Google Git
Sign in
chromium / external / github.com / emscripten-core / emscripten-fastcomp-clang / refs/heads/back-to-llvm / . / lib / CodeGen / TargetInfo.cpp
blob: 55d8deeadbc6ff7050aadcdfa90f3fee0219a2bc [file] [log] [blame]
//===---- TargetInfo.cpp - Encapsulate target details -----------*- C++ -*-===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// These classes wrap the information about a call or function
// definition used to handle ABI compliancy.
//
//===----------------------------------------------------------------------===//
#include "TargetInfo.h"
#include "ABIInfo.h"
#include "CGCXXABI.h"
#include "CGValue.h"
#include "CodeGenFunction.h"
#include "clang/AST/RecordLayout.h"
#include "clang/CodeGen/CGFunctionInfo.h"
#include "clang/Frontend/CodeGenOptions.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/ADT/Triple.h"
#include "llvm/IR/DataLayout.h"
#include "llvm/IR/Type.h"
#include "llvm/Support/raw_ostream.h"
#include <algorithm> // std::sort
using namespace clang;
using namespace CodeGen;
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; ++I) {
llvm::Value *Cell =
Builder.CreateConstInBoundsGEP1_32(Builder.getInt8Ty(), Array, I);
Builder.CreateStore(Value, Cell);
}
}
static bool isAggregateTypeForABI(QualType T) {
return !CodeGenFunction::hasScalarEvaluationKind(T) ||
T->isMemberFunctionPointerType();
}
ABIInfo::~ABIInfo() {}
static CGCXXABI::RecordArgABI getRecordArgABI(const RecordType *RT,
CGCXXABI &CXXABI) {
const CXXRecordDecl *RD = dyn_cast<CXXRecordDecl>(RT->getDecl());
if (!RD)
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);
}
/// 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();
}
bool ABIInfo::isHomogeneousAggregateBaseType(QualType Ty) const {
return false;
}
bool ABIInfo::isHomogeneousAggregateSmallEnough(const Type *Base,
uint64_t Members) const {
return false;
}
bool ABIInfo::shouldSignExtUnsignedType(QualType Ty) const {
return false;
}
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()
<< " ByVal=" << getIndirectByVal()
<< " Realign=" << getIndirectRealign();
break;
case Expand:
OS << "Expand";
break;
}
OS << ")\n";
}
TargetCodeGenInfo::~TargetCodeGenInfo() { delete Info; }
// 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;
}
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.
if (AllowArrays)
while (const ConstantArrayType *AT = Context.getAsConstantArrayType(FT)) {
if (AT->getSize() == 0)
return true;
FT = AT->getElementType();
}
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.
if (isa<CXXRecordDecl>(RT->getDecl()))
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 0;
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()) {
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 *AT = 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))
return nullptr;
return Found;
}
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() &&
!Ty->isEnumeralType() && !Ty->isBlockPointerType())
return false;
uint64_t Size = Context.getTypeSize(Ty);
return Size == 32 || Size == 64;
}
/// canExpandIndirectArgument - 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.
///
// FIXME: This predicate is missing many cases, currently it just follows
// llvm-gcc (checks that all fields are 32-bit or 64-bit primitive types). We
// should probably make this smarter, or better yet make the LLVM backend
// capable of handling it.
static bool canExpandIndirectArgument(QualType Ty, ASTContext &Context) {
// We can only expand structure types.
const RecordType *RT = Ty->getAs<RecordType>();
if (!RT)
return false;
// We can only expand (C) structures.
//
// FIXME: This needs to be generalized to handle classes as well.
const RecordDecl *RD = RT->getDecl();
if (!RD->isStruct())
return false;
// We try to expand CLike CXXRecordDecl.
if (const CXXRecordDecl *CXXRD = dyn_cast<CXXRecordDecl>(RD)) {
if (!CXXRD->isCLike())
return false;
}
uint64_t Size = 0;
for (const auto *FD : RD->fields()) {
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());
}
// Make sure there are not any holes in the struct.
if (Size != Context.getTypeSize(Ty))
return false;
return true;
}
namespace {
/// 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);
}
llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
CodeGenFunction &CGF) const override;
};
class DefaultTargetCodeGenInfo : public TargetCodeGenInfo {
public:
DefaultTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT)
: TargetCodeGenInfo(new DefaultABIInfo(CGT)) {}
};
llvm::Value *DefaultABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
CodeGenFunction &CGF) const {
return nullptr;
}
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 ABIArgInfo::getIndirect(0, RAA == CGCXXABI::RAA_DirectInMemory);
return ABIArgInfo::getIndirect(0);
}
// Treat an enum type as its underlying type.
if (const EnumType *EnumTy = Ty->getAs<EnumType>())
Ty = EnumTy->getDecl()->getIntegerType();
return (Ty->isPromotableIntegerType() ?
