/Users/buildslave/jenkins/workspace/coverage/llvm-project/clang/lib/CodeGen/TargetInfo.cpp

Source (jump to first uncovered line)
//===---- TargetInfo.cpp - Encapsulate target details -----------*- C++ -*-===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// These classes wrap the information about a call or function
// definition used to handle ABI compliancy.
//
//===----------------------------------------------------------------------===//

#include "TargetInfo.h"
#include "ABIInfo.h"
#include "CGBlocks.h"
#include "CGCXXABI.h"
#include "CGValue.h"
#include "CodeGenFunction.h"
#include "clang/AST/Attr.h"
#include "clang/AST/RecordLayout.h"
#include "clang/Basic/CodeGenOptions.h"
#include "clang/Basic/DiagnosticFrontend.h"
#include "clang/CodeGen/CGFunctionInfo.h"
#include "clang/CodeGen/SwiftCallingConv.h"
#include "llvm/ADT/SmallBitVector.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/StringSwitch.h"
#include "llvm/ADT/Triple.h"
#include "llvm/ADT/Twine.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/IntrinsicsNVPTX.h"
#include "llvm/IR/Type.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm> // std::sort

using namespace clang;
using namespace CodeGen;

// Helper for coercing an aggregate argument or return value into an integer
// array of the same size (including padding) and alignment.  This alternate
// coercion happens only for the RenderScript ABI and can be removed after
// runtimes that rely on it are no longer supported.
//
// RenderScript assumes that the size of the argument / return value in the IR
// is the same as the size of the corresponding qualified type. This helper
// coerces the aggregate type into an array of the same size (including
// padding).  This coercion is used in lieu of expansion of struct members or
// other canonical coercions that return a coerced-type of larger size.
//
// Ty          - The argument / return value type
// Context     - The associated ASTContext
// LLVMContext - The associated LLVMContext
static ABIArgInfo coerceToIntArray(QualType Ty,
                                   ASTContext &Context,
                                   llvm::LLVMContext &LLVMContext) {
  // Alignment and Size are measured in bits.
  const uint64_t Size = Context.getTypeSize(Ty);
  const uint64_t Alignment = Context.getTypeAlign(Ty);
  llvm::Type *IntType = llvm::Type::getIntNTy(LLVMContext, Alignment);
  const uint64_t NumElements = (Size + Alignment - 1) / Alignment;
  return ABIArgInfo::getDirect(llvm::ArrayType::get(IntType, NumElements));
}

static void AssignToArrayRange(CodeGen::CGBuilderTy &Builder,
                               llvm::Value *Array,
                               llvm::Value *Value,
                               unsigned FirstIndex,
                               unsigned LastIndex) {
  // Alternatively, we could emit this as a loop in the source.
  for (unsigned I = FirstIndex; I <= LastIndex; ++I541) {
    llvm::Value *Cell =
        Builder.CreateConstInBoundsGEP1_32(Builder.getInt8Ty(), Array, I);
    Builder.CreateAlignedStore(Value, Cell, CharUnits::One());
  }
}

static bool isAggregateTypeForABI(QualType T) {
  return !CodeGenFunction::hasScalarEvaluationKind(T) ||
         T->isMemberFunctionPointerType();
}

ABIArgInfo ABIInfo::getNaturalAlignIndirect(QualType Ty, bool ByVal,
                                            bool Realign,
                                            llvm::Type *Padding) const {
  return ABIArgInfo::getIndirect(getContext().getTypeAlignInChars(Ty), ByVal,
                                 Realign, Padding);
}

ABIArgInfo
ABIInfo::getNaturalAlignIndirectInReg(QualType Ty, bool Realign) const {
  return ABIArgInfo::getIndirectInReg(getContext().getTypeAlignInChars(Ty),
                                      /*ByVal*/ false, Realign);
}

Address ABIInfo::EmitMSVAArg(CodeGenFunction &CGF, Address VAListAddr,
                             QualType Ty) const {
  return Address::invalid();
}

bool ABIInfo::isPromotableIntegerTypeForABI(QualType Ty) const {
  if (Ty->isPromotableIntegerType())
    return true;

  if (const auto *EIT = Ty->getAs<ExtIntType>())
    if (EIT->getNumBits() < getContext().getTypeSize(getContext().IntTy))
      return true;

  return false;
}

ABIInfo::~ABIInfo() {}

/// Does the given lowering require more than the given number of
/// registers when expanded?
///
/// This is intended to be the basis of a reasonable basic implementation
/// of should{Pass,Return}IndirectlyForSwift.
///
/// For most targets, a limit of four total registers is reasonable; this
/// limits the amount of code required in order to move around the value
/// in case it wasn't produced immediately prior to the call by the caller
/// (or wasn't produced in exactly the right registers) or isn't used
/// immediately within the callee.  But some targets may need to further
/// limit the register count due to an inability to support that many
/// return registers.
static bool occupiesMoreThan(CodeGenTypes &cgt,
                             ArrayRef<llvm::Type*> scalarTypes,
                             unsigned maxAllRegisters) {
  unsigned intCount = 0, fpCount = 0;
  for (llvm::Type *type : scalarTypes) {
    if (type->isPointerTy()) {
      intCount++;
    } else if (auto intTy = dyn_cast<llvm::IntegerType>(type)) {
      auto ptrWidth = cgt.getTarget().getPointerWidth(0);
      intCount += (intTy->getBitWidth() + ptrWidth - 1) / ptrWidth;
    } else {
      assert(type->isVectorTy() || type->isFloatingPointTy());
      fpCount++;
    }
  }

  return (intCount + fpCount > maxAllRegisters);
}

bool SwiftABIInfo::isLegalVectorTypeForSwift(CharUnits vectorSize,
                                             llvm::Type *eltTy,
                                             unsigned numElts) const {
  // The default implementation of this assumes that the target guarantees
  // 128-bit SIMD support but nothing more.
  return (vectorSize.getQuantity() > 8 && vectorSize.getQuantity() <= 16432);
}

static CGCXXABI::RecordArgABI getRecordArgABI(const RecordType *RT,
                                              CGCXXABI &CXXABI) {
  const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl());
  if (!RD) {
    if (!RT->getDecl()->canPassInRegisters())
      return CGCXXABI::RAA_Indirect;
    return CGCXXABI::RAA_Default;
  }
  return CXXABI.getRecordArgABI(RD);
}

static CGCXXABI::RecordArgABI getRecordArgABI(QualType T,
                                              CGCXXABI &CXXABI) {
  const RecordType *RT = T->getAs<RecordType>();
  if (!RT)
    return CGCXXABI::RAA_Default;
  return getRecordArgABI(RT, CXXABI);
}

static bool classifyReturnType(const CGCXXABI &CXXABI, CGFunctionInfo &FI,
                               const ABIInfo &Info) {
  QualType Ty = FI.getReturnType();

  if (const auto *RT = Ty->getAs<RecordType>())
    if (!isa<CXXRecordDecl>(RT->getDecl()) &&
        !RT->getDecl()->canPassInRegisters()) {
      FI.getReturnInfo() = Info.getNaturalAlignIndirect(Ty);
      return true;
    }

  return CXXABI.classifyReturnType(FI);
}

/// Pass transparent unions as if they were the type of the first element. Sema
/// should ensure that all elements of the union have the same "machine type".
static QualType useFirstFieldIfTransparentUnion(QualType Ty) {
  if (const RecordType *UT = Ty->getAsUnionType()) {
    const RecordDecl *UD = UT->getDecl();
    if (UD->hasAttr<TransparentUnionAttr>()) {
      assert(!UD->field_empty() && "sema created an empty transparent union");
      return UD->field_begin()->getType();
    }
  }
  return Ty;
}

CGCXXABI &ABIInfo::getCXXABI() const {
  return CGT.getCXXABI();
}

ASTContext &ABIInfo::getContext() const {
  return CGT.getContext();
}

llvm::LLVMContext &ABIInfo::getVMContext() const {
  return CGT.getLLVMContext();
}

const llvm::DataLayout &ABIInfo::getDataLayout() const {
  return CGT.getDataLayout();
}

const TargetInfo &ABIInfo::getTarget() const {
  return CGT.getTarget();
}

const CodeGenOptions &ABIInfo::getCodeGenOpts() const {
  return CGT.getCodeGenOpts();
}

bool ABIInfo::isAndroid() const { return getTarget().getTriple().isAndroid(); }

bool ABIInfo::isHomogeneousAggregateBaseType(QualType Ty) const {
  return false;
}

bool ABIInfo::isHomogeneousAggregateSmallEnough(const Type *Base,
                                                uint64_t Members) const {
  return false;
}

LLVM_DUMP_METHOD void ABIArgInfo::dump() const {
  raw_ostream &OS = llvm::errs();
  OS << "(ABIArgInfo Kind=";
  switch (TheKind) {
  case Direct:
    OS << "Direct Type=";
    if (llvm::Type *Ty = getCoerceToType())
      Ty->print(OS);
    else
      OS << "null";
    break;
  case Extend:
    OS << "Extend";
    break;
  case Ignore:
    OS << "Ignore";
    break;
  case InAlloca:
    OS << "InAlloca Offset=" << getInAllocaFieldIndex();
    break;
  case Indirect:
    OS << "Indirect Align=" << getIndirectAlign().getQuantity()
       << " ByVal=" << getIndirectByVal()
       << " Realign=" << getIndirectRealign();
    break;
  case IndirectAliased:
    OS << "Indirect Align=" << getIndirectAlign().getQuantity()
       << " AadrSpace=" << getIndirectAddrSpace()
       << " Realign=" << getIndirectRealign();
    break;
  case Expand:
    OS << "Expand";
    break;
  case CoerceAndExpand:
    OS << "CoerceAndExpand Type=";
    getCoerceAndExpandType()->print(OS);
    break;
  }
  OS << ")\n";
}

// Dynamically round a pointer up to a multiple of the given alignment.
static llvm::Value *emitRoundPointerUpToAlignment(CodeGenFunction &CGF,
                                                  llvm::Value *Ptr,
                                                  CharUnits Align) {
  llvm::Value *PtrAsInt = Ptr;
  // OverflowArgArea = (OverflowArgArea + Align - 1) & -Align;
  PtrAsInt = CGF.Builder.CreatePtrToInt(PtrAsInt, CGF.IntPtrTy);
  PtrAsInt = CGF.Builder.CreateAdd(PtrAsInt,
        llvm::ConstantInt::get(CGF.IntPtrTy, Align.getQuantity() - 1));
  PtrAsInt = CGF.Builder.CreateAnd(PtrAsInt,
           llvm::ConstantInt::get(CGF.IntPtrTy, -Align.getQuantity()));
  PtrAsInt = CGF.Builder.CreateIntToPtr(PtrAsInt,
                                        Ptr->getType(),
                                        Ptr->getName() + ".aligned");
  return PtrAsInt;
}

/// Emit va_arg for a platform using the common void* representation,
/// where arguments are simply emitted in an array of slots on the stack.
///
/// This version implements the core direct-value passing rules.
///
/// \param SlotSize - The size and alignment of a stack slot.
///   Each argument will be allocated to a multiple of this number of
///   slots, and all the slots will be aligned to this value.
/// \param AllowHigherAlign - The slot alignment is not a cap;
///   an argument type with an alignment greater than the slot size
///   will be emitted on a higher-alignment address, potentially
///   leaving one or more empty slots behind as padding.  If this
///   is false, the returned address might be less-aligned than
///   DirectAlign.
static Address emitVoidPtrDirectVAArg(CodeGenFunction &CGF,
                                      Address VAListAddr,
                                      llvm::Type *DirectTy,
                                      CharUnits DirectSize,
                                      CharUnits DirectAlign,
                                      CharUnits SlotSize,
                                      bool AllowHigherAlign) {
  // Cast the element type to i8* if necessary.  Some platforms define
  // va_list as a struct containing an i8* instead of just an i8*.
  if (VAListAddr.getElementType() != CGF.Int8PtrTy)
    VAListAddr = CGF.Builder.CreateElementBitCast(VAListAddr, CGF.Int8PtrTy);

  llvm::Value *Ptr = CGF.Builder.CreateLoad(VAListAddr, "argp.cur");

  // If the CC aligns values higher than the slot size, do so if needed.
  Address Addr = Address::invalid();
  if (AllowHigherAlign && DirectAlign > SlotSize245) {
    Addr = Address(emitRoundPointerUpToAlignment(CGF, Ptr, DirectAlign),
                                                 DirectAlign);
  } else {
    Addr = Address(Ptr, SlotSize);
  }

  // Advance the pointer past the argument, then store that back.
  CharUnits FullDirectSize = DirectSize.alignTo(SlotSize);
  Address NextPtr =
      CGF.Builder.CreateConstInBoundsByteGEP(Addr, FullDirectSize, "argp.next");
  CGF.Builder.CreateStore(NextPtr.getPointer(), VAListAddr);

  // If the argument is smaller than a slot, and this is a big-endian
  // target, the argument will be right-adjusted in its slot.
  if (DirectSize < SlotSize && CGF.CGM.getDataLayout().isBigEndian()37 &&
      !DirectTy->isStructTy()) {
    Addr = CGF.Builder.CreateConstInBoundsByteGEP(Addr, SlotSize - DirectSize);
  }

  Addr = CGF.Builder.CreateElementBitCast(Addr, DirectTy);
  return Addr;
}

/// Emit va_arg for a platform using the common void* representation,
/// where arguments are simply emitted in an array of slots on the stack.
///
/// \param IsIndirect - Values of this type are passed indirectly.
/// \param ValueInfo - The size and alignment of this type, generally
///   computed with getContext().getTypeInfoInChars(ValueTy).
/// \param SlotSizeAndAlign - The size and alignment of a stack slot.
///   Each argument will be allocated to a multiple of this number of
///   slots, and all the slots will be aligned to this value.
/// \param AllowHigherAlign - The slot alignment is not a cap;
///   an argument type with an alignment greater than the slot size
///   will be emitted on a higher-alignment address, potentially
///   leaving one or more empty slots behind as padding.
static Address emitVoidPtrVAArg(CodeGenFunction &CGF, Address VAListAddr,
                                QualType ValueTy, bool IsIndirect,
                                TypeInfoChars ValueInfo,
                                CharUnits SlotSizeAndAlign,
                                bool AllowHigherAlign) {
  // The size and alignment of the value that was passed directly.
  CharUnits DirectSize, DirectAlign;
  if (IsIndirect) {
    DirectSize = CGF.getPointerSize();
    DirectAlign = CGF.getPointerAlign();
  } else {
    DirectSize = ValueInfo.Width;
    DirectAlign = ValueInfo.Align;
  }

  // Cast the address we've calculated to the right type.
  llvm::Type *DirectTy = CGF.ConvertTypeForMem(ValueTy);
  if (IsIndirect)
    DirectTy = DirectTy->getPointerTo(0);

  Address Addr = emitVoidPtrDirectVAArg(CGF, VAListAddr, DirectTy,
                                        DirectSize, DirectAlign,
                                        SlotSizeAndAlign,
                                        AllowHigherAlign);

  if (IsIndirect) {
    Addr = Address(CGF.Builder.CreateLoad(Addr), ValueInfo.Align);
  }

  return Addr;

}

static Address emitMergePHI(CodeGenFunction &CGF,
                            Address Addr1, llvm::BasicBlock *Block1,
                            Address Addr2, llvm::BasicBlock *Block2,
                            const llvm::Twine &Name = "") {
  assert(Addr1.getType() == Addr2.getType());
  llvm::PHINode *PHI = CGF.Builder.CreatePHI(Addr1.getType(), 2, Name);
  PHI->addIncoming(Addr1.getPointer(), Block1);
  PHI->addIncoming(Addr2.getPointer(), Block2);
  CharUnits Align = std::min(Addr1.getAlignment(), Addr2.getAlignment());
  return Address(PHI, Align);
}

TargetCodeGenInfo::~TargetCodeGenInfo() = default;

// If someone can figure out a general rule for this, that would be great.
// It's probably just doomed to be platform-dependent, though.
unsigned TargetCodeGenInfo::getSizeOfUnwindException() const {
  // Verified for:
  //   x86-64     FreeBSD, Linux, Darwin
  //   x86-32     FreeBSD, Linux, Darwin
  //   PowerPC    Linux, Darwin
  //   ARM        Darwin (*not* EABI)
  //   AArch64    Linux
  return 32;
}

bool TargetCodeGenInfo::isNoProtoCallVariadic(const CallArgList &args,
                                     const FunctionNoProtoType *fnType) const {
  // The following conventions are known to require this to be false:
  //   x86_stdcall
  //   MIPS
  // For everything else, we just prefer false unless we opt out.
  return false;
}

void
TargetCodeGenInfo::getDependentLibraryOption(llvm::StringRef Lib,
                                             llvm::SmallString<24> &Opt) const {
  // This assumes the user is passing a library name like "rt" instead of a
  // filename like "librt.a/so", and that they don't care whether it's static or
  // dynamic.
  Opt = "-l";
  Opt += Lib;
}

unsigned TargetCodeGenInfo::getOpenCLKernelCallingConv() const {
  // OpenCL kernels are called via an explicit runtime API with arguments
  // set with clSetKernelArg(), not as normal sub-functions.
  // Return SPIR_KERNEL by default as the kernel calling convention to
  // ensure the fingerprint is fixed such way that each OpenCL argument
  // gets one matching argument in the produced kernel function argument
  // list to enable feasible implementation of clSetKernelArg() with
  // aggregates etc. In case we would use the default C calling conv here,
  // clSetKernelArg() might break depending on the target-specific
  // conventions; different targets might split structs passed as values
  // to multiple function arguments etc.
  return llvm::CallingConv::SPIR_KERNEL;
}

llvm::Constant *TargetCodeGenInfo::getNullPointer(const CodeGen::CodeGenModule &CGM,
    llvm::PointerType *T, QualType QT) const {
  return llvm::ConstantPointerNull::get(T);
}

LangAS TargetCodeGenInfo::getGlobalVarAddressSpace(CodeGenModule &CGM,
                                                   const VarDecl *D) const {
  assert(!CGM.getLangOpts().OpenCL &&
         !(CGM.getLangOpts().CUDA && CGM.getLangOpts().CUDAIsDevice) &&
         "Address space agnostic languages only");
  return D ? D->getType().getAddressSpace() : LangAS::Default1.94k;
}

llvm::Value *TargetCodeGenInfo::performAddrSpaceCast(
    CodeGen::CodeGenFunction &CGF, llvm::Value *Src, LangAS SrcAddr,
    LangAS DestAddr, llvm::Type *DestTy, bool isNonNull) const {
  // Since target may map different address spaces in AST to the same address
  // space, an address space conversion may end up as a bitcast.
  if (auto *C = dyn_cast<llvm::Constant>(Src))
    return performAddrSpaceCast(CGF.CGM, C, SrcAddr, DestAddr, DestTy);
  // Try to preserve the source's name to make IR more readable.
  return CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(
      Src, DestTy, Src->hasName() ? Src->getName() + ".ascast" : ""149);
}

llvm::Constant *
TargetCodeGenInfo::performAddrSpaceCast(CodeGenModule &CGM, llvm::Constant *Src,
                                        LangAS SrcAddr, LangAS DestAddr,
                                        llvm::Type *DestTy) const {
  // Since target may map different address spaces in AST to the same address
  // space, an address space conversion may end up as a bitcast.
  return llvm::ConstantExpr::getPointerCast(Src, DestTy);
}

llvm::SyncScope::ID
TargetCodeGenInfo::getLLVMSyncScopeID(const LangOptions &LangOpts,
                                      SyncScope Scope,
                                      llvm::AtomicOrdering Ordering,
                                      llvm::LLVMContext &Ctx) const {
  return Ctx.getOrInsertSyncScopeID(""); /* default sync scope */
}

static bool isEmptyRecord(ASTContext &Context, QualType T, bool AllowArrays);

/// isEmptyField - Return true iff a the field is "empty", that is it
/// is an unnamed bit-field or an (array of) empty record(s).
static bool isEmptyField(ASTContext &Context, const FieldDecl *FD,
                         bool AllowArrays) {
  if (FD->isUnnamedBitfield())
    return true;

  QualType FT = FD->getType();

  // Constant arrays of empty records count as empty, strip them off.
  // Constant arrays of zero length always count as empty.
  bool WasArray = false;
  if (AllowArrays)
    while (const ConstantArrayType *5.15kAT = Context.getAsConstantArrayType(FT)) {
      if (AT->getSize() == 0)
        return true;
      FT = AT->getElementType();
      // The [[no_unique_address]] special case below does not apply to
      // arrays of C++ empty records, so we need to remember this fact.
      WasArray = true;
    }

  const RecordType *RT = FT->getAs<RecordType>();
  if (!RT)
    return false;

  // C++ record fields are never empty, at least in the Itanium ABI.
  //
  // FIXME: We should use a predicate for whether this behavior is true in the
  // current ABI.
  //
  // The exception to the above rule are fields marked with the
  // [[no_unique_address]] attribute (since C++20).  Those do count as empty
  // according to the Itanium ABI.  The exception applies only to records,
  // not arrays of records, so we must also check whether we stripped off an
  // array type above.
  if (isa<CXXRecordDecl>(RT->getDecl()) &&
      (WasArray || !FD->hasAttr<NoUniqueAddressAttr>()68))
    return false;

  return isEmptyRecord(Context, FT, AllowArrays);
}

/// isEmptyRecord - Return true iff a structure contains only empty
/// fields. Note that a structure with a flexible array member is not
/// considered empty.
static bool isEmptyRecord(ASTContext &Context, QualType T, bool AllowArrays) {
  const RecordType *RT = T->getAs<RecordType>();
  if (!RT)
    return false;
  const RecordDecl *RD = RT->getDecl();
  if (RD->hasFlexibleArrayMember())
    return false;

  // If this is a C++ record, check the bases first.
  if (const CXXRecordDecl *CXXRD = dyn_cast<CXXRecordDecl>(RD))
    for (const auto &I : CXXRD->bases())
      if (!isEmptyRecord(Context, I.getType(), true))
        return false;

  for (const auto *I : RD->fields())
    if (!isEmptyField(Context, I, AllowArrays))
      return false;
  return true;
}

/// isSingleElementStruct - Determine if a structure is a "single
/// element struct", i.e. it has exactly one non-empty field or
/// exactly one field which is itself a single element
/// struct. Structures with flexible array members are never
/// considered single element structs.
///
/// \return The field declaration for the single non-empty field, if
/// it exists.
static const Type *isSingleElementStruct(QualType T, ASTContext &Context) {
  const RecordType *RT = T->getAs<RecordType>();
  if (!RT)
    return nullptr;

  const RecordDecl *RD = RT->getDecl();
  if (RD->hasFlexibleArrayMember())
    return nullptr;

  const Type *Found = nullptr;

  // If this is a C++ record, check the bases first.
  if (const CXXRecordDecl *CXXRD = dyn_cast<CXXRecordDecl>(RD)) {
    for (const auto &I : CXXRD->bases()) {
      // Ignore empty records.
      if (isEmptyRecord(Context, I.getType(), true))
        continue;

      // If we already found an element then this isn't a single-element struct.
      if (Found)
        return nullptr;

      // If this is non-empty and not a single element struct, the composite
      // cannot be a single element struct.
      Found = isSingleElementStruct(I.getType(), Context);
      if (!Found)
        return nullptr;
    }
  }

  // Check for single element.
  for (const auto *FD : RD->fields())812 {
    QualType FT = FD->getType();

    // Ignore empty fields.
    if (isEmptyField(Context, FD, true))
      continue;

    // If we already found an element then this isn't a single-element
    // struct.
    if (Found)
      return nullptr;

    // Treat single element arrays as the element.
    while (const ConstantArrayType *778AT = Context.getAsConstantArrayType(FT)) {
      if (AT->getSize().getZExtValue() != 1)
        break;
      FT = AT->getElementType();
    }

    if (!isAggregateTypeForABI(FT)) {
      Found = FT.getTypePtr();
    } else {
      Found = isSingleElementStruct(FT, Context);
      if (!Found)
        return nullptr;
    }
  }

  // We don't consider a struct a single-element struct if it has
  // padding beyond the element type.
  if (Found && Context.getTypeSize(Found) != Context.getTypeSize(T)263)
    return nullptr;

  return Found;
}

namespace {
Address EmitVAArgInstr(CodeGenFunction &CGF, Address VAListAddr, QualType Ty,
                       const ABIArgInfo &AI) {
  // This default implementation defers to the llvm backend's va_arg
  // instruction. It can handle only passing arguments directly
  // (typically only handled in the backend for primitive types), or
  // aggregates passed indirectly by pointer (NOTE: if the "byval"
  // flag has ABI impact in the callee, this implementation cannot
  // work.)

