SDValue ARM64SelectionDAGInfo::EmitTargetCodeForMemset(
SelectionDAG &DAG, SDLoc dl, SDValue Chain, SDValue Dst, SDValue Src,
SDValue Size, unsigned Align, bool isVolatile,
MachinePointerInfo DstPtrInfo) const {
// Check to see if there is a specialized entry-point for memory zeroing.
ConstantSDNode *V = dyn_cast<ConstantSDNode>(Src);
ConstantSDNode *SizeValue = dyn_cast<ConstantSDNode>(Size);
const char *bzeroEntry =
(V && V->isNullValue()) ? Subtarget->getBZeroEntry() : 0;
// For small size (< 256), it is not beneficial to use bzero
// instead of memset.
if (bzeroEntry && (!SizeValue || SizeValue->getZExtValue() > 256)) {
const ARM64TargetLowering &TLI = *static_cast<const ARM64TargetLowering *>(
DAG.getTarget().getTargetLowering());
EVT IntPtr = TLI.getPointerTy();
Type *IntPtrTy = getDataLayout()->getIntPtrType(*DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Node = Dst;
Entry.Ty = IntPtrTy;
Args.push_back(Entry);
Entry.Node = Size;
Args.push_back(Entry);
TargetLowering::CallLoweringInfo CLI(
Chain, Type::getVoidTy(*DAG.getContext()), false, false, false, false,
0, CallingConv::C, /*isTailCall=*/false,
/*doesNotRet=*/false, /*isReturnValueUsed=*/false,
DAG.getExternalSymbol(bzeroEntry, IntPtr), Args, DAG, dl);
std::pair<SDValue, SDValue> CallResult = TLI.LowerCallTo(CLI);
return CallResult.second;
}
return SDValue();
}
SDValue XCoreSelectionDAGInfo::
EmitTargetCodeForMemcpy(SelectionDAG &DAG, SDLoc dl, SDValue Chain,
SDValue Dst, SDValue Src, SDValue Size, unsigned Align,
bool isVolatile, bool AlwaysInline,
MachinePointerInfo DstPtrInfo,
MachinePointerInfo SrcPtrInfo) const
{
unsigned SizeBitWidth = Size.getValueType().getSizeInBits();
// Call __memcpy_4 if the src, dst and size are all 4 byte aligned.
if (!AlwaysInline && (Align & 3) == 0 &&
DAG.MaskedValueIsZero(Size, APInt(SizeBitWidth, 3))) {
const TargetLowering &TLI = *DAG.getTarget().getTargetLowering();
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Ty = TLI.getDataLayout()->getIntPtrType(*DAG.getContext());
Entry.Node = Dst; Args.push_back(Entry);
Entry.Node = Src; Args.push_back(Entry);
Entry.Node = Size; Args.push_back(Entry);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(Chain)
.setCallee(TLI.getLibcallCallingConv(RTLIB::MEMCPY),
Type::getVoidTy(*DAG.getContext()),
DAG.getExternalSymbol("__memcpy_4", TLI.getPointerTy()),
&Args, 0)
.setDiscardResult();
std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
return CallResult.second;
}
// Otherwise have the target-independent code call memcpy.
return SDValue();
}
// Adjust parameters for memset, EABI uses format (ptr, size, value),
// GNU library uses (ptr, value, size)
// See RTABI section 4.3.4
SDValue ARMSelectionDAGInfo::
EmitTargetCodeForMemset(SelectionDAG &DAG, SDLoc dl,
SDValue Chain, SDValue Dst,
SDValue Src, SDValue Size,
unsigned Align, bool isVolatile,
MachinePointerInfo DstPtrInfo) const {
// Use default for non AAPCS (or Darwin) subtargets
if (!Subtarget->isAAPCS_ABI() || Subtarget->isTargetDarwin())
return SDValue();
const ARMTargetLowering &TLI =
*static_cast<const ARMTargetLowering*>(DAG.getTarget().getTargetLowering());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
// First argument: data pointer
Type *IntPtrTy = TLI.getDataLayout()->getIntPtrType(*DAG.getContext());
Entry.Node = Dst;
Entry.Ty = IntPtrTy;
Args.push_back(Entry);
// Second argument: buffer size
Entry.Node = Size;
Entry.Ty = IntPtrTy;
Entry.isSExt = false;
Args.push_back(Entry);
// Extend or truncate the argument to be an i32 value for the call.
if (Src.getValueType().bitsGT(MVT::i32))
Src = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, Src);
else
Src = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Src);
// Third argument: value to fill
Entry.Node = Src;
Entry.Ty = Type::getInt32Ty(*DAG.getContext());
Entry.isSExt = true;
Args.push_back(Entry);
// Emit __eabi_memset call
TargetLowering::CallLoweringInfo CLI(Chain,
Type::getVoidTy(*DAG.getContext()), // return type
false, // return sign ext
false, // return zero ext
false, // is var arg
false, // is in regs
0, // number of fixed arguments
TLI.getLibcallCallingConv(RTLIB::MEMSET), // call conv
false, // is tail call
false, // does not return
false, // is return val used
DAG.getExternalSymbol(TLI.getLibcallName(RTLIB::MEMSET),
TLI.getPointerTy()), // callee
Args, DAG, dl);
std::pair<SDValue,SDValue> CallResult =
TLI.LowerCallTo(CLI);
return CallResult.second;
}
SDValue
AVM2TargetLowering::LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG,
const AVM2Subtarget *Subtarget) const
{
unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
DebugLoc dl = Op.getDebugLoc();
switch (IntNo) {
default:
return SDValue(); // Don't custom lower most intrinsics.
