LoongArchISelLowering.cpp

Go to the documentation of this file.

//=- LoongArchISelLowering.cpp - LoongArch DAG Lowering Implementation  ---===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
//
// This file defines the interfaces that LoongArch uses to lower LLVM code into
// a selection DAG.
//
//===----------------------------------------------------------------------===//
 
#include "LoongArchISelLowering.h"
#include "LoongArch.h"
#include "LoongArchMachineFunctionInfo.h"
#include "LoongArchRegisterInfo.h"
#include "LoongArchSubtarget.h"
#include "LoongArchTargetMachine.h"
#include "MCTargetDesc/LoongArchBaseInfo.h"
#include "MCTargetDesc/LoongArchMCTargetDesc.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/StringExtras.h"
#include "llvm/CodeGen/ISDOpcodes.h"
#include "llvm/CodeGen/RuntimeLibcallUtil.h"
#include "llvm/CodeGen/SelectionDAGNodes.h"
#include "llvm/IR/IRBuilder.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/IntrinsicsLoongArch.h"
#include "llvm/Support/CodeGen.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/ErrorHandling.h"
#include "llvm/Support/KnownBits.h"
#include "llvm/Support/MathExtras.h"
 
using namespace llvm;
 
#define DEBUG_TYPE "loongarch-isel-lowering"
 
STATISTIC(NumTailCalls, "Number of tail calls");
 
static cl::opt<bool> ZeroDivCheck("loongarch-check-zero-division", cl::Hidden,
                                  cl::desc("Trap on integer division by zero."),
                                  cl::init(false));
 
LoongArchTargetLowering::LoongArchTargetLowering(const TargetMachine &TM,
                                                 const LoongArchSubtarget &STI)
    : TargetLowering(TM), Subtarget(STI) {
 
  MVT GRLenVT = Subtarget.getGRLenVT();
 
  // Set up the register classes.
 
  addRegisterClass(GRLenVT, &LoongArch::GPRRegClass);
  if (Subtarget.hasBasicF())
    addRegisterClass(MVT::f32, &LoongArch::FPR32RegClass);
  if (Subtarget.hasBasicD())
    addRegisterClass(MVT::f64, &LoongArch::FPR64RegClass);
 
  static const MVT::SimpleValueType LSXVTs[] = {
      MVT::v16i8, MVT::v8i16, MVT::v4i32, MVT::v2i64, MVT::v4f32, MVT::v2f64};
  static const MVT::SimpleValueType LASXVTs[] = {
      MVT::v32i8, MVT::v16i16, MVT::v8i32, MVT::v4i64, MVT::v8f32, MVT::v4f64};
 
  if (Subtarget.hasExtLSX())
    for (MVT VT : LSXVTs)
      addRegisterClass(VT, &LoongArch::LSX128RegClass);
 
  if (Subtarget.hasExtLASX())
    for (MVT VT : LASXVTs)
      addRegisterClass(VT, &LoongArch::LASX256RegClass);
 
  // Set operations for LA32 and LA64.
 
  setLoadExtAction({ISD::EXTLOAD, ISD::SEXTLOAD, ISD::ZEXTLOAD}, GRLenVT,
                   MVT::i1, Promote);
 
  setOperationAction(ISD::SHL_PARTS, GRLenVT, Custom);
  setOperationAction(ISD::SRA_PARTS, GRLenVT, Custom);
  setOperationAction(ISD::SRL_PARTS, GRLenVT, Custom);
  setOperationAction(ISD::FP_TO_SINT, GRLenVT, Custom);
  setOperationAction(ISD::ROTL, GRLenVT, Expand);
  setOperationAction(ISD::CTPOP, GRLenVT, Expand);
 
  setOperationAction({ISD::GlobalAddress, ISD::BlockAddress, ISD::ConstantPool,
                      ISD::JumpTable, ISD::GlobalTLSAddress},
                     GRLenVT, Custom);
 
  setOperationAction(ISD::EH_DWARF_CFA, GRLenVT, Custom);
 
  setOperationAction(ISD::DYNAMIC_STACKALLOC, GRLenVT, Expand);
  setOperationAction({ISD::STACKSAVE, ISD::STACKRESTORE}, MVT::Other, Expand);
  setOperationAction(ISD::VASTART, MVT::Other, Custom);
  setOperationAction({ISD::VAARG, ISD::VACOPY, ISD::VAEND}, MVT::Other, Expand);
 
  setOperationAction(ISD::DEBUGTRAP, MVT::Other, Legal);
  setOperationAction(ISD::TRAP, MVT::Other, Legal);
 
  setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
  setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
  setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
 
  // Expand bitreverse.i16 with native-width bitrev and shift for now, before
  // we get to know which of sll and revb.2h is faster.
  setOperationAction(ISD::BITREVERSE, MVT::i8, Custom);
  setOperationAction(ISD::BITREVERSE, GRLenVT, Legal);
 
  // LA32 does not have REVB.2W and REVB.D due to the 64-bit operands, and
  // the narrower REVB.W does not exist. But LA32 does have REVB.2H, so i16
  // and i32 could still be byte-swapped relatively cheaply.
  setOperationAction(ISD::BSWAP, MVT::i16, Custom);
 
  setOperationAction(ISD::BR_JT, MVT::Other, Expand);
  setOperationAction(ISD::BR_CC, GRLenVT, Expand);
  setOperationAction(ISD::SELECT_CC, GRLenVT, Expand);
  setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
  setOperationAction({ISD::SMUL_LOHI, ISD::UMUL_LOHI}, GRLenVT, Expand);
 
  setOperationAction(ISD::FP_TO_UINT, GRLenVT, Custom);
  setOperationAction(ISD::UINT_TO_FP, GRLenVT, Expand);
 
  // Set operations for LA64 only.
 
  if (Subtarget.is64Bit()) {
    setOperationAction(ISD::ADD, MVT::i32, Custom);
    setOperationAction(ISD::SUB, MVT::i32, Custom);
    setOperationAction(ISD::SHL, MVT::i32, Custom);
    setOperationAction(ISD::SRA, MVT::i32, Custom);
    setOperationAction(ISD::SRL, MVT::i32, Custom);
    setOperationAction(ISD::FP_TO_SINT, MVT::i32, Custom);
    setOperationAction(ISD::BITCAST, MVT::i32, Custom);
    setOperationAction(ISD::ROTR, MVT::i32, Custom);
    setOperationAction(ISD::ROTL, MVT::i32, Custom);
    setOperationAction(ISD::CTTZ, MVT::i32, Custom);
    setOperationAction(ISD::CTLZ, MVT::i32, Custom);
    setOperationAction(ISD::EH_DWARF_CFA, MVT::i32, Custom);
    setOperationAction(ISD::READ_REGISTER, MVT::i32, Custom);
    setOperationAction(ISD::WRITE_REGISTER, MVT::i32, Custom);
    setOperationAction(ISD::INTRINSIC_VOID, MVT::i32, Custom);
    setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::i32, Custom);
    setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::i32, Custom);
 
    setOperationAction(ISD::BITREVERSE, MVT::i32, Custom);
    setOperationAction(ISD::BSWAP, MVT::i32, Custom);
    setOperationAction({ISD::UDIV, ISD::UREM}, MVT::i32, Custom);
  }
 
  // Set operations for LA32 only.
 
  if (!Subtarget.is64Bit()) {
    setOperationAction(ISD::READ_REGISTER, MVT::i64, Custom);
    setOperationAction(ISD::WRITE_REGISTER, MVT::i64, Custom);
    setOperationAction(ISD::INTRINSIC_VOID, MVT::i64, Custom);
    setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::i64, Custom);
    setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::i64, Custom);
  }
 
  setOperationAction(ISD::ATOMIC_FENCE, MVT::Other, Custom);
 
  static const ISD::CondCode FPCCToExpand[] = {
      ISD::SETOGT, ISD::SETOGE, ISD::SETUGT, ISD::SETUGE,
      ISD::SETGE,  ISD::SETNE,  ISD::SETGT};
 
  // Set operations for 'F' feature.
 
  if (Subtarget.hasBasicF()) {
    setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
    setTruncStoreAction(MVT::f32, MVT::f16, Expand);
    setCondCodeAction(FPCCToExpand, MVT::f32, Expand);
 
    setOperationAction(ISD::SELECT_CC, MVT::f32, Expand);
    setOperationAction(ISD::BR_CC, MVT::f32, Expand);
    setOperationAction(ISD::FMA, MVT::f32, Legal);
    setOperationAction(ISD::FMINNUM_IEEE, MVT::f32, Legal);
    setOperationAction(ISD::FMAXNUM_IEEE, MVT::f32, Legal);
    setOperationAction(ISD::STRICT_FSETCCS, MVT::f32, Legal);
    setOperationAction(ISD::STRICT_FSETCC, MVT::f32, Legal);
    setOperationAction(ISD::IS_FPCLASS, MVT::f32, Legal);
    setOperationAction(ISD::FSIN, MVT::f32, Expand);
    setOperationAction(ISD::FCOS, MVT::f32, Expand);
    setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
    setOperationAction(ISD::FPOW, MVT::f32, Expand);
    setOperationAction(ISD::FREM, MVT::f32, Expand);
    setOperationAction(ISD::FP16_TO_FP, MVT::f32, Expand);
    setOperationAction(ISD::FP_TO_FP16, MVT::f32, Expand);
 
    if (Subtarget.is64Bit())
      setOperationAction(ISD::FRINT, MVT::f32, Legal);
 
    if (!Subtarget.hasBasicD()) {
      setOperationAction(ISD::FP_TO_UINT, MVT::i32, Custom);
      if (Subtarget.is64Bit()) {
        setOperationAction(ISD::SINT_TO_FP, MVT::i64, Custom);
        setOperationAction(ISD::UINT_TO_FP, MVT::i64, Custom);
      }
    }
  }
 
  // Set operations for 'D' feature.
 
  if (Subtarget.hasBasicD()) {
    setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Expand);
    setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f32, Expand);
    setTruncStoreAction(MVT::f64, MVT::f16, Expand);
    setTruncStoreAction(MVT::f64, MVT::f32, Expand);
    setCondCodeAction(FPCCToExpand, MVT::f64, Expand);
 
    setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
    setOperationAction(ISD::BR_CC, MVT::f64, Expand);
    setOperationAction(ISD::STRICT_FSETCCS, MVT::f64, Legal);
    setOperationAction(ISD::STRICT_FSETCC, MVT::f64, Legal);
    setOperationAction(ISD::FMA, MVT::f64, Legal);
    setOperationAction(ISD::FMINNUM_IEEE, MVT::f64, Legal);
    setOperationAction(ISD::FMAXNUM_IEEE, MVT::f64, Legal);
    setOperationAction(ISD::IS_FPCLASS, MVT::f64, Legal);
    setOperationAction(ISD::FSIN, MVT::f64, Expand);
    setOperationAction(ISD::FCOS, MVT::f64, Expand);
    setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
    setOperationAction(ISD::FPOW, MVT::f64, Expand);
    setOperationAction(ISD::FREM, MVT::f64, Expand);
    setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
    setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
 
    if (Subtarget.is64Bit())
      setOperationAction(ISD::FRINT, MVT::f64, Legal);
  }
 
  // Set operations for 'LSX' feature.
 
  if (Subtarget.hasExtLSX()) {
    for (MVT VT : MVT::fixedlen_vector_valuetypes()) {
      // Expand all truncating stores and extending loads.
      for (MVT InnerVT : MVT::fixedlen_vector_valuetypes()) {
        setTruncStoreAction(VT, InnerVT, Expand);
        setLoadExtAction(ISD::SEXTLOAD, VT, InnerVT, Expand);
        setLoadExtAction(ISD::ZEXTLOAD, VT, InnerVT, Expand);
        setLoadExtAction(ISD::EXTLOAD, VT, InnerVT, Expand);
      }
      // By default everything must be expanded. Then we will selectively turn
      // on ones that can be effectively codegen'd.
      for (unsigned Op = 0; Op < ISD::BUILTIN_OP_END; ++Op)
        setOperationAction(Op, VT, Expand);
    }
 
    for (MVT VT : LSXVTs) {
      setOperationAction({ISD::LOAD, ISD::STORE}, VT, Legal);
      setOperationAction(ISD::BITCAST, VT, Legal);
      setOperationAction(ISD::UNDEF, VT, Legal);
 
      setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
      setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Legal);
      setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
 
      setOperationAction(ISD::SETCC, VT, Legal);
      setOperationAction(ISD::VSELECT, VT, Legal);
      setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
    }
    for (MVT VT : {MVT::v16i8, MVT::v8i16, MVT::v4i32, MVT::v2i64}) {
      setOperationAction({ISD::ADD, ISD::SUB}, VT, Legal);
      setOperationAction({ISD::UMAX, ISD::UMIN, ISD::SMAX, ISD::SMIN}, VT,
                         Legal);
      setOperationAction({ISD::MUL, ISD::SDIV, ISD::SREM, ISD::UDIV, ISD::UREM},
                         VT, Legal);
      setOperationAction({ISD::AND, ISD::OR, ISD::XOR}, VT, Legal);
      setOperationAction({ISD::SHL, ISD::SRA, ISD::SRL}, VT, Legal);
      setOperationAction({ISD::CTPOP, ISD::CTLZ}, VT, Legal);
      setOperationAction({ISD::MULHS, ISD::MULHU}, VT, Legal);
      setCondCodeAction(
          {ISD::SETNE, ISD::SETGE, ISD::SETGT, ISD::SETUGE, ISD::SETUGT}, VT,
          Expand);
    }
    for (MVT VT : {MVT::v4i32, MVT::v2i64}) {
      setOperationAction({ISD::SINT_TO_FP, ISD::UINT_TO_FP}, VT, Legal);
      setOperationAction({ISD::FP_TO_SINT, ISD::FP_TO_UINT}, VT, Legal);
    }
    for (MVT VT : {MVT::v4f32, MVT::v2f64}) {
      setOperationAction({ISD::FADD, ISD::FSUB}, VT, Legal);
      setOperationAction({ISD::FMUL, ISD::FDIV}, VT, Legal);
      setOperationAction(ISD::FMA, VT, Legal);
      setOperationAction(ISD::FSQRT, VT, Legal);
      setOperationAction(ISD::FNEG, VT, Legal);
      setCondCodeAction({ISD::SETGE, ISD::SETGT, ISD::SETOGE, ISD::SETOGT,
                         ISD::SETUGE, ISD::SETUGT},
                        VT, Expand);
    }
  }
 