  // Only a few cases are covered here at the moment -- those needed
  // by the default abi.
  llvm::Value *Val;

  if (AI.isIndirect()) {
    assert(!AI.getPaddingType() &&
           "Unexpected PaddingType seen in arginfo in generic VAArg emitter!");
    assert(
        !AI.getIndirectRealign() &&
        "Unexpected IndirectRealign seen in arginfo in generic VAArg emitter!");

    auto TyInfo = CGF.getContext().getTypeInfoInChars(Ty);
    CharUnits TyAlignForABI = TyInfo.Align;

    llvm::Type *BaseTy =
        llvm::PointerType::getUnqual(CGF.ConvertTypeForMem(Ty));
    llvm::Value *Addr =
        CGF.Builder.CreateVAArg(VAListAddr.getPointer(), BaseTy);
    return Address(Addr, TyAlignForABI);
  } else {
    assert((AI.isDirect() || AI.isExtend()) &&
           "Unexpected ArgInfo Kind in generic VAArg emitter!");

    assert(!AI.getInReg() &&
           "Unexpected InReg seen in arginfo in generic VAArg emitter!");
    assert(!AI.getPaddingType() &&
           "Unexpected PaddingType seen in arginfo in generic VAArg emitter!");
    assert(!AI.getDirectOffset() &&
           "Unexpected DirectOffset seen in arginfo in generic VAArg emitter!");
    assert(!AI.getCoerceToType() &&
           "Unexpected CoerceToType seen in arginfo in generic VAArg emitter!");

    Address Temp = CGF.CreateMemTemp(Ty, "varet");
    Val = CGF.Builder.CreateVAArg(VAListAddr.getPointer(), CGF.ConvertType(Ty));
    CGF.Builder.CreateStore(Val, Temp);
    return Temp;
  }
}

/// DefaultABIInfo - The default implementation for ABI specific
/// details. This implementation provides information which results in
/// self-consistent and sensible LLVM IR generation, but does not
/// conform to any particular ABI.
class DefaultABIInfo : public ABIInfo {
public:
  DefaultABIInfo(CodeGen::CodeGenTypes &CGT) : ABIInfo(CGT) {}

  ABIArgInfo classifyReturnType(QualType RetTy) const;
  ABIArgInfo classifyArgumentType(QualType RetTy) const;

  void computeInfo(CGFunctionInfo &FI) const override {
    if (!getCXXABI().classifyReturnType(FI))
      FI.getReturnInfo() = classifyReturnType(FI.getReturnType());
    for (auto &I : FI.arguments())
      I.info = classifyArgumentType(I.type);
  }

  Address EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                    QualType Ty) const override {
    return EmitVAArgInstr(CGF, VAListAddr, Ty, classifyArgumentType(Ty));
  }
};

class DefaultTargetCodeGenInfo : public TargetCodeGenInfo {
public:
  DefaultTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT)
      : TargetCodeGenInfo(std::make_unique<DefaultABIInfo>(CGT)) {}
};

ABIArgInfo DefaultABIInfo::classifyArgumentType(QualType Ty) const {
  Ty = useFirstFieldIfTransparentUnion(Ty);

  if (isAggregateTypeForABI(Ty)) {
    // Records with non-trivial destructors/copy-constructors should not be
    // passed by value.
    if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI()))
      return getNaturalAlignIndirect(Ty, RAA == CGCXXABI::RAA_DirectInMemory);

    return getNaturalAlignIndirect(Ty);
  }

  // Treat an enum type as its underlying type.
  if (const EnumType *EnumTy = Ty->getAs<EnumType>())
    Ty = EnumTy->getDecl()->getIntegerType();

  ASTContext &Context = getContext();
  if (const auto *EIT = Ty->getAs<ExtIntType>())
    if (EIT->getNumBits() >
        Context.getTypeSize(Context.getTargetInfo().hasInt128Type()
                                ? Context.Int128Ty
                                : Context.LongLongTy))
      return getNaturalAlignIndirect(Ty);

  return (isPromotableIntegerTypeForABI(Ty) ? ABIArgInfo::getExtend(Ty)142
                                            : ABIArgInfo::getDirect());
}

ABIArgInfo DefaultABIInfo::classifyReturnType(QualType RetTy) const {
  if (RetTy->isVoidType())
    return ABIArgInfo::getIgnore();

  if (isAggregateTypeForABI(RetTy))
    return getNaturalAlignIndirect(RetTy);

  // Treat an enum type as its underlying type.
  if (const EnumType *EnumTy = RetTy->getAs<EnumType>())
    RetTy = EnumTy->getDecl()->getIntegerType();

  if (const auto *EIT = RetTy->getAs<ExtIntType>())
    if (EIT->getNumBits() >
        getContext().getTypeSize(getContext().getTargetInfo().hasInt128Type()
                                     ? getContext().Int128Ty
                                     : getContext().LongLongTy))
      return getNaturalAlignIndirect(RetTy);

  return (isPromotableIntegerTypeForABI(RetTy) ? ABIArgInfo::getExtend(RetTy)79
                                               : ABIArgInfo::getDirect());
}

//===----------------------------------------------------------------------===//
// WebAssembly ABI Implementation
//
// This is a very simple ABI that relies a lot on DefaultABIInfo.
//===----------------------------------------------------------------------===//

class WebAssemblyABIInfo final : public SwiftABIInfo {
public:
  enum ABIKind {
    MVP = 0,
    ExperimentalMV = 1,
  };

private:
  DefaultABIInfo defaultInfo;
  ABIKind Kind;

public:
  explicit WebAssemblyABIInfo(CodeGen::CodeGenTypes &CGT, ABIKind Kind)
      : SwiftABIInfo(CGT), defaultInfo(CGT), Kind(Kind) {}

private:
  ABIArgInfo classifyReturnType(QualType RetTy) const;
  ABIArgInfo classifyArgumentType(QualType Ty) const;

  // DefaultABIInfo's classifyReturnType and classifyArgumentType are
  // non-virtual, but computeInfo and EmitVAArg are virtual, so we
  // overload them.
  void computeInfo(CGFunctionInfo &FI) const override {
    if (!getCXXABI().classifyReturnType(FI))
      FI.getReturnInfo() = classifyReturnType(FI.getReturnType());
    for (auto &Arg : FI.arguments())
      Arg.info = classifyArgumentType(Arg.type);
  }

  Address EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                    QualType Ty) const override;

  bool shouldPassIndirectlyForSwift(ArrayRef<llvm::Type*> scalars,
                                    bool asReturnValue) const override {
    return occupiesMoreThan(CGT, scalars, /*total*/ 4);
  }

  bool isSwiftErrorInRegister() const override {
    return false;
  }
};

class WebAssemblyTargetCodeGenInfo final : public TargetCodeGenInfo {
public:
  explicit WebAssemblyTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT,
                                        WebAssemblyABIInfo::ABIKind K)
      : TargetCodeGenInfo(std::make_unique<WebAssemblyABIInfo>(CGT, K)) {}

  void setTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
                           CodeGen::CodeGenModule &CGM) const override {
    TargetCodeGenInfo::setTargetAttributes(D, GV, CGM);
    if (const auto *FD = dyn_cast_or_null<FunctionDecl>(D)) {
      if (const auto *Attr = FD->getAttr<WebAssemblyImportModuleAttr>()) {
        llvm::Function *Fn = cast<llvm::Function>(GV);
        llvm::AttrBuilder B;
        B.addAttribute("wasm-import-module", Attr->getImportModule());
        Fn->addAttributes(llvm::AttributeList::FunctionIndex, B);
      }
      if (const auto *Attr = FD->getAttr<WebAssemblyImportNameAttr>()) {
        llvm::Function *Fn = cast<llvm::Function>(GV);
        llvm::AttrBuilder B;
        B.addAttribute("wasm-import-name", Attr->getImportName());
        Fn->addAttributes(llvm::AttributeList::FunctionIndex, B);
      }
      if (const auto *Attr = FD->getAttr<WebAssemblyExportNameAttr>()) {
        llvm::Function *Fn = cast<llvm::Function>(GV);
        llvm::AttrBuilder B;
        B.addAttribute("wasm-export-name", Attr->getExportName());
        Fn->addAttributes(llvm::AttributeList::FunctionIndex, B);
      }
    }

    if (auto *FD = dyn_cast_or_null<FunctionDecl>(D)) {
      llvm::Function *Fn = cast<llvm::Function>(GV);
      if (!FD->doesThisDeclarationHaveABody() && !FD->hasPrototype()73)
        Fn->addFnAttr("no-prototype");
    }
  }
};

/// Classify argument of given type \p Ty.
ABIArgInfo WebAssemblyABIInfo::classifyArgumentType(QualType Ty) const {
  Ty = useFirstFieldIfTransparentUnion(Ty);

  if (isAggregateTypeForABI(Ty)) {
    // Records with non-trivial destructors/copy-constructors should not be
    // passed by value.
    if (auto RAA = getRecordArgABI(Ty, getCXXABI()))
      return getNaturalAlignIndirect(Ty, RAA == CGCXXABI::RAA_DirectInMemory);
    // Ignore empty structs/unions.
    if (isEmptyRecord(getContext(), Ty, true))
      return ABIArgInfo::getIgnore();
    // Lower single-element structs to just pass a regular value. TODO: We
    // could do reasonable-size multiple-element structs too, using getExpand(),
    // though watch out for things like bitfields.
    if (const Type *SeltTy = isSingleElementStruct(Ty, getContext()))
      return ABIArgInfo::getDirect(CGT.ConvertType(QualType(SeltTy, 0)));
    // For the experimental multivalue ABI, fully expand all other aggregates
    if (Kind == ABIKind::ExperimentalMV) {
      const RecordType *RT = Ty->getAs<RecordType>();
      assert(RT);
      bool HasBitField = false;
      for (auto *Field : RT->getDecl()->fields()) {
        if (Field->isBitField()) {
          HasBitField = true;
          break;
        }
      }
      if (!HasBitField)
        return ABIArgInfo::getExpand();
    }
  }

  // Otherwise just do the default thing.
  return defaultInfo.classifyArgumentType(Ty);
}

ABIArgInfo WebAssemblyABIInfo::classifyReturnType(QualType RetTy) const {
  if (isAggregateTypeForABI(RetTy)) {
    // Records with non-trivial destructors/copy-constructors should not be
    // returned by value.
    if (!getRecordArgABI(RetTy, getCXXABI())) {
      // Ignore empty structs/unions.
      if (isEmptyRecord(getContext(), RetTy, true))
        return ABIArgInfo::getIgnore();
      // Lower single-element structs to just return a regular value. TODO: We
      // could do reasonable-size multiple-element structs too, using
      // ABIArgInfo::getDirect().
      if (const Type *SeltTy = isSingleElementStruct(RetTy, getContext()))
        return ABIArgInfo::getDirect(CGT.ConvertType(QualType(SeltTy, 0)));
      // For the experimental multivalue ABI, return all other aggregates
      if (Kind == ABIKind::ExperimentalMV)
        return ABIArgInfo::getDirect();
    }
  }

  // Otherwise just do the default thing.
  return defaultInfo.classifyReturnType(RetTy);
}

Address WebAssemblyABIInfo::EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                                      QualType Ty) const {
  bool IsIndirect = isAggregateTypeForABI(Ty) &&
                    !isEmptyRecord(getContext(), Ty, true) &&
                    !isSingleElementStruct(Ty, getContext());
  return emitVoidPtrVAArg(CGF, VAListAddr, Ty, IsIndirect,
                          getContext().getTypeInfoInChars(Ty),
                          CharUnits::fromQuantity(4),
                          /*AllowHigherAlign=*/true);
}

//===----------------------------------------------------------------------===//
// le32/PNaCl bitcode ABI Implementation
//
// This is a simplified version of the x86_32 ABI.  Arguments and return values
// are always passed on the stack.
//===----------------------------------------------------------------------===//

class PNaClABIInfo : public ABIInfo {
 public:
  PNaClABIInfo(CodeGen::CodeGenTypes &CGT) : ABIInfo(CGT) {}

  ABIArgInfo classifyReturnType(QualType RetTy) const;
  ABIArgInfo classifyArgumentType(QualType RetTy) const;

  void computeInfo(CGFunctionInfo &FI) const override;
  Address EmitVAArg(CodeGenFunction &CGF,
                    Address VAListAddr, QualType Ty) const override;
};

class PNaClTargetCodeGenInfo : public TargetCodeGenInfo {
 public:
   PNaClTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT)
       : TargetCodeGenInfo(std::make_unique<PNaClABIInfo>(CGT)) {}
};

void PNaClABIInfo::computeInfo(CGFunctionInfo &FI) const {
  if (!getCXXABI().classifyReturnType(FI))
    FI.getReturnInfo() = classifyReturnType(FI.getReturnType());

  for (auto &I : FI.arguments())
    I.info = classifyArgumentType(I.type);
}

Address PNaClABIInfo::EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                                QualType Ty) const {
  // The PNaCL ABI is a bit odd, in that varargs don't use normal
  // function classification. Structs get passed directly for varargs
  // functions, through a rewriting transform in
  // pnacl-llvm/lib/Transforms/NaCl/ExpandVarArgs.cpp, which allows
  // this target to actually support a va_arg instructions with an
  // aggregate type, unlike other targets.
  return EmitVAArgInstr(CGF, VAListAddr, Ty, ABIArgInfo::getDirect());
}

/// Classify argument of given type \p Ty.
ABIArgInfo PNaClABIInfo::classifyArgumentType(QualType Ty) const {
  if (isAggregateTypeForABI(Ty)) {
    if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI()))
      return getNaturalAlignIndirect(Ty, RAA == CGCXXABI::RAA_DirectInMemory);
    return getNaturalAlignIndirect(Ty);
  } else if (const EnumType *EnumTy = Ty->getAs<EnumType>()) {
    // Treat an enum type as its underlying type.
    Ty = EnumTy->getDecl()->getIntegerType();
  } else if (Ty->isFloatingType()) {
    // Floating-point types don't go inreg.
    return ABIArgInfo::getDirect();
  } else if (const auto *EIT = Ty->getAs<ExtIntType>()) {
    // Treat extended integers as integers if <=64, otherwise pass indirectly.
    if (EIT->getNumBits() > 64)
      return getNaturalAlignIndirect(Ty);
    return ABIArgInfo::getDirect();
  }

  return (isPromotableIntegerTypeForABI(Ty) ? ABIArgInfo::getExtend(Ty)5
                                            : ABIArgInfo::getDirect());
}

ABIArgInfo PNaClABIInfo::classifyReturnType(QualType RetTy) const {
  if (RetTy->isVoidType())
    return ABIArgInfo::getIgnore();

  // In the PNaCl ABI we always return records/structures on the stack.
  if (isAggregateTypeForABI(RetTy))
    return getNaturalAlignIndirect(RetTy);

  // Treat extended integers as integers if <=64, otherwise pass indirectly.
  if (const auto *EIT = RetTy->getAs<ExtIntType>()) {
    if (EIT->getNumBits() > 64)
      return getNaturalAlignIndirect(RetTy);
    return ABIArgInfo::getDirect();
  }

  // Treat an enum type as its underlying type.
  if (const EnumType *EnumTy = RetTy->getAs<EnumType>())
    RetTy = EnumTy->getDecl()->getIntegerType();

  return (isPromotableIntegerTypeForABI(RetTy) ? ABIArgInfo::getExtend(RetTy)
                                               : ABIArgInfo::getDirect());
}

/// IsX86_MMXType - Return true if this is an MMX type.
bool IsX86_MMXType(llvm::Type *IRType) {
  // Return true if the type is an MMX type <2 x i32>, <4 x i16>, or <8 x i8>.
  return IRType->isVectorTy() && IRType->getPrimitiveSizeInBits() == 64 &&
    cast<llvm::VectorType>(IRType)->getElementType()->isIntegerTy() &&
    IRType->getScalarSizeInBits() != 64;
}

static llvm::Type* X86AdjustInlineAsmType(CodeGen::CodeGenFunction &CGF,
                                          StringRef Constraint,
                                          llvm::Type* Ty) {
  bool IsMMXCons = llvm::StringSwitch<bool>(Constraint)
                     .Cases("y", "&y", "^Ym", true)
                     .Default(false);
  if (IsMMXCons && Ty->isVectorTy()18) {
    if (cast<llvm::VectorType>(Ty)->getPrimitiveSizeInBits().getFixedSize() !=
        64) {
      // Invalid MMX constraint
      return nullptr;
    }

    return llvm::Type::getX86_MMXTy(CGF.getLLVMContext());
  }

  // No operation needed
  return Ty;
}

/// Returns true if this type can be passed in SSE registers with the
/// X86_VectorCall calling convention. Shared between x86_32 and x86_64.
static bool isX86VectorTypeForVectorCall(ASTContext &Context, QualType Ty) {
  if (const BuiltinType *BT = Ty->getAs<BuiltinType>()) {
    if (BT->isFloatingPoint() && BT->getKind() != BuiltinType::Half281) {
      if (BT->getKind() == BuiltinType::LongDouble) {
        if (&Context.getTargetInfo().getLongDoubleFormat() ==
            &llvm::APFloat::x87DoubleExtended())
          return false;
      }
      return true;
    }
  } else if (const VectorType *VT = Ty->getAs<VectorType>()) {
    // vectorcall can pass XMM, YMM, and ZMM vectors. We don't pass SSE1 MMX
    // registers specially.
    unsigned VecSize = Context.getTypeSize(VT);
    if (VecSize == 128 || VecSize == 2560 || VecSize == 5120)
      return true;
  }
  return false;
}

/// Returns true if this aggregate is small enough to be passed in SSE registers
/// in the X86_VectorCall calling convention. Shared between x86_32 and x86_64.
static bool isX86VectorCallAggregateSmallEnough(uint64_t NumMembers) {
  return NumMembers <= 4;
}

/// Returns a Homogeneous Vector Aggregate ABIArgInfo, used in X86.
static ABIArgInfo getDirectX86Hva(llvm::Type* T = nullptr) {
  auto AI = ABIArgInfo::getDirect(T);
  AI.setInReg(true);
  AI.setCanBeFlattened(false);
  return AI;
}

//===----------------------------------------------------------------------===//
// X86-32 ABI Implementation
//===----------------------------------------------------------------------===//

/// Similar to llvm::CCState, but for Clang.
struct CCState {
  CCState(CGFunctionInfo &FI)
      : IsPreassigned(FI.arg_size()), CC(FI.getCallingConvention()) {}

  llvm::SmallBitVector IsPreassigned;
  unsigned CC = CallingConv::CC_C;
  unsigned FreeRegs = 0;
  unsigned FreeSSERegs = 0;
};

/// X86_32ABIInfo - The X86-32 ABI information.
class X86_32ABIInfo : public SwiftABIInfo {
  enum Class {
    Integer,
    Float
  };

  static const unsigned MinABIStackAlignInBytes = 4;

  bool IsDarwinVectorABI;
  bool IsRetSmallStructInRegABI;
  bool IsWin32StructABI;
  bool IsSoftFloatABI;
  bool IsMCUABI;
  unsigned DefaultNumRegisterParameters;

  static bool isRegisterSize(unsigned Size) {
    return (Size == 8 || Size == 16502 || Size == 32483 || Size == 64241);
  }

  bool isHomogeneousAggregateBaseType(QualType Ty) const override {
    // FIXME: Assumes vectorcall is in use.
    return isX86VectorTypeForVectorCall(getContext(), Ty);
  }

  bool isHomogeneousAggregateSmallEnough(const Type *Ty,
                                         uint64_t NumMembers) const override {
    // FIXME: Assumes vectorcall is in use.
    return isX86VectorCallAggregateSmallEnough(NumMembers);
  }

  bool shouldReturnTypeInRegister(QualType Ty, ASTContext &Context) const;

  /// getIndirectResult - Give a source type \arg Ty, return a suitable result
  /// such that the argument will be passed in memory.
  ABIArgInfo getIndirectResult(QualType Ty, bool ByVal, CCState &State) const;

  ABIArgInfo getIndirectReturnResult(QualType Ty, CCState &State) const;

  /// Return the alignment to use for the given type on the stack.
  unsigned getTypeStackAlignInBytes(QualType Ty, unsigned Align) const;

  Class classify(QualType Ty) const;
  ABIArgInfo classifyReturnType(QualType RetTy, CCState &State) const;
  ABIArgInfo classifyArgumentType(QualType RetTy, CCState &State) const;

  /// Updates the number of available free registers, returns
  /// true if any registers were allocated.
  bool updateFreeRegs(QualType Ty, CCState &State) const;

  bool shouldAggregateUseDirect(QualType Ty, CCState &State, bool &InReg,
                                bool &NeedsPadding) const;
  bool shouldPrimitiveUseInReg(QualType Ty, CCState &State) const;

  bool canExpandIndirectArgument(QualType Ty) const;

  /// Rewrite the function info so that all memory arguments use
  /// inalloca.
  void rewriteWithInAlloca(CGFunctionInfo &FI) const;

  void addFieldToArgStruct(SmallVector<llvm::Type *, 6> &FrameFields,
                           CharUnits &StackOffset, ABIArgInfo &Info,
                           QualType Type) const;
  void runVectorCallFirstPass(CGFunctionInfo &FI, CCState &State) const;

public:

  void computeInfo(CGFunctionInfo &FI) const override;
  Address EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                    QualType Ty) const override;

  X86_32ABIInfo(CodeGen::CodeGenTypes &CGT, bool DarwinVectorABI,
                bool RetSmallStructInRegABI, bool Win32StructABI,
                unsigned NumRegisterParameters, bool SoftFloatABI)
    : SwiftABIInfo(CGT), IsDarwinVectorABI(DarwinVectorABI),
      IsRetSmallStructInRegABI(RetSmallStructInRegABI),
      IsWin32StructABI(Win32StructABI),
      IsSoftFloatABI(SoftFloatABI),
      IsMCUABI(CGT.getTarget().getTriple().isOSIAMCU()),
      DefaultNumRegisterParameters(NumRegisterParameters) {}

  bool shouldPassIndirectlyForSwift(ArrayRef<llvm::Type*> scalars,
                                    bool asReturnValue) const override {
    // LLVM's x86-32 lowering currently only assigns up to three
    // integer registers and three fp registers.  Oddly, it'll use up to
    // four vector registers for vectors, but those can overlap with the
    // scalar registers.
    return occupiesMoreThan(CGT, scalars, /*total*/ 3);
  }

  bool isSwiftErrorInRegister() const override {
    // x86-32 lowering does not support passing swifterror in a register.
    return false;
  }
};

class X86_32TargetCodeGenInfo : public TargetCodeGenInfo {
public:
  X86_32TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, bool DarwinVectorABI,
                          bool RetSmallStructInRegABI, bool Win32StructABI,
                          unsigned NumRegisterParameters, bool SoftFloatABI)
      : TargetCodeGenInfo(std::make_unique<X86_32ABIInfo>(
            CGT, DarwinVectorABI, RetSmallStructInRegABI, Win32StructABI,
            NumRegisterParameters, SoftFloatABI)) {}

  static bool isStructReturnInRegABI(
      const llvm::Triple &Triple, const CodeGenOptions &Opts);

  void setTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
                           CodeGen::CodeGenModule &CGM) const override;

  int getDwarfEHStackPointer(CodeGen::CodeGenModule &CGM) const override {
    // Darwin uses different dwarf register numbers for EH.
    if (CGM.getTarget().getTriple().isOSDarwin()) return 5;
    return 4;
  }

  bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                               llvm::Value *Address) const override;

  llvm::Type* adjustInlineAsmType(CodeGen::CodeGenFunction &CGF,
                                  StringRef Constraint,
                                  llvm::Type* Ty) const override {
    return X86AdjustInlineAsmType(CGF, Constraint, Ty);
  }

  void addReturnRegisterOutputs(CodeGenFunction &CGF, LValue ReturnValue,
                                std::string &Constraints,
                                std::vector<llvm::Type *> &ResultRegTypes,
                                std::vector<llvm::Type *> &ResultTruncRegTypes,
                                std::vector<LValue> &ResultRegDests,
                                std::string &AsmString,
                                unsigned NumOutputs) const override;

  llvm::Constant *
  getUBSanFunctionSignature(CodeGen::CodeGenModule &CGM) const override {
    unsigned Sig = (0xeb << 0) |  // jmp rel8
                   (0x06 << 8) |  //           .+0x08
                   ('v' << 16) |
                   ('2' << 24);
    return llvm::ConstantInt::get(CGM.Int32Ty, Sig);
  }

  StringRef getARCRetainAutoreleasedReturnValueMarker() const override {
    return "movl\t%ebp, %ebp"
           "\t\t// marker for objc_retainAutoreleaseReturnValue";
  }
};

}

/// Rewrite input constraint references after adding some output constraints.
/// In the case where there is one output and one input and we add one output,
/// we need to replace all operand references greater than or equal to 1:
///     mov $0, $1
///     mov eax, $1
/// The result will be:
///     mov $0, $2
///     mov eax, $2
static void rewriteInputConstraintReferences(unsigned FirstIn,
                                             unsigned NumNewOuts,
                                             std::string &AsmString) {
  std::string Buf;
  llvm::raw_string_ostream OS(Buf);
  size_t Pos = 0;
  while (Pos < AsmString.size()) {
    size_t DollarStart = AsmString.find('$', Pos);
    if (DollarStart == std::string::npos)
      DollarStart = AsmString.size();
    size_t DollarEnd = AsmString.find_first_not_of('$', DollarStart);
    if (DollarEnd == std::string::npos)
      DollarEnd = AsmString.size();
    OS << StringRef(&AsmString[Pos], DollarEnd - Pos);
    Pos = DollarEnd;
    size_t NumDollars = DollarEnd - DollarStart;
    if (NumDollars % 2 != 0 && Pos < AsmString.size()18) {
      // We have an operand reference.
      size_t DigitStart = Pos;
      if (AsmString[DigitStart] == '{') {
        OS << '{';
        ++DigitStart;
      }
      size_t DigitEnd = AsmString.find_first_not_of("0123456789", DigitStart);
      if (DigitEnd == std::string::npos)
        DigitEnd = AsmString.size();
      StringRef OperandStr(&AsmString[DigitStart], DigitEnd - DigitStart);
      unsigned OperandIndex;
      if (!OperandStr.getAsInteger(10, OperandIndex)) {
        if (OperandIndex >= FirstIn)
          OperandIndex += NumNewOuts;
        OS << OperandIndex;
      } else {
        OS << OperandStr;
      }
      Pos = DigitEnd;
    }
  }
  AsmString = std::move(OS.str());
}

/// Add output constraints for EAX:EDX because they are return registers.
void X86_32TargetCodeGenInfo::addReturnRegisterOutputs(
    CodeGenFunction &CGF, LValue ReturnSlot, std::string &Constraints,
    std::vector<llvm::Type *> &ResultRegTypes,
    std::vector<llvm::Type *> &ResultTruncRegTypes,
    std::vector<LValue> &ResultRegDests, std::string &AsmString,
    unsigned NumOutputs) const {
  uint64_t RetWidth = CGF.getContext().getTypeSize(ReturnSlot.getType());

  // Use the EAX constraint if the width is 32 or smaller and EAX:EDX if it is
  // larger.
  if (!Constraints.empty())
    Constraints += ',';
  if (RetWidth <= 32) {
    Constraints += "={eax}";
    ResultRegTypes.push_back(CGF.Int32Ty);
  } else {
    // Use the 'A' constraint for EAX:EDX.
    Constraints += "=A";
    ResultRegTypes.push_back(CGF.Int64Ty);
  }

  // Truncate EAX or EAX:EDX to an integer of the appropriate size.
  llvm::Type *CoerceTy = llvm::IntegerType::get(CGF.getLLVMContext(), RetWidth);
  ResultTruncRegTypes.push_back(CoerceTy);

  // Coerce the integer by bitcasting the return slot pointer.
  ReturnSlot.setAddress(CGF.Builder.CreateBitCast(ReturnSlot.getAddress(CGF),
                                                  CoerceTy->getPointerTo()));
  ResultRegDests.push_back(ReturnSlot);

  rewriteInputConstraintReferences(NumOutputs, 1, AsmString);
}

/// shouldReturnTypeInRegister - Determine if the given type should be
/// returned in a register (for the Darwin and MCU ABI).
bool X86_32ABIInfo::shouldReturnTypeInRegister(QualType Ty,
                                               ASTContext &Context) const {
  uint64_t Size = Context.getTypeSize(Ty);

  // For i386, type must be register sized.
  // For the MCU ABI, it only needs to be <= 8-byte
  if ((IsMCUABI && Size > 6419) || (549!IsMCUABI549 && !isRegisterSize(Size)533))
   return false;

  if (Ty->isVectorType()) {
    // 64- and 128- bit vectors inside structures are not returned in
    // registers.
    if (Size == 64 || Size == 1281)
      return false;

    return true;
  }

  // If this is a builtin, pointer, enum, complex type, member pointer, or
  // member function pointer it is ok.
  if (Ty->getAs<BuiltinType>() || Ty->hasPointerRepresentation()384 ||
      Ty->isAnyComplexType() || Ty->isEnumeralType()355 ||
      Ty->isBlockPointerType() || Ty->isMemberPointerType())
    return true;

  // Arrays are treated like records.
  if (const ConstantArrayType *AT = Context.getAsConstantArrayType(Ty))
    return shouldReturnTypeInRegister(AT->getElementType(), Context);

  // Otherwise, it must be a record type.
  const RecordType *RT = Ty->getAs<RecordType>();
  if (!RT) return false0;

  // FIXME: Traverse bases here too.