case Intrinsic::eh_sjlj_lsda: {
MachineFunction &MF = DAG.getMachineFunction();
EVT VT(MVT::i32);
SmallString<128> Str;
raw_svector_ostream O(Str);
O << getTargetMachine().getMCAsmInfo()->getPrivateGlobalPrefix() << "_LSDA_" << MF.getFunctionNumber();
O.flush();
// TODO strdup is probably a leak
return DAG.getExternalSymbol(strdup(Str.str().str().c_str()), VT);
}
}
}
// Emit, if possible, a specialized version of the given Libcall. Typically this
// means selecting the appropriately aligned version, but we also convert memset
// of 0 into memclr.
SDValue ARMSelectionDAGInfo::EmitSpecializedLibcall(
SelectionDAG &DAG, const SDLoc &dl, SDValue Chain, SDValue Dst, SDValue Src,
SDValue Size, unsigned Align, RTLIB::Libcall LC) const {
const ARMSubtarget &Subtarget =
DAG.getMachineFunction().getSubtarget<ARMSubtarget>();
const ARMTargetLowering *TLI = Subtarget.getTargetLowering();
// Only use a specialized AEABI function if the default version of this
// Libcall is an AEABI function.
if (std::strncmp(TLI->getLibcallName(LC), "__aeabi", 7) != 0)
return SDValue();
// Translate RTLIB::Libcall to AEABILibcall. We only do this in order to be
// able to translate memset to memclr and use the value to index the function
// name array.
enum {
AEABI_MEMCPY = 0,
AEABI_MEMMOVE,
AEABI_MEMSET,
AEABI_MEMCLR
} AEABILibcall;
switch (LC) {
case RTLIB::MEMCPY:
AEABILibcall = AEABI_MEMCPY;
break;
case RTLIB::MEMMOVE:
AEABILibcall = AEABI_MEMMOVE;
break;
case RTLIB::MEMSET:
AEABILibcall = AEABI_MEMSET;
if (ConstantSDNode *ConstantSrc = dyn_cast<ConstantSDNode>(Src))
if (ConstantSrc->getZExtValue() == 0)
AEABILibcall = AEABI_MEMCLR;
break;
default:
return SDValue();
}
// Choose the most-aligned libcall variant that we can
enum {
ALIGN1 = 0,
ALIGN4,
ALIGN8
} AlignVariant;
if ((Align & 7) == 0)
AlignVariant = ALIGN8;
else if ((Align & 3) == 0)
AlignVariant = ALIGN4;
else
AlignVariant = ALIGN1;
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Ty = DAG.getDataLayout().getIntPtrType(*DAG.getContext());
Entry.Node = Dst;
Args.push_back(Entry);
if (AEABILibcall == AEABI_MEMCLR) {
Entry.Node = Size;
Args.push_back(Entry);
} else if (AEABILibcall == AEABI_MEMSET) {
// Adjust parameters for memset, EABI uses format (ptr, size, value),
// GNU library uses (ptr, value, size)
// See RTABI section 4.3.4
Entry.Node = Size;
Args.push_back(Entry);
// Extend or truncate the argument to be an i32 value for the call.
if (Src.getValueType().bitsGT(MVT::i32))
Src = DAG.getNode(ISD::TRUNCATE, dl, MVT::i32, Src);
else if (Src.getValueType().bitsLT(MVT::i32))
Src = DAG.getNode(ISD::ZERO_EXTEND, dl, MVT::i32, Src);
Entry.Node = Src;
Entry.Ty = Type::getInt32Ty(*DAG.getContext());
Entry.IsSExt = false;
Args.push_back(Entry);
} else {
Entry.Node = Src;
Args.push_back(Entry);
Entry.Node = Size;
Args.push_back(Entry);
}
char const *FunctionNames[4][3] = {
{ "__aeabi_memcpy", "__aeabi_memcpy4", "__aeabi_memcpy8" },
{ "__aeabi_memmove", "__aeabi_memmove4", "__aeabi_memmove8" },
{ "__aeabi_memset", "__aeabi_memset4", "__aeabi_memset8" },
{ "__aeabi_memclr", "__aeabi_memclr4", "__aeabi_memclr8" }
};
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl)
.setChain(Chain)
.setLibCallee(
TLI->getLibcallCallingConv(LC), Type::getVoidTy(*DAG.getContext()),
DAG.getExternalSymbol(FunctionNames[AEABILibcall][AlignVariant],
TLI->getPointerTy(DAG.getDataLayout())),
std::move(Args))
.setDiscardResult();
std::pair<SDValue,SDValue> CallResult = TLI->LowerCallTo(CLI);
return CallResult.second;
}
SDValue
X86SelectionDAGInfo::EmitTargetCodeForMemset(SelectionDAG &DAG, SDLoc dl,
SDValue Chain,
SDValue Dst, SDValue Src,
SDValue Size, unsigned Align,
bool isVolatile,
MachinePointerInfo DstPtrInfo) const {
ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
// If to a segment-relative address space, use the default lowering.
if (DstPtrInfo.getAddrSpace() >= 256)
return SDValue();
// If not DWORD aligned or size is more than the threshold, call the library.
