, /*IsImplicit=*/false);
-}
+    if (const auto *Method = dyn_cast<CXXMethodDecl>(DC);
+        Method && Method->isExplicitObjectMemberFunction()) {
+      return Diag(Loc, diag::err_invalid_this_use) << 1;
+    }
 
-bool Sema::CheckCXXThisType(SourceLocation Loc, QualType Type) {
-  if (!Type.isNull())
-    return false;
+    if (isLambdaCallWithExplicitObjectParameter(CurContext))
+      return Diag(Loc, diag::err_invalid_this_use) << 1;
 
-  // C++20 [expr.prim.this]p3:
-  //   If a declaration declares a member function or member function template
-  //   of a class X, the expression this is a prvalue of type
-  //   "pointer to cv-qualifier-seq X" wherever X is the current class between
-  //   the optional cv-qualifier-seq and the end of the function-definition,
-  //   member-declarator, or declarator. It shall not appear within the
-  //   declaration of either a static member function or an explicit object
-  //   member function of the current class (although its type and value
-  //   category are defined within such member functions as they are within
-  //   an implicit object member function).
-  DeclContext *DC = getFunctionLevelDeclContext();
-  if (const auto *Method = dyn_cast<CXXMethodDecl>(DC);
-      Method && Method->isExplicitObjectMemberFunction()) {
-    Diag(Loc, diag::err_invalid_this_use) << 1;
-  } else if (isLambdaCallWithExplicitObjectParameter(CurContext)) {
-    Diag(Loc, diag::err_invalid_this_use) << 1;
-  } else {
-    Diag(Loc, diag::err_invalid_this_use) << 0;
+    return Diag(Loc, diag::err_invalid_this_use) << 0;
   }
-  return true;
+
+  return BuildCXXThisExpr(Loc, ThisTy, /*IsImplicit=*/false);
 }
 
 Expr *Sema::BuildCXXThisExpr(SourceLocation Loc, QualType Type,
@@ -1708,17 +1693,17 @@ bool Sema::isUsualDeallocationFunction(const CXXMethodDecl *Method) {
   // [CUDA] Ignore this function, if we can't call it.
   const FunctionDecl *Caller = getCurFunctionDecl(/*AllowLambda=*/true);
   if (getLangOpts().CUDA) {
-    auto CallPreference = IdentifyCUDAPreference(Caller, Method);
+    auto CallPreference = CUDA().IdentifyPreference(Caller, Method);
     // If it's not callable at all, it's not the right function.
-    if (CallPreference < CFP_WrongSide)
+    if (CallPreference < SemaCUDA::CFP_WrongSide)
       return false;
-    if (CallPreference == CFP_WrongSide) {
+    if (CallPreference == SemaCUDA::CFP_WrongSide) {
       // Maybe. We have to check if there are better alternatives.
       DeclContext::lookup_result R =
           Method->getDeclContext()->lookup(Method->getDeclName());
       for (const auto *D : R) {
         if (const auto *FD = dyn_cast<FunctionDecl>(D)) {
-          if (IdentifyCUDAPreference(Caller, FD) > CFP_WrongSide)
+          if (CUDA().IdentifyPreference(Caller, FD) > SemaCUDA::CFP_WrongSide)
             return false;
         }
       }
@@ -1737,7 +1722,7 @@ bool Sema::isUsualDeallocationFunction(const CXXMethodDecl *Method) {
   return llvm::none_of(PreventedBy, [&](const FunctionDecl *FD) {
     assert(FD->getNumParams() == 1 &&
            "Only single-operand functions should be in PreventedBy");
-    return IdentifyCUDAPreference(Caller, FD) >= CFP_HostDevice;
+    return CUDA().IdentifyPreference(Caller, FD) >= SemaCUDA::CFP_HostDevice;
   });
 }
 
@@ -1774,7 +1759,7 @@ namespace {
     UsualDeallocFnInfo(Sema &S, DeclAccessPair Found)
         : Found(Found), FD(dyn_cast<FunctionDecl>(Found->getUnderlyingDecl())),
           Destroying(false), HasSizeT(false), HasAlignValT(false),
-          CUDAPref(Sema::CFP_Native) {
+          CUDAPref(SemaCUDA::CFP_Native) {
       // A function template declaration is never a usual deallocation function.
       if (!FD)
         return;
@@ -1800,7 +1785,7 @@ namespace {
 