ABIArgInfo::getExtend() : ABIArgInfo::getDirect());
}
ABIArgInfo DefaultABIInfo::classifyReturnType(QualType RetTy) const {
if (RetTy->isVoidType())
return ABIArgInfo::getIgnore();
if (isAggregateTypeForABI(RetTy))
return ABIArgInfo::getIndirect(0);
// Treat an enum type as its underlying type.
if (const EnumType *EnumTy = RetTy->getAs<EnumType>())
RetTy = EnumTy->getDecl()->getIntegerType();
return (RetTy->isPromotableIntegerType() ?
ABIArgInfo::getExtend() : ABIArgInfo::getDirect());
}
// @LOCALMOD-START Emscripten
//===----------------------------------------------------------------------===//
// Emscripten ABI Implementation
//
// This is a very simple ABI that relies a lot on DefaultABIInfo.
//===----------------------------------------------------------------------===//
class EmscriptenABIInfo : public DefaultABIInfo {
public:
explicit EmscriptenABIInfo(CodeGen::CodeGenTypes &CGT)
: DefaultABIInfo(CGT) {}
ABIArgInfo classifyReturnType(QualType RetTy) const;
ABIArgInfo classifyArgumentType(QualType Ty) const;
// DefaultABIInfo's classifyReturnType and classifyArgumentType are
// non-virtual, but computeInfo is virtual, so we overload that.
void computeInfo(CGFunctionInfo &FI) const override {
FI.getReturnInfo() = classifyReturnType(FI.getReturnType());
for (auto &Arg : FI.arguments())
Arg.info = classifyArgumentType(Arg.type);
}
};
class EmscriptenTargetCodeGenInfo : public TargetCodeGenInfo {
public:
explicit EmscriptenTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT)
: TargetCodeGenInfo(new EmscriptenABIInfo(CGT)) {}
};
/// \brief Classify argument of given type \p Ty.
ABIArgInfo EmscriptenABIInfo::classifyArgumentType(QualType Ty) const {
if (isAggregateTypeForABI(Ty)) {
unsigned TypeAlign = getContext().getTypeAlignInChars(Ty).getQuantity();
if (auto RAA = getRecordArgABI(Ty, getCXXABI()))
return ABIArgInfo::getIndirect(TypeAlign,
RAA == CGCXXABI::RAA_DirectInMemory);
return ABIArgInfo::getIndirect(TypeAlign);
}
// Otherwise just do the default thing.
return DefaultABIInfo::classifyArgumentType(Ty);
}
ABIArgInfo EmscriptenABIInfo::classifyReturnType(QualType RetTy) const {
if (isAggregateTypeForABI(RetTy)) {
// As an optimization, lower single-element structs to just return a regular
// value.
if (const Type *SeltTy = isSingleElementStruct(RetTy, getContext()))
return ABIArgInfo::getDirect(CGT.ConvertType(QualType(SeltTy, 0)));
}
// Otherwise just do the default thing.
return DefaultABIInfo::classifyReturnType(RetTy);
}
// @LOCALMOD-END Emscripten
//===----------------------------------------------------------------------===//
// 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;
llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
CodeGenFunction &CGF) const override;
};
class PNaClTargetCodeGenInfo : public TargetCodeGenInfo {
public:
PNaClTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT)
: TargetCodeGenInfo(new 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);
}
llvm::Value *PNaClABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
CodeGenFunction &CGF) const {
return nullptr;
}
/// \brief Classify argument of given type \p Ty.
ABIArgInfo PNaClABIInfo::classifyArgumentType(QualType Ty) const {
if (isAggregateTypeForABI(Ty)) {
if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI()))
return ABIArgInfo::getIndirect(0, RAA == CGCXXABI::RAA_DirectInMemory);
return ABIArgInfo::getIndirect(0);
} 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();
}
return (Ty->isPromotableIntegerType() ?
