Google Git
Sign in
llvm / clang / e8264bb9815647b4c6558aeb045c65965cdb26d3 / . / lib / CodeGen / TargetInfo.cpp
blob: f2352a373401205594cd407fee818b1c8464f6f7 [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 "CGBlocks.h"
#include "CGCXXABI.h"
#include "CGValue.h"
#include "CodeGenFunction.h"
#include "clang/AST/RecordLayout.h"
#include "clang/CodeGen/CGFunctionInfo.h"
#include "clang/CodeGen/SwiftCallingConv.h"
#include "clang/Frontend/CodeGenOptions.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/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; ++I) {
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 ByRef, bool Realign,
llvm::Type *Padding) const {
return ABIArgInfo::getIndirect(getContext().getTypeAlignInChars(Ty),
ByRef, Realign, Padding);
}
ABIArgInfo
ABIInfo::getNaturalAlignIndirectInReg(QualType Ty, bool Realign) const {
return ABIArgInfo::getIndirectInReg(getContext().getTypeAlignInChars(Ty),
/*ByRef*/ false, Realign);
}
Address ABIInfo::EmitMSVAArg(CodeGenFunction &CGF, Address VAListAddr,
QualType Ty) const {
return Address::invalid();
}
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() <= 16);
}
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();
}
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;
}
bool ABIInfo::shouldSignExtUnsignedType(QualType Ty) 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 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 > SlotSize) {
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);
llvm::Value *NextPtr =
CGF.Builder.CreateConstInBoundsByteGEP(Addr.getPointer(), FullDirectSize,
"argp.next");
CGF.Builder.CreateStore(NextPtr, 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() &&
!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,
std::pair<CharUnits, CharUnits> 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.first;
DirectAlign = ValueInfo.second;
}
// 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.second);
}
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() { 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;
}
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::Default;
}
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);
return CGF.Builder.CreatePointerBitCastOrAddrSpaceCast(Src, DestTy);
}
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(SyncScope S, llvm::LLVMContext &C) const {
return C.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.
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 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()) {
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;
}
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.second;
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(new 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();
return (Ty->isPromotableIntegerType() ?
ABIArgInfo::getExtend() : 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();
return (RetTy->isPromotableIntegerType() ?
ABIArgInfo::getExtend() : ABIArgInfo::getDirect());
}
//===----------------------------------------------------------------------===//
// WebAssembly ABI Implementation
//
// This is a very simple ABI that relies a lot on DefaultABIInfo.
//===----------------------------------------------------------------------===//
class WebAssemblyABIInfo final : public DefaultABIInfo {
public:
explicit WebAssemblyABIInfo(CodeGen::CodeGenTypes &CGT)
: DefaultABIInfo(CGT) {}
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;
};
class WebAssemblyTargetCodeGenInfo final : public TargetCodeGenInfo {
public:
explicit WebAssemblyTargetCodeGenInfo(CodeGen::CodeGenTypes &CGT)
: TargetCodeGenInfo(new WebAssemblyABIInfo(CGT)) {}
};
/// \brief 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)));
}
// Otherwise just do the default thing.
return DefaultABIInfo::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)));
}
}
// Otherwise just do the default thing.
return DefaultABIInfo::classifyReturnType(RetTy);
}
Address WebAssemblyABIInfo::EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
QualType Ty) const {
return emitVoidPtrVAArg(CGF, VAListAddr, Ty, /*Indirect=*/ false,
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(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);
}
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());
}
/// \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 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();
}
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 getNaturalAlignIndirect(RetTy);
// 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) {
bool IsMMXCons = llvm::StringSwitch<bool>(Constraint)
.Cases("y", "&y", "^Ym", true)
.Default(false);
if (IsMMXCons && 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) {
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 == 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;
}
/// 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
//===----------------------------------------------------------------------===//
/// \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;
};
enum {
// Vectorcall only allows the first 6 parameters to be passed in registers.
VectorcallMaxParamNumAsReg = 6
};
/// 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 == 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(QualType Ty, 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;
/// \brief 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;
/// \brief 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 computeVectorCallArgs(CGFunctionInfo &FI, CCState &State,
bool &UsedInAlloca) 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(CharUnits totalSize,
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(new 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,
ForDefinition_t IsForDefinition) 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()) {
// 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
/// 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 > 64) || (!IsMCUABI && !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;
}
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;
}
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::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(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())
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())
|| 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();
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.
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::Double)
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 > 2)
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))
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_RegCall) {
if (getContext().getTypeSize(Ty) > 32)
return false;
return (Ty->isIntegralOrEnumerationType() || Ty->isPointerType() ||
Ty->isReferenceType());
}
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);
}
}
// Regcall uses 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_RegCall &&
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)) {
// Structures with flexible arrays are always indirect.
// FIXME: This should not be byval!
if (RT && RT->getDecl()->hasFlexibleArrayMember())
return getIndirectResult(Ty, true, State);
// Ignore empty structs/unions on non-Windows.
if (!IsWin32StructABI && isEmptyRecord(getContext(), Ty, true))
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 = (getContext().getTypeSize(Ty) + 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 ? 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.
// Don't do this for the MCU if there are still free integer registers
// (see X86_64 ABI for full explanation).
if (getContext().getTypeSize(Ty) <= 4 * 32 &&
(!IsMCUABI || State.FreeRegs == 0) && canExpandIndirectArgument(Ty))
return ABIArgInfo::getExpandWithPadding(
State.CC == llvm::CallingConv::X86_FastCall ||
State.CC == llvm::CallingConv::X86_VectorCall ||
State.CC == llvm::CallingConv::X86_RegCall,
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 InReg = shouldPrimitiveUseInReg(Ty, State);
if (Ty->isPromotableIntegerType()) {
if (InReg)
return ABIArgInfo::getExtendInReg();
return ABIArgInfo::getExtend();
}
if (InReg)
return ABIArgInfo::getDirectInReg();
return ABIArgInfo::getDirect();
}
void X86_32ABIInfo::computeVectorCallArgs(CGFunctionInfo &FI, CCState &State,
bool &UsedInAlloca) 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 YMM0-5 or XMM0-5 registers.
// This differs from the x64 implementation, where the first 6 by INDEX get
// registers.
// After that, integers AND HVAs are assigned Left to Right in the same pass.
// Integers are passed as ECX/EDX if one is available (in order). HVAs will
// first take up the remaining YMM/XMM registers. If insufficient registers
// remain but an integer register (ECX/EDX) is available, it will be passed
// in that, else, on the stack.
for (auto &I : FI.arguments()) {
// First pass do all the vector types.
const Type *Base = nullptr;
uint64_t NumElts = 0;
const QualType& Ty = I.type;
if ((Ty->isVectorType() || Ty->isBuiltinType()) &&
isHomogeneousAggregate(Ty, Base, NumElts)) {
if (State.FreeSSERegs >= NumElts) {
State.FreeSSERegs -= NumElts;
I.info = ABIArgInfo::getDirect();
} else {
I.info = classifyArgumentType(Ty, State);
}
UsedInAlloca |= (I.info.getKind() == ABIArgInfo::InAlloca);
}
}
for (auto &I : FI.arguments()) {
// Second pass, do the rest!
const Type *Base = nullptr;
uint64_t NumElts = 0;
const QualType& Ty = I.type;
bool IsHva = isHomogeneousAggregate(Ty, Base, NumElts);
if (IsHva && !Ty->isVectorType() && !Ty->isBuiltinType()) {
// Assign true HVAs (non vector/native FP types).
if (State.FreeSSERegs >= NumElts) {
State.FreeSSERegs -= NumElts;
I.info = getDirectX86Hva();
} else {
I.info = getIndirectResult(Ty, /*ByVal=*/false, State);
}
} else if (!IsHva) {
// Assign all Non-HVAs, so this will exclude Vector/FP args.
I.info = classifyArgumentType(Ty, State);
UsedInAlloca |= (I.info.getKind() == ABIArgInfo::InAlloca);
}
}
}
void X86_32ABIInfo::computeInfo(CGFunctionInfo &FI) const {
CCState State(FI.getCallingConvention());
if (IsMCUABI)
State.FreeRegs = 3;
else 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 if (State.CC == llvm::CallingConv::X86_RegCall) {
State.FreeRegs = 5;
State.FreeSSERegs = 8;
} 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.
if (!IsMCUABI)
FI.getReturnInfo().setInReg(true);
}
}
// The chain argument effectively gives us another free register.
if (FI.isChainCall())
++State.FreeRegs;
bool UsedInAlloca = false;
if (State.CC == llvm::CallingConv::X86_VectorCall) {
computeVectorCallArgs(FI, State, UsedInAlloca);
} else {
// If not vectorcall, revert to normal behavior.
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,
CharUnits &StackOffset, ABIArgInfo &Info,
QualType Type) const {
// Arguments are always 4-byte-aligned.
CharUnits FieldAlign = CharUnits::fromQuantity(4);
assert(StackOffset.isMultipleOf(FieldAlign) && "unaligned inalloca struct");
Info = ABIArgInfo::getInAlloca(FrameFields.size());
FrameFields.push_back(CGT.ConvertTypeForMem(Type));
StackOffset += getContext().getTypeSizeInChars(Type);
// Insert padding bytes to respect alignment.
CharUnits FieldEnd = StackOffset;
StackOffset = FieldEnd.alignTo(FieldAlign);
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::Indirect:
assert(Info.getIndirectByVal());
return true;
case ABIArgInfo::Ignore:
return false;
case ABIArgInfo::Direct:
case ABIArgInfo::Extend:
if (Info.getInReg())
return false;
return true;
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() && !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),
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.second = CharUnits::fromQuantity(
getTypeStackAlignInBytes(Ty, TypeInfo.second.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())
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;
}
}
void X86_32TargetCodeGenInfo::setTargetAttributes(
const Decl *D, llvm::GlobalValue *GV, CodeGen::CodeGenModule &CGM,
ForDefinition_t IsForDefinition) const {
if (!IsForDefinition)
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");
}
if (FD->hasAttr<AnyX86InterruptAttr>()) {
llvm::Function *Fn = cast<llvm::Function>(GV);
Fn->setCallingConv(llvm::CallingConv::X86_INTR);
}
}
}
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 (getCodeGenOpts().getClangABICompat() <=
CodeGenOptions::ClangABI::Ver3_8)
return false;
const llvm::Triple &Triple = getTarget().getTriple();
if (Triple.isOSDarwin() || Triple.getOS() == llvm::Triple::PS4)
return false;
if (Triple.isOSFreeBSD() && Triple.getOSMajorVersion() >= 10)
return false;
return true;
}
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->getBitWidth() > 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(CharUnits totalSize,
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)
: SwiftABIInfo(CGT),
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(CharUnits totalSize,
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 reclassifyHvaArgType(QualType Ty, unsigned &FreeSSERegs,
const ABIArgInfo &current) const;
void computeVectorCallArgs(CGFunctionInfo &FI, unsigned FreeSSERegs,
bool IsVectorCall, bool IsRegCall) const;
bool IsMingw64;
};
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 = (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,
ForDefinition_t IsForDefinition) const override {
if (!IsForDefinition)
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");
}
if (FD->hasAttr<AnyX86InterruptAttr>()) {
llvm::Function *Fn = cast<llvm::Function>(GV);
Fn->setCallingConv(llvm::CallingConv::X86_INTR);
}
}
}
};
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 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,
ForDefinition_t IsForDefinition) 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 (D && 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,
ForDefinition_t IsForDefinition) const {
X86_32TargetCodeGenInfo::setTargetAttributes(D, GV, CGM, IsForDefinition);
if (!IsForDefinition)
return;
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,
ForDefinition_t IsForDefinition) 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,
ForDefinition_t IsForDefinition) const {
TargetCodeGenInfo::setTargetAttributes(D, GV, CGM, IsForDefinition);
if (!IsForDefinition)
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");
}
if (FD->hasAttr<AnyX86InterruptAttr>()) {
llvm::Function *Fn = cast<llvm::Function>(GV);
Fn->setCallingConv(llvm::CallingConv::X86_INTR);
}
}
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) {
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 == 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) {
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 != 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) {
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_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 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)))
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>()) {