void RISCVDAGToDAGISel::Select()

in llvm/lib/Target/RISCV/RISCVISelDAGToDAG.cpp [499:1573]


void RISCVDAGToDAGISel::Select(SDNode *Node) {
  // If we have a custom node, we have already selected.
  if (Node->isMachineOpcode()) {
    LLVM_DEBUG(dbgs() << "== "; Node->dump(CurDAG); dbgs() << "\n");
    Node->setNodeId(-1);
    return;
  }

  // Instruction Selection not handled by the auto-generated tablegen selection
  // should be handled here.
  unsigned Opcode = Node->getOpcode();
  MVT XLenVT = Subtarget->getXLenVT();
  SDLoc DL(Node);
  MVT VT = Node->getSimpleValueType(0);

  switch (Opcode) {
  case ISD::Constant: {
    auto *ConstNode = cast<ConstantSDNode>(Node);
    if (VT == XLenVT && ConstNode->isZero()) {
      SDValue New =
          CurDAG->getCopyFromReg(CurDAG->getEntryNode(), DL, RISCV::X0, XLenVT);
      ReplaceNode(Node, New.getNode());
      return;
    }
    int64_t Imm = ConstNode->getSExtValue();
    // If the upper XLen-16 bits are not used, try to convert this to a simm12
    // by sign extending bit 15.
    if (isUInt<16>(Imm) && isInt<12>(SignExtend64(Imm, 16)) &&
        hasAllHUsers(Node))
      Imm = SignExtend64(Imm, 16);
    // If the upper 32-bits are not used try to convert this into a simm32 by
    // sign extending bit 32.
    if (!isInt<32>(Imm) && isUInt<32>(Imm) && hasAllWUsers(Node))
      Imm = SignExtend64(Imm, 32);

    ReplaceNode(Node, selectImm(CurDAG, DL, VT, Imm, *Subtarget));
    return;
  }
  case ISD::FrameIndex: {
    SDValue Imm = CurDAG->getTargetConstant(0, DL, XLenVT);
    int FI = cast<FrameIndexSDNode>(Node)->getIndex();
    SDValue TFI = CurDAG->getTargetFrameIndex(FI, VT);
    ReplaceNode(Node, CurDAG->getMachineNode(RISCV::ADDI, DL, VT, TFI, Imm));
    return;
  }
  case ISD::SRL: {
    // Optimize (srl (and X, C2), C) ->
    //          (srli (slli X, (XLen-C3), (XLen-C3) + C)
    // Where C2 is a mask with C3 trailing ones.
    // Taking into account that the C2 may have had lower bits unset by
    // SimplifyDemandedBits. This avoids materializing the C2 immediate.
    // This pattern occurs when type legalizing right shifts for types with
    // less than XLen bits.
    auto *N1C = dyn_cast<ConstantSDNode>(Node->getOperand(1));
    if (!N1C)
      break;
    SDValue N0 = Node->getOperand(0);
    if (N0.getOpcode() != ISD::AND || !N0.hasOneUse() ||
        !isa<ConstantSDNode>(N0.getOperand(1)))
      break;
    unsigned ShAmt = N1C->getZExtValue();
    uint64_t Mask = N0.getConstantOperandVal(1);
    Mask |= maskTrailingOnes<uint64_t>(ShAmt);
    if (!isMask_64(Mask))
      break;
    unsigned TrailingOnes = countTrailingOnes(Mask);
    // 32 trailing ones should use srliw via tablegen pattern.
    if (TrailingOnes == 32 || ShAmt >= TrailingOnes)
      break;
    unsigned LShAmt = Subtarget->getXLen() - TrailingOnes;
    SDNode *SLLI =
        CurDAG->getMachineNode(RISCV::SLLI, DL, VT, N0->getOperand(0),
                               CurDAG->getTargetConstant(LShAmt, DL, VT));
    SDNode *SRLI = CurDAG->getMachineNode(
        RISCV::SRLI, DL, VT, SDValue(SLLI, 0),
        CurDAG->getTargetConstant(LShAmt + ShAmt, DL, VT));
    ReplaceNode(Node, SRLI);
    return;
  }
  case ISD::SRA: {
    // Optimize (sra (sext_inreg X, i16), C) ->
    //          (srai (slli X, (XLen-16), (XLen-16) + C)
    // And      (sra (sext_inreg X, i8), C) ->
    //          (srai (slli X, (XLen-8), (XLen-8) + C)
    // This can occur when Zbb is enabled, which makes sext_inreg i16/i8 legal.
    // This transform matches the code we get without Zbb. The shifts are more
    // compressible, and this can help expose CSE opportunities in the sdiv by
    // constant optimization.
    auto *N1C = dyn_cast<ConstantSDNode>(Node->getOperand(1));
    if (!N1C)
      break;
    SDValue N0 = Node->getOperand(0);
    if (N0.getOpcode() != ISD::SIGN_EXTEND_INREG || !N0.hasOneUse())
      break;
    unsigned ShAmt = N1C->getZExtValue();
    unsigned ExtSize =
        cast<VTSDNode>(N0.getOperand(1))->getVT().getSizeInBits();
    // ExtSize of 32 should use sraiw via tablegen pattern.
    if (ExtSize >= 32 || ShAmt >= ExtSize)
      break;
    unsigned LShAmt = Subtarget->getXLen() - ExtSize;
    SDNode *SLLI =
        CurDAG->getMachineNode(RISCV::SLLI, DL, VT, N0->getOperand(0),
                               CurDAG->getTargetConstant(LShAmt, DL, VT));
    SDNode *SRAI = CurDAG->getMachineNode(
        RISCV::SRAI, DL, VT, SDValue(SLLI, 0),
        CurDAG->getTargetConstant(LShAmt + ShAmt, DL, VT));
    ReplaceNode(Node, SRAI);
    return;
  }
  case ISD::AND: {
    auto *N1C = dyn_cast<ConstantSDNode>(Node->getOperand(1));
    if (!N1C)
      break;

    SDValue N0 = Node->getOperand(0);

    bool LeftShift = N0.getOpcode() == ISD::SHL;
    if (!LeftShift && N0.getOpcode() != ISD::SRL)
      break;

    auto *C = dyn_cast<ConstantSDNode>(N0.getOperand(1));
    if (!C)
      break;
    uint64_t C2 = C->getZExtValue();
    unsigned XLen = Subtarget->getXLen();
    if (!C2 || C2 >= XLen)
      break;

    uint64_t C1 = N1C->getZExtValue();

