"rather than promotion."),
cl::Hidden);
-// Forward declarations.
-static SDValue getMOVL(SelectionDAG &DAG, SDLoc dl, EVT VT, SDValue V1,
- SDValue V2);
-
X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
const X86Subtarget &STI)
: TargetLowering(TM), Subtarget(&STI) {
X86ScalarSSEf64 = Subtarget->hasSSE2();
X86ScalarSSEf32 = Subtarget->hasSSE1();
- TD = TM.getDataLayout();
+ MVT PtrVT = MVT::getIntegerVT(8 * TM.getPointerSize());
// Set up the TargetLowering object.
static const MVT IntVTs[] = { MVT::i8, MVT::i16, MVT::i32, MVT::i64 };
setLibcallCallingConv(RTLIB::SREM_I64, CallingConv::X86_StdCall);
setLibcallCallingConv(RTLIB::UREM_I64, CallingConv::X86_StdCall);
setLibcallCallingConv(RTLIB::MUL_I64, CallingConv::X86_StdCall);
-
- // The _ftol2 runtime function has an unusual calling conv, which
- // is modeled by a special pseudo-instruction.
- setLibcallName(RTLIB::FPTOUINT_F64_I64, nullptr);
- setLibcallName(RTLIB::FPTOUINT_F32_I64, nullptr);
- setLibcallName(RTLIB::FPTOUINT_F64_I32, nullptr);
- setLibcallName(RTLIB::FPTOUINT_F32_I32, nullptr);
}
if (Subtarget->isTargetDarwin()) {
setOperationAction(ISD::FP_TO_UINT , MVT::i16 , Promote);
if (Subtarget->is64Bit()) {
- setOperationAction(ISD::FP_TO_UINT , MVT::i64 , Expand);
- setOperationAction(ISD::FP_TO_UINT , MVT::i32 , Promote);
+ if (!Subtarget->useSoftFloat() && Subtarget->hasAVX512()) {
+ // FP_TO_UINT-i32/i64 is legal for f32/f64, but custom for f80.
+ setOperationAction(ISD::FP_TO_UINT , MVT::i32 , Custom);
+ setOperationAction(ISD::FP_TO_UINT , MVT::i64 , Custom);
+ } else {
+ setOperationAction(ISD::FP_TO_UINT , MVT::i32 , Promote);
+ setOperationAction(ISD::FP_TO_UINT , MVT::i64 , Expand);
+ }
} else if (!Subtarget->useSoftFloat()) {
// Since AVX is a superset of SSE3, only check for SSE here.
if (Subtarget->hasSSE1() && !Subtarget->hasSSE3())
// the optimal thing for SSE vs. the default expansion in the legalizer.
setOperationAction(ISD::FP_TO_UINT , MVT::i32 , Expand);
else
+ // With AVX512 we can use vcvts[ds]2usi for f32/f64->i32, f80 is custom.
// With SSE3 we can use fisttpll to convert to a signed i64; without
// SSE, we're stuck with a fistpll.
setOperationAction(ISD::FP_TO_UINT , MVT::i32 , Custom);
- }
- if (isTargetFTOL()) {
- // Use the _ftol2 runtime function, which has a pseudo-instruction
- // to handle its weird calling convention.
setOperationAction(ISD::FP_TO_UINT , MVT::i64 , Custom);
}
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i8 , Legal);
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1 , Expand);
setOperationAction(ISD::FP_ROUND_INREG , MVT::f32 , Expand);
- setOperationAction(ISD::FREM , MVT::f32 , Expand);
+
+ if (Subtarget->is32Bit() && Subtarget->isTargetKnownWindowsMSVC()) {
+ // On 32 bit MSVC, `fmodf(f32)` is not defined - only `fmod(f64)`
+ // is. We should promote the value to 64-bits to solve this.
+ // This is what the CRT headers do - `fmodf` is an inline header
+ // function casting to f64 and calling `fmod`.
+ setOperationAction(ISD::FREM , MVT::f32 , Promote);
+ } else {
+ setOperationAction(ISD::FREM , MVT::f32 , Expand);
+ }
+
setOperationAction(ISD::FREM , MVT::f64 , Expand);
setOperationAction(ISD::FREM , MVT::f80 , Expand);
setOperationAction(ISD::FLT_ROUNDS_ , MVT::i32 , Custom);
setOperationAction(ISD::SETCC , MVT::i64 , Custom);
}
setOperationAction(ISD::EH_RETURN , MVT::Other, Custom);
+ setOperationAction(ISD::CATCHRET , MVT::Other, Custom);
// NOTE: EH_SJLJ_SETJMP/_LONGJMP supported here is NOT intended to support
// SjLj exception handling but a light-weight setjmp/longjmp replacement to
// support continuation, user-level threading, and etc.. As a result, no
setOperationAction(ISD::EH_LABEL, MVT::Other, Expand);
}
- if (Subtarget->is64Bit()) {
+ if (Subtarget->isTarget64BitLP64()) {
setExceptionPointerRegister(X86::RAX);
setExceptionSelectorRegister(X86::RDX);
} else {
// VASTART needs to be custom lowered to use the VarArgsFrameIndex
setOperationAction(ISD::VASTART , MVT::Other, Custom);
setOperationAction(ISD::VAEND , MVT::Other, Expand);
- if (Subtarget->is64Bit() && !Subtarget->isTargetWin64()) {
- // TargetInfo::X86_64ABIBuiltinVaList
+ if (Subtarget->is64Bit()) {
setOperationAction(ISD::VAARG , MVT::Other, Custom);
setOperationAction(ISD::VACOPY , MVT::Other, Custom);
} else {
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
- setOperationAction(ISD::DYNAMIC_STACKALLOC, getPointerTy(*TD), Custom);
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, PtrVT, Custom);
// GC_TRANSITION_START and GC_TRANSITION_END need custom lowering.
setOperationAction(ISD::GC_TRANSITION_START, MVT::Other, Custom);
setOperationAction(ISD::MUL, MVT::v8i32, Custom);
setOperationAction(ISD::MUL, MVT::v16i16, Custom);
setOperationAction(ISD::MUL, MVT::v32i8, Custom);
+
+ setOperationAction(ISD::SMAX, MVT::v32i8, Custom);
+ setOperationAction(ISD::SMAX, MVT::v16i16, Custom);
+ setOperationAction(ISD::SMAX, MVT::v8i32, Custom);
+ setOperationAction(ISD::UMAX, MVT::v32i8, Custom);
+ setOperationAction(ISD::UMAX, MVT::v16i16, Custom);
+ setOperationAction(ISD::UMAX, MVT::v8i32, Custom);
+ setOperationAction(ISD::SMIN, MVT::v32i8, Custom);
+ setOperationAction(ISD::SMIN, MVT::v16i16, Custom);
+ setOperationAction(ISD::SMIN, MVT::v8i32, Custom);
+ setOperationAction(ISD::UMIN, MVT::v32i8, Custom);
+ setOperationAction(ISD::UMIN, MVT::v16i16, Custom);
+ setOperationAction(ISD::UMIN, MVT::v8i32, Custom);
}
// In the customized shift lowering, the legal cases in AVX2 will be
setOperationAction(ISD::FMA, MVT::v8f64, Legal);
setOperationAction(ISD::FMA, MVT::v16f32, Legal);
- setOperationAction(ISD::FP_TO_SINT, MVT::i32, Legal);
- setOperationAction(ISD::FP_TO_UINT, MVT::i32, Legal);
+ // FIXME: [US]INT_TO_FP are not legal for f80.
setOperationAction(ISD::SINT_TO_FP, MVT::i32, Legal);
setOperationAction(ISD::UINT_TO_FP, MVT::i32, Legal);
if (Subtarget->is64Bit()) {
- setOperationAction(ISD::FP_TO_UINT, MVT::i64, Legal);
- setOperationAction(ISD::FP_TO_SINT, MVT::i64, Legal);
setOperationAction(ISD::SINT_TO_FP, MVT::i64, Legal);
setOperationAction(ISD::UINT_TO_FP, MVT::i64, Legal);
}
setOperationAction(ISD::FP_ROUND, MVT::v8f32, Legal);
setOperationAction(ISD::FP_EXTEND, MVT::v8f32, Legal);
+ setTruncStoreAction(MVT::v8i64, MVT::v8i8, Legal);
+ setTruncStoreAction(MVT::v8i64, MVT::v8i16, Legal);
+ setTruncStoreAction(MVT::v8i64, MVT::v8i32, Legal);
+ setTruncStoreAction(MVT::v16i32, MVT::v16i8, Legal);
+ setTruncStoreAction(MVT::v16i32, MVT::v16i16, Legal);
+ if (Subtarget->hasVLX()){
+ setTruncStoreAction(MVT::v4i64, MVT::v4i8, Legal);
+ setTruncStoreAction(MVT::v4i64, MVT::v4i16, Legal);
+ setTruncStoreAction(MVT::v4i64, MVT::v4i32, Legal);
+ setTruncStoreAction(MVT::v8i32, MVT::v8i8, Legal);
+ setTruncStoreAction(MVT::v8i32, MVT::v8i16, Legal);
+
+ setTruncStoreAction(MVT::v2i64, MVT::v2i8, Legal);
+ setTruncStoreAction(MVT::v2i64, MVT::v2i16, Legal);
+ setTruncStoreAction(MVT::v2i64, MVT::v2i32, Legal);
+ setTruncStoreAction(MVT::v4i32, MVT::v4i8, Legal);
+ setTruncStoreAction(MVT::v4i32, MVT::v4i16, Legal);
+ }
setOperationAction(ISD::TRUNCATE, MVT::i1, Custom);
setOperationAction(ISD::TRUNCATE, MVT::v16i8, Custom);
setOperationAction(ISD::TRUNCATE, MVT::v8i32, Custom);
setOperationAction(ISD::VSELECT, MVT::v64i8, Legal);
setOperationAction(ISD::TRUNCATE, MVT::v32i1, Custom);
setOperationAction(ISD::TRUNCATE, MVT::v64i1, Custom);
+ setOperationAction(ISD::TRUNCATE, MVT::v32i8, Custom);
setOperationAction(ISD::SMAX, MVT::v64i8, Legal);
setOperationAction(ISD::SMAX, MVT::v32i16, Legal);
setOperationAction(ISD::UMIN, MVT::v64i8, Legal);
setOperationAction(ISD::UMIN, MVT::v32i16, Legal);
+ setTruncStoreAction(MVT::v32i16, MVT::v32i8, Legal);
+ setTruncStoreAction(MVT::v16i16, MVT::v16i8, Legal);
+ if (Subtarget->hasVLX())
+ setTruncStoreAction(MVT::v8i16, MVT::v8i8, Legal);
+
for (int i = MVT::v32i8; i != MVT::v8i64; ++i) {
const MVT VT = (MVT::SimpleValueType)i;
setTargetDAGCombine(ISD::SINT_TO_FP);
setTargetDAGCombine(ISD::UINT_TO_FP);
setTargetDAGCombine(ISD::SETCC);
- setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
setTargetDAGCombine(ISD::BUILD_VECTOR);
setTargetDAGCombine(ISD::MUL);
setTargetDAGCombine(ISD::XOR);
computeRegisterProperties(Subtarget->getRegisterInfo());
- // On Darwin, -Os means optimize for size without hurting performance,
- // do not reduce the limit.
MaxStoresPerMemset = 16; // For @llvm.memset -> sequence of stores
- MaxStoresPerMemsetOptSize = Subtarget->isTargetDarwin() ? 16 : 8;
+ MaxStoresPerMemsetOptSize = 8;
MaxStoresPerMemcpy = 8; // For @llvm.memcpy -> sequence of stores
- MaxStoresPerMemcpyOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
+ MaxStoresPerMemcpyOptSize = 4;
MaxStoresPerMemmove = 8; // For @llvm.memmove -> sequence of stores
- MaxStoresPerMemmoveOptSize = Subtarget->isTargetDarwin() ? 8 : 4;
+ MaxStoresPerMemmoveOptSize = 4;
setPrefLoopAlignment(4); // 2^4 bytes.
// Predictable cmov don't hurt on atom because it's in-order.
if (EltAlign > MaxAlign)
MaxAlign = EltAlign;
} else if (StructType *STy = dyn_cast<StructType>(Ty)) {
- for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
+ for (auto *EltTy : STy->elements()) {
unsigned EltAlign = 0;
- getMaxByValAlign(STy->getElementType(i), EltAlign);
+ getMaxByValAlign(EltTy, EltAlign);
if (EltAlign > MaxAlign)
MaxAlign = EltAlign;
if (MaxAlign == 16)
if ((!IsMemset || ZeroMemset) &&
!F->hasFnAttribute(Attribute::NoImplicitFloat)) {
if (Size >= 16 &&
- (Subtarget->isUnalignedMemAccessFast() ||
+ (!Subtarget->isUnalignedMem16Slow() ||
((DstAlign == 0 || DstAlign >= 16) &&
(SrcAlign == 0 || SrcAlign >= 16)))) {
if (Size >= 32) {
+ // FIXME: Check if unaligned 32-byte accesses are slow.
if (Subtarget->hasInt256())
return MVT::v8i32;
if (Subtarget->hasFp256())
return MVT::f64;
}
}
+ // This is a compromise. If we reach here, unaligned accesses may be slow on
+ // this target. However, creating smaller, aligned accesses could be even
+ // slower and would certainly be a lot more code.
if (Subtarget->is64Bit() && Size >= 8)
return MVT::i64;
return MVT::i32;
unsigned,
unsigned,
bool *Fast) const {
- if (Fast)
- *Fast = Subtarget->isUnalignedMemAccessFast();
+ if (Fast) {
+ if (VT.getSizeInBits() == 256)
+ *Fast = !Subtarget->isUnalignedMem32Slow();
+ else
+ // FIXME: We should always return that 8-byte and under accesses are fast.
+ // That is what other x86 lowering code assumes.
+ *Fast = !Subtarget->isUnalignedMem16Slow();
+ }
return true;
}
int FI = MFI->CreateFixedObject(ValVT.getSizeInBits()/8,
VA.getLocMemOffset(), isImmutable);
SDValue FIN = DAG.getFrameIndex(FI, getPointerTy(DAG.getDataLayout()));
- SDValue Val = DAG.getLoad(ValVT, dl, Chain, FIN,
- MachinePointerInfo::getFixedStack(FI),
- false, false, false, 0);
+ SDValue Val = DAG.getLoad(
+ ValVT, dl, Chain, FIN,
+ MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI), false,
+ false, false, 0);
return ExtendedInMem ?
DAG.getNode(ISD::TRUNCATE, dl, VA.getValVT(), Val) : Val;
}
SDValue FIN = DAG.getNode(ISD::ADD, dl, getPointerTy(DAG.getDataLayout()),
RSFIN, DAG.getIntPtrConstant(Offset, dl));
SDValue Store =
- DAG.getStore(Val.getValue(1), dl, Val, FIN,
- MachinePointerInfo::getFixedStack(
- FuncInfo->getRegSaveFrameIndex(), Offset),
- false, false, 0);
+ DAG.getStore(Val.getValue(1), dl, Val, FIN,
+ MachinePointerInfo::getFixedStack(
+ DAG.getMachineFunction(),
+ FuncInfo->getRegSaveFrameIndex(), Offset),
+ false, false, 0);
MemOps.push_back(Store);
Offset += 8;
}
SDValue Val = DAG.getCopyFromReg(Chain, dl, GPR, MVT::i64);
Chain = DAG.getStore(
Val.getValue(1), dl, Val, RSFIN,
- MachinePointerInfo::getFixedStack(FuncInfo->getRegSaveFrameIndex()),
+ MachinePointerInfo::getFixedStack(DAG.getMachineFunction(),
+ FuncInfo->getRegSaveFrameIndex()),
/*isVolatile=*/true, /*isNonTemporal=*/false, /*Alignment=*/0);
}
MMI.getWinEHFuncInfo(MF.getFunction()).UnwindHelpFrameIdx = UnwindHelpFI;
SDValue Neg2 = DAG.getConstant(-2, dl, MVT::i64);
Chain = DAG.getStore(Chain, dl, Neg2, StackSlot,
- MachinePointerInfo::getFixedStack(UnwindHelpFI),
+ MachinePointerInfo::getFixedStack(
+ DAG.getMachineFunction(), UnwindHelpFI),
/*isVolatile=*/true,
/*isNonTemporal=*/false, /*Alignment=*/0);
} else {
if (Flags.isByVal())
return CreateCopyOfByValArgument(Arg, PtrOff, Chain, Flags, DAG, dl);
- return DAG.getStore(Chain, dl, Arg, PtrOff,
- MachinePointerInfo::getStack(LocMemOffset),
- false, false, 0);
+ return DAG.getStore(
+ Chain, dl, Arg, PtrOff,
+ MachinePointerInfo::getStack(DAG.getMachineFunction(), LocMemOffset),
+ false, false, 0);
}
/// Emit a load of return address if tail call
false);
SDValue NewRetAddrFrIdx = DAG.getFrameIndex(NewReturnAddrFI, PtrVT);
Chain = DAG.getStore(Chain, dl, RetAddrFrIdx, NewRetAddrFrIdx,
- MachinePointerInfo::getFixedStack(NewReturnAddrFI),
+ MachinePointerInfo::getFixedStack(
+ DAG.getMachineFunction(), NewReturnAddrFI),
false, false, 0);
return Chain;
}
+/// Returns a vector_shuffle mask for an movs{s|d}, movd
+/// operation of specified width.
