//
//===----------------------------------------------------------------------===//
-#define DEBUG_TYPE "x86-isel"
#include "X86ISelLowering.h"
#include "Utils/X86ShuffleDecode.h"
-#include "X86.h"
#include "X86CallingConv.h"
#include "X86InstrBuilder.h"
+#include "X86MachineFunctionInfo.h"
#include "X86TargetMachine.h"
#include "X86TargetObjectFile.h"
#include "llvm/ADT/SmallSet.h"
#include <cctype>
using namespace llvm;
+#define DEBUG_TYPE "x86-isel"
+
STATISTIC(NumTailCalls, "Number of tail calls");
// Forward declarations.
// If the input is a buildvector just emit a smaller one.
if (Vec.getOpcode() == ISD::BUILD_VECTOR)
return DAG.getNode(ISD::BUILD_VECTOR, dl, ResultVT,
- Vec->op_begin()+NormalizedIdxVal, ElemsPerChunk);
+ ArrayRef<SDUse>(Vec->op_begin()+NormalizedIdxVal,
+ ElemsPerChunk));
SDValue VecIdx = DAG.getIntPtrConstant(NormalizedIdxVal);
SDValue Result = DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, ResultVT, Vec,
return new X86LinuxTargetObjectFile();
if (Subtarget->isTargetELF())
return new TargetLoweringObjectFileELF();
- if (Subtarget->isTargetWindows())
+ if (Subtarget->isTargetKnownWindowsMSVC())
return new X86WindowsTargetObjectFile();
if (Subtarget->isTargetCOFF())
return new TargetLoweringObjectFileCOFF();
addBypassSlowDiv(64, 16);
}
- if (Subtarget->isTargetWindows() && !Subtarget->isTargetCygMing()) {
+ if (Subtarget->isTargetKnownWindowsMSVC()) {
// Setup Windows compiler runtime calls.
setLibcallName(RTLIB::SDIV_I64, "_alldiv");
setLibcallName(RTLIB::UDIV_I64, "_aulldiv");
// The _ftol2 runtime function has an unusual calling conv, which
// is modeled by a special pseudo-instruction.
- setLibcallName(RTLIB::FPTOUINT_F64_I64, 0);
- setLibcallName(RTLIB::FPTOUINT_F32_I64, 0);
- setLibcallName(RTLIB::FPTOUINT_F64_I32, 0);
- setLibcallName(RTLIB::FPTOUINT_F32_I32, 0);
+ 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()) {
// Darwin should use _setjmp/_longjmp instead of setjmp/longjmp.
setUseUnderscoreSetJmp(false);
setUseUnderscoreLongJmp(false);
- } else if (Subtarget->isTargetMingw()) {
+ } else if (Subtarget->isTargetWindowsGNU()) {
// MS runtime is weird: it exports _setjmp, but longjmp!
setUseUnderscoreSetJmp(true);
setUseUnderscoreLongJmp(false);
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
- if (Subtarget->isOSWindows() && !Subtarget->isTargetMacho())
- setOperationAction(ISD::DYNAMIC_STACKALLOC, Subtarget->is64Bit() ?
- MVT::i64 : MVT::i32, Custom);
- else if (TM.Options.EnableSegmentedStacks)
- setOperationAction(ISD::DYNAMIC_STACKALLOC, Subtarget->is64Bit() ?
- MVT::i64 : MVT::i32, Custom);
- else
- setOperationAction(ISD::DYNAMIC_STACKALLOC, Subtarget->is64Bit() ?
- MVT::i64 : MVT::i32, Expand);
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, Subtarget->is64Bit() ?
+ MVT::i64 : MVT::i32, Custom);
if (!TM.Options.UseSoftFloat && X86ScalarSSEf64) {
// f32 and f64 use SSE.
setOperationAction(ISD::FRINT, VT, Expand);
setOperationAction(ISD::FNEARBYINT, VT, Expand);
setOperationAction(ISD::SMUL_LOHI, VT, Expand);
+ setOperationAction(ISD::MULHS, VT, Expand);
setOperationAction(ISD::UMUL_LOHI, VT, Expand);
+ setOperationAction(ISD::MULHU, VT, Expand);
setOperationAction(ISD::SDIVREM, VT, Expand);
setOperationAction(ISD::UDIVREM, VT, Expand);
setOperationAction(ISD::FPOW, VT, Expand);
setOperationAction(ISD::ADD, MVT::v2i64, Legal);
setOperationAction(ISD::MUL, MVT::v4i32, Custom);
setOperationAction(ISD::MUL, MVT::v2i64, Custom);
+ setOperationAction(ISD::UMUL_LOHI, MVT::v4i32, Custom);
+ setOperationAction(ISD::SMUL_LOHI, MVT::v4i32, Custom);
+ setOperationAction(ISD::MULHU, MVT::v8i16, Legal);
+ setOperationAction(ISD::MULHS, MVT::v8i16, Legal);
setOperationAction(ISD::SUB, MVT::v16i8, Legal);
setOperationAction(ISD::SUB, MVT::v8i16, Legal);
setOperationAction(ISD::SUB, MVT::v4i32, Legal);
setOperationAction(ISD::SHL, MVT::v4i32, Custom);
setOperationAction(ISD::SRA, MVT::v4i32, Custom);
-
- setOperationAction(ISD::SDIV, MVT::v8i16, Custom);
- setOperationAction(ISD::SDIV, MVT::v4i32, Custom);
}
if (!TM.Options.UseSoftFloat && Subtarget->hasFp256()) {
setOperationAction(ISD::SRA, MVT::v16i16, Custom);
setOperationAction(ISD::SRA, MVT::v32i8, Custom);
- setOperationAction(ISD::SDIV, MVT::v16i16, Custom);
-
setOperationAction(ISD::SETCC, MVT::v32i8, Custom);
setOperationAction(ISD::SETCC, MVT::v16i16, Custom);
setOperationAction(ISD::SETCC, MVT::v8i32, Custom);
setOperationAction(ISD::MUL, MVT::v16i16, Legal);
// Don't lower v32i8 because there is no 128-bit byte mul
- setOperationAction(ISD::VSELECT, MVT::v32i8, Legal);
+ setOperationAction(ISD::UMUL_LOHI, MVT::v8i32, Custom);
+ setOperationAction(ISD::SMUL_LOHI, MVT::v8i32, Custom);
+ setOperationAction(ISD::MULHU, MVT::v16i16, Legal);
+ setOperationAction(ISD::MULHS, MVT::v16i16, Legal);
- setOperationAction(ISD::SDIV, MVT::v8i32, Custom);
+ setOperationAction(ISD::VSELECT, MVT::v32i8, Legal);
} else {
setOperationAction(ISD::ADD, MVT::v4i64, Custom);
setOperationAction(ISD::ADD, MVT::v8i32, Custom);
setOperationAction(ISD::FNEG, MVT::v8f64, Custom);
setOperationAction(ISD::FMA, MVT::v8f64, Legal);
setOperationAction(ISD::FMA, MVT::v16f32, Legal);
- setOperationAction(ISD::SDIV, MVT::v16i32, Custom);
setOperationAction(ISD::FP_TO_SINT, MVT::i32, Legal);
setOperationAction(ISD::FP_TO_UINT, MVT::i32, Legal);
setOperationAction(ISD::FP_TO_SINT, MVT::v16i32, Legal);
setOperationAction(ISD::FP_TO_UINT, MVT::v16i32, Legal);
setOperationAction(ISD::FP_TO_UINT, MVT::v8i32, Legal);
+ setOperationAction(ISD::FP_TO_UINT, MVT::v4i32, Legal);
setOperationAction(ISD::SINT_TO_FP, MVT::v16i32, Legal);
setOperationAction(ISD::UINT_TO_FP, MVT::v16i32, Legal);
setOperationAction(ISD::UINT_TO_FP, MVT::v8i32, Legal);
+ setOperationAction(ISD::UINT_TO_FP, MVT::v4i32, Legal);
setOperationAction(ISD::FP_ROUND, MVT::v8f32, Legal);
setOperationAction(ISD::FP_EXTEND, MVT::v8f32, Legal);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v8i1, Custom);
setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v16i1, Custom);
+ setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v16i1, Custom);
+ setOperationAction(ISD::INSERT_VECTOR_ELT, MVT::v8i1, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v8i1, Custom);
setOperationAction(ISD::BUILD_VECTOR, MVT::v16i1, Custom);
setOperationAction(ISD::SELECT, MVT::v8f64, Custom);
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
+ if (!Subtarget->is64Bit())
+ setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::i64, Custom);
// Only custom-lower 64-bit SADDO and friends on 64-bit because we don't
// handle type legalization for these operations here.
if (!Subtarget->is64Bit()) {
// These libcalls are not available in 32-bit.
- setLibcallName(RTLIB::SHL_I128, 0);
- setLibcallName(RTLIB::SRL_I128, 0);
- setLibcallName(RTLIB::SRA_I128, 0);
+ setLibcallName(RTLIB::SHL_I128, nullptr);
+ setLibcallName(RTLIB::SRL_I128, nullptr);
+ setLibcallName(RTLIB::SRA_I128, nullptr);
}
// Combine sin / cos into one node or libcall if possible.
// FIXME: Why this routine is here? Move to RegInfo!
std::pair<const TargetRegisterClass*, uint8_t>
X86TargetLowering::findRepresentativeClass(MVT VT) const{
- const TargetRegisterClass *RRC = 0;
+ const TargetRegisterClass *RRC = nullptr;
uint8_t Cost = 1;
switch (VT.SimpleTy) {
default:
return CCInfo.CheckReturn(Outs, RetCC_X86);
}
-const uint16_t *X86TargetLowering::getScratchRegisters(CallingConv::ID) const {
- static const uint16_t ScratchRegs[] = { X86::R11, 0 };
+const MCPhysReg *X86TargetLowering::getScratchRegisters(CallingConv::ID) const {
+ static const MCPhysReg ScratchRegs[] = { X86::R11, 0 };
return ScratchRegs;
}
// We saved the argument into a virtual register in the entry block,
// so now we copy the value out and into %rax/%eax.
if (DAG.getMachineFunction().getFunction()->hasStructRetAttr() &&
- (Subtarget->is64Bit() || Subtarget->isTargetWindows())) {
+ (Subtarget->is64Bit() || Subtarget->isTargetKnownWindowsMSVC())) {
MachineFunction &MF = DAG.getMachineFunction();
X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
unsigned Reg = FuncInfo->getSRetReturnReg();
if (Flag.getNode())
RetOps.push_back(Flag);
- return DAG.getNode(X86ISD::RET_FLAG, dl,
- MVT::Other, &RetOps[0], RetOps.size());
+ return DAG.getNode(X86ISD::RET_FLAG, dl, MVT::Other, RetOps);
}
bool X86TargetLowering::isUsedByReturnOnly(SDNode *N, SDValue &Chain) const {
// Save the argument into a virtual register so that we can access it
// from the return points.
if (MF.getFunction()->hasStructRetAttr() &&
- (Subtarget->is64Bit() || Subtarget->isTargetWindows())) {
+ (Subtarget->is64Bit() || Subtarget->isTargetKnownWindowsMSVC())) {
X86MachineFunctionInfo *FuncInfo = MF.getInfo<X86MachineFunctionInfo>();
unsigned Reg = FuncInfo->getSRetReturnReg();
if (!Reg) {
unsigned TotalNumIntRegs = 0, TotalNumXMMRegs = 0;
// FIXME: We should really autogenerate these arrays
- static const uint16_t GPR64ArgRegsWin64[] = {
+ static const MCPhysReg GPR64ArgRegsWin64[] = {
X86::RCX, X86::RDX, X86::R8, X86::R9
};
- static const uint16_t GPR64ArgRegs64Bit[] = {
+ static const MCPhysReg GPR64ArgRegs64Bit[] = {
X86::RDI, X86::RSI, X86::RDX, X86::RCX, X86::R8, X86::R9
};
- static const uint16_t XMMArgRegs64Bit[] = {
+ static const MCPhysReg XMMArgRegs64Bit[] = {
X86::XMM0, X86::XMM1, X86::XMM2, X86::XMM3,
X86::XMM4, X86::XMM5, X86::XMM6, X86::XMM7
};
- const uint16_t *GPR64ArgRegs;
+ const MCPhysReg *GPR64ArgRegs;
unsigned NumXMMRegs = 0;
if (IsWin64) {
SaveXMMOps.push_back(Val);
}
MemOps.push_back(DAG.getNode(X86ISD::VASTART_SAVE_XMM_REGS, dl,
- MVT::Other,
- &SaveXMMOps[0], SaveXMMOps.size()));
+ MVT::Other, SaveXMMOps));
}
if (!MemOps.empty())
- Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
- &MemOps[0], MemOps.size());
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOps);
}
}
MF.getFunction()->hasStructRetAttr(), CLI.RetTy,
Outs, OutVals, Ins, DAG);
+ if (!isTailCall && CLI.CS && CLI.CS->isMustTailCall())
+ report_fatal_error("failed to perform tail call elimination on a call "
+ "site marked musttail");
+
// Sibcalls are automatically detected tailcalls which do not require
// ABI changes.
if (!MF.getTarget().Options.GuaranteedTailCallOpt && isTailCall)
}
} else if (!IsSibcall && (!isTailCall || isByVal)) {
assert(VA.isMemLoc());
- if (StackPtr.getNode() == 0)
+ if (!StackPtr.getNode())
StackPtr = DAG.getCopyFromReg(Chain, dl, RegInfo->getStackRegister(),
getPointerTy());
MemOpChains.push_back(LowerMemOpCallTo(Chain, StackPtr, Arg,
}
if (!MemOpChains.empty())
- Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
- &MemOpChains[0], MemOpChains.size());
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
if (Subtarget->isPICStyleGOT()) {
// ELF / PIC requires GOT in the EBX register before function calls via PLT
// registers used and is in the range 0 - 8 inclusive.
// Count the number of XMM registers allocated.
- static const uint16_t XMMArgRegs[] = {
+ static const MCPhysReg XMMArgRegs[] = {
X86::XMM0, X86::XMM1, X86::XMM2, X86::XMM3,
X86::XMM4, X86::XMM5, X86::XMM6, X86::XMM7
};
if (Flags.isByVal()) {
// Copy relative to framepointer.