// The libc version is likely to be faster for these cases. It can use the
// address value and run time information about the CPU.
if ((Align & 3) != 0 ||
!ConstantSize ||
ConstantSize->getZExtValue() >
Subtarget->getMaxInlineSizeThreshold()) {
// Check to see if there is a specialized entry-point for memory zeroing.
ConstantSDNode *V = dyn_cast<ConstantSDNode>(Src);
if (const char *bzeroEntry = V &&
V->isNullValue() ? Subtarget->getBZeroEntry() : nullptr) {
EVT IntPtr = TLI.getPointerTy();
Type *IntPtrTy = getDataLayout()->getIntPtrType(*DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Node = Dst;
Entry.Ty = IntPtrTy;
Args.push_back(Entry);
Entry.Node = Size;
Args.push_back(Entry);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl).setChain(Chain)
.setCallee(CallingConv::C, Type::getVoidTy(*DAG.getContext()),
DAG.getExternalSymbol(bzeroEntry, IntPtr), &Args, 0)
.setDiscardResult();
std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
return CallResult.second;
}
// Otherwise have the target-independent code call memset.
return SDValue();
}
uint64_t SizeVal = ConstantSize->getZExtValue();
SDValue InFlag;
EVT AVT;
SDValue Count;
ConstantSDNode *ValC = dyn_cast<ConstantSDNode>(Src);
unsigned BytesLeft = 0;
bool TwoRepStos = false;
if (ValC) {
unsigned ValReg;
uint64_t Val = ValC->getZExtValue() & 255;
// If the value is a constant, then we can potentially use larger sets.
switch (Align & 3) {
case 2: // WORD aligned
AVT = MVT::i16;
ValReg = X86::AX;
Val = (Val << 8) | Val;
break;
case 0: // DWORD aligned
AVT = MVT::i32;
ValReg = X86::EAX;
Val = (Val << 8) | Val;
Val = (Val << 16) | Val;
if (Subtarget->is64Bit() && ((Align & 0x7) == 0)) { // QWORD aligned
AVT = MVT::i64;
ValReg = X86::RAX;
Val = (Val << 32) | Val;
}
break;
default: // Byte aligned
AVT = MVT::i8;
ValReg = X86::AL;
Count = DAG.getIntPtrConstant(SizeVal);
break;
}
if (AVT.bitsGT(MVT::i8)) {
unsigned UBytes = AVT.getSizeInBits() / 8;
Count = DAG.getIntPtrConstant(SizeVal / UBytes);
BytesLeft = SizeVal % UBytes;
}
Chain = DAG.getCopyToReg(Chain, dl, ValReg, DAG.getConstant(Val, AVT),
InFlag);
InFlag = Chain.getValue(1);
} else {
AVT = MVT::i8;
Count = DAG.getIntPtrConstant(SizeVal);
Chain = DAG.getCopyToReg(Chain, dl, X86::AL, Src, InFlag);
InFlag = Chain.getValue(1);
}
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RCX :
X86::ECX,
Count, InFlag);
InFlag = Chain.getValue(1);
Chain = DAG.getCopyToReg(Chain, dl, Subtarget->is64Bit() ? X86::RDI :
X86::EDI,
Dst, InFlag);
InFlag = Chain.getValue(1);
SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, DAG.getValueType(AVT), InFlag };
Chain = DAG.getNode(X86ISD::REP_STOS, dl, Tys, Ops);
if (TwoRepStos) {
InFlag = Chain.getValue(1);
Count = Size;
EVT CVT = Count.getValueType();
SDValue Left = DAG.getNode(ISD::AND, dl, CVT, Count,
DAG.getConstant((AVT == MVT::i64) ? 7 : 3, CVT));
Chain = DAG.getCopyToReg(Chain, dl, (CVT == MVT::i64) ? X86::RCX :
X86::ECX,
Left, InFlag);
InFlag = Chain.getValue(1);
Tys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, DAG.getValueType(MVT::i8), InFlag };
Chain = DAG.getNode(X86ISD::REP_STOS, dl, Tys, Ops);
} else if (BytesLeft) {
// Handle the last 1 - 7 bytes.
unsigned Offset = SizeVal - BytesLeft;
EVT AddrVT = Dst.getValueType();
EVT SizeVT = Size.getValueType();
Chain = DAG.getMemset(Chain, dl,
DAG.getNode(ISD::ADD, dl, AddrVT, Dst,
DAG.getConstant(Offset, AddrVT)),
Src,
DAG.getConstant(BytesLeft, SizeVT),
Align, isVolatile, DstPtrInfo.getWithOffset(Offset));
}
// TODO: Use a Tokenfactor, as in memcpy, instead of a single chain.
return Chain;
}
/// LowerOperation - Provide custom lowering hooks for some operations.
