ElemsPerChunk));
SDValue VecIdx = DAG.getIntPtrConstant(NormalizedIdxVal);
- SDValue Result = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, ResultVT, Vec,
- VecIdx);
-
- return Result;
+ return DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, ResultVT, Vec, VecIdx);
}
/// Generate a DAG to grab 128-bits from a vector > 128 bits. This
setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4i32, Custom);
setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v4f32, Custom);
+ // Only provide customized ctpop vector bit twiddling for vector types we
+ // know to perform better than using the popcnt instructions on each vector
+ // element. If popcnt isn't supported, always provide the custom version.
+ if (!Subtarget->hasPOPCNT()) {
+ setOperationAction(ISD::CTPOP, MVT::v4i32, Custom);
+ setOperationAction(ISD::CTPOP, MVT::v2i64, Custom);
+ }
+
// Custom lower build_vector, vector_shuffle, and extract_vector_elt.
for (int i = MVT::v16i8; i != MVT::v2i64; ++i) {
MVT VT = (MVT::SimpleValueType)i;
// The custom lowering for UINT_TO_FP for v8i32 becomes interesting
// when we have a 256bit-wide blend with immediate.
setOperationAction(ISD::UINT_TO_FP, MVT::v8i32, Custom);
+
+ // Only provide customized ctpop vector bit twiddling for vector types we
+ // know to perform better than using the popcnt instructions on each
+ // vector element. If popcnt isn't supported, always provide the custom
+ // version.
+ if (!Subtarget->hasPOPCNT())
+ setOperationAction(ISD::CTPOP, MVT::v4i64, Custom);
+
+ // Custom CTPOP always performs better on natively supported v8i32
+ setOperationAction(ISD::CTPOP, MVT::v8i32, Custom);
} else {
setOperationAction(ISD::ADD, MVT::v4i64, Custom);
setOperationAction(ISD::ADD, MVT::v8i32, Custom);
i <= MVT::LAST_VECTOR_VALUETYPE; ++i) {
MVT VT = (MVT::SimpleValueType)i;
+ if (VT.getScalarSizeInBits() >= 32) {
+ setOperationAction(ISD::MLOAD, VT, Legal);
+ setOperationAction(ISD::MSTORE, VT, Legal);
+ }
// Extract subvector is special because the value type
// (result) is 128-bit but the source is 256-bit wide.
if (VT.is128BitVector()) {
- if (VT.getScalarSizeInBits() >= 32) {
- setOperationAction(ISD::MLOAD, VT, Custom);
- setOperationAction(ISD::MSTORE, VT, Custom);
- }
setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Custom);
}
// Do not attempt to custom lower other non-256-bit vectors
if (!VT.is256BitVector())
continue;
- if (VT.getScalarSizeInBits() >= 32) {
- setOperationAction(ISD::MLOAD, VT, Legal);
- setOperationAction(ISD::MSTORE, VT, Legal);
- }
setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
// (result) is 256/128-bit but the source is 512-bit wide.
if (VT.is128BitVector() || VT.is256BitVector()) {
setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Custom);
- if ( EltSize >= 32) {
- setOperationAction(ISD::MLOAD, VT, Legal);
- setOperationAction(ISD::MSTORE, VT, Legal);
- }
}
if (VT.getVectorElementType() == MVT::i1)
setOperationAction(ISD::EXTRACT_SUBVECTOR, VT, Legal);
setOperationAction(ISD::LOAD, MVT::v64i8, Legal);
setOperationAction(ISD::SETCC, MVT::v32i1, Custom);
setOperationAction(ISD::SETCC, MVT::v64i1, Custom);
+ setOperationAction(ISD::ADD, MVT::v32i16, Legal);
+ setOperationAction(ISD::ADD, MVT::v64i8, Legal);
+ setOperationAction(ISD::SUB, MVT::v32i16, Legal);
+ setOperationAction(ISD::SUB, MVT::v64i8, Legal);
+ setOperationAction(ISD::MUL, MVT::v32i16, Legal);
for (int i = MVT::v32i8; i != MVT::v8i64; ++i) {
const MVT VT = (MVT::SimpleValueType)i;
setOperationAction(ISD::SETCC, MVT::v4i1, Custom);
setOperationAction(ISD::SETCC, MVT::v2i1, Custom);
setOperationAction(ISD::INSERT_SUBVECTOR, MVT::v8i1, Legal);
+
+ setOperationAction(ISD::AND, MVT::v8i32, Legal);
+ setOperationAction(ISD::OR, MVT::v8i32, Legal);
+ setOperationAction(ISD::XOR, MVT::v8i32, Legal);
+ setOperationAction(ISD::AND, MVT::v4i32, Legal);
+ setOperationAction(ISD::OR, MVT::v4i32, Legal);
+ setOperationAction(ISD::XOR, MVT::v4i32, Legal);
}
// SIGN_EXTEND_INREGs are evaluated by the extend type. Handle the expansion
// Predictable cmov don't hurt on atom because it's in-order.
