setSchedulingPreference(Sched::RegPressure);
setStackPointerRegisterToSaveRestore(X86StackPtr);
+ if (Subtarget->isTargetWindows() && !Subtarget->isTargetCygMing()) {
+ // Setup Windows compiler runtime calls.
+ setLibcallName(RTLIB::SDIV_I64, "_alldiv");
+ setLibcallName(RTLIB::UDIV_I64, "_aulldiv");
+ setLibcallName(RTLIB::FPTOUINT_F64_I64, "_ftol2");
+ setLibcallName(RTLIB::FPTOUINT_F32_I64, "_ftol2");
+ setLibcallCallingConv(RTLIB::SDIV_I64, CallingConv::X86_StdCall);
+ setLibcallCallingConv(RTLIB::UDIV_I64, CallingConv::X86_StdCall);
+ setLibcallCallingConv(RTLIB::FPTOUINT_F64_I64, CallingConv::C);
+ setLibcallCallingConv(RTLIB::FPTOUINT_F32_I64, CallingConv::C);
+ }
+
if (Subtarget->isTargetDarwin()) {
// Darwin should use _setjmp/_longjmp instead of setjmp/longjmp.
setUseUnderscoreSetJmp(false);
// TODO: when we have SSE, these could be more efficient, by using movd/movq.
if (!X86ScalarSSEf64) {
- setOperationAction(ISD::BIT_CONVERT , MVT::f32 , Expand);
- setOperationAction(ISD::BIT_CONVERT , MVT::i32 , Expand);
+ setOperationAction(ISD::BITCAST , MVT::f32 , Expand);
+ setOperationAction(ISD::BITCAST , MVT::i32 , Expand);
if (Subtarget->is64Bit()) {
- setOperationAction(ISD::BIT_CONVERT , MVT::f64 , Expand);
+ setOperationAction(ISD::BITCAST , MVT::f64 , Expand);
// Without SSE, i64->f64 goes through memory.
- setOperationAction(ISD::BIT_CONVERT , MVT::i64 , Expand);
+ setOperationAction(ISD::BITCAST , MVT::i64 , Expand);
}
}
setOperationAction(ISD::FREM , MVT::f80 , Expand);
setOperationAction(ISD::FLT_ROUNDS_ , MVT::i32 , Custom);
- setOperationAction(ISD::CTPOP , MVT::i8 , Expand);
setOperationAction(ISD::CTTZ , MVT::i8 , Custom);
setOperationAction(ISD::CTLZ , MVT::i8 , Custom);
- setOperationAction(ISD::CTPOP , MVT::i16 , Expand);
setOperationAction(ISD::CTTZ , MVT::i16 , Custom);
setOperationAction(ISD::CTLZ , MVT::i16 , Custom);
- setOperationAction(ISD::CTPOP , MVT::i32 , Expand);
setOperationAction(ISD::CTTZ , MVT::i32 , Custom);
setOperationAction(ISD::CTLZ , MVT::i32 , Custom);
if (Subtarget->is64Bit()) {
- setOperationAction(ISD::CTPOP , MVT::i64 , Expand);
setOperationAction(ISD::CTTZ , MVT::i64 , Custom);
setOperationAction(ISD::CTLZ , MVT::i64 , Custom);
}
+ if (Subtarget->hasPOPCNT()) {
+ setOperationAction(ISD::CTPOP , MVT::i8 , Promote);
+ } else {
+ setOperationAction(ISD::CTPOP , MVT::i8 , Expand);
+ setOperationAction(ISD::CTPOP , MVT::i16 , Expand);
+ setOperationAction(ISD::CTPOP , MVT::i32 , Expand);
+ if (Subtarget->is64Bit())
+ setOperationAction(ISD::CTPOP , MVT::i64 , Expand);
+ }
+
setOperationAction(ISD::READCYCLECOUNTER , MVT::i64 , Custom);
setOperationAction(ISD::BSWAP , MVT::i16 , Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
if (Subtarget->is64Bit())
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64, Expand);
- if (Subtarget->isTargetCygMing())
+ if (Subtarget->isTargetCygMing() || Subtarget->isTargetWindows())
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Custom);
else
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
setOperationAction(ISD::UNDEF, MVT::f80, Expand);
setOperationAction(ISD::FCOPYSIGN, MVT::f80, Expand);
{
- bool ignored;
- APFloat TmpFlt(+0.0);
- TmpFlt.convert(APFloat::x87DoubleExtended, APFloat::rmNearestTiesToEven,
- &ignored);
+ APFloat TmpFlt = APFloat::getZero(APFloat::x87DoubleExtended);
addLegalFPImmediate(TmpFlt); // FLD0
TmpFlt.changeSign();
addLegalFPImmediate(TmpFlt); // FLD0/FCHS
+
+ bool ignored;
APFloat TmpFlt2(+1.0);
TmpFlt2.convert(APFloat::x87DoubleExtended, APFloat::rmNearestTiesToEven,
&ignored);
// FIXME: In order to prevent SSE instructions being expanded to MMX ones
// with -msoft-float, disable use of MMX as well.
if (!UseSoftFloat && !DisableMMX && Subtarget->hasMMX()) {
- addRegisterClass(MVT::x86mmx, X86::VR64RegisterClass, false);
+ addRegisterClass(MVT::x86mmx, X86::VR64RegisterClass);
// No operations on x86mmx supported, everything uses intrinsics.
}
setOperationAction(ISD::SELECT, MVT::v4i16, Expand);
setOperationAction(ISD::SELECT, MVT::v2i32, Expand);
setOperationAction(ISD::SELECT, MVT::v1i64, Expand);
- setOperationAction(ISD::BIT_CONVERT, MVT::v8i8, Expand);
- setOperationAction(ISD::BIT_CONVERT, MVT::v4i16, Expand);
- setOperationAction(ISD::BIT_CONVERT, MVT::v2i32, Expand);
- setOperationAction(ISD::BIT_CONVERT, MVT::v1i64, Expand);
+ setOperationAction(ISD::BITCAST, MVT::v8i8, Expand);
+ setOperationAction(ISD::BITCAST, MVT::v4i16, Expand);
+ setOperationAction(ISD::BITCAST, MVT::v2i32, Expand);
+ setOperationAction(ISD::BITCAST, MVT::v1i64, Expand);
if (!UseSoftFloat && Subtarget->hasSSE1()) {
addRegisterClass(MVT::v4f32, X86::VR128RegisterClass);
return TargetLowering::getJumpTableEncoding();
}
-/// getPICBaseSymbol - Return the X86-32 PIC base.
-MCSymbol *
-X86TargetLowering::getPICBaseSymbol(const MachineFunction *MF,
- MCContext &Ctx) const {
- const MCAsmInfo &MAI = *getTargetMachine().getMCAsmInfo();
- return Ctx.GetOrCreateSymbol(Twine(MAI.getPrivateGlobalPrefix())+
- Twine(MF->getFunctionNumber())+"$pb");
-}
-
-
const MCExpr *
X86TargetLowering::LowerCustomJumpTableEntry(const MachineJumpTableInfo *MJTI,
const MachineBasicBlock *MBB,
return TargetLowering::getPICJumpTableRelocBaseExpr(MF, JTI, Ctx);
// Otherwise, the reference is relative to the PIC base.