    // Keep track of whether this is a andi, zext.h, or zext.w.
    bool ZExtOrANDI = isInt<12>(N1C->getSExtValue());
    if (C1 == UINT64_C(0xFFFF) &&
        (Subtarget->hasStdExtZbb() || Subtarget->hasStdExtZbp()))
      ZExtOrANDI = true;
    if (C1 == UINT64_C(0xFFFFFFFF) && Subtarget->hasStdExtZba())
      ZExtOrANDI = true;

    // Clear irrelevant bits in the mask.
    if (LeftShift)
      C1 &= maskTrailingZeros<uint64_t>(C2);
    else
      C1 &= maskTrailingOnes<uint64_t>(XLen - C2);

    // Some transforms should only be done if the shift has a single use or
    // the AND would become (srli (slli X, 32), 32)
    bool OneUseOrZExtW = N0.hasOneUse() || C1 == UINT64_C(0xFFFFFFFF);

    SDValue X = N0.getOperand(0);

    // Turn (and (srl x, c2) c1) -> (srli (slli x, c3-c2), c3) if c1 is a mask
    // with c3 leading zeros.
    if (!LeftShift && isMask_64(C1)) {
      uint64_t C3 = XLen - (64 - countLeadingZeros(C1));
      if (C2 < C3) {
        // If the number of leading zeros is C2+32 this can be SRLIW.
        if (C2 + 32 == C3) {
          SDNode *SRLIW =
              CurDAG->getMachineNode(RISCV::SRLIW, DL, XLenVT, X,
                                     CurDAG->getTargetConstant(C2, DL, XLenVT));
          ReplaceNode(Node, SRLIW);
          return;
        }

        // (and (srl (sexti32 Y), c2), c1) -> (srliw (sraiw Y, 31), c3 - 32) if
        // c1 is a mask with c3 leading zeros and c2 >= 32 and c3-c2==1.
        //
        // This pattern occurs when (i32 (srl (sra 31), c3 - 32)) is type
        // legalized and goes through DAG combine.
        SDValue Y;
        if (C2 >= 32 && (C3 - C2) == 1 && N0.hasOneUse() &&
            selectSExti32(X, Y)) {
          SDNode *SRAIW =
              CurDAG->getMachineNode(RISCV::SRAIW, DL, XLenVT, Y,
                                     CurDAG->getTargetConstant(31, DL, XLenVT));
          SDNode *SRLIW = CurDAG->getMachineNode(
              RISCV::SRLIW, DL, XLenVT, SDValue(SRAIW, 0),
              CurDAG->getTargetConstant(C3 - 32, DL, XLenVT));
          ReplaceNode(Node, SRLIW);
          return;
        }

        // (srli (slli x, c3-c2), c3).
        if (OneUseOrZExtW && !ZExtOrANDI) {
          SDNode *SLLI = CurDAG->getMachineNode(
              RISCV::SLLI, DL, XLenVT, X,
              CurDAG->getTargetConstant(C3 - C2, DL, XLenVT));
          SDNode *SRLI =
              CurDAG->getMachineNode(RISCV::SRLI, DL, XLenVT, SDValue(SLLI, 0),
                                     CurDAG->getTargetConstant(C3, DL, XLenVT));
          ReplaceNode(Node, SRLI);
          return;
        }
      }
    }

    // Turn (and (shl x, c2), c1) -> (srli (slli c2+c3), c3) if c1 is a mask
    // shifted by c2 bits with c3 leading zeros.
    if (LeftShift && isShiftedMask_64(C1)) {
      uint64_t C3 = XLen - (64 - countLeadingZeros(C1));

      if (C2 + C3 < XLen &&
          C1 == (maskTrailingOnes<uint64_t>(XLen - (C2 + C3)) << C2)) {
        // Use slli.uw when possible.
        if ((XLen - (C2 + C3)) == 32 && Subtarget->hasStdExtZba()) {
          SDNode *SLLIUW =
              CurDAG->getMachineNode(RISCV::SLLIUW, DL, XLenVT, X,
                                     CurDAG->getTargetConstant(C2, DL, XLenVT));
          ReplaceNode(Node, SLLIUW);
          return;
        }

        // (srli (slli c2+c3), c3)
        if (OneUseOrZExtW && !ZExtOrANDI) {
          SDNode *SLLI = CurDAG->getMachineNode(
              RISCV::SLLI, DL, XLenVT, X,
              CurDAG->getTargetConstant(C2 + C3, DL, XLenVT));
          SDNode *SRLI =
              CurDAG->getMachineNode(RISCV::SRLI, DL, XLenVT, SDValue(SLLI, 0),
                                     CurDAG->getTargetConstant(C3, DL, XLenVT));
          ReplaceNode(Node, SRLI);
          return;
        }
      }
    }

    // Turn (and (shr x, c2), c1) -> (slli (srli x, c2+c3), c3) if c1 is a
    // shifted mask with c2 leading zeros and c3 trailing zeros.
    if (!LeftShift && isShiftedMask_64(C1)) {
      uint64_t Leading = XLen - (64 - countLeadingZeros(C1));
      uint64_t C3 = countTrailingZeros(C1);
      if (Leading == C2 && C2 + C3 < XLen && OneUseOrZExtW && !ZExtOrANDI) {
        SDNode *SRLI = CurDAG->getMachineNode(
            RISCV::SRLI, DL, XLenVT, X,
            CurDAG->getTargetConstant(C2 + C3, DL, XLenVT));
        SDNode *SLLI =
            CurDAG->getMachineNode(RISCV::SLLI, DL, XLenVT, SDValue(SRLI, 0),
                                   CurDAG->getTargetConstant(C3, DL, XLenVT));
        ReplaceNode(Node, SLLI);
        return;
      }
      // If the leading zero count is C2+32, we can use SRLIW instead of SRLI.
      if (Leading > 32 && (Leading - 32) == C2 && C2 + C3 < 32 &&
          OneUseOrZExtW && !ZExtOrANDI) {
        SDNode *SRLIW = CurDAG->getMachineNode(
            RISCV::SRLIW, DL, XLenVT, X,
            CurDAG->getTargetConstant(C2 + C3, DL, XLenVT));
        SDNode *SLLI =
            CurDAG->getMachineNode(RISCV::SLLI, DL, XLenVT, SDValue(SRLIW, 0),
                                   CurDAG->getTargetConstant(C3, DL, XLenVT));
        ReplaceNode(Node, SLLI);
        return;
      }
    }