///
SDValue AlphaTargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) {
DebugLoc dl = Op.getDebugLoc();
switch (Op.getOpcode()) {
default: llvm_unreachable("Wasn't expecting to be able to lower this!");
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
case ISD::INTRINSIC_WO_CHAIN: {
unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(0))->getZExtValue();
switch (IntNo) {
default: break; // Don't custom lower most intrinsics.
case Intrinsic::alpha_umulh:
return DAG.getNode(ISD::MULHU, dl, MVT::i64,
Op.getOperand(1), Op.getOperand(2));
}
}
case ISD::SRL_PARTS: {
SDValue ShOpLo = Op.getOperand(0);
SDValue ShOpHi = Op.getOperand(1);
SDValue ShAmt = Op.getOperand(2);
SDValue bm = DAG.getNode(ISD::SUB, dl, MVT::i64,
DAG.getConstant(64, MVT::i64), ShAmt);
SDValue BMCC = DAG.getSetCC(dl, MVT::i64, bm,
DAG.getConstant(0, MVT::i64), ISD::SETLE);
// if 64 - shAmt <= 0
SDValue Hi_Neg = DAG.getConstant(0, MVT::i64);
SDValue ShAmt_Neg = DAG.getNode(ISD::SUB, dl, MVT::i64,
DAG.getConstant(0, MVT::i64), bm);
SDValue Lo_Neg = DAG.getNode(ISD::SRL, dl, MVT::i64, ShOpHi, ShAmt_Neg);
// else
SDValue carries = DAG.getNode(ISD::SHL, dl, MVT::i64, ShOpHi, bm);
SDValue Hi_Pos = DAG.getNode(ISD::SRL, dl, MVT::i64, ShOpHi, ShAmt);
SDValue Lo_Pos = DAG.getNode(ISD::SRL, dl, MVT::i64, ShOpLo, ShAmt);
Lo_Pos = DAG.getNode(ISD::OR, dl, MVT::i64, Lo_Pos, carries);
// Merge
SDValue Hi = DAG.getNode(ISD::SELECT, dl, MVT::i64, BMCC, Hi_Neg, Hi_Pos);
SDValue Lo = DAG.getNode(ISD::SELECT, dl, MVT::i64, BMCC, Lo_Neg, Lo_Pos);
SDValue Ops[2] = { Lo, Hi };
return DAG.getMergeValues(Ops, 2, dl);
}
// case ISD::SRA_PARTS:
// case ISD::SHL_PARTS:
case ISD::SINT_TO_FP: {
assert(Op.getOperand(0).getValueType() == MVT::i64 &&
"Unhandled SINT_TO_FP type in custom expander!");
SDValue LD;
bool isDouble = Op.getValueType() == MVT::f64;
LD = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f64, Op.getOperand(0));
SDValue FP = DAG.getNode(isDouble?AlphaISD::CVTQT_:AlphaISD::CVTQS_, dl,
isDouble?MVT::f64:MVT::f32, LD);
return FP;
}
case ISD::FP_TO_SINT: {
bool isDouble = Op.getOperand(0).getValueType() == MVT::f64;
SDValue src = Op.getOperand(0);
if (!isDouble) //Promote
src = DAG.getNode(ISD::FP_EXTEND, dl, MVT::f64, src);
src = DAG.getNode(AlphaISD::CVTTQ_, dl, MVT::f64, src);
return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, src);
}
case ISD::ConstantPool: {
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
Constant *C = CP->getConstVal();
SDValue CPI = DAG.getTargetConstantPool(C, MVT::i64, CP->getAlignment());
// FIXME there isn't really any debug info here
SDValue Hi = DAG.getNode(AlphaISD::GPRelHi, dl, MVT::i64, CPI,
DAG.getGLOBAL_OFFSET_TABLE(MVT::i64));
SDValue Lo = DAG.getNode(AlphaISD::GPRelLo, dl, MVT::i64, CPI, Hi);
return Lo;
}
case ISD::GlobalTLSAddress:
llvm_unreachable("TLS not implemented for Alpha.");
case ISD::GlobalAddress: {
GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
GlobalValue *GV = GSDN->getGlobal();
SDValue GA = DAG.getTargetGlobalAddress(GV, MVT::i64, GSDN->getOffset());
// FIXME there isn't really any debug info here
// if (!GV->hasWeakLinkage() && !GV->isDeclaration() && !GV->hasLinkOnceLinkage()) {
if (GV->hasLocalLinkage()) {
SDValue Hi = DAG.getNode(AlphaISD::GPRelHi, dl, MVT::i64, GA,
DAG.getGLOBAL_OFFSET_TABLE(MVT::i64));
SDValue Lo = DAG.getNode(AlphaISD::GPRelLo, dl, MVT::i64, GA, Hi);
return Lo;
} else
return DAG.getNode(AlphaISD::RelLit, dl, MVT::i64, GA,
DAG.getGLOBAL_OFFSET_TABLE(MVT::i64));
}
case ISD::ExternalSymbol: {
return DAG.getNode(AlphaISD::RelLit, dl, MVT::i64,
DAG.getTargetExternalSymbol(cast<ExternalSymbolSDNode>(Op)
->getSymbol(), MVT::i64),
DAG.getGLOBAL_OFFSET_TABLE(MVT::i64));
}
case ISD::UREM:
case ISD::SREM:
//Expand only on constant case
if (Op.getOperand(1).getOpcode() == ISD::Constant) {
EVT VT = Op.getNode()->getValueType(0);
SDValue Tmp1 = Op.getNode()->getOpcode() == ISD::UREM ?