- return MCSymbolRefExpr::Create(getPICBaseSymbol(MF, Ctx), Ctx);
+ return MCSymbolRefExpr::Create(MF->getPICBaseSymbol(), Ctx);
}
/// getFunctionAlignment - Return the Log2 alignment of this function.
unsigned
X86TargetLowering::getRegPressureLimit(const TargetRegisterClass *RC,
MachineFunction &MF) const {
- unsigned FPDiff = RegInfo->hasFP(MF) ? 1 : 0;
+ const TargetFrameInfo *TFI = MF.getTarget().getFrameInfo();
+
+ unsigned FPDiff = TFI->hasFP(MF) ? 1 : 0;
switch (RC->getID()) {
default:
return 0;
if (Subtarget->is64Bit()) {
if (ValVT == MVT::x86mmx) {
if (VA.getLocReg() == X86::XMM0 || VA.getLocReg() == X86::XMM1) {
- ValToCopy = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, ValToCopy);
+ ValToCopy = DAG.getNode(ISD::BITCAST, dl, MVT::i64, ValToCopy);
ValToCopy = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64,
ValToCopy);
// If we don't have SSE2 available, convert to v4f32 so the generated
// register is legal.
if (!Subtarget->hasSSE2())
- ValToCopy = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v4f32,ValToCopy);
+ ValToCopy = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32,ValToCopy);
}
}
}
MVT::Other, &RetOps[0], RetOps.size());
}
+bool X86TargetLowering::isUsedByReturnOnly(SDNode *N) const {
+ if (N->getNumValues() != 1)
+ return false;
+ if (!N->hasNUsesOfValue(1, 0))
+ return false;
+
+ SDNode *Copy = *N->use_begin();
+ if (Copy->getOpcode() != ISD::CopyToReg &&
+ Copy->getOpcode() != ISD::FP_EXTEND)
+ return false;
+
+ bool HasRet = false;
+ for (SDNode::use_iterator UI = Copy->use_begin(), UE = Copy->use_end();
+ UI != UE; ++UI) {
+ if (UI->getOpcode() != X86ISD::RET_FLAG)
+ return false;
+ HasRet = true;
+ }
+
+ return HasRet;
+}
+
/// LowerCallResult - Lower the result values of a call into the
/// appropriate copies out of appropriate physical registers.
///
MVT::i64, InFlag).getValue(1);
Val = Chain.getValue(0);
}
- Val = DAG.getNode(ISD::BIT_CONVERT, dl, CopyVT, Val);
+ Val = DAG.getNode(ISD::BITCAST, dl, CopyVT, Val);
} else {
Chain = DAG.getCopyFromReg(Chain, dl, VA.getLocReg(),
CopyVT, InFlag).getValue(1);
return Ins[0].Flags.isSRet();
}
-/// CCAssignFnForNode - Selects the correct CCAssignFn for a the
-/// given CallingConvention value.
-CCAssignFn *X86TargetLowering::CCAssignFnForNode(CallingConv::ID CC) const {
- if (Subtarget->is64Bit()) {
- if (CC == CallingConv::GHC)
- return CC_X86_64_GHC;
- else if (Subtarget->isTargetWin64())
- return CC_X86_Win64_C;
- else
- return CC_X86_64_C;
- }
-
- if (CC == CallingConv::X86_FastCall)
- return CC_X86_32_FastCall;
- else if (CC == CallingConv::X86_ThisCall)
- return CC_X86_32_ThisCall;
- else if (CC == CallingConv::Fast)
- return CC_X86_32_FastCC;
- else if (CC == CallingConv::GHC)
- return CC_X86_32_GHC;
- else
- return CC_X86_32_C;
-}
-
/// CreateCopyOfByValArgument - Make a copy of an aggregate at address specified
/// by "Src" to address "Dst" with size and alignment information specified by
/// the specific parameter attribute. The copy will be passed as a byval
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
- CCInfo.AnalyzeFormalArguments(Ins, CCAssignFnForNode(CallConv));
+ CCInfo.AnalyzeFormalArguments(Ins, CC_X86);
unsigned LastVal = ~0U;
SDValue ArgValue;
ArgValue = DAG.getNode(ISD::AssertZext, dl, RegVT, ArgValue,
DAG.getValueType(VA.getValVT()));
else if (VA.getLocInfo() == CCValAssign::BCvt)
- ArgValue = DAG.getNode(ISD::BIT_CONVERT, dl, VA.getValVT(), ArgValue);
+ ArgValue = DAG.getNode(ISD::BITCAST, dl, VA.getValVT(), ArgValue);
if (VA.isExtInLoc()) {
// Handle MMX values passed in XMM regs.
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CallConv, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
- CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForNode(CallConv));
+ CCInfo.AnalyzeCallOperands(Outs, CC_X86);
// Get a count of how many bytes are to be pushed on the stack.
unsigned NumBytes = CCInfo.getNextStackOffset();
case CCValAssign::AExt:
if (RegVT.isVector() && RegVT.getSizeInBits() == 128) {
// Special case: passing MMX values in XMM registers.
- Arg = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::i64, Arg);
+ Arg = DAG.getNode(ISD::BITCAST, dl, MVT::i64, Arg);
Arg = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64, Arg);
Arg = getMOVL(DAG, dl, MVT::v2i64, DAG.getUNDEF(MVT::v2i64), Arg);
} else
Arg = DAG.getNode(ISD::ANY_EXTEND, dl, RegVT, Arg);
break;
case CCValAssign::BCvt:
- Arg = DAG.getNode(ISD::BIT_CONVERT, dl, RegVT, Arg);
+ Arg = DAG.getNode(ISD::BITCAST, dl, RegVT, Arg);
break;
case CCValAssign::Indirect: {
// Store the argument.
} else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
unsigned char OpFlags = 0;
- // On ELF targets, in either X86-64 or X86-32 mode, direct calls to external
- // symbols should go through the PLT.
+ // On ELF targets, in either X86-64 or X86-32 mode, direct calls to
+ // external symbols should go through the PLT.
if (Subtarget->isTargetELF() &&
getTargetMachine().getRelocationModel() == Reloc::PIC_) {
OpFlags = X86II::MO_PLT;
SmallVector<CCValAssign, 16> ArgLocs;
CCState CCInfo(CalleeCC, isVarArg, getTargetMachine(),
ArgLocs, *DAG.getContext());
- CCInfo.AnalyzeCallOperands(Outs, CCAssignFnForNode(CalleeCC));
+ CCInfo.AnalyzeCallOperands(Outs, CC_X86);
if (CCInfo.getNextStackOffset()) {
MachineFunction &MF = DAG.getMachineFunction();
if (MF.getInfo<X86MachineFunctionInfo>()->getBytesToPopOnReturn())
}
}
+ // An stdcall caller is expected to clean up its arguments; the callee
+ // isn't going to do that.
+ if (!CCMatch && CallerCC==CallingConv::X86_StdCall)
+ return false;
+
return true;
}
SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8f32, Ops, 8);
}
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT, Vec);
+ return DAG.getNode(ISD::BITCAST, dl, VT, Vec);
}
/// getOnesVector - Returns a vector of specified type with all bits set.
SDValue Cst = DAG.getTargetConstant(~0U, MVT::i32);
SDValue Vec;
Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst);
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT, Vec);
+ return DAG.getNode(ISD::BITCAST, dl, VT, Vec);
}
// Perform the splat.
int SplatMask[4] = { EltNo, EltNo, EltNo, EltNo };
- V1 = DAG.getNode(ISD::BIT_CONVERT, dl, PVT, V1);
+ V1 = DAG.getNode(ISD::BITCAST, dl, PVT, V1);
V1 = DAG.getVectorShuffle(PVT, dl, V1, DAG.getUNDEF(PVT), &SplatMask[0]);
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT, V1);
+ return DAG.getNode(ISD::BITCAST, dl, VT, V1);
}
/// getShuffleVectorZeroOrUndef - Return a vector_shuffle of the specified
}
// Actual nodes that may contain scalar elements
- if (Opcode == ISD::BIT_CONVERT) {
+ if (Opcode == ISD::BITCAST) {
V = V.getOperand(0);
EVT SrcVT = V.getValueType();
unsigned NumElems = VT.getVectorNumElements();
}
}
- return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, V);
+ return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, V);
}
/// LowerBuildVectorv8i16 - Custom lower build_vector of v8i16.
const TargetLowering &TLI, DebugLoc dl) {
EVT ShVT = MVT::v2i64;
unsigned Opc = isLeft ? X86ISD::VSHL : X86ISD::VSRL;
- SrcOp = DAG.getNode(ISD::BIT_CONVERT, dl, ShVT, SrcOp);
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT,
+ SrcOp = DAG.getNode(ISD::BITCAST, dl, ShVT, SrcOp);
+ return DAG.getNode(ISD::BITCAST, dl, VT,
DAG.getNode(Opc, dl, ShVT, SrcOp,
DAG.getConstant(NumBits, TLI.getShiftAmountTy())));
}
LD->getPointerInfo().getWithOffset(StartOffset),
false, false, 0);
// Canonicalize it to a v4i32 shuffle.