    // Turn (and (shl x, c2), c1) -> (slli (srli x, c3-c2), c3) if c1 is a
    // shifted mask with no leading zeros and c3 trailing zeros.
    if (LeftShift && isShiftedMask_64(C1)) {
      uint64_t Leading = XLen - (64 - countLeadingZeros(C1));
      uint64_t C3 = countTrailingZeros(C1);
      if (Leading == 0 && C2 < C3 && OneUseOrZExtW && !ZExtOrANDI) {
        SDNode *SRLI = CurDAG->getMachineNode(
            RISCV::SRLI, DL, XLenVT, X,
            CurDAG->getTargetConstant(C3 - C2, DL, XLenVT));
        SDNode *SLLI =
            CurDAG->getMachineNode(RISCV::SLLI, DL, XLenVT, SDValue(SRLI, 0),
                                   CurDAG->getTargetConstant(C3, DL, XLenVT));
        ReplaceNode(Node, SLLI);
        return;
      }
      // If we have (32-C2) leading zeros, we can use SRLIW instead of SRLI.
      if (C2 < C3 && Leading + C2 == 32 && OneUseOrZExtW && !ZExtOrANDI) {
        SDNode *SRLIW = CurDAG->getMachineNode(
            RISCV::SRLIW, DL, XLenVT, X,
            CurDAG->getTargetConstant(C3 - C2, DL, XLenVT));
        SDNode *SLLI =
            CurDAG->getMachineNode(RISCV::SLLI, DL, XLenVT, SDValue(SRLIW, 0),
                                   CurDAG->getTargetConstant(C3, DL, XLenVT));
        ReplaceNode(Node, SLLI);
        return;
      }
    }

    break;
  }
  case ISD::MUL: {
    // Special case for calculating (mul (and X, C2), C1) where the full product
    // fits in XLen bits. We can shift X left by the number of leading zeros in
    // C2 and shift C1 left by XLen-lzcnt(C2). This will ensure the final
    // product has XLen trailing zeros, putting it in the output of MULHU. This
    // can avoid materializing a constant in a register for C2.

    // RHS should be a constant.
    auto *N1C = dyn_cast<ConstantSDNode>(Node->getOperand(1));
    if (!N1C || !N1C->hasOneUse())
      break;

    // LHS should be an AND with constant.
    SDValue N0 = Node->getOperand(0);
    if (N0.getOpcode() != ISD::AND || !isa<ConstantSDNode>(N0.getOperand(1)))
      break;

    uint64_t C2 = cast<ConstantSDNode>(N0.getOperand(1))->getZExtValue();

    // Constant should be a mask.
    if (!isMask_64(C2))
      break;

    // This should be the only use of the AND unless we will use
    // (SRLI (SLLI X, 32), 32). We don't use a shift pair for other AND
    // constants.
    if (!N0.hasOneUse() && C2 != UINT64_C(0xFFFFFFFF))
      break;

    // If this can be an ANDI, ZEXT.H or ZEXT.W we don't need to do this
    // optimization.
    if (isInt<12>(C2) ||
        (C2 == UINT64_C(0xFFFF) &&
         (Subtarget->hasStdExtZbb() || Subtarget->hasStdExtZbp())) ||
        (C2 == UINT64_C(0xFFFFFFFF) && Subtarget->hasStdExtZba()))
      break;

    // We need to shift left the AND input and C1 by a total of XLen bits.

    // How far left do we need to shift the AND input?
    unsigned XLen = Subtarget->getXLen();
    unsigned LeadingZeros = XLen - (64 - countLeadingZeros(C2));

    // The constant gets shifted by the remaining amount unless that would
    // shift bits out.
    uint64_t C1 = N1C->getZExtValue();
    unsigned ConstantShift = XLen - LeadingZeros;
    if (ConstantShift > (XLen - (64 - countLeadingZeros(C1))))
      break;

    uint64_t ShiftedC1 = C1 << ConstantShift;
    // If this RV32, we need to sign extend the constant.
    if (XLen == 32)
      ShiftedC1 = SignExtend64(ShiftedC1, 32);

    // Create (mulhu (slli X, lzcnt(C2)), C1 << (XLen - lzcnt(C2))).
    SDNode *Imm = selectImm(CurDAG, DL, VT, ShiftedC1, *Subtarget);
    SDNode *SLLI =
        CurDAG->getMachineNode(RISCV::SLLI, DL, VT, N0.getOperand(0),
                               CurDAG->getTargetConstant(LeadingZeros, DL, VT));
    SDNode *MULHU = CurDAG->getMachineNode(RISCV::MULHU, DL, VT,
                                           SDValue(SLLI, 0), SDValue(Imm, 0));
    ReplaceNode(Node, MULHU);
    return;
  }
  case ISD::INTRINSIC_WO_CHAIN: {
    unsigned IntNo = Node->getConstantOperandVal(0);
    switch (IntNo) {
      // By default we do not custom select any intrinsic.
    default:
      break;
    case Intrinsic::riscv_vmsgeu:
    case Intrinsic::riscv_vmsge: {
      SDValue Src1 = Node->getOperand(1);
      SDValue Src2 = Node->getOperand(2);
      // Only custom select scalar second operand.
      if (Src2.getValueType() != XLenVT)
        break;
      // Small constants are handled with patterns.
      if (auto *C = dyn_cast<ConstantSDNode>(Src2)) {
        int64_t CVal = C->getSExtValue();
        if (CVal >= -15 && CVal <= 16)
          break;
      }
      bool IsUnsigned = IntNo == Intrinsic::riscv_vmsgeu;
      MVT Src1VT = Src1.getSimpleValueType();
      unsigned VMSLTOpcode, VMNANDOpcode;
      switch (RISCVTargetLowering::getLMUL(Src1VT)) {
      default:
        llvm_unreachable("Unexpected LMUL!");
#define CASE_VMSLT_VMNAND_OPCODES(lmulenum, suffix) \
      case RISCVII::VLMUL::lmulenum: \
        VMSLTOpcode = IsUnsigned ? RISCV::PseudoVMSLTU_VX_##suffix \
                                 : RISCV::PseudoVMSLT_VX_##suffix; \
        VMNANDOpcode = RISCV::PseudoVMNAND_MM_##suffix; \
        break;
      CASE_VMSLT_VMNAND_OPCODES(LMUL_F8, MF8)
      CASE_VMSLT_VMNAND_OPCODES(LMUL_F4, MF4)
      CASE_VMSLT_VMNAND_OPCODES(LMUL_F2, MF2)
      CASE_VMSLT_VMNAND_OPCODES(LMUL_1, M1)
      CASE_VMSLT_VMNAND_OPCODES(LMUL_2, M2)
      CASE_VMSLT_VMNAND_OPCODES(LMUL_4, M4)
      CASE_VMSLT_VMNAND_OPCODES(LMUL_8, M8)
#undef CASE_VMSLT_VMNAND_OPCODES
      }
      SDValue SEW = CurDAG->getTargetConstant(
          Log2_32(Src1VT.getScalarSizeInBits()), DL, XLenVT);
      SDValue VL;
      selectVLOp(Node->getOperand(3), VL);

