SemaHLSL.cpp

Go to the documentation of this file.

//===- SemaHLSL.cpp - Semantic Analysis for HLSL constructs ---------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
// This implements Semantic Analysis for HLSL constructs.
//===----------------------------------------------------------------------===//
 
#include "clang/Sema/SemaHLSL.h"
#include "clang/AST/ASTConsumer.h"
#include "clang/AST/ASTContext.h"
#include "clang/AST/Attr.h"
#include "clang/AST/Decl.h"
#include "clang/AST/DeclBase.h"
#include "clang/AST/DeclCXX.h"
#include "clang/AST/DeclarationName.h"
#include "clang/AST/DynamicRecursiveASTVisitor.h"
#include "clang/AST/Expr.h"
#include "clang/AST/HLSLResource.h"
#include "clang/AST/MatrixUtils.h"
#include "clang/AST/Type.h"
#include "clang/AST/TypeBase.h"
#include "clang/AST/TypeLoc.h"
#include "clang/Basic/Builtins.h"
#include "clang/Basic/DiagnosticSema.h"
#include "clang/Basic/IdentifierTable.h"
#include "clang/Basic/LLVM.h"
#include "clang/Basic/SourceLocation.h"
#include "clang/Basic/Specifiers.h"
#include "clang/Basic/TargetInfo.h"
#include "clang/Sema/Initialization.h"
#include "clang/Sema/Lookup.h"
#include "clang/Sema/ParsedAttr.h"
#include "clang/Sema/Sema.h"
#include "clang/Sema/Template.h"
#include "llvm/ADT/ArrayRef.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/StringRef.h"
#include "llvm/ADT/Twine.h"
#include "llvm/Frontend/HLSL/HLSLBinding.h"
#include "llvm/Frontend/HLSL/RootSignatureValidations.h"
#include "llvm/Support/Casting.h"
#include "llvm/Support/DXILABI.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/FormatVariadic.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/TargetParser/Triple.h"
#include <algorithm>
#include <cmath>
#include <cstddef>
#include <iterator>
#include <utility>
 
using namespace clang;
using namespace clang::hlsl;
using llvm::hlsl::InterpolationModifier;
using llvm::hlsl::IOType;
using llvm::hlsl::SemanticStageInfo;
using SemanticKind = llvm::dxbc::PSV::SemanticKind;
using RegisterType = HLSLResourceBindingAttr::RegisterType;
 
static CXXRecordDecl *createHostLayoutStruct(Sema &S,
                                             CXXRecordDecl *StructDecl);
 
static QualType getScalarComponentType(QualType T) {
  if (const auto *VT = T->getAs<VectorType>())
    return VT->getElementType();
  if (const auto *MT = T->getAs<MatrixType>())
    return MT->getElementType();
  return T;
}
 
static RegisterType getRegisterType(ResourceClass RC) {
  switch (RC) {
  case ResourceClass::SRV:
    return RegisterType::SRV;
  case ResourceClass::UAV:
    return RegisterType::UAV;
  case ResourceClass::CBuffer:
    return RegisterType::CBuffer;
  case ResourceClass::Sampler:
    return RegisterType::Sampler;
  }
  llvm_unreachable("unexpected ResourceClass value");
}
 
static RegisterType getRegisterType(const HLSLAttributedResourceType *ResTy) {
  return getRegisterType(ResTy->getAttrs().ResourceClass);
}
 
static LangAS getLangASFromResourceClass(ResourceClass RC) {
  switch (RC) {
  case ResourceClass::SRV:
  case ResourceClass::UAV:
    return LangAS::hlsl_device;
  case ResourceClass::CBuffer:
    return LangAS::hlsl_constant;
  case ResourceClass::Sampler:
    return LangAS::hlsl_device;
  }
  llvm_unreachable("unexpected ResourceClass value");
}
 
// Converts the first letter of string Slot to RegisterType.
// Returns false if the letter does not correspond to a valid register type.
static bool convertToRegisterType(StringRef Slot, RegisterType *RT) {
  assert(RT != nullptr);
  switch (Slot[0]) {
  case 't':
  case 'T':
    *RT = RegisterType::SRV;
    return true;
  case 'u':
  case 'U':
    *RT = RegisterType::UAV;
    return true;
  case 'b':
  case 'B':
    *RT = RegisterType::CBuffer;
    return true;
  case 's':
  case 'S':
    *RT = RegisterType::Sampler;
    return true;
  case 'c':
  case 'C':
    *RT = RegisterType::C;
    return true;
  case 'i':
  case 'I':
    *RT = RegisterType::I;
    return true;
  default:
    return false;
  }
}
 
static char getRegisterTypeChar(RegisterType RT) {
  switch (RT) {
  case RegisterType::SRV:
    return 't';
  case RegisterType::UAV:
    return 'u';
  case RegisterType::CBuffer:
    return 'b';
  case RegisterType::Sampler:
    return 's';
  case RegisterType::C:
    return 'c';
  case RegisterType::I:
    return 'i';
  }
  llvm_unreachable("unexpected RegisterType value");
}
 
static ResourceClass getResourceClass(RegisterType RT) {
  switch (RT) {
  case RegisterType::SRV:
    return ResourceClass::SRV;
  case RegisterType::UAV:
    return ResourceClass::UAV;
  case RegisterType::CBuffer:
    return ResourceClass::CBuffer;
  case RegisterType::Sampler:
    return ResourceClass::Sampler;
  case RegisterType::C:
  case RegisterType::I:
    // Deliberately falling through to the unreachable below.
    break;
  }
  llvm_unreachable("unexpected RegisterType value");
}
 
static Builtin::ID getSpecConstBuiltinId(const Type *Type) {
  const auto *BT = dyn_cast<BuiltinType>(Type);
  if (!BT) {
    if (!Type->isEnumeralType())
      return Builtin::NotBuiltin;
    return Builtin::BI__builtin_get_spirv_spec_constant_int;
  }
 
  switch (BT->getKind()) {
  case BuiltinType::Bool:
    return Builtin::BI__builtin_get_spirv_spec_constant_bool;
  case BuiltinType::Short:
    return Builtin::BI__builtin_get_spirv_spec_constant_short;
  case BuiltinType::Int:
    return Builtin::BI__builtin_get_spirv_spec_constant_int;
  case BuiltinType::LongLong:
    return Builtin::BI__builtin_get_spirv_spec_constant_longlong;
  case BuiltinType::UShort:
    return Builtin::BI__builtin_get_spirv_spec_constant_ushort;
  case BuiltinType::UInt:
    return Builtin::BI__builtin_get_spirv_spec_constant_uint;
  case BuiltinType::ULongLong:
    return Builtin::BI__builtin_get_spirv_spec_constant_ulonglong;
  case BuiltinType::Half:
    return Builtin::BI__builtin_get_spirv_spec_constant_half;
  case BuiltinType::Float:
    return Builtin::BI__builtin_get_spirv_spec_constant_float;
  case BuiltinType::Double:
    return Builtin::BI__builtin_get_spirv_spec_constant_double;
  default:
    return Builtin::NotBuiltin;
  }
}
 
static StringRef createRegisterString(ASTContext &AST, RegisterType RegType,
                                      unsigned N) {
  llvm::SmallString<16> Buffer;
  llvm::raw_svector_ostream OS(Buffer);
  OS << getRegisterTypeChar(RegType);
  OS << N;
  return AST.backupStr(OS.str());
}
 
DeclBindingInfo *ResourceBindings::addDeclBindingInfo(const VarDecl *VD,
                                                      ResourceClass ResClass) {
  assert(getDeclBindingInfo(VD, ResClass) == nullptr &&
         "DeclBindingInfo already added");
  assert(!hasBindingInfoForDecl(VD) || BindingsList.back().Decl == VD);
  // VarDecl may have multiple entries for different resource classes.
  // DeclToBindingListIndex stores the index of the first binding we saw
  // for this decl. If there are any additional ones then that index
  // shouldn't be updated.
  DeclToBindingListIndex.try_emplace(VD, BindingsList.size());
  return &BindingsList.emplace_back(VD, ResClass);
}
 
DeclBindingInfo *ResourceBindings::getDeclBindingInfo(const VarDecl *VD,
                                                      ResourceClass ResClass) {
  auto Entry = DeclToBindingListIndex.find(VD);
  if (Entry != DeclToBindingListIndex.end()) {
    for (unsigned Index = Entry->getSecond();
         Index < BindingsList.size() && BindingsList[Index].Decl == VD;
         ++Index) {
      if (BindingsList[Index].ResClass == ResClass)
        return &BindingsList[Index];
    }
  }
  return nullptr;
}
 
bool ResourceBindings::hasBindingInfoForDecl(const VarDecl *VD) const {
  return DeclToBindingListIndex.contains(VD);
}
 
SemaHLSL::SemaHLSL(Sema &S) : SemaBase(S) {}
 
Decl *SemaHLSL::ActOnStartBuffer(Scope *BufferScope, bool CBuffer,
                                 SourceLocation KwLoc, IdentifierInfo *Ident,
                                 SourceLocation IdentLoc,
                                 SourceLocation LBrace) {
  // For anonymous namespace, take the location of the left brace.
  DeclContext *LexicalParent = SemaRef.getCurLexicalContext();
  HLSLBufferDecl *Result = HLSLBufferDecl::Create(
      getASTContext(), LexicalParent, CBuffer, KwLoc, Ident, IdentLoc, LBrace);
 
  // if CBuffer is false, then it's a TBuffer
  auto RC = CBuffer ? llvm::hlsl::ResourceClass::CBuffer
                    : llvm::hlsl::ResourceClass::SRV;
  Result->addAttr(HLSLResourceClassAttr::CreateImplicit(getASTContext(), RC));
 
  SemaRef.PushOnScopeChains(Result, BufferScope);
  SemaRef.PushDeclContext(BufferScope, Result);
 
  return Result;
}
 
static unsigned calculateLegacyCbufferFieldAlign(const ASTContext &Context,
                                                 QualType T) {
  // Arrays, Matrices, and Structs are always aligned to new buffer rows
  if (T->isArrayType() || T->isStructureType() || T->isConstantMatrixType())
    return 16;
 
  // Vectors are aligned to the type they contain
  if (const VectorType *VT = T->getAs<VectorType>())
    return calculateLegacyCbufferFieldAlign(Context, VT->getElementType());
 
  assert(Context.getTypeSize(T) <= 64 &&
         "Scalar bit widths larger than 64 not supported");
 
  // Scalar types are aligned to their byte width
  return Context.getTypeSize(T) / 8;
}
 
// Calculate the size of a legacy cbuffer type in bytes based on
// https://learn.microsoft.com/en-us/windows/win32/direct3dhlsl/dx-graphics-hlsl-packing-rules
static unsigned calculateLegacyCbufferSize(const ASTContext &Context,
                                           QualType T) {
  constexpr unsigned CBufferAlign = 16;
  if (const auto *RD = T->getAsRecordDecl()) {
    unsigned Size = 0;
    for (const FieldDecl *Field : RD->fields()) {
      QualType Ty = Field->getType();
      unsigned FieldSize = calculateLegacyCbufferSize(Context, Ty);
      unsigned FieldAlign = calculateLegacyCbufferFieldAlign(Context, Ty);
 
      // If the field crosses the row boundary after alignment it drops to the
      // next row
      unsigned AlignSize = llvm::alignTo(Size, FieldAlign);
      if ((AlignSize % CBufferAlign) + FieldSize > CBufferAlign) {
        FieldAlign = CBufferAlign;
      }
 
      Size = llvm::alignTo(Size, FieldAlign);
      Size += FieldSize;
    }
    return Size;
  }
 
  if (const ConstantArrayType *AT = Context.getAsConstantArrayType(T)) {
    unsigned ElementCount = AT->getSize().getZExtValue();
    if (ElementCount == 0)
      return 0;
 
    unsigned ElementSize =
        calculateLegacyCbufferSize(Context, AT->getElementType());
    unsigned AlignedElementSize = llvm::alignTo(ElementSize, CBufferAlign);
    return AlignedElementSize * (ElementCount - 1) + ElementSize;
  }
 
  if (const VectorType *VT = T->getAs<VectorType>()) {
    unsigned ElementCount = VT->getNumElements();
    unsigned ElementSize =
        calculateLegacyCbufferSize(Context, VT->getElementType());
    return ElementSize * ElementCount;
  }
 
  return Context.getTypeSize(T) / 8;
}
 
// Validate packoffset:
// - if packoffset it used it must be set on all declarations inside the buffer
// - packoffset ranges must not overlap
static void validatePackoffset(Sema &S, HLSLBufferDecl *BufDecl) {
  llvm::SmallVector<std::pair<VarDecl *, HLSLPackOffsetAttr *>> PackOffsetVec;
 
  // Make sure the packoffset annotations are either on all declarations
  // or on none.
  bool HasPackOffset = false;
  bool HasNonPackOffset = false;
  for (auto *Field : BufDecl->buffer_decls()) {
    VarDecl *Var = dyn_cast<VarDecl>(Field);
    if (!Var)
      continue;
    if (Field->hasAttr<HLSLPackOffsetAttr>()) {
      PackOffsetVec.emplace_back(Var, Field->getAttr<HLSLPackOffsetAttr>());
      HasPackOffset = true;
    } else {
      HasNonPackOffset = true;
    }
  }
 
  if (!HasPackOffset)
    return;
 
  if (HasNonPackOffset)
    S.Diag(BufDecl->getLocation(), diag::warn_hlsl_packoffset_mix);
 
  // Make sure there is no overlap in packoffset - sort PackOffsetVec by offset
  // and compare adjacent values.
  bool IsValid = true;
  ASTContext &Context = S.getASTContext();
  std::sort(PackOffsetVec.begin(), PackOffsetVec.end(),
            [](const std::pair<VarDecl *, HLSLPackOffsetAttr *> &LHS,
               const std::pair<VarDecl *, HLSLPackOffsetAttr *> &RHS) {
              return LHS.second->getOffsetInBytes() <
                     RHS.second->getOffsetInBytes();
            });
  for (unsigned i = 0; i < PackOffsetVec.size() - 1; i++) {
    VarDecl *Var = PackOffsetVec[i].first;
    HLSLPackOffsetAttr *Attr = PackOffsetVec[i].second;
    unsigned Size = calculateLegacyCbufferSize(Context, Var->getType());
    unsigned Begin = Attr->getOffsetInBytes();
    unsigned End = Begin + Size;
    unsigned NextBegin = PackOffsetVec[i + 1].second->getOffsetInBytes();
    if (End > NextBegin) {
      VarDecl *NextVar = PackOffsetVec[i + 1].first;
      S.Diag(NextVar->getLocation(), diag::err_hlsl_packoffset_overlap)
          << NextVar << Var;
      IsValid = false;
    }
  }
  BufDecl->setHasValidPackoffset(IsValid);
}
 
// Returns true if the array has a zero size = if any of the dimensions is 0
static bool isZeroSizedArray(const ConstantArrayType *CAT) {
  while (CAT && !CAT->isZeroSize())
    CAT = dyn_cast<ConstantArrayType>(
        CAT->getElementType()->getUnqualifiedDesugaredType());
  return CAT != nullptr;
}
 
static bool isResourceRecordTypeOrArrayOf(QualType Ty) {
  return Ty->isHLSLResourceRecord() || Ty->isHLSLResourceRecordArray();
}
 
static bool isResourceRecordTypeOrArrayOf(VarDecl *VD) {
  return isResourceRecordTypeOrArrayOf(VD->getType());
}
 
static const HLSLAttributedResourceType *
getResourceArrayHandleType(QualType QT) {
  assert(QT->isHLSLResourceRecordArray() &&
         "expected array of resource records");
  const Type *Ty = QT->getUnqualifiedDesugaredType();
  while (const ArrayType *AT = dyn_cast<ArrayType>(Ty))
    Ty = AT->getArrayElementTypeNoTypeQual()->getUnqualifiedDesugaredType();
  return HLSLAttributedResourceType::findHandleTypeOnResource(Ty);
}
 
static const HLSLAttributedResourceType *
getResourceArrayHandleType(VarDecl *VD) {
  return getResourceArrayHandleType(VD->getType());
}
 
// Returns true if the type is a leaf element type that is not valid to be
// included in HLSL Buffer, such as a resource class, empty struct, zero-sized
// array, or a builtin intangible type. Returns false it is a valid leaf element
// type or if it is a record type that needs to be inspected further.
static bool isInvalidConstantBufferLeafElementType(const Type *Ty) {
  Ty = Ty->getUnqualifiedDesugaredType();
  if (Ty->isHLSLResourceRecord() || Ty->isHLSLResourceRecordArray())
    return true;
  if (const auto *RD = Ty->getAsCXXRecordDecl())
    return RD->isEmpty();
  if (Ty->isConstantArrayType() &&
      isZeroSizedArray(cast<ConstantArrayType>(Ty)))
    return true;
  if (Ty->isHLSLBuiltinIntangibleType() || Ty->isHLSLAttributedResourceType())
    return true;
  return false;
}
 
// Returns true if the struct contains at least one element that prevents it
// from being included inside HLSL Buffer as is, such as an intangible type,
// empty struct, or zero-sized array. If it does, a new implicit layout struct
// needs to be created for HLSL Buffer use that will exclude these unwanted
// declarations (see createHostLayoutStruct function).
static bool requiresImplicitBufferLayoutStructure(const CXXRecordDecl *RD) {
  if (RD->isHLSLIntangible() || RD->isEmpty())
    return true;
  // check fields
  for (const FieldDecl *Field : RD->fields()) {
    QualType Ty = Field->getType();
    if (isInvalidConstantBufferLeafElementType(Ty.getTypePtr()))
      return true;
    if (const auto *RD = Ty->getAsCXXRecordDecl();
        RD && requiresImplicitBufferLayoutStructure(RD))
      return true;
  }
  // check bases
  for (const CXXBaseSpecifier &Base : RD->bases())
    if (requiresImplicitBufferLayoutStructure(
            Base.getType()->castAsCXXRecordDecl()))
      return true;
  return false;
}
 
static CXXRecordDecl *findRecordDeclInContext(IdentifierInfo *II,
                                              DeclContext *DC) {
  CXXRecordDecl *RD = nullptr;
  for (NamedDecl *Decl :
       DC->getNonTransparentContext()->lookup(DeclarationName(II))) {
    if (CXXRecordDecl *FoundRD = dyn_cast<CXXRecordDecl>(Decl)) {
      assert(RD == nullptr &&
             "there should be at most 1 record by a given name in a scope");
      RD = FoundRD;
    }
  }
  return RD;
}
 
// Creates a name for buffer layout struct using the provide name base.
// If the name must be unique (not previously defined), a suffix is added
// until a unique name is found.
static IdentifierInfo *getHostLayoutStructName(Sema &S, NamedDecl *BaseDecl,
                                               bool MustBeUnique) {
  ASTContext &AST = S.getASTContext();
 
  IdentifierInfo *NameBaseII = BaseDecl->getIdentifier();
  llvm::SmallString<64> Name("__cblayout_");
  if (NameBaseII) {
    Name.append(NameBaseII->getName());
  } else {
    // anonymous struct
    Name.append("anon");
    MustBeUnique = true;
  }
 
  size_t NameLength = Name.size();
  IdentifierInfo *II = &AST.Idents.get(Name, tok::TokenKind::identifier);
  if (!MustBeUnique)
    return II;
 
  unsigned suffix = 0;
  while (true) {
    if (suffix != 0) {
      Name.append("_");
      Name.append(llvm::Twine(suffix).str());
      II = &AST.Idents.get(Name, tok::TokenKind::identifier);
    }
    if (!findRecordDeclInContext(II, BaseDecl->getDeclContext()))
      return II;
    // declaration with that name already exists - increment suffix and try
    // again until unique name is found
    suffix++;
    Name.truncate(NameLength);
  };
}
 
static const Type *createHostLayoutType(Sema &S, const Type *Ty) {
  ASTContext &AST = S.getASTContext();
  if (auto *RD = Ty->getAsCXXRecordDecl()) {
    if (!requiresImplicitBufferLayoutStructure(RD))
      return Ty;
    RD = createHostLayoutStruct(S, RD);
    if (!RD)
      return nullptr;
    return AST.getCanonicalTagType(RD)->getTypePtr();
  }
 
  if (const auto *CAT = dyn_cast<ConstantArrayType>(Ty)) {
    const Type *ElementTy = createHostLayoutType(
        S, CAT->getElementType()->getUnqualifiedDesugaredType());
    if (!ElementTy)
      return nullptr;
    return AST
        .getConstantArrayType(QualType(ElementTy, 0), CAT->getSize(), nullptr,
                              CAT->getSizeModifier(),
                              CAT->getIndexTypeCVRQualifiers())
        .getTypePtr();
  }
  return Ty;
}
 
// Returns the type to use for a host layout struct field. For most types this
// is the unqualified desugared type. Matrix types, however, retain their sugar
// so that the row_major/column_major orientation (carried as an AttributedType)
// is preserved; the orientation determines the in-memory cbuffer layout.
static const Type *getHostLayoutFieldType(QualType QT) {
  const Type *Desugared = QT->getUnqualifiedDesugaredType();
  if (Desugared->isConstantMatrixType())
    return QT.getTypePtr();
  return Desugared;
}
 
// Creates a field declaration of given name and type for HLSL buffer layout
// struct. Returns nullptr if the type cannot be use in HLSL Buffer layout.
static FieldDecl *createFieldForHostLayoutStruct(Sema &S, const Type *Ty,
                                                 IdentifierInfo *II,
                                                 CXXRecordDecl *LayoutStruct) {
  if (isInvalidConstantBufferLeafElementType(Ty))
    return nullptr;
 
  Ty = createHostLayoutType(S, Ty);
  if (!Ty)
    return nullptr;
 
  QualType QT = QualType(Ty, 0);
  ASTContext &AST = S.getASTContext();
  TypeSourceInfo *TSI = AST.getTrivialTypeSourceInfo(QT, SourceLocation());
  auto *Field = FieldDecl::Create(AST, LayoutStruct, SourceLocation(),
                                  SourceLocation(), II, QT, TSI, nullptr, false,
                                  InClassInitStyle::ICIS_NoInit);
  Field->setAccess(AccessSpecifier::AS_public);
  return Field;
}
 
// Creates host layout struct for a struct included in HLSL Buffer.
// The layout struct will include only fields that are allowed in HLSL buffer.
// These fields will be filtered out:
// - resource classes
// - empty structs
// - zero-sized arrays
// Returns nullptr if the resulting layout struct would be empty.
static CXXRecordDecl *createHostLayoutStruct(Sema &S,
                                             CXXRecordDecl *StructDecl) {
  assert(requiresImplicitBufferLayoutStructure(StructDecl) &&
         "struct is already HLSL buffer compatible");
 
  ASTContext &AST = S.getASTContext();
  DeclContext *DC = StructDecl->getDeclContext();
  IdentifierInfo *II = getHostLayoutStructName(S, StructDecl, false);
 
  // reuse existing if the layout struct if it already exists
  if (CXXRecordDecl *RD = findRecordDeclInContext(II, DC))
    return RD;
 
  CXXRecordDecl *LS =
      CXXRecordDecl::Create(AST, TagDecl::TagKind::Struct, DC, SourceLocation(),
                            SourceLocation(), II);
  LS->setImplicit(true);
  LS->addAttr(PackedAttr::CreateImplicit(AST));
  LS->startDefinition();
 
  // copy base struct, create HLSL Buffer compatible version if needed
  if (unsigned NumBases = StructDecl->getNumBases()) {
    assert(NumBases == 1 && "HLSL supports only one base type");
    (void)NumBases;
    CXXBaseSpecifier Base = *StructDecl->bases_begin();
    CXXRecordDecl *BaseDecl = Base.getType()->castAsCXXRecordDecl();
    if (requiresImplicitBufferLayoutStructure(BaseDecl)) {
      BaseDecl = createHostLayoutStruct(S, BaseDecl);
      if (BaseDecl) {
        TypeSourceInfo *TSI =
            AST.getTrivialTypeSourceInfo(AST.getCanonicalTagType(BaseDecl));
        Base = CXXBaseSpecifier(SourceRange(), false, StructDecl->isClass(),
                                AS_none, TSI, SourceLocation());
      }
    }
    if (BaseDecl) {
      const CXXBaseSpecifier *BasesArray[1] = {&Base};
      LS->setBases(BasesArray, 1);
    }
  }
 
  // filter struct fields
  for (const FieldDecl *FD : StructDecl->fields()) {
    const Type *Ty = getHostLayoutFieldType(FD->getType());
    if (FieldDecl *NewFD =
            createFieldForHostLayoutStruct(S, Ty, FD->getIdentifier(), LS))
      LS->addDecl(NewFD);
  }
  LS->completeDefinition();
 
  if (LS->field_empty() && LS->getNumBases() == 0)
    return nullptr;
 
  DC->addDecl(LS);
  return LS;
}
 
// Creates host layout struct for HLSL Buffer. The struct will include only
// fields of types that are allowed in HLSL buffer and it will filter out:
// - static or groupshared variable declarations
// - resource classes
// - empty structs
// - zero-sized arrays
// - non-variable declarations
// The layout struct will be added to the HLSLBufferDecl declarations.
static void createHostLayoutStructForBuffer(Sema &S, HLSLBufferDecl *BufDecl) {
  ASTContext &AST = S.getASTContext();
  IdentifierInfo *II = getHostLayoutStructName(S, BufDecl, true);
 
  CXXRecordDecl *LS =
      CXXRecordDecl::Create(AST, TagDecl::TagKind::Struct, BufDecl,
                            SourceLocation(), SourceLocation(), II);
  LS->addAttr(PackedAttr::CreateImplicit(AST));
  LS->setImplicit(true);
  LS->startDefinition();
 
  for (Decl *D : BufDecl->buffer_decls()) {
    VarDecl *VD = dyn_cast<VarDecl>(D);
    if (!VD || VD->getStorageClass() == SC_Static ||
        VD->getType().getAddressSpace() == LangAS::hlsl_groupshared)
      continue;
    const Type *Ty = getHostLayoutFieldType(VD->getType());
 
    FieldDecl *FD =
        createFieldForHostLayoutStruct(S, Ty, VD->getIdentifier(), LS);
    // Declarations collected for the default $Globals constant buffer have
    // already been checked to have non-empty cbuffer layout, so
    // createFieldForHostLayoutStruct should always succeed. These declarations
    // already have their address space set to hlsl_constant.
    // For declarations in a named cbuffer block
    // createFieldForHostLayoutStruct can still return nullptr if the type
    // is empty (does not have a cbuffer layout).
    assert((FD || VD->getType().getAddressSpace() != LangAS::hlsl_constant) &&
           "host layout field for $Globals decl failed to be created");
    if (FD) {
      // Add the field decl to the layout struct.
      LS->addDecl(FD);
      if (VD->getType().getAddressSpace() != LangAS::hlsl_constant) {
        // Update address space of the original decl to hlsl_constant.
        QualType NewTy =
            AST.getAddrSpaceQualType(VD->getType(), LangAS::hlsl_constant);
        VD->setType(NewTy);
      }
    }
  }
  LS->completeDefinition();
  BufDecl->addLayoutStruct(LS);
}
 
static void addImplicitBindingAttrToDecl(Sema &S, Decl *D, RegisterType RT,
                                         uint32_t ImplicitBindingOrderID) {
  auto *Attr =
      HLSLResourceBindingAttr::CreateImplicit(S.getASTContext(), "", "0", {});
  Attr->setBinding(RT, std::nullopt, 0);
  Attr->setImplicitBindingOrderID(ImplicitBindingOrderID);
  D->addAttr(Attr);
}
 
// Handle end of cbuffer/tbuffer declaration
void SemaHLSL::ActOnFinishBuffer(Decl *Dcl, SourceLocation RBrace) {
  auto *BufDecl = cast<HLSLBufferDecl>(Dcl);
  BufDecl->setRBraceLoc(RBrace);
 
  validatePackoffset(SemaRef, BufDecl);
 
  createHostLayoutStructForBuffer(SemaRef, BufDecl);
 
  // Handle implicit binding if needed.
  ResourceBindingAttrs ResourceAttrs(Dcl);
  if (!ResourceAttrs.isExplicit()) {
    SemaRef.Diag(Dcl->getLocation(), diag::warn_hlsl_implicit_binding);
    // Use HLSLResourceBindingAttr to transfer implicit binding order_ID
    // to codegen. If it does not exist, create an implicit attribute.
    uint32_t OrderID = getNextImplicitBindingOrderID();
    if (ResourceAttrs.hasBinding())
      ResourceAttrs.setImplicitOrderID(OrderID);
    else
      addImplicitBindingAttrToDecl(SemaRef, BufDecl,
                                   BufDecl->isCBuffer() ? RegisterType::CBuffer
                                                        : RegisterType::SRV,
                                   OrderID);
  }
 
  SemaRef.PopDeclContext();
}
 
HLSLNumThreadsAttr *SemaHLSL::mergeNumThreadsAttr(Decl *D,
                                                  const AttributeCommonInfo &AL,
                                                  int X, int Y, int Z) {
  if (HLSLNumThreadsAttr *NT = D->getAttr<HLSLNumThreadsAttr>()) {
    if (NT->getX() != X || NT->getY() != Y || NT->getZ() != Z) {
      Diag(NT->getLocation(), diag::err_hlsl_attribute_param_mismatch) << AL;
      Diag(AL.getLoc(), diag::note_conflicting_attribute);
    }
    return nullptr;
  }
  return ::new (getASTContext())
      HLSLNumThreadsAttr(getASTContext(), AL, X, Y, Z);
}
 
HLSLWaveSizeAttr *SemaHLSL::mergeWaveSizeAttr(Decl *D,
                                              const AttributeCommonInfo &AL,
                                              int Min, int Max, int Preferred,
                                              int SpelledArgsCount) {
  if (HLSLWaveSizeAttr *WS = D->getAttr<HLSLWaveSizeAttr>()) {
    if (WS->getMin() != Min || WS->getMax() != Max ||
        WS->getPreferred() != Preferred ||
        WS->getSpelledArgsCount() != SpelledArgsCount) {
      Diag(WS->getLocation(), diag::err_hlsl_attribute_param_mismatch) << AL;
      Diag(AL.getLoc(), diag::note_conflicting_attribute);
    }
    return nullptr;
  }
  HLSLWaveSizeAttr *Result = ::new (getASTContext())
      HLSLWaveSizeAttr(getASTContext(), AL, Min, Max, Preferred);
  Result->setSpelledArgsCount(SpelledArgsCount);
  return Result;
}
 
HLSLVkConstantIdAttr *
SemaHLSL::mergeVkConstantIdAttr(Decl *D, const AttributeCommonInfo &AL,
                                int Id) {
 
  auto &TargetInfo = getASTContext().getTargetInfo();
  if (TargetInfo.getTriple().getArch() != llvm::Triple::spirv) {
    Diag(AL.getLoc(), diag::warn_attribute_ignored) << AL;
    return nullptr;
  }
 
  auto *VD = cast<VarDecl>(D);
 
  if (getSpecConstBuiltinId(VD->getType()->getUnqualifiedDesugaredType()) ==
      Builtin::NotBuiltin) {
    Diag(VD->getLocation(), diag::err_specialization_const);
    return nullptr;
  }
 
  if (!VD->getType().isConstQualified()) {
    Diag(VD->getLocation(), diag::err_specialization_const);
    return nullptr;
  }
 
  if (HLSLVkConstantIdAttr *CI = D->getAttr<HLSLVkConstantIdAttr>()) {
    if (CI->getId() != Id) {
      Diag(CI->getLocation(), diag::err_hlsl_attribute_param_mismatch) << AL;
      Diag(AL.getLoc(), diag::note_conflicting_attribute);
    }
    return nullptr;
  }
 
  HLSLVkConstantIdAttr *Result =
      ::new (getASTContext()) HLSLVkConstantIdAttr(getASTContext(), AL, Id);
  return Result;
}
 
HLSLShaderAttr *
SemaHLSL::mergeShaderAttr(Decl *D, const AttributeCommonInfo &AL,
                          llvm::Triple::EnvironmentType ShaderType) {
  if (HLSLShaderAttr *NT = D->getAttr<HLSLShaderAttr>()) {
    if (NT->getType() != ShaderType) {
      Diag(NT->getLocation(), diag::err_hlsl_attribute_param_mismatch) << AL;
      Diag(AL.getLoc(), diag::note_conflicting_attribute);
    }
    return nullptr;
  }
  return HLSLShaderAttr::Create(getASTContext(), ShaderType, AL);
}
 
HLSLParamModifierAttr *
SemaHLSL::mergeParamModifierAttr(Decl *D, const AttributeCommonInfo &AL,
                                 HLSLParamModifierAttr::Spelling Spelling) {
  // We can only merge an `in` attribute with an `out` attribute. All other
  // combinations of duplicated attributes are ill-formed.
  if (HLSLParamModifierAttr *PA = D->getAttr<HLSLParamModifierAttr>()) {
    if ((PA->isIn() && Spelling == HLSLParamModifierAttr::Keyword_out) ||
        (PA->isOut() && Spelling == HLSLParamModifierAttr::Keyword_in)) {
      D->dropAttr<HLSLParamModifierAttr>();
      SourceRange AdjustedRange = {PA->getLocation(), AL.getRange().getEnd()};
      return HLSLParamModifierAttr::Create(
          getASTContext(), /*MergedSpelling=*/true, AdjustedRange,
          HLSLParamModifierAttr::Keyword_inout);
    }
    Diag(AL.getLoc(), diag::err_hlsl_duplicate_parameter_modifier) << AL;
    Diag(PA->getLocation(), diag::note_conflicting_attribute);
    return nullptr;
  }
  return HLSLParamModifierAttr::Create(getASTContext(), AL);
}
 
void SemaHLSL::handleInterpolationModifierAttr(Decl *D, const ParsedAttr &AL) {
  InterpolationModifier Modifier;
  switch (static_cast<HLSLInterpolationModifierAttr::Spelling>(
      AL.getSemanticSpelling())) {
  case HLSLInterpolationModifierAttr::Keyword_nointerpolation:
    Modifier = InterpolationModifier::NoInterpolation;
    break;
  case HLSLInterpolationModifierAttr::Keyword_linear:
    Modifier = InterpolationModifier::Linear;
    break;
  case HLSLInterpolationModifierAttr::Keyword_centroid:
    Modifier = InterpolationModifier::Centroid;
    break;
  case HLSLInterpolationModifierAttr::Keyword_noperspective:
    Modifier = InterpolationModifier::NoPerspective;
    break;
  case HLSLInterpolationModifierAttr::Keyword_sample:
    Modifier = InterpolationModifier::Sample;
    break;
  case HLSLInterpolationModifierAttr::Keyword_center:
    Modifier = InterpolationModifier::Center;
    break;
  case HLSLInterpolationModifierAttr::SpellingNotCalculated:
    llvm_unreachable("interpolation modifier spelling was not calculated");
  }
 
  InterpolationModifier Modifiers = Modifier;
  if (auto *Previous = D->getAttr<HLSLInterpolationModifierAttr>()) {
    auto Old = static_cast<InterpolationModifier>(Previous->getModifiers());
    Modifiers |= Old;
    if (any(Old & Modifier)) {
      Diag(AL.getLoc(), diag::warn_hlsl_duplicate_interpolation) << AL;
    } else if (llvm::hlsl::getInterpolationMode(Modifiers) ==
                   llvm::dxbc::PSV::InterpolationMode::Invalid &&
               llvm::hlsl::getInterpolationMode(Old) !=
                   llvm::dxbc::PSV::InterpolationMode::Invalid) {
      Diag(AL.getLoc(), diag::err_hlsl_interpolation_conflict);
      Diag(Previous->getLocation(), diag::note_conflicting_attribute);
      D->setInvalidDecl();
    } else {
      InterpolationModifier OldLocation =
          llvm::hlsl::getInterpolationSamplingLocation(Old);
      InterpolationModifier NewLocation =
          llvm::hlsl::getInterpolationSamplingLocation(Modifier);
      if (any(OldLocation) && any(NewLocation)) {
        Diag(AL.getLoc(), diag::warn_hlsl_interpolation_override)
            << (std::max(OldLocation, NewLocation) ==
                InterpolationModifier::Sample)
            << (std::min(OldLocation, NewLocation) ==
                InterpolationModifier::Centroid);
      }
    }
    D->dropAttr<HLSLInterpolationModifierAttr>();
  }
  D->addAttr(HLSLInterpolationModifierAttr::Create(
      getASTContext(), static_cast<unsigned>(Modifiers), AL));
}
 
bool SemaHLSL::checkInterpolationModifiers(
    const DeclaratorDecl *D, const HLSLInterpolationModifierAttr *Inherited,
    const HLSLParsedSemanticAttr *Semantic) {
  if (D->isInvalidDecl())
    return false;
  const auto *A = D->getAttr<HLSLInterpolationModifierAttr>();
  if (!A)
    A = Inherited;
  if (!Semantic)
    Semantic = D->getAttr<HLSLParsedSemanticAttr>();
 
  const auto *FD = dyn_cast<FunctionDecl>(D);
  QualType T = FD ? FD->getReturnType() : D->getType();
  T = getASTContext().getBaseElementType(T.getNonReferenceType());
  if (T->isDependentType())
    return true;
  if (const auto *RT = T->getAs<RecordType>()) {
    const RecordDecl *RD = RT->getDecl()->getDefinition();
    if (!RD)
      return true;
    bool Valid = true;
    for (const FieldDecl *Field : RD->fields())
      Valid &= checkInterpolationModifiers(Field, A, Semantic);
    return Valid;
  }
  if (!A)
    return true;
 
  auto Modifiers = static_cast<InterpolationModifier>(A->getModifiers());
  if (Modifiers == InterpolationModifier::NoInterpolation) {
    if (!Semantic)
      return true;
    SemanticKind Kind =
        llvm::hlsl::getSemanticKind(Semantic->getSemanticName());
    if (Kind != SemanticKind::Position && Kind != SemanticKind::Barycentrics)
      return true;
    Diag(A->getLocation(), diag::err_hlsl_interpolation_semantic)
        << Semantic->getSemanticName();
    Diag(Semantic->getLocation(), diag::note_conflicting_attribute);
    return false;
  }
 
  T = getScalarComponentType(T);
  if (T->isIntegerType() ||
      (T->isRealFloatingType() && getASTContext().getTypeSize(T) > 32)) {
    Diag(A->getLocation(), diag::err_hlsl_interpolation_type) << T;
    return false;
  }
  return true;
}
 
void SemaHLSL::ActOnTopLevelFunction(FunctionDecl *FD) {
  auto &TargetInfo = getASTContext().getTargetInfo();
 
  if (FD->getName() != TargetInfo.getTargetOpts().HLSLEntry)
    return;
 