BuildUDIV(Op.getNode(), DAG, NULL) :
BuildSDIV(Op.getNode(), DAG, NULL);
Tmp1 = DAG.getNode(ISD::MUL, dl, VT, Tmp1, Op.getOperand(1));
Tmp1 = DAG.getNode(ISD::SUB, dl, VT, Op.getOperand(0), Tmp1);
return Tmp1;
}
//fall through
case ISD::SDIV:
case ISD::UDIV:
if (Op.getValueType().isInteger()) {
if (Op.getOperand(1).getOpcode() == ISD::Constant)
return Op.getOpcode() == ISD::SDIV ? BuildSDIV(Op.getNode(), DAG, NULL)
: BuildUDIV(Op.getNode(), DAG, NULL);
const char* opstr = 0;
switch (Op.getOpcode()) {
case ISD::UREM: opstr = "__remqu"; break;
case ISD::SREM: opstr = "__remq"; break;
case ISD::UDIV: opstr = "__divqu"; break;
case ISD::SDIV: opstr = "__divq"; break;
}
SDValue Tmp1 = Op.getOperand(0),
Tmp2 = Op.getOperand(1),
Addr = DAG.getExternalSymbol(opstr, MVT::i64);
return DAG.getNode(AlphaISD::DivCall, dl, MVT::i64, Addr, Tmp1, Tmp2);
}
break;
case ISD::VAARG: {
SDValue Chain, DataPtr;
LowerVAARG(Op.getNode(), Chain, DataPtr, DAG);
SDValue Result;
if (Op.getValueType() == MVT::i32)
Result = DAG.getExtLoad(ISD::SEXTLOAD, dl, MVT::i64, Chain, DataPtr,
NULL, 0, MVT::i32);
else
Result = DAG.getLoad(Op.getValueType(), dl, Chain, DataPtr, NULL, 0);
return Result;
}
case ISD::VACOPY: {
SDValue Chain = Op.getOperand(0);
SDValue DestP = Op.getOperand(1);
SDValue SrcP = Op.getOperand(2);
const Value *DestS = cast<SrcValueSDNode>(Op.getOperand(3))->getValue();
const Value *SrcS = cast<SrcValueSDNode>(Op.getOperand(4))->getValue();
SDValue Val = DAG.getLoad(getPointerTy(), dl, Chain, SrcP, SrcS, 0);
SDValue Result = DAG.getStore(Val.getValue(1), dl, Val, DestP, DestS, 0);
SDValue NP = DAG.getNode(ISD::ADD, dl, MVT::i64, SrcP,
DAG.getConstant(8, MVT::i64));
Val = DAG.getExtLoad(ISD::SEXTLOAD, dl, MVT::i64, Result,
NP, NULL,0, MVT::i32);
SDValue NPD = DAG.getNode(ISD::ADD, dl, MVT::i64, DestP,
DAG.getConstant(8, MVT::i64));
return DAG.getTruncStore(Val.getValue(1), dl, Val, NPD, NULL, 0, MVT::i32);
}
case ISD::VASTART: {
SDValue Chain = Op.getOperand(0);
SDValue VAListP = Op.getOperand(1);
const Value *VAListS = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
// vastart stores the address of the VarArgsBase and VarArgsOffset
SDValue FR = DAG.getFrameIndex(VarArgsBase, MVT::i64);
SDValue S1 = DAG.getStore(Chain, dl, FR, VAListP, VAListS, 0);
SDValue SA2 = DAG.getNode(ISD::ADD, dl, MVT::i64, VAListP,
DAG.getConstant(8, MVT::i64));
return DAG.getTruncStore(S1, dl, DAG.getConstant(VarArgsOffset, MVT::i64),
SA2, NULL, 0, MVT::i32);
}
case ISD::RETURNADDR:
return DAG.getNode(AlphaISD::GlobalRetAddr, DebugLoc::getUnknownLoc(),
MVT::i64);
//FIXME: implement
case ISD::FRAMEADDR: break;
}
return SDValue();
}
/// LowerOperation - Provide custom lowering hooks for some operations.
///
SDOperand AlphaTargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
switch (Op.getOpcode()) {
default: assert(0 && "Wasn't expecting to be able to lower this!");
case ISD::FORMAL_ARGUMENTS: return LowerFORMAL_ARGUMENTS(Op, DAG,
VarArgsBase,
VarArgsOffset);
case ISD::RET: return LowerRET(Op,DAG);
case ISD::JumpTable: return LowerJumpTable(Op, DAG);
case ISD::SINT_TO_FP: {
assert(MVT::i64 == Op.getOperand(0).getValueType() &&
"Unhandled SINT_TO_FP type in custom expander!");
SDOperand LD;
bool isDouble = MVT::f64 == Op.getValueType();
LD = DAG.getNode(ISD::BIT_CONVERT, MVT::f64, Op.getOperand(0));
SDOperand FP = DAG.getNode(isDouble?AlphaISD::CVTQT_:AlphaISD::CVTQS_,
isDouble?MVT::f64:MVT::f32, LD);
return FP;
}
case ISD::FP_TO_SINT: {
bool isDouble = MVT::f64 == Op.getOperand(0).getValueType();
SDOperand src = Op.getOperand(0);
if (!isDouble) //Promote
src = DAG.getNode(ISD::FP_EXTEND, MVT::f64, src);
src = DAG.getNode(AlphaISD::CVTTQ_, MVT::f64, src);
return DAG.getNode(ISD::BIT_CONVERT, MVT::i64, src);
}
case ISD::ConstantPool: {
ConstantPoolSDNode *CP = cast<ConstantPoolSDNode>(Op);
Constant *C = CP->getConstVal();
SDOperand CPI = DAG.getTargetConstantPool(C, MVT::i64, CP->getAlignment());
SDOperand Hi = DAG.getNode(AlphaISD::GPRelHi, MVT::i64, CPI,
DAG.getNode(ISD::GLOBAL_OFFSET_TABLE, MVT::i64));
SDOperand Lo = DAG.getNode(AlphaISD::GPRelLo, MVT::i64, CPI, Hi);
return Lo;
}
case ISD::GlobalAddress: {
GlobalAddressSDNode *GSDN = cast<GlobalAddressSDNode>(Op);
GlobalValue *GV = GSDN->getGlobal();
SDOperand GA = DAG.getTargetGlobalAddress(GV, MVT::i64, GSDN->getOffset());
// if (!GV->hasWeakLinkage() && !GV->isDeclaration() && !GV->hasLinkOnceLinkage()) {
if (GV->hasInternalLinkage()) {
SDOperand Hi = DAG.getNode(AlphaISD::GPRelHi, MVT::i64, GA,
DAG.getNode(ISD::GLOBAL_OFFSET_TABLE, MVT::i64));
SDOperand Lo = DAG.getNode(AlphaISD::GPRelLo, MVT::i64, GA, Hi);
return Lo;
} else
return DAG.getNode(AlphaISD::RelLit, MVT::i64, GA,
DAG.getNode(ISD::GLOBAL_OFFSET_TABLE, MVT::i64));
}
case ISD::ExternalSymbol: {
return DAG.getNode(AlphaISD::RelLit, MVT::i64,
DAG.getTargetExternalSymbol(cast<ExternalSymbolSDNode>(Op)
->getSymbol(), MVT::i64),
DAG.getNode(ISD::GLOBAL_OFFSET_TABLE, MVT::i64));
}
case ISD::UREM:
case ISD::SREM:
//Expand only on constant case
if (Op.getOperand(1).getOpcode() == ISD::Constant) {
MVT::ValueType VT = Op.Val->getValueType(0);
SDOperand Tmp1 = Op.Val->getOpcode() == ISD::UREM ?
BuildUDIV(Op.Val, DAG, NULL) :
BuildSDIV(Op.Val, DAG, NULL);
Tmp1 = DAG.getNode(ISD::MUL, VT, Tmp1, Op.getOperand(1));
Tmp1 = DAG.getNode(ISD::SUB, VT, Op.getOperand(0), Tmp1);
return Tmp1;
}
//fall through
case ISD::SDIV:
case ISD::UDIV:
if (MVT::isInteger(Op.getValueType())) {
if (Op.getOperand(1).getOpcode() == ISD::Constant)
return Op.getOpcode() == ISD::SDIV ? BuildSDIV(Op.Val, DAG, NULL)
: BuildUDIV(Op.Val, DAG, NULL);
const char* opstr = 0;
switch (Op.getOpcode()) {
case ISD::UREM: opstr = "__remqu"; break;
case ISD::SREM: opstr = "__remq"; break;
case ISD::UDIV: opstr = "__divqu"; break;
case ISD::SDIV: opstr = "__divq"; break;
}
SDOperand Tmp1 = Op.getOperand(0),
Tmp2 = Op.getOperand(1),
Addr = DAG.getExternalSymbol(opstr, MVT::i64);
return DAG.getNode(AlphaISD::DivCall, MVT::i64, Addr, Tmp1, Tmp2);
}
break;
case ISD::VAARG: {
SDOperand Chain = Op.getOperand(0);
SDOperand VAListP = Op.getOperand(1);
SrcValueSDNode *VAListS = cast<SrcValueSDNode>(Op.getOperand(2));
SDOperand Base = DAG.getLoad(MVT::i64, Chain, VAListP, VAListS->getValue(),
VAListS->getOffset());
SDOperand Tmp = DAG.getNode(ISD::ADD, MVT::i64, VAListP,
DAG.getConstant(8, MVT::i64));
SDOperand Offset = DAG.getExtLoad(ISD::SEXTLOAD, MVT::i64, Base.getValue(1),
Tmp, NULL, 0, MVT::i32);
SDOperand DataPtr = DAG.getNode(ISD::ADD, MVT::i64, Base, Offset);
if (MVT::isFloatingPoint(Op.getValueType()))
{
//if fp && Offset < 6*8, then subtract 6*8 from DataPtr
SDOperand FPDataPtr = DAG.getNode(ISD::SUB, MVT::i64, DataPtr,
DAG.getConstant(8*6, MVT::i64));
SDOperand CC = DAG.getSetCC(MVT::i64, Offset,
DAG.getConstant(8*6, MVT::i64), ISD::SETLT);
DataPtr = DAG.getNode(ISD::SELECT, MVT::i64, CC, FPDataPtr, DataPtr);
}
SDOperand NewOffset = DAG.getNode(ISD::ADD, MVT::i64, Offset,
DAG.getConstant(8, MVT::i64));
SDOperand Update = DAG.getTruncStore(Offset.getValue(1), NewOffset,
Tmp, NULL, 0, MVT::i32);
SDOperand Result;
if (Op.getValueType() == MVT::i32)
Result = DAG.getExtLoad(ISD::SEXTLOAD, MVT::i64, Update, DataPtr,
NULL, 0, MVT::i32);
else
Result = DAG.getLoad(Op.getValueType(), Update, DataPtr, NULL, 0);
return Result;
}
case ISD::VACOPY: {
SDOperand Chain = Op.getOperand(0);
SDOperand DestP = Op.getOperand(1);
SDOperand SrcP = Op.getOperand(2);
SrcValueSDNode *DestS = cast<SrcValueSDNode>(Op.getOperand(3));
SrcValueSDNode *SrcS = cast<SrcValueSDNode>(Op.getOperand(4));
SDOperand Val = DAG.getLoad(getPointerTy(), Chain, SrcP,
SrcS->getValue(), SrcS->getOffset());
SDOperand Result = DAG.getStore(Val.getValue(1), Val, DestP, DestS->getValue(),
DestS->getOffset());
SDOperand NP = DAG.getNode(ISD::ADD, MVT::i64, SrcP,
DAG.getConstant(8, MVT::i64));
Val = DAG.getExtLoad(ISD::SEXTLOAD, MVT::i64, Result, NP, NULL,0, MVT::i32);
SDOperand NPD = DAG.getNode(ISD::ADD, MVT::i64, DestP,
DAG.getConstant(8, MVT::i64));
return DAG.getTruncStore(Val.getValue(1), Val, NPD, NULL, 0, MVT::i32);
}
case ISD::VASTART: {
SDOperand Chain = Op.getOperand(0);
SDOperand VAListP = Op.getOperand(1);
SrcValueSDNode *VAListS = cast<SrcValueSDNode>(Op.getOperand(2));
// vastart stores the address of the VarArgsBase and VarArgsOffset
SDOperand FR = DAG.getFrameIndex(VarArgsBase, MVT::i64);
SDOperand S1 = DAG.getStore(Chain, FR, VAListP, VAListS->getValue(),
VAListS->getOffset());
SDOperand SA2 = DAG.getNode(ISD::ADD, MVT::i64, VAListP,
DAG.getConstant(8, MVT::i64));
return DAG.getTruncStore(S1, DAG.getConstant(VarArgsOffset, MVT::i64),
SA2, NULL, 0, MVT::i32);
}
case ISD::RETURNADDR:
return DAG.getNode(AlphaISD::GlobalRetAddr, MVT::i64);
//FIXME: implement
case ISD::FRAMEADDR: break;
}
return SDOperand();
}
SDValue Cpu0TargetLowering::
LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const
{
// If the relocation model is PIC, use the General Dynamic TLS Model or
// Local Dynamic TLS model, otherwise use the Initial Exec or
// Local Exec TLS Model.
GlobalAddressSDNode *GA = cast<GlobalAddressSDNode>(Op);
DebugLoc dl = GA->getDebugLoc();
const GlobalValue *GV = GA->getGlobal();
EVT PtrVT = getPointerTy();
if (getTargetMachine().getRelocationModel() == Reloc::PIC_) {
// General Dynamic TLS Model
bool LocalDynamic = GV->hasInternalLinkage();
unsigned Flag = LocalDynamic ? Cpu0II::MO_TLSLDM :Cpu0II::MO_TLSGD;
SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0, Flag);
SDValue Argument = DAG.getNode(Cpu0ISD::Wrapper, dl, PtrVT,
GetGlobalReg(DAG, PtrVT), TGA);
unsigned PtrSize = PtrVT.getSizeInBits();
IntegerType *PtrTy = Type::getIntNTy(*DAG.getContext(), PtrSize);
SDValue TlsGetAddr = DAG.getExternalSymbol("__tls_get_addr", PtrVT);
ArgListTy Args;
ArgListEntry Entry;
Entry.Node = Argument;
Entry.Ty = PtrTy;
Args.push_back(Entry);
std::pair<SDValue, SDValue> CallResult =
LowerCallTo(DAG.getEntryNode(), PtrTy,
false, false, false, false, 0, CallingConv::C,
/*isTailCall=*/false, /*doesNotRet=*/false,
/*isReturnValueUsed=*/true,
TlsGetAddr, Args, DAG, dl);
SDValue Ret = CallResult.first;
if (!LocalDynamic)
return Ret;
SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
Cpu0II::MO_DTPREL_HI);
SDValue Hi = DAG.getNode(Cpu0ISD::Hi, dl, PtrVT, TGAHi);
SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
Cpu0II::MO_DTPREL_LO);
SDValue Lo = DAG.getNode(Cpu0ISD::Lo, dl, PtrVT, TGALo);
SDValue Add = DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Ret);
return DAG.getNode(ISD::ADD, dl, PtrVT, Add, Lo);
}
SDValue Offset;
if (GV->isDeclaration()) {
// Initial Exec TLS Model
SDValue TGA = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
Cpu0II::MO_GOTTPREL);
TGA = DAG.getNode(Cpu0ISD::Wrapper, dl, PtrVT, GetGlobalReg(DAG, PtrVT),
TGA);
Offset = DAG.getLoad(PtrVT, dl,
DAG.getEntryNode(), TGA, MachinePointerInfo(),
false, false, false, 0);
} else {
// Local Exec TLS Model
SDValue TGAHi = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
Cpu0II::MO_TPREL_HI);
SDValue TGALo = DAG.getTargetGlobalAddress(GV, dl, PtrVT, 0,
Cpu0II::MO_TPREL_LO);
SDValue Hi = DAG.getNode(Cpu0ISD::Hi, dl, PtrVT, TGAHi);
SDValue Lo = DAG.getNode(Cpu0ISD::Lo, dl, PtrVT, TGALo);
Offset = DAG.getNode(ISD::ADD, dl, PtrVT, Hi, Lo);
}
SDValue ThreadPointer = DAG.getNode(Cpu0ISD::ThreadPointer, dl, PtrVT);
return DAG.getNode(ISD::ADD, dl, PtrVT, ThreadPointer, Offset);
}
SDValue X86SelectionDAGInfo::EmitTargetCodeForMemset(
SelectionDAG &DAG, const SDLoc &dl, SDValue Chain, SDValue Dst, SDValue Val,
SDValue Size, unsigned Align, bool isVolatile,
MachinePointerInfo DstPtrInfo) const {
ConstantSDNode *ConstantSize = dyn_cast<ConstantSDNode>(Size);
const X86Subtarget &Subtarget =
DAG.getMachineFunction().getSubtarget<X86Subtarget>();
#ifndef NDEBUG
// If the base register might conflict with our physical registers, bail out.