      // Expand to
      // vmslt{u}.vx vd, va, x; vmnand.mm vd, vd, vd
      SDValue Cmp = SDValue(
          CurDAG->getMachineNode(VMSLTOpcode, DL, VT, {Src1, Src2, VL, SEW}),
          0);
      ReplaceNode(Node, CurDAG->getMachineNode(VMNANDOpcode, DL, VT,
                                               {Cmp, Cmp, VL, SEW}));
      return;
    }
    case Intrinsic::riscv_vmsgeu_mask:
    case Intrinsic::riscv_vmsge_mask: {
      SDValue Src1 = Node->getOperand(2);
      SDValue Src2 = Node->getOperand(3);
      // Only custom select scalar second operand.
      if (Src2.getValueType() != XLenVT)
        break;
      // Small constants are handled with patterns.
      if (auto *C = dyn_cast<ConstantSDNode>(Src2)) {
        int64_t CVal = C->getSExtValue();
        if (CVal >= -15 && CVal <= 16)
          break;
      }
      bool IsUnsigned = IntNo == Intrinsic::riscv_vmsgeu_mask;
      MVT Src1VT = Src1.getSimpleValueType();
      unsigned VMSLTOpcode, VMSLTMaskOpcode, VMXOROpcode, VMANDNOpcode;
      switch (RISCVTargetLowering::getLMUL(Src1VT)) {
      default:
        llvm_unreachable("Unexpected LMUL!");
#define CASE_VMSLT_OPCODES(lmulenum, suffix) \
      case RISCVII::VLMUL::lmulenum: \
        VMSLTOpcode = IsUnsigned ? RISCV::PseudoVMSLTU_VX_##suffix \
                                 : RISCV::PseudoVMSLT_VX_##suffix; \
        VMSLTMaskOpcode = IsUnsigned ? RISCV::PseudoVMSLTU_VX_##suffix##_MASK \
                                     : RISCV::PseudoVMSLT_VX_##suffix##_MASK; \
        break;
      CASE_VMSLT_OPCODES(LMUL_F8, MF8)
      CASE_VMSLT_OPCODES(LMUL_F4, MF4)
      CASE_VMSLT_OPCODES(LMUL_F2, MF2)
      CASE_VMSLT_OPCODES(LMUL_1, M1)
      CASE_VMSLT_OPCODES(LMUL_2, M2)
      CASE_VMSLT_OPCODES(LMUL_4, M4)
      CASE_VMSLT_OPCODES(LMUL_8, M8)
#undef CASE_VMSLT_OPCODES
      }
      // Mask operations use the LMUL from the mask type.
      switch (RISCVTargetLowering::getLMUL(VT)) {
      default:
        llvm_unreachable("Unexpected LMUL!");
#define CASE_VMXOR_VANDN_OPCODES(lmulenum, suffix) \
      case RISCVII::VLMUL::lmulenum: \
        VMXOROpcode = RISCV::PseudoVMXOR_MM_##suffix; \
        VMANDNOpcode = RISCV::PseudoVMANDN_MM_##suffix; \
        break;
      CASE_VMXOR_VANDN_OPCODES(LMUL_F8, MF8)
      CASE_VMXOR_VANDN_OPCODES(LMUL_F4, MF4)
      CASE_VMXOR_VANDN_OPCODES(LMUL_F2, MF2)
      CASE_VMXOR_VANDN_OPCODES(LMUL_1, M1)
      CASE_VMXOR_VANDN_OPCODES(LMUL_2, M2)
      CASE_VMXOR_VANDN_OPCODES(LMUL_4, M4)
      CASE_VMXOR_VANDN_OPCODES(LMUL_8, M8)
#undef CASE_VMXOR_VANDN_OPCODES
      }
      SDValue SEW = CurDAG->getTargetConstant(
          Log2_32(Src1VT.getScalarSizeInBits()), DL, XLenVT);
      SDValue MaskSEW = CurDAG->getTargetConstant(0, DL, XLenVT);
      SDValue VL;
      selectVLOp(Node->getOperand(5), VL);
      SDValue MaskedOff = Node->getOperand(1);
      SDValue Mask = Node->getOperand(4);
      // If the MaskedOff value and the Mask are the same value use
      // vmslt{u}.vx vt, va, x;  vmandn.mm vd, vd, vt
      // This avoids needing to copy v0 to vd before starting the next sequence.
      if (Mask == MaskedOff) {
        SDValue Cmp = SDValue(
            CurDAG->getMachineNode(VMSLTOpcode, DL, VT, {Src1, Src2, VL, SEW}),
            0);
        ReplaceNode(Node, CurDAG->getMachineNode(VMANDNOpcode, DL, VT,
                                                 {Mask, Cmp, VL, MaskSEW}));
        return;
      }

      // Mask needs to be copied to V0.
      SDValue Chain = CurDAG->getCopyToReg(CurDAG->getEntryNode(), DL,
                                           RISCV::V0, Mask, SDValue());
      SDValue Glue = Chain.getValue(1);
      SDValue V0 = CurDAG->getRegister(RISCV::V0, VT);

      // Otherwise use
      // vmslt{u}.vx vd, va, x, v0.t; vmxor.mm vd, vd, v0
      SDValue Cmp = SDValue(
          CurDAG->getMachineNode(VMSLTMaskOpcode, DL, VT,
                                 {MaskedOff, Src1, Src2, V0, VL, SEW, Glue}),
          0);
      ReplaceNode(Node, CurDAG->getMachineNode(VMXOROpcode, DL, VT,
                                               {Cmp, Mask, VL, MaskSEW}));
      return;
    }
    }
    break;
  }
  case ISD::INTRINSIC_W_CHAIN: {
    unsigned IntNo = cast<ConstantSDNode>(Node->getOperand(1))->getZExtValue();
    switch (IntNo) {
      // By default we do not custom select any intrinsic.
    default:
      break;

    case Intrinsic::riscv_vsetvli:
    case Intrinsic::riscv_vsetvlimax: {
      if (!Subtarget->hasVInstructions())
        break;

      bool VLMax = IntNo == Intrinsic::riscv_vsetvlimax;
      unsigned Offset = VLMax ? 2 : 3;

      assert(Node->getNumOperands() == Offset + 2 &&
             "Unexpected number of operands");

      unsigned SEW =
          RISCVVType::decodeVSEW(Node->getConstantOperandVal(Offset) & 0x7);
      RISCVII::VLMUL VLMul = static_cast<RISCVII::VLMUL>(
          Node->getConstantOperandVal(Offset + 1) & 0x7);

      unsigned VTypeI = RISCVVType::encodeVTYPE(
          VLMul, SEW, /*TailAgnostic*/ true, /*MaskAgnostic*/ false);
      SDValue VTypeIOp = CurDAG->getTargetConstant(VTypeI, DL, XLenVT);