SDValue Source = DAG.getIntPtrConstant(VA.getLocMemOffset());
- if (StackPtr.getNode() == 0)
+ if (!StackPtr.getNode())
StackPtr = DAG.getCopyFromReg(Chain, dl,
RegInfo->getStackRegister(),
getPointerTy());
}
if (!MemOpChains2.empty())
- Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other,
- &MemOpChains2[0], MemOpChains2.size());
+ Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains2);
// Store the return address to the appropriate stack slot.
Chain = EmitTailCallStoreRetAddr(DAG, MF, Chain, RetAddrFrIdx,
// This isn't right, although it's probably harmless on x86; liveouts
// should be computed from returns not tail calls. Consider a void
// function making a tail call to a function returning int.
- return DAG.getNode(X86ISD::TC_RETURN, dl, NodeTys, &Ops[0], Ops.size());
+ return DAG.getNode(X86ISD::TC_RETURN, dl, NodeTys, Ops);
}
- Chain = DAG.getNode(X86ISD::CALL, dl, NodeTys, &Ops[0], Ops.size());
+ Chain = DAG.getNode(X86ISD::CALL, dl, NodeTys, Ops);
InFlag = Chain.getValue(1);
// Create the CALLSEQ_END node.
return true;
}
+/// isINSERTPSMask - Return true if the specified VECTOR_SHUFFLE operand
+/// specifies a shuffle of elements that is suitable for input to INSERTPS.
+/// i. e: If all but one element come from the same vector.
+static bool isINSERTPSMask(ArrayRef<int> Mask, MVT VT) {
+ // TODO: Deal with AVX's VINSERTPS
+ if (!VT.is128BitVector() || (VT != MVT::v4f32 && VT != MVT::v4i32))
+ return false;
+
+ unsigned CorrectPosV1 = 0;
+ unsigned CorrectPosV2 = 0;
+ for (int i = 0, e = (int)VT.getVectorNumElements(); i != e; ++i)
+ if (Mask[i] == i)
+ ++CorrectPosV1;
+ else if (Mask[i] == i + 4)
+ ++CorrectPosV2;
+
+ if (CorrectPosV1 == 3 || CorrectPosV2 == 3)
+ // We have 3 elements from one vector, and one from another.
+ return true;
+
+ return false;
+}
+
//
// Some special combinations that can be optimized.
//
/// isScalarLoadToVector - Returns true if the node is a scalar load that
/// is promoted to a vector. It also returns the LoadSDNode by reference if
/// required.
-static bool isScalarLoadToVector(SDNode *N, LoadSDNode **LD = NULL) {
+static bool isScalarLoadToVector(SDNode *N, LoadSDNode **LD = nullptr) {
if (N->getOpcode() != ISD::SCALAR_TO_VECTOR)
return false;
N = N->getOperand(0).getNode();
if (Subtarget->hasInt256()) { // AVX2
SDValue Cst = DAG.getTargetConstant(0, MVT::i32);
SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
- Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops,
- array_lengthof(Ops));
+ Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops);
} else {
// 256-bit logic and arithmetic instructions in AVX are all
// floating-point, no support for integer ops. Emit fp zeroed vectors.
SDValue Cst = DAG.getTargetConstantFP(+0.0, MVT::f32);
SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
- Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8f32, Ops,
- array_lengthof(Ops));
+ Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8f32, Ops);
}
} else if (VT.is512BitVector()) { // AVX-512
SDValue Cst = DAG.getTargetConstant(0, MVT::i32);
SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst,
Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
- Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v16i32, Ops, 16);
+ Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v16i32, Ops);
} else if (VT.getScalarType() == MVT::i1) {
assert(VT.getVectorNumElements() <= 16 && "Unexpected vector type");
SDValue Cst = DAG.getTargetConstant(0, MVT::i1);
- SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst,
- Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
- return DAG.getNode(ISD::BUILD_VECTOR, dl, VT,
- Ops, VT.getVectorNumElements());
+ SmallVector<SDValue, 16> Ops(VT.getVectorNumElements(), Cst);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Ops);
} else
llvm_unreachable("Unexpected vector type");
if (VT.is256BitVector()) {
if (HasInt256) { // AVX2
SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
- Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops,
- array_lengthof(Ops));
+ Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v8i32, Ops);
} else { // AVX
Vec = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, Cst, Cst, Cst, Cst);
Vec = Concat128BitVectors(Vec, Vec, MVT::v8i32, 8, DAG, dl);
return SDValue();
SDLoc dl(Op);
- SDValue V(0, 0);
+ SDValue V;
bool First = true;
for (unsigned i = 0; i < 16; ++i) {
bool ThisIsNonZero = (NonZeros & (1 << i)) != 0;
}
if ((i & 1) != 0) {
- SDValue ThisElt(0, 0), LastElt(0, 0);
+ SDValue ThisElt, LastElt;
bool LastIsNonZero = (NonZeros & (1 << (i-1))) != 0;
if (LastIsNonZero) {
LastElt = DAG.getNode(ISD::ZERO_EXTEND, dl,
return SDValue();
SDLoc dl(Op);
- SDValue V(0, 0);
+ SDValue V;
bool First = true;
for (unsigned i = 0; i < 8; ++i) {
bool isNonZero = (NonZeros & (1 << i)) != 0;
EVT EltVT = VT.getVectorElementType();
unsigned NumElems = Elts.size();
- LoadSDNode *LDBase = NULL;
+ LoadSDNode *LDBase = nullptr;
unsigned LastLoadedElt = -1U;
// For each element in the initializer, see if we've found a load or an undef.
SDVTList Tys = DAG.getVTList(MVT::v2i64, MVT::Other);
SDValue Ops[] = { LDBase->getChain(), LDBase->getBasePtr() };
SDValue ResNode =
- DAG.getMemIntrinsicNode(X86ISD::VZEXT_LOAD, DL, Tys, Ops,
- array_lengthof(Ops), MVT::i64,
+ DAG.getMemIntrinsicNode(X86ISD::VZEXT_LOAD, DL, Tys, Ops, MVT::i64,
LDBase->getPointerInfo(),
LDBase->getAlignment(),
false/*isVolatile*/, true/*ReadMem*/,
unsigned ScalarSize = CVT.getSizeInBits();
if (ScalarSize == 32 || (IsGE256 && ScalarSize == 64)) {
- const Constant *C = 0;
+ const Constant *C = nullptr;
if (ConstantSDNode *CI = dyn_cast<ConstantSDNode>(Ld))
C = CI->getConstantIntValue();
else if (ConstantFPSDNode *CF = dyn_cast<ConstantFPSDNode>(Ld))
return SDValue();
}
+/// \brief For an EXTRACT_VECTOR_ELT with a constant index return the real
+/// underlying vector and index.
+///
+/// Modifies \p ExtractedFromVec to the real vector and returns the real
+/// index.
+static int getUnderlyingExtractedFromVec(SDValue &ExtractedFromVec,
+ SDValue ExtIdx) {
+ int Idx = cast<ConstantSDNode>(ExtIdx)->getZExtValue();
+ if (!isa<ShuffleVectorSDNode>(ExtractedFromVec))
+ return Idx;
+
+ // For 256-bit vectors, LowerEXTRACT_VECTOR_ELT_SSE4 may have already
+ // lowered this:
+ // (extract_vector_elt (v8f32 %vreg1), Constant<6>)
+ // to:
+ // (extract_vector_elt (vector_shuffle<2,u,u,u>
+ // (extract_subvector (v8f32 %vreg0), Constant<4>),
+ // undef)
+ // Constant<0>)
+ // In this case the vector is the extract_subvector expression and the index
+ // is 2, as specified by the shuffle.
+ ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(ExtractedFromVec);
+ SDValue ShuffleVec = SVOp->getOperand(0);
+ MVT ShuffleVecVT = ShuffleVec.getSimpleValueType();
+ assert(ShuffleVecVT.getVectorElementType() ==
+ ExtractedFromVec.getSimpleValueType().getVectorElementType());
+
+ int ShuffleIdx = SVOp->getMaskElt(Idx);
+ if (isUndefOrInRange(ShuffleIdx, 0, ShuffleVecVT.getVectorNumElements())) {
+ ExtractedFromVec = ShuffleVec;
+ return ShuffleIdx;
+ }
+ return Idx;
+}
+
static SDValue buildFromShuffleMostly(SDValue Op, SelectionDAG &DAG) {
MVT VT = Op.getSimpleValueType();
SDValue ExtractedFromVec = Op.getOperand(i).getOperand(0);
SDValue ExtIdx = Op.getOperand(i).getOperand(1);
+ // Quit if non-constant index.
+ if (!isa<ConstantSDNode>(ExtIdx))
+ return SDValue();
+ int Idx = getUnderlyingExtractedFromVec(ExtractedFromVec, ExtIdx);
// Quit if extracted from vector of different type.
if (ExtractedFromVec.getValueType() != VT)
return SDValue();
- // Quit if non-constant index.
- if (!isa<ConstantSDNode>(ExtIdx))
- return SDValue();
-
- if (VecIn1.getNode() == 0)
+ if (!VecIn1.getNode())
VecIn1 = ExtractedFromVec;
else if (VecIn1 != ExtractedFromVec) {
- if (VecIn2.getNode() == 0)
+ if (!VecIn2.getNode())
VecIn2 = ExtractedFromVec;
else if (VecIn2 != ExtractedFromVec)
// Quit if more than 2 vectors to shuffle
return SDValue();
}
- unsigned Idx = cast<ConstantSDNode>(ExtIdx)->getZExtValue();
-
if (ExtractedFromVec == VecIn1)
Mask[i] = Idx;
else if (ExtractedFromVec == VecIn2)
Mask[i] = Idx + NumElems;
}
- if (VecIn1.getNode() == 0)
+ if (!VecIn1.getNode())
return SDValue();
VecIn2 = VecIn2.getNode() ? VecIn2 : DAG.getUNDEF(VT);
SDLoc dl(Op);
if (ISD::isBuildVectorAllZeros(Op.getNode())) {
SDValue Cst = DAG.getTargetConstant(0, MVT::i1);
- SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst,
- Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
- return DAG.getNode(ISD::BUILD_VECTOR, dl, VT,
- Ops, VT.getVectorNumElements());
+ SmallVector<SDValue, 16> Ops(VT.getVectorNumElements(), Cst);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Ops);
}
if (ISD::isBuildVectorAllOnes(Op.getNode())) {
SDValue Cst = DAG.getTargetConstant(1, MVT::i1);
- SDValue Ops[] = { Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst,
- Cst, Cst, Cst, Cst, Cst, Cst, Cst, Cst };
- return DAG.getNode(ISD::BUILD_VECTOR, dl, VT,
- Ops, VT.getVectorNumElements());
+ SmallVector<SDValue, 16> Ops(VT.getVectorNumElements(), Cst);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Ops);
}
bool AllContants = true;
uint64_t Immediate = 0;
int NonConstIdx = -1;
bool IsSplat = true;
+ unsigned NumNonConsts = 0;
+ unsigned NumConsts = 0;
for (unsigned idx = 0, e = Op.getNumOperands(); idx < e; ++idx) {
SDValue In = Op.getOperand(idx);
if (In.getOpcode() == ISD::UNDEF)
if (!isa<ConstantSDNode>(In)) {
AllContants = false;
NonConstIdx = idx;
+ NumNonConsts++;
}
- else if (cast<ConstantSDNode>(In)->getZExtValue())
+ else {
+ NumConsts++;
+ if (cast<ConstantSDNode>(In)->getZExtValue())
Immediate |= (1ULL << idx);
+ }
if (In != Op.getOperand(0))
IsSplat = false;
}
DAG.getIntPtrConstant(0));
}
+ if (NumNonConsts == 1 && NonConstIdx != 0) {
+ SDValue DstVec;
+ if (NumConsts) {
+ SDValue VecAsImm = DAG.getConstant(Immediate,
+ MVT::getIntegerVT(VT.getSizeInBits()));
+ DstVec = DAG.getNode(ISD::BITCAST, dl, VT, VecAsImm);
+ }
+ else
+ DstVec = DAG.getUNDEF(VT);
+ return DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, DstVec,
+ Op.getOperand(NonConstIdx),
+ DAG.getIntPtrConstant(NonConstIdx));
+ }
if (!IsSplat && (NonConstIdx != 0))
llvm_unreachable("Unsupported BUILD_VECTOR operation");
MVT SelectVT = (VT == MVT::v16i1)? MVT::i16 : MVT::i8;
EVT HVT = EVT::getVectorVT(*DAG.getContext(), ExtVT, NumElems/2);
// Build both the lower and upper subvector.
- SDValue Lower = DAG.getNode(ISD::BUILD_VECTOR, dl, HVT, &V[0], NumElems/2);
- SDValue Upper = DAG.getNode(ISD::BUILD_VECTOR, dl, HVT, &V[NumElems / 2],
- NumElems/2);
+ SDValue Lower = DAG.getNode(ISD::BUILD_VECTOR, dl, HVT,
+ ArrayRef<SDValue>(&V[0], NumElems/2));
+ SDValue Upper = DAG.getNode(ISD::BUILD_VECTOR, dl, HVT,
+ ArrayRef<SDValue>(&V[NumElems / 2],
+ NumElems/2));
// Recreate the wider vector with the lower and upper part.
if (VT.is256BitVector())
return DAG.getNode(ISD::BITCAST, dl, VT, Ret);
}
+/// In vector type \p VT, return true if the element at index \p InputIdx
+/// falls on a different 128-bit lane than \p OutputIdx.
+static bool ShuffleCrosses128bitLane(MVT VT, unsigned InputIdx,
+ unsigned OutputIdx) {
+ unsigned EltSize = VT.getVectorElementType().getSizeInBits();
+ return InputIdx * EltSize / 128 != OutputIdx * EltSize / 128;
+}
+
+/// Generate a PSHUFB if possible. Selects elements from \p V1 according to
+/// \p MaskVals. MaskVals[OutputIdx] = InputIdx specifies that we want to
+/// shuffle the element at InputIdx in V1 to OutputIdx in the result. If \p
+/// MaskVals refers to elements outside of \p V1 or is undef (-1), insert a
+/// zero.