       // In CUDA, determine how much we'd like / dislike to call this.
       if (S.getLangOpts().CUDA)
-        CUDAPref = S.IdentifyCUDAPreference(
+        CUDAPref = S.CUDA().IdentifyPreference(
             S.getCurFunctionDecl(/*AllowLambda=*/true), FD);
     }
 
@@ -1831,7 +1816,7 @@ namespace {
     DeclAccessPair Found;
     FunctionDecl *FD;
     bool Destroying, HasSizeT, HasAlignValT;
-    Sema::CUDAFunctionPreference CUDAPref;
+    SemaCUDA::CUDAFunctionPreference CUDAPref;
   };
 }
 
@@ -1855,7 +1840,7 @@ static UsualDeallocFnInfo resolveDeallocationOverload(
   for (auto I = R.begin(), E = R.end(); I != E; ++I) {
     UsualDeallocFnInfo Info(S, I.getPair());
     if (!Info || !isNonPlacementDeallocationFunction(S, Info.FD) ||
-        Info.CUDAPref == Sema::CFP_Never)
+        Info.CUDAPref == SemaCUDA::CFP_Never)
       continue;
 
     if (!Best) {
@@ -2715,13 +2700,9 @@ static bool resolveAllocationOverload(
     return true;
 
   case OR_Deleted: {
-    if (Diagnose) {
-      Candidates.NoteCandidates(
-          PartialDiagnosticAt(R.getNameLoc(),
-                              S.PDiag(diag::err_ovl_deleted_call)
-                                  << R.getLookupName() << Range),
-          S, OCD_AllCandidates, Args);
-    }
+    if (Diagnose)
+      S.DiagnoseUseOfDeletedFunction(R.getNameLoc(), Range, R.getLookupName(),
+                                     Candidates, Best->Function, Args);
     return true;
   }
   }
@@ -2956,8 +2937,8 @@ bool Sema::FindAllocationFunctions(SourceLocation StartLoc, SourceRange Range,
     }
 
     if (getLangOpts().CUDA)
-      EraseUnwantedCUDAMatches(getCurFunctionDecl(/*AllowLambda=*/true),
-                               Matches);
+      CUDA().EraseUnwantedMatches(getCurFunctionDecl(/*AllowLambda=*/true),
+                                  Matches);
   } else {
     // C++1y [expr.new]p22:
     //   For a non-placement allocation function, the normal deallocation
@@ -3375,7 +3356,9 @@ bool Sema::FindDeallocationFunction(SourceLocation StartLoc, CXXRecordDecl *RD,
     // FIXME: DiagnoseUseOfDecl?
     if (Operator->isDeleted()) {
       if (Diagnose) {
-        Diag(StartLoc, diag::err_deleted_function_use);
+        StringLiteral *Msg = Operator->getDeletedMessage();
+        Diag(StartLoc, diag::err_deleted_function_use)
+            << (Msg != nullptr) << (Msg ? Msg->getString() : StringRef());
         NoteDeletedFunction(Operator);
       }
       return true;
@@ -3979,14 +3962,11 @@ static bool resolveBuiltinNewDeleteOverload(Sema &S, CallExpr *TheCall,
         S, OCD_AmbiguousCandidates, Args);
     return true;
 
-  case OR_Deleted: {
-    Candidates.NoteCandidates(
-        PartialDiagnosticAt(R.getNameLoc(), S.PDiag(diag::err_ovl_deleted_call)
-                                                << R.getLookupName() << Range),
-        S, OCD_AllCandidates, Args);
+  case OR_Deleted:
+    S.DiagnoseUseOfDeletedFunction(R.getNameLoc(), Range, R.getLookupName(),
+                                   Candidates, Best->Function, Args);
     return true;
   }
-  }
   llvm_unreachable("Unreachable, bad result from BestViableFunction");
 }
 
@@ -5011,6 +4991,20 @@ Sema::PerformImplicitConversion(Expr *From, QualType ToType,
   return From;
 }
 
+/// Checks that type T is not a VLA.
+///
+/// @returns @c true if @p T is VLA and a diagnostic was emitted,
+/// @c false otherwise.
+static bool DiagnoseVLAInCXXTypeTrait(Sema &S, const TypeSourceInfo *T,
+                                      clang::tok::TokenKind TypeTraitID) {
+  if (!T->getType()->isVariableArrayType())
+    return false;
+
+  S.Diag(T->getTypeLoc().getBeginLoc(), diag::err_vla_unsupported)
+      << 1 << TypeTraitID;
+  return true;
+}
+
 /// Check the completeness of a type in a unary type trait.
 ///
 /// If the particular type trait requires a complete type, tries to complete
@@ -5187,7 +5181,9 @@ static bool HasNoThrowOperator(const RecordType *RT, OverloadedOperatorKind Op,
 }
 
 static bool EvaluateUnaryTypeTrait(Sema &Self, TypeTrait UTT,
-                                   SourceLocation KeyLoc, QualType T) {
+                                   SourceLocation KeyLoc,
+                                   TypeSourceInfo *TInfo) {
+  QualType T = TInfo->getType();
   assert(!T->isDependentType() && "Cannot evaluate traits of dependent type");
 