  // If we have specified a root signature to override the entry function then
  // attach it now
  HLSLRootSignatureDecl *SignatureDecl =
      lookupRootSignatureOverrideDecl(FD->getDeclContext());
  if (SignatureDecl) {
    FD->dropAttr<RootSignatureAttr>();
    // We could look up the SourceRange of the macro here as well
    AttributeCommonInfo AL(RootSigOverrideIdent, AttributeScopeInfo(),
                           SourceRange(), ParsedAttr::Form::Microsoft());
    FD->addAttr(::new (getASTContext()) RootSignatureAttr(
        getASTContext(), AL, RootSigOverrideIdent, SignatureDecl));
  }
 
  llvm::Triple::EnvironmentType Env = TargetInfo.getTriple().getEnvironment();
  if (HLSLShaderAttr::isValidShaderType(Env) && Env != llvm::Triple::Library) {
    if (const auto *Shader = FD->getAttr<HLSLShaderAttr>()) {
      // The entry point is already annotated - check that it matches the
      // triple.
      if (Shader->getType() != Env) {
        Diag(Shader->getLocation(), diag::err_hlsl_entry_shader_attr_mismatch)
            << Shader;
        FD->setInvalidDecl();
      }
    } else {
      // Implicitly add the shader attribute if the entry function isn't
      // explicitly annotated.
      FD->addAttr(HLSLShaderAttr::CreateImplicit(getASTContext(), Env,
                                                 FD->getBeginLoc()));
    }
  } else {
    switch (Env) {
    case llvm::Triple::UnknownEnvironment:
    case llvm::Triple::Library:
      break;
    case llvm::Triple::RootSignature:
      llvm_unreachable("rootsig environment has no functions");
    default:
      llvm_unreachable("Unhandled environment in triple");
    }
  }
}
 
static bool isVkPipelineBuiltin(const ASTContext &AstContext, FunctionDecl *FD,
                                const DeclaratorDecl *D,
                                HLSLAppliedSemanticAttr *Semantic,
                                bool IsInput) {
  // Explicitly annotated built-ins skip location assignment
  if (D->hasAttr<HLSLVkBuiltInAttr>())
    return true;
 
  if (AstContext.getTargetInfo().getTriple().getOS() != llvm::Triple::Vulkan)
    return false;
 
  const auto *ShaderAttr = FD->getAttr<HLSLShaderAttr>();
  assert(ShaderAttr && "Entry point has no shader attribute");
  llvm::Triple::EnvironmentType ST = ShaderAttr->getType();
  SemanticKind Kind = llvm::hlsl::getSemanticKind(Semantic->getSemanticName());
 
  switch (Kind) {
  case SemanticKind::Position:
    // The SV_Position semantic is lowered to:
    //  - Position built-in for vertex output.
    //  - FragCoord built-in for fragment input.
    return (ST == llvm::Triple::Vertex && !IsInput) ||
           (ST == llvm::Triple::Pixel && IsInput);
  case SemanticKind::VertexID:
    return true;
  case SemanticKind::InstanceID:
    return ST == llvm::Triple::Vertex && IsInput;
  case SemanticKind::IsFrontFace:
    return ST == llvm::Triple::Pixel && IsInput;
  case SemanticKind::ShadingRate:
    // Lowered to:
    //  - ShadingRateKHR built-in for fragment input.
    //  - PrimitiveShadingRateKHR built-in for vertex output.
    // TODO: extend to mesh
    return (ST == llvm::Triple::Pixel && IsInput) ||
           (ST == llvm::Triple::Vertex && !IsInput);
  case SemanticKind::Barycentrics:
  case SemanticKind::SampleIndex:
  case SemanticKind::PrimitiveID:
    return ST == llvm::Triple::Pixel && IsInput;
  case SemanticKind::StartVertexLocation:
  case SemanticKind::StartInstanceLocation:
    return ST == llvm::Triple::Vertex && IsInput;
  case SemanticKind::ViewID:
    return IsInput;
  default:
    return false;
  }
}
 
static void validateVkBuiltinUsage(Sema &S, FunctionDecl *FD,
                                   const DeclaratorDecl *D,
                                   const HLSLVkBuiltInAttr *Attr,
                                   IOType CurrentIOType) {
  const ASTContext &AstContext = S.getASTContext();
  if (AstContext.getTargetInfo().getTriple().getOS() != llvm::Triple::Vulkan)
    return;
 
  StringRef BuiltIn = Attr->getBuiltIn();
 
  QualType T =
      (D == FD ? FD->getReturnType() : D->getType()).getNonReferenceType();
  const auto *ShaderAttr = FD->getAttr<HLSLShaderAttr>();
  assert(ShaderAttr && "Entry point has no shader attribute");
  llvm::Triple::EnvironmentType Stage = ShaderAttr->getType();
  bool IsIn = any(CurrentIOType & IOType::In);
  bool IsOut = any(CurrentIOType & IOType::Out);
 
  if (BuiltIn == "PointSize") {
    // Vulkan requires PointSize to be a 32-bit float
    // (VUID-PointSize-PointSize-04317). Without native 16-bit types, `half` is
    // lowered to a 32-bit float, so it is accepted as well.
    bool Is32BitFloat = T->isSpecificBuiltinType(BuiltinType::Float) ||
                        (T->isSpecificBuiltinType(BuiltinType::Half) &&
                         !AstContext.getLangOpts().NativeHalfType);
    if (!Is32BitFloat)
      S.Diag(Attr->getLocation(), diag::err_hlsl_vk_builtin_wrong_type)
          << BuiltIn << S.getASTContext().FloatTy;
 
    // PointSize is a per-vertex value, so patch-constant (hull/domain)
    // and per-primitive (mesh) data never carries it.
    // Note: Sema does not set this bit yet.
    bool IsPatchConstantOrPrimitive =
        any(CurrentIOType & IOType::PatchConstantOrPrimitive);
    bool LegalStage =
        !IsPatchConstantOrPrimitive &&
        ((Stage == llvm::Triple::Vertex && IsOut) ||
         Stage == llvm::Triple::Hull || Stage == llvm::Triple::Domain ||
         Stage == llvm::Triple::Geometry ||
         (Stage == llvm::Triple::Mesh && IsOut));
    if (!LegalStage) {
      StringRef IOTypeName = (IsIn && IsOut) ? "inputs and outputs"
                             : IsIn          ? "inputs"
                                             : "outputs";
      S.Diag(Attr->getLocation(), diag::err_hlsl_vk_builtin_wrong_stage)
          << BuiltIn << llvm::Triple::getEnvironmentTypeName(Stage)
          << IOTypeName;
    }
  }
}
 
bool SemaHLSL::determineActiveSemanticOnScalar(FunctionDecl *FD,
                                               DeclaratorDecl *OutputDecl,
                                               DeclaratorDecl *D,
                                               SemanticInfo &ActiveSemantic,
                                               SemaHLSL::SemanticContext &SC) {
  if (ActiveSemantic.Semantic == nullptr) {
    ActiveSemantic.Semantic = D->getAttr<HLSLParsedSemanticAttr>();
    if (ActiveSemantic.Semantic)
      ActiveSemantic.Index = ActiveSemantic.Semantic->getSemanticIndex();
  }
 
  if (!ActiveSemantic.Semantic) {
    Diag(D->getLocation(), diag::err_hlsl_missing_semantic_annotation);
    return false;
  }
 
  auto *A = ::new (getASTContext())
      HLSLAppliedSemanticAttr(getASTContext(), *ActiveSemantic.Semantic,
                              ActiveSemantic.Semantic->getAttrName()->getName(),
                              ActiveSemantic.Index.value_or(0));
  if (!A)
    return false;
 
  // Each array element occupies a separate semantic index.
  QualType T = D == FD ? FD->getReturnType() : D->getType();
  const ConstantArrayType *AT =
      getASTContext().getAsConstantArrayType(T.getNonReferenceType());
  if (isZeroSizedArray(AT)) {
    Diag(A->getLoc(), diag::err_hlsl_semantic_zero_sized_array)
        << A->getAttrName();
    return false;
  }
  unsigned ElementCount = AT ? ASTContext::getConstantArrayElementCount(AT) : 1;
 
  // Each matrix row (or column for column_major) also occupies a separate
  // semantic index.
  QualType BaseTy = getASTContext().getBaseElementType(T.getNonReferenceType());
  if (const auto *MT = BaseTy->getAs<ConstantMatrixType>())
    ElementCount *= isMatrixRowMajor(getLangOpts(), BaseTy)
                        ? MT->getNumRows()
                        : MT->getNumColumns();
 
  checkSemanticAnnotation(FD, D, A, SC, ElementCount);
  OutputDecl->addAttr(A);
 
  unsigned Location = ActiveSemantic.Index.value_or(0);
 
  bool IsBuiltin = isVkPipelineBuiltin(getASTContext(), FD, D, A,
                                       any(SC.CurrentIOType & IOType::In));
  if (IsBuiltin) {
    if (const auto *BuiltInAttr = D->getAttr<HLSLVkBuiltInAttr>())
      validateVkBuiltinUsage(SemaRef, FD, D, BuiltInAttr, SC.CurrentIOType);
  }
 
  if (!IsBuiltin) {
    bool HasVkLocation = false;
    if (auto *A = D->getAttr<HLSLVkLocationAttr>()) {
      HasVkLocation = true;
      Location = A->getLocation();
    }
 
    if (SC.UsesExplicitVkLocations.value_or(HasVkLocation) != HasVkLocation) {
      Diag(D->getLocation(), diag::err_hlsl_semantic_partial_explicit_indexing);
      return false;
    }
    SC.UsesExplicitVkLocations = HasVkLocation;
  }
 
  ActiveSemantic.Index = Location + ElementCount;
 
  StringRef BaseName = ActiveSemantic.Semantic->getAttrName()->getName();
  std::string LowerName = BaseName.lower();
  for (unsigned I = 0; I < ElementCount; ++I) {
    auto [It, Inserted] = SC.ActiveSemantics.try_emplace(
        (Twine(LowerName) + Twine(Location + I)).str(), D->getLocation());
    if (!Inserted) {
      Diag(D->getLocation(), diag::err_hlsl_semantic_index_overlap)
          << (BaseName + Twine(Location + I)).str();
      Diag(It->second, diag::note_previous_use);
      return false;
    }
  }
 
  return true;
}
 
bool SemaHLSL::determineActiveSemantic(FunctionDecl *FD,
                                       DeclaratorDecl *OutputDecl,
                                       DeclaratorDecl *D,
                                       SemanticInfo &ActiveSemantic,
                                       SemaHLSL::SemanticContext &SC) {
  if (ActiveSemantic.Semantic == nullptr) {
    ActiveSemantic.Semantic = D->getAttr<HLSLParsedSemanticAttr>();
    if (ActiveSemantic.Semantic)
      ActiveSemantic.Index = ActiveSemantic.Semantic->getSemanticIndex();
  }
 
  const Type *T = D == FD ? &*FD->getReturnType() : &*D->getType();
  T = T->getUnqualifiedDesugaredType();
 
  const RecordType *RT = dyn_cast<RecordType>(T);
  if (!RT)
    return determineActiveSemanticOnScalar(FD, OutputDecl, D, ActiveSemantic,
                                           SC);
 
  const RecordDecl *RD = RT->getDecl();
  for (FieldDecl *Field : RD->fields()) {
    SemanticInfo Info = ActiveSemantic;
    if (!determineActiveSemantic(FD, OutputDecl, Field, Info, SC)) {
      Diag(Field->getLocation(), diag::note_hlsl_semantic_used_here) << Field;
      return false;
    }
    if (ActiveSemantic.Semantic)
      ActiveSemantic = Info;
  }
 
  return true;
}
 
void SemaHLSL::CheckEntryPoint(FunctionDecl *FD) {
  const auto *ShaderAttr = FD->getAttr<HLSLShaderAttr>();
  assert(ShaderAttr && "Entry point has no shader attribute");
  llvm::Triple::EnvironmentType ST = ShaderAttr->getType();
  auto &TargetInfo = getASTContext().getTargetInfo();
  VersionTuple Ver = TargetInfo.getTriple().getOSVersion();
  switch (ST) {
  case llvm::Triple::Pixel:
  case llvm::Triple::Vertex:
  case llvm::Triple::Geometry:
  case llvm::Triple::Hull:
  case llvm::Triple::Domain:
  case llvm::Triple::RayGeneration:
  case llvm::Triple::Intersection:
  case llvm::Triple::AnyHit:
  case llvm::Triple::ClosestHit:
  case llvm::Triple::Miss:
  case llvm::Triple::Callable:
    if (const auto *NT = FD->getAttr<HLSLNumThreadsAttr>()) {
      diagnoseAttrStageMismatch(NT, ST,
                                {llvm::Triple::Compute,
                                 llvm::Triple::Amplification,
                                 llvm::Triple::Mesh});
      FD->setInvalidDecl();
    }
    if (const auto *WS = FD->getAttr<HLSLWaveSizeAttr>()) {
      diagnoseAttrStageMismatch(WS, ST,
                                {llvm::Triple::Compute,
                                 llvm::Triple::Amplification,
                                 llvm::Triple::Mesh});
      FD->setInvalidDecl();
    }
    break;
 
  case llvm::Triple::Compute:
  case llvm::Triple::Amplification:
  case llvm::Triple::Mesh:
    if (!FD->hasAttr<HLSLNumThreadsAttr>()) {
      Diag(FD->getLocation(), diag::err_hlsl_missing_numthreads)
          << llvm::Triple::getEnvironmentTypeName(ST);
      FD->setInvalidDecl();
    }
    if (const auto *WS = FD->getAttr<HLSLWaveSizeAttr>()) {
      if (TargetInfo.getTriple().isSPIRV()) {
        Diag(WS->getLocation(), diag::warn_hlsl_wavesize_unsupported_spirv);
      } else if (Ver < VersionTuple(6, 6)) {
        Diag(WS->getLocation(), diag::err_hlsl_attribute_in_wrong_shader_model)
            << WS << "6.6";
        FD->setInvalidDecl();
      } else if (WS->getSpelledArgsCount() > 1 && Ver < VersionTuple(6, 8)) {
        Diag(
            WS->getLocation(),
            diag::err_hlsl_attribute_number_arguments_insufficient_shader_model)
            << WS << WS->getSpelledArgsCount() << "6.8";
        FD->setInvalidDecl();
      }
    }
    break;
  case llvm::Triple::RootSignature:
    llvm_unreachable("rootsig environment has no function entry point");
  default:
    llvm_unreachable("Unhandled environment in triple");
  }
 
  SemaHLSL::SemanticContext InputSC = {};
  InputSC.CurrentIOType = IOType::In;
  SemaHLSL::SemanticContext OutputSC = {};
  OutputSC.CurrentIOType = IOType::Out;
 
  for (ParmVarDecl *Param : FD->parameters()) {
    SemanticInfo ActiveSemantic;
    ActiveSemantic.Semantic = Param->getAttr<HLSLParsedSemanticAttr>();
    if (ActiveSemantic.Semantic)
      ActiveSemantic.Index = ActiveSemantic.Semantic->getSemanticIndex();
 
    // FIXME: An `inout` parameter is part of both signatures, but it is only
    // verified against the output one here.
    const auto *MA = Param->getAttr<HLSLParamModifierAttr>();
    SemanticContext &SC = MA && MA->isAnyOut() ? OutputSC : InputSC;
 
    // Interpolation applies to pixel inputs and vertex outputs, including the
    // corresponding side of inout parameters.
    if (((ST == llvm::Triple::Pixel && (!MA || MA->isAnyIn())) ||
         (ST == llvm::Triple::Vertex && MA && MA->isAnyOut())) &&
        !checkInterpolationModifiers(Param, nullptr, nullptr))
      FD->setInvalidDecl();
 
    if (!determineActiveSemantic(FD, Param, Param, ActiveSemantic, SC)) {
      Diag(Param->getLocation(), diag::note_previous_decl) << Param;
      FD->setInvalidDecl();
    }
  }
 
  SemanticInfo ActiveSemantic;
  ActiveSemantic.Semantic = FD->getAttr<HLSLParsedSemanticAttr>();
  if (ActiveSemantic.Semantic)
    ActiveSemantic.Index = ActiveSemantic.Semantic->getSemanticIndex();
  if (!FD->getReturnType()->isVoidType()) {
    if (ST == llvm::Triple::Vertex &&
        !checkInterpolationModifiers(FD, nullptr, nullptr))
      FD->setInvalidDecl();
    determineActiveSemantic(FD, FD, FD, ActiveSemantic, OutputSC);
  }
}
 
void SemaHLSL::checkSemanticAnnotation(
    FunctionDecl *EntryPoint, const Decl *Param,
    const HLSLAppliedSemanticAttr *SemanticAttr, const SemanticContext &SC,
    unsigned ElementCount) {
  auto *ShaderAttr = EntryPoint->getAttr<HLSLShaderAttr>();
  assert(ShaderAttr && "Entry point has no shader attribute");
  llvm::Triple::EnvironmentType ST = ShaderAttr->getType();
 
  SemanticKind Kind =
      llvm::hlsl::getSemanticKind(SemanticAttr->getSemanticName());
  llvm::hlsl::SemanticInterpretation Interpretation =
      llvm::hlsl::getInterpretationKind(Kind, ST, SC.CurrentIOType);
  if (Interpretation == llvm::hlsl::SemanticInterpretation::Invalid) {
    diagnoseSemanticStageMismatch(SemanticAttr, ST, SC.CurrentIOType, Kind);
    return;
  }
 
  // A system-value name can have an arbitrary interpretation, for example
  // SV_Position on a vertex input. Only the general type restrictions apply.
  if (Interpretation == llvm::hlsl::SemanticInterpretation::Arbitrary) {
    diagnoseSemanticType(Param, SemanticAttr, SemanticKind::Arbitrary);
    return;
  }
 
  diagnoseSystemSemanticIndex(SemanticAttr, Kind, ElementCount);
  diagnoseSemanticType(Param, SemanticAttr, Kind);
  diagnoseSemanticShaderModel(SemanticAttr, Kind);
}
 
void SemaHLSL::diagnoseSemanticShaderModel(const HLSLAppliedSemanticAttr *A,
                                           SemanticKind Kind) {
  const auto &Triple = getASTContext().getTargetInfo().getTriple();
  if (!Triple.isDXIL())
    return;
 
  switch (Kind) {
  case SemanticKind::Barycentrics:
  case SemanticKind::ViewID:
    if (Triple.getOSVersion() < VersionTuple(6, 1))
      Diag(A->getLoc(), diag::err_hlsl_semantic_in_wrong_shader_model)
          << A->getAttrName() << "6.1";
    return;
  case SemanticKind::ShadingRate:
    if (Triple.getOSVersion() < VersionTuple(6, 4))
      Diag(A->getLoc(), diag::err_hlsl_semantic_in_wrong_shader_model)
          << A->getAttrName() << "6.4";
    return;
  case SemanticKind::StartVertexLocation:
  case SemanticKind::StartInstanceLocation:
    if (Triple.getOSVersion() < VersionTuple(6, 8))
      Diag(A->getLoc(), diag::err_hlsl_semantic_in_wrong_shader_model)
          << A->getAttrName() << "6.8";
    return;
  default:
    return;
  }
}
 
void SemaHLSL::diagnoseSystemSemanticIndex(const HLSLAppliedSemanticAttr *A,
                                           SemanticKind Kind,
                                           unsigned ElementCount) {
  assert(Kind != SemanticKind::Invalid && Kind != SemanticKind::Arbitrary &&
         "expected a recognized system semantic");
  assert(ElementCount > 0 && "a semantic covers at least one element");
  // The attribute stores the index in an int. Recover its unsigned value
  // before widening the arithmetic to detect overflow of the semantic range.
  uint32_t FirstIndex = A->getSemanticIndex();
  uint64_t LastIndex = uint64_t(FirstIndex) + ElementCount - 1;
  constexpr uint32_t MaxSemanticIndex = std::numeric_limits<uint32_t>::max();
  if (LastIndex > MaxSemanticIndex) {
    Diag(A->getLoc(), diag::err_hlsl_semantic_index_out_of_range)
        << A->getAttrName() << LastIndex << MaxSemanticIndex;
    return;
  }
  if (LastIndex == 0)
    return;
 
  switch (Kind) {
  // These semantics are limited by signature packing, not semantic indices.
  case SemanticKind::ClipDistance:
  case SemanticKind::CullDistance:
    return;
  case SemanticKind::Target: {
    constexpr unsigned MaxTargetIndex = 7;
    if (LastIndex > MaxTargetIndex)
      Diag(A->getLoc(), diag::err_hlsl_semantic_index_out_of_range)
          << A->getAttrName() << LastIndex << MaxTargetIndex;
    return;
  }
  default:
    Diag(A->getLoc(), diag::err_hlsl_semantic_indexing_not_supported)
        << A->getAttrName();
    return;
  }
}
 
static QualType getElementTypeOf(QualType T, bool IncludeMatrix) {
  if (const auto *VT = T->getAs<clang::VectorType>())
    return VT->getElementType();
  if (IncludeMatrix)
    if (const auto *MT = T->getAs<clang::MatrixType>())
      return MT->getElementType();
  return T;
}
 
static unsigned getComponentCountOf(QualType T) {
  if (const auto *VT = T->getAs<clang::VectorType>())
    return VT->getNumElements();
  return 1;
}
 
static bool isFloatOrHalfElement(QualType Elem) {
  return Elem->isHalfType() || Elem->isFloat16Type() || Elem->isFloat32Type();
}
 
// System-value integer types exclude bool.
static bool isIntElementOfWidth(const ASTContext &Ctx, QualType Elem,
                                uint64_t Width) {
  if (!Elem->isIntegerType() || Elem->isBooleanType())
    return false;
  return Ctx.getTypeSize(Elem) == Width;
}
 
static bool isIntUpTo32Element(const ASTContext &Ctx, QualType Elem) {
  return isIntElementOfWidth(Ctx, Elem, 16) ||
         isIntElementOfWidth(Ctx, Elem, 32);
}
 
void SemaHLSL::diagnoseSemanticType(const Decl *D,
                                    const HLSLAppliedSemanticAttr *A,
                                    SemanticKind Kind) {
  assert(Kind != SemanticKind::Invalid && "expected a valid semantic");
  ASTContext &Ctx = getASTContext();
 
  QualType T;
  if (const auto *FD = dyn_cast<FunctionDecl>(D))
    T = FD->getReturnType();
  else
    T = cast<ValueDecl>(D)->getType();
 
  // `out` and `inout` parameters are passed by reference.
  T = T.getNonReferenceType();
 
  // Array semantics constrain each element's type.
  QualType DeclaredTy = T;
  while (const ConstantArrayType *AT = Ctx.getAsConstantArrayType(T))
    T = AT->getElementType();
 
  QualType ElemTy = getElementTypeOf(T, /*IncludeMatrix=*/false);
  unsigned Components = getComponentCountOf(T);
 
  bool IsSPIRV = getASTContext().getTargetInfo().getTriple().isSPIRV();
 
  switch (Kind) {
  case SemanticKind::Barycentrics:
    if (DeclaredTy != T || !isFloatOrHalfElement(ElemTy) || Components != 3)
      Diag(A->getLoc(), diag::err_hlsl_semantic_invalid_type)
          << A->getAttrName() << /* vector of */ 2 << 3
          << /* 16 or 32 bit floating-point */ 2 << DeclaredTy;
    return;
  case SemanticKind::DispatchThreadID:
  case SemanticKind::GroupID:
  case SemanticKind::GroupThreadID:
    if (!isIntUpTo32Element(Ctx, ElemTy) || Components > 3)
      Diag(A->getLoc(), diag::err_hlsl_semantic_invalid_type)
          << A->getAttrName() << /* scalar or vector of up to */ 1 << 3
          << /* 16 or 32 bit integer */ 0 << DeclaredTy;
    return;
  case SemanticKind::StartVertexLocation:
  case SemanticKind::StartInstanceLocation:
  case SemanticKind::ViewID:
    if (DeclaredTy != T || !T->isScalarType() ||
        !isIntElementOfWidth(Ctx, T, 32))
      Diag(A->getLoc(), diag::err_hlsl_semantic_invalid_type)
          << A->getAttrName() << /* scalar */ 0 << 1 << /* 32 bit integer */ 1
          << DeclaredTy;
    return;
  case SemanticKind::GroupIndex:
    if (!isIntElementOfWidth(Ctx, ElemTy, 32) || Components != 1)
      Diag(A->getLoc(), diag::err_hlsl_semantic_invalid_type)
          << A->getAttrName() << /* scalar */ 0 << 1 << /* 32 bit integer */ 1
          << DeclaredTy;
    return;
  case SemanticKind::VertexID:
    if (!isIntUpTo32Element(Ctx, ElemTy) || Components != 1)
      Diag(A->getLoc(), diag::err_hlsl_semantic_invalid_type)
          << A->getAttrName() << /* scalar */ 0 << 1
          << /* 16 or 32 bit integer */ 0 << DeclaredTy;
    return;
  case SemanticKind::Position:
  case SemanticKind::Target:
    if (!isFloatOrHalfElement(ElemTy) || Components > 4)
      Diag(A->getLoc(), diag::err_hlsl_semantic_invalid_type)
          << A->getAttrName() << /* scalar or vector of up to */ 1 << 4
          << /* 16 or 32 bit floating-point */ 2 << DeclaredTy;
    return;
  case SemanticKind::SampleIndex:
  case SemanticKind::PrimitiveID:
    // SPIR-V declares these built-ins as 32-bit integers and converts 16-bit
    // declarations.
    if (!DeclaredTy->isScalarType() || !isIntUpTo32Element(Ctx, DeclaredTy))
      Diag(A->getLoc(), diag::err_hlsl_semantic_invalid_type)
          << A->getAttrName() << /* scalar */ 0 << 1
          << /* 16 or 32 bit integer */ 0 << DeclaredTy;
    return;
  case SemanticKind::InstanceID:
  case SemanticKind::ShadingRate:
    // DXIL permits U32 or U16. SPIR-V requires a 32-bit scalar per:
    // - VUID-InstanceIndex-InstanceIndex-04265
    // - VUID-ShadingRateKHR-ShadingRateKHR-04492
    if (!T->isUnsignedIntegerType() ||
        !(isIntElementOfWidth(Ctx, T, 32) ||
          (!IsSPIRV && isIntElementOfWidth(Ctx, T, 16))))
      Diag(A->getLoc(), diag::err_hlsl_semantic_invalid_type)
          << A->getAttrName() << /* scalar */ 0 << 1
          << /* 16 or 32 bit unsigned integer / 32 bit unsigned integer */
          (IsSPIRV ? 4 : 3) << DeclaredTy;
    return;
  case SemanticKind::IsFrontFace:
    // Accept bool or int; SPIR-V FrontFacing BuiltIn is always a bool
    // and CodeGen converts it to the declared type.
    if (DeclaredTy != T || !T->isScalarType() ||
        !(T->isBooleanType() || isIntElementOfWidth(Ctx, T, 32)))
      Diag(A->getLoc(), diag::err_hlsl_semantic_invalid_type)
          << A->getAttrName() << /* scalar */ 0 << 1
          << /* bool or 32 bit integer */ 5 << DeclaredTy;
    return;
  default:
    // Other semantics only have the general signature type restrictions.
    break;
  }
 
  // DXIL signatures cannot carry 64-bit components, even for arbitrary
  // semantics. SPIR-V interfaces do support these types.
  QualType ScalarTy = getElementTypeOf(T, /*IncludeMatrix=*/true);
  if (Ctx.getTargetInfo().getTriple().isDXIL() &&
      (ScalarTy->isSpecificBuiltinType(BuiltinType::Double) ||
       isIntElementOfWidth(Ctx, ScalarTy, 64)))
    Diag(A->getLoc(), diag::err_hlsl_semantic_64bit_type)
        << A->getAttrName() << DeclaredTy;
}
 
void SemaHLSL::diagnoseAttrStageMismatch(
    const Attr *A, llvm::Triple::EnvironmentType Stage,
    std::initializer_list<llvm::Triple::EnvironmentType> AllowedStages) {
  SmallVector<StringRef, 8> StageStrings;
  llvm::transform(AllowedStages, std::back_inserter(StageStrings),
                  [](llvm::Triple::EnvironmentType ST) {
                    return StringRef(
                        HLSLShaderAttr::ConvertEnvironmentTypeToStr(ST));
                  });
  Diag(A->getLoc(), diag::err_hlsl_attr_unsupported_in_stage)
      << A->getAttrName() << llvm::Triple::getEnvironmentTypeName(Stage)
      << (AllowedStages.size() != 1) << join(StageStrings, ", ");
}
 
void SemaHLSL::diagnoseSemanticStageMismatch(
    const Attr *A, llvm::Triple::EnvironmentType Stage, IOType CurrentIOType,
    SemanticKind Kind) {
 
  ArrayRef<SemanticStageInfo> Allowed = llvm::hlsl::getAvailableStages(Kind);
  auto It = llvm::find_if(Allowed, [&Stage](const SemanticStageInfo &Info) {
    return Info.Stage == Stage;
  });
 
  StringRef CurrentIOTypeName = "patch constants or primitives";
  if (any(CurrentIOType & IOType::In))
    CurrentIOTypeName = "inputs";
  else if (any(CurrentIOType & IOType::Out))
    CurrentIOTypeName = "outputs";
 
  // The semantic is not available in this shader stage at all.
  if (It == Allowed.end()) {
    Diag(A->getLoc(), diag::err_hlsl_semantic_unsupported_iotype_for_stage)
        << A->getAttrName() << llvm::Triple::getEnvironmentTypeName(Stage)
        << CurrentIOTypeName;
    return;
  }
 
  IOType AllowedIOTypes = It->AllowedIOTypesMask;
  if (!(AllowedIOTypes & CurrentIOType)) {
    Diag(A->getLoc(), diag::err_hlsl_semantic_unsupported_iotype_for_stage)
        << A->getAttrName() << llvm::Triple::getEnvironmentTypeName(Stage)
        << CurrentIOTypeName;
    return;
  }
}
 
template <CastKind Kind>
static void castVector(Sema &S, ExprResult &E, QualType &Ty, unsigned Sz) {
  if (const auto *VTy = Ty->getAs<VectorType>())
    Ty = VTy->getElementType();
  Ty = S.getASTContext().getExtVectorType(Ty, Sz);
  E = S.ImpCastExprToType(E.get(), Ty, Kind);
}
 
template <CastKind Kind>
static QualType castElement(Sema &S, ExprResult &E, QualType Ty) {
  E = S.ImpCastExprToType(E.get(), Ty, Kind);
  return Ty;
}
 
static QualType handleFloatVectorBinOpConversion(
    Sema &SemaRef, ExprResult &LHS, ExprResult &RHS, QualType LHSType,
    QualType RHSType, QualType LElTy, QualType RElTy, bool IsCompAssign) {
  bool LHSFloat = LElTy->isRealFloatingType();
  bool RHSFloat = RElTy->isRealFloatingType();
 
  if (LHSFloat && RHSFloat) {
    if (IsCompAssign ||
        SemaRef.getASTContext().getFloatingTypeOrder(LElTy, RElTy) > 0)
      return castElement<CK_FloatingCast>(SemaRef, RHS, LHSType);
 
    return castElement<CK_FloatingCast>(SemaRef, LHS, RHSType);
  }
 
  if (LHSFloat)
    return castElement<CK_IntegralToFloating>(SemaRef, RHS, LHSType);
 
  assert(RHSFloat);
  if (IsCompAssign)
    return castElement<clang::CK_FloatingToIntegral>(SemaRef, RHS, LHSType);
 
  return castElement<CK_IntegralToFloating>(SemaRef, LHS, RHSType);
}
 
static QualType handleIntegerVectorBinOpConversion(
    Sema &SemaRef, ExprResult &LHS, ExprResult &RHS, QualType LHSType,
    QualType RHSType, QualType LElTy, QualType RElTy, bool IsCompAssign) {
 
  int IntOrder = SemaRef.Context.getIntegerTypeOrder(LElTy, RElTy);
  bool LHSSigned = LElTy->hasSignedIntegerRepresentation();
  bool RHSSigned = RElTy->hasSignedIntegerRepresentation();
  auto &Ctx = SemaRef.getASTContext();
 
  // If both types have the same signedness, use the higher ranked type.
  if (LHSSigned == RHSSigned) {
    if (IsCompAssign || IntOrder >= 0)
      return castElement<CK_IntegralCast>(SemaRef, RHS, LHSType);
 
    return castElement<CK_IntegralCast>(SemaRef, LHS, RHSType);
  }
 
  // If the unsigned type has greater than or equal rank of the signed type, use
  // the unsigned type.
  if (IntOrder != (LHSSigned ? 1 : -1)) {
    if (IsCompAssign || RHSSigned)
      return castElement<CK_IntegralCast>(SemaRef, RHS, LHSType);
    return castElement<CK_IntegralCast>(SemaRef, LHS, RHSType);
  }
 
  // At this point the signed type has higher rank than the unsigned type, which
  // means it will be the same size or bigger. If the signed type is bigger, it
  // can represent all the values of the unsigned type, so select it.
  if (Ctx.getIntWidth(LElTy) != Ctx.getIntWidth(RElTy)) {
    if (IsCompAssign || LHSSigned)
      return castElement<CK_IntegralCast>(SemaRef, RHS, LHSType);
    return castElement<CK_IntegralCast>(SemaRef, LHS, RHSType);
  }
 
  // This is a bit of an odd duck case in HLSL. It shouldn't happen, but can due
  // to C/C++ leaking through. The place this happens today is long vs long
  // long. When arguments are vector<unsigned long, N> and vector<long long, N>,
  // the long long has higher rank than long even though they are the same size.
 
  // If this is a compound assignment cast the right hand side to the left hand
  // side's type.
  if (IsCompAssign)
    return castElement<CK_IntegralCast>(SemaRef, RHS, LHSType);
 
  // If this isn't a compound assignment we convert to unsigned long long.
  QualType ElTy = Ctx.getCorrespondingUnsignedType(LHSSigned ? LElTy : RElTy);
  QualType NewTy = Ctx.getExtVectorType(
      ElTy, RHSType->castAs<VectorType>()->getNumElements());
  (void)castElement<CK_IntegralCast>(SemaRef, RHS, NewTy);
 
  return castElement<CK_IntegralCast>(SemaRef, LHS, NewTy);
}
 
static CastKind getScalarCastKind(ASTContext &Ctx, QualType DestTy,
                                  QualType SrcTy) {
  if (DestTy->isRealFloatingType() && SrcTy->isRealFloatingType())
    return CK_FloatingCast;
  if (DestTy->isIntegralType(Ctx) && SrcTy->isIntegralType(Ctx))
    return CK_IntegralCast;
  if (DestTy->isRealFloatingType())
    return CK_IntegralToFloating;
  assert(SrcTy->isRealFloatingType() && DestTy->isIntegralType(Ctx));
  return CK_FloatingToIntegral;
}
 
QualType SemaHLSL::handleVectorBinOpConversion(ExprResult &LHS, ExprResult &RHS,
                                               QualType LHSType,
                                               QualType RHSType,
                                               bool IsCompAssign) {
  const auto *LVecTy = LHSType->getAs<VectorType>();
  const auto *RVecTy = RHSType->getAs<VectorType>();
  auto &Ctx = getASTContext();
 
  // If the LHS is not a vector and this is a compound assignment, we truncate
  // the argument to a scalar then convert it to the LHS's type.
  if (!LVecTy && IsCompAssign) {
    QualType RElTy = RHSType->castAs<VectorType>()->getElementType();
    RHS = SemaRef.ImpCastExprToType(RHS.get(), RElTy, CK_HLSLVectorTruncation);
    RHSType = RHS.get()->getType();
    if (Ctx.hasSameUnqualifiedType(LHSType, RHSType))
      return LHSType;
    RHS = SemaRef.ImpCastExprToType(RHS.get(), LHSType,
                                    getScalarCastKind(Ctx, LHSType, RHSType));
    return LHSType;
  }
 
  unsigned EndSz = std::numeric_limits<unsigned>::max();
  unsigned LSz = 0;
  if (LVecTy)
    LSz = EndSz = LVecTy->getNumElements();
  if (RVecTy)
    EndSz = std::min(RVecTy->getNumElements(), EndSz);
  assert(EndSz != std::numeric_limits<unsigned>::max() &&
         "one of the above should have had a value");
 
  // In a compound assignment, the left operand does not change type, the right
  // operand is converted to the type of the left operand.
  if (IsCompAssign && LSz != EndSz) {
    Diag(LHS.get()->getBeginLoc(),
         diag::err_hlsl_vector_compound_assignment_truncation)
        << LHSType << RHSType;
    return QualType();
  }
 
  if (RVecTy && RVecTy->getNumElements() > EndSz)
    castVector<CK_HLSLVectorTruncation>(SemaRef, RHS, RHSType, EndSz);
  if (!IsCompAssign && LVecTy && LVecTy->getNumElements() > EndSz)
    castVector<CK_HLSLVectorTruncation>(SemaRef, LHS, LHSType, EndSz);
 
  if (!RVecTy)
    castVector<CK_VectorSplat>(SemaRef, RHS, RHSType, EndSz);
  if (!IsCompAssign && !LVecTy)
    castVector<CK_VectorSplat>(SemaRef, LHS, LHSType, EndSz);
 
  // If we're at the same type after resizing we can stop here.
  if (Ctx.hasSameUnqualifiedType(LHSType, RHSType))
    return Ctx.getCommonSugaredType(LHSType, RHSType);
 
  QualType LElTy = LHSType->castAs<VectorType>()->getElementType();
  QualType RElTy = RHSType->castAs<VectorType>()->getElementType();
 
  // Handle conversion for floating point vectors.
  if (LElTy->isRealFloatingType() || RElTy->isRealFloatingType())
    return handleFloatVectorBinOpConversion(SemaRef, LHS, RHS, LHSType, RHSType,
                                            LElTy, RElTy, IsCompAssign);
 
  assert(LElTy->isIntegralType(Ctx) && RElTy->isIntegralType(Ctx) &&
         "HLSL Vectors can only contain integer or floating point types");
  return handleIntegerVectorBinOpConversion(SemaRef, LHS, RHS, LHSType, RHSType,
                                            LElTy, RElTy, IsCompAssign);
}
 
void SemaHLSL::emitLogicalOperatorFixIt(Expr *LHS, Expr *RHS,
                                        BinaryOperatorKind Opc) {
  assert((Opc == BO_LOr || Opc == BO_LAnd) &&
         "Called with non-logical operator");
  llvm::SmallVector<char, 256> Buff;
  llvm::raw_svector_ostream OS(Buff);
  PrintingPolicy PP(SemaRef.getLangOpts());
  StringRef NewFnName = Opc == BO_LOr ? "or" : "and";
  OS << NewFnName << "(";
  LHS->printPretty(OS, nullptr, PP);
  OS << ", ";
  RHS->printPretty(OS, nullptr, PP);
  OS << ")";
  SourceRange FullRange = SourceRange(LHS->getBeginLoc(), RHS->getEndLoc());
  SemaRef.Diag(LHS->getBeginLoc(), diag::note_function_suggestion)
      << NewFnName << FixItHint::CreateReplacement(FullRange, OS.str());
}
 
std::pair<IdentifierInfo *, bool>
SemaHLSL::ActOnStartRootSignatureDecl(StringRef Signature) {
  llvm::hash_code Hash = llvm::hash_value(Signature);
  std::string IdStr = "__hlsl_rootsig_decl_" + std::to_string(Hash);
  IdentifierInfo *DeclIdent = &(getASTContext().Idents.get(IdStr));
 
  // Check if we have already found a decl of the same name.
  LookupResult R(SemaRef, DeclIdent, SourceLocation(),
                 Sema::LookupOrdinaryName);
  bool Found = SemaRef.LookupQualifiedName(R, SemaRef.CurContext);
  return {DeclIdent, Found};
}
 
void SemaHLSL::ActOnFinishRootSignatureDecl(
    SourceLocation Loc, IdentifierInfo *DeclIdent,
    ArrayRef<hlsl::RootSignatureElement> RootElements) {
 
  if (handleRootSignatureElements(RootElements))
    return;
 
  SmallVector<llvm::hlsl::rootsig::RootElement> Elements;
  for (auto &RootSigElement : RootElements)
    Elements.push_back(RootSigElement.getElement());
 
  auto *SignatureDecl = HLSLRootSignatureDecl::Create(
      SemaRef.getASTContext(), /*DeclContext=*/SemaRef.CurContext, Loc,
      DeclIdent, SemaRef.getLangOpts().HLSLRootSigVer, Elements);
 