+static SDValue getPSHUFB(ArrayRef<int> MaskVals, SDValue V1, SDLoc &dl,
+ SelectionDAG &DAG) {
+ MVT VT = V1.getSimpleValueType();
+ assert(VT.is128BitVector() || VT.is256BitVector());
+
+ MVT EltVT = VT.getVectorElementType();
+ unsigned EltSizeInBytes = EltVT.getSizeInBits() / 8;
+ unsigned NumElts = VT.getVectorNumElements();
+
+ SmallVector<SDValue, 32> PshufbMask;
+ for (unsigned OutputIdx = 0; OutputIdx < NumElts; ++OutputIdx) {
+ int InputIdx = MaskVals[OutputIdx];
+ unsigned InputByteIdx;
+
+ if (InputIdx < 0 || NumElts <= (unsigned)InputIdx)
+ InputByteIdx = 0x80;
+ else {
+ // Cross lane is not allowed.
+ if (ShuffleCrosses128bitLane(VT, InputIdx, OutputIdx))
+ return SDValue();
+ InputByteIdx = InputIdx * EltSizeInBytes;
+ // Index is an byte offset within the 128-bit lane.
+ InputByteIdx &= 0xf;
+ }
+
+ for (unsigned j = 0; j < EltSizeInBytes; ++j) {
+ PshufbMask.push_back(DAG.getConstant(InputByteIdx, MVT::i8));
+ if (InputByteIdx != 0x80)
+ ++InputByteIdx;
+ }
+ }
+
+ MVT ShufVT = MVT::getVectorVT(MVT::i8, PshufbMask.size());
+ if (ShufVT != VT)
+ V1 = DAG.getNode(ISD::BITCAST, dl, ShufVT, V1);
+ return DAG.getNode(X86ISD::PSHUFB, dl, ShufVT, V1,
+ DAG.getNode(ISD::BUILD_VECTOR, dl, ShufVT, PshufbMask));
+}
+
// v8i16 shuffles - Prefer shuffles in the following order:
// 1. [all] pshuflw, pshufhw, optional move
// 2. [ssse3] 1 x pshufb
// Determine if more than 1 of the words in each of the low and high quadwords
// of the result come from the same quadword of one of the two inputs. Undef
// mask values count as coming from any quadword, for better codegen.
+ //
+ // Lo/HiQuad[i] = j indicates how many words from the ith quad of the input
+ // feeds this quad. For i, 0 and 1 refer to V1, 2 and 3 refer to V2.
unsigned LoQuad[] = { 0, 0, 0, 0 };
unsigned HiQuad[] = { 0, 0, 0, 0 };
+ // Indices of quads used.
std::bitset<4> InputQuads;
for (unsigned i = 0; i < 8; ++i) {
unsigned *Quad = i < 4 ? LoQuad : HiQuad;
// For SSSE3, If all 8 words of the result come from only 1 quadword of each
// of the two input vectors, shuffle them into one input vector so only a
- // single pshufb instruction is necessary. If There are more than 2 input
+ // single pshufb instruction is necessary. If there are more than 2 input
// quads, disable the next transformation since it does not help SSSE3.
bool V1Used = InputQuads[0] || InputQuads[1];
bool V2Used = InputQuads[2] || InputQuads[3];
// mask, and elements that come from V1 in the V2 mask, so that the two
// results can be OR'd together.
bool TwoInputs = V1Used && V2Used;
- for (unsigned i = 0; i != 8; ++i) {
- int EltIdx = MaskVals[i] * 2;
- int Idx0 = (TwoInputs && (EltIdx >= 16)) ? 0x80 : EltIdx;
- int Idx1 = (TwoInputs && (EltIdx >= 16)) ? 0x80 : EltIdx+1;
- pshufbMask.push_back(DAG.getConstant(Idx0, MVT::i8));
- pshufbMask.push_back(DAG.getConstant(Idx1, MVT::i8));
- }
- V1 = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, V1);
- V1 = DAG.getNode(X86ISD::PSHUFB, dl, MVT::v16i8, V1,
- DAG.getNode(ISD::BUILD_VECTOR, dl,
- MVT::v16i8, &pshufbMask[0], 16));
+ V1 = getPSHUFB(MaskVals, V1, dl, DAG);
if (!TwoInputs)
return DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, V1);
// Calculate the shuffle mask for the second input, shuffle it, and
// OR it with the first shuffled input.
- pshufbMask.clear();
- for (unsigned i = 0; i != 8; ++i) {
- int EltIdx = MaskVals[i] * 2;
- int Idx0 = (EltIdx < 16) ? 0x80 : EltIdx - 16;
- int Idx1 = (EltIdx < 16) ? 0x80 : EltIdx - 15;
- pshufbMask.push_back(DAG.getConstant(Idx0, MVT::i8));
- pshufbMask.push_back(DAG.getConstant(Idx1, MVT::i8));
- }
- V2 = DAG.getNode(ISD::BITCAST, dl, MVT::v16i8, V2);
- V2 = DAG.getNode(X86ISD::PSHUFB, dl, MVT::v16i8, V2,
- DAG.getNode(ISD::BUILD_VECTOR, dl,
- MVT::v16i8, &pshufbMask[0], 16));
+ CommuteVectorShuffleMask(MaskVals, 8);
+ V2 = getPSHUFB(MaskVals, V2, dl, DAG);
V1 = DAG.getNode(ISD::OR, dl, MVT::v16i8, V1, V2);
return DAG.getNode(ISD::BITCAST, dl, MVT::v8i16, V1);
}
return NewV;
}
+/// \brief v16i16 shuffles
+///
+/// FIXME: We only support generation of a single pshufb currently. We can
+/// generalize the other applicable cases from LowerVECTOR_SHUFFLEv8i16 as
+/// well (e.g 2 x pshufb + 1 x por).
+static SDValue
+LowerVECTOR_SHUFFLEv16i16(SDValue Op, SelectionDAG &DAG) {
+ ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
+ SDValue V1 = SVOp->getOperand(0);
+ SDValue V2 = SVOp->getOperand(1);
+ SDLoc dl(SVOp);
+
+ if (V2.getOpcode() != ISD::UNDEF)
+ return SDValue();
+
+ SmallVector<int, 16> MaskVals(SVOp->getMask().begin(), SVOp->getMask().end());
+ return getPSHUFB(MaskVals, V1, dl, DAG);
+}
+
// v16i8 shuffles - Prefer shuffles in the following order:
// 1. [ssse3] 1 x pshufb
// 2. [ssse3] 2 x pshufb + 1 x por
}
V1 = DAG.getNode(X86ISD::PSHUFB, dl, MVT::v16i8, V1,
DAG.getNode(ISD::BUILD_VECTOR, dl,
- MVT::v16i8, &pshufbMask[0], 16));
+ MVT::v16i8, pshufbMask));
// As PSHUFB will zero elements with negative indices, it's safe to ignore
// the 2nd operand if it's undefined or zero.
}
V2 = DAG.getNode(X86ISD::PSHUFB, dl, MVT::v16i8, V2,
DAG.getNode(ISD::BUILD_VECTOR, dl,
- MVT::v16i8, &pshufbMask[0], 16));
+ MVT::v16i8, pshufbMask));
return DAG.getNode(ISD::OR, dl, MVT::v16i8, V1, V2);
}
CommuteVectorShuffleMask(MaskVals, 32);
V1 = V2;
}
- SmallVector<SDValue, 32> pshufbMask;
- for (unsigned i = 0; i != 32; i++) {
- int EltIdx = MaskVals[i];
- if (EltIdx < 0 || EltIdx >= 32)
- EltIdx = 0x80;
- else {
- if ((EltIdx >= 16 && i < 16) || (EltIdx < 16 && i >= 16))
- // Cross lane is not allowed.
- return SDValue();
- EltIdx &= 0xf;
- }
- pshufbMask.push_back(DAG.getConstant(EltIdx, MVT::i8));
- }
- return DAG.getNode(X86ISD::PSHUFB, dl, MVT::v32i8, V1,
- DAG.getNode(ISD::BUILD_VECTOR, dl,
- MVT::v32i8, &pshufbMask[0], 32));
+ return getPSHUFB(MaskVals, V1, dl, DAG);
}
/// RewriteAsNarrowerShuffle - Try rewriting v8i16 and v16i8 shuffles as 4 wide
SDValue SrcOp, SelectionDAG &DAG,
const X86Subtarget *Subtarget, SDLoc dl) {
if (VT == MVT::v2f64 || VT == MVT::v4f32) {
- LoadSDNode *LD = NULL;
+ LoadSDNode *LD = nullptr;
if (!isScalarLoadToVector(SrcOp.getNode(), &LD))
LD = dyn_cast<LoadSDNode>(SrcOp);
if (!LD) {
}
// Construct the output using a BUILD_VECTOR.
- Output[l] = DAG.getNode(ISD::BUILD_VECTOR, dl, NVT, &SVOps[0],
- SVOps.size());
+ Output[l] = DAG.getNode(ISD::BUILD_VECTOR, dl, NVT, SVOps);
} else if (InputUsed[0] < 0) {
// No input vectors were used! The result is undefined.
Output[l] = DAG.getUNDEF(NVT);
getShuffleSHUFImmediate(SVOp), DAG);
}
+// It is only safe to call this function if isINSERTPSMask is true for
+// this shufflevector mask.
+static SDValue getINSERTPS(ShuffleVectorSDNode *SVOp, SDLoc &dl,
+ SelectionDAG &DAG) {
+ // Generate an insertps instruction when inserting an f32 from memory onto a
+ // v4f32 or when copying a member from one v4f32 to another.
+ // We also use it for transferring i32 from one register to another,
+ // since it simply copies the same bits.
+ // If we're transfering an i32 from memory to a specific element in a
+ // register, we output a generic DAG that will match the PINSRD
+ // instruction.
+ // TODO: Optimize for AVX cases too (VINSERTPS)
+ MVT VT = SVOp->getSimpleValueType(0);
+ MVT EVT = VT.getVectorElementType();
+ SDValue V1 = SVOp->getOperand(0);
+ SDValue V2 = SVOp->getOperand(1);
+ auto Mask = SVOp->getMask();
+ assert((VT == MVT::v4f32 || VT == MVT::v4i32) &&
+ "unsupported vector type for insertps/pinsrd");
+
+ int FromV1 = std::count_if(Mask.begin(), Mask.end(),
+ [](const int &i) { return i < 4; });
+
+ SDValue From;
+ SDValue To;
+ unsigned DestIndex;
+ if (FromV1 == 1) {
+ From = V1;
+ To = V2;
+ DestIndex = std::find_if(Mask.begin(), Mask.end(),
+ [](const int &i) { return i < 4; }) -
+ Mask.begin();
+ } else {
+ From = V2;
+ To = V1;
+ DestIndex = std::find_if(Mask.begin(), Mask.end(),
+ [](const int &i) { return i >= 4; }) -
+ Mask.begin();
+ }
+
+ if (MayFoldLoad(From)) {
+ // Trivial case, when From comes from a load and is only used by the
+ // shuffle. Make it use insertps from the vector that we need from that
+ // load.
+ SDValue Addr = From.getOperand(1);
+ SDValue NewAddr =
+ DAG.getNode(ISD::ADD, dl, Addr.getSimpleValueType(), Addr,
+ DAG.getConstant(DestIndex * EVT.getStoreSize(),
+ Addr.getSimpleValueType()));
+
+ LoadSDNode *Load = cast<LoadSDNode>(From);
+ SDValue NewLoad =
+ DAG.getLoad(EVT, dl, Load->getChain(), NewAddr,
+ DAG.getMachineFunction().getMachineMemOperand(
+ Load->getMemOperand(), 0, EVT.getStoreSize()));
+
+ if (EVT == MVT::f32) {
+ // Create this as a scalar to vector to match the instruction pattern.
+ SDValue LoadScalarToVector =
+ DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VT, NewLoad);
+ SDValue InsertpsMask = DAG.getIntPtrConstant(DestIndex << 4);
+ return DAG.getNode(X86ISD::INSERTPS, dl, VT, To, LoadScalarToVector,
+ InsertpsMask);
+ } else { // EVT == MVT::i32
+ // If we're getting an i32 from memory, use an INSERT_VECTOR_ELT
+ // instruction, to match the PINSRD instruction, which loads an i32 to a
+ // certain vector element.
+ return DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, VT, To, NewLoad,
+ DAG.getConstant(DestIndex, MVT::i32));
+ }
+ }
+
+ // Vector-element-to-vector
+ unsigned SrcIndex = Mask[DestIndex] % 4;
+ SDValue InsertpsMask = DAG.getIntPtrConstant(DestIndex << 4 | SrcIndex << 6);
+ return DAG.getNode(X86ISD::INSERTPS, dl, VT, To, From, InsertpsMask);
+}
+
// Reduce a vector shuffle to zext.
static SDValue LowerVectorIntExtend(SDValue Op, const X86Subtarget *Subtarget,
SelectionDAG &DAG) {
if (BlendOp.getNode())
return BlendOp;
+ if (Subtarget->hasSSE41() && isINSERTPSMask(M, VT))
+ return getINSERTPS(SVOp, dl, DAG);
+
unsigned Imm8;
if (V2IsUndef && HasInt256 && isPermImmMask(M, VT, Imm8))
return getTargetShuffleNode(X86ISD::VPERMI, dl, VT, V1, Imm8, DAG);
permclMask.push_back(DAG.getConstant((M[i]>=0) ? M[i] : 0, MaskEltVT));
}
- SDValue Mask = DAG.getNode(ISD::BUILD_VECTOR, dl, MaskVectorVT,
- &permclMask[0], NumElems);
+ SDValue Mask = DAG.getNode(ISD::BUILD_VECTOR, dl, MaskVectorVT, permclMask);
if (V2IsUndef)
// Bitcast is for VPERMPS since mask is v8i32 but node takes v8f32
return DAG.getNode(X86ISD::VPERMV, dl, VT,
return NewOp;
}
+ if (VT == MVT::v16i16 && Subtarget->hasInt256()) {
+ SDValue NewOp = LowerVECTOR_SHUFFLEv16i16(Op, DAG);
+ if (NewOp.getNode())
+ return NewOp;
+ }
+
if (VT == MVT::v16i8) {
SDValue NewOp = LowerVECTOR_SHUFFLEv16i8(SVOp, Subtarget, DAG);
if (NewOp.getNode())
return SDValue();
}
+/// Insert one bit to mask vector, like v16i1 or v8i1.