  // Structure types are passed in register if all fields would be
  // passed in a register.
  for (const auto *FD : RT->getDecl()->fields())85 {
    // Empty fields are ignored.
    if (isEmptyField(Context, FD, true))
      continue;

    // Check fields recursively.
    if (!shouldReturnTypeInRegister(FD->getType(), Context))
      return false;
  }
  return true;
}

static bool is32Or64BitBasicType(QualType Ty, ASTContext &Context) {
  // Treat complex types as the element type.
  if (const ComplexType *CTy = Ty->getAs<ComplexType>())
    Ty = CTy->getElementType();

  // Check for a type which we know has a simple scalar argument-passing
  // convention without any padding.  (We're specifically looking for 32
  // and 64-bit integer and integer-equivalents, float, and double.)
  if (!Ty->getAs<BuiltinType>() && !Ty->hasPointerRepresentation()36 &&
      !Ty->isEnumeralType() && !Ty->isBlockPointerType()28)
    return false;

  uint64_t Size = Context.getTypeSize(Ty);
  return Size == 32 || Size == 6428;
}

static bool addFieldSizes(ASTContext &Context, const RecordDecl *RD,
                          uint64_t &Size) {
  for (const auto *FD : RD->fields()) {
    // Scalar arguments on the stack get 4 byte alignment on x86. If the
    // argument is smaller than 32-bits, expanding the struct will create
    // alignment padding.
    if (!is32Or64BitBasicType(FD->getType(), Context))
      return false;

    // FIXME: Reject bit-fields wholesale; there are two problems, we don't know
    // how to expand them yet, and the predicate for telling if a bitfield still
    // counts as "basic" is more complicated than what we were doing previously.
    if (FD->isBitField())
      return false;

    Size += Context.getTypeSize(FD->getType());
  }
  return true;
}

static bool addBaseAndFieldSizes(ASTContext &Context, const CXXRecordDecl *RD,
                                 uint64_t &Size) {
  // Don't do this if there are any non-empty bases.
  for (const CXXBaseSpecifier &Base : RD->bases()) {
    if (!addBaseAndFieldSizes(Context, Base.getType()->getAsCXXRecordDecl(),
                              Size))
      return false;
  }
  if (!addFieldSizes(Context, RD, Size))
    return false;
  return true;
}

/// Test whether an argument type which is to be passed indirectly (on the
/// stack) would have the equivalent layout if it was expanded into separate
/// arguments. If so, we prefer to do the latter to avoid inhibiting
/// optimizations.
bool X86_32ABIInfo::canExpandIndirectArgument(QualType Ty) const {
  // We can only expand structure types.
  const RecordType *RT = Ty->getAs<RecordType>();
  if (!RT)
    return false;
  const RecordDecl *RD = RT->getDecl();
  uint64_t Size = 0;
  if (const CXXRecordDecl *CXXRD = dyn_cast<CXXRecordDecl>(RD)) {
    if (!IsWin32StructABI) {
      // On non-Windows, we have to conservatively match our old bitcode
      // prototypes in order to be ABI-compatible at the bitcode level.
      if (!CXXRD->isCLike())
        return false;
    } else {
      // Don't do this for dynamic classes.
      if (CXXRD->isDynamicClass())
        return false;
    }
    if (!addBaseAndFieldSizes(getContext(), CXXRD, Size))
      return false;
  } else {
    if (!addFieldSizes(getContext(), RD, Size))
      return false;
  }

  // We can do this if there was no alignment padding.
  return Size == getContext().getTypeSize(Ty);
}

ABIArgInfo X86_32ABIInfo::getIndirectReturnResult(QualType RetTy, CCState &State) const {
  // If the return value is indirect, then the hidden argument is consuming one
  // integer register.
  if (State.FreeRegs) {
    --State.FreeRegs;
    if (!IsMCUABI)
      return getNaturalAlignIndirectInReg(RetTy);
  }
  return getNaturalAlignIndirect(RetTy, /*ByVal=*/false);
}

ABIArgInfo X86_32ABIInfo::classifyReturnType(QualType RetTy,
                                             CCState &State) const {
  if (RetTy->isVoidType())
    return ABIArgInfo::getIgnore();

  const Type *Base = nullptr;
  uint64_t NumElts = 0;
  if ((State.CC == llvm::CallingConv::X86_VectorCall ||
       State.CC == llvm::CallingConv::X86_RegCall) &&
      isHomogeneousAggregate(RetTy, Base, NumElts)) {
    // The LLVM struct type for such an aggregate should lower properly.
    return ABIArgInfo::getDirect();
  }

  if (const VectorType *VT = RetTy->getAs<VectorType>()) {
    // On Darwin, some vectors are returned in registers.
    if (IsDarwinVectorABI) {
      uint64_t Size = getContext().getTypeSize(RetTy);

      // 128-bit vectors are a special case; they are returned in
      // registers and we need to make sure to pick a type the LLVM
      // backend will like.
      if (Size == 128)
        return ABIArgInfo::getDirect(llvm::FixedVectorType::get(
            llvm::Type::getInt64Ty(getVMContext()), 2));

      // Always return in register if it fits in a general purpose
      // register, or if it is 64 bits and has a single element.
      if ((Size == 8 || Size == 16 || Size == 32) ||
          (Size == 64 && VT->getNumElements() == 1))
        return ABIArgInfo::getDirect(llvm::IntegerType::get(getVMContext(),
                                                            Size));

      return getIndirectReturnResult(RetTy, State);
    }

    return ABIArgInfo::getDirect();
  }

  if (isAggregateTypeForABI(RetTy)) {
    if (const RecordType *RT = RetTy->getAs<RecordType>()) {
      // Structures with flexible arrays are always indirect.
      if (RT->getDecl()->hasFlexibleArrayMember())
        return getIndirectReturnResult(RetTy, State);
    }

    // If specified, structs and unions are always indirect.
    if (!IsRetSmallStructInRegABI && !RetTy->isAnyComplexType()116)
      return getIndirectReturnResult(RetTy, State);

    // Ignore empty structs/unions.
    if (isEmptyRecord(getContext(), RetTy, true))
      return ABIArgInfo::getIgnore();

    // Small structures which are register sized are generally returned
    // in a register.
    if (shouldReturnTypeInRegister(RetTy, getContext())) {
      uint64_t Size = getContext().getTypeSize(RetTy);

      // As a special-case, if the struct is a "single-element" struct, and
      // the field is of type "float" or "double", return it in a
      // floating-point register. (MSVC does not apply this special case.)
      // We apply a similar transformation for pointer types to improve the
      // quality of the generated IR.
      if (const Type *SeltTy = isSingleElementStruct(RetTy, getContext()))
        if ((!IsWin32StructABI && SeltTy->isRealFloatingType()42)
            || SeltTy->hasPointerRepresentation())
          return ABIArgInfo::getDirect(CGT.ConvertType(QualType(SeltTy, 0)));

      // FIXME: We should be able to narrow this integer in cases with dead
      // padding.
      return ABIArgInfo::getDirect(llvm::IntegerType::get(getVMContext(),Size));
    }

    return getIndirectReturnResult(RetTy, State);
  }

  // Treat an enum type as its underlying type.
  if (const EnumType *EnumTy = RetTy->getAs<EnumType>())
    RetTy = EnumTy->getDecl()->getIntegerType();

  if (const auto *EIT = RetTy->getAs<ExtIntType>())
    if (EIT->getNumBits() > 64)
      return getIndirectReturnResult(RetTy, State);

  return (isPromotableIntegerTypeForABI(RetTy) ? ABIArgInfo::getExtend(RetTy)130
                                               : ABIArgInfo::getDirect());
}

static bool isSIMDVectorType(ASTContext &Context, QualType Ty) {
  return Ty->getAs<VectorType>() && Context.getTypeSize(Ty) == 12814;
}

static bool isRecordWithSIMDVectorType(ASTContext &Context, QualType Ty) {
  const RecordType *RT = Ty->getAs<RecordType>();
  if (!RT)
    return 0;
  const RecordDecl *RD = RT->getDecl();

  // If this is a C++ record, check the bases first.
  if (const CXXRecordDecl *CXXRD = dyn_cast<CXXRecordDecl>(RD))
    for (const auto &I : CXXRD->bases())
      if (!isRecordWithSIMDVectorType(Context, I.getType()))
        return false;

  for (const auto *i : RD->fields())42 {
    QualType FT = i->getType();

    if (isSIMDVectorType(Context, FT))
      return true;

    if (isRecordWithSIMDVectorType(Context, FT))
      return true;
  }

  return false;
}

unsigned X86_32ABIInfo::getTypeStackAlignInBytes(QualType Ty,
                                                 unsigned Align) const {
  // Otherwise, if the alignment is less than or equal to the minimum ABI
  // alignment, just use the default; the backend will handle this.
  if (Align <= MinABIStackAlignInBytes)
    return 0; // Use default alignment.

  // On non-Darwin, the stack type alignment is always 4.
  if (!IsDarwinVectorABI) {
    // Set explicit alignment, since we may need to realign the top.
    return MinABIStackAlignInBytes;
  }

  // Otherwise, if the type contains an SSE vector type, the alignment is 16.
  if (Align >= 16 && (17isSIMDVectorType(getContext(), Ty)17 ||
                      isRecordWithSIMDVectorType(getContext(), Ty)))
    return 16;

  return MinABIStackAlignInBytes;
}

ABIArgInfo X86_32ABIInfo::getIndirectResult(QualType Ty, bool ByVal,
                                            CCState &State) const {
  if (!ByVal) {
    if (State.FreeRegs) {
      --State.FreeRegs; // Non-byval indirects just use one pointer.
      if (!IsMCUABI)
        return getNaturalAlignIndirectInReg(Ty);
    }
    return getNaturalAlignIndirect(Ty, false);
  }

  // Compute the byval alignment.
  unsigned TypeAlign = getContext().getTypeAlign(Ty) / 8;
  unsigned StackAlign = getTypeStackAlignInBytes(Ty, TypeAlign);
  if (StackAlign == 0)
    return ABIArgInfo::getIndirect(CharUnits::fromQuantity(4), /*ByVal=*/true);

  // If the stack alignment is less than the type alignment, realign the
  // argument.
  bool Realign = TypeAlign > StackAlign;
  return ABIArgInfo::getIndirect(CharUnits::fromQuantity(StackAlign),
                                 /*ByVal=*/true, Realign);
}

X86_32ABIInfo::Class X86_32ABIInfo::classify(QualType Ty) const {
  const Type *T = isSingleElementStruct(Ty, getContext());
  if (!T)
    T = Ty.getTypePtr();

  if (const BuiltinType *BT = T->getAs<BuiltinType>()) {
    BuiltinType::Kind K = BT->getKind();
    if (K == BuiltinType::Float || K == BuiltinType::Double10.4k)
      return Float;
  }
  return Integer;
}

bool X86_32ABIInfo::updateFreeRegs(QualType Ty, CCState &State) const {
  if (!IsSoftFloatABI) {
    Class C = classify(Ty);
    if (C == Float)
      return false;
  }

  unsigned Size = getContext().getTypeSize(Ty);
  unsigned SizeInRegs = (Size + 31) / 32;

  if (SizeInRegs == 0)
    return false;

  if (!IsMCUABI) {
    if (SizeInRegs > State.FreeRegs) {
      State.FreeRegs = 0;
      return false;
    }
  } else {
    // The MCU psABI allows passing parameters in-reg even if there are
    // earlier parameters that are passed on the stack. Also,
    // it does not allow passing >8-byte structs in-register,
    // even if there are 3 free registers available.
    if (SizeInRegs > State.FreeRegs || SizeInRegs > 234)
      return false;
  }

  State.FreeRegs -= SizeInRegs;
  return true;
}

bool X86_32ABIInfo::shouldAggregateUseDirect(QualType Ty, CCState &State,
                                             bool &InReg,
                                             bool &NeedsPadding) const {
  // On Windows, aggregates other than HFAs are never passed in registers, and
  // they do not consume register slots. Homogenous floating-point aggregates
  // (HFAs) have already been dealt with at this point.
  if (IsWin32StructABI && isAggregateTypeForABI(Ty)168)
    return false;

  NeedsPadding = false;
  InReg = !IsMCUABI;

  if (!updateFreeRegs(Ty, State))
    return false;

  if (IsMCUABI)
    return true;

  if (State.CC == llvm::CallingConv::X86_FastCall ||
      State.CC == llvm::CallingConv::X86_VectorCall ||
      State.CC == llvm::CallingConv::X86_RegCall) {
    if (getContext().getTypeSize(Ty) <= 32 && State.FreeRegs)
      NeedsPadding = true;

    return false;
  }

  return true;
}

bool X86_32ABIInfo::shouldPrimitiveUseInReg(QualType Ty, CCState &State) const {
  if (!updateFreeRegs(Ty, State))
    return false;

  if (IsMCUABI)
    return false;

  if (State.CC == llvm::CallingConv::X86_FastCall ||
      State.CC == llvm::CallingConv::X86_VectorCall ||
      State.CC == llvm::CallingConv::X86_RegCall94) {
    if (getContext().getTypeSize(Ty) > 32)
      return false;

    return (Ty->isIntegralOrEnumerationType() || Ty->isPointerType()27 ||
        Ty->isReferenceType());
  }

  return true;
}

void X86_32ABIInfo::runVectorCallFirstPass(CGFunctionInfo &FI, CCState &State) const {
  // Vectorcall x86 works subtly different than in x64, so the format is
  // a bit different than the x64 version.  First, all vector types (not HVAs)
  // are assigned, with the first 6 ending up in the [XYZ]MM0-5 registers.
  // This differs from the x64 implementation, where the first 6 by INDEX get
  // registers.
  // In the second pass over the arguments, HVAs are passed in the remaining
  // vector registers if possible, or indirectly by address. The address will be
  // passed in ECX/EDX if available. Any other arguments are passed according to
  // the usual fastcall rules.
  MutableArrayRef<CGFunctionInfoArgInfo> Args = FI.arguments();
  for (int I = 0, E = Args.size(); I < E; ++I114) {
    const Type *Base = nullptr;
    uint64_t NumElts = 0;
    const QualType &Ty = Args[I].type;
    if ((Ty->isVectorType() || Ty->isBuiltinType()93) &&
        isHomogeneousAggregate(Ty, Base, NumElts)) {
      if (State.FreeSSERegs >= NumElts) {
        State.FreeSSERegs -= NumElts;
        Args[I].info = ABIArgInfo::getDirectInReg();
        State.IsPreassigned.set(I);
      }
    }
  }
}

ABIArgInfo X86_32ABIInfo::classifyArgumentType(QualType Ty,
                                               CCState &State) const {
  // FIXME: Set alignment on indirect arguments.
  bool IsFastCall = State.CC == llvm::CallingConv::X86_FastCall;
  bool IsRegCall = State.CC == llvm::CallingConv::X86_RegCall;
  bool IsVectorCall = State.CC == llvm::CallingConv::X86_VectorCall;

  Ty = useFirstFieldIfTransparentUnion(Ty);
  TypeInfo TI = getContext().getTypeInfo(Ty);

  // Check with the C++ ABI first.
  const RecordType *RT = Ty->getAs<RecordType>();
  if (RT) {
    CGCXXABI::RecordArgABI RAA = getRecordArgABI(RT, getCXXABI());
    if (RAA == CGCXXABI::RAA_Indirect) {
      return getIndirectResult(Ty, false, State);
    } else if (RAA == CGCXXABI::RAA_DirectInMemory) {
      // The field index doesn't matter, we'll fix it up later.
      return ABIArgInfo::getInAlloca(/*FieldIndex=*/0);
    }
  }

  // Regcall uses the concept of a homogenous vector aggregate, similar
  // to other targets.
  const Type *Base = nullptr;
  uint64_t NumElts = 0;
  if ((IsRegCall || IsVectorCall32.2k) &&
      isHomogeneousAggregate(Ty, Base, NumElts)) {
    if (State.FreeSSERegs >= NumElts) {
      State.FreeSSERegs -= NumElts;

      // Vectorcall passes HVAs directly and does not flatten them, but regcall
      // does.
      if (IsVectorCall)
        return getDirectX86Hva();

      if (Ty->isBuiltinType() || Ty->isVectorType()35)
        return ABIArgInfo::getDirect();
      return ABIArgInfo::getExpand();
    }
    return getIndirectResult(Ty, /*ByVal=*/false, State);
  }

  if (isAggregateTypeForABI(Ty)) {
    // Structures with flexible arrays are always indirect.
    // FIXME: This should not be byval!
    if (RT && RT->getDecl()->hasFlexibleArrayMember()348)
      return getIndirectResult(Ty, true, State);

    // Ignore empty structs/unions on non-Windows.
    if (!IsWin32StructABI && isEmptyRecord(getContext(), Ty, true)255)
      return ABIArgInfo::getIgnore();

    llvm::LLVMContext &LLVMContext = getVMContext();
    llvm::IntegerType *Int32 = llvm::Type::getInt32Ty(LLVMContext);
    bool NeedsPadding = false;
    bool InReg;
    if (shouldAggregateUseDirect(Ty, State, InReg, NeedsPadding)) {
      unsigned SizeInRegs = (TI.Width + 31) / 32;
      SmallVector<llvm::Type*, 3> Elements(SizeInRegs, Int32);
      llvm::Type *Result = llvm::StructType::get(LLVMContext, Elements);
      if (InReg)
        return ABIArgInfo::getDirectInReg(Result);
      else
        return ABIArgInfo::getDirect(Result);
    }
    llvm::IntegerType *PaddingType = NeedsPadding ? Int323 : nullptr369;

    // Pass over-aligned aggregates on Windows indirectly. This behavior was
    // added in MSVC 2015.
    if (IsWin32StructABI && TI.AlignIsRequired168 && TI.Align > 321)
      return getIndirectResult(Ty, /*ByVal=*/false, State);

    // Expand small (<= 128-bit) record types when we know that the stack layout
    // of those arguments will match the struct. This is important because the
    // LLVM backend isn't smart enough to remove byval, which inhibits many
    // optimizations.
    // Don't do this for the MCU if there are still free integer registers
    // (see X86_64 ABI for full explanation).
    if (TI.Width <= 4 * 32 && (325!IsMCUABI325 || State.FreeRegs == 05) &&
        canExpandIndirectArgument(Ty))
      return ABIArgInfo::getExpandWithPadding(
          IsFastCall || IsVectorCall81 || IsRegCall80, PaddingType);

    return getIndirectResult(Ty, true, State);
  }

  if (const VectorType *VT = Ty->getAs<VectorType>()) {
    // On Windows, vectors are passed directly if registers are available, or
    // indirectly if not. This avoids the need to align argument memory. Pass
    // user-defined vector types larger than 512 bits indirectly for simplicity.
    if (IsWin32StructABI) {
      if (TI.Width <= 512 && State.FreeSSERegs > 040) {
        --State.FreeSSERegs;
        return ABIArgInfo::getDirectInReg();
      }
      return getIndirectResult(Ty, /*ByVal=*/false, State);
    }

    // On Darwin, some vectors are passed in memory, we handle this by passing
    // it as an i8/i16/i32/i64.
    if (IsDarwinVectorABI) {
      if ((TI.Width == 8 || TI.Width == 16 || TI.Width == 32) ||
          (TI.Width == 64 && VT->getNumElements() == 110))
        return ABIArgInfo::getDirect(
            llvm::IntegerType::get(getVMContext(), TI.Width));
    }

    if (IsX86_MMXType(CGT.ConvertType(Ty)))
      return ABIArgInfo::getDirect(llvm::IntegerType::get(getVMContext(), 64));

    return ABIArgInfo::getDirect();
  }


  if (const EnumType *EnumTy = Ty->getAs<EnumType>())
    Ty = EnumTy->getDecl()->getIntegerType();

  bool InReg = shouldPrimitiveUseInReg(Ty, State);

  if (isPromotableIntegerTypeForABI(Ty)) {
    if (InReg)
      return ABIArgInfo::getExtendInReg(Ty);
    return ABIArgInfo::getExtend(Ty);
  }

  if (const auto * EIT = Ty->getAs<ExtIntType>()) {
    if (EIT->getNumBits() <= 64) {
      if (InReg)
        return ABIArgInfo::getDirectInReg();
      return ABIArgInfo::getDirect();
    }
    return getIndirectResult(Ty, /*ByVal=*/false, State);
  }

  if (InReg)
    return ABIArgInfo::getDirectInReg();
  return ABIArgInfo::getDirect();
}

void X86_32ABIInfo::computeInfo(CGFunctionInfo &FI) const {
  CCState State(FI);
  if (IsMCUABI)
    State.FreeRegs = 3;
  else if (State.CC == llvm::CallingConv::X86_FastCall) {
    State.FreeRegs = 2;
    State.FreeSSERegs = 3;
  } else if (State.CC == llvm::CallingConv::X86_VectorCall) {
    State.FreeRegs = 2;
    State.FreeSSERegs = 6;
  } else if (FI.getHasRegParm())
    State.FreeRegs = FI.getRegParm();
  else if (State.CC == llvm::CallingConv::X86_RegCall) {
    State.FreeRegs = 5;
    State.FreeSSERegs = 8;
  } else if (IsWin32StructABI) {
    // Since MSVC 2015, the first three SSE vectors have been passed in
    // registers. The rest are passed indirectly.
    State.FreeRegs = DefaultNumRegisterParameters;
    State.FreeSSERegs = 3;
  } else
    State.FreeRegs = DefaultNumRegisterParameters;

  if (!::classifyReturnType(getCXXABI(), FI, *this)) {
    FI.getReturnInfo() = classifyReturnType(FI.getReturnType(), State);
  } else if (FI.getReturnInfo().isIndirect()) {
    // The C++ ABI is not aware of register usage, so we have to check if the
    // return value was sret and put it in a register ourselves if appropriate.
    if (State.FreeRegs) {
      --State.FreeRegs;  // The sret parameter consumes a register.
      if (!IsMCUABI)
        FI.getReturnInfo().setInReg(true);
    }
  }

  // The chain argument effectively gives us another free register.
  if (FI.isChainCall())
    ++State.FreeRegs;

  // For vectorcall, do a first pass over the arguments, assigning FP and vector
  // arguments to XMM registers as available.
  if (State.CC == llvm::CallingConv::X86_VectorCall)
    runVectorCallFirstPass(FI, State);

  bool UsedInAlloca = false;
  MutableArrayRef<CGFunctionInfoArgInfo> Args = FI.arguments();
  for (int I = 0, E = Args.size(); I < E; ++I32.5k) {
    // Skip arguments that have already been assigned.
    if (State.IsPreassigned.test(I))
      continue;

    Args[I].info = classifyArgumentType(Args[I].type, State);
    UsedInAlloca |= (Args[I].info.getKind() == ABIArgInfo::InAlloca);
  }

  // If we needed to use inalloca for any argument, do a second pass and rewrite
  // all the memory arguments to use inalloca.
  if (UsedInAlloca)
    rewriteWithInAlloca(FI);
}

void
X86_32ABIInfo::addFieldToArgStruct(SmallVector<llvm::Type *, 6> &FrameFields,
                                   CharUnits &StackOffset, ABIArgInfo &Info,
                                   QualType Type) const {
  // Arguments are always 4-byte-aligned.
  CharUnits WordSize = CharUnits::fromQuantity(4);
  assert(StackOffset.isMultipleOf(WordSize) && "unaligned inalloca struct");

  // sret pointers and indirect things will require an extra pointer
  // indirection, unless they are byval. Most things are byval, and will not
  // require this indirection.
  bool IsIndirect = false;
  if (Info.isIndirect() && !Info.getIndirectByVal()17)
    IsIndirect = true;
  Info = ABIArgInfo::getInAlloca(FrameFields.size(), IsIndirect);
  llvm::Type *LLTy = CGT.ConvertTypeForMem(Type);
  if (IsIndirect)
    LLTy = LLTy->getPointerTo(0);
  FrameFields.push_back(LLTy);
  StackOffset += IsIndirect ? WordSize14 : getContext().getTypeSizeInChars(Type);

  // Insert padding bytes to respect alignment.
  CharUnits FieldEnd = StackOffset;
  StackOffset = FieldEnd.alignTo(WordSize);
  if (StackOffset != FieldEnd) {
    CharUnits NumBytes = StackOffset - FieldEnd;
    llvm::Type *Ty = llvm::Type::getInt8Ty(getVMContext());
    Ty = llvm::ArrayType::get(Ty, NumBytes.getQuantity());
    FrameFields.push_back(Ty);
  }
}

static bool isArgInAlloca(const ABIArgInfo &Info) {
  // Leave ignored and inreg arguments alone.
  switch (Info.getKind()) {
  case ABIArgInfo::InAlloca:
    return true;
  case ABIArgInfo::Ignore:
  case ABIArgInfo::IndirectAliased:
    return false;
  case ABIArgInfo::Indirect:
  case ABIArgInfo::Direct:
  case ABIArgInfo::Extend:
    return !Info.getInReg();
  case ABIArgInfo::Expand:
  case ABIArgInfo::CoerceAndExpand:
    // These are aggregate types which are never passed in registers when
    // inalloca is involved.
    return true;
  }
  llvm_unreachable("invalid enum");
}

void X86_32ABIInfo::rewriteWithInAlloca(CGFunctionInfo &FI) const {
  assert(IsWin32StructABI && "inalloca only supported on win32");

