diff --git a/clang-tools-extra/docs/ReleaseNotes.rst b/clang-tools-extra/docs/ReleaseNotes.rst
@@ -97,6 +97,10 @@ The improvements are...
 Improvements to clang-tidy
 --------------------------
 
+- Improved :program:`run-clang-tidy.py` script. Added argument `-source-filter`
+  to filter source files from the compilation database, via a RegEx. In a
+  similar fashion to what `-header-filter` does for header files.
+
 New checks
 ^^^^^^^^^^
 
@@ -179,6 +183,12 @@ Changes in existing checks
   <clang-tidy/checks/modernize/use-override>` check to also remove any trailing
   whitespace when deleting the ``virtual`` keyword.
 
+- Improved :doc:`performance-unnecessary-copy-initialization
+  <clang-tidy/checks/performance/unnecessary-copy-initialization>` check by
+  detecting more cases of constant access. In particular, pointers can be
+  analyzed, se the check now handles the common patterns
+  `const auto e = (*vector_ptr)[i]` and `const auto e = vector_ptr->at(i);`.
+
 - Improved :doc:`readability-implicit-bool-conversion
   <clang-tidy/checks/readability/implicit-bool-conversion>` check to provide
   valid fix suggestions for ``static_cast`` without a preceding space and

diff --git a/clang-tools-extra/test/clang-tidy/checkers/performance/unnecessary-copy-initialization.cpp b/clang-tools-extra/test/clang-tidy/checkers/performance/unnecessary-copy-initialization.cpp
@@ -41,6 +41,10 @@ struct Container {
   ConstIterator<T> end() const;
   void nonConstMethod();
   bool constMethod() const;
+
+  const T& at(int) const;
+  T& at(int);
+
 };
 
 using ExpensiveToCopyContainerAlias = Container<ExpensiveToCopyType>;
@@ -225,25 +229,25 @@ void PositiveOperatorCallConstValueParamAlias(const ExpensiveToCopyContainerAlia
   VarCopyConstructed.constMethod();
 }
 
-void PositiveOperatorCallConstValueParam(const Container<ExpensiveToCopyType>* C) {
+void PositiveOperatorCallConstPtrParam(const Container<ExpensiveToCopyType>* C) {
   const auto AutoAssigned = (*C)[42];
-  // TODO-MESSAGES: [[@LINE-1]]:14: warning: the const qualified variable 'AutoAssigned'
-  // TODO-FIXES: const auto& AutoAssigned = (*C)[42];
+  // CHECK-MESSAGES: [[@LINE-1]]:14: warning: the const qualified variable 'AutoAssigned'
+  // CHECK-FIXES: const auto& AutoAssigned = (*C)[42];
   AutoAssigned.constMethod();
 
   const auto AutoCopyConstructed((*C)[42]);
-  // TODO-MESSAGES: [[@LINE-1]]:14: warning: the const qualified variable 'AutoCopyConstructed'
-  // TODO-FIXES: const auto& AutoCopyConstructed((*C)[42]);
+  // CHECK-MESSAGES: [[@LINE-1]]:14: warning: the const qualified variable 'AutoCopyConstructed'
+  // CHECK-FIXES: const auto& AutoCopyConstructed((*C)[42]);
   AutoCopyConstructed.constMethod();
 
-  const ExpensiveToCopyType VarAssigned = C->operator[](42);
-  // TODO-MESSAGES: [[@LINE-1]]:29: warning: the const qualified variable 'VarAssigned'
-  // TODO-FIXES: const ExpensiveToCopyType& VarAssigned = C->operator[](42);
+  const ExpensiveToCopyType VarAssigned = C->at(42);
+  // CHECK-MESSAGES: [[@LINE-1]]:29: warning: the const qualified variable 'VarAssigned'
+  // CHECK-FIXES: const ExpensiveToCopyType& VarAssigned = C->at(42);
   VarAssigned.constMethod();
 
-  const ExpensiveToCopyType VarCopyConstructed(C->operator[](42));
-  // TODO-MESSAGES: [[@LINE-1]]:29: warning: the const qualified variable 'VarCopyConstructed'
-  // TODO-FIXES: const ExpensiveToCopyType& VarCopyConstructed(C->operator[](42));
+  const ExpensiveToCopyType VarCopyConstructed(C->at(42));
+  // CHECK-MESSAGES: [[@LINE-1]]:29: warning: the const qualified variable 'VarCopyConstructed'
+  // CHECK-FIXES: const ExpensiveToCopyType& VarCopyConstructed(C->at(42));
   VarCopyConstructed.constMethod();
 }
 
@@ -876,3 +880,4 @@ void negativeNonConstMemberExpr() {
     mutate(&Copy.Member);
   }
 }
+
diff --git a/clang-tools-extra/unittests/clang-tidy/DeclRefExprUtilsTest.cpp b/clang-tools-extra/unittests/clang-tidy/DeclRefExprUtilsTest.cpp
diff --git a/clang/docs/ClangFormatStyleOptions.rst b/clang/docs/ClangFormatStyleOptions.rst
@@ -955,6 +955,146 @@ the configuration (without a prefix: ``Auto``).
       }
 
 
+.. _AlignConsecutiveTableGenCondOperatorColons:
+
+**AlignConsecutiveTableGenCondOperatorColons** (``AlignConsecutiveStyle``) :versionbadge:`clang-format 19` :ref:`¶ <AlignConsecutiveTableGenCondOperatorColons>`
+  Style of aligning consecutive TableGen cond operator colons.
+  Align the colons of cases inside !cond operators.
+
+  .. code-block:: c++
+
+    !cond(!eq(size, 1) : 1,
+          !eq(size, 16): 1,
+          true         : 0)
+
+  Nested configuration flags:
+
+  Alignment options.
+
+  They can also be read as a whole for compatibility. The choices are:
+  - None
+  - Consecutive
+  - AcrossEmptyLines
+  - AcrossComments
+  - AcrossEmptyLinesAndComments
+
+  For example, to align across empty lines and not across comments, either
+  of these work.
+
+  .. code-block:: c++
+
+    AlignConsecutiveMacros: AcrossEmptyLines
+
+    AlignConsecutiveMacros:
+      Enabled: true
+      AcrossEmptyLines: true
+      AcrossComments: false
+
+  * ``bool Enabled`` Whether aligning is enabled.
+
+    .. code-block:: c++
+
+      #define SHORT_NAME       42
+      #define LONGER_NAME      0x007f
+      #define EVEN_LONGER_NAME (2)
+      #define foo(x)           (x * x)
+      #define bar(y, z)        (y + z)
+
+      int a            = 1;
+      int somelongname = 2;
+      double c         = 3;
+
+      int aaaa : 1;
+      int b    : 12;
+      int ccc  : 8;
+
+      int         aaaa = 12;
+      float       b = 23;
+      std::string ccc;
+
+  * ``bool AcrossEmptyLines`` Whether to align across empty lines.
+
+    .. code-block:: c++
+
+      true:
+      int a            = 1;
+      int somelongname = 2;
+      double c         = 3;
+
+      int d            = 3;
+
+      false:
+      int a            = 1;
+      int somelongname = 2;
+      double c         = 3;
+
+      int d = 3;
+
+  * ``bool AcrossComments`` Whether to align across comments.
+
+    .. code-block:: c++
+
+      true:
+      int d    = 3;
+      /* A comment. */
+      double e = 4;
+
+      false:
+      int d = 3;
+      /* A comment. */
+      double e = 4;
+
+  * ``bool AlignCompound`` Only for ``AlignConsecutiveAssignments``.  Whether compound assignments
+    like ``+=`` are aligned along with ``=``.
+
+    .. code-block:: c++
+
+      true:
+      a   &= 2;
+      bbb  = 2;
+
+      false:
+      a &= 2;
+      bbb = 2;
+
+  * ``bool AlignFunctionPointers`` Only for ``AlignConsecutiveDeclarations``. Whether function pointers are
+    aligned.
+
+    .. code-block:: c++
+
+      true:
+      unsigned i;
+      int     &r;
+      int     *p;
+      int      (*f)();
+
+      false:
+      unsigned i;
+      int     &r;
+      int     *p;
+      int (*f)();
+
+  * ``bool PadOperators`` Only for ``AlignConsecutiveAssignments``.  Whether short assignment
+    operators are left-padded to the same length as long ones in order to
+    put all assignment operators to the right of the left hand side.
+
+    .. code-block:: c++
+
+      true:
+      a   >>= 2;
+      bbb   = 2;
+
+      a     = 2;
+      bbb >>= 2;
+
+      false:
+      a >>= 2;
+      bbb = 2;
+
+      a     = 2;
+      bbb >>= 2;
+
+
 .. _AlignEscapedNewlines:
 
 **AlignEscapedNewlines** (``EscapedNewlineAlignmentStyle``) :versionbadge:`clang-format 5` :ref:`¶ <AlignEscapedNewlines>`
@@ -4844,7 +4984,8 @@ the configuration (without a prefix: ``Auto``).
 .. _RemoveSemicolon:
 
 **RemoveSemicolon** (``Boolean``) :versionbadge:`clang-format 16` :ref:`¶ <RemoveSemicolon>`
-  Remove semicolons after the closing brace of a non-empty function.
+  Remove semicolons after the closing braces of functions and
+  constructors/destructors.
 
   .. warning::
 

diff --git a/clang/docs/HLSL/ExpectedDifferences.rst b/clang/docs/HLSL/ExpectedDifferences.rst
@@ -93,7 +93,7 @@ behavior between Clang and DXC. Some examples include:
     fma(X, Y, Z); // DXC: Fails to resolve no known conversion from float to double.
                   // Clang: Resolves to fma(double,double,double).
   #endif
-    
+
     double D = dot(A, B); // DXC: Resolves to dot(double3, double3), fails DXIL Validation.
                           // FXC: Expands to compute double dot product with fmul/fadd
                           // Clang: Resolves to dot(float3, float3), emits conversion warnings.
@@ -102,7 +102,7 @@ behavior between Clang and DXC. Some examples include:
 
 .. note::
 
-  In Clang, a conscious decision was made to exclude the ``dot(vector<double,N>, vector<double,N>)`` 
+  In Clang, a conscious decision was made to exclude the ``dot(vector<double,N>, vector<double,N>)``
   overload and allow overload resolution to resolve the
   ``vector<float,N>`` overload. This approach provides ``-Wconversion``
   diagnostic notifying the user of the conversion rather than silently altering

diff --git a/clang/docs/ReleaseNotes.rst b/clang/docs/ReleaseNotes.rst
@@ -281,6 +281,8 @@ Bug Fixes to C++ Support
   a requires-clause lie at the same depth as those of the surrounding lambda. This,
   in turn, results in the wrong template argument substitution during constraint checking.
   (`#78524 <https://github.com/llvm/llvm-project/issues/78524>`_)
+- Clang no longer instantiates the exception specification of discarded candidate function
+  templates when determining the primary template of an explicit specialization.
 
 Bug Fixes to AST Handling
 ^^^^^^^^^^^^^^^^^^^^^^^^^

diff --git a/clang/include/clang/Analysis/FlowSensitive/DataflowEnvironment.h b/clang/include/clang/Analysis/FlowSensitive/DataflowEnvironment.h
@@ -753,12 +753,9 @@ RecordStorageLocation *getImplicitObjectLocation(const CXXMemberCallExpr &MCE,
 RecordStorageLocation *getBaseObjectLocation(const MemberExpr &ME,
                                              const Environment &Env);
 
-/// Returns the fields of a `RecordDecl` that are initialized by an
-/// `InitListExpr`, in the order in which they appear in
-/// `InitListExpr::inits()`.
-/// `Init->getType()` must be a record type.
-std::vector<const FieldDecl *>
-getFieldsForInitListExpr(const InitListExpr *InitList);
+/// Returns the fields of `RD` that are initialized by an `InitListExpr`, in the
+/// order in which they appear in `InitListExpr::inits()`.
+std::vector<FieldDecl *> getFieldsForInitListExpr(const RecordDecl *RD);
 
 /// Associates a new `RecordValue` with `Loc` and returns the new value.
 RecordValue &refreshRecordValue(RecordStorageLocation &Loc, Environment &Env);

diff --git a/clang/include/clang/Analysis/FlowSensitive/DataflowValues.h b/clang/include/clang/Analysis/FlowSensitive/DataflowValues.h
@@ -45,12 +45,12 @@ class DataflowValues {
   //===--------------------------------------------------------------------===//
 
 public:
-  typedef typename ValueTypes::ValTy               ValTy;
-  typedef typename ValueTypes::AnalysisDataTy      AnalysisDataTy;
-  typedef _AnalysisDirTag                          AnalysisDirTag;
-  typedef llvm::DenseMap<ProgramPoint, ValTy>      EdgeDataMapTy;
-  typedef llvm::DenseMap<const CFGBlock*, ValTy>   BlockDataMapTy;
-  typedef llvm::DenseMap<const Stmt*, ValTy>       StmtDataMapTy;
+  using ValTy = typename ValueTypes::ValTy;
+  using AnalysisDataTy = typename ValueTypes::AnalysisDataTy;
+  using AnalysisDirTag = _AnalysisDirTag;
+  using EdgeDataMapTy = llvm::DenseMap<ProgramPoint, ValTy>;
+  using BlockDataMapTy = llvm::DenseMap<const CFGBlock *, ValTy>;
+  using StmtDataMapTy = llvm::DenseMap<const Stmt *, ValTy>;
 
   //===--------------------------------------------------------------------===//
   // Predicates.