   ASTContext &C = Self.Context;
@@ -5204,21 +5200,13 @@ static bool EvaluateUnaryTypeTrait(Sema &Self, TypeTrait UTT,
   case UTT_IsArray:
     return T->isArrayType();
   case UTT_IsBoundedArray:
-    if (!T->isVariableArrayType()) {
-      return T->isArrayType() && !T->isIncompleteArrayType();
-    }
-
-    Self.Diag(KeyLoc, diag::err_vla_unsupported)
-        << 1 << tok::kw___is_bounded_array;
-    return false;
+    if (DiagnoseVLAInCXXTypeTrait(Self, TInfo, tok::kw___is_bounded_array))
+      return false;
+    return T->isArrayType() && !T->isIncompleteArrayType();
   case UTT_IsUnboundedArray:
-    if (!T->isVariableArrayType()) {
-      return T->isIncompleteArrayType();
-    }
-
-    Self.Diag(KeyLoc, diag::err_vla_unsupported)
-        << 1 << tok::kw___is_unbounded_array;
-    return false;
+    if (DiagnoseVLAInCXXTypeTrait(Self, TInfo, tok::kw___is_unbounded_array))
+      return false;
+    return T->isIncompleteArrayType();
   case UTT_IsPointer:
     return T->isAnyPointerType();
   case UTT_IsNullPointer:
@@ -5630,7 +5618,7 @@ static bool EvaluateBooleanTypeTrait(Sema &S, TypeTrait Kind,
     return false;
 
   if (Kind <= UTT_Last)
-    return EvaluateUnaryTypeTrait(S, Kind, KWLoc, Args[0]->getType());
+    return EvaluateUnaryTypeTrait(S, Kind, KWLoc, Args[0]);
 
   // Evaluate ReferenceBindsToTemporary and ReferenceConstructsFromTemporary
   // alongside the IsConstructible traits to avoid duplication.
@@ -6092,13 +6080,24 @@ static bool EvaluateBinaryTypeTrait(Sema &Self, TypeTrait BTT, const TypeSourceI
       Self.RequireCompleteType(Rhs->getTypeLoc().getBeginLoc(), RhsT,
                                diag::err_incomplete_type);
 
-    if (LhsT->isVariableArrayType())
-      Self.Diag(Lhs->getTypeLoc().getBeginLoc(), diag::err_vla_unsupported)
-          << 1 << tok::kw___is_layout_compatible;
-    if (RhsT->isVariableArrayType())
-      Self.Diag(Rhs->getTypeLoc().getBeginLoc(), diag::err_vla_unsupported)
-          << 1 << tok::kw___is_layout_compatible;
+    DiagnoseVLAInCXXTypeTrait(Self, Lhs, tok::kw___is_layout_compatible);
+    DiagnoseVLAInCXXTypeTrait(Self, Rhs, tok::kw___is_layout_compatible);
+
     return Self.IsLayoutCompatible(LhsT, RhsT);
+  }
+  case BTT_IsPointerInterconvertibleBaseOf: {
+    if (LhsT->isStructureOrClassType() && RhsT->isStructureOrClassType() &&
+        !Self.getASTContext().hasSameUnqualifiedType(LhsT, RhsT)) {
+      Self.RequireCompleteType(Rhs->getTypeLoc().getBeginLoc(), RhsT,
+                               diag::err_incomplete_type);
+    }
+
+    DiagnoseVLAInCXXTypeTrait(Self, Lhs,
+                              tok::kw___is_pointer_interconvertible_base_of);
+    DiagnoseVLAInCXXTypeTrait(Self, Rhs,
+                              tok::kw___is_pointer_interconvertible_base_of);
+
+    return Self.IsPointerInterconvertibleBaseOf(Lhs, Rhs);
   }
     default: llvm_unreachable("not a BTT");
   }
@@ -8430,7 +8429,7 @@ ExprResult Sema::IgnoredValueConversions(Expr *E) {
     // unnecessary temporary objects. If we skip this step, IR generation is
     // able to synthesize the storage for itself in the aggregate case, and
     // adding the extra node to the AST is just clutter.
-    if (isInMaterializeTemporaryObjectContext() && getLangOpts().CPlusPlus17 &&
+    if (isInLifetimeExtendingContext() && getLangOpts().CPlusPlus17 &&
         E->isPRValue() && !E->getType()->isVoidType()) {
       ExprResult Res = TemporaryMaterializationConversion(E);
       if (Res.isInvalid())
@@ -8643,8 +8642,21 @@ static ExprResult attemptRecovery(Sema &SemaRef,
 