  SignatureDecl->setImplicit();
  SemaRef.PushOnScopeChains(SignatureDecl, SemaRef.getCurScope());
}
 
HLSLRootSignatureDecl *
SemaHLSL::lookupRootSignatureOverrideDecl(DeclContext *DC) const {
  if (RootSigOverrideIdent) {
    LookupResult R(SemaRef, RootSigOverrideIdent, SourceLocation(),
                   Sema::LookupOrdinaryName);
    if (SemaRef.LookupQualifiedName(R, DC))
      return dyn_cast<HLSLRootSignatureDecl>(R.getFoundDecl());
  }
 
  return nullptr;
}
 
namespace {
 
struct PerVisibilityBindingChecker {
  SemaHLSL *S;
  // We need one builder per `llvm::dxbc::ShaderVisibility` value.
  std::array<llvm::hlsl::BindingInfoBuilder, 8> Builders;
 
  struct ElemInfo {
    const hlsl::RootSignatureElement *Elem;
    llvm::dxbc::ShaderVisibility Vis;
    bool Diagnosed;
  };
  llvm::SmallVector<ElemInfo> ElemInfoMap;
 
  PerVisibilityBindingChecker(SemaHLSL *S) : S(S) {}
 
  void trackBinding(llvm::dxbc::ShaderVisibility Visibility,
                    llvm::dxil::ResourceClass RC, uint32_t Space,
                    uint32_t LowerBound, uint32_t UpperBound,
                    const hlsl::RootSignatureElement *Elem) {
    uint32_t BuilderIndex = llvm::to_underlying(Visibility);
    assert(BuilderIndex < Builders.size() &&
           "Not enough builders for visibility type");
    Builders[BuilderIndex].trackBinding(RC, Space, LowerBound, UpperBound,
                                        static_cast<const void *>(Elem));
 
    static_assert(llvm::to_underlying(llvm::dxbc::ShaderVisibility::All) == 0,
                  "'All' visibility must come first");
    if (Visibility == llvm::dxbc::ShaderVisibility::All)
      for (size_t I = 1, E = Builders.size(); I < E; ++I)
        Builders[I].trackBinding(RC, Space, LowerBound, UpperBound,
                                 static_cast<const void *>(Elem));
 
    ElemInfoMap.push_back({Elem, Visibility, false});
  }
 
  ElemInfo &getInfo(const hlsl::RootSignatureElement *Elem) {
    auto It = llvm::lower_bound(
        ElemInfoMap, Elem,
        [](const auto &LHS, const auto &RHS) { return LHS.Elem < RHS; });
    assert(It->Elem == Elem && "Element not in map");
    return *It;
  }
 
  bool checkOverlap() {
    llvm::sort(ElemInfoMap, [](const auto &LHS, const auto &RHS) {
      return LHS.Elem < RHS.Elem;
    });
 
    bool HadOverlap = false;
 
    using llvm::hlsl::BindingInfoBuilder;
    auto ReportOverlap = [this,
                          &HadOverlap](const BindingInfoBuilder &Builder,
                                       const llvm::hlsl::Binding &Reported) {
      HadOverlap = true;
 
      const auto *Elem =
          static_cast<const hlsl::RootSignatureElement *>(Reported.Cookie);
      const llvm::hlsl::Binding &Previous = Builder.findOverlapping(Reported);
      const auto *PrevElem =
          static_cast<const hlsl::RootSignatureElement *>(Previous.Cookie);
 
      ElemInfo &Info = getInfo(Elem);
      // We will have already diagnosed this binding if there's overlap in the
      // "All" visibility as well as any particular visibility.
      if (Info.Diagnosed)
        return;
      Info.Diagnosed = true;
 
      ElemInfo &PrevInfo = getInfo(PrevElem);
      llvm::dxbc::ShaderVisibility CommonVis =
          Info.Vis == llvm::dxbc::ShaderVisibility::All ? PrevInfo.Vis
                                                        : Info.Vis;
 
      this->S->Diag(Elem->getLocation(), diag::err_hlsl_resource_range_overlap)
          << llvm::to_underlying(Reported.RC) << Reported.LowerBound
          << Reported.isUnbounded() << Reported.UpperBound
          << llvm::to_underlying(Previous.RC) << Previous.LowerBound
          << Previous.isUnbounded() << Previous.UpperBound << Reported.Space
          << CommonVis;
 
      this->S->Diag(PrevElem->getLocation(),
                    diag::note_hlsl_resource_range_here);
    };
 
    for (BindingInfoBuilder &Builder : Builders)
      Builder.calculateBindingInfo(ReportOverlap);
 
    return HadOverlap;
  }
};
 
static CXXMethodDecl *lookupMethod(Sema &S, CXXRecordDecl *RecordDecl,
                                   StringRef Name, SourceLocation Loc) {
  DeclarationName DeclName(&S.getASTContext().Idents.get(Name));
  LookupResult Result(S, DeclName, Loc, Sema::LookupMemberName);
  if (!S.LookupQualifiedName(Result, static_cast<DeclContext *>(RecordDecl)))
    return nullptr;
  return cast<CXXMethodDecl>(Result.getFoundDecl());
}
 
} // end anonymous namespace
 
bool SemaHLSL::handleRootSignatureElements(
    ArrayRef<hlsl::RootSignatureElement> Elements) {
  // Define some common error handling functions
  bool HadError = false;
  auto ReportError = [this, &HadError](SourceLocation Loc, uint32_t LowerBound,
                                       uint32_t UpperBound) {
    HadError = true;
    this->Diag(Loc, diag::err_hlsl_invalid_rootsig_value)
        << LowerBound << UpperBound;
  };
 
  auto ReportFloatError = [this, &HadError](SourceLocation Loc,
                                            float LowerBound,
                                            float UpperBound) {
    HadError = true;
    this->Diag(Loc, diag::err_hlsl_invalid_rootsig_value)
        << llvm::formatv("{0:f}", LowerBound).sstr<6>()
        << llvm::formatv("{0:f}", UpperBound).sstr<6>();
  };
 
  auto VerifyRegister = [ReportError](SourceLocation Loc, uint32_t Register) {
    if (!llvm::hlsl::rootsig::verifyRegisterValue(Register))
      ReportError(Loc, 0, 0xfffffffe);
  };
 
  auto VerifySpace = [ReportError](SourceLocation Loc, uint32_t Space) {
    if (!llvm::hlsl::rootsig::verifyRegisterSpace(Space))
      ReportError(Loc, 0, 0xffffffef);
  };
 
  const uint32_t Version =
      llvm::to_underlying(SemaRef.getLangOpts().HLSLRootSigVer);
  const uint32_t VersionEnum = Version - 1;
  auto ReportFlagError = [this, &HadError, VersionEnum](SourceLocation Loc) {
    HadError = true;
    this->Diag(Loc, diag::err_hlsl_invalid_rootsig_flag)
        << /*version minor*/ VersionEnum;
  };
 
  // Iterate through the elements and do basic validations
  for (const hlsl::RootSignatureElement &RootSigElem : Elements) {
    SourceLocation Loc = RootSigElem.getLocation();
    const llvm::hlsl::rootsig::RootElement &Elem = RootSigElem.getElement();
    if (const auto *Descriptor =
            std::get_if<llvm::hlsl::rootsig::RootDescriptor>(&Elem)) {
      VerifyRegister(Loc, Descriptor->Reg.Number);
      VerifySpace(Loc, Descriptor->Space);
 
      if (!llvm::hlsl::rootsig::verifyRootDescriptorFlag(Version,
                                                         Descriptor->Flags))
        ReportFlagError(Loc);
    } else if (const auto *Constants =
                   std::get_if<llvm::hlsl::rootsig::RootConstants>(&Elem)) {
      VerifyRegister(Loc, Constants->Reg.Number);
      VerifySpace(Loc, Constants->Space);
    } else if (const auto *Sampler =
                   std::get_if<llvm::hlsl::rootsig::StaticSampler>(&Elem)) {
      VerifyRegister(Loc, Sampler->Reg.Number);
      VerifySpace(Loc, Sampler->Space);
 
      assert(!std::isnan(Sampler->MaxLOD) && !std::isnan(Sampler->MinLOD) &&
             "By construction, parseFloatParam can't produce a NaN from a "
             "float_literal token");
 
      if (!llvm::hlsl::rootsig::verifyMaxAnisotropy(Sampler->MaxAnisotropy))
        ReportError(Loc, 0, 16);
      if (!llvm::hlsl::rootsig::verifyMipLODBias(Sampler->MipLODBias))
        ReportFloatError(Loc, -16.f, 15.99f);
    } else if (const auto *Clause =
                   std::get_if<llvm::hlsl::rootsig::DescriptorTableClause>(
                       &Elem)) {
      VerifyRegister(Loc, Clause->Reg.Number);
      VerifySpace(Loc, Clause->Space);
 
      if (!llvm::hlsl::rootsig::verifyNumDescriptors(Clause->NumDescriptors)) {
        // NumDescriptor could techincally be ~0u but that is reserved for
        // unbounded, so the diagnostic will not report that as a valid int
        // value
        ReportError(Loc, 1, 0xfffffffe);
      }
 
      if (!llvm::hlsl::rootsig::verifyDescriptorRangeFlag(Version, Clause->Type,
                                                          Clause->Flags))
        ReportFlagError(Loc);
    }
  }
 
  PerVisibilityBindingChecker BindingChecker(this);
  SmallVector<std::pair<const llvm::hlsl::rootsig::DescriptorTableClause *,
                        const hlsl::RootSignatureElement *>>
      UnboundClauses;
 
  for (const hlsl::RootSignatureElement &RootSigElem : Elements) {
    const llvm::hlsl::rootsig::RootElement &Elem = RootSigElem.getElement();
    if (const auto *Descriptor =
            std::get_if<llvm::hlsl::rootsig::RootDescriptor>(&Elem)) {
      uint32_t LowerBound(Descriptor->Reg.Number);
      uint32_t UpperBound(LowerBound); // inclusive range
 
      BindingChecker.trackBinding(
          Descriptor->Visibility,
          static_cast<llvm::dxil::ResourceClass>(Descriptor->Type),
          Descriptor->Space, LowerBound, UpperBound, &RootSigElem);
    } else if (const auto *Constants =
                   std::get_if<llvm::hlsl::rootsig::RootConstants>(&Elem)) {
      uint32_t LowerBound(Constants->Reg.Number);
      uint32_t UpperBound(LowerBound); // inclusive range
 
      BindingChecker.trackBinding(
          Constants->Visibility, llvm::dxil::ResourceClass::CBuffer,
          Constants->Space, LowerBound, UpperBound, &RootSigElem);
    } else if (const auto *Sampler =
                   std::get_if<llvm::hlsl::rootsig::StaticSampler>(&Elem)) {
      uint32_t LowerBound(Sampler->Reg.Number);
      uint32_t UpperBound(LowerBound); // inclusive range
 
      BindingChecker.trackBinding(
          Sampler->Visibility, llvm::dxil::ResourceClass::Sampler,
          Sampler->Space, LowerBound, UpperBound, &RootSigElem);
    } else if (const auto *Clause =
                   std::get_if<llvm::hlsl::rootsig::DescriptorTableClause>(
                       &Elem)) {
      // We'll process these once we see the table element.
      UnboundClauses.emplace_back(Clause, &RootSigElem);
    } else if (const auto *Table =
                   std::get_if<llvm::hlsl::rootsig::DescriptorTable>(&Elem)) {
      assert(UnboundClauses.size() == Table->NumClauses &&
             "Number of unbound elements must match the number of clauses");
      bool HasAnySampler = false;
      bool HasAnyNonSampler = false;
      uint64_t Offset = 0;
      bool IsPrevUnbound = false;
      for (const auto &[Clause, ClauseElem] : UnboundClauses) {
        SourceLocation Loc = ClauseElem->getLocation();
        if (Clause->Type == llvm::dxil::ResourceClass::Sampler)
          HasAnySampler = true;
        else
          HasAnyNonSampler = true;
 
        if (HasAnySampler && HasAnyNonSampler)
          Diag(Loc, diag::err_hlsl_invalid_mixed_resources);
 
        // Relevant error will have already been reported above and needs to be
        // fixed before we can conduct further analysis, so shortcut error
        // return
        if (Clause->NumDescriptors == 0)
          return true;
 
        bool IsAppending =
            Clause->Offset == llvm::hlsl::rootsig::DescriptorTableOffsetAppend;
        if (!IsAppending)
          Offset = Clause->Offset;
 
        uint64_t RangeBound = llvm::hlsl::rootsig::computeRangeBound(
            Offset, Clause->NumDescriptors);
 
        if (IsPrevUnbound && IsAppending)
          Diag(Loc, diag::err_hlsl_appending_onto_unbound);
        else if (!llvm::hlsl::rootsig::verifyNoOverflowedOffset(RangeBound))
          Diag(Loc, diag::err_hlsl_offset_overflow) << Offset << RangeBound;
 
        // Update offset to be 1 past this range's bound
        Offset = RangeBound + 1;
        IsPrevUnbound = Clause->NumDescriptors ==
                        llvm::hlsl::rootsig::NumDescriptorsUnbounded;
 
        // Compute the register bounds and track resource binding
        uint32_t LowerBound(Clause->Reg.Number);
        uint32_t UpperBound = llvm::hlsl::rootsig::computeRangeBound(
            LowerBound, Clause->NumDescriptors);
 
        BindingChecker.trackBinding(
            Table->Visibility,
            static_cast<llvm::dxil::ResourceClass>(Clause->Type), Clause->Space,
            LowerBound, UpperBound, ClauseElem);
      }
      UnboundClauses.clear();
    }
  }
 
  return BindingChecker.checkOverlap();
}
 
void SemaHLSL::handleRootSignatureAttr(Decl *D, const ParsedAttr &AL) {
  if (AL.getNumArgs() != 1) {
    Diag(AL.getLoc(), diag::err_attribute_wrong_number_arguments) << AL << 1;
    return;
  }
 
  IdentifierInfo *Ident = AL.getArgAsIdent(0)->getIdentifierInfo();
  if (auto *RS = D->getAttr<RootSignatureAttr>()) {
    if (RS->getSignatureIdent() != Ident) {
      Diag(AL.getLoc(), diag::err_disallowed_duplicate_attribute) << RS;
      return;
    }
 
    Diag(AL.getLoc(), diag::warn_duplicate_attribute_exact) << RS;
    return;
  }
 
  LookupResult R(SemaRef, Ident, SourceLocation(), Sema::LookupOrdinaryName);
  if (SemaRef.LookupQualifiedName(R, D->getDeclContext()))
    if (auto *SignatureDecl =
            dyn_cast<HLSLRootSignatureDecl>(R.getFoundDecl())) {
      D->addAttr(::new (getASTContext()) RootSignatureAttr(
          getASTContext(), AL, Ident, SignatureDecl));
    }
}
 
void SemaHLSL::handleNumThreadsAttr(Decl *D, const ParsedAttr &AL) {
  llvm::VersionTuple SMVersion =
      getASTContext().getTargetInfo().getTriple().getOSVersion();
  bool IsDXIL = getASTContext().getTargetInfo().getTriple().getArch() ==
                llvm::Triple::dxil;
 
  uint32_t ZMax = 1024;
  uint32_t ThreadMax = 1024;
  if (IsDXIL && SMVersion.getMajor() <= 4) {
    ZMax = 1;
    ThreadMax = 768;
  } else if (IsDXIL && SMVersion.getMajor() == 5) {
    ZMax = 64;
    ThreadMax = 1024;
  }
 
  uint32_t X;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), X))
    return;
  if (X > 1024) {
    Diag(AL.getArgAsExpr(0)->getExprLoc(),
         diag::err_hlsl_numthreads_argument_oor)
        << 0 << 1024;
    return;
  }
  uint32_t Y;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(1), Y))
    return;
  if (Y > 1024) {
    Diag(AL.getArgAsExpr(1)->getExprLoc(),
         diag::err_hlsl_numthreads_argument_oor)
        << 1 << 1024;
    return;
  }
  uint32_t Z;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(2), Z))
    return;
  if (Z > ZMax) {
    SemaRef.Diag(AL.getArgAsExpr(2)->getExprLoc(),
                 diag::err_hlsl_numthreads_argument_oor)
        << 2 << ZMax;
    return;
  }
 
  if (X * Y * Z > ThreadMax) {
    Diag(AL.getLoc(), diag::err_hlsl_numthreads_invalid) << ThreadMax;
    return;
  }
 
  HLSLNumThreadsAttr *NewAttr = mergeNumThreadsAttr(D, AL, X, Y, Z);
  if (NewAttr)
    D->addAttr(NewAttr);
}
 
static bool isValidWaveSizeValue(unsigned Value) {
  return llvm::isPowerOf2_32(Value) && Value >= 4 && Value <= 128;
}
 
void SemaHLSL::handleWaveSizeAttr(Decl *D, const ParsedAttr &AL) {
  // validate that the wavesize argument is a power of 2 between 4 and 128
  // inclusive
  unsigned SpelledArgsCount = AL.getNumArgs();
  if (SpelledArgsCount == 0 || SpelledArgsCount > 3)
    return;
 
  uint32_t Min;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), Min))
    return;
 
  uint32_t Max = 0;
  if (SpelledArgsCount > 1 &&
      !SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(1), Max))
    return;
 
  uint32_t Preferred = 0;
  if (SpelledArgsCount > 2 &&
      !SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(2), Preferred))
    return;
 
  if (SpelledArgsCount > 2) {
    if (!isValidWaveSizeValue(Preferred)) {
      Diag(AL.getArgAsExpr(2)->getExprLoc(),
           diag::err_attribute_power_of_two_in_range)
          << AL << llvm::dxil::MinWaveSize << llvm::dxil::MaxWaveSize
          << Preferred;
      return;
    }
    // Preferred not in range.
    if (Preferred < Min || Preferred > Max) {
      Diag(AL.getArgAsExpr(2)->getExprLoc(),
           diag::err_attribute_power_of_two_in_range)
          << AL << Min << Max << Preferred;
      return;
    }
  } else if (SpelledArgsCount > 1) {
    if (!isValidWaveSizeValue(Max)) {
      Diag(AL.getArgAsExpr(1)->getExprLoc(),
           diag::err_attribute_power_of_two_in_range)
          << AL << llvm::dxil::MinWaveSize << llvm::dxil::MaxWaveSize << Max;
      return;
    }
    if (Max < Min) {
      Diag(AL.getLoc(), diag::err_attribute_argument_invalid) << AL << 1;
      return;
    } else if (Max == Min) {
      Diag(AL.getLoc(), diag::warn_attr_min_eq_max) << AL;
    }
  } else {
    if (!isValidWaveSizeValue(Min)) {
      Diag(AL.getArgAsExpr(0)->getExprLoc(),
           diag::err_attribute_power_of_two_in_range)
          << AL << llvm::dxil::MinWaveSize << llvm::dxil::MaxWaveSize << Min;
      return;
    }
  }
 
  HLSLWaveSizeAttr *NewAttr =
      mergeWaveSizeAttr(D, AL, Min, Max, Preferred, SpelledArgsCount);
  if (NewAttr)
    D->addAttr(NewAttr);
}
 
void SemaHLSL::handleVkExtBuiltinInputAttr(Decl *D, const ParsedAttr &AL) {
  uint32_t ID;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), ID))
    return;
  D->addAttr(::new (getASTContext())
                 HLSLVkExtBuiltinInputAttr(getASTContext(), AL, ID));
}
 
void SemaHLSL::handleVkExtBuiltinOutputAttr(Decl *D, const ParsedAttr &AL) {
  uint32_t ID;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), ID))
    return;
  D->addAttr(::new (getASTContext())
                 HLSLVkExtBuiltinOutputAttr(getASTContext(), AL, ID));
}
 
void SemaHLSL::handleVkPushConstantAttr(Decl *D, const ParsedAttr &AL) {
  D->addAttr(::new (getASTContext())
                 HLSLVkPushConstantAttr(getASTContext(), AL));
}
 
void SemaHLSL::handleVkConstantIdAttr(Decl *D, const ParsedAttr &AL) {
  uint32_t Id;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), Id))
    return;
  HLSLVkConstantIdAttr *NewAttr = mergeVkConstantIdAttr(D, AL, Id);
  if (NewAttr)
    D->addAttr(NewAttr);
}
 
void SemaHLSL::handleVkBindingAttr(Decl *D, const ParsedAttr &AL) {
  uint32_t Binding = 0;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), Binding))
    return;
  uint32_t Set = 0;
  if (AL.getNumArgs() > 1 &&
      !SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(1), Set))
    return;
 
  D->addAttr(::new (getASTContext())
                 HLSLVkBindingAttr(getASTContext(), AL, Binding, Set));
}
 
void SemaHLSL::handleVkLocationAttr(Decl *D, const ParsedAttr &AL) {
  uint32_t Location;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), Location))
    return;
 
  // vk::builtin and vk::location mutually exclusive
  if (const auto *BuiltIn = D->getAttr<HLSLVkBuiltInAttr>()) {
    Diag(AL.getLoc(), diag::err_attributes_are_not_compatible)
        << AL << BuiltIn
        << (AL.isRegularKeywordAttribute() ||
            BuiltIn->isRegularKeywordAttribute());
    Diag(BuiltIn->getLocation(), diag::note_conflicting_attribute);
  }
 
  D->addAttr(::new (getASTContext())
                 HLSLVkLocationAttr(getASTContext(), AL, Location));
}
 
void SemaHLSL::handleVkBuiltInAttr(Decl *D, const ParsedAttr &AL) {
  StringRef BuiltIn;
  SourceLocation ArgLoc;
  if (!SemaRef.checkStringLiteralArgumentAttr(AL, 0, BuiltIn, &ArgLoc))
    return;
 
  static constexpr StringRef SupportedBuiltIns[] = {
      "PointSize", // TODO: uncomment build-ins as they're fully implemented
                   // "HelperInvocation", "BaseVertex", "BaseInstance",
                   // "DrawIndex", "DeviceIndex", "ViewportMaskNV",
  };
  if (!llvm::is_contained(SupportedBuiltIns, BuiltIn)) {
    std::string Quoted;
    llvm::raw_string_ostream OS(Quoted);
    llvm::interleaveComma(SupportedBuiltIns, OS,
                          [&](StringRef S) { OS << "'" << S << "'"; });
    Diag(ArgLoc, diag::err_hlsl_vk_builtin_unsupported_value)
        << BuiltIn << Quoted;
    return;
  }
 
  // vk::builtin and vk::location mutually exclusive.
  if (const auto *Location = D->getAttr<HLSLVkLocationAttr>()) {
    Diag(AL.getLoc(), diag::err_attributes_are_not_compatible)
        << AL << Location
        << (AL.isRegularKeywordAttribute() ||
            Location->isRegularKeywordAttribute());
    Diag(Location->getLoc(), diag::note_conflicting_attribute);
  }
 
  D->addAttr(::new (getASTContext())
                 HLSLVkBuiltInAttr(getASTContext(), AL, BuiltIn));
}
 
void SemaHLSL::handleSemanticAttr(Decl *D, const ParsedAttr &AL) {
  uint32_t IndexValue(0), ExplicitIndex(0);
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), IndexValue) ||
      !SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(1), ExplicitIndex)) {
    assert(0 && "HLSLUnparsedSemantic is expected to have 2 int arguments.");
  }
  assert(IndexValue > 0 ? ExplicitIndex : true);
 
  SemanticKind Kind = llvm::hlsl::getSemanticKind(AL.getAttrName()->getName());
  if (Kind == SemanticKind::Invalid) {
    Diag(AL.getLoc(), diag::err_hlsl_unknown_semantic) << AL;
    return;
  }
 
  switch (Kind) {
  // FIXME: These semantics do not yet have CodeGen support.
  case SemanticKind::RenderTargetArrayIndex:
  case SemanticKind::ViewPortArrayIndex:
  case SemanticKind::ClipDistance:
  case SemanticKind::CullDistance:
  case SemanticKind::OutputControlPointID:
  case SemanticKind::DomainLocation:
  case SemanticKind::GSInstanceID:
  case SemanticKind::Coverage:
  case SemanticKind::InnerCoverage:
  case SemanticKind::Depth:
  case SemanticKind::DepthLessEqual:
  case SemanticKind::DepthGreaterEqual:
  case SemanticKind::StencilRef:
  case SemanticKind::TessFactor:
  case SemanticKind::InsideTessFactor:
  case SemanticKind::CullPrimitive:
    Diag(AL.getLoc(), diag::err_hlsl_unknown_semantic) << AL;
    return;
  default:
    break;
  }
 
  D->addAttr(HLSLParsedSemanticAttr::Create(
      getASTContext(), AL.getAttrName()->getName(), IndexValue, AL));
}
 
void SemaHLSL::handlePackOffsetAttr(Decl *D, const ParsedAttr &AL) {
  if (!isa<VarDecl>(D) || !isa<HLSLBufferDecl>(D->getDeclContext())) {
    Diag(AL.getLoc(), diag::err_hlsl_attr_invalid_ast_node)
        << AL << "shader constant in a constant buffer";
    return;
  }
 
  uint32_t SubComponent;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(0), SubComponent))
    return;
  uint32_t Component;
  if (!SemaRef.checkUInt32Argument(AL, AL.getArgAsExpr(1), Component))
    return;
 
  QualType T = cast<VarDecl>(D)->getType().getCanonicalType();
  // Check if T is an array or struct type.
  // TODO: mark matrix type as aggregate type.
  bool IsAggregateTy = (T->isArrayType() || T->isStructureType());
 
  // Check Component is valid for T.
  if (Component) {
    unsigned Size = getASTContext().getTypeSize(T);
    if (IsAggregateTy) {
      Diag(AL.getLoc(), diag::err_hlsl_invalid_register_or_packoffset);
      return;
    } else {
      // Make sure Component + sizeof(T) <= 4.
      if ((Component * 32 + Size) > 128) {
        Diag(AL.getLoc(), diag::err_hlsl_packoffset_cross_reg_boundary);
        return;
      }
      QualType EltTy = T;
      if (const auto *VT = T->getAs<VectorType>())
        EltTy = VT->getElementType();
      unsigned Align = getASTContext().getTypeAlign(EltTy);
      if (Align > 32 && Component == 1) {
        // NOTE: Component 3 will hit err_hlsl_packoffset_cross_reg_boundary.
        // So we only need to check Component 1 here.
        Diag(AL.getLoc(), diag::err_hlsl_packoffset_alignment_mismatch)
            << Align << EltTy;
        return;
      }
    }
  }
 
  D->addAttr(::new (getASTContext()) HLSLPackOffsetAttr(
      getASTContext(), AL, SubComponent, Component));
}
 
void SemaHLSL::handleShaderAttr(Decl *D, const ParsedAttr &AL) {
  StringRef Str;
  SourceLocation ArgLoc;
  if (!SemaRef.checkStringLiteralArgumentAttr(AL, 0, Str, &ArgLoc))
    return;
 
  llvm::Triple::EnvironmentType ShaderType;
  if (!HLSLShaderAttr::ConvertStrToEnvironmentType(Str, ShaderType)) {
    Diag(AL.getLoc(), diag::warn_attribute_type_not_supported)
        << AL << Str << ArgLoc;
    return;
  }
 
  // FIXME: check function match the shader stage.
 
  HLSLShaderAttr *NewAttr = mergeShaderAttr(D, AL, ShaderType);
  if (NewAttr)
    D->addAttr(NewAttr);
}
 
bool clang::CreateHLSLAttributedResourceType(
    Sema &S, QualType Wrapped, ArrayRef<const Attr *> AttrList,
    QualType &ResType, HLSLAttributedResourceLocInfo *LocInfo,
    Expr *SampleCountExpr) {
  assert(AttrList.size() && "expected list of resource attributes");
 
  QualType ContainedTy = QualType();
  TypeSourceInfo *ContainedTyInfo = nullptr;
  SourceLocation LocBegin = AttrList[0]->getRange().getBegin();
  SourceLocation LocEnd = AttrList[0]->getRange().getEnd();
 
  HLSLAttributedResourceType::Attributes ResAttrs;
 
  bool HasResourceClass = false;
  bool HasResourceDimension = false;
  for (const Attr *A : AttrList) {
    if (!A)
      continue;
    LocEnd = A->getRange().getEnd();
    switch (A->getKind()) {
    case attr::HLSLResourceClass: {
      ResourceClass RC = cast<HLSLResourceClassAttr>(A)->getResourceClass();
      if (HasResourceClass) {
        S.Diag(A->getLocation(), ResAttrs.ResourceClass == RC
                                     ? diag::warn_duplicate_attribute_exact
                                     : diag::warn_duplicate_attribute)
            << A;
        return false;
      }
      ResAttrs.ResourceClass = RC;
      HasResourceClass = true;
      break;
    }
    case attr::HLSLResourceDimension: {
      llvm::dxil::ResourceDimension RD =
          cast<HLSLResourceDimensionAttr>(A)->getDimension();
      if (HasResourceDimension) {
        S.Diag(A->getLocation(), ResAttrs.ResourceDimension == RD
                                     ? diag::warn_duplicate_attribute_exact
                                     : diag::warn_duplicate_attribute)
            << A;
        return false;
      }
      ResAttrs.ResourceDimension = RD;
      HasResourceDimension = true;
      break;
    }
    case attr::HLSLIsROV:
      if (ResAttrs.IsROV) {
        S.Diag(A->getLocation(), diag::warn_duplicate_attribute_exact) << A;
        return false;
      }
      ResAttrs.IsROV = true;
      break;
    case attr::HLSLRawBuffer:
      if (ResAttrs.RawBuffer) {
        S.Diag(A->getLocation(), diag::warn_duplicate_attribute_exact) << A;
        return false;
      }
      ResAttrs.RawBuffer = true;
      break;
    case attr::HLSLIsArray:
      if (ResAttrs.IsArray) {
        S.Diag(A->getLocation(), diag::warn_duplicate_attribute_exact) << A;
        return false;
      }
      ResAttrs.IsArray = true;
      break;
    case attr::HLSLIsMultiSampled:
      if (ResAttrs.SampleCountExpr) {
        S.Diag(A->getLocation(), diag::warn_duplicate_attribute_exact) << A;
        return false;
      }
      // A bare [[hlsl::is_ms]] carries no count, so default it to 0, the same
      // value Texture2DMS<T> gets from its template parameter.
      ResAttrs.SampleCountExpr =
          SampleCountExpr
              ? SampleCountExpr
              : IntegerLiteral::Create(S.Context, llvm::APInt(32, 0),
                                       S.Context.IntTy, A->getLocation());
      break;
    case attr::HLSLIsCounter:
      if (ResAttrs.IsCounter) {
        S.Diag(A->getLocation(), diag::warn_duplicate_attribute_exact) << A;
        return false;
      }
      ResAttrs.IsCounter = true;
      break;
    case attr::HLSLContainedType: {
      const HLSLContainedTypeAttr *CTAttr = cast<HLSLContainedTypeAttr>(A);
      QualType Ty = CTAttr->getType();
      if (!ContainedTy.isNull()) {
        S.Diag(A->getLocation(), ContainedTy == Ty
                                     ? diag::warn_duplicate_attribute_exact
                                     : diag::warn_duplicate_attribute)
            << A;
        return false;
      }
      ContainedTy = Ty;
      ContainedTyInfo = CTAttr->getTypeLoc();
      break;
    }
    default:
      llvm_unreachable("unhandled resource attribute type");
    }
  }
 
  if (!HasResourceClass) {
    S.Diag(AttrList.back()->getRange().getEnd(),
           diag::err_hlsl_missing_resource_class);
    return false;
  }
 
  ResType = S.getASTContext().getHLSLAttributedResourceType(
      Wrapped, ContainedTy, ResAttrs);
 
  if (LocInfo && ContainedTyInfo) {
    LocInfo->Range = SourceRange(LocBegin, LocEnd);
    LocInfo->ContainedTyInfo = ContainedTyInfo;
  }
  return true;
}
 
// Validates and creates an HLSL attribute that is applied as type attribute on
// HLSL resource. The attributes are collected in HLSLResourcesTypeAttrs and at
// the end of the declaration they are applied to the declaration type by
// wrapping it in HLSLAttributedResourceType.
bool SemaHLSL::handleResourceTypeAttr(QualType T, const ParsedAttr &AL) {
  // only allow resource type attributes on intangible types
  if (!T->isHLSLResourceType()) {
    Diag(AL.getLoc(), diag::err_hlsl_attribute_needs_intangible_type)
        << AL << getASTContext().HLSLResourceTy;
    return false;
  }
 
  // validate number of arguments
  if (!AL.checkExactlyNumArgs(SemaRef, AL.getMinArgs()))
    return false;
 
  Attr *A = nullptr;
 
  AttributeCommonInfo ACI(
      AL.getLoc(), AttributeScopeInfo(AL.getScopeName(), AL.getScopeLoc()),
      AttributeCommonInfo::NoSemaHandlerAttribute,
      {
          AttributeCommonInfo::AS_CXX11, 0, false /*IsAlignas*/,
          false /*IsRegularKeywordAttribute*/
      });
 
  switch (AL.getKind()) {
  case ParsedAttr::AT_HLSLResourceClass: {
    StringRef Identifier;
    SourceLocation ArgLoc;
    if (!SemaRef.checkStringLiteralArgumentAttr(AL, 0, Identifier, &ArgLoc))
      return false;
 
    // Validate resource class value
    ResourceClass RC;
    if (!HLSLResourceClassAttr::ConvertStrToResourceClass(Identifier, RC)) {
      Diag(ArgLoc, diag::warn_attribute_type_not_supported)
          << "ResourceClass" << Identifier;
      return false;
    }
    A = HLSLResourceClassAttr::Create(getASTContext(), RC, ACI);
    break;
  }
 
  case ParsedAttr::AT_HLSLResourceDimension: {
    StringRef Identifier;
    SourceLocation ArgLoc;
    if (!SemaRef.checkStringLiteralArgumentAttr(AL, 0, Identifier, &ArgLoc))
      return false;
 
    // Validate resource dimension value
    llvm::dxil::ResourceDimension RD;
    if (!HLSLResourceDimensionAttr::ConvertStrToResourceDimension(Identifier,
                                                                  RD)) {
      Diag(ArgLoc, diag::warn_attribute_type_not_supported)
          << "ResourceDimension" << Identifier;
      return false;
    }
    A = HLSLResourceDimensionAttr::Create(getASTContext(), RD, ACI);
    break;
  }
 
  case ParsedAttr::AT_HLSLIsROV:
    A = HLSLIsROVAttr::Create(getASTContext(), ACI);
    break;
 
  case ParsedAttr::AT_HLSLRawBuffer:
    A = HLSLRawBufferAttr::Create(getASTContext(), ACI);
    break;
 
  case ParsedAttr::AT_HLSLIsCounter:
    A = HLSLIsCounterAttr::Create(getASTContext(), ACI);
    break;
 
  case ParsedAttr::AT_HLSLIsArray:
    A = HLSLIsArrayAttr::Create(getASTContext(), ACI);
    break;
 
  case ParsedAttr::AT_HLSLIsMultiSampled:
    A = HLSLIsMultiSampledAttr::Create(getASTContext(), ACI);
    break;
 
  case ParsedAttr::AT_HLSLContainedType: {
    if (AL.getNumArgs() != 1 && !AL.hasParsedType()) {
      Diag(AL.getLoc(), diag::err_attribute_wrong_number_arguments) << AL << 1;
      return false;
    }
 
    TypeSourceInfo *TSI = nullptr;
    QualType QT = SemaRef.GetTypeFromParser(AL.getTypeArg(), &TSI);
    assert(TSI && "no type source info for attribute argument");
    if (SemaRef.RequireCompleteType(TSI->getTypeLoc().getBeginLoc(), QT,
                                    diag::err_incomplete_type))
      return false;
    A = HLSLContainedTypeAttr::Create(getASTContext(), TSI, ACI);
    break;
  }
 
  default:
    llvm_unreachable("unhandled HLSL attribute");
  }
 
  HLSLResourcesTypeAttrs.emplace_back(A);
  return true;
}
 
// Combines all resource type attributes and creates HLSLAttributedResourceType.
QualType SemaHLSL::ProcessResourceTypeAttributes(QualType CurrentType) {
  if (!HLSLResourcesTypeAttrs.size())
    return CurrentType;
 
  QualType QT = CurrentType;
  HLSLAttributedResourceLocInfo LocInfo;
  if (CreateHLSLAttributedResourceType(SemaRef, CurrentType,
                                       HLSLResourcesTypeAttrs, QT, &LocInfo)) {
    const HLSLAttributedResourceType *RT =
        cast<HLSLAttributedResourceType>(QT.getTypePtr());
 
    // Temporarily store TypeLoc information for the new type.
    // It will be transferred to HLSLAttributesResourceTypeLoc
    // shortly after the type is created by TypeSpecLocFiller which
    // will call the TakeLocForHLSLAttribute method below.
    LocsForHLSLAttributedResources.insert(std::pair(RT, LocInfo));
  }
  HLSLResourcesTypeAttrs.clear();
  return QT;
}
 
// Returns source location for the HLSLAttributedResourceType
HLSLAttributedResourceLocInfo
SemaHLSL::TakeLocForHLSLAttribute(const HLSLAttributedResourceType *RT) {
  HLSLAttributedResourceLocInfo LocInfo = {};
  auto I = LocsForHLSLAttributedResources.find(RT);
  if (I != LocsForHLSLAttributedResources.end()) {
    LocInfo = I->second;
    LocsForHLSLAttributedResources.erase(I);
    return LocInfo;
  }
  LocInfo.Range = SourceRange();
  return LocInfo;
}
 
// Walks though the global variable declaration, collects all resource binding
// requirements and adds them to Bindings
void SemaHLSL::collectResourceBindingsOnUserRecordDecl(const VarDecl *VD,
                                                       const RecordType *RT) {
  const RecordDecl *RD = RT->getDecl()->getDefinitionOrSelf();
  for (FieldDecl *FD : RD->fields()) {
    const Type *Ty = FD->getType()->getUnqualifiedDesugaredType();
 
    // Unwrap arrays
    // FIXME: Calculate array size while unwrapping
    assert(!Ty->isIncompleteArrayType() &&
           "incomplete arrays inside user defined types are not supported");
    while (Ty->isConstantArrayType()) {
      const ConstantArrayType *CAT = cast<ConstantArrayType>(Ty);
      Ty = CAT->getElementType()->getUnqualifiedDesugaredType();
    }
 
    if (!Ty->isRecordType())
      continue;
 
    if (const HLSLAttributedResourceType *AttrResType =
            HLSLAttributedResourceType::findHandleTypeOnResource(Ty)) {
      // Add a new DeclBindingInfo to Bindings if it does not already exist
      ResourceClass RC = AttrResType->getAttrs().ResourceClass;
      DeclBindingInfo *DBI = Bindings.getDeclBindingInfo(VD, RC);
      if (!DBI)
        Bindings.addDeclBindingInfo(VD, RC);
    } else if (const RecordType *RT = dyn_cast<RecordType>(Ty)) {
      // Recursively scan embedded struct or class; it would be nice to do this
      // without recursion, but tricky to correctly calculate the size of the
      // binding, which is something we are probably going to need to do later
      // on. Hopefully nesting of structs in structs too many levels is
      // unlikely.
      collectResourceBindingsOnUserRecordDecl(VD, RT);
    }
  }
}
 
// Diagnose localized register binding errors for a single binding; does not
// diagnose resource binding on user record types, that will be done later
// in processResourceBindingOnDecl based on the information collected in
// collectResourceBindingsOnVarDecl.
// Returns false if the register binding is not valid.
static bool DiagnoseLocalRegisterBinding(Sema &S, SourceLocation &ArgLoc,
                                         Decl *D, RegisterType RegType,
                                         bool SpecifiedSpace) {
  int RegTypeNum = static_cast<int>(RegType);
 