  // Build a packed struct type for all of the arguments in memory.
  SmallVector<llvm::Type *, 6> FrameFields;

  // The stack alignment is always 4.
  CharUnits StackAlign = CharUnits::fromQuantity(4);

  CharUnits StackOffset;
  CGFunctionInfo::arg_iterator I = FI.arg_begin(), E = FI.arg_end();

  // Put 'this' into the struct before 'sret', if necessary.
  bool IsThisCall =
      FI.getCallingConvention() == llvm::CallingConv::X86_ThisCall;
  ABIArgInfo &Ret = FI.getReturnInfo();
  if (Ret.isIndirect() && Ret.isSRetAfterThis()9 && !IsThisCall7 &&
      isArgInAlloca(I->info)) {
    addFieldToArgStruct(FrameFields, StackOffset, I->info, I->type);
    ++I;
  }

  // Put the sret parameter into the inalloca struct if it's in memory.
  if (Ret.isIndirect() && !Ret.getInReg()9) {
    addFieldToArgStruct(FrameFields, StackOffset, Ret, FI.getReturnType());
    // On Windows, the hidden sret parameter is always returned in eax.
    Ret.setInAllocaSRet(IsWin32StructABI);
  }

  // Skip the 'this' parameter in ecx.
  if (IsThisCall)
    ++I;

  // Put arguments passed in memory into the struct.
  for (; I != E; ++I169) {
    if (isArgInAlloca(I->info))
      addFieldToArgStruct(FrameFields, StackOffset, I->info, I->type);
  }

  FI.setArgStruct(llvm::StructType::get(getVMContext(), FrameFields,
                                        /*isPacked=*/true),
                  StackAlign);
}

Address X86_32ABIInfo::EmitVAArg(CodeGenFunction &CGF,
                                 Address VAListAddr, QualType Ty) const {

  auto TypeInfo = getContext().getTypeInfoInChars(Ty);

  // x86-32 changes the alignment of certain arguments on the stack.
  //
  // Just messing with TypeInfo like this works because we never pass
  // anything indirectly.
  TypeInfo.Align = CharUnits::fromQuantity(
                getTypeStackAlignInBytes(Ty, TypeInfo.Align.getQuantity()));

  return emitVoidPtrVAArg(CGF, VAListAddr, Ty, /*Indirect*/ false,
                          TypeInfo, CharUnits::fromQuantity(4),
                          /*AllowHigherAlign*/ true);
}

bool X86_32TargetCodeGenInfo::isStructReturnInRegABI(
    const llvm::Triple &Triple, const CodeGenOptions &Opts) {
  assert(Triple.getArch() == llvm::Triple::x86);

  switch (Opts.getStructReturnConvention()) {
  case CodeGenOptions::SRCK_Default:
    break;
  case CodeGenOptions::SRCK_OnStack:  // -fpcc-struct-return
    return false;
  case CodeGenOptions::SRCK_InRegs:  // -freg-struct-return
    return true;
  }

  if (Triple.isOSDarwin() || Triple.isOSIAMCU()2.81k)
    return true;

  switch (Triple.getOS()) {
  case llvm::Triple::DragonFly:
  case llvm::Triple::FreeBSD:
  case llvm::Triple::OpenBSD:
  case llvm::Triple::Win32:
    return true;
  default:
    return false;
  }
}

static void addX86InterruptAttrs(const FunctionDecl *FD, llvm::GlobalValue *GV,
                                 CodeGen::CodeGenModule &CGM) {
  if (!FD->hasAttr<AnyX86InterruptAttr>())
    return;

  llvm::Function *Fn = cast<llvm::Function>(GV);
  Fn->setCallingConv(llvm::CallingConv::X86_INTR);
  if (FD->getNumParams() == 0)
    return;

  auto PtrTy = cast<PointerType>(FD->getParamDecl(0)->getType());
  llvm::Type *ByValTy = CGM.getTypes().ConvertType(PtrTy->getPointeeType());
  llvm::Attribute NewAttr = llvm::Attribute::getWithByValType(
    Fn->getContext(), ByValTy);
  Fn->addParamAttr(0, NewAttr);
}

void X86_32TargetCodeGenInfo::setTargetAttributes(
    const Decl *D, llvm::GlobalValue *GV, CodeGen::CodeGenModule &CGM) const {
  if (GV->isDeclaration())
    return;
  if (const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D)) {
    if (FD->hasAttr<X86ForceAlignArgPointerAttr>()) {
      llvm::Function *Fn = cast<llvm::Function>(GV);
      Fn->addFnAttr("stackrealign");
    }

    addX86InterruptAttrs(FD, GV, CGM);
  }
}

bool X86_32TargetCodeGenInfo::initDwarfEHRegSizeTable(
                                               CodeGen::CodeGenFunction &CGF,
                                               llvm::Value *Address) const {
  CodeGen::CGBuilderTy &Builder = CGF.Builder;

  llvm::Value *Four8 = llvm::ConstantInt::get(CGF.Int8Ty, 4);

  // 0-7 are the eight integer registers;  the order is different
  //   on Darwin (for EH), but the range is the same.
  // 8 is %eip.
  AssignToArrayRange(Builder, Address, Four8, 0, 8);

  if (CGF.CGM.getTarget().getTriple().isOSDarwin()) {
    // 12-16 are st(0..4).  Not sure why we stop at 4.
    // These have size 16, which is sizeof(long double) on
    // platforms with 8-byte alignment for that type.
    llvm::Value *Sixteen8 = llvm::ConstantInt::get(CGF.Int8Ty, 16);
    AssignToArrayRange(Builder, Address, Sixteen8, 12, 16);

  } else {
    // 9 is %eflags, which doesn't get a size on Darwin for some
    // reason.
    Builder.CreateAlignedStore(
        Four8, Builder.CreateConstInBoundsGEP1_32(CGF.Int8Ty, Address, 9),
                               CharUnits::One());

    // 11-16 are st(0..5).  Not sure why we stop at 5.
    // These have size 12, which is sizeof(long double) on
    // platforms with 4-byte alignment for that type.
    llvm::Value *Twelve8 = llvm::ConstantInt::get(CGF.Int8Ty, 12);
    AssignToArrayRange(Builder, Address, Twelve8, 11, 16);
  }

  return false;
}

//===----------------------------------------------------------------------===//
// X86-64 ABI Implementation
//===----------------------------------------------------------------------===//


namespace {
/// The AVX ABI level for X86 targets.
enum class X86AVXABILevel {
  None,
  AVX,
  AVX512
};

/// \p returns the size in bits of the largest (native) vector for \p AVXLevel.
static unsigned getNativeVectorSizeForAVXABI(X86AVXABILevel AVXLevel) {
  switch (AVXLevel) {
  case X86AVXABILevel::AVX512:
    return 512;
  case X86AVXABILevel::AVX:
    return 256;
  case X86AVXABILevel::None:
    return 128;
  }
  llvm_unreachable("Unknown AVXLevel");
}

/// X86_64ABIInfo - The X86_64 ABI information.
class X86_64ABIInfo : public SwiftABIInfo {
  enum Class {
    Integer = 0,
    SSE,
    SSEUp,
    X87,
    X87Up,
    ComplexX87,
    NoClass,
    Memory
  };

  /// merge - Implement the X86_64 ABI merging algorithm.
  ///
  /// Merge an accumulating classification \arg Accum with a field
  /// classification \arg Field.
  ///
  /// \param Accum - The accumulating classification. This should
  /// always be either NoClass or the result of a previous merge
  /// call. In addition, this should never be Memory (the caller
  /// should just return Memory for the aggregate).
  static Class merge(Class Accum, Class Field);

  /// postMerge - Implement the X86_64 ABI post merging algorithm.
  ///
  /// Post merger cleanup, reduces a malformed Hi and Lo pair to
  /// final MEMORY or SSE classes when necessary.
  ///
  /// \param AggregateSize - The size of the current aggregate in
  /// the classification process.
  ///
  /// \param Lo - The classification for the parts of the type
  /// residing in the low word of the containing object.
  ///
  /// \param Hi - The classification for the parts of the type
  /// residing in the higher words of the containing object.
  ///
  void postMerge(unsigned AggregateSize, Class &Lo, Class &Hi) const;

  /// classify - Determine the x86_64 register classes in which the
  /// given type T should be passed.
  ///
  /// \param Lo - The classification for the parts of the type
  /// residing in the low word of the containing object.
  ///
  /// \param Hi - The classification for the parts of the type
  /// residing in the high word of the containing object.
  ///
  /// \param OffsetBase - The bit offset of this type in the
  /// containing object.  Some parameters are classified different
  /// depending on whether they straddle an eightbyte boundary.
  ///
  /// \param isNamedArg - Whether the argument in question is a "named"
  /// argument, as used in AMD64-ABI 3.5.7.
  ///
  /// If a word is unused its result will be NoClass; if a type should
  /// be passed in Memory then at least the classification of \arg Lo
  /// will be Memory.
  ///
  /// The \arg Lo class will be NoClass iff the argument is ignored.
  ///
  /// If the \arg Lo class is ComplexX87, then the \arg Hi class will
  /// also be ComplexX87.
  void classify(QualType T, uint64_t OffsetBase, Class &Lo, Class &Hi,
                bool isNamedArg) const;

  llvm::Type *GetByteVectorType(QualType Ty) const;
  llvm::Type *GetSSETypeAtOffset(llvm::Type *IRType,
                                 unsigned IROffset, QualType SourceTy,
                                 unsigned SourceOffset) const;
  llvm::Type *GetINTEGERTypeAtOffset(llvm::Type *IRType,
                                     unsigned IROffset, QualType SourceTy,
                                     unsigned SourceOffset) const;

  /// getIndirectResult - Give a source type \arg Ty, return a suitable result
  /// such that the argument will be returned in memory.
  ABIArgInfo getIndirectReturnResult(QualType Ty) const;

  /// getIndirectResult - Give a source type \arg Ty, return a suitable result
  /// such that the argument will be passed in memory.
  ///
  /// \param freeIntRegs - The number of free integer registers remaining
  /// available.
  ABIArgInfo getIndirectResult(QualType Ty, unsigned freeIntRegs) const;

  ABIArgInfo classifyReturnType(QualType RetTy) const;

  ABIArgInfo classifyArgumentType(QualType Ty, unsigned freeIntRegs,
                                  unsigned &neededInt, unsigned &neededSSE,
                                  bool isNamedArg) const;

  ABIArgInfo classifyRegCallStructType(QualType Ty, unsigned &NeededInt,
                                       unsigned &NeededSSE) const;

  ABIArgInfo classifyRegCallStructTypeImpl(QualType Ty, unsigned &NeededInt,
                                           unsigned &NeededSSE) const;

  bool IsIllegalVectorType(QualType Ty) const;

  /// The 0.98 ABI revision clarified a lot of ambiguities,
  /// unfortunately in ways that were not always consistent with
  /// certain previous compilers.  In particular, platforms which
  /// required strict binary compatibility with older versions of GCC
  /// may need to exempt themselves.
  bool honorsRevision0_98() const {
    return !getTarget().getTriple().isOSDarwin();
  }

  /// GCC classifies <1 x long long> as SSE but some platform ABIs choose to
  /// classify it as INTEGER (for compatibility with older clang compilers).
  bool classifyIntegerMMXAsSSE() const {
    // Clang <= 3.8 did not do this.
    if (getContext().getLangOpts().getClangABICompat() <=
        LangOptions::ClangABI::Ver3_8)
      return false;

    const llvm::Triple &Triple = getTarget().getTriple();
    if (Triple.isOSDarwin() || Triple.getOS() == llvm::Triple::PS464)
      return false;
    if (Triple.isOSFreeBSD() && Triple.getOSMajorVersion() >= 100)
      return false;
    return true;
  }

  // GCC classifies vectors of __int128 as memory.
  bool passInt128VectorsInMem() const {
    // Clang <= 9.0 did not do this.
    if (getContext().getLangOpts().getClangABICompat() <=
        LangOptions::ClangABI::Ver9)
      return false;

    const llvm::Triple &T = getTarget().getTriple();
    return T.isOSLinux() || T.isOSNetBSD()9.97k;
  }

  X86AVXABILevel AVXLevel;
  // Some ABIs (e.g. X32 ABI and Native Client OS) use 32 bit pointers on
  // 64-bit hardware.
  bool Has64BitPointers;

public:
  X86_64ABIInfo(CodeGen::CodeGenTypes &CGT, X86AVXABILevel AVXLevel) :
      SwiftABIInfo(CGT), AVXLevel(AVXLevel),
      Has64BitPointers(CGT.getDataLayout().getPointerSize(0) == 8) {
  }

  bool isPassedUsingAVXType(QualType type) const {
    unsigned neededInt, neededSSE;
    // The freeIntRegs argument doesn't matter here.
    ABIArgInfo info = classifyArgumentType(type, 0, neededInt, neededSSE,
                                           /*isNamedArg*/true);
    if (info.isDirect()) {
      llvm::Type *ty = info.getCoerceToType();
      if (llvm::VectorType *vectorTy = dyn_cast_or_null<llvm::VectorType>(ty))
        return vectorTy->getPrimitiveSizeInBits().getFixedSize() > 128;
    }
    return false;
  }

  void computeInfo(CGFunctionInfo &FI) const override;

  Address EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                    QualType Ty) const override;
  Address EmitMSVAArg(CodeGenFunction &CGF, Address VAListAddr,
                      QualType Ty) const override;

  bool has64BitPointers() const {
    return Has64BitPointers;
  }

  bool shouldPassIndirectlyForSwift(ArrayRef<llvm::Type*> scalars,
                                    bool asReturnValue) const override {
    return occupiesMoreThan(CGT, scalars, /*total*/ 4);
  }
  bool isSwiftErrorInRegister() const override {
    return true;
  }
};

/// WinX86_64ABIInfo - The Windows X86_64 ABI information.
class WinX86_64ABIInfo : public SwiftABIInfo {
public:
  WinX86_64ABIInfo(CodeGen::CodeGenTypes &CGT, X86AVXABILevel AVXLevel)
      : SwiftABIInfo(CGT), AVXLevel(AVXLevel),
        IsMingw64(getTarget().getTriple().isWindowsGNUEnvironment()) {}

  void computeInfo(CGFunctionInfo &FI) const override;

  Address EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                    QualType Ty) const override;

  bool isHomogeneousAggregateBaseType(QualType Ty) const override {
    // FIXME: Assumes vectorcall is in use.
    return isX86VectorTypeForVectorCall(getContext(), Ty);
  }

  bool isHomogeneousAggregateSmallEnough(const Type *Ty,
                                         uint64_t NumMembers) const override {
    // FIXME: Assumes vectorcall is in use.
    return isX86VectorCallAggregateSmallEnough(NumMembers);
  }

  bool shouldPassIndirectlyForSwift(ArrayRef<llvm::Type *> scalars,
                                    bool asReturnValue) const override {
    return occupiesMoreThan(CGT, scalars, /*total*/ 4);
  }

  bool isSwiftErrorInRegister() const override {
    return true;
  }

private:
  ABIArgInfo classify(QualType Ty, unsigned &FreeSSERegs, bool IsReturnType,
                      bool IsVectorCall, bool IsRegCall) const;
  ABIArgInfo reclassifyHvaArgForVectorCall(QualType Ty, unsigned &FreeSSERegs,
                                           const ABIArgInfo &current) const;

  X86AVXABILevel AVXLevel;

  bool IsMingw64;
};

class X86_64TargetCodeGenInfo : public TargetCodeGenInfo {
public:
  X86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, X86AVXABILevel AVXLevel)
      : TargetCodeGenInfo(std::make_unique<X86_64ABIInfo>(CGT, AVXLevel)) {}

  const X86_64ABIInfo &getABIInfo() const {
    return static_cast<const X86_64ABIInfo&>(TargetCodeGenInfo::getABIInfo());
  }

  /// Disable tail call on x86-64. The epilogue code before the tail jump blocks
  /// autoreleaseRV/retainRV and autoreleaseRV/unsafeClaimRV optimizations.
  bool markARCOptimizedReturnCallsAsNoTail() const override { return true; }

  int getDwarfEHStackPointer(CodeGen::CodeGenModule &CGM) const override {
    return 7;
  }

  bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                               llvm::Value *Address) const override {
    llvm::Value *Eight8 = llvm::ConstantInt::get(CGF.Int8Ty, 8);

    // 0-15 are the 16 integer registers.
    // 16 is %rip.
    AssignToArrayRange(CGF.Builder, Address, Eight8, 0, 16);
    return false;
  }

  llvm::Type* adjustInlineAsmType(CodeGen::CodeGenFunction &CGF,
                                  StringRef Constraint,
                                  llvm::Type* Ty) const override {
    return X86AdjustInlineAsmType(CGF, Constraint, Ty);
  }

  bool isNoProtoCallVariadic(const CallArgList &args,
                             const FunctionNoProtoType *fnType) const override {
    // The default CC on x86-64 sets %al to the number of SSA
    // registers used, and GCC sets this when calling an unprototyped
    // function, so we override the default behavior.  However, don't do
    // that when AVX types are involved: the ABI explicitly states it is
    // undefined, and it doesn't work in practice because of how the ABI
    // defines varargs anyway.
    if (fnType->getCallConv() == CC_C) {
      bool HasAVXType = false;
      for (CallArgList::const_iterator
             it = args.begin(), ie = args.end(); it != ie; ++it87) {
        if (getABIInfo().isPassedUsingAVXType(it->Ty)) {
          HasAVXType = true;
          break;
        }
      }

      if (!HasAVXType)
        return true;
    }

    return TargetCodeGenInfo::isNoProtoCallVariadic(args, fnType);
  }

  llvm::Constant *
  getUBSanFunctionSignature(CodeGen::CodeGenModule &CGM) const override {
    unsigned Sig = (0xeb << 0) | // jmp rel8
                   (0x06 << 8) | //           .+0x08
                   ('v' << 16) |
                   ('2' << 24);
    return llvm::ConstantInt::get(CGM.Int32Ty, Sig);
  }

  void setTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
                           CodeGen::CodeGenModule &CGM) const override {
    if (GV->isDeclaration())
      return;
    if (const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D)) {
      if (FD->hasAttr<X86ForceAlignArgPointerAttr>()) {
        llvm::Function *Fn = cast<llvm::Function>(GV);
        Fn->addFnAttr("stackrealign");
      }

      addX86InterruptAttrs(FD, GV, CGM);
    }
  }

  void checkFunctionCallABI(CodeGenModule &CGM, SourceLocation CallLoc,
                            const FunctionDecl *Caller,
                            const FunctionDecl *Callee,
                            const CallArgList &Args) const override;
};

static void initFeatureMaps(const ASTContext &Ctx,
                            llvm::StringMap<bool> &CallerMap,
                            const FunctionDecl *Caller,
                            llvm::StringMap<bool> &CalleeMap,
                            const FunctionDecl *Callee) {
  if (CalleeMap.empty() && CallerMap.empty()5.55k) {
    // The caller is potentially nullptr in the case where the call isn't in a
    // function.  In this case, the getFunctionFeatureMap ensures we just get
    // the TU level setting (since it cannot be modified by 'target'..
    Ctx.getFunctionFeatureMap(CallerMap, Caller);
    Ctx.getFunctionFeatureMap(CalleeMap, Callee);
  }
}

static bool checkAVXParamFeature(DiagnosticsEngine &Diag,
                                 SourceLocation CallLoc,
                                 const llvm::StringMap<bool> &CallerMap,
                                 const llvm::StringMap<bool> &CalleeMap,
                                 QualType Ty, StringRef Feature,
                                 bool IsArgument) {
  bool CallerHasFeat = CallerMap.lookup(Feature);
  bool CalleeHasFeat = CalleeMap.lookup(Feature);
  if (!CallerHasFeat && !CalleeHasFeat62)
    return Diag.Report(CallLoc, diag::warn_avx_calling_convention)
           << IsArgument << Ty << Feature;

  // Mixing calling conventions here is very clearly an error.
  if (!CallerHasFeat || !CalleeHasFeat12.0k)
    return Diag.Report(CallLoc, diag::err_avx_calling_convention)
           << IsArgument << Ty << Feature;

  // Else, both caller and callee have the required feature, so there is no need
  // to diagnose.
  return false;
}

static bool checkAVXParam(DiagnosticsEngine &Diag, ASTContext &Ctx,
                          SourceLocation CallLoc,
                          const llvm::StringMap<bool> &CallerMap,
                          const llvm::StringMap<bool> &CalleeMap, QualType Ty,
                          bool IsArgument) {
  uint64_t Size = Ctx.getTypeSize(Ty);
  if (Size > 256)
    return checkAVXParamFeature(Diag, CallLoc, CallerMap, CalleeMap, Ty,
                                "avx512f", IsArgument);

  if (Size > 128)
    return checkAVXParamFeature(Diag, CallLoc, CallerMap, CalleeMap, Ty, "avx",
                                IsArgument);

  return false;
}

void X86_64TargetCodeGenInfo::checkFunctionCallABI(
    CodeGenModule &CGM, SourceLocation CallLoc, const FunctionDecl *Caller,
    const FunctionDecl *Callee, const CallArgList &Args) const {
  llvm::StringMap<bool> CallerMap;
  llvm::StringMap<bool> CalleeMap;
  unsigned ArgIndex = 0;

  // We need to loop through the actual call arguments rather than the the
  // function's parameters, in case this variadic.
  for (const CallArg &Arg : Args) {
    // The "avx" feature changes how vectors >128 in size are passed. "avx512f"
    // additionally changes how vectors >256 in size are passed. Like GCC, we
    // warn when a function is called with an argument where this will change.
    // Unlike GCC, we also error when it is an obvious ABI mismatch, that is,
    // the caller and callee features are mismatched.
    // Unfortunately, we cannot do this diagnostic in SEMA, since the callee can
    // change its ABI with attribute-target after this call.
    if (Arg.getType()->isVectorType() &&
        CGM.getContext().getTypeSize(Arg.getType()) > 128) {
      initFeatureMaps(CGM.getContext(), CallerMap, Caller, CalleeMap, Callee);
      QualType Ty = Arg.getType();
      // The CallArg seems to have desugared the type already, so for clearer
      // diagnostics, replace it with the type in the FunctionDecl if possible.
      if (ArgIndex < Callee->getNumParams())
        Ty = Callee->getParamDecl(ArgIndex)->getType();

      if (checkAVXParam(CGM.getDiags(), CGM.getContext(), CallLoc, CallerMap,
                        CalleeMap, Ty, /*IsArgument*/ true))
        return;
    }
    ++ArgIndex;
  }

  // Check return always, as we don't have a good way of knowing in codegen
  // whether this value is used, tail-called, etc.
  if (Callee->getReturnType()->isVectorType() &&
      CGM.getContext().getTypeSize(Callee->getReturnType()) > 128) {
    initFeatureMaps(CGM.getContext(), CallerMap, Caller, CalleeMap, Callee);
    checkAVXParam(CGM.getDiags(), CGM.getContext(), CallLoc, CallerMap,
                  CalleeMap, Callee->getReturnType(),
                  /*IsArgument*/ false);
  }
}

static std::string qualifyWindowsLibrary(llvm::StringRef Lib) {
  // If the argument does not end in .lib, automatically add the suffix.
  // If the argument contains a space, enclose it in quotes.
  // This matches the behavior of MSVC.
  bool Quote = (Lib.find(' ') != StringRef::npos);
  std::string ArgStr = Quote ? "\""4 : "";
  ArgStr += Lib;
  if (!Lib.endswith_lower(".lib") && !Lib.endswith_lower(".a")63)
    ArgStr += ".lib";
  ArgStr += Quote ? "\""4 : "";
  return ArgStr;
}

class WinX86_32TargetCodeGenInfo : public X86_32TargetCodeGenInfo {
public:
  WinX86_32TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT,
        bool DarwinVectorABI, bool RetSmallStructInRegABI, bool Win32StructABI,
        unsigned NumRegisterParameters)
    : X86_32TargetCodeGenInfo(CGT, DarwinVectorABI, RetSmallStructInRegABI,
        Win32StructABI, NumRegisterParameters, false) {}

  void setTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
                           CodeGen::CodeGenModule &CGM) const override;

  void getDependentLibraryOption(llvm::StringRef Lib,
                                 llvm::SmallString<24> &Opt) const override {
    Opt = "/DEFAULTLIB:";
    Opt += qualifyWindowsLibrary(Lib);
  }

  void getDetectMismatchOption(llvm::StringRef Name,
                               llvm::StringRef Value,
                               llvm::SmallString<32> &Opt) const override {
    Opt = "/FAILIFMISMATCH:\"" + Name.str() + "=" + Value.str() + "\"";
  }
};

static void addStackProbeTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
                                          CodeGen::CodeGenModule &CGM) {
  if (llvm::Function *Fn = dyn_cast_or_null<llvm::Function>(GV)) {

    if (CGM.getCodeGenOpts().StackProbeSize != 4096)
      Fn->addFnAttr("stack-probe-size",
                    llvm::utostr(CGM.getCodeGenOpts().StackProbeSize));
    if (CGM.getCodeGenOpts().NoStackArgProbe)
      Fn->addFnAttr("no-stack-arg-probe");
  }
}

void WinX86_32TargetCodeGenInfo::setTargetAttributes(
    const Decl *D, llvm::GlobalValue *GV, CodeGen::CodeGenModule &CGM) const {
  X86_32TargetCodeGenInfo::setTargetAttributes(D, GV, CGM);
  if (GV->isDeclaration())
    return;
  addStackProbeTargetAttributes(D, GV, CGM);
}

class WinX86_64TargetCodeGenInfo : public TargetCodeGenInfo {
public:
  WinX86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT,
                             X86AVXABILevel AVXLevel)
      : TargetCodeGenInfo(std::make_unique<WinX86_64ABIInfo>(CGT, AVXLevel)) {}

  void setTargetAttributes(const Decl *D, llvm::GlobalValue *GV,
                           CodeGen::CodeGenModule &CGM) const override;

  int getDwarfEHStackPointer(CodeGen::CodeGenModule &CGM) const override {
    return 7;
  }

  bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                               llvm::Value *Address) const override {
    llvm::Value *Eight8 = llvm::ConstantInt::get(CGF.Int8Ty, 8);

    // 0-15 are the 16 integer registers.
    // 16 is %rip.
    AssignToArrayRange(CGF.Builder, Address, Eight8, 0, 16);
    return false;
  }

  void getDependentLibraryOption(llvm::StringRef Lib,
                                 llvm::SmallString<24> &Opt) const override {
    Opt = "/DEFAULTLIB:";
    Opt += qualifyWindowsLibrary(Lib);
  }

  void getDetectMismatchOption(llvm::StringRef Name,
                               llvm::StringRef Value,
                               llvm::SmallString<32> &Opt) const override {
    Opt = "/FAILIFMISMATCH:\"" + Name.str() + "=" + Value.str() + "\"";
  }
};

void WinX86_64TargetCodeGenInfo::setTargetAttributes(
    const Decl *D, llvm::GlobalValue *GV, CodeGen::CodeGenModule &CGM) const {
  TargetCodeGenInfo::setTargetAttributes(D, GV, CGM);
  if (GV->isDeclaration())
    return;
  if (const FunctionDecl *FD = dyn_cast_or_null<FunctionDecl>(D)) {
    if (FD->hasAttr<X86ForceAlignArgPointerAttr>()) {
      llvm::Function *Fn = cast<llvm::Function>(GV);
      Fn->addFnAttr("stackrealign");
    }

    addX86InterruptAttrs(FD, GV, CGM);
  }

  addStackProbeTargetAttributes(D, GV, CGM);
}
}

void X86_64ABIInfo::postMerge(unsigned AggregateSize, Class &Lo,
                              Class &Hi) const {
  // AMD64-ABI 3.2.3p2: Rule 5. Then a post merger cleanup is done:
  //
  // (a) If one of the classes is Memory, the whole argument is passed in
  //     memory.
  //
  // (b) If X87UP is not preceded by X87, the whole argument is passed in
  //     memory.
  //
  // (c) If the size of the aggregate exceeds two eightbytes and the first
  //     eightbyte isn't SSE or any other eightbyte isn't SSEUP, the whole
  //     argument is passed in memory. NOTE: This is necessary to keep the
  //     ABI working for processors that don't support the __m256 type.
  //
  // (d) If SSEUP is not preceded by SSE or SSEUP, it is converted to SSE.
  //
  // Some of these are enforced by the merging logic.  Others can arise
  // only with unions; for example:
  //   union { _Complex double; unsigned; }
  //
  // Note that clauses (b) and (c) were added in 0.98.
  //
  if (Hi == Memory)
    Lo = Memory;
  if (Hi == X87Up && Lo != X87348 && honorsRevision0_98()17)
    Lo = Memory;
  if (AggregateSize > 128 && (323Lo != SSE323 || Hi != SSEUp24))
    Lo = Memory;
  if (Hi == SSEUp && Lo != SSE33)
    Hi = SSE;
}

X86_64ABIInfo::Class X86_64ABIInfo::merge(Class Accum, Class Field) {
  // AMD64-ABI 3.2.3p2: Rule 4. Each field of an object is
  // classified recursively so that always two fields are
  // considered. The resulting class is calculated according to
  // the classes of the fields in the eightbyte:
  //
  // (a) If both classes are equal, this is the resulting class.
  //
  // (b) If one of the classes is NO_CLASS, the resulting class is
  // the other class.
  //
  // (c) If one of the classes is MEMORY, the result is the MEMORY
  // class.
  //
  // (d) If one of the classes is INTEGER, the result is the
  // INTEGER.
  //
  // (e) If one of the classes is X87, X87UP, COMPLEX_X87 class,
  // MEMORY is used as class.
  //
  // (f) Otherwise class SSE is used.