+/// AVX-512 feature.
+SDValue
+X86TargetLowering::InsertBitToMaskVector(SDValue Op, SelectionDAG &DAG) const {
+ SDLoc dl(Op);
+ SDValue Vec = Op.getOperand(0);
+ SDValue Elt = Op.getOperand(1);
+ SDValue Idx = Op.getOperand(2);
+ MVT VecVT = Vec.getSimpleValueType();
+
+ if (!isa<ConstantSDNode>(Idx)) {
+ // Non constant index. Extend source and destination,
+ // insert element and then truncate the result.
+ MVT ExtVecVT = (VecVT == MVT::v8i1 ? MVT::v8i64 : MVT::v16i32);
+ MVT ExtEltVT = (VecVT == MVT::v8i1 ? MVT::i64 : MVT::i32);
+ SDValue ExtOp = DAG.getNode(ISD::INSERT_VECTOR_ELT, dl, ExtVecVT,
+ DAG.getNode(ISD::ZERO_EXTEND, dl, ExtVecVT, Vec),
+ DAG.getNode(ISD::ZERO_EXTEND, dl, ExtEltVT, Elt), Idx);
+ return DAG.getNode(ISD::TRUNCATE, dl, VecVT, ExtOp);
+ }
+
+ unsigned IdxVal = cast<ConstantSDNode>(Idx)->getZExtValue();
+ SDValue EltInVec = DAG.getNode(ISD::SCALAR_TO_VECTOR, dl, VecVT, Elt);
+ if (Vec.getOpcode() == ISD::UNDEF)
+ return DAG.getNode(X86ISD::VSHLI, dl, VecVT, EltInVec,
+ DAG.getConstant(IdxVal, MVT::i8));
+ const TargetRegisterClass* rc = getRegClassFor(VecVT);
+ unsigned MaxSift = rc->getSize()*8 - 1;
+ EltInVec = DAG.getNode(X86ISD::VSHLI, dl, VecVT, EltInVec,
+ DAG.getConstant(MaxSift, MVT::i8));
+ EltInVec = DAG.getNode(X86ISD::VSRLI, dl, VecVT, EltInVec,
+ DAG.getConstant(MaxSift - IdxVal, MVT::i8));
+ return DAG.getNode(ISD::OR, dl, VecVT, Vec, EltInVec);
+}
SDValue
X86TargetLowering::LowerINSERT_VECTOR_ELT(SDValue Op, SelectionDAG &DAG) const {
MVT VT = Op.getSimpleValueType();
MVT EltVT = VT.getVectorElementType();
+
+ if (EltVT == MVT::i1)
+ return InsertBitToMaskVector(Op, DAG);
SDLoc dl(Op);
SDValue N0 = Op.getOperand(0);
if (InFlag) {
SDValue Ops[] = { Chain, TGA, *InFlag };
- Chain = DAG.getNode(CallType, dl, NodeTys, Ops, array_lengthof(Ops));
+ Chain = DAG.getNode(CallType, dl, NodeTys, Ops);
} else {
SDValue Ops[] = { Chain, TGA };
- Chain = DAG.getNode(CallType, dl, NodeTys, Ops, array_lengthof(Ops));
+ Chain = DAG.getNode(CallType, dl, NodeTys, Ops);
}
// TLSADDR will be codegen'ed as call. Inform MFI that function has calls.
static SDValue
LowerToTLSGeneralDynamicModel64(GlobalAddressSDNode *GA, SelectionDAG &DAG,
const EVT PtrVT) {
- return GetTLSADDR(DAG, DAG.getEntryNode(), GA, NULL, PtrVT,
+ return GetTLSADDR(DAG, DAG.getEntryNode(), GA, nullptr, PtrVT,
X86::RAX, X86II::MO_TLSGD);
}
SDValue Base;
if (is64Bit) {
- Base = GetTLSADDR(DAG, DAG.getEntryNode(), GA, NULL, PtrVT, X86::RAX,
+ Base = GetTLSADDR(DAG, DAG.getEntryNode(), GA, nullptr, PtrVT, X86::RAX,
X86II::MO_TLSLD, /*LocalDynamic=*/true);
} else {
SDValue InFlag;
SDValue Chain = DAG.getEntryNode();
SDVTList NodeTys = DAG.getVTList(MVT::Other, MVT::Glue);
SDValue Args[] = { Chain, Offset };
- Chain = DAG.getNode(X86ISD::TLSCALL, DL, NodeTys, Args, 2);
+ Chain = DAG.getNode(X86ISD::TLSCALL, DL, NodeTys, Args);
// TLSCALL will be codegen'ed as call. Inform MFI that function has calls.
MachineFrameInfo *MFI = DAG.getMachineFunction().getFrameInfo();
Chain.getValue(1));
}
- if (Subtarget->isTargetWindows() || Subtarget->isTargetMingw()) {
+ if (Subtarget->isTargetKnownWindowsMSVC() ||
+ Subtarget->isTargetWindowsGNU()) {
// Just use the implicit TLS architecture
// Need to generate someting similar to:
// mov rdx, qword [gs:abs 58H]; Load pointer to ThreadLocalStorage
// If GV is an alias then use the aliasee for determining
// thread-localness.
if (const GlobalAlias *GA = dyn_cast<GlobalAlias>(GV))
- GV = GA->resolveAliasedGlobal(false);
+ GV = GA->getAliasedGlobal();
SDLoc dl(GA);
SDValue Chain = DAG.getEntryNode();
: Type::getInt32PtrTy(*DAG.getContext(),
257));
- SDValue TlsArray = Subtarget->is64Bit() ? DAG.getIntPtrConstant(0x58) :
- (Subtarget->isTargetMingw() ? DAG.getIntPtrConstant(0x2C) :
- DAG.getExternalSymbol("_tls_array", getPointerTy()));
+ SDValue TlsArray =
+ Subtarget->is64Bit()
+ ? DAG.getIntPtrConstant(0x58)
+ : (Subtarget->isTargetWindowsGNU()
+ ? DAG.getIntPtrConstant(0x2C)
+ : DAG.getExternalSymbol("_tls_array", getPointerTy()));
- SDValue ThreadPointer = DAG.getLoad(getPointerTy(), dl, Chain, TlsArray,
- MachinePointerInfo(Ptr),
- false, false, false, 0);
+ SDValue ThreadPointer =
+ DAG.getLoad(getPointerTy(), dl, Chain, TlsArray,
+ MachinePointerInfo(Ptr), false, false, false, 0);
// Load the _tls_index variable
SDValue IDX = DAG.getExternalSymbol("_tls_index", getPointerTy());
SDValue Ops1[4] = { Tmp3, Tmp1, CC, Cond };
if (Op.getOpcode() == ISD::SHL_PARTS) {
- Hi = DAG.getNode(X86ISD::CMOV, dl, VT, Ops0, 4);
- Lo = DAG.getNode(X86ISD::CMOV, dl, VT, Ops1, 4);
+ Hi = DAG.getNode(X86ISD::CMOV, dl, VT, Ops0);
+ Lo = DAG.getNode(X86ISD::CMOV, dl, VT, Ops1);
} else {
- Lo = DAG.getNode(X86ISD::CMOV, dl, VT, Ops0, 4);
- Hi = DAG.getNode(X86ISD::CMOV, dl, VT, Ops1, 4);
+ Lo = DAG.getNode(X86ISD::CMOV, dl, VT, Ops0);
+ Hi = DAG.getNode(X86ISD::CMOV, dl, VT, Ops1);
}
SDValue Ops[2] = { Lo, Hi };
- return DAG.getMergeValues(Ops, array_lengthof(Ops), dl);
+ return DAG.getMergeValues(Ops, dl);
}
SDValue X86TargetLowering::LowerSINT_TO_FP(SDValue Op,
SDValue Ops[] = { Chain, StackSlot, DAG.getValueType(SrcVT) };
SDValue Result = DAG.getMemIntrinsicNode(useSSE ? X86ISD::FILD_FLAG :
X86ISD::FILD, DL,
- Tys, Ops, array_lengthof(Ops),
- SrcVT, MMO);
+ Tys, Ops, SrcVT, MMO);
if (useSSE) {
Chain = Result.getValue(1);
MachineMemOperand::MOStore, SSFISize, SSFISize);
Chain = DAG.getMemIntrinsicNode(X86ISD::FST, DL, Tys,
- Ops, array_lengthof(Ops),
- Op.getValueType(), MMO);
+ Ops, Op.getValueType(), MMO);
Result = DAG.getLoad(Op.getValueType(), DL, Chain, StackSlot,
MachinePointerInfo::getFixedStack(SSFI),
false, false, false, 0);
SDVTList Tys = DAG.getVTList(MVT::f80, MVT::Other);
SDValue Ops[] = { Store, StackSlot, DAG.getValueType(MVT::i64) };
SDValue Fild = DAG.getMemIntrinsicNode(X86ISD::FILD, dl, Tys, Ops,
- array_lengthof(Ops), MVT::i64, MMO);
+ MVT::i64, MMO);
APInt FF(32, 0x5F800000ULL);
MachineMemOperand *MMO =
MF.getMachineMemOperand(MachinePointerInfo::getFixedStack(SSFI),
MachineMemOperand::MOLoad, MemSize, MemSize);
- Value = DAG.getMemIntrinsicNode(X86ISD::FLD, DL, Tys, Ops,
- array_lengthof(Ops), DstTy, MMO);
+ Value = DAG.getMemIntrinsicNode(X86ISD::FLD, DL, Tys, Ops, DstTy, MMO);
Chain = Value.getValue(1);
SSFI = MF.getFrameInfo()->CreateStackObject(MemSize, MemSize, false);
StackSlot = DAG.getFrameIndex(SSFI, getPointerTy());
// Build the FP_TO_INT*_IN_MEM
SDValue Ops[] = { Chain, Value, StackSlot };
SDValue FIST = DAG.getMemIntrinsicNode(Opc, DL, DAG.getVTList(MVT::Other),
- Ops, array_lengthof(Ops), DstTy,
- MMO);
+ Ops, DstTy, MMO);
return std::make_pair(FIST, StackSlot);
} else {
SDValue ftol = DAG.getNode(X86ISD::WIN_FTOL, DL,
MVT::i32, eax.getValue(2));
SDValue Ops[] = { eax, edx };
SDValue pair = IsReplace
- ? DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Ops, array_lengthof(Ops))
- : DAG.getMergeValues(Ops, array_lengthof(Ops), DL);
+ ? DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Ops)
+ : DAG.getMergeValues(Ops, DL);
return std::make_pair(pair, SDValue());
}
}
for (unsigned j = 0; j < 8; ++j)
pshufbMask.push_back(DAG.getConstant(0x80, MVT::i8));
}
- SDValue BV = DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v32i8,
- &pshufbMask[0], 32);
+ SDValue BV = DAG.getNode(ISD::BUILD_VECTOR, DL, MVT::v32i8, pshufbMask);
In = DAG.getNode(X86ISD::PSHUFB, DL, MVT::v32i8, In, BV);
In = DAG.getNode(ISD::BITCAST, DL, MVT::v4i64, In);
/*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() == 0) return Op;
+ if (!FIST.getNode()) return Op;
if (StackSlot.getNode())
// Load the result.
SmallVector<SDValue, 8> Opnds;
DenseMap<SDValue, unsigned> VecInMap;
+ SmallVector<SDValue, 8> VecIns;
EVT VT = MVT::Other;
// Recognize a special case where a vector is casted into wide integer to
VT != VecInMap.begin()->first.getValueType())
return SDValue();
M = VecInMap.insert(std::make_pair(ExtractedFromVec, 0)).first;
+ VecIns.push_back(ExtractedFromVec);
}
M->second |= 1U << cast<ConstantSDNode>(Idx)->getZExtValue();
}
"Not extracted from 128-/256-bit vector.");
unsigned FullMask = (1U << VT.getVectorNumElements()) - 1U;
- SmallVector<SDValue, 8> VecIns;
for (DenseMap<SDValue, unsigned>::const_iterator
I = VecInMap.begin(), E = VecInMap.end(); I != E; ++I) {
// Quit if not all elements are used.
if (I->second != FullMask)
return SDValue();
- VecIns.push_back(I->first);
}
EVT TestVT = VT.is128BitVector() ? MVT::v2i64 : MVT::v4i64;
VecIns.back(), VecIns.back());
}
+/// \brief return true if \c Op has a use that doesn't just read flags.
+static bool hasNonFlagsUse(SDValue Op) {
+ for (SDNode::use_iterator UI = Op->use_begin(), UE = Op->use_end(); UI != UE;
+ ++UI) {
+ SDNode *User = *UI;
+ unsigned UOpNo = UI.getOperandNo();
+ if (User->getOpcode() == ISD::TRUNCATE && User->hasOneUse()) {
+ // Look pass truncate.
+ UOpNo = User->use_begin().getOperandNo();
+ User = *User->use_begin();
+ }
+
+ if (User->getOpcode() != ISD::BRCOND && User->getOpcode() != ISD::SETCC &&
+ !(User->getOpcode() == ISD::SELECT && UOpNo == 0))
+ return true;
+ }
+ return false;
+}
+
/// Emit nodes that will be selected as "test Op0,Op0", or something
/// equivalent.
-SDValue X86TargetLowering::EmitTest(SDValue Op, unsigned X86CC,
+SDValue X86TargetLowering::EmitTest(SDValue Op, unsigned X86CC, SDLoc dl,
SelectionDAG &DAG) const {
- SDLoc dl(Op);
-
if (Op.getValueType() == MVT::i1)
// KORTEST instruction should be selected
return DAG.getNode(X86ISD::CMP, dl, MVT::i32, Op,
Opcode = X86ISD::ADD;
NumOperands = 2;
break;
- case ISD::AND: {
- // If the primary and result isn't used, don't bother using X86ISD::AND,
- // because a TEST instruction will be better.