       // Detect and handle the case where the decl might be an implicit
       // member.
-      if (SemaRef.isPotentialImplicitMemberAccess(
-              NewSS, R, Consumer.isAddressOfOperand()))
+      bool MightBeImplicitMember;
+      if (!Consumer.isAddressOfOperand())
+        MightBeImplicitMember = true;
+      else if (!NewSS.isEmpty())
+        MightBeImplicitMember = false;
+      else if (R.isOverloadedResult())
+        MightBeImplicitMember = false;
+      else if (R.isUnresolvableResult())
+        MightBeImplicitMember = true;
+      else
+        MightBeImplicitMember = isa<FieldDecl>(ND) ||
+                                isa<IndirectFieldDecl>(ND) ||
+                                isa<MSPropertyDecl>(ND);
+
+      if (MightBeImplicitMember)
         return SemaRef.BuildPossibleImplicitMemberExpr(
             NewSS, /*TemplateKWLoc*/ SourceLocation(), R,
             /*TemplateArgs*/ nullptr, /*S*/ nullptr);

diff --git a/clang/lib/Sema/SemaExprMember.cpp b/clang/lib/Sema/SemaExprMember.cpp
@@ -9,7 +9,6 @@
 //  This file implements semantic analysis member access expressions.
 //
 //===----------------------------------------------------------------------===//
-#include "clang/Sema/Overload.h"
 #include "clang/AST/ASTLambda.h"
 #include "clang/AST/DeclCXX.h"
 #include "clang/AST/DeclObjC.h"
@@ -18,9 +17,11 @@
 #include "clang/AST/ExprObjC.h"
 #include "clang/Lex/Preprocessor.h"
 #include "clang/Sema/Lookup.h"
+#include "clang/Sema/Overload.h"
 #include "clang/Sema/Scope.h"
 #include "clang/Sema/ScopeInfo.h"
 #include "clang/Sema/SemaInternal.h"
+#include "clang/Sema/SemaOpenMP.h"
 
 using namespace clang;
 using namespace sema;
@@ -61,10 +62,6 @@ enum IMAKind {
   /// The reference is a contextually-permitted abstract member reference.
   IMA_Abstract,
 
-  /// Whether the context is static is dependent on the enclosing template (i.e.
-  /// in a dependent class scope explicit specialization).
-  IMA_Dependent,
-
   /// The reference may be to an unresolved using declaration and the
   /// context is not an instance method.
   IMA_Unresolved_StaticOrExplicitContext,
@@ -95,18 +92,10 @@ static IMAKind ClassifyImplicitMemberAccess(Sema &SemaRef,
 
   DeclContext *DC = SemaRef.getFunctionLevelDeclContext();
 
-  bool couldInstantiateToStatic = false;
-  bool isStaticOrExplicitContext = SemaRef.CXXThisTypeOverride.isNull();
-
-  if (auto *MD = dyn_cast<CXXMethodDecl>(DC)) {
-    if (MD->isImplicitObjectMemberFunction()) {
-      isStaticOrExplicitContext = false;
-      // A dependent class scope function template explicit specialization
-      // that is neither declared 'static' nor with an explicit object
-      // parameter could instantiate to a static or non-static member function.
-      couldInstantiateToStatic = MD->getDependentSpecializationInfo();
-    }
-  }
+  bool isStaticOrExplicitContext =
+      SemaRef.CXXThisTypeOverride.isNull() &&
+      (!isa<CXXMethodDecl>(DC) || cast<CXXMethodDecl>(DC)->isStatic() ||
+       cast<CXXMethodDecl>(DC)->isExplicitObjectMemberFunction());
 
   if (R.isUnresolvableResult())
     return isStaticOrExplicitContext ? IMA_Unresolved_StaticOrExplicitContext
@@ -135,9 +124,6 @@ static IMAKind ClassifyImplicitMemberAccess(Sema &SemaRef,
   if (Classes.empty())
     return IMA_Static;
 