PredictableSelectIsExpensive = !Subtarget->isAtom();
-
+ EnableExtLdPromotion = true;
setPrefFunctionAlignment(4); // 2^4 bytes.
verifyIntrinsicTables();
return VT.changeVectorElementTypeToInteger();
}
-/// getMaxByValAlign - Helper for getByValTypeAlignment to determine
+/// Helper for getByValTypeAlignment to determine
/// the desired ByVal argument alignment.
static void getMaxByValAlign(Type *Ty, unsigned &MaxAlign) {
if (MaxAlign == 16)
}
}
-/// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
+/// Return the desired alignment for ByVal aggregate
/// function arguments in the caller parameter area. For X86, aggregates
/// that contain SSE vectors are placed at 16-byte boundaries while the rest
/// are at 4-byte boundaries.
return Align;
}
-/// getOptimalMemOpType - Returns the target specific optimal type for load
+/// Returns the target specific optimal type for load
/// and store operations as a result of memset, memcpy, and memmove
/// lowering. If DstAlign is zero that means it's safe to destination
/// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
return true;
}
-/// getJumpTableEncoding - Return the entry encoding for a jump table in the
+/// Return the entry encoding for a jump table in the
/// current function. The returned value is a member of the
/// MachineJumpTableInfo::JTEntryKind enum.
unsigned X86TargetLowering::getJumpTableEncoding() const {
MCSymbolRefExpr::VK_GOTOFF, Ctx);
}
-/// getPICJumpTableRelocaBase - Returns relocation base for the given PIC
-/// jumptable.
+/// Returns relocation base for the given PIC jumptable.
SDValue X86TargetLowering::getPICJumpTableRelocBase(SDValue Table,
SelectionDAG &DAG) const {
if (!Subtarget->is64Bit())
return Table;
}
-/// getPICJumpTableRelocBaseExpr - This returns the relocation base for the
-/// given PIC jumptable, the same as getPICJumpTableRelocBase, but as an
-/// MCExpr.
+/// This returns the relocation base for the given PIC jumptable,
+/// the same as getPICJumpTableRelocBase, but as an MCExpr.
const MCExpr *X86TargetLowering::
getPICJumpTableRelocBaseExpr(const MachineFunction *MF, unsigned JTI,
MCContext &Ctx) const {
return VT.bitsLT(MinVT) ? MinVT : VT;
}
-/// LowerCallResult - Lower the result values of a call into the
+/// Lower the result values of a call into the
/// appropriate copies out of appropriate physical registers.
///
SDValue
return StackStructReturn;
}
-/// ArgsAreStructReturn - Determines whether a function uses struct
-/// return semantics.
+/// Determines whether a function uses struct return semantics.
static StructReturnType
argsAreStructReturn(const SmallVectorImpl<ISD::InputArg> &Ins) {
if (Ins.empty())
return StackStructReturn;
}
-/// 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
-/// function parameter.
+/// 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 function parameter.
static SDValue
CreateCopyOfByValArgument(SDValue Src, SDValue Dst, SDValue Chain,
ISD::ArgFlagsTy Flags, SelectionDAG &DAG,
MachinePointerInfo(), MachinePointerInfo());
}
-/// IsTailCallConvention - Return true if the calling convention is one that
+/// Return true if the calling convention is one that
/// supports tail call optimization.
static bool IsTailCallConvention(CallingConv::ID CC) {
return (CC == CallingConv::Fast || CC == CallingConv::GHC ||
return true;
}
-/// FuncIsMadeTailCallSafe - Return true if the function is being made into
+/// Return true if the function is being made into
/// a tailcall target by changing its ABI.
static bool FuncIsMadeTailCallSafe(CallingConv::ID CC,
bool GuaranteedTailCallOpt) {
MFI->CreateFixedObject(1, StackSize, true));
}
+ // Figure out if XMM registers are in use.
+ bool HaveXMMArgs = Is64Bit && !IsWin64;
+ bool NoImplicitFloatOps = Fn->getAttributes().hasAttribute(
+ AttributeSet::FunctionIndex, Attribute::NoImplicitFloat);
+ assert(!(MF.getTarget().Options.UseSoftFloat && NoImplicitFloatOps) &&
+ "SSE register cannot be used when SSE is disabled!");
+ if (MF.getTarget().Options.UseSoftFloat || NoImplicitFloatOps ||
+ !Subtarget->hasSSE1())
+ HaveXMMArgs = false;
+
// 64-bit calling conventions support varargs and register parameters, so we
- // have to do extra work to spill them in the prologue or forward them to
- // musttail calls.
- if (Is64Bit && isVarArg &&
- (MFI->hasVAStart() || MFI->hasMustTailInVarArgFunc())) {
+ // have to do extra work to spill them in the prologue.
+ if (Is64Bit && isVarArg && MFI->hasVAStart()) {
// Find the first unallocated argument registers.
ArrayRef<MCPhysReg> ArgGPRs = get64BitArgumentGPRs(CallConv, Subtarget);
ArrayRef<MCPhysReg> ArgXMMs = get64BitArgumentXMMs(MF, CallConv, Subtarget);
}
}
- // Store them to the va_list returned by va_start.
- if (MFI->hasVAStart()) {
- if (IsWin64) {
- const TargetFrameLowering &TFI = *MF.getSubtarget().getFrameLowering();
- // Get to the caller-allocated home save location. Add 8 to account
- // for the return address.