- V1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v4i32, V1);
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT,
+ V1 = DAG.getNode(ISD::BITCAST, dl, MVT::v4i32, V1);
+ return DAG.getNode(ISD::BITCAST, dl, VT,
DAG.getVectorShuffle(MVT::v4i32, dl, V1,
DAG.getUNDEF(MVT::v4i32),&Mask[0]));
}
SDValue ResNode = DAG.getMemIntrinsicNode(X86ISD::VZEXT_LOAD, DL, Tys,
Ops, 2, MVT::i32,
LDBase->getMemOperand());
- return DAG.getNode(ISD::BIT_CONVERT, DL, VT, ResNode);
+ return DAG.getNode(ISD::BITCAST, DL, VT, ResNode);
}
return SDValue();
}
DAG.getUNDEF(Item.getValueType()),
&Mask[0]);
}
- return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(), Item);
+ return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(), Item);
}
}
Item = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MiddleVT, Item);
Item = getShuffleVectorZeroOrUndef(Item, 0, true,
Subtarget->hasSSE2(), DAG);
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT, Item);
+ return DAG.getNode(ISD::BITCAST, dl, VT, Item);
}
}
assert(ResVT == MVT::v2i64 || ResVT == MVT::v4i32 ||
ResVT == MVT::v8i16 || ResVT == MVT::v16i8);
int Mask[2];
- SDValue InVec = DAG.getNode(ISD::BIT_CONVERT,dl, MVT::v1i64, Op.getOperand(0));
+ SDValue InVec = DAG.getNode(ISD::BITCAST,dl, MVT::v1i64, Op.getOperand(0));
SDValue VecOp = DAG.getNode(X86ISD::MOVQ2DQ, dl, MVT::v2i64, InVec);
InVec = Op.getOperand(1);
if (InVec.getOpcode() == ISD::SCALAR_TO_VECTOR) {
unsigned NumElts = ResVT.getVectorNumElements();
- VecOp = DAG.getNode(ISD::BIT_CONVERT, dl, ResVT, VecOp);
+ VecOp = DAG.getNode(ISD::BITCAST, dl, ResVT, VecOp);
VecOp = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, ResVT, VecOp,
InVec.getOperand(0), DAG.getIntPtrConstant(NumElts/2+1));
} else {
- InVec = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v1i64, InVec);
+ InVec = DAG.getNode(ISD::BITCAST, dl, MVT::v1i64, InVec);
SDValue VecOp2 = DAG.getNode(X86ISD::MOVQ2DQ, dl, MVT::v2i64, InVec);
Mask[0] = 0; Mask[1] = 2;
VecOp = DAG.getVectorShuffle(MVT::v2i64, dl, VecOp, VecOp2, Mask);
}
- return DAG.getNode(ISD::BIT_CONVERT, dl, ResVT, VecOp);
+ return DAG.getNode(ISD::BITCAST, dl, ResVT, VecOp);
}
// v8i16 shuffles - Prefer shuffles in the following order:
MaskV.push_back(BestLoQuad < 0 ? 0 : BestLoQuad);
MaskV.push_back(BestHiQuad < 0 ? 1 : BestHiQuad);
NewV = DAG.getVectorShuffle(MVT::v2i64, dl,
- DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2i64, V1),
- DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2i64, V2), &MaskV[0]);
- NewV = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v8i16, NewV);
+ DAG.getNode(ISD::BITCAST, dl, MVT::v2i64, V1),
+ DAG.getNode(ISD::BITCAST, dl, MVT::v2i64, V2), &MaskV[0]);
+ NewV = DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, NewV);
// Rewrite the MaskVals and assign NewV to V1 if NewV now contains all the
// source words for the shuffle, to aid later transformations.
pshufbMask.push_back(DAG.getConstant(EltIdx, MVT::i8));
pshufbMask.push_back(DAG.getConstant(EltIdx+1, MVT::i8));
}
- V1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, V1);
+ V1 = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, V1);
V1 = DAG.getNode(X86ISD::PSHUFB, dl, MVT::v16i8, V1,
DAG.getNode(ISD::BUILD_VECTOR, dl,
MVT::v16i8, &pshufbMask[0], 16));
if (!TwoInputs)
- return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v8i16, V1);
+ return DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, V1);
// Calculate the shuffle mask for the second input, shuffle it, and
// OR it with the first shuffled input.
pshufbMask.push_back(DAG.getConstant(EltIdx - 16, MVT::i8));
pshufbMask.push_back(DAG.getConstant(EltIdx - 15, MVT::i8));
}
- V2 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, V2);
+ V2 = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, V2);
V2 = DAG.getNode(X86ISD::PSHUFB, dl, MVT::v16i8, V2,
DAG.getNode(ISD::BUILD_VECTOR, dl,
MVT::v16i8, &pshufbMask[0], 16));
V1 = DAG.getNode(ISD::OR, dl, MVT::v16i8, V1, V2);
- return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v8i16, V1);
+ return DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, V1);
}
// If BestLoQuad >= 0, generate a pshuflw to put the low elements in order,
// No SSSE3 - Calculate in place words and then fix all out of place words
// With 0-16 extracts & inserts. Worst case is 16 bytes out of order from
// the 16 different words that comprise the two doublequadword input vectors.
- V1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v8i16, V1);
- V2 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v8i16, V2);
+ V1 = DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, V1);
+ V2 = DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, V2);
SDValue NewV = V2Only ? V2 : V1;
for (int i = 0; i != 8; ++i) {
int Elt0 = MaskVals[i*2];
NewV = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, MVT::v8i16, NewV, InsElt,
DAG.getIntPtrConstant(i));
}
- return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v16i8, NewV);
+ return DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, NewV);
}
/// RewriteAsNarrowerShuffle - Try rewriting v8i16 and v16i8 shuffles as 4 wide
MaskVec.push_back(StartIdx / Scale);
}
- V1 = DAG.getNode(ISD::BIT_CONVERT, dl, NewVT, V1);
- V2 = DAG.getNode(ISD::BIT_CONVERT, dl, NewVT, V2);
+ V1 = DAG.getNode(ISD::BITCAST, dl, NewVT, V1);
+ V2 = DAG.getNode(ISD::BITCAST, dl, NewVT, V2);
return DAG.getVectorShuffle(NewVT, dl, V1, V2, &MaskVec[0]);
}
// movssrr and movsdrr do not clear top bits. Try to use movd, movq
// instead.
MVT ExtVT = (OpVT == MVT::v2f64) ? MVT::i64 : MVT::i32;
- if ((ExtVT.SimpleTy != MVT::i64 || Subtarget->is64Bit()) &&
+ if ((ExtVT != MVT::i64 || Subtarget->is64Bit()) &&
SrcOp.getOpcode() == ISD::SCALAR_TO_VECTOR &&
- SrcOp.getOperand(0).getOpcode() == ISD::BIT_CONVERT &&
+ SrcOp.getOperand(0).getOpcode() == ISD::BITCAST &&
SrcOp.getOperand(0).getOperand(0).getValueType() == ExtVT) {
// PR2108
OpVT = (OpVT == MVT::v2f64) ? MVT::v2i64 : MVT::v4i32;
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT,
+ return DAG.getNode(ISD::BITCAST, dl, VT,
DAG.getNode(X86ISD::VZEXT_MOVL, dl, OpVT,
DAG.getNode(ISD::SCALAR_TO_VECTOR, dl,
OpVT,
}
}
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT,
+ return DAG.getNode(ISD::BITCAST, dl, VT,
DAG.getNode(X86ISD::VZEXT_MOVL, dl, OpVT,
- DAG.getNode(ISD::BIT_CONVERT, dl,
+ DAG.getNode(ISD::BITCAST, dl,
OpVT, SrcOp)));
}
}
static bool MayFoldVectorLoad(SDValue V) {
- if (V.hasOneUse() && V.getOpcode() == ISD::BIT_CONVERT)
+ if (V.hasOneUse() && V.getOpcode() == ISD::BITCAST)
V = V.getOperand(0);
if (V.hasOneUse() && V.getOpcode() == ISD::SCALAR_TO_VECTOR)
V = V.getOperand(0);
// one use. Remove this version after this bug get fixed.