  // check if the decl type is groupshared
  if (D->hasAttr<HLSLGroupSharedAddressSpaceAttr>()) {
    S.Diag(ArgLoc, diag::err_hlsl_binding_type_mismatch) << RegTypeNum;
    return false;
  }
 
  // Cbuffers and Tbuffers are HLSLBufferDecl types
  if (HLSLBufferDecl *CBufferOrTBuffer = dyn_cast<HLSLBufferDecl>(D)) {
    ResourceClass RC = CBufferOrTBuffer->isCBuffer() ? ResourceClass::CBuffer
                                                     : ResourceClass::SRV;
    if (RegType == getRegisterType(RC))
      return true;
 
    S.Diag(D->getLocation(), diag::err_hlsl_binding_type_mismatch)
        << RegTypeNum;
    return false;
  }
 
  // Samplers, UAVs, and SRVs are VarDecl types
  assert(isa<VarDecl>(D) && "D is expected to be VarDecl or HLSLBufferDecl");
  VarDecl *VD = cast<VarDecl>(D);
 
  // Resource
  if (const HLSLAttributedResourceType *AttrResType =
          HLSLAttributedResourceType::findHandleTypeOnResource(
              VD->getType().getTypePtr())) {
    if (RegType == getRegisterType(AttrResType))
      return true;
 
    S.Diag(D->getLocation(), diag::err_hlsl_binding_type_mismatch)
        << RegTypeNum;
    return false;
  }
 
  const clang::Type *Ty = VD->getType().getTypePtr();
  while (Ty->isArrayType())
    Ty = Ty->getArrayElementTypeNoTypeQual();
 
  // Basic types
  if (Ty->isArithmeticType() || Ty->isVectorType()) {
    bool DeclaredInCOrTBuffer = isa<HLSLBufferDecl>(D->getDeclContext());
    if (SpecifiedSpace && !DeclaredInCOrTBuffer)
      S.Diag(ArgLoc, diag::err_hlsl_space_on_global_constant);
 
    if (!DeclaredInCOrTBuffer && (Ty->isIntegralType(S.getASTContext()) ||
                                  Ty->isFloatingType() || Ty->isVectorType())) {
      // Register annotation on default constant buffer declaration ($Globals)
      if (RegType == RegisterType::CBuffer)
        S.Diag(ArgLoc, diag::warn_hlsl_deprecated_register_type_b);
      else if (RegType != RegisterType::C)
        S.Diag(ArgLoc, diag::err_hlsl_binding_type_mismatch) << RegTypeNum;
      else
        return true;
    } else {
      if (RegType == RegisterType::C)
        S.Diag(ArgLoc, diag::warn_hlsl_register_type_c_packoffset);
      else
        S.Diag(ArgLoc, diag::err_hlsl_binding_type_mismatch) << RegTypeNum;
    }
    return false;
  }
  if (Ty->isRecordType())
    // RecordTypes will be diagnosed in processResourceBindingOnDecl
    // that is called from ActOnVariableDeclarator
    return true;
 
  // Anything else is an error
  S.Diag(ArgLoc, diag::err_hlsl_binding_type_mismatch) << RegTypeNum;
  return false;
}
 
static bool ValidateMultipleRegisterAnnotations(Sema &S, Decl *TheDecl,
                                                RegisterType regType) {
  // make sure that there are no two register annotations
  // applied to the decl with the same register type
  bool RegisterTypesDetected[5] = {false};
  RegisterTypesDetected[static_cast<int>(regType)] = true;
 
  for (auto it = TheDecl->attr_begin(); it != TheDecl->attr_end(); ++it) {
    if (HLSLResourceBindingAttr *attr =
            dyn_cast<HLSLResourceBindingAttr>(*it)) {
 
      RegisterType otherRegType = attr->getRegisterType();
      if (RegisterTypesDetected[static_cast<int>(otherRegType)]) {
        int otherRegTypeNum = static_cast<int>(otherRegType);
        S.Diag(TheDecl->getLocation(),
               diag::err_hlsl_duplicate_register_annotation)
            << otherRegTypeNum;
        return false;
      }
      RegisterTypesDetected[static_cast<int>(otherRegType)] = true;
    }
  }
  return true;
}
 
static bool DiagnoseHLSLRegisterAttribute(Sema &S, SourceLocation &ArgLoc,
                                          Decl *D, RegisterType RegType,
                                          bool SpecifiedSpace) {
 
  // exactly one of these two types should be set
  assert(((isa<VarDecl>(D) && !isa<HLSLBufferDecl>(D)) ||
          (!isa<VarDecl>(D) && isa<HLSLBufferDecl>(D))) &&
         "expecting VarDecl or HLSLBufferDecl");
 
  // check if the declaration contains resource matching the register type
  if (!DiagnoseLocalRegisterBinding(S, ArgLoc, D, RegType, SpecifiedSpace))
    return false;
 
  // next, if multiple register annotations exist, check that none conflict.
  return ValidateMultipleRegisterAnnotations(S, D, RegType);
}
 
// return false if the slot count exceeds the limit, true otherwise
static bool AccumulateHLSLResourceSlots(QualType Ty, uint64_t &StartSlot,
                                        const uint64_t &Limit,
                                        const ResourceClass ResClass,
                                        ASTContext &Ctx,
                                        uint64_t ArrayCount = 1) {
  Ty = Ty.getCanonicalType();
  const Type *T = Ty.getTypePtr();
 
  // Early exit if already overflowed
  if (StartSlot > Limit)
    return false;
 
  // Case 1: array type
  if (const auto *AT = dyn_cast<ArrayType>(T)) {
    uint64_t Count = 1;
 
    if (const auto *CAT = dyn_cast<ConstantArrayType>(AT))
      Count = CAT->getSize().getZExtValue();
 
    QualType ElemTy = AT->getElementType();
    return AccumulateHLSLResourceSlots(ElemTy, StartSlot, Limit, ResClass, Ctx,
                                       ArrayCount * Count);
  }
 
  // Case 2: resource leaf
  if (auto ResTy = dyn_cast<HLSLAttributedResourceType>(T)) {
    // First ensure this resource counts towards the corresponding
    // register type limit.
    if (ResTy->getAttrs().ResourceClass != ResClass)
      return true;
 
    // Validate highest slot used
    uint64_t EndSlot = StartSlot + ArrayCount - 1;
    if (EndSlot > Limit)
      return false;
 
    // Advance SlotCount past the consumed range
    StartSlot = EndSlot + 1;
    return true;
  }
 
  // Case 3: struct / record
  if (const auto *RT = dyn_cast<RecordType>(T)) {
    const RecordDecl *RD = RT->getDecl();
 
    if (const auto *CXXRD = dyn_cast<CXXRecordDecl>(RD)) {
      for (const CXXBaseSpecifier &Base : CXXRD->bases()) {
        if (!AccumulateHLSLResourceSlots(Base.getType(), StartSlot, Limit,
                                         ResClass, Ctx, ArrayCount))
          return false;
      }
    }
 
    for (const FieldDecl *Field : RD->fields()) {
      if (!AccumulateHLSLResourceSlots(Field->getType(), StartSlot, Limit,
                                       ResClass, Ctx, ArrayCount))
        return false;
    }
 
    return true;
  }
 
  // Case 4: everything else
  return true;
}
 
// return true if there is something invalid, false otherwise
static bool ValidateRegisterNumber(uint64_t SlotNum, Decl *TheDecl,
                                   ASTContext &Ctx, RegisterType RegTy) {
  const uint64_t Limit = UINT32_MAX;
  if (SlotNum > Limit)
    return true;
 
  // after verifying the number doesn't exceed uint32max, we don't need
  // to look further into c or i register types
  if (RegTy == RegisterType::C || RegTy == RegisterType::I)
    return false;
 
  if (VarDecl *VD = dyn_cast<VarDecl>(TheDecl)) {
    uint64_t BaseSlot = SlotNum;
 
    if (!AccumulateHLSLResourceSlots(VD->getType(), SlotNum, Limit,
                                     getResourceClass(RegTy), Ctx))
      return true;
 
    // After AccumulateHLSLResourceSlots runs, SlotNum is now
    // the first free slot; last used was SlotNum - 1
    return (BaseSlot > Limit);
  }
  // handle the cbuffer/tbuffer case
  if (isa<HLSLBufferDecl>(TheDecl))
    // resources cannot be put within a cbuffer, so no need
    // to analyze the structure since the register number
    // won't be pushed any higher.
    return (SlotNum > Limit);
 
  // we don't expect any other decl type, so fail
  llvm_unreachable("unexpected decl type");
}
 
void SemaHLSL::handleResourceBindingAttr(Decl *TheDecl, const ParsedAttr &AL) {
  if (VarDecl *VD = dyn_cast<VarDecl>(TheDecl)) {
    QualType Ty = VD->getType();
    if (const auto *IAT = dyn_cast<IncompleteArrayType>(Ty))
      Ty = IAT->getElementType();
    if (SemaRef.RequireCompleteType(TheDecl->getBeginLoc(), Ty,
                                    diag::err_incomplete_type))
      return;
  }
 
  StringRef Slot = "";
  StringRef Space = "";
  SourceLocation SlotLoc, SpaceLoc;
 
  if (!AL.isArgIdent(0)) {
    Diag(AL.getLoc(), diag::err_attribute_argument_type)
        << AL << AANT_ArgumentIdentifier;
    return;
  }
  IdentifierLoc *Loc = AL.getArgAsIdent(0);
 
  if (AL.getNumArgs() == 2) {
    Slot = Loc->getIdentifierInfo()->getName();
    SlotLoc = Loc->getLoc();
    if (!AL.isArgIdent(1)) {
      Diag(AL.getLoc(), diag::err_attribute_argument_type)
          << AL << AANT_ArgumentIdentifier;
      return;
    }
    Loc = AL.getArgAsIdent(1);
    Space = Loc->getIdentifierInfo()->getName();
    SpaceLoc = Loc->getLoc();
  } else {
    StringRef Str = Loc->getIdentifierInfo()->getName();
    if (Str.starts_with("space")) {
      Space = Str;
      SpaceLoc = Loc->getLoc();
    } else {
      Slot = Str;
      SlotLoc = Loc->getLoc();
      Space = "space0";
    }
  }
 
  RegisterType RegType = RegisterType::SRV;
  std::optional<unsigned> SlotNum;
  unsigned SpaceNum = 0;
 
  // Validate slot
  if (!Slot.empty()) {
    if (!convertToRegisterType(Slot, &RegType)) {
      Diag(SlotLoc, diag::err_hlsl_binding_type_invalid) << Slot.substr(0, 1);
      return;
    }
    if (RegType == RegisterType::I) {
      Diag(SlotLoc, diag::warn_hlsl_deprecated_register_type_i);
      return;
    }
    const StringRef SlotNumStr = Slot.substr(1);
 
    uint64_t N;
 
    // validate that the slot number is a non-empty number
    if (SlotNumStr.getAsInteger(10, N)) {
      Diag(SlotLoc, diag::err_hlsl_unsupported_register_number);
      return;
    }
 
    // Validate register number. It should not exceed UINT32_MAX,
    // including if the resource type is an array that starts
    // before UINT32_MAX, but ends afterwards.
    if (ValidateRegisterNumber(N, TheDecl, getASTContext(), RegType)) {
      Diag(SlotLoc, diag::err_hlsl_register_number_too_large);
      return;
    }
 
    // the slot number has been validated and does not exceed UINT32_MAX
    SlotNum = (unsigned)N;
  }
 
  // Validate space
  if (!Space.starts_with("space")) {
    Diag(SpaceLoc, diag::err_hlsl_expected_space) << Space;
    return;
  }
  StringRef SpaceNumStr = Space.substr(5);
  if (SpaceNumStr.getAsInteger(10, SpaceNum)) {
    Diag(SpaceLoc, diag::err_hlsl_expected_space) << Space;
    return;
  }
 
  // If we have slot, diagnose it is the right register type for the decl
  if (SlotNum.has_value())
    if (!DiagnoseHLSLRegisterAttribute(SemaRef, SlotLoc, TheDecl, RegType,
                                       !SpaceLoc.isInvalid()))
      return;
 
  HLSLResourceBindingAttr *NewAttr =
      HLSLResourceBindingAttr::Create(getASTContext(), Slot, Space, AL);
  if (NewAttr) {
    NewAttr->setBinding(RegType, SlotNum, SpaceNum);
    TheDecl->addAttr(NewAttr);
  }
}
 
void SemaHLSL::handleParamModifierAttr(Decl *D, const ParsedAttr &AL) {
  HLSLParamModifierAttr *NewAttr = mergeParamModifierAttr(
      D, AL,
      static_cast<HLSLParamModifierAttr::Spelling>(AL.getSemanticSpelling()));
  if (NewAttr)
    D->addAttr(NewAttr);
}
 
static bool isMatrixType(QualType QT) {
  const Type *Ty = QT->getUnqualifiedDesugaredType();
  return Ty->isDependentType() || Ty->isConstantMatrixType();
}
 
/// Walks the existing AttributedType sugar of \p T looking for a previously
/// applied HLSLRowMajor/HLSLColumnMajor marker. If one is found, populates
/// \p ExistingKind with its attr::Kind and returns true.
static bool findExistingMatrixLayoutMarker(QualType T,
                                           attr::Kind &ExistingKind) {
  QualType Cur = T;
  while (const auto *AT = Cur->getAs<AttributedType>()) {
    attr::Kind K = AT->getAttrKind();
    if (K == attr::HLSLRowMajor || K == attr::HLSLColumnMajor) {
      ExistingKind = K;
      return true;
    }
    Cur = AT->getModifiedType();
  }
  return false;
}
 
Attr *SemaHLSL::buildMatrixLayoutTypeAttr(QualType T, const ParsedAttr &AL) {
  if (T.isNull())
    return nullptr;
 
  ASTContext &Ctx = getASTContext();
  attr::Kind AttrK = AL.getKind() == ParsedAttr::AT_HLSLRowMajor
                         ? attr::HLSLRowMajor
                         : attr::HLSLColumnMajor;
 
  // For non-dependent types, the operand must be a matrix.
  if (!T->isDependentType() && !isMatrixType(T)) {
    Diag(AL.getLoc(), diag::err_hlsl_matrix_layout_non_matrix)
        << AL.getAttrName();
    AL.setInvalid();
    return nullptr;
  }
 
  // Conflict / duplicate detection by walking existing sugar.
  attr::Kind ExistingKind;
  if (findExistingMatrixLayoutMarker(T, ExistingKind)) {
    if (ExistingKind == AttrK) {
      Diag(AL.getLoc(), diag::warn_duplicate_attribute_exact)
          << AL.getAttrName();
      Diag(AL.getLoc(), diag::note_previous_attribute);
      return nullptr;
    }
    IdentifierInfo *ExistingII = &Ctx.Idents.get(
        ExistingKind == attr::HLSLRowMajor ? "row_major" : "column_major");
    Diag(AL.getLoc(), diag::err_hlsl_matrix_layout_conflict)
        << AL.getAttrName() << ExistingII;
    Diag(AL.getLoc(), diag::note_conflicting_attribute);
    AL.setInvalid();
    return nullptr;
  }
 
  if (AttrK == attr::HLSLRowMajor)
    return ::new (Ctx) HLSLRowMajorAttr(Ctx, AL);
  return ::new (Ctx) HLSLColumnMajorAttr(Ctx, AL);
}
 
// Re-validates an HLSL `row_major` / `column_major` attribute after template
// substitution. The parse-time check in `buildMatrixLayoutTypeAttr` is skipped
// for dependent types; `TransformAttributedType` calls this once the type is
// concrete. Returns `true` (and emits a diagnostic) if the substituted type is
// not a matrix or array of matrices, signaling the caller to abort the
// transform.
bool SemaHLSL::diagnoseMatrixLayoutInstantiation(attr::Kind K, QualType T,
                                                 SourceLocation Loc) {
  if (K != attr::HLSLRowMajor && K != attr::HLSLColumnMajor)
    return false;
  if (T.isNull() || T->isDependentType())
    return false;
  if (isMatrixType(T))
    return false;
  IdentifierInfo *II = &getASTContext().Idents.get(
      K == attr::HLSLRowMajor ? "row_major" : "column_major");
  Diag(Loc, diag::err_hlsl_matrix_layout_non_matrix) << II;
  return true;
}
 
// Transpose and matrix mul need to read the destination layout.
// Elementwise builtins reuse the operand layout instead.
namespace {
 
using llvm::dxil::BarrierMemoryTypeFlag;
using llvm::dxil::BarrierSemanticFlag;
 
template <typename T> constexpr uint64_t barrierFlagValue(T Flag) {
  return llvm::to_underlying(Flag);
}
 
/// This class implements reachable HLSL diagnostics.
///
/// It diagnoses unavailable APIs in default and relaxed availability modes.
/// It also validates Barrier calls in all availability modes.
///
/// This is done by traversing the AST of all shader entry point functions
/// and of all exported functions, and any functions that are referenced
/// from this AST. In other words, any functions that are reachable from
/// the entry points.
class DiagnoseHLSLAvailability : public DynamicRecursiveASTVisitor {
  Sema &SemaRef;
  bool DiagnoseAvailability;
 
  // Stack of functions to be scaned
  llvm::SmallVector<const FunctionDecl *, 8> DeclsToScan;
 
  // Tracks which environments functions have been scanned in.
  //
  // Maps FunctionDecl to an unsigned number that represents the set of shader
  // environments the function has been scanned for.
  // The llvm::Triple::EnvironmentType enum values for shader stages guaranteed
  // to be numbered from llvm::Triple::Pixel to llvm::Triple::Amplification
  // (verified by static_asserts in Triple.cpp), we can use it to index
  // individual bits in the set, as long as we shift the values to start with 0
  // by subtracting the value of llvm::Triple::Pixel first.
  //
  // The N'th bit in the set will be set if the function has been scanned
  // in shader environment whose llvm::Triple::EnvironmentType integer value
  // equals (llvm::Triple::Pixel + N).
  //
  // For example, if a function has been scanned in compute and pixel stage
  // environment, the value will be 0x21 (100001 binary) because:
  //
  //   (int)(llvm::Triple::Pixel - llvm::Triple::Pixel) == 0
  //   (int)(llvm::Triple::Compute - llvm::Triple::Pixel) == 5
  //
  // A FunctionDecl is mapped to 0 (or not included in the map) if it has not
  // been scanned in any environment.
  llvm::DenseMap<const FunctionDecl *, unsigned> ScannedDecls;
 
  // Do not access these directly, use the get/set methods below to make
  // sure the values are in sync
  llvm::Triple::EnvironmentType CurrentShaderEnvironment;
  unsigned CurrentShaderStageBit;
 
  // True if scanning a function that was already scanned in a different
  // shader stage context. Suppress stage-independent diagnostics because
  // they were reported during the first scan.
  bool ReportOnlyShaderStageIssues;
 
  // Helper methods for dealing with current stage context / environment
  void SetShaderStageContext(llvm::Triple::EnvironmentType ShaderType) {
    static_assert(sizeof(unsigned) >= 4);
    assert(HLSLShaderAttr::isValidShaderType(ShaderType));
    assert((unsigned)(ShaderType - llvm::Triple::Pixel) < 31 &&
           "ShaderType is too big for this bitmap"); // 31 is reserved for
                                                     // "unknown"
 
    unsigned bitmapIndex = ShaderType - llvm::Triple::Pixel;
    CurrentShaderEnvironment = ShaderType;
    CurrentShaderStageBit = (1 << bitmapIndex);
  }
 
  void SetUnknownShaderStageContext() {
    CurrentShaderEnvironment = llvm::Triple::UnknownEnvironment;
    CurrentShaderStageBit = (1 << 31);
  }
 
  llvm::Triple::EnvironmentType GetCurrentShaderEnvironment() const {
    return CurrentShaderEnvironment;
  }
 
  bool InUnknownShaderStageContext() const {
    return CurrentShaderEnvironment == llvm::Triple::UnknownEnvironment;
  }
 
  // Helper methods for dealing with shader stage bitmap
  void AddToScannedFunctions(const FunctionDecl *FD) {
    unsigned &ScannedStages = ScannedDecls[FD];
    ScannedStages |= CurrentShaderStageBit;
  }
 
  unsigned GetScannedStages(const FunctionDecl *FD) { return ScannedDecls[FD]; }
 
  bool WasAlreadyScannedInCurrentStage(const FunctionDecl *FD) {
    return WasAlreadyScannedInCurrentStage(GetScannedStages(FD));
  }
 
  bool WasAlreadyScannedInCurrentStage(unsigned ScannerStages) {
    return ScannerStages & CurrentShaderStageBit;
  }
 
  static bool NeverBeenScanned(unsigned ScannedStages) {
    return ScannedStages == 0;
  }
 
  // Scanning methods
  void HandleFunctionOrMethodRef(FunctionDecl *FD, Expr *RefExpr);
  void CheckDeclAvailability(NamedDecl *D, const AvailabilityAttr *AA,
                             SourceRange Range);
  const AvailabilityAttr *FindAvailabilityAttr(const Decl *D);
  bool HasMatchingEnvironmentOrNone(const AvailabilityAttr *AA);
  void DiagnoseBarrierCall(CallExpr *CE);
  uint64_t DiagnoseBarrierGroupMemory(Expr *MemoryArg, uint64_t MemoryFlags,
                                      bool HasVisibleGroup, bool IsAllMemory);
  uint64_t DiagnoseBarrierNodeMemory(Expr *MemoryArg, uint64_t MemoryFlags,
                                     bool HasKnownStage, bool IsAllMemory);
  void DiagnoseBarrierGroupSemantic(Expr *SemanticArg, uint64_t SemanticFlags,
                                    bool HasVisibleGroup);
  void DiagnoseBarrierScope(Expr *SemanticArg, uint64_t MemoryFlags,
                            uint64_t SemanticFlags);
 
public:
  DiagnoseHLSLAvailability(Sema &SemaRef, bool DiagnoseAvailability)
      : SemaRef(SemaRef), DiagnoseAvailability(DiagnoseAvailability),
        CurrentShaderEnvironment(llvm::Triple::UnknownEnvironment),
        CurrentShaderStageBit(0), ReportOnlyShaderStageIssues(false) {}
 
  // AST traversal methods
  void RunOnTranslationUnit(const TranslationUnitDecl *TU);
  void RunOnFunction(const FunctionDecl *FD);
 
  bool VisitDeclRefExpr(DeclRefExpr *DRE) override {
    FunctionDecl *FD = llvm::dyn_cast<FunctionDecl>(DRE->getDecl());
    if (FD)
      HandleFunctionOrMethodRef(FD, DRE);
    return true;
  }
 
  bool VisitMemberExpr(MemberExpr *ME) override {
    FunctionDecl *FD = llvm::dyn_cast<FunctionDecl>(ME->getMemberDecl());
    if (FD)
      HandleFunctionOrMethodRef(FD, ME);
    return true;
  }
 
  bool VisitCallExpr(CallExpr *CE) override {
    DiagnoseBarrierCall(CE);
    return true;
  }
};
 
uint64_t DiagnoseHLSLAvailability::DiagnoseBarrierGroupMemory(
    Expr *MemoryArg, uint64_t MemoryFlags, bool HasVisibleGroup,
    bool IsAllMemory) {
  const uint64_t GroupSharedMemory =
      barrierFlagValue(BarrierMemoryTypeFlag::GroupSharedMemory);
  if (HasVisibleGroup || (MemoryFlags & GroupSharedMemory) == 0)
    return MemoryFlags;
 
  if (!IsAllMemory) {
    SemaRef.Diag(MemoryArg->getExprLoc(),
                 diag::err_hlsl_barrier_flag_requires_group)
        << 0;
    return MemoryFlags;
  }
 
  return MemoryFlags & ~GroupSharedMemory;
}
 
uint64_t DiagnoseHLSLAvailability::DiagnoseBarrierNodeMemory(
    Expr *MemoryArg, uint64_t MemoryFlags, bool HasKnownStage,
    bool IsAllMemory) {
  const uint64_t NodeMemory =
      barrierFlagValue(BarrierMemoryTypeFlag::NodeMemory);
  if (!HasKnownStage || (MemoryFlags & NodeMemory) == 0)
    return MemoryFlags;
 
  if (!IsAllMemory) {
    SemaRef.Diag(MemoryArg->getExprLoc(),
                 diag::err_hlsl_barrier_node_memory_requires_node);
    return MemoryFlags;
  }
 
  return MemoryFlags & ~NodeMemory;
}
 
void DiagnoseHLSLAvailability::DiagnoseBarrierGroupSemantic(
    Expr *SemanticArg, uint64_t SemanticFlags, bool HasVisibleGroup) {
  if (HasVisibleGroup ||
      (SemanticFlags & barrierFlagValue(BarrierSemanticFlag::GroupFlags)) == 0)
    return;
 
  SemaRef.Diag(SemanticArg->getExprLoc(),
               diag::err_hlsl_barrier_flag_requires_group)
      << ((SemanticFlags & barrierFlagValue(BarrierSemanticFlag::GroupSync)) !=
                  0
              ? 1
              : 2);
}
 
void DiagnoseHLSLAvailability::DiagnoseBarrierScope(Expr *SemanticArg,
                                                    uint64_t MemoryFlags,
                                                    uint64_t SemanticFlags) {
  if (ReportOnlyShaderStageIssues)
    return;
 
  const uint64_t DeviceScopeMemory =
      barrierFlagValue(BarrierMemoryTypeFlag::UAVMemory) |
      barrierFlagValue(BarrierMemoryTypeFlag::NodeInputMemory);
  if ((SemanticFlags & barrierFlagValue(BarrierSemanticFlag::DeviceScope)) !=
          0 &&
      (MemoryFlags & DeviceScopeMemory) == 0)
    SemaRef.Diag(SemanticArg->getExprLoc(),
                 diag::err_hlsl_barrier_scope_requires_memory)
        << 1;
  if ((SemanticFlags & barrierFlagValue(BarrierSemanticFlag::GroupScope)) !=
          0 &&
      MemoryFlags == 0)
    SemaRef.Diag(SemanticArg->getExprLoc(),
                 diag::err_hlsl_barrier_scope_requires_memory)
        << 0;
}
 
void DiagnoseHLSLAvailability::DiagnoseBarrierCall(CallExpr *CE) {
  const FunctionDecl *FD = CE->getDirectCallee();
  if (!FD || FD->getBuiltinID() != Builtin::BI__builtin_hlsl_barrier)
    return;
 
  const llvm::Triple::EnvironmentType Stage = GetCurrentShaderEnvironment();
  const bool HasKnownStage = !InUnknownShaderStageContext();
  const bool HasVisibleGroup =
      !HasKnownStage || Stage == llvm::Triple::Compute ||
      Stage == llvm::Triple::Mesh || Stage == llvm::Triple::Amplification;
 
  uint64_t MemoryFlags = barrierFlagValue(BarrierMemoryTypeFlag::ValidMask);
  Expr *MemoryArg = CE->getArg(0);
  if (MemoryArg->getType()->isUnsignedIntegerType()) {
    std::optional<llvm::APSInt> Value =
        MemoryArg->getIntegerConstantExpr(SemaRef.Context);
    if (!Value)
      return;
    MemoryFlags = Value->getZExtValue();
    const bool IsAllMemory =
        MemoryFlags == barrierFlagValue(BarrierMemoryTypeFlag::ValidMask);
 
    MemoryFlags = DiagnoseBarrierGroupMemory(MemoryArg, MemoryFlags,
                                             HasVisibleGroup, IsAllMemory);
    MemoryFlags = DiagnoseBarrierNodeMemory(MemoryArg, MemoryFlags,
                                            HasKnownStage, IsAllMemory);
  } else if (!HasVisibleGroup) {
    SemaRef.Diag(MemoryArg->getExprLoc(),
                 diag::err_hlsl_barrier_resource_requires_group);
    return;
  }
 
  Expr *SemanticArg = CE->getArg(1);
  std::optional<llvm::APSInt> Value =
      SemanticArg->getIntegerConstantExpr(SemaRef.Context);
  if (!Value)
    return;
  const uint64_t SemanticFlags = Value->getZExtValue();
 
  DiagnoseBarrierGroupSemantic(SemanticArg, SemanticFlags, HasVisibleGroup);
 
  if (MemoryArg->getType()->isUnsignedIntegerType())
    DiagnoseBarrierScope(SemanticArg, MemoryFlags, SemanticFlags);
}
 
void DiagnoseHLSLAvailability::HandleFunctionOrMethodRef(FunctionDecl *FD,
                                                         Expr *RefExpr) {
  assert((isa<DeclRefExpr>(RefExpr) || isa<MemberExpr>(RefExpr)) &&
         "expected DeclRefExpr or MemberExpr");
 
  if (DiagnoseAvailability)
    if (const AvailabilityAttr *AA = FindAvailabilityAttr(FD))
      CheckDeclAvailability(
          FD, AA, SourceRange(RefExpr->getBeginLoc(), RefExpr->getEndLoc()));
 
  // has a definition -> add to stack to be scanned
  const FunctionDecl *FDWithBody = nullptr;
  if (FD->hasBody(FDWithBody) && !WasAlreadyScannedInCurrentStage(FDWithBody))
    DeclsToScan.push_back(FDWithBody);
}
 
void DiagnoseHLSLAvailability::RunOnTranslationUnit(
    const TranslationUnitDecl *TU) {
  const TargetInfo &TargetInfo = SemaRef.getASTContext().getTargetInfo();
  std::string &EntryName = TargetInfo.getTargetOpts().HLSLEntry;
  bool IsLibraryShader = TargetInfo.getTriple().getEnvironment() ==
                         llvm::Triple::EnvironmentType::Library;
  SourceLocation EntryLoc{};
 
  // Iterate over all shader entry functions and library exports, and for those
  // that have a body (definiton), run diag scan on each, setting appropriate
  // shader environment context based on whether it is a shader entry function
  // or an exported function. Exported functions can be in namespaces and in
  // export declarations so we need to scan those declaration contexts as well.
  llvm::SmallVector<const DeclContext *, 8> DeclContextsToScan;
  DeclContextsToScan.push_back(TU);
 
  while (!DeclContextsToScan.empty()) {
    const DeclContext *DC = DeclContextsToScan.pop_back_val();
    for (auto &D : DC->decls()) {
      // do not scan implicit declaration generated by the implementation
      if (D->isImplicit())
        continue;
 
      // for namespace or export declaration add the context to the list to be
      // scanned later
      if (llvm::dyn_cast<NamespaceDecl>(D) || llvm::dyn_cast<ExportDecl>(D)) {
        DeclContextsToScan.push_back(llvm::dyn_cast<DeclContext>(D));
        continue;
      }
 
      // skip over other decls or function decls without body
      const FunctionDecl *FD = llvm::dyn_cast<FunctionDecl>(D);
      if (!FD || !FD->isThisDeclarationADefinition())
        continue;
 
      // shader entry point
      if (HLSLShaderAttr *ShaderAttr = FD->getAttr<HLSLShaderAttr>()) {
        if (!IsLibraryShader && FD->getName() == EntryName) {
          if (EntryLoc.isValid()) {
            SemaRef.Diag(FD->getLocation(),
                         diag::err_hlsl_ambiguous_entry_point)
                << EntryName;
            SemaRef.Diag(EntryLoc, diag::note_previous_declaration_as)
                << EntryName;
            return;
          }
          EntryLoc = FD->getLocation();
        }
        SetShaderStageContext(ShaderAttr->getType());
        RunOnFunction(FD);
        continue;
      }
      // exported library function
      // FIXME: replace this loop with external linkage check once issue #92071
      // is resolved
      bool isExport = FD->isInExportDeclContext();
      if (!isExport) {
        for (const auto *Redecl : FD->redecls()) {
          if (Redecl->isInExportDeclContext()) {
            isExport = true;
            break;
          }
        }
      }
      if (isExport) {
        SetUnknownShaderStageContext();
        RunOnFunction(FD);
        continue;
      }
    }
  }
 
  if (!IsLibraryShader && EntryLoc.isInvalid()) {
    SemaRef.Diag(TU->getLocation(), diag::err_hlsl_missing_entry_point)
        << EntryName;
    return;
  }
}
 
void DiagnoseHLSLAvailability::RunOnFunction(const FunctionDecl *FD) {
  assert(DeclsToScan.empty() && "DeclsToScan should be empty");
  DeclsToScan.push_back(FD);
 
  while (!DeclsToScan.empty()) {
    // Take one decl from the stack and check it by traversing its AST.
    // For any CallExpr found during the traversal add it's callee to the top of
    // the stack to be processed next. Functions already processed are stored in
    // ScannedDecls.
    const FunctionDecl *FD = DeclsToScan.pop_back_val();
 
    // Decl was already scanned
    const unsigned ScannedStages = GetScannedStages(FD);
    if (WasAlreadyScannedInCurrentStage(ScannedStages))
      continue;
 
    ReportOnlyShaderStageIssues = !NeverBeenScanned(ScannedStages);
 
    AddToScannedFunctions(FD);
    TraverseStmt(FD->getBody());
  }
}
 
bool DiagnoseHLSLAvailability::HasMatchingEnvironmentOrNone(
    const AvailabilityAttr *AA) {
  const IdentifierInfo *IIEnvironment = AA->getEnvironment();
  if (!IIEnvironment)
    return true;
 
  llvm::Triple::EnvironmentType CurrentEnv = GetCurrentShaderEnvironment();
  if (CurrentEnv == llvm::Triple::UnknownEnvironment)
    return false;
 
  llvm::Triple::EnvironmentType AttrEnv =
      AvailabilityAttr::getEnvironmentType(IIEnvironment->getName());
 
  return CurrentEnv == AttrEnv;
}
 
const AvailabilityAttr *
DiagnoseHLSLAvailability::FindAvailabilityAttr(const Decl *D) {
  AvailabilityAttr const *PartialMatch = nullptr;
  // Check each AvailabilityAttr to find the one for this platform.
  // For multiple attributes with the same platform try to find one for this
  // environment.
  for (const auto *A : D->attrs()) {
    if (const auto *Avail = dyn_cast<AvailabilityAttr>(A)) {
      const AvailabilityAttr *EffectiveAvail = Avail->getEffectiveAttr();
      StringRef AttrPlatform = EffectiveAvail->getPlatform()->getName();
      StringRef TargetPlatform =
          SemaRef.getASTContext().getTargetInfo().getPlatformName();
 
      // Match the platform name.
      if (AttrPlatform == TargetPlatform) {
        // Find the best matching attribute for this environment
        if (HasMatchingEnvironmentOrNone(EffectiveAvail))
          return Avail;
        PartialMatch = Avail;
      }
    }
  }
  return PartialMatch;
}
 
// Check availability against target shader model version and current shader
// stage and emit diagnostic
void DiagnoseHLSLAvailability::CheckDeclAvailability(NamedDecl *D,
                                                     const AvailabilityAttr *AA,
                                                     SourceRange Range) {
 
  const IdentifierInfo *IIEnv = AA->getEnvironment();
 
  if (!IIEnv) {
    // The availability attribute does not have environment -> it depends only
    // on shader model version and not on specific the shader stage.
 