const MCPhysReg ClobberSet[] = {X86::RCX, X86::RAX, X86::RDI,
X86::ECX, X86::EAX, X86::EDI};
assert(!isBaseRegConflictPossible(DAG, ClobberSet));
#endif
// If to a segment-relative address space, use the default lowering.
if (DstPtrInfo.getAddrSpace() >= 256)
return SDValue();
// If not DWORD aligned or size is more than the threshold, call the library.
// The libc version is likely to be faster for these cases. It can use the
// address value and run time information about the CPU.
if ((Align & 3) != 0 || !ConstantSize ||
ConstantSize->getZExtValue() > Subtarget.getMaxInlineSizeThreshold()) {
// Check to see if there is a specialized entry-point for memory zeroing.
ConstantSDNode *ValC = dyn_cast<ConstantSDNode>(Val);
if (const char *bzeroName = (ValC && ValC->isNullValue())
? DAG.getTargetLoweringInfo().getLibcallName(RTLIB::BZERO)
: nullptr) {
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
EVT IntPtr = TLI.getPointerTy(DAG.getDataLayout());
Type *IntPtrTy = DAG.getDataLayout().getIntPtrType(*DAG.getContext());
TargetLowering::ArgListTy Args;
TargetLowering::ArgListEntry Entry;
Entry.Node = Dst;
Entry.Ty = IntPtrTy;
Args.push_back(Entry);
Entry.Node = Size;
Args.push_back(Entry);
TargetLowering::CallLoweringInfo CLI(DAG);
CLI.setDebugLoc(dl)
.setChain(Chain)
.setLibCallee(CallingConv::C, Type::getVoidTy(*DAG.getContext()),
DAG.getExternalSymbol(bzeroName, IntPtr),
std::move(Args))
.setDiscardResult();
std::pair<SDValue,SDValue> CallResult = TLI.LowerCallTo(CLI);
return CallResult.second;
}
// Otherwise have the target-independent code call memset.
return SDValue();
}
uint64_t SizeVal = ConstantSize->getZExtValue();
SDValue InFlag;
EVT AVT;
SDValue Count;
ConstantSDNode *ValC = dyn_cast<ConstantSDNode>(Val);
unsigned BytesLeft = 0;
if (ValC) {
unsigned ValReg;
uint64_t Val = ValC->getZExtValue() & 255;
// If the value is a constant, then we can potentially use larger sets.
switch (Align & 3) {
case 2: // WORD aligned
AVT = MVT::i16;
ValReg = X86::AX;
Val = (Val << 8) | Val;
break;
case 0: // DWORD aligned
AVT = MVT::i32;
ValReg = X86::EAX;
Val = (Val << 8) | Val;
Val = (Val << 16) | Val;
if (Subtarget.is64Bit() && ((Align & 0x7) == 0)) { // QWORD aligned
AVT = MVT::i64;
ValReg = X86::RAX;
Val = (Val << 32) | Val;
}
break;
default: // Byte aligned
AVT = MVT::i8;
ValReg = X86::AL;
Count = DAG.getIntPtrConstant(SizeVal, dl);
break;
}
if (AVT.bitsGT(MVT::i8)) {
unsigned UBytes = AVT.getSizeInBits() / 8;
Count = DAG.getIntPtrConstant(SizeVal / UBytes, dl);
BytesLeft = SizeVal % UBytes;
}
Chain = DAG.getCopyToReg(Chain, dl, ValReg, DAG.getConstant(Val, dl, AVT),
InFlag);
InFlag = Chain.getValue(1);
} else {
AVT = MVT::i8;
Count = DAG.getIntPtrConstant(SizeVal, dl);
Chain = DAG.getCopyToReg(Chain, dl, X86::AL, Val, InFlag);
InFlag = Chain.getValue(1);
}
bool Use64BitRegs = Subtarget.isTarget64BitLP64();
Chain = DAG.getCopyToReg(Chain, dl, Use64BitRegs ? X86::RCX : X86::ECX,
Count, InFlag);
InFlag = Chain.getValue(1);
Chain = DAG.getCopyToReg(Chain, dl, Use64BitRegs ? X86::RDI : X86::EDI,
Dst, InFlag);
InFlag = Chain.getValue(1);
SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Ops[] = { Chain, DAG.getValueType(AVT), InFlag };
Chain = DAG.getNode(X86ISD::REP_STOS, dl, Tys, Ops);
if (BytesLeft) {
// Handle the last 1 - 7 bytes.
unsigned Offset = SizeVal - BytesLeft;
EVT AddrVT = Dst.getValueType();
EVT SizeVT = Size.getValueType();
Chain = DAG.getMemset(Chain, dl,
DAG.getNode(ISD::ADD, dl, AddrVT, Dst,
DAG.getConstant(Offset, dl, AddrVT)),
Val,
DAG.getConstant(BytesLeft, dl, SizeVT),
Align, isVolatile, false,
DstPtrInfo.getWithOffset(Offset));
}
// TODO: Use a Tokenfactor, as in memcpy, instead of a single chain.
return Chain;
}