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
@@ -4524,6 +4524,12 @@ def HLSLCreateHandle : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void*(unsigned char)";
 }
 
+def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 // Builtins for XRay.
 def XRayCustomEvent : Builtin {
   let Spellings = ["__xray_customevent"];

diff --git a/clang/include/clang/Basic/DiagnosticSemaKinds.td b/clang/include/clang/Basic/DiagnosticSemaKinds.td
@@ -10760,6 +10760,13 @@ def warn_imp_cast_drops_unaligned : Warning<
 
 } // end of sema category
 
+let CategoryName = "API Notes Issue" in {
+
+def err_incompatible_replacement_type : Error<
+  "API notes replacement type %0 has a different size from original type %1">;
+
+} // end of API Notes category
+
 let CategoryName = "OpenMP Issue" in {
 // OpenMP support.
 def err_omp_expected_var_arg : Error<
@@ -12203,6 +12210,7 @@ def warn_acc_clause_unimplemented
 def err_acc_construct_appertainment
     : Error<"OpenACC construct '%0' cannot be used here; it can only "
             "be used in a statement context">;
-def err_acc_branch_in_out
-    : Error<"invalid branch %select{out of|into}0 OpenACC Compute Construct">;
+def err_acc_branch_in_out_compute_construct
+    : Error<"invalid %select{branch|return}0 %select{out of|into}1 OpenACC "
+            "Compute Construct">;
 } // end of sema component.
diff --git a/clang/include/clang/Basic/DiagnosticSerializationKinds.td b/clang/include/clang/Basic/DiagnosticSerializationKinds.td
@@ -44,7 +44,9 @@ def err_pch_diagopt_mismatch : Error<"%0 is currently enabled, but was not in "
   "the PCH file">;
 def err_pch_modulecache_mismatch : Error<"PCH was compiled with module cache "
   "path '%0', but the path is currently '%1'">;
-def err_pch_vfsoverlay_mismatch : Error<"PCH was compiled with different VFS overlay files than are currently in use">;
+def warn_pch_vfsoverlay_mismatch : Warning<
+  "PCH was compiled with different VFS overlay files than are currently in use">,
+  InGroup<DiagGroup<"pch-vfs-diff">>;
 def note_pch_vfsoverlay_files : Note<"%select{PCH|current translation unit}0 has the following VFS overlays:\n%1">;
 def note_pch_vfsoverlay_empty : Note<"%select{PCH|current translation unit}0 has no VFS overlays">;
 

diff --git a/clang/include/clang/Format/Format.h b/clang/include/clang/Format/Format.h
@@ -414,6 +414,16 @@ struct FormatStyle {
   /// \version 17
   ShortCaseStatementsAlignmentStyle AlignConsecutiveShortCaseStatements;
 
+  /// Style of aligning consecutive TableGen cond operator colons.
+  /// Align the colons of cases inside !cond operators.
+  /// \code
+  ///   !cond(!eq(size, 1) : 1,
+  ///         !eq(size, 16): 1,
+  ///         true         : 0)
+  /// \endcode
+  /// \version 19
+  AlignConsecutiveStyle AlignConsecutiveTableGenCondOperatorColons;
+
   /// Different styles for aligning escaped newlines.
   enum EscapedNewlineAlignmentStyle : int8_t {
     /// Don't align escaped newlines.
@@ -3771,7 +3781,8 @@ struct FormatStyle {
   /// \version 17
   RemoveParenthesesStyle RemoveParentheses;
 
-  /// Remove semicolons after the closing brace of a non-empty function.
+  /// Remove semicolons after the closing braces of functions and
+  /// constructors/destructors.
   /// \warning
   ///  Setting this option to ``true`` could lead to incorrect code formatting
   ///  due to clang-format's lack of complete semantic information. As such,
@@ -4804,6 +4815,8 @@ struct FormatStyle {
            AlignConsecutiveMacros == R.AlignConsecutiveMacros &&
            AlignConsecutiveShortCaseStatements ==
                R.AlignConsecutiveShortCaseStatements &&
+           AlignConsecutiveTableGenCondOperatorColons ==
+               R.AlignConsecutiveTableGenCondOperatorColons &&
            AlignEscapedNewlines == R.AlignEscapedNewlines &&
            AlignOperands == R.AlignOperands &&
            AlignTrailingComments == R.AlignTrailingComments &&

diff --git a/clang/include/clang/Sema/Scope.h b/clang/include/clang/Sema/Scope.h
@@ -521,6 +521,19 @@ class Scope {
     return getFlags() & Scope::OpenACCComputeConstructScope;
   }
 
+  bool isInOpenACCComputeConstructScope() const {
+    for (const Scope *S = this; S; S = S->getParent()) {
+      if (S->getFlags() & Scope::OpenACCComputeConstructScope)
+        return true;
+      else if (S->getFlags() &
+               (Scope::FnScope | Scope::ClassScope | Scope::BlockScope |
+                Scope::TemplateParamScope | Scope::FunctionPrototypeScope |
+                Scope::AtCatchScope | Scope::ObjCMethodScope))
+        return false;
+    }
+    return false;
+  }
+
   /// Determine whether this scope is a while/do/for statement, which can have
   /// continue statements embedded into it.
   bool isContinueScope() const {

diff --git a/clang/include/clang/Sema/Sema.h b/clang/include/clang/Sema/Sema.h
@@ -4883,6 +4883,12 @@ class Sema final {
   bool checkCommonAttributeFeatures(const Stmt *S, const ParsedAttr &A,
                                     bool SkipArgCountCheck = false);
 
+  /// Map any API notes provided for this declaration to attributes on the
+  /// declaration.
+  ///
+  /// Triggered by declaration-attribute processing.
+  void ProcessAPINotes(Decl *D);
+
   /// Determine if type T is a valid subject for a nonnull and similar
   /// attributes. By default, we look through references (the behavior used by
   /// nonnull), but if the second parameter is true, then we treat a reference
@@ -14057,6 +14063,7 @@ class Sema final {
   bool CheckPPCBuiltinFunctionCall(const TargetInfo &TI, unsigned BuiltinID,
                                    CallExpr *TheCall);
   bool CheckAMDGCNBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
+  bool CheckHLSLBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
   bool CheckRISCVLMUL(CallExpr *TheCall, unsigned ArgNum);
   bool CheckRISCVBuiltinFunctionCall(const TargetInfo &TI, unsigned BuiltinID,
                                      CallExpr *TheCall);
@@ -14122,6 +14129,8 @@ class Sema final {
 
   bool CheckPPCMMAType(QualType Type, SourceLocation TypeLoc);
 
+  bool SemaBuiltinVectorMath(CallExpr *TheCall, QualType &Res);
+  bool SemaBuiltinVectorToScalarMath(CallExpr *TheCall);
   bool SemaBuiltinElementwiseMath(CallExpr *TheCall);
   bool SemaBuiltinElementwiseTernaryMath(CallExpr *TheCall);
   bool PrepareBuiltinElementwiseMathOneArgCall(CallExpr *TheCall);

diff --git a/clang/include/clang/Tooling/DependencyScanning/DependencyScanningService.h b/clang/include/clang/Tooling/DependencyScanning/DependencyScanningService.h
@@ -60,7 +60,10 @@ enum class ScanningOptimizations {
   /// Remove unused -ivfsoverlay arguments.
   VFS = 4,
 
-  DSS_LAST_BITMASK_ENUM(VFS),
+  /// Canonicalize -D and -U options.
+  Macros = 8,
+
+  DSS_LAST_BITMASK_ENUM(Macros),
   Default = All
 };
 

diff --git a/clang/include/clang/Tooling/DependencyScanning/ModuleDepCollector.h b/clang/include/clang/Tooling/DependencyScanning/ModuleDepCollector.h
@@ -149,6 +149,8 @@ struct ModuleDeps {
       BuildInfo;
 };
 
+using PrebuiltModuleVFSMapT = llvm::StringMap<llvm::StringSet<>>;
+
 class ModuleDepCollector;
 
 /// Callback that records textual includes and direct modular includes/imports
@@ -214,6 +216,7 @@ class ModuleDepCollector final : public DependencyCollector {
                      CompilerInstance &ScanInstance, DependencyConsumer &C,
                      DependencyActionController &Controller,
                      CompilerInvocation OriginalCI,
+                     PrebuiltModuleVFSMapT PrebuiltModuleVFSMap,
                      ScanningOptimizations OptimizeArgs, bool EagerLoadModules,
                      bool IsStdModuleP1689Format);
 
@@ -233,6 +236,8 @@ class ModuleDepCollector final : public DependencyCollector {
   DependencyConsumer &Consumer;
   /// Callbacks for computing dependency information.
   DependencyActionController &Controller;
+  /// Mapping from prebuilt AST files to their sorted list of VFS overlay files.
+  PrebuiltModuleVFSMapT PrebuiltModuleVFSMap;
   /// Path to the main source file.
   std::string MainFile;
   /// Hash identifying the compilation conditions of the current TU.

diff --git a/clang/lib/AST/CMakeLists.txt b/clang/lib/AST/CMakeLists.txt
@@ -88,6 +88,7 @@ add_clang_library(clangAST
   Interp/Record.cpp
   Interp/Source.cpp
   Interp/State.cpp
+  Interp/InterpShared.cpp
   ItaniumCXXABI.cpp
   ItaniumMangle.cpp
   JSONNodeDumper.cpp

diff --git a/clang/lib/AST/Interp/ByteCodeExprGen.cpp b/clang/lib/AST/Interp/ByteCodeExprGen.cpp
@@ -13,8 +13,10 @@
 #include "Context.h"
 #include "Floating.h"
 #include "Function.h"
+#include "InterpShared.h"
 #include "PrimType.h"
 #include "Program.h"
+#include "clang/AST/Attr.h"
 
 using namespace clang;
 using namespace clang::interp;
@@ -658,19 +660,16 @@ bool ByteCodeExprGen<Emitter>::VisitComplexBinOp(const BinaryOperator *E) {
       return false;
   }
 
+  // Both LHS and RHS might _not_ be of complex type, but one of them
+  // needs to be.
   const Expr *LHS = E->getLHS();
   const Expr *RHS = E->getRHS();
-  PrimType LHSElemT = this->classifyComplexElementType(LHS->getType());
-  PrimType RHSElemT = this->classifyComplexElementType(RHS->getType());
 
-  unsigned LHSOffset = this->allocateLocalPrimitive(LHS, PT_Ptr, true, false);
-  unsigned RHSOffset = this->allocateLocalPrimitive(RHS, PT_Ptr, true, false);
+  PrimType ResultElemT = this->classifyComplexElementType(E->getType());
   unsigned ResultOffset = ~0u;
-  if (!this->DiscardResult)
+  if (!DiscardResult)
     ResultOffset = this->allocateLocalPrimitive(E, PT_Ptr, true, false);
 
-  assert(LHSElemT == RHSElemT);
-
   // Save result pointer in ResultOffset
   if (!this->DiscardResult) {
     if (!this->emitDupPtr(E))
@@ -680,16 +679,64 @@ bool ByteCodeExprGen<Emitter>::VisitComplexBinOp(const BinaryOperator *E) {
   }
 
   // Evaluate LHS and save value to LHSOffset.
-  if (!this->visit(LHS))
-    return false;
-  if (!this->emitSetLocal(PT_Ptr, LHSOffset, E))
-    return false;
+  bool LHSIsComplex;
+  unsigned LHSOffset;
+  if (LHS->getType()->isAnyComplexType()) {
+    LHSIsComplex = true;
+    LHSOffset = this->allocateLocalPrimitive(LHS, PT_Ptr, true, false);
+    if (!this->visit(LHS))
+      return false;
+    if (!this->emitSetLocal(PT_Ptr, LHSOffset, E))
+      return false;
+  } else {
+    LHSIsComplex = false;
+    PrimType LHST = classifyPrim(LHS->getType());
+    LHSOffset = this->allocateLocalPrimitive(LHS, LHST, true, false);
+    if (!this->visit(LHS))
+      return false;
+    if (!this->emitSetLocal(LHST, LHSOffset, E))
+      return false;
+  }
 
   // Same with RHS.
-  if (!this->visit(RHS))
-    return false;
-  if (!this->emitSetLocal(PT_Ptr, RHSOffset, E))
-    return false;
+  bool RHSIsComplex;
+  unsigned RHSOffset;
+  if (RHS->getType()->isAnyComplexType()) {
+    RHSIsComplex = true;
+    RHSOffset = this->allocateLocalPrimitive(RHS, PT_Ptr, true, false);
+    if (!this->visit(RHS))
+      return false;
+    if (!this->emitSetLocal(PT_Ptr, RHSOffset, E))
+      return false;
+  } else {
+    RHSIsComplex = false;
+    PrimType RHST = classifyPrim(RHS->getType());
+    RHSOffset = this->allocateLocalPrimitive(RHS, RHST, true, false);
+    if (!this->visit(RHS))
+      return false;
+    if (!this->emitSetLocal(RHST, RHSOffset, E))
+      return false;
+  }
+
+  // For both LHS and RHS, either load the value from the complex pointer, or
+  // directly from the local variable. For index 1 (i.e. the imaginary part),
+  // just load 0 and do the operation anyway.
+  auto loadComplexValue = [this](bool IsComplex, unsigned ElemIndex,
+                                 unsigned Offset, const Expr *E) -> bool {
+    if (IsComplex) {
+      if (!this->emitGetLocal(PT_Ptr, Offset, E))
+        return false;
+      if (!this->emitConstUint8(ElemIndex, E))
+        return false;
+      if (!this->emitArrayElemPtrPopUint8(E))
+        return false;
+      return this->emitLoadPop(classifyComplexElementType(E->getType()), E);
+    }
+    if (ElemIndex == 0)
+      return this->emitGetLocal(classifyPrim(E->getType()), Offset, E);
+    return this->visitZeroInitializer(classifyPrim(E->getType()), E->getType(),
+                                      E);
+  };
 
   // Now we can get pointers to the LHS and RHS from the offsets above.
   BinaryOperatorKind Op = E->getOpcode();
@@ -700,41 +747,29 @@ bool ByteCodeExprGen<Emitter>::VisitComplexBinOp(const BinaryOperator *E) {
         return false;
     }
 