- bool NonFlagUse = false;
- for (SDNode::use_iterator UI = Op.getNode()->use_begin(),
- UE = Op.getNode()->use_end(); UI != UE; ++UI) {
- SDNode *User = *UI;
- unsigned UOpNo = UI.getOperandNo();
- if (User->getOpcode() == ISD::TRUNCATE && User->hasOneUse()) {
- // Look pass truncate.
- UOpNo = User->use_begin().getOperandNo();
- User = *User->use_begin();
- }
-
- if (User->getOpcode() != ISD::BRCOND &&
- User->getOpcode() != ISD::SETCC &&
- !(User->getOpcode() == ISD::SELECT && UOpNo == 0)) {
- NonFlagUse = true;
+ case ISD::SHL:
+ case ISD::SRL:
+ // If we have a constant logical shift that's only used in a comparison
+ // against zero turn it into an equivalent AND. This allows turning it into
+ // a TEST instruction later.
+ if ((X86CC == X86::COND_E || X86CC == X86::COND_NE) &&
+ isa<ConstantSDNode>(Op->getOperand(1)) && !hasNonFlagsUse(Op)) {
+ EVT VT = Op.getValueType();
+ unsigned BitWidth = VT.getSizeInBits();
+ unsigned ShAmt = Op->getConstantOperandVal(1);
+ if (ShAmt >= BitWidth) // Avoid undefined shifts.
break;
- }
+ APInt Mask = ArithOp.getOpcode() == ISD::SRL
+ ? APInt::getHighBitsSet(BitWidth, BitWidth - ShAmt)
+ : APInt::getLowBitsSet(BitWidth, BitWidth - ShAmt);
+ if (!Mask.isSignedIntN(32)) // Avoid large immediates.
+ break;
+ SDValue New = DAG.getNode(ISD::AND, dl, VT, Op->getOperand(0),
+ DAG.getConstant(Mask, VT));
+ DAG.ReplaceAllUsesWith(Op, New);
+ Op = New;
}
+ break;
- if (!NonFlagUse)
+ case ISD::AND:
+ // If the primary and result isn't used, don't bother using X86ISD::AND,
+ // because a TEST instruction will be better.
+ if (!hasNonFlagsUse(Op))
break;
- }
// FALL THROUGH
case ISD::SUB:
case ISD::OR:
for (unsigned i = 0; i != NumOperands; ++i)
Ops.push_back(Op.getOperand(i));
- SDValue New = DAG.getNode(Opcode, dl, VTs, &Ops[0], NumOperands);
+ SDValue New = DAG.getNode(Opcode, dl, VTs, Ops);
DAG.ReplaceAllUsesWith(Op, New);
return SDValue(New.getNode(), 1);
}
/// Emit nodes that will be selected as "cmp Op0,Op1", or something
/// equivalent.
SDValue X86TargetLowering::EmitCmp(SDValue Op0, SDValue Op1, unsigned X86CC,
- SelectionDAG &DAG) const {
- SDLoc dl(Op0);
+ SDLoc dl, SelectionDAG &DAG) const {
if (ConstantSDNode *C = dyn_cast<ConstantSDNode>(Op1)) {
if (C->getAPIntValue() == 0)
- return EmitTest(Op0, X86CC, DAG);
+ return EmitTest(Op0, X86CC, dl, DAG);
if (Op0.getValueType() == MVT::i1)
llvm_unreachable("Unexpected comparison operation for MVT::i1 operands");
if ((Op0.getValueType() == MVT::i8 || Op0.getValueType() == MVT::i16 ||
Op0.getValueType() == MVT::i32 || Op0.getValueType() == MVT::i64)) {
- // Do the comparison at i32 if it's smaller. This avoids subregister
- // aliasing issues. Keep the smaller reference if we're optimizing for
- // size, however, as that'll allow better folding of memory operations.
+ // Do the comparison at i32 if it's smaller, besides the Atom case.
+ // This avoids subregister aliasing issues. Keep the smaller reference
+ // 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()->getAttributes().hasAttribute(
- AttributeSet::FunctionIndex, Attribute::MinSize)) {
+ AttributeSet::FunctionIndex, Attribute::MinSize) &&
+ !Subtarget->isAtom()) {
unsigned ExtendOp =
isX86CCUnsigned(X86CC) ? ISD::ZERO_EXTEND : ISD::SIGN_EXTEND;
Op0 = DAG.getNode(ExtendOp, dl, MVT::i32, Op0);
/// \brief Try to turn a VSETULT into a VSETULE by modifying its second
/// operand \p Op1. If non-trivial (for example because it's not constant)
/// return an empty value.
-static SDValue ChangeVSETULTtoVSETULE(SDValue Op1, SelectionDAG &DAG)
+static SDValue ChangeVSETULTtoVSETULE(SDLoc dl, SDValue Op1, SelectionDAG &DAG)
{
BuildVectorSDNode *BV = dyn_cast<BuildVectorSDNode>(Op1.getNode());
if (!BV)
ULTOp1.push_back(DAG.getConstant(Val - 1, EVT));
}
- return DAG.getNode(ISD::BUILD_VECTOR, SDLoc(Op1), VT, ULTOp1.data(),
- ULTOp1.size());
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, ULTOp1);
}
static SDValue LowerVSETCC(SDValue Op, const X86Subtarget *Subtarget,
// Only do this pre-AVX since vpcmp* is no longer destructive.
if (Subtarget->hasAVX())
break;
- SDValue ULEOp1 = ChangeVSETULTtoVSETULE(Op1, DAG);
+ SDValue ULEOp1 = ChangeVSETULTtoVSETULE(dl, Op1, DAG);
if (ULEOp1.getNode()) {
Op1 = ULEOp1;
Subus = true; Invert = false; Swap = false;
if (X86CC == X86::COND_INVALID)
return SDValue();
- SDValue EFLAGS = EmitCmp(Op0, Op1, X86CC, DAG);
+ SDValue EFLAGS = EmitCmp(Op0, Op1, X86CC, dl, DAG);
EFLAGS = ConvertCmpIfNecessary(EFLAGS, DAG);
SDValue SetCC = DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
DAG.getConstant(X86CC, MVT::i8), EFLAGS);
Res = DAG.getNOT(DL, Res, Res.getValueType());
ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(Op2);
- if (N2C == 0 || !N2C->isNullValue())
+ if (!N2C || !N2C->isNullValue())
Res = DAG.getNode(ISD::OR, DL, Res.getValueType(), Res, Y);
return Res;
}
if (addTest) {
CC = DAG.getConstant(X86::COND_NE, MVT::i8);
- Cond = EmitTest(Cond, X86::COND_NE, DAG);
+ Cond = EmitTest(Cond, X86::COND_NE, DL, DAG);
}
// a < b ? -1 : 0 -> RES = ~setcc_carry
// condition is true.
SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::Glue);
SDValue Ops[] = { Op2, Op1, CC, Cond };
- return DAG.getNode(X86ISD::CMOV, DL, VTs, Ops, array_lengthof(Ops));
+ return DAG.getNode(X86ISD::CMOV, DL, VTs, Ops);
}
static SDValue LowerSIGN_EXTEND_AVX512(SDValue Op, SelectionDAG &DAG) {
if (addTest) {
CC = DAG.getConstant(X86::COND_NE, MVT::i8);
- Cond = EmitTest(Cond, X86::COND_NE, DAG);
+ Cond = EmitTest(Cond, X86::COND_NE, dl, DAG);
}
Cond = ConvertCmpIfNecessary(Cond, DAG);
return DAG.getNode(X86ISD::BRCOND, dl, Op.getValueType(),
SDValue
X86TargetLowering::LowerDYNAMIC_STACKALLOC(SDValue Op,
SelectionDAG &DAG) const {
- assert((Subtarget->isTargetCygMing() || Subtarget->isTargetWindows() ||
- getTargetMachine().Options.EnableSegmentedStacks) &&
- "This should be used only on Windows targets or when segmented stacks "
- "are being used");
- assert(!Subtarget->isTargetMacho() && "Not implemented");
+ MachineFunction &MF = DAG.getMachineFunction();
+ bool SplitStack = MF.shouldSplitStack();
+ bool Lower = (Subtarget->isOSWindows() && !Subtarget->isTargetMacho()) ||
+ SplitStack;
SDLoc dl(Op);
+ if (!Lower) {
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+ SDNode* Node = Op.getNode();
+
+ unsigned SPReg = TLI.getStackPointerRegisterToSaveRestore();
+ assert(SPReg && "Target cannot require DYNAMIC_STACKALLOC expansion and"
+ " not tell us which reg is the stack pointer!");
+ EVT VT = Node->getValueType(0);
+ SDValue Tmp1 = SDValue(Node, 0);
+ SDValue Tmp2 = SDValue(Node, 1);
+ SDValue Tmp3 = Node->getOperand(2);
+ SDValue Chain = Tmp1.getOperand(0);
+
+ // Chain the dynamic stack allocation so that it doesn't modify the stack
+ // pointer when other instructions are using the stack.
+ Chain = DAG.getCALLSEQ_START(Chain, DAG.getIntPtrConstant(0, true),
+ SDLoc(Node));
+
+ SDValue Size = Tmp2.getOperand(1);
+ SDValue SP = DAG.getCopyFromReg(Chain, dl, SPReg, VT);
+ Chain = SP.getValue(1);
+ unsigned Align = cast<ConstantSDNode>(Tmp3)->getZExtValue();
+ const TargetFrameLowering &TFI = *getTargetMachine().getFrameLowering();
+ unsigned StackAlign = TFI.getStackAlignment();
+ Tmp1 = DAG.getNode(ISD::SUB, dl, VT, SP, Size); // Value
+ if (Align > StackAlign)
+ Tmp1 = DAG.getNode(ISD::AND, dl, VT, Tmp1,
+ DAG.getConstant(-(uint64_t)Align, VT));
+ Chain = DAG.getCopyToReg(Chain, dl, SPReg, Tmp1); // Output chain
+
+ Tmp2 = DAG.getCALLSEQ_END(Chain, DAG.getIntPtrConstant(0, true),
+ DAG.getIntPtrConstant(0, true), SDValue(),
+ SDLoc(Node));
+
+ SDValue Ops[2] = { Tmp1, Tmp2 };
+ return DAG.getMergeValues(Ops, dl);
+ }
+
// Get the inputs.
SDValue Chain = Op.getOperand(0);
SDValue Size = Op.getOperand(1);
bool Is64Bit = Subtarget->is64Bit();
EVT SPTy = Is64Bit ? MVT::i64 : MVT::i32;
- if (getTargetMachine().Options.EnableSegmentedStacks) {
- MachineFunction &MF = DAG.getMachineFunction();
+ if (SplitStack) {
MachineRegisterInfo &MRI = MF.getRegInfo();
if (Is64Bit) {
SDValue Value = DAG.getNode(X86ISD::SEG_ALLOCA, dl, SPTy, Chain,
DAG.getRegister(Vreg, SPTy));
SDValue Ops1[2] = { Value, Chain };
- return DAG.getMergeValues(Ops1, 2, dl);
+ return DAG.getMergeValues(Ops1, dl);
} else {
SDValue Flag;
unsigned Reg = (Subtarget->is64Bit() ? X86::RAX : X86::EAX);
}
SDValue Ops1[2] = { SP, Chain };
- return DAG.getMergeValues(Ops1, 2, dl);
+ return DAG.getMergeValues(Ops1, dl);
}
}
Store = DAG.getStore(Op.getOperand(0), DL, RSFIN, FIN,
MachinePointerInfo(SV, 16), false, false, 0);
MemOps.push_back(Store);
- return DAG.getNode(ISD::TokenFactor, DL, MVT::Other,
- &MemOps[0], MemOps.size());
+ return DAG.getNode(ISD::TokenFactor, DL, MVT::Other, MemOps);
}
SDValue X86TargetLowering::LowerVAARG(SDValue Op, SelectionDAG &DAG) const {
InstOps.push_back(DAG.getConstant(Align, MVT::i32));
SDVTList VTs = DAG.getVTList(getPointerTy(), MVT::Other);
SDValue VAARG = DAG.getMemIntrinsicNode(X86ISD::VAARG_64, dl,
- VTs, &InstOps[0], InstOps.size(),
- MVT::i64,
+ VTs, InstOps, MVT::i64,
MachinePointerInfo(SV),
/*Align=*/0,
/*Volatile=*/false,
ConstantSDNode *ND;
switch(Opc) {
- default: llvm_unreachable(0);
+ default: llvm_unreachable(nullptr);
case X86ISD::VSHLI:
for (unsigned i=0; i!=NumElts; ++i) {
SDValue CurrentOp = SrcOp->getOperand(i);
break;
}
- return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Elts[0], NumElts);
+ return DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Elts);
}
return DAG.getNode(Opc, dl, VT, SrcOp, DAG.getConstant(ShiftAmt, MVT::i8));
ShOps[0] = ShAmt;
ShOps[1] = DAG.getConstant(0, MVT::i32);
ShOps[2] = ShOps[3] = DAG.getUNDEF(MVT::i32);
- ShAmt = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, &ShOps[0], 4);
+ ShAmt = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v4i32, ShOps);
// The return type has to be a 128-bit type with the same element
// type as the input type.