+static SDValue getMOVL(SelectionDAG &DAG, SDLoc dl, EVT VT, SDValue V1,
+ SDValue V2) {
+ unsigned NumElems = VT.getVectorNumElements();
+ SmallVector<int, 8> Mask;
+ Mask.push_back(NumElems);
+ for (unsigned i = 1; i != NumElems; ++i)
+ Mask.push_back(i);
+ return DAG.getVectorShuffle(VT, dl, V1, V2, &Mask[0]);
+}
+
SDValue
X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
SmallVectorImpl<SDValue> &InVals) const {
// Store the argument.
SDValue SpillSlot = DAG.CreateStackTemporary(VA.getValVT());
int FI = cast<FrameIndexSDNode>(SpillSlot)->getIndex();
- Chain = DAG.getStore(Chain, dl, Arg, SpillSlot,
- MachinePointerInfo::getFixedStack(FI),
- false, false, 0);
+ Chain = DAG.getStore(
+ Chain, dl, Arg, SpillSlot,
+ MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI),
+ false, false, 0);
Arg = SpillSlot;
break;
}
Flags, DAG, dl));
} else {
// Store relative to framepointer.
- MemOpChains2.push_back(
- DAG.getStore(ArgChain, dl, Arg, FIN,
- MachinePointerInfo::getFixedStack(FI),
- false, false, 0));
+ MemOpChains2.push_back(DAG.getStore(
+ ArgChain, dl, Arg, FIN,
+ MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), FI),
+ false, false, 0));
}
}
if (ExtraLoad)
Callee = DAG.getLoad(
getPointerTy(DAG.getDataLayout()), dl, DAG.getEntryNode(), Callee,
- MachinePointerInfo::getGOT(), false, false, false, 0);
+ MachinePointerInfo::getGOT(DAG.getMachineFunction()), false, false,
+ false, 0);
}
} else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
unsigned char OpFlags = 0;
// EDI
// local1 ..
-/// GetAlignedArgumentStackSize - Make the stack size align e.g 16n + 12 aligned
-/// for a 16 byte align requirement.
+/// Make the stack size align e.g 16n + 12 aligned for a 16-byte align
+/// requirement.
unsigned
X86TargetLowering::GetAlignedArgumentStackSize(unsigned StackSize,
SelectionDAG& DAG) const {
return Offset;
}
-/// MatchingStackOffset - Return true if the given stack call argument is
-/// already available in the same position (relatively) of the caller's
-/// incoming argument stack.
+/// Return true if the given stack call argument is already available in the
+/// same position (relatively) of the caller's incoming argument stack.
static
bool MatchingStackOffset(SDValue Arg, unsigned Offset, ISD::ArgFlagsTy Flags,
MachineFrameInfo *MFI, const MachineRegisterInfo *MRI,
return Offset == MFI->getObjectOffset(FI) && Bytes == MFI->getObjectSize(FI);
}
-/// IsEligibleForTailCallOptimization - Check whether the call is eligible
-/// for tail call optimization. Targets which want to do tail call
-/// optimization should implement this function.
+/// Check whether the call is eligible for tail call optimization. Targets
+/// that want to do tail call optimization should implement this function.
bool
X86TargetLowering::IsEligibleForTailCallOptimization(SDValue Callee,
CallingConv::ID CalleeCC,
return false;
}
-/// isCalleePop - Determines whether the callee is required to pop its
-/// own arguments. Callee pop is necessary to support tail calls.
+/// Determines whether the callee is required to pop its own arguments.
+/// Callee pop is necessary to support tail calls.
bool X86::isCalleePop(CallingConv::ID CallingConv,
bool is64Bit, bool IsVarArg, bool TailCallOpt) {
switch (CallingConv) {
llvm_unreachable("covered switch fell through?!");
}
-/// TranslateX86CC - do a one to one translation of a ISD::CondCode to the X86
-/// specific condition code, returning the condition code and the LHS/RHS of the
+/// Do a one-to-one translation of a ISD::CondCode to the X86-specific
+/// condition code, returning the condition code and the LHS/RHS of the
/// comparison to make.
static unsigned TranslateX86CC(ISD::CondCode SetCCOpcode, SDLoc DL, bool isFP,
SDValue &LHS, SDValue &RHS, SelectionDAG &DAG) {
}
}
-/// hasFPCMov - is there a floating point cmov for the specific X86 condition
-/// code. Current x86 isa includes the following FP cmov instructions:
+/// Is there a floating point cmov for the specific X86 condition code?
+/// Current x86 isa includes the following FP cmov instructions:
/// fcmovb, fcomvbe, fcomve, fcmovu, fcmovae, fcmova, fcmovne, fcmovnu.
static bool hasFPCMov(unsigned X86CC) {
switch (X86CC) {
}
}
-/// isFPImmLegal - Returns true if the target can instruction select the
+/// Returns true if the target can instruction select the
/// specified FP immediate natively. If false, the legalizer will
/// materialize the FP immediate as a load from a constant pool.
bool X86TargetLowering::isFPImmLegal(const APFloat &Imm, EVT VT) const {
return Subtarget->hasLZCNT();
}
-/// isUndefInRange - Return true if every element in Mask, beginning
+/// Return true if every element in Mask, beginning
/// from position Pos and ending in Pos+Size is undef.
static bool isUndefInRange(ArrayRef<int> Mask, unsigned Pos, unsigned Size) {
for (unsigned i = Pos, e = Pos + Size; i != e; ++i)
return true;
}
-/// isUndefOrInRange - Return true if Val is undef or if its value falls within
-/// the specified range (L, H].
+/// 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) {
return (Val < 0) || (Val >= Low && Val < Hi);
}
-/// isUndefOrEqual - Val is either less than zero (undef) or equal to the
-/// specified value.
+/// Val is either less than zero (undef) or equal to the specified value.
static bool isUndefOrEqual(int Val, int CmpVal) {
return (Val < 0 || Val == CmpVal);
}
-/// isSequentialOrUndefInRange - Return true if every element in Mask, beginning
+/// Return true if every element in Mask, beginning
/// from position Pos and ending in Pos+Size, falls within the specified
/// sequential range (Low, Low+Size]. or is undef.
static bool isSequentialOrUndefInRange(ArrayRef<int> Mask,
return true;
}
-/// isVEXTRACTIndex - Return true if the specified
-/// EXTRACT_SUBVECTOR operand specifies a vector extract that is
-/// suitable for instruction that extract 128 or 256 bit vectors
+/// Return true if the specified EXTRACT_SUBVECTOR operand specifies a vector
+/// extract that is suitable for instruction that extract 128 or 256 bit vectors
static bool isVEXTRACTIndex(SDNode *N, unsigned vecWidth) {
assert((vecWidth == 128 || vecWidth == 256) && "Unexpected vector width");
if (!isa<ConstantSDNode>(N->getOperand(1).getNode()))
return Result;
}
-/// isVINSERTIndex - Return true if the specified INSERT_SUBVECTOR
+/// Return true if the specified INSERT_SUBVECTOR
/// operand specifies a subvector insert that is suitable for input to
/// insertion of 128 or 256-bit subvectors
static bool isVINSERTIndex(SDNode *N, unsigned vecWidth) {
return Index / NumElemsPerChunk;
}
-/// getExtractVEXTRACT128Immediate - Return the appropriate immediate
-/// to extract the specified EXTRACT_SUBVECTOR index with VEXTRACTF128
-/// and VINSERTI128 instructions.
+/// Return the appropriate immediate to extract the specified
+/// EXTRACT_SUBVECTOR index with VEXTRACTF128 and VINSERTI128 instructions.
unsigned X86::getExtractVEXTRACT128Immediate(SDNode *N) {
return getExtractVEXTRACTImmediate(N, 128);
}
-/// getExtractVEXTRACT256Immediate - Return the appropriate immediate
-/// to extract the specified EXTRACT_SUBVECTOR index with VEXTRACTF64x4
-/// and VINSERTI64x4 instructions.
+/// Return the appropriate immediate to extract the specified
+/// EXTRACT_SUBVECTOR index with VEXTRACTF64x4 and VINSERTI64x4 instructions.
unsigned X86::getExtractVEXTRACT256Immediate(SDNode *N) {
return getExtractVEXTRACTImmediate(N, 256);
}
-/// getInsertVINSERT128Immediate - Return the appropriate immediate
-/// to insert at the specified INSERT_SUBVECTOR index with VINSERTF128
-/// and VINSERTI128 instructions.
+/// Return the appropriate immediate to insert at the specified
+/// INSERT_SUBVECTOR index with VINSERTF128 and VINSERTI128 instructions.
unsigned X86::getInsertVINSERT128Immediate(SDNode *N) {
return getInsertVINSERTImmediate(N, 128);
}
-/// getInsertVINSERT256Immediate - Return the appropriate immediate
-/// to insert at the specified INSERT_SUBVECTOR index with VINSERTF46x4
-/// and VINSERTI64x4 instructions.
+/// Return the appropriate immediate to insert at the specified
+/// INSERT_SUBVECTOR index with VINSERTF46x4 and VINSERTI64x4 instructions.
unsigned X86::getInsertVINSERT256Immediate(SDNode *N) {
return getInsertVINSERTImmediate(N, 256);
}
-/// isZero - Returns true if Elt is a constant integer zero
+/// Returns true if Elt is a constant integer zero
static bool isZero(SDValue V) {
ConstantSDNode *C = dyn_cast<ConstantSDNode>(V);
return C && C->isNullValue();
}
-/// isZeroNode - Returns true if Elt is a constant zero or a floating point
-/// constant +0.0.
+/// Returns true if Elt is a constant zero or a floating point constant +0.0.
bool X86::isZeroNode(SDValue Elt) {
if (isZero(Elt))
return true;
return false;
}
-/// getZeroVector - Returns a vector of specified type with all zero elements.
-///
+/// Returns a vector of specified type with all zero elements.
static SDValue getZeroVector(EVT VT, const X86Subtarget *Subtarget,
SelectionDAG &DAG, SDLoc dl) {
assert(VT.isVector() && "Expected a vector type");
return Insert256BitVector(V, V2, NumElems/2, DAG, dl);
}
-/// getOnesVector - Returns a vector of specified type with all bits set.
+/// Returns a vector of specified type with all bits set.
/// Always build ones vectors as <4 x i32> or <8 x i32>. For 256-bit types with
/// no AVX2 supprt, use two <4 x i32> inserted in a <8 x i32> appropriately.
/// Then bitcast to their original type, ensuring they get CSE'd.
return DAG.getBitcast(VT, Vec);
}
-/// getMOVLMask - Returns a vector_shuffle mask for an movs{s|d}, movd
-/// operation of specified width.
-static SDValue getMOVL(SelectionDAG &DAG, SDLoc dl, EVT VT, SDValue V1,
- SDValue V2) {
- unsigned NumElems = VT.getVectorNumElements();
- SmallVector<int, 8> Mask;
- Mask.push_back(NumElems);
- for (unsigned i = 1; i != NumElems; ++i)
- Mask.push_back(i);
- return DAG.getVectorShuffle(VT, dl, V1, V2, &Mask[0]);
-}
-
-/// getUnpackl - Returns a vector_shuffle node for an unpackl operation.
+/// Returns a vector_shuffle node for an unpackl operation.
static SDValue getUnpackl(SelectionDAG &DAG, SDLoc dl, MVT VT, SDValue V1,
SDValue V2) {
unsigned NumElems = VT.getVectorNumElements();
return DAG.getVectorShuffle(VT, dl, V1, V2, &Mask[0]);
}
-/// getUnpackh - Returns a vector_shuffle node for an unpackh operation.
+/// Returns a vector_shuffle node for an unpackh operation.
static SDValue getUnpackh(SelectionDAG &DAG, SDLoc dl, MVT VT, SDValue V1,
SDValue V2) {
unsigned NumElems = VT.getVectorNumElements();
return DAG.getVectorShuffle(VT, dl, V1, V2, &Mask[0]);
}
-/// getShuffleVectorZeroOrUndef - Return a vector_shuffle of the specified
-/// vector of zero or undef vector. This produces a shuffle where the low
-/// element of V2 is swizzled into the zero/undef vector, landing at element
-/// Idx. This produces a shuffle mask like 4,1,2,3 (idx=0) or 0,1,2,4 (idx=3).
+/// Return a vector_shuffle of the specified vector of zero or undef vector.
+/// This produces a shuffle where the low element of V2 is swizzled into the
+/// zero/undef vector, landing at element Idx.
+/// This produces a shuffle mask like 4,1,2,3 (idx=0) or 0,1,2,4 (idx=3).
static SDValue getShuffleVectorZeroOrUndef(SDValue V2, unsigned Idx,
bool IsZero,
const X86Subtarget *Subtarget,
return DAG.getVectorShuffle(VT, SDLoc(V2), V1, V2, &MaskVec[0]);
}
-/// getTargetShuffleMask - Calculates the shuffle mask corresponding to the
-/// target specific opcode. Returns true if the Mask could be calculated. Sets
-/// IsUnary to true if only uses one source. Note that this will set IsUnary for
-/// shuffles which use a single input multiple times, and in those cases it will
+/// Calculates the shuffle mask corresponding to the target-specific opcode.
+/// Returns true if the Mask could be calculated. Sets IsUnary to true if only
+/// uses one source. Note that this will set IsUnary for shuffles which use a
+/// single input multiple times, and in those cases it will
/// adjust the mask to only have indices within that single input.
/// FIXME: Add support for Decode*Mask functions that return SM_SentinelZero.
static bool getTargetShuffleMask(SDNode *N, MVT VT,
return true;
}
-/// getShuffleScalarElt - Returns the scalar element that will make up the ith
+/// Returns the scalar element that will make up the ith
/// element of the result of the vector shuffle.
static SDValue getShuffleScalarElt(SDNode *N, unsigned Index, SelectionDAG &DAG,
unsigned Depth) {
return SDValue();
}
-/// LowerBuildVectorv16i8 - Custom lower build_vector of v16i8.
-///
+/// Custom lower build_vector of v16i8.
static SDValue LowerBuildVectorv16i8(SDValue Op, unsigned NonZeros,
unsigned NumNonZero, unsigned NumZero,
SelectionDAG &DAG,
return DAG.getBitcast(MVT::v16i8, V);
}
-/// LowerBuildVectorv8i16 - Custom lower build_vector of v8i16.
-///
+/// Custom lower build_vector of v8i16.
static SDValue LowerBuildVectorv8i16(SDValue Op, unsigned NonZeros,
unsigned NumNonZero, unsigned NumZero,
SelectionDAG &DAG,
return V;
}
-/// LowerBuildVectorv4x32 - Custom lower build_vector of v4i32 or v4f32.
+/// Custom lower build_vector of v4i32 or v4f32.
static SDValue LowerBuildVectorv4x32(SDValue Op, SelectionDAG &DAG,
const X86Subtarget *Subtarget,
const TargetLowering &TLI) {
return SDValue();
if ((Offset % RequiredAlign) & 3)
return SDValue();
- int64_t StartOffset = Offset & ~(RequiredAlign-1);
+ int64_t StartOffset = Offset & ~int64_t(RequiredAlign - 1);
if (StartOffset) {
SDLoc DL(Ptr);
Ptr = DAG.getNode(ISD::ADD, DL, Ptr.getValueType(), Ptr,
// TODO: If multiple splats are generated to load the same constant,
// it may be detrimental to overall size. There needs to be a way to detect
// that condition to know if this is truly a size win.
- const Function *F = DAG.getMachineFunction().getFunction();
- bool OptForSize = F->hasFnAttribute(Attribute::OptimizeForSize);
+ bool OptForSize = DAG.getMachineFunction().getFunction()->optForSize();
// Handle broadcasting a single constant scalar from the constant pool
// into a vector.
SDValue CP =
DAG.getConstantPool(C, TLI.getPointerTy(DAG.getDataLayout()));
unsigned Alignment = cast<ConstantPoolSDNode>(CP)->getAlignment();
- Ld = DAG.getLoad(CVT, dl, DAG.getEntryNode(), CP,
- MachinePointerInfo::getConstantPool(),
- false, false, false, Alignment);
+ Ld = DAG.getLoad(
+ CVT, dl, DAG.getEntryNode(), CP,
+ MachinePointerInfo::getConstantPool(DAG.getMachineFunction()), false,
+ false, false, Alignment);
return DAG.getNode(X86ISD::VBROADCAST, dl, VT, Ld);
}
return SDValue();
}
-// LowerAVXCONCAT_VECTORS - 256-bit AVX can use the vinsertf128 instruction
+// 256-bit AVX can use the vinsertf128 instruction
// to create 256-bit vectors from two other 128-bit ones.
static SDValue LowerAVXCONCAT_VECTORS(SDValue Op, SelectionDAG &DAG) {
SDLoc dl(Op);
return DAG.getConstant(Imm, DL, MVT::i8);
}
+/// \brief Compute whether each element of a shuffle is zeroable.
+///
+/// A "zeroable" vector shuffle element is one which can be lowered to zero.
+/// Either it is an undef element in the shuffle mask, the element of the input
+/// referenced is undef, or the element of the input referenced is known to be
+/// zero. Many x86 shuffles can zero lanes cheaply and we often want to handle
+/// as many lanes with this technique as possible to simplify the remaining
+/// shuffle.