// rdar://8434668, PR8156
static bool RelaxedMayFoldVectorLoad(SDValue V) {
- if (V.hasOneUse() && V.getOpcode() == ISD::BIT_CONVERT)
+ if (V.hasOneUse() && V.getOpcode() == ISD::BITCAST)
V = V.getOperand(0);
if (V.hasOneUse() && V.getOpcode() == ISD::SCALAR_TO_VECTOR)
V = V.getOperand(0);
// If the bit convert changed the number of elements, it is unsafe
// to examine the mask.
bool HasShuffleIntoBitcast = false;
- if (V.getOpcode() == ISD::BIT_CONVERT) {
+ if (V.getOpcode() == ISD::BITCAST) {
EVT SrcVT = V.getOperand(0).getValueType();
if (SrcVT.getVectorNumElements() != VT.getVectorNumElements())
return false;
V = (Idx < (int)NumElems) ? V.getOperand(0) : V.getOperand(1);
// Skip one more bit_convert if necessary
- if (V.getOpcode() == ISD::BIT_CONVERT)
+ if (V.getOpcode() == ISD::BITCAST)
V = V.getOperand(0);
if (ISD::isNormalLoad(V.getNode())) {
EVT VT = Op.getValueType();
// Canonizalize to v2f64.
- V1 = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2f64, V1);
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT,
+ V1 = DAG.getNode(ISD::BITCAST, dl, MVT::v2f64, V1);
+ return DAG.getNode(ISD::BITCAST, dl, VT,
getTargetShuffleNode(X86ISD::MOVDDUP, dl, MVT::v2f64,
V1, DAG));
}
if (VT == MVT::v8i16 || VT == MVT::v16i8) {
SDValue NewOp = RewriteAsNarrowerShuffle(SVOp, DAG, dl);
if (NewOp.getNode())
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT, NewOp);
+ return DAG.getNode(ISD::BITCAST, dl, VT, NewOp);
} else if ((VT == MVT::v4i32 || (VT == MVT::v4f32 && Subtarget->hasSSE2()))) {
// FIXME: Figure out a cleaner way to do this.
// Try to make use of movq to zero out the top part.
if (Idx == 0)
return DAG.getNode(ISD::TRUNCATE, dl, MVT::i16,
DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i32,
- DAG.getNode(ISD::BIT_CONVERT, dl,
+ DAG.getNode(ISD::BITCAST, dl,
MVT::v4i32,
Op.getOperand(0)),
Op.getOperand(1)));
if ((User->getOpcode() != ISD::STORE ||
(isa<ConstantSDNode>(Op.getOperand(1)) &&
cast<ConstantSDNode>(Op.getOperand(1))->isNullValue())) &&
- (User->getOpcode() != ISD::BIT_CONVERT ||
+ (User->getOpcode() != ISD::BITCAST ||
User->getValueType(0) != MVT::i32))
return SDValue();
SDValue Extract = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i32,
- DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v4i32,
+ DAG.getNode(ISD::BITCAST, dl, MVT::v4i32,
Op.getOperand(0)),
Op.getOperand(1));
- return DAG.getNode(ISD::BIT_CONVERT, dl, MVT::f32, Extract);
+ return DAG.getNode(ISD::BITCAST, dl, MVT::f32, Extract);
} else if (VT == MVT::i32) {
// ExtractPS works with constant index.
if (isa<ConstantSDNode>(Op.getOperand(1)))
if (Idx == 0)
return DAG.getNode(ISD::TRUNCATE, dl, MVT::i16,
DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i32,
- DAG.getNode(ISD::BIT_CONVERT, dl,
+ DAG.getNode(ISD::BITCAST, dl,
MVT::v4i32, Vec),
Op.getOperand(1)));
// Transform it so it match pextrw which produces a 32-bit result.
SDValue AnyExt = DAG.getNode(ISD::ANY_EXTEND, dl, MVT::i32, Op.getOperand(0));
assert(Op.getValueType().getSimpleVT().getSizeInBits() == 128 &&
"Expected an SSE type!");
- return DAG.getNode(ISD::BIT_CONVERT, dl, Op.getValueType(),
+ return DAG.getNode(ISD::BITCAST, dl, Op.getValueType(),
DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v4i32,AnyExt));
}
Result = DAG.getNode(WrapperKind, DL, getPointerTy(), Result);
// With PIC, the address is actually $g + Offset.
- if (OpFlag) {
+ if (OpFlag)
Result = DAG.getNode(ISD::ADD, DL, getPointerTy(),
DAG.getNode(X86ISD::GlobalBaseReg,
DebugLoc(), getPointerTy()),
Result);
- }
return Result;
}
MachinePointerInfo::getConstantPool(),
false, false, 16);
SDValue Unpck2 = getUnpackl(DAG, dl, MVT::v4i32, Unpck1, CLod0);
- SDValue XR2F = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2f64, Unpck2);
+ SDValue XR2F = DAG.getNode(ISD::BITCAST, dl, MVT::v2f64, Unpck2);
SDValue CLod1 = DAG.getLoad(MVT::v2f64, dl, CLod0.getValue(1), CPIdx1,
MachinePointerInfo::getConstantPool(),
false, false, 16);
DAG.getIntPtrConstant(0)));
Load = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64,
- DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2f64, Load),
+ DAG.getNode(ISD::BITCAST, dl, MVT::v2f64, Load),
DAG.getIntPtrConstant(0));
// Or the load with the bias.
SDValue Or = DAG.getNode(ISD::OR, dl, MVT::v2i64,
- DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2i64,
+ DAG.getNode(ISD::BITCAST, dl, MVT::v2i64,
DAG.getNode(ISD::SCALAR_TO_VECTOR, dl,
MVT::v2f64, Load)),
- DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2i64,
+ DAG.getNode(ISD::BITCAST, dl, MVT::v2i64,
DAG.getNode(ISD::SCALAR_TO_VECTOR, dl,
MVT::v2f64, Bias)));
Or = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64,
- DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2f64, Or),
+ DAG.getNode(ISD::BITCAST, dl, MVT::v2f64, Or),
DAG.getIntPtrConstant(0));
// Subtract the bias.
MachinePointerInfo::getConstantPool(),
false, false, 16);
if (VT.isVector()) {
- return DAG.getNode(ISD::BIT_CONVERT, dl, VT,
+ return DAG.getNode(ISD::BITCAST, dl, VT,
DAG.getNode(ISD::XOR, dl, MVT::v2i64,
- DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2i64,
+ DAG.getNode(ISD::BITCAST, dl, MVT::v2i64,
Op.getOperand(0)),
- DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v2i64, Mask)));
+ DAG.getNode(ISD::BITCAST, dl, MVT::v2i64, Mask)));
} else {
return DAG.getNode(X86ISD::FXOR, dl, VT, Op.getOperand(0), Mask);
}
SignBit = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2f64, SignBit);
SignBit = DAG.getNode(X86ISD::FSRL, dl, MVT::v2f64, SignBit,
DAG.getConstant(32, MVT::i32));
- SignBit = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v4f32, SignBit);
+ SignBit = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32, SignBit);
SignBit = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f32, SignBit,
DAG.getIntPtrConstant(0));
}
SDValue
X86TargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op,
SelectionDAG &DAG) const {
- assert(Subtarget->isTargetCygMing() &&
- "This should be used only on Cygwin/Mingw targets");
+ assert((Subtarget->isTargetCygMing() || Subtarget->isTargetWindows()) &&
+ "This should be used only on Windows targets");
DebugLoc dl = Op.getDebugLoc();
// Get the inputs.