return DAG.getNode(X86ISD::PMULUDQ, dl, Op.getValueType(),
Op.getOperand(1), Op.getOperand(2));
+ case Intrinsic::x86_sse41_pmuldq:
+ case Intrinsic::x86_avx2_pmul_dq:
+ return DAG.getNode(X86ISD::PMULDQ, dl, Op.getValueType(),
+ Op.getOperand(1), Op.getOperand(2));
+
+ case Intrinsic::x86_sse2_pmulhu_w:
+ case Intrinsic::x86_avx2_pmulhu_w:
+ return DAG.getNode(ISD::MULHU, dl, Op.getValueType(),
+ Op.getOperand(1), Op.getOperand(2));
+
+ case Intrinsic::x86_sse2_pmulh_w:
+ case Intrinsic::x86_avx2_pmulh_w:
+ return DAG.getNode(ISD::MULHS, dl, Op.getValueType(),
+ Op.getOperand(1), Op.getOperand(2));
+
// SSE2/AVX2 sub with unsigned saturation intrinsics
case Intrinsic::x86_sse2_psubus_b:
case Intrinsic::x86_sse2_psubus_w:
}
SmallVector<SDValue, 5> NewOps(Op->op_begin()+1, Op->op_end());
SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::i32);
- SDValue PCMP = DAG.getNode(Opcode, dl, VTs, NewOps.data(), NewOps.size());
+ SDValue PCMP = DAG.getNode(Opcode, dl, VTs, NewOps);
SDValue SetCC = DAG.getNode(X86ISD::SETCC, dl, MVT::i8,
DAG.getConstant(X86CC, MVT::i8),
SDValue(PCMP.getNode(), 1));
SmallVector<SDValue, 5> NewOps(Op->op_begin()+1, Op->op_end());
SDVTList VTs = DAG.getVTList(Op.getValueType(), MVT::i32);
- return DAG.getNode(Opcode, dl, VTs, NewOps.data(), NewOps.size());
+ return DAG.getNode(Opcode, dl, VTs, NewOps);
}
case Intrinsic::x86_fma_vfmadd_ps:
case Intrinsic::x86_fma_vfmadd_pd:
SDValue Ops[] = {Src, MaskInReg, Base, Scale, Index, Disp, Segment, Chain};
SDNode *Res = DAG.getMachineNode(Opc, dl, VTs, Ops);
SDValue RetOps[] = { SDValue(Res, 0), SDValue(Res, 2) };
- return DAG.getMergeValues(RetOps, array_lengthof(RetOps), dl);
+ return DAG.getMergeValues(RetOps, dl);
}
static SDValue getMGatherNode(unsigned Opc, SDValue Op, SelectionDAG &DAG,
SDValue Ops[] = {Src, MaskInReg, Base, Scale, Index, Disp, Segment, Chain};
SDNode *Res = DAG.getMachineNode(Opc, dl, VTs, Ops);
SDValue RetOps[] = { SDValue(Res, 0), SDValue(Res, 2) };
- return DAG.getMergeValues(RetOps, array_lengthof(RetOps), dl);
+ return DAG.getMergeValues(RetOps, dl);
}
static SDValue getScatterNode(unsigned Opc, SDValue Op, SelectionDAG &DAG,
return SDValue(Res, 1);
}
+// getReadTimeStampCounter - Handles the lowering of builtin intrinsics that
+// read the time stamp counter (x86_rdtsc and x86_rdtscp). This function is
+// also used to custom lower READCYCLECOUNTER nodes.
+static void getReadTimeStampCounter(SDNode *N, SDLoc DL, unsigned Opcode,
+ SelectionDAG &DAG, const X86Subtarget *Subtarget,
+ SmallVectorImpl<SDValue> &Results) {
+ SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
+ SDValue rd = DAG.getNode(Opcode, DL, Tys, N->getOperand(0));
+ SDValue LO, HI;
+
+ // The processor's time-stamp counter (a 64-bit MSR) is stored into the
+ // EDX:EAX registers. EDX is loaded with the high-order 32 bits of the MSR
+ // and the EAX register is loaded with the low-order 32 bits.
+ if (Subtarget->is64Bit()) {
+ LO = DAG.getCopyFromReg(rd, DL, X86::RAX, MVT::i64, rd.getValue(1));
+ HI = DAG.getCopyFromReg(LO.getValue(1), DL, X86::RDX, MVT::i64,
+ LO.getValue(2));
+ } else {
+ LO = DAG.getCopyFromReg(rd, DL, X86::EAX, MVT::i32, rd.getValue(1));
+ HI = DAG.getCopyFromReg(LO.getValue(1), DL, X86::EDX, MVT::i32,
+ LO.getValue(2));
+ }
+ SDValue Chain = HI.getValue(1);
+
+ if (Opcode == X86ISD::RDTSCP_DAG) {
+ assert(N->getNumOperands() == 3 && "Unexpected number of operands!");
+
+ // Instruction RDTSCP loads the IA32:TSC_AUX_MSR (address C000_0103H) into
+ // the ECX register. Add 'ecx' explicitly to the chain.
+ SDValue ecx = DAG.getCopyFromReg(Chain, DL, X86::ECX, MVT::i32,
+ HI.getValue(2));
+ // Explicitly store the content of ECX at the location passed in input
+ // to the 'rdtscp' intrinsic.
+ Chain = DAG.getStore(ecx.getValue(1), DL, ecx, N->getOperand(2),
+ MachinePointerInfo(), false, false, 0);
+ }
+
+ if (Subtarget->is64Bit()) {
+ // The EDX register is loaded with the high-order 32 bits of the MSR, and
+ // the EAX register is loaded with the low-order 32 bits.
+ SDValue Tmp = DAG.getNode(ISD::SHL, DL, MVT::i64, HI,
+ DAG.getConstant(32, MVT::i8));
+ Results.push_back(DAG.getNode(ISD::OR, DL, MVT::i64, LO, Tmp));
+ Results.push_back(Chain);
+ return;
+ }
+
+ // Use a buildpair to merge the two 32-bit values into a 64-bit one.
+ SDValue Ops[] = { LO, HI };
+ SDValue Pair = DAG.getNode(ISD::BUILD_PAIR, DL, MVT::i64, Ops);
+ Results.push_back(Pair);
+ Results.push_back(Chain);
+}
+
+static SDValue LowerREADCYCLECOUNTER(SDValue Op, const X86Subtarget *Subtarget,
+ SelectionDAG &DAG) {
+ SmallVector<SDValue, 2> Results;
+ SDLoc DL(Op);
+ getReadTimeStampCounter(Op.getNode(), DL, X86ISD::RDTSC_DAG, DAG, Subtarget,
+ Results);
+ return DAG.getMergeValues(Results, DL);
+}
+
static SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, const X86Subtarget *Subtarget,
SelectionDAG &DAG) {
SDLoc dl(Op);
SDValue(Result.getNode(), 1) };
SDValue isValid = DAG.getNode(X86ISD::CMOV, dl,
DAG.getVTList(Op->getValueType(1), MVT::Glue),
- Ops, array_lengthof(Ops));
+ Ops);
// Return { result, isValid, chain }.
return DAG.getNode(ISD::MERGE_VALUES, dl, Op->getVTList(), Result, isValid,
SDValue Scale = Op.getOperand(6);
return getMScatterNode(Opc, Op, DAG, Src, Mask, Base, Index, Scale, Chain);
}
+ // Read Time Stamp Counter (RDTSC).
+ case Intrinsic::x86_rdtsc:
+ // Read Time Stamp Counter and Processor ID (RDTSCP).
+ case Intrinsic::x86_rdtscp: {
+ unsigned Opc;
+ switch (IntNo) {
+ default: llvm_unreachable("Impossible intrinsic"); // Can't reach here.
+ case Intrinsic::x86_rdtsc:
+ Opc = X86ISD::RDTSC_DAG; break;
+ case Intrinsic::x86_rdtscp:
+ Opc = X86ISD::RDTSCP_DAG; break;
+ }
+ SmallVector<SDValue, 2> Results;
+ getReadTimeStampCounter(Op.getNode(), dl, Opc, DAG, Subtarget, Results);
+ return DAG.getMergeValues(Results, dl);
+ }
// XTEST intrinsics.
case Intrinsic::x86_xtest: {
SDVTList VTs = DAG.getVTList(Op->getValueType(0), MVT::Other);
MachinePointerInfo(TrmpAddr, 22),
false, false, 0);
- return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains, 6);
+ return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains);
} else {
const Function *Func =
cast<Function>(cast<SrcValueSDNode>(Op.getOperand(5))->getValue());
MachinePointerInfo(TrmpAddr, 6),
false, false, 1);
- return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains, 4);
+ return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains);
}
}
SDValue Ops[] = { DAG.getEntryNode(), StackSlot };
SDValue Chain = DAG.getMemIntrinsicNode(X86ISD::FNSTCW16m, DL,
DAG.getVTList(MVT::Other),
- Ops, array_lengthof(Ops), MVT::i16,
- MMO);
+ Ops, MVT::i16, MMO);
// Load FP Control Word from stack slot
SDValue CWD = DAG.getLoad(MVT::i16, DL, Chain, StackSlot,
DAG.getConstant(X86::COND_E, MVT::i8),
Op.getValue(1)
};
- Op = DAG.getNode(X86ISD::CMOV, dl, OpVT, Ops, array_lengthof(Ops));
+ Op = DAG.getNode(X86ISD::CMOV, dl, OpVT, Ops);
// Finally xor with NumBits-1.
Op = DAG.getNode(ISD::XOR, dl, OpVT, Op, DAG.getConstant(NumBits-1, OpVT));
DAG.getConstant(X86::COND_E, MVT::i8),
Op.getValue(1)
};
- return DAG.getNode(X86ISD::CMOV, dl, VT, Ops, array_lengthof(Ops));
+ return DAG.getNode(X86ISD::CMOV, dl, VT, Ops);
}
// Lower256IntArith - Break a 256-bit integer operation into two new 128-bit
return DAG.getNode(ISD::ADD, dl, VT, Res, AhiBlo);
}
-static SDValue LowerSDIV(SDValue Op, SelectionDAG &DAG) {
- MVT VT = Op.getSimpleValueType();
- MVT EltTy = VT.getVectorElementType();
- unsigned NumElts = VT.getVectorNumElements();
- SDValue N0 = Op.getOperand(0);
+static SDValue LowerMUL_LOHI(SDValue Op, const X86Subtarget *Subtarget,
+ SelectionDAG &DAG) {
+ SDValue Op0 = Op.getOperand(0), Op1 = Op.getOperand(1);
+ EVT VT = Op0.getValueType();
SDLoc dl(Op);
- // Lower sdiv X, pow2-const.
- BuildVectorSDNode *C = dyn_cast<BuildVectorSDNode>(Op.getOperand(1));
- if (!C)
- return SDValue();
+ assert((VT == MVT::v4i32 && Subtarget->hasSSE2()) ||
+ (VT == MVT::v8i32 && Subtarget->hasInt256()));
- APInt SplatValue, SplatUndef;
- unsigned SplatBitSize;
- bool HasAnyUndefs;
- if (!C->isConstantSplat(SplatValue, SplatUndef, SplatBitSize,
- HasAnyUndefs) ||
- EltTy.getSizeInBits() < SplatBitSize)
- return SDValue();
+ // Get the high parts.
+ const int Mask[] = {1, 2, 3, 4, 5, 6, 7, 8};
+ SDValue Hi0 = DAG.getVectorShuffle(VT, dl, Op0, Op0, Mask);
+ SDValue Hi1 = DAG.getVectorShuffle(VT, dl, Op1, Op1, Mask);
- if ((SplatValue != 0) &&
- (SplatValue.isPowerOf2() || (-SplatValue).isPowerOf2())) {
- unsigned Lg2 = SplatValue.countTrailingZeros();
- // Splat the sign bit.
- SmallVector<SDValue, 16> Sz(NumElts,
- DAG.getConstant(EltTy.getSizeInBits() - 1,
- EltTy));
- SDValue SGN = DAG.getNode(ISD::SRA, dl, VT, N0,
- DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Sz[0],
- NumElts));
- // Add (N0 < 0) ? abs2 - 1 : 0;
- SmallVector<SDValue, 16> Amt(NumElts,
- DAG.getConstant(EltTy.getSizeInBits() - Lg2,
- EltTy));
- SDValue SRL = DAG.getNode(ISD::SRL, dl, VT, SGN,
- DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Amt[0],
- NumElts));
- SDValue ADD = DAG.getNode(ISD::ADD, dl, VT, N0, SRL);
- SmallVector<SDValue, 16> Lg2Amt(NumElts, DAG.getConstant(Lg2, EltTy));
- SDValue SRA = DAG.getNode(ISD::SRA, dl, VT, ADD,
- DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Lg2Amt[0],
- NumElts));
-
- // If we're dividing by a positive value, we're done. Otherwise, we must
- // negate the result.
- if (SplatValue.isNonNegative())
- return SRA;
-
- SmallVector<SDValue, 16> V(NumElts, DAG.getConstant(0, EltTy));
- SDValue Zero = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &V[0], NumElts);
- return DAG.getNode(ISD::SUB, dl, VT, Zero, SRA);
+ // Emit two multiplies, one for the lower 2 ints and one for the higher 2
+ // ints.
+ MVT MulVT = VT == MVT::v4i32 ? MVT::v2i64 : MVT::v4i64;
+ bool IsSigned = Op->getOpcode() == ISD::SMUL_LOHI;
+ unsigned Opcode =
+ (!IsSigned || !Subtarget->hasSSE41()) ? X86ISD::PMULUDQ : X86ISD::PMULDQ;
+ SDValue Mul1 = DAG.getNode(ISD::BITCAST, dl, VT,
+ DAG.getNode(Opcode, dl, MulVT, Op0, Op1));
+ SDValue Mul2 = DAG.getNode(ISD::BITCAST, dl, VT,
+ DAG.getNode(Opcode, dl, MulVT, Hi0, Hi1));
+
+ // Shuffle it back into the right order.
+ const int HighMask[] = {1, 5, 3, 7, 9, 13, 11, 15};
+ SDValue Highs = DAG.getVectorShuffle(VT, dl, Mul1, Mul2, HighMask);
+ const int LowMask[] = {0, 4, 2, 6, 8, 12, 10, 14};
+ SDValue Lows = DAG.getVectorShuffle(VT, dl, Mul1, Mul2, LowMask);
+
+ // If we have a signed multiply but no PMULDQ fix up the high parts of a
+ // unsigned multiply.