- int HomeOffset = TFI.getOffsetOfLocalArea() + 8;
- FuncInfo->setRegSaveFrameIndex(
+ if (IsWin64) {
+ const TargetFrameLowering &TFI = *MF.getSubtarget().getFrameLowering();
+ // Get to the caller-allocated home save location. Add 8 to account
+ // for the return address.
+ int HomeOffset = TFI.getOffsetOfLocalArea() + 8;
+ FuncInfo->setRegSaveFrameIndex(
MFI->CreateFixedObject(1, NumIntRegs * 8 + HomeOffset, false));
- // Fixup to set vararg frame on shadow area (4 x i64).
- if (NumIntRegs < 4)
- FuncInfo->setVarArgsFrameIndex(FuncInfo->getRegSaveFrameIndex());
- } else {
- // For X86-64, if there are vararg parameters that are passed via
- // registers, then we must store them to their spots on the stack so
- // they may be loaded by deferencing the result of va_next.
- FuncInfo->setVarArgsGPOffset(NumIntRegs * 8);
- FuncInfo->setVarArgsFPOffset(ArgGPRs.size() * 8 + NumXMMRegs * 16);
- FuncInfo->setRegSaveFrameIndex(MFI->CreateStackObject(
- ArgGPRs.size() * 8 + ArgXMMs.size() * 16, 16, false));
- }
-
- // Store the integer parameter registers.
- SmallVector<SDValue, 8> MemOps;
- SDValue RSFIN = DAG.getFrameIndex(FuncInfo->getRegSaveFrameIndex(),
- getPointerTy());
- unsigned Offset = FuncInfo->getVarArgsGPOffset();
- for (SDValue Val : LiveGPRs) {
- SDValue FIN = DAG.getNode(ISD::ADD, dl, getPointerTy(), RSFIN,
- DAG.getIntPtrConstant(Offset));
- SDValue Store =
- DAG.getStore(Val.getValue(1), dl, Val, FIN,
- MachinePointerInfo::getFixedStack(
- FuncInfo->getRegSaveFrameIndex(), Offset),
- false, false, 0);
- MemOps.push_back(Store);
- Offset += 8;
- }
-
- if (!ArgXMMs.empty() && NumXMMRegs != ArgXMMs.size()) {
- // Now store the XMM (fp + vector) parameter registers.
- SmallVector<SDValue, 12> SaveXMMOps;
- SaveXMMOps.push_back(Chain);
- SaveXMMOps.push_back(ALVal);
- SaveXMMOps.push_back(DAG.getIntPtrConstant(
- FuncInfo->getRegSaveFrameIndex()));
- SaveXMMOps.push_back(DAG.getIntPtrConstant(
- FuncInfo->getVarArgsFPOffset()));
- SaveXMMOps.insert(SaveXMMOps.end(), LiveXMMRegs.begin(),
- LiveXMMRegs.end());
- MemOps.push_back(DAG.getNode(X86ISD::VASTART_SAVE_XMM_REGS, dl,
- MVT::Other, SaveXMMOps));
- }
-
- if (!MemOps.empty())
- Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOps);
+ // Fixup to set vararg frame on shadow area (4 x i64).
+ if (NumIntRegs < 4)
+ FuncInfo->setVarArgsFrameIndex(FuncInfo->getRegSaveFrameIndex());
} else {
- // Add all GPRs, al, and XMMs to the list of forwards. We will add then
- // to the liveout set on a musttail call.
- assert(MFI->hasMustTailInVarArgFunc());
- auto &Forwards = FuncInfo->getForwardedMustTailRegParms();
- typedef X86MachineFunctionInfo::Forward Forward;
-
- for (unsigned I = 0, E = LiveGPRs.size(); I != E; ++I) {
- unsigned VReg =
- MF.getRegInfo().createVirtualRegister(&X86::GR64RegClass);
- Chain = DAG.getCopyToReg(Chain, dl, VReg, LiveGPRs[I]);
- Forwards.push_back(Forward(VReg, ArgGPRs[NumIntRegs + I], MVT::i64));
- }
-
- if (!ArgXMMs.empty()) {
- unsigned ALVReg =
- MF.getRegInfo().createVirtualRegister(&X86::GR8RegClass);
- Chain = DAG.getCopyToReg(Chain, dl, ALVReg, ALVal);
- Forwards.push_back(Forward(ALVReg, X86::AL, MVT::i8));
-
- for (unsigned I = 0, E = LiveXMMRegs.size(); I != E; ++I) {
- unsigned VReg =
- MF.getRegInfo().createVirtualRegister(&X86::VR128RegClass);
- Chain = DAG.getCopyToReg(Chain, dl, VReg, LiveXMMRegs[I]);
- Forwards.push_back(
- Forward(VReg, ArgXMMs[NumXMMRegs + I], MVT::v4f32));
- }
- }
+ // For X86-64, if there are vararg parameters that are passed via
+ // registers, then we must store them to their spots on the stack so
+ // they may be loaded by deferencing the result of va_next.
+ FuncInfo->setVarArgsGPOffset(NumIntRegs * 8);
+ FuncInfo->setVarArgsFPOffset(ArgGPRs.size() * 8 + NumXMMRegs * 16);
+ FuncInfo->setRegSaveFrameIndex(MFI->CreateStackObject(
+ ArgGPRs.size() * 8 + ArgXMMs.size() * 16, 16, false));
+ }
+
+ // Store the integer parameter registers.