-    if (!this->emitGetLocal(PT_Ptr, LHSOffset, E))
-      return false;
-    if (!this->emitConstUint8(ElemIndex, E))
-      return false;
-    if (!this->emitArrayElemPtrPopUint8(E))
-      return false;
-    if (!this->emitLoadPop(LHSElemT, E))
+    if (!loadComplexValue(LHSIsComplex, ElemIndex, LHSOffset, LHS))
       return false;
 
-    if (!this->emitGetLocal(PT_Ptr, RHSOffset, E))
-      return false;
-    if (!this->emitConstUint8(ElemIndex, E))
-      return false;
-    if (!this->emitArrayElemPtrPopUint8(E))
-      return false;
-    if (!this->emitLoadPop(RHSElemT, E))
+    if (!loadComplexValue(RHSIsComplex, ElemIndex, RHSOffset, RHS))
       return false;
 
     // The actual operation.
     switch (Op) {
     case BO_Add:
-      if (LHSElemT == PT_Float) {
+      if (ResultElemT == PT_Float) {
         if (!this->emitAddf(getRoundingMode(E), E))
           return false;
       } else {
-        if (!this->emitAdd(LHSElemT, E))
+        if (!this->emitAdd(ResultElemT, E))
           return false;
       }
       break;
     case BO_Sub:
-      if (LHSElemT == PT_Float) {
+      if (ResultElemT == PT_Float) {
         if (!this->emitSubf(getRoundingMode(E), E))
           return false;
       } else {
-        if (!this->emitSub(LHSElemT, E))
+        if (!this->emitSub(ResultElemT, E))
           return false;
       }
       break;
@@ -745,10 +780,10 @@ bool ByteCodeExprGen<Emitter>::VisitComplexBinOp(const BinaryOperator *E) {
 
     if (!this->DiscardResult) {
       // Initialize array element with the value we just computed.
-      if (!this->emitInitElemPop(LHSElemT, ElemIndex, E))
+      if (!this->emitInitElemPop(ResultElemT, ElemIndex, E))
         return false;
     } else {
-      if (!this->emitPop(LHSElemT, E))
+      if (!this->emitPop(ResultElemT, E))
         return false;
     }
   }
@@ -1644,13 +1679,24 @@ bool ByteCodeExprGen<Emitter>::VisitCompoundLiteralExpr(
 
   std::optional<PrimType> T = classify(E->getType());
   if (E->isFileScope()) {
+    // Avoid creating a variable if this is a primitive RValue anyway.
+    if (T && !E->isLValue())
+      return this->delegate(Init);
+
     if (std::optional<unsigned> GlobalIndex = P.createGlobal(E)) {
-      if (classify(E->getType()))
-        return this->visit(Init);
       if (!this->emitGetPtrGlobal(*GlobalIndex, E))
         return false;
+
+      if (T) {
+        if (!this->visit(Init))
+          return false;
+        return this->emitInitGlobal(*T, *GlobalIndex, E);
+      }
+
       return this->visitInitializer(Init);
     }
+
+    return false;
   }
 
   // Otherwise, use a local variable.
@@ -1696,6 +1742,9 @@ bool ByteCodeExprGen<Emitter>::VisitLambdaExpr(const LambdaExpr *E) {
     const Expr *Init = *CaptureInitIt;
     ++CaptureInitIt;
 
+    if (!Init)
+      continue;
+
     if (std::optional<PrimType> T = classify(Init)) {
       if (!this->visit(Init))
         return false;
@@ -2656,6 +2705,7 @@ bool ByteCodeExprGen<Emitter>::VisitCallExpr(const CallExpr *E) {
   QualType ReturnType = E->getCallReturnType(Ctx.getASTContext());
   std::optional<PrimType> T = classify(ReturnType);
   bool HasRVO = !ReturnType->isVoidType() && !T;
+  const FunctionDecl *FuncDecl = E->getDirectCallee();
 
   if (HasRVO) {
     if (DiscardResult) {
@@ -2667,23 +2717,29 @@ bool ByteCodeExprGen<Emitter>::VisitCallExpr(const CallExpr *E) {
           return false;
       }
     } else {
-      assert(Initializing);
+      // We need the result. Prepare a pointer to return or
+      // dup the current one.
+      if (!Initializing) {
+        if (std::optional<unsigned> LocalIndex = allocateLocal(E)) {
+          if (!this->emitGetPtrLocal(*LocalIndex, E))
+            return false;
+        }
+      }
       if (!this->emitDupPtr(E))
         return false;
     }
   }
 
-  auto Args = E->arguments();
+  auto Args = llvm::ArrayRef(E->getArgs(), E->getNumArgs());
   // Calling a static operator will still
   // pass the instance, but we don't need it.
   // Discard it here.
   if (isa<CXXOperatorCallExpr>(E)) {
-    if (const auto *MD =
-            dyn_cast_if_present<CXXMethodDecl>(E->getDirectCallee());
+    if (const auto *MD = dyn_cast_if_present<CXXMethodDecl>(FuncDecl);
         MD && MD->isStatic()) {
       if (!this->discard(E->getArg(0)))
         return false;
-      Args = drop_begin(Args, 1);
+      Args = Args.drop_front();
     }
   }
 
@@ -2693,13 +2749,25 @@ bool ByteCodeExprGen<Emitter>::VisitCallExpr(const CallExpr *E) {
       return false;
   }
 
+  llvm::BitVector NonNullArgs = collectNonNullArgs(FuncDecl, Args);
   // Put arguments on the stack.
+  unsigned ArgIndex = 0;
   for (const auto *Arg : Args) {
     if (!this->visit(Arg))
       return false;
+
+    // If we know the callee already, check the known parametrs for nullability.
+    if (FuncDecl && NonNullArgs[ArgIndex]) {
+      PrimType ArgT = classify(Arg).value_or(PT_Ptr);
+      if (ArgT == PT_Ptr || ArgT == PT_FnPtr) {
+        if (!this->emitCheckNonNullArg(ArgT, Arg))
+          return false;
+      }
+    }
+    ++ArgIndex;
   }
 
-  if (const FunctionDecl *FuncDecl = E->getDirectCallee()) {
+  if (FuncDecl) {
     const Function *Func = getFunction(FuncDecl);
     if (!Func)
       return false;
@@ -2748,7 +2816,7 @@ bool ByteCodeExprGen<Emitter>::VisitCallExpr(const CallExpr *E) {
     if (!this->visit(E->getCallee()))
       return false;
 
-    if (!this->emitCallPtr(ArgSize, E))
+    if (!this->emitCallPtr(ArgSize, E, E))
       return false;
   }
 

diff --git a/clang/lib/AST/Interp/Context.cpp b/clang/lib/AST/Interp/Context.cpp
@@ -160,7 +160,7 @@ std::optional<PrimType> Context::classify(QualType T) const {
   if (T->isReferenceType() || T->isPointerType())
     return PT_Ptr;
 
-  if (const auto *AT = dyn_cast<AtomicType>(T))
+  if (const auto *AT = T->getAs<AtomicType>())
     return classify(AT->getValueType());
 
   if (const auto *DT = dyn_cast<DecltypeType>(T))

diff --git a/clang/lib/AST/Interp/Function.h b/clang/lib/AST/Interp/Function.h
@@ -15,9 +15,10 @@
 #ifndef LLVM_CLANG_AST_INTERP_FUNCTION_H
 #define LLVM_CLANG_AST_INTERP_FUNCTION_H
 
-#include "Source.h"
 #include "Descriptor.h"
+#include "Source.h"
 #include "clang/AST/ASTLambda.h"
+#include "clang/AST/Attr.h"
 #include "clang/AST/Decl.h"
 #include "llvm/Support/raw_ostream.h"
 
@@ -108,6 +109,8 @@ class Function final {
   /// Checks if the first argument is a RVO pointer.
   bool hasRVO() const { return HasRVO; }
 
+  bool hasNonNullAttr() const { return getDecl()->hasAttr<NonNullAttr>(); }
+
   /// Range over the scope blocks.
   llvm::iterator_range<llvm::SmallVector<Scope, 2>::const_iterator>
   scopes() const {

diff --git a/clang/lib/AST/Interp/FunctionPointer.h b/clang/lib/AST/Interp/FunctionPointer.h
@@ -26,6 +26,7 @@ class FunctionPointer final {
   FunctionPointer(const Function *Func) : Func(Func) { assert(Func); }
 
   const Function *getFunction() const { return Func; }
+  bool isZero() const { return !Func; }
 
   APValue toAPValue() const {
     if (!Func)

diff --git a/clang/lib/AST/Interp/Interp.cpp b/clang/lib/AST/Interp/Interp.cpp
@@ -7,10 +7,9 @@
 //===----------------------------------------------------------------------===//
 
 #include "Interp.h"
-#include <limits>
-#include <vector>
 #include "Function.h"
 #include "InterpFrame.h"
+#include "InterpShared.h"
 #include "InterpStack.h"
 #include "Opcode.h"
 #include "PrimType.h"
@@ -22,6 +21,10 @@
 #include "clang/AST/Expr.h"
 #include "clang/AST/ExprCXX.h"
 #include "llvm/ADT/APSInt.h"
+#include <limits>
+#include <vector>
+
+using namespace clang;
 
 using namespace clang;
 using namespace clang::interp;
@@ -622,6 +625,28 @@ bool CheckDeclRef(InterpState &S, CodePtr OpPC, const DeclRefExpr *DR) {
   return false;
 }
 
+bool CheckNonNullArgs(InterpState &S, CodePtr OpPC, const Function *F,
+                      const CallExpr *CE, unsigned ArgSize) {
+  auto Args = llvm::ArrayRef(CE->getArgs(), CE->getNumArgs());
+  auto NonNullArgs = collectNonNullArgs(F->getDecl(), Args);
+  unsigned Offset = 0;
+  unsigned Index = 0;
+  for (const Expr *Arg : Args) {
+    if (NonNullArgs[Index] && Arg->getType()->isPointerType()) {
+      const Pointer &ArgPtr = S.Stk.peek<Pointer>(ArgSize - Offset);
+      if (ArgPtr.isZero()) {
+        const SourceLocation &Loc = S.Current->getLocation(OpPC);
+        S.CCEDiag(Loc, diag::note_non_null_attribute_failed);
+        return false;
+      }
+    }
+
+    Offset += align(primSize(S.Ctx.classify(Arg).value_or(PT_Ptr)));
+    ++Index;
+  }
+  return true;
+}
+
 bool Interpret(InterpState &S, APValue &Result) {
   // The current stack frame when we started Interpret().
   // This is being used by the ops to determine wheter

diff --git a/clang/lib/AST/Interp/Interp.h b/clang/lib/AST/Interp/Interp.h
@@ -113,6 +113,10 @@ bool CheckThis(InterpState &S, CodePtr OpPC, const Pointer &This);
 /// Checks if a method is pure virtual.
 bool CheckPure(InterpState &S, CodePtr OpPC, const CXXMethodDecl *MD);
 
+/// Checks if all the arguments annotated as 'nonnull' are in fact not null.
+bool CheckNonNullArgs(InterpState &S, CodePtr OpPC, const Function *F,
+                      const CallExpr *CE, unsigned ArgSize);
+
 /// Sets the given integral value to the pointer, which is of
 /// a std::{weak,partial,strong}_ordering type.
 bool SetThreeWayComparisonField(InterpState &S, CodePtr OpPC,
@@ -804,7 +808,8 @@ bool CMP3(InterpState &S, CodePtr OpPC, const ComparisonCategoryInfo *CmpInfo) {
   }
 
   assert(CmpInfo);
-  const auto *CmpValueInfo = CmpInfo->getValueInfo(CmpResult);
+  const auto *CmpValueInfo =
+      CmpInfo->getValueInfo(CmpInfo->makeWeakResult(CmpResult));
   assert(CmpValueInfo);
   assert(CmpValueInfo->hasValidIntValue());
   const APSInt &IntValue = CmpValueInfo->getIntValue();
@@ -1043,7 +1048,7 @@ bool InitGlobal(InterpState &S, CodePtr OpPC, uint32_t I) {
 
 /// 1) Converts the value on top of the stack to an APValue
 /// 2) Sets that APValue on \Temp
-/// 3) Initialized global with index \I with that
+/// 3) Initializes global with index \I with that
 template <PrimType Name, class T = typename PrimConv<Name>::T>
 bool InitGlobalTemp(InterpState &S, CodePtr OpPC, uint32_t I,
                     const LifetimeExtendedTemporaryDecl *Temp) {
@@ -1980,6 +1985,7 @@ inline bool CallVar(InterpState &S, CodePtr OpPC, const Function *Func,
 
   return false;
 }
+
 inline bool Call(InterpState &S, CodePtr OpPC, const Function *Func,
                  uint32_t VarArgSize) {
   if (Func->hasThisPointer()) {
@@ -2083,12 +2089,25 @@ inline bool CallBI(InterpState &S, CodePtr &PC, const Function *Func,
   return false;
 }
 
-inline bool CallPtr(InterpState &S, CodePtr OpPC, uint32_t ArgSize) {
+inline bool CallPtr(InterpState &S, CodePtr OpPC, uint32_t ArgSize,
+                    const CallExpr *CE) {
   const FunctionPointer &FuncPtr = S.Stk.pop<FunctionPointer>();
 
   const Function *F = FuncPtr.getFunction();
+  if (!F) {
+    const Expr *E = S.Current->getExpr(OpPC);
+    S.FFDiag(E, diag::note_constexpr_null_callee)
+        << const_cast<Expr *>(E) << E->getSourceRange();
+    return false;
+  }
   assert(F);
 
+  // Check argument nullability state.
+  if (F->hasNonNullAttr()) {
+    if (!CheckNonNullArgs(S, OpPC, F, CE, ArgSize))
+      return false;
+  }
+
   assert(ArgSize >= F->getWrittenArgSize());
   uint32_t VarArgSize = ArgSize - F->getWrittenArgSize();
 
@@ -2145,6 +2164,18 @@ inline bool OffsetOf(InterpState &S, CodePtr OpPC, const OffsetOfExpr *E) {
   return true;
 }
 
+template <PrimType Name, class T = typename PrimConv<Name>::T>
+inline bool CheckNonNullArg(InterpState &S, CodePtr OpPC) {
+  const T &Arg = S.Stk.peek<T>();
+  if (!Arg.isZero())
+    return true;
+
+  const SourceLocation &Loc = S.Current->getLocation(OpPC);
+  S.CCEDiag(Loc, diag::note_non_null_attribute_failed);
+
+  return false;
+}
+
 //===----------------------------------------------------------------------===//
 // Read opcode arguments
 //===----------------------------------------------------------------------===//

diff --git a/clang/lib/AST/Interp/InterpBuiltin.cpp b/clang/lib/AST/Interp/InterpBuiltin.cpp
diff --git a/clang/lib/AST/Interp/InterpShared.cpp b/clang/lib/AST/Interp/InterpShared.cpp
@@ -0,0 +1,42 @@
+//===--- InterpShared.cpp ---------------------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#include "InterpShared.h"
+#include "clang/AST/Attr.h"
+#include "llvm/ADT/BitVector.h"
+
+namespace clang {
+namespace interp {
+
+llvm::BitVector collectNonNullArgs(const FunctionDecl *F,
+                                   const llvm::ArrayRef<const Expr *> &Args) {
+  llvm::BitVector NonNullArgs;
+  if (!F)
+    return NonNullArgs;
+
+  assert(F);
+  NonNullArgs.resize(Args.size());
+
+  for (const auto *Attr : F->specific_attrs<NonNullAttr>()) {
+    if (!Attr->args_size()) {
+      NonNullArgs.set();
+      break;
+    } else
+      for (auto Idx : Attr->args()) {
+        unsigned ASTIdx = Idx.getASTIndex();
+        if (ASTIdx >= Args.size())
+          continue;
+        NonNullArgs[ASTIdx] = true;
+      }
+  }
+
+  return NonNullArgs;
+}
+
+} // namespace interp
+} // namespace clang
diff --git a/clang/lib/AST/Interp/InterpShared.h b/clang/lib/AST/Interp/InterpShared.h
@@ -0,0 +1,26 @@
+//===--- InterpShared.h -----------------------------------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CLANG_LIB_AST_INTERP_SHARED_H
+#define LLVM_CLANG_LIB_AST_INTERP_SHARED_H
+
+#include "llvm/ADT/BitVector.h"
+
+namespace clang {
+class FunctionDecl;
+class Expr;
+
+namespace interp {
+
+llvm::BitVector collectNonNullArgs(const FunctionDecl *F,
+                                   const llvm::ArrayRef<const Expr *> &Args);
+
+} // namespace interp
+} // namespace clang
+
+#endif
diff --git a/clang/lib/AST/Interp/Opcodes.td b/clang/lib/AST/Interp/Opcodes.td
@@ -206,7 +206,7 @@ def CallBI : Opcode {
 }
 
 def CallPtr : Opcode {
-  let Args = [ArgUint32];
+  let Args = [ArgUint32, ArgCallExpr];
   let Types = [];
 }
 
@@ -706,3 +706,8 @@ def InvalidDeclRef : Opcode {
 }
 
 def ArrayDecay : Opcode;
+
+def CheckNonNullArg : Opcode {
+  let Types = [PtrTypeClass];
+  let HasGroup = 1;
+}
diff --git a/clang/lib/Analysis/FlowSensitive/DataflowEnvironment.cpp b/clang/lib/Analysis/FlowSensitive/DataflowEnvironment.cpp
@@ -361,8 +361,8 @@ getFieldsGlobalsAndFuncs(const Stmt &S, FieldSet &Fields,
     if (const auto *FD = dyn_cast<FieldDecl>(VD))
       Fields.insert(FD);
   } else if (auto *InitList = dyn_cast<InitListExpr>(&S)) {
-    if (InitList->getType()->isRecordType())
-      for (const auto *FD : getFieldsForInitListExpr(InitList))
+    if (RecordDecl *RD = InitList->getType()->getAsRecordDecl())
+      for (const auto *FD : getFieldsForInitListExpr(RD))
         Fields.insert(FD);
   }
 }
@@ -1104,22 +1104,12 @@ RecordStorageLocation *getBaseObjectLocation(const MemberExpr &ME,
   return Env.get<RecordStorageLocation>(*Base);
 }
 
-std::vector<const FieldDecl *>
-getFieldsForInitListExpr(const InitListExpr *InitList) {
-  const RecordDecl *RD = InitList->getType()->getAsRecordDecl();
-  assert(RD != nullptr);
-
-  std::vector<const FieldDecl *> Fields;
-
-  if (InitList->getType()->isUnionType()) {
-    Fields.push_back(InitList->getInitializedFieldInUnion());
-    return Fields;
-  }
-
+std::vector<FieldDecl *> getFieldsForInitListExpr(const RecordDecl *RD) {
   // Unnamed bitfields are only used for padding and do not appear in
   // `InitListExpr`'s inits. However, those fields do appear in `RecordDecl`'s
   // field list, and we thus need to remove them before mapping inits to
   // fields to avoid mapping inits to the wrongs fields.
+  std::vector<FieldDecl *> Fields;
   llvm::copy_if(
       RD->fields(), std::back_inserter(Fields),
       [](const FieldDecl *Field) { return !Field->isUnnamedBitfield(); });

diff --git a/clang/lib/Analysis/FlowSensitive/Transfer.cpp b/clang/lib/Analysis/FlowSensitive/Transfer.cpp
@@ -663,10 +663,17 @@ class TransferVisitor : public ConstStmtVisitor<TransferVisitor> {
   void VisitInitListExpr(const InitListExpr *S) {
     QualType Type = S->getType();
 
-    if (!Type->isRecordType()) {
-      // Until array initialization is implemented, we don't need to care about
-      // cases where `getNumInits() > 1`.
-      if (S->getNumInits() == 1)
+    if (Type->isUnionType()) {
+      // FIXME: Initialize unions properly.
+      if (auto *Val = Env.createValue(Type))
+        Env.setValue(*S, *Val);
+      return;
+    }
+
+    if (!Type->isStructureOrClassType()) {
+      // Until array initialization is implemented, we skip arrays and don't
+      // need to care about cases where `getNumInits() > 1`.
+      if (!Type->isArrayType() && S->getNumInits() == 1)
         propagateValueOrStorageLocation(*S->getInit(0), *S, Env);
       return;
     }
@@ -681,9 +688,10 @@ class TransferVisitor : public ConstStmtVisitor<TransferVisitor> {
     llvm::DenseMap<const ValueDecl *, StorageLocation *> FieldLocs;
 
     // This only contains the direct fields for the given type.
-    std::vector<const FieldDecl *> FieldsForInit = getFieldsForInitListExpr(S);
+    std::vector<FieldDecl *> FieldsForInit =
+        getFieldsForInitListExpr(Type->getAsRecordDecl());
 
-    // `S->inits()` contains all the initializer expressions, including the
+    // `S->inits()` contains all the initializer epressions, including the
     // ones for direct base classes.
     auto Inits = S->inits();
     size_t InitIdx = 0;
@@ -723,17 +731,6 @@ class TransferVisitor : public ConstStmtVisitor<TransferVisitor> {
       FieldLocs.insert({Field, &Loc});
     }
 
-    // In the case of a union, we don't in general have initializers for all
-    // of the fields. Create storage locations for the remaining fields (but
-    // don't associate them with values).
-    if (Type->isUnionType()) {
-      for (const FieldDecl *Field :
-           Env.getDataflowAnalysisContext().getModeledFields(Type)) {
-        if (auto [it, inserted] = FieldLocs.insert({Field, nullptr}); inserted)
-          it->second = &Env.createStorageLocation(Field->getType());
-      }
-    }
-
     // Check that we satisfy the invariant that a `RecordStorageLoation`
     // contains exactly the set of modeled fields for that type.
     // `ModeledFields` includes fields from all the bases, but only the

diff --git a/clang/lib/Basic/Targets/AMDGPU.h b/clang/lib/Basic/Targets/AMDGPU.h
@@ -414,8 +414,10 @@ class LLVM_LIBRARY_VISIBILITY AMDGPUTargetInfo final : public TargetInfo {
   // value ~0.
   uint64_t getNullPointerValue(LangAS AS) const override {
     // FIXME: Also should handle region.
-    return (AS == LangAS::opencl_local || AS == LangAS::opencl_private)
-      ? ~0 : 0;
+    return (AS == LangAS::opencl_local || AS == LangAS::opencl_private ||
+            AS == LangAS::sycl_local || AS == LangAS::sycl_private)
+               ? ~0
+               : 0;
   }
 
   void setAuxTarget(const TargetInfo *Aux) override;

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -44,6 +44,7 @@
 #include "llvm/IR/IntrinsicsAMDGPU.h"
 #include "llvm/IR/IntrinsicsARM.h"
 #include "llvm/IR/IntrinsicsBPF.h"
+#include "llvm/IR/IntrinsicsDirectX.h"
 #include "llvm/IR/IntrinsicsHexagon.h"
 #include "llvm/IR/IntrinsicsNVPTX.h"
 #include "llvm/IR/IntrinsicsPowerPC.h"
@@ -5982,6 +5983,10 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
     llvm_unreachable("Bad evaluation kind in EmitBuiltinExpr");
   }
 
+  // EmitHLSLBuiltinExpr will check getLangOpts().HLSL
+  if (Value *V = EmitHLSLBuiltinExpr(BuiltinID, E))
+    return RValue::get(V);
+
   if (getLangOpts().HIPStdPar && getLangOpts().CUDAIsDevice)
     return EmitHipStdParUnsupportedBuiltin(this, FD);
 
@@ -17959,6 +17964,52 @@ llvm::Value *CodeGenFunction::EmitScalarOrConstFoldImmArg(unsigned ICEArguments,
   return Arg;
 }
 
+Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
+                                            const CallExpr *E) {
+  if (!getLangOpts().HLSL)
+    return nullptr;
+
+  switch (BuiltinID) {
+  case Builtin::BI__builtin_hlsl_dot: {
+    Value *Op0 = EmitScalarExpr(E->getArg(0));
+    Value *Op1 = EmitScalarExpr(E->getArg(1));
+    llvm::Type *T0 = Op0->getType();
+    llvm::Type *T1 = Op1->getType();
+    if (!T0->isVectorTy() && !T1->isVectorTy()) {
+      if (T0->isFloatingPointTy())
+        return Builder.CreateFMul(Op0, Op1, "dx.dot");
+
+      if (T0->isIntegerTy())
+        return Builder.CreateMul(Op0, Op1, "dx.dot");
+
+      // Bools should have been promoted
+      llvm_unreachable(
+          "Scalar dot product is only supported on ints and floats.");
+    }
+    // A VectorSplat should have happened
+    assert(T0->isVectorTy() && T1->isVectorTy() &&
+           "Dot product of vector and scalar is not supported.");
+
+    // A vector sext or sitofp should have happened
+    assert(T0->getScalarType() == T1->getScalarType() &&
+           "Dot product of vectors need the same element types.");
+
+    [[maybe_unused]] auto *VecTy0 =
+        E->getArg(0)->getType()->getAs<VectorType>();
+    [[maybe_unused]] auto *VecTy1 =
+        E->getArg(1)->getType()->getAs<VectorType>();
+    // A HLSLVectorTruncation should have happend
+    assert(VecTy0->getNumElements() == VecTy1->getNumElements() &&
+           "Dot product requires vectors to be of the same size.");
+
+    return Builder.CreateIntrinsic(
+        /*ReturnType*/ T0->getScalarType(), Intrinsic::dx_dot,
+        ArrayRef<Value *>{Op0, Op1}, nullptr, "dx.dot");
+  } break;
+  }
+  return nullptr;
+}
+
 Value *CodeGenFunction::EmitAMDGPUBuiltinExpr(unsigned BuiltinID,
                                               const CallExpr *E) {
   llvm::AtomicOrdering AO = llvm::AtomicOrdering::SequentiallyConsistent;

diff --git a/clang/lib/CodeGen/CGRecordLayoutBuilder.cpp b/clang/lib/CodeGen/CGRecordLayoutBuilder.cpp
@@ -95,7 +95,7 @@ struct CGRecordLowering {
   CGRecordLowering(CodeGenTypes &Types, const RecordDecl *D, bool Packed);
   // Short helper routines.
   /// Constructs a MemberInfo instance from an offset and llvm::Type *.
-  MemberInfo StorageInfo(CharUnits Offset, llvm::Type *Data) {
+  static MemberInfo StorageInfo(CharUnits Offset, llvm::Type *Data) {
     return MemberInfo(Offset, MemberInfo::Field, Data);
   }
 
@@ -104,7 +104,7 @@ struct CGRecordLowering {
   /// fields of the same formal type.  We want to emit a layout with
   /// these discrete storage units instead of combining them into a
   /// continuous run.
-  bool isDiscreteBitFieldABI() {
+  bool isDiscreteBitFieldABI() const {
     return Context.getTargetInfo().getCXXABI().isMicrosoft() ||
            D->isMsStruct(Context);
   }
@@ -121,60 +121,60 @@ struct CGRecordLowering {
   /// other bases, which complicates layout in specific ways.
   ///
   /// Note specifically that the ms_struct attribute doesn't change this.
-  bool isOverlappingVBaseABI() {
+  bool isOverlappingVBaseABI() const {
     return !Context.getTargetInfo().getCXXABI().isMicrosoft();
   }
 
   /// Wraps llvm::Type::getIntNTy with some implicit arguments.
-  llvm::Type *getIntNType(uint64_t NumBits) {
+  llvm::Type *getIntNType(uint64_t NumBits) const {
     unsigned AlignedBits = llvm::alignTo(NumBits, Context.getCharWidth());
     return llvm::Type::getIntNTy(Types.getLLVMContext(), AlignedBits);
   }
   /// Get the LLVM type sized as one character unit.
-  llvm::Type *getCharType() {
+  llvm::Type *getCharType() const {
     return llvm::Type::getIntNTy(Types.getLLVMContext(),
                                  Context.getCharWidth());
   }
   /// Gets an llvm type of size NumChars and alignment 1.
-  llvm::Type *getByteArrayType(CharUnits NumChars) {
+  llvm::Type *getByteArrayType(CharUnits NumChars) const {
     assert(!NumChars.isZero() && "Empty byte arrays aren't allowed.");
     llvm::Type *Type = getCharType();
     return NumChars == CharUnits::One() ? Type :
         (llvm::Type *)llvm::ArrayType::get(Type, NumChars.getQuantity());
   }
   /// Gets the storage type for a field decl and handles storage
   /// for itanium bitfields that are smaller than their declared type.
-  llvm::Type *getStorageType(const FieldDecl *FD) {
+  llvm::Type *getStorageType(const FieldDecl *FD) const {
     llvm::Type *Type = Types.ConvertTypeForMem(FD->getType());
     if (!FD->isBitField()) return Type;
     if (isDiscreteBitFieldABI()) return Type;
     return getIntNType(std::min(FD->getBitWidthValue(Context),
                              (unsigned)Context.toBits(getSize(Type))));
   }
   /// Gets the llvm Basesubobject type from a CXXRecordDecl.
-  llvm::Type *getStorageType(const CXXRecordDecl *RD) {
+  llvm::Type *getStorageType(const CXXRecordDecl *RD) const {
     return Types.getCGRecordLayout(RD).getBaseSubobjectLLVMType();
   }
-  CharUnits bitsToCharUnits(uint64_t BitOffset) {
+  CharUnits bitsToCharUnits(uint64_t BitOffset) const {
     return Context.toCharUnitsFromBits(BitOffset);
   }
-  CharUnits getSize(llvm::Type *Type) {
+  CharUnits getSize(llvm::Type *Type) const {
     return CharUnits::fromQuantity(DataLayout.getTypeAllocSize(Type));
   }
-  CharUnits getAlignment(llvm::Type *Type) {
+  CharUnits getAlignment(llvm::Type *Type) const {
     return CharUnits::fromQuantity(DataLayout.getABITypeAlign(Type));
   }
-  bool isZeroInitializable(const FieldDecl *FD) {
+  bool isZeroInitializable(const FieldDecl *FD) const {
     return Types.isZeroInitializable(FD->getType());
   }
-  bool isZeroInitializable(const RecordDecl *RD) {
+  bool isZeroInitializable(const RecordDecl *RD) const {
     return Types.isZeroInitializable(RD);
   }
   void appendPaddingBytes(CharUnits Size) {
     if (!Size.isZero())
       FieldTypes.push_back(getByteArrayType(Size));
   }
-  uint64_t getFieldBitOffset(const FieldDecl *FD) {
+  uint64_t getFieldBitOffset(const FieldDecl *FD) const {
     return Layout.getFieldOffset(FD->getFieldIndex());
   }
   // Layout routines.

diff --git a/clang/lib/CodeGen/CodeGenFunction.h b/clang/lib/CodeGen/CodeGenFunction.h
@@ -4405,6 +4405,7 @@ class CodeGenFunction : public CodeGenTypeCache {
   llvm::Value *EmitX86BuiltinExpr(unsigned BuiltinID, const CallExpr *E);
   llvm::Value *EmitPPCBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
   llvm::Value *EmitAMDGPUBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
+  llvm::Value *EmitHLSLBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
   llvm::Value *EmitScalarOrConstFoldImmArg(unsigned ICEArguments, unsigned Idx,
                                            const CallExpr *E);
   llvm::Value *EmitSystemZBuiltinExpr(unsigned BuiltinID, const CallExpr *E);

diff --git a/clang/lib/CodeGen/CodeGenPGO.cpp b/clang/lib/CodeGen/CodeGenPGO.cpp
@@ -165,6 +165,7 @@ struct MapRegionCounters : public RecursiveASTVisitor<MapRegionCounters> {
   llvm::DenseMap<const Stmt *, unsigned> &CounterMap;
   /// The next bitmap byte index to assign.
   unsigned NextMCDCBitmapIdx;
+  /// The state of MC/DC Coverage in this function.
   MCDC::State &MCDCState;
   /// Maximum number of supported MC/DC conditions in a boolean expression.
   unsigned MCDCMaxCond;
@@ -311,7 +312,7 @@ struct MapRegionCounters : public RecursiveASTVisitor<MapRegionCounters> {
 
           // Otherwise, allocate the number of bytes required for the bitmap
           // based on the number of conditions. Must be at least 1-byte long.
-          MCDCState.BitmapMap[BinOp] = NextMCDCBitmapIdx;
+          MCDCState.DecisionByStmt[BinOp].BitmapIdx = NextMCDCBitmapIdx;
           unsigned SizeInBits = std::max<unsigned>(1L << NumCond, CHAR_BIT);
           NextMCDCBitmapIdx += SizeInBits / CHAR_BIT;
         }
@@ -1034,7 +1035,7 @@ void CodeGenPGO::emitCounterRegionMapping(const Decl *D) {
 
   std::string CoverageMapping;
   llvm::raw_string_ostream OS(CoverageMapping);
-  RegionCondIDMap.reset(new llvm::DenseMap<const Stmt *, int16_t>);
+  RegionMCDCState->BranchByStmt.clear();
   CoverageMappingGen MappingGen(
       *CGM.getCoverageMapping(), CGM.getContext().getSourceManager(),
       CGM.getLangOpts(), RegionCounterMap.get(), RegionMCDCState.get());
@@ -1142,12 +1143,12 @@ void CodeGenPGO::emitMCDCTestVectorBitmapUpdate(CGBuilderTy &Builder,
 
   S = S->IgnoreParens();
 
-  auto ExprMCDCBitmapMapIterator = RegionMCDCState->BitmapMap.find(S);
-  if (ExprMCDCBitmapMapIterator == RegionMCDCState->BitmapMap.end())
+  auto DecisionStateIter = RegionMCDCState->DecisionByStmt.find(S);
+  if (DecisionStateIter == RegionMCDCState->DecisionByStmt.end())
     return;
 
-  // Extract the ID of the global bitmap associated with this expression.
-  unsigned MCDCTestVectorBitmapID = ExprMCDCBitmapMapIterator->second;
+  // Extract the offset of the global bitmap associated with this expression.
+  unsigned MCDCTestVectorBitmapOffset = DecisionStateIter->second.BitmapIdx;
   auto *I8PtrTy = llvm::PointerType::getUnqual(CGM.getLLVMContext());
 
   // Emit intrinsic responsible for updating the global bitmap corresponding to
@@ -1158,7 +1159,7 @@ void CodeGenPGO::emitMCDCTestVectorBitmapUpdate(CGBuilderTy &Builder,
   llvm::Value *Args[5] = {llvm::ConstantExpr::getBitCast(FuncNameVar, I8PtrTy),
                           Builder.getInt64(FunctionHash),
                           Builder.getInt32(RegionMCDCState->BitmapBytes),
-                          Builder.getInt32(MCDCTestVectorBitmapID),
+                          Builder.getInt32(MCDCTestVectorBitmapOffset),
                           MCDCCondBitmapAddr.getPointer()};
   Builder.CreateCall(
       CGM.getIntrinsic(llvm::Intrinsic::instrprof_mcdc_tvbitmap_update), Args);
@@ -1171,7 +1172,7 @@ void CodeGenPGO::emitMCDCCondBitmapReset(CGBuilderTy &Builder, const Expr *S,
 
   S = S->IgnoreParens();
 
-  if (!RegionMCDCState->BitmapMap.contains(S))
+  if (!RegionMCDCState->DecisionByStmt.contains(S))
     return;
 
   // Emit intrinsic that resets a dedicated temporary value on the stack to 0.
@@ -1193,13 +1194,13 @@ void CodeGenPGO::emitMCDCCondBitmapUpdate(CGBuilderTy &Builder, const Expr *S,
   // also make debugging a bit easier.
   S = CodeGenFunction::stripCond(S);
 
-  auto ExprMCDCConditionIDMapIterator = RegionMCDCState->CondIDMap.find(S);
-  if (ExprMCDCConditionIDMapIterator == RegionMCDCState->CondIDMap.end())
+  auto BranchStateIter = RegionMCDCState->BranchByStmt.find(S);
+  if (BranchStateIter == RegionMCDCState->BranchByStmt.end())
     return;
 
   // Extract the ID of the condition we are setting in the bitmap.
-  auto CondID = ExprMCDCConditionIDMapIterator->second;
-  assert(CondID >= 0 && "Condition has no ID!");
+  const auto &Branch = BranchStateIter->second;
+  assert(Branch.ID >= 0 && "Condition has no ID!");
 
   auto *I8PtrTy = llvm::PointerType::getUnqual(CGM.getLLVMContext());
 
@@ -1208,7 +1209,7 @@ void CodeGenPGO::emitMCDCCondBitmapUpdate(CGBuilderTy &Builder, const Expr *S,
   // the resulting value is used to update the boolean expression's bitmap.
   llvm::Value *Args[5] = {llvm::ConstantExpr::getBitCast(FuncNameVar, I8PtrTy),
                           Builder.getInt64(FunctionHash),
-                          Builder.getInt32(CondID),
+                          Builder.getInt32(Branch.ID),
                           MCDCCondBitmapAddr.getPointer(), Val};
   Builder.CreateCall(
       CGM.getIntrinsic(llvm::Intrinsic::instrprof_mcdc_condbitmap_update),

diff --git a/clang/lib/CodeGen/CodeGenPGO.h b/clang/lib/CodeGen/CodeGenPGO.h
@@ -36,8 +36,6 @@ class CodeGenPGO {
   unsigned NumRegionCounters;
   uint64_t FunctionHash;
   std::unique_ptr<llvm::DenseMap<const Stmt *, unsigned>> RegionCounterMap;
-  std::unique_ptr<llvm::DenseMap<const Stmt *, unsigned>> RegionMCDCBitmapMap;
-  std::unique_ptr<llvm::DenseMap<const Stmt *, int16_t>> RegionCondIDMap;
   std::unique_ptr<llvm::DenseMap<const Stmt *, uint64_t>> StmtCountMap;
   std::unique_ptr<llvm::InstrProfRecord> ProfRecord;
   std::unique_ptr<MCDC::State> RegionMCDCState;

diff --git a/clang/lib/CodeGen/CoverageMappingGen.cpp b/clang/lib/CodeGen/CoverageMappingGen.cpp
@@ -684,7 +684,6 @@ struct MCDCCoverageBuilder {
 
   llvm::SmallVector<mcdc::ConditionIDs> DecisionStack;
   MCDC::State &MCDCState;
-  llvm::DenseMap<const Stmt *, mcdc::ConditionID> &CondIDs;
   mcdc::ConditionID NextID = 0;
   bool NotMapped = false;
 
@@ -699,8 +698,8 @@ struct MCDCCoverageBuilder {
 
 public:
   MCDCCoverageBuilder(CodeGenModule &CGM, MCDC::State &MCDCState)
-      : CGM(CGM), DecisionStack(1, DecisionStackSentinel), MCDCState(MCDCState),
-        CondIDs(MCDCState.CondIDMap) {}
+      : CGM(CGM), DecisionStack(1, DecisionStackSentinel),
+        MCDCState(MCDCState) {}
 
   /// Return whether the build of the control flow map is at the top-level
   /// (root) of a logical operator nest in a boolean expression prior to the
@@ -714,16 +713,16 @@ struct MCDCCoverageBuilder {
 
   /// Set the given condition's ID.
   void setCondID(const Expr *Cond, mcdc::ConditionID ID) {
-    CondIDs[CodeGenFunction::stripCond(Cond)] = ID;
+    MCDCState.BranchByStmt[CodeGenFunction::stripCond(Cond)].ID = ID;
   }
 
   /// Return the ID of a given condition.
   mcdc::ConditionID getCondID(const Expr *Cond) const {
-    auto I = CondIDs.find(CodeGenFunction::stripCond(Cond));
-    if (I == CondIDs.end())
+    auto I = MCDCState.BranchByStmt.find(CodeGenFunction::stripCond(Cond));
+    if (I == MCDCState.BranchByStmt.end())
       return -1;
     else
-      return I->second;
+      return I->second.ID;
   }
 
   /// Return the LHS Decision ([0,0] if not set).
@@ -738,7 +737,7 @@ struct MCDCCoverageBuilder {
 
     // If binary expression is disqualified, don't do mapping.
     if (!isBuilding() &&
-        !MCDCState.BitmapMap.contains(CodeGenFunction::stripCond(E)))
+        !MCDCState.DecisionByStmt.contains(CodeGenFunction::stripCond(E)))
       NotMapped = true;
 
     // Don't go any further if we don't need to map condition IDs.
@@ -750,7 +749,7 @@ struct MCDCCoverageBuilder {
     // If the operator itself has an assigned ID, this means it represents a
     // larger subtree.  In this case, assign that ID to its LHS node.  Its RHS
     // will receive a new ID below. Otherwise, assign ID+1 to LHS.
-    if (CondIDs.contains(CodeGenFunction::stripCond(E)))
+    if (MCDCState.BranchByStmt.contains(CodeGenFunction::stripCond(E)))
       setCondID(E->getLHS(), getCondID(E));
     else
       setCondID(E->getLHS(), NextID++);
@@ -853,8 +852,6 @@ struct CounterCoverageMappingBuilder
     return Counter::getCounter(CounterMap[S]);
   }
 
-  unsigned getRegionBitmap(const Stmt *S) { return MCDCState.BitmapMap[S]; }
-
   /// Push a region onto the stack.
   ///
   /// Returns the index on the stack where the region was pushed. This can be
@@ -887,12 +884,11 @@ struct CounterCoverageMappingBuilder
     return RegionStack.size() - 1;
   }
 
-  size_t pushRegion(unsigned BitmapIdx, uint16_t Conditions,
+  size_t pushRegion(const mcdc::DecisionParameters &DecisionParams,
                     std::optional<SourceLocation> StartLoc = std::nullopt,
                     std::optional<SourceLocation> EndLoc = std::nullopt) {
 
-    RegionStack.emplace_back(mcdc::DecisionParameters{BitmapIdx, Conditions},
-                             StartLoc, EndLoc);
+    RegionStack.emplace_back(DecisionParams, StartLoc, EndLoc);
 
     return RegionStack.size() - 1;
   }
@@ -1058,8 +1054,9 @@ struct CounterCoverageMappingBuilder
   /// Create a Decision Region with a BitmapIdx and number of Conditions. This
   /// type of region "contains" branch regions, one for each of the conditions.
   /// The visualization tool will group everything together.
-  void createDecisionRegion(const Expr *C, unsigned BitmapIdx, unsigned Conds) {
-    popRegions(pushRegion(BitmapIdx, Conds, getStart(C), getEnd(C)));
+  void createDecisionRegion(const Expr *C,
+                            const mcdc::DecisionParameters &DecisionParams) {
+    popRegions(pushRegion(DecisionParams, getStart(C), getEnd(C)));
   }
 
   /// Create a Branch Region around a SwitchCase for code coverage
@@ -1311,7 +1308,7 @@ struct CounterCoverageMappingBuilder
       return;
     assert(SpellingRegion(SM, NewStartLoc, EndLoc).isInSourceOrder());
     handleFileExit(NewStartLoc);
-    size_t Index = pushRegion({}, NewStartLoc, EndLoc);
+    size_t Index = pushRegion(Counter{}, NewStartLoc, EndLoc);
     getRegion().setSkipped(true);
     handleFileExit(EndLoc);
     popRegions(Index);
@@ -1961,6 +1958,20 @@ struct CounterCoverageMappingBuilder
                        subtractCounters(ParentCount, TrueCount));
   }
 
+  void createDecision(const BinaryOperator *E) {
+    unsigned NumConds = MCDCBuilder.getTotalConditionsAndReset(E);
+    if (NumConds == 0)
+      return;
+
+    auto DecisionParams = mcdc::DecisionParameters{
+        MCDCState.DecisionByStmt[E].BitmapIdx,
+        NumConds,
+    };
+
+    // Create MCDC Decision Region.
+    createDecisionRegion(E, DecisionParams);
+  }
+
   void VisitBinLAnd(const BinaryOperator *E) {
     bool IsRootNode = MCDCBuilder.isIdle();
 
@@ -1981,11 +1992,6 @@ struct CounterCoverageMappingBuilder
     // Track RHS True/False Decision.
     const auto DecisionRHS = MCDCBuilder.back();
 
-    // Create MCDC Decision Region if at top-level (root).
-    unsigned NumConds = 0;
-    if (IsRootNode && (NumConds = MCDCBuilder.getTotalConditionsAndReset(E)))
-      createDecisionRegion(E, getRegionBitmap(E), NumConds);
-
     // Extract the RHS's Execution Counter.
     Counter RHSExecCnt = getRegionCounter(E);
 
@@ -2002,6 +2008,10 @@ struct CounterCoverageMappingBuilder
     // Create Branch Region around RHS condition.
     createBranchRegion(E->getRHS(), RHSTrueCnt,
                        subtractCounters(RHSExecCnt, RHSTrueCnt), DecisionRHS);
+
+    // Create MCDC Decision Region if at top-level (root).
+    if (IsRootNode)
+      createDecision(E);
   }
 
   // Determine whether the right side of OR operation need to be visited.
@@ -2034,11 +2044,6 @@ struct CounterCoverageMappingBuilder
     // Track RHS True/False Decision.
     const auto DecisionRHS = MCDCBuilder.back();
 
-    // Create MCDC Decision Region if at top-level (root).
-    unsigned NumConds = 0;
-    if (IsRootNode && (NumConds = MCDCBuilder.getTotalConditionsAndReset(E)))
-      createDecisionRegion(E, getRegionBitmap(E), NumConds);
-
     // Extract the RHS's Execution Counter.
     Counter RHSExecCnt = getRegionCounter(E);
 
@@ -2059,6 +2064,10 @@ struct CounterCoverageMappingBuilder
     // Create Branch Region around RHS condition.
     createBranchRegion(E->getRHS(), subtractCounters(RHSExecCnt, RHSFalseCnt),
                        RHSFalseCnt, DecisionRHS);
+
+    // Create MCDC Decision Region if at top-level (root).
+    if (IsRootNode)
+      createDecision(E);
   }
 
   void VisitLambdaExpr(const LambdaExpr *LE) {

diff --git a/clang/lib/CodeGen/MCDCState.h b/clang/lib/CodeGen/MCDCState.h
@@ -27,8 +27,18 @@ using namespace llvm::coverage::mcdc;
 /// Per-Function MC/DC state
 struct State {
   unsigned BitmapBytes = 0;
-  llvm::DenseMap<const Stmt *, unsigned> BitmapMap;
-  llvm::DenseMap<const Stmt *, ConditionID> CondIDMap;
+
+  struct Decision {
+    unsigned BitmapIdx;
+  };
+
+  llvm::DenseMap<const Stmt *, Decision> DecisionByStmt;
+
+  struct Branch {
+    ConditionID ID;
+  };
+
+  llvm::DenseMap<const Stmt *, Branch> BranchByStmt;
 };
 
 } // namespace clang::CodeGen::MCDC

diff --git a/clang/lib/CodeGen/Targets/NVPTX.cpp b/clang/lib/CodeGen/Targets/NVPTX.cpp
@@ -47,6 +47,10 @@ class NVPTXTargetCodeGenInfo : public TargetCodeGenInfo {
                            CodeGen::CodeGenModule &M) const override;
   bool shouldEmitStaticExternCAliases() const override;
 
+  llvm::Constant *getNullPointer(const CodeGen::CodeGenModule &CGM,
+                                 llvm::PointerType *T,
+                                 QualType QT) const override;
+
   llvm::Type *getCUDADeviceBuiltinSurfaceDeviceType() const override {
     // On the device side, surface reference is represented as an object handle
     // in 64-bit integer.
@@ -285,6 +289,20 @@ void NVPTXTargetCodeGenInfo::addNVVMMetadata(llvm::GlobalValue *GV,
 bool NVPTXTargetCodeGenInfo::shouldEmitStaticExternCAliases() const {
   return false;
 }
+
+llvm::Constant *
+NVPTXTargetCodeGenInfo::getNullPointer(const CodeGen::CodeGenModule &CGM,
+                                       llvm::PointerType *PT,
+                                       QualType QT) const {
+  auto &Ctx = CGM.getContext();
+  if (PT->getAddressSpace() != Ctx.getTargetAddressSpace(LangAS::opencl_local))
+    return llvm::ConstantPointerNull::get(PT);
+
+  auto NPT = llvm::PointerType::get(
+      PT->getContext(), Ctx.getTargetAddressSpace(LangAS::opencl_generic));
+  return llvm::ConstantExpr::getAddrSpaceCast(
+      llvm::ConstantPointerNull::get(NPT), PT);
+}
 }
 
 void CodeGenModule::handleCUDALaunchBoundsAttr(llvm::Function *F,

diff --git a/clang/lib/Driver/ToolChain.cpp b/clang/lib/Driver/ToolChain.cpp
@@ -200,7 +200,12 @@ static void getAArch64MultilibFlags(const Driver &D,
   for (const auto &Ext : AArch64::Extensions)
     if (FeatureSet.contains(Ext.NegFeature))
       MArch.push_back(("no" + Ext.Name).str());
-  MArch.insert(MArch.begin(), ("-march=" + Triple.getArchName()).str());
+  StringRef ArchName;
+  for (const auto &ArchInfo : AArch64::ArchInfos)
+    if (FeatureSet.contains(ArchInfo->ArchFeature))
+      ArchName = ArchInfo->Name;
+  assert(!ArchName.empty() && "at least one architecture should be found");
+  MArch.insert(MArch.begin(), ("-march=" + ArchName).str());
   Result.push_back(llvm::join(MArch, "+"));
 }
 

diff --git a/clang/lib/Driver/ToolChains/AMDGPU.cpp b/clang/lib/Driver/ToolChains/AMDGPU.cpp
@@ -625,6 +625,7 @@ void amdgpu::Linker::ConstructJob(Compilation &C, const JobAction &JA,
 
   addLinkerCompressDebugSectionsOption(getToolChain(), Args, CmdArgs);
   Args.AddAllArgs(CmdArgs, options::OPT_L);
+  getToolChain().AddFilePathLibArgs(Args, CmdArgs);
   AddLinkerInputs(getToolChain(), Inputs, Args, CmdArgs, JA);
   if (C.getDriver().isUsingLTO())
     addLTOOptions(getToolChain(), Args, CmdArgs, Output, Inputs[0],

diff --git a/clang/lib/Driver/ToolChains/Clang.cpp b/clang/lib/Driver/ToolChains/Clang.cpp
@@ -1111,8 +1111,8 @@ void Clang::AddPreprocessingOptions(Compilation &C, const JobAction &JA,
         C.getActiveOffloadKinds() == Action::OFK_None) {
       SmallString<128> P(llvm::sys::path::parent_path(D.InstalledDir));
       llvm::sys::path::append(P, "include");
-      llvm::sys::path::append(P, "gpu-none-llvm");
-      CmdArgs.push_back("-c-isystem");
+      llvm::sys::path::append(P, getToolChain().getTripleString());
+      CmdArgs.push_back("-internal-isystem");
       CmdArgs.push_back(Args.MakeArgString(P));
     } else if (C.getActiveOffloadKinds() == Action::OFK_OpenMP) {
       // TODO: CUDA / HIP include their own headers for some common functions

diff --git a/clang/lib/Driver/ToolChains/Cuda.cpp b/clang/lib/Driver/ToolChains/Cuda.cpp
@@ -609,6 +609,10 @@ void NVPTX::Linker::ConstructJob(Compilation &C, const JobAction &JA,
   // Add paths specified in LIBRARY_PATH environment variable as -L options.
   addDirectoryList(Args, CmdArgs, "-L", "LIBRARY_PATH");
 
+  // Add standard library search paths passed on the command line.
+  Args.AddAllArgs(CmdArgs, options::OPT_L);
+  getToolChain().AddFilePathLibArgs(Args, CmdArgs);
+
   // Add paths for the default clang library path.
   SmallString<256> DefaultLibPath =
       llvm::sys::path::parent_path(TC.getDriver().Dir);

diff --git a/clang/lib/Format/Format.cpp b/clang/lib/Format/Format.cpp
@@ -915,6 +915,8 @@ template <> struct MappingTraits<FormatStyle> {
     IO.mapOptional("AlignConsecutiveMacros", Style.AlignConsecutiveMacros);
     IO.mapOptional("AlignConsecutiveShortCaseStatements",
                    Style.AlignConsecutiveShortCaseStatements);
+    IO.mapOptional("AlignConsecutiveTableGenCondOperatorColons",
+                   Style.AlignConsecutiveTableGenCondOperatorColons);
     IO.mapOptional("AlignEscapedNewlines", Style.AlignEscapedNewlines);
     IO.mapOptional("AlignOperands", Style.AlignOperands);
     IO.mapOptional("AlignTrailingComments", Style.AlignTrailingComments);
@@ -1420,6 +1422,7 @@ FormatStyle getLLVMStyle(FormatStyle::LanguageKind Language) {
   LLVMStyle.AlignConsecutiveDeclarations = {};
   LLVMStyle.AlignConsecutiveMacros = {};
   LLVMStyle.AlignConsecutiveShortCaseStatements = {};
+  LLVMStyle.AlignConsecutiveTableGenCondOperatorColons = {};
   LLVMStyle.AlignEscapedNewlines = FormatStyle::ENAS_Right;
   LLVMStyle.AlignOperands = FormatStyle::OAS_Align;
   LLVMStyle.AlignTrailingComments = {};
@@ -1591,7 +1594,7 @@ FormatStyle getLLVMStyle(FormatStyle::LanguageKind Language) {
   LLVMStyle.PenaltyBreakScopeResolution = 500;
   LLVMStyle.PenaltyBreakString = 1000;
   LLVMStyle.PenaltyBreakTemplateDeclaration = prec::Relational;
-  LLVMStyle.PenaltyExcessCharacter = 1000000;
+  LLVMStyle.PenaltyExcessCharacter = 1'000'000;
   LLVMStyle.PenaltyIndentedWhitespace = 0;
   LLVMStyle.PenaltyReturnTypeOnItsOwnLine = 60;
 
@@ -1914,9 +1917,12 @@ FormatStyle getClangFormatStyle() {
   FormatStyle Style = getLLVMStyle();
   Style.InsertBraces = true;
   Style.InsertNewlineAtEOF = true;
+  Style.IntegerLiteralSeparator.Decimal = 3;
+  Style.IntegerLiteralSeparator.DecimalMinDigits = 5;
   Style.LineEnding = FormatStyle::LE_LF;
   Style.RemoveBracesLLVM = true;
   Style.RemoveParentheses = FormatStyle::RPS_ReturnStatement;
+  Style.RemoveSemicolon = true;
   return Style;
 }
 

diff --git a/clang/lib/Format/FormatToken.cpp b/clang/lib/Format/FormatToken.cpp
@@ -137,7 +137,7 @@ unsigned CommaSeparatedList::formatAfterToken(LineState &State,
   // bin-packed. Add a severe penalty to this so that column layouts are
   // preferred if possible.
   if (!Format)
-    return 10000;
+    return 10'000;
 
   // Format the entire list.
   unsigned Penalty = 0;

diff --git a/clang/lib/Format/UnwrappedLineFormatter.cpp b/clang/lib/Format/UnwrappedLineFormatter.cpp
@@ -1221,7 +1221,7 @@ class OptimizingLineFormatter : public LineFormatter {
     // While not empty, take first element and follow edges.
     while (!Queue.empty()) {
       // Quit if we still haven't found a solution by now.
-      if (Count > 25000000)
+      if (Count > 25'000'000)
         return 0;
 
       Penalty = Queue.top().first.first;
@@ -1235,7 +1235,7 @@ class OptimizingLineFormatter : public LineFormatter {
 
       // Cut off the analysis of certain solutions if the analysis gets too
       // complex. See description of IgnoreStackForComparison.
-      if (Count > 50000)
+      if (Count > 50'000)
         Node->State.IgnoreStackForComparison = true;
 
       if (!Seen.insert(&Node->State).second) {

diff --git a/clang/lib/Format/WhitespaceManager.cpp b/clang/lib/Format/WhitespaceManager.cpp
@@ -111,6 +111,8 @@ const tooling::Replacements &WhitespaceManager::generateReplacements() {
   alignConsecutiveDeclarations();
   alignConsecutiveBitFields();
   alignConsecutiveAssignments();
+  if (Style.isTableGen())
+    alignConsecutiveTableGenCondOperatorColons();
   alignChainedConditionals();
   alignTrailingComments();
   alignEscapedNewlines();
@@ -849,7 +851,12 @@ void WhitespaceManager::alignConsecutiveAssignments() {
 }
 
 void WhitespaceManager::alignConsecutiveBitFields() {
-  if (!Style.AlignConsecutiveBitFields.Enabled)
+  alignConsecutiveColons(Style.AlignConsecutiveBitFields, TT_BitFieldColon);
+}
+
+void WhitespaceManager::alignConsecutiveColons(
+    const FormatStyle::AlignConsecutiveStyle &AlignStyle, TokenType Type) {
+  if (!AlignStyle.Enabled)
     return;
 
   AlignTokens(
@@ -863,9 +870,9 @@ void WhitespaceManager::alignConsecutiveBitFields() {
         if (&C != &Changes.back() && (&C + 1)->NewlinesBefore > 0)
           return false;
 
-        return C.Tok->is(TT_BitFieldColon);
+        return C.Tok->is(Type);
       },
-      Changes, /*StartAt=*/0, Style.AlignConsecutiveBitFields);
+      Changes, /*StartAt=*/0, AlignStyle);
 }
 
 void WhitespaceManager::alignConsecutiveShortCaseStatements() {
@@ -972,6 +979,11 @@ void WhitespaceManager::alignConsecutiveShortCaseStatements() {
                              Changes);
 }
 
+void WhitespaceManager::alignConsecutiveTableGenCondOperatorColons() {
+  alignConsecutiveColons(Style.AlignConsecutiveTableGenCondOperatorColons,
+                         TT_TableGenCondOperatorColon);
+}
+
 void WhitespaceManager::alignConsecutiveDeclarations() {
   if (!Style.AlignConsecutiveDeclarations.Enabled)
     return;

diff --git a/clang/lib/Format/WhitespaceManager.h b/clang/lib/Format/WhitespaceManager.h
@@ -226,6 +226,11 @@ class WhitespaceManager {
   /// Align consecutive bitfields over all \c Changes.
   void alignConsecutiveBitFields();
 
+  /// Align consecutive colon. For bitfields, TableGen DAGArgs and defintions.
+  void
+  alignConsecutiveColons(const FormatStyle::AlignConsecutiveStyle &AlignStyle,
+                         TokenType Type);
+
   /// Align consecutive declarations over all \c Changes.
   void alignConsecutiveDeclarations();
 
@@ -235,6 +240,9 @@ class WhitespaceManager {
   /// Align consecutive short case statements over all \c Changes.
   void alignConsecutiveShortCaseStatements();
 
+  /// Align consecutive TableGen cond operator colon over all \c Changes.
+  void alignConsecutiveTableGenCondOperatorColons();
+
   /// Align trailing comments over all \c Changes.
   void alignTrailingComments();
 

diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -179,6 +179,104 @@ double3 cos(double3);
 _HLSL_BUILTIN_ALIAS(__builtin_elementwise_cos)
 double4 cos(double4);
 
+//===----------------------------------------------------------------------===//
+// dot product builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t4, uint16_t4);
+#endif
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+float dot(float, float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+float dot(float2, float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+float dot(float3, float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+float dot(float4, float4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+double dot(double, double);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int dot(int, int);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int dot(int2, int2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int dot(int3, int3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int dot(int4, int4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint dot(uint, uint);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint dot(uint2, uint2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint dot(uint3, uint3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint dot(uint4, uint4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int64_t dot(int64_t, int64_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int64_t dot(int64_t2, int64_t2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int64_t dot(int64_t3, int64_t3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int64_t dot(int64_t4, int64_t4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint64_t dot(uint64_t, uint64_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint64_t dot(uint64_t2, uint64_t2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint64_t dot(uint64_t3, uint64_t3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint64_t dot(uint64_t4, uint64_t4);
+
 //===----------------------------------------------------------------------===//
 // floor builtins
 //===----------------------------------------------------------------------===//

diff --git a/clang/lib/Sema/CMakeLists.txt b/clang/lib/Sema/CMakeLists.txt
@@ -27,6 +27,7 @@ add_clang_library(clangSema
   Sema.cpp
   SemaAccess.cpp
   SemaAttr.cpp
+  SemaAPINotes.cpp
   SemaAvailability.cpp
   SemaCXXScopeSpec.cpp
   SemaCast.cpp

diff --git a/clang/lib/Sema/SemaAPINotes.cpp b/clang/lib/Sema/SemaAPINotes.cpp
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
@@ -2120,10 +2120,11 @@ bool Sema::CheckTSBuiltinFunctionCall(const TargetInfo &TI, unsigned BuiltinID,
 // not a valid type, emit an error message and return true. Otherwise return
 // false.
 static bool checkMathBuiltinElementType(Sema &S, SourceLocation Loc,
-                                        QualType Ty) {
-  if (!Ty->getAs<VectorType>() && !ConstantMatrixType::isValidElementType(Ty)) {
+                                        QualType ArgTy, int ArgIndex) {
+  if (!ArgTy->getAs<VectorType>() &&
+      !ConstantMatrixType::isValidElementType(ArgTy)) {
     return S.Diag(Loc, diag::err_builtin_invalid_arg_type)
-           << 1 << /* vector, integer or float ty*/ 0 << Ty;
+           << ArgIndex << /* vector, integer or float ty*/ 0 << ArgTy;
   }
 
   return false;
@@ -2961,6 +2962,9 @@ Sema::CheckBuiltinFunctionCall(FunctionDecl *FDecl, unsigned BuiltinID,
   }
   }
 
+  if (getLangOpts().HLSL && CheckHLSLBuiltinFunctionCall(BuiltinID, TheCall))
+    return ExprError();
+
   // Since the target specific builtins for each arch overlap, only check those
   // of the arch we are compiling for.
   if (Context.BuiltinInfo.isTSBuiltin(BuiltinID)) {
@@ -5161,6 +5165,70 @@ bool Sema::CheckPPCMMAType(QualType Type, SourceLocation TypeLoc) {
   return false;
 }
 
+// Helper function for CheckHLSLBuiltinFunctionCall
+bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall) {
+  assert(TheCall->getNumArgs() > 1);
+  ExprResult A = TheCall->getArg(0);
+  ExprResult B = TheCall->getArg(1);
+  QualType ArgTyA = A.get()->getType();
+  QualType ArgTyB = B.get()->getType();
+  auto *VecTyA = ArgTyA->getAs<VectorType>();
+  auto *VecTyB = ArgTyB->getAs<VectorType>();
+  SourceLocation BuiltinLoc = TheCall->getBeginLoc();
+  if (VecTyA == nullptr && VecTyB == nullptr)
+    return false;
+
+  if (VecTyA && VecTyB) {
+    bool retValue = false;
+    if (VecTyA->getElementType() != VecTyB->getElementType()) {
+      // Note: type promotion is intended to be handeled via the intrinsics
+      //  and not the builtin itself.
+      S->Diag(TheCall->getBeginLoc(), diag::err_vec_builtin_incompatible_vector)
+          << TheCall->getDirectCallee()
+          << SourceRange(A.get()->getBeginLoc(), B.get()->getEndLoc());
+      retValue = true;
+    }
+    if (VecTyA->getNumElements() != VecTyB->getNumElements()) {
+      // if we get here a HLSLVectorTruncation is needed.
+      S->Diag(BuiltinLoc, diag::err_vec_builtin_incompatible_vector)
+          << TheCall->getDirectCallee()
+          << SourceRange(TheCall->getArg(0)->getBeginLoc(),
+                         TheCall->getArg(1)->getEndLoc());
+      retValue = true;
+    }
+
+    if (retValue)
+      TheCall->setType(VecTyA->getElementType());
+
+    return retValue;
+  }
+
+  // Note: if we get here one of the args is a scalar which
+  // requires a VectorSplat on Arg0 or Arg1
+  S->Diag(BuiltinLoc, diag::err_vec_builtin_non_vector)
+      << TheCall->getDirectCallee()
+      << SourceRange(TheCall->getArg(0)->getBeginLoc(),
+                     TheCall->getArg(1)->getEndLoc());
+  return true;
+}
+
+// Note: returning true in this case results in CheckBuiltinFunctionCall
+// returning an ExprError
+bool Sema::CheckHLSLBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
+  switch (BuiltinID) {
+  case Builtin::BI__builtin_hlsl_dot: {
+    if (checkArgCount(*this, TheCall, 2))
+      return true;
+    if (CheckVectorElementCallArgs(this, TheCall))
+      return true;
+    if (SemaBuiltinVectorToScalarMath(TheCall))
+      return true;
+    break;
+  }
+  }
+  return false;
+}
+
 bool Sema::CheckAMDGCNBuiltinFunctionCall(unsigned BuiltinID,
                                           CallExpr *TheCall) {
   // position of memory order and scope arguments in the builtin
@@ -19594,23 +19662,43 @@ bool Sema::PrepareBuiltinElementwiseMathOneArgCall(CallExpr *TheCall) {
   TheCall->setArg(0, A.get());
   QualType TyA = A.get()->getType();
 
-  if (checkMathBuiltinElementType(*this, A.get()->getBeginLoc(), TyA))
+  if (checkMathBuiltinElementType(*this, A.get()->getBeginLoc(), TyA, 1))
     return true;
 
   TheCall->setType(TyA);
   return false;
 }
 
 bool Sema::SemaBuiltinElementwiseMath(CallExpr *TheCall) {
+  QualType Res;
+  if (SemaBuiltinVectorMath(TheCall, Res))
+    return true;
+  TheCall->setType(Res);
+  return false;
+}
+
+bool Sema::SemaBuiltinVectorToScalarMath(CallExpr *TheCall) {
+  QualType Res;
+  if (SemaBuiltinVectorMath(TheCall, Res))
+    return true;
+
+  if (auto *VecTy0 = Res->getAs<VectorType>())
+    TheCall->setType(VecTy0->getElementType());
+  else
+    TheCall->setType(Res);
+
+  return false;
+}
+
+bool Sema::SemaBuiltinVectorMath(CallExpr *TheCall, QualType &Res) {
   if (checkArgCount(*this, TheCall, 2))
     return true;
 
   ExprResult A = TheCall->getArg(0);
   ExprResult B = TheCall->getArg(1);
   // Do standard promotions between the two arguments, returning their common
   // type.
-  QualType Res =
-      UsualArithmeticConversions(A, B, TheCall->getExprLoc(), ACK_Comparison);
+  Res = UsualArithmeticConversions(A, B, TheCall->getExprLoc(), ACK_Comparison);
   if (A.isInvalid() || B.isInvalid())
     return true;
 
@@ -19622,12 +19710,11 @@ bool Sema::SemaBuiltinElementwiseMath(CallExpr *TheCall) {
                 diag::err_typecheck_call_different_arg_types)
            << TyA << TyB;
 
-  if (checkMathBuiltinElementType(*this, A.get()->getBeginLoc(), TyA))
+  if (checkMathBuiltinElementType(*this, A.get()->getBeginLoc(), TyA, 1))
     return true;
 
   TheCall->setArg(0, A.get());
   TheCall->setArg(1, B.get());
-  TheCall->setType(Res);
   return false;
 }
 

diff --git a/clang/lib/Sema/SemaDecl.cpp b/clang/lib/Sema/SemaDecl.cpp
@@ -16424,6 +16424,7 @@ void Sema::ActOnFinishDelayedAttribute(Scope *S, Decl *D,
   if (TemplateDecl *TD = dyn_cast<TemplateDecl>(D))
     D = TD->getTemplatedDecl();
   ProcessDeclAttributeList(S, D, Attrs);
+  ProcessAPINotes(D);
 
   if (CXXMethodDecl *Method = dyn_cast_or_null<CXXMethodDecl>(D))
     if (Method->isStatic())
@@ -19683,6 +19684,7 @@ void Sema::ActOnFields(Scope *S, SourceLocation RecLoc, Decl *EnclosingDecl,
       CDecl->setIvarRBraceLoc(RBrac);
     }
   }
+  ProcessAPINotes(Record);
 }
 
 /// Determine whether the given integral value is representable within
@@ -19997,6 +19999,7 @@ Decl *Sema::ActOnEnumConstant(Scope *S, Decl *theEnumDecl, Decl *lastEnumConst,
   // Process attributes.
   ProcessDeclAttributeList(S, New, Attrs);
   AddPragmaAttributes(S, New);
+  ProcessAPINotes(New);
 
   // Register this decl in the current scope stack.
   New->setAccess(TheEnumDecl->getAccess());
@@ -20195,6 +20198,7 @@ void Sema::ActOnEnumBody(SourceLocation EnumLoc, SourceRange BraceRange,
   QualType EnumType = Context.getTypeDeclType(Enum);
 
   ProcessDeclAttributeList(S, Enum, Attrs);
+  ProcessAPINotes(Enum);
 
   if (Enum->isDependentType()) {
     for (unsigned i = 0, e = Elements.size(); i != e; ++i) {

diff --git a/clang/lib/Sema/SemaDeclAttr.cpp b/clang/lib/Sema/SemaDeclAttr.cpp
@@ -10146,6 +10146,9 @@ void Sema::ProcessDeclAttributes(Scope *S, Decl *D, const Declarator &PD) {
 
   // Apply additional attributes specified by '#pragma clang attribute'.
   AddPragmaAttributes(S, D);
+
+  // Look for API notes that map to attributes.
+  ProcessAPINotes(D);
 }
 
 /// Is the given declaration allowed to use a forbidden type?

diff --git a/clang/lib/Sema/SemaDeclCXX.cpp b/clang/lib/Sema/SemaDeclCXX.cpp
@@ -11724,6 +11724,7 @@ Decl *Sema::ActOnStartNamespaceDef(Scope *NamespcScope,
 
   ProcessDeclAttributeList(DeclRegionScope, Namespc, AttrList);
   AddPragmaAttributes(DeclRegionScope, Namespc);
+  ProcessAPINotes(Namespc);
 
   // FIXME: Should we be merging attributes?
   if (const VisibilityAttr *Attr = Namespc->getAttr<VisibilityAttr>())
@@ -12270,8 +12271,10 @@ Decl *Sema::ActOnUsingDirective(Scope *S, SourceLocation UsingLoc,
     Diag(IdentLoc, diag::err_expected_namespace_name) << SS.getRange();
   }
 
-  if (UDir)
+  if (UDir) {
     ProcessDeclAttributeList(S, UDir, AttrList);
+    ProcessAPINotes(UDir);
+  }
 
   return UDir;
 }
@@ -13563,6 +13566,7 @@ Decl *Sema::ActOnAliasDeclaration(Scope *S, AccessSpecifier AS,
 
   ProcessDeclAttributeList(S, NewTD, AttrList);
   AddPragmaAttributes(S, NewTD);
+  ProcessAPINotes(NewTD);
 
   CheckTypedefForVariablyModifiedType(S, NewTD);
   Invalid |= NewTD->isInvalidDecl();

diff --git a/clang/lib/Sema/SemaDeclObjC.cpp b/clang/lib/Sema/SemaDeclObjC.cpp
@@ -1072,6 +1072,7 @@ ObjCInterfaceDecl *Sema::ActOnStartClassInterface(
 
   ProcessDeclAttributeList(TUScope, IDecl, AttrList);
   AddPragmaAttributes(TUScope, IDecl);
+  ProcessAPINotes(IDecl);
 
   // Merge attributes from previous declarations.
   if (PrevIDecl)
@@ -1273,6 +1274,7 @@ ObjCProtocolDecl *Sema::ActOnStartProtocolInterface(
 
   ProcessDeclAttributeList(TUScope, PDecl, AttrList);
   AddPragmaAttributes(TUScope, PDecl);
+  ProcessAPINotes(PDecl);
 
   // Merge attributes from previous declarations.
   if (PrevDecl)
@@ -4807,6 +4809,7 @@ Decl *Sema::ActOnMethodDeclaration(
     // Apply the attributes to the parameter.
     ProcessDeclAttributeList(TUScope, Param, ArgInfo[i].ArgAttrs);
     AddPragmaAttributes(TUScope, Param);
+    ProcessAPINotes(Param);
 
     if (Param->hasAttr<BlocksAttr>()) {
       Diag(Param->getLocation(), diag::err_block_on_nonlocal);
@@ -4837,6 +4840,7 @@ Decl *Sema::ActOnMethodDeclaration(
 
   ProcessDeclAttributeList(TUScope, ObjCMethod, AttrList);
   AddPragmaAttributes(TUScope, ObjCMethod);
+  ProcessAPINotes(ObjCMethod);
 
   // Add the method now.
   const ObjCMethodDecl *PrevMethod = nullptr;

diff --git a/clang/lib/Sema/SemaObjCProperty.cpp b/clang/lib/Sema/SemaObjCProperty.cpp
@@ -2509,6 +2509,8 @@ void Sema::ProcessPropertyDecl(ObjCPropertyDecl *property) {
       GetterMethod->addAttr(SectionAttr::CreateImplicit(
           Context, SA->getName(), Loc, SectionAttr::GNU_section));
 
+    ProcessAPINotes(GetterMethod);
+
     if (getLangOpts().ObjCAutoRefCount)
       CheckARCMethodDecl(GetterMethod);
   } else
@@ -2578,6 +2580,9 @@ void Sema::ProcessPropertyDecl(ObjCPropertyDecl *property) {
       if (const SectionAttr *SA = property->getAttr<SectionAttr>())
         SetterMethod->addAttr(SectionAttr::CreateImplicit(
             Context, SA->getName(), Loc, SectionAttr::GNU_section));
+
+      ProcessAPINotes(SetterMethod);
+
       // It's possible for the user to have set a very odd custom
       // setter selector that causes it to have a method family.
       if (getLangOpts().ObjCAutoRefCount)

diff --git a/clang/lib/Sema/SemaStmt.cpp b/clang/lib/Sema/SemaStmt.cpp
@@ -3361,8 +3361,9 @@ Sema::ActOnContinueStmt(SourceLocation ContinueLoc, Scope *CurScope) {
   // of a compute construct counts as 'branching out of' the compute construct,
   // so diagnose here.
   if (S->isOpenACCComputeConstructScope())
-    return StmtError(Diag(ContinueLoc, diag::err_acc_branch_in_out)
-                     << /*out of */ 0);
+    return StmtError(
+        Diag(ContinueLoc, diag::err_acc_branch_in_out_compute_construct)
+        << /*branch*/ 0 << /*out of */ 0);
 
   CheckJumpOutOfSEHFinally(*this, ContinueLoc, *S);
 
@@ -3390,8 +3391,9 @@ Sema::ActOnBreakStmt(SourceLocation BreakLoc, Scope *CurScope) {
   if (S->isOpenACCComputeConstructScope() ||
       (S->isLoopScope() && S->getParent() &&
        S->getParent()->isOpenACCComputeConstructScope()))
-    return StmtError(Diag(BreakLoc, diag::err_acc_branch_in_out)
-                     << /*out of */ 0);
+    return StmtError(
+        Diag(BreakLoc, diag::err_acc_branch_in_out_compute_construct)
+        << /*branch*/ 0 << /*out of */ 0);
 
   CheckJumpOutOfSEHFinally(*this, BreakLoc, *S);
 
@@ -3947,6 +3949,12 @@ Sema::ActOnReturnStmt(SourceLocation ReturnLoc, Expr *RetValExp,
       RetValExp, nullptr, /*RecoverUncorrectedTypos=*/true);
   if (RetVal.isInvalid())
     return StmtError();
+
+  if (getCurScope()->isInOpenACCComputeConstructScope())
+    return StmtError(
+        Diag(ReturnLoc, diag::err_acc_branch_in_out_compute_construct)
+        << /*return*/ 1 << /*out of */ 0);
+
   StmtResult R =
       BuildReturnStmt(ReturnLoc, RetVal.get(), /*AllowRecovery=*/true);
   if (R.isInvalid() || ExprEvalContexts.back().isDiscardedStatementContext())

diff --git a/clang/lib/Sema/SemaTemplate.cpp b/clang/lib/Sema/SemaTemplate.cpp
@@ -2158,6 +2158,7 @@ DeclResult Sema::CheckClassTemplate(
     NewClass->startDefinition();
 
   ProcessDeclAttributeList(S, NewClass, Attr);
+  ProcessAPINotes(NewClass);
 
   if (PrevClassTemplate)
     mergeDeclAttributes(NewClass, PrevClassTemplate->getTemplatedDecl());
@@ -9112,6 +9113,7 @@ DeclResult Sema::ActOnClassTemplateSpecialization(
   }
 
   ProcessDeclAttributeList(S, Specialization, Attr);
+  ProcessAPINotes(Specialization);
 
   // Add alignment attributes if necessary; these attributes are checked when
   // the ASTContext lays out the structure.
@@ -9707,6 +9709,40 @@ bool Sema::CheckFunctionTemplateSpecialization(
   // Ignore access information;  it doesn't figure into redeclaration checking.
   FunctionDecl *Specialization = cast<FunctionDecl>(*Result);
 
+  // C++23 [except.spec]p13:
+  //   An exception specification is considered to be needed when:
+  //   - [...]
+  //   - the exception specification is compared to that of another declaration
+  //     (e.g., an explicit specialization or an overriding virtual function);
+  //   - [...]
+  //
+  //  The exception specification of a defaulted function is evaluated as
+  //  described above only when needed; similarly, the noexcept-specifier of a
+  //  specialization of a function template or member function of a class
+  //  template is instantiated only when needed.
+  //
+  // The standard doesn't specify what the "comparison with another declaration"
+  // entails, nor the exact circumstances in which it occurs. Moreover, it does
+  // not state which properties of an explicit specialization must match the
+  // primary template.
+  //
+  // We assume that an explicit specialization must correspond with (per
+  // [basic.scope.scope]p4) and declare the same entity as (per [basic.link]p8)
+  // the declaration produced by substitution into the function template.
+  //
+  // Since the determination whether two function declarations correspond does
+  // not consider exception specification, we only need to instantiate it once
+  // we determine the primary template when comparing types per
+  // [basic.link]p11.1.
+  auto *SpecializationFPT =
+      Specialization->getType()->castAs<FunctionProtoType>();
+  // If the function has a dependent exception specification, resolve it after
+  // we have selected the primary template so we can check whether it matches.
+  if (getLangOpts().CPlusPlus17 &&
+      isUnresolvedExceptionSpec(SpecializationFPT->getExceptionSpecType()) &&
+      !ResolveExceptionSpec(FD->getLocation(), SpecializationFPT))
+    return true;
+
   FunctionTemplateSpecializationInfo *SpecInfo
     = Specialization->getTemplateSpecializationInfo();
   assert(SpecInfo && "Function template specialization info missing?");
@@ -10346,6 +10382,7 @@ DeclResult Sema::ActOnExplicitInstantiation(
 
   bool PreviouslyDLLExported = Specialization->hasAttr<DLLExportAttr>();
   ProcessDeclAttributeList(S, Specialization, Attr);
+  ProcessAPINotes(Specialization);
 
   // Add the explicit instantiation into its lexical context. However,
   // since explicit instantiations are never found by name lookup, we
@@ -10756,6 +10793,9 @@ DeclResult Sema::ActOnExplicitInstantiation(Scope *S,
       Prev->setTemplateSpecializationKind(TSK, D.getIdentifierLoc());
       // Merge attributes.
       ProcessDeclAttributeList(S, Prev, D.getDeclSpec().getAttributes());
+      if (PrevTemplate)
+        ProcessAPINotes(Prev);
+
       if (TSK == TSK_ExplicitInstantiationDefinition)
         InstantiateVariableDefinition(D.getIdentifierLoc(), Prev);
     }
@@ -10931,6 +10971,7 @@ DeclResult Sema::ActOnExplicitInstantiation(Scope *S,
   }
 
   ProcessDeclAttributeList(S, Specialization, D.getDeclSpec().getAttributes());
+  ProcessAPINotes(Specialization);
 
   // In MSVC mode, dllimported explicit instantiation definitions are treated as
   // instantiation declarations.

diff --git a/clang/lib/Sema/SemaTemplateDeduction.cpp b/clang/lib/Sema/SemaTemplateDeduction.cpp
@@ -4632,11 +4632,9 @@ TemplateDeductionResult Sema::DeduceTemplateArguments(
                                                Info.getLocation()))
     return TemplateDeductionResult::MiscellaneousDeductionFailure;
 
-  // If the function has a dependent exception specification, resolve it now,
-  // so we can check that the exception specification matches.
   auto *SpecializationFPT =
       Specialization->getType()->castAs<FunctionProtoType>();
-  if (getLangOpts().CPlusPlus17 &&
+  if (IsAddressOfFunction && getLangOpts().CPlusPlus17 &&
       isUnresolvedExceptionSpec(SpecializationFPT->getExceptionSpecType()) &&
       !ResolveExceptionSpec(Info.getLocation(), SpecializationFPT))
     return TemplateDeductionResult::MiscellaneousDeductionFailure;
@@ -4662,11 +4660,11 @@ TemplateDeductionResult Sema::DeduceTemplateArguments(
   // specialization with respect to arguments of compatible pointer to function
   // types, template argument deduction fails.
   if (!ArgFunctionType.isNull()) {
-    if (IsAddressOfFunction
-            ? !isSameOrCompatibleFunctionType(
-                  Context.getCanonicalType(SpecializationType),
-                  Context.getCanonicalType(ArgFunctionType))
-            : !Context.hasSameType(SpecializationType, ArgFunctionType)) {
+    if (IsAddressOfFunction ? !isSameOrCompatibleFunctionType(
+                                  Context.getCanonicalType(SpecializationType),
+                                  Context.getCanonicalType(ArgFunctionType))
+                            : !Context.hasSameFunctionTypeIgnoringExceptionSpec(
+                                  SpecializationType, ArgFunctionType)) {
       Info.FirstArg = TemplateArgument(SpecializationType);
       Info.SecondArg = TemplateArgument(ArgFunctionType);
       return TemplateDeductionResult::NonDeducedMismatch;