+ if (IsSigned && !Subtarget->hasSSE41()) {
+ SDValue ShAmt =
+ DAG.getConstant(31, DAG.getTargetLoweringInfo().getShiftAmountTy(VT));
+ SDValue T1 = DAG.getNode(ISD::AND, dl, VT,
+ DAG.getNode(ISD::SRA, dl, VT, Op0, ShAmt), Op1);
+ SDValue T2 = DAG.getNode(ISD::AND, dl, VT,
+ DAG.getNode(ISD::SRA, dl, VT, Op1, ShAmt), Op0);
+
+ SDValue Fixup = DAG.getNode(ISD::ADD, dl, VT, T1, T2);
+ Highs = DAG.getNode(ISD::SUB, dl, VT, Highs, Fixup);
}
- return SDValue();
+
+ return DAG.getNode(ISD::MERGE_VALUES, dl, Op.getValueType(), Highs, Lows);
}
static SDValue LowerScalarImmediateShift(SDValue Op, SelectionDAG &DAG,
DAG.getConstant(uint8_t(-1U << ShiftAmt),
MVT::i8));
return DAG.getNode(ISD::AND, dl, VT, SHL,
- DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &V[0], 16));
+ DAG.getNode(ISD::BUILD_VECTOR, dl, VT, V));
}
if (Op.getOpcode() == ISD::SRL) {
// Make a large shift.
DAG.getConstant(uint8_t(-1U) >> ShiftAmt,
MVT::i8));
return DAG.getNode(ISD::AND, dl, VT, SRL,
- DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &V[0], 16));
+ DAG.getNode(ISD::BUILD_VECTOR, dl, VT, V));
}
if (Op.getOpcode() == ISD::SRA) {
if (ShiftAmt == 7) {
SDValue Res = DAG.getNode(ISD::SRL, dl, VT, R, Amt);
SmallVector<SDValue, 16> V(16, DAG.getConstant(128 >> ShiftAmt,
MVT::i8));
- SDValue Mask = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &V[0], 16);
+ SDValue Mask = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, V);
Res = DAG.getNode(ISD::XOR, dl, VT, Res, Mask);
Res = DAG.getNode(ISD::SUB, dl, VT, Res, Mask);
return Res;
DAG.getConstant(uint8_t(-1U << ShiftAmt),
MVT::i8));
return DAG.getNode(ISD::AND, dl, VT, SHL,
- DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &V[0], 32));
+ DAG.getNode(ISD::BUILD_VECTOR, dl, VT, V));
}
if (Op.getOpcode() == ISD::SRL) {
// Make a large shift.
DAG.getConstant(uint8_t(-1U) >> ShiftAmt,
MVT::i8));
return DAG.getNode(ISD::AND, dl, VT, SRL,
- DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &V[0], 32));
+ DAG.getNode(ISD::BUILD_VECTOR, dl, VT, V));
}
if (Op.getOpcode() == ISD::SRA) {
if (ShiftAmt == 7) {
SDValue Res = DAG.getNode(ISD::SRL, dl, VT, R, Amt);
SmallVector<SDValue, 32> V(32, DAG.getConstant(128 >> ShiftAmt,
MVT::i8));
- SDValue Mask = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &V[0], 32);
+ SDValue Mask = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, V);
Res = DAG.getNode(ISD::XOR, dl, VT, Res, Mask);
Res = DAG.getNode(ISD::SUB, dl, VT, Res, Mask);
return Res;
uint64_t ShiftAmt = 0;
for (unsigned i = 0; i != Ratio; ++i) {
ConstantSDNode *C = dyn_cast<ConstantSDNode>(Amt.getOperand(i));
- if (C == 0)
+ if (!C)
return SDValue();
// 6 == Log2(64)
ShiftAmt |= C->getZExtValue() << (i * (1 << (6 - RatioInLog2)));
for (unsigned j = 0; j != Ratio; ++j) {
ConstantSDNode *C =
dyn_cast<ConstantSDNode>(Amt.getOperand(i + j));
- if (C == 0)
+ if (!C)
return SDValue();
// 6 == Log2(64)
ShAmt |= C->getZExtValue() << (j * (1 << (6 - RatioInLog2)));
BaseShAmt = InVec.getOperand(1);
}
}
- if (BaseShAmt.getNode() == 0)
+ if (!BaseShAmt.getNode())
BaseShAmt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, EltVT, Amt,
DAG.getIntPtrConstant(0));
}
}
Elts.push_back(DAG.getConstant(One.shl(ShAmt), SVT));
}
- SDValue BV = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, &Elts[0], NumElems);
+ SDValue BV = DAG.getNode(ISD::BUILD_VECTOR, dl, VT, Elts);
return DAG.getNode(ISD::MUL, dl, VT, R, BV);
}
return DAG.getNode(ISD::MUL, dl, VT, Op, R);
}
+ // If possible, lower this shift as a sequence of two shifts by
+ // constant plus a MOVSS/MOVSD instead of scalarizing it.
+ // Example:
+ // (v4i32 (srl A, (build_vector < X, Y, Y, Y>)))
+ //
+ // Could be rewritten as:
+ // (v4i32 (MOVSS (srl A, <Y,Y,Y,Y>), (srl A, <X,X,X,X>)))
+ //
+ // The advantage is that the two shifts from the example would be
+ // lowered as X86ISD::VSRLI nodes. This would be cheaper than scalarizing
+ // the vector shift into four scalar shifts plus four pairs of vector
+ // insert/extract.
+ if ((VT == MVT::v8i16 || VT == MVT::v4i32) &&
+ ISD::isBuildVectorOfConstantSDNodes(Amt.getNode())) {
+ unsigned TargetOpcode = X86ISD::MOVSS;
+ bool CanBeSimplified;
+ // The splat value for the first packed shift (the 'X' from the example).
+ SDValue Amt1 = Amt->getOperand(0);
+ // The splat value for the second packed shift (the 'Y' from the example).
+ SDValue Amt2 = (VT == MVT::v4i32) ? Amt->getOperand(1) :
+ Amt->getOperand(2);
+
+ // See if it is possible to replace this node with a sequence of
+ // two shifts followed by a MOVSS/MOVSD
+ if (VT == MVT::v4i32) {
+ // Check if it is legal to use a MOVSS.
+ CanBeSimplified = Amt2 == Amt->getOperand(2) &&
+ Amt2 == Amt->getOperand(3);
+ if (!CanBeSimplified) {
+ // Otherwise, check if we can still simplify this node using a MOVSD.
+ CanBeSimplified = Amt1 == Amt->getOperand(1) &&
+ Amt->getOperand(2) == Amt->getOperand(3);
+ TargetOpcode = X86ISD::MOVSD;
+ Amt2 = Amt->getOperand(2);
+ }
+ } else {
+ // Do similar checks for the case where the machine value type
+ // is MVT::v8i16.
+ CanBeSimplified = Amt1 == Amt->getOperand(1);
+ for (unsigned i=3; i != 8 && CanBeSimplified; ++i)
+ CanBeSimplified = Amt2 == Amt->getOperand(i);
+
+ if (!CanBeSimplified) {
+ TargetOpcode = X86ISD::MOVSD;
+ CanBeSimplified = true;
+ Amt2 = Amt->getOperand(4);
+ for (unsigned i=0; i != 4 && CanBeSimplified; ++i)
+ CanBeSimplified = Amt1 == Amt->getOperand(i);
+ for (unsigned j=4; j != 8 && CanBeSimplified; ++j)
+ CanBeSimplified = Amt2 == Amt->getOperand(j);
+ }
+ }
+
+ if (CanBeSimplified && isa<ConstantSDNode>(Amt1) &&
+ isa<ConstantSDNode>(Amt2)) {
+ // Replace this node with two shifts followed by a MOVSS/MOVSD.
+ EVT CastVT = MVT::v4i32;
+ SDValue Splat1 =
+ DAG.getConstant(cast<ConstantSDNode>(Amt1)->getAPIntValue(), VT);
+ SDValue Shift1 = DAG.getNode(Op->getOpcode(), dl, VT, R, Splat1);
+ SDValue Splat2 =
+ DAG.getConstant(cast<ConstantSDNode>(Amt2)->getAPIntValue(), VT);
+ SDValue Shift2 = DAG.getNode(Op->getOpcode(), dl, VT, R, Splat2);
+ if (TargetOpcode == X86ISD::MOVSD)
+ CastVT = MVT::v2i64;
+ SDValue BitCast1 = DAG.getNode(ISD::BITCAST, dl, CastVT, Shift1);
+ SDValue BitCast2 = DAG.getNode(ISD::BITCAST, dl, CastVT, Shift2);
+ SDValue Result = getTargetShuffleNode(TargetOpcode, dl, CastVT, BitCast2,
+ BitCast1, DAG);
+ return DAG.getNode(ISD::BITCAST, dl, VT, Result);
+ }
+ }
+
if (VT == MVT::v16i8 && Op->getOpcode() == ISD::SHL) {
assert(Subtarget->hasSSE2() && "Need SSE2 for pslli/pcmpeq.");
for (unsigned i = NumElems/2; i != NumElems; ++i)
Amt2Csts.push_back(Amt->getOperand(i));
- Amt1 = DAG.getNode(ISD::BUILD_VECTOR, dl, NewVT,
- &Amt1Csts[0], NumElems/2);
- Amt2 = DAG.getNode(ISD::BUILD_VECTOR, dl, NewVT,
- &Amt2Csts[0], NumElems/2);
+ Amt1 = DAG.getNode(ISD::BUILD_VECTOR, dl, NewVT, Amt1Csts);
+ Amt2 = DAG.getNode(ISD::BUILD_VECTOR, dl, NewVT, Amt2Csts);
} else {
// Variable shift amount
Amt1 = Extract128BitVector(Amt, 0, DAG, dl);
SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
MachineMemOperand *MMO = cast<AtomicSDNode>(Op)->getMemOperand();
SDValue Result = DAG.getMemIntrinsicNode(X86ISD::LCMPXCHG_DAG, DL, Tys,
- Ops, array_lengthof(Ops), T, MMO);
+ Ops, T, MMO);
SDValue cpOut =
DAG.getCopyFromReg(Result.getValue(0), DL, Reg, T, Result.getValue(1));
return cpOut;
}
-static SDValue LowerREADCYCLECOUNTER(SDValue Op, const X86Subtarget *Subtarget,
- SelectionDAG &DAG) {
- assert(Subtarget->is64Bit() && "Result not type legalized?");
- SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
- SDValue TheChain = Op.getOperand(0);
- SDLoc dl(Op);
- SDValue rd = DAG.getNode(X86ISD::RDTSC_DAG, dl, Tys, &TheChain, 1);
- SDValue rax = DAG.getCopyFromReg(rd, dl, X86::RAX, MVT::i64, rd.getValue(1));
- SDValue rdx = DAG.getCopyFromReg(rax.getValue(1), dl, X86::RDX, MVT::i64,
- rax.getValue(2));
- SDValue Tmp = DAG.getNode(ISD::SHL, dl, MVT::i64, rdx,
- DAG.getConstant(32, MVT::i8));
- SDValue Ops[] = {
- DAG.getNode(ISD::OR, dl, MVT::i64, rax, Tmp),
- rdx.getValue(1)
- };
- return DAG.getMergeValues(Ops, array_lengthof(Ops), dl);
-}
-
static SDValue LowerBITCAST(SDValue Op, const X86Subtarget *Subtarget,
SelectionDAG &DAG) {
MVT SrcVT = Op.getOperand(0).getSimpleValueType();
cast<AtomicSDNode>(Node)->getMemoryVT(),
Node->getOperand(0),
Node->getOperand(1), negOp,
- cast<AtomicSDNode>(Node)->getSrcValue(),
- cast<AtomicSDNode>(Node)->getAlignment(),
+ cast<AtomicSDNode>(Node)->getMemOperand(),
cast<AtomicSDNode>(Node)->getOrdering(),
cast<AtomicSDNode>(Node)->getSynchScope());
}
case ISD::CTLZ_ZERO_UNDEF: return LowerCTLZ_ZERO_UNDEF(Op, DAG);
case ISD::CTTZ: return LowerCTTZ(Op, DAG);
case ISD::MUL: return LowerMUL(Op, Subtarget, DAG);
+ case ISD::UMUL_LOHI:
+ case ISD::SMUL_LOHI: return LowerMUL_LOHI(Op, Subtarget, DAG);
case ISD::SRA:
case ISD::SRL:
case ISD::SHL: return LowerShift(Op, Subtarget, 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::SDIV: return LowerSDIV(Op, DAG);
case ISD::FSINCOS: return LowerFSINCOS(Op, Subtarget, DAG);
}
}
SDValue Ops[] = { Chain, In1, In2L, In2H };
SDVTList Tys = DAG.getVTList(MVT::i32, MVT::i32, MVT::Other);
SDValue Result =
- DAG.getMemIntrinsicNode(NewOp, dl, Tys, Ops, array_lengthof(Ops), MVT::i64,
+ DAG.getMemIntrinsicNode(NewOp, dl, Tys, Ops, MVT::i64,
cast<MemSDNode>(Node)->getMemOperand());
SDValue OpsF[] = { Result.getValue(0), Result.getValue(1)};
- Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, OpsF, 2));
+ Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, OpsF));
Results.push_back(Result.getValue(2));
}
std::pair<SDValue,SDValue> Vals =
FP_TO_INTHelper(SDValue(N, 0), DAG, IsSigned, /*IsReplace=*/ true);
SDValue FIST = Vals.first, StackSlot = Vals.second;
- if (FIST.getNode() != 0) {
+ if (FIST.getNode()) {
EVT VT = N->getValueType(0);
// Return a load from the stack slot.
- if (StackSlot.getNode() != 0)
+ if (StackSlot.getNode())
Results.push_back(DAG.getLoad(VT, dl, FIST, StackSlot,
MachinePointerInfo(),
false, false, false, 0));
Results.push_back(V);
return;
}
+ case ISD::INTRINSIC_W_CHAIN: {
+ unsigned IntNo = cast<ConstantSDNode>(N->getOperand(1))->getZExtValue();
+ switch (IntNo) {
+ default : llvm_unreachable("Do not know how to custom type "
+ "legalize this intrinsic operation!");
+ case Intrinsic::x86_rdtsc:
+ return getReadTimeStampCounter(N, dl, X86ISD::RDTSC_DAG, DAG, Subtarget,
+ Results);
+ case Intrinsic::x86_rdtscp:
+ return getReadTimeStampCounter(N, dl, X86ISD::RDTSCP_DAG, DAG, Subtarget,
+ Results);
+ }
+ }
case ISD::READCYCLECOUNTER: {
- SDVTList Tys = DAG.getVTList(MVT::Other, MVT::Glue);
- SDValue TheChain = N->getOperand(0);
- SDValue rd = DAG.getNode(X86ISD::RDTSC_DAG, dl, Tys, &TheChain, 1);
- SDValue eax = DAG.getCopyFromReg(rd, dl, X86::EAX, MVT::i32,
- rd.getValue(1));
- SDValue edx = DAG.getCopyFromReg(eax.getValue(1), dl, X86::EDX, MVT::i32,
- eax.getValue(2));
- // Use a buildpair to merge the two 32-bit values into a 64-bit one.
- SDValue Ops[] = { eax, edx };
- Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, MVT::i64, Ops,
- array_lengthof(Ops)));
- Results.push_back(edx.getValue(1));
- return;
+ return getReadTimeStampCounter(N, dl, X86ISD::RDTSC_DAG, DAG, Subtarget,
+ Results);
}
case ISD::ATOMIC_CMP_SWAP: {
EVT T = N->getValueType(0);
MachineMemOperand *MMO = cast<AtomicSDNode>(N)->getMemOperand();
unsigned Opcode = Regs64bit ? X86ISD::LCMPXCHG16_DAG :
X86ISD::LCMPXCHG8_DAG;
- SDValue Result = DAG.getMemIntrinsicNode(Opcode, dl, Tys,
- Ops, array_lengthof(Ops), T, MMO);
+ SDValue Result = DAG.getMemIntrinsicNode(Opcode, dl, Tys, Ops, T, MMO);
SDValue cpOutL = DAG.getCopyFromReg(Result.getValue(0), dl,
Regs64bit ? X86::RAX : X86::EAX,
HalfT, Result.getValue(1));
Regs64bit ? X86::RDX : X86::EDX,
HalfT, cpOutL.getValue(2));
SDValue OpsF[] = { cpOutL.getValue(0), cpOutH.getValue(0)};
- Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, T, OpsF, 2));
+ Results.push_back(DAG.getNode(ISD::BUILD_PAIR, dl, T, OpsF));
Results.push_back(cpOutH.getValue(1));
return;
}
const char *X86TargetLowering::getTargetNodeName(unsigned Opcode) const {
switch (Opcode) {
- default: return NULL;
+ default: return nullptr;
case X86ISD::BSF: return "X86ISD::BSF";
case X86ISD::BSR: return "X86ISD::BSR";
case X86ISD::SHLD: return "X86ISD::SHLD";
case X86ISD::OR: return "X86ISD::OR";
case X86ISD::XOR: return "X86ISD::XOR";
case X86ISD::AND: return "X86ISD::AND";
- case X86ISD::BZHI: return "X86ISD::BZHI";
case X86ISD::BEXTR: return "X86ISD::BEXTR";
case X86ISD::MUL_IMM: return "X86ISD::MUL_IMM";
case X86ISD::PTEST: return "X86ISD::PTEST";
case X86ISD::UNPCKH: return "X86ISD::UNPCKH";
case X86ISD::VBROADCAST: return "X86ISD::VBROADCAST";
case X86ISD::VBROADCASTM: return "X86ISD::VBROADCASTM";
+ case X86ISD::VEXTRACT: return "X86ISD::VEXTRACT";
case X86ISD::VPERMILP: return "X86ISD::VPERMILP";
case X86ISD::VPERM2X128: return "X86ISD::VPERM2X128";
case X86ISD::VPERMV: return "X86ISD::VPERMV";
case X86ISD::VPERMIV3: return "X86ISD::VPERMIV3";
case X86ISD::VPERMI: return "X86ISD::VPERMI";
case X86ISD::PMULUDQ: return "X86ISD::PMULUDQ";
+ case X86ISD::PMULDQ: return "X86ISD::PMULDQ";
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";
Reloc::Model R = getTargetMachine().getRelocationModel();
// X86 allows a sign-extended 32-bit immediate field as a displacement.
- if (!X86::isOffsetSuitableForCodeModel(AM.BaseOffs, M, AM.BaseGV != NULL))
+ if (!X86::isOffsetSuitableForCodeModel(AM.BaseOffs, M, AM.BaseGV != nullptr))
return false;
if (AM.BaseGV) {
OffsetDestReg = 0; // unused
OverflowDestReg = DestReg;
- offsetMBB = NULL;
+ offsetMBB = nullptr;
overflowMBB = thisMBB;
endMBB = thisMBB;
} else {
MachineFunction *MF = BB->getParent();
const BasicBlock *LLVM_BB = BB->getBasicBlock();
- assert(getTargetMachine().Options.EnableSegmentedStacks);
+ assert(MF->shouldSplitStack());
unsigned TlsReg = Is64Bit ? X86::FS : X86::GS;
unsigned TlsOffset = Is64Bit ? 0x70 : 0x30;
}
} else {
const char *StackProbeSymbol =
- Subtarget->isTargetWindows() ? "_chkstk" : "_alloca";
+ Subtarget->isTargetKnownWindowsMSVC() ? "_chkstk" : "_alloca";
BuildMI(*BB, MI, DL, TII->get(X86::CALLpcrel32))
.addExternalSymbol(StackProbeSymbol)
}
}
-unsigned X86TargetLowering::ComputeNumSignBitsForTargetNode(SDValue Op,
- unsigned Depth) const {
+unsigned X86TargetLowering::ComputeNumSignBitsForTargetNode(
+ SDValue Op,
+ const SelectionDAG &,
+ unsigned Depth) const {
// SETCC_CARRY sets the dest to ~0 for true or 0 for false.
if (Op.getOpcode() == X86ISD::SETCC_CARRY)
return Op.getValueType().getScalarType().getSizeInBits();
SDValue Ops[] = { Ld->getChain(), Ld->getBasePtr() };
SDValue ResNode =
DAG.getMemIntrinsicNode(X86ISD::VZEXT_LOAD, dl, Tys, Ops,
- array_lengthof(Ops),
Ld->getMemoryVT(),
Ld->getPointerInfo(),
Ld->getAlignment(),
SDValue Op2 = Cmp.getOperand(1);
SDValue SetCC;
- const ConstantSDNode* C = 0;
+ const ConstantSDNode* C = nullptr;
bool needOppositeCond = (CC == X86::COND_E);
bool checkAgainstTrue = false; // Is it a comparison against 1?
(FalseOp.getValueType() != MVT::f80 || hasFPCMov(CC))) {
SDValue Ops[] = { FalseOp, TrueOp,
DAG.getConstant(CC, MVT::i8), Flags };
- return DAG.getNode(X86ISD::CMOV, DL, N->getVTList(),
- Ops, array_lengthof(Ops));
+ return DAG.getNode(X86ISD::CMOV, DL, N->getVTList(), Ops);
}
// If this is a select between two integer constants, try to do some
// the DCI.xxxx conditions are provided to postpone the optimization as
// late as possible.
- ConstantSDNode *CmpAgainst = 0;
+ ConstantSDNode *CmpAgainst = nullptr;
if ((Cond.getOpcode() == X86ISD::CMP || Cond.getOpcode() == X86ISD::SUB) &&
(CmpAgainst = dyn_cast<ConstantSDNode>(Cond.getOperand(1))) &&
!isa<ConstantSDNode>(Cond.getOperand(0))) {
CmpAgainst == dyn_cast<ConstantSDNode>(TrueOp)) {
SDValue Ops[] = { FalseOp, Cond.getOperand(0),
DAG.getConstant(CC, MVT::i8), Cond };
- return DAG.getNode(X86ISD::CMOV, DL, N->getVTList (), Ops,
- array_lengthof(Ops));
+ return DAG.getNode(X86ISD::CMOV, DL, N->getVTList (), Ops);
}
}
}
N1 = DAG.getNode(ISD::ZERO_EXTEND, DL, WideVT.getScalarType(),
N1->getOperand(0));
SmallVector<SDValue, 8> C(WideVT.getVectorNumElements(), N1);
- N1 = DAG.getNode(ISD::BUILD_VECTOR, DL, WideVT, &C[0], C.size());
+ N1 = DAG.getNode(ISD::BUILD_VECTOR, DL, WideVT, C);
} else if (RHSTrunc) {
N1 = N1->getOperand(0);
}
if (R.getNode())
return R;
- // Create BEXTR and BZHI instructions
- // BZHI is X & ((1 << Y) - 1)
+ // Create BEXTR instructions
// BEXTR is ((X >> imm) & (2**size-1))
if (VT == MVT::i32 || VT == MVT::i64) {
SDValue N0 = N->getOperand(0);
SDValue N1 = N->getOperand(1);
SDLoc DL(N);
- if (Subtarget->hasBMI2()) {
- // Check for (and (add (shl 1, Y), -1), X)
- if (N0.getOpcode() == ISD::ADD && isAllOnes(N0.getOperand(1))) {
- SDValue N00 = N0.getOperand(0);
- if (N00.getOpcode() == ISD::SHL) {
- SDValue N001 = N00.getOperand(1);
- assert(N001.getValueType() == MVT::i8 && "unexpected type");
- ConstantSDNode *C = dyn_cast<ConstantSDNode>(N00.getOperand(0));
- if (C && C->getZExtValue() == 1)
- return DAG.getNode(X86ISD::BZHI, DL, VT, N1, N001);
- }
- }
-
- // Check for (and X, (add (shl 1, Y), -1))
- if (N1.getOpcode() == ISD::ADD && isAllOnes(N1.getOperand(1))) {
- SDValue N10 = N1.getOperand(0);
- if (N10.getOpcode() == ISD::SHL) {
- SDValue N101 = N10.getOperand(1);
- assert(N101.getValueType() == MVT::i8 && "unexpected type");
- ConstantSDNode *C = dyn_cast<ConstantSDNode>(N10.getOperand(0));
- if (C && C->getZExtValue() == 1)
- return DAG.getNode(X86ISD::BZHI, DL, VT, N0, N101);
- }
- }
- }
-
// Check for BEXTR.
if ((Subtarget->hasBMI() || Subtarget->hasTBM()) &&
(N0.getOpcode() == ISD::SRA || N0.getOpcode() == ISD::SRL)) {
SDValue Ops[] = { N0.getOperand(0), Neg,
DAG.getConstant(X86::COND_GE, MVT::i8),
SDValue(Neg.getNode(), 1) };
- return DAG.getNode(X86ISD::CMOV, DL, DAG.getVTList(VT, MVT::Glue),
- Ops, array_lengthof(Ops));
+ return DAG.getNode(X86ISD::CMOV, DL, DAG.getVTList(VT, MVT::Glue), Ops);
}
return SDValue();
}
Ptr = DAG.getNode(ISD::ADD, dl, Ptr.getValueType(), Ptr, Increment);
}
- SDValue TF = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &Chains[0],
- Chains.size());
+ SDValue TF = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Chains);
// Bitcast the loaded value to a vector of the original element type, in
// the size of the target vector type.
Chains.push_back(Ch);
}
- return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, &Chains[0],
- Chains.size());
+ return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Chains);
}
// Turn load->store of MMX types into GPR load/stores. This avoids clobbering
!cast<LoadSDNode>(St->getValue())->isVolatile() &&
St->getChain().hasOneUse() && !St->isVolatile()) {
SDNode* LdVal = St->getValue().getNode();
- LoadSDNode *Ld = 0;
+ LoadSDNode *Ld = nullptr;
int TokenFactorIndex = -1;
SmallVector<SDValue, 8> Ops;
SDNode* ChainVal = St->getChain().getNode();
SDValue NewChain = NewLd.getValue(1);
if (TokenFactorIndex != -1) {
Ops.push_back(NewChain);
- NewChain = DAG.getNode(ISD::TokenFactor, LdDL, MVT::Other, &Ops[0],
- Ops.size());
+ NewChain = DAG.getNode(ISD::TokenFactor, LdDL, MVT::Other, Ops);
}
return DAG.getStore(NewChain, StDL, NewLd, St->getBasePtr(),
St->getPointerInfo(),
if (TokenFactorIndex != -1) {
Ops.push_back(LoLd);
Ops.push_back(HiLd);
- NewChain = DAG.getNode(ISD::TokenFactor, LdDL, MVT::Other, &Ops[0],
- Ops.size());
+ NewChain = DAG.getNode(ISD::TokenFactor, LdDL, MVT::Other, Ops);
}
LoAddr = St->getBasePtr();
Value *CallOperandVal = info.CallOperandVal;
// If we don't have a value, we can't do a match,
// but allow it at the lowest weight.
- if (CallOperandVal == NULL)
+ if (!CallOperandVal)
return CW_Default;
Type *type = CallOperandVal->getType();
// Look at the constraint type.
std::string &Constraint,
std::vector<SDValue>&Ops,
SelectionDAG &DAG) const {
- SDValue Result(0, 0);
+ SDValue Result;
// Only support length 1 constraints for now.
if (Constraint.length() > 1) return;
// If we are in non-pic codegen mode, we allow the address of a global (with
// an optional displacement) to be used with 'i'.
- GlobalAddressSDNode *GA = 0;
+ GlobalAddressSDNode *GA = nullptr;
int64_t Offset = 0;
// Match either (GA), (GA+C), (GA+C1+C2), etc.
Res = TargetLowering::getRegForInlineAsmConstraint(Constraint, VT);
// Not found as a standard register?
- if (Res.second == 0) {
+ if (!Res.second) {
// Map st(0) -> st(7) -> ST0
if (Constraint.size() == 7 && Constraint[0] == '{' &&
tolower(Constraint[1]) == 's' &&
return Res;
}
+
+int X86TargetLowering::getScalingFactorCost(const AddrMode &AM,
+ Type *Ty) const {
+ // Scaling factors are not free at all.
+ // An indexed folded instruction, i.e., inst (reg1, reg2, scale),
+ // will take 2 allocations instead of 1 for plain addressing mode,
+ // i.e. inst (reg1).
+ if (isLegalAddressingMode(AM, Ty))
+ // Scale represents reg2 * scale, thus account for 1
+ // as soon as we use a second register.
+ return AM.Scale != 0;
+ return -1;
+}
projects / oota-llvm.git / blobdiff