      SDValue VLOperand;
      unsigned Opcode = RISCV::PseudoVSETVLI;
      if (VLMax) {
        VLOperand = CurDAG->getRegister(RISCV::X0, XLenVT);
        Opcode = RISCV::PseudoVSETVLIX0;
      } else {
        VLOperand = Node->getOperand(2);

        if (auto *C = dyn_cast<ConstantSDNode>(VLOperand)) {
          uint64_t AVL = C->getZExtValue();
          if (isUInt<5>(AVL)) {
            SDValue VLImm = CurDAG->getTargetConstant(AVL, DL, XLenVT);
            ReplaceNode(
                Node, CurDAG->getMachineNode(RISCV::PseudoVSETIVLI, DL, XLenVT,
                                             MVT::Other, VLImm, VTypeIOp,
                                             /* Chain */ Node->getOperand(0)));
            return;
          }
        }
      }

      ReplaceNode(Node,
                  CurDAG->getMachineNode(Opcode, DL, XLenVT,
                                         MVT::Other, VLOperand, VTypeIOp,
                                         /* Chain */ Node->getOperand(0)));
      return;
    }
    case Intrinsic::riscv_vlseg2:
    case Intrinsic::riscv_vlseg3:
    case Intrinsic::riscv_vlseg4:
    case Intrinsic::riscv_vlseg5:
    case Intrinsic::riscv_vlseg6:
    case Intrinsic::riscv_vlseg7:
    case Intrinsic::riscv_vlseg8: {
      selectVLSEG(Node, /*IsMasked*/ false, /*IsStrided*/ false);
      return;
    }
    case Intrinsic::riscv_vlseg2_mask:
    case Intrinsic::riscv_vlseg3_mask:
    case Intrinsic::riscv_vlseg4_mask:
    case Intrinsic::riscv_vlseg5_mask:
    case Intrinsic::riscv_vlseg6_mask:
    case Intrinsic::riscv_vlseg7_mask:
    case Intrinsic::riscv_vlseg8_mask: {
      selectVLSEG(Node, /*IsMasked*/ true, /*IsStrided*/ false);
      return;
    }
    case Intrinsic::riscv_vlsseg2:
    case Intrinsic::riscv_vlsseg3:
    case Intrinsic::riscv_vlsseg4:
    case Intrinsic::riscv_vlsseg5:
    case Intrinsic::riscv_vlsseg6:
    case Intrinsic::riscv_vlsseg7:
    case Intrinsic::riscv_vlsseg8: {
      selectVLSEG(Node, /*IsMasked*/ false, /*IsStrided*/ true);
      return;
    }
    case Intrinsic::riscv_vlsseg2_mask:
    case Intrinsic::riscv_vlsseg3_mask:
    case Intrinsic::riscv_vlsseg4_mask:
    case Intrinsic::riscv_vlsseg5_mask:
    case Intrinsic::riscv_vlsseg6_mask:
    case Intrinsic::riscv_vlsseg7_mask:
    case Intrinsic::riscv_vlsseg8_mask: {
      selectVLSEG(Node, /*IsMasked*/ true, /*IsStrided*/ true);
      return;
    }
    case Intrinsic::riscv_vloxseg2:
    case Intrinsic::riscv_vloxseg3:
    case Intrinsic::riscv_vloxseg4:
    case Intrinsic::riscv_vloxseg5:
    case Intrinsic::riscv_vloxseg6:
    case Intrinsic::riscv_vloxseg7:
    case Intrinsic::riscv_vloxseg8:
      selectVLXSEG(Node, /*IsMasked*/ false, /*IsOrdered*/ true);
      return;
    case Intrinsic::riscv_vluxseg2:
    case Intrinsic::riscv_vluxseg3:
    case Intrinsic::riscv_vluxseg4:
    case Intrinsic::riscv_vluxseg5:
    case Intrinsic::riscv_vluxseg6:
    case Intrinsic::riscv_vluxseg7:
    case Intrinsic::riscv_vluxseg8:
      selectVLXSEG(Node, /*IsMasked*/ false, /*IsOrdered*/ false);
      return;
    case Intrinsic::riscv_vloxseg2_mask:
    case Intrinsic::riscv_vloxseg3_mask:
    case Intrinsic::riscv_vloxseg4_mask:
    case Intrinsic::riscv_vloxseg5_mask:
    case Intrinsic::riscv_vloxseg6_mask:
    case Intrinsic::riscv_vloxseg7_mask:
    case Intrinsic::riscv_vloxseg8_mask:
      selectVLXSEG(Node, /*IsMasked*/ true, /*IsOrdered*/ true);
      return;
    case Intrinsic::riscv_vluxseg2_mask:
    case Intrinsic::riscv_vluxseg3_mask:
    case Intrinsic::riscv_vluxseg4_mask:
    case Intrinsic::riscv_vluxseg5_mask:
    case Intrinsic::riscv_vluxseg6_mask:
    case Intrinsic::riscv_vluxseg7_mask:
    case Intrinsic::riscv_vluxseg8_mask:
      selectVLXSEG(Node, /*IsMasked*/ true, /*IsOrdered*/ false);
      return;
    case Intrinsic::riscv_vlseg8ff:
    case Intrinsic::riscv_vlseg7ff:
    case Intrinsic::riscv_vlseg6ff:
    case Intrinsic::riscv_vlseg5ff:
    case Intrinsic::riscv_vlseg4ff:
    case Intrinsic::riscv_vlseg3ff:
    case Intrinsic::riscv_vlseg2ff: {
      selectVLSEGFF(Node, /*IsMasked*/ false);
      return;
    }
    case Intrinsic::riscv_vlseg8ff_mask:
    case Intrinsic::riscv_vlseg7ff_mask:
    case Intrinsic::riscv_vlseg6ff_mask:
    case Intrinsic::riscv_vlseg5ff_mask:
    case Intrinsic::riscv_vlseg4ff_mask:
    case Intrinsic::riscv_vlseg3ff_mask:
    case Intrinsic::riscv_vlseg2ff_mask: {
      selectVLSEGFF(Node, /*IsMasked*/ true);
      return;
    }
    case Intrinsic::riscv_vloxei:
    case Intrinsic::riscv_vloxei_mask:
    case Intrinsic::riscv_vluxei:
    case Intrinsic::riscv_vluxei_mask: {
      bool IsMasked = IntNo == Intrinsic::riscv_vloxei_mask ||
                      IntNo == Intrinsic::riscv_vluxei_mask;
      bool IsOrdered = IntNo == Intrinsic::riscv_vloxei ||
                       IntNo == Intrinsic::riscv_vloxei_mask;