-  if (couldInstantiateToStatic)
-    return IMA_Dependent;
-
   // C++11 [expr.prim.general]p12:
   //   An id-expression that denotes a non-static data member or non-static
   //   member function of a class can only be used:
@@ -278,52 +264,32 @@ static void diagnoseInstanceReference(Sema &SemaRef,
   }
 }
 
-bool Sema::isPotentialImplicitMemberAccess(const CXXScopeSpec &SS,
-                                           LookupResult &R,
-                                           bool IsAddressOfOperand) {
-  if (!getLangOpts().CPlusPlus)
-    return false;
-  else if (R.empty() || !R.begin()->isCXXClassMember())
-    return false;
-  else if (!IsAddressOfOperand)
-    return true;
-  else if (!SS.isEmpty())
-    return false;
-  else if (R.isOverloadedResult())
-    return false;
-  else if (R.isUnresolvableResult())
-    return true;
-  else
-    return isa<FieldDecl, IndirectFieldDecl, MSPropertyDecl>(R.getFoundDecl());
-}
-
 /// Builds an expression which might be an implicit member expression.
 ExprResult Sema::BuildPossibleImplicitMemberExpr(
     const CXXScopeSpec &SS, SourceLocation TemplateKWLoc, LookupResult &R,
-    const TemplateArgumentListInfo *TemplateArgs, const Scope *S) {
-  switch (IMAKind Classification = ClassifyImplicitMemberAccess(*this, R)) {
+    const TemplateArgumentListInfo *TemplateArgs, const Scope *S,
+    UnresolvedLookupExpr *AsULE) {
+  switch (ClassifyImplicitMemberAccess(*this, R)) {
   case IMA_Instance:
+    return BuildImplicitMemberExpr(SS, TemplateKWLoc, R, TemplateArgs, true, S);
+
   case IMA_Mixed:
   case IMA_Mixed_Unrelated:
   case IMA_Unresolved:
-    return BuildImplicitMemberExpr(
-        SS, TemplateKWLoc, R, TemplateArgs,
-        /*IsKnownInstance=*/Classification == IMA_Instance, S);
+    return BuildImplicitMemberExpr(SS, TemplateKWLoc, R, TemplateArgs, false,
+                                   S);
+
   case IMA_Field_Uneval_Context:
     Diag(R.getNameLoc(), diag::warn_cxx98_compat_non_static_member_use)
       << R.getLookupNameInfo().getName();
     [[fallthrough]];
   case IMA_Static:
   case IMA_Abstract:
-  case IMA_Dependent:
   case IMA_Mixed_StaticOrExplicitContext:
   case IMA_Unresolved_StaticOrExplicitContext:
     if (TemplateArgs || TemplateKWLoc.isValid())
-      return BuildTemplateIdExpr(SS, TemplateKWLoc, R, /*RequiresADL=*/false,
-                                 TemplateArgs);
-    return BuildDeclarationNameExpr(
-        SS, R, /*NeedsADL=*/false, /*AcceptInvalidDecl=*/false,
-        /*NeedUnresolved=*/Classification == IMA_Dependent);
+      return BuildTemplateIdExpr(SS, TemplateKWLoc, R, false, TemplateArgs);
+    return AsULE ? AsULE : BuildDeclarationNameExpr(SS, R, false);
 
   case IMA_Error_StaticOrExplicitContext:
   case IMA_Error_Unrelated:
@@ -1935,9 +1901,9 @@ Sema::BuildFieldReferenceExpr(Expr *BaseExpr, bool IsArrow,
   if (getLangOpts().OpenMP && IsArrow &&
       !CurContext->isDependentContext() &&
       isa<CXXThisExpr>(Base.get()->IgnoreParenImpCasts())) {
-    if (auto *PrivateCopy = isOpenMPCapturedDecl(Field)) {
-      return getOpenMPCapturedExpr(PrivateCopy, VK, OK,
-                                   MemberNameInfo.getLoc());
+    if (auto *PrivateCopy = OpenMP().isOpenMPCapturedDecl(Field)) {
+      return OpenMP().getOpenMPCapturedExpr(PrivateCopy, VK, OK,
+                                            MemberNameInfo.getLoc());
     }
   }
 