  // Accum should never be memory (we should have returned) or
  // ComplexX87 (because this cannot be passed in a structure).
  assert((Accum != Memory && Accum != ComplexX87) &&
         "Invalid accumulated classification during merge.");
  if (Accum == Field || Field == NoClass10.0k)
    return Accum;
  if (Field == Memory)
    return Memory;
  if (Accum == NoClass)
    return Field;
  if (Accum == Integer || Field == Integer73)
    return Integer;
  if (Field == X87 || Field == X87Up || Field == ComplexX87 ||
      Accum == X87 || Accum == X87Up0)
    return Memory;
  return SSE;
}

void X86_64ABIInfo::classify(QualType Ty, uint64_t OffsetBase,
                             Class &Lo, Class &Hi, bool isNamedArg) const {
  // FIXME: This code can be simplified by introducing a simple value class for
  // Class pairs with appropriate constructor methods for the various
  // situations.

  // FIXME: Some of the split computations are wrong; unaligned vectors
  // shouldn't be passed in registers for example, so there is no chance they
  // can straddle an eightbyte. Verify & simplify.

  Lo = Hi = NoClass;

  Class &Current = OffsetBase < 64 ? Lo : Hi1.62k;
  Current = Memory;

  if (const BuiltinType *BT = Ty->getAs<BuiltinType>()) {
    BuiltinType::Kind k = BT->getKind();

    if (k == BuiltinType::Void) {
      Current = NoClass;
    } else if (k == BuiltinType::Int128 || k == BuiltinType::UInt12893.1k) {
      Lo = Integer;
      Hi = Integer;
    } else if (k >= BuiltinType::Bool && k <= BuiltinType::LongLong93.1k) {
      Current = Integer;
    } else if (k == BuiltinType::Float || k == BuiltinType::Double2.93k) {
      Current = SSE;
    } else if (k == BuiltinType::LongDouble) {
      const llvm::fltSemantics *LDF = &getTarget().getLongDoubleFormat();
      if (LDF == &llvm::APFloat::IEEEquad()) {
        Lo = SSE;
        Hi = SSEUp;
      } else if (LDF == &llvm::APFloat::x87DoubleExtended()) {
        Lo = X87;
        Hi = X87Up;
      } else if (LDF == &llvm::APFloat::IEEEdouble()) {
        Current = SSE;
      } else
        llvm_unreachable("unexpected long double representation!");
    }
    // FIXME: _Decimal32 and _Decimal64 are SSE.
    // FIXME: _float128 and _Decimal128 are (SSE, SSEUp).
    return;
  }

  if (const EnumType *ET = Ty->getAs<EnumType>()) {
    // Classify the underlying integer type.
    classify(ET->getDecl()->getIntegerType(), OffsetBase, Lo, Hi, isNamedArg);
    return;
  }

  if (Ty->hasPointerRepresentation()) {
    Current = Integer;
    return;
  }

  if (Ty->isMemberPointerType()) {
    if (Ty->isMemberFunctionPointerType()) {
      if (Has64BitPointers) {
        // If Has64BitPointers, this is an {i64, i64}, so classify both
        // Lo and Hi now.
        Lo = Hi = Integer;
      } else {
        // Otherwise, with 32-bit pointers, this is an {i32, i32}. If that
        // straddles an eightbyte boundary, Hi should be classified as well.
        uint64_t EB_FuncPtr = (OffsetBase) / 64;
        uint64_t EB_ThisAdj = (OffsetBase + 64 - 1) / 64;
        if (EB_FuncPtr != EB_ThisAdj) {
          Lo = Hi = Integer;
        } else {
          Current = Integer;
        }
      }
    } else {
      Current = Integer;
    }
    return;
  }

  if (const VectorType *VT = Ty->getAs<VectorType>()) {
    uint64_t Size = getContext().getTypeSize(VT);
    if (Size == 1 || Size == 8 || Size == 165.87k || Size == 325.87k) {
      // gcc passes the following as integer:
      // 4 bytes - <4 x char>, <2 x short>, <1 x int>, <1 x float>
      // 2 bytes - <2 x char>, <1 x short>
      // 1 byte  - <1 x char>
      Current = Integer;

      // If this type crosses an eightbyte boundary, it should be
      // split.
      uint64_t EB_Lo = (OffsetBase) / 64;
      uint64_t EB_Hi = (OffsetBase + Size - 1) / 64;
      if (EB_Lo != EB_Hi)
        Hi = Lo;
    } else if (Size == 64) {
      QualType ElementType = VT->getElementType();

      // gcc passes <1 x double> in memory. :(
      if (ElementType->isSpecificBuiltinType(BuiltinType::Double))
        return;

      // gcc passes <1 x long long> as SSE but clang used to unconditionally
      // pass them as integer.  For platforms where clang is the de facto
      // platform compiler, we must continue to use integer.
      if (!classifyIntegerMMXAsSSE() &&
          (ElementType->isSpecificBuiltinType(BuiltinType::LongLong) ||
           ElementType->isSpecificBuiltinType(BuiltinType::ULongLong) ||
           ElementType->isSpecificBuiltinType(BuiltinType::Long) ||
           ElementType->isSpecificBuiltinType(BuiltinType::ULong)))
        Current = Integer;
      else
        Current = SSE;

      // If this type crosses an eightbyte boundary, it should be
      // split.
      if (OffsetBase && OffsetBase != 640)
        Hi = Lo;
    } else if (Size == 128 ||
               (3.42kisNamedArg3.42k && Size <= getNativeVectorSizeForAVXABI(AVXLevel)3.39k)) {
      QualType ElementType = VT->getElementType();

      // gcc passes 256 and 512 bit <X x __int128> vectors in memory. :(
      if (passInt128VectorsInMem() && Size != 128516 &&
          (ElementType->isSpecificBuiltinType(BuiltinType::Int128) ||
           ElementType->isSpecificBuiltinType(BuiltinType::UInt128)))
        return;

      // Arguments of 256-bits are split into four eightbyte chunks. The
      // least significant one belongs to class SSE and all the others to class
      // SSEUP. The original Lo and Hi design considers that types can't be
      // greater than 128-bits, so a 64-bit split in Hi and Lo makes sense.
      // This design isn't correct for 256-bits, but since there're no cases
      // where the upper parts would need to be inspected, avoid adding
      // complexity and just consider Hi to match the 64-256 part.
      //
      // Note that per 3.5.7 of AMD64-ABI, 256-bit args are only passed in
      // registers if they are "named", i.e. not part of the "..." of a
      // variadic function.
      //
      // Similarly, per 3.2.3. of the AVX512 draft, 512-bits ("named") args are
      // split into eight eightbyte chunks, one SSE and seven SSEUP.
      Lo = SSE;
      Hi = SSEUp;
    }
    return;
  }

  if (const ComplexType *CT = Ty->getAs<ComplexType>()) {
    QualType ET = getContext().getCanonicalType(CT->getElementType());

    uint64_t Size = getContext().getTypeSize(Ty);
    if (ET->isIntegralOrEnumerationType()) {
      if (Size <= 64)
        Current = Integer;
      else if (Size <= 128)
        Lo = Hi = Integer;
    } else if (ET == getContext().FloatTy) {
      Current = SSE;
    } else if (ET == getContext().DoubleTy) {
      Lo = Hi = SSE;
    } else if (ET == getContext().LongDoubleTy) {
      const llvm::fltSemantics *LDF = &getTarget().getLongDoubleFormat();
      if (LDF == &llvm::APFloat::IEEEquad())
        Current = Memory;
      else if (LDF == &llvm::APFloat::x87DoubleExtended())
        Current = ComplexX87;
      else if (LDF == &llvm::APFloat::IEEEdouble())
        Lo = Hi = SSE;
      else
        llvm_unreachable("unexpected long double representation!");
    }

    // If this complex type crosses an eightbyte boundary then it
    // should be split.
    uint64_t EB_Real = (OffsetBase) / 64;
    uint64_t EB_Imag = (OffsetBase + getContext().getTypeSize(ET)) / 64;
    if (Hi == NoClass && EB_Real != EB_Imag267)
      Hi = Lo;

    return;
  }

  if (const auto *EITy = Ty->getAs<ExtIntType>()) {
    if (EITy->getNumBits() <= 64)
      Current = Integer;
    else if (EITy->getNumBits() <= 128)
      Lo = Hi = Integer;
    // Larger values need to get passed in memory.
    return;
  }

  if (const ConstantArrayType *AT = getContext().getAsConstantArrayType(Ty)) {
    // Arrays are treated like structures.

    uint64_t Size = getContext().getTypeSize(Ty);

    // AMD64-ABI 3.2.3p2: Rule 1. If the size of an object is larger
    // than eight eightbytes, ..., it has class MEMORY.
    if (Size > 512)
      return;

    // AMD64-ABI 3.2.3p2: Rule 1. If ..., or it contains unaligned
    // fields, it has class MEMORY.
    //
    // Only need to check alignment of array base.
    if (OffsetBase % getContext().getTypeAlign(AT->getElementType()))
      return;

    // Otherwise implement simplified merge. We could be smarter about
    // this, but it isn't worth it and would be harder to verify.
    Current = NoClass;
    uint64_t EltSize = getContext().getTypeSize(AT->getElementType());
    uint64_t ArraySize = AT->getSize().getZExtValue();

    // The only case a 256-bit wide vector could be used is when the array
    // contains a single 256-bit element. Since Lo and Hi logic isn't extended
    // to work for sizes wider than 128, early check and fallback to memory.
    //
    if (Size > 128 &&
        (Size != EltSize || Size > getNativeVectorSizeForAVXABI(AVXLevel)4))
      return;

    for (uint64_t i=0, Offset=OffsetBase; 43i<ArraySize; ++i, Offset += EltSize153) {
      Class FieldLo, FieldHi;
      classify(AT->getElementType(), Offset, FieldLo, FieldHi, isNamedArg);
      Lo = merge(Lo, FieldLo);
      Hi = merge(Hi, FieldHi);
      if (Lo == Memory || Hi == Memory)
        break;
    }

    postMerge(Size, Lo, Hi);
    assert((Hi != SSEUp || Lo == SSE) && "Invalid SSEUp array classification.");
    return;
  }

  if (const RecordType *RT = Ty->getAs<RecordType>()) {
    uint64_t Size = getContext().getTypeSize(Ty);

    // AMD64-ABI 3.2.3p2: Rule 1. If the size of an object is larger
    // than eight eightbytes, ..., it has class MEMORY.
    if (Size > 512)
      return;

    // AMD64-ABI 3.2.3p2: Rule 2. If a C++ object has either a non-trivial
    // copy constructor or a non-trivial destructor, it is passed by invisible
    // reference.
    if (getRecordArgABI(RT, getCXXABI()))
      return;

    const RecordDecl *RD = RT->getDecl();

    // Assume variable sized types are passed in memory.
    if (RD->hasFlexibleArrayMember())
      return;

    const ASTRecordLayout &Layout = getContext().getASTRecordLayout(RD);

    // Reset Lo class, this will be recomputed.
    Current = NoClass;

    // If this is a C++ record, classify the bases first.
    if (const CXXRecordDecl *CXXRD = dyn_cast<CXXRecordDecl>(RD)) {
      for (const auto &I : CXXRD->bases()) {
        assert(!I.isVirtual() && !I.getType()->isDependentType() &&
               "Unexpected base class!");
        const auto *Base =
            cast<CXXRecordDecl>(I.getType()->castAs<RecordType>()->getDecl());

        // Classify this field.
        //
        // AMD64-ABI 3.2.3p2: Rule 3. If the size of the aggregate exceeds a
        // single eightbyte, each is classified separately. Each eightbyte gets
        // initialized to class NO_CLASS.
        Class FieldLo, FieldHi;
        uint64_t Offset =
          OffsetBase + getContext().toBits(Layout.getBaseClassOffset(Base));
        classify(I.getType(), Offset, FieldLo, FieldHi, isNamedArg);
        Lo = merge(Lo, FieldLo);
        Hi = merge(Hi, FieldHi);
        if (Lo == Memory || Hi == Memory1.12k) {
          postMerge(Size, Lo, Hi);
          return;
        }
      }
    }

    // Classify the fields one at a time, merging the results.
    unsigned idx = 0;
    bool UseClang11Compat = getContext().getLangOpts().getClangABICompat() <=
                                LangOptions::ClangABI::Ver11 ||
                            getContext().getTargetInfo().getTriple().isPS4();
    bool IsUnion = RT->isUnionType() && !UseClang11Compat61;

    for (RecordDecl::field_iterator i = RD->field_begin(), e = RD->field_end();
           i != e; ++i, ++idx8.20k) {
      uint64_t Offset = OffsetBase + Layout.getFieldOffset(idx);
      bool BitField = i->isBitField();

      // Ignore padding bit-fields.
      if (BitField && i->isUnnamedBitfield()42)
        continue;

      // AMD64-ABI 3.2.3p2: Rule 1. If the size of an object is larger than
      // eight eightbytes, or it contains unaligned fields, it has class MEMORY.
      //
      // The only case a 256-bit or a 512-bit wide vector could be used is when
      // the struct contains a single 256-bit or 512-bit element. Early check
      // and fallback to memory.
      //
      // FIXME: Extended the Lo and Hi logic properly to work for size wider
      // than 128.
      if (Size > 128 &&
          ((!IsUnion && Size != getContext().getTypeSize(i->getType())309) ||
           Size > getNativeVectorSizeForAVXABI(AVXLevel)84)) {
        Lo = Memory;
        postMerge(Size, Lo, Hi);
        return;
      }
      // Note, skip this test for bit-fields, see below.
      if (!BitField && Offset % getContext().getTypeAlign(i->getType())8.19k) {
        Lo = Memory;
        postMerge(Size, Lo, Hi);
        return;
      }

      // Classify this field.
      //
      // AMD64-ABI 3.2.3p2: Rule 3. If the size of the aggregate
      // exceeds a single eightbyte, each is classified
      // separately. Each eightbyte gets initialized to class
      // NO_CLASS.
      Class FieldLo, FieldHi;

      // Bit-fields require special handling, they do not force the
      // structure to be passed in memory even if unaligned, and
      // therefore they can straddle an eightbyte.
      if (BitField) {
        assert(!i->isUnnamedBitfield());
        uint64_t Offset = OffsetBase + Layout.getFieldOffset(idx);
        uint64_t Size = i->getBitWidthValue(getContext());

        uint64_t EB_Lo = Offset / 64;
        uint64_t EB_Hi = (Offset + Size - 1) / 64;

        if (EB_Lo) {
          assert(EB_Hi == EB_Lo && "Invalid classification, type > 16 bytes.");
          FieldLo = NoClass;
          FieldHi = Integer;
        } else {
          FieldLo = Integer;
          FieldHi = EB_Hi ? Integer0 : NoClass;
        }
      } else
        classify(i->getType(), Offset, FieldLo, FieldHi, isNamedArg);
      Lo = merge(Lo, FieldLo);
      Hi = merge(Hi, FieldHi);
      if (Lo == Memory || Hi == Memory8.19k)
        break;
    }

    postMerge(Size, Lo, Hi);
  }
}

ABIArgInfo X86_64ABIInfo::getIndirectReturnResult(QualType Ty) const {
  // If this is a scalar LLVM value then assume LLVM will pass it in the right
  // place naturally.
  if (!isAggregateTypeForABI(Ty)) {
    // Treat an enum type as its underlying type.
    if (const EnumType *EnumTy = Ty->getAs<EnumType>())
      Ty = EnumTy->getDecl()->getIntegerType();

    if (Ty->isExtIntType())
      return getNaturalAlignIndirect(Ty);

    return (isPromotableIntegerTypeForABI(Ty) ? ABIArgInfo::getExtend(Ty)0
                                              : ABIArgInfo::getDirect());
  }

  return getNaturalAlignIndirect(Ty);
}

bool X86_64ABIInfo::IsIllegalVectorType(QualType Ty) const {
  if (const VectorType *VecTy = Ty->getAs<VectorType>()) {
    uint64_t Size = getContext().getTypeSize(VecTy);
    unsigned LargestVector = getNativeVectorSizeForAVXABI(AVXLevel);
    if (Size <= 64 || Size > LargestVector132)
      return true;
    QualType EltTy = VecTy->getElementType();
    if (passInt128VectorsInMem() &&
        (EltTy->isSpecificBuiltinType(BuiltinType::Int128) ||
         EltTy->isSpecificBuiltinType(BuiltinType::UInt128)))
      return true;
  }

  return false;
}

ABIArgInfo X86_64ABIInfo::getIndirectResult(QualType Ty,
                                            unsigned freeIntRegs) const {
  // If this is a scalar LLVM value then assume LLVM will pass it in the right
  // place naturally.
  //
  // This assumption is optimistic, as there could be free registers available
  // when we need to pass this argument in memory, and LLVM could try to pass
  // the argument in the free register. This does not seem to happen currently,
  // but this code would be much safer if we could mark the argument with
  // 'onstack'. See PR12193.
  if (!isAggregateTypeForABI(Ty) && !IsIllegalVectorType(Ty)2.78k &&
      !Ty->isExtIntType()) {
    // Treat an enum type as its underlying type.
    if (const EnumType *EnumTy = Ty->getAs<EnumType>())
      Ty = EnumTy->getDecl()->getIntegerType();

    return (isPromotableIntegerTypeForABI(Ty) ? ABIArgInfo::getExtend(Ty)
                                              : ABIArgInfo::getDirect());
  }

  if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI()))
    return getNaturalAlignIndirect(Ty, RAA == CGCXXABI::RAA_DirectInMemory);

  // Compute the byval alignment. We specify the alignment of the byval in all
  // cases so that the mid-level optimizer knows the alignment of the byval.
  unsigned Align = std::max(getContext().getTypeAlign(Ty) / 8, 8U);

  // Attempt to avoid passing indirect results using byval when possible. This
  // is important for good codegen.
  //
  // We do this by coercing the value into a scalar type which the backend can
  // handle naturally (i.e., without using byval).
  //
  // For simplicity, we currently only do this when we have exhausted all of the
  // free integer registers. Doing this when there are free integer registers
  // would require more care, as we would have to ensure that the coerced value
  // did not claim the unused register. That would require either reording the
  // arguments to the function (so that any subsequent inreg values came first),
  // or only doing this optimization when there were no following arguments that
  // might be inreg.
  //
  // We currently expect it to be rare (particularly in well written code) for
  // arguments to be passed on the stack when there are still free integer
  // registers available (this would typically imply large structs being passed
  // by value), so this seems like a fair tradeoff for now.
  //
  // We can revisit this if the backend grows support for 'onstack' parameter
  // attributes. See PR12193.
  if (freeIntRegs == 0) {
    uint64_t Size = getContext().getTypeSize(Ty);

    // If this type fits in an eightbyte, coerce it into the matching integral
    // type, which will end up on the stack (with alignment 8).
    if (Align == 8 && Size <= 6422)
      return ABIArgInfo::getDirect(llvm::IntegerType::get(getVMContext(),
                                                          Size));
  }

  return ABIArgInfo::getIndirect(CharUnits::fromQuantity(Align));
}

/// The ABI specifies that a value should be passed in a full vector XMM/YMM
/// register. Pick an LLVM IR type that will be passed as a vector register.
llvm::Type *X86_64ABIInfo::GetByteVectorType(QualType Ty) const {
  // Wrapper structs/arrays that only contain vectors are passed just like
  // vectors; strip them off if present.
  if (const Type *InnerTy = isSingleElementStruct(Ty, getContext()))
    Ty = QualType(InnerTy, 0);

  llvm::Type *IRType = CGT.ConvertType(Ty);
  if (isa<llvm::VectorType>(IRType)) {
    // Don't pass vXi128 vectors in their native type, the backend can't
    // legalize them.
    if (passInt128VectorsInMem() &&
        cast<llvm::VectorType>(IRType)->getElementType()->isIntegerTy(128)) {
      // Use a vXi64 vector.
      uint64_t Size = getContext().getTypeSize(Ty);
      return llvm::FixedVectorType::get(llvm::Type::getInt64Ty(getVMContext()),
                                        Size / 64);
    }

    return IRType;
  }

  if (IRType->getTypeID() == llvm::Type::FP128TyID)
    return IRType;

  // We couldn't find the preferred IR vector type for 'Ty'.
  uint64_t Size = getContext().getTypeSize(Ty);
  assert((Size == 128 || Size == 256 || Size == 512) && "Invalid type found!");


  // Return a LLVM IR vector type based on the size of 'Ty'.
  return llvm::FixedVectorType::get(llvm::Type::getDoubleTy(getVMContext()),
                                    Size / 64);
}

/// BitsContainNoUserData - Return true if the specified [start,end) bit range
/// is known to either be off the end of the specified type or being in
/// alignment padding.  The user type specified is known to be at most 128 bits
/// in size, and have passed through X86_64ABIInfo::classify with a successful
/// classification that put one of the two halves in the INTEGER class.
///
/// It is conservatively correct to return false.
static bool BitsContainNoUserData(QualType Ty, unsigned StartBit,
                                  unsigned EndBit, ASTContext &Context) {
  // If the bytes being queried are off the end of the type, there is no user
  // data hiding here.  This handles analysis of builtins, vectors and other
  // types that don't contain interesting padding.
  unsigned TySize = (unsigned)Context.getTypeSize(Ty);
  if (TySize <= StartBit)
    return true;

  if (const ConstantArrayType *AT = Context.getAsConstantArrayType(Ty)) {
    unsigned EltSize = (unsigned)Context.getTypeSize(AT->getElementType());
    unsigned NumElts = (unsigned)AT->getSize().getZExtValue();

    // Check each element to see if the element overlaps with the queried range.
    for (unsigned i = 0; i != NumElts; ++i57) {
      // If the element is after the span we care about, then we're done..
      unsigned EltOffset = i*EltSize;
      if (EltOffset >= EndBit) break0;

      unsigned EltStart = EltOffset < StartBit ? StartBit-EltOffset64 :022;
      if (!BitsContainNoUserData(AT->getElementType(), EltStart,
                                 EndBit-EltOffset, Context))
        return false;
    }
    // If it overlaps no elements, then it is safe to process as padding.
    return true;
  }

  if (const RecordType *RT = Ty->getAs<RecordType>()) {
    const RecordDecl *RD = RT->getDecl();
    const ASTRecordLayout &Layout = Context.getASTRecordLayout(RD);

    // If this is a C++ record, check the bases first.
    if (const CXXRecordDecl *CXXRD = dyn_cast<CXXRecordDecl>(RD)) {
      for (const auto &I : CXXRD->bases()) {
        assert(!I.isVirtual() && !I.getType()->isDependentType() &&
               "Unexpected base class!");
        const auto *Base =
            cast<CXXRecordDecl>(I.getType()->castAs<RecordType>()->getDecl());

        // If the base is after the span we care about, ignore it.
        unsigned BaseOffset = Context.toBits(Layout.getBaseClassOffset(Base));
        if (BaseOffset >= EndBit) continue0;

        unsigned BaseStart = BaseOffset < StartBit ? StartBit-BaseOffset26 :03;
        if (!BitsContainNoUserData(I.getType(), BaseStart,
                                   EndBit-BaseOffset, Context))
          return false;
      }
    }