      MVT VT = Node->getSimpleValueType(0);
      unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());

      unsigned CurOp = 2;
      SmallVector<SDValue, 8> Operands;
      if (IsMasked)
        Operands.push_back(Node->getOperand(CurOp++));

      MVT IndexVT;
      addVectorLoadStoreOperands(Node, Log2SEW, DL, CurOp, IsMasked,
                                 /*IsStridedOrIndexed*/ true, Operands,
                                 /*IsLoad=*/true, &IndexVT);

      assert(VT.getVectorElementCount() == IndexVT.getVectorElementCount() &&
             "Element count mismatch");

      RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
      RISCVII::VLMUL IndexLMUL = RISCVTargetLowering::getLMUL(IndexVT);
      unsigned IndexLog2EEW = Log2_32(IndexVT.getScalarSizeInBits());
      if (IndexLog2EEW == 6 && !Subtarget->is64Bit()) {
        report_fatal_error("The V extension does not support EEW=64 for index "
                           "values when XLEN=32");
      }
      const RISCV::VLX_VSXPseudo *P = RISCV::getVLXPseudo(
          IsMasked, IsOrdered, IndexLog2EEW, static_cast<unsigned>(LMUL),
          static_cast<unsigned>(IndexLMUL));
      MachineSDNode *Load =
          CurDAG->getMachineNode(P->Pseudo, DL, Node->getVTList(), Operands);

      if (auto *MemOp = dyn_cast<MemSDNode>(Node))
        CurDAG->setNodeMemRefs(Load, {MemOp->getMemOperand()});

      ReplaceNode(Node, Load);
      return;
    }
    case Intrinsic::riscv_vlm:
    case Intrinsic::riscv_vle:
    case Intrinsic::riscv_vle_mask:
    case Intrinsic::riscv_vlse:
    case Intrinsic::riscv_vlse_mask: {
      bool IsMasked = IntNo == Intrinsic::riscv_vle_mask ||
                      IntNo == Intrinsic::riscv_vlse_mask;
      bool IsStrided =
          IntNo == Intrinsic::riscv_vlse || IntNo == Intrinsic::riscv_vlse_mask;

      MVT VT = Node->getSimpleValueType(0);
      unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());

      unsigned CurOp = 2;
      SmallVector<SDValue, 8> Operands;
      if (IsMasked)
        Operands.push_back(Node->getOperand(CurOp++));

      addVectorLoadStoreOperands(Node, Log2SEW, DL, CurOp, IsMasked, IsStrided,
                                 Operands, /*IsLoad=*/true);

      RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
      const RISCV::VLEPseudo *P =
          RISCV::getVLEPseudo(IsMasked, IsStrided, /*FF*/ false, Log2SEW,
                              static_cast<unsigned>(LMUL));
      MachineSDNode *Load =
          CurDAG->getMachineNode(P->Pseudo, DL, Node->getVTList(), Operands);

      if (auto *MemOp = dyn_cast<MemSDNode>(Node))
        CurDAG->setNodeMemRefs(Load, {MemOp->getMemOperand()});

      ReplaceNode(Node, Load);
      return;
    }
    case Intrinsic::riscv_vleff:
    case Intrinsic::riscv_vleff_mask: {
      bool IsMasked = IntNo == Intrinsic::riscv_vleff_mask;

      MVT VT = Node->getSimpleValueType(0);
      unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());

      unsigned CurOp = 2;
      SmallVector<SDValue, 7> Operands;
      if (IsMasked)
        Operands.push_back(Node->getOperand(CurOp++));

      addVectorLoadStoreOperands(Node, Log2SEW, DL, CurOp, IsMasked,
                                 /*IsStridedOrIndexed*/ false, Operands,
                                 /*IsLoad=*/true);

      RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
      const RISCV::VLEPseudo *P =
          RISCV::getVLEPseudo(IsMasked, /*Strided*/ false, /*FF*/ true, Log2SEW,
                              static_cast<unsigned>(LMUL));
      MachineSDNode *Load =
          CurDAG->getMachineNode(P->Pseudo, DL, Node->getValueType(0),
                                 MVT::Other, MVT::Glue, Operands);
      SDNode *ReadVL = CurDAG->getMachineNode(RISCV::PseudoReadVL, DL, XLenVT,
                                              /*Glue*/ SDValue(Load, 2));

      if (auto *MemOp = dyn_cast<MemSDNode>(Node))
        CurDAG->setNodeMemRefs(Load, {MemOp->getMemOperand()});