+ SmallVector<SDValue, 8> MemOps;
+ SDValue RSFIN = DAG.getFrameIndex(FuncInfo->getRegSaveFrameIndex(),
+ getPointerTy());
+ unsigned Offset = FuncInfo->getVarArgsGPOffset();
+ for (SDValue Val : LiveGPRs) {
+ SDValue FIN = DAG.getNode(ISD::ADD, dl, getPointerTy(), RSFIN,
+ DAG.getIntPtrConstant(Offset));
+ SDValue Store =
+ DAG.getStore(Val.getValue(1), dl, Val, FIN,
+ MachinePointerInfo::getFixedStack(
+ FuncInfo->getRegSaveFrameIndex(), Offset),
+ false, false, 0);
+ MemOps.push_back(Store);
+ Offset += 8;
+ }
+
+ if (!ArgXMMs.empty() && NumXMMRegs != ArgXMMs.size()) {
+ // Now store the XMM (fp + vector) parameter registers.
+ SmallVector<SDValue, 12> SaveXMMOps;
+ SaveXMMOps.push_back(Chain);
+ SaveXMMOps.push_back(ALVal);
+ SaveXMMOps.push_back(DAG.getIntPtrConstant(
+ FuncInfo->getRegSaveFrameIndex()));
+ SaveXMMOps.push_back(DAG.getIntPtrConstant(
+ FuncInfo->getVarArgsFPOffset()));
+ SaveXMMOps.insert(SaveXMMOps.end(), LiveXMMRegs.begin(),
+ LiveXMMRegs.end());
+ MemOps.push_back(DAG.getNode(X86ISD::VASTART_SAVE_XMM_REGS, dl,
+ MVT::Other, SaveXMMOps));
+ }
+
+ if (!MemOps.empty())
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOps);
+ }
+
+ if (isVarArg && MFI->hasMustTailInVarArgFunc()) {
+ // Find the largest legal vector type.
+ MVT VecVT = MVT::Other;
+ // FIXME: Only some x86_32 calling conventions support AVX512.
+ if (Subtarget->hasAVX512() &&
+ (Is64Bit || (CallConv == CallingConv::X86_VectorCall ||
+ CallConv == CallingConv::Intel_OCL_BI)))
+ VecVT = MVT::v16f32;
+ else if (Subtarget->hasAVX())
+ VecVT = MVT::v8f32;
+ else if (Subtarget->hasSSE2())
+ VecVT = MVT::v4f32;
+
+ // We forward some GPRs and some vector types.
+ SmallVector<MVT, 2> RegParmTypes;
+ MVT IntVT = Is64Bit ? MVT::i64 : MVT::i32;
+ RegParmTypes.push_back(IntVT);
+ if (VecVT != MVT::Other)
+ RegParmTypes.push_back(VecVT);
+
+ // Compute the set of forwarded registers. The rest are scratch.
+ SmallVectorImpl<ForwardedRegister> &Forwards =
+ FuncInfo->getForwardedMustTailRegParms();
+ CCInfo.analyzeMustTailForwardedRegisters(Forwards, RegParmTypes, CC_X86);
+
+ // Conservatively forward AL on x86_64, since it might be used for varargs.
+ if (Is64Bit && !CCInfo.isAllocated(X86::AL)) {
+ unsigned ALVReg = MF.addLiveIn(X86::AL, &X86::GR8RegClass);
+ Forwards.push_back(ForwardedRegister(ALVReg, X86::AL, MVT::i8));
+ }
+
+ // Copy all forwards from physical to virtual registers.
+ for (ForwardedRegister &F : Forwards) {
+ // FIXME: Can we use a less constrained schedule?
+ SDValue RegVal = DAG.getCopyFromReg(Chain, dl, F.VReg, F.VT);
+ F.VReg = MF.getRegInfo().createVirtualRegister(getRegClassFor(F.VT));
+ Chain = DAG.getCopyToReg(Chain, dl, F.VReg, RegVal);
}
}
false, false, 0);
}
-/// EmitTailCallLoadRetAddr - Emit a load of return address if tail call
+/// Emit a load of return address if tail call
/// optimization is performed and it is required.
SDValue
X86TargetLowering::EmitTailCallLoadRetAddr(SelectionDAG &DAG,
return SDValue(OutRetAddr.getNode(), 1);
}
-/// EmitTailCallStoreRetAddr - Emit a store of the return address if tail call
+/// Emit a store of the return address if tail call
/// optimization is performed and it is required (FPDiff!=0).
static SDValue EmitTailCallStoreRetAddr(SelectionDAG &DAG, MachineFunction &MF,
SDValue Chain, SDValue RetAddrFrIdx,
DAG.getConstant(NumXMMRegs, MVT::i8)));
}
- if (Is64Bit && isVarArg && IsMustTail) {
+ if (isVarArg && IsMustTail) {
const auto &Forwards = X86Info->getForwardedMustTailRegParms();
for (const auto &F : Forwards) {
SDValue Val = DAG.getCopyFromReg(Chain, dl, F.VReg, F.VT);
return true;
}
+bool X86TargetLowering::isExtractSubvectorCheap(EVT ResVT,
+ unsigned Index) const {
+ if (!isOperationLegalOrCustom(ISD::EXTRACT_SUBVECTOR, ResVT))
+ return false;
+
+ return (Index == 0 || Index == ResVT.getVectorNumElements());
+}
+
/// isUndefOrInRange - Return true if Val is undef or if its value falls within
/// the specified range (L, H].