  // Set operations for 'LASX' feature.
 
  if (Subtarget.hasExtLASX()) {
    for (MVT VT : LASXVTs) {
      setOperationAction({ISD::LOAD, ISD::STORE}, VT, Legal);
      setOperationAction(ISD::BITCAST, VT, Legal);
      setOperationAction(ISD::UNDEF, VT, Legal);
 
      setOperationAction(ISD::INSERT_VECTOR_ELT, VT, Custom);
      setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
      setOperationAction(ISD::BUILD_VECTOR, VT, Custom);
 
      setOperationAction(ISD::SETCC, VT, Legal);
      setOperationAction(ISD::VSELECT, VT, Legal);
      setOperationAction(ISD::VECTOR_SHUFFLE, VT, Custom);
    }
    for (MVT VT : {MVT::v4i64, MVT::v8i32, MVT::v16i16, MVT::v32i8}) {
      setOperationAction({ISD::ADD, ISD::SUB}, VT, Legal);
      setOperationAction({ISD::UMAX, ISD::UMIN, ISD::SMAX, ISD::SMIN}, VT,
                         Legal);
      setOperationAction({ISD::MUL, ISD::SDIV, ISD::SREM, ISD::UDIV, ISD::UREM},
                         VT, Legal);
      setOperationAction({ISD::AND, ISD::OR, ISD::XOR}, VT, Legal);
      setOperationAction({ISD::SHL, ISD::SRA, ISD::SRL}, VT, Legal);
      setOperationAction({ISD::CTPOP, ISD::CTLZ}, VT, Legal);
      setOperationAction({ISD::MULHS, ISD::MULHU}, VT, Legal);
      setCondCodeAction(
          {ISD::SETNE, ISD::SETGE, ISD::SETGT, ISD::SETUGE, ISD::SETUGT}, VT,
          Expand);
    }
    for (MVT VT : {MVT::v8i32, MVT::v4i32, MVT::v4i64}) {
      setOperationAction({ISD::SINT_TO_FP, ISD::UINT_TO_FP}, VT, Legal);
      setOperationAction({ISD::FP_TO_SINT, ISD::FP_TO_UINT}, VT, Legal);
    }
    for (MVT VT : {MVT::v8f32, MVT::v4f64}) {
      setOperationAction({ISD::FADD, ISD::FSUB}, VT, Legal);
      setOperationAction({ISD::FMUL, ISD::FDIV}, VT, Legal);
      setOperationAction(ISD::FMA, VT, Legal);
      setOperationAction(ISD::FSQRT, VT, Legal);
      setOperationAction(ISD::FNEG, VT, Legal);
      setCondCodeAction({ISD::SETGE, ISD::SETGT, ISD::SETOGE, ISD::SETOGT,
                         ISD::SETUGE, ISD::SETUGT},
                        VT, Expand);
    }
  }
 
  // Set DAG combine for LA32 and LA64.
 
  setTargetDAGCombine(ISD::AND);
  setTargetDAGCombine(ISD::OR);
  setTargetDAGCombine(ISD::SRL);
  setTargetDAGCombine(ISD::SETCC);
 
  // Set DAG combine for 'LSX' feature.
 
  if (Subtarget.hasExtLSX())
    setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
 
  // Compute derived properties from the register classes.
  computeRegisterProperties(Subtarget.getRegisterInfo());
 
  setStackPointerRegisterToSaveRestore(LoongArch::R3);
 
  setBooleanContents(ZeroOrOneBooleanContent);
  setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
 
  setMaxAtomicSizeInBitsSupported(Subtarget.getGRLen());
 
  setMinCmpXchgSizeInBits(32);
 
  // Function alignments.
  setMinFunctionAlignment(Align(4));
  // Set preferred alignments.
  setPrefFunctionAlignment(Subtarget.getPrefFunctionAlignment());
  setPrefLoopAlignment(Subtarget.getPrefLoopAlignment());
  setMaxBytesForAlignment(Subtarget.getMaxBytesForAlignment());
}
 
bool LoongArchTargetLowering::isOffsetFoldingLegal(
    const GlobalAddressSDNode *GA) const {
  // In order to maximise the opportunity for common subexpression elimination,
  // keep a separate ADD node for the global address offset instead of folding
  // it in the global address node. Later peephole optimisations may choose to
  // fold it back in when profitable.
  return false;
}
 
SDValue LoongArchTargetLowering::LowerOperation(SDValue Op,
                                                SelectionDAG &DAG) const {
  switch (Op.getOpcode()) {
  case ISD::ATOMIC_FENCE:
    return lowerATOMIC_FENCE(Op, DAG);
  case ISD::EH_DWARF_CFA:
    return lowerEH_DWARF_CFA(Op, DAG);
  case ISD::GlobalAddress:
    return lowerGlobalAddress(Op, DAG);
  case ISD::GlobalTLSAddress:
    return lowerGlobalTLSAddress(Op, DAG);
  case ISD::INTRINSIC_WO_CHAIN:
    return lowerINTRINSIC_WO_CHAIN(Op, DAG);
  case ISD::INTRINSIC_W_CHAIN:
    return lowerINTRINSIC_W_CHAIN(Op, DAG);
  case ISD::INTRINSIC_VOID:
    return lowerINTRINSIC_VOID(Op, DAG);
  case ISD::BlockAddress:
    return lowerBlockAddress(Op, DAG);
  case ISD::JumpTable:
    return lowerJumpTable(Op, DAG);
  case ISD::SHL_PARTS:
    return lowerShiftLeftParts(Op, DAG);
  case ISD::SRA_PARTS:
    return lowerShiftRightParts(Op, DAG, true);
  case ISD::SRL_PARTS:
    return lowerShiftRightParts(Op, DAG, false);
  case ISD::ConstantPool:
    return lowerConstantPool(Op, DAG);
  case ISD::FP_TO_SINT:
    return lowerFP_TO_SINT(Op, DAG);
  case ISD::BITCAST:
    return lowerBITCAST(Op, DAG);
  case ISD::UINT_TO_FP:
    return lowerUINT_TO_FP(Op, DAG);
  case ISD::SINT_TO_FP:
    return lowerSINT_TO_FP(Op, DAG);
  case ISD::VASTART:
    return lowerVASTART(Op, DAG);
  case ISD::FRAMEADDR:
    return lowerFRAMEADDR(Op, DAG);
  case ISD::RETURNADDR:
    return lowerRETURNADDR(Op, DAG);
  case ISD::WRITE_REGISTER:
    return lowerWRITE_REGISTER(Op, DAG);
  case ISD::INSERT_VECTOR_ELT:
    return lowerINSERT_VECTOR_ELT(Op, DAG);
  case ISD::EXTRACT_VECTOR_ELT:
    return lowerEXTRACT_VECTOR_ELT(Op, DAG);
  case ISD::BUILD_VECTOR:
    return lowerBUILD_VECTOR(Op, DAG);
  case ISD::VECTOR_SHUFFLE:
    return lowerVECTOR_SHUFFLE(Op, DAG);
  }
  return SDValue();
}
 
/// Determine whether a range fits a regular pattern of values.
/// This function accounts for the possibility of jumping over the End iterator.
template <typename ValType>
static bool
fitsRegularPattern(typename SmallVectorImpl<ValType>::const_iterator Begin,
                   unsigned CheckStride,
                   typename SmallVectorImpl<ValType>::const_iterator End,
                   ValType ExpectedIndex, unsigned ExpectedIndexStride) {
  auto &I = Begin;
 
  while (I != End) {
    if (*I != -1 && *I != ExpectedIndex)
      return false;
    ExpectedIndex += ExpectedIndexStride;
 
    // Incrementing past End is undefined behaviour so we must increment one
    // step at a time and check for End at each step.
    for (unsigned n = 0; n < CheckStride && I != End; ++n, ++I)
      ; // Empty loop body.
  }
  return true;
}
 
/// Lower VECTOR_SHUFFLE into VREPLVEI (if possible).
///
/// VREPLVEI performs vector broadcast based on an element specified by an
/// integer immediate, with its mask being similar to:
///   <x, x, x, ...>
/// where x is any valid index.
///
/// When undef's appear in the mask they are treated as if they were whatever
/// value is necessary in order to fit the above form.
static SDValue lowerVECTOR_SHUFFLE_VREPLVEI(const SDLoc &DL, ArrayRef<int> Mask,
                                            MVT VT, SDValue V1, SDValue V2,
                                            SelectionDAG &DAG) {
  int SplatIndex = -1;
  for (const auto &M : Mask) {
    if (M != -1) {
      SplatIndex = M;
      break;
    }
  }
 
  if (SplatIndex == -1)
    return DAG.getUNDEF(VT);
 
  assert(SplatIndex < (int)Mask.size() && "Out of bounds mask index");
  if (fitsRegularPattern<int>(Mask.begin(), 1, Mask.end(), SplatIndex, 0)) {
    APInt Imm(64, SplatIndex);
    return DAG.getNode(LoongArchISD::VREPLVEI, DL, VT, V1,
                       DAG.getConstant(Imm, DL, MVT::i64));
  }
 
  return SDValue();
}
 
/// Lower VECTOR_SHUFFLE into VSHUF4I (if possible).
///
/// VSHUF4I splits the vector into blocks of four elements, then shuffles these
/// elements according to a <4 x i2> constant (encoded as an integer immediate).
///
/// It is therefore possible to lower into VSHUF4I when the mask takes the form:
///   <a, b, c, d, a+4, b+4, c+4, d+4, a+8, b+8, c+8, d+8, ...>
/// When undef's appear they are treated as if they were whatever value is
/// necessary in order to fit the above forms.
///
/// For example:
///   %2 = shufflevector <8 x i16> %0, <8 x i16> undef,
///                      <8 x i32> <i32 3, i32 2, i32 1, i32 0,
///                                 i32 7, i32 6, i32 5, i32 4>
/// is lowered to:
///   (VSHUF4I_H $v0, $v1, 27)
/// where the 27 comes from:
///   3 + (2 << 2) + (1 << 4) + (0 << 6)
static SDValue lowerVECTOR_SHUFFLE_VSHUF4I(const SDLoc &DL, ArrayRef<int> Mask,
                                           MVT VT, SDValue V1, SDValue V2,
                                           SelectionDAG &DAG) {
 
  // When the size is less than 4, lower cost instructions may be used.
  if (Mask.size() < 4)
    return SDValue();
 
  int SubMask[4] = {-1, -1, -1, -1};
  for (unsigned i = 0; i < 4; ++i) {
    for (unsigned j = i; j < Mask.size(); j += 4) {
      int Idx = Mask[j];
 
      // Convert from vector index to 4-element subvector index
      // If an index refers to an element outside of the subvector then give up
      if (Idx != -1) {
        Idx -= 4 * (j / 4);
        if (Idx < 0 || Idx >= 4)
          return SDValue();
      }
 
      // If the mask has an undef, replace it with the current index.
      // Note that it might still be undef if the current index is also undef
      if (SubMask[i] == -1)
        SubMask[i] = Idx;
      // Check that non-undef values are the same as in the mask. If they
      // aren't then give up
      else if (Idx != -1 && Idx != SubMask[i])
        return SDValue();
    }
  }
 
  // Calculate the immediate. Replace any remaining undefs with zero
  APInt Imm(64, 0);
  for (int i = 3; i >= 0; --i) {
    int Idx = SubMask[i];
 
    if (Idx == -1)
      Idx = 0;
 
    Imm <<= 2;
    Imm |= Idx & 0x3;
  }
 
  return DAG.getNode(LoongArchISD::VSHUF4I, DL, VT, V1,
                     DAG.getConstant(Imm, DL, MVT::i64));
}
 
/// Lower VECTOR_SHUFFLE into VPACKEV (if possible).
///
/// VPACKEV interleaves the even elements from each vector.
///
/// It is possible to lower into VPACKEV when the mask consists of two of the
/// following forms interleaved:
///   <0, 2, 4, ...>
///   <n, n+2, n+4, ...>
/// where n is the number of elements in the vector.
/// For example:
///   <0, 0, 2, 2, 4, 4, ...>
///   <0, n, 2, n+2, 4, n+4, ...>
///
/// When undef's appear in the mask they are treated as if they were whatever
/// value is necessary in order to fit the above forms.
static SDValue lowerVECTOR_SHUFFLE_VPACKEV(const SDLoc &DL, ArrayRef<int> Mask,
                                           MVT VT, SDValue V1, SDValue V2,
                                           SelectionDAG &DAG) {
 
  const auto &Begin = Mask.begin();
  const auto &End = Mask.end();
  SDValue OriV1 = V1, OriV2 = V2;
 
  if (fitsRegularPattern<int>(Begin, 2, End, 0, 2))
    V1 = OriV1;
  else if (fitsRegularPattern<int>(Begin, 2, End, Mask.size(), 2))
    V1 = OriV2;
  else
    return SDValue();
 
  if (fitsRegularPattern<int>(Begin + 1, 2, End, 0, 2))
    V2 = OriV1;
  else if (fitsRegularPattern<int>(Begin + 1, 2, End, Mask.size(), 2))
    V2 = OriV2;
  else
    return SDValue();
 
  return DAG.getNode(LoongArchISD::VPACKEV, DL, VT, V2, V1);
}
 
/// Lower VECTOR_SHUFFLE into VPACKOD (if possible).
///
/// VPACKOD interleaves the odd elements from each vector.
///
/// It is possible to lower into VPACKOD when the mask consists of two of the
/// following forms interleaved:
///   <1, 3, 5, ...>
///   <n+1, n+3, n+5, ...>
/// where n is the number of elements in the vector.
/// For example:
///   <1, 1, 3, 3, 5, 5, ...>
///   <1, n+1, 3, n+3, 5, n+5, ...>
///
/// When undef's appear in the mask they are treated as if they were whatever
/// value is necessary in order to fit the above forms.
static SDValue lowerVECTOR_SHUFFLE_VPACKOD(const SDLoc &DL, ArrayRef<int> Mask,
                                           MVT VT, SDValue V1, SDValue V2,
                                           SelectionDAG &DAG) {
 
  const auto &Begin = Mask.begin();
  const auto &End = Mask.end();
  SDValue OriV1 = V1, OriV2 = V2;
 
  if (fitsRegularPattern<int>(Begin, 2, End, 1, 2))
    V1 = OriV1;
  else if (fitsRegularPattern<int>(Begin, 2, End, Mask.size() + 1, 2))
    V1 = OriV2;
  else
    return SDValue();
 