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Flag);
- Chain = DAG.getNode(X86ISD::MINGW_ALLOCA, dl, NodeTys, Chain, Flag);
+ Chain = DAG.getNode(X86ISD::WIN_ALLOCA, dl, NodeTys, Chain, Flag);
Flag = Chain.getValue(1);
Chain = DAG.getCopyFromReg(Chain, dl, X86StackPtr, SPTy).getValue(1);
}
SDValue X86TargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG) const {
- // X86-64 va_list is a struct { i32, i32, i8*, i8* }.
- assert(Subtarget->is64Bit() && "This code only handles 64-bit va_arg!");
+ assert(Subtarget->is64Bit() &&
+ "LowerVAARG only handles 64-bit va_arg!");
+ assert((Subtarget->isTargetLinux() ||
+ Subtarget->isTargetDarwin()) &&
+ "Unhandled target in LowerVAARG");
+ assert(Op.getNode()->getNumOperands() == 4);
+ SDValue Chain = Op.getOperand(0);
+ SDValue SrcPtr = Op.getOperand(1);
+ const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
+ unsigned Align = Op.getConstantOperandVal(3);
+ DebugLoc dl = Op.getDebugLoc();
- report_fatal_error("VAArgInst is not yet implemented for x86-64!");
- return SDValue();
+ EVT ArgVT = Op.getNode()->getValueType(0);
+ const Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
+ uint32_t ArgSize = getTargetData()->getTypeAllocSize(ArgTy);
+ uint8_t ArgMode;
+
+ // Decide which area this value should be read from.
+ // TODO: Implement the AMD64 ABI in its entirety. This simple
+ // selection mechanism works only for the basic types.
+ if (ArgVT == MVT::f80) {
+ llvm_unreachable("va_arg for f80 not yet implemented");
+ } else if (ArgVT.isFloatingPoint() && ArgSize <= 16 /*bytes*/) {
+ ArgMode = 2; // Argument passed in XMM register. Use fp_offset.
+ } else if (ArgVT.isInteger() && ArgSize <= 32 /*bytes*/) {
+ ArgMode = 1; // Argument passed in GPR64 register(s). Use gp_offset.
+ } else {
+ llvm_unreachable("Unhandled argument type in LowerVAARG");
+ }
+
+ if (ArgMode == 2) {
+ // Sanity Check: Make sure using fp_offset makes sense.
+ assert(!UseSoftFloat &&
+ !(DAG.getMachineFunction()
+ .getFunction()->hasFnAttr(Attribute::NoImplicitFloat)) &&
+ Subtarget->hasSSE1());
+ }
+
+ // Insert VAARG_64 node into the DAG
+ // VAARG_64 returns two values: Variable Argument Address, Chain
+ SmallVector<SDValue, 11> InstOps;
+ InstOps.push_back(Chain);
+ InstOps.push_back(SrcPtr);
+ InstOps.push_back(DAG.getConstant(ArgSize, MVT::i32));
+ InstOps.push_back(DAG.getConstant(ArgMode, MVT::i8));
+ InstOps.push_back(DAG.getConstant(Align, MVT::i32));
+ SDVTList VTs = DAG.getVTList(getPointerTy(), MVT::Other);
+ SDValue VAARG = DAG.getMemIntrinsicNode(X86ISD::VAARG_64, dl,
+ VTs, &InstOps[0], InstOps.size(),
+ MVT::i64,
+ MachinePointerInfo(SV),
+ /*Align=*/0,
+ /*Volatile=*/false,
+ /*ReadMem=*/true,
+ /*WriteMem=*/true);
+ Chain = VAARG.getValue(1);
+
+ // Load the next argument and return it
+ return DAG.getLoad(ArgVT, dl,
+ Chain,
+ VAARG,
+ MachinePointerInfo(),
+ false, false, 0);
}
SDValue X86TargetLowering::LowerVACOPY(SDValue Op, SelectionDAG &DAG) const {
}
EVT VT = Op.getValueType();
- ShAmt = DAG.getNode(ISD::BIT_CONVERT, dl, VT, ShAmt);
+ ShAmt = DAG.getNode(ISD::BITCAST, dl, VT, ShAmt);
return DAG.getNode(ISD::INTRINSIC_WO_CHAIN, dl, VT,
DAG.getConstant(NewIntNo, MVT::i32),
Op.getOperand(1), ShAmt);
false, false, 16);
Op = DAG.getNode(ISD::ADD, dl, VT, Op, Addend);
- Op = DAG.getNode(ISD::BIT_CONVERT, dl, MVT::v4f32, Op);
+ Op = DAG.getNode(ISD::BITCAST, dl, MVT::v4f32, Op);
Op = DAG.getNode(ISD::FP_TO_SINT, dl, VT, Op);
return DAG.getNode(ISD::MUL, dl, VT, Op, R);
}
return DAG.getMergeValues(Ops, 2, dl);
}
-SDValue X86TargetLowering::LowerBIT_CONVERT(SDValue Op,
+SDValue X86TargetLowering::LowerBITCAST(SDValue Op,
SelectionDAG &DAG) const {
EVT SrcVT = Op.getOperand(0).getValueType();
EVT DstVT = Op.getValueType();
assert((Subtarget->is64Bit() && !Subtarget->hasSSE2() &&
Subtarget->hasMMX() && !DisableMMX) &&
- "Unexpected custom BIT_CONVERT");
+ "Unexpected custom BITCAST");
assert((DstVT == MVT::i64 ||
(DstVT.isVector() && DstVT.getSizeInBits()==64)) &&
- "Unexpected custom BIT_CONVERT");
+ "Unexpected custom BITCAST");
// i64 <=> MMX conversions are Legal.
if (SrcVT==MVT::i64 && DstVT.isVector())
return Op;
case ISD::SMULO:
case ISD::UMULO: return LowerXALUO(Op, DAG);
case ISD::READCYCLECOUNTER: return LowerREADCYCLECOUNTER(Op, DAG);
- case ISD::BIT_CONVERT: return LowerBIT_CONVERT(Op, DAG);
+ case ISD::BITCAST: return LowerBITCAST(Op, DAG);
}
}
N->getOperand(1),
swapInH.getValue(1) };
SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Flag);
- SDValue Result = DAG.getNode(X86ISD::LCMPXCHG8_DAG, dl, Tys, Ops, 3);
+ MachineMemOperand *MMO = cast<AtomicSDNode>(N)->getMemOperand();
+ SDValue Result = DAG.getMemIntrinsicNode(X86ISD::LCMPXCHG8_DAG, dl, Tys,
+ Ops, 3, T, MMO);
SDValue cpOutL = DAG.getCopyFromReg(Result.getValue(0), dl, X86::EAX,
MVT::i32, Result.getValue(1));
SDValue cpOutH = DAG.getCopyFromReg(cpOutL.getValue(1), dl, X86::EDX,
case X86ISD::PUNPCKHDQ: return "X86ISD::PUNPCKHDQ";
case X86ISD::PUNPCKHQDQ: return "X86ISD::PUNPCKHQDQ";
case X86ISD::VASTART_SAVE_XMM_REGS: return "X86ISD::VASTART_SAVE_XMM_REGS";
- case X86ISD::MINGW_ALLOCA: return "X86ISD::MINGW_ALLOCA";
+ case X86ISD::VAARG_64: return "X86ISD::VAARG_64";
+ case X86ISD::WIN_ALLOCA: return "X86ISD::WIN_ALLOCA";
}
}
MachineBasicBlock *
X86TargetLowering::EmitPCMP(MachineInstr *MI, MachineBasicBlock *BB,
unsigned numArgs, bool memArg) const {
-
assert((Subtarget->hasSSE42() || Subtarget->hasAVX()) &&
"Target must have SSE4.2 or AVX features enabled");
DebugLoc dl = MI->getDebugLoc();
const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
-
unsigned Opc;
-
if (!Subtarget->hasAVX()) {
if (memArg)
Opc = numArgs == 3 ? X86::PCMPISTRM128rm : X86::PCMPESTRM128rm;
Opc = numArgs == 3 ? X86::VPCMPISTRM128rr : X86::VPCMPESTRM128rr;
}
- MachineInstrBuilder MIB = BuildMI(BB, dl, TII->get(Opc));
-
+ MachineInstrBuilder MIB = BuildMI(*BB, MI, dl, TII->get(Opc));
for (unsigned i = 0; i < numArgs; ++i) {
MachineOperand &Op = MI->getOperand(i+1);
-
if (!(Op.isReg() && Op.isImplicit()))
MIB.addOperand(Op);
}
-
- BuildMI(BB, dl, TII->get(X86::MOVAPSrr), MI->getOperand(0).getReg())
+ BuildMI(*BB, MI, dl, TII->get(X86::MOVAPSrr), MI->getOperand(0).getReg())
.addReg(X86::XMM0);
MI->eraseFromParent();
+ return BB;
+}
+
+MachineBasicBlock *
+X86TargetLowering::EmitMonitor(MachineInstr *MI, MachineBasicBlock *BB) const {
+ DebugLoc dl = MI->getDebugLoc();
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+
+ // Address into RAX/EAX, other two args into ECX, EDX.