      ReplaceUses(SDValue(Node, 0), SDValue(Load, 0));
      ReplaceUses(SDValue(Node, 1), SDValue(ReadVL, 0)); // VL
      ReplaceUses(SDValue(Node, 2), SDValue(Load, 1));   // Chain
      CurDAG->RemoveDeadNode(Node);
      return;
    }
    }
    break;
  }
  case ISD::INTRINSIC_VOID: {
    unsigned IntNo = cast<ConstantSDNode>(Node->getOperand(1))->getZExtValue();
    switch (IntNo) {
    case Intrinsic::riscv_vsseg2:
    case Intrinsic::riscv_vsseg3:
    case Intrinsic::riscv_vsseg4:
    case Intrinsic::riscv_vsseg5:
    case Intrinsic::riscv_vsseg6:
    case Intrinsic::riscv_vsseg7:
    case Intrinsic::riscv_vsseg8: {
      selectVSSEG(Node, /*IsMasked*/ false, /*IsStrided*/ false);
      return;
    }
    case Intrinsic::riscv_vsseg2_mask:
    case Intrinsic::riscv_vsseg3_mask:
    case Intrinsic::riscv_vsseg4_mask:
    case Intrinsic::riscv_vsseg5_mask:
    case Intrinsic::riscv_vsseg6_mask:
    case Intrinsic::riscv_vsseg7_mask:
    case Intrinsic::riscv_vsseg8_mask: {
      selectVSSEG(Node, /*IsMasked*/ true, /*IsStrided*/ false);
      return;
    }
    case Intrinsic::riscv_vssseg2:
    case Intrinsic::riscv_vssseg3:
    case Intrinsic::riscv_vssseg4:
    case Intrinsic::riscv_vssseg5:
    case Intrinsic::riscv_vssseg6:
    case Intrinsic::riscv_vssseg7:
    case Intrinsic::riscv_vssseg8: {
      selectVSSEG(Node, /*IsMasked*/ false, /*IsStrided*/ true);
      return;
    }
    case Intrinsic::riscv_vssseg2_mask:
    case Intrinsic::riscv_vssseg3_mask:
    case Intrinsic::riscv_vssseg4_mask:
    case Intrinsic::riscv_vssseg5_mask:
    case Intrinsic::riscv_vssseg6_mask:
    case Intrinsic::riscv_vssseg7_mask:
    case Intrinsic::riscv_vssseg8_mask: {
      selectVSSEG(Node, /*IsMasked*/ true, /*IsStrided*/ true);
      return;
    }
    case Intrinsic::riscv_vsoxseg2:
    case Intrinsic::riscv_vsoxseg3:
    case Intrinsic::riscv_vsoxseg4:
    case Intrinsic::riscv_vsoxseg5:
    case Intrinsic::riscv_vsoxseg6:
    case Intrinsic::riscv_vsoxseg7:
    case Intrinsic::riscv_vsoxseg8:
      selectVSXSEG(Node, /*IsMasked*/ false, /*IsOrdered*/ true);
      return;
    case Intrinsic::riscv_vsuxseg2:
    case Intrinsic::riscv_vsuxseg3:
    case Intrinsic::riscv_vsuxseg4:
    case Intrinsic::riscv_vsuxseg5:
    case Intrinsic::riscv_vsuxseg6:
    case Intrinsic::riscv_vsuxseg7:
    case Intrinsic::riscv_vsuxseg8:
      selectVSXSEG(Node, /*IsMasked*/ false, /*IsOrdered*/ false);
      return;
    case Intrinsic::riscv_vsoxseg2_mask:
    case Intrinsic::riscv_vsoxseg3_mask:
    case Intrinsic::riscv_vsoxseg4_mask:
    case Intrinsic::riscv_vsoxseg5_mask:
    case Intrinsic::riscv_vsoxseg6_mask:
    case Intrinsic::riscv_vsoxseg7_mask:
    case Intrinsic::riscv_vsoxseg8_mask:
      selectVSXSEG(Node, /*IsMasked*/ true, /*IsOrdered*/ true);
      return;
    case Intrinsic::riscv_vsuxseg2_mask:
    case Intrinsic::riscv_vsuxseg3_mask:
    case Intrinsic::riscv_vsuxseg4_mask:
    case Intrinsic::riscv_vsuxseg5_mask:
    case Intrinsic::riscv_vsuxseg6_mask:
    case Intrinsic::riscv_vsuxseg7_mask:
    case Intrinsic::riscv_vsuxseg8_mask:
      selectVSXSEG(Node, /*IsMasked*/ true, /*IsOrdered*/ false);
      return;
    case Intrinsic::riscv_vsoxei:
    case Intrinsic::riscv_vsoxei_mask:
    case Intrinsic::riscv_vsuxei:
    case Intrinsic::riscv_vsuxei_mask: {
      bool IsMasked = IntNo == Intrinsic::riscv_vsoxei_mask ||
                      IntNo == Intrinsic::riscv_vsuxei_mask;
      bool IsOrdered = IntNo == Intrinsic::riscv_vsoxei ||
                       IntNo == Intrinsic::riscv_vsoxei_mask;

      MVT VT = Node->getOperand(2)->getSimpleValueType(0);
      unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());

      unsigned CurOp = 2;
      SmallVector<SDValue, 8> Operands;
      Operands.push_back(Node->getOperand(CurOp++)); // Store value.

      MVT IndexVT;
      addVectorLoadStoreOperands(Node, Log2SEW, DL, CurOp, IsMasked,
                                 /*IsStridedOrIndexed*/ true, Operands,
                                 /*IsLoad=*/false, &IndexVT);

      assert(VT.getVectorElementCount() == IndexVT.getVectorElementCount() &&
             "Element count mismatch");

      RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
      RISCVII::VLMUL IndexLMUL = RISCVTargetLowering::getLMUL(IndexVT);
      unsigned IndexLog2EEW = Log2_32(IndexVT.getScalarSizeInBits());
      if (IndexLog2EEW == 6 && !Subtarget->is64Bit()) {
        report_fatal_error("The V extension does not support EEW=64 for index "
                           "values when XLEN=32");
      }
      const RISCV::VLX_VSXPseudo *P = RISCV::getVSXPseudo(
          IsMasked, IsOrdered, IndexLog2EEW, static_cast<unsigned>(LMUL),
          static_cast<unsigned>(IndexLMUL));
      MachineSDNode *Store =
          CurDAG->getMachineNode(P->Pseudo, DL, Node->getVTList(), Operands);

      if (auto *MemOp = dyn_cast<MemSDNode>(Node))
        CurDAG->setNodeMemRefs(Store, {MemOp->getMemOperand()});

      ReplaceNode(Node, Store);
      return;
    }
    case Intrinsic::riscv_vsm:
    case Intrinsic::riscv_vse:
    case Intrinsic::riscv_vse_mask:
    case Intrinsic::riscv_vsse:
    case Intrinsic::riscv_vsse_mask: {
      bool IsMasked = IntNo == Intrinsic::riscv_vse_mask ||
                      IntNo == Intrinsic::riscv_vsse_mask;
      bool IsStrided =
          IntNo == Intrinsic::riscv_vsse || IntNo == Intrinsic::riscv_vsse_mask;

      MVT VT = Node->getOperand(2)->getSimpleValueType(0);
      unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());

      unsigned CurOp = 2;
      SmallVector<SDValue, 8> Operands;
      Operands.push_back(Node->getOperand(CurOp++)); // Store value.

      addVectorLoadStoreOperands(Node, Log2SEW, DL, CurOp, IsMasked, IsStrided,
                                 Operands);

      RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
      const RISCV::VSEPseudo *P = RISCV::getVSEPseudo(
          IsMasked, IsStrided, Log2SEW, static_cast<unsigned>(LMUL));
      MachineSDNode *Store =
          CurDAG->getMachineNode(P->Pseudo, DL, Node->getVTList(), Operands);
      if (auto *MemOp = dyn_cast<MemSDNode>(Node))
        CurDAG->setNodeMemRefs(Store, {MemOp->getMemOperand()});

      ReplaceNode(Node, Store);
      return;
    }
    }
    break;
  }
  case ISD::BITCAST: {
    MVT SrcVT = Node->getOperand(0).getSimpleValueType();
    // Just drop bitcasts between vectors if both are fixed or both are
    // scalable.
    if ((VT.isScalableVector() && SrcVT.isScalableVector()) ||
        (VT.isFixedLengthVector() && SrcVT.isFixedLengthVector())) {
      ReplaceUses(SDValue(Node, 0), Node->getOperand(0));
      CurDAG->RemoveDeadNode(Node);
      return;
    }
    break;
  }
  case ISD::INSERT_SUBVECTOR: {
    SDValue V = Node->getOperand(0);
    SDValue SubV = Node->getOperand(1);
    SDLoc DL(SubV);
    auto Idx = Node->getConstantOperandVal(2);
    MVT SubVecVT = SubV.getSimpleValueType();

    const RISCVTargetLowering &TLI = *Subtarget->getTargetLowering();
    MVT SubVecContainerVT = SubVecVT;
    // Establish the correct scalable-vector types for any fixed-length type.
    if (SubVecVT.isFixedLengthVector())
      SubVecContainerVT = TLI.getContainerForFixedLengthVector(SubVecVT);
    if (VT.isFixedLengthVector())
      VT = TLI.getContainerForFixedLengthVector(VT);

    const auto *TRI = Subtarget->getRegisterInfo();
    unsigned SubRegIdx;
    std::tie(SubRegIdx, Idx) =
        RISCVTargetLowering::decomposeSubvectorInsertExtractToSubRegs(
            VT, SubVecContainerVT, Idx, TRI);