static bool isUndefOrInRange(int Val, int Low, int Hi) {
SDValue NewLd = SDValue();
- if (DAG.InferPtrAlignment(LDBase->getBasePtr()) >= 16)
- NewLd = DAG.getLoad(VT, DL, LDBase->getChain(), LDBase->getBasePtr(),
- LDBase->getPointerInfo(),
- LDBase->isVolatile(), LDBase->isNonTemporal(),
- LDBase->isInvariant(), 0);
NewLd = DAG.getLoad(VT, DL, LDBase->getChain(), LDBase->getBasePtr(),
- LDBase->getPointerInfo(),
- LDBase->isVolatile(), LDBase->isNonTemporal(),
- LDBase->isInvariant(), LDBase->getAlignment());
+ LDBase->getPointerInfo(), LDBase->isVolatile(),
+ LDBase->isNonTemporal(), LDBase->isInvariant(),
+ LDBase->getAlignment());
if (LDBase->hasAnyUseOfValue(1)) {
SDValue NewChain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other,
return NewLd;
}
- if (NumElems == 4 && LastLoadedElt == 1 &&
+
+ //TODO: The code below fires only for for loading the low v2i32 / v2f32
+ //of a v4i32 / v4f32. It's probably worth generalizing.
+ if (NumElems == 4 && LastLoadedElt == 1 && (EltVT.getSizeInBits() == 32) &&
DAG.getTargetLoweringInfo().isTypeLegal(MVT::v2i64)) {
SDVTList Tys = DAG.getVTList(MVT::v2i64, MVT::Other);
SDValue Ops[] = { LDBase->getChain(), LDBase->getBasePtr() };
if (!IsLoad)
return SDValue();
- if (ScalarSize == 32 || (IsGE256 && ScalarSize == 64))
+ if (ScalarSize == 32 || (IsGE256 && ScalarSize == 64) ||
+ (Subtarget->hasVLX() && ScalarSize == 64))
return DAG.getNode(X86ISD::VBROADCAST, dl, VT, Ld);
// The integer check is needed for the 64-bit into 128-bit so it doesn't match
cast<ConstantSDNode>(Op1)->isNullValue() &&
(CC == ISD::SETEQ || CC == ISD::SETNE)) {
SDValue NewSetCC = LowerToBT(Op0, CC, dl, DAG);
- if (NewSetCC.getNode())
+ if (NewSetCC.getNode()) {
+ if (VT == MVT::i1)
+ return DAG.getNode(ISD::TRUNCATE, dl, MVT::i1, NewSetCC);
return NewSetCC;
+ }
}
// Look for X == 0, X == 1, X != 0, or X != 1. We can simplify some forms of
return DAG.getNode(ISD::VSELECT, dl, VT, VMask, Op, PreservedSrc);
}
+/// \brief Creates an SDNode for a predicated scalar operation.
+/// \returns (X86vselect \p Mask, \p Op, \p PreservedSrc).
+/// The mask is comming as MVT::i8 and it should be truncated
+/// to MVT::i1 while lowering masking intrinsics.
+/// The main difference between ScalarMaskingNode and VectorMaskingNode is using
+/// "X86select" instead of "vselect". We just can't create the "vselect" node for
+/// a scalar instruction.
static SDValue getScalarMaskingNode(SDValue Op, SDValue Mask,
SDValue PreservedSrc,
const X86Subtarget *Subtarget,
return getTargetVShiftNode(IntrData->Opc0, dl, Op.getSimpleValueType(),
Op.getOperand(1), Op.getOperand(2), DAG);
case VSHIFT_MASK:
- return getVectorMaskingNode(getTargetVShiftNode(IntrData->Opc0, dl, Op.getSimpleValueType(),
- Op.getOperand(1), Op.getOperand(2), DAG),
- Op.getOperand(4), Op.getOperand(3), Subtarget, DAG);
+ return getVectorMaskingNode(getTargetVShiftNode(IntrData->Opc0, dl,
+ Op.getSimpleValueType(),
+ Op.getOperand(1),
+ Op.getOperand(2), DAG),
+ Op.getOperand(4), Op.getOperand(3), Subtarget,
+ DAG);
+ case COMPRESS_EXPAND_IN_REG: {
+ SDValue Mask = Op.getOperand(3);
+ SDValue DataToCompress = Op.getOperand(1);
+ SDValue PassThru = Op.getOperand(2);
+ if (isAllOnes(Mask)) // return data as is
+ return Op.getOperand(1);
+ EVT VT = Op.getValueType();
+ EVT MaskVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
+ VT.getVectorNumElements());
+ EVT BitcastVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
+ Mask.getValueType().getSizeInBits());
+ SDLoc dl(Op);
+ SDValue VMask = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MaskVT,
+ DAG.getNode(ISD::BITCAST, dl, BitcastVT, Mask),
+ DAG.getIntPtrConstant(0));
+
+ return DAG.getNode(IntrData->Opc0, dl, VT, VMask, DataToCompress,
+ PassThru);
+ }
+ case BLEND: {
+ SDValue Mask = Op.getOperand(3);
+ EVT VT = Op.getValueType();
+ EVT MaskVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
+ VT.getVectorNumElements());