+ unsigned MemOpc = Subtarget->is64Bit() ? X86::LEA64r : X86::LEA32r;
+ unsigned MemReg = Subtarget->is64Bit() ? X86::RAX : X86::EAX;
+ MachineInstrBuilder MIB = BuildMI(*BB, MI, dl, TII->get(MemOpc), MemReg);
+ for (int i = 0; i < X86::AddrNumOperands; ++i)
+ MIB.addOperand(MI->getOperand(i));
+
+ unsigned ValOps = X86::AddrNumOperands;
+ BuildMI(*BB, MI, dl, TII->get(TargetOpcode::COPY), X86::ECX)
+ .addReg(MI->getOperand(ValOps).getReg());
+ BuildMI(*BB, MI, dl, TII->get(TargetOpcode::COPY), X86::EDX)
+ .addReg(MI->getOperand(ValOps+1).getReg());
+
+ // The instruction doesn't actually take any operands though.
+ BuildMI(*BB, MI, dl, TII->get(X86::MONITORrrr));
+
+ MI->eraseFromParent(); // The pseudo is gone now.
+ return BB;
+}
+MachineBasicBlock *
+X86TargetLowering::EmitMwait(MachineInstr *MI, MachineBasicBlock *BB) const {
+ DebugLoc dl = MI->getDebugLoc();
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+
+ // First arg in ECX, the second in EAX.
+ BuildMI(*BB, MI, dl, TII->get(TargetOpcode::COPY), X86::ECX)
+ .addReg(MI->getOperand(0).getReg());
+ BuildMI(*BB, MI, dl, TII->get(TargetOpcode::COPY), X86::EAX)
+ .addReg(MI->getOperand(1).getReg());
+
+ // The instruction doesn't actually take any operands though.
+ BuildMI(*BB, MI, dl, TII->get(X86::MWAITrr));
+
+ MI->eraseFromParent(); // The pseudo is gone now.
return BB;
}
+MachineBasicBlock *
+X86TargetLowering::EmitVAARG64WithCustomInserter(
+ MachineInstr *MI,
+ MachineBasicBlock *MBB) const {
+ // Emit va_arg instruction on X86-64.
+
+ // Operands to this pseudo-instruction:
+ // 0 ) Output : destination address (reg)
+ // 1-5) Input : va_list address (addr, i64mem)
+ // 6 ) ArgSize : Size (in bytes) of vararg type
+ // 7 ) ArgMode : 0=overflow only, 1=use gp_offset, 2=use fp_offset
+ // 8 ) Align : Alignment of type
+ // 9 ) EFLAGS (implicit-def)
+
+ assert(MI->getNumOperands() == 10 && "VAARG_64 should have 10 operands!");
+ assert(X86::AddrNumOperands == 5 && "VAARG_64 assumes 5 address operands");
+
+ unsigned DestReg = MI->getOperand(0).getReg();
+ MachineOperand &Base = MI->getOperand(1);
+ MachineOperand &Scale = MI->getOperand(2);
+ MachineOperand &Index = MI->getOperand(3);
+ MachineOperand &Disp = MI->getOperand(4);
+ MachineOperand &Segment = MI->getOperand(5);
+ unsigned ArgSize = MI->getOperand(6).getImm();
+ unsigned ArgMode = MI->getOperand(7).getImm();
+ unsigned Align = MI->getOperand(8).getImm();
+
+ // Memory Reference
+ assert(MI->hasOneMemOperand() && "Expected VAARG_64 to have one memoperand");
+ MachineInstr::mmo_iterator MMOBegin = MI->memoperands_begin();
+ MachineInstr::mmo_iterator MMOEnd = MI->memoperands_end();
+
+ // Machine Information
+ const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
+ MachineRegisterInfo &MRI = MBB->getParent()->getRegInfo();
+ const TargetRegisterClass *AddrRegClass = getRegClassFor(MVT::i64);
+ const TargetRegisterClass *OffsetRegClass = getRegClassFor(MVT::i32);
+ DebugLoc DL = MI->getDebugLoc();
+
+ // struct va_list {
+ // i32 gp_offset
+ // i32 fp_offset
+ // i64 overflow_area (address)
+ // i64 reg_save_area (address)
+ // }
+ // sizeof(va_list) = 24
+ // alignment(va_list) = 8
+
+ unsigned TotalNumIntRegs = 6;
+ unsigned TotalNumXMMRegs = 8;
+ bool UseGPOffset = (ArgMode == 1);
+ bool UseFPOffset = (ArgMode == 2);
+ unsigned MaxOffset = TotalNumIntRegs * 8 +
+ (UseFPOffset ? TotalNumXMMRegs * 16 : 0);
+
+ /* Align ArgSize to a multiple of 8 */
+ unsigned ArgSizeA8 = (ArgSize + 7) & ~7;
+ bool NeedsAlign = (Align > 8);
+
+ MachineBasicBlock *thisMBB = MBB;
+ MachineBasicBlock *overflowMBB;
+ MachineBasicBlock *offsetMBB;
+ MachineBasicBlock *endMBB;
+
+ unsigned OffsetDestReg = 0; // Argument address computed by offsetMBB
+ unsigned OverflowDestReg = 0; // Argument address computed by overflowMBB
+ unsigned OffsetReg = 0;
+
+ if (!UseGPOffset && !UseFPOffset) {
+ // If we only pull from the overflow region, we don't create a branch.
+ // We don't need to alter control flow.
+ OffsetDestReg = 0; // unused
+ OverflowDestReg = DestReg;
+
+ offsetMBB = NULL;
+ overflowMBB = thisMBB;
+ endMBB = thisMBB;
+ } else {
+ // First emit code to check if gp_offset (or fp_offset) is below the bound.
+ // If so, pull the argument from reg_save_area. (branch to offsetMBB)
+ // If not, pull from overflow_area. (branch to overflowMBB)
+ //
+ // thisMBB
+ // | .
+ // | .
+ // offsetMBB overflowMBB
+ // | .
+ // | .
+ // endMBB
+
+ // Registers for the PHI in endMBB
+ OffsetDestReg = MRI.createVirtualRegister(AddrRegClass);
+ OverflowDestReg = MRI.createVirtualRegister(AddrRegClass);
+
+ const BasicBlock *LLVM_BB = MBB->getBasicBlock();
+ MachineFunction *MF = MBB->getParent();
+ overflowMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+ offsetMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+ endMBB = MF->CreateMachineBasicBlock(LLVM_BB);
+
+ MachineFunction::iterator MBBIter = MBB;
+ ++MBBIter;
+
+ // Insert the new basic blocks
+ MF->insert(MBBIter, offsetMBB);
+ MF->insert(MBBIter, overflowMBB);
+ MF->insert(MBBIter, endMBB);
+
+ // Transfer the remainder of MBB and its successor edges to endMBB.