diff --git a/clang/lib/Sema/SemaInit.cpp b/clang/lib/Sema/SemaInit.cpp
@@ -9763,12 +9763,15 @@ bool InitializationSequence::Diagnose(Sema &S,
       break;
     }
     case OR_Deleted: {
-      S.Diag(Kind.getLocation(), diag::err_typecheck_deleted_function)
-        << OnlyArg->getType() << DestType.getNonReferenceType()
-        << Args[0]->getSourceRange();
       OverloadCandidateSet::iterator Best;
       OverloadingResult Ovl
         = FailedCandidateSet.BestViableFunction(S, Kind.getLocation(), Best);
+
+      StringLiteral *Msg = Best->Function->getDeletedMessage();
+      S.Diag(Kind.getLocation(), diag::err_typecheck_deleted_function)
+          << OnlyArg->getType() << DestType.getNonReferenceType()
+          << (Msg != nullptr) << (Msg ? Msg->getString() : StringRef())
+          << Args[0]->getSourceRange();
       if (Ovl == OR_Deleted) {
         S.NoteDeletedFunction(Best->Function);
       } else {
@@ -10027,9 +10030,12 @@ bool InitializationSequence::Diagnose(Sema &S,
               << llvm::to_underlying(
                      S.getSpecialMember(cast<CXXMethodDecl>(Best->Function)))
               << DestType << ArgsRange;
-        else
+        else {
+          StringLiteral *Msg = Best->Function->getDeletedMessage();
           S.Diag(Kind.getLocation(), diag::err_ovl_deleted_init)
-              << DestType << ArgsRange;
+              << DestType << (Msg != nullptr)
+              << (Msg ? Msg->getString() : StringRef()) << ArgsRange;
+        }
 
         S.NoteDeletedFunction(Best->Function);
         break;
@@ -11061,6 +11067,9 @@ QualType Sema::DeduceTemplateSpecializationFromInitializer(
   }
 
   case OR_Deleted: {
+    // FIXME: There are no tests for this diagnostic, and it doesn't seem
+    // like we ever get here; attempts to trigger this seem to yield a
+    // generic c'all to deleted function' diagnostic instead.
     Diag(Kind.getLocation(), diag::err_deduced_class_template_deleted)
       << TemplateName;
     NoteDeletedFunction(Best->Function);

diff --git a/clang/lib/Sema/SemaLambda.cpp b/clang/lib/Sema/SemaLambda.cpp
@@ -9,17 +9,19 @@
 //  This file implements semantic analysis for C++ lambda expressions.
 //
 //===----------------------------------------------------------------------===//
-#include "clang/Sema/DeclSpec.h"
+#include "clang/Sema/SemaLambda.h"
 #include "TypeLocBuilder.h"
 #include "clang/AST/ASTLambda.h"
 #include "clang/AST/ExprCXX.h"
 #include "clang/Basic/TargetInfo.h"
+#include "clang/Sema/DeclSpec.h"
 #include "clang/Sema/Initialization.h"
 #include "clang/Sema/Lookup.h"
 #include "clang/Sema/Scope.h"
 #include "clang/Sema/ScopeInfo.h"
+#include "clang/Sema/SemaCUDA.h"
 #include "clang/Sema/SemaInternal.h"
-#include "clang/Sema/SemaLambda.h"
+#include "clang/Sema/SemaOpenMP.h"
 #include "clang/Sema/Template.h"
 #include "llvm/ADT/STLExtras.h"
 #include <optional>
@@ -1393,11 +1395,11 @@ void Sema::ActOnStartOfLambdaDefinition(LambdaIntroducer &Intro,
 
   // CUDA lambdas get implicit host and device attributes.
   if (getLangOpts().CUDA)
-    CUDASetLambdaAttrs(Method);
+    CUDA().SetLambdaAttrs(Method);
 
   // OpenMP lambdas might get assumumption attributes.
   if (LangOpts.OpenMP)
-    ActOnFinishedFunctionDefinitionInOpenMPAssumeScope(Method);
+    OpenMP().ActOnFinishedFunctionDefinitionInOpenMPAssumeScope(Method);
 
   handleLambdaNumbering(Class, Method);
 
@@ -2136,7 +2138,7 @@ ExprResult Sema::BuildLambdaExpr(SourceLocation StartLoc, SourceLocation EndLoc,
       CaptureInits.push_back(Init.get());
 
       if (LangOpts.CUDA)
-        CUDACheckLambdaCapture(CallOperator, From);
+        CUDA().CheckLambdaCapture(CallOperator, From);
     }
 
     Class->setCaptures(Context, Captures);