+ EVT BitcastVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
+ Mask.getValueType().getSizeInBits());
+ SDLoc dl(Op);
+ SDValue VMask = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MaskVT,
+ DAG.getNode(ISD::BITCAST, dl, BitcastVT, Mask),
+ DAG.getIntPtrConstant(0));
+ return DAG.getNode(IntrData->Opc0, dl, VT, VMask, Op.getOperand(1),
+ Op.getOperand(2));
+ }
+ case FMA_OP_MASK:
+ {
+ return getVectorMaskingNode(DAG.getNode(IntrData->Opc0,
+ dl, Op.getValueType(),
+ Op.getOperand(1),
+ Op.getOperand(2),
+ Op.getOperand(3)),
+ Op.getOperand(4), Op.getOperand(1),
+ Subtarget, DAG);
+ }
default:
break;
}
Results.push_back(Store);
return DAG.getMergeValues(Results, dl);
}
+ case COMPRESS_TO_MEM: {
+ SDLoc dl(Op);
+ SDValue Mask = Op.getOperand(4);
+ SDValue DataToCompress = Op.getOperand(3);
+ SDValue Addr = Op.getOperand(2);
+ SDValue Chain = Op.getOperand(0);
+
+ if (isAllOnes(Mask)) // return just a store
+ return DAG.getStore(Chain, dl, DataToCompress, Addr,
+ MachinePointerInfo(), false, false, 0);
+
+ EVT VT = DataToCompress.getValueType();
+ EVT MaskVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
+ VT.getVectorNumElements());
+ EVT BitcastVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
+ Mask.getValueType().getSizeInBits());
+ SDValue VMask = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MaskVT,
+ DAG.getNode(ISD::BITCAST, dl, BitcastVT, Mask),
+ DAG.getIntPtrConstant(0));
+
+ SDValue Compressed = DAG.getNode(IntrData->Opc0, dl, VT, VMask,
+ DataToCompress, DAG.getUNDEF(VT));
+ return DAG.getStore(Chain, dl, Compressed, Addr,
+ MachinePointerInfo(), false, false, 0);
+ }
+ case EXPAND_FROM_MEM: {
+ SDLoc dl(Op);
+ SDValue Mask = Op.getOperand(4);
+ SDValue PathThru = Op.getOperand(3);
+ SDValue Addr = Op.getOperand(2);
+ SDValue Chain = Op.getOperand(0);
+ EVT VT = Op.getValueType();
+
+ if (isAllOnes(Mask)) // return just a load
+ return DAG.getLoad(VT, dl, Chain, Addr, MachinePointerInfo(), false, false,
+ false, 0);
+ EVT MaskVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
+ VT.getVectorNumElements());
+ EVT BitcastVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
+ Mask.getValueType().getSizeInBits());
+ SDValue VMask = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MaskVT,
+ DAG.getNode(ISD::BITCAST, dl, BitcastVT, Mask),
+ DAG.getIntPtrConstant(0));
+
+ SDValue DataToExpand = DAG.getLoad(VT, dl, Chain, Addr, MachinePointerInfo(),
+ false, false, false, 0);
+
+ SmallVector<SDValue, 2> Results;
+ Results.push_back(DAG.getNode(IntrData->Opc0, dl, VT, VMask, DataToExpand,
+ PathThru));
+ Results.push_back(Chain);
+ return DAG.getMergeValues(Results, dl);
+ }
}
}
return SDValue();
}
+static SDValue LowerCTPOP(SDValue Op, const X86Subtarget *Subtarget,
+ SelectionDAG &DAG) {
+ SDNode *Node = Op.getNode();
+ SDLoc dl(Node);
+
+ Op = Op.getOperand(0);
+ EVT VT = Op.getValueType();
+ assert((VT.is128BitVector() || VT.is256BitVector()) &&
+ "CTPOP lowering only implemented for 128/256-bit wide vector types");
+
+ unsigned NumElts = VT.getVectorNumElements();
+ EVT EltVT = VT.getVectorElementType();
+ unsigned Len = EltVT.getSizeInBits();
+
+ // This is the vectorized version of the "best" algorithm from
+ // http://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel
+ // with a minor tweak to use a series of adds + shifts instead of vector
+ // multiplications. Implemented for the v2i64, v4i64, v4i32, v8i32 types:
+ //
+ // v2i64, v4i64, v4i32 => Only profitable w/ popcnt disabled
+ // v8i32 => Always profitable
+ //
+ // FIXME: There a couple of possible improvements:
+ //
+ // 1) Support for i8 and i16 vectors (needs measurements if popcnt enabled).
+ // 2) Use strategies from http://wm.ite.pl/articles/sse-popcount.html
+ //
+ assert(EltVT.isInteger() && (Len == 32 || Len == 64) && Len % 8 == 0 &&
+ "CTPOP not implemented for this vector element type.");
+
+ // X86 canonicalize ANDs to vXi64, generate the appropriate bitcasts to avoid
+ // extra legalization.