+ endMBB->splice(endMBB->begin(), thisMBB,
+ llvm::next(MachineBasicBlock::iterator(MI)),
+ thisMBB->end());
+ endMBB->transferSuccessorsAndUpdatePHIs(thisMBB);
+
+ // Make offsetMBB and overflowMBB successors of thisMBB
+ thisMBB->addSuccessor(offsetMBB);
+ thisMBB->addSuccessor(overflowMBB);
+
+ // endMBB is a successor of both offsetMBB and overflowMBB
+ offsetMBB->addSuccessor(endMBB);
+ overflowMBB->addSuccessor(endMBB);
+
+ // Load the offset value into a register
+ OffsetReg = MRI.createVirtualRegister(OffsetRegClass);
+ BuildMI(thisMBB, DL, TII->get(X86::MOV32rm), OffsetReg)
+ .addOperand(Base)
+ .addOperand(Scale)
+ .addOperand(Index)
+ .addDisp(Disp, UseFPOffset ? 4 : 0)
+ .addOperand(Segment)
+ .setMemRefs(MMOBegin, MMOEnd);
+
+ // Check if there is enough room left to pull this argument.
+ BuildMI(thisMBB, DL, TII->get(X86::CMP32ri))
+ .addReg(OffsetReg)
+ .addImm(MaxOffset + 8 - ArgSizeA8);
+
+ // Branch to "overflowMBB" if offset >= max
+ // Fall through to "offsetMBB" otherwise
+ BuildMI(thisMBB, DL, TII->get(X86::GetCondBranchFromCond(X86::COND_AE)))
+ .addMBB(overflowMBB);
+ }
+
+ // In offsetMBB, emit code to use the reg_save_area.
+ if (offsetMBB) {
+ assert(OffsetReg != 0);
+
+ // Read the reg_save_area address.
+ unsigned RegSaveReg = MRI.createVirtualRegister(AddrRegClass);
+ BuildMI(offsetMBB, DL, TII->get(X86::MOV64rm), RegSaveReg)
+ .addOperand(Base)
+ .addOperand(Scale)
+ .addOperand(Index)
+ .addDisp(Disp, 16)
+ .addOperand(Segment)
+ .setMemRefs(MMOBegin, MMOEnd);
+
+ // Zero-extend the offset
+ unsigned OffsetReg64 = MRI.createVirtualRegister(AddrRegClass);
+ BuildMI(offsetMBB, DL, TII->get(X86::SUBREG_TO_REG), OffsetReg64)
+ .addImm(0)
+ .addReg(OffsetReg)
+ .addImm(X86::sub_32bit);
+
+ // Add the offset to the reg_save_area to get the final address.
+ BuildMI(offsetMBB, DL, TII->get(X86::ADD64rr), OffsetDestReg)
+ .addReg(OffsetReg64)
+ .addReg(RegSaveReg);
+
+ // Compute the offset for the next argument
+ unsigned NextOffsetReg = MRI.createVirtualRegister(OffsetRegClass);
+ BuildMI(offsetMBB, DL, TII->get(X86::ADD32ri), NextOffsetReg)
+ .addReg(OffsetReg)
+ .addImm(UseFPOffset ? 16 : 8);
+
+ // Store it back into the va_list.
+ BuildMI(offsetMBB, DL, TII->get(X86::MOV32mr))
+ .addOperand(Base)
+ .addOperand(Scale)
+ .addOperand(Index)
+ .addDisp(Disp, UseFPOffset ? 4 : 0)
+ .addOperand(Segment)
+ .addReg(NextOffsetReg)
+ .setMemRefs(MMOBegin, MMOEnd);
+
+ // Jump to endMBB
+ BuildMI(offsetMBB, DL, TII->get(X86::JMP_4))
+ .addMBB(endMBB);
+ }
+
+ //
+ // Emit code to use overflow area
+ //
+
+ // Load the overflow_area address into a register.
+ unsigned OverflowAddrReg = MRI.createVirtualRegister(AddrRegClass);
+ BuildMI(overflowMBB, DL, TII->get(X86::MOV64rm), OverflowAddrReg)
+ .addOperand(Base)
+ .addOperand(Scale)
+ .addOperand(Index)
+ .addDisp(Disp, 8)
+ .addOperand(Segment)
+ .setMemRefs(MMOBegin, MMOEnd);
+
+ // If we need to align it, do so. Otherwise, just copy the address
+ // to OverflowDestReg.
+ if (NeedsAlign) {
+ // Align the overflow address
+ assert((Align & (Align-1)) == 0 && "Alignment must be a power of 2");
+ unsigned TmpReg = MRI.createVirtualRegister(AddrRegClass);
+
+ // aligned_addr = (addr + (align-1)) & ~(align-1)
+ BuildMI(overflowMBB, DL, TII->get(X86::ADD64ri32), TmpReg)
+ .addReg(OverflowAddrReg)
+ .addImm(Align-1);
+
+ BuildMI(overflowMBB, DL, TII->get(X86::AND64ri32), OverflowDestReg)
+ .addReg(TmpReg)
+ .addImm(~(uint64_t)(Align-1));
+ } else {
+ BuildMI(overflowMBB, DL, TII->get(TargetOpcode::COPY), OverflowDestReg)
+ .addReg(OverflowAddrReg);
+ }
+
+ // Compute the next overflow address after this argument.
+ // (the overflow address should be kept 8-byte aligned)
+ unsigned NextAddrReg = MRI.createVirtualRegister(AddrRegClass);
+ BuildMI(overflowMBB, DL, TII->get(X86::ADD64ri32), NextAddrReg)
+ .addReg(OverflowDestReg)
+ .addImm(ArgSizeA8);
+
+ // Store the new overflow address.
+ BuildMI(overflowMBB, DL, TII->get(X86::MOV64mr))
+ .addOperand(Base)
+ .addOperand(Scale)
+ .addOperand(Index)
+ .addDisp(Disp, 8)
+ .addOperand(Segment)
+ .addReg(NextAddrReg)
+ .setMemRefs(MMOBegin, MMOEnd);
+
+ // If we branched, emit the PHI to the front of endMBB.
+ if (offsetMBB) {
+ BuildMI(*endMBB, endMBB->begin(), DL,
+ TII->get(X86::PHI), DestReg)
+ .addReg(OffsetDestReg).addMBB(offsetMBB)
+ .addReg(OverflowDestReg).addMBB(overflowMBB);
+ }
+
+ // Erase the pseudo instruction
+ MI->eraseFromParent();
+
+ return endMBB;
+}
+
MachineBasicBlock *
X86TargetLowering::EmitVAStartSaveXMMRegsWithCustomInserter(
MachineInstr *MI,
}
MachineBasicBlock *
-X86TargetLowering::EmitLoweredMingwAlloca(MachineInstr *MI,
+X86TargetLowering::EmitLoweredWinAlloca(MachineInstr *MI,
MachineBasicBlock *BB) const {
const TargetInstrInfo *TII = getTargetMachine().getInstrInfo();
DebugLoc DL = MI->getDebugLoc();
// FIXME: The code should be tweaked as soon as we'll try to do codegen for
// mingw-w64.
+ const char *StackProbeSymbol =
+ Subtarget->isTargetWindows() ? "_chkstk" : "_alloca";
+
BuildMI(*BB, MI, DL, TII->get(X86::CALLpcrel32))
- .addExternalSymbol("_alloca")
+ .addExternalSymbol(StackProbeSymbol)
.addReg(X86::EAX, RegState::Implicit)
.addReg(X86::ESP, RegState::Implicit)
.addReg(X86::EAX, RegState::Define | RegState::Implicit)
MachineBasicBlock *BB) const {
switch (MI->getOpcode()) {
default: assert(false && "Unexpected instr type to insert");
- case X86::MINGW_ALLOCA:
- return EmitLoweredMingwAlloca(MI, BB);
+ case X86::WIN_ALLOCA:
+ return EmitLoweredWinAlloca(MI, BB);
case X86::TLSCall_32:
case X86::TLSCall_64:
return EmitLoweredTLSCall(MI, BB);
case X86::VPCMPESTRM128MEM:
return EmitPCMP(MI, BB, 5, true /* in mem */);
+ // Thread synchronization.