    // If the Idx hasn't been completely eliminated then this is a subvector
    // insert which doesn't naturally align to a vector register. These must
    // be handled using instructions to manipulate the vector registers.
    if (Idx != 0)
      break;

    RISCVII::VLMUL SubVecLMUL = RISCVTargetLowering::getLMUL(SubVecContainerVT);
    bool IsSubVecPartReg = SubVecLMUL == RISCVII::VLMUL::LMUL_F2 ||
                           SubVecLMUL == RISCVII::VLMUL::LMUL_F4 ||
                           SubVecLMUL == RISCVII::VLMUL::LMUL_F8;
    (void)IsSubVecPartReg; // Silence unused variable warning without asserts.
    assert((!IsSubVecPartReg || V.isUndef()) &&
           "Expecting lowering to have created legal INSERT_SUBVECTORs when "
           "the subvector is smaller than a full-sized register");

    // If we haven't set a SubRegIdx, then we must be going between
    // equally-sized LMUL groups (e.g. VR -> VR). This can be done as a copy.
    if (SubRegIdx == RISCV::NoSubRegister) {
      unsigned InRegClassID = RISCVTargetLowering::getRegClassIDForVecVT(VT);
      assert(RISCVTargetLowering::getRegClassIDForVecVT(SubVecContainerVT) ==
                 InRegClassID &&
             "Unexpected subvector extraction");
      SDValue RC = CurDAG->getTargetConstant(InRegClassID, DL, XLenVT);
      SDNode *NewNode = CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS,
                                               DL, VT, SubV, RC);
      ReplaceNode(Node, NewNode);
      return;
    }

    SDValue Insert = CurDAG->getTargetInsertSubreg(SubRegIdx, DL, VT, V, SubV);
    ReplaceNode(Node, Insert.getNode());
    return;
  }
  case ISD::EXTRACT_SUBVECTOR: {
    SDValue V = Node->getOperand(0);
    auto Idx = Node->getConstantOperandVal(1);
    MVT InVT = V.getSimpleValueType();
    SDLoc DL(V);

    const RISCVTargetLowering &TLI = *Subtarget->getTargetLowering();
    MVT SubVecContainerVT = VT;
    // Establish the correct scalable-vector types for any fixed-length type.
    if (VT.isFixedLengthVector())
      SubVecContainerVT = TLI.getContainerForFixedLengthVector(VT);
    if (InVT.isFixedLengthVector())
      InVT = TLI.getContainerForFixedLengthVector(InVT);

    const auto *TRI = Subtarget->getRegisterInfo();
    unsigned SubRegIdx;
    std::tie(SubRegIdx, Idx) =
        RISCVTargetLowering::decomposeSubvectorInsertExtractToSubRegs(
            InVT, SubVecContainerVT, Idx, TRI);

    // If the Idx hasn't been completely eliminated then this is a subvector
    // extract which doesn't naturally align to a vector register. These must
    // be handled using instructions to manipulate the vector registers.
    if (Idx != 0)
      break;

    // If we haven't set a SubRegIdx, then we must be going between
    // equally-sized LMUL types (e.g. VR -> VR). This can be done as a copy.
    if (SubRegIdx == RISCV::NoSubRegister) {
      unsigned InRegClassID = RISCVTargetLowering::getRegClassIDForVecVT(InVT);
      assert(RISCVTargetLowering::getRegClassIDForVecVT(SubVecContainerVT) ==
                 InRegClassID &&
             "Unexpected subvector extraction");
      SDValue RC = CurDAG->getTargetConstant(InRegClassID, DL, XLenVT);
      SDNode *NewNode =
          CurDAG->getMachineNode(TargetOpcode::COPY_TO_REGCLASS, DL, VT, V, RC);
      ReplaceNode(Node, NewNode);
      return;
    }

    SDValue Extract = CurDAG->getTargetExtractSubreg(SubRegIdx, DL, VT, V);
    ReplaceNode(Node, Extract.getNode());
    return;
  }
  case ISD::SPLAT_VECTOR:
  case RISCVISD::VMV_V_X_VL:
  case RISCVISD::VFMV_V_F_VL: {
    // Try to match splat of a scalar load to a strided load with stride of x0.
    SDValue Src = Node->getOperand(0);
    auto *Ld = dyn_cast<LoadSDNode>(Src);
    if (!Ld)
      break;
    EVT MemVT = Ld->getMemoryVT();
    // The memory VT should be the same size as the element type.
    if (MemVT.getStoreSize() != VT.getVectorElementType().getStoreSize())
      break;
    if (!IsProfitableToFold(Src, Node, Node) ||
        !IsLegalToFold(Src, Node, Node, TM.getOptLevel()))
      break;

    SDValue VL;
    if (Node->getOpcode() == ISD::SPLAT_VECTOR)
      VL = CurDAG->getTargetConstant(RISCV::VLMaxSentinel, DL, XLenVT);
    else
      selectVLOp(Node->getOperand(1), VL);

    unsigned Log2SEW = Log2_32(VT.getScalarSizeInBits());
    SDValue SEW = CurDAG->getTargetConstant(Log2SEW, DL, XLenVT);

    SDValue Operands[] = {Ld->getBasePtr(),
                          CurDAG->getRegister(RISCV::X0, XLenVT), VL, SEW,
                          Ld->getChain()};

    RISCVII::VLMUL LMUL = RISCVTargetLowering::getLMUL(VT);
    const RISCV::VLEPseudo *P = RISCV::getVLEPseudo(
        /*IsMasked*/ false, /*IsStrided*/ true, /*FF*/ false, Log2SEW,
        static_cast<unsigned>(LMUL));
    MachineSDNode *Load =
        CurDAG->getMachineNode(P->Pseudo, DL, Node->getVTList(), Operands);

    if (auto *MemOp = dyn_cast<MemSDNode>(Node))
      CurDAG->setNodeMemRefs(Load, {MemOp->getMemOperand()});

    ReplaceNode(Node, Load);
    return;
  }
  }

  // Select the default instruction.
  SelectCode(Node);
}