+ bool NeedsBitcast = EltVT == MVT::i32;
+ MVT BitcastVT = VT.is256BitVector() ? MVT::v4i64 : MVT::v2i64;
+
+ SDValue Cst55 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x55)), EltVT);
+ SDValue Cst33 = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x33)), EltVT);
+ SDValue Cst0F = DAG.getConstant(APInt::getSplat(Len, APInt(8, 0x0F)), EltVT);
+
+ // v = v - ((v >> 1) & 0x55555555...)
+ SmallVector<SDValue, 8> Ones(NumElts, DAG.getConstant(1, EltVT));
+ SDValue OnesV = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Ones);
+ SDValue Srl = DAG.getNode(ISD::SRL, dl, VT, Op, OnesV);
+ if (NeedsBitcast)
+ Srl = DAG.getNode(ISD::BITCAST, dl, BitcastVT, Srl);
+
+ SmallVector<SDValue, 8> Mask55(NumElts, Cst55);
+ SDValue M55 = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Mask55);
+ if (NeedsBitcast)
+ M55 = DAG.getNode(ISD::BITCAST, dl, BitcastVT, M55);
+
+ SDValue And = DAG.getNode(ISD::AND, dl, Srl.getValueType(), Srl, M55);
+ if (VT != And.getValueType())
+ And = DAG.getNode(ISD::BITCAST, dl, VT, And);
+ SDValue Sub = DAG.getNode(ISD::SUB, dl, VT, Op, And);
+
+ // v = (v & 0x33333333...) + ((v >> 2) & 0x33333333...)
+ SmallVector<SDValue, 8> Mask33(NumElts, Cst33);
+ SDValue M33 = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Mask33);
+ SmallVector<SDValue, 8> Twos(NumElts, DAG.getConstant(2, EltVT));
+ SDValue TwosV = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Twos);
+
+ Srl = DAG.getNode(ISD::SRL, dl, VT, Sub, TwosV);
+ if (NeedsBitcast) {
+ Srl = DAG.getNode(ISD::BITCAST, dl, BitcastVT, Srl);
+ M33 = DAG.getNode(ISD::BITCAST, dl, BitcastVT, M33);
+ Sub = DAG.getNode(ISD::BITCAST, dl, BitcastVT, Sub);
+ }
+
+ SDValue AndRHS = DAG.getNode(ISD::AND, dl, M33.getValueType(), Srl, M33);
+ SDValue AndLHS = DAG.getNode(ISD::AND, dl, M33.getValueType(), Sub, M33);
+ if (VT != AndRHS.getValueType()) {
+ AndRHS = DAG.getNode(ISD::BITCAST, dl, VT, AndRHS);
+ AndLHS = DAG.getNode(ISD::BITCAST, dl, VT, AndLHS);
+ }
+ SDValue Add = DAG.getNode(ISD::ADD, dl, VT, AndLHS, AndRHS);
+
+ // v = (v + (v >> 4)) & 0x0F0F0F0F...
+ SmallVector<SDValue, 8> Fours(NumElts, DAG.getConstant(4, EltVT));
+ SDValue FoursV = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Fours);
+ Srl = DAG.getNode(ISD::SRL, dl, VT, Add, FoursV);
+ Add = DAG.getNode(ISD::ADD, dl, VT, Add, Srl);
+
+ SmallVector<SDValue, 8> Mask0F(NumElts, Cst0F);
+ SDValue M0F = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Mask0F);
+ if (NeedsBitcast) {
+ Add = DAG.getNode(ISD::BITCAST, dl, BitcastVT, Add);
+ M0F = DAG.getNode(ISD::BITCAST, dl, BitcastVT, M0F);
+ }
+ And = DAG.getNode(ISD::AND, dl, M0F.getValueType(), Add, M0F);
+ if (VT != And.getValueType())
+ And = DAG.getNode(ISD::BITCAST, dl, VT, And);
+
+ // The algorithm mentioned above uses:
+ // v = (v * 0x01010101...) >> (Len - 8)
+ //
+ // Change it to use vector adds + vector shifts which yield faster results on
+ // Haswell than using vector integer multiplication.
+ //
+ // For i32 elements:
+ // v = v + (v >> 8)
+ // v = v + (v >> 16)
+ //
+ // For i64 elements:
+ // v = v + (v >> 8)
+ // v = v + (v >> 16)
+ // v = v + (v >> 32)
+ //
+ Add = And;
+ SmallVector<SDValue, 8> Csts;
+ for (unsigned i = 8; i <= Len/2; i *= 2) {
+ Csts.assign(NumElts, DAG.getConstant(i, EltVT));
+ SDValue CstsV = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Csts);
+ Srl = DAG.getNode(ISD::SRL, dl, VT, Add, CstsV);
+ Add = DAG.getNode(ISD::ADD, dl, VT, Add, Srl);
+ Csts.clear();
+ }
+
+ // The result is on the least significant 6-bits on i32 and 7-bits on i64.