ABIArgInfo::getExtend() : 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 ABIArgInfo::getIndirect(0);
// Treat an enum type as its underlying type.
if (const EnumType *EnumTy = RetTy->getAs<EnumType>())
RetTy = EnumTy->getDecl()->getIntegerType();
return (RetTy->isPromotableIntegerType() ?
ABIArgInfo::getExtend() : 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) {
if ((Constraint == "y" || Constraint == "&y") && Ty->isVectorTy()) {
if (cast<llvm::VectorType>(Ty)->getBitWidth() != 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::Half)
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 == 256 || VecSize == 512)
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;
}
//===----------------------------------------------------------------------===//
// X86-32 ABI Implementation
//===----------------------------------------------------------------------===//
/// \brief Similar to llvm::CCState, but for Clang.
struct CCState {
CCState(unsigned CC) : CC(CC), FreeRegs(0), FreeSSERegs(0) {}
unsigned CC;
unsigned FreeRegs;
unsigned FreeSSERegs;
};
/// X86_32ABIInfo - The X86-32 ABI information.
class X86_32ABIInfo : public ABIInfo {
enum Class {
Integer,
Float
};
static const unsigned MinABIStackAlignInBytes = 4;
bool IsDarwinVectorABI;
bool IsSmallStructInRegABI;
bool IsWin32StructABI;
unsigned DefaultNumRegisterParameters;
static bool isRegisterSize(unsigned Size) {
return (Size == 8 || Size == 16 || Size == 32 || Size == 64);
}
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(CCState &State) const;
/// \brief 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;
bool shouldUseInReg(QualType Ty, CCState &State, bool &NeedsPadding) const;
/// \brief Rewrite the function info so that all memory arguments use
/// inalloca.
void rewriteWithInAlloca(CGFunctionInfo &FI) const;
void addFieldToArgStruct(SmallVector<llvm::Type *, 6> &FrameFields,
unsigned &StackOffset, ABIArgInfo &Info,
QualType Type) const;
public:
void computeInfo(CGFunctionInfo &FI) const override;
llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
CodeGenFunction &CGF) const override;
X86_32ABIInfo(CodeGen::CodeGenTypes &CGT, bool d, bool p, bool w,
unsigned r)
: ABIInfo(CGT), IsDarwinVectorABI(d), IsSmallStructInRegABI(p),
IsWin32StructABI(w), DefaultNumRegisterParameters(r) {}
};
class X86_32TargetCodeGenInfo : public TargetCodeGenInfo {
public:
X86_32TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT,
bool d, bool p, bool w, unsigned r)
:TargetCodeGenInfo(new X86_32ABIInfo(CGT, d, p, w, r)) {}
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
('F' << 16) |
('T' << 24);
return llvm::ConstantInt::get(CGM.Int32Ty, Sig);
}
};
}
/// 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()) {
// We have an operand reference.
size_t DigitStart = Pos;
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(),
CoerceTy->getPointerTo()));
ResultRegDests.push_back(ReturnSlot);
rewriteInputConstraintReferences(NumOutputs, 1, AsmString);
}
/// shouldReturnTypeInRegister - Determine if the given type should be
/// passed in a register (for the Darwin ABI).
bool X86_32ABIInfo::shouldReturnTypeInRegister(QualType Ty,
ASTContext &Context) const {
uint64_t Size = Context.getTypeSize(Ty);
// Type must be register sized.
if (!isRegisterSize(Size))
return false;
if (Ty->isVectorType()) {
// 64- and 128- bit vectors inside structures are not returned in
// registers.
if (Size == 64 || Size == 128)
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() ||
Ty->isAnyComplexType() || Ty->isEnumeralType() ||
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 false;
// 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()) {
// Empty fields are ignored.
if (isEmptyField(Context, FD, true))
continue;
// Check fields recursively.
if (!shouldReturnTypeInRegister(FD->getType(), Context))
return false;
}
return true;
}
ABIArgInfo X86_32ABIInfo::getIndirectReturnResult(CCState &State) const {
// If the return value is indirect, then the hidden argument is consuming one
// integer register.
if (State.FreeRegs) {
--State.FreeRegs;
return ABIArgInfo::getIndirectInReg(/*Align=*/0, /*ByVal=*/false);
}
return ABIArgInfo::getIndirect(/*Align=*/0, /*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 &&
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::VectorType::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(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(State);
}
// If specified, structs and unions are always indirect.
if (!IsSmallStructInRegABI && !RetTy->isAnyComplexType())
return getIndirectReturnResult(State);
// 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())
|| 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(State);
}
// Treat an enum type as its underlying type.
if (const EnumType *EnumTy = RetTy->getAs<EnumType>())
RetTy = EnumTy->getDecl()->getIntegerType();
return (RetTy->isPromotableIntegerType() ?
ABIArgInfo::getExtend() : ABIArgInfo::getDirect());
}
static bool isSSEVectorType(ASTContext &Context, QualType Ty) {
return Ty->getAs<VectorType>() && Context.getTypeSize(Ty) == 128;
}
static bool isRecordWithSSEVectorType(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 (!isRecordWithSSEVectorType(Context, I.getType()))
return false;
for (const auto *i : RD->fields()) {
QualType FT = i->getType();
if (isSSEVectorType(Context, FT))
return true;
if (isRecordWithSSEVectorType(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 && (isSSEVectorType(getContext(), Ty) ||
isRecordWithSSEVectorType(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.
return ABIArgInfo::getIndirectInReg(0, false);
}
return ABIArgInfo::getIndirect(0, false);
}
// Compute the byval alignment.
unsigned TypeAlign = getContext().getTypeAlign(Ty) / 8;
unsigned StackAlign = getTypeStackAlignInBytes(Ty, TypeAlign);
if (StackAlign == 0)
return ABIArgInfo::getIndirect(4, /*ByVal=*/true);
// If the stack alignment is less than the type alignment, realign the
// argument.
bool Realign = TypeAlign > StackAlign;
return ABIArgInfo::getIndirect(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::Double)
return Float;
}
return Integer;
}
bool X86_32ABIInfo::shouldUseInReg(QualType Ty, CCState &State,
bool &NeedsPadding) const {
NeedsPadding = false;
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 (SizeInRegs > State.FreeRegs) {
State.FreeRegs = 0;
return false;
}
State.FreeRegs -= SizeInRegs;
if (State.CC == llvm::CallingConv::X86_FastCall ||
State.CC == llvm::CallingConv::X86_VectorCall) {
if (Size > 32)
return false;
if (Ty->isIntegralOrEnumerationType())
return true;
if (Ty->isPointerType())
return true;
if (Ty->isReferenceType())
return true;
if (State.FreeRegs)
NeedsPadding = true;
return false;
}
return true;
}
ABIArgInfo X86_32ABIInfo::classifyArgumentType(QualType Ty,
CCState &State) const {
// FIXME: Set alignment on indirect arguments.
Ty = useFirstFieldIfTransparentUnion(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);
}
}
// vectorcall adds the concept of a homogenous vector aggregate, similar
// to other targets.
const Type *Base = nullptr;
uint64_t NumElts = 0;
if (State.CC == llvm::CallingConv::X86_VectorCall &&
isHomogeneousAggregate(Ty, Base, NumElts)) {
if (State.FreeSSERegs >= NumElts) {
State.FreeSSERegs -= NumElts;
if (Ty->isBuiltinType() || Ty->isVectorType())
return ABIArgInfo::getDirect();
return ABIArgInfo::getExpand();
}
return getIndirectResult(Ty, /*ByVal=*/false, State);
}
if (isAggregateTypeForABI(Ty)) {
if (RT) {
// Structs are always byval on win32, regardless of what they contain.
if (IsWin32StructABI)
return getIndirectResult(Ty, true, State);
// Structures with flexible arrays are always indirect.
if (RT->getDecl()->hasFlexibleArrayMember())
return getIndirectResult(Ty, true, State);
}
// Ignore empty structs/unions.
if (isEmptyRecord(getContext(), Ty, true))
return ABIArgInfo::getIgnore();
llvm::LLVMContext &LLVMContext = getVMContext();
llvm::IntegerType *Int32 = llvm::Type::getInt32Ty(LLVMContext);
bool NeedsPadding;
if (shouldUseInReg(Ty, State, NeedsPadding)) {
unsigned SizeInRegs = (getContext().getTypeSize(Ty) + 31) / 32;
SmallVector<llvm::Type*, 3> Elements(SizeInRegs, Int32);
llvm::Type *Result = llvm::StructType::get(LLVMContext, Elements);
return ABIArgInfo::getDirectInReg(Result);
}
llvm::IntegerType *PaddingType = NeedsPadding ? Int32 : nullptr;
// 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.
if (getContext().getTypeSize(Ty) <= 4*32 &&
canExpandIndirectArgument(Ty, getContext()))
return ABIArgInfo::getExpandWithPadding(
State.CC == llvm::CallingConv::X86_FastCall ||
State.CC == llvm::CallingConv::X86_VectorCall,
PaddingType);
return getIndirectResult(Ty, true, State);
}
if (const VectorType *VT = Ty->getAs<VectorType>()) {
// On Darwin, some vectors are passed in memory, we handle this by passing
// it as an i8/i16/i32/i64.
if (IsDarwinVectorABI) {
uint64_t Size = getContext().getTypeSize(Ty);
if ((Size == 8 || Size == 16 || Size == 32) ||
(Size == 64 && VT->getNumElements() == 1))
return ABIArgInfo::getDirect(llvm::IntegerType::get(getVMContext(),
Size));
}
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 NeedsPadding;
bool InReg = shouldUseInReg(Ty, State, NeedsPadding);
if (Ty->isPromotableIntegerType()) {
if (InReg)
return ABIArgInfo::getExtendInReg();
return ABIArgInfo::getExtend();
}
if (InReg)
return ABIArgInfo::getDirectInReg();
return ABIArgInfo::getDirect();
}
void X86_32ABIInfo::computeInfo(CGFunctionInfo &FI) const {
CCState State(FI.getCallingConvention());
if (State.CC == llvm::CallingConv::X86_FastCall)
State.FreeRegs = 2;
else if (State.CC == llvm::CallingConv::X86_VectorCall) {
State.FreeRegs = 2;
State.FreeSSERegs = 6;
} else if (FI.getHasRegParm())
State.FreeRegs = FI.getRegParm();
else
State.FreeRegs = DefaultNumRegisterParameters;
if (!getCXXABI().classifyReturnType(FI)) {
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.
FI.getReturnInfo().setInReg(true);
}
}
// The chain argument effectively gives us another free register.
if (FI.isChainCall())
++State.FreeRegs;
bool UsedInAlloca = false;
for (auto &I : FI.arguments()) {
I.info = classifyArgumentType(I.type, State);
UsedInAlloca |= (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,
unsigned &StackOffset,
ABIArgInfo &Info, QualType Type) const {
assert(StackOffset % 4U == 0 && "unaligned inalloca struct");
Info = ABIArgInfo::getInAlloca(FrameFields.size());
FrameFields.push_back(CGT.ConvertTypeForMem(Type));
StackOffset += getContext().getTypeSizeInChars(Type).getQuantity();
// Insert padding bytes to respect alignment. For x86_32, each argument is 4
// byte aligned.
if (StackOffset % 4U) {
unsigned OldOffset = StackOffset;
StackOffset = llvm::RoundUpToAlignment(StackOffset, 4U);
unsigned NumBytes = StackOffset - OldOffset;
assert(NumBytes);
llvm::Type *Ty = llvm::Type::getInt8Ty(getVMContext());
Ty = llvm::ArrayType::get(Ty, NumBytes);
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::Indirect:
assert(Info.getIndirectByVal());
return true;
case ABIArgInfo::Ignore:
return false;
case ABIArgInfo::Direct:
case ABIArgInfo::Extend:
case ABIArgInfo::Expand:
if (Info.getInReg())
return false;
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;
unsigned StackOffset = 0;
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() && !IsThisCall &&
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()) {
CanQualType PtrTy = getContext().getPointerType(FI.getReturnType());
addFieldToArgStruct(FrameFields, StackOffset, Ret, PtrTy);
// 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; ++I) {
if (isArgInAlloca(I->info))
addFieldToArgStruct(FrameFields, StackOffset, I->info, I->type);
}
FI.setArgStruct(llvm::StructType::get(getVMContext(), FrameFields,
/*isPacked=*/true));
}
llvm::Value *X86_32ABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
CodeGenFunction &CGF) const {
llvm::Type *BPP = CGF.Int8PtrPtrTy;
CGBuilderTy &Builder = CGF.Builder;
llvm::Value *VAListAddrAsBPP = Builder.CreateBitCast(VAListAddr, BPP,
"ap");
llvm::Value *Addr = Builder.CreateLoad(VAListAddrAsBPP, "ap.cur");
// Compute if the address needs to be aligned
unsigned Align = CGF.getContext().getTypeAlignInChars(Ty).getQuantity();
Align = getTypeStackAlignInBytes(Ty, Align);
Align = std::max(Align, 4U);
if (Align > 4) {
// addr = (addr + align - 1) & -align;
llvm::Value *Offset =
llvm::ConstantInt::get(CGF.Int32Ty, Align - 1);
Addr = CGF.Builder.CreateGEP(Addr, Offset);
llvm::Value *AsInt = CGF.Builder.CreatePtrToInt(Addr,
CGF.Int32Ty);
llvm::Value *Mask = llvm::ConstantInt::get(CGF.Int32Ty, -Align);
Addr = CGF.Builder.CreateIntToPtr(CGF.Builder.CreateAnd(AsInt, Mask),
Addr->getType(),
"ap.cur.aligned");
}
llvm::Type *PTy =
llvm::PointerType::getUnqual(CGF.ConvertType(Ty));
llvm::Value *AddrTyped = Builder.CreateBitCast(Addr, PTy);
uint64_t Offset =
llvm::RoundUpToAlignment(CGF.getContext().getTypeSize(Ty) / 8, Align);
llvm::Value *NextAddr =
Builder.CreateGEP(Addr, llvm::ConstantInt::get(CGF.Int32Ty, Offset),
"ap.next");
Builder.CreateStore(NextAddr, VAListAddrAsBPP);
return AddrTyped;
}
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())
return true;
switch (Triple.getOS()) {
case llvm::Triple::DragonFly:
case llvm::Triple::FreeBSD:
case llvm::Triple::OpenBSD:
case llvm::Triple::Bitrig:
case llvm::Triple::Win32:
return true;
default:
return false;
}
}
void X86_32TargetCodeGenInfo::setTargetAttributes(const Decl *D,
llvm::GlobalValue *GV,
CodeGen::CodeGenModule &CGM) const {
if (const FunctionDecl *FD = dyn_cast<FunctionDecl>(D)) {
if (FD->hasAttr<X86ForceAlignArgPointerAttr>()) {
// Get the LLVM function.
llvm::Function *Fn = cast<llvm::Function>(GV);
// Now add the 'alignstack' attribute with a value of 16.
llvm::AttrBuilder B;
B.addStackAlignmentAttr(16);
Fn->addAttributes(llvm::AttributeSet::FunctionIndex,
llvm::AttributeSet::get(CGM.getLLVMContext(),
llvm::AttributeSet::FunctionIndex,
B));
}
}
}
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.CreateStore(
Four8, Builder.CreateConstInBoundsGEP1_32(CGF.Int8Ty, Address, 9));
// 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 ABIInfo {
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;
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();
}
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) :
ABIInfo(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->getBitWidth() > 128);
}
return false;
}
void computeInfo(CGFunctionInfo &FI) const override;
llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
CodeGenFunction &CGF) const override;
bool has64BitPointers() const {
return Has64BitPointers;
}
};
/// WinX86_64ABIInfo - The Windows X86_64 ABI information.
class WinX86_64ABIInfo : public ABIInfo {
ABIArgInfo classify(QualType Ty, unsigned &FreeSSERegs,
bool IsReturnType) const;
public:
WinX86_64ABIInfo(CodeGen::CodeGenTypes &CGT) : ABIInfo(CGT) {}
void computeInfo(CGFunctionInfo &FI) const override;
llvm::Value *EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
CodeGenFunction &CGF) 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);
}
};
class X86_64TargetCodeGenInfo : public TargetCodeGenInfo {
public:
X86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, X86AVXABILevel AVXLevel)
: TargetCodeGenInfo(new X86_64ABIInfo(CGT, AVXLevel)) {}
const X86_64ABIInfo &getABIInfo() const {
return static_cast<const X86_64ABIInfo&>(TargetCodeGenInfo::getABIInfo());
}
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; ++it) {
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;
if (getABIInfo().has64BitPointers())
Sig = (0xeb << 0) | // jmp rel8
(0x0a << 8) | // .+0x0c
('F' << 16) |
('T' << 24);
else
Sig = (0xeb << 0) | // jmp rel8
(0x06 << 8) | // .+0x08
('F' << 16) |
('T' << 24);
return llvm::ConstantInt::get(CGM.Int32Ty, Sig);
}
};
class PS4TargetCodeGenInfo : public X86_64TargetCodeGenInfo {
public:
PS4TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT, X86AVXABILevel AVXLevel)
: X86_64TargetCodeGenInfo(CGT, AVXLevel) {}
void getDependentLibraryOption(llvm::StringRef Lib,
llvm::SmallString<24> &Opt) const override {
Opt = "\01";
// If the argument contains a space, enclose it in quotes.