+static SmallBitVector computeZeroableShuffleElements(ArrayRef<int> Mask,
+ SDValue V1, SDValue V2) {
+ SmallBitVector Zeroable(Mask.size(), false);
+
+ while (V1.getOpcode() == ISD::BITCAST)
+ V1 = V1->getOperand(0);
+ while (V2.getOpcode() == ISD::BITCAST)
+ V2 = V2->getOperand(0);
+
+ bool V1IsZero = ISD::isBuildVectorAllZeros(V1.getNode());
+ bool V2IsZero = ISD::isBuildVectorAllZeros(V2.getNode());
+
+ for (int i = 0, Size = Mask.size(); i < Size; ++i) {
+ int M = Mask[i];
+ // Handle the easy cases.
+ if (M < 0 || (M >= 0 && M < Size && V1IsZero) || (M >= Size && V2IsZero)) {
+ Zeroable[i] = true;
+ continue;
+ }
+
+ // If this is an index into a build_vector node (which has the same number
+ // of elements), dig out the input value and use it.
+ SDValue V = M < Size ? V1 : V2;
+ if (V.getOpcode() != ISD::BUILD_VECTOR || Size != (int)V.getNumOperands())
+ continue;
+
+ SDValue Input = V.getOperand(M % Size);
+ // The UNDEF opcode check really should be dead code here, but not quite
+ // worth asserting on (it isn't invalid, just unexpected).
+ if (Input.getOpcode() == ISD::UNDEF || X86::isZeroNode(Input))
+ Zeroable[i] = true;
+ }
+
+ return Zeroable;
+}
+
+/// \brief Try to emit a bitmask instruction for a shuffle.
+///
+/// This handles cases where we can model a blend exactly as a bitmask due to
+/// one of the inputs being zeroable.
+static SDValue lowerVectorShuffleAsBitMask(SDLoc DL, MVT VT, SDValue V1,
+ SDValue V2, ArrayRef<int> Mask,
+ SelectionDAG &DAG) {
+ MVT EltVT = VT.getScalarType();
+ int NumEltBits = EltVT.getSizeInBits();
+ MVT IntEltVT = MVT::getIntegerVT(NumEltBits);
+ SDValue Zero = DAG.getConstant(0, DL, IntEltVT);
+ SDValue AllOnes = DAG.getConstant(APInt::getAllOnesValue(NumEltBits), DL,
+ IntEltVT);
+ if (EltVT.isFloatingPoint()) {
+ Zero = DAG.getBitcast(EltVT, Zero);
+ AllOnes = DAG.getBitcast(EltVT, AllOnes);
+ }
+ SmallVector<SDValue, 16> VMaskOps(Mask.size(), Zero);
+ SmallBitVector Zeroable = computeZeroableShuffleElements(Mask, V1, V2);
+ SDValue V;
+ for (int i = 0, Size = Mask.size(); i < Size; ++i) {
+ if (Zeroable[i])
+ continue;
+ if (Mask[i] % Size != i)
+ return SDValue(); // Not a blend.
+ if (!V)
+ V = Mask[i] < Size ? V1 : V2;
+ else if (V != (Mask[i] < Size ? V1 : V2))
+ return SDValue(); // Can only let one input through the mask.
+
+ VMaskOps[i] = AllOnes;
+ }
+ if (!V)
+ return SDValue(); // No non-zeroable elements!
+
+ SDValue VMask = DAG.getNode(ISD::BUILD_VECTOR, DL, VT, VMaskOps);
+ V = DAG.getNode(VT.isFloatingPoint()
+ ? (unsigned) X86ISD::FAND : (unsigned) ISD::AND,
+ DL, VT, V, VMask);
+ return V;
+}
+
/// \brief Try to emit a blend instruction for a shuffle using bit math.
///
/// This is used as a fallback approach when first class blend instructions are
assert((VT.getSizeInBits() == 128 || Subtarget->hasAVX2()) &&
"256-bit byte-blends require AVX2 support!");
+ // Attempt to lower to a bitmask if we can. VPAND is faster than VPBLENDVB.
+ if (SDValue Masked = lowerVectorShuffleAsBitMask(DL, VT, V1, V2, Mask, DAG))
+ return Masked;
+
// Scale the blend by the number of bytes per element.
int Scale = VT.getScalarSizeInBits() / 8;
Hi = DAG.getBitcast(AlignVT, Hi);
return DAG.getBitcast(
- VT, DAG.getNode(X86ISD::PALIGNR, DL, AlignVT, Hi, Lo,
+ VT, DAG.getNode(X86ISD::PALIGNR, DL, AlignVT, Lo, Hi,
DAG.getConstant(Rotation * Scale, DL, MVT::i8)));
}
DAG.getNode(ISD::OR, DL, MVT::v2i64, LoShift, HiShift));
}
-/// \brief Compute whether each element of a shuffle is zeroable.
-///
-/// A "zeroable" vector shuffle element is one which can be lowered to zero.
-/// Either it is an undef element in the shuffle mask, the element of the input
-/// referenced is undef, or the element of the input referenced is known to be
-/// zero. Many x86 shuffles can zero lanes cheaply and we often want to handle
-/// as many lanes with this technique as possible to simplify the remaining
-/// shuffle.
-static SmallBitVector computeZeroableShuffleElements(ArrayRef<int> Mask,
- SDValue V1, SDValue V2) {
- SmallBitVector Zeroable(Mask.size(), false);
-
- while (V1.getOpcode() == ISD::BITCAST)
- V1 = V1->getOperand(0);
- while (V2.getOpcode() == ISD::BITCAST)
- V2 = V2->getOperand(0);
-
- bool V1IsZero = ISD::isBuildVectorAllZeros(V1.getNode());
- bool V2IsZero = ISD::isBuildVectorAllZeros(V2.getNode());
-
- for (int i = 0, Size = Mask.size(); i < Size; ++i) {
- int M = Mask[i];
- // Handle the easy cases.
- if (M < 0 || (M >= 0 && M < Size && V1IsZero) || (M >= Size && V2IsZero)) {
- Zeroable[i] = true;
- continue;
- }
-
- // If this is an index into a build_vector node (which has the same number
- // of elements), dig out the input value and use it.
- SDValue V = M < Size ? V1 : V2;
- if (V.getOpcode() != ISD::BUILD_VECTOR || Size != (int)V.getNumOperands())
- continue;
-
- SDValue Input = V.getOperand(M % Size);
- // The UNDEF opcode check really should be dead code here, but not quite
- // worth asserting on (it isn't invalid, just unexpected).
- if (Input.getOpcode() == ISD::UNDEF || X86::isZeroNode(Input))
- Zeroable[i] = true;
- }
-
- return Zeroable;
-}
-
-/// \brief Try to emit a bitmask instruction for a shuffle.
-///
-/// This handles cases where we can model a blend exactly as a bitmask due to
-/// one of the inputs being zeroable.
-static SDValue lowerVectorShuffleAsBitMask(SDLoc DL, MVT VT, SDValue V1,
- SDValue V2, ArrayRef<int> Mask,
- SelectionDAG &DAG) {
- MVT EltVT = VT.getScalarType();
- int NumEltBits = EltVT.getSizeInBits();
- MVT IntEltVT = MVT::getIntegerVT(NumEltBits);
- SDValue Zero = DAG.getConstant(0, DL, IntEltVT);
- SDValue AllOnes = DAG.getConstant(APInt::getAllOnesValue(NumEltBits), DL,
- IntEltVT);
- if (EltVT.isFloatingPoint()) {
- Zero = DAG.getBitcast(EltVT, Zero);
- AllOnes = DAG.getBitcast(EltVT, AllOnes);
- }
- SmallVector<SDValue, 16> VMaskOps(Mask.size(), Zero);
- SmallBitVector Zeroable = computeZeroableShuffleElements(Mask, V1, V2);
- SDValue V;
- for (int i = 0, Size = Mask.size(); i < Size; ++i) {
- if (Zeroable[i])
- continue;
- if (Mask[i] % Size != i)
- return SDValue(); // Not a blend.
- if (!V)
- V = Mask[i] < Size ? V1 : V2;
- else if (V != (Mask[i] < Size ? V1 : V2))
- return SDValue(); // Can only let one input through the mask.
-
- VMaskOps[i] = AllOnes;
- }
- if (!V)
- return SDValue(); // No non-zeroable elements!
-
- SDValue VMask = DAG.getNode(ISD::BUILD_VECTOR, DL, VT, VMaskOps);
- V = DAG.getNode(VT.isFloatingPoint()
- ? (unsigned) X86ISD::FAND : (unsigned) ISD::AND,
- DL, VT, V, VMask);
- return V;
-}
-
/// \brief Try to lower a vector shuffle as a bit shift (shifts in zeros).
///
/// Attempts to match a shuffle mask against the PSLL(W/D/Q/DQ) and
// all the smarts here sunk into that routine. However, the current
// lowering of BUILD_VECTOR makes that nearly impossible until the old
// vector shuffle lowering is dead.
- if (SDValue V2S = getScalarValueForVectorElement(
- V2, Mask[V2Index] - Mask.size(), DAG)) {
+ SDValue V2S = getScalarValueForVectorElement(V2, Mask[V2Index] - Mask.size(),
+ DAG);
+ if (V2S && DAG.getTargetLoweringInfo().isTypeLegal(V2S.getValueType())) {
// We need to zext the scalar if it is smaller than an i32.
V2S = DAG.getBitcast(EltVT, V2S);
if (EltVT == MVT::i8 || EltVT == MVT::i16) {
// Check if this is a broadcast of a scalar. We special case lowering
// for scalars so that we can more effectively fold with loads.
+ // First, look through bitcast: if the original value has a larger element
+ // type than the shuffle, the broadcast element is in essence truncated.
+ // Make that explicit to ease folding.
+ if (V.getOpcode() == ISD::BITCAST && VT.isInteger()) {
+ EVT EltVT = VT.getVectorElementType();
+ SDValue V0 = V.getOperand(0);
+ EVT V0VT = V0.getValueType();
+
+ if (V0VT.isInteger() && V0VT.getVectorElementType().bitsGT(EltVT) &&
+ ((V0.getOpcode() == ISD::BUILD_VECTOR ||
+ (V0.getOpcode() == ISD::SCALAR_TO_VECTOR && BroadcastIdx == 0)))) {
+ V = DAG.getNode(ISD::TRUNCATE, DL, EltVT, V0.getOperand(BroadcastIdx));
+ BroadcastIdx = 0;
+ }
+ }
+
+ // Also check the simpler case, where we can directly reuse the scalar.
if (V.getOpcode() == ISD::BUILD_VECTOR ||
(V.getOpcode() == ISD::SCALAR_TO_VECTOR && BroadcastIdx == 0)) {
V = V.getOperand(BroadcastIdx);
assert(AToAInputs.size() + BToAInputs.size() == 4 &&
"Must call this with either 3:1 or 1:3 inputs (summing to 4).");
+ bool ThreeAInputs = AToAInputs.size() == 3;
+
// Compute the index of dword with only one word among the three inputs in
// a half by taking the sum of the half with three inputs and subtracting
// the sum of the actual three inputs. The difference is the remaining
// slot.
int ADWord, BDWord;
- int &TripleDWord = AToAInputs.size() == 3 ? ADWord : BDWord;
- int &OneInputDWord = AToAInputs.size() == 3 ? BDWord : ADWord;
- int TripleInputOffset = AToAInputs.size() == 3 ? AOffset : BOffset;
- ArrayRef<int> TripleInputs = AToAInputs.size() == 3 ? AToAInputs : BToAInputs;
- int OneInput = AToAInputs.size() == 3 ? BToAInputs[0] : AToAInputs[0];
+ int &TripleDWord = ThreeAInputs ? ADWord : BDWord;
+ int &OneInputDWord = ThreeAInputs ? BDWord : ADWord;
+ int TripleInputOffset = ThreeAInputs ? AOffset : BOffset;
+ ArrayRef<int> TripleInputs = ThreeAInputs ? AToAInputs : BToAInputs;
+ int OneInput = ThreeAInputs ? BToAInputs[0] : AToAInputs[0];
int TripleInputSum = 0 + 1 + 2 + 3 + (4 * TripleInputOffset);
int TripleNonInputIdx =
TripleInputSum - std::accumulate(TripleInputs.begin(), TripleInputs.end(), 0);
FixFlippedInputs(BPinnedIdx, BDWord, BToBInputs);
} else {
assert(NumFlippedAToBInputs != 0 && "Impossible given predicates!");
- int APinnedIdx =
- AToAInputs.size() == 3 ? TripleNonInputIdx : OneInput;
+ int APinnedIdx = ThreeAInputs ? TripleNonInputIdx : OneInput;
FixFlippedInputs(APinnedIdx, ADWord, AToBInputs);
}
}
return V;
}
+ if (SDValue Masked =
+ lowerVectorShuffleAsBitMask(DL, MVT::v16i8, V1, V2, Mask, DAG))
+ return Masked;
+
// Use dedicated unpack instructions for masks that match their pattern.
if (isShuffleEquivalent(V1, V2, Mask, {// Low half.
0, 16, 1, 17, 2, 18, 3, 19,
unsigned NumLanes = (NumElems - 1) / 8 + 1;
unsigned NumElemsInLane = NumElems / NumLanes;
- // Blend for v16i16 should be symetric for the both lanes.
+ // Blend for v16i16 should be symmetric for the both lanes.
for (unsigned i = 0; i < NumElemsInLane; ++i) {
SDValue EltCond = BuildVector->getOperand(i);
SDValue SndLaneEltCond =
// Bits [3:0] of the constant are the zero mask. The DAG Combiner may
// combine either bitwise AND or insert of float 0.0 to set these bits.
- const Function *F = DAG.getMachineFunction().getFunction();
- bool MinSize = F->hasFnAttribute(Attribute::MinSize);
+ bool MinSize = DAG.getMachineFunction().getFunction()->optForMinSize();
if (IdxVal == 0 && (!MinSize || !MayFoldLoad(N1))) {
// If this is an insertion of 32-bits into the low 32-bits of
// a vector, we prefer to generate a blend with immediate rather
// --> load32 addr
if ((IdxVal == OpVT.getVectorNumElements() / 2) &&
Vec.getOpcode() == ISD::INSERT_SUBVECTOR &&
- OpVT.is256BitVector() && SubVecVT.is128BitVector() &&
- !Subtarget->isUnalignedMem32Slow()) {
- SDValue SubVec2 = Vec.getOperand(1);
- if (auto *Idx2 = dyn_cast<ConstantSDNode>(Vec.getOperand(2))) {
- if (Idx2->getZExtValue() == 0) {
- SDValue Ops[] = { SubVec2, SubVec };
- if (SDValue Ld = EltsFromConsecutiveLoads(OpVT, Ops, dl, DAG, false))
- return Ld;
+ OpVT.is256BitVector() && SubVecVT.is128BitVector()) {
+ auto *Idx2 = dyn_cast<ConstantSDNode>(Vec.getOperand(2));
+ if (Idx2 && Idx2->getZExtValue() == 0) {
+ SDValue SubVec2 = Vec.getOperand(1);
+ // If needed, look through a bitcast to get to the load.
+ if (SubVec2.getNode() && SubVec2.getOpcode() == ISD::BITCAST)
+ SubVec2 = SubVec2.getOperand(0);
+
+ if (auto *FirstLd = dyn_cast<LoadSDNode>(SubVec2)) {
+ bool Fast;
+ unsigned Alignment = FirstLd->getAlignment();
+ unsigned AS = FirstLd->getAddressSpace();
+ const X86TargetLowering *TLI = Subtarget->getTargetLowering();
+ if (TLI->allowsMemoryAccess(*DAG.getContext(), DAG.getDataLayout(),
+ OpVT, AS, Alignment, &Fast) && Fast) {
+ SDValue Ops[] = { SubVec2, SubVec };
+ if (SDValue Ld = EltsFromConsecutiveLoads(OpVT, Ops, dl, DAG, false))
+ return Ld;
+ }
}
}
}
// load.
if (isGlobalStubReference(OpFlag))
Result = DAG.getLoad(PtrVT, DL, DAG.getEntryNode(), Result,
- MachinePointerInfo::getGOT(), false, false, false, 0);
+ MachinePointerInfo::getGOT(DAG.getMachineFunction()),
+ false, false, false, 0);
return Result;
}
// load.
if (isGlobalStubReference(OpFlags))
Result = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Result,
- MachinePointerInfo::getGOT(), false, false, false, 0);
+ MachinePointerInfo::getGOT(DAG.getMachineFunction()),
+ false, false, false, 0);
// If there was a non-zero offset that we didn't fold, create an explicit
// addition for it.
}
Offset = DAG.getLoad(PtrVT, dl, DAG.getEntryNode(), Offset,
- MachinePointerInfo::getGOT(), false, false, false, 0);
+ MachinePointerInfo::getGOT(DAG.getMachineFunction()),
+ false, false, false, 0);
}
// The address of the thread local variable is the add of the thread
auto PtrVT = getPointerTy(DAG.getDataLayout());
if (Subtarget->isTargetELF()) {
+ if (DAG.getTarget().Options.EmulatedTLS)
+ return LowerToTLSEmulatedModel(GA, DAG);
TLSModel::Model model = DAG.getTarget().getTLSModel(GV);
switch (model) {
case TLSModel::GeneralDynamic:
auto PtrVT = getPointerTy(MF.getDataLayout());
int SSFI = MF.getFrameInfo()->CreateStackObject(Size, Size, false);
SDValue StackSlot = DAG.getFrameIndex(SSFI, PtrVT);
- SDValue Chain = DAG.getStore(DAG.getEntryNode(), dl, Op.getOperand(0),
- StackSlot,
- MachinePointerInfo::getFixedStack(SSFI),
- false, false, 0);
+ SDValue Chain = DAG.getStore(
+ DAG.getEntryNode(), dl, Op.getOperand(0), StackSlot,
+ MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SSFI), false,
+ false, 0);
return BuildFILD(Op, SrcVT, Chain, StackSlot, DAG);
}
MachineMemOperand *MMO;
if (FI) {
int SSFI = FI->getIndex();
- MMO =
- DAG.getMachineFunction()
- .getMachineMemOperand(MachinePointerInfo::getFixedStack(SSFI),
- MachineMemOperand::MOLoad, ByteSize, ByteSize);
+ MMO = DAG.getMachineFunction().getMachineMemOperand(
+ MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SSFI),
+ MachineMemOperand::MOLoad, ByteSize, ByteSize);
} else {
MMO = cast<LoadSDNode>(StackSlot)->getMemOperand();
StackSlot = StackSlot.getOperand(1);
SDValue Ops[] = {
Chain, Result, StackSlot, DAG.getValueType(Op.getValueType()), InFlag
};
- MachineMemOperand *MMO =
- DAG.getMachineFunction()
- .getMachineMemOperand(MachinePointerInfo::getFixedStack(SSFI),
- MachineMemOperand::MOStore, SSFISize, SSFISize);
+ MachineMemOperand *MMO = DAG.getMachineFunction().getMachineMemOperand(
+ MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SSFI),
+ MachineMemOperand::MOStore, SSFISize, SSFISize);
Chain = DAG.getMemIntrinsicNode(X86ISD::FST, DL, Tys,
Ops, Op.getValueType(), MMO);
- Result = DAG.getLoad(Op.getValueType(), DL, Chain, StackSlot,
- MachinePointerInfo::getFixedStack(SSFI),
- false, false, false, 0);
+ Result = DAG.getLoad(
+ Op.getValueType(), DL, Chain, StackSlot,
+ MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SSFI),
+ false, false, false, 0);
}
return Result;
// Load the 64-bit value into an XMM register.
SDValue XR1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, MVT::v2i64,
Op.getOperand(0));
- SDValue CLod0 = DAG.getLoad(MVT::v4i32, dl, DAG.getEntryNode(), CPIdx0,
- MachinePointerInfo::getConstantPool(),
- false, false, false, 16);
+ SDValue CLod0 =
+ DAG.getLoad(MVT::v4i32, dl, DAG.getEntryNode(), CPIdx0,
+ MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
+ false, false, false, 16);
SDValue Unpck1 =
getUnpackl(DAG, dl, MVT::v4i32, DAG.getBitcast(MVT::v4i32, XR1), CLod0);
- SDValue CLod1 = DAG.getLoad(MVT::v2f64, dl, CLod0.getValue(1), CPIdx1,
- MachinePointerInfo::getConstantPool(),
- false, false, false, 16);
+ SDValue CLod1 =
+ DAG.getLoad(MVT::v2f64, dl, CLod0.getValue(1), CPIdx1,
+ MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
+ false, false, false, 16);
SDValue XR2F = DAG.getBitcast(MVT::v2f64, Unpck1);
SDValue Sub = DAG.getNode(ISD::FSUB, dl, MVT::v2f64, XR2F, CLod1);
SDValue Result;
// we must be careful to do the computation in x87 extended precision, not
// in SSE. (The generic code can't know it's OK to do this, or how to.)
int SSFI = cast<FrameIndexSDNode>(StackSlot)->getIndex();
- MachineMemOperand *MMO =
- DAG.getMachineFunction()
- .getMachineMemOperand(MachinePointerInfo::getFixedStack(SSFI),
- MachineMemOperand::MOLoad, 8, 8);
+ MachineMemOperand *MMO = DAG.getMachineFunction().getMachineMemOperand(
+ MachinePointerInfo::getFixedStack(DAG.getMachineFunction(), SSFI),
+ MachineMemOperand::MOLoad, 8, 8);
SDVTList Tys = DAG.getVTList(MVT::f80, MVT::Other);
SDValue Ops[] = { Store, StackSlot, DAG.getValueType(MVT::i64) };
// Load the value out, extending it from f32 to f80.
// FIXME: Avoid the extend by constructing the right constant pool?
- SDValue Fudge = DAG.getExtLoad(ISD::EXTLOAD, dl, MVT::f80, DAG.getEntryNode(),
- FudgePtr, MachinePointerInfo::getConstantPool(),
- MVT::f32, false, false, false, 4);
+ SDValue Fudge = DAG.getExtLoad(
+ ISD::EXTLOAD, dl, MVT::f80, DAG.getEntryNode(), FudgePtr,
+ MachinePointerInfo::getConstantPool(DAG.getMachineFunction()), MVT::f32,
+ false, false, false, 4);
// Extend everything to 80 bits to force it to be done on x87.
SDValue Add = DAG.getNode(ISD::FADD, dl, MVT::f80, Fild, Fudge);
return DAG.getNode(ISD::FP_ROUND, dl, DstVT, Add,
DAG.getIntPtrConstant(0, dl));
}
+// If the given FP_TO_SINT (IsSigned) or FP_TO_UINT (!IsSigned) operation
+// is legal, or has an f16 source (which needs to be promoted to f32),
+// just return an <SDValue(), SDValue()> pair.
+// Otherwise it is assumed to be a conversion from one of f32, f64 or f80
+// to i16, i32 or i64, and we lower it to a legal sequence.
+// If lowered to the final integer result we return a <result, SDValue()> pair.
+// Otherwise we lower it to a sequence ending with a FIST, return a
+// <FIST, StackSlot> pair, and the caller is responsible for loading
+// the final integer result from StackSlot.
std::pair<SDValue,SDValue>
-X86TargetLowering:: FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG,
- bool IsSigned, bool IsReplace) const {
+X86TargetLowering::FP_TO_INTHelper(SDValue Op, SelectionDAG &DAG,
+ bool IsSigned, bool IsReplace) const {
SDLoc DL(Op);
EVT DstTy = Op.getValueType();
+ EVT TheVT = Op.getOperand(0).getValueType();
auto PtrVT = getPointerTy(DAG.getDataLayout());
- if (!IsSigned && !isIntegerTypeFTOL(DstTy)) {
+ if (TheVT == MVT::f16)
+ // We need to promote the f16 to f32 before using the lowering
+ // in this routine.
+ return std::make_pair(SDValue(), SDValue());
+
+ assert((TheVT == MVT::f32 ||
+ TheVT == MVT::f64 ||
+ TheVT == MVT::f80) &&
+ "Unexpected FP operand type in FP_TO_INTHelper");
+
+ // If using FIST to compute an unsigned i64, we'll need some fixup
+ // to handle values above the maximum signed i64. A FIST is always
+ // used for the 32-bit subtarget, but also for f80 on a 64-bit target.
+ bool UnsignedFixup = !IsSigned &&
+ DstTy == MVT::i64 &&
+ (!Subtarget->is64Bit() ||
+ !isScalarFPTypeInSSEReg(TheVT));
+
+ if (!IsSigned && DstTy != MVT::i64 && !Subtarget->hasAVX512()) {
+ // Replace the fp-to-uint32 operation with an fp-to-sint64 FIST.
+ // The low 32 bits of the fist result will have the correct uint32 result.
assert(DstTy == MVT::i32 && "Unexpected FP_TO_UINT");
DstTy = MVT::i64;
}
isScalarFPTypeInSSEReg(Op.getOperand(0).getValueType()))
return std::make_pair(SDValue(), SDValue());
- // We lower FP->int64 either into FISTP64 followed by a load from a temporary
- // stack slot, or into the FTOL runtime function.
+ // We lower FP->int64 into FISTP64 followed by a load from a temporary
+ // stack slot.
MachineFunction &MF = DAG.getMachineFunction();
unsigned MemSize = DstTy.getSizeInBits()/8;
int SSFI = MF.getFrameInfo()->CreateStackObject(MemSize, MemSize, false);
SDValue StackSlot = DAG.getFrameIndex(SSFI, PtrVT);
unsigned Opc;
- if (!IsSigned && isIntegerTypeFTOL(DstTy))
- Opc = X86ISD::WIN_FTOL;
- else
- switch (DstTy.getSimpleVT().SimpleTy) {
- default: llvm_unreachable("Invalid FP_TO_SINT to lower!");
- case MVT::i16: Opc = X86ISD::FP_TO_INT16_IN_MEM; break;
- case MVT::i32: Opc = X86ISD::FP_TO_INT32_IN_MEM; break;
- case MVT::i64: Opc = X86ISD::FP_TO_INT64_IN_MEM; break;
- }
+ switch (DstTy.getSimpleVT().SimpleTy) {
+ default: llvm_unreachable("Invalid FP_TO_SINT to lower!");
+ case MVT::i16: Opc = X86ISD::FP_TO_INT16_IN_MEM; break;
+ case MVT::i32: Opc = X86ISD::FP_TO_INT32_IN_MEM; break;
+ case MVT::i64: Opc = X86ISD::FP_TO_INT64_IN_MEM; break;
+ }
SDValue Chain = DAG.getEntryNode();
SDValue Value = Op.getOperand(0);
- EVT TheVT = Op.getOperand(0).getValueType();
+ SDValue Adjust; // 0x0 or 0x80000000, for result sign bit adjustment.
+
+ if (UnsignedFixup) {
+ //
+ // Conversion to unsigned i64 is implemented with a select,
+ // depending on whether the source value fits in the range
+ // of a signed i64. Let Thresh be the FP equivalent of
+ // 0x8000000000000000ULL.
+ //
+ // Adjust i32 = (Value < Thresh) ? 0 : 0x80000000;
+ // FistSrc = (Value < Thresh) ? Value : (Value - Thresh);
+ // Fist-to-mem64 FistSrc
+ // Add 0 or 0x800...0ULL to the 64-bit result, which is equivalent
+ // to XOR'ing the high 32 bits with Adjust.
+ //
+ // Being a power of 2, Thresh is exactly representable in all FP formats.
+ // For X87 we'd like to use the smallest FP type for this constant, but
+ // for DAG type consistency we have to match the FP operand type.
+
+ APFloat Thresh(APFloat::IEEEsingle, APInt(32, 0x5f000000));
+ APFloat::opStatus Status = APFloat::opOK;
+ bool LosesInfo = false;
+ if (TheVT == MVT::f64)
+ // The rounding mode is irrelevant as the conversion should be exact.
+ Status = Thresh.convert(APFloat::IEEEdouble, APFloat::rmNearestTiesToEven,
+ &LosesInfo);
+ else if (TheVT == MVT::f80)
+ Status = Thresh.convert(APFloat::x87DoubleExtended,
+ APFloat::rmNearestTiesToEven, &LosesInfo);
+
+ assert(Status == APFloat::opOK && !LosesInfo &&
+ "FP conversion should have been exact");
+
+ SDValue ThreshVal = DAG.getConstantFP(Thresh, DL, TheVT);
+
+ SDValue Cmp = DAG.getSetCC(DL,
+ getSetCCResultType(DAG.getDataLayout(),
+ *DAG.getContext(), TheVT),
+ Value, ThreshVal, ISD::SETLT);
+ Adjust = DAG.getSelect(DL, MVT::i32, Cmp,
+ DAG.getConstant(0, DL, MVT::i32),
+ DAG.getConstant(0x80000000, DL, MVT::i32));
+ SDValue Sub = DAG.getNode(ISD::FSUB, DL, TheVT, Value, ThreshVal);
+ Cmp = DAG.getSetCC(DL, getSetCCResultType(DAG.getDataLayout(),
+ *DAG.getContext(), TheVT),
+ Value, ThreshVal, ISD::SETLT);
+ Value = DAG.getSelect(DL, TheVT, Cmp, Value, Sub);
+ }
+
// FIXME This causes a redundant load/store if the SSE-class value is already
// in memory, such as if it is on the callstack.
if (isScalarFPTypeInSSEReg(TheVT)) {
assert(DstTy == MVT::i64 && "Invalid FP_TO_SINT to lower!");
Chain = DAG.getStore(Chain, DL, Value, StackSlot,
- MachinePointerInfo::getFixedStack(SSFI),
- false, false, 0);
+ MachinePointerInfo::getFixedStack(MF, SSFI), false,
+ false, 0);
SDVTList Tys = DAG.getVTList(Op.getOperand(0).getValueType(), MVT::Other);
SDValue Ops[] = {
Chain, StackSlot, DAG.getValueType(TheVT)
};
MachineMemOperand *MMO =
- MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(SSFI),
- MachineMemOperand::MOLoad, MemSize, MemSize);
+ MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(MF, SSFI),
+ MachineMemOperand::MOLoad, MemSize, MemSize);
Value = DAG.getMemIntrinsicNode(X86ISD::FLD, DL, Tys, Ops, DstTy, MMO);
Chain = Value.getValue(1);
SSFI = MF.getFrameInfo()->CreateStackObject(MemSize, MemSize, false);
}
MachineMemOperand *MMO =
- MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(SSFI),
- MachineMemOperand::MOStore, MemSize, MemSize);
+ MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(MF, SSFI),
+ MachineMemOperand::MOStore, MemSize, MemSize);
- if (Opc != X86ISD::WIN_FTOL) {
- // Build the FP_TO_INT*_IN_MEM
- SDValue Ops[] = { Chain, Value, StackSlot };
+ if (UnsignedFixup) {
+
+ // Insert the FIST, load its result as two i32's,
+ // and XOR the high i32 with Adjust.
+
+ SDValue FistOps[] = { Chain, Value, StackSlot };
SDValue FIST = DAG.getMemIntrinsicNode(Opc, DL, DAG.getVTList(MVT::Other),
- Ops, DstTy, MMO);
- return std::make_pair(FIST, StackSlot);
- } else {
- SDValue ftol = DAG.getNode(X86ISD::WIN_FTOL, DL,
- DAG.getVTList(MVT::Other, MVT::Glue),
- Chain, Value);
- SDValue eax = DAG.getCopyFromReg(ftol, DL, X86::EAX,
- MVT::i32, ftol.getValue(1));
- SDValue edx = DAG.getCopyFromReg(eax.getValue(1), DL, X86::EDX,
- MVT::i32, eax.getValue(2));
- SDValue Ops[] = { eax, edx };
+ FistOps, DstTy, MMO);
+
+ SDValue Low32 = DAG.getLoad(MVT::i32, DL, FIST, StackSlot,
+ MachinePointerInfo(),
+ false, false, false, 0);
+ SDValue HighAddr = DAG.getNode(ISD::ADD, DL, PtrVT, StackSlot,
+ DAG.getConstant(4, DL, PtrVT));
+
+ SDValue High32 = DAG.getLoad(MVT::i32, DL, FIST, HighAddr,
+ MachinePointerInfo(),
+ false, false, false, 0);
+ High32 = DAG.getNode(ISD::XOR, DL, MVT::i32, High32, Adjust);
+
+ if (Subtarget->is64Bit()) {
+ // Join High32 and Low32 into a 64-bit result.
+ // (High32 << 32) | Low32
+ Low32 = DAG.getNode(ISD::ZERO_EXTEND, DL, MVT::i64, Low32);
+ High32 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, High32);
+ High32 = DAG.getNode(ISD::SHL, DL, MVT::i64, High32,
+ DAG.getConstant(32, DL, MVT::i8));
+ SDValue Result = DAG.getNode(ISD::OR, DL, MVT::i64, High32, Low32);
+ return std::make_pair(Result, SDValue());
+ }
+
+ SDValue ResultOps[] = { Low32, High32 };
+
SDValue pair = IsReplace
- ? DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Ops)
- : DAG.getMergeValues(Ops, DL);
+ ? DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, ResultOps)
+ : DAG.getMergeValues(ResultOps, DL);
return std::make_pair(pair, SDValue());
+ } else {
+ // Build the FP_TO_INT*_IN_MEM
+ SDValue Ops[] = { Chain, Value, StackSlot };
+ SDValue FIST = DAG.getMemIntrinsicNode(Opc, DL, DAG.getVTList(MVT::Other),
+ Ops, DstTy, MMO);
+ return std::make_pair(FIST, StackSlot);
}
}
Subtarget->hasDQI() && Subtarget->hasVLX())
return Op; // legal, will go to VPMOVB2M, VPMOVQ2M
}
- if (InVT.is512BitVector() || VT.getVectorElementType() == MVT::i1) {
- if (VT.getVectorElementType().getSizeInBits() >=8)
- return DAG.getNode(X86ISD::VTRUNC, DL, VT, In);
+ if (VT.getVectorElementType() == MVT::i1) {
assert(VT.getVectorElementType() == MVT::i1 && "Unexpected vector type");
unsigned NumElts = InVT.getVectorNumElements();
assert ((NumElts == 8 || NumElts == 16) && "Unexpected vector type");
return DAG.getNode(X86ISD::TESTM, DL, VT, And, And);
}
+ // vpmovqb/w/d, vpmovdb/w, vpmovwb
+ if (((!InVT.is512BitVector() && Subtarget->hasVLX()) || InVT.is512BitVector()) &&
+ (InVT.getVectorElementType() != MVT::i16 || Subtarget->hasBWI()))
+ return DAG.getNode(X86ISD::VTRUNC, DL, VT, In);
+
if ((VT == MVT::v4i32) && (InVT == MVT::v4i64)) {
// On AVX2, v4i64 -> v4i32 becomes VPERMD.
if (Subtarget->hasInt256()) {
/*IsSigned=*/ true, /*IsReplace=*/ false);
SDValue FIST = Vals.first, StackSlot = Vals.second;
// If FP_TO_INTHelper failed, the node is actually supposed to be Legal.
- if (!FIST.getNode()) return Op;
+ if (!FIST.getNode())
+ return Op;
if (StackSlot.getNode())
// Load the result.
std::pair<SDValue,SDValue> Vals = FP_TO_INTHelper(Op, DAG,
/*IsSigned=*/ false, /*IsReplace=*/ false);
SDValue FIST = Vals.first, StackSlot = Vals.second;
- assert(FIST.getNode() && "Unexpected failure");
+ // If FP_TO_INTHelper failed, the node is actually supposed to be Legal.
+ if (!FIST.getNode())
+ return Op;
if (StackSlot.getNode())
// Load the result.
if (User->getOpcode() == ISD::FNEG)
return Op;
- SDValue Op0 = Op.getOperand(0);
- bool IsFNABS = !IsFABS && (Op0.getOpcode() == ISD::FABS);
-
SDLoc dl(Op);
MVT VT = Op.getSimpleValueType();
- // Assume scalar op for initialization; update for vector if needed.
- // Note that there are no scalar bitwise logical SSE/AVX instructions, so we
- // generate a 16-byte vector constant and logic op even for the scalar case.
- // Using a 16-byte mask allows folding the load of the mask with
- // the logic op, so it can save (~4 bytes) on code size.
- MVT EltVT = VT;
- unsigned NumElts = VT == MVT::f64 ? 2 : 4;
+
// FIXME: Use function attribute "OptimizeForSize" and/or CodeGenOpt::Level to
// decide if we should generate a 16-byte constant mask when we only need 4 or
// 8 bytes for the scalar case.
+
+ MVT LogicVT;
+ MVT EltVT;
+ unsigned NumElts;
+
if (VT.isVector()) {
+ LogicVT = VT;
EltVT = VT.getVectorElementType();
NumElts = VT.getVectorNumElements();
+ } else {
+ // There are no scalar bitwise logical SSE/AVX instructions, so we
+ // generate a 16-byte vector constant and logic op even for the scalar case.
+ // Using a 16-byte mask allows folding the load of the mask with
+ // the logic op, so it can save (~4 bytes) on code size.
+ LogicVT = (VT == MVT::f64) ? MVT::v2f64 : MVT::v4f32;
+ EltVT = VT;
+ NumElts = (VT == MVT::f64) ? 2 : 4;
}
unsigned EltBits = EltVT.getSizeInBits();
const TargetLowering &TLI = DAG.getTargetLoweringInfo();
SDValue CPIdx = DAG.getConstantPool(C, TLI.getPointerTy(DAG.getDataLayout()));
unsigned Alignment = cast<ConstantPoolSDNode>(CPIdx)->getAlignment();
- SDValue Mask = DAG.getLoad(VT, dl, DAG.getEntryNode(), CPIdx,
- MachinePointerInfo::getConstantPool(),
- false, false, false, Alignment);
+ SDValue Mask =
+ DAG.getLoad(LogicVT, dl, DAG.getEntryNode(), CPIdx,
+ MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
+ false, false, false, Alignment);
- if (VT.isVector()) {
- // For a vector, cast operands to a vector type, perform the logic op,
- // and cast the result back to the original value type.
- MVT VecVT = MVT::getVectorVT(MVT::i64, VT.getSizeInBits() / 64);
- SDValue MaskCasted = DAG.getBitcast(VecVT, Mask);
- SDValue Operand = IsFNABS ? DAG.getBitcast(VecVT, Op0.getOperand(0))
- : DAG.getBitcast(VecVT, Op0);
- unsigned BitOp = IsFABS ? ISD::AND : IsFNABS ? ISD::OR : ISD::XOR;
- return DAG.getBitcast(VT,
- DAG.getNode(BitOp, dl, VecVT, Operand, MaskCasted));
- }
-
- // If not vector, then scalar.
- unsigned BitOp = IsFABS ? X86ISD::FAND : IsFNABS ? X86ISD::FOR : X86ISD::FXOR;
+ SDValue Op0 = Op.getOperand(0);
+ bool IsFNABS = !IsFABS && (Op0.getOpcode() == ISD::FABS);
+ unsigned LogicOp =
+ IsFABS ? X86ISD::FAND : IsFNABS ? X86ISD::FOR : X86ISD::FXOR;
SDValue Operand = IsFNABS ? Op0.getOperand(0) : Op0;
- return DAG.getNode(BitOp, dl, VT, Operand, Mask);
+
+ if (VT.isVector())
+ return DAG.getNode(LogicOp, dl, LogicVT, Operand, Mask);
+
+ // For the scalar case extend to a 128-bit vector, perform the logic op,
+ // and extract the scalar result back out.
+ Operand = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, LogicVT, Operand);
+ SDValue LogicNode = DAG.getNode(LogicOp, dl, LogicVT, Operand, Mask);
+ return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, VT, LogicNode,
+ DAG.getIntPtrConstant(0, dl));
}
static SDValue LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) {
Constant *C = ConstantVector::get(CV);
auto PtrVT = TLI.getPointerTy(DAG.getDataLayout());
SDValue CPIdx = DAG.getConstantPool(C, PtrVT, 16);
- SDValue Mask1 = DAG.getLoad(SrcVT, dl, DAG.getEntryNode(), CPIdx,
- MachinePointerInfo::getConstantPool(),
- false, false, false, 16);
- SDValue SignBit = DAG.getNode(X86ISD::FAND, dl, SrcVT, Op1, Mask1);
+
+ // Perform all logic operations as 16-byte vectors because there are no
+ // scalar FP logic instructions in SSE. This allows load folding of the
+ // constants into the logic instructions.
+ MVT LogicVT = (VT == MVT::f64) ? MVT::v2f64 : MVT::v4f32;
+ SDValue Mask1 =
+ DAG.getLoad(LogicVT, dl, DAG.getEntryNode(), CPIdx,
+ MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
+ false, false, false, 16);
+ Op1 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, LogicVT, Op1);
+ SDValue SignBit = DAG.getNode(X86ISD::FAND, dl, LogicVT, Op1, Mask1);
// Next, clear the sign bit from the first operand (magnitude).
// If it's a constant, we can clear it here.
APFloat APF = Op0CN->getValueAPF();
// If the magnitude is a positive zero, the sign bit alone is enough.
if (APF.isPosZero())
- return SignBit;
+ return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, SrcVT, SignBit,
+ DAG.getIntPtrConstant(0, dl));
APF.clearSign();
CV[0] = ConstantFP::get(*Context, APF);
} else {
}
C = ConstantVector::get(CV);
CPIdx = DAG.getConstantPool(C, PtrVT, 16);
- SDValue Val = DAG.getLoad(VT, dl, DAG.getEntryNode(), CPIdx,
- MachinePointerInfo::getConstantPool(),
- false, false, false, 16);
+ SDValue Val =
+ DAG.getLoad(LogicVT, dl, DAG.getEntryNode(), CPIdx,
+ MachinePointerInfo::getConstantPool(DAG.getMachineFunction()),
+ false, false, false, 16);
// If the magnitude operand wasn't a constant, we need to AND out the sign.
- if (!isa<ConstantFPSDNode>(Op0))
- Val = DAG.getNode(X86ISD::FAND, dl, VT, Op0, Val);
-
+ if (!isa<ConstantFPSDNode>(Op0)) {
+ Op0 = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, LogicVT, Op0);
+ Val = DAG.getNode(X86ISD::FAND, dl, LogicVT, Op0, Val);
+ }
// OR the magnitude value with the sign bit.
- return DAG.getNode(X86ISD::FOR, dl, VT, Val, SignBit);
+ Val = DAG.getNode(X86ISD::FOR, dl, LogicVT, Val, SignBit);
+ return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, SrcVT, Val,
+ DAG.getIntPtrConstant(0, dl));
}
static SDValue LowerFGETSIGN(SDValue Op, SelectionDAG &DAG) {
// if we're optimizing for size, however, as that'll allow better folding
// of memory operations.
if (Op0.getValueType() != MVT::i32 && Op0.getValueType() != MVT::i64 &&
- !DAG.getMachineFunction().getFunction()->hasFnAttribute(
- Attribute::MinSize) &&
+ !DAG.getMachineFunction().getFunction()->optForMinSize() &&
!Subtarget->isAtom()) {
unsigned ExtendOp =
isX86CCUnsigned(X86CC) ? ISD::ZERO_EXTEND : ISD::SIGN_EXTEND;
/// This is because we still need one division to calculate the reciprocal and
/// then we need two multiplies by that reciprocal as replacements for the
/// original divisions.
-bool X86TargetLowering::combineRepeatedFPDivisors(unsigned NumUsers) const {
- return NumUsers > 1;
+unsigned X86TargetLowering::combineRepeatedFPDivisors() const {
+ return 2;
}
static bool isAllOnes(SDValue V) {
}
if (addTest) {
- // Look pass the truncate if the high bits are known zero.
+ // Look past the truncate if the high bits are known zero.
if (isTruncWithZeroHighBitsInput(Cond, DAG))
- Cond = Cond.getOperand(0);
+ Cond = Cond.getOperand(0);
// We know the result of AND is compared against zero. Try to match
// it to BT.
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
SDLoc DL(Op);
- if (!Subtarget->is64Bit() || Subtarget->isTargetWin64()) {
+ if (!Subtarget->is64Bit() ||
+ Subtarget->isCallingConvWin64(MF.getFunction()->getCallingConv())) {
// vastart just stores the address of the VarArgsFrameIndex slot into the
// memory location argument.
SDValue FR = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(), PtrVT);
SDValue X86TargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG) const {
assert(Subtarget->is64Bit() &&
"LowerVAARG only handles 64-bit va_arg!");
- assert((Subtarget->isTargetLinux() ||
- Subtarget->isTargetDarwin()) &&
- "Unhandled target in LowerVAARG");
assert(Op.getNode()->getNumOperands() == 4);
+
+ MachineFunction &MF = DAG.getMachineFunction();
+ if (Subtarget->isCallingConvWin64(MF.getFunction()->getCallingConv()))
+ // The Win64 ABI uses char* instead of a structure.
+ return DAG.expandVAArg(Op.getNode());
+
SDValue Chain = Op.getOperand(0);
SDValue SrcPtr = Op.getOperand(1);
const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
if (ArgMode == 2) {
// Sanity Check: Make sure using fp_offset makes sense.
assert(!Subtarget->useSoftFloat() &&
- !(DAG.getMachineFunction().getFunction()->hasFnAttribute(
- Attribute::NoImplicitFloat)) &&
+ !(MF.getFunction()->hasFnAttribute(Attribute::NoImplicitFloat)) &&
Subtarget->hasSSE1());
}
static SDValue LowerVACOPY(SDValue Op, const X86Subtarget *Subtarget,
SelectionDAG &DAG) {
- // X86-64 va_list is a struct { i32, i32, i8*, i8* }.
+ // X86-64 va_list is a struct { i32, i32, i8*, i8* }, except on Windows,
+ // where a va_list is still an i8*.
assert(Subtarget->is64Bit() && "This code only handles 64-bit va_copy!");
+ if (Subtarget->isCallingConvWin64(
+ DAG.getMachineFunction().getFunction()->getCallingConv()))
+ // Probably a Win64 va_copy.
+ return DAG.expandVACopy(Op.getNode());
+
SDValue Chain = Op.getOperand(0);
SDValue DstPtr = Op.getOperand(1);
SDValue SrcPtr = Op.getOperand(2);
/// \brief Return (and \p Op, \p Mask) for compare instructions or
/// (vselect \p Mask, \p Op, \p PreservedSrc) for others along with the
-/// necessary casting for \p Mask when lowering masking intrinsics.
+/// necessary casting or extending for \p Mask when lowering masking intrinsics
static SDValue getVectorMaskingNode(SDValue Op, SDValue Mask,
SDValue PreservedSrc,
const X86Subtarget *Subtarget,
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());
+ SDValue VMask = SDValue();
+ unsigned OpcodeSelect = ISD::VSELECT;
SDLoc dl(Op);
assert(MaskVT.isSimple() && "invalid mask type");
if (isAllOnes(Mask))
return Op;
- // In case when MaskVT equals v2i1 or v4i1, low 2 or 4 elements
- // are extracted by EXTRACT_SUBVECTOR.
- SDValue VMask = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MaskVT,
- DAG.getBitcast(BitcastVT, Mask),
- DAG.getIntPtrConstant(0, dl));
+ if (MaskVT.bitsGT(Mask.getValueType())) {
+ EVT newMaskVT = EVT::getIntegerVT(*DAG.getContext(),
+ MaskVT.getSizeInBits());
+ VMask = DAG.getBitcast(MaskVT,
+ DAG.getNode(ISD::ANY_EXTEND, dl, newMaskVT, Mask));
+ } else {
+ EVT BitcastVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
+ Mask.getValueType().getSizeInBits());
+ // In case when MaskVT equals v2i1 or v4i1, low 2 or 4 elements
+ // are extracted by EXTRACT_SUBVECTOR.
+ VMask = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MaskVT,
+ DAG.getBitcast(BitcastVT, Mask),
+ DAG.getIntPtrConstant(0, dl));
+ }
switch (Op.getOpcode()) {
default: break;
case X86ISD::CMPM:
case X86ISD::CMPMU:
return DAG.getNode(ISD::AND, dl, VT, Op, VMask);
+ case X86ISD::VTRUNC:
+ case X86ISD::VTRUNCS:
+ case X86ISD::VTRUNCUS:
+ // We can't use ISD::VSELECT here because it is not always "Legal"
+ // for the destination type. For example vpmovqb require only AVX512
+ // and vselect that can operate on byte element type require BWI
+ OpcodeSelect = X86ISD::SELECT;
+ break;
}
if (PreservedSrc.getOpcode() == ISD::UNDEF)
PreservedSrc = getZeroVector(VT, Subtarget, DAG, dl);
- return DAG.getNode(ISD::VSELECT, dl, VT, VMask, Op, PreservedSrc);
+ return DAG.getNode(OpcodeSelect, 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
+/// The mask is coming 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.
+/// "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,
SDValue Mask = Op.getOperand(3);
SDValue RoundingMode;
// We allways add rounding mode to the Node.
- // If the rounding mode is not specified, we add the
+ // If the rounding mode is not specified, we add the
// "current direction" mode.
if (Op.getNumOperands() == 4)
RoundingMode =
SDValue Src0 = Op.getOperand(3);
SDValue Mask = Op.getOperand(4);
// There are 2 kinds of intrinsics in this group:
- // (1) With supress-all-exceptions (sae) or rounding mode- 6 operands
+ // (1) With suppress-all-exceptions (sae) or rounding mode- 6 operands
// (2) With rounding mode and sae - 7 operands.
if (Op.getNumOperands() == 6) {
SDValue Sae = Op.getOperand(5);
Src1, Src2, Imm, Rnd),
Mask, PassThru, Subtarget, DAG);
}
+ case INTR_TYPE_3OP_IMM8_MASK:
case INTR_TYPE_3OP_MASK: {
SDValue Src1 = Op.getOperand(1);
SDValue Src2 = Op.getOperand(2);
SDValue Src3 = Op.getOperand(3);
SDValue PassThru = Op.getOperand(4);
SDValue Mask = Op.getOperand(5);
+
+ if (IntrData->Type == INTR_TYPE_3OP_IMM8_MASK)
+ Src3 = DAG.getNode(ISD::TRUNCATE, dl, MVT::i8, Src3);
// We specify 2 possible opcodes for intrinsics with rounding modes.
// First, we check if the intrinsic may have non-default rounding mode,
// (IntrData->Opc1 != 0), then we check the rounding mode operand.
return Chain;
}
+/// \brief Lower intrinsics for TRUNCATE_TO_MEM case
+/// return truncate Store/MaskedStore Node
+static SDValue LowerINTRINSIC_TRUNCATE_TO_MEM(const SDValue & Op,
+ SelectionDAG &DAG,
+ MVT ElementType) {
+ SDLoc dl(Op);
+ SDValue Mask = Op.getOperand(4);
+ SDValue DataToTruncate = Op.getOperand(3);
+ SDValue Addr = Op.getOperand(2);
+ SDValue Chain = Op.getOperand(0);
+
+ EVT VT = DataToTruncate.getValueType();
+ EVT SVT = EVT::getVectorVT(*DAG.getContext(),
+ ElementType, VT.getVectorNumElements());
+
+ if (isAllOnes(Mask)) // return just a truncate store
+ return DAG.getTruncStore(Chain, dl, DataToTruncate, Addr,
+ MachinePointerInfo(), SVT, false, false,
+ SVT.getScalarSizeInBits()/8);
+
+ EVT MaskVT = EVT::getVectorVT(*DAG.getContext(),
+ MVT::i1, VT.getVectorNumElements());
+ EVT BitcastVT = EVT::getVectorVT(*DAG.getContext(), MVT::i1,
+ Mask.getValueType().getSizeInBits());
+ // In case when MaskVT equals v2i1 or v4i1, low 2 or 4 elements
+ // are extracted by EXTRACT_SUBVECTOR.
+ SDValue VMask = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, MaskVT,
+ DAG.getBitcast(BitcastVT, Mask),
+ DAG.getIntPtrConstant(0, dl));
+
+ MachineMemOperand *MMO = DAG.getMachineFunction().
+ getMachineMemOperand(MachinePointerInfo(),
+ MachineMemOperand::MOStore, SVT.getStoreSize(),
+ SVT.getScalarSizeInBits()/8);
+
+ return DAG.getMaskedStore(Chain, dl, DataToTruncate, Addr,
+ VMask, SVT, MMO, true);
+}
+
static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, const X86Subtarget *Subtarget,
SelectionDAG &DAG) {
unsigned IntNo = cast<ConstantSDNode>(Op.getOperand(1))->getZExtValue();
MachinePointerInfo(), false, false,
VT.getScalarSizeInBits()/8);
}
+ case TRUNCATE_TO_MEM_VI8:
+ return LowerINTRINSIC_TRUNCATE_TO_MEM(Op, DAG, MVT::i8);
+ case TRUNCATE_TO_MEM_VI16:
+ return LowerINTRINSIC_TRUNCATE_TO_MEM(Op, DAG, MVT::i16);
+ case TRUNCATE_TO_MEM_VI32:
+ return LowerINTRINSIC_TRUNCATE_TO_MEM(Op, DAG, MVT::i32);
case EXPAND_FROM_MEM: {
SDLoc dl(Op);
SDValue Mask = Op.getOperand(4);
DAG.getRegister(StoreAddrReg, PtrVT));
}
+SDValue X86TargetLowering::LowerCATCHRET(SDValue Op, SelectionDAG &DAG) const {
+ SDValue Chain = Op.getOperand(0);
+ SDValue Dest = Op.getOperand(1);
+ SDLoc DL(Op);
+
+ MVT PtrVT = getPointerTy(DAG.getDataLayout());
+ unsigned ReturnReg = (PtrVT == MVT::i64 ? X86::RAX : X86::EAX);
+
+ // Load the address of the destination block.
+ MachineBasicBlock *DestMBB = cast<BasicBlockSDNode>(Dest)->getBasicBlock();
+ SDValue BlockPtr = DAG.getMCSymbol(DestMBB->getSymbol(), PtrVT);
+ unsigned WrapperKind =
+ Subtarget->isPICStyleRIPRel() ? X86ISD::WrapperRIP : X86ISD::Wrapper;
+ SDValue WrappedPtr = DAG.getNode(WrapperKind, DL, PtrVT, BlockPtr);
+ Chain = DAG.getCopyToReg(Chain, DL, ReturnReg, WrappedPtr);
+ return DAG.getNode(X86ISD::CATCHRET, DL, MVT::Other, Chain,
+ DAG.getRegister(ReturnReg, PtrVT));
+}
+
SDValue X86TargetLowering::lowerEH_SJLJ_SETJMP(SDValue Op,
SelectionDAG &DAG) const {
SDLoc DL(Op);
for (FunctionType::param_iterator I = FTy->param_begin(),
E = FTy->param_end(); I != E; ++I, ++Idx)
- if (Attrs.hasAttribute(Idx, Attribute::InReg))
+ if (Attrs.hasAttribute(Idx, Attribute::InReg)) {
+ auto &DL = DAG.getDataLayout();
// FIXME: should only count parameters that are lowered to integers.
- InRegCount += (TD->getTypeSizeInBits(*I) + 31) / 32;
+ InRegCount += (DL.getTypeSizeInBits(*I) + 31) / 32;
+ }
if (InRegCount > 2) {
report_fatal_error("Nest register in use - reduce number of inreg"
DAG.getFrameIndex(SSFI, getPointerTy(DAG.getDataLayout()));
MachineMemOperand *MMO =
- MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(SSFI),
- MachineMemOperand::MOStore, 2, 2);
+ MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(MF, SSFI),
+ MachineMemOperand::MOStore, 2, 2);
SDValue Ops[] = { DAG.getEntryNode(), StackSlot };
SDValue Chain = DAG.getMemIntrinsicNode(X86ISD::FNSTCW16m, DL,
return Lower256IntArith(Op, DAG);
}
+static SDValue LowerMINMAX(SDValue Op, SelectionDAG &DAG) {
+ assert(Op.getSimpleValueType().is256BitVector() &&
+ Op.getSimpleValueType().isInteger() &&
+ "Only handle AVX 256-bit vector integer operation");
+ return Lower256IntArith(Op, DAG);
+}
+
static SDValue LowerMUL(SDValue Op, const X86Subtarget *Subtarget,
SelectionDAG &DAG) {
SDLoc dl(Op);
}
if (Op.getOpcode() == ISD::SRA) {
if (ShiftAmt == 7) {
- // R s>> 7 === R s< 0
+ // ashr(R, 7) === cmp_slt(R, 0)
SDValue Zeros = getZeroVector(VT, Subtarget, DAG, dl);
return DAG.getNode(X86ISD::PCMPGT, dl, VT, Zeros, R);
}
- // R s>> a === ((R u>> a) ^ m) - m
+ // ashr(R, Amt) === sub(xor(lshr(R, Amt), Mask), Mask)
SDValue Res = DAG.getNode(ISD::SRL, dl, VT, R, Amt);
SmallVector<SDValue, 32> V(NumElts,
DAG.getConstant(128 >> ShiftAmt, dl,
// Special case in 32-bit mode, where i64 is expanded into high and low parts.
if (!Subtarget->is64Bit() &&
- (VT == MVT::v2i64 || (Subtarget->hasInt256() && VT == MVT::v4i64)) &&
- Amt.getOpcode() == ISD::BITCAST &&
- Amt.getOperand(0).getOpcode() == ISD::BUILD_VECTOR) {
+ (VT == MVT::v2i64 || (Subtarget->hasInt256() && VT == MVT::v4i64))) {
+
+ // Peek through any splat that was introduced for i64 shift vectorization.
+ int SplatIndex = -1;
+ if (ShuffleVectorSDNode *SVN = dyn_cast<ShuffleVectorSDNode>(Amt.getNode()))
+ if (SVN->isSplat()) {
+ SplatIndex = SVN->getSplatIndex();
+ Amt = Amt.getOperand(0);
+ assert(SplatIndex < (int)VT.getVectorNumElements() &&
+ "Splat shuffle referencing second operand");
+ }
+
+ if (Amt.getOpcode() != ISD::BITCAST ||
+ Amt.getOperand(0).getOpcode() != ISD::BUILD_VECTOR)
+ return SDValue();
+
Amt = Amt.getOperand(0);
unsigned Ratio = Amt.getSimpleValueType().getVectorNumElements() /
VT.getVectorNumElements();
unsigned RatioInLog2 = Log2_32_Ceil(Ratio);
uint64_t ShiftAmt = 0;
+ unsigned BaseOp = (SplatIndex < 0 ? 0 : SplatIndex * Ratio);
for (unsigned i = 0; i != Ratio; ++i) {
- ConstantSDNode *C = dyn_cast<ConstantSDNode>(Amt.getOperand(i));
+ ConstantSDNode *C = dyn_cast<ConstantSDNode>(Amt.getOperand(i + BaseOp));
if (!C)
return SDValue();
// 6 == Log2(64)
ShiftAmt |= C->getZExtValue() << (i * (1 << (6 - RatioInLog2)));
}
- // Check remaining shift amounts.
- for (unsigned i = Ratio; i != Amt.getNumOperands(); i += Ratio) {
- uint64_t ShAmt = 0;
- for (unsigned j = 0; j != Ratio; ++j) {
- ConstantSDNode *C =
- dyn_cast<ConstantSDNode>(Amt.getOperand(i + j));
- if (!C)
+
+ // Check remaining shift amounts (if not a splat).
+ if (SplatIndex < 0) {
+ for (unsigned i = Ratio; i != Amt.getNumOperands(); i += Ratio) {
+ uint64_t ShAmt = 0;
+ for (unsigned j = 0; j != Ratio; ++j) {
+ ConstantSDNode *C = dyn_cast<ConstantSDNode>(Amt.getOperand(i + j));
+ if (!C)
+ return SDValue();
+ // 6 == Log2(64)
+ ShAmt |= C->getZExtValue() << (j * (1 << (6 - RatioInLog2)));
+ }
+ if (ShAmt != ShiftAmt)
return SDValue();
- // 6 == Log2(64)
- ShAmt |= C->getZExtValue() << (j * (1 << (6 - RatioInLog2)));
}
- if (ShAmt != ShiftAmt)
- return SDValue();
}
if (SupportedVectorShiftWithImm(VT, Subtarget, Op.getOpcode()))
return DAG.getVectorShuffle(VT, dl, R0, R1, {0, 3});
}
+ // i64 vector arithmetic shift can be emulated with the transform:
+ // M = lshr(SIGN_BIT, Amt)
+ // ashr(R, Amt) === sub(xor(lshr(R, Amt), M), M)
+ if ((VT == MVT::v2i64 || (VT == MVT::v4i64 && Subtarget->hasInt256())) &&
+ Op.getOpcode() == ISD::SRA) {
+ SDValue S = DAG.getConstant(APInt::getSignBit(64), dl, VT);
+ SDValue M = DAG.getNode(ISD::SRL, dl, VT, S, Amt);
+ R = DAG.getNode(ISD::SRL, dl, VT, R, Amt);
+ R = DAG.getNode(ISD::XOR, dl, VT, R, M);
+ R = DAG.getNode(ISD::SUB, dl, VT, R, M);
+ return R;
+ }
+
// If possible, lower this packed shift into a vector multiply instead of
// expanding it into a sequence of scalar shifts.
// Do this only if the vector shift count is a constant build_vector.
/// the corresponding cmpxchg8b or cmpxchg16b instruction is available.
/// Used to know whether to use cmpxchg8/16b when expanding atomic operations
/// (otherwise we leave them alone to become __sync_fetch_and_... calls).
-bool X86TargetLowering::needsCmpXchgNb(const Type *MemType) const {
+bool X86TargetLowering::needsCmpXchgNb(Type *MemType) const {
unsigned OpWidth = MemType->getPrimitiveSizeInBits();
if (OpWidth == 64)
TargetLoweringBase::AtomicRMWExpansionKind
X86TargetLowering::shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const {
unsigned NativeWidth = Subtarget->is64Bit() ? 64 : 32;
- const Type *MemType = AI->getType();
+ Type *MemType = AI->getType();
// If the operand is too big, we must see if cmpxchg8/16b is available
// and default to library calls otherwise.
LoadInst *
X86TargetLowering::lowerIdempotentRMWIntoFencedLoad(AtomicRMWInst *AI) const {
unsigned NativeWidth = Subtarget->is64Bit() ? 64 : 32;
- const Type *MemType = AI->getType();
+ Type *MemType = AI->getType();
// Accesses larger than the native width are turned into cmpxchg/libcalls, so
// there is no benefit in turning such RMWs into loads, and it is actually
// harmful as it introduces a mfence.
// lowered to just a load without a fence. A mfence flushes the store buffer,
// making the optimization clearly correct.
// FIXME: it is required if isAtLeastRelease(Order) but it is not clear
- // otherwise, we might be able to be more agressive on relaxed idempotent
+ // otherwise, we might be able to be more aggressive on relaxed idempotent
// rmw. In practice, they do not look useful, so we don't try to be
// especially clever.
if (SynchScope == SingleThread)
return LowerFRAME_TO_ARGS_OFFSET(Op, DAG);
case ISD::DYNAMIC_STACKALLOC: return LowerDYNAMIC_STACKALLOC(Op, DAG);
case ISD::EH_RETURN: return LowerEH_RETURN(Op, DAG);
+ case ISD::CATCHRET: return LowerCATCHRET(Op, DAG);
case ISD::EH_SJLJ_SETJMP: return lowerEH_SJLJ_SETJMP(Op, DAG);
case ISD::EH_SJLJ_LONGJMP: return lowerEH_SJLJ_LONGJMP(Op, DAG);
case ISD::INIT_TRAMPOLINE: return LowerINIT_TRAMPOLINE(Op, DAG);
case ISD::SUBE: return LowerADDC_ADDE_SUBC_SUBE(Op, DAG);
case ISD::ADD: return LowerADD(Op, DAG);
case ISD::SUB: return LowerSUB(Op, DAG);
+ case ISD::SMAX:
+ case ISD::SMIN:
+ case ISD::UMAX:
+ case ISD::UMIN: return LowerMINMAX(Op, DAG);
case ISD::FSINCOS: return LowerFSINCOS(Op, Subtarget, DAG);
case ISD::MGATHER: return LowerMGATHER(Op, Subtarget, DAG);
case ISD::MSCATTER: return LowerMSCATTER(Op, Subtarget, DAG);
return;
}
case ISD::FP_TO_SINT:
- // FP_TO_INT*_IN_MEM is not legal for f16 inputs. Do not convert
- // (FP_TO_SINT (load f16)) to FP_TO_INT*.
- if (N->getOperand(0).getValueType() == MVT::f16)
- break;
- // fallthrough
case ISD::FP_TO_UINT: {
bool IsSigned = N->getOpcode() == ISD::FP_TO_SINT;
- if (!IsSigned && !isIntegerTypeFTOL(SDValue(N, 0).getValueType()))
- return;
-
std::pair<SDValue,SDValue> Vals =
FP_TO_INTHelper(SDValue(N, 0), DAG, IsSigned, /*IsReplace=*/ true);
SDValue FIST = Vals.first, StackSlot = Vals.second;
case X86ISD::EH_SJLJ_SETJMP: return "X86ISD::EH_SJLJ_SETJMP";
case X86ISD::EH_SJLJ_LONGJMP: return "X86ISD::EH_SJLJ_LONGJMP";
case X86ISD::EH_RETURN: return "X86ISD::EH_RETURN";
+ case X86ISD::CATCHRET: return "X86ISD::CATCHRET";
case X86ISD::TC_RETURN: return "X86ISD::TC_RETURN";
case X86ISD::FNSTCW16m: return "X86ISD::FNSTCW16m";
case X86ISD::FNSTSW16r: return "X86ISD::FNSTSW16r";
case X86ISD::VZEXT: return "X86ISD::VZEXT";
case X86ISD::VSEXT: return "X86ISD::VSEXT";
case X86ISD::VTRUNC: return "X86ISD::VTRUNC";
- case X86ISD::VTRUNCM: return "X86ISD::VTRUNCM";
+ case X86ISD::VTRUNCS: return "X86ISD::VTRUNCS";
+ case X86ISD::VTRUNCUS: return "X86ISD::VTRUNCUS";
case X86ISD::VINSERT: return "X86ISD::VINSERT";
case X86ISD::VFPEXT: return "X86ISD::VFPEXT";
case X86ISD::VFPROUND: return "X86ISD::VFPROUND";
case X86ISD::TESTM: return "X86ISD::TESTM";
case X86ISD::TESTNM: return "X86ISD::TESTNM";
case X86ISD::KORTEST: return "X86ISD::KORTEST";
+ case X86ISD::KTEST: return "X86ISD::KTEST";
case X86ISD::PACKSS: return "X86ISD::PACKSS";
case X86ISD::PACKUS: return "X86ISD::PACKUS";
case X86ISD::PALIGNR: return "X86ISD::PALIGNR";
case X86ISD::PMULUDQ: return "X86ISD::PMULUDQ";
case X86ISD::PMULDQ: return "X86ISD::PMULDQ";
case X86ISD::PSADBW: return "X86ISD::PSADBW";
+ case X86ISD::DBPSADBW: return "X86ISD::DBPSADBW";
case X86ISD::VASTART_SAVE_XMM_REGS: return "X86ISD::VASTART_SAVE_XMM_REGS";
case X86ISD::VAARG_64: return "X86ISD::VAARG_64";
case X86ISD::WIN_ALLOCA: return "X86ISD::WIN_ALLOCA";
case X86ISD::SFENCE: return "X86ISD::SFENCE";
case X86ISD::LFENCE: return "X86ISD::LFENCE";
case X86ISD::SEG_ALLOCA: return "X86ISD::SEG_ALLOCA";
- case X86ISD::WIN_FTOL: return "X86ISD::WIN_FTOL";
case X86ISD::SAHF: return "X86ISD::SAHF";
case X86ISD::RDRAND: return "X86ISD::RDRAND";
case X86ISD::RDSEED: return "X86ISD::RDSEED";
int64_t RegSaveFrameIndex = MI->getOperand(1).getImm();
int64_t VarArgsFPOffset = MI->getOperand(2).getImm();
- if (!Subtarget->isTargetWin64()) {
+ if (!Subtarget->isCallingConvWin64(F->getFunction()->getCallingConv())) {
// If %al is 0, branch around the XMM save block.
BuildMI(MBB, DL, TII->get(X86::TEST8rr)).addReg(CountReg).addReg(CountReg);
BuildMI(MBB, DL, TII->get(X86::JE_1)).addMBB(EndMBB);
// In the XMM save block, save all the XMM argument registers.
for (int i = 3, e = MI->getNumOperands() - 1; i != e; ++i) {
int64_t Offset = (i - 3) * 16 + VarArgsFPOffset;
- MachineMemOperand *MMO =
- F->getMachineMemOperand(
- MachinePointerInfo::getFixedStack(RegSaveFrameIndex, Offset),
+ MachineMemOperand *MMO = F->getMachineMemOperand(
+ MachinePointerInfo::getFixedStack(*F, RegSaveFrameIndex, Offset),
MachineMemOperand::MOStore,
/*Size=*/16, /*Align=*/16);
BuildMI(XMMSaveMBB, DL, TII->get(MOVOpc))
return true;
}
+// Return true if it is OK for this CMOV pseudo-opcode to be cascaded
+// together with other CMOV pseudo-opcodes into a single basic-block with
+// conditional jump around it.
+static bool isCMOVPseudo(MachineInstr *MI) {
+ switch (MI->getOpcode()) {
+ case X86::CMOV_FR32:
+ case X86::CMOV_FR64:
+ case X86::CMOV_GR8:
+ case X86::CMOV_GR16:
+ case X86::CMOV_GR32:
+ case X86::CMOV_RFP32:
+ case X86::CMOV_RFP64:
+ case X86::CMOV_RFP80:
+ case X86::CMOV_V2F64:
+ case X86::CMOV_V2I64:
+ case X86::CMOV_V4F32:
+ case X86::CMOV_V4F64:
+ case X86::CMOV_V4I64:
+ case X86::CMOV_V16F32:
+ case X86::CMOV_V8F32:
+ case X86::CMOV_V8F64:
+ case X86::CMOV_V8I64:
+ case X86::CMOV_V8I1:
+ case X86::CMOV_V16I1:
+ case X86::CMOV_V32I1:
+ case X86::CMOV_V64I1:
+ return true;
+
+ default:
+ return false;
+ }
+}
+
MachineBasicBlock *
X86TargetLowering::EmitLoweredSelect(MachineInstr *MI,
MachineBasicBlock *BB) const {
MachineBasicBlock *thisMBB = BB;
MachineFunction *F = BB->getParent();
- // We also lower double CMOVs:
+ // This code lowers all pseudo-CMOV instructions. Generally it lowers these
+ // as described above, by inserting a BB, and then making a PHI at the join
+ // point to select the true and false operands of the CMOV in the PHI.
+ //
+ // The code also handles two different cases of multiple CMOV opcodes
+ // in a row.
+ //
+ // Case 1:
+ // In this case, there are multiple CMOVs in a row, all which are based on
+ // the same condition setting (or the exact opposite condition setting).
+ // In this case we can lower all the CMOVs using a single inserted BB, and
+ // then make a number of PHIs at the join point to model the CMOVs. The only
+ // trickiness here, is that in a case like:
+ //
+ // t2 = CMOV cond1 t1, f1
+ // t3 = CMOV cond1 t2, f2
+ //
+ // when rewriting this into PHIs, we have to perform some renaming on the
+ // temps since you cannot have a PHI operand refer to a PHI result earlier
+ // in the same block. The "simple" but wrong lowering would be:
+ //
+ // t2 = PHI t1(BB1), f1(BB2)
+ // t3 = PHI t2(BB1), f2(BB2)
+ //
+ // but clearly t2 is not defined in BB1, so that is incorrect. The proper
+ // renaming is to note that on the path through BB1, t2 is really just a
+ // copy of t1, and do that renaming, properly generating:
+ //
+ // t2 = PHI t1(BB1), f1(BB2)
+ // t3 = PHI t1(BB1), f2(BB2)
+ //
+ // Case 2, we lower cascaded CMOVs such as
+ //
// (CMOV (CMOV F, T, cc1), T, cc2)
+ //
// to two successives branches. For that, we look for another CMOV as the
// following instruction.
//
// .LBB5_4:
// retq
//
- MachineInstr *NextCMOV = nullptr;
+ MachineInstr *CascadedCMOV = nullptr;
+ MachineInstr *LastCMOV = MI;
+ X86::CondCode CC = X86::CondCode(MI->getOperand(3).getImm());
+ X86::CondCode OppCC = X86::GetOppositeBranchCondition(CC);
MachineBasicBlock::iterator NextMIIt =
std::next(MachineBasicBlock::iterator(MI));
- if (NextMIIt != BB->end() && NextMIIt->getOpcode() == MI->getOpcode() &&
+
+ // Check for case 1, where there are multiple CMOVs with the same condition
+ // first. Of the two cases of multiple CMOV lowerings, case 1 reduces the
+ // number of jumps the most.
+
+ if (isCMOVPseudo(MI)) {
+ // See if we have a string of CMOVS with the same condition.
+ while (NextMIIt != BB->end() &&
+ isCMOVPseudo(NextMIIt) &&
+ (NextMIIt->getOperand(3).getImm() == CC ||
+ NextMIIt->getOperand(3).getImm() == OppCC)) {
+ LastCMOV = &*NextMIIt;
+ ++NextMIIt;
+ }
+ }
+
+ // This checks for case 2, but only do this if we didn't already find
+ // case 1, as indicated by LastCMOV == MI.
+ if (LastCMOV == MI &&
+ NextMIIt != BB->end() && NextMIIt->getOpcode() == MI->getOpcode() &&
NextMIIt->getOperand(2).getReg() == MI->getOperand(2).getReg() &&
- NextMIIt->getOperand(1).getReg() == MI->getOperand(0).getReg())
- NextCMOV = &*NextMIIt;
+ NextMIIt->getOperand(1).getReg() == MI->getOperand(0).getReg()) {
+ CascadedCMOV = &*NextMIIt;
+ }
MachineBasicBlock *jcc1MBB = nullptr;
- // If we have a double CMOV, we lower it to two successive branches to
+ // If we have a cascaded CMOV, we lower it to two successive branches to
// the same block. EFLAGS is used by both, so mark it as live in the second.
- if (NextCMOV) {
+ if (CascadedCMOV) {
jcc1MBB = F->CreateMachineBasicBlock(LLVM_BB);
F->insert(It, jcc1MBB);
jcc1MBB->addLiveIn(X86::EFLAGS);
// live into the sink and copy blocks.
const TargetRegisterInfo *TRI = Subtarget->getRegisterInfo();
- MachineInstr *LastEFLAGSUser = NextCMOV ? NextCMOV : MI;
+ MachineInstr *LastEFLAGSUser = CascadedCMOV ? CascadedCMOV : LastCMOV;
if (!LastEFLAGSUser->killsRegister(X86::EFLAGS) &&
!checkAndUpdateEFLAGSKill(LastEFLAGSUser, BB, TRI)) {
copy0MBB->addLiveIn(X86::EFLAGS);
// Transfer the remainder of BB and its successor edges to sinkMBB.
sinkMBB->splice(sinkMBB->begin(), BB,
- std::next(MachineBasicBlock::iterator(MI)), BB->end());
+ std::next(MachineBasicBlock::iterator(LastCMOV)), BB->end());
sinkMBB->transferSuccessorsAndUpdatePHIs(BB);
// Add the true and fallthrough blocks as its successors.
- if (NextCMOV) {
- // The fallthrough block may be jcc1MBB, if we have a double CMOV.
+ if (CascadedCMOV) {
+ // The fallthrough block may be jcc1MBB, if we have a cascaded CMOV.
BB->addSuccessor(jcc1MBB);
// In that case, jcc1MBB will itself fallthrough the copy0MBB, and
BB->addSuccessor(sinkMBB);
// Create the conditional branch instruction.
- unsigned Opc =
- X86::GetCondBranchFromCond((X86::CondCode)MI->getOperand(3).getImm());
+ unsigned Opc = X86::GetCondBranchFromCond(CC);
BuildMI(BB, DL, TII->get(Opc)).addMBB(sinkMBB);
- if (NextCMOV) {
+ if (CascadedCMOV) {
unsigned Opc2 = X86::GetCondBranchFromCond(
- (X86::CondCode)NextCMOV->getOperand(3).getImm());
+ (X86::CondCode)CascadedCMOV->getOperand(3).getImm());
BuildMI(jcc1MBB, DL, TII->get(Opc2)).addMBB(sinkMBB);
}
// sinkMBB:
// %Result = phi [ %FalseValue, copy0MBB ], [ %TrueValue, thisMBB ]
// ...
- MachineInstrBuilder MIB =
- BuildMI(*sinkMBB, sinkMBB->begin(), DL, TII->get(X86::PHI),
- MI->getOperand(0).getReg())
- .addReg(MI->getOperand(1).getReg()).addMBB(copy0MBB)
- .addReg(MI->getOperand(2).getReg()).addMBB(thisMBB);
+ MachineBasicBlock::iterator MIItBegin = MachineBasicBlock::iterator(MI);
+ MachineBasicBlock::iterator MIItEnd =
+ std::next(MachineBasicBlock::iterator(LastCMOV));
+ MachineBasicBlock::iterator SinkInsertionPoint = sinkMBB->begin();
+ DenseMap<unsigned, std::pair<unsigned, unsigned>> RegRewriteTable;
+ MachineInstrBuilder MIB;
+
+ // As we are creating the PHIs, we have to be careful if there is more than
+ // one. Later CMOVs may reference the results of earlier CMOVs, but later
+ // PHIs have to reference the individual true/false inputs from earlier PHIs.
+ // That also means that PHI construction must work forward from earlier to
+ // later, and that the code must maintain a mapping from earlier PHI's
+ // destination registers, and the registers that went into the PHI.
+
+ for (MachineBasicBlock::iterator MIIt = MIItBegin; MIIt != MIItEnd; ++MIIt) {
+ unsigned DestReg = MIIt->getOperand(0).getReg();
+ unsigned Op1Reg = MIIt->getOperand(1).getReg();
+ unsigned Op2Reg = MIIt->getOperand(2).getReg();
+
+ // If this CMOV we are generating is the opposite condition from
+ // the jump we generated, then we have to swap the operands for the
+ // PHI that is going to be generated.
+ if (MIIt->getOperand(3).getImm() == OppCC)
+ std::swap(Op1Reg, Op2Reg);
- // If we have a double CMOV, the second Jcc provides the same incoming
+ if (RegRewriteTable.find(Op1Reg) != RegRewriteTable.end())
+ Op1Reg = RegRewriteTable[Op1Reg].first;
+
+ if (RegRewriteTable.find(Op2Reg) != RegRewriteTable.end())
+ Op2Reg = RegRewriteTable[Op2Reg].second;
+
+ MIB = BuildMI(*sinkMBB, SinkInsertionPoint, DL,
+ TII->get(X86::PHI), DestReg)
+ .addReg(Op1Reg).addMBB(copy0MBB)
+ .addReg(Op2Reg).addMBB(thisMBB);
+
+ // Add this PHI to the rewrite table.
+ RegRewriteTable[DestReg] = std::make_pair(Op1Reg, Op2Reg);
+ }
+
+ // If we have a cascaded CMOV, the second Jcc provides the same incoming
// value as the first Jcc (the True operand of the SELECT_CC/CMOV nodes).
- if (NextCMOV) {
+ if (CascadedCMOV) {
MIB.addReg(MI->getOperand(2).getReg()).addMBB(jcc1MBB);
// Copy the PHI result to the register defined by the second CMOV.
BuildMI(*sinkMBB, std::next(MachineBasicBlock::iterator(MIB.getInstr())),
- DL, TII->get(TargetOpcode::COPY), NextCMOV->getOperand(0).getReg())
+ DL, TII->get(TargetOpcode::COPY),
+ CascadedCMOV->getOperand(0).getReg())
.addReg(MI->getOperand(0).getReg());
- NextCMOV->eraseFromParent();
+ CascadedCMOV->eraseFromParent();
}
- MI->eraseFromParent(); // The pseudo instruction is gone now.
+ // Now remove the CMOV(s).
+ for (MachineBasicBlock::iterator MIIt = MIItBegin; MIIt != MIItEnd; )
+ (MIIt++)->eraseFromParent();
+
return sinkMBB;
}
+MachineBasicBlock *
+X86TargetLowering::EmitLoweredAtomicFP(MachineInstr *MI,
+ MachineBasicBlock *BB) const {
+ // Combine the following atomic floating-point modification pattern:
+ // a.store(reg OP a.load(acquire), release)
+ // Transform them into:
+ // OPss (%gpr), %xmm
+ // movss %xmm, (%gpr)
+ // Or sd equivalent for 64-bit operations.
+ unsigned MOp, FOp;
+ switch (MI->getOpcode()) {
+ default: llvm_unreachable("unexpected instr type for EmitLoweredAtomicFP");
+ case X86::RELEASE_FADD32mr: MOp = X86::MOVSSmr; FOp = X86::ADDSSrm; break;
+ case X86::RELEASE_FADD64mr: MOp = X86::MOVSDmr; FOp = X86::ADDSDrm; break;
+ }
+ const X86InstrInfo *TII = Subtarget->getInstrInfo();
+ DebugLoc DL = MI->getDebugLoc();
+ MachineRegisterInfo &MRI = BB->getParent()->getRegInfo();
+ unsigned MSrc = MI->getOperand(0).getReg();
+ unsigned VSrc = MI->getOperand(5).getReg();
+ MachineInstrBuilder MIM = BuildMI(*BB, MI, DL, TII->get(MOp))
+ .addReg(/*Base=*/MSrc)
+ .addImm(/*Scale=*/1)
+ .addReg(/*Index=*/0)
+ .addImm(0)
+ .addReg(0);
+ MachineInstr *MIO = BuildMI(*BB, (MachineInstr *)MIM, DL, TII->get(FOp),
+ MRI.createVirtualRegister(MRI.getRegClass(VSrc)))
+ .addReg(VSrc)
+ .addReg(/*Base=*/MSrc)
+ .addImm(/*Scale=*/1)
+ .addReg(/*Index=*/0)
+ .addImm(/*Disp=*/0)
+ .addReg(/*Segment=*/0);
+ MIM.addReg(MIO->getOperand(0).getReg(), RegState::Kill);
+ MI->eraseFromParent(); // The pseudo instruction is gone now.
+ return BB;
+}
+
MachineBasicBlock *
X86TargetLowering::EmitLoweredSegAlloca(MachineInstr *MI,
MachineBasicBlock *BB) const {
case X86::TLSCall_32:
case X86::TLSCall_64:
return EmitLoweredTLSCall(MI, BB);
- case X86::CMOV_GR8:
case X86::CMOV_FR32:
case X86::CMOV_FR64:
- case X86::CMOV_V4F32:
+ case X86::CMOV_GR8:
+ case X86::CMOV_GR16:
+ case X86::CMOV_GR32:
+ case X86::CMOV_RFP32:
+ case X86::CMOV_RFP64:
+ case X86::CMOV_RFP80:
case X86::CMOV_V2F64:
case X86::CMOV_V2I64:
- case X86::CMOV_V8F32:
+ case X86::CMOV_V4F32:
case X86::CMOV_V4F64:
case X86::CMOV_V4I64:
case X86::CMOV_V16F32:
+ case X86::CMOV_V8F32:
case X86::CMOV_V8F64:
case X86::CMOV_V8I64:
- case X86::CMOV_GR16:
- case X86::CMOV_GR32:
- case X86::CMOV_RFP32:
- case X86::CMOV_RFP64:
- case X86::CMOV_RFP80:
case X86::CMOV_V8I1:
case X86::CMOV_V16I1:
case X86::CMOV_V32I1:
case X86::CMOV_V64I1:
return EmitLoweredSelect(MI, BB);
+ case X86::RELEASE_FADD32mr:
+ case X86::RELEASE_FADD64mr:
+ return EmitLoweredAtomicFP(MI, BB);
+
case X86::FP32_TO_INT16_IN_MEM:
case X86::FP32_TO_INT32_IN_MEM:
case X86::FP32_TO_INT64_IN_MEM:
return SDValue();
}
-/// \brief Matches a VSELECT onto min/max or return 0 if the node doesn't match.
-static std::pair<unsigned, bool>
-matchIntegerMINMAX(SDValue Cond, EVT VT, SDValue LHS, SDValue RHS,
- SelectionDAG &DAG, const X86Subtarget *Subtarget) {
- if (!VT.isVector())
- return std::make_pair(0, false);
-
- 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:
- if (!Subtarget->hasAVX2())
- NeedSplit = true;
- if (!Subtarget->hasAVX())
- return std::make_pair(0, false);
- break;
- case MVT::v16i8:
- case MVT::v8i16:
- case MVT::v4i32:
- if (!Subtarget->hasSSE2())
- return std::make_pair(0, false);
- }
-
- // SSE2 has only a small subset of the operations.
- bool hasUnsigned = Subtarget->hasSSE41() ||
- (Subtarget->hasSSE2() && VT == MVT::v16i8);
- bool hasSigned = Subtarget->hasSSE41() ||
- (Subtarget->hasSSE2() && VT == MVT::v8i16);
-
- ISD::CondCode CC = cast<CondCodeSDNode>(Cond.getOperand(2))->get();
-
- unsigned Opc = 0;
- // Check for x CC y ? x : y.
- if (DAG.isEqualTo(LHS, Cond.getOperand(0)) &&
- DAG.isEqualTo(RHS, Cond.getOperand(1))) {
- switch (CC) {
- default: break;
- case ISD::SETULT:
- case ISD::SETULE:
- Opc = hasUnsigned ? ISD::UMIN : 0; break;
- case ISD::SETUGT:
- case ISD::SETUGE:
- Opc = hasUnsigned ? ISD::UMAX : 0; break;
- case ISD::SETLT:
- case ISD::SETLE:
- Opc = hasSigned ? ISD::SMIN : 0; break;
- case ISD::SETGT:
- case ISD::SETGE:
- Opc = hasSigned ? ISD::SMAX : 0; break;
- }
- // Check for x CC y ? y : x -- a min/max with reversed arms.
- } else if (DAG.isEqualTo(LHS, Cond.getOperand(1)) &&
- DAG.isEqualTo(RHS, Cond.getOperand(0))) {
- switch (CC) {
- default: break;
- case ISD::SETULT:
- case ISD::SETULE:
- Opc = hasUnsigned ? ISD::UMAX : 0; break;
- case ISD::SETUGT:
- case ISD::SETUGE:
- Opc = hasUnsigned ? ISD::UMIN : 0; break;
- case ISD::SETLT:
- case ISD::SETLE:
- Opc = hasSigned ? ISD::SMAX : 0; break;
- case ISD::SETGT:
- case ISD::SETGE:
- Opc = hasSigned ? ISD::SMIN : 0; break;
- }
- }
-
- return std::make_pair(Opc, NeedSplit);
-}
-
static SDValue
transformVSELECTtoBlendVECTOR_SHUFFLE(SDNode *N, SelectionDAG &DAG,
const X86Subtarget *Subtarget) {
}
}
- // Try to match a min/max vector operation.
- if (N->getOpcode() == ISD::VSELECT && Cond.getOpcode() == ISD::SETCC) {
- std::pair<unsigned, bool> ret = matchIntegerMINMAX(Cond, VT, LHS, RHS, DAG, Subtarget);
- unsigned Opc = ret.first;
- bool NeedSplit = ret.second;
-
- if (Opc && NeedSplit) {
- unsigned NumElems = VT.getVectorNumElements();
- // Extract the LHS vectors
- SDValue LHS1 = Extract128BitVector(LHS, 0, DAG, DL);
- SDValue LHS2 = Extract128BitVector(LHS, NumElems/2, DAG, DL);
-
- // Extract the RHS vectors
- SDValue RHS1 = Extract128BitVector(RHS, 0, DAG, DL);
- SDValue RHS2 = Extract128BitVector(RHS, NumElems/2, DAG, DL);
-
- // Create min/max for each subvector
- LHS = DAG.getNode(Opc, DL, LHS1.getValueType(), LHS1, RHS1);
- RHS = DAG.getNode(Opc, DL, LHS2.getValueType(), LHS2, RHS2);
-
- // Merge the result
- return DAG.getNode(ISD::CONCAT_VECTORS, DL, VT, LHS, RHS);
- } else if (Opc)
- return DAG.getNode(Opc, DL, VT, LHS, RHS);
- }
-
// Simplify vector selection if condition value type matches vselect
// operand type
if (N->getOpcode() == ISD::VSELECT && CondVT == VT) {
return SDValue();
}
-static SDValue PerformINTRINSIC_WO_CHAINCombine(SDNode *N, SelectionDAG &DAG,
- const X86Subtarget *Subtarget) {
- unsigned IntNo = cast<ConstantSDNode>(N->getOperand(0))->getZExtValue();
- switch (IntNo) {
- default: return SDValue();
- // SSE/AVX/AVX2 blend intrinsics.
- case Intrinsic::x86_avx2_pblendvb:
- // Don't try to simplify this intrinsic if we don't have AVX2.
- if (!Subtarget->hasAVX2())
- return SDValue();
- // FALL-THROUGH
- case Intrinsic::x86_avx_blendv_pd_256:
- case Intrinsic::x86_avx_blendv_ps_256:
- // Don't try to simplify this intrinsic if we don't have AVX.
- if (!Subtarget->hasAVX())
- return SDValue();
- // FALL-THROUGH
- case Intrinsic::x86_sse41_blendvps:
- case Intrinsic::x86_sse41_blendvpd:
- case Intrinsic::x86_sse41_pblendvb: {
- SDValue Op0 = N->getOperand(1);
- SDValue Op1 = N->getOperand(2);
- SDValue Mask = N->getOperand(3);
-
- // Don't try to simplify this intrinsic if we don't have SSE4.1.
- if (!Subtarget->hasSSE41())
- return SDValue();
-
- // fold (blend A, A, Mask) -> A
- if (Op0 == Op1)
- return Op0;
- // fold (blend A, B, allZeros) -> A
- if (ISD::isBuildVectorAllZeros(Mask.getNode()))
- return Op0;
- // fold (blend A, B, allOnes) -> B
- if (ISD::isBuildVectorAllOnes(Mask.getNode()))
- return Op1;
-
- // Simplify the case where the mask is a constant i32 value.
- if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Mask)) {
- if (C->isNullValue())
- return Op0;
- if (C->isAllOnesValue())
- return Op1;
- }
-
- return SDValue();
- }
-
- // Packed SSE2/AVX2 arithmetic shift immediate intrinsics.
- case Intrinsic::x86_sse2_psrai_w:
- case Intrinsic::x86_sse2_psrai_d:
- case Intrinsic::x86_avx2_psrai_w:
- case Intrinsic::x86_avx2_psrai_d:
- case Intrinsic::x86_sse2_psra_w:
- case Intrinsic::x86_sse2_psra_d:
- case Intrinsic::x86_avx2_psra_w:
- case Intrinsic::x86_avx2_psra_d: {
- SDValue Op0 = N->getOperand(1);
- SDValue Op1 = N->getOperand(2);
- EVT VT = Op0.getValueType();
- assert(VT.isVector() && "Expected a vector type!");
-
- if (isa<BuildVectorSDNode>(Op1))
- Op1 = Op1.getOperand(0);
-
- if (!isa<ConstantSDNode>(Op1))
- return SDValue();
-
- EVT SVT = VT.getVectorElementType();
- unsigned SVTBits = SVT.getSizeInBits();
-
- ConstantSDNode *CND = cast<ConstantSDNode>(Op1);
- const APInt &C = APInt(SVTBits, CND->getAPIntValue().getZExtValue());
- uint64_t ShAmt = C.getZExtValue();
-
- // Don't try to convert this shift into a ISD::SRA if the shift
- // count is bigger than or equal to the element size.
- if (ShAmt >= SVTBits)
- return SDValue();
-
- // Trivial case: if the shift count is zero, then fold this
- // into the first operand.
- if (ShAmt == 0)
- return Op0;
-
- // Replace this packed shift intrinsic with a target independent
- // shift dag node.
- SDLoc DL(N);
- SDValue Splat = DAG.getConstant(C, DL, VT);
- return DAG.getNode(ISD::SRA, DL, VT, Op0, Splat);
- }
- }
-}
-
/// PerformMulCombine - Optimize a single multiply with constant into two
/// in order to implement it with two cheaper instructions, e.g.
/// LEA + SHL, LEA + LEA.
static SDValue PerformMulCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI) {
+ // An imul is usually smaller than the alternative sequence.
+ if (DAG.getMachineFunction().getFunction()->optForMinSize())
+ return SDValue();
+
if (DCI.isBeforeLegalize() || DCI.isCalledByLegalizer())
return SDValue();
N1C && N0.getOpcode() == ISD::AND &&
N0.getOperand(1).getOpcode() == ISD::Constant) {
SDValue N00 = N0.getOperand(0);
- if (N00.getOpcode() == X86ISD::SETCC_CARRY ||
- ((N00.getOpcode() == ISD::ANY_EXTEND ||
- N00.getOpcode() == ISD::ZERO_EXTEND) &&
- N00.getOperand(0).getOpcode() == X86ISD::SETCC_CARRY)) {
- APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
- APInt ShAmt = N1C->getAPIntValue();
- Mask = Mask.shl(ShAmt);
- if (Mask != 0) {
- SDLoc DL(N);
- return DAG.getNode(ISD::AND, DL, VT,
- N00, DAG.getConstant(Mask, DL, VT));
- }
+ APInt Mask = cast<ConstantSDNode>(N0.getOperand(1))->getAPIntValue();
+ APInt ShAmt = N1C->getAPIntValue();
+ Mask = Mask.shl(ShAmt);
+ bool MaskOK = false;
+ // We can handle cases concerning bit-widening nodes containing setcc_c if
+ // we carefully interrogate the mask to make sure we are semantics
+ // preserving.
+ // The transform is not safe if the result of C1 << C2 exceeds the bitwidth
+ // of the underlying setcc_c operation if the setcc_c was zero extended.
+ // Consider the following example:
+ // zext(setcc_c) -> i32 0x0000FFFF
+ // c1 -> i32 0x0000FFFF
+ // c2 -> i32 0x00000001
+ // (shl (and (setcc_c), c1), c2) -> i32 0x0001FFFE
+ // (and setcc_c, (c1 << c2)) -> i32 0x0000FFFE
+ if (N00.getOpcode() == X86ISD::SETCC_CARRY) {
+ MaskOK = true;
+ } else if (N00.getOpcode() == ISD::SIGN_EXTEND &&
+ N00.getOperand(0).getOpcode() == X86ISD::SETCC_CARRY) {
+ MaskOK = true;
+ } else if ((N00.getOpcode() == ISD::ZERO_EXTEND ||
+ N00.getOpcode() == ISD::ANY_EXTEND) &&
+ N00.getOperand(0).getOpcode() == X86ISD::SETCC_CARRY) {
+ MaskOK = Mask.isIntN(N00.getOperand(0).getValueSizeInBits());
+ }
+ if (MaskOK && Mask != 0) {
+ SDLoc DL(N);
+ return DAG.getNode(ISD::AND, DL, VT, N00, DAG.getConstant(Mask, DL, VT));
}
}
return SDValue();
// fold (or (x << c) | (y >> (64 - c))) ==> (shld64 x, y, c)
- MachineFunction &MF = DAG.getMachineFunction();
- bool OptForSize =
- MF.getFunction()->hasFnAttribute(Attribute::OptimizeForSize);
+ bool OptForSize = DAG.getMachineFunction().getFunction()->optForSize();
// SHLD/SHRD instructions have lower register pressure, but on some
// platforms they have higher latency than the equivalent
return SDValue();
}
-// PerformXorCombine - Attempts to turn XOR nodes into BLSMSK nodes
+// Try to turn tests against the signbit in the form of:
+// XOR(TRUNCATE(SRL(X, size(X)-1)), 1)
+// into:
+// SETGT(X, -1)
+static SDValue foldXorTruncShiftIntoCmp(SDNode *N, SelectionDAG &DAG) {
+ // This is only worth doing if the output type is i8.
+ if (N->getValueType(0) != MVT::i8)
+ return SDValue();
+
+ SDValue N0 = N->getOperand(0);
+ SDValue N1 = N->getOperand(1);
+
+ // We should be performing an xor against a truncated shift.
+ if (N0.getOpcode() != ISD::TRUNCATE || !N0.hasOneUse())
+ return SDValue();
+
+ // Make sure we are performing an xor against one.
+ if (!isa<ConstantSDNode>(N1) || !cast<ConstantSDNode>(N1)->isOne())
+ return SDValue();
+
+ // SetCC on x86 zero extends so only act on this if it's a logical shift.
+ SDValue Shift = N0.getOperand(0);
+ if (Shift.getOpcode() != ISD::SRL || !Shift.hasOneUse())
+ return SDValue();
+
+ // Make sure we are truncating from one of i16, i32 or i64.
+ EVT ShiftTy = Shift.getValueType();
+ if (ShiftTy != MVT::i16 && ShiftTy != MVT::i32 && ShiftTy != MVT::i64)
+ return SDValue();
+
+ // Make sure the shift amount extracts the sign bit.
+ if (!isa<ConstantSDNode>(Shift.getOperand(1)) ||
+ Shift.getConstantOperandVal(1) != ShiftTy.getSizeInBits() - 1)
+ return SDValue();
+
+ // Create a greater-than comparison against -1.
+ // N.B. Using SETGE against 0 works but we want a canonical looking
+ // comparison, using SETGT matches up with what TranslateX86CC.
+ SDLoc DL(N);
+ SDValue ShiftOp = Shift.getOperand(0);
+ EVT ShiftOpTy = ShiftOp.getValueType();
+ SDValue Cond = DAG.getSetCC(DL, MVT::i8, ShiftOp,
+ DAG.getConstant(-1, DL, ShiftOpTy), ISD::SETGT);
+ return Cond;
+}
+
static SDValue PerformXorCombine(SDNode *N, SelectionDAG &DAG,
TargetLowering::DAGCombinerInfo &DCI,
const X86Subtarget *Subtarget) {
if (DCI.isBeforeLegalizeOps())
return SDValue();
+ if (SDValue RV = foldXorTruncShiftIntoCmp(N, DAG))
+ return RV;
+
if (Subtarget->hasCMov())
if (SDValue RV = performIntegerAbsCombine(N, DAG))
return RV;
// For chips with slow 32-byte unaligned loads, break the 32-byte operation
// into two 16-byte operations.
ISD::LoadExtType Ext = Ld->getExtensionType();
+ bool Fast;
+ unsigned AddressSpace = Ld->getAddressSpace();
unsigned Alignment = Ld->getAlignment();
- bool IsAligned = Alignment == 0 || Alignment >= MemVT.getSizeInBits()/8;
- if (RegVT.is256BitVector() && Subtarget->isUnalignedMem32Slow() &&
- !DCI.isBeforeLegalizeOps() && !IsAligned && Ext == ISD::NON_EXTLOAD) {
+ if (RegVT.is256BitVector() && !DCI.isBeforeLegalizeOps() &&
+ Ext == ISD::NON_EXTLOAD &&
+ TLI.allowsMemoryAccess(*DAG.getContext(), DAG.getDataLayout(), RegVT,
+ AddressSpace, Alignment, &Fast) && !Fast) {
unsigned NumElems = RegVT.getVectorNumElements();
if (NumElems < 2)
return SDValue();
unsigned FromSz = VT.getVectorElementType().getSizeInBits();
unsigned ToSz = StVT.getVectorElementType().getSizeInBits();
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+
+ // The truncating store is legal in some cases. For example
+ // vpmovqb, vpmovqw, vpmovqd, vpmovdb, vpmovdw
+ // are designated for truncate store.
+ // In this case we don't need any further transformations.
+ if (TLI.isTruncStoreLegal(VT, StVT))
+ return SDValue();
+
// From, To sizes and ElemCount must be pow of two
assert (isPowerOf2_32(NumElems * FromSz * ToSz) &&
"Unexpected size for truncating masked store");
// If we are saving a concatenation of two XMM registers and 32-byte stores
// are slow, such as on Sandy Bridge, perform two 16-byte stores.
+ bool Fast;
+ unsigned AddressSpace = St->getAddressSpace();
unsigned Alignment = St->getAlignment();
- bool IsAligned = Alignment == 0 || Alignment >= VT.getSizeInBits()/8;
- if (VT.is256BitVector() && Subtarget->isUnalignedMem32Slow() &&
- StVT == VT && !IsAligned) {
+ if (VT.is256BitVector() && StVT == VT &&
+ TLI.allowsMemoryAccess(*DAG.getContext(), DAG.getDataLayout(), VT,
+ AddressSpace, Alignment, &Fast) && !Fast) {
unsigned NumElems = VT.getVectorNumElements();
if (NumElems < 2)
return SDValue();
unsigned FromSz = VT.getVectorElementType().getSizeInBits();
unsigned ToSz = StVT.getVectorElementType().getSizeInBits();
+ // The truncating store is legal in some cases. For example
+ // vpmovqb, vpmovqw, vpmovqd, vpmovdb, vpmovdw
+ // are designated for truncate store.
+ // In this case we don't need any further transformations.
+ if (TLI.isTruncStoreLegal(VT, StVT))
+ return SDValue();
+
// From, To sizes and ElemCount must be pow of two
if (!isPowerOf2_32(NumElems * FromSz * ToSz)) return SDValue();
// We are going to use the original vector elt for storing.
}
// Check if we can bypass extracting and re-inserting an element of an input
- // vector. Essentialy:
+ // vector. Essentially:
// (bitcast (sclr2vec (ext_vec_elt x))) -> (bitcast x)
if (V.getOpcode() == ISD::SCALAR_TO_VECTOR &&
V.getOperand(0).getOpcode() == ISD::EXTRACT_VECTOR_ELT &&
case X86ISD::VPERM2X128:
case ISD::VECTOR_SHUFFLE: return PerformShuffleCombine(N, DAG, DCI,Subtarget);
case ISD::FMA: return PerformFMACombine(N, DAG, Subtarget);
- case ISD::INTRINSIC_WO_CHAIN:
- return PerformINTRINSIC_WO_CHAINCombine(N, DAG, Subtarget);
case X86ISD::INSERTPS: {
if (getTargetMachine().getOptLevel() > CodeGenOpt::None)
return PerformINSERTPSCombine(N, DAG, Subtarget);
return -1;
}
-bool X86TargetLowering::isTargetFTOL() const {
- return Subtarget->isTargetKnownWindowsMSVC() && !Subtarget->is64Bit();
+bool X86TargetLowering::isIntDivCheap(EVT VT, AttributeSet Attr) const {
+ // Integer division on x86 is expensive. However, when aggressively optimizing
+ // for code size, we prefer to use a div instruction, as it is usually smaller
+ // than the alternative sequence.
+ // The exception to this is vector division. Since x86 doesn't have vector
+ // integer division, leaving the division as-is is a loss even in terms of
+ // size, because it will have to be scalarized, while the alternative code
+ // sequence can be performed in vector form.
+ bool OptSize = Attr.hasAttribute(AttributeSet::FunctionIndex,
+ Attribute::MinSize);
+ return OptSize && !VT.isVector();
}
projects / oota-llvm.git / blobdiff