+ SDValue Cst3F = DAG.getConstant(APInt(Len, Len == 32 ? 0x3F : 0x7F), EltVT);
+ SmallVector<SDValue, 8> Cst3FV(NumElts, Cst3F);
+ SDValue M3F = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Cst3FV);
+ if (NeedsBitcast) {
+ Add = DAG.getNode(ISD::BITCAST, dl, BitcastVT, Add);
+ M3F = DAG.getNode(ISD::BITCAST, dl, BitcastVT, M3F);
+ }
+ And = DAG.getNode(ISD::AND, dl, M3F.getValueType(), Add, M3F);
+ if (VT != And.getValueType())
+ And = DAG.getNode(ISD::BITCAST, dl, VT, And);
+
+ return And;
+}
+
static SDValue LowerLOAD_SUB(SDValue Op, SelectionDAG &DAG) {
SDNode *Node = Op.getNode();
SDLoc dl(Node);
case ISD::ATOMIC_FENCE: return LowerATOMIC_FENCE(Op, Subtarget, DAG);
case ISD::ATOMIC_CMP_SWAP_WITH_SUCCESS:
return LowerCMP_SWAP(Op, Subtarget, DAG);
+ case ISD::CTPOP: return LowerCTPOP(Op, Subtarget, DAG);
case ISD::ATOMIC_LOAD_SUB: return LowerLOAD_SUB(Op,DAG);
case ISD::ATOMIC_STORE: return LowerATOMIC_STORE(Op,DAG);
case ISD::BUILD_VECTOR: return LowerBUILD_VECTOR(Op, DAG);
case X86ISD::PCMPESTRI: return "X86ISD::PCMPESTRI";
case X86ISD::PCMPISTRI: return "X86ISD::PCMPISTRI";
case X86ISD::XTEST: return "X86ISD::XTEST";
+ case X86ISD::COMPRESS: return "X86ISD::COMPRESS";
+ case X86ISD::EXPAND: return "X86ISD::EXPAND";
+ case X86ISD::SELECT: return "X86ISD::SELECT";
}
}
bool NeedSplit = false;
switch (VT.getSimpleVT().SimpleTy) {
default: return std::make_pair(0, false);
+ case MVT::v4i64:
+ case MVT::v2i64:
+ if (!Subtarget->hasVLX())
+ return std::make_pair(0, false);
+ break;
+ case MVT::v64i8:
+ case MVT::v32i16:
+ if (!Subtarget->hasBWI())
+ return std::make_pair(0, false);
+ break;
+ case MVT::v16i32:
+ case MVT::v8i64:
+ if (!Subtarget->hasAVX512())
+ return std::make_pair(0, false);
+ break;
case MVT::v32i8:
case MVT::v16i16:
case MVT::v8i32:
return SDValue();
}
-/// isHorizontalBinOp - Return 'true' if this vector operation is "horizontal"
+/// Return 'true' if this vector operation is "horizontal"
/// and return the operands for the horizontal operation in LHS and RHS. A
/// horizontal operation performs the binary operation on successive elements
/// of its first operand, then on successive elements of its second operand,
return true;
}
-/// PerformFADDCombine - Do target-specific dag combines on floating point adds.
+/// Do target-specific dag combines on floating point adds.
static SDValue PerformFADDCombine(SDNode *N, SelectionDAG &DAG,
const X86Subtarget *Subtarget) {
EVT VT = N->getValueType(0);
return SDValue();
}
-/// PerformFSUBCombine - Do target-specific dag combines on floating point subs.
+/// Do target-specific dag combines on floating point subs.
static SDValue PerformFSUBCombine(SDNode *N, SelectionDAG &DAG,
const X86Subtarget *Subtarget) {
EVT VT = N->getValueType(0);
return SDValue();
}
-/// PerformFORCombine - Do target-specific dag combines on X86ISD::FOR and
-/// X86ISD::FXOR nodes.
+/// Do target-specific dag combines on X86ISD::FOR and X86ISD::FXOR nodes.
static SDValue PerformFORCombine(SDNode *N, SelectionDAG &DAG) {
assert(N->getOpcode() == X86ISD::FOR || N->getOpcode() == X86ISD::FXOR);
// F[X]OR(0.0, x) -> x
return SDValue();
}
-/// PerformFMinFMaxCombine - Do target-specific dag combines on X86ISD::FMIN and
-/// X86ISD::FMAX nodes.
+/// Do target-specific dag combines on X86ISD::FMIN and X86ISD::FMAX nodes.
static SDValue PerformFMinFMaxCombine(SDNode *N, SelectionDAG &DAG) {
assert(N->getOpcode() == X86ISD::FMIN || N->getOpcode() == X86ISD::FMAX);
N->getOperand(0), N->getOperand(1));
}
-/// PerformFANDCombine - Do target-specific dag combines on X86ISD::FAND nodes.
+/// Do target-specific dag combines on X86ISD::FAND nodes.
static SDValue PerformFANDCombine(SDNode *N, SelectionDAG &DAG) {
// FAND(0.0, x) -> 0.0
// FAND(x, 0.0) -> 0.0
return SDValue();
}
-/// PerformFANDNCombine - Do target-specific dag combines on X86ISD::FANDN nodes
+/// Do target-specific dag combines on X86ISD::FANDN nodes
static SDValue PerformFANDNCombine(SDNode *N, SelectionDAG &DAG) {
// FANDN(x, 0.0) -> 0.0
// FANDN(0.0, x) -> x