if (Lib.find(" ") != StringRef::npos)
Opt += "\"" + Lib.str() + "\"";
else
Opt += Lib;
}
};
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 ? "\"" : "";
ArgStr += Lib;
if (!Lib.endswith_lower(".lib"))
ArgStr += ".lib";
ArgStr += Quote ? "\"" : "";
return ArgStr;
}
class WinX86_32TargetCodeGenInfo : public X86_32TargetCodeGenInfo {
public:
WinX86_32TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT,
bool d, bool p, bool w, unsigned RegParms)
: X86_32TargetCodeGenInfo(CGT, d, p, w, RegParms) {}
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 addStackProbeSizeTargetAttribute(const Decl *D,
llvm::GlobalValue *GV,
CodeGen::CodeGenModule &CGM) {
if (isa<FunctionDecl>(D)) {
if (CGM.getCodeGenOpts().StackProbeSize != 4096) {
llvm::Function *Fn = cast<llvm::Function>(GV);
Fn->addFnAttr("stack-probe-size",
llvm::utostr(CGM.getCodeGenOpts().StackProbeSize));
}
}
}
void WinX86_32TargetCodeGenInfo::setTargetAttributes(const Decl *D,
llvm::GlobalValue *GV,
CodeGen::CodeGenModule &CGM) const {
X86_32TargetCodeGenInfo::setTargetAttributes(D, GV, CGM);
addStackProbeSizeTargetAttribute(D, GV, CGM);
}
class WinX86_64TargetCodeGenInfo : public TargetCodeGenInfo {
public:
WinX86_64TargetCodeGenInfo(CodeGen::CodeGenTypes &CGT,
X86AVXABILevel AVXLevel)
: TargetCodeGenInfo(new WinX86_64ABIInfo(CGT)) {}
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);
addStackProbeSizeTargetAttribute(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 != X87 && honorsRevision0_98())
Lo = Memory;
if (AggregateSize > 128 && (Lo != SSE || Hi != SSEUp))
Lo = Memory;
if (Hi == SSEUp && Lo != SSE)
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 == NoClass)
return Accum;
if (Field == Memory)
return Memory;
if (Accum == NoClass)
return Field;
if (Accum == Integer || Field == Integer)
return Integer;
if (Field == X87 || Field == X87Up || Field == ComplexX87 ||
Accum == X87 || Accum == X87Up)
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 : Hi;
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::UInt128) {
Lo = Integer;
Hi = Integer;
} else if (k >= BuiltinType::Bool && k <= BuiltinType::LongLong) {
Current = Integer;
} else if ((k == BuiltinType::Float || k == BuiltinType::Double) ||
(k == BuiltinType::LongDouble &&
getTarget().getTriple().isOSNaCl())) {
Current = SSE;
} else if (k == BuiltinType::LongDouble) {
Lo = X87;
Hi = X87Up;
}
// 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 == 16 || Size == 32) {
// 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) {
// gcc passes <1 x double> in memory. :(
if (VT->getElementType()->isSpecificBuiltinType(BuiltinType::Double))
return;
// gcc passes <1 x long long> as INTEGER.
if (VT->getElementType()->isSpecificBuiltinType(BuiltinType::LongLong) ||
VT->getElementType()->isSpecificBuiltinType(BuiltinType::ULongLong) ||
VT->getElementType()->isSpecificBuiltinType(BuiltinType::Long) ||
VT->getElementType()->isSpecificBuiltinType(BuiltinType::ULong))
Current = Integer;
else
Current = SSE;
// If this type crosses an eightbyte boundary, it should be
// split.
if (OffsetBase && OffsetBase != 64)
Hi = Lo;
} else if (Size == 128 ||
(isNamedArg && Size <= getNativeVectorSizeForAVXABI(AVXLevel))) {
// 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 ||
(ET == getContext().LongDoubleTy &&
getTarget().getTriple().isOSNaCl()))
Lo = Hi = SSE;
else if (ET == getContext().LongDoubleTy)
Current = ComplexX87;
// 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_Imag)
Hi = Lo;
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 four eightbytes, ..., it has class MEMORY.
if (Size > 256)
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 && EltSize != 256)
return;
for (uint64_t i=0, Offset=OffsetBase; i<ArraySize; ++i, Offset += EltSize) {
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 four eightbytes, ..., it has class MEMORY.
if (Size > 256)
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 CXXRecordDecl *Base =
cast<CXXRecordDecl>(I.getType()->getAs<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 == Memory) {
postMerge(Size, Lo, Hi);
return;
}
}
}
// Classify the fields one at a time, merging the results.
unsigned idx = 0;
for (RecordDecl::field_iterator i = RD->field_begin(), e = RD->field_end();
i != e; ++i, ++idx) {
uint64_t Offset = OffsetBase + Layout.getFieldOffset(idx);
bool BitField = i->isBitField();
// AMD64-ABI 3.2.3p2: Rule 1. If the size of an object is larger than
// four eightbytes, or it contains unaligned fields, it has class MEMORY.
//
// The only case a 256-bit wide vector could be used is when the struct
// 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 && getContext().getTypeSize(i->getType()) != 256) {
Lo = Memory;
postMerge(Size, Lo, Hi);
return;
}
// Note, skip this test for bit-fields, see below.
if (!BitField && Offset % getContext().getTypeAlign(i->getType())) {
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) {
// Ignore padding bit-fields.
if (i->isUnnamedBitfield())
continue;
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 ? Integer : NoClass;
}
} else
classify(i->getType(), Offset, FieldLo, FieldHi, isNamedArg);
Lo = merge(Lo, FieldLo);
Hi = merge(Hi, FieldHi);
if (Lo == Memory || Hi == Memory)
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();
return (Ty->isPromotableIntegerType() ?
ABIArgInfo::getExtend() : ABIArgInfo::getDirect());
}
return ABIArgInfo::getIndirect(0);
}
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 > LargestVector)
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)) {
// Treat an enum type as its underlying type.
if (const EnumType *EnumTy = Ty->getAs<EnumType>())
Ty = EnumTy->getDecl()->getIntegerType();
return (Ty->isPromotableIntegerType() ?
ABIArgInfo::getExtend() : ABIArgInfo::getDirect());
}
if (CGCXXABI::RecordArgABI RAA = getRecordArgABI(Ty, getCXXABI()))
return ABIArgInfo::getIndirect(0, 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 <= 64)
return ABIArgInfo::getDirect(llvm::IntegerType::get(getVMContext(),
Size));
}
return ABIArgInfo::getIndirect(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))
return IRType;
// We couldn't find the preferred IR vector type for 'Ty'.
uint64_t Size = getContext().getTypeSize(Ty);
assert((Size == 128 || Size == 256) && "Invalid type found!");
// Return a LLVM IR vector type based on the size of 'Ty'.
return llvm::VectorType::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; ++i) {
// If the element is after the span we care about, then we're done..
unsigned EltOffset = i*EltSize;
if (EltOffset >= EndBit) break;
unsigned EltStart = EltOffset < StartBit ? StartBit-EltOffset :0;
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 CXXRecordDecl *Base =
cast<CXXRecordDecl>(I.getType()->getAs<RecordType>()->getDecl());
// If the base is after the span we care about, ignore it.
unsigned BaseOffset = Context.toBits(Layout.getBaseClassOffset(Base));
if (BaseOffset >= EndBit) continue;
unsigned BaseStart = BaseOffset < StartBit ? StartBit-BaseOffset :0;
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, ++idx) {
unsigned FieldOffset = (unsigned)Layout.getFieldOffset(idx);
// If we found a field after the region we care about, then we're done.
if (FieldOffset >= EndBit) break;
unsigned FieldStart = FieldOffset < StartBit ? StartBit-FieldOffset :0;
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())
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::VectorType::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) && Has64BitPointers) ||
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) ||
IRType->isIntegerTy(32) ||
(isa<llvm::PointerType>(IRType) && !Has64BitPointers)) {
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::RoundUpToAlignment(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, nullptr);
// 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)) {
// Treat an enum type as its underlying type.
if (const EnumType *EnumTy = RetTy->getAs<EnumType>())
RetTy = EnumTy->getDecl()->getIntegerType();
if (RetTy->isIntegralOrEnumerationType() &&
RetTy->isPromotableIntegerType())
return ABIArgInfo::getExtend();
}
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()),
nullptr);
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