    // Verify that no field has data that overlaps the region of interest.  Yes
    // this could be sped up a lot by being smarter about queried fields,
    // however we're only looking at structs up to 16 bytes, so we don't care
    // much.
    unsigned idx = 0;
    for (RecordDecl::field_iterator i = RD->field_begin(), e = RD->field_end();
         i != e; ++i, ++idx1.41k) {
      unsigned FieldOffset = (unsigned)Layout.getFieldOffset(idx);

      // If we found a field after the region we care about, then we're done.
      if (FieldOffset >= EndBit) break140;

      unsigned FieldStart = FieldOffset < StartBit ? StartBit-FieldOffset1.67k :0587;
      if (!BitsContainNoUserData(i->getType(), FieldStart, EndBit-FieldOffset,
                                 Context))
        return false;
    }

    // If nothing in this record overlapped the area of interest, then we're
    // clean.
    return true;
  }

  return false;
}

/// ContainsFloatAtOffset - Return true if the specified LLVM IR type has a
/// float member at the specified offset.  For example, {int,{float}} has a
/// float at offset 4.  It is conservatively correct for this routine to return
/// false.
static bool ContainsFloatAtOffset(llvm::Type *IRType, unsigned IROffset,
                                  const llvm::DataLayout &TD) {
  // Base case if we find a float.
  if (IROffset == 0 && IRType->isFloatTy()8.63k)
    return true;

  // If this is a struct, recurse into the field at the specified offset.
  if (llvm::StructType *STy = dyn_cast<llvm::StructType>(IRType)) {
    const llvm::StructLayout *SL = TD.getStructLayout(STy);
    unsigned Elt = SL->getElementContainingOffset(IROffset);
    IROffset -= SL->getElementOffset(Elt);
    return ContainsFloatAtOffset(STy->getElementType(Elt), IROffset, TD);
  }

  // If this is an array, recurse into the field at the specified offset.
  if (llvm::ArrayType *ATy = dyn_cast<llvm::ArrayType>(IRType)) {
    llvm::Type *EltTy = ATy->getElementType();
    unsigned EltSize = TD.getTypeAllocSize(EltTy);
    IROffset -= IROffset/EltSize*EltSize;
    return ContainsFloatAtOffset(EltTy, IROffset, TD);
  }

  return false;
}


/// GetSSETypeAtOffset - Return a type that will be passed by the backend in the
/// low 8 bytes of an XMM register, corresponding to the SSE class.
llvm::Type *X86_64ABIInfo::
GetSSETypeAtOffset(llvm::Type *IRType, unsigned IROffset,
                   QualType SourceTy, unsigned SourceOffset) const {
  // The only three choices we have are either double, <2 x float>, or float. We
  // pass as float if the last 4 bytes is just padding.  This happens for
  // structs that contain 3 floats.
  if (BitsContainNoUserData(SourceTy, SourceOffset*8+32,
                            SourceOffset*8+64, getContext()))
    return llvm::Type::getFloatTy(getVMContext());

  // We want to pass as <2 x float> if the LLVM IR type contains a float at
  // offset+0 and offset+4.  Walk the LLVM IR type to find out if this is the
  // case.
  if (ContainsFloatAtOffset(IRType, IROffset, getDataLayout()) &&
      ContainsFloatAtOffset(IRType, IROffset+4, getDataLayout()))
    return llvm::FixedVectorType::get(llvm::Type::getFloatTy(getVMContext()),
                                      2);

  return llvm::Type::getDoubleTy(getVMContext());
}


/// GetINTEGERTypeAtOffset - The ABI specifies that a value should be passed in
/// an 8-byte GPR.  This means that we either have a scalar or we are talking
/// about the high or low part of an up-to-16-byte struct.  This routine picks
/// the best LLVM IR type to represent this, which may be i64 or may be anything
/// else that the backend will pass in a GPR that works better (e.g. i8, %foo*,
/// etc).
///
/// PrefType is an LLVM IR type that corresponds to (part of) the IR type for
/// the source type.  IROffset is an offset in bytes into the LLVM IR type that
/// the 8-byte value references.  PrefType may be null.
///
/// SourceTy is the source-level type for the entire argument.  SourceOffset is
/// an offset into this that we're processing (which is always either 0 or 8).
///
llvm::Type *X86_64ABIInfo::
GetINTEGERTypeAtOffset(llvm::Type *IRType, unsigned IROffset,
                       QualType SourceTy, unsigned SourceOffset) const {
  // If we're dealing with an un-offset LLVM IR type, then it means that we're
  // returning an 8-byte unit starting with it.  See if we can safely use it.
  if (IROffset == 0) {
    // Pointers and int64's always fill the 8-byte unit.
    if ((isa<llvm::PointerType>(IRType) && Has64BitPointers223k) ||
        IRType->isIntegerTy(64))
      return IRType;

    // If we have a 1/2/4-byte integer, we can use it only if the rest of the
    // goodness in the source type is just tail padding.  This is allowed to
    // kick in for struct {double,int} on the int, but not on
    // struct{double,int,int} because we wouldn't return the second int.  We
    // have to do this analysis on the source type because we can't depend on
    // unions being lowered a specific way etc.
    if (IRType->isIntegerTy(8) || IRType->isIntegerTy(16)70.9k ||
        IRType->isIntegerTy(32) ||
        (9.36kisa<llvm::PointerType>(IRType)9.36k && !Has64BitPointers92)) {
      unsigned BitWidth = isa<llvm::PointerType>(IRType) ? 32 :
          cast<llvm::IntegerType>(IRType)->getBitWidth();

      if (BitsContainNoUserData(SourceTy, SourceOffset*8+BitWidth,
                                SourceOffset*8+64, getContext()))
        return IRType;
    }
  }

  if (llvm::StructType *STy = dyn_cast<llvm::StructType>(IRType)) {
    // If this is a struct, recurse into the field at the specified offset.
    const llvm::StructLayout *SL = getDataLayout().getStructLayout(STy);
    if (IROffset < SL->getSizeInBytes()) {
      unsigned FieldIdx = SL->getElementContainingOffset(IROffset);
      IROffset -= SL->getElementOffset(FieldIdx);

      return GetINTEGERTypeAtOffset(STy->getElementType(FieldIdx), IROffset,
                                    SourceTy, SourceOffset);
    }
  }

  if (llvm::ArrayType *ATy = dyn_cast<llvm::ArrayType>(IRType)) {
    llvm::Type *EltTy = ATy->getElementType();
    unsigned EltSize = getDataLayout().getTypeAllocSize(EltTy);
    unsigned EltOffset = IROffset/EltSize*EltSize;
    return GetINTEGERTypeAtOffset(EltTy, IROffset-EltOffset, SourceTy,
                                  SourceOffset);
  }

  // Okay, we don't have any better idea of what to pass, so we pass this in an
  // integer register that isn't too big to fit the rest of the struct.
  unsigned TySizeInBytes =
    (unsigned)getContext().getTypeSizeInChars(SourceTy).getQuantity();

  assert(TySizeInBytes != SourceOffset && "Empty field?");

  // It is always safe to classify this as an integer type up to i64 that
  // isn't larger than the structure.
  return llvm::IntegerType::get(getVMContext(),
                                std::min(TySizeInBytes-SourceOffset, 8U)*8);
}


/// GetX86_64ByValArgumentPair - Given a high and low type that can ideally
/// be used as elements of a two register pair to pass or return, return a
/// first class aggregate to represent them.  For example, if the low part of
/// a by-value argument should be passed as i32* and the high part as float,
/// return {i32*, float}.
static llvm::Type *
GetX86_64ByValArgumentPair(llvm::Type *Lo, llvm::Type *Hi,
                           const llvm::DataLayout &TD) {
  // In order to correctly satisfy the ABI, we need to the high part to start
  // at offset 8.  If the high and low parts we inferred are both 4-byte types
  // (e.g. i32 and i32) then the resultant struct type ({i32,i32}) won't have
  // the second element at offset 8.  Check for this:
  unsigned LoSize = (unsigned)TD.getTypeAllocSize(Lo);
  unsigned HiAlign = TD.getABITypeAlignment(Hi);
  unsigned HiStart = llvm::alignTo(LoSize, HiAlign);
  assert(HiStart != 0 && HiStart <= 8 && "Invalid x86-64 argument pair!");

  // To handle this, we have to increase the size of the low part so that the
  // second element will start at an 8 byte offset.  We can't increase the size
  // of the second element because it might make us access off the end of the
  // struct.
  if (HiStart != 8) {
    // There are usually two sorts of types the ABI generation code can produce
    // for the low part of a pair that aren't 8 bytes in size: float or
    // i8/i16/i32.  This can also include pointers when they are 32-bit (X32 and
    // NaCl).
    // Promote these to a larger type.
    if (Lo->isFloatTy())
      Lo = llvm::Type::getDoubleTy(Lo->getContext());
    else {
      assert((Lo->isIntegerTy() || Lo->isPointerTy())
             && "Invalid/unknown lo type");
      Lo = llvm::Type::getInt64Ty(Lo->getContext());
    }
  }

  llvm::StructType *Result = llvm::StructType::get(Lo, Hi);

  // Verify that the second element is at an 8-byte offset.
  assert(TD.getStructLayout(Result)->getElementOffset(1) == 8 &&
         "Invalid x86-64 argument pair!");
  return Result;
}

ABIArgInfo X86_64ABIInfo::
classifyReturnType(QualType RetTy) const {
  // AMD64-ABI 3.2.3p4: Rule 1. Classify the return type with the
  // classification algorithm.
  X86_64ABIInfo::Class Lo, Hi;
  classify(RetTy, 0, Lo, Hi, /*isNamedArg*/ true);

  // Check some invariants.
  assert((Hi != Memory || Lo == Memory) && "Invalid memory classification.");
  assert((Hi != SSEUp || Lo == SSE) && "Invalid SSEUp classification.");

  llvm::Type *ResType = nullptr;
  switch (Lo) {
  case NoClass:
    if (Hi == NoClass)
      return ABIArgInfo::getIgnore();
    // If the low part is just padding, it takes no register, leave ResType
    // null.
    assert((Hi == SSE || Hi == Integer || Hi == X87Up) &&
           "Unknown missing lo part");
    break;

  case SSEUp:
  case X87Up:
    llvm_unreachable("Invalid classification for lo word.");

    // AMD64-ABI 3.2.3p4: Rule 2. Types of class memory are returned via
    // hidden argument.
  case Memory:
    return getIndirectReturnResult(RetTy);

    // AMD64-ABI 3.2.3p4: Rule 3. If the class is INTEGER, the next
    // available register of the sequence %rax, %rdx is used.
  case Integer:
    ResType = GetINTEGERTypeAtOffset(CGT.ConvertType(RetTy), 0, RetTy, 0);

    // If we have a sign or zero extended integer, make sure to return Extend
    // so that the parameter gets the right LLVM IR attributes.
    if (Hi == NoClass && isa<llvm::IntegerType>(ResType)77.1k) {
      // Treat an enum type as its underlying type.
      if (const EnumType *EnumTy = RetTy->getAs<EnumType>())
        RetTy = EnumTy->getDecl()->getIntegerType();

      if (RetTy->isIntegralOrEnumerationType() &&
          isPromotableIntegerTypeForABI(RetTy))
        return ABIArgInfo::getExtend(RetTy);
    }
    break;

    // AMD64-ABI 3.2.3p4: Rule 4. If the class is SSE, the next
    // available SSE register of the sequence %xmm0, %xmm1 is used.
  case SSE:
    ResType = GetSSETypeAtOffset(CGT.ConvertType(RetTy), 0, RetTy, 0);
    break;

    // AMD64-ABI 3.2.3p4: Rule 6. If the class is X87, the value is
    // returned on the X87 stack in %st0 as 80-bit x87 number.
  case X87:
    ResType = llvm::Type::getX86_FP80Ty(getVMContext());
    break;

    // AMD64-ABI 3.2.3p4: Rule 8. If the class is COMPLEX_X87, the real
    // part of the value is returned in %st0 and the imaginary part in
    // %st1.
  case ComplexX87:
    assert(Hi == ComplexX87 && "Unexpected ComplexX87 classification.");
    ResType = llvm::StructType::get(llvm::Type::getX86_FP80Ty(getVMContext()),
                                    llvm::Type::getX86_FP80Ty(getVMContext()));
    break;
  }

  llvm::Type *HighPart = nullptr;
  switch (Hi) {
    // Memory was handled previously and X87 should
    // never occur as a hi class.
  case Memory:
  case X87:
    llvm_unreachable("Invalid classification for hi word.");

  case ComplexX87: // Previously handled.
  case NoClass:
    break;

  case Integer:
    HighPart = GetINTEGERTypeAtOffset(CGT.ConvertType(RetTy), 8, RetTy, 8);
    if (Lo == NoClass)  // Return HighPart at offset 8 in memory.
      return ABIArgInfo::getDirect(HighPart, 8);
    break;
  case SSE:
    HighPart = GetSSETypeAtOffset(CGT.ConvertType(RetTy), 8, RetTy, 8);
    if (Lo == NoClass)  // Return HighPart at offset 8 in memory.
      return ABIArgInfo::getDirect(HighPart, 8);
    break;

    // AMD64-ABI 3.2.3p4: Rule 5. If the class is SSEUP, the eightbyte
    // is passed in the next available eightbyte chunk if the last used
    // vector register.
    //
    // SSEUP should always be preceded by SSE, just widen.
  case SSEUp:
    assert(Lo == SSE && "Unexpected SSEUp classification.");
    ResType = GetByteVectorType(RetTy);
    break;

    // AMD64-ABI 3.2.3p4: Rule 7. If the class is X87UP, the value is
    // returned together with the previous X87 value in %st0.
  case X87Up:
    // If X87Up is preceded by X87, we don't need to do
    // anything. However, in some cases with unions it may not be
    // preceded by X87. In such situations we follow gcc and pass the
    // extra bits in an SSE reg.
    if (Lo != X87) {
      HighPart = GetSSETypeAtOffset(CGT.ConvertType(RetTy), 8, RetTy, 8);
      if (Lo == NoClass)  // Return HighPart at offset 8 in memory.
        return ABIArgInfo::getDirect(HighPart, 8);
    }
    break;
  }

  // If a high part was specified, merge it together with the low part.  It is
  // known to pass in the high eightbyte of the result.  We do this by forming a
  // first class struct aggregate with the high and low part: {low, high}
  if (HighPart)
    ResType = GetX86_64ByValArgumentPair(ResType, HighPart, getDataLayout());

  return ABIArgInfo::getDirect(ResType);
}

ABIArgInfo X86_64ABIInfo::classifyArgumentType(
  QualType Ty, unsigned freeIntRegs, unsigned &neededInt, unsigned &neededSSE,
  bool isNamedArg)
  const
{
  Ty = useFirstFieldIfTransparentUnion(Ty);

  X86_64ABIInfo::Class Lo, Hi;
  classify(Ty, 0, Lo, Hi, isNamedArg);

  // Check some invariants.
  // FIXME: Enforce these by construction.
  assert((Hi != Memory || Lo == Memory) && "Invalid memory classification.");
  assert((Hi != SSEUp || Lo == SSE) && "Invalid SSEUp classification.");

  neededInt = 0;
  neededSSE = 0;
  llvm::Type *ResType = nullptr;
  switch (Lo) {
  case NoClass:
    if (Hi == NoClass)
      return ABIArgInfo::getIgnore();
    // If the low part is just padding, it takes no register, leave ResType
    // null.
    assert((Hi == SSE || Hi == Integer || Hi == X87Up) &&
           "Unknown missing lo part");
    break;

    // AMD64-ABI 3.2.3p3: Rule 1. If the class is MEMORY, pass the argument
    // on the stack.
  case Memory:

    // AMD64-ABI 3.2.3p3: Rule 5. If the class is X87, X87UP or
    // COMPLEX_X87, it is passed in memory.
  case X87:
  case ComplexX87:
    if (getRecordArgABI(Ty, getCXXABI()) == CGCXXABI::RAA_Indirect)
      ++neededInt;
    return getIndirectResult(Ty, freeIntRegs);

  case SSEUp:
  case X87Up:
    llvm_unreachable("Invalid classification for lo word.");

    // AMD64-ABI 3.2.3p3: Rule 2. If the class is INTEGER, the next
    // available register of the sequence %rdi, %rsi, %rdx, %rcx, %r8
    // and %r9 is used.
  case Integer:
    ++neededInt;

    // Pick an 8-byte type based on the preferred type.
    ResType = GetINTEGERTypeAtOffset(CGT.ConvertType(Ty), 0, Ty, 0);

    // If we have a sign or zero extended integer, make sure to return Extend
    // so that the parameter gets the right LLVM IR attributes.
    if (Hi == NoClass && isa<llvm::IntegerType>(ResType)231k) {
      // Treat an enum type as its underlying type.
      if (const EnumType *EnumTy = Ty->getAs<EnumType>())
        Ty = EnumTy->getDecl()->getIntegerType();

      if (Ty->isIntegralOrEnumerationType() &&
          isPromotableIntegerTypeForABI(Ty))
        return ABIArgInfo::getExtend(Ty);
    }

    break;

    // AMD64-ABI 3.2.3p3: Rule 3. If the class is SSE, the next
    // available SSE register is used, the registers are taken in the
    // order from %xmm0 to %xmm7.
  case SSE: {
    llvm::Type *IRType = CGT.ConvertType(Ty);
    ResType = GetSSETypeAtOffset(IRType, 0, Ty, 0);
    ++neededSSE;
    break;
  }
  }

  llvm::Type *HighPart = nullptr;
  switch (Hi) {
    // Memory was handled previously, ComplexX87 and X87 should
    // never occur as hi classes, and X87Up must be preceded by X87,
    // which is passed in memory.
  case Memory:
  case X87:
  case ComplexX87:
    llvm_unreachable("Invalid classification for hi word.");

  case NoClass: break;

  case Integer:
    ++neededInt;
    // Pick an 8-byte type based on the preferred type.
    HighPart = GetINTEGERTypeAtOffset(CGT.ConvertType(Ty), 8, Ty, 8);

    if (Lo == NoClass)  // Pass HighPart at offset 8 in memory.
      return ABIArgInfo::getDirect(HighPart, 8);
    break;

    // X87Up generally doesn't occur here (long double is passed in
    // memory), except in situations involving unions.
  case X87Up:
  case SSE:
    HighPart = GetSSETypeAtOffset(CGT.ConvertType(Ty), 8, Ty, 8);

    if (Lo == NoClass)  // Pass HighPart at offset 8 in memory.
      return ABIArgInfo::getDirect(HighPart, 8);

    ++neededSSE;
    break;

    // AMD64-ABI 3.2.3p3: Rule 4. If the class is SSEUP, the
    // eightbyte is passed in the upper half of the last used SSE
    // register.  This only happens when 128-bit vectors are passed.
  case SSEUp:
    assert(Lo == SSE && "Unexpected SSEUp classification");
    ResType = GetByteVectorType(Ty);
    break;
  }

  // If a high part was specified, merge it together with the low part.  It is
  // known to pass in the high eightbyte of the result.  We do this by forming a
  // first class struct aggregate with the high and low part: {low, high}
  if (HighPart)
    ResType = GetX86_64ByValArgumentPair(ResType, HighPart, getDataLayout());

  return ABIArgInfo::getDirect(ResType);
}

ABIArgInfo
X86_64ABIInfo::classifyRegCallStructTypeImpl(QualType Ty, unsigned &NeededInt,
                                             unsigned &NeededSSE) const {
  auto RT = Ty->getAs<RecordType>();
  assert(RT && "classifyRegCallStructType only valid with struct types");

  if (RT->getDecl()->hasFlexibleArrayMember())
    return getIndirectReturnResult(Ty);

  // Sum up bases
  if (auto CXXRD = dyn_cast<CXXRecordDecl>(RT->getDecl())) {
    if (CXXRD->isDynamicClass()) {
      NeededInt = NeededSSE = 0;
      return getIndirectReturnResult(Ty);
    }

    for (const auto &I : CXXRD->bases())
      if (classifyRegCallStructTypeImpl(I.getType(), NeededInt, NeededSSE)
              .isIndirect()) {
        NeededInt = NeededSSE = 0;
        return getIndirectReturnResult(Ty);
      }
  }

  // Sum up members
  for (const auto *FD : RT->getDecl()->fields())18 {
    if (FD->getType()->isRecordType() && !FD->getType()->isUnionType()0) {
      if (classifyRegCallStructTypeImpl(FD->getType(), NeededInt, NeededSSE)
              .isIndirect()) {
        NeededInt = NeededSSE = 0;
        return getIndirectReturnResult(Ty);
      }
    } else {
      unsigned LocalNeededInt, LocalNeededSSE;
      if (classifyArgumentType(FD->getType(), UINT_MAX, LocalNeededInt,
                               LocalNeededSSE, true)
              .isIndirect()) {
        NeededInt = NeededSSE = 0;
        return getIndirectReturnResult(Ty);
      }
      NeededInt += LocalNeededInt;
      NeededSSE += LocalNeededSSE;
    }
  }

  return ABIArgInfo::getDirect();
}

ABIArgInfo X86_64ABIInfo::classifyRegCallStructType(QualType Ty,
                                                    unsigned &NeededInt,
                                                    unsigned &NeededSSE) const {

  NeededInt = 0;
  NeededSSE = 0;

  return classifyRegCallStructTypeImpl(Ty, NeededInt, NeededSSE);
}

void X86_64ABIInfo::computeInfo(CGFunctionInfo &FI) const {

  const unsigned CallingConv = FI.getCallingConvention();
  // It is possible to force Win64 calling convention on any x86_64 target by
  // using __attribute__((ms_abi)). In such case to correctly emit Win64
  // compatible code delegate this call to WinX86_64ABIInfo::computeInfo.
  if (CallingConv == llvm::CallingConv::Win64) {
    WinX86_64ABIInfo Win64ABIInfo(CGT, AVXLevel);
    Win64ABIInfo.computeInfo(FI);
    return;
  }

  bool IsRegCall = CallingConv == llvm::CallingConv::X86_RegCall;

  // Keep track of the number of assigned registers.
  unsigned FreeIntRegs = IsRegCall ? 1124 : 6;
  unsigned FreeSSERegs = IsRegCall ? 1624 : 8;
  unsigned NeededInt, NeededSSE;

  if (!::classifyReturnType(getCXXABI(), FI, *this)) {
    if (IsRegCall && FI.getReturnType()->getTypePtr()->isRecordType()23 &&
        !FI.getReturnType()->getTypePtr()->isUnionType()) {
      FI.getReturnInfo() =
          classifyRegCallStructType(FI.getReturnType(), NeededInt, NeededSSE);
      if (FreeIntRegs >= NeededInt && FreeSSERegs >= NeededSSE) {
        FreeIntRegs -= NeededInt;
        FreeSSERegs -= NeededSSE;
      } else {
        FI.getReturnInfo() = getIndirectReturnResult(FI.getReturnType());
      }
    } else if (IsRegCall && FI.getReturnType()->getAs<ComplexType>()21 &&
               getContext().getCanonicalType(FI.getReturnType()
                                                 ->getAs<ComplexType>()
                                                 ->getElementType()) ==
                   getContext().LongDoubleTy)
      // Complex Long Double Type is passed in Memory when Regcall
      // calling convention is used.
      FI.getReturnInfo() = getIndirectReturnResult(FI.getReturnType());
    else
      FI.getReturnInfo() = classifyReturnType(FI.getReturnType());
  }

  // If the return value is indirect, then the hidden argument is consuming one
  // integer register.
  if (FI.getReturnInfo().isIndirect())
    --FreeIntRegs;

  // The chain argument effectively gives us another free register.
  if (FI.isChainCall())
    ++FreeIntRegs;

  unsigned NumRequiredArgs = FI.getNumRequiredArgs();
  // AMD64-ABI 3.2.3p3: Once arguments are classified, the registers
  // get assigned (in left-to-right order) for passing as follows...
  unsigned ArgNo = 0;
  for (CGFunctionInfo::arg_iterator it = FI.arg_begin(), ie = FI.arg_end();
       it != ie; ++it, ++ArgNo243k) {
    bool IsNamedArg = ArgNo < NumRequiredArgs;

    if (IsRegCall && it->type->isStructureOrClassType()54)
      it->info = classifyRegCallStructType(it->type, NeededInt, NeededSSE);
    else
      it->info = classifyArgumentType(it->type, FreeIntRegs, NeededInt,
                                      NeededSSE, IsNamedArg);

    // AMD64-ABI 3.2.3p3: If there are no registers available for any
    // eightbyte of an argument, the whole argument is passed on the
    // stack. If registers have already been assigned for some
    // eightbytes of such an argument, the assignments get reverted.
    if (FreeIntRegs >= NeededInt && FreeSSERegs >= NeededSSE241k) {
      FreeIntRegs -= NeededInt;
      FreeSSERegs -= NeededSSE;
    } else {
      it->info = getIndirectResult(it->type, FreeIntRegs);
    }
  }
}

static Address EmitX86_64VAArgFromMemory(CodeGenFunction &CGF,
                                         Address VAListAddr, QualType Ty) {
  Address overflow_arg_area_p =
      CGF.Builder.CreateStructGEP(VAListAddr, 2, "overflow_arg_area_p");
  llvm::Value *overflow_arg_area =
    CGF.Builder.CreateLoad(overflow_arg_area_p, "overflow_arg_area");

  // AMD64-ABI 3.5.7p5: Step 7. Align l->overflow_arg_area upwards to a 16
  // byte boundary if alignment needed by type exceeds 8 byte boundary.
  // It isn't stated explicitly in the standard, but in practice we use
  // alignment greater than 16 where necessary.
  CharUnits Align = CGF.getContext().getTypeAlignInChars(Ty);
  if (Align > CharUnits::fromQuantity(8)) {
    overflow_arg_area = emitRoundPointerUpToAlignment(CGF, overflow_arg_area,
                                                      Align);
  }

  // AMD64-ABI 3.5.7p5: Step 8. Fetch type from l->overflow_arg_area.
  llvm::Type *LTy = CGF.ConvertTypeForMem(Ty);
  llvm::Value *Res =
    CGF.Builder.CreateBitCast(overflow_arg_area,
                              llvm::PointerType::getUnqual(LTy));

  // AMD64-ABI 3.5.7p5: Step 9. Set l->overflow_arg_area to:
  // l->overflow_arg_area + sizeof(type).
  // AMD64-ABI 3.5.7p5: Step 10. Align l->overflow_arg_area upwards to
  // an 8 byte boundary.

  uint64_t SizeInBytes = (CGF.getContext().getTypeSize(Ty) + 7) / 8;
  llvm::Value *Offset =
      llvm::ConstantInt::get(CGF.Int32Ty, (SizeInBytes + 7)  & ~7);
  overflow_arg_area = CGF.Builder.CreateGEP(overflow_arg_area, Offset,
                                            "overflow_arg_area.next");
  CGF.Builder.CreateStore(overflow_arg_area, overflow_arg_area_p);

  // AMD64-ABI 3.5.7p5: Step 11. Return the fetched type.
  return Address(Res, Align);
}

Address X86_64ABIInfo::EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                                 QualType Ty) const {
  // Assume that va_list type is correct; should be pointer to LLVM type:
  // struct {
  //   i32 gp_offset;
  //   i32 fp_offset;
  //   i8* overflow_arg_area;
  //   i8* reg_save_area;
  // };
  unsigned neededInt, neededSSE;

  Ty = getContext().getCanonicalType(Ty);
  ABIArgInfo AI = classifyArgumentType(Ty, 0, neededInt, neededSSE,
                                       /*isNamedArg*/false);

  // AMD64-ABI 3.5.7p5: Step 1. Determine whether type may be passed
  // in the registers. If not go to step 7.
  if (!neededInt && !neededSSE34)
    return EmitX86_64VAArgFromMemory(CGF, VAListAddr, Ty);

  // AMD64-ABI 3.5.7p5: Step 2. Compute num_gp to hold the number of
  // general purpose registers needed to pass type and num_fp to hold
  // the number of floating point registers needed.

  // AMD64-ABI 3.5.7p5: Step 3. Verify whether arguments fit into
  // registers. In the case: l->gp_offset > 48 - num_gp * 8 or
  // l->fp_offset > 304 - num_fp * 16 go to step 7.
  //
  // NOTE: 304 is a typo, there are (6 * 8 + 8 * 16) = 176 bytes of
  // register save space).

  llvm::Value *InRegs = nullptr;
  Address gp_offset_p = Address::invalid(), fp_offset_p = Address::invalid();
  llvm::Value *gp_offset = nullptr, *fp_offset = nullptr;
  if (neededInt) {
    gp_offset_p = CGF.Builder.CreateStructGEP(VAListAddr, 0, "gp_offset_p");
    gp_offset = CGF.Builder.CreateLoad(gp_offset_p, "gp_offset");
    InRegs = llvm::ConstantInt::get(CGF.Int32Ty, 48 - neededInt * 8);
    InRegs = CGF.Builder.CreateICmpULE(gp_offset, InRegs, "fits_in_gp");
  }

  if (neededSSE) {
    fp_offset_p = CGF.Builder.CreateStructGEP(VAListAddr, 1, "fp_offset_p");
    fp_offset = CGF.Builder.CreateLoad(fp_offset_p, "fp_offset");
    llvm::Value *FitsInFP =
      llvm::ConstantInt::get(CGF.Int32Ty, 176 - neededSSE * 16);
    FitsInFP = CGF.Builder.CreateICmpULE(fp_offset, FitsInFP, "fits_in_fp");
    InRegs = InRegs ? CGF.Builder.CreateAnd(InRegs, FitsInFP)4 : FitsInFP;
  }

  llvm::BasicBlock *InRegBlock = CGF.createBasicBlock("vaarg.in_reg");
  llvm::BasicBlock *InMemBlock = CGF.createBasicBlock("vaarg.in_mem");
  llvm::BasicBlock *ContBlock = CGF.createBasicBlock("vaarg.end");
  CGF.Builder.CreateCondBr(InRegs, InRegBlock, InMemBlock);

  // Emit code to load the value if it was passed in registers.

  CGF.EmitBlock(InRegBlock);

  // AMD64-ABI 3.5.7p5: Step 4. Fetch type from l->reg_save_area with
  // an offset of l->gp_offset and/or l->fp_offset. This may require
  // copying to a temporary location in case the parameter is passed
  // in different register classes or requires an alignment greater
  // than 8 for general purpose registers and 16 for XMM registers.
  //
  // FIXME: This really results in shameful code when we end up needing to
  // collect arguments from different places; often what should result in a
  // simple assembling of a structure from scattered addresses has many more
  // loads than necessary. Can we clean this up?
  llvm::Type *LTy = CGF.ConvertTypeForMem(Ty);
  llvm::Value *RegSaveArea = CGF.Builder.CreateLoad(
      CGF.Builder.CreateStructGEP(VAListAddr, 3), "reg_save_area");

  Address RegAddr = Address::invalid();
  if (neededInt && neededSSE63) {
    // FIXME: Cleanup.
    assert(AI.isDirect() && "Unexpected ABI info for mixed regs");
    llvm::StructType *ST = cast<llvm::StructType>(AI.getCoerceToType());
    Address Tmp = CGF.CreateMemTemp(Ty);
    Tmp = CGF.Builder.CreateElementBitCast(Tmp, ST);
    assert(ST->getNumElements() == 2 && "Unexpected ABI info for mixed regs");
    llvm::Type *TyLo = ST->getElementType(0);
    llvm::Type *TyHi = ST->getElementType(1);
    assert((TyLo->isFPOrFPVectorTy() ^ TyHi->isFPOrFPVectorTy()) &&
           "Unexpected ABI info for mixed regs");
    llvm::Type *PTyLo = llvm::PointerType::getUnqual(TyLo);
    llvm::Type *PTyHi = llvm::PointerType::getUnqual(TyHi);
    llvm::Value *GPAddr = CGF.Builder.CreateGEP(RegSaveArea, gp_offset);
    llvm::Value *FPAddr = CGF.Builder.CreateGEP(RegSaveArea, fp_offset);
    llvm::Value *RegLoAddr = TyLo->isFPOrFPVectorTy() ? FPAddr1 : GPAddr;
    llvm::Value *RegHiAddr = TyLo->isFPOrFPVectorTy() ? GPAddr1 : FPAddr;

    // Copy the first element.
    // FIXME: Our choice of alignment here and below is probably pessimistic.
    llvm::Value *V = CGF.Builder.CreateAlignedLoad(
        TyLo, CGF.Builder.CreateBitCast(RegLoAddr, PTyLo),
        CharUnits::fromQuantity(getDataLayout().getABITypeAlignment(TyLo)));
    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 0));

    // Copy the second element.
    V = CGF.Builder.CreateAlignedLoad(
        TyHi, CGF.Builder.CreateBitCast(RegHiAddr, PTyHi),
        CharUnits::fromQuantity(getDataLayout().getABITypeAlignment(TyHi)));
    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 1));

    RegAddr = CGF.Builder.CreateElementBitCast(Tmp, LTy);
  } else if (neededInt) {
    RegAddr = Address(CGF.Builder.CreateGEP(RegSaveArea, gp_offset),
                      CharUnits::fromQuantity(8));
    RegAddr = CGF.Builder.CreateElementBitCast(RegAddr, LTy);

    // Copy to a temporary if necessary to ensure the appropriate alignment.
    auto TInfo = getContext().getTypeInfoInChars(Ty);
    uint64_t TySize = TInfo.Width.getQuantity();
    CharUnits TyAlign = TInfo.Align;

    // Copy into a temporary if the type is more aligned than the
    // register save area.
    if (TyAlign.getQuantity() > 8) {
      Address Tmp = CGF.CreateMemTemp(Ty);
      CGF.Builder.CreateMemCpy(Tmp, RegAddr, TySize, false);
      RegAddr = Tmp;
    }

  } else if (neededSSE == 1) {
    RegAddr = Address(CGF.Builder.CreateGEP(RegSaveArea, fp_offset),
                      CharUnits::fromQuantity(16));
    RegAddr = CGF.Builder.CreateElementBitCast(RegAddr, LTy);
  } else {
    assert(neededSSE == 2 && "Invalid number of needed registers!");
    // SSE registers are spaced 16 bytes apart in the register save
    // area, we need to collect the two eightbytes together.
    // The ABI isn't explicit about this, but it seems reasonable
    // to assume that the slots are 16-byte aligned, since the stack is
    // naturally 16-byte aligned and the prologue is expected to store
    // all the SSE registers to the RSA.
    Address RegAddrLo = Address(CGF.Builder.CreateGEP(RegSaveArea, fp_offset),
                                CharUnits::fromQuantity(16));
    Address RegAddrHi =
      CGF.Builder.CreateConstInBoundsByteGEP(RegAddrLo,
                                             CharUnits::fromQuantity(16));
    llvm::Type *ST = AI.canHaveCoerceToType()
                         ? AI.getCoerceToType()
                         : llvm::StructType::get(CGF.DoubleTy, CGF.DoubleTy);
    llvm::Value *V;
    Address Tmp = CGF.CreateMemTemp(Ty);
    Tmp = CGF.Builder.CreateElementBitCast(Tmp, ST);
    V = CGF.Builder.CreateLoad(CGF.Builder.CreateElementBitCast(
        RegAddrLo, ST->getStructElementType(0)));
    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 0));
    V = CGF.Builder.CreateLoad(CGF.Builder.CreateElementBitCast(
        RegAddrHi, ST->getStructElementType(1)));
    CGF.Builder.CreateStore(V, CGF.Builder.CreateStructGEP(Tmp, 1));

    RegAddr = CGF.Builder.CreateElementBitCast(Tmp, LTy);
  }

  // AMD64-ABI 3.5.7p5: Step 5. Set:
  // l->gp_offset = l->gp_offset + num_gp * 8
  // l->fp_offset = l->fp_offset + num_fp * 16.
  if (neededInt) {
    llvm::Value *Offset = llvm::ConstantInt::get(CGF.Int32Ty, neededInt * 8);
    CGF.Builder.CreateStore(CGF.Builder.CreateAdd(gp_offset, Offset),
                            gp_offset_p);
  }
  if (neededSSE) {
    llvm::Value *Offset = llvm::ConstantInt::get(CGF.Int32Ty, neededSSE * 16);
    CGF.Builder.CreateStore(CGF.Builder.CreateAdd(fp_offset, Offset),
                            fp_offset_p);
  }
  CGF.EmitBranch(ContBlock);

  // Emit code to load the value if it was passed in memory.

  CGF.EmitBlock(InMemBlock);
  Address MemAddr = EmitX86_64VAArgFromMemory(CGF, VAListAddr, Ty);

  // Return the appropriate result.

  CGF.EmitBlock(ContBlock);
  Address ResAddr = emitMergePHI(CGF, RegAddr, InRegBlock, MemAddr, InMemBlock,
                                 "vaarg.addr");
  return ResAddr;
}

Address X86_64ABIInfo::EmitMSVAArg(CodeGenFunction &CGF, Address VAListAddr,
                                   QualType Ty) const {
  return emitVoidPtrVAArg(CGF, VAListAddr, Ty, /*indirect*/ false,
                          CGF.getContext().getTypeInfoInChars(Ty),
                          CharUnits::fromQuantity(8),
                          /*allowHigherAlign*/ false);
}

ABIArgInfo WinX86_64ABIInfo::reclassifyHvaArgForVectorCall(
    QualType Ty, unsigned &FreeSSERegs, const ABIArgInfo &current) const {
  const Type *Base = nullptr;
  uint64_t NumElts = 0;

  if (!Ty->isBuiltinType() && !Ty->isVectorType()49 &&
      isHomogeneousAggregate(Ty, Base, NumElts) && FreeSSERegs >= NumElts27) {
    FreeSSERegs -= NumElts;
    return getDirectX86Hva();
  }
  return current;
}

ABIArgInfo WinX86_64ABIInfo::classify(QualType Ty, unsigned &FreeSSERegs,
                                      bool IsReturnType, bool IsVectorCall,
                                      bool IsRegCall) const {

  if (Ty->isVoidType())
    return ABIArgInfo::getIgnore();

  if (const EnumType *EnumTy = Ty->getAs<EnumType>())
    Ty = EnumTy->getDecl()->getIntegerType();

  TypeInfo Info = getContext().getTypeInfo(Ty);
  uint64_t Width = Info.Width;
  CharUnits Align = getContext().toCharUnitsFromBits(Info.Align);

  const RecordType *RT = Ty->getAs<RecordType>();
  if (RT) {
    if (!IsReturnType) {
      if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(RT, getCXXABI()))
        return getNaturalAlignIndirect(Ty, RAA == CGCXXABI::RAA_DirectInMemory);
    }

    if (RT->getDecl()->hasFlexibleArrayMember())
      return getNaturalAlignIndirect(Ty, /*ByVal=*/false);

  }

  const Type *Base = nullptr;
  uint64_t NumElts = 0;
  // vectorcall adds the concept of a homogenous vector aggregate, similar to
  // other targets.
  if ((IsVectorCall || IsRegCall7.96k) &&
      isHomogeneousAggregate(Ty, Base, NumElts)) {
    if (IsRegCall) {
      if (FreeSSERegs >= NumElts) {
        FreeSSERegs -= NumElts;
        if (IsReturnType || Ty->isBuiltinType()27 || Ty->isVectorType()21)
          return ABIArgInfo::getDirect();
        return ABIArgInfo::getExpand();
      }
      return ABIArgInfo::getIndirect(Align, /*ByVal=*/false);
    } else if (IsVectorCall) {
      if (FreeSSERegs >= NumElts &&
          (IsReturnType || Ty->isBuiltinType()45 || Ty->isVectorType()36)) {
        FreeSSERegs -= NumElts;
        return ABIArgInfo::getDirect();
      } else if (IsReturnType) {
        return ABIArgInfo::getExpand();
      } else if (!Ty->isBuiltinType() && !Ty->isVectorType()27) {
        // HVAs are delayed and reclassified in the 2nd step.
        return ABIArgInfo::getIndirect(Align, /*ByVal=*/false);
      }
    }
  }

  if (Ty->isMemberPointerType()) {
    // If the member pointer is represented by an LLVM int or ptr, pass it
    // directly.
    llvm::Type *LLTy = CGT.ConvertType(Ty);
    if (LLTy->isPointerTy() || LLTy->isIntegerTy()76)
      return ABIArgInfo::getDirect();
  }

  if (RT || Ty->isAnyComplexType()7.72k || Ty->isMemberPointerType()7.70k) {
    // MS x64 ABI requirement: "Any argument that doesn't fit in 8 bytes, or is
    // not 1, 2, 4, or 8 bytes, must be passed by reference."
    if (Width > 64 || !llvm::isPowerOf2_64(Width)148)
      return getNaturalAlignIndirect(Ty, /*ByVal=*/false);

    // Otherwise, coerce it to a small integer.
    return ABIArgInfo::getDirect(llvm::IntegerType::get(getVMContext(), Width));
  }

  if (const BuiltinType *BT = Ty->getAs<BuiltinType>()) {
    switch (BT->getKind()) {
    case BuiltinType::Bool:
      // Bool type is always extended to the ABI, other builtin types are not
      // extended.
      return ABIArgInfo::getExtend(Ty);

    case BuiltinType::LongDouble:
      // Mingw64 GCC uses the old 80 bit extended precision floating point
      // unit. It passes them indirectly through memory.
      if (IsMingw64) {
        const llvm::fltSemantics *LDF = &getTarget().getLongDoubleFormat();
        if (LDF == &llvm::APFloat::x87DoubleExtended())
          return ABIArgInfo::getIndirect(Align, /*ByVal=*/false);
      }
      break;

    case BuiltinType::Int128:
    case BuiltinType::UInt128:
      // If it's a parameter type, the normal ABI rule is that arguments larger
      // than 8 bytes are passed indirectly. GCC follows it. We follow it too,
      // even though it isn't particularly efficient.
      if (!IsReturnType)
        return ABIArgInfo::getIndirect(Align, /*ByVal=*/false);

      // Mingw64 GCC returns i128 in XMM0. Coerce to v2i64 to handle that.
      // Clang matches them for compatibility.
      return ABIArgInfo::getDirect(llvm::FixedVectorType::get(
          llvm::Type::getInt64Ty(getVMContext()), 2));

    default:
      break;
    }
  }

  if (Ty->isExtIntType()) {
    // MS x64 ABI requirement: "Any argument that doesn't fit in 8 bytes, or is
    // not 1, 2, 4, or 8 bytes, must be passed by reference."
    // However, non-power-of-two _ExtInts will be passed as 1,2,4 or 8 bytes
    // anyway as long is it fits in them, so we don't have to check the power of
    // 2.
    if (Width <= 64)
      return ABIArgInfo::getDirect();
    return ABIArgInfo::getIndirect(Align, /*ByVal=*/false);
  }

  return ABIArgInfo::getDirect();
}

void WinX86_64ABIInfo::computeInfo(CGFunctionInfo &FI) const {
  const unsigned CC = FI.getCallingConvention();
  bool IsVectorCall = CC == llvm::CallingConv::X86_VectorCall;
  bool IsRegCall = CC == llvm::CallingConv::X86_RegCall;

  // If __attribute__((sysv_abi)) is in use, use the SysV argument
  // classification rules.
  if (CC == llvm::CallingConv::X86_64_SysV) {
    X86_64ABIInfo SysVABIInfo(CGT, AVXLevel);
    SysVABIInfo.computeInfo(FI);
    return;
  }

  unsigned FreeSSERegs = 0;
  if (IsVectorCall) {
    // We can use up to 4 SSE return registers with vectorcall.
    FreeSSERegs = 4;
  } else if (IsRegCall) {
    // RegCall gives us 16 SSE registers.
    FreeSSERegs = 16;
  }

  if (!getCXXABI().classifyReturnType(FI))
    FI.getReturnInfo() = classify(FI.getReturnType(), FreeSSERegs, true,
                                  IsVectorCall, IsRegCall);

  if (IsVectorCall) {
    // We can use up to 6 SSE register parameters with vectorcall.
    FreeSSERegs = 6;
  } else if (IsRegCall) {
    // RegCall gives us 16 SSE registers, we can reuse the return registers.
    FreeSSERegs = 16;
  }

  unsigned ArgNum = 0;
  unsigned ZeroSSERegs = 0;
  for (auto &I : FI.arguments()) {
    // Vectorcall in x64 only permits the first 6 arguments to be passed as
    // XMM/YMM registers. After the sixth argument, pretend no vector
    // registers are left.
    unsigned *MaybeFreeSSERegs =
        (IsVectorCall && ArgNum >= 697) ? &ZeroSSERegs5 : &FreeSSERegs5.78k;
    I.info =
        classify(I.type, *MaybeFreeSSERegs, false, IsVectorCall, IsRegCall);
    ++ArgNum;
  }

  if (IsVectorCall) {
    // For vectorcall, assign aggregate HVAs to any free vector registers in a
    // second pass.
    for (auto &I : FI.arguments())
      I.info = reclassifyHvaArgForVectorCall(I.type, FreeSSERegs, I.info);
  }
}

Address WinX86_64ABIInfo::EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                                    QualType Ty) const {

  bool IsIndirect = false;

  // MS x64 ABI requirement: "Any argument that doesn't fit in 8 bytes, or is
  // not 1, 2, 4, or 8 bytes, must be passed by reference."
  if (isAggregateTypeForABI(Ty) || Ty->isMemberPointerType()13) {
    uint64_t Width = getContext().getTypeSize(Ty);
    IsIndirect = Width > 64 || !llvm::isPowerOf2_64(Width)0;
  }

  return emitVoidPtrVAArg(CGF, VAListAddr, Ty, IsIndirect,
                          CGF.getContext().getTypeInfoInChars(Ty),
                          CharUnits::fromQuantity(8),
                          /*allowHigherAlign*/ false);
}

static bool PPC_initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                                        llvm::Value *Address, bool Is64Bit,
                                        bool IsAIX) {
  // This is calculated from the LLVM and GCC tables and verified
  // against gcc output.  AFAIK all PPC ABIs use the same encoding.

  CodeGen::CGBuilderTy &Builder = CGF.Builder;

  llvm::IntegerType *i8 = CGF.Int8Ty;
  llvm::Value *Four8 = llvm::ConstantInt::get(i8, 4);
  llvm::Value *Eight8 = llvm::ConstantInt::get(i8, 8);
  llvm::Value *Sixteen8 = llvm::ConstantInt::get(i8, 16);

  // 0-31: r0-31, the 4-byte or 8-byte general-purpose registers
  AssignToArrayRange(Builder, Address, Is64Bit ? Eight8 : Four8, 0, 31);

  // 32-63: fp0-31, the 8-byte floating-point registers
  AssignToArrayRange(Builder, Address, Eight8, 32, 63);

  // 64-67 are various 4-byte or 8-byte special-purpose registers:
  // 64: mq
  // 65: lr
  // 66: ctr
  // 67: ap
  AssignToArrayRange(Builder, Address, Is64Bit ? Eight8 : Four8, 64, 67);

  // 68-76 are various 4-byte special-purpose registers:
  // 68-75 cr0-7
  // 76: xer
  AssignToArrayRange(Builder, Address, Four8, 68, 76);

  // 77-108: v0-31, the 16-byte vector registers
  AssignToArrayRange(Builder, Address, Sixteen8, 77, 108);

  // 109: vrsave
  // 110: vscr
  AssignToArrayRange(Builder, Address, Is64Bit ? Eight8 : Four8, 109, 110);

  // AIX does not utilize the rest of the registers.
  if (IsAIX)
    return false;

  // 111: spe_acc
  // 112: spefscr
  // 113: sfp
  AssignToArrayRange(Builder, Address, Is64Bit ? Eight81 : Four81, 111, 113);

  if (!Is64Bit)
    return false;

  // TODO: Need to verify if these registers are used on 64 bit AIX with Power8
  // or above CPU.
  // 64-bit only registers:
  // 114: tfhar
  // 115: tfiar
  // 116: texasr
  AssignToArrayRange(Builder, Address, Eight8, 114, 116);

  return false;
}

// AIX
namespace {
/// AIXABIInfo - The AIX XCOFF ABI information.
class AIXABIInfo : public ABIInfo {
  const bool Is64Bit;
  const unsigned PtrByteSize;
  CharUnits getParamTypeAlignment(QualType Ty) const;

public:
  AIXABIInfo(CodeGen::CodeGenTypes &CGT, bool Is64Bit)
      : ABIInfo(CGT), Is64Bit(Is64Bit), PtrByteSize(Is64Bit ? 8 : 4) {}

  bool isPromotableTypeForABI(QualType Ty) const;

  ABIArgInfo classifyReturnType(QualType RetTy) const;
  ABIArgInfo classifyArgumentType(QualType Ty) const;

  void computeInfo(CGFunctionInfo &FI) const override {
    if (!getCXXABI().classifyReturnType(FI))
      FI.getReturnInfo() = classifyReturnType(FI.getReturnType());

    for (auto &I : FI.arguments())
      I.info = classifyArgumentType(I.type);
  }

  Address EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                    QualType Ty) const override;
};

class AIXTargetCodeGenInfo : public TargetCodeGenInfo {
  const bool Is64Bit;

public:
  AIXTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, bool Is64Bit)
      : TargetCodeGenInfo(std::make_unique<AIXABIInfo>(CGT, Is64Bit)),
        Is64Bit(Is64Bit) {}
  int getDwarfEHStackPointer(CodeGen::CodeGenModule &M) const override {
    return 1; // r1 is the dedicated stack pointer
  }

  bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                               llvm::Value *Address) const override;
};
} // namespace

// Return true if the ABI requires Ty to be passed sign- or zero-
// extended to 32/64 bits.
bool AIXABIInfo::isPromotableTypeForABI(QualType Ty) const {
  // Treat an enum type as its underlying type.
  if (const EnumType *EnumTy = Ty->getAs<EnumType>())
    Ty = EnumTy->getDecl()->getIntegerType();

  // Promotable integer types are required to be promoted by the ABI.
  if (Ty->isPromotableIntegerType())
    return true;

  if (!Is64Bit)
    return false;

  // For 64 bit mode, in addition to the usual promotable integer types, we also
  // need to extend all 32-bit types, since the ABI requires promotion to 64
  // bits.
  if (const BuiltinType *BT = Ty->getAs<BuiltinType>())
    switch (BT->getKind()) {
    case BuiltinType::Int:
    case BuiltinType::UInt:
      return true;
    default:
      break;
    }

  return false;
}

ABIArgInfo AIXABIInfo::classifyReturnType(QualType RetTy) const {
  if (RetTy->isAnyComplexType())
    return ABIArgInfo::getDirect();

  if (RetTy->isVectorType())
    return ABIArgInfo::getDirect();

  if (RetTy->isVoidType())
    return ABIArgInfo::getIgnore();

  if (isAggregateTypeForABI(RetTy))
    return getNaturalAlignIndirect(RetTy);

  return (isPromotableTypeForABI(RetTy) ? ABIArgInfo::getExtend(RetTy)20
                                        : ABIArgInfo::getDirect());
}

ABIArgInfo AIXABIInfo::classifyArgumentType(QualType Ty) const {
  Ty = useFirstFieldIfTransparentUnion(Ty);

  if (Ty->isAnyComplexType())
    return ABIArgInfo::getDirect();

  if (Ty->isVectorType())
    return ABIArgInfo::getDirect();

  if (isAggregateTypeForABI(Ty)) {
    // Records with non-trivial destructors/copy-constructors should not be
    // passed by value.
    if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI()))
      return getNaturalAlignIndirect(Ty, RAA == CGCXXABI::RAA_DirectInMemory);

    CharUnits CCAlign = getParamTypeAlignment(Ty);
    CharUnits TyAlign = getContext().getTypeAlignInChars(Ty);

    return ABIArgInfo::getIndirect(CCAlign, /*ByVal*/ true,
                                   /*Realign*/ TyAlign > CCAlign);
  }

  return (isPromotableTypeForABI(Ty) ? ABIArgInfo::getExtend(Ty)14
                                     : ABIArgInfo::getDirect());
}

CharUnits AIXABIInfo::getParamTypeAlignment(QualType Ty) const {
  // Complex types are passed just like their elements.
  if (const ComplexType *CTy = Ty->getAs<ComplexType>())
    Ty = CTy->getElementType();

  if (Ty->isVectorType())
    return CharUnits::fromQuantity(16);