+ case X86::MONITOR:
+ return EmitMonitor(MI, BB);
+ case X86::MWAIT:
+ return EmitMwait(MI, BB);
+
// Atomic Lowering.
case X86::ATOMAND32:
return EmitAtomicBitwiseWithCustomInserter(MI, BB, X86::AND32rr,
false);
case X86::VASTART_SAVE_XMM_REGS:
return EmitVAStartSaveXMMRegsWithCustomInserter(MI, BB);
+
+ case X86::VAARG_64:
+ return EmitVAARG64WithCustomInserter(MI, BB);
}
}
static SDValue PerformVZEXT_MOVLCombine(SDNode *N, SelectionDAG &DAG) {
SDValue Op = N->getOperand(0);
- if (Op.getOpcode() == ISD::BIT_CONVERT)
+ if (Op.getOpcode() == ISD::BITCAST)
Op = Op.getOperand(0);
EVT VT = N->getValueType(0), OpVT = Op.getValueType();
if (Op.getOpcode() == X86ISD::VZEXT_LOAD &&
VT.getVectorElementType().getSizeInBits() ==
OpVT.getVectorElementType().getSizeInBits()) {
- return DAG.getNode(ISD::BIT_CONVERT, N->getDebugLoc(), VT, Op);
+ return DAG.getNode(ISD::BITCAST, N->getDebugLoc(), VT, Op);
}
return SDValue();
}
bool X86TargetLowering::ExpandInlineAsm(CallInst *CI) const {
InlineAsm *IA = cast<InlineAsm>(CI->getCalledValue());
- std::vector<InlineAsm::ConstraintInfo> Constraints = IA->ParseConstraints();
+ InlineAsm::ConstraintInfoVector Constraints = IA->ParseConstraints();
std::string AsmStr = IA->getAsmString();
// TODO: should remove alternatives from the asmstring: "foo {a|b}" -> "foo a"
SmallVector<StringRef, 4> AsmPieces;
- SplitString(AsmStr, AsmPieces, "\n"); // ; as separator?
+ SplitString(AsmStr, AsmPieces, ";\n");
switch (AsmPieces.size()) {
default: return false;
}
break;
case 3:
+ if (CI->getType()->isIntegerTy(32) &&
+ IA->getConstraintString().compare(0, 5, "=r,0,") == 0) {
+ SmallVector<StringRef, 4> Words;
+ SplitString(AsmPieces[0], Words, " \t,");
+ if (Words.size() == 3 && Words[0] == "rorw" && Words[1] == "$$8" &&
+ Words[2] == "${0:w}") {
+ Words.clear();
+ SplitString(AsmPieces[1], Words, " \t,");
+ if (Words.size() == 3 && Words[0] == "rorl" && Words[1] == "$$16" &&
+ Words[2] == "$0") {
+ Words.clear();
+ SplitString(AsmPieces[2], Words, " \t,");
+ if (Words.size() == 3 && Words[0] == "rorw" && Words[1] == "$$8" &&
+ Words[2] == "${0:w}") {
+ AsmPieces.clear();
+ const std::string &Constraints = IA->getConstraintString();
+ SplitString(StringRef(Constraints).substr(5), AsmPieces, ",");
+ std::sort(AsmPieces.begin(), AsmPieces.end());
+ if (AsmPieces.size() == 4 &&
+ AsmPieces[0] == "~{cc}" &&
+ AsmPieces[1] == "~{dirflag}" &&
+ AsmPieces[2] == "~{flags}" &&
+ AsmPieces[3] == "~{fpsr}") {
+ return LowerToBSwap(CI);
+ }
+ }
+ }
+ }
+ }
if (CI->getType()->isIntegerTy(64) &&
Constraints.size() >= 2 &&
Constraints[0].Codes.size() == 1 && Constraints[0].Codes[0] == "A" &&
X86TargetLowering::getConstraintType(const std::string &Constraint) const {
if (Constraint.size() == 1) {
switch (Constraint[0]) {
- case 'A':
- return C_Register;
- case 'f':
- case 'r':
case 'R':
- case 'l':
case 'q':
case 'Q':
- case 'x':
+ case 'f':
+ case 't':
+ case 'u':
case 'y':
+ case 'x':
case 'Y':
return C_RegisterClass;
+ case 'a':
+ case 'b':
+ case 'c':
+ case 'd':
+ case 'S':
+ case 'D':
+ case 'A':
+ return C_Register;
+ case 'I':
+ case 'J':
+ case 'K':
+ case 'L':
+ case 'M':
+ case 'N':
+ case 'G':
+ case 'C':
case 'e':
case 'Z':
return C_Other;
return TargetLowering::getConstraintType(Constraint);
}
-/// Examine constraint type and operand type and determine a weight value,
-/// where: -1 = invalid match, and 0 = so-so match to 3 = good match.
+/// Examine constraint type and operand type and determine a weight value.
/// This object must already have been set up with the operand type
/// and the current alternative constraint selected.
-int X86TargetLowering::getSingleConstraintMatchWeight(
+TargetLowering::ConstraintWeight
+ X86TargetLowering::getSingleConstraintMatchWeight(
AsmOperandInfo &info, const char *constraint) const {
- int weight = -1;
+ ConstraintWeight weight = CW_Invalid;
Value *CallOperandVal = info.CallOperandVal;
// If we don't have a value, we can't do a match,
// but allow it at the lowest weight.
if (CallOperandVal == NULL)
- return 0;
+ return CW_Default;
+ const Type *type = CallOperandVal->getType();
// Look at the constraint type.
switch (*constraint) {
default:
- return TargetLowering::getSingleConstraintMatchWeight(info, constraint);
+ weight = TargetLowering::getSingleConstraintMatchWeight(info, constraint);
+ case 'R':
+ case 'q':
+ case 'Q':
+ case 'a':
+ case 'b':
+ case 'c':
+ case 'd':
+ case 'S':
+ case 'D':
+ case 'A':
+ if (CallOperandVal->getType()->isIntegerTy())
+ weight = CW_SpecificReg;
+ break;
+ case 'f':
+ case 't':
+ case 'u':
+ if (type->isFloatingPointTy())
+ weight = CW_SpecificReg;
+ break;
+ case 'y':
+ if (type->isX86_MMXTy() && !DisableMMX && Subtarget->hasMMX())
+ weight = CW_SpecificReg;
+ break;
+ case 'x':
+ case 'Y':
+ if ((type->getPrimitiveSizeInBits() == 128) && Subtarget->hasSSE1())
+ weight = CW_Register;
break;
case 'I':
if (ConstantInt *C = dyn_cast<ConstantInt>(info.CallOperandVal)) {
if (C->getZExtValue() <= 31)
- weight = 3;
+ weight = CW_Constant;
+ }
+ break;
+ case 'J':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if (C->getZExtValue() <= 63)
+ weight = CW_Constant;
+ }
+ break;
+ case 'K':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if ((C->getSExtValue() >= -0x80) && (C->getSExtValue() <= 0x7f))
+ weight = CW_Constant;
+ }
+ break;
+ case 'L':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if ((C->getZExtValue() == 0xff) || (C->getZExtValue() == 0xffff))
+ weight = CW_Constant;
+ }
+ break;
+ case 'M':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if (C->getZExtValue() <= 3)
+ weight = CW_Constant;
+ }
+ break;
+ case 'N':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if (C->getZExtValue() <= 0xff)
+ weight = CW_Constant;
+ }
+ break;
+ case 'G':
+ case 'C':
+ if (dyn_cast<ConstantFP>(CallOperandVal)) {
+ weight = CW_Constant;
+ }
+ break;
+ case 'e':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if ((C->getSExtValue() >= -0x80000000LL) &&
+ (C->getSExtValue() <= 0x7fffffffLL))
+ weight = CW_Constant;
+ }
+ break;
+ case 'Z':
+ if (ConstantInt *C = dyn_cast<ConstantInt>(CallOperandVal)) {
+ if (C->getZExtValue() <= 0xffffffff)
+ weight = CW_Constant;
}
break;
- // etc.
}
return weight;
}
projects / oota-llvm.git / blobdiff