    // Skip emitting the diagnostics if the diagnostic mode is set to
    // strict (-fhlsl-strict-availability) because all relevant diagnostics
    // were already emitted in the DiagnoseUnguardedAvailability scan
    // (SemaAvailability.cpp).
    if (SemaRef.getLangOpts().HLSLStrictAvailability)
      return;
 
    // Do not report shader-stage-independent issues if scanning a function
    // that was already scanned in a different shader stage context (they would
    // be duplicate)
    if (ReportOnlyShaderStageIssues)
      return;
 
  } else {
    // The availability attribute has environment -> we need to know
    // the current stage context to property diagnose it.
    if (InUnknownShaderStageContext())
      return;
  }
 
  // Check introduced version and if environment matches
  bool EnvironmentMatches = HasMatchingEnvironmentOrNone(AA);
  VersionTuple Introduced = AA->getIntroduced();
  VersionTuple TargetVersion =
      SemaRef.Context.getTargetInfo().getPlatformMinVersion();
 
  if (TargetVersion >= Introduced && EnvironmentMatches)
    return;
 
  // Emit diagnostic message
  const TargetInfo &TI = SemaRef.getASTContext().getTargetInfo();
  llvm::StringRef PlatformName(
      AvailabilityAttr::getPrettyPlatformName(TI.getPlatformName()));
 
  llvm::StringRef CurrentEnvStr =
      llvm::Triple::getEnvironmentTypeName(GetCurrentShaderEnvironment());
 
  llvm::StringRef AttrEnvStr =
      AA->getEnvironment() ? AA->getEnvironment()->getName() : "";
  bool UseEnvironment = !AttrEnvStr.empty();
 
  if (EnvironmentMatches) {
    SemaRef.Diag(Range.getBegin(), diag::warn_hlsl_availability)
        << Range << D << PlatformName << Introduced.getAsString()
        << UseEnvironment << CurrentEnvStr;
  } else {
    SemaRef.Diag(Range.getBegin(), diag::warn_hlsl_availability_unavailable)
        << Range << D;
  }
 
  SemaRef.Diag(D->getLocation(), diag::note_partial_availability_specified_here)
      << D << PlatformName << Introduced.getAsString()
      << SemaRef.Context.getTargetInfo().getPlatformMinVersion().getAsString()
      << UseEnvironment << AttrEnvStr << CurrentEnvStr;
}
 
} // namespace
 
void SemaHLSL::ActOnEndOfTranslationUnit(TranslationUnitDecl *TU) {
  // process default CBuffer - create buffer layout struct and invoke codegenCGH
  if (!DefaultCBufferDecls.empty()) {
    HLSLBufferDecl *DefaultCBuffer = HLSLBufferDecl::CreateDefaultCBuffer(
        SemaRef.getASTContext(), SemaRef.getCurLexicalContext(),
        DefaultCBufferDecls);
    addImplicitBindingAttrToDecl(SemaRef, DefaultCBuffer, RegisterType::CBuffer,
                                 getNextImplicitBindingOrderID());
    SemaRef.getCurLexicalContext()->addDecl(DefaultCBuffer);
    createHostLayoutStructForBuffer(SemaRef, DefaultCBuffer);
 
    // Set HasValidPackoffset if any of the decls has a register(c#) annotation;
    for (const Decl *VD : DefaultCBufferDecls) {
      const HLSLResourceBindingAttr *RBA =
          VD->getAttr<HLSLResourceBindingAttr>();
      if (RBA && RBA->hasRegisterSlot() &&
          RBA->getRegisterType() == HLSLResourceBindingAttr::RegisterType::C) {
        DefaultCBuffer->setHasValidPackoffset(true);
        break;
      }
    }
 
    DeclGroupRef DG(DefaultCBuffer);
    SemaRef.Consumer.HandleTopLevelDecl(DG);
  }
  diagnoseAvailabilityViolations(TU);
}
 
// For resource member access through a global struct array, verify that the
// array index selecting the struct element is a constant integer expression.
// Returns false if the member expression is invalid.
bool SemaHLSL::ActOnResourceMemberAccessExpr(MemberExpr *ME) {
  assert((ME->getType()->isHLSLResourceRecord() ||
          ME->getType()->isHLSLResourceRecordArray()) &&
         "expected member expr to have resource record type or array of them");
 
  // Walk the AST from MemberExpr to the VarDecl of the parent struct instance
  // and take note of any non-constant array indexing along the way. If the
  // VarDecl we find is a global variable, report error if there was any
  // non-constant array index in the resource member access along the way.
  const Expr *NonConstIndexExpr = nullptr;
  const Expr *E = ME->getBase();
  while (E) {
    if (const DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E)) {
      if (!NonConstIndexExpr)
        return true;
 
      const VarDecl *VD = cast<VarDecl>(DRE->getDecl());
      if (!VD->hasGlobalStorage())
        return true;
 
      SemaRef.Diag(NonConstIndexExpr->getExprLoc(),
                   diag::err_hlsl_resource_member_array_access_not_constant);
      return false;
    }
 
    if (const auto *ASE = dyn_cast<ArraySubscriptExpr>(E)) {
      const Expr *IdxExpr = ASE->getIdx();
      if (!IdxExpr->isIntegerConstantExpr(SemaRef.getASTContext()))
        NonConstIndexExpr = IdxExpr;
      E = ASE->getBase();
    } else if (const auto *SubME = dyn_cast<MemberExpr>(E)) {
      E = SubME->getBase();
    } else if (const auto *ICE = dyn_cast<ImplicitCastExpr>(E)) {
      E = ICE->getSubExpr();
    } else if (isa<CXXThisExpr>(E)) {
      return true;
    } else {
      llvm_unreachable("unexpected expr type in resource member access");
    }
  }
  return true;
}
 
NamedDecl *SemaHLSL::getConstantBufferConversionFunction(QualType Type,
                                                         CXXRecordDecl *RD) {
  QualType AddrSpaceType =
      SemaRef.Context.getCanonicalType(SemaRef.Context.getAddrSpaceQualType(
          Type.withConst(), LangAS::hlsl_constant));
  QualType ReturnTy = SemaRef.Context.getCanonicalType(
      SemaRef.Context.getLValueReferenceType(AddrSpaceType));
 
  DeclarationName ConvName =
      SemaRef.Context.DeclarationNames.getCXXConversionFunctionName(
          CanQualType::CreateUnsafe(ReturnTy));
  LookupResult ConvR(SemaRef, ConvName, SourceLocation(),
                     Sema::LookupOrdinaryName);
  [[maybe_unused]] bool LookupSucceeded =
      SemaRef.LookupQualifiedName(ConvR, RD);
  assert(LookupSucceeded);
 
  for (NamedDecl *D : ConvR) {
    if (isa<CXXConversionDecl>(D->getUnderlyingDecl()))
      return D;
  }
  return nullptr;
}
 
std::optional<ExprResult>
SemaHLSL::tryPerformConstantBufferConversion(Expr *BaseExpr) {
  QualType BaseType = BaseExpr->getType();
  const HLSLAttributedResourceType *ResTy =
      HLSLAttributedResourceType::findHandleTypeOnResource(
          BaseType.getTypePtr());
  if (!ResTy ||
      ResTy->getAttrs().ResourceClass != llvm::dxil::ResourceClass::CBuffer)
    return std::nullopt;
 
  QualType TemplateType = ResTy->getContainedType();
 
  NamedDecl *NamedConversionDecl = getConstantBufferConversionFunction(
      TemplateType, BaseType->getAsCXXRecordDecl());
  assert(NamedConversionDecl &&
         "Could not find conversion function for ConstantBuffer.");
  auto *ConversionDecl =
      cast<CXXConversionDecl>(NamedConversionDecl->getUnderlyingDecl());
 
  return SemaRef.BuildCXXMemberCallExpr(BaseExpr, NamedConversionDecl,
                                        ConversionDecl,
                                        /*HadMultipleCandidates=*/false);
}
 
void SemaHLSL::diagnoseAvailabilityViolations(TranslationUnitDecl *TU) {
  // Strict mode diagnoses availability during the
  // DiagnoseUnguardedAvailability scan in SemaAvailability.cpp. The reachable
  // function scan must still run to validate Barrier calls.
  const TargetInfo &TI = SemaRef.getASTContext().getTargetInfo();
  const bool DiagnoseAvailability =
      !SemaRef.getLangOpts().HLSLStrictAvailability ||
      TI.getTriple().getEnvironment() == llvm::Triple::EnvironmentType::Library;
  DiagnoseHLSLAvailability(SemaRef, DiagnoseAvailability)
      .RunOnTranslationUnit(TU);
}
 
static bool CheckAllArgsHaveSameType(Sema *S, CallExpr *TheCall) {
  assert(TheCall->getNumArgs() > 1);
  QualType ArgTy0 = TheCall->getArg(0)->getType();
 
  for (unsigned I = 1, N = TheCall->getNumArgs(); I < N; ++I) {
    if (!S->getASTContext().hasSameUnqualifiedType(
            ArgTy0, TheCall->getArg(I)->getType())) {
      S->Diag(TheCall->getBeginLoc(), diag::err_vec_builtin_incompatible_vector)
          << TheCall->getDirectCallee() << /*useAllTerminology*/ true
          << SourceRange(TheCall->getArg(0)->getBeginLoc(),
                         TheCall->getArg(N - 1)->getEndLoc());
      return true;
    }
  }
  return false;
}
 
static bool CheckArgTypeMatches(Sema *S, Expr *Arg, QualType ExpectedType) {
  QualType ArgType = Arg->getType();
  if (!S->getASTContext().hasSameUnqualifiedType(ArgType, ExpectedType)) {
    S->Diag(Arg->getBeginLoc(), diag::err_typecheck_convert_incompatible)
        << ArgType << ExpectedType << 1 << 0 << 0;
    return true;
  }
  return false;
}
 
static bool CheckAllArgTypesAreCorrect(
    Sema *S, CallExpr *TheCall,
    llvm::function_ref<bool(Sema *S, SourceLocation Loc, int ArgOrdinal,
                            clang::QualType PassedType)>
        Check) {
  for (unsigned I = 0; I < TheCall->getNumArgs(); ++I) {
    Expr *Arg = TheCall->getArg(I);
    if (Check(S, Arg->getBeginLoc(), I + 1, Arg->getType()))
      return true;
  }
  return false;
}
 
static bool CheckFloatRepresentation(Sema *S, SourceLocation Loc,
                                     int ArgOrdinal,
                                     clang::QualType PassedType) {
  clang::QualType BaseType =
      getElementTypeOf(PassedType, /*IncludeMatrix=*/false);
  if (!BaseType->isFloat32Type())
    return S->Diag(Loc, diag::err_builtin_invalid_arg_type)
           << ArgOrdinal << /* scalar or vector of */ 5 << /* no int */ 0
           << /* float */ 1 << PassedType;
  return false;
}
 
static bool CheckFloatOrHalfRepresentation(Sema *S, SourceLocation Loc,
                                           int ArgOrdinal,
                                           clang::QualType PassedType) {
  QualType BaseType = getScalarComponentType(PassedType);
 
  if (!BaseType->isHalfType() && !BaseType->isFloat32Type())
    return S->Diag(Loc, diag::err_builtin_invalid_arg_type)
           << ArgOrdinal << /* scalar or vector of */ 5 << /* no int */ 0
           << /* half or float */ 2 << PassedType;
  return false;
}
 
static bool CheckAnyDoubleRepresentation(Sema *S, SourceLocation Loc,
                                         int ArgOrdinal,
                                         clang::QualType PassedType) {
  QualType BaseType = getScalarComponentType(PassedType);
  if (!BaseType->isDoubleType()) {
    // FIXME: adopt standard `err_builtin_invalid_arg_type` instead of using
    // this custom error.
    return S->Diag(Loc, diag::err_builtin_requires_double_type)
           << ArgOrdinal << PassedType;
  }
 
  return false;
}
 
static bool CheckModifiableLValue(Sema *S, CallExpr *TheCall,
                                  unsigned ArgIndex) {
  auto *Arg = TheCall->getArg(ArgIndex);
  SourceLocation OrigLoc = Arg->getExprLoc();
  if (Arg->IgnoreCasts()->isModifiableLvalue(S->Context, &OrigLoc) ==
      Expr::MLV_Valid)
    return false;
  S->Diag(OrigLoc, diag::error_hlsl_inout_lvalue) << Arg << 0;
  return true;
}
 
// Verifies that the argument at `ArgIndex` of `TheCall` refers to memory in
// one of `AllowedSpaces`. Intended for HLSL builtins (e.g. atomics).
static bool CheckArgAddrSpaceOneOf(Sema *S, CallExpr *TheCall,
                                   unsigned ArgIndex,
                                   ArrayRef<LangAS> AllowedSpaces) {
  Expr *Arg = TheCall->getArg(ArgIndex);
  QualType LValueTy = Arg->IgnoreCasts()->getType();
  if (llvm::is_contained(AllowedSpaces, LValueTy.getAddressSpace()))
    return false;
  S->Diag(Arg->getBeginLoc(), diag::err_hlsl_atomic_arg_addr_space)
      << (ArgIndex + 1) << LValueTy;
  return true;
}
 
static bool CheckNoDoubleVectors(Sema *S, SourceLocation Loc, int ArgOrdinal,
                                 clang::QualType PassedType) {
  const auto *VecTy = PassedType->getAs<VectorType>();
  if (!VecTy)
    return false;
 
  if (VecTy->getElementType()->isDoubleType())
    return S->Diag(Loc, diag::err_builtin_invalid_arg_type)
           << ArgOrdinal << /* scalar */ 1 << /* no int */ 0 << /* fp */ 1
           << PassedType;
  return false;
}
 
static bool CheckFloatingOrIntRepresentation(Sema *S, SourceLocation Loc,
                                             int ArgOrdinal,
                                             clang::QualType PassedType) {
  if (!PassedType->hasIntegerRepresentation() &&
      !PassedType->hasFloatingRepresentation())
    return S->Diag(Loc, diag::err_builtin_invalid_arg_type)
           << ArgOrdinal << /* scalar or vector of */ 5 << /* integer */ 1
           << /* fp */ 1 << PassedType;
  return false;
}
 
static bool CheckUnsignedIntVecRepresentation(Sema *S, SourceLocation Loc,
                                              int ArgOrdinal,
                                              clang::QualType PassedType) {
  if (auto *VecTy = PassedType->getAs<VectorType>())
    if (VecTy->getElementType()->isUnsignedIntegerType())
      return false;
 
  return S->Diag(Loc, diag::err_builtin_invalid_arg_type)
         << ArgOrdinal << /* vector of */ 4 << /* uint */ 3 << /* no fp */ 0
         << PassedType;
}
 
// checks for unsigned ints of all sizes
static bool CheckUnsignedIntRepresentation(Sema *S, SourceLocation Loc,
                                           int ArgOrdinal,
                                           clang::QualType PassedType) {
  if (!PassedType->hasUnsignedIntegerRepresentation())
    return S->Diag(Loc, diag::err_builtin_invalid_arg_type)
           << ArgOrdinal << /* scalar or vector of */ 5 << /* unsigned int */ 3
           << /* no fp */ 0 << PassedType;
  return false;
}
 
static bool CheckExpectedBitWidth(Sema *S, CallExpr *TheCall,
                                  unsigned ArgOrdinal, unsigned Width) {
  QualType ArgTy = TheCall->getArg(0)->getType();
  if (auto *VTy = ArgTy->getAs<VectorType>())
    ArgTy = VTy->getElementType();
  // ensure arg type has expected bit width
  uint64_t ElementBitCount =
      S->getASTContext().getTypeSizeInChars(ArgTy).getQuantity() * 8;
  if (ElementBitCount != Width) {
    S->Diag(TheCall->getArg(0)->getBeginLoc(),
            diag::err_integer_incorrect_bit_count)
        << Width << ElementBitCount;
    return true;
  }
  return false;
}
 
static void SetElementTypeAsReturnType(Sema *S, CallExpr *TheCall,
                                       QualType ReturnType) {
  if (auto *VecTyA = TheCall->getArg(0)->getType()->getAs<VectorType>())
    ReturnType =
        S->Context.getExtVectorType(ReturnType, VecTyA->getNumElements());
  else if (auto *MatTyA =
               TheCall->getArg(0)->getType()->getAs<ConstantMatrixType>())
    ReturnType = S->Context.getConstantMatrixType(
        ReturnType, MatTyA->getNumRows(), MatTyA->getNumColumns());
 
  TheCall->setType(ReturnType);
}
 
static bool CheckScalarOrVector(Sema *S, CallExpr *TheCall, QualType Scalar,
                                unsigned ArgIndex) {
  assert(TheCall->getNumArgs() >= ArgIndex);
  QualType ArgType = TheCall->getArg(ArgIndex)->getType();
  auto *VTy = ArgType->getAs<VectorType>();
  // not the scalar or vector<scalar>
  if (!(S->Context.hasSameUnqualifiedType(ArgType, Scalar) ||
        (VTy &&
         S->Context.hasSameUnqualifiedType(VTy->getElementType(), Scalar)))) {
    S->Diag(TheCall->getArg(0)->getBeginLoc(),
            diag::err_typecheck_expect_scalar_or_vector)
        << ArgType << Scalar;
    return true;
  }
  return false;
}
 
static bool CheckScalarOrVectorOrMatrix(Sema *S, CallExpr *TheCall,
                                        QualType Scalar, unsigned ArgIndex) {
  assert(TheCall->getNumArgs() > ArgIndex);
 
  Expr *Arg = TheCall->getArg(ArgIndex);
  QualType ArgType = Arg->getType();
 
  // Scalar: T
  if (S->Context.hasSameUnqualifiedType(ArgType, Scalar))
    return false;
 
  // Vector: vector<T>
  if (const auto *VTy = ArgType->getAs<VectorType>()) {
    if (S->Context.hasSameUnqualifiedType(VTy->getElementType(), Scalar))
      return false;
  }
 
  // Matrix: ConstantMatrixType with element type T
  if (const auto *MTy = ArgType->getAs<ConstantMatrixType>()) {
    if (S->Context.hasSameUnqualifiedType(MTy->getElementType(), Scalar))
      return false;
  }
 
  // Not a scalar/vector/matrix-of-scalar
  S->Diag(Arg->getBeginLoc(),
          diag::err_typecheck_expect_scalar_or_vector_or_matrix)
      << ArgType << Scalar;
  return true;
}
 
static bool CheckAnyScalarOrVector(Sema *S, CallExpr *TheCall,
                                   unsigned ArgIndex) {
  assert(TheCall->getNumArgs() >= ArgIndex);
  QualType ArgType = TheCall->getArg(ArgIndex)->getType();
  auto *VTy = ArgType->getAs<VectorType>();
  // not the scalar or vector<scalar>
  if (!(ArgType->isScalarType() ||
        (VTy && VTy->getElementType()->isScalarType()))) {
    S->Diag(TheCall->getArg(0)->getBeginLoc(),
            diag::err_typecheck_expect_any_scalar_or_vector_or_matrix)
        << ArgType << 1;
    return true;
  }
  return false;
}
 
static bool CheckAnyScalarOrVectorOrMatrix(Sema *S, CallExpr *TheCall,
                                           unsigned ArgIndex) {
  assert(TheCall->getNumArgs() > ArgIndex);
  QualType ArgType = TheCall->getArg(ArgIndex)->getType();
  if (ArgType->isDependentType())
    return false;
 
  QualType ElementType = ArgType;
  if (const auto *VectorTy = ArgType->getAs<VectorType>())
    ElementType = VectorTy->getElementType();
  else if (const auto *MatrixTy = ArgType->getAs<ConstantMatrixType>())
    ElementType = MatrixTy->getElementType();
 
  if (ElementType->isBooleanType())
    return false;
 
  if (ElementType->isIntegerType() || ElementType->isRealFloatingType()) {
    unsigned BitWidth = S->Context.getTypeSize(ElementType);
    if (BitWidth == 16 || BitWidth == 32 || BitWidth == 64)
      return false;
  }
 
  S->Diag(TheCall->getArg(ArgIndex)->getBeginLoc(),
          diag::err_typecheck_expect_any_scalar_or_vector_or_matrix)
      << ArgType << 2;
  return true;
}
 
// Check that the argument is not a bool or vector<bool>
// Returns true on error
static bool CheckNotBoolScalarOrVector(Sema *S, CallExpr *TheCall,
                                       unsigned ArgIndex) {
  QualType BoolType = S->getASTContext().BoolTy;
  assert(ArgIndex < TheCall->getNumArgs());
  QualType ArgType = TheCall->getArg(ArgIndex)->getType();
  auto *VTy = ArgType->getAs<VectorType>();
  // is the bool or vector<bool>
  if (S->Context.hasSameUnqualifiedType(ArgType, BoolType) ||
      (VTy &&
       S->Context.hasSameUnqualifiedType(VTy->getElementType(), BoolType))) {
    S->Diag(TheCall->getArg(0)->getBeginLoc(),
            diag::err_typecheck_expect_any_scalar_or_vector_or_matrix)
        << ArgType << 0;
    return true;
  }
  return false;
}
 
static bool CheckWaveActive(Sema *S, CallExpr *TheCall) {
  if (CheckNotBoolScalarOrVector(S, TheCall, 0))
    return true;
  return false;
}
 
static bool CheckWavePrefix(Sema *S, CallExpr *TheCall) {
  if (CheckNotBoolScalarOrVector(S, TheCall, 0))
    return true;
  return false;
}
 
static bool CheckWaveMultiPrefixMask(Sema *S, CallExpr *TheCall) {
  QualType MaskTy = TheCall->getArg(1)->getType();
  QualType UInt4Ty = S->Context.getExtVectorType(S->Context.UnsignedIntTy, 4);
  if (S->Context.hasSameUnqualifiedType(MaskTy, UInt4Ty))
    return false;
 
  S->Diag(TheCall->getArg(1)->getBeginLoc(),
          diag::err_typecheck_convert_incompatible)
      << MaskTy << UInt4Ty << 1 << 0 << 0;
  return true;
}
 
static bool CheckBoolSelect(Sema *S, CallExpr *TheCall) {
  assert(TheCall->getNumArgs() == 3);
  Expr *Arg1 = TheCall->getArg(1);
  Expr *Arg2 = TheCall->getArg(2);
  if (!S->Context.hasSameUnqualifiedType(Arg1->getType(), Arg2->getType())) {
    S->Diag(TheCall->getBeginLoc(),
            diag::err_typecheck_call_different_arg_types)
        << Arg1->getType() << Arg2->getType() << Arg1->getSourceRange()
        << Arg2->getSourceRange();
    return true;
  }
 
  TheCall->setType(Arg1->getType());
  return false;
}
 
static bool CheckVectorSelect(Sema *S, CallExpr *TheCall) {
  assert(TheCall->getNumArgs() == 3);
  Expr *Arg1 = TheCall->getArg(1);
  QualType Arg1Ty = Arg1->getType();
  Expr *Arg2 = TheCall->getArg(2);
  QualType Arg2Ty = Arg2->getType();
 
  QualType Arg1ScalarTy = Arg1Ty;
  if (auto VTy = Arg1ScalarTy->getAs<VectorType>())
    Arg1ScalarTy = VTy->getElementType();
 
  QualType Arg2ScalarTy = Arg2Ty;
  if (auto VTy = Arg2ScalarTy->getAs<VectorType>())
    Arg2ScalarTy = VTy->getElementType();
 
  if (!S->Context.hasSameUnqualifiedType(Arg1ScalarTy, Arg2ScalarTy))
    S->Diag(Arg1->getBeginLoc(), diag::err_hlsl_builtin_scalar_vector_mismatch)
        << /* second and third */ 1 << TheCall->getCallee() << Arg1Ty << Arg2Ty;
 
  QualType Arg0Ty = TheCall->getArg(0)->getType();
  unsigned Arg0Length = Arg0Ty->getAs<VectorType>()->getNumElements();
  unsigned Arg1Length = Arg1Ty->isVectorType()
                            ? Arg1Ty->getAs<VectorType>()->getNumElements()
                            : 0;
  unsigned Arg2Length = Arg2Ty->isVectorType()
                            ? Arg2Ty->getAs<VectorType>()->getNumElements()
                            : 0;
  if (Arg1Length > 0 && Arg0Length != Arg1Length) {
    S->Diag(TheCall->getBeginLoc(),
            diag::err_typecheck_vector_lengths_not_equal)
        << Arg0Ty << Arg1Ty << TheCall->getArg(0)->getSourceRange()
        << Arg1->getSourceRange();
    return true;
  }
 
  if (Arg2Length > 0 && Arg0Length != Arg2Length) {
    S->Diag(TheCall->getBeginLoc(),
            diag::err_typecheck_vector_lengths_not_equal)
        << Arg0Ty << Arg2Ty << TheCall->getArg(0)->getSourceRange()
        << Arg2->getSourceRange();
    return true;
  }
 
  TheCall->setType(
      S->getASTContext().getExtVectorType(Arg1ScalarTy, Arg0Length));
  return false;
}
 
static bool CheckMatrixSelect(Sema *S, CallExpr *TheCall) {
  assert(TheCall->getNumArgs() == 3);
  Expr *Arg1 = TheCall->getArg(1);
  QualType Arg1Ty = Arg1->getType();
  Expr *Arg2 = TheCall->getArg(2);
  QualType Arg2Ty = Arg2->getType();
 
  QualType Arg1ScalarTy = Arg1Ty;
  if (auto MTy = Arg1ScalarTy->getAs<ConstantMatrixType>())
    Arg1ScalarTy = MTy->getElementType();
 
  QualType Arg2ScalarTy = Arg2Ty;
  if (auto MTy = Arg2ScalarTy->getAs<ConstantMatrixType>())
    Arg2ScalarTy = MTy->getElementType();
 
  if (!S->Context.hasSameUnqualifiedType(Arg1ScalarTy, Arg2ScalarTy))
    S->Diag(Arg1->getBeginLoc(), diag::err_hlsl_builtin_scalar_vector_mismatch)
        << /* second and third */ 1 << TheCall->getCallee() << Arg1Ty << Arg2Ty;
 
  QualType Arg0Ty = TheCall->getArg(0)->getType();
  auto *Arg0MatTy = Arg0Ty->getAs<ConstantMatrixType>();
  unsigned Arg0Rows = Arg0MatTy->getNumRows();
  unsigned Arg0Cols = Arg0MatTy->getNumColumns();
 
  for (Expr *Arg : {Arg1, Arg2}) {
    auto *MTy = Arg->getType()->getAs<ConstantMatrixType>();
    if (MTy &&
        (MTy->getNumRows() != Arg0Rows || MTy->getNumColumns() != Arg0Cols)) {
      S->Diag(TheCall->getBeginLoc(),
              diag::err_typecheck_vector_lengths_not_equal)
          << Arg0Ty << Arg->getType() << TheCall->getArg(0)->getSourceRange()
          << Arg->getSourceRange();
      return true;
    }
  }
 
  TheCall->setType(
      S->Context.getConstantMatrixType(Arg1ScalarTy, Arg0Rows, Arg0Cols));
  return false;
}
 
static QualType getVectorOrScalarType(Sema &S, QualType BaseType,
                                      unsigned Count) {
  return Count > 1 ? S.Context.getExtVectorType(BaseType, Count) : BaseType;
}
 
static bool CheckScalarFloatOperand(Sema &S, CallExpr *TheCall,
                                    unsigned ArgIndex) {
  return CheckArgTypeMatches(&S, TheCall->getArg(ArgIndex), S.Context.FloatTy);
}
 
static bool CheckIndexType(Sema *S, CallExpr *TheCall, unsigned IndexArgIndex) {
  assert(TheCall->getNumArgs() > IndexArgIndex && "Index argument missing");
  QualType ArgType = TheCall->getArg(IndexArgIndex)->getType();
  QualType IndexTy = ArgType;
  unsigned int ActualDim = 1;
  if (const auto *VTy = IndexTy->getAs<VectorType>()) {
    ActualDim = VTy->getNumElements();
    IndexTy = VTy->getElementType();
  }
  if (!IndexTy->isIntegerType()) {
    S->Diag(TheCall->getArg(IndexArgIndex)->getBeginLoc(),
            diag::err_typecheck_expect_int)
        << ArgType;
    return true;
  }
 
  QualType ResourceArgTy = TheCall->getArg(0)->getType();
  const HLSLAttributedResourceType *ResTy =
      ResourceArgTy.getTypePtr()->getAs<HLSLAttributedResourceType>();
  assert(ResTy && "Resource argument must be a resource");
  HLSLAttributedResourceType::Attributes ResAttrs = ResTy->getAttrs();
 
  unsigned int ExpectedDim = 1;
  if (ResAttrs.ResourceDimension != llvm::dxil::ResourceDimension::Unknown)
    ExpectedDim = getResourceDimensions(ResAttrs.ResourceDimension) +
                  (ResAttrs.IsArray ? 1 : 0);
 
  if (ActualDim != ExpectedDim) {
    S->Diag(TheCall->getArg(IndexArgIndex)->getBeginLoc(),
            diag::err_hlsl_builtin_resource_coordinate_dimension_mismatch)
        << cast<NamedDecl>(TheCall->getCalleeDecl()) << ExpectedDim
        << ActualDim;
    return true;
  }
 
  return false;
}
 
static bool CheckResourceHandle(
    Sema *S, CallExpr *TheCall, unsigned ArgIndex,
    llvm::function_ref<bool(const HLSLAttributedResourceType *ResType)> Check =
        nullptr) {
  assert(TheCall->getNumArgs() >= ArgIndex);
  QualType ArgType = TheCall->getArg(ArgIndex)->getType();
  const HLSLAttributedResourceType *ResTy =
      ArgType.getTypePtr()->getAs<HLSLAttributedResourceType>();
  if (!ResTy) {
    S->Diag(TheCall->getArg(ArgIndex)->getBeginLoc(),
            diag::err_typecheck_expect_hlsl_resource)
        << ArgType;
    return true;
  }
  if (Check && Check(ResTy)) {
    S->Diag(TheCall->getArg(ArgIndex)->getExprLoc(),
            diag::err_invalid_hlsl_resource_type)
        << ArgType;
    return true;
  }
  return false;
}
 
static QualType createCounterHandleType(ASTContext &AST,
                                        QualType MainHandleTy) {
  assert(MainHandleTy->isHLSLAttributedResourceType() &&
         "expected resource handle type");
  auto *MainResType = MainHandleTy->getAs<HLSLAttributedResourceType>();
  auto MainAttrs = MainResType->getAttrs();
  assert(!MainAttrs.IsCounter && "cannot create a counter from a counter");
  MainAttrs.IsCounter = true;
  return AST.getHLSLAttributedResourceType(MainResType->getWrappedType(),
                                           MainResType->getContainedType(),
                                           MainAttrs);
}
 
enum class SampleKind { Sample, Bias, Grad, Level, Cmp, CmpLevelZero };
 
static StringRef getSampleMethodName(SampleKind Kind) {
  switch (Kind) {
  case SampleKind::Sample:
    return "Sample";
  case SampleKind::Bias:
    return "SampleBias";
  case SampleKind::Grad:
    return "SampleGrad";
  case SampleKind::Level:
    return "SampleLevel";
  case SampleKind::Cmp:
    return "SampleCmp";
  case SampleKind::CmpLevelZero:
    return "SampleCmpLevelZero";
  }
  llvm_unreachable("Invalid SampleKind");
}
 
// Returns the name of the resource method whose body the sampling or gather
// builtin is being emitted into, which is the name the user called. This
// matters for methods that share a builtin, like 'Gather' and 'GatherRed'.
// Falls back to DefaultName if the builtin is used outside of a resource
// method.
static StringRef getCurrentResourceMethodName(Sema &S, StringRef DefaultName) {
  const auto *MD = dyn_cast_if_present<CXXMethodDecl>(S.getCurFunctionDecl());
  if (!MD || !MD->getDeclName().isIdentifier())
    return DefaultName;
 
  QualType RecordTy = S.Context.getCanonicalTagType(MD->getParent());
  if (!RecordTy->isHLSLResourceRecord())
    return DefaultName;
 
  return MD->getName();
}
 
// Sampling from and gathering on resources with a 'double' element type is not
// supported. Such resources are still valid declarations whose contents can be
// accessed by other means, like Load or the subscript operator.
static bool CheckNoDoubleElementType(Sema &S, CallExpr *TheCall,
                                     QualType ContainedType,
                                     StringRef DefaultName) {
  QualType EltTy = getElementTypeOf(ContainedType, /*IncludeMatrix=*/false);
  if (!EltTy->isSpecificBuiltinType(BuiltinType::Double))
    return false;
 
  S.Diag(TheCall->getBeginLoc(), diag::err_hlsl_sample_double_element_type)
      << getCurrentResourceMethodName(S, DefaultName) << ContainedType;
  return true;
}
 
// Sampling textures with an integer element type was introduced in SM 6.7 as
// part of Advanced Texture Operations. The shader model only applies to DirectX
// targets; Vulkan has no such restriction.
static bool CheckIntegerElementTypeShaderModel(Sema &S, CallExpr *TheCall,
                                               QualType ContainedType,
                                               SampleKind Kind) {
  // Comparison sampling requires a floating point element type at every shader
  // model, which the caller diagnoses.
  if (Kind == SampleKind::Cmp || Kind == SampleKind::CmpLevelZero)
    return false;
 
  // 'bool' is an integer type in HLSL, but sampling bool resources is never
  // allowed, so it must not be reported as requiring shader model 6.7.
  QualType EltTy = getElementTypeOf(ContainedType, /*IncludeMatrix=*/false);
  if (!EltTy->isIntegerType() || EltTy->isBooleanType())
    return false;
 
  const TargetInfo &TI = S.Context.getTargetInfo();
  if (!TI.getTriple().isDXIL())
    return false;
 
  VersionTuple SMVersion = TI.getPlatformMinVersion();
  if (SMVersion >= VersionTuple(6, 7))
    return false;
 
  S.Diag(TheCall->getBeginLoc(), diag::err_hlsl_sample_integer_element_type)
      << getCurrentResourceMethodName(S, getSampleMethodName(Kind))
      << ContainedType << SMVersion.getAsString();
  return true;
}
 
static bool CheckTextureSamplerAndLocation(Sema &S, CallExpr *TheCall,
                                           bool IncludeArraySlice = true) {
  // Check the texture handle.
  if (CheckResourceHandle(&S, TheCall, 0,
                          [](const HLSLAttributedResourceType *ResType) {
                            return ResType->getAttrs().ResourceDimension ==
                                   llvm::dxil::ResourceDimension::Unknown;
                          }))
    return true;
 
  // Check the sampler handle.
  if (CheckResourceHandle(&S, TheCall, 1,
                          [](const HLSLAttributedResourceType *ResType) {
                            return ResType->getAttrs().ResourceClass !=
                                   llvm::hlsl::ResourceClass::Sampler;
                          }))
    return true;
 
  auto *ResourceTy =
      TheCall->getArg(0)->getType()->castAs<HLSLAttributedResourceType>();
 
  // Check the location.
  unsigned ExpectedDim =
      getResourceDimensions(ResourceTy->getAttrs().ResourceDimension) +
      (IncludeArraySlice && ResourceTy->getAttrs().IsArray ? 1 : 0);
  if (CheckArgTypeMatches(
          &S, TheCall->getArg(2),
          getVectorOrScalarType(S, S.Context.FloatTy, ExpectedDim)))
    return true;
 
  return false;
}
 
static bool CheckCalculateLodBuiltin(Sema &S, CallExpr *TheCall) {
  if (S.checkArgCount(TheCall, 3))
    return true;
 
  // CalculateLevelOfDetail location uses resource dimension only (e.g. float2
  // for 2D), not an extra array slice component like Sample/Gather.
  if (CheckTextureSamplerAndLocation(S, TheCall, /*IncludeArraySlice=*/false))
    return true;
 
  TheCall->setType(S.Context.FloatTy);
  return false;
}
 
static bool CheckGatherBuiltin(Sema &S, CallExpr *TheCall, bool IsCmp) {
  if (S.checkArgCountRange(TheCall, IsCmp ? 5 : 4, IsCmp ? 6 : 5))
    return true;
 
  if (CheckTextureSamplerAndLocation(S, TheCall))
    return true;
 
  unsigned NextIdx = 3;
  if (IsCmp) {
    // Check the compare value.
    if (CheckScalarFloatOperand(S, TheCall, NextIdx))
      return true;
    NextIdx++;
  }
 
  // Check the component operand.
  if (CheckArgTypeMatches(&S, TheCall->getArg(NextIdx),
                          S.Context.UnsignedIntTy))
    return true;
  Expr *ComponentArg = TheCall->getArg(NextIdx);
 
  // GatherCmp operations on Vulkan target must use component 0 (Red).
  if (IsCmp && S.getASTContext().getTargetInfo().getTriple().isSPIRV()) {
    std::optional<llvm::APSInt> ComponentOpt =
        ComponentArg->getIntegerConstantExpr(S.getASTContext());
    if (ComponentOpt) {
      int64_t ComponentVal = ComponentOpt->getSExtValue();
      if (ComponentVal != 0) {
        // Issue an error if the component is not 0 (Red).
        // 0 -> Red, 1 -> Green, 2 -> Blue, 3 -> Alpha
        assert(ComponentVal >= 0 && ComponentVal <= 3 &&
               "The component is not in the expected range.");
        S.Diag(ComponentArg->getBeginLoc(),
               diag::err_hlsl_gathercmp_invalid_component)
            << ComponentVal;
        return true;
      }
    }
  }
 
  NextIdx++;
 
  // Check the offset operand.
  const HLSLAttributedResourceType *ResourceTy =
      TheCall->getArg(0)->getType()->castAs<HLSLAttributedResourceType>();
  if (TheCall->getNumArgs() > NextIdx) {
    unsigned ExpectedDim =
        getResourceDimensions(ResourceTy->getAttrs().ResourceDimension);
    if (CheckArgTypeMatches(
            &S, TheCall->getArg(NextIdx),
            getVectorOrScalarType(S, S.Context.IntTy, ExpectedDim)))
      return true;
    NextIdx++;
  }
 
  assert(ResourceTy->hasContainedType() &&
         "Expecting a contained type for resource with a dimension "
         "attribute.");
  QualType ReturnType = ResourceTy->getContainedType();
 
  if (CheckNoDoubleElementType(S, TheCall, ReturnType,
                               IsCmp ? "GatherCmp" : "Gather"))
    return true;
 
  if (IsCmp) {
    if (!ReturnType->hasFloatingRepresentation()) {
      S.Diag(TheCall->getBeginLoc(), diag::err_hlsl_samplecmp_requires_float);
      return true;
    }
  }
 
  if (const auto *VecTy = ReturnType->getAs<VectorType>())
    ReturnType = VecTy->getElementType();
  ReturnType = S.Context.getExtVectorType(ReturnType, 4);
 
  TheCall->setType(ReturnType);
 
  return false;
}
static bool CheckLoadLevelBuiltin(Sema &S, CallExpr *TheCall) {
  if (S.checkArgCountRange(TheCall, 2, 3))
    return true;
 
  // Check the texture handle.
  if (CheckResourceHandle(&S, TheCall, 0,
                          [](const HLSLAttributedResourceType *ResType) {
                            return ResType->getAttrs().ResourceDimension ==
                                   llvm::dxil::ResourceDimension::Unknown;
                          }))
    return true;
 
  auto *ResourceTy =
      TheCall->getArg(0)->getType()->castAs<HLSLAttributedResourceType>();
 
  // A UAV descriptor binds a single mip slice, so a RWTexture location has no
  // mip component to select, and TextureLoad on a UAV takes no offset.
  bool IsUAV =
      ResourceTy->getAttrs().ResourceClass == llvm::dxil::ResourceClass::UAV;
  if (IsUAV && S.checkArgCount(TheCall, 2))
    return true;
 
  // Check the location: int3 for Texture2D and int4 for Texture2DArray, which
  // both carry a trailing mip level; int2 and int3 for the RWTexture forms,
  // which do not.
  unsigned ResourceDim =
      getResourceDimensions(ResourceTy->getAttrs().ResourceDimension);
  unsigned LocationDim = ResourceDim + (ResourceTy->getAttrs().IsArray ? 1 : 0);
  if (!IsUAV)
    ++LocationDim;
  if (CheckArgTypeMatches(
          &S, TheCall->getArg(1),
          getVectorOrScalarType(S, S.Context.IntTy, LocationDim)))
    return true;
 
  // Check the offset operand (int2 for 2D textures; no array slice).
  if (TheCall->getNumArgs() > 2) {
    if (CheckArgTypeMatches(
            &S, TheCall->getArg(2),
            getVectorOrScalarType(S, S.Context.IntTy, ResourceDim)))
      return true;
  }
 
  TheCall->setType(ResourceTy->getContainedType());
  return false;
}
 
static bool CheckLoadMSBuiltin(Sema &S, CallExpr *TheCall) {
  if (S.checkArgCountRange(TheCall, 3, 4))
    return true;
 
  // Check the multisampled texture handle.
  if (CheckResourceHandle(&S, TheCall, 0,
                          [](const HLSLAttributedResourceType *ResType) {
                            return !ResType->isMultiSampled();
                          }))
    return true;
 
  auto *ResourceTy =
      TheCall->getArg(0)->getType()->castAs<HLSLAttributedResourceType>();
 
  // Check the location (int2 for Texture2DMS, int3 for Texture2DMSArray).
  // Unlike Load on regular textures, there is no mip/LOD component.
  unsigned ResourceDim =
      getResourceDimensions(ResourceTy->getAttrs().ResourceDimension);
  unsigned LocationDim = ResourceDim + (ResourceTy->getAttrs().IsArray ? 1 : 0);
  if (CheckArgTypeMatches(
          &S, TheCall->getArg(1),
          getVectorOrScalarType(S, S.Context.IntTy, LocationDim)))
    return true;
 
  // Check the sample index operand (scalar int).
  if (CheckArgTypeMatches(&S, TheCall->getArg(2), S.Context.IntTy))
    return true;
 
  // Check the offset operand (int2 for 2D textures; no array slice).
  if (TheCall->getNumArgs() > 3) {
    if (CheckArgTypeMatches(
            &S, TheCall->getArg(3),
            getVectorOrScalarType(S, S.Context.IntTy, ResourceDim)))
      return true;
  }
 
  TheCall->setType(ResourceTy->getContainedType());
  return false;
}
 
static bool CheckSamplingBuiltin(Sema &S, CallExpr *TheCall, SampleKind Kind) {
  unsigned MinArgs, MaxArgs;
  if (Kind == SampleKind::Sample) {
    MinArgs = 3;
    MaxArgs = 5;
  } else if (Kind == SampleKind::Bias) {
    MinArgs = 4;
    MaxArgs = 6;
  } else if (Kind == SampleKind::Grad) {
    MinArgs = 5;
    MaxArgs = 7;
  } else if (Kind == SampleKind::Level) {
    MinArgs = 4;
    MaxArgs = 5;
  } else if (Kind == SampleKind::Cmp) {
    MinArgs = 4;
    MaxArgs = 6;
  } else {
    assert(Kind == SampleKind::CmpLevelZero);
    MinArgs = 4;
    MaxArgs = 5;
  }
 
  if (S.checkArgCountRange(TheCall, MinArgs, MaxArgs))
    return true;
 
  if (CheckTextureSamplerAndLocation(S, TheCall))
    return true;
 
  const HLSLAttributedResourceType *ResourceTy =
      TheCall->getArg(0)->getType()->castAs<HLSLAttributedResourceType>();
  unsigned ExpectedDim =
      getResourceDimensions(ResourceTy->getAttrs().ResourceDimension);
 
  unsigned NextIdx = 3;
  if (Kind == SampleKind::Bias || Kind == SampleKind::Level ||
      Kind == SampleKind::Cmp || Kind == SampleKind::CmpLevelZero) {
    // Check the bias, lod level, or compare value, depending on the kind.
    // All of them must be a scalar float value.
    if (CheckScalarFloatOperand(S, TheCall, NextIdx))
      return true;
    NextIdx++;
  } else if (Kind == SampleKind::Grad) {
    QualType GradTy = getVectorOrScalarType(S, S.Context.FloatTy, ExpectedDim);
 
    // Check the DDX operand.
    if (CheckArgTypeMatches(&S, TheCall->getArg(NextIdx), GradTy))
      return true;
 
    // Check the DDY operand.
    if (CheckArgTypeMatches(&S, TheCall->getArg(NextIdx + 1), GradTy))
      return true;
    NextIdx += 2;
  }
 
  // Check the offset operand (if applicable).
  if (hasResourceOffset(ResourceTy->getAttrs().ResourceDimension) &&
      TheCall->getNumArgs() > NextIdx) {
    if (CheckArgTypeMatches(
            &S, TheCall->getArg(NextIdx),
            getVectorOrScalarType(S, S.Context.IntTy, ExpectedDim)))
      return true;
    NextIdx++;
  }
 
  // Check the clamp operand.
  if (Kind != SampleKind::Level && Kind != SampleKind::CmpLevelZero &&
      TheCall->getNumArgs() > NextIdx) {
    if (CheckScalarFloatOperand(S, TheCall, NextIdx))
      return true;
  }
 
  assert(ResourceTy->hasContainedType() &&
         "Expecting a contained type for resource with a dimension "
         "attribute.");
  QualType ReturnType = ResourceTy->getContainedType();
 
  if (CheckNoDoubleElementType(S, TheCall, ReturnType,
                               getSampleMethodName(Kind)))
    return true;
 
  if (CheckIntegerElementTypeShaderModel(S, TheCall, ReturnType, Kind))
    return true;
 
  if (Kind == SampleKind::Cmp || Kind == SampleKind::CmpLevelZero) {
    if (!ReturnType->hasFloatingRepresentation()) {
      S.Diag(TheCall->getBeginLoc(), diag::err_hlsl_samplecmp_requires_float);
      return true;
    }
    ReturnType = S.Context.FloatTy;
  }
  TheCall->setType(ReturnType);
 
  return false;
}
 
/// The `dest` types an interlocked operation accepts. Float is 32-bit only.
enum class InterlockedDest { Int, IntOrFloat, Float };
 
/// Check a call to an HLSL interlocked builtin. The builtins are variadic, so
/// this is the only check a direct call gets. Overload resolution checks the
/// calls that come through the `InterlockedOp` overload sets.
static bool CheckInterlockedBuiltin(Sema &S, CallExpr *TheCall,
                                    unsigned MinArgs, unsigned MaxArgs,
                                    InterlockedDest Dest,
                                    bool ReportsOriginalValue) {
  if (MinArgs == MaxArgs) {
    if (S.checkArgCount(TheCall, MinArgs))
      return true;
  } else if (TheCall->getNumArgs() < MinArgs) {
    S.Diag(TheCall->getEndLoc(), diag::err_typecheck_call_too_few_args_at_least)
        << /*callee_type=*/0 << /*min_arg_count=*/MinArgs
        << TheCall->getNumArgs() << /*is_non_object=*/0
        << TheCall->getSourceRange();
    return true;
  } else if (S.checkArgCountAtMost(TheCall, MaxArgs)) {
    return true;
  }
 
  QualType DestTy = TheCall->getArg(0)->getType().getUnqualifiedType();
  const bool DestIsOK =
      DestTy->isSpecificBuiltinType(BuiltinType::Float)
          ? Dest != InterlockedDest::Int
          : Dest != InterlockedDest::Float && DestTy->isIntegerType();
  if (!DestIsOK) {
    S.Diag(TheCall->getArg(0)->getBeginLoc(),
           diag::err_builtin_invalid_arg_type)
        << /*ordinal=*/1 << /*scalar*/ 1
        << /*integer*/ (Dest == InterlockedDest::Float ? 0 : 1)
        << /*32 bit floating-point*/ (Dest == InterlockedDest::Int ? 0 : 3)
        << DestTy;
    return true;
  }
 
  // 64-bit interlocked ops require SM 6.6 on DXIL. The synthesized wrapper
  // methods (e.g. RWByteAddressBuffer::InterlockedAdd64) are only declared on
  // SM 6.6+, so this defensive check only fires for direct builtin calls; skip
  // synthetic invocations (invalid source location).
  const TargetInfo &TI = S.Context.getTargetInfo();
  if (TheCall->getBeginLoc().isValid() &&
      TI.getTriple().getArch() == llvm::Triple::dxil &&
      S.Context.getTypeSize(DestTy) == 64 &&
      TI.getPlatformMinVersion() < VersionTuple(6, 6)) {
    S.Diag(TheCall->getBeginLoc(), diag::err_hlsl_builtin_requires_sm)
        << TheCall->getDirectCallee() << VersionTuple(6, 6).getAsString();
    return true;
  }
 
  if (CheckModifiableLValue(&S, TheCall, 0))
    return true;
 
  if (CheckArgAddrSpaceOneOf(&S, TheCall, 0,
                             {LangAS::hlsl_groupshared, LangAS::hlsl_device}))
    return true;
 
  // Every argument after `dest` has the destination's type.
  for (unsigned I = 1, E = TheCall->getNumArgs(); I != E; ++I)
    if (CheckArgTypeMatches(&S, TheCall->getArg(I), DestTy))
      return true;
 
  // Operations that report the previous value write it back through their last
  // argument.
  const unsigned NumArgs = TheCall->getNumArgs();
  if (ReportsOriginalValue && NumArgs == MaxArgs &&
      CheckModifiableLValue(&S, TheCall, NumArgs - 1))
    return true;
 
  TheCall->setType(S.Context.VoidTy);
  return false;
}
 
// Note: returning true in this case results in CheckBuiltinFunctionCall
// returning an ExprError
bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
  switch (BuiltinID) {
  case Builtin::BI__builtin_hlsl_barrier: {
    if (SemaRef.checkArgCount(TheCall, 2))
      return true;
 
    if (SemaRef.Context.getTargetInfo().getTriple().getArch() !=
        llvm::Triple::dxil) {
      SemaRef.Diag(TheCall->getExprLoc(), diag::err_hlsl_dxil_only)
          << "Barrier";
      return true;
    }
 
    Expr *MemoryArg = TheCall->getArg(0);
    if (MemoryArg->getType()->isUnsignedIntegerType()) {
      std::optional<llvm::APSInt> MemoryFlags =
          MemoryArg->getIntegerConstantExpr(SemaRef.Context);
      if (!MemoryFlags) {
        SemaRef.Diag(MemoryArg->getExprLoc(),
                     diag::err_constant_integer_arg_type)
            << "Barrier";
        return true;
      }
      if ((MemoryFlags->getZExtValue() &
           ~barrierFlagValue(BarrierMemoryTypeFlag::ValidMask)) != 0) {
        SemaRef.Diag(MemoryArg->getExprLoc(),
                     diag::err_hlsl_invalid_barrier_memory_flags);
        return true;
      }
    } else {
      const HLSLAttributedResourceType *ResTy =
          HLSLAttributedResourceType::findHandleTypeOnResource(
              MemoryArg->getType().getTypePtr());
      if (!ResTy) {
        SemaRef.Diag(MemoryArg->getExprLoc(),
                     diag::err_typecheck_expect_hlsl_resource)
            << MemoryArg->getType();
        return true;
      }
      if (ResTy->getAttrs().ResourceClass != ResourceClass::UAV) {
        SemaRef.Diag(MemoryArg->getExprLoc(),
                     diag::err_invalid_hlsl_resource_type)
            << MemoryArg->getType();
        return true;
      }
    }
 
    Expr *SemanticArg = TheCall->getArg(1);
    std::optional<llvm::APSInt> SemanticFlags =
        SemanticArg->getIntegerConstantExpr(SemaRef.Context);
    if (!SemanticFlags) {
      SemaRef.Diag(SemanticArg->getExprLoc(),
                   diag::err_constant_integer_arg_type)
          << "Barrier";
      return true;
    }
    if ((SemanticFlags->getZExtValue() &
         ~barrierFlagValue(BarrierSemanticFlag::ValidMask)) != 0) {
      SemaRef.Diag(SemanticArg->getExprLoc(),
                   diag::err_hlsl_invalid_barrier_semantic_flags);
      return true;
    }
 
    TheCall->setType(SemaRef.Context.VoidTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_adduint64: {
    if (SemaRef.checkArgCount(TheCall, 2))
      return true;
 
    if (CheckAllArgTypesAreCorrect(&SemaRef, TheCall,
                                   CheckUnsignedIntVecRepresentation))
      return true;
 
    // ensure arg integers are 32-bits
    if (CheckExpectedBitWidth(&SemaRef, TheCall, 0, 32))
      return true;
 
    // ensure both args are vectors of total bit size of a multiple of 64
    auto *VTy = TheCall->getArg(0)->getType()->getAs<VectorType>();
    int NumElementsArg = VTy->getNumElements();
    if (NumElementsArg != 2 && NumElementsArg != 4) {
      SemaRef.Diag(TheCall->getBeginLoc(), diag::err_vector_incorrect_bit_count)
          << 1 /*a multiple of*/ << 64 << NumElementsArg * 32;
      return true;
    }
 
    // ensure first arg and second arg have the same type
    if (CheckAllArgsHaveSameType(&SemaRef, TheCall))
      return true;
 
    ExprResult A = TheCall->getArg(0);
    QualType ArgTyA = A.get()->getType();
    // return type is the same as the input type
    TheCall->setType(ArgTyA);
    break;
  }
  case Builtin::BI__builtin_hlsl_resource_getpointer: {
    if (SemaRef.checkArgCountRange(TheCall, 1, 2) ||
        CheckResourceHandle(&SemaRef, TheCall, 0) ||
        (TheCall->getNumArgs() == 2 && CheckIndexType(&SemaRef, TheCall, 1)))
      return true;
 
    auto *ResourceTy =
        TheCall->getArg(0)->getType()->castAs<HLSLAttributedResourceType>();
    QualType ContainedTy = ResourceTy->getContainedType();
    auto ReturnType = SemaRef.Context.getAddrSpaceQualType(
        ContainedTy,
        getLangASFromResourceClass(ResourceTy->getAttrs().ResourceClass));
    ReturnType = SemaRef.Context.getPointerType(ReturnType);
    TheCall->setType(ReturnType);
 
    break;
  }
  case Builtin::BI__builtin_hlsl_resource_getpointer_typed: {
    if (SemaRef.checkArgCount(TheCall, 3) ||
        CheckResourceHandle(&SemaRef, TheCall, 0) ||
        CheckIndexType(&SemaRef, TheCall, 1))
      return true;
 
    QualType ElementTy = TheCall->getArg(2)->getType();
    assert(ElementTy->isPointerType() &&
           "expected pointer type for second argument");
    ElementTy = ElementTy->getPointeeType();
 
    // Reject array types
    if (ElementTy->isArrayType())
      return SemaRef.Diag(
          cast<FunctionDecl>(SemaRef.CurContext)->getPointOfInstantiation(),
          diag::err_invalid_use_of_array_type);
 
    auto *ResourceTy =
        TheCall->getArg(0)->getType()->castAs<HLSLAttributedResourceType>();
    auto ReturnType = SemaRef.Context.getAddrSpaceQualType(
        ElementTy,
        getLangASFromResourceClass(ResourceTy->getAttrs().ResourceClass));
    ReturnType = SemaRef.Context.getPointerType(ReturnType);
    TheCall->setType(ReturnType);
 
    break;
  }
  case Builtin::BI__builtin_hlsl_transpose_if_memory_is_row_major: {
    if (SemaRef.checkArgCount(TheCall, 2) ||
        CheckArgTypeMatches(&SemaRef, TheCall->getArg(1),
                            SemaRef.getASTContext().IntTy))
      return true;
 
    TheCall->setType(TheCall->getArg(0)->getType());
 
    break;
  }
  case Builtin::BI__builtin_hlsl_resource_load_with_status: {
    if (SemaRef.checkArgCount(TheCall, 3) ||
        CheckResourceHandle(&SemaRef, TheCall, 0) ||
        CheckArgTypeMatches(&SemaRef, TheCall->getArg(1),
                            SemaRef.getASTContext().UnsignedIntTy) ||
        CheckArgTypeMatches(&SemaRef, TheCall->getArg(2),
                            SemaRef.getASTContext().UnsignedIntTy) ||
        CheckModifiableLValue(&SemaRef, TheCall, 2))
      return true;
 
    auto *ResourceTy =
        TheCall->getArg(0)->getType()->castAs<HLSLAttributedResourceType>();
    QualType ReturnType = ResourceTy->getContainedType();
    TheCall->setType(ReturnType);
 
    break;
  }
  case Builtin::BI__builtin_hlsl_resource_load_with_status_typed: {
    if (SemaRef.checkArgCount(TheCall, 4) ||
        CheckResourceHandle(&SemaRef, TheCall, 0) ||
        CheckArgTypeMatches(&SemaRef, TheCall->getArg(1),
                            SemaRef.getASTContext().UnsignedIntTy) ||
        CheckArgTypeMatches(&SemaRef, TheCall->getArg(2),
                            SemaRef.getASTContext().UnsignedIntTy) ||
        CheckModifiableLValue(&SemaRef, TheCall, 2))
      return true;
 
    QualType ReturnType = TheCall->getArg(3)->getType();
    assert(ReturnType->isPointerType() &&
           "expected pointer type for second argument");
    ReturnType = ReturnType->getPointeeType();
 
    // Reject array types
    if (ReturnType->isArrayType())
      return SemaRef.Diag(
          cast<FunctionDecl>(SemaRef.CurContext)->getPointOfInstantiation(),
          diag::err_invalid_use_of_array_type);
 
    TheCall->setType(ReturnType);
 
    break;
  }
  case Builtin::BI__builtin_hlsl_resource_load_level:
    return CheckLoadLevelBuiltin(SemaRef, TheCall);
  case Builtin::BI__builtin_hlsl_resource_load_ms:
    return CheckLoadMSBuiltin(SemaRef, TheCall);
  case Builtin::BI__builtin_hlsl_resource_sample:
    return CheckSamplingBuiltin(SemaRef, TheCall, SampleKind::Sample);
  case Builtin::BI__builtin_hlsl_resource_sample_bias:
    return CheckSamplingBuiltin(SemaRef, TheCall, SampleKind::Bias);
  case Builtin::BI__builtin_hlsl_resource_sample_grad:
    return CheckSamplingBuiltin(SemaRef, TheCall, SampleKind::Grad);
  case Builtin::BI__builtin_hlsl_resource_sample_level:
    return CheckSamplingBuiltin(SemaRef, TheCall, SampleKind::Level);
  case Builtin::BI__builtin_hlsl_resource_sample_cmp:
    return CheckSamplingBuiltin(SemaRef, TheCall, SampleKind::Cmp);
  case Builtin::BI__builtin_hlsl_resource_sample_cmp_level_zero:
    return CheckSamplingBuiltin(SemaRef, TheCall, SampleKind::CmpLevelZero);
  case Builtin::BI__builtin_hlsl_resource_calculate_lod:
  case Builtin::BI__builtin_hlsl_resource_calculate_lod_unclamped:
    return CheckCalculateLodBuiltin(SemaRef, TheCall);
  case Builtin::BI__builtin_hlsl_resource_gather:
    return CheckGatherBuiltin(SemaRef, TheCall, /*IsCmp=*/false);
  case Builtin::BI__builtin_hlsl_resource_gather_cmp:
    return CheckGatherBuiltin(SemaRef, TheCall, /*IsCmp=*/true);
  case Builtin::BI__builtin_hlsl_resource_uninitializedhandle: {
    assert(TheCall->getNumArgs() == 1 && "expected 1 arg");
    // Update return type to be the attributed resource type from arg0.
    QualType ResourceTy = TheCall->getArg(0)->getType();
    TheCall->setType(ResourceTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_resource_handlefrombinding: {
    assert(TheCall->getNumArgs() == 6 && "expected 6 args");
    // Update return type to be the attributed resource type from arg0.
    QualType ResourceTy = TheCall->getArg(0)->getType();
    TheCall->setType(ResourceTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_resource_handlefromimplicitbinding: {
    assert(TheCall->getNumArgs() == 6 && "expected 6 args");
    // Update return type to be the attributed resource type from arg0.
    QualType ResourceTy = TheCall->getArg(0)->getType();
    TheCall->setType(ResourceTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_resource_counterhandlefromimplicitbinding: {
    assert(TheCall->getNumArgs() == 3 && "expected 3 args");
    // Update return type to be the attributed resource type from arg0
    // with added IsCounter flag.
    QualType MainHandleTy = TheCall->getArg(0)->getType();
    QualType CounterHandleTy =
        createCounterHandleType(SemaRef.getASTContext(), MainHandleTy);
    TheCall->setType(CounterHandleTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_resource_handlefromheap: {
    if (SemaRef.checkArgCount(TheCall, 2) ||
        CheckResourceHandle(&SemaRef, TheCall, 0) ||
        CheckArgTypeMatches(&SemaRef, TheCall->getArg(1),
                            SemaRef.getASTContext().UnsignedIntTy))
      return true;
 
    // Update return type to be the attributed resource type from arg0.
    QualType ResourceTy = TheCall->getArg(0)->getType();
    TheCall->setType(ResourceTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_resource_counterhandlefromheap: {
    if (SemaRef.checkArgCount(TheCall, 1) ||
        CheckResourceHandle(&SemaRef, TheCall, 0))
      return true;
    // Update return type to be the attributed resource type from arg0
    // with added IsCounter flag.
    QualType MainHandleTy = TheCall->getArg(0)->getType();
    QualType CounterHandleTy =
        createCounterHandleType(SemaRef.getASTContext(), MainHandleTy);
    TheCall->setType(CounterHandleTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_and:
  case Builtin::BI__builtin_hlsl_or: {
    if (SemaRef.checkArgCount(TheCall, 2))
      return true;
    if (CheckScalarOrVectorOrMatrix(&SemaRef, TheCall, getASTContext().BoolTy,
                                    0))
      return true;
    if (CheckAllArgsHaveSameType(&SemaRef, TheCall))
      return true;
 
    ExprResult A = TheCall->getArg(0);
    QualType ArgTyA = A.get()->getType();
    // return type is the same as the input type
    TheCall->setType(ArgTyA);
    break;
  }
  case Builtin::BI__builtin_hlsl_asdouble: {
    if (SemaRef.checkArgCount(TheCall, 2))
      return true;
    if (CheckScalarOrVector(
            &SemaRef, TheCall,
            /*only check for uint*/ SemaRef.Context.UnsignedIntTy,
            /* arg index */ 0))
      return true;
    if (CheckScalarOrVector(
            &SemaRef, TheCall,
            /*only check for uint*/ SemaRef.Context.UnsignedIntTy,
            /* arg index */ 1))
      return true;
    if (CheckAllArgsHaveSameType(&SemaRef, TheCall))
      return true;
 
    SetElementTypeAsReturnType(&SemaRef, TheCall, getASTContext().DoubleTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_elementwise_clamp: {
    if (SemaRef.BuiltinElementwiseTernaryMath(
            TheCall, /*ArgTyRestr=*/
            Sema::EltwiseBuiltinArgTyRestriction::None))
      return true;
    break;
  }
  case Builtin::BI__builtin_hlsl_dot: {
    // arg count is checked by BuiltinVectorToScalarMath
    if (SemaRef.BuiltinVectorToScalarMath(TheCall))
      return true;
    if (CheckAllArgTypesAreCorrect(&SemaRef, TheCall, CheckNoDoubleVectors))
      return true;
    break;
  }
  case Builtin::BI__builtin_hlsl_elementwise_firstbithigh:
  case Builtin::BI__builtin_hlsl_elementwise_firstbitlow: {
    if (SemaRef.PrepareBuiltinElementwiseMathOneArgCall(TheCall))
      return true;
 
    const Expr *Arg = TheCall->getArg(0);
    QualType ArgTy = Arg->getType();
    QualType EltTy = ArgTy;
 
    QualType ResTy = SemaRef.Context.UnsignedIntTy;
 
    if (auto *VecTy = EltTy->getAs<VectorType>()) {
      EltTy = VecTy->getElementType();
      ResTy = SemaRef.Context.getExtVectorType(ResTy, VecTy->getNumElements());
    }
 
    if (!EltTy->isIntegerType()) {
      Diag(Arg->getBeginLoc(), diag::err_builtin_invalid_arg_type)
          << 1 << /* scalar or vector of */ 5 << /* integer ty */ 1
          << /* no fp */ 0 << ArgTy;
      return true;
    }
 
    TheCall->setType(ResTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_select: {
    if (SemaRef.checkArgCount(TheCall, 3))
      return true;
    if (CheckScalarOrVectorOrMatrix(&SemaRef, TheCall, getASTContext().BoolTy,
                                    0))
      return true;
    QualType ArgTy = TheCall->getArg(0)->getType();
    if (ArgTy->isBooleanType() && CheckBoolSelect(&SemaRef, TheCall))
      return true;
    auto *VTy = ArgTy->getAs<VectorType>();
    if (VTy && VTy->getElementType()->isBooleanType() &&
        CheckVectorSelect(&SemaRef, TheCall))
      return true;
    auto *MTy = ArgTy->getAs<ConstantMatrixType>();
    if (MTy && MTy->getElementType()->isBooleanType() &&
        CheckMatrixSelect(&SemaRef, TheCall))
      return true;
    break;
  }
  case Builtin::BI__builtin_hlsl_elementwise_saturate:
  case Builtin::BI__builtin_hlsl_elementwise_rcp: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
    if (!TheCall->getArg(0)
             ->getType()
             ->hasFloatingRepresentation()) // half or float or double
      return SemaRef.Diag(TheCall->getArg(0)->getBeginLoc(),
                          diag::err_builtin_invalid_arg_type)
             << /* ordinal */ 1 << /* scalar or vector */ 5 << /* no int */ 0
             << /* fp */ 1 << TheCall->getArg(0)->getType();
    if (SemaRef.PrepareBuiltinElementwiseMathOneArgCall(TheCall))
      return true;
    break;
  }
  case Builtin::BI__builtin_hlsl_elementwise_rsqrt:
  case Builtin::BI__builtin_hlsl_elementwise_frac:
  case Builtin::BI__builtin_hlsl_elementwise_ddx_coarse:
  case Builtin::BI__builtin_hlsl_elementwise_ddy_coarse:
  case Builtin::BI__builtin_hlsl_elementwise_ddx_fine:
  case Builtin::BI__builtin_hlsl_elementwise_ddy_fine: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
    if (CheckAllArgTypesAreCorrect(&SemaRef, TheCall,
                                   CheckFloatOrHalfRepresentation))
      return true;
    if (SemaRef.PrepareBuiltinElementwiseMathOneArgCall(TheCall))
      return true;
    break;
  }
  case Builtin::BI__builtin_hlsl_elementwise_isfinite:
  case Builtin::BI__builtin_hlsl_elementwise_isinf:
  case Builtin::BI__builtin_hlsl_elementwise_isnan: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
    if (CheckAllArgTypesAreCorrect(&SemaRef, TheCall,
                                   CheckFloatOrHalfRepresentation))
      return true;
    if (SemaRef.PrepareBuiltinElementwiseMathOneArgCall(TheCall))
      return true;
    SetElementTypeAsReturnType(&SemaRef, TheCall, getASTContext().BoolTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_mad: {
    if (SemaRef.BuiltinElementwiseTernaryMath(
            TheCall, /*ArgTyRestr=*/
            Sema::EltwiseBuiltinArgTyRestriction::None))
      return true;
    break;
  }
  case Builtin::BI__builtin_hlsl_mul: {
    if (SemaRef.checkArgCount(TheCall, 2))
      return true;
 
    Expr *Arg0 = TheCall->getArg(0);
    Expr *Arg1 = TheCall->getArg(1);
    QualType Ty0 = Arg0->getType();
    QualType Ty1 = Arg1->getType();
 
    auto getElemType = [](QualType T) -> QualType {
      if (const auto *VTy = T->getAs<VectorType>())
        return VTy->getElementType();
      if (const auto *MTy = T->getAs<ConstantMatrixType>())
        return MTy->getElementType();
      return T;
    };
 
    QualType EltTy0 = getElemType(Ty0);
 
    bool IsVec0 = Ty0->isVectorType();
    bool IsMat0 = Ty0->isConstantMatrixType();
    bool IsVec1 = Ty1->isVectorType();
    bool IsMat1 = Ty1->isConstantMatrixType();
 
    QualType RetTy;
 
    if (IsVec0 && IsMat1) {
      auto *MatTy = Ty1->castAs<ConstantMatrixType>();
      RetTy = getASTContext().getExtVectorType(EltTy0, MatTy->getNumColumns());
    } else if (IsMat0 && IsVec1) {
      auto *MatTy = Ty0->castAs<ConstantMatrixType>();
      RetTy = getASTContext().getExtVectorType(EltTy0, MatTy->getNumRows());
    } else {
      assert(IsMat0 && IsMat1);
      auto *MatTy0 = Ty0->castAs<ConstantMatrixType>();
      auto *MatTy1 = Ty1->castAs<ConstantMatrixType>();
      RetTy = getASTContext().getConstantMatrixType(
          EltTy0, MatTy0->getNumRows(), MatTy1->getNumColumns());
    }
 
    TheCall->setType(RetTy);
    break;
  }
  case Builtin::BI__builtin_elementwise_fma: {
    if (SemaRef.checkArgCount(TheCall, 3) ||
        CheckAllArgsHaveSameType(&SemaRef, TheCall)) {
      return true;
    }
 
    if (CheckAllArgTypesAreCorrect(&SemaRef, TheCall,
                                   CheckAnyDoubleRepresentation))
      return true;
 
    ExprResult A = TheCall->getArg(0);
    QualType ArgTyA = A.get()->getType();
    // return type is the same as input type
    TheCall->setType(ArgTyA);
    break;
  }
  case Builtin::BI__builtin_hlsl_transpose: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
 
    Expr *Arg = TheCall->getArg(0);
    QualType ArgTy = Arg->getType();
 
    const auto *MatTy = ArgTy->getAs<ConstantMatrixType>();
    if (!MatTy) {
      SemaRef.Diag(Arg->getBeginLoc(), diag::err_builtin_invalid_arg_type)
          << 1 << /* matrix */ 3 << /* no int */ 0 << /* no fp */ 0 << ArgTy;
      return true;
    }
 
    QualType RetTy = getASTContext().getConstantMatrixType(
        MatTy->getElementType(), MatTy->getNumColumns(), MatTy->getNumRows());
    TheCall->setType(RetTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_elementwise_sign: {
    if (SemaRef.PrepareBuiltinElementwiseMathOneArgCall(TheCall))
      return true;
    if (CheckAllArgTypesAreCorrect(&SemaRef, TheCall,
                                   CheckFloatingOrIntRepresentation))
      return true;
    SetElementTypeAsReturnType(&SemaRef, TheCall, getASTContext().IntTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_wave_active_all_equal: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
 
    // Ensure input expr type is a scalar/vector
    if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0))
      return true;
 
    QualType InputTy = TheCall->getArg(0)->getType();
    ASTContext &Ctx = getASTContext();
 
    QualType RetTy;
 
    // If vector, construct bool vector of same size
    if (const auto *VecTy = InputTy->getAs<ExtVectorType>()) {
      unsigned NumElts = VecTy->getNumElements();
      RetTy = Ctx.getExtVectorType(Ctx.BoolTy, NumElts);
    } else {
      // Scalar case
      RetTy = Ctx.BoolTy;
    }
 
    TheCall->setType(RetTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_wave_active_max:
  case Builtin::BI__builtin_hlsl_wave_active_min:
  case Builtin::BI__builtin_hlsl_wave_active_sum:
  case Builtin::BI__builtin_hlsl_wave_active_product: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
 
    // Ensure input expr type is a scalar/vector and the same as the return type
    if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0))
      return true;
    if (CheckWaveActive(&SemaRef, TheCall))
      return true;
    ExprResult Expr = TheCall->getArg(0);
    QualType ArgTyExpr = Expr.get()->getType();
    TheCall->setType(ArgTyExpr);
    break;
  }
  case Builtin::BI__builtin_hlsl_wave_active_bit_or:
  case Builtin::BI__builtin_hlsl_wave_active_bit_xor:
  case Builtin::BI__builtin_hlsl_wave_active_bit_and: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
 
    // Ensure input expr type is a scalar/vector
    if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0))
      return true;
 
    if (CheckWaveActive(&SemaRef, TheCall))
      return true;
 
    // Ensure the expr type is interpretable as a uint or vector<uint>
    ExprResult Expr = TheCall->getArg(0);
    QualType ArgTyExpr = Expr.get()->getType();
    auto *VTy = ArgTyExpr->getAs<VectorType>();
    if (!(ArgTyExpr->isIntegerType() ||
          (VTy && VTy->getElementType()->isIntegerType()))) {
      SemaRef.Diag(TheCall->getArg(0)->getBeginLoc(),
                   diag::err_builtin_invalid_arg_type)
          << ArgTyExpr << SemaRef.Context.UnsignedIntTy << 1 << 0 << 0;
      return true;
    }
 
    // Ensure input expr type is the same as the return type
    TheCall->setType(ArgTyExpr);
    break;
  }
  case Builtin::BI__builtin_hlsl_interlocked_add:
  case Builtin::BI__builtin_hlsl_interlocked_and:
  case Builtin::BI__builtin_hlsl_interlocked_max:
  case Builtin::BI__builtin_hlsl_interlocked_min:
  case Builtin::BI__builtin_hlsl_interlocked_or:
  case Builtin::BI__builtin_hlsl_interlocked_xor:
    if (CheckInterlockedBuiltin(SemaRef, TheCall, /*MinArgs=*/2, /*MaxArgs=*/3,
                                InterlockedDest::Int,
                                /*ReportsOriginalValue=*/true))
      return true;
    break;
  case Builtin::BI__builtin_hlsl_interlocked_exchange:
    if (CheckInterlockedBuiltin(SemaRef, TheCall, /*MinArgs=*/3, /*MaxArgs=*/3,
                                InterlockedDest::IntOrFloat,
                                /*ReportsOriginalValue=*/true))
      return true;
    break;
  case Builtin::BI__builtin_hlsl_interlocked_compare_store:
    if (CheckInterlockedBuiltin(SemaRef, TheCall, /*MinArgs=*/3, /*MaxArgs=*/3,
                                InterlockedDest::Int,
                                /*ReportsOriginalValue=*/false))
      return true;
    break;
  case Builtin::BI__builtin_hlsl_interlocked_compare_store_float_bitwise:
    if (CheckInterlockedBuiltin(SemaRef, TheCall, /*MinArgs=*/3, /*MaxArgs=*/3,
                                InterlockedDest::Float,
                                /*ReportsOriginalValue=*/false))
      return true;
    break;
  case Builtin::BI__builtin_hlsl_interlocked_compare_exchange:
    if (CheckInterlockedBuiltin(SemaRef, TheCall, /*MinArgs=*/4, /*MaxArgs=*/4,
                                InterlockedDest::Int,
                                /*ReportsOriginalValue=*/true))
      return true;
    break;
  case Builtin::BI__builtin_hlsl_interlocked_compare_exchange_float_bitwise:
    if (CheckInterlockedBuiltin(SemaRef, TheCall, /*MinArgs=*/4, /*MaxArgs=*/4,
                                InterlockedDest::Float,
                                /*ReportsOriginalValue=*/true))
      return true;
    break;
  // Note these are llvm builtins that we want to catch invalid intrinsic
  // generation. Normal handling of these builtins will occur elsewhere.
  case Builtin::BI__builtin_elementwise_bitreverse: {
    // does not include a check for number of arguments
    // because that is done previously
    if (CheckAllArgTypesAreCorrect(&SemaRef, TheCall,
                                   CheckUnsignedIntRepresentation))
      return true;
    break;
  }
  case Builtin::BI__builtin_hlsl_wave_prefix_count_bits:
  case Builtin::BI__builtin_hlsl_wave_multi_prefix_count_bits: {
    bool IsMultiPrefix =
        BuiltinID == Builtin::BI__builtin_hlsl_wave_multi_prefix_count_bits;
    if (SemaRef.checkArgCount(TheCall, IsMultiPrefix ? 2 : 1))
      return true;
 
    QualType ArgType = TheCall->getArg(0)->getType();
 
    if (!(ArgType->isScalarType())) {
      SemaRef.Diag(TheCall->getArg(0)->getBeginLoc(),
                   diag::err_typecheck_expect_any_scalar_or_vector_or_matrix)
          << ArgType << 0;
      return true;
    }
 
    if (!(ArgType->isBooleanType())) {
      SemaRef.Diag(TheCall->getArg(0)->getBeginLoc(),
                   diag::err_typecheck_expect_any_scalar_or_vector_or_matrix)
          << ArgType << 0;
      return true;
    }
 
    if (IsMultiPrefix) {
      if (CheckWaveMultiPrefixMask(&SemaRef, TheCall))
        return true;
 
      // Alias validation uses calls with an invalid source location.
      if (TheCall->getBeginLoc().isValid() &&
          !getASTContext().getTargetInfo().getTriple().isDXIL()) {
        SemaRef.Diag(TheCall->getBeginLoc(),
                     diag::err_builtin_target_unsupported)
            << TheCall->getSourceRange();
        return true;
      }
    }
 
    break;
  }
  case Builtin::BI__builtin_hlsl_wave_read_lane_at: {
    if (SemaRef.checkArgCount(TheCall, 2))
      return true;
 
    // Ensure index parameter type can be interpreted as a uint
    ExprResult Index = TheCall->getArg(1);
    QualType ArgTyIndex = Index.get()->getType();
    if (!ArgTyIndex->isIntegerType()) {
      SemaRef.Diag(TheCall->getArg(1)->getBeginLoc(),
                   diag::err_typecheck_convert_incompatible)
          << ArgTyIndex << SemaRef.Context.UnsignedIntTy << 1 << 0 << 0;
      return true;
    }
 
    // Ensure input expr type is a scalar/vector and the same as the return type
    if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0))
      return true;
 
    ExprResult Expr = TheCall->getArg(0);
    QualType ArgTyExpr = Expr.get()->getType();
    TheCall->setType(ArgTyExpr);
    break;
  }
  case Builtin::BI__builtin_hlsl_wave_read_lane_first: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
 
    if (CheckAnyScalarOrVectorOrMatrix(&SemaRef, TheCall, 0))
      return true;
 
    TheCall->setType(TheCall->getArg(0)->getType());
    break;
  }
  case Builtin::BI__builtin_hlsl_wave_get_lane_index: {
    if (SemaRef.checkArgCount(TheCall, 0))
      return true;
    break;
  }
  case Builtin::BI__builtin_hlsl_wave_prefix_sum:
  case Builtin::BI__builtin_hlsl_wave_prefix_product: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
 
    // Ensure input expr type is a scalar/vector and the same as the return type
    if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0))
      return true;
    if (CheckWavePrefix(&SemaRef, TheCall))
      return true;
    ExprResult Expr = TheCall->getArg(0);
    QualType ArgTyExpr = Expr.get()->getType();
    TheCall->setType(ArgTyExpr);
    break;
  }
  case Builtin::BI__builtin_hlsl_wave_multi_prefix_bit_and:
  case Builtin::BI__builtin_hlsl_wave_multi_prefix_bit_or:
  case Builtin::BI__builtin_hlsl_wave_multi_prefix_bit_xor:
  case Builtin::BI__builtin_hlsl_wave_multi_prefix_product:
  case Builtin::BI__builtin_hlsl_wave_multi_prefix_sum: {
    if (SemaRef.checkArgCount(TheCall, 2))
      return true;
 
    if (CheckAnyScalarOrVectorOrMatrix(&SemaRef, TheCall, 0) ||
        CheckNotBoolScalarOrVector(&SemaRef, TheCall, 0))
      return true;
 
    QualType ValueTy = TheCall->getArg(0)->getType();
    QualType ElementTy = getScalarComponentType(ValueTy);
    bool IsArithmetic =
        BuiltinID == Builtin::BI__builtin_hlsl_wave_multi_prefix_product ||
        BuiltinID == Builtin::BI__builtin_hlsl_wave_multi_prefix_sum;
    if ((!ElementTy->isIntegerType() &&
         !(IsArithmetic && ElementTy->hasFloatingRepresentation())) ||
        ElementTy->isBooleanType()) {
      SemaRef.Diag(TheCall->getArg(0)->getBeginLoc(),
                   diag::err_builtin_invalid_arg_type)
          << 1 << (ValueTy->isMatrixType() ? 3 : 5) << 1 << IsArithmetic
          << ValueTy;
      return true;
    }
 
    if (CheckWaveMultiPrefixMask(&SemaRef, TheCall))
      return true;
 
    TheCall->setType(ValueTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_quad_read_across_x:
  case Builtin::BI__builtin_hlsl_quad_read_across_y:
  case Builtin::BI__builtin_hlsl_quad_read_across_diagonal: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
 
    if (CheckAnyScalarOrVector(&SemaRef, TheCall, 0))
      return true;
    if (CheckNotBoolScalarOrVector(&SemaRef, TheCall, 0))
      return true;
    ExprResult Expr = TheCall->getArg(0);
    QualType ArgTyExpr = Expr.get()->getType();
    TheCall->setType(ArgTyExpr);
    break;
  }
  case Builtin::BI__builtin_hlsl_elementwise_splitdouble: {
    if (SemaRef.checkArgCount(TheCall, 3))
      return true;
 
    if (CheckScalarOrVectorOrMatrix(&SemaRef, TheCall, SemaRef.Context.DoubleTy,
                                    0) ||
        CheckScalarOrVectorOrMatrix(&SemaRef, TheCall,
                                    SemaRef.Context.UnsignedIntTy, 1) ||
        CheckScalarOrVectorOrMatrix(&SemaRef, TheCall,
                                    SemaRef.Context.UnsignedIntTy, 2))
      return true;
 
    if (CheckModifiableLValue(&SemaRef, TheCall, 1) ||
        CheckModifiableLValue(&SemaRef, TheCall, 2))
      return true;
    break;
  }
  case Builtin::BI__builtin_hlsl_elementwise_clip: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
 
    if (CheckScalarOrVector(&SemaRef, TheCall, SemaRef.Context.FloatTy, 0))
      return true;
    break;
  }
  case Builtin::BI__builtin_elementwise_acos:
  case Builtin::BI__builtin_elementwise_asin:
  case Builtin::BI__builtin_elementwise_atan:
  case Builtin::BI__builtin_elementwise_atan2:
  case Builtin::BI__builtin_elementwise_ceil:
  case Builtin::BI__builtin_elementwise_cos:
  case Builtin::BI__builtin_elementwise_cosh:
  case Builtin::BI__builtin_elementwise_exp:
  case Builtin::BI__builtin_elementwise_exp2:
  case Builtin::BI__builtin_elementwise_exp10:
  case Builtin::BI__builtin_elementwise_floor:
  case Builtin::BI__builtin_elementwise_fmod:
  case Builtin::BI__builtin_elementwise_log:
  case Builtin::BI__builtin_elementwise_log2:
  case Builtin::BI__builtin_elementwise_log10:
  case Builtin::BI__builtin_elementwise_pow:
  case Builtin::BI__builtin_elementwise_roundeven:
  case Builtin::BI__builtin_elementwise_sin:
  case Builtin::BI__builtin_elementwise_sinh:
  case Builtin::BI__builtin_elementwise_sqrt:
  case Builtin::BI__builtin_elementwise_tan:
  case Builtin::BI__builtin_elementwise_tanh:
  case Builtin::BI__builtin_elementwise_trunc: {
    if (CheckAllArgTypesAreCorrect(&SemaRef, TheCall,
                                   CheckFloatOrHalfRepresentation))
      return true;
    break;
  }
  case Builtin::BI__builtin_hlsl_buffer_update_counter: {
    assert(TheCall->getNumArgs() == 2 && "expected 2 args");
    auto checkResTy = [](const HLSLAttributedResourceType *ResTy) -> bool {
      return !(ResTy->getAttrs().ResourceClass == ResourceClass::UAV &&
               ResTy->getAttrs().RawBuffer && ResTy->hasContainedType());
    };
    if (CheckResourceHandle(&SemaRef, TheCall, 0, checkResTy))
      return true;
    Expr *OffsetExpr = TheCall->getArg(1);
    std::optional<llvm::APSInt> Offset =
        OffsetExpr->getIntegerConstantExpr(SemaRef.getASTContext());
    if (!Offset.has_value() || std::abs(Offset->getExtValue()) != 1) {
      SemaRef.Diag(TheCall->getArg(1)->getBeginLoc(),
                   diag::err_hlsl_expect_arg_const_int_one_or_neg_one)
          << 1;
      return true;
    }
    break;
  }
  case Builtin::BI__builtin_hlsl_elementwise_f16tof32: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
    if (CheckAllArgTypesAreCorrect(&SemaRef, TheCall,
                                   CheckUnsignedIntRepresentation))
      return true;
    // ensure arg integers are 32 bits
    if (CheckExpectedBitWidth(&SemaRef, TheCall, 0, 32))
      return true;
    // check it wasn't a bool type
    QualType ArgTy = TheCall->getArg(0)->getType();
    if (auto *VTy = ArgTy->getAs<VectorType>())
      ArgTy = VTy->getElementType();
    if (ArgTy->isBooleanType()) {
      SemaRef.Diag(TheCall->getArg(0)->getBeginLoc(),
                   diag::err_builtin_invalid_arg_type)
          << 1 << /* scalar or vector of */ 5 << /* unsigned int */ 3
          << /* no fp */ 0 << TheCall->getArg(0)->getType();
      return true;
    }
 
    SetElementTypeAsReturnType(&SemaRef, TheCall, getASTContext().FloatTy);
    break;
  }
  case Builtin::BI__builtin_hlsl_elementwise_f32tof16: {
    if (SemaRef.checkArgCount(TheCall, 1))
      return true;
    if (CheckAllArgTypesAreCorrect(&SemaRef, TheCall, CheckFloatRepresentation))
      return true;
    SetElementTypeAsReturnType(&SemaRef, TheCall,
                               getASTContext().UnsignedIntTy);
    break;
  }
  }
  return false;
}
 
static void BuildFlattenedTypeList(QualType BaseTy,
                                   llvm::SmallVectorImpl<QualType> &List) {
  llvm::SmallVector<QualType, 16> WorkList;
  WorkList.push_back(BaseTy);
  while (!WorkList.empty()) {
    QualType T = WorkList.pop_back_val();
    T = T.getCanonicalType().getUnqualifiedType();
    if (const auto *AT = dyn_cast<ConstantArrayType>(T)) {
      llvm::SmallVector<QualType, 16> ElementFields;
      // Generally I've avoided recursion in this algorithm, but arrays of
      // structs could be time-consuming to flatten and churn through on the
      // work list. Hopefully nesting arrays of structs containing arrays
      // of structs too many levels deep is unlikely.
      BuildFlattenedTypeList(AT->getElementType(), ElementFields);
      // Repeat the element's field list n times.
      for (uint64_t Ct = 0; Ct < AT->getZExtSize(); ++Ct)
        llvm::append_range(List, ElementFields);
      continue;
    }
    // Vectors can only have element types that are builtin types, so this can
    // add directly to the list instead of to the WorkList.
    if (const auto *VT = dyn_cast<VectorType>(T)) {
      List.insert(List.end(), VT->getNumElements(), VT->getElementType());
      continue;
    }
    if (const auto *MT = dyn_cast<ConstantMatrixType>(T)) {
      List.insert(List.end(), MT->getNumElementsFlattened(),
                  MT->getElementType());
      continue;
    }
    if (const auto *RD = T->getAsCXXRecordDecl()) {
      if (RD->isStandardLayout())
        RD = RD->getStandardLayoutBaseWithFields();
 
      // For types that we shouldn't decompose (unions and non-aggregates), just
      // add the type itself to the list.
      if (RD->isUnion() || !RD->isAggregate()) {
        List.push_back(T);
        continue;
      }
 
      llvm::SmallVector<QualType, 16> FieldTypes;
      for (const auto *FD : RD->fields())
        if (!FD->isUnnamedBitField())
          FieldTypes.push_back(FD->getType());
      // Reverse the newly added sub-range.
      std::reverse(FieldTypes.begin(), FieldTypes.end());
      llvm::append_range(WorkList, FieldTypes);
 
      // If this wasn't a standard layout type we may also have some base
      // classes to deal with.
      if (!RD->isStandardLayout()) {
        FieldTypes.clear();
        for (const auto &Base : RD->bases())
          FieldTypes.push_back(Base.getType());
        std::reverse(FieldTypes.begin(), FieldTypes.end());
        llvm::append_range(WorkList, FieldTypes);
      }
      continue;
    }
    List.push_back(T);
  }
}
 
bool SemaHLSL::IsConstantBufferElementCompatible(clang::QualType QT) {
  if (QT.isNull())
    return false;
 
  // Must be a class/struct.
  const auto *RD = QT->getAsCXXRecordDecl();
  if (!RD || RD->isUnion())
    return false;
 
  // Cannot be a resource type or contain one.
  return !QT->isHLSLIntangibleType();
}
 
bool SemaHLSL::IsTypedResourceElementCompatible(clang::QualType QT) {
  // null and array types are not allowed.
  if (QT.isNull() || QT->isArrayType())
    return false;
 
  // UDT types are not allowed
  if (QT->isRecordType())
    return false;
 
  if (QT->isBooleanType() || QT->isEnumeralType())
    return false;
 
  // the only other valid builtin types are scalars or vectors
  if (QT->isArithmeticType()) {
    if (SemaRef.Context.getTypeSize(QT) / 8 > 16)
      return false;
    return true;
  }
 
  if (const VectorType *VT = QT->getAs<VectorType>()) {
    int ArraySize = VT->getNumElements();
 
    if (ArraySize > 4)
      return false;
 
    QualType ElTy = VT->getElementType();
    if (ElTy->isBooleanType())
      return false;
 
    if (SemaRef.Context.getTypeSize(QT) / 8 > 16)
      return false;
    return true;
  }
 
  return false;
}
 
bool SemaHLSL::IsScalarizedLayoutCompatible(QualType T1, QualType T2) const {
  if (T1.isNull() || T2.isNull())
    return false;
 
  T1 = T1.getCanonicalType().getUnqualifiedType();
  T2 = T2.getCanonicalType().getUnqualifiedType();
 
  // If both types are the same canonical type, they're obviously compatible.
  if (SemaRef.getASTContext().hasSameType(T1, T2))
    return true;
 
  llvm::SmallVector<QualType, 16> T1Types;
  BuildFlattenedTypeList(T1, T1Types);
  llvm::SmallVector<QualType, 16> T2Types;
  BuildFlattenedTypeList(T2, T2Types);
 
  // Check the flattened type list
  return llvm::equal(T1Types, T2Types,
                     [this](QualType LHS, QualType RHS) -> bool {
                       return SemaRef.IsLayoutCompatible(LHS, RHS);
                     });
}
 
bool SemaHLSL::CheckCompatibleParameterABI(FunctionDecl *New,
                                           FunctionDecl *Old) {
  if (New->getNumParams() != Old->getNumParams())
    return true;
 
  bool HadError = false;
 
  for (unsigned i = 0, e = New->getNumParams(); i != e; ++i) {
    ParmVarDecl *NewParam = New->getParamDecl(i);
    ParmVarDecl *OldParam = Old->getParamDecl(i);
 
    // HLSL parameter declarations for inout and out must match between
    // declarations. In HLSL inout and out are ambiguous at the call site,
    // but have different calling behavior, so you cannot overload a
    // method based on a difference between inout and out annotations.
    const auto *NDAttr = NewParam->getAttr<HLSLParamModifierAttr>();
    unsigned NSpellingIdx = (NDAttr ? NDAttr->getSpellingListIndex() : 0);
    const auto *ODAttr = OldParam->getAttr<HLSLParamModifierAttr>();
    unsigned OSpellingIdx = (ODAttr ? ODAttr->getSpellingListIndex() : 0);
 
    if (NSpellingIdx != OSpellingIdx) {
      SemaRef.Diag(NewParam->getLocation(),
                   diag::err_hlsl_param_qualifier_mismatch)
          << NDAttr << NewParam;
      SemaRef.Diag(OldParam->getLocation(), diag::note_previous_declaration_as)
          << ODAttr;
      HadError = true;
    }
  }
  return HadError;
}
 
// Generally follows PerformScalarCast, with cases reordered for
// clarity of what types are supported
bool SemaHLSL::CanPerformScalarCast(QualType SrcTy, QualType DestTy) {
 
  if (!SrcTy->isScalarType() || !DestTy->isScalarType())
    return false;
 
  if (SemaRef.getASTContext().hasSameUnqualifiedType(SrcTy, DestTy))
    return true;
 
  switch (SrcTy->getScalarTypeKind()) {
  case Type::STK_Bool: // casting from bool is like casting from an integer
  case Type::STK_Integral:
    switch (DestTy->getScalarTypeKind()) {
    case Type::STK_Bool:
    case Type::STK_Integral:
    case Type::STK_Floating:
      return true;
    case Type::STK_CPointer:
    case Type::STK_ObjCObjectPointer:
    case Type::STK_BlockPointer:
    case Type::STK_MemberPointer:
      llvm_unreachable("HLSL doesn't support pointers.");
    case Type::STK_IntegralComplex:
    case Type::STK_FloatingComplex:
      llvm_unreachable("HLSL doesn't support complex types.");
    case Type::STK_FixedPoint:
      llvm_unreachable("HLSL doesn't support fixed point types.");
    }
    llvm_unreachable("Should have returned before this");
 
  case Type::STK_Floating:
    switch (DestTy->getScalarTypeKind()) {
    case Type::STK_Floating:
    case Type::STK_Bool:
    case Type::STK_Integral:
      return true;
    case Type::STK_FloatingComplex:
    case Type::STK_IntegralComplex:
      llvm_unreachable("HLSL doesn't support complex types.");
    case Type::STK_FixedPoint:
      llvm_unreachable("HLSL doesn't support fixed point types.");
    case Type::STK_CPointer:
    case Type::STK_ObjCObjectPointer:
    case Type::STK_BlockPointer:
    case Type::STK_MemberPointer:
      llvm_unreachable("HLSL doesn't support pointers.");
    }
    llvm_unreachable("Should have returned before this");
 
  case Type::STK_MemberPointer:
  case Type::STK_CPointer:
  case Type::STK_BlockPointer:
  case Type::STK_ObjCObjectPointer:
    llvm_unreachable("HLSL doesn't support pointers.");
 
  case Type::STK_FixedPoint:
    llvm_unreachable("HLSL doesn't support fixed point types.");
 
  case Type::STK_FloatingComplex:
  case Type::STK_IntegralComplex:
    llvm_unreachable("HLSL doesn't support complex types.");
  }
 
  llvm_unreachable("Unhandled scalar cast");
}
 
// Can perform an HLSL Aggregate splat cast if the Dest is an aggregate and the
// Src is a scalar, a vector of length 1, or a 1x1 matrix
// Or if Dest is a vector and Src is a vector of length 1 or a 1x1 matrix
bool SemaHLSL::CanPerformAggregateSplatCast(Expr *Src, QualType DestTy) {
 
  QualType SrcTy = Src->getType();
  // Not a valid HLSL Aggregate Splat cast if Dest is a scalar or if this is
  // going to be a vector splat from a scalar.
  if ((SrcTy->isScalarType() && DestTy->isVectorType()) ||
      DestTy->isScalarType())
    return false;
 
  const VectorType *SrcVecTy = SrcTy->getAs<VectorType>();
  const ConstantMatrixType *SrcMatTy = SrcTy->getAs<ConstantMatrixType>();
 
  // Src isn't a scalar, a vector of length 1, or a 1x1 matrix
  if (!SrcTy->isScalarType() &&
      !(SrcVecTy && SrcVecTy->getNumElements() == 1) &&
      !(SrcMatTy && SrcMatTy->getNumElementsFlattened() == 1))
    return false;
 
  if (SrcVecTy)
    SrcTy = SrcVecTy->getElementType();
  else if (SrcMatTy)
    SrcTy = SrcMatTy->getElementType();
 
  llvm::SmallVector<QualType> DestTypes;
  BuildFlattenedTypeList(DestTy, DestTypes);
 
  for (unsigned I = 0, Size = DestTypes.size(); I < Size; ++I) {
    if (DestTypes[I]->isUnionType())
      return false;
    if (!CanPerformScalarCast(SrcTy, DestTypes[I]))
      return false;
  }
  return true;
}
 
// Can we perform an HLSL Elementwise cast?
bool SemaHLSL::CanPerformElementwiseCast(Expr *Src, QualType DestTy) {
 
  // Don't handle casts where LHS and RHS are any combination of scalar/vector
  // There must be an aggregate somewhere
  QualType SrcTy = Src->getType();
  if (SrcTy->isScalarType()) // always a splat and this cast doesn't handle that
    return false;
 
  if (SrcTy->isVectorType() &&
      (DestTy->isScalarType() || DestTy->isVectorType()))
    return false;
 
  if (SrcTy->isConstantMatrixType() &&
      (DestTy->isScalarType() || DestTy->isConstantMatrixType()))
    return false;
 
  llvm::SmallVector<QualType> DestTypes;
  BuildFlattenedTypeList(DestTy, DestTypes);
  llvm::SmallVector<QualType> SrcTypes;
  BuildFlattenedTypeList(SrcTy, SrcTypes);
 
  // Usually the size of SrcTypes must be greater than or equal to the size of
  // DestTypes.
  if (SrcTypes.size() < DestTypes.size())
    return false;
 
  unsigned SrcSize = SrcTypes.size();
  unsigned DstSize = DestTypes.size();
  unsigned I;
  for (I = 0; I < DstSize && I < SrcSize; I++) {
    if (SrcTypes[I]->isUnionType() || DestTypes[I]->isUnionType())
      return false;
    if (!CanPerformScalarCast(SrcTypes[I], DestTypes[I])) {
      return false;
    }
  }
 
  // check the rest of the source type for unions.
  for (; I < SrcSize; I++) {
    if (SrcTypes[I]->isUnionType())
      return false;
  }
  return true;
}
 
bool SemaHLSL::CanPerformPackedTypeCast(Expr *Src, QualType DestTy) {
  ASTContext &Ctx = SemaRef.getASTContext();
  QualType UIntTy = Ctx.UnsignedIntTy;
  QualType SrcTy = Src->getType();
 
  return (SrcTy->isHLSLBuiltinPackedType() &&
          DestTy->isHLSLBuiltinPackedType()) ||
         (SrcTy->isHLSLBuiltinPackedType() &&
          Ctx.hasSameUnqualifiedType(DestTy, UIntTy)) ||
         (DestTy->isHLSLBuiltinPackedType() &&
          Ctx.hasSameUnqualifiedType(SrcTy, UIntTy));
}
 
ExprResult SemaHLSL::CreateOutArgExpr(Expr *Arg, QualType ArgTy,
                                      QualType ParamTy,
                                      InitializedEntity Entity,
                                      SourceLocation Loc, bool IsInOut) {
  ASTContext &Ctx = SemaRef.getASTContext();
 
  QualType Ty = ParamTy.getNonLValueExprType(Ctx);
 
  auto *ArgOpV =
      new (Ctx) OpaqueValueExpr(Loc, ArgTy, VK_LValue, OK_Ordinary, Arg);
 
  // Parameters are initialized via copy initialization. This allows for
  // overload resolution of argument constructors.
  ExprResult Res = SemaRef.PerformCopyInitialization(Entity, Loc, ArgOpV);
  if (Res.isInvalid())
    return ExprError();
  Expr *Base = Res.get();
  // After the cast, drop the reference type when creating the exprs.
  Ty = Ty.getNonLValueExprType(Ctx);
  auto *OpV = new (Ctx) OpaqueValueExpr(Loc, Ty, VK_LValue, OK_Ordinary, Base);
 
  // Writebacks are performed with `=` binary operator, which allows for
  // overload resolution on writeback result expressions.
  Res = SemaRef.ActOnBinOp(SemaRef.getCurScope(), Arg->getBeginLoc(),
                           tok::equal, ArgOpV, OpV);
 
  if (Res.isInvalid())
    return ExprError();
  Expr *Writeback = Res.get();
  auto *OutExpr =
      HLSLOutArgExpr::Create(Ctx, Ty, ArgOpV, OpV, Writeback, IsInOut);
  return ExprResult(OutExpr);
}
 
ExprResult SemaHLSL::ActOnOutParamExpr(ParmVarDecl *Param, Expr *Arg) {
  ASTContext &Ctx = SemaRef.getASTContext();
  assert(Param->hasAttr<HLSLParamModifierAttr>() &&
         "We should not get here without a parameter modifier expression");
  const auto *Attr = Param->getAttr<HLSLParamModifierAttr>();
  if (Attr->getABI() == ParameterABI::Ordinary)
    return ExprResult(Arg);
 
  bool IsInOut = Attr->getABI() == ParameterABI::HLSLInOut;
  if (!Arg->isLValue()) {
    SemaRef.Diag(Arg->getBeginLoc(), diag::error_hlsl_inout_lvalue)
        << Arg << (IsInOut ? 1 : 0);
    return ExprError();
  }
 
  QualType Ty = Param->getType().getNonLValueExprType(Ctx);
 
  // HLSL allows implicit conversions from scalars to vectors, but not the
  // inverse, so we need to disallow `inout` with scalar->vector or
  // scalar->matrix conversions.
  if (Arg->getType()->isScalarType() != Ty->isScalarType()) {
    SemaRef.Diag(Arg->getBeginLoc(), diag::error_hlsl_inout_scalar_extension)
        << Arg << (IsInOut ? 1 : 0);
    return ExprError();
  }
 
  InitializedEntity Entity =
      InitializedEntity::InitializeParameter(Ctx, Ty, false);
 
  return CreateOutArgExpr(Arg, Arg->getType(), Param->getType(), Entity,
                          Param->getBeginLoc(), IsInOut);
}
 
QualType SemaHLSL::getInoutParameterType(QualType Ty) {
  // If HLSL gains support for references, all the cites that use this will need
  // to be updated with semantic checking to produce errors for
  // pointers/references.
  assert(!Ty->isReferenceType() &&
         "Pointer and reference types cannot be inout or out parameters");
  Ty = SemaRef.getASTContext().getLValueReferenceType(Ty);
  Ty.addRestrict();
  return Ty;
}
 
// Returns true if the type has a non-empty constant buffer layout (if it is
// scalar, vector or matrix, or if it contains any of these.
static bool hasConstantBufferLayout(QualType QT) {
  const Type *Ty = QT->getUnqualifiedDesugaredType();
  if (Ty->isScalarType() || Ty->isVectorType() || Ty->isMatrixType())
    return true;
 
  if (Ty->isHLSLResourceRecord() || Ty->isHLSLResourceRecordArray())
    return false;
 
  if (const auto *RD = Ty->getAsCXXRecordDecl()) {
    for (const auto *FD : RD->fields()) {
      if (hasConstantBufferLayout(FD->getType()))
        return true;
    }
    assert(RD->getNumBases() <= 1 &&
           "HLSL doesn't support multiple inheritance");
    return RD->getNumBases()
               ? hasConstantBufferLayout(RD->bases_begin()->getType())
               : false;
  }
 
  if (const auto *AT = dyn_cast<ArrayType>(Ty)) {
    if (const auto *CAT = dyn_cast<ConstantArrayType>(AT))
      if (isZeroSizedArray(CAT))
        return false;
    return hasConstantBufferLayout(AT->getElementType());
  }
 
  return false;
}
 
static bool IsDefaultBufferConstantDecl(const ASTContext &Ctx, VarDecl *VD) {
  bool IsVulkan =
      Ctx.getTargetInfo().getTriple().getOS() == llvm::Triple::Vulkan;
  bool IsVKPushConstant = IsVulkan && VD->hasAttr<HLSLVkPushConstantAttr>();
  QualType QT = VD->getType();
  return VD->getDeclContext()->isTranslationUnit() &&
         QT.getAddressSpace() == LangAS::Default &&
         VD->getStorageClass() != SC_Static &&
         !VD->hasAttr<HLSLVkConstantIdAttr>() && !IsVKPushConstant &&
         hasConstantBufferLayout(QT);
}
 
void SemaHLSL::deduceAddressSpace(VarDecl *Decl) {
  // The variable already has an address space (groupshared for ex).
  if (Decl->getType().hasAddressSpace())
    return;
 
  if (Decl->getType()->isDependentType())
    return;
 
  QualType Type = Decl->getType();
 
  if (Decl->hasAttr<HLSLVkExtBuiltinInputAttr>()) {
    LangAS ImplAS = LangAS::hlsl_input;
    Type = SemaRef.getASTContext().getAddrSpaceQualType(Type, ImplAS);
    Decl->setType(Type);
    return;
  }
 
  if (Decl->hasAttr<HLSLVkExtBuiltinOutputAttr>()) {
    LangAS ImplAS = LangAS::hlsl_output;
    Type = SemaRef.getASTContext().getAddrSpaceQualType(Type, ImplAS);
    Decl->setType(Type);
 
    // HLSL uses `static` differently than C++. For BuiltIn output, the static
    // does not imply private to the module scope.
    // Marking it as external to reflect the semantic this attribute brings.
    // See https://github.com/microsoft/hlsl-specs/issues/350
    Decl->setStorageClass(SC_Extern);
    return;
  }
 
  bool IsVulkan = getASTContext().getTargetInfo().getTriple().getOS() ==
                  llvm::Triple::Vulkan;
  if (IsVulkan && Decl->hasAttr<HLSLVkPushConstantAttr>()) {
    if (HasDeclaredAPushConstant)
      SemaRef.Diag(Decl->getLocation(), diag::err_hlsl_push_constant_unique);
 
    LangAS ImplAS = LangAS::hlsl_push_constant;
    Type = SemaRef.getASTContext().getAddrSpaceQualType(Type, ImplAS);
    Decl->setType(Type);
    HasDeclaredAPushConstant = true;
    return;
  }
 
  if (Type->isSamplerT() || Type->isVoidType())
    return;
 
  // Resource handles.
  if (Type->isHLSLResourceRecord() || Type->isHLSLResourceRecordArray())
    return;
 
  // Only static globals belong to the Private address space.
  // Non-static globals belongs to the cbuffer.
  if (Decl->getStorageClass() != SC_Static && !Decl->isStaticDataMember())
    return;
 
  LangAS ImplAS = LangAS::hlsl_private;
  Type = SemaRef.getASTContext().getAddrSpaceQualType(Type, ImplAS);
  Decl->setType(Type);
}
 
namespace {
 
// Helper class for assigning bindings to resources declared within a struct.
// It keeps track of all binding attributes declared on a struct instance, and
// the offsets for each register type that have been assigned so far.
// Handles both explicit and implicit bindings.
class StructBindingContext {
  // Bindings and offsets per register type. We only need to support four
  // register types - SRV (u), UAV (t), CBuffer (c), and Sampler (s).
  HLSLResourceBindingAttr *RegBindingsAttrs[4];
  unsigned RegBindingOffset[4];
 
  // Make sure the RegisterType values are what we expect
  static_assert(static_cast<unsigned>(RegisterType::SRV) == 0 &&
                    static_cast<unsigned>(RegisterType::UAV) == 1 &&
                    static_cast<unsigned>(RegisterType::CBuffer) == 2 &&
                    static_cast<unsigned>(RegisterType::Sampler) == 3,
                "unexpected register type values");
 
  // Vulkan binding attribute does not vary by register type.
  HLSLVkBindingAttr *VkBindingAttr;
  unsigned VkBindingOffset;
 
public:
  // Constructor: gather all binding attributes on a struct instance and
  // initialize offsets.
  StructBindingContext(VarDecl *VD) {
    for (unsigned i = 0; i < 4; ++i) {
      RegBindingsAttrs[i] = nullptr;
      RegBindingOffset[i] = 0;
    }
    VkBindingAttr = nullptr;
    VkBindingOffset = 0;
 
    ASTContext &AST = VD->getASTContext();
    bool IsSpirv = AST.getTargetInfo().getTriple().isSPIRV();
 
    for (Attr *A : VD->attrs()) {
      if (auto *RBA = dyn_cast<HLSLResourceBindingAttr>(A)) {
        RegisterType RegType = RBA->getRegisterType();
        unsigned RegTypeIdx = static_cast<unsigned>(RegType);
        // Ignore unsupported register annotations, such as 'c' or 'i'.
        if (RegTypeIdx < 4)
          RegBindingsAttrs[RegTypeIdx] = RBA;
        continue;
      }
      // Gather the Vulkan binding attributes only if the target is SPIR-V.
      if (IsSpirv) {
        if (auto *VBA = dyn_cast<HLSLVkBindingAttr>(A))
          VkBindingAttr = VBA;
      }
    }
  }
 
  // Creates a binding attribute for a resource based on the gathered attributes
  // and the required register type and range.
  Attr *createBindingAttr(SemaHLSL &S, ASTContext &AST, RegisterType RegType,
                          unsigned Range, bool HasCounter) {
    assert(static_cast<unsigned>(RegType) < 4 && "unexpected register type");
 
    if (VkBindingAttr) {
      unsigned Offset = VkBindingOffset;
      VkBindingOffset += Range;
      return HLSLVkBindingAttr::CreateImplicit(
          AST, VkBindingAttr->getBinding() + Offset, VkBindingAttr->getSet(),
          VkBindingAttr->getRange());
    }
 
    HLSLResourceBindingAttr *RBA =
        RegBindingsAttrs[static_cast<unsigned>(RegType)];
    HLSLResourceBindingAttr *NewAttr = nullptr;
 
    if (RBA && RBA->hasRegisterSlot()) {
      // Explicit binding - create a new attribute with offseted slot number
      // based on the required register type.
      unsigned Offset = RegBindingOffset[static_cast<unsigned>(RegType)];
      RegBindingOffset[static_cast<unsigned>(RegType)] += Range;
 
      unsigned NewSlotNumber = RBA->getSlotNumber() + Offset;
      StringRef NewSlotNumberStr =
          createRegisterString(AST, RBA->getRegisterType(), NewSlotNumber);
      NewAttr = HLSLResourceBindingAttr::CreateImplicit(
          AST, NewSlotNumberStr, RBA->getSpace(), RBA->getRange());
      NewAttr->setBinding(RegType, NewSlotNumber, RBA->getSpaceNumber());
    } else {
      // No binding attribute or space-only binding - create a binding
      // attribute for implicit binding.
      NewAttr = HLSLResourceBindingAttr::CreateImplicit(AST, "", "0", {});
      NewAttr->setBinding(RegType, std::nullopt,
                          RBA ? RBA->getSpaceNumber() : 0);
      NewAttr->setImplicitBindingOrderID(S.getNextImplicitBindingOrderID());
    }
    if (HasCounter)
      NewAttr->setImplicitCounterBindingOrderID(
          S.getNextImplicitBindingOrderID());
    return NewAttr;
  }
};
 
// Creates a global variable declaration for a resource field embedded in a
// struct, assigns it a binding, initializes it, and associates it with the
// struct declaration via an HLSLAssociatedResourceDeclAttr.
static void createGlobalResourceDeclForStruct(
    Sema &S, VarDecl *ParentVD, SourceLocation Loc, IdentifierInfo *Id,
    QualType ResTy, StructBindingContext &BindingCtx) {
  assert(isResourceRecordTypeOrArrayOf(ResTy) &&
         "expected resource type or array of resources");
 
  DeclContext *DC = ParentVD->getNonTransparentDeclContext();
  assert(DC->isTranslationUnit() && "expected translation unit decl context");
 
  ASTContext &AST = S.getASTContext();
  VarDecl *ResDecl =
      VarDecl::Create(AST, DC, Loc, Loc, Id, ResTy, nullptr, SC_None);
 
  unsigned Range = 1;
  const Type *SingleResTy = ResTy.getTypePtr()->getUnqualifiedDesugaredType();
  while (const auto *AT = dyn_cast<ArrayType>(SingleResTy)) {
    const auto *CAT = dyn_cast<ConstantArrayType>(AT);
    Range = CAT ? (Range * CAT->getSize().getZExtValue()) : 0;
    SingleResTy =
        AT->getArrayElementTypeNoTypeQual()->getUnqualifiedDesugaredType();
  }
  const HLSLAttributedResourceType *ResHandleTy =
      HLSLAttributedResourceType::findHandleTypeOnResource(SingleResTy);
 
  // Add a binding attribute to the global resource declaration.
  bool HasCounter = hasCounterHandle(SingleResTy->getAsCXXRecordDecl());
  Attr *BindingAttr = BindingCtx.createBindingAttr(
      S.HLSL(), AST, getRegisterType(ResHandleTy), Range, HasCounter);
  ResDecl->addAttr(BindingAttr);
  ResDecl->addAttr(InternalLinkageAttr::CreateImplicit(AST));
  ResDecl->setImplicit();
 
  if (Range == 1)
    S.HLSL().initGlobalResourceDecl(ResDecl);
  else
    S.HLSL().initGlobalResourceArrayDecl(ResDecl);
 
  ParentVD->addAttr(
      HLSLAssociatedResourceDeclAttr::CreateImplicit(AST, ResDecl));
  DC->addDecl(ResDecl);
 
  DeclGroupRef DG(ResDecl);
  S.Consumer.HandleTopLevelDecl(DG);
}
 
static void handleArrayOfStructWithResources(
    Sema &S, VarDecl *ParentVD, const ConstantArrayType *CAT,
    EmbeddedResourceNameBuilder &NameBuilder, StructBindingContext &BindingCtx);
 
// Scans base and all fields of a struct/class type to find all embedded
// resources or resource arrays. Creates a global variable for each resource
// found.
static void handleStructWithResources(Sema &S, VarDecl *ParentVD,
                                      const CXXRecordDecl *RD,
                                      EmbeddedResourceNameBuilder &NameBuilder,
                                      StructBindingContext &BindingCtx) {
 
  // Scan the base classes.
  assert(RD->getNumBases() <= 1 && "HLSL doesn't support multiple inheritance");
  const auto *BasesIt = RD->bases_begin();
  if (BasesIt != RD->bases_end()) {
    QualType QT = BasesIt->getType();
    if (QT->isHLSLIntangibleType()) {
      CXXRecordDecl *BaseRD = QT->getAsCXXRecordDecl();
      NameBuilder.pushBaseName(BaseRD->getName());
      handleStructWithResources(S, ParentVD, BaseRD, NameBuilder, BindingCtx);
      NameBuilder.pop();
    }
  }
  // Process this class fields.
  for (const FieldDecl *FD : RD->fields()) {
    QualType FDTy = FD->getType().getCanonicalType();
    if (!FDTy->isHLSLIntangibleType())
      continue;
 
    NameBuilder.pushName(FD->getName());
 
    if (isResourceRecordTypeOrArrayOf(FDTy)) {
      IdentifierInfo *II = NameBuilder.getNameAsIdentifier(S.getASTContext());
      createGlobalResourceDeclForStruct(S, ParentVD, FD->getLocation(), II,
                                        FDTy, BindingCtx);
    } else if (const auto *RD = FDTy->getAsCXXRecordDecl()) {
      handleStructWithResources(S, ParentVD, RD, NameBuilder, BindingCtx);
 
    } else if (const auto *ArrayTy = dyn_cast<ConstantArrayType>(FDTy)) {
      assert(!FDTy->isHLSLResourceRecordArray() &&
             "resource arrays should have been already handled");
      handleArrayOfStructWithResources(S, ParentVD, ArrayTy, NameBuilder,
                                       BindingCtx);
    }
    NameBuilder.pop();
  }
}
 
// Processes array of structs with resources.
static void
handleArrayOfStructWithResources(Sema &S, VarDecl *ParentVD,
                                 const ConstantArrayType *CAT,
                                 EmbeddedResourceNameBuilder &NameBuilder,
                                 StructBindingContext &BindingCtx) {
 
  QualType ElementTy = CAT->getElementType().getCanonicalType();
  assert(ElementTy->isHLSLIntangibleType() && "Expected HLSL intangible type");
 
  const ConstantArrayType *SubCAT = dyn_cast<ConstantArrayType>(ElementTy);
  const CXXRecordDecl *ElementRD = ElementTy->getAsCXXRecordDecl();
 
  if (!SubCAT && !ElementRD)
    return;
 
  for (unsigned I = 0, E = CAT->getSize().getZExtValue(); I < E; ++I) {
    NameBuilder.pushArrayIndex(I);
    if (ElementRD)
      handleStructWithResources(S, ParentVD, ElementRD, NameBuilder,
                                BindingCtx);
    else
      handleArrayOfStructWithResources(S, ParentVD, SubCAT, NameBuilder,
                                       BindingCtx);
    NameBuilder.pop();
  }
}
 
} // namespace
 
// Scans all fields of a user-defined struct (or array of structs)
// to find all embedded resources or resource arrays. For each resource
// a global variable of the resource type is created and associated
// with the parent declaration (VD) through a HLSLAssociatedResourceDeclAttr
// attribute.
void SemaHLSL::handleGlobalStructOrArrayOfWithResources(VarDecl *VD) {
  EmbeddedResourceNameBuilder NameBuilder(VD->getName());
  StructBindingContext BindingCtx(VD);
 
  const Type *VDTy = VD->getType().getTypePtr();
  assert(VDTy->isHLSLIntangibleType() && !isResourceRecordTypeOrArrayOf(VD) &&
         "Expected non-resource struct or array type");
 
  if (const CXXRecordDecl *RD = VDTy->getAsCXXRecordDecl()) {
    handleStructWithResources(SemaRef, VD, RD, NameBuilder, BindingCtx);
    return;
  }
 
  if (const auto *CAT = dyn_cast<ConstantArrayType>(VDTy)) {
    handleArrayOfStructWithResources(SemaRef, VD, CAT, NameBuilder, BindingCtx);
    return;
  }
}
 
void SemaHLSL::ActOnVariableDeclarator(VarDecl *VD) {
  if (VD->hasGlobalStorage()) {
    // make sure the declaration has a complete type
    if (SemaRef.RequireCompleteType(
            VD->getLocation(),
            SemaRef.getASTContext().getBaseElementType(VD->getType()),
            diag::err_typecheck_decl_incomplete_type)) {
      VD->setInvalidDecl();
      deduceAddressSpace(VD);
      return;
    }
 
    // Global variables outside a cbuffer block that are not a resource, static,
    // groupshared, or an empty array or struct belong to the default constant
    // buffer $Globals (to be created at the end of the translation unit).
    if (IsDefaultBufferConstantDecl(getASTContext(), VD)) {
      // update address space to hlsl_constant
      QualType NewTy = getASTContext().getAddrSpaceQualType(
          VD->getType(), LangAS::hlsl_constant);
      VD->setType(NewTy);
      DefaultCBufferDecls.push_back(VD);
    }
 
    // find all resources bindings on decl
    if (VD->getType()->isHLSLIntangibleType())
      collectResourceBindingsOnVarDecl(VD);
 
    if (VD->hasAttr<HLSLVkConstantIdAttr>())
      VD->setStorageClass(StorageClass::SC_Static);
 
    if (isResourceRecordTypeOrArrayOf(VD) &&
        VD->getStorageClass() != SC_Static) {
      // Add internal linkage attribute to non-static resource variables. The
      // global externally visible storage is accessed through the handle, which
      // is a member. The variable itself is not externally visible.
      VD->addAttr(InternalLinkageAttr::CreateImplicit(getASTContext()));
    }
 
    // process explicit bindings
    processExplicitBindingsOnDecl(VD);
 
    // Add implicit binding attribute to non-static resource arrays.
    if (VD->getType()->isHLSLResourceRecordArray() &&
        VD->getStorageClass() != SC_Static) {
      // If the resource array does not have an explicit binding attribute,
      // create an implicit one. It will be used to transfer implicit binding
      // order_ID to codegen.
      ResourceBindingAttrs Binding(VD);
      if (!Binding.isExplicit()) {
        uint32_t OrderID = getNextImplicitBindingOrderID();
        if (Binding.hasBinding())
          Binding.setImplicitOrderID(OrderID);
        else {
          addImplicitBindingAttrToDecl(
              SemaRef, VD, getRegisterType(getResourceArrayHandleType(VD)),
              OrderID);
          // Re-create the binding object to pick up the new attribute.
          Binding = ResourceBindingAttrs(VD);
        }
      }
 
      // Get to the base type of a potentially multi-dimensional array.
      QualType Ty = getASTContext().getBaseElementType(VD->getType());
 
      const CXXRecordDecl *RD = Ty->getAsCXXRecordDecl();
      if (hasCounterHandle(RD)) {
        if (!Binding.hasCounterImplicitOrderID()) {
          uint32_t OrderID = getNextImplicitBindingOrderID();
          Binding.setCounterImplicitOrderID(OrderID);
        }
      }
    }
 
    // Process resources in user-defined structs, or arrays of such structs.
    if (VD->getStorageClass() != SC_Static &&
        isStructOrArrayOfWithResources(VD->getType()))
      handleGlobalStructOrArrayOfWithResources(VD);
 
    // Mark groupshared variables as extern so they will have
    // external storage and won't be default initialized
    if (VD->hasAttr<HLSLGroupSharedAddressSpaceAttr>())
      VD->setStorageClass(StorageClass::SC_Extern);
  }
 
  deduceAddressSpace(VD);
}
 
bool SemaHLSL::initGlobalResourceDecl(VarDecl *VD) {
  assert(VD->getType()->isHLSLResourceRecord() &&
         "expected resource record type");
 
  ASTContext &AST = SemaRef.getASTContext();
  uint64_t UIntTySize = AST.getTypeSize(AST.UnsignedIntTy);
  uint64_t IntTySize = AST.getTypeSize(AST.IntTy);
 
  // Gather resource binding attributes.
  ResourceBindingAttrs Binding(VD);
 
  // Find correct initialization method and create its arguments.
  QualType ResourceTy = VD->getType();
  CXXRecordDecl *ResourceDecl = ResourceTy->getAsCXXRecordDecl();
  CXXMethodDecl *CreateMethod = nullptr;
  llvm::SmallVector<Expr *> Args;
 
  bool HasCounter = hasCounterHandle(ResourceDecl);
  const char *CreateMethodName;
  if (Binding.isExplicit())
    CreateMethodName = HasCounter ? "__createFromBindingWithImplicitCounter"
                                  : "__createFromBinding";
  else
    CreateMethodName = HasCounter
                           ? "__createFromImplicitBindingWithImplicitCounter"
                           : "__createFromImplicitBinding";
 
  CreateMethod =
      lookupMethod(SemaRef, ResourceDecl, CreateMethodName, VD->getLocation());
 
  if (!CreateMethod) {
    // This can happen if someone creates a struct that looks like an HLSL
    // resource record but does not have the required static create method.
    // No binding will be generated for it.
    assert(!ResourceDecl->isImplicit() &&
           "create method lookup should always succeed for built-in resource "
           "records");
    return false;
  }
 
  if (Binding.isExplicit()) {
    IntegerLiteral *RegSlot =
        IntegerLiteral::Create(AST, llvm::APInt(UIntTySize, Binding.getSlot()),
                               AST.UnsignedIntTy, SourceLocation());
    Args.push_back(RegSlot);
  } else {
    uint32_t OrderID = (Binding.hasImplicitOrderID())
                           ? Binding.getImplicitOrderID()
                           : getNextImplicitBindingOrderID();
    IntegerLiteral *OrderId =
        IntegerLiteral::Create(AST, llvm::APInt(UIntTySize, OrderID),
                               AST.UnsignedIntTy, SourceLocation());
    Args.push_back(OrderId);
  }
 
  IntegerLiteral *Space =
      IntegerLiteral::Create(AST, llvm::APInt(UIntTySize, Binding.getSpace()),
                             AST.UnsignedIntTy, SourceLocation());
  Args.push_back(Space);
 
  IntegerLiteral *RangeSize = IntegerLiteral::Create(
      AST, llvm::APInt(IntTySize, 1), AST.IntTy, SourceLocation());
  Args.push_back(RangeSize);
 
  IntegerLiteral *Index = IntegerLiteral::Create(
      AST, llvm::APInt(UIntTySize, 0), AST.UnsignedIntTy, SourceLocation());
  Args.push_back(Index);
 
  StringRef VarName = VD->getName();
  StringLiteral *Name = StringLiteral::Create(
      AST, VarName, StringLiteralKind::Ordinary, false,
      AST.getStringLiteralArrayType(AST.CharTy.withConst(), VarName.size()),
      SourceLocation());
  ImplicitCastExpr *NameCast = ImplicitCastExpr::Create(
      AST, AST.getPointerType(AST.CharTy.withConst()), CK_ArrayToPointerDecay,
      Name, nullptr, VK_PRValue, FPOptionsOverride());
  Args.push_back(NameCast);
 
  if (HasCounter) {
    // Will this be in the correct order?
    uint32_t CounterOrderID = getNextImplicitBindingOrderID();
    IntegerLiteral *CounterId =
        IntegerLiteral::Create(AST, llvm::APInt(UIntTySize, CounterOrderID),
                               AST.UnsignedIntTy, SourceLocation());
    Args.push_back(CounterId);
  }
 
  // Make sure the create method template is instantiated and emitted.
  if (!CreateMethod->isDefined() && CreateMethod->isTemplateInstantiation())
    SemaRef.InstantiateFunctionDefinition(VD->getLocation(), CreateMethod,
                                          true);
 
  // Create CallExpr with a call to the static method and set it as the decl
  // initialization.
  DeclRefExpr *DRE = DeclRefExpr::Create(
      AST, NestedNameSpecifierLoc(), SourceLocation(), CreateMethod, false,
      CreateMethod->getNameInfo(), CreateMethod->getType(), VK_PRValue);
 
  auto *ImpCast = ImplicitCastExpr::Create(
      AST, AST.getPointerType(CreateMethod->getType()),
      CK_FunctionToPointerDecay, DRE, nullptr, VK_PRValue, FPOptionsOverride());
 
  CallExpr *InitExpr =
      CallExpr::Create(AST, ImpCast, Args, ResourceTy, VK_PRValue,
                       SourceLocation(), FPOptionsOverride());
  VD->setInit(InitExpr);
  VD->setInitStyle(VarDecl::CallInit);
  SemaRef.CheckCompleteVariableDeclaration(VD);
  return true;
}
 
bool SemaHLSL::initGlobalResourceArrayDecl(VarDecl *VD) {
  assert(VD->getType()->isHLSLResourceRecordArray() &&
         "expected array of resource records");
 
  // Individual resources in a resource array are not initialized here. They
  // are initialized later on during codegen when the individual resources are
  // accessed. Codegen will emit a call to the resource initialization method
  // with the specified array index. We need to make sure though that the method
  // for the specific resource type is instantiated, so codegen can emit a call
  // to it when the array element is accessed.
 
  // Find correct initialization method based on the resource binding
  // information.
  ASTContext &AST = SemaRef.getASTContext();
  QualType ResElementTy = AST.getBaseElementType(VD->getType());
  CXXRecordDecl *ResourceDecl = ResElementTy->getAsCXXRecordDecl();
  CXXMethodDecl *CreateMethod = nullptr;
 
  bool HasCounter = hasCounterHandle(ResourceDecl);
  ResourceBindingAttrs ResourceAttrs(VD);
  if (ResourceAttrs.isExplicit())
    // Resource has explicit binding.
    CreateMethod =
        lookupMethod(SemaRef, ResourceDecl,
                     HasCounter ? "__createFromBindingWithImplicitCounter"
                                : "__createFromBinding",
                     VD->getLocation());
  else
    // Resource has implicit binding.
    CreateMethod = lookupMethod(
        SemaRef, ResourceDecl,
        HasCounter ? "__createFromImplicitBindingWithImplicitCounter"
                   : "__createFromImplicitBinding",
        VD->getLocation());
 
  if (!CreateMethod)
    return false;
 
  // Make sure the create method template is instantiated and emitted.
  if (!CreateMethod->isDefined() && CreateMethod->isTemplateInstantiation())
    SemaRef.InstantiateFunctionDefinition(VD->getLocation(), CreateMethod,
                                          true);
  return true;
}
 
// Returns true if the initialization has been handled.
// Returns false to use default initialization.
bool SemaHLSL::ActOnUninitializedVarDecl(VarDecl *VD) {
  // Objects in the hlsl_constant address space are initialized
  // externally, so don't synthesize an implicit initializer.
  if (VD->getType().getAddressSpace() == LangAS::hlsl_constant)
    return true;
 
  if (VD->hasGlobalStorage() && VD->getStorageClass() != SC_Static) {
    const Type *Ty = VD->getType().getTypePtr();
    if (Ty->isHLSLResourceRecord() && initGlobalResourceDecl(VD))
      return true;
    if (Ty->isHLSLResourceRecordArray() && initGlobalResourceArrayDecl(VD))
      return true;
  }
 
  // User-defined structs/classes do not have constructors.
  // When declared at a global scope, they are part of the constant buffer
  // and should not be initialized by the compiler.
  // When declared at a local scope, they are not initialized.
  // Also applies to arrays of user-defined structs/classes.
  const Type *Ty = VD->getType()->getUnqualifiedDesugaredType();
  while (Ty->isArrayType())
    Ty = Ty->getArrayElementTypeNoTypeQual()->getUnqualifiedDesugaredType();
  if (CXXRecordDecl *RD = Ty->getAsCXXRecordDecl())
    return !RD->isHLSLBuiltinRecord();
 
  return false;
}
 
std::optional<const DeclBindingInfo *> SemaHLSL::inferGlobalBinding(Expr *E) {
  if (auto *Ternary = dyn_cast<ConditionalOperator>(E)) {
    auto TrueInfo = inferGlobalBinding(Ternary->getTrueExpr());
    auto FalseInfo = inferGlobalBinding(Ternary->getFalseExpr());
    if (!TrueInfo || !FalseInfo)
      return std::nullopt;
    if (*TrueInfo != *FalseInfo)
      return std::nullopt;
    return TrueInfo;
  }
 
  if (auto *ASE = dyn_cast<ArraySubscriptExpr>(E))
    E = ASE->getBase()->IgnoreParenImpCasts();
 
  if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E->IgnoreParens()))
    if (VarDecl *VD = dyn_cast<VarDecl>(DRE->getDecl())) {
      const Type *Ty = VD->getType()->getUnqualifiedDesugaredType();
      if (Ty->isArrayType())
        Ty = Ty->getArrayElementTypeNoTypeQual();
 
      if (const auto *AttrResType =
              HLSLAttributedResourceType::findHandleTypeOnResource(Ty)) {
        ResourceClass RC = AttrResType->getAttrs().ResourceClass;
        return Bindings.getDeclBindingInfo(VD, RC);
      }
    }
 
  return nullptr;
}
 
void SemaHLSL::trackLocalResource(VarDecl *VD, Expr *E) {
  std::optional<const DeclBindingInfo *> ExprBinding = inferGlobalBinding(E);
  if (!ExprBinding) {
    SemaRef.Diag(E->getBeginLoc(),
                 diag::warn_hlsl_assigning_local_resource_is_not_unique)
        << E << VD;
    return; // Expr use multiple resources
  }
 
  if (*ExprBinding == nullptr)
    return; // No binding could be inferred to track, return without error
 
  auto PrevBinding = Assigns.find(VD);
  if (PrevBinding == Assigns.end()) {
    // No previous binding recorded, simply record the new assignment
    Assigns.insert({VD, *ExprBinding});
    return;
  }
 
  // Otherwise, warn if the assignment implies different resource bindings
  if (*ExprBinding != PrevBinding->second) {
    SemaRef.Diag(E->getBeginLoc(),
                 diag::warn_hlsl_assigning_local_resource_is_not_unique)
        << E << VD;
    SemaRef.Diag(VD->getLocation(), diag::note_var_declared_here) << VD;
    return;
  }
 
  return;
}
 
bool SemaHLSL::CheckResourceBinOp(BinaryOperatorKind Opc, Expr *LHSExpr,
                                  Expr *RHSExpr, SourceLocation Loc) {
  assert((LHSExpr->getType()->isHLSLResourceRecord() ||
          LHSExpr->getType()->isHLSLResourceRecordArray()) &&
         "expected LHS to be a resource record or array of resource records");
  if (Opc != BO_Assign)
    return true;
 
  // If LHS is an array subscript, get the underlying declaration.
  Expr *E = LHSExpr;
  while (auto *ASE = dyn_cast<ArraySubscriptExpr>(E))
    E = ASE->getBase()->IgnoreParenImpCasts();
 
  // Report error if LHS is a non-static resource declared at a global scope.
  if (DeclRefExpr *DRE = dyn_cast<DeclRefExpr>(E->IgnoreParens())) {
    if (VarDecl *VD = dyn_cast<VarDecl>(DRE->getDecl())) {
      if (VD->hasGlobalStorage() && VD->getStorageClass() != SC_Static) {
        // assignment to global resource is not allowed
        SemaRef.Diag(Loc, diag::err_hlsl_assign_to_global_resource) << VD;
        SemaRef.Diag(VD->getLocation(), diag::note_var_declared_here) << VD;
        return false;
      }
 
      trackLocalResource(VD, RHSExpr);
    }
  }
  return true;
}
 
// Returns true if the given type can have an overload of the given
// binary operator.
bool SemaHLSL::canHaveOverloadedBinOp(QualType LHSTy, BinaryOperatorKind Opc) {
  CXXRecordDecl *RD = LHSTy->getAsCXXRecordDecl();
  if (!RD)
    return true;
  return RD->isHLSLBuiltinRecord() || Opc != BO_Assign;
}
 
// Walks though the global variable declaration, collects all resource binding
// requirements and adds them to Bindings
void SemaHLSL::collectResourceBindingsOnVarDecl(VarDecl *VD) {
  assert(VD->hasGlobalStorage() && VD->getType()->isHLSLIntangibleType() &&
         "expected global variable that contains HLSL resource");
 
  // Cbuffers and Tbuffers are HLSLBufferDecl types
  if (const HLSLBufferDecl *CBufferOrTBuffer = dyn_cast<HLSLBufferDecl>(VD)) {
    Bindings.addDeclBindingInfo(VD, CBufferOrTBuffer->isCBuffer()
                                        ? ResourceClass::CBuffer
                                        : ResourceClass::SRV);
    return;
  }
 
  // Unwrap arrays
  // FIXME: Calculate array size while unwrapping
  const Type *Ty = VD->getType()->getUnqualifiedDesugaredType();
  while (Ty->isArrayType()) {
    const ArrayType *AT = cast<ArrayType>(Ty);
    Ty = AT->getElementType()->getUnqualifiedDesugaredType();
  }
 
  // Resource (or array of resources)
  if (const HLSLAttributedResourceType *AttrResType =
          HLSLAttributedResourceType::findHandleTypeOnResource(Ty)) {
    Bindings.addDeclBindingInfo(VD, AttrResType->getAttrs().ResourceClass);
    return;
  }
 
  // User defined record type
  if (const RecordType *RT = dyn_cast<RecordType>(Ty))
    collectResourceBindingsOnUserRecordDecl(VD, RT);
}
 
// Walks though the explicit resource binding attributes on the declaration,
// and makes sure there is a resource that matched the binding and updates
// DeclBindingInfoLists
void SemaHLSL::processExplicitBindingsOnDecl(VarDecl *VD) {
  assert(VD->hasGlobalStorage() && "expected global variable");
 
  bool HasBinding = false;
  for (Attr *A : VD->attrs()) {
    if (isa<HLSLVkBindingAttr>(A)) {
      HasBinding = true;
      if (auto PA = VD->getAttr<HLSLVkPushConstantAttr>())
        Diag(PA->getLoc(), diag::err_hlsl_attr_incompatible) << A << PA;
    }
 
    HLSLResourceBindingAttr *RBA = dyn_cast<HLSLResourceBindingAttr>(A);
    if (!RBA || !RBA->hasRegisterSlot())
      continue;
    HasBinding = true;
 
    RegisterType RT = RBA->getRegisterType();
    assert(RT != RegisterType::I && "invalid or obsolete register type should "
                                    "never have an attribute created");
 
    if (RT == RegisterType::C) {
      if (Bindings.hasBindingInfoForDecl(VD))
        SemaRef.Diag(VD->getLocation(),
                     diag::warn_hlsl_user_defined_type_missing_member)
            << static_cast<int>(RT);
      continue;
    }
 
    // Find DeclBindingInfo for this binding and update it, or report error
    // if it does not exist (user type does to contain resources with the
    // expected resource class).
    ResourceClass RC = getResourceClass(RT);
    if (DeclBindingInfo *BI = Bindings.getDeclBindingInfo(VD, RC)) {
      // update binding info
      BI->setBindingAttribute(RBA, BindingType::Explicit);
    } else {
      SemaRef.Diag(VD->getLocation(),
                   diag::warn_hlsl_user_defined_type_missing_member)
          << static_cast<int>(RT);
    }
  }
 
  if (!HasBinding && isResourceRecordTypeOrArrayOf(VD))
    SemaRef.Diag(VD->getLocation(), diag::warn_hlsl_implicit_binding);
}
namespace {
class InitListTransformer {
  Sema &S;
  ASTContext &Ctx;
  QualType InitTy;
  QualType *DstIt = nullptr;
  Expr **ArgIt = nullptr;
  // Is wrapping the destination type iterator required? This is only used for
  // incomplete array types where we loop over the destination type since we
  // don't know the full number of elements from the declaration.
  bool Wrap;
 
  bool castInitializer(Expr *E) {
    assert(DstIt && "This should always be something!");
    if (DstIt == DestTypes.end()) {
      if (!Wrap) {
        ArgExprs.push_back(E);
        // This is odd, but it isn't technically a failure due to conversion, we
        // handle mismatched counts of arguments differently.
        return true;
      }
      DstIt = DestTypes.begin();
    }
    InitializedEntity Entity = InitializedEntity::InitializeParameter(
        Ctx, *DstIt, /* Consumed (ObjC) */ false);
    ExprResult Res = S.PerformCopyInitialization(Entity, E->getBeginLoc(), E);
    if (Res.isInvalid())
      return false;
    Expr *Init = Res.get();
    ArgExprs.push_back(Init);
    DstIt++;
    return true;
  }
 
  bool buildInitializerListImpl(Expr *E) {
    // If this is an initialization list, traverse the sub initializers.
    if (auto *Init = dyn_cast<InitListExpr>(E)) {
      for (auto *SubInit : Init->inits())
        if (!buildInitializerListImpl(SubInit))
          return false;
      return true;
    }
 
    // If this is a scalar type, just enqueue the expression.
    QualType Ty = E->getType().getDesugaredType(Ctx);
 
    if (Ty->isScalarType() || (Ty->isRecordType() && !Ty->isAggregateType()) ||
        Ty->isHLSLAttributedResourceType())
      return castInitializer(E);
 
    // If this is an aggregate type and a prvalue, create an xvalue temporary
    // so the member accesses will be xvalues. Wrap it in OpaqueExpr to make
    // sure codegen will not generate duplicate copies.
    if (E->isPRValue() && Ty->isAggregateType()) {
      ExprResult TmpExpr = S.TemporaryMaterializationConversion(E);
      if (TmpExpr.isInvalid())
        return false;
      E = TmpExpr.get();
      E = new (Ctx) OpaqueValueExpr(E->getBeginLoc(), E->getType(),
                                    E->getValueKind(), E->getObjectKind(), E);
    }
 
    if (auto *VecTy = Ty->getAs<VectorType>()) {
      uint64_t Size = VecTy->getNumElements();
 
      QualType SizeTy = Ctx.getSizeType();
      uint64_t SizeTySize = Ctx.getTypeSize(SizeTy);
      for (uint64_t I = 0; I < Size; ++I) {
        auto *Idx = IntegerLiteral::Create(Ctx, llvm::APInt(SizeTySize, I),
                                           SizeTy, SourceLocation());
 
        ExprResult ElExpr = S.CreateBuiltinArraySubscriptExpr(
            E, E->getBeginLoc(), Idx, E->getEndLoc());
        if (ElExpr.isInvalid())
          return false;
        if (!castInitializer(ElExpr.get()))
          return false;
      }
      return true;
    }
    if (auto *MTy = Ty->getAs<ConstantMatrixType>()) {
      unsigned Rows = MTy->getNumRows();
      unsigned Cols = MTy->getNumColumns();
      QualType ElemTy = MTy->getElementType();
 
      for (unsigned R = 0; R < Rows; ++R) {
        for (unsigned C = 0; C < Cols; ++C) {
          // row index literal
          Expr *RowIdx = IntegerLiteral::Create(
              Ctx, llvm::APInt(Ctx.getIntWidth(Ctx.IntTy), R), Ctx.IntTy,
              E->getBeginLoc());
          // column index literal
          Expr *ColIdx = IntegerLiteral::Create(
              Ctx, llvm::APInt(Ctx.getIntWidth(Ctx.IntTy), C), Ctx.IntTy,
              E->getBeginLoc());
          ExprResult ElExpr = S.CreateBuiltinMatrixSubscriptExpr(
              E, RowIdx, ColIdx, E->getEndLoc());
          if (ElExpr.isInvalid())
            return false;
          if (!castInitializer(ElExpr.get()))
            return false;
          ElExpr.get()->setType(ElemTy);
        }
      }
      return true;
    }
 
    if (auto *ArrTy = dyn_cast<ConstantArrayType>(Ty.getTypePtr())) {
      uint64_t Size = ArrTy->getZExtSize();
      QualType SizeTy = Ctx.getSizeType();
      uint64_t SizeTySize = Ctx.getTypeSize(SizeTy);
      for (uint64_t I = 0; I < Size; ++I) {
        auto *Idx = IntegerLiteral::Create(Ctx, llvm::APInt(SizeTySize, I),
                                           SizeTy, SourceLocation());
        ExprResult ElExpr = S.CreateBuiltinArraySubscriptExpr(
            E, E->getBeginLoc(), Idx, E->getEndLoc());
        if (ElExpr.isInvalid())
          return false;
        if (!buildInitializerListImpl(ElExpr.get()))
          return false;
      }
      return true;
    }
 
    if (auto *RD = Ty->getAsCXXRecordDecl()) {
      llvm::SmallVector<CXXRecordDecl *> RecordDecls;
      RecordDecls.push_back(RD);
      while (RecordDecls.back()->getNumBases()) {
        CXXRecordDecl *D = RecordDecls.back();
        assert(D->getNumBases() == 1 &&
               "HLSL doesn't support multiple inheritance");
        RecordDecls.push_back(
            D->bases_begin()->getType()->castAsCXXRecordDecl());
      }
      while (!RecordDecls.empty()) {
        CXXRecordDecl *RD = RecordDecls.pop_back_val();
        for (auto *FD : RD->fields()) {
          if (FD->isUnnamedBitField())
            continue;
          DeclAccessPair Found = DeclAccessPair::make(FD, FD->getAccess());
          DeclarationNameInfo NameInfo(FD->getDeclName(), E->getBeginLoc());
          ExprResult Res = S.BuildFieldReferenceExpr(
              E, false, E->getBeginLoc(), CXXScopeSpec(), FD, Found, NameInfo);
          if (Res.isInvalid())
            return false;
          if (!buildInitializerListImpl(Res.get()))
            return false;
        }
      }
    }
    return true;
  }
 
  Expr *generateInitListsImpl(QualType Ty) {
    Ty = Ty.getDesugaredType(Ctx);
    assert(ArgIt != ArgExprs.end() && "Something is off in iteration!");
    if (Ty->isScalarType() || (Ty->isRecordType() && !Ty->isAggregateType()) ||
        Ty->isHLSLAttributedResourceType())
      return *(ArgIt++);
 
    llvm::SmallVector<Expr *> Inits;
    if (Ty->isVectorType() || Ty->isConstantArrayType() ||
        Ty->isConstantMatrixType()) {
      QualType ElTy;
      uint64_t Size = 0;
      if (auto *ATy = Ty->getAs<VectorType>()) {
        ElTy = ATy->getElementType();
        Size = ATy->getNumElements();
      } else if (auto *CMTy = Ty->getAs<ConstantMatrixType>()) {
        ElTy = CMTy->getElementType();
        Size = CMTy->getNumElementsFlattened();
      } else {
        auto *VTy = cast<ConstantArrayType>(Ty.getTypePtr());
        ElTy = VTy->getElementType();
        Size = VTy->getZExtSize();
      }
      for (uint64_t I = 0; I < Size; ++I)
        Inits.push_back(generateInitListsImpl(ElTy));
    }
    if (auto *RD = Ty->getAsCXXRecordDecl()) {
      llvm::SmallVector<CXXRecordDecl *> RecordDecls;
      RecordDecls.push_back(RD);
      while (RecordDecls.back()->getNumBases()) {
        CXXRecordDecl *D = RecordDecls.back();
        assert(D->getNumBases() == 1 &&
               "HLSL doesn't support multiple inheritance");
        RecordDecls.push_back(
            D->bases_begin()->getType()->castAsCXXRecordDecl());
      }
      while (!RecordDecls.empty()) {
        CXXRecordDecl *RD = RecordDecls.pop_back_val();
        for (auto *FD : RD->fields())
          if (!FD->isUnnamedBitField())
            Inits.push_back(generateInitListsImpl(FD->getType()));
      }
    }
    auto *NewInit =
        new (Ctx) InitListExpr(Ctx, Inits.front()->getBeginLoc(), Inits,
                               Inits.back()->getEndLoc(), /*isExplicit=*/false);
    NewInit->setType(Ty);
    return NewInit;
  }
 
public:
  llvm::SmallVector<QualType, 16> DestTypes;
  llvm::SmallVector<Expr *, 16> ArgExprs;
  InitListTransformer(Sema &SemaRef, const InitializedEntity &Entity)
      : S(SemaRef), Ctx(SemaRef.getASTContext()),
        Wrap(Entity.getType()->isIncompleteArrayType()) {
    InitTy = Entity.getType().getNonReferenceType();
    // When we're generating initializer lists for incomplete array types we
    // need to wrap around both when building the initializers and when
    // generating the final initializer lists.
    if (Wrap) {
      assert(InitTy->isIncompleteArrayType());
      const IncompleteArrayType *IAT = Ctx.getAsIncompleteArrayType(InitTy);
      InitTy = IAT->getElementType();
    }
    BuildFlattenedTypeList(InitTy, DestTypes);
    DstIt = DestTypes.begin();
  }
 
  bool buildInitializerList(Expr *E) { return buildInitializerListImpl(E); }
 
  Expr *generateInitLists() {
    assert(!ArgExprs.empty() &&
           "Call buildInitializerList to generate argument expressions.");
    ArgIt = ArgExprs.begin();
    if (!Wrap)
      return generateInitListsImpl(InitTy);
    llvm::SmallVector<Expr *> Inits;
    while (ArgIt != ArgExprs.end())
      Inits.push_back(generateInitListsImpl(InitTy));
 
    auto *NewInit =
        new (Ctx) InitListExpr(Ctx, Inits.front()->getBeginLoc(), Inits,
                               Inits.back()->getEndLoc(), /*isExplicit=*/false);
    llvm::APInt ArySize(64, Inits.size());
    NewInit->setType(Ctx.getConstantArrayType(InitTy, ArySize, nullptr,
                                              ArraySizeModifier::Normal, 0));
    return NewInit;
  }
};
} // namespace
 
// Recursively detect any incomplete array anywhere in the type graph,
// including arrays, struct fields, and base classes.
static bool containsIncompleteArrayType(QualType Ty) {
  Ty = Ty.getCanonicalType();
 
  // Array types
  if (const ArrayType *AT = dyn_cast<ArrayType>(Ty)) {
    if (isa<IncompleteArrayType>(AT))
      return true;
    return containsIncompleteArrayType(AT->getElementType());
  }
 
  // Record (struct/class) types
  if (const auto *RT = Ty->getAs<RecordType>()) {
    const RecordDecl *RD = RT->getDecl();
 
    // Walk base classes (for C++ / HLSL structs with inheritance)
    if (const auto *CXXRD = dyn_cast<CXXRecordDecl>(RD)) {
      for (const CXXBaseSpecifier &Base : CXXRD->bases()) {
        if (containsIncompleteArrayType(Base.getType()))
          return true;
      }
    }
 
    // Walk fields
    for (const FieldDecl *F : RD->fields()) {
      if (containsIncompleteArrayType(F->getType()))
        return true;
    }
  }
 
  return false;
}
 
bool SemaHLSL::transformInitList(const InitializedEntity &Entity,
                                 InitListExpr *Init) {
  // If the initializer is a scalar, just return it.
  if (Init->getType()->isScalarType())
    return true;
  ASTContext &Ctx = SemaRef.getASTContext();
  InitListTransformer ILT(SemaRef, Entity);
 
  for (unsigned I = 0; I < Init->getNumInits(); ++I) {
    Expr *E = Init->getInit(I);
    if (E->HasSideEffects(Ctx)) {
      QualType Ty = E->getType();
      if (Ty->isRecordType())
        E = new (Ctx) MaterializeTemporaryExpr(Ty, E, E->isLValue());
      E = new (Ctx) OpaqueValueExpr(E->getBeginLoc(), Ty, E->getValueKind(),
                                    E->getObjectKind(), E);
      Init->setInit(I, E);
    }
    if (!ILT.buildInitializerList(E))
      return false;
  }
  size_t ExpectedSize = ILT.DestTypes.size();
  size_t ActualSize = ILT.ArgExprs.size();
  if (ExpectedSize == 0 && ActualSize == 0)
    return true;
 
  // Reject empty initializer if *any* incomplete array exists structurally
  if (ActualSize == 0 && containsIncompleteArrayType(Entity.getType())) {
    QualType InitTy = Entity.getType().getNonReferenceType();
    if (InitTy.hasAddressSpace())
      InitTy = SemaRef.getASTContext().removeAddrSpaceQualType(InitTy);
 
    SemaRef.Diag(Init->getBeginLoc(), diag::err_hlsl_incorrect_num_initializers)
        << /*TooManyOrFew=*/(int)(ExpectedSize < ActualSize) << InitTy
        << /*ExpectedSize=*/ExpectedSize << /*ActualSize=*/ActualSize;
    return false;
  }
 
  // We infer size after validating legality.
  // For incomplete arrays it is completely arbitrary to choose whether we think
  // the user intended fewer or more elements. This implementation assumes that
  // the user intended more, and errors that there are too few initializers to
  // complete the final element.
  if (Entity.getType()->isIncompleteArrayType()) {
    assert(ExpectedSize > 0 &&
           "The expected size of an incomplete array type must be at least 1.");
    ExpectedSize =
        ((ActualSize + ExpectedSize - 1) / ExpectedSize) * ExpectedSize;
  }
 
  // An initializer list might be attempting to initialize a reference or
  // rvalue-reference. When checking the initializer we should look through
  // the reference.
  QualType InitTy = Entity.getType().getNonReferenceType();
  if (InitTy.hasAddressSpace())
    InitTy = SemaRef.getASTContext().removeAddrSpaceQualType(InitTy);
  if (ExpectedSize != ActualSize) {
    int TooManyOrFew = ActualSize > ExpectedSize ? 1 : 0;
    SemaRef.Diag(Init->getBeginLoc(), diag::err_hlsl_incorrect_num_initializers)
        << TooManyOrFew << InitTy << ExpectedSize << ActualSize;
    return false;
  }
 
  // generateInitListsImpl will always return an InitListExpr here, because the
  // scalar case is handled above.
  auto *NewInit = cast<InitListExpr>(ILT.generateInitLists());
  Init->resizeInits(Ctx, NewInit->getNumInits());
  for (unsigned I = 0; I < NewInit->getNumInits(); ++I)
    Init->updateInit(Ctx, I, NewInit->getInit(I));
  return true;
}
 
static QualType ReportMatrixInvalidMember(Sema &S, StringRef Name,
                                          StringRef Expected,
                                          SourceLocation OpLoc,
                                          SourceLocation CompLoc) {
  S.Diag(OpLoc, diag::err_builtin_matrix_invalid_member)
      << Name << Expected << SourceRange(CompLoc);
  return QualType();
}
 
QualType SemaHLSL::checkMatrixComponent(Sema &S, QualType baseType,
                                        ExprValueKind &VK, SourceLocation OpLoc,
                                        const IdentifierInfo *CompName,
                                        SourceLocation CompLoc) {
  const auto *MT = baseType->castAs<ConstantMatrixType>();
  StringRef AccessorName = CompName->getName();
  assert(!AccessorName.empty() && "Matrix Accessor must have a name");
 
  unsigned Rows = MT->getNumRows();
  unsigned Cols = MT->getNumColumns();
  bool IsZeroBasedAccessor = false;
  unsigned ChunkLen = 0;
  if (AccessorName.size() < 2)
    return ReportMatrixInvalidMember(S, AccessorName,
                                     "length 4 for zero based: \'_mRC\' or "
                                     "length 3 for one-based: \'_RC\' accessor",
                                     OpLoc, CompLoc);
 
  if (AccessorName[0] == '_') {
    if (AccessorName[1] == 'm') {
      IsZeroBasedAccessor = true;
      ChunkLen = 4; // zero-based: "_mRC"
    } else {
      ChunkLen = 3; // one-based: "_RC"
    }
  } else
    return ReportMatrixInvalidMember(
        S, AccessorName, "zero based: \'_mRC\' or one-based: \'_RC\' accessor",
        OpLoc, CompLoc);
 
  if (AccessorName.size() % ChunkLen != 0) {
    const llvm::StringRef Expected = IsZeroBasedAccessor
                                         ? "zero based: '_mRC' accessor"
                                         : "one-based: '_RC' accessor";
 
    return ReportMatrixInvalidMember(S, AccessorName, Expected, OpLoc, CompLoc);
  }
 
  auto isDigit = [](char c) { return c >= '0' && c <= '9'; };
  auto isZeroBasedIndex = [](unsigned i) { return i <= 3; };
  auto isOneBasedIndex = [](unsigned i) { return i >= 1 && i <= 4; };
 
  bool HasRepeated = false;
  SmallVector<bool, 16> Seen(Rows * Cols, false);
  unsigned NumComponents = 0;
  const char *Begin = AccessorName.data();
 
  for (unsigned I = 0, E = AccessorName.size(); I < E; I += ChunkLen) {
    const char *Chunk = Begin + I;
    char RowChar = 0, ColChar = 0;
    if (IsZeroBasedAccessor) {
      // Zero-based: "_mRC"
      if (Chunk[0] != '_' || Chunk[1] != 'm') {
        char Bad = (Chunk[0] != '_') ? Chunk[0] : Chunk[1];
        return ReportMatrixInvalidMember(
            S, StringRef(&Bad, 1), "\'_m\' prefix",
            OpLoc.getLocWithOffset(I + (Bad == Chunk[0] ? 1 : 2)), CompLoc);
      }
      RowChar = Chunk[2];
      ColChar = Chunk[3];
    } else {
      // One-based: "_RC"
      if (Chunk[0] != '_')
        return ReportMatrixInvalidMember(
            S, StringRef(&Chunk[0], 1), "\'_\' prefix",
            OpLoc.getLocWithOffset(I + 1), CompLoc);
      RowChar = Chunk[1];
      ColChar = Chunk[2];
    }
 
    // Must be digits.
    bool IsDigitsError = false;
    if (!isDigit(RowChar)) {
      unsigned BadPos = IsZeroBasedAccessor ? 2 : 1;
      ReportMatrixInvalidMember(S, StringRef(&RowChar, 1), "row as integer",
                                OpLoc.getLocWithOffset(I + BadPos + 1),
                                CompLoc);
      IsDigitsError = true;
    }
 
    if (!isDigit(ColChar)) {
      unsigned BadPos = IsZeroBasedAccessor ? 3 : 2;
      ReportMatrixInvalidMember(S, StringRef(&ColChar, 1), "column as integer",
                                OpLoc.getLocWithOffset(I + BadPos + 1),
                                CompLoc);
      IsDigitsError = true;
    }
    if (IsDigitsError)
      return QualType();
 
    unsigned Row = RowChar - '0';
    unsigned Col = ColChar - '0';
 
    bool HasIndexingError = false;
    if (IsZeroBasedAccessor) {
      // 0-based [0..3]
      if (!isZeroBasedIndex(Row)) {
        S.Diag(OpLoc, diag::err_hlsl_matrix_element_not_in_bounds)
            << /*row*/ 0 << /*zero-based*/ 0 << SourceRange(CompLoc);
        HasIndexingError = true;
      }
      if (!isZeroBasedIndex(Col)) {
        S.Diag(OpLoc, diag::err_hlsl_matrix_element_not_in_bounds)
            << /*col*/ 1 << /*zero-based*/ 0 << SourceRange(CompLoc);
        HasIndexingError = true;
      }
    } else {
      // 1-based [1..4]
      if (!isOneBasedIndex(Row)) {
        S.Diag(OpLoc, diag::err_hlsl_matrix_element_not_in_bounds)
            << /*row*/ 0 << /*one-based*/ 1 << SourceRange(CompLoc);
        HasIndexingError = true;
      }
      if (!isOneBasedIndex(Col)) {
        S.Diag(OpLoc, diag::err_hlsl_matrix_element_not_in_bounds)
            << /*col*/ 1 << /*one-based*/ 1 << SourceRange(CompLoc);
        HasIndexingError = true;
      }
      // Convert to 0-based after range checking.
      --Row;
      --Col;
    }
 
    if (HasIndexingError)
      return QualType();
 
    // Note: matrix swizzle index is hard coded. That means Row and Col can
    // potentially be larger than Rows and Cols if matrix size is less than
    // the max index size.
    bool HasBoundsError = false;
    if (Row >= Rows) {
      Diag(OpLoc, diag::err_hlsl_matrix_index_out_of_bounds)
          << /*Row*/ 0 << Row << Rows << SourceRange(CompLoc);
      HasBoundsError = true;
    }
    if (Col >= Cols) {
      Diag(OpLoc, diag::err_hlsl_matrix_index_out_of_bounds)
          << /*Col*/ 1 << Col << Cols << SourceRange(CompLoc);
      HasBoundsError = true;
    }
    if (HasBoundsError)
      return QualType();
 
    unsigned FlatIndex = Row * Cols + Col;
    if (Seen[FlatIndex])
      HasRepeated = true;
    Seen[FlatIndex] = true;
    ++NumComponents;
  }
  if (NumComponents == 0 || NumComponents > 4) {
    S.Diag(OpLoc, diag::err_hlsl_matrix_swizzle_invalid_length)
        << NumComponents << SourceRange(CompLoc);
    return QualType();
  }
 
  QualType ElemTy = MT->getElementType();
  if (NumComponents == 1)
    return ElemTy;
  QualType VT = S.Context.getExtVectorType(ElemTy, NumComponents);
  if (HasRepeated)
    VK = VK_PRValue;
 
  for (Sema::ExtVectorDeclsType::iterator
           I = S.ExtVectorDecls.begin(S.getExternalSource()),
           E = S.ExtVectorDecls.end();
       I != E; ++I) {
    if ((*I)->getUnderlyingType() == VT)
      return S.Context.getTypedefType(ElaboratedTypeKeyword::None,
                                      /*Qualifier=*/std::nullopt, *I);
  }
 
  return VT;
}
 
bool SemaHLSL::handleInitialization(VarDecl *VDecl, Expr *&Init) {
  // If initializing a local resource, track the resource binding it is using
  if (VDecl->getType()->isHLSLResourceRecord() && !VDecl->hasGlobalStorage())
    trackLocalResource(VDecl, Init);
 
  const HLSLVkConstantIdAttr *ConstIdAttr =
      VDecl->getAttr<HLSLVkConstantIdAttr>();
  if (!ConstIdAttr)
    return true;
 
  ASTContext &Context = SemaRef.getASTContext();
 
  APValue InitValue;
  if (!Init->isCXX11ConstantExpr(Context, InitValue)) {
    Diag(VDecl->getLocation(), diag::err_specialization_const);
    VDecl->setInvalidDecl();
    return false;
  }
 
  Builtin::ID BID =
      getSpecConstBuiltinId(VDecl->getType()->getUnqualifiedDesugaredType());
 
  // Argument 1: The ID from the attribute
  int ConstantID = ConstIdAttr->getId();
  llvm::APInt IDVal(Context.getIntWidth(Context.IntTy), ConstantID);
  Expr *IdExpr = IntegerLiteral::Create(Context, IDVal, Context.IntTy,
                                        ConstIdAttr->getLocation());
 
  SmallVector<Expr *, 2> Args = {IdExpr, Init};
  Expr *C = SemaRef.BuildBuiltinCallExpr(Init->getExprLoc(), BID, Args);
  if (C->getType()->getCanonicalTypeUnqualified() !=
      VDecl->getType()->getCanonicalTypeUnqualified()) {
    C = SemaRef
            .BuildCStyleCastExpr(SourceLocation(),
                                 Context.getTrivialTypeSourceInfo(
                                     Init->getType(), Init->getExprLoc()),
                                 SourceLocation(), C)
            .get();
  }
  Init = C;
  return true;
}
 
QualType SemaHLSL::ActOnTemplateShorthand(TemplateDecl *Template,
                                          SourceLocation NameLoc) {
  if (!Template)
    return QualType();
 
  DeclContext *DC = Template->getDeclContext();
  if (!DC->isNamespace() || !cast<NamespaceDecl>(DC)->getIdentifier() ||
      cast<NamespaceDecl>(DC)->getName() != "hlsl")
    return QualType();
 
  TemplateParameterList *Params = Template->getTemplateParameters();
  if (!Params || Params->size() != 1)
    return QualType();
 
  if (!Template->isImplicit())
    return QualType();
 
  // We manually extract default arguments here instead of letting
  // CheckTemplateIdType handle it. This ensures that for resource types that
  // lack a default argument (like Buffer), we return a null QualType, which
  // triggers the "requires template arguments" error rather than a less
  // descriptive "too few template arguments" error.
  TemplateArgumentListInfo TemplateArgs(NameLoc, NameLoc);
  for (NamedDecl *P : *Params) {
    if (auto *TTP = dyn_cast<TemplateTypeParmDecl>(P)) {
      if (TTP->hasDefaultArgument()) {
        TemplateArgs.addArgument(TTP->getDefaultArgument());
        continue;
      }
    } else if (auto *NTTP = dyn_cast<NonTypeTemplateParmDecl>(P)) {
      if (NTTP->hasDefaultArgument()) {
        TemplateArgs.addArgument(NTTP->getDefaultArgument());
        continue;
      }
    } else if (auto *TTPD = dyn_cast<TemplateTemplateParmDecl>(P)) {
      if (TTPD->hasDefaultArgument()) {
        TemplateArgs.addArgument(TTPD->getDefaultArgument());
        continue;
      }
    }
    return QualType();
  }
 
  return SemaRef.CheckTemplateIdType(
      ElaboratedTypeKeyword::None, TemplateName(Template), NameLoc,
      TemplateArgs, nullptr, /*ForNestedNameSpecifier=*/false);
}