  // If the structure contains a vector type, the alignment is 16.
  if (isRecordWithSIMDVectorType(getContext(), Ty))
    return CharUnits::fromQuantity(16);

  return CharUnits::fromQuantity(PtrByteSize);
}

Address AIXABIInfo::EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                              QualType Ty) const {
  if (Ty->isAnyComplexType())
    llvm::report_fatal_error("complex type is not supported on AIX yet");

  if (Ty->isVectorType())
    llvm::report_fatal_error(
        "vector types are not yet supported for variadic functions on AIX");

  auto TypeInfo = getContext().getTypeInfoInChars(Ty);
  TypeInfo.Align = getParamTypeAlignment(Ty);

  CharUnits SlotSize = CharUnits::fromQuantity(PtrByteSize);

  return emitVoidPtrVAArg(CGF, VAListAddr, Ty, /*Indirect*/ false, TypeInfo,
                          SlotSize, /*AllowHigher*/ true);
}

bool AIXTargetCodeGenInfo::initDwarfEHRegSizeTable(
    CodeGen::CodeGenFunction &CGF, llvm::Value *Address) const {
  return PPC_initDwarfEHRegSizeTable(CGF, Address, Is64Bit, /*IsAIX*/ true);
}

// PowerPC-32
namespace {
/// PPC32_SVR4_ABIInfo - The 32-bit PowerPC ELF (SVR4) ABI information.
class PPC32_SVR4_ABIInfo : public DefaultABIInfo {
  bool IsSoftFloatABI;
  bool IsRetSmallStructInRegABI;

  CharUnits getParamTypeAlignment(QualType Ty) const;

public:
  PPC32_SVR4_ABIInfo(CodeGen::CodeGenTypes &CGT, bool SoftFloatABI,
                     bool RetSmallStructInRegABI)
      : DefaultABIInfo(CGT), IsSoftFloatABI(SoftFloatABI),
        IsRetSmallStructInRegABI(RetSmallStructInRegABI) {}

  ABIArgInfo classifyReturnType(QualType RetTy) const;

  void computeInfo(CGFunctionInfo &FI) const override {
    if (!getCXXABI().classifyReturnType(FI))
      FI.getReturnInfo() = classifyReturnType(FI.getReturnType());
    for (auto &I : FI.arguments())
      I.info = classifyArgumentType(I.type);
  }

  Address EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                    QualType Ty) const override;
};

class PPC32TargetCodeGenInfo : public TargetCodeGenInfo {
public:
  PPC32TargetCodeGenInfo(CodeGenTypes &CGT, bool SoftFloatABI,
                         bool RetSmallStructInRegABI)
      : TargetCodeGenInfo(std::make_unique<PPC32_SVR4_ABIInfo>(
            CGT, SoftFloatABI, RetSmallStructInRegABI)) {}

  static bool isStructReturnInRegABI(const llvm::Triple &Triple,
                                     const CodeGenOptions &Opts);

  int getDwarfEHStackPointer(CodeGen::CodeGenModule &M) const override {
    // This is recovered from gcc output.
    return 1; // r1 is the dedicated stack pointer
  }

  bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                               llvm::Value *Address) const override;
};
}

CharUnits PPC32_SVR4_ABIInfo::getParamTypeAlignment(QualType Ty) const {
  // Complex types are passed just like their elements.
  if (const ComplexType *CTy = Ty->getAs<ComplexType>())
    Ty = CTy->getElementType();

  if (Ty->isVectorType())
    return CharUnits::fromQuantity(getContext().getTypeSize(Ty) == 128 ? 16
                                                                       : 4);

  // For single-element float/vector structs, we consider the whole type
  // to have the same alignment requirements as its single element.
  const Type *AlignTy = nullptr;
  if (const Type *EltType = isSingleElementStruct(Ty, getContext())) {
    const BuiltinType *BT = EltType->getAs<BuiltinType>();
    if ((EltType->isVectorType() && getContext().getTypeSize(EltType) == 128) ||
        (BT && BT->isFloatingPoint()))
      AlignTy = EltType;
  }

  if (AlignTy)
    return CharUnits::fromQuantity(AlignTy->isVectorType() ? 16 : 4);
  return CharUnits::fromQuantity(4);
}

ABIArgInfo PPC32_SVR4_ABIInfo::classifyReturnType(QualType RetTy) const {
  uint64_t Size;

  // -msvr4-struct-return puts small aggregates in GPR3 and GPR4.
  if (isAggregateTypeForABI(RetTy) && IsRetSmallStructInRegABI141 &&
      (Size = getContext().getTypeSize(RetTy)) <= 64) {
    // System V ABI (1995), page 3-22, specified:
    // > A structure or union whose size is less than or equal to 8 bytes
    // > shall be returned in r3 and r4, as if it were first stored in the
    // > 8-byte aligned memory area and then the low addressed word were
    // > loaded into r3 and the high-addressed word into r4.  Bits beyond
    // > the last member of the structure or union are not defined.
    //
    // GCC for big-endian PPC32 inserts the pad before the first member,
    // not "beyond the last member" of the struct.  To stay compatible
    // with GCC, we coerce the struct to an integer of the same size.
    // LLVM will extend it and return i32 in r3, or i64 in r3:r4.
    if (Size == 0)
      return ABIArgInfo::getIgnore();
    else {
      llvm::Type *CoerceTy = llvm::Type::getIntNTy(getVMContext(), Size);
      return ABIArgInfo::getDirect(CoerceTy);
    }
  }

  return DefaultABIInfo::classifyReturnType(RetTy);
}

// TODO: this implementation is now likely redundant with
// DefaultABIInfo::EmitVAArg.
Address PPC32_SVR4_ABIInfo::EmitVAArg(CodeGenFunction &CGF, Address VAList,
                                      QualType Ty) const {
  if (getTarget().getTriple().isOSDarwin()) {
    auto TI = getContext().getTypeInfoInChars(Ty);
    TI.Align = getParamTypeAlignment(Ty);

    CharUnits SlotSize = CharUnits::fromQuantity(4);
    return emitVoidPtrVAArg(CGF, VAList, Ty,
                            classifyArgumentType(Ty).isIndirect(), TI, SlotSize,
                            /*AllowHigherAlign=*/true);
  }

  const unsigned OverflowLimit = 8;
  if (const ComplexType *CTy = Ty->getAs<ComplexType>()) {
    // TODO: Implement this. For now ignore.
    (void)CTy;
    return Address::invalid(); // FIXME?
  }

  // struct __va_list_tag {
  //   unsigned char gpr;
  //   unsigned char fpr;
  //   unsigned short reserved;
  //   void *overflow_arg_area;
  //   void *reg_save_area;
  // };

  bool isI64 = Ty->isIntegerType() && getContext().getTypeSize(Ty) == 641;
  bool isInt =
      Ty->isIntegerType() || Ty->isPointerType()2 || Ty->isAggregateType()2;
  bool isF64 = Ty->isFloatingType() && getContext().getTypeSize(Ty) == 641;

  // All aggregates are passed indirectly?  That doesn't seem consistent
  // with the argument-lowering code.
  bool isIndirect = Ty->isAggregateType();

  CGBuilderTy &Builder = CGF.Builder;

  // The calling convention either uses 1-2 GPRs or 1 FPR.
  Address NumRegsAddr = Address::invalid();
  if (isInt || IsSoftFloatABI1) {
    NumRegsAddr = Builder.CreateStructGEP(VAList, 0, "gpr");
  } else {
    NumRegsAddr = Builder.CreateStructGEP(VAList, 1, "fpr");
  }

  llvm::Value *NumRegs = Builder.CreateLoad(NumRegsAddr, "numUsedRegs");

  // "Align" the register count when TY is i64.
  if (isI64 || (isF64 && IsSoftFloatABI1)) {
    NumRegs = Builder.CreateAdd(NumRegs, Builder.getInt8(1));
    NumRegs = Builder.CreateAnd(NumRegs, Builder.getInt8((uint8_t) ~1U));
  }

  llvm::Value *CC =
      Builder.CreateICmpULT(NumRegs, Builder.getInt8(OverflowLimit), "cond");

  llvm::BasicBlock *UsingRegs = CGF.createBasicBlock("using_regs");
  llvm::BasicBlock *UsingOverflow = CGF.createBasicBlock("using_overflow");
  llvm::BasicBlock *Cont = CGF.createBasicBlock("cont");

  Builder.CreateCondBr(CC, UsingRegs, UsingOverflow);

  llvm::Type *DirectTy = CGF.ConvertType(Ty);
  if (isIndirect) DirectTy = DirectTy->getPointerTo(0)1;

  // Case 1: consume registers.
  Address RegAddr = Address::invalid();
  {
    CGF.EmitBlock(UsingRegs);

    Address RegSaveAreaPtr = Builder.CreateStructGEP(VAList, 4);
    RegAddr = Address(Builder.CreateLoad(RegSaveAreaPtr),
                      CharUnits::fromQuantity(8));
    assert(RegAddr.getElementType() == CGF.Int8Ty);

    // Floating-point registers start after the general-purpose registers.
    if (!(isInt || IsSoftFloatABI1)) {
      RegAddr = Builder.CreateConstInBoundsByteGEP(RegAddr,
                                                   CharUnits::fromQuantity(32));
    }

    // Get the address of the saved value by scaling the number of
    // registers we've used by the number of
    CharUnits RegSize = CharUnits::fromQuantity((isInt || IsSoftFloatABI1) ? 4 : 80);
    llvm::Value *RegOffset =
      Builder.CreateMul(NumRegs, Builder.getInt8(RegSize.getQuantity()));
    RegAddr = Address(Builder.CreateInBoundsGEP(CGF.Int8Ty,
                                            RegAddr.getPointer(), RegOffset),
                      RegAddr.getAlignment().alignmentOfArrayElement(RegSize));
    RegAddr = Builder.CreateElementBitCast(RegAddr, DirectTy);

    // Increase the used-register count.
    NumRegs =
      Builder.CreateAdd(NumRegs,
                        Builder.getInt8((isI64 || (isF64 && IsSoftFloatABI1)) ? 21 : 12));
    Builder.CreateStore(NumRegs, NumRegsAddr);

    CGF.EmitBranch(Cont);
  }

  // Case 2: consume space in the overflow area.
  Address MemAddr = Address::invalid();
  {
    CGF.EmitBlock(UsingOverflow);

    Builder.CreateStore(Builder.getInt8(OverflowLimit), NumRegsAddr);

    // Everything in the overflow area is rounded up to a size of at least 4.
    CharUnits OverflowAreaAlign = CharUnits::fromQuantity(4);

    CharUnits Size;
    if (!isIndirect) {
      auto TypeInfo = CGF.getContext().getTypeInfoInChars(Ty);
      Size = TypeInfo.Width.alignTo(OverflowAreaAlign);
    } else {
      Size = CGF.getPointerSize();
    }

    Address OverflowAreaAddr = Builder.CreateStructGEP(VAList, 3);
    Address OverflowArea(Builder.CreateLoad(OverflowAreaAddr, "argp.cur"),
                         OverflowAreaAlign);
    // Round up address of argument to alignment
    CharUnits Align = CGF.getContext().getTypeAlignInChars(Ty);
    if (Align > OverflowAreaAlign) {
      llvm::Value *Ptr = OverflowArea.getPointer();
      OverflowArea = Address(emitRoundPointerUpToAlignment(CGF, Ptr, Align),
                                                           Align);
    }

    MemAddr = Builder.CreateElementBitCast(OverflowArea, DirectTy);

    // Increase the overflow area.
    OverflowArea = Builder.CreateConstInBoundsByteGEP(OverflowArea, Size);
    Builder.CreateStore(OverflowArea.getPointer(), OverflowAreaAddr);
    CGF.EmitBranch(Cont);
  }

  CGF.EmitBlock(Cont);

  // Merge the cases with a phi.
  Address Result = emitMergePHI(CGF, RegAddr, UsingRegs, MemAddr, UsingOverflow,
                                "vaarg.addr");

  // Load the pointer if the argument was passed indirectly.
  if (isIndirect) {
    Result = Address(Builder.CreateLoad(Result, "aggr"),
                     getContext().getTypeAlignInChars(Ty));
  }

  return Result;
}

bool PPC32TargetCodeGenInfo::isStructReturnInRegABI(
    const llvm::Triple &Triple, const CodeGenOptions &Opts) {
  assert(Triple.isPPC32());

  switch (Opts.getStructReturnConvention()) {
  case CodeGenOptions::SRCK_Default:
    break;
  case CodeGenOptions::SRCK_OnStack: // -maix-struct-return
    return false;
  case CodeGenOptions::SRCK_InRegs: // -msvr4-struct-return
    return true;
  }

  if (Triple.isOSBinFormatELF() && !Triple.isOSLinux())
    return true;

  return false;
}

bool
PPC32TargetCodeGenInfo::initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                                                llvm::Value *Address) const {
  return PPC_initDwarfEHRegSizeTable(CGF, Address, /*Is64Bit*/ false,
                                     /*IsAIX*/ false);
}

// PowerPC-64

namespace {
/// PPC64_SVR4_ABIInfo - The 64-bit PowerPC ELF (SVR4) ABI information.
class PPC64_SVR4_ABIInfo : public SwiftABIInfo {
public:
  enum ABIKind {
    ELFv1 = 0,
    ELFv2
  };

private:
  static const unsigned GPRBits = 64;
  ABIKind Kind;
  bool IsSoftFloatABI;

public:
  PPC64_SVR4_ABIInfo(CodeGen::CodeGenTypes &CGT, ABIKind Kind,
                     bool SoftFloatABI)
      : SwiftABIInfo(CGT), Kind(Kind), IsSoftFloatABI(SoftFloatABI) {}

  bool isPromotableTypeForABI(QualType Ty) const;
  CharUnits getParamTypeAlignment(QualType Ty) const;

  ABIArgInfo classifyReturnType(QualType RetTy) const;
  ABIArgInfo classifyArgumentType(QualType Ty) const;

  bool isHomogeneousAggregateBaseType(QualType Ty) const override;
  bool isHomogeneousAggregateSmallEnough(const Type *Ty,
                                         uint64_t Members) const override;

  // TODO: We can add more logic to computeInfo to improve performance.
  // Example: For aggregate arguments that fit in a register, we could
  // use getDirectInReg (as is done below for structs containing a single
  // floating-point value) to avoid pushing them to memory on function
  // entry.  This would require changing the logic in PPCISelLowering
  // when lowering the parameters in the caller and args in the callee.
  void computeInfo(CGFunctionInfo &FI) const override {
    if (!getCXXABI().classifyReturnType(FI))
      FI.getReturnInfo() = classifyReturnType(FI.getReturnType());
    for (auto &I : FI.arguments()) {
      // We rely on the default argument classification for the most part.
      // One exception:  An aggregate containing a single floating-point
      // or vector item must be passed in a register if one is available.
      const Type *T = isSingleElementStruct(I.type, getContext());
      if (T) {
        const BuiltinType *BT = T->getAs<BuiltinType>();
        if ((T->isVectorType() && getContext().getTypeSize(T) == 1284) ||
            (BT && BT->isFloatingPoint()19)) {
          QualType QT(T, 0);
          I.info = ABIArgInfo::getDirectInReg(CGT.ConvertType(QT));
          continue;
        }
      }
      I.info = classifyArgumentType(I.type);
    }
  }

  Address EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
                    QualType Ty) const override;

  bool shouldPassIndirectlyForSwift(ArrayRef<llvm::Type*> scalars,
                                    bool asReturnValue) const override {
    return occupiesMoreThan(CGT, scalars, /*total*/ 4);
  }

  bool isSwiftErrorInRegister() const override {
    return false;
  }
};

class PPC64_SVR4_TargetCodeGenInfo : public TargetCodeGenInfo {

public:
  PPC64_SVR4_TargetCodeGenInfo(CodeGenTypes &CGT,
                               PPC64_SVR4_ABIInfo::ABIKind Kind,
                               bool SoftFloatABI)
      : TargetCodeGenInfo(
            std::make_unique<PPC64_SVR4_ABIInfo>(CGT, Kind, SoftFloatABI)) {}

  int getDwarfEHStackPointer(CodeGen::CodeGenModule &M) const override {
    // This is recovered from gcc output.
    return 1; // r1 is the dedicated stack pointer
  }

  bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                               llvm::Value *Address) const override;
};

class PPC64TargetCodeGenInfo : public DefaultTargetCodeGenInfo {
public:
  PPC64TargetCodeGenInfo(CodeGenTypes &CGT) : DefaultTargetCodeGenInfo(CGT) {}

  int getDwarfEHStackPointer(CodeGen::CodeGenModule &M) const override {
    // This is recovered from gcc output.
    return 1; // r1 is the dedicated stack pointer
  }

  bool initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                               llvm::Value *Address) const override;
};

}

// Return true if the ABI requires Ty to be passed sign- or zero-
// extended to 64 bits.
bool
PPC64_SVR4_ABIInfo::isPromotableTypeForABI(QualType Ty) const {
  // Treat an enum type as its underlying type.
  if (const EnumType *EnumTy = Ty->getAs<EnumType>())
    Ty = EnumTy->getDecl()->getIntegerType();

  // Promotable integer types are required to be promoted by the ABI.
  if (isPromotableIntegerTypeForABI(Ty))
    return true;

  // In addition to the usual promotable integer types, we also need to
  // extend all 32-bit types, since the ABI requires promotion to 64 bits.
  if (const BuiltinType *BT = Ty->getAs<BuiltinType>())
    switch (BT->getKind()) {
    case BuiltinType::Int:
    case BuiltinType::UInt:
      return true;
    default:
      break;
    }

  if (const auto *EIT = Ty->getAs<ExtIntType>())
    if (EIT->getNumBits() < 64)
      return true;

  return false;
}

/// isAlignedParamType - Determine whether a type requires 16-byte or
/// higher alignment in the parameter area.  Always returns at least 8.
CharUnits PPC64_SVR4_ABIInfo::getParamTypeAlignment(QualType Ty) const {
  // Complex types are passed just like their elements.
  if (const ComplexType *CTy = Ty->getAs<ComplexType>())
    Ty = CTy->getElementType();

  // Only vector types of size 16 bytes need alignment (larger types are
  // passed via reference, smaller types are not aligned).
  if (Ty->isVectorType()) {
    return CharUnits::fromQuantity(getContext().getTypeSize(Ty) == 128 ? 16 : 8);
  } else if (Ty->isRealFloatingType() &&
             &getContext().getFloatTypeSemantics(Ty) ==
                 &llvm::APFloat::IEEEquad()) {
    // According to ABI document section 'Optional Save Areas': If extended
    // precision floating-point values in IEEE BINARY 128 QUADRUPLE PRECISION
    // format are supported, map them to a single quadword, quadword aligned.
    return CharUnits::fromQuantity(16);
  }

  // For single-element float/vector structs, we consider the whole type
  // to have the same alignment requirements as its single element.
  const Type *AlignAsType = nullptr;
  const Type *EltType = isSingleElementStruct(Ty, getContext());
  if (EltType) {
    const BuiltinType *BT = EltType->getAs<BuiltinType>();
    if ((EltType->isVectorType() && getContext().getTypeSize(EltType) == 1282) ||
        (BT && BT->isFloatingPoint()1))
      AlignAsType = EltType;
  }

  // Likewise for ELFv2 homogeneous aggregates.
  const Type *Base = nullptr;
  uint64_t Members = 0;
  if (!AlignAsType && Kind == ELFv2127 &&
      isAggregateTypeForABI(Ty) && isHomogeneousAggregate(Ty, Base, Members)82)
    AlignAsType = Base;

  // With special case aggregates, only vector base types need alignment.
  if (AlignAsType) {
    return CharUnits::fromQuantity(AlignAsType->isVectorType() ? 1619 : 8);
  }

  // Otherwise, we only need alignment for any aggregate type that
  // has an alignment requirement of >= 16 bytes.
  if (isAggregateTypeForABI(Ty) && getContext().getTypeAlign(Ty) >= 12863) {
    return CharUnits::fromQuantity(16);
  }

  return CharUnits::fromQuantity(8);
}

/// isHomogeneousAggregate - Return true if a type is an ELFv2 homogeneous
/// aggregate.  Base is set to the base element type, and Members is set
/// to the number of base elements.
bool ABIInfo::isHomogeneousAggregate(QualType Ty, const Type *&Base,
                                     uint64_t &Members) const {
  if (const ConstantArrayType *AT = getContext().getAsConstantArrayType(Ty)) {
    uint64_t NElements = AT->getSize().getZExtValue();
    if (NElements == 0)
      return false;
    if (!isHomogeneousAggregate(AT->getElementType(), Base, Members))
      return false;
    Members *= NElements;
  } else if (const RecordType *RT = Ty->getAs<RecordType>()) {
    const RecordDecl *RD = RT->getDecl();
    if (RD->hasFlexibleArrayMember())
      return false;

    Members = 0;

    // If this is a C++ record, check the properties of the record such as
    // bases and ABI specific restrictions
    if (const CXXRecordDecl *CXXRD = dyn_cast<CXXRecordDecl>(RD)) {
      if (!getCXXABI().isPermittedToBeHomogeneousAggregate(CXXRD))
        return false;

      for (const auto &I : CXXRD->bases()) {
        // Ignore empty records.
        if (isEmptyRecord(getContext(), I.getType(), true))
          continue;

        uint64_t FldMembers;
        if (!isHomogeneousAggregate(I.getType(), Base, FldMembers))
          return false;

        Members += FldMembers;
      }
    }

    for (const auto *FD : RD->fields())1.94k {
      // Ignore (non-zero arrays of) empty records.
      QualType FT = FD->getType();
      while (const ConstantArrayType *AT =
             getContext().getAsConstantArrayType(FT)) {
        if (AT->getSize().getZExtValue() == 0)
          return false;
        FT = AT->getElementType();
      }
      if (isEmptyRecord(getContext(), FT, true))
        continue;

      // For compatibility with GCC, ignore empty bitfields in C++ mode.
      if (getContext().getLangOpts().CPlusPlus &&
          FD->isZeroLengthBitField(getContext()))
        continue;

      uint64_t FldMembers;
      if (!isHomogeneousAggregate(FD->getType(), Base, FldMembers))
        return false;

      Members = (RD->isUnion() ?
                 std::max(Members, FldMembers)27 : Members + FldMembers);
    }

    if (!Base)
      return false;

    // Ensure there is no padding.
    if (getContext().getTypeSize(Base) * Members !=
        getContext().getTypeSize(Ty))
      return false;
  } else {
    Members = 1;
    if (const ComplexType *CT = Ty->getAs<ComplexType>()) {
      Members = 2;
      Ty = CT->getElementType();
    }

    // Most ABIs only support float, double, and some vector type widths.
    if (!isHomogeneousAggregateBaseType(Ty))
      return false;

    // The base type must be the same for all members.  Types that
    // agree in both total size and mode (float vs. vector) are
    // treated as being equivalent here.
    const Type *TyPtr = Ty.getTypePtr();
    if (!Base) {
      Base = TyPtr;
      // If it's a non-power-of-2 vector, its size is already a power-of-2,
      // so make sure to widen it explicitly.
      if (const VectorType *VT = Base->getAs<VectorType>()) {
        QualType EltTy = VT->getElementType();
        unsigned NumElements =
            getContext().getTypeSize(VT) / getContext().getTypeSize(EltTy);
        Base = getContext()
                   .getVectorType(EltTy, NumElements, VT->getVectorKind())
                   .getTypePtr();
      }
    }

    if (Base->isVectorType() != TyPtr->isVectorType() ||
        getContext().getTypeSize(Base) != getContext().getTypeSize(TyPtr))
      return false;
  }
  return Members > 0 && isHomogeneousAggregateSmallEnough(Base, Members);
}

bool PPC64_SVR4_ABIInfo::isHomogeneousAggregateBaseType(QualType Ty) const {
  // Homogeneous aggregates for ELFv2 must have base types of float,
  // double, long double, or 128-bit vectors.
  if (const BuiltinType *BT = Ty->getAs<BuiltinType>()) {
    if (BT->getKind() == BuiltinType::Float ||
        BT->getKind() == BuiltinType::Double ||
        BT->getKind() == BuiltinType::LongDouble ||
        (getContext().getTargetInfo().hasFloat128Type() &&
          (BT->getKind() == BuiltinType::Float128)46)) {
      if (IsSoftFloatABI)
        return false;
      return true;
    }
  }
  if (const VectorType *VT = Ty->getAs<VectorType>()) {
    if (getContext().getTypeSize(VT) == 128)
      return true;
  }
  return false;
}

bool PPC64_SVR4_ABIInfo::isHomogeneousAggregateSmallEnough(
    const Type *Base, uint64_t Members) const {
  // Vector and fp128 types require one register, other floating point types
  // require one or two registers depending on their size.
  uint32_t NumRegs =
      ((getContext().getTargetInfo().hasFloat128Type() &&
          Base->isFloat128Type()) ||
        Base->isVectorType()) ? 1294
                              : (getContext().getTypeSize(Base) + 63) / 64;

  // Homogeneous Aggregates may occupy at most 8 registers.
  return Members * NumRegs <= 8;
}

ABIArgInfo
PPC64_SVR4_ABIInfo::classifyArgumentType(QualType Ty) const {
  Ty = useFirstFieldIfTransparentUnion(Ty);

  if (Ty->isAnyComplexType())
    return ABIArgInfo::getDirect();

  // Non-Altivec vector types are passed in GPRs (smaller than 16 bytes)
  // or via reference (larger than 16 bytes).
  if (Ty->isVectorType()) {
    uint64_t Size = getContext().getTypeSize(Ty);
    if (Size > 128)
      return getNaturalAlignIndirect(Ty, /*ByVal=*/false);
    else if (Size < 128) {
      llvm::Type *CoerceTy = llvm::IntegerType::get(getVMContext(), Size);
      return ABIArgInfo::getDirect(CoerceTy);
    }
  }

  if (const auto *EIT = Ty->getAs<ExtIntType>())
    if (EIT->getNumBits() > 128)
      return getNaturalAlignIndirect(Ty, /*ByVal=*/true);

  if (isAggregateTypeForABI(Ty)) {
    if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI()))
      return getNaturalAlignIndirect(Ty, RAA == CGCXXABI::RAA_DirectInMemory);

    uint64_t ABIAlign = getParamTypeAlignment(Ty).getQuantity();
    uint64_t TyAlign = getContext().getTypeAlignInChars(Ty).getQuantity();

    // ELFv2 homogeneous aggregates are passed as array types.
    const Type *Base = nullptr;
    uint64_t Members = 0;
    if (Kind == ELFv2 &&
        isHomogeneousAggregate(Ty, Base, Members)) {
      llvm::Type *BaseTy = CGT.ConvertType(QualType(Base, 0));
5226

Coverage Report

Created: 2021-01-19 06:58