  if (fitsRegularPattern<int>(Begin + 1, 2, End, 1, 2))
    V2 = OriV1;
  else if (fitsRegularPattern<int>(Begin + 1, 2, End, Mask.size() + 1, 2))
    V2 = OriV2;
  else
    return SDValue();
 
  return DAG.getNode(LoongArchISD::VPACKOD, DL, VT, V2, V1);
}
 
/// Lower VECTOR_SHUFFLE into VILVH (if possible).
///
/// VILVH interleaves consecutive elements from the left (highest-indexed) half
/// of each vector.
///
/// It is possible to lower into VILVH when the mask consists of two of the
/// following forms interleaved:
///   <x, x+1, x+2, ...>
///   <n+x, n+x+1, n+x+2, ...>
/// where n is the number of elements in the vector and x is half n.
/// For example:
///   <x, x, x+1, x+1, x+2, x+2, ...>
///   <x, n+x, x+1, n+x+1, x+2, n+x+2, ...>
///
/// When undef's appear in the mask they are treated as if they were whatever
/// value is necessary in order to fit the above forms.
static SDValue lowerVECTOR_SHUFFLE_VILVH(const SDLoc &DL, ArrayRef<int> Mask,
                                         MVT VT, SDValue V1, SDValue V2,
                                         SelectionDAG &DAG) {
 
  const auto &Begin = Mask.begin();
  const auto &End = Mask.end();
  unsigned HalfSize = Mask.size() / 2;
  SDValue OriV1 = V1, OriV2 = V2;
 
  if (fitsRegularPattern<int>(Begin, 2, End, HalfSize, 1))
    V1 = OriV1;
  else if (fitsRegularPattern<int>(Begin, 2, End, Mask.size() + HalfSize, 1))
    V1 = OriV2;
  else
    return SDValue();
 
  if (fitsRegularPattern<int>(Begin + 1, 2, End, HalfSize, 1))
    V2 = OriV1;
  else if (fitsRegularPattern<int>(Begin + 1, 2, End, Mask.size() + HalfSize,
                                   1))
    V2 = OriV2;
  else
    return SDValue();
 
  return DAG.getNode(LoongArchISD::VILVH, DL, VT, V2, V1);
}
 
/// Lower VECTOR_SHUFFLE into VILVL (if possible).
///
/// VILVL interleaves consecutive elements from the right (lowest-indexed) half
/// of each vector.
///
/// It is possible to lower into VILVL when the mask consists of two of the
/// following forms interleaved:
///   <0, 1, 2, ...>
///   <n, n+1, n+2, ...>
/// where n is the number of elements in the vector.
/// For example:
///   <0, 0, 1, 1, 2, 2, ...>
///   <0, n, 1, n+1, 2, n+2, ...>
///
/// When undef's appear in the mask they are treated as if they were whatever
/// value is necessary in order to fit the above forms.
static SDValue lowerVECTOR_SHUFFLE_VILVL(const SDLoc &DL, ArrayRef<int> Mask,
                                         MVT VT, SDValue V1, SDValue V2,
                                         SelectionDAG &DAG) {
 
  const auto &Begin = Mask.begin();
  const auto &End = Mask.end();
  SDValue OriV1 = V1, OriV2 = V2;
 
  if (fitsRegularPattern<int>(Begin, 2, End, 0, 1))
    V1 = OriV1;
  else if (fitsRegularPattern<int>(Begin, 2, End, Mask.size(), 1))
    V1 = OriV2;
  else
    return SDValue();
 
  if (fitsRegularPattern<int>(Begin + 1, 2, End, 0, 1))
    V2 = OriV1;
  else if (fitsRegularPattern<int>(Begin + 1, 2, End, Mask.size(), 1))
    V2 = OriV2;
  else
    return SDValue();
 
  return DAG.getNode(LoongArchISD::VILVL, DL, VT, V2, V1);
}
 
/// Lower VECTOR_SHUFFLE into VPICKEV (if possible).
///
/// VPICKEV copies the even elements of each vector into the result vector.
///
/// It is possible to lower into VPICKEV when the mask consists of two of the
/// following forms concatenated:
///   <0, 2, 4, ...>
///   <n, n+2, n+4, ...>
/// where n is the number of elements in the vector.
/// For example:
///   <0, 2, 4, ..., 0, 2, 4, ...>
///   <0, 2, 4, ..., n, n+2, n+4, ...>
///
/// When undef's appear in the mask they are treated as if they were whatever
/// value is necessary in order to fit the above forms.
static SDValue lowerVECTOR_SHUFFLE_VPICKEV(const SDLoc &DL, ArrayRef<int> Mask,
                                           MVT VT, SDValue V1, SDValue V2,
                                           SelectionDAG &DAG) {
 
  const auto &Begin = Mask.begin();
  const auto &Mid = Mask.begin() + Mask.size() / 2;
  const auto &End = Mask.end();
  SDValue OriV1 = V1, OriV2 = V2;
 
  if (fitsRegularPattern<int>(Begin, 1, Mid, 0, 2))
    V1 = OriV1;
  else if (fitsRegularPattern<int>(Begin, 1, Mid, Mask.size(), 2))
    V1 = OriV2;
  else
    return SDValue();
 
  if (fitsRegularPattern<int>(Mid, 1, End, 0, 2))
    V2 = OriV1;
  else if (fitsRegularPattern<int>(Mid, 1, End, Mask.size(), 2))
    V2 = OriV2;
 
  else
    return SDValue();
 
  return DAG.getNode(LoongArchISD::VPICKEV, DL, VT, V2, V1);
}
 
/// Lower VECTOR_SHUFFLE into VPICKOD (if possible).
///
/// VPICKOD copies the odd elements of each vector into the result vector.
///
/// It is possible to lower into VPICKOD when the mask consists of two of the
/// following forms concatenated:
///   <1, 3, 5, ...>
///   <n+1, n+3, n+5, ...>
/// where n is the number of elements in the vector.
/// For example:
///   <1, 3, 5, ..., 1, 3, 5, ...>
///   <1, 3, 5, ..., n+1, n+3, n+5, ...>
///
/// When undef's appear in the mask they are treated as if they were whatever
/// value is necessary in order to fit the above forms.
static SDValue lowerVECTOR_SHUFFLE_VPICKOD(const SDLoc &DL, ArrayRef<int> Mask,
                                           MVT VT, SDValue V1, SDValue V2,
                                           SelectionDAG &DAG) {
 
  const auto &Begin = Mask.begin();
  const auto &Mid = Mask.begin() + Mask.size() / 2;
  const auto &End = Mask.end();
  SDValue OriV1 = V1, OriV2 = V2;
 
  if (fitsRegularPattern<int>(Begin, 1, Mid, 1, 2))
    V1 = OriV1;
  else if (fitsRegularPattern<int>(Begin, 1, Mid, Mask.size() + 1, 2))
    V1 = OriV2;
  else
    return SDValue();
 
  if (fitsRegularPattern<int>(Mid, 1, End, 1, 2))
    V2 = OriV1;
  else if (fitsRegularPattern<int>(Mid, 1, End, Mask.size() + 1, 2))
    V2 = OriV2;
  else
    return SDValue();
 
  return DAG.getNode(LoongArchISD::VPICKOD, DL, VT, V2, V1);
}
 
/// Lower VECTOR_SHUFFLE into VSHUF.
///
/// This mostly consists of converting the shuffle mask into a BUILD_VECTOR and
/// adding it as an operand to the resulting VSHUF.
static SDValue lowerVECTOR_SHUFFLE_VSHUF(const SDLoc &DL, ArrayRef<int> Mask,
                                         MVT VT, SDValue V1, SDValue V2,
                                         SelectionDAG &DAG) {
 
  SmallVector<SDValue, 16> Ops;
  for (auto M : Mask)
    Ops.push_back(DAG.getConstant(M, DL, MVT::i64));
 
  EVT MaskVecTy = VT.changeVectorElementTypeToInteger();
  SDValue MaskVec = DAG.getBuildVector(MaskVecTy, DL, Ops);
 
  // VECTOR_SHUFFLE concatenates the vectors in an vectorwise fashion.
  // <0b00, 0b01> + <0b10, 0b11> -> <0b00, 0b01, 0b10, 0b11>
  // VSHF concatenates the vectors in a bitwise fashion:
  // <0b00, 0b01> + <0b10, 0b11> ->
  // 0b0100       + 0b1110       -> 0b01001110
  //                                <0b10, 0b11, 0b00, 0b01>
  // We must therefore swap the operands to get the correct result.
  return DAG.getNode(LoongArchISD::VSHUF, DL, VT, MaskVec, V2, V1);
}
 
/// Dispatching routine to lower various 128-bit LoongArch vector shuffles.
///
/// This routine breaks down the specific type of 128-bit shuffle and
/// dispatches to the lowering routines accordingly.
static SDValue lower128BitShuffle(const SDLoc &DL, ArrayRef<int> Mask, MVT VT,
                                  SDValue V1, SDValue V2, SelectionDAG &DAG) {
  assert((VT.SimpleTy == MVT::v16i8 || VT.SimpleTy == MVT::v8i16 ||
          VT.SimpleTy == MVT::v4i32 || VT.SimpleTy == MVT::v2i64 ||
          VT.SimpleTy == MVT::v4f32 || VT.SimpleTy == MVT::v2f64) &&
         "Vector type is unsupported for lsx!");
  assert(V1.getSimpleValueType() == V2.getSimpleValueType() &&
         "Two operands have different types!");
  assert(VT.getVectorNumElements() == Mask.size() &&
         "Unexpected mask size for shuffle!");
  assert(Mask.size() % 2 == 0 && "Expected even mask size.");
 
  SDValue Result;
  // TODO: Add more comparison patterns.
  if (V2.isUndef()) {
    if ((Result = lowerVECTOR_SHUFFLE_VREPLVEI(DL, Mask, VT, V1, V2, DAG)))
      return Result;
    if ((Result = lowerVECTOR_SHUFFLE_VSHUF4I(DL, Mask, VT, V1, V2, DAG)))
      return Result;
 
    // TODO: This comment may be enabled in the future to better match the
    // pattern for instruction selection.
    /* V2 = V1; */
  }
 
  // It is recommended not to change the pattern comparison order for better
  // performance.
  if ((Result = lowerVECTOR_SHUFFLE_VPACKEV(DL, Mask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_VPACKOD(DL, Mask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_VILVH(DL, Mask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_VILVL(DL, Mask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_VPICKEV(DL, Mask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_VPICKOD(DL, Mask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_VSHUF(DL, Mask, VT, V1, V2, DAG)))
    return Result;
 
  return SDValue();
}
 
/// Lower VECTOR_SHUFFLE into XVREPLVEI (if possible).
///
/// It is a XVREPLVEI when the mask is:
///   <x, x, x, ..., x+n, x+n, x+n, ...>
/// where the number of x is equal to n and n is half the length of vector.
///
/// When undef's appear in the mask they are treated as if they were whatever
/// value is necessary in order to fit the above form.
static SDValue lowerVECTOR_SHUFFLE_XVREPLVEI(const SDLoc &DL,
                                             ArrayRef<int> Mask, MVT VT,
                                             SDValue V1, SDValue V2,
                                             SelectionDAG &DAG) {
  int SplatIndex = -1;
  for (const auto &M : Mask) {
    if (M != -1) {
      SplatIndex = M;
      break;
    }
  }
 
  if (SplatIndex == -1)
    return DAG.getUNDEF(VT);
 
  const auto &Begin = Mask.begin();
  const auto &End = Mask.end();
  unsigned HalfSize = Mask.size() / 2;
 
  assert(SplatIndex < (int)Mask.size() && "Out of bounds mask index");
  if (fitsRegularPattern<int>(Begin, 1, End - HalfSize, SplatIndex, 0) &&
      fitsRegularPattern<int>(Begin + HalfSize, 1, End, SplatIndex + HalfSize,
                              0)) {
    APInt Imm(64, SplatIndex);
    return DAG.getNode(LoongArchISD::VREPLVEI, DL, VT, V1,
                       DAG.getConstant(Imm, DL, MVT::i64));
  }
 
  return SDValue();
}
 
/// Lower VECTOR_SHUFFLE into XVSHUF4I (if possible).
static SDValue lowerVECTOR_SHUFFLE_XVSHUF4I(const SDLoc &DL, ArrayRef<int> Mask,
                                            MVT VT, SDValue V1, SDValue V2,
                                            SelectionDAG &DAG) {
  // When the size is less than or equal to 4, lower cost instructions may be
  // used.
  if (Mask.size() <= 4)
    return SDValue();
  return lowerVECTOR_SHUFFLE_VSHUF4I(DL, Mask, VT, V1, V2, DAG);
}
 
/// Lower VECTOR_SHUFFLE into XVPACKEV (if possible).
static SDValue lowerVECTOR_SHUFFLE_XVPACKEV(const SDLoc &DL, ArrayRef<int> Mask,
                                            MVT VT, SDValue V1, SDValue V2,
                                            SelectionDAG &DAG) {
  return lowerVECTOR_SHUFFLE_VPACKEV(DL, Mask, VT, V1, V2, DAG);
}
 
/// Lower VECTOR_SHUFFLE into XVPACKOD (if possible).
static SDValue lowerVECTOR_SHUFFLE_XVPACKOD(const SDLoc &DL, ArrayRef<int> Mask,
                                            MVT VT, SDValue V1, SDValue V2,
                                            SelectionDAG &DAG) {
  return lowerVECTOR_SHUFFLE_VPACKOD(DL, Mask, VT, V1, V2, DAG);
}
 
/// Lower VECTOR_SHUFFLE into XVILVH (if possible).
static SDValue lowerVECTOR_SHUFFLE_XVILVH(const SDLoc &DL, ArrayRef<int> Mask,
                                          MVT VT, SDValue V1, SDValue V2,
                                          SelectionDAG &DAG) {
 
  const auto &Begin = Mask.begin();
  const auto &End = Mask.end();
  unsigned HalfSize = Mask.size() / 2;
  unsigned LeftSize = HalfSize / 2;
  SDValue OriV1 = V1, OriV2 = V2;
 
  if (fitsRegularPattern<int>(Begin, 2, End - HalfSize, HalfSize - LeftSize,
                              1) &&
      fitsRegularPattern<int>(Begin + HalfSize, 2, End, HalfSize + LeftSize, 1))
    V1 = OriV1;
  else if (fitsRegularPattern<int>(Begin, 2, End - HalfSize,
                                   Mask.size() + HalfSize - LeftSize, 1) &&
           fitsRegularPattern<int>(Begin + HalfSize, 2, End,
                                   Mask.size() + HalfSize + LeftSize, 1))
    V1 = OriV2;
  else
    return SDValue();
 
  if (fitsRegularPattern<int>(Begin + 1, 2, End - HalfSize, HalfSize - LeftSize,
                              1) &&
      fitsRegularPattern<int>(Begin + 1 + HalfSize, 2, End, HalfSize + LeftSize,
                              1))
    V2 = OriV1;
  else if (fitsRegularPattern<int>(Begin + 1, 2, End - HalfSize,
                                   Mask.size() + HalfSize - LeftSize, 1) &&
           fitsRegularPattern<int>(Begin + 1 + HalfSize, 2, End,
                                   Mask.size() + HalfSize + LeftSize, 1))
    V2 = OriV2;
  else
    return SDValue();
 
  return DAG.getNode(LoongArchISD::VILVH, DL, VT, V2, V1);
}
 
/// Lower VECTOR_SHUFFLE into XVILVL (if possible).
static SDValue lowerVECTOR_SHUFFLE_XVILVL(const SDLoc &DL, ArrayRef<int> Mask,
                                          MVT VT, SDValue V1, SDValue V2,
                                          SelectionDAG &DAG) {
 
  const auto &Begin = Mask.begin();
  const auto &End = Mask.end();
  unsigned HalfSize = Mask.size() / 2;
  SDValue OriV1 = V1, OriV2 = V2;
 
  if (fitsRegularPattern<int>(Begin, 2, End - HalfSize, 0, 1) &&
      fitsRegularPattern<int>(Begin + HalfSize, 2, End, HalfSize, 1))
    V1 = OriV1;
  else if (fitsRegularPattern<int>(Begin, 2, End - HalfSize, Mask.size(), 1) &&
           fitsRegularPattern<int>(Begin + HalfSize, 2, End,
                                   Mask.size() + HalfSize, 1))
    V1 = OriV2;
  else
    return SDValue();
 
  if (fitsRegularPattern<int>(Begin + 1, 2, End - HalfSize, 0, 1) &&
      fitsRegularPattern<int>(Begin + 1 + HalfSize, 2, End, HalfSize, 1))
    V2 = OriV1;
  else if (fitsRegularPattern<int>(Begin + 1, 2, End - HalfSize, Mask.size(),
                                   1) &&
           fitsRegularPattern<int>(Begin + 1 + HalfSize, 2, End,
                                   Mask.size() + HalfSize, 1))
    V2 = OriV2;
  else
    return SDValue();
 
  return DAG.getNode(LoongArchISD::VILVL, DL, VT, V2, V1);
}
 
/// Lower VECTOR_SHUFFLE into XVPICKEV (if possible).
static SDValue lowerVECTOR_SHUFFLE_XVPICKEV(const SDLoc &DL, ArrayRef<int> Mask,
                                            MVT VT, SDValue V1, SDValue V2,
                                            SelectionDAG &DAG) {
 
  const auto &Begin = Mask.begin();
  const auto &LeftMid = Mask.begin() + Mask.size() / 4;
  const auto &Mid = Mask.begin() + Mask.size() / 2;
  const auto &RightMid = Mask.end() - Mask.size() / 4;
  const auto &End = Mask.end();
  unsigned HalfSize = Mask.size() / 2;
  SDValue OriV1 = V1, OriV2 = V2;
 
  if (fitsRegularPattern<int>(Begin, 1, LeftMid, 0, 2) &&
      fitsRegularPattern<int>(Mid, 1, RightMid, HalfSize, 2))
    V1 = OriV1;
  else if (fitsRegularPattern<int>(Begin, 1, LeftMid, Mask.size(), 2) &&
           fitsRegularPattern<int>(Mid, 1, RightMid, Mask.size() + HalfSize, 2))
    V1 = OriV2;
  else
    return SDValue();
 
  if (fitsRegularPattern<int>(LeftMid, 1, Mid, 0, 2) &&
      fitsRegularPattern<int>(RightMid, 1, End, HalfSize, 2))
    V2 = OriV1;
  else if (fitsRegularPattern<int>(LeftMid, 1, Mid, Mask.size(), 2) &&
           fitsRegularPattern<int>(RightMid, 1, End, Mask.size() + HalfSize, 2))
    V2 = OriV2;
 
  else
    return SDValue();
 
  return DAG.getNode(LoongArchISD::VPICKEV, DL, VT, V2, V1);
}
 
/// Lower VECTOR_SHUFFLE into XVPICKOD (if possible).
static SDValue lowerVECTOR_SHUFFLE_XVPICKOD(const SDLoc &DL, ArrayRef<int> Mask,
                                            MVT VT, SDValue V1, SDValue V2,
                                            SelectionDAG &DAG) {
 
  const auto &Begin = Mask.begin();
  const auto &LeftMid = Mask.begin() + Mask.size() / 4;
  const auto &Mid = Mask.begin() + Mask.size() / 2;
  const auto &RightMid = Mask.end() - Mask.size() / 4;
  const auto &End = Mask.end();
  unsigned HalfSize = Mask.size() / 2;
  SDValue OriV1 = V1, OriV2 = V2;
 
  if (fitsRegularPattern<int>(Begin, 1, LeftMid, 1, 2) &&
      fitsRegularPattern<int>(Mid, 1, RightMid, HalfSize + 1, 2))
    V1 = OriV1;
  else if (fitsRegularPattern<int>(Begin, 1, LeftMid, Mask.size() + 1, 2) &&
           fitsRegularPattern<int>(Mid, 1, RightMid, Mask.size() + HalfSize + 1,
                                   2))
    V1 = OriV2;
  else
    return SDValue();
 
  if (fitsRegularPattern<int>(LeftMid, 1, Mid, 1, 2) &&
      fitsRegularPattern<int>(RightMid, 1, End, HalfSize + 1, 2))
    V2 = OriV1;
  else if (fitsRegularPattern<int>(LeftMid, 1, Mid, Mask.size() + 1, 2) &&
           fitsRegularPattern<int>(RightMid, 1, End, Mask.size() + HalfSize + 1,
                                   2))
    V2 = OriV2;
  else
    return SDValue();
 
  return DAG.getNode(LoongArchISD::VPICKOD, DL, VT, V2, V1);
}
 
/// Lower VECTOR_SHUFFLE into XVSHUF (if possible).
static SDValue lowerVECTOR_SHUFFLE_XVSHUF(const SDLoc &DL, ArrayRef<int> Mask,
                                          MVT VT, SDValue V1, SDValue V2,
                                          SelectionDAG &DAG) {
 
  int MaskSize = Mask.size();
  int HalfSize = Mask.size() / 2;
  const auto &Begin = Mask.begin();
  const auto &Mid = Mask.begin() + HalfSize;
  const auto &End = Mask.end();
 
  // VECTOR_SHUFFLE concatenates the vectors:
  //  <0, 1, 2, 3, 4, 5, 6, 7> + <8, 9, 10, 11, 12, 13, 14, 15>
  //  shuffling ->
  //  <0, 1, 2, 3, 8, 9, 10, 11> <4, 5, 6, 7, 12, 13, 14, 15>
  //
  // XVSHUF concatenates the vectors:
  //  <a0, a1, a2, a3, b0, b1, b2, b3> + <a4, a5, a6, a7, b4, b5, b6, b7>
  //  shuffling ->
  //  <a0, a1, a2, a3, a4, a5, a6, a7> + <b0, b1, b2, b3, b4, b5, b6, b7>
  SmallVector<SDValue, 8> MaskAlloc;
  for (auto it = Begin; it < Mid; it++) {
    if (*it < 0) // UNDEF
      MaskAlloc.push_back(DAG.getTargetConstant(0, DL, MVT::i64));
    else if ((*it >= 0 && *it < HalfSize) ||
             (*it >= MaskSize && *it <= MaskSize + HalfSize)) {
      int M = *it < HalfSize ? *it : *it - HalfSize;
      MaskAlloc.push_back(DAG.getTargetConstant(M, DL, MVT::i64));
    } else
      return SDValue();
  }
  assert((int)MaskAlloc.size() == HalfSize && "xvshuf convert failed!");
 
  for (auto it = Mid; it < End; it++) {
    if (*it < 0) // UNDEF
      MaskAlloc.push_back(DAG.getTargetConstant(0, DL, MVT::i64));
    else if ((*it >= HalfSize && *it < MaskSize) ||
             (*it >= MaskSize + HalfSize && *it < MaskSize * 2)) {
      int M = *it < MaskSize ? *it - HalfSize : *it - MaskSize;
      MaskAlloc.push_back(DAG.getTargetConstant(M, DL, MVT::i64));
    } else
      return SDValue();
  }
  assert((int)MaskAlloc.size() == MaskSize && "xvshuf convert failed!");
 
  EVT MaskVecTy = VT.changeVectorElementTypeToInteger();
  SDValue MaskVec = DAG.getBuildVector(MaskVecTy, DL, MaskAlloc);
  return DAG.getNode(LoongArchISD::VSHUF, DL, VT, MaskVec, V2, V1);
}
 
/// Shuffle vectors by lane to generate more optimized instructions.
/// 256-bit shuffles are always considered as 2-lane 128-bit shuffles.
///
/// Therefore, except for the following four cases, other cases are regarded
/// as cross-lane shuffles, where optimization is relatively limited.
///
/// - Shuffle high, low lanes of two inputs vector
///   <0, 1, 2, 3> + <4, 5, 6, 7> --- <0, 5, 3, 6>
/// - Shuffle low, high lanes of two inputs vector
///   <0, 1, 2, 3> + <4, 5, 6, 7> --- <3, 6, 0, 5>
/// - Shuffle low, low lanes of two inputs vector
///   <0, 1, 2, 3> + <4, 5, 6, 7> --- <3, 6, 3, 6>
/// - Shuffle high, high lanes of two inputs vector
///   <0, 1, 2, 3> + <4, 5, 6, 7> --- <0, 5, 0, 5>
///
/// The first case is the closest to LoongArch instructions and the other
/// cases need to be converted to it for processing.
///
/// This function may modify V1, V2 and Mask
static void canonicalizeShuffleVectorByLane(const SDLoc &DL,
                                            MutableArrayRef<int> Mask, MVT VT,
                                            SDValue &V1, SDValue &V2,
                                            SelectionDAG &DAG) {
 
  enum HalfMaskType { HighLaneTy, LowLaneTy, None };
 
  int MaskSize = Mask.size();
  int HalfSize = Mask.size() / 2;
 
  HalfMaskType preMask = None, postMask = None;
 
  if (std::all_of(Mask.begin(), Mask.begin() + HalfSize, [&](int M) {
        return M < 0 || (M >= 0 && M < HalfSize) ||
               (M >= MaskSize && M < MaskSize + HalfSize);
      }))
    preMask = HighLaneTy;
  else if (std::all_of(Mask.begin(), Mask.begin() + HalfSize, [&](int M) {
             return M < 0 || (M >= HalfSize && M < MaskSize) ||
                    (M >= MaskSize + HalfSize && M < MaskSize * 2);
           }))
    preMask = LowLaneTy;
 
  if (std::all_of(Mask.begin() + HalfSize, Mask.end(), [&](int M) {
        return M < 0 || (M >= 0 && M < HalfSize) ||
               (M >= MaskSize && M < MaskSize + HalfSize);
      }))
    postMask = HighLaneTy;
  else if (std::all_of(Mask.begin() + HalfSize, Mask.end(), [&](int M) {
             return M < 0 || (M >= HalfSize && M < MaskSize) ||
                    (M >= MaskSize + HalfSize && M < MaskSize * 2);
           }))
    postMask = LowLaneTy;
 
  // The pre-half of mask is high lane type, and the post-half of mask
  // is low lane type, which is closest to the LoongArch instructions.
  //
  // Note: In the LoongArch architecture, the high lane of mask corresponds
  // to the lower 128-bit of vector register, and the low lane of mask
  // corresponds the higher 128-bit of vector register.
  if (preMask == HighLaneTy && postMask == LowLaneTy) {
    return;
  }
  if (preMask == LowLaneTy && postMask == HighLaneTy) {
    V1 = DAG.getBitcast(MVT::v4i64, V1);
    V1 = DAG.getNode(LoongArchISD::XVPERMI, DL, MVT::v4i64, V1,
                     DAG.getConstant(0b01001110, DL, MVT::i64));
    V1 = DAG.getBitcast(VT, V1);
 
    if (!V2.isUndef()) {
      V2 = DAG.getBitcast(MVT::v4i64, V2);
      V2 = DAG.getNode(LoongArchISD::XVPERMI, DL, MVT::v4i64, V2,
                       DAG.getConstant(0b01001110, DL, MVT::i64));
      V2 = DAG.getBitcast(VT, V2);
    }
 
    for (auto it = Mask.begin(); it < Mask.begin() + HalfSize; it++) {
      *it = *it < 0 ? *it : *it - HalfSize;
    }
    for (auto it = Mask.begin() + HalfSize; it < Mask.end(); it++) {
      *it = *it < 0 ? *it : *it + HalfSize;
    }
  } else if (preMask == LowLaneTy && postMask == LowLaneTy) {
    V1 = DAG.getBitcast(MVT::v4i64, V1);
    V1 = DAG.getNode(LoongArchISD::XVPERMI, DL, MVT::v4i64, V1,
                     DAG.getConstant(0b11101110, DL, MVT::i64));
    V1 = DAG.getBitcast(VT, V1);
 
    if (!V2.isUndef()) {
      V2 = DAG.getBitcast(MVT::v4i64, V2);
      V2 = DAG.getNode(LoongArchISD::XVPERMI, DL, MVT::v4i64, V2,
                       DAG.getConstant(0b11101110, DL, MVT::i64));
      V2 = DAG.getBitcast(VT, V2);
    }
 
    for (auto it = Mask.begin(); it < Mask.begin() + HalfSize; it++) {
      *it = *it < 0 ? *it : *it - HalfSize;
    }
  } else if (preMask == HighLaneTy && postMask == HighLaneTy) {
    V1 = DAG.getBitcast(MVT::v4i64, V1);
    V1 = DAG.getNode(LoongArchISD::XVPERMI, DL, MVT::v4i64, V1,
                     DAG.getConstant(0b01000100, DL, MVT::i64));
    V1 = DAG.getBitcast(VT, V1);
 
    if (!V2.isUndef()) {
      V2 = DAG.getBitcast(MVT::v4i64, V2);
      V2 = DAG.getNode(LoongArchISD::XVPERMI, DL, MVT::v4i64, V2,
                       DAG.getConstant(0b01000100, DL, MVT::i64));
      V2 = DAG.getBitcast(VT, V2);
    }
 
    for (auto it = Mask.begin() + HalfSize; it < Mask.end(); it++) {
      *it = *it < 0 ? *it : *it + HalfSize;
    }
  } else { // cross-lane
    return;
  }
}
 
/// Dispatching routine to lower various 256-bit LoongArch vector shuffles.
///
/// This routine breaks down the specific type of 256-bit shuffle and
/// dispatches to the lowering routines accordingly.
static SDValue lower256BitShuffle(const SDLoc &DL, ArrayRef<int> Mask, MVT VT,
                                  SDValue V1, SDValue V2, SelectionDAG &DAG) {
  assert((VT.SimpleTy == MVT::v32i8 || VT.SimpleTy == MVT::v16i16 ||
          VT.SimpleTy == MVT::v8i32 || VT.SimpleTy == MVT::v4i64 ||
          VT.SimpleTy == MVT::v8f32 || VT.SimpleTy == MVT::v4f64) &&
         "Vector type is unsupported for lasx!");
  assert(V1.getSimpleValueType() == V2.getSimpleValueType() &&
         "Two operands have different types!");
  assert(VT.getVectorNumElements() == Mask.size() &&
         "Unexpected mask size for shuffle!");
  assert(Mask.size() % 2 == 0 && "Expected even mask size.");
  assert(Mask.size() >= 4 && "Mask size is less than 4.");
 
  // canonicalize non cross-lane shuffle vector
  SmallVector<int> NewMask(Mask);
  canonicalizeShuffleVectorByLane(DL, NewMask, VT, V1, V2, DAG);
 
  SDValue Result;
  // TODO: Add more comparison patterns.
  if (V2.isUndef()) {
    if ((Result = lowerVECTOR_SHUFFLE_XVREPLVEI(DL, NewMask, VT, V1, V2, DAG)))
      return Result;
    if ((Result = lowerVECTOR_SHUFFLE_XVSHUF4I(DL, NewMask, VT, V1, V2, DAG)))
      return Result;
 
    // TODO: This comment may be enabled in the future to better match the
    // pattern for instruction selection.
    /* V2 = V1; */
  }
 
  // It is recommended not to change the pattern comparison order for better
  // performance.
  if ((Result = lowerVECTOR_SHUFFLE_XVPACKEV(DL, NewMask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_XVPACKOD(DL, NewMask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_XVILVH(DL, NewMask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_XVILVL(DL, NewMask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_XVPICKEV(DL, NewMask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_XVPICKOD(DL, NewMask, VT, V1, V2, DAG)))
    return Result;
  if ((Result = lowerVECTOR_SHUFFLE_XVSHUF(DL, NewMask, VT, V1, V2, DAG)))
    return Result;
 
  return SDValue();
}
 
SDValue LoongArchTargetLowering::lowerVECTOR_SHUFFLE(SDValue Op,
                                                     SelectionDAG &DAG) const {
  ShuffleVectorSDNode *SVOp = cast<ShuffleVectorSDNode>(Op);
  ArrayRef<int> OrigMask = SVOp->getMask();
  SDValue V1 = Op.getOperand(0);
  SDValue V2 = Op.getOperand(1);
  MVT VT = Op.getSimpleValueType();
  int NumElements = VT.getVectorNumElements();
  SDLoc DL(Op);
 
  bool V1IsUndef = V1.isUndef();
  bool V2IsUndef = V2.isUndef();
  if (V1IsUndef && V2IsUndef)
    return DAG.getUNDEF(VT);
 
  // When we create a shuffle node we put the UNDEF node to second operand,
  // but in some cases the first operand may be transformed to UNDEF.
  // In this case we should just commute the node.
  if (V1IsUndef)
    return DAG.getCommutedVectorShuffle(*SVOp);
 
  // Check for non-undef masks pointing at an undef vector and make the masks
  // undef as well. This makes it easier to match the shuffle based solely on
  // the mask.
  if (V2IsUndef &&
      any_of(OrigMask, [NumElements](int M) { return M >= NumElements; })) {
    SmallVector<int, 8> NewMask(OrigMask);
    for (int &M : NewMask)
      if (M >= NumElements)
        M = -1;
    return DAG.getVectorShuffle(VT, DL, V1, V2, NewMask);
  }
 
  // Check for illegal shuffle mask element index values.
  int MaskUpperLimit = OrigMask.size() * (V2IsUndef ? 1 : 2);
  (void)MaskUpperLimit;
  assert(llvm::all_of(OrigMask,
                      [&](int M) { return -1 <= M && M < MaskUpperLimit; }) &&
         "Out of bounds shuffle index");
 
  // For each vector width, delegate to a specialized lowering routine.
  if (VT.is128BitVector())
    return lower128BitShuffle(DL, OrigMask, VT, V1, V2, DAG);
 
  if (VT.is256BitVector())
    return lower256BitShuffle(DL, OrigMask, VT, V1, V2, DAG);
 
  return SDValue();
}
 
static bool isConstantOrUndef(const SDValue Op) {
  if (Op->isUndef())
    return true;
  if (isa<ConstantSDNode>(Op))
    return true;
  if (isa<ConstantFPSDNode>(Op))
    return true;
  return false;
}
 
static bool isConstantOrUndefBUILD_VECTOR(const BuildVectorSDNode *Op) {
  for (unsigned i = 0; i < Op->getNumOperands(); ++i)
    if (isConstantOrUndef(Op->getOperand(i)))
      return true;
  return false;
}
 
SDValue LoongArchTargetLowering::lowerBUILD_VECTOR(SDValue Op,
                                                   SelectionDAG &DAG) const {
  BuildVectorSDNode *Node = cast<BuildVectorSDNode>(Op);
  EVT ResTy = Op->getValueType(0);
  SDLoc DL(Op);
  APInt SplatValue, SplatUndef;
  unsigned SplatBitSize;
  bool HasAnyUndefs;
  bool Is128Vec = ResTy.is128BitVector();
  bool Is256Vec = ResTy.is256BitVector();
 
  if ((!Subtarget.hasExtLSX() || !Is128Vec) &&
      (!Subtarget.hasExtLASX() || !Is256Vec))
    return SDValue();
 
  if (Node->isConstantSplat(SplatValue, SplatUndef, SplatBitSize, HasAnyUndefs,
                            /*MinSplatBits=*/8) &&
      SplatBitSize <= 64) {
    // We can only cope with 8, 16, 32, or 64-bit elements.
    if (SplatBitSize != 8 && SplatBitSize != 16 && SplatBitSize != 32 &&
        SplatBitSize != 64)
      return SDValue();
 
    EVT ViaVecTy;
 
    switch (SplatBitSize) {
    default:
      return SDValue();
    case 8:
      ViaVecTy = Is128Vec ? MVT::v16i8 : MVT::v32i8;
      break;
    case 16:
      ViaVecTy = Is128Vec ? MVT::v8i16 : MVT::v16i16;
      break;
    case 32:
      ViaVecTy = Is128Vec ? MVT::v4i32 : MVT::v8i32;
      break;
    case 64:
      ViaVecTy = Is128Vec ? MVT::v2i64 : MVT::v4i64;
      break;
    }
 
    // SelectionDAG::getConstant will promote SplatValue appropriately.
    SDValue Result = DAG.getConstant(SplatValue, DL, ViaVecTy);
 
    // Bitcast to the type we originally wanted.
    if (ViaVecTy != ResTy)
      Result = DAG.getNode(ISD::BITCAST, SDLoc(Node), ResTy, Result);
 
    return Result;
  }
 
  if (DAG.isSplatValue(Op, /*AllowUndefs=*/false))
    return Op;
 
  if (!isConstantOrUndefBUILD_VECTOR(Node)) {
    // Use INSERT_VECTOR_ELT operations rather than expand to stores.
    // The resulting code is the same length as the expansion, but it doesn't
    // use memory operations.
    EVT ResTy = Node->getValueType(0);
 
    assert(ResTy.isVector());
 
    unsigned NumElts = ResTy.getVectorNumElements();
    SDValue Vector = DAG.getUNDEF(ResTy);
    for (unsigned i = 0; i < NumElts; ++i) {
      Vector = DAG.getNode(ISD::INSERT_VECTOR_ELT, DL, ResTy, Vector,
                           Node->getOperand(i),
                           DAG.getConstant(i, DL, Subtarget.getGRLenVT()));
    }
    return Vector;
  }
 
  return SDValue();
}
 
SDValue
LoongArchTargetLowering::lowerEXTRACT_VECTOR_ELT(SDValue Op,
                                                 SelectionDAG &DAG) const {
  EVT VecTy = Op->getOperand(0)->getValueType(0);
  SDValue Idx = Op->getOperand(1);
  EVT EltTy = VecTy.getVectorElementType();
  unsigned NumElts = VecTy.getVectorNumElements();
 
  if (isa<ConstantSDNode>(Idx) &&
      (EltTy == MVT::i32 || EltTy == MVT::i64 || EltTy == MVT::f32 ||
       EltTy == MVT::f64 || Idx->getAsZExtVal() < NumElts / 2))
    return Op;
 
  return SDValue();
}
 
SDValue
LoongArchTargetLowering::lowerINSERT_VECTOR_ELT(SDValue Op,
                                                SelectionDAG &DAG) const {
  if (isa<ConstantSDNode>(Op->getOperand(2)))
    return Op;
  return SDValue();
}
 
SDValue LoongArchTargetLowering::lowerATOMIC_FENCE(SDValue Op,
                                                   SelectionDAG &DAG) const {
  SDLoc DL(Op);
  SyncScope::ID FenceSSID =
      static_cast<SyncScope::ID>(Op.getConstantOperandVal(2));
 
  // singlethread fences only synchronize with signal handlers on the same
  // thread and thus only need to preserve instruction order, not actually
  // enforce memory ordering.
  if (FenceSSID == SyncScope::SingleThread)
    // MEMBARRIER is a compiler barrier; it codegens to a no-op.
    return DAG.getNode(ISD::MEMBARRIER, DL, MVT::Other, Op.getOperand(0));
 
  return Op;
}
 
SDValue LoongArchTargetLowering::lowerWRITE_REGISTER(SDValue Op,
                                                     SelectionDAG &DAG) const {
 
  if (Subtarget.is64Bit() && Op.getOperand(2).getValueType() == MVT::i32) {
    DAG.getContext()->emitError(
        "On LA64, only 64-bit registers can be written.");
    return Op.getOperand(0);
  }
 
  if (!Subtarget.is64Bit() && Op.getOperand(2).getValueType() == MVT::i64) {
    DAG.getContext()->emitError(
        "On LA32, only 32-bit registers can be written.");
    return Op.getOperand(0);
  }
 
  return Op;
}
 
SDValue LoongArchTargetLowering::lowerFRAMEADDR(SDValue Op,
                                                SelectionDAG &DAG) const {
  if (!isa<ConstantSDNode>(Op.getOperand(0))) {
    DAG.getContext()->emitError("argument to '__builtin_frame_address' must "
                                "be a constant integer");
    return SDValue();
  }
 
  MachineFunction &MF = DAG.getMachineFunction();
  MF.getFrameInfo().setFrameAddressIsTaken(true);
  Register FrameReg = Subtarget.getRegisterInfo()->getFrameRegister(MF);
  EVT VT = Op.getValueType();
  SDLoc DL(Op);
  SDValue FrameAddr = DAG.getCopyFromReg(DAG.getEntryNode(), DL, FrameReg, VT);
  unsigned Depth = Op.getConstantOperandVal(0);
  int GRLenInBytes = Subtarget.getGRLen() / 8;
 
  while (Depth--) {
    int Offset = -(GRLenInBytes * 2);
    SDValue Ptr = DAG.getNode(ISD::ADD, DL, VT, FrameAddr,
                              DAG.getIntPtrConstant(Offset, DL));
    FrameAddr =
        DAG.getLoad(VT, DL, DAG.getEntryNode(), Ptr, MachinePointerInfo());
  }
  return FrameAddr;
}
 
SDValue LoongArchTargetLowering::lowerRETURNADDR(SDValue Op,
                                                 SelectionDAG &DAG) const {
  if (verifyReturnAddressArgumentIsConstant(Op, DAG))
    return SDValue();
 
  // Currently only support lowering return address for current frame.
  if (Op.getConstantOperandVal(0) != 0) {
    DAG.getContext()->emitError(
        "return address can only be determined for the current frame");
    return SDValue();
  }
 
  MachineFunction &MF = DAG.getMachineFunction();
  MF.getFrameInfo().setReturnAddressIsTaken(true);
  MVT GRLenVT = Subtarget.getGRLenVT();
 
  // Return the value of the return address register, marking it an implicit
  // live-in.
  Register Reg = MF.addLiveIn(Subtarget.getRegisterInfo()->getRARegister(),
                              getRegClassFor(GRLenVT));
  return DAG.getCopyFromReg(DAG.getEntryNode(), SDLoc(Op), Reg, GRLenVT);
}
 
SDValue LoongArchTargetLowering::lowerEH_DWARF_CFA(SDValue Op,
                                                   SelectionDAG &DAG) const {
  MachineFunction &MF = DAG.getMachineFunction();
  auto Size = Subtarget.getGRLen() / 8;
  auto FI = MF.getFrameInfo().CreateFixedObject(Size, 0, false);
  return DAG.getFrameIndex(FI, getPointerTy(DAG.getDataLayout()));
}
 
SDValue LoongArchTargetLowering::lowerVASTART(SDValue Op,
                                              SelectionDAG &DAG) const {
  MachineFunction &MF = DAG.getMachineFunction();
  auto *FuncInfo = MF.getInfo<LoongArchMachineFunctionInfo>();
 
  SDLoc DL(Op);
  SDValue FI = DAG.getFrameIndex(FuncInfo->getVarArgsFrameIndex(),
                                 getPointerTy(MF.getDataLayout()));
 
  // vastart just stores the address of the VarArgsFrameIndex slot into the
  // memory location argument.
  const Value *SV = cast<SrcValueSDNode>(Op.getOperand(2))->getValue();
  return DAG.getStore(Op.getOperand(0), DL, FI, Op.getOperand(1),
                      MachinePointerInfo(SV));
}
 
SDValue LoongArchTargetLowering::lowerUINT_TO_FP(SDValue Op,
                                                 SelectionDAG &DAG) const {
  assert(Subtarget.is64Bit() && Subtarget.hasBasicF() &&
         !Subtarget.hasBasicD() && "unexpected target features");
 
  SDLoc DL(Op);
  SDValue Op0 = Op.getOperand(0);
  if (Op0->getOpcode() == ISD::AND) {
    auto *C = dyn_cast<ConstantSDNode>(Op0.getOperand(1));
    if (C && C->getZExtValue() < UINT64_C(0xFFFFFFFF))
      return Op;
  }
 
  if (Op0->getOpcode() == LoongArchISD::BSTRPICK &&
      Op0.getConstantOperandVal(1) < UINT64_C(0X1F) &&
      Op0.getConstantOperandVal(2) == UINT64_C(0))
    return Op;
 
  if (Op0.getOpcode() == ISD::AssertZext &&
      dyn_cast<VTSDNode>(Op0.getOperand(1))->getVT().bitsLT(MVT::i32))
    return Op;
 
  EVT OpVT = Op0.getValueType();
  EVT RetVT = Op.getValueType();
  RTLIB::Libcall LC = RTLIB::getUINTTOFP(OpVT, RetVT);
  MakeLibCallOptions CallOptions;
  CallOptions.setTypeListBeforeSoften(OpVT, RetVT, true);
  SDValue Chain = SDValue();
  SDValue Result;
  std::tie(Result, Chain) =
      makeLibCall(DAG, LC, Op.getValueType(), Op0, CallOptions, DL, Chain);
  return Result;
}
 
SDValue LoongArchTargetLowering::lowerSINT_TO_FP(SDValue Op,
                                                 SelectionDAG &DAG) const {
  assert(Subtarget.is64Bit() && Subtarget.hasBasicF() &&
         !Subtarget.hasBasicD() && "unexpected target features");
 
  SDLoc DL(Op);
  SDValue Op0 = Op.getOperand(0);
 
  if ((Op0.getOpcode() == ISD::AssertSext ||
       Op0.getOpcode() == ISD::SIGN_EXTEND_INREG) &&
      dyn_cast<VTSDNode>(Op0.getOperand(1))->getVT().bitsLE(MVT::i32))
    return Op;
 
  EVT OpVT = Op0.getValueType();
  EVT RetVT = Op.getValueType();
  RTLIB::Libcall LC = RTLIB::getSINTTOFP(OpVT, RetVT);
  MakeLibCallOptions CallOptions;
  CallOptions.setTypeListBeforeSoften(OpVT, RetVT, true);
  SDValue Chain = SDValue();
  SDValue Result;
  std::tie(Result, Chain) =
      makeLibCall(DAG, LC, Op.getValueType(), Op0, CallOptions, DL, Chain);
  return Result;
}
 
SDValue LoongArchTargetLowering::lowerBITCAST(SDValue Op,
                                              SelectionDAG &DAG) const {
 
  SDLoc DL(Op);
  SDValue Op0 = Op.getOperand(0);
 
  if (Op.getValueType() == MVT::f32 && Op0.getValueType() == MVT::i32 &&
      Subtarget.is64Bit() && Subtarget.hasBasicF()) {
    SDValue NewOp0 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op0);
    return DAG.getNode(LoongArchISD::MOVGR2FR_W_LA64, DL, MVT::f32, NewOp0);
  }
  return Op;
}
 
SDValue LoongArchTargetLowering::lowerFP_TO_SINT(SDValue Op,
                                                 SelectionDAG &DAG) const {
 
  SDLoc DL(Op);
 
  if (Op.getValueSizeInBits() > 32 && Subtarget.hasBasicF() &&
      !Subtarget.hasBasicD()) {
    SDValue Dst =
        DAG.getNode(LoongArchISD::FTINT, DL, MVT::f32, Op.getOperand(0));
    return DAG.getNode(LoongArchISD::MOVFR2GR_S_LA64, DL, MVT::i64, Dst);
  }
 
  EVT FPTy = EVT::getFloatingPointVT(Op.getValueSizeInBits());
  SDValue Trunc = DAG.getNode(LoongArchISD::FTINT, DL, FPTy, Op.getOperand(0));
  return DAG.getNode(ISD::BITCAST, DL, Op.getValueType(), Trunc);
}
 
static SDValue getTargetNode(GlobalAddressSDNode *N, SDLoc DL, EVT Ty,
                             SelectionDAG &DAG, unsigned Flags) {
  return DAG.getTargetGlobalAddress(N->getGlobal(), DL, Ty, 0, Flags);
}
 
static SDValue getTargetNode(BlockAddressSDNode *N, SDLoc DL, EVT Ty,
                             SelectionDAG &DAG, unsigned Flags) {
  return DAG.getTargetBlockAddress(N->getBlockAddress(), Ty, N->getOffset(),
                                   Flags);
}
 
static SDValue getTargetNode(ConstantPoolSDNode *N, SDLoc DL, EVT Ty,
                             SelectionDAG &DAG, unsigned Flags) {
  return DAG.getTargetConstantPool(N->getConstVal(), Ty, N->getAlign(),
                                   N->getOffset(), Flags);
}
 
static SDValue getTargetNode(JumpTableSDNode *N, SDLoc DL, EVT Ty,
                             SelectionDAG &DAG, unsigned Flags) {
  return DAG.getTargetJumpTable(N->getIndex(), Ty, Flags);
}
 
template <class NodeTy>
SDValue LoongArchTargetLowering::getAddr(NodeTy *N, SelectionDAG &DAG,
                                         CodeModel::Model M,
                                         bool IsLocal) const {
  SDLoc DL(N);
  EVT Ty = getPointerTy(DAG.getDataLayout());
  SDValue Addr = getTargetNode(N, DL, Ty, DAG, 0);
  SDValue Load;
 
  switch (M) {
  default:
    report_fatal_error("Unsupported code model");
 
  case CodeModel::Large: {
    assert(Subtarget.is64Bit() && "Large code model requires LA64");
 
    // This is not actually used, but is necessary for successfully matching
    // the PseudoLA_*_LARGE nodes.
    SDValue Tmp = DAG.getConstant(0, DL, Ty);
    if (IsLocal) {
      // This generates the pattern (PseudoLA_PCREL_LARGE tmp sym), that
      // eventually becomes the desired 5-insn code sequence.
      Load = SDValue(DAG.getMachineNode(LoongArch::PseudoLA_PCREL_LARGE, DL, Ty,
                                        Tmp, Addr),
                     0);
    } else {
      // This generates the pattern (PseudoLA_GOT_LARGE tmp sym), that
      // eventually becomes the desired 5-insn code sequence.
      Load = SDValue(
          DAG.getMachineNode(LoongArch::PseudoLA_GOT_LARGE, DL, Ty, Tmp, Addr),
          0);
    }
    break;
  }
 
  case CodeModel::Small:
  case CodeModel::Medium:
    if (IsLocal) {
      // This generates the pattern (PseudoLA_PCREL sym), which expands to
      // (addi.w/d (pcalau12i %pc_hi20(sym)) %pc_lo12(sym)).
      Load = SDValue(
          DAG.getMachineNode(LoongArch::PseudoLA_PCREL, DL, Ty, Addr), 0);
    } else {
      // This generates the pattern (PseudoLA_GOT sym), which expands to (ld.w/d
      // (pcalau12i %got_pc_hi20(sym)) %got_pc_lo12(sym)).
      Load =
          SDValue(DAG.getMachineNode(LoongArch::PseudoLA_GOT, DL, Ty, Addr), 0);
    }
  }
 
  if (!IsLocal) {
    // Mark the load instruction as invariant to enable hoisting in MachineLICM.
    MachineFunction &MF = DAG.getMachineFunction();
    MachineMemOperand *MemOp = MF.getMachineMemOperand(
        MachinePointerInfo::getGOT(MF),
        MachineMemOperand::MOLoad | MachineMemOperand::MODereferenceable |
            MachineMemOperand::MOInvariant,
        LLT(Ty.getSimpleVT()), Align(Ty.getFixedSizeInBits() / 8));
    DAG.setNodeMemRefs(cast<MachineSDNode>(Load.getNode()), {MemOp});
  }
 
  return Load;
}
 
SDValue LoongArchTargetLowering::lowerBlockAddress(SDValue Op,
                                                   SelectionDAG &DAG) const {
  return getAddr(cast<BlockAddressSDNode>(Op), DAG,
                 DAG.getTarget().getCodeModel());
}
 
SDValue LoongArchTargetLowering::lowerJumpTable(SDValue Op,
                                                SelectionDAG &DAG) const {
  return getAddr(cast<JumpTableSDNode>(Op), DAG,
                 DAG.getTarget().getCodeModel());
}
 
SDValue LoongArchTargetLowering::lowerConstantPool(SDValue Op,
                                                   SelectionDAG &DAG) const {
  return getAddr(cast<ConstantPoolSDNode>(Op), DAG,
                 DAG.getTarget().getCodeModel());
}
 
SDValue LoongArchTargetLowering::lowerGlobalAddress(SDValue Op,
                                                    SelectionDAG &DAG) const {
  GlobalAddressSDNode *N = cast<GlobalAddressSDNode>(Op);
  assert(N->getOffset() == 0 && "unexpected offset in global node");
  auto CM = DAG.getTarget().getCodeModel();
  const GlobalValue *GV = N->getGlobal();
 
  if (GV->isDSOLocal() && isa<GlobalVariable>(GV)) {
    if (auto GCM = dyn_cast<GlobalVariable>(GV)->getCodeModel())
      CM = *GCM;
  }
 
  return getAddr(N, DAG, CM, GV->isDSOLocal());
}
 
SDValue LoongArchTargetLowering::getStaticTLSAddr(GlobalAddressSDNode *N,
                                                  SelectionDAG &DAG,
                                                  unsigned Opc, bool UseGOT,
                                                  bool Large) const {
  SDLoc DL(N);
  EVT Ty = getPointerTy(DAG.getDataLayout());
  MVT GRLenVT = Subtarget.getGRLenVT();
 
  // This is not actually used, but is necessary for successfully matching the
  // PseudoLA_*_LARGE nodes.
  SDValue Tmp = DAG.getConstant(0, DL, Ty);
  SDValue Addr = DAG.getTargetGlobalAddress(N->getGlobal(), DL, Ty, 0, 0);
  SDValue Offset = Large
                       ? SDValue(DAG.getMachineNode(Opc, DL, Ty, Tmp, Addr), 0)
                       : SDValue(DAG.getMachineNode(Opc, DL, Ty, Addr), 0);
  if (UseGOT) {
    // Mark the load instruction as invariant to enable hoisting in MachineLICM.
    MachineFunction &MF = DAG.getMachineFunction();
    MachineMemOperand *MemOp = MF.getMachineMemOperand(
        MachinePointerInfo::getGOT(MF),
        MachineMemOperand::MOLoad | MachineMemOperand::MODereferenceable |
            MachineMemOperand::MOInvariant,
        LLT(Ty.getSimpleVT()), Align(Ty.getFixedSizeInBits() / 8));
    DAG.setNodeMemRefs(cast<MachineSDNode>(Offset.getNode()), {MemOp});
  }
 
  // Add the thread pointer.
  return DAG.getNode(ISD::ADD, DL, Ty, Offset,
                     DAG.getRegister(LoongArch::R2, GRLenVT));
}
 
SDValue LoongArchTargetLowering::getDynamicTLSAddr(GlobalAddressSDNode *N,
                                                   SelectionDAG &DAG,
                                                   unsigned Opc,
                                                   bool Large) const {
  SDLoc DL(N);
  EVT Ty = getPointerTy(DAG.getDataLayout());
  IntegerType *CallTy = Type::getIntNTy(*DAG.getContext(), Ty.getSizeInBits());
 
  // This is not actually used, but is necessary for successfully matching the
  // PseudoLA_*_LARGE nodes.
  SDValue Tmp = DAG.getConstant(0, DL, Ty);
 
  // Use a PC-relative addressing mode to access the dynamic GOT address.
  SDValue Addr = DAG.getTargetGlobalAddress(N->getGlobal(), DL, Ty, 0, 0);
  SDValue Load = Large ? SDValue(DAG.getMachineNode(Opc, DL, Ty, Tmp, Addr), 0)
                       : SDValue(DAG.getMachineNode(Opc, DL, Ty, Addr), 0);
 
  // Prepare argument list to generate call.
  ArgListTy Args;
  ArgListEntry Entry;
  Entry.Node = Load;
  Entry.Ty = CallTy;
  Args.push_back(Entry);
 
  // Setup call to __tls_get_addr.
  TargetLowering::CallLoweringInfo CLI(DAG);
  CLI.setDebugLoc(DL)
      .setChain(DAG.getEntryNode())
      .setLibCallee(CallingConv::C, CallTy,
                    DAG.getExternalSymbol("__tls_get_addr", Ty),
                    std::move(Args));
 
  return LowerCallTo(CLI).first;
}
 
SDValue LoongArchTargetLowering::getTLSDescAddr(GlobalAddressSDNode *N,
                                                SelectionDAG &DAG, unsigned Opc,
                                                bool Large) const {
  SDLoc DL(N);
  EVT Ty = getPointerTy(DAG.getDataLayout());
  const GlobalValue *GV = N->getGlobal();
 
  // This is not actually used, but is necessary for successfully matching the
  // PseudoLA_*_LARGE nodes.
  SDValue Tmp = DAG.getConstant(0, DL, Ty);
 
  // Use a PC-relative addressing mode to access the global dynamic GOT address.
  // This generates the pattern (PseudoLA_TLS_DESC_PC{,LARGE} sym).
  SDValue Addr = DAG.getTargetGlobalAddress(GV, DL, Ty, 0, 0);
  return Large ? SDValue(DAG.getMachineNode(Opc, DL, Ty, Tmp, Addr), 0)
               : SDValue(DAG.getMachineNode(Opc, DL, Ty, Addr), 0);
}
 
SDValue
LoongArchTargetLowering::lowerGlobalTLSAddress(SDValue Op,
                                               SelectionDAG &DAG) const {
  if (DAG.getMachineFunction().getFunction().getCallingConv() ==
      CallingConv::GHC)
    report_fatal_error("In GHC calling convention TLS is not supported");
 
  bool Large = DAG.getTarget().getCodeModel() == CodeModel::Large;
  assert((!Large || Subtarget.is64Bit()) && "Large code model requires LA64");
 
  GlobalAddressSDNode *N = cast<GlobalAddressSDNode>(Op);
  assert(N->getOffset() == 0 && "unexpected offset in global node");
 
  if (DAG.getTarget().useEmulatedTLS())
    report_fatal_error("the emulated TLS is prohibited",
                       /*GenCrashDiag=*/false);
 
  bool IsDesc = DAG.getTarget().useTLSDESC();
 
  switch (getTargetMachine().getTLSModel(N->getGlobal())) {
  case TLSModel::GeneralDynamic:
    // In this model, application code calls the dynamic linker function
    // __tls_get_addr to locate TLS offsets into the dynamic thread vector at
    // runtime.
    if (!IsDesc)
      return getDynamicTLSAddr(N, DAG,
                               Large ? LoongArch::PseudoLA_TLS_GD_LARGE
                                     : LoongArch::PseudoLA_TLS_GD,
                               Large);
    break;
  case TLSModel::LocalDynamic:
    // Same as GeneralDynamic, except for assembly modifiers and relocation
    // records.
    if (!IsDesc)
      return getDynamicTLSAddr(N, DAG,
                               Large ? LoongArch::PseudoLA_TLS_LD_LARGE
                                     : LoongArch::PseudoLA_TLS_LD,
                               Large);
    break;
  case TLSModel::InitialExec:
    // This model uses the GOT to resolve TLS offsets.
    return getStaticTLSAddr(N, DAG,
                            Large ? LoongArch::PseudoLA_TLS_IE_LARGE
                                  : LoongArch::PseudoLA_TLS_IE,
                            /*UseGOT=*/true, Large);
  case TLSModel::LocalExec:
    // This model is used when static linking as the TLS offsets are resolved
    // during program linking.
    //
    // This node doesn't need an extra argument for the large code model.
    return getStaticTLSAddr(N, DAG, LoongArch::PseudoLA_TLS_LE,
                            /*UseGOT=*/false);
  }
 
  return getTLSDescAddr(N, DAG,
                        Large ? LoongArch::PseudoLA_TLS_DESC_PC_LARGE
                              : LoongArch::PseudoLA_TLS_DESC_PC,
                        Large);
}
 
template <unsigned N>
static SDValue checkIntrinsicImmArg(SDValue Op, unsigned ImmOp,
                                    SelectionDAG &DAG, bool IsSigned = false) {
  auto *CImm = cast<ConstantSDNode>(Op->getOperand(ImmOp));
  // Check the ImmArg.
  if ((IsSigned && !isInt<N>(CImm->getSExtValue())) ||
      (!IsSigned && !isUInt<N>(CImm->getZExtValue()))) {
    DAG.getContext()->emitError(Op->getOperationName(0) +
                                ": argument out of range.");
    return DAG.getNode(ISD::UNDEF, SDLoc(Op), Op.getValueType());
  }
  return SDValue();
}
 
SDValue
LoongArchTargetLowering::lowerINTRINSIC_WO_CHAIN(SDValue Op,
                                                 SelectionDAG &DAG) const {
  SDLoc DL(Op);
  switch (Op.getConstantOperandVal(0)) {
  default:
    return SDValue(); // Don't custom lower most intrinsics.
  case Intrinsic::thread_pointer: {
    EVT PtrVT = getPointerTy(DAG.getDataLayout());
    return DAG.getRegister(LoongArch::R2, PtrVT);
  }
  case Intrinsic::loongarch_lsx_vpickve2gr_d:
  case Intrinsic::loongarch_lsx_vpickve2gr_du:
  case Intrinsic::loongarch_lsx_vreplvei_d:
  case Intrinsic::loongarch_lasx_xvrepl128vei_d:
    return checkIntrinsicImmArg<1>(Op, 2, DAG);
  case Intrinsic::loongarch_lsx_vreplvei_w:
  case Intrinsic::loongarch_lasx_xvrepl128vei_w:
  case Intrinsic::loongarch_lasx_xvpickve2gr_d:
  case Intrinsic::loongarch_lasx_xvpickve2gr_du:
  case Intrinsic::loongarch_lasx_xvpickve_d:
  case Intrinsic::loongarch_lasx_xvpickve_d_f:
    return checkIntrinsicImmArg<2>(Op, 2, DAG);
  case Intrinsic::loongarch_lasx_xvinsve0_d:
    return checkIntrinsicImmArg<2>(Op, 3, DAG);
  case Intrinsic::loongarch_lsx_vsat_b:
  case Intrinsic::loongarch_lsx_vsat_bu:
  case Intrinsic::loongarch_lsx_vrotri_b:
  case Intrinsic::loongarch_lsx_vsllwil_h_b:
  case Intrinsic::loongarch_lsx_vsllwil_hu_bu:
  case Intrinsic::loongarch_lsx_vsrlri_b:
  case Intrinsic::loongarch_lsx_vsrari_b:
  case Intrinsic::loongarch_lsx_vreplvei_h:
  case Intrinsic::loongarch_lasx_xvsat_b:
  case Intrinsic::loongarch_lasx_xvsat_bu:
  case Intrinsic::loongarch_lasx_xvrotri_b:
  case Intrinsic::loongarch_lasx_xvsllwil_h_b:
  case Intrinsic::loongarch_lasx_xvsllwil_hu_bu:
  case Intrinsic::loongarch_lasx_xvsrlri_b:
  case Intrinsic::loongarch_lasx_xvsrari_b:
  case Intrinsic::loongarch_lasx_xvrepl128vei_h:
  case Intrinsic::loongarch_lasx_xvpickve_w:
  case Intrinsic::loongarch_lasx_xvpickve_w_f:
    return checkIntrinsicImmArg<3>(Op, 2, DAG);
  case Intrinsic::loongarch_lasx_xvinsve0_w:
    return checkIntrinsicImmArg<3>(Op, 3, DAG);
  case Intrinsic::loongarch_lsx_vsat_h:
  case Intrinsic::loongarch_lsx_vsat_hu:
  case Intrinsic::loongarch_lsx_vrotri_h:
  case Intrinsic::loongarch_lsx_vsllwil_w_h:
  case Intrinsic::loongarch_lsx_vsllwil_wu_hu:
  case Intrinsic::loongarch_lsx_vsrlri_h:
  case Intrinsic::loongarch_lsx_vsrari_h:
  case Intrinsic::loongarch_lsx_vreplvei_b:
  case Intrinsic::loongarch_lasx_xvsat_h:
  case Intrinsic::loongarch_lasx_xvsat_hu:
  case Intrinsic::loongarch_lasx_xvrotri_h:
  case Intrinsic::loongarch_lasx_xvsllwil_w_h:
  case Intrinsic::loongarch_lasx_xvsllwil_wu_hu:
  case Intrinsic::loongarch_lasx_xvsrlri_h:
  case Intrinsic::loongarch_lasx_xvsrari_h:
  case Intrinsic::loongarch_lasx_xvrepl128vei_b:
    return checkIntrinsicImmArg<4>(Op, 2, DAG);
  case Intrinsic::loongarch_lsx_vsrlni_b_h:
  case Intrinsic::loongarch_lsx_vsrani_b_h:
  case Intrinsic::loongarch_lsx_vsrlrni_b_h:
  case Intrinsic::loongarch_lsx_vsrarni_b_h:
  case Intrinsic::loongarch_lsx_vssrlni_b_h:
  case Intrinsic::loongarch_lsx_vssrani_b_h:
  case Intrinsic::loongarch_lsx_vssrlni_bu_h:
  case Intrinsic::loongarch_lsx_vssrani_bu_h:
  case Intrinsic::loongarch_lsx_vssrlrni_b_h:
  case Intrinsic::loongarch_lsx_vssrarni_b_h:
  case Intrinsic::loongarch_lsx_vssrlrni_bu_h:
  case Intrinsic::loongarch_lsx_vssrarni_bu_h:
  case Intrinsic::loongarch_lasx_xvsrlni_b_h:
  case Intrinsic::loongarch_lasx_xvsrani_b_h:
  case Intrinsic::loongarch_lasx_xvsrlrni_b_h:
  case Intrinsic::loongarch_lasx_xvsrarni_b_h:
  case Intrinsic::loongarch_lasx_xvssrlni_b_h:
  case Intrinsic::loongarch_lasx_xvssrani_b_h:
  case Intrinsic::loongarch_lasx_xvssrlni_bu_h:
  case Intrinsic::loongarch_lasx_xvssrani_bu_h:
  case Intrinsic::loongarch_lasx_xvssrlrni_b_h:
  case Intrinsic::loongarch_lasx_xvssrarni_b_h:
  case Intrinsic::loongarch_lasx_xvssrlrni_bu_h:
  case Intrinsic::loongarch_lasx_xvssrarni_bu_h:
    return checkIntrinsicImmArg<4>(Op, 3, DAG);
  case Intrinsic::loongarch_lsx_vsat_w:
  case Intrinsic::loongarch_lsx_vsat_wu:
  case Intrinsic::loongarch_lsx_vrotri_w:
  case Intrinsic::loongarch_lsx_vsllwil_d_w:
  case Intrinsic::loongarch_lsx_vsllwil_du_wu:
  case Intrinsic::loongarch_lsx_vsrlri_w:
  case Intrinsic::loongarch_lsx_vsrari_w:
  case Intrinsic::loongarch_lsx_vslei_bu:
  case Intrinsic::loongarch_lsx_vslei_hu:
  case Intrinsic::loongarch_lsx_vslei_wu:
  case Intrinsic::loongarch_lsx_vslei_du:
  case Intrinsic::loongarch_lsx_vslti_bu:
  case Intrinsic::loongarch_lsx_vslti_hu:
  case Intrinsic::loongarch_lsx_vslti_wu:
  case Intrinsic::loongarch_lsx_vslti_du:
  case Intrinsic::loongarch_lsx_vbsll_v:
  case Intrinsic::loongarch_lsx_vbsrl_v:
  case Intrinsic::loongarch_lasx_xvsat_w:
  case Intrinsic::loongarch_lasx_xvsat_wu:
  case Intrinsic::loongarch_lasx_xvrotri_w:
  case Intrinsic::loongarch_lasx_xvsllwil_d_w:
  case Intrinsic::loongarch_lasx_xvsllwil_du_wu:
  case Intrinsic::loongarch_lasx_xvsrlri_w:
  case Intrinsic::loongarch_lasx_xvsrari_w:
  case Intrinsic::loongarch_lasx_xvslei_bu:
  case Intrinsic::loongarch_lasx_xvslei_hu:
  case Intrinsic::loongarch_lasx_xvslei_wu:
  case Intrinsic::loongarch_lasx_xvslei_du:
  case Intrinsic::loongarch_lasx_xvslti_bu:
  case Intrinsic::loongarch_lasx_xvslti_hu:
  case Intrinsic::loongarch_lasx_xvslti_wu:
  case Intrinsic::loongarch_lasx_xvslti_du:
  case Intrinsic::loongarch_lasx_xvbsll_v:
  case Intrinsic::loongarch_lasx_xvbsrl_v:
    return checkIntrinsicImmArg<5>(Op, 2, DAG);
  case Intrinsic::loongarch_lsx_vseqi_b:
  case Intrinsic::loongarch_lsx_vseqi_h:
  case Intrinsic::loongarch_lsx_vseqi_w:
  case Intrinsic::loongarch_lsx_vseqi_d:
  case Intrinsic::loongarch_lsx_vslei_b:
  case Intrinsic::loongarch_lsx_vslei_h:
  case Intrinsic::loongarch_lsx_vslei_w:
  case Intrinsic::loongarch_lsx_vslei_d:
  case Intrinsic::loongarch_lsx_vslti_b:
  case Intrinsic::loongarch_lsx_vslti_h:
  case Intrinsic::loongarch_lsx_vslti_w:
  case Intrinsic::loongarch_lsx_vslti_d:
  case Intrinsic::loongarch_lasx_xvseqi_b:
  case Intrinsic::loongarch_lasx_xvseqi_h:
  case Intrinsic::loongarch_lasx_xvseqi_w:
  case Intrinsic::loongarch_lasx_xvseqi_d:
  case Intrinsic::loongarch_lasx_xvslei_b:
  case Intrinsic::loongarch_lasx_xvslei_h:
  case Intrinsic::loongarch_lasx_xvslei_w:
  case Intrinsic::loongarch_lasx_xvslei_d:
  case Intrinsic::loongarch_lasx_xvslti_b:
  case Intrinsic::loongarch_lasx_xvslti_h:
  case Intrinsic::loongarch_lasx_xvslti_w:
  case Intrinsic::loongarch_lasx_xvslti_d:
    return checkIntrinsicImmArg<5>(Op, 2, DAG, /*IsSigned=*/true);
  case Intrinsic::loongarch_lsx_vsrlni_h_w:
  case Intrinsic::loongarch_lsx_vsrani_h_w:
  case Intrinsic::loongarch_lsx_vsrlrni_h_w:
  case Intrinsic::loongarch_lsx_vsrarni_h_w:
  case Intrinsic::loongarch_lsx_vssrlni_h_w:
  case Intrinsic::loongarch_lsx_vssrani_h_w:
  case Intrinsic::loongarch_lsx_vssrlni_hu_w:
  case Intrinsic::loongarch_lsx_vssrani_hu_w:
  case Intrinsic::loongarch_lsx_vssrlrni_h_w:
  case Intrinsic::loongarch_lsx_vssrarni_h_w:
  case Intrinsic::loongarch_lsx_vssrlrni_hu_w:
  case Intrinsic::loongarch_lsx_vssrarni_hu_w:
  case Intrinsic::loongarch_lsx_vfrstpi_b:
  case Intrinsic::loongarch_lsx_vfrstpi_h:
  case Intrinsic::loongarch_lasx_xvsrlni_h_w:
  case Intrinsic::loongarch_lasx_xvsrani_h_w:
  case Intrinsic::loongarch_lasx_xvsrlrni_h_w:
  case Intrinsic::loongarch_lasx_xvsrarni_h_w:
  case Intrinsic::loongarch_lasx_xvssrlni_h_w:
  case Intrinsic::loongarch_lasx_xvssrani_h_w:
  case Intrinsic::loongarch_lasx_xvssrlni_hu_w:
  case Intrinsic::loongarch_lasx_xvssrani_hu_w:
  case Intrinsic::loongarch_lasx_xvssrlrni_h_w:
  case Intrinsic::loongarch_lasx_xvssrarni_h_w:
  case Intrinsic::loongarch_lasx_xvssrlrni_hu_w:
  case Intrinsic::loongarch_lasx_xvssrarni_hu_w:
  case Intrinsic::loongarch_lasx_xvfrstpi_b:
  case Intrinsic::loongarch_lasx_xvfrstpi_h:
    return checkIntrinsicImmArg<5>(Op, 3, DAG);
  case Intrinsic::loongarch_lsx_vsat_d:
  case Intrinsic::loongarch_lsx_vsat_du:
  case Intrinsic::loongarch_lsx_vrotri_d:
  case Intrinsic::loongarch_lsx_vsrlri_d:
  case Intrinsic::loongarch_lsx_vsrari_d:
  case Intrinsic::loongarch_lasx_xvsat_d:
  case Intrinsic::loongarch_lasx_xvsat_du:
  case Intrinsic::loongarch_lasx_xvrotri_d:
  case Intrinsic::loongarch_lasx_xvsrlri_d:
  case Intrinsic::loongarch_lasx_xvsrari_d:
    return checkIntrinsicImmArg<6>(Op, 2, DAG);
  case Intrinsic::loongarch_lsx_vsrlni_w_d:
  case Intrinsic::loongarch_lsx_vsrani_w_d:
  case Intrinsic::loongarch_lsx_vsrlrni_w_d:
  case Intrinsic::loongarch_lsx_vsrarni_w_d:
  case Intrinsic::loongarch_lsx_vssrlni_w_d:
  case Intrinsic::loongarch_lsx_vssrani_w_d:
  case Intrinsic::loongarch_lsx_vssrlni_wu_d:
  case Intrinsic::loongarch_lsx_vssrani_wu_d:
  case Intrinsic::loongarch_lsx_vssrlrni_w_d:
  case Intrinsic::loongarch_lsx_vssrarni_w_d:
  case Intrinsic::loongarch_lsx_vssrlrni_wu_d:
  case Intrinsic::loongarch_lsx_vssrarni_wu_d:
  case Intrinsic::loongarch_lasx_xvsrlni_w_d:
  case Intrinsic::loongarch_lasx_xvsrani_w_d:
  case Intrinsic::loongarch_lasx_xvsrlrni_w_d:
  case Intrinsic::loongarch_lasx_xvsrarni_w_d:
  case Intrinsic::loongarch_lasx_xvssrlni_w_d:
  case Intrinsic::loongarch_lasx_xvssrani_w_d:
  case Intrinsic::loongarch_lasx_xvssrlni_wu_d:
  case Intrinsic::loongarch_lasx_xvssrani_wu_d:
  case Intrinsic::loongarch_lasx_xvssrlrni_w_d:
  case Intrinsic::loongarch_lasx_xvssrarni_w_d:
  case Intrinsic::loongarch_lasx_xvssrlrni_wu_d:
  case Intrinsic::loongarch_lasx_xvssrarni_wu_d:
    return checkIntrinsicImmArg<6>(Op, 3, DAG);
  case Intrinsic::loongarch_lsx_vsrlni_d_q:
  case Intrinsic::loongarch_lsx_vsrani_d_q:
  case Intrinsic::loongarch_lsx_vsrlrni_d_q:
  case Intrinsic::loongarch_lsx_vsrarni_d_q:
  case Intrinsic::loongarch_lsx_vssrlni_d_q:
  case Intrinsic::loongarch_lsx_vssrani_d_q:
  case Intrinsic::loongarch_lsx_vssrlni_du_q:
  case Intrinsic::loongarch_lsx_vssrani_du_q:
  case Intrinsic::loongarch_lsx_vssrlrni_d_q:
  case Intrinsic::loongarch_lsx_vssrarni_d_q:
  case Intrinsic::loongarch_lsx_vssrlrni_du_q:
  case Intrinsic::loongarch_lsx_vssrarni_du_q:
  case Intrinsic::loongarch_lasx_xvsrlni_d_q:
  case Intrinsic::loongarch_lasx_xvsrani_d_q:
  case Intrinsic::loongarch_lasx_xvsrlrni_d_q:
  case Intrinsic::loongarch_lasx_xvsrarni_d_q:
  case Intrinsic::loongarch_lasx_xvssrlni_d_q:
  case Intrinsic::loongarch_lasx_xvssrani_d_q:
  case Intrinsic::loongarch_lasx_xvssrlni_du_q:
  case Intrinsic::loongarch_lasx_xvssrani_du_q:
  case Intrinsic::loongarch_lasx_xvssrlrni_d_q:
  case Intrinsic::loongarch_lasx_xvssrarni_d_q:
  case Intrinsic::loongarch_lasx_xvssrlrni_du_q:
  case Intrinsic::loongarch_lasx_xvssrarni_du_q:
    return checkIntrinsicImmArg<7>(Op, 3, DAG);
  case Intrinsic::loongarch_lsx_vnori_b:
  case Intrinsic::loongarch_lsx_vshuf4i_b:
  case Intrinsic::loongarch_lsx_vshuf4i_h:
  case Intrinsic::loongarch_lsx_vshuf4i_w:
  case Intrinsic::loongarch_lasx_xvnori_b:
  case Intrinsic::loongarch_lasx_xvshuf4i_b:
  case Intrinsic::loongarch_lasx_xvshuf4i_h:
  case Intrinsic::loongarch_lasx_xvshuf4i_w:
  case Intrinsic::loongarch_lasx_xvpermi_d:
    return checkIntrinsicImmArg<8>(Op, 2, DAG);
  case Intrinsic::loongarch_lsx_vshuf4i_d:
  case Intrinsic::loongarch_lsx_vpermi_w:
  case Intrinsic::loongarch_lsx_vbitseli_b:
  case Intrinsic::loongarch_lsx_vextrins_b:
  case Intrinsic::loongarch_lsx_vextrins_h:
  case Intrinsic::loongarch_lsx_vextrins_w:
  case Intrinsic::loongarch_lsx_vextrins_d:
  case Intrinsic::loongarch_lasx_xvshuf4i_d:
  case Intrinsic::loongarch_lasx_xvpermi_w:
  case Intrinsic::loongarch_lasx_xvpermi_q:
  case Intrinsic::loongarch_lasx_xvbitseli_b:
  case Intrinsic::loongarch_lasx_xvextrins_b:
  case Intrinsic::loongarch_lasx_xvextrins_h:
  case Intrinsic::loongarch_lasx_xvextrins_w:
  case Intrinsic::loongarch_lasx_xvextrins_d:
    return checkIntrinsicImmArg<8>(Op, 3, DAG);
  case Intrinsic::loongarch_lsx_vrepli_b:
  case Intrinsic::loongarch_lsx_vrepli_h:
  case Intrinsic::loongarch_lsx_vrepli_w:
  case Intrinsic::loongarch_lsx_vrepli_d:
  case Intrinsic::loongarch_lasx_xvrepli_b:
  case Intrinsic::loongarch_lasx_xvrepli_h:
  case Intrinsic::loongarch_lasx_xvrepli_w:
  case Intrinsic::loongarch_lasx_xvrepli_d:
    return checkIntrinsicImmArg<10>(Op, 1, DAG, /*IsSigned=*/true);
  case Intrinsic::loongarch_lsx_vldi:
  case Intrinsic::loongarch_lasx_xvldi:
    return checkIntrinsicImmArg<13>(Op, 1, DAG, /*IsSigned=*/true);
  }
}
 
// Helper function that emits error message for intrinsics with chain and return
// merge values of a UNDEF and the chain.
static SDValue emitIntrinsicWithChainErrorMessage(SDValue Op,
                                                  StringRef ErrorMsg,
                                                  SelectionDAG &DAG) {
  DAG.getContext()->emitError(Op->getOperationName(0) + ": " + ErrorMsg + ".");
  return DAG.getMergeValues({DAG.getUNDEF(Op.getValueType()), Op.getOperand(0)},
                            SDLoc(Op));
}
 
SDValue
LoongArchTargetLowering::lowerINTRINSIC_W_CHAIN(SDValue Op,
                                                SelectionDAG &DAG) const {
  SDLoc DL(Op);
  MVT GRLenVT = Subtarget.getGRLenVT();
  EVT VT = Op.getValueType();
  SDValue Chain = Op.getOperand(0);
  const StringRef ErrorMsgOOR = "argument out of range";
  const StringRef ErrorMsgReqLA64 = "requires loongarch64";
  const StringRef ErrorMsgReqF = "requires basic 'f' target feature";
 
  switch (Op.getConstantOperandVal(1)) {
  default:
    return Op;
  case Intrinsic::loongarch_crc_w_b_w:
  case Intrinsic::loongarch_crc_w_h_w:
  case Intrinsic::loongarch_crc_w_w_w:
  case Intrinsic::loongarch_crc_w_d_w:
  case Intrinsic::loongarch_crcc_w_b_w:
  case Intrinsic::loongarch_crcc_w_h_w:
  case Intrinsic::loongarch_crcc_w_w_w:
  case Intrinsic::loongarch_crcc_w_d_w:
    return emitIntrinsicWithChainErrorMessage(Op, ErrorMsgReqLA64, DAG);
  case Intrinsic::loongarch_csrrd_w:
  case Intrinsic::loongarch_csrrd_d: {
    unsigned Imm = Op.getConstantOperandVal(2);
    return !isUInt<14>(Imm)
               ? emitIntrinsicWithChainErrorMessage(Op, ErrorMsgOOR, DAG)
               : DAG.getNode(LoongArchISD::CSRRD, DL, {GRLenVT, MVT::Other},
                             {Chain, DAG.getConstant(Imm, DL, GRLenVT)});
  }
  case Intrinsic::loongarch_csrwr_w:
  case Intrinsic::loongarch_csrwr_d: {
    unsigned Imm = Op.getConstantOperandVal(3);
    return !isUInt<14>(Imm)
               ? emitIntrinsicWithChainErrorMessage(Op, ErrorMsgOOR, DAG)
               : DAG.getNode(LoongArchISD::CSRWR, DL, {GRLenVT, MVT::Other},
                             {Chain, Op.getOperand(2),
                              DAG.getConstant(Imm, DL, GRLenVT)});
  }
  case Intrinsic::loongarch_csrxchg_w:
  case Intrinsic::loongarch_csrxchg_d: {
    unsigned Imm = Op.getConstantOperandVal(4);
    return !isUInt<14>(Imm)
               ? emitIntrinsicWithChainErrorMessage(Op, ErrorMsgOOR, DAG)
               : DAG.getNode(LoongArchISD::CSRXCHG, DL, {GRLenVT, MVT::Other},
                             {Chain, Op.getOperand(2), Op.getOperand(3),
                              DAG.getConstant(Imm, DL, GRLenVT)});
  }
  case Intrinsic::loongarch_iocsrrd_d: {
    return DAG.getNode(
        LoongArchISD::IOCSRRD_D, DL, {GRLenVT, MVT::Other},
        {Chain, DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op.getOperand(2))});
  }
#define IOCSRRD_CASE(NAME, NODE)                                               \
  case Intrinsic::loongarch_##NAME: {                                          \
    return DAG.getNode(LoongArchISD::NODE, DL, {GRLenVT, MVT::Other},          \
                       {Chain, Op.getOperand(2)});                             \
  }
    IOCSRRD_CASE(iocsrrd_b, IOCSRRD_B);
    IOCSRRD_CASE(iocsrrd_h, IOCSRRD_H);
    IOCSRRD_CASE(iocsrrd_w, IOCSRRD_W);
#undef IOCSRRD_CASE
  case Intrinsic::loongarch_cpucfg: {
    return DAG.getNode(LoongArchISD::CPUCFG, DL, {GRLenVT, MVT::Other},
                       {Chain, Op.getOperand(2)});
  }
  case Intrinsic::loongarch_lddir_d: {
    unsigned Imm = Op.getConstantOperandVal(3);
    return !isUInt<8>(Imm)
               ? emitIntrinsicWithChainErrorMessage(Op, ErrorMsgOOR, DAG)
               : Op;
  }
  case Intrinsic::loongarch_movfcsr2gr: {
    if (!Subtarget.hasBasicF())
      return emitIntrinsicWithChainErrorMessage(Op, ErrorMsgReqF, DAG);
    unsigned Imm = Op.getConstantOperandVal(2);
    return !isUInt<2>(Imm)
               ? emitIntrinsicWithChainErrorMessage(Op, ErrorMsgOOR, DAG)
               : DAG.getNode(LoongArchISD::MOVFCSR2GR, DL, {VT, MVT::Other},
                             {Chain, DAG.getConstant(Imm, DL, GRLenVT)});
  }
  case Intrinsic::loongarch_lsx_vld:
  case Intrinsic::loongarch_lsx_vldrepl_b:
  case Intrinsic::loongarch_lasx_xvld:
  case Intrinsic::loongarch_lasx_xvldrepl_b:
    return !isInt<12>(cast<ConstantSDNode>(Op.getOperand(3))->getSExtValue())
               ? emitIntrinsicWithChainErrorMessage(Op, ErrorMsgOOR, DAG)
               : SDValue();
  case Intrinsic::loongarch_lsx_vldrepl_h:
  case Intrinsic::loongarch_lasx_xvldrepl_h:
    return !isShiftedInt<11, 1>(
               cast<ConstantSDNode>(Op.getOperand(3))->getSExtValue())
               ? emitIntrinsicWithChainErrorMessage(
                     Op, "argument out of range or not a multiple of 2", DAG)
               : SDValue();
  case Intrinsic::loongarch_lsx_vldrepl_w:
  case Intrinsic::loongarch_lasx_xvldrepl_w:
    return !isShiftedInt<10, 2>(
               cast<ConstantSDNode>(Op.getOperand(3))->getSExtValue())
               ? emitIntrinsicWithChainErrorMessage(
                     Op, "argument out of range or not a multiple of 4", DAG)
               : SDValue();
  case Intrinsic::loongarch_lsx_vldrepl_d:
  case Intrinsic::loongarch_lasx_xvldrepl_d:
    return !isShiftedInt<9, 3>(
               cast<ConstantSDNode>(Op.getOperand(3))->getSExtValue())
               ? emitIntrinsicWithChainErrorMessage(
                     Op, "argument out of range or not a multiple of 8", DAG)
               : SDValue();
  }
}
 
// Helper function that emits error message for intrinsics with void return
// value and return the chain.
static SDValue emitIntrinsicErrorMessage(SDValue Op, StringRef ErrorMsg,
                                         SelectionDAG &DAG) {
 
  DAG.getContext()->emitError(Op->getOperationName(0) + ": " + ErrorMsg + ".");
  return Op.getOperand(0);
}
 
SDValue LoongArchTargetLowering::lowerINTRINSIC_VOID(SDValue Op,
                                                     SelectionDAG &DAG) const {
  SDLoc DL(Op);
  MVT GRLenVT = Subtarget.getGRLenVT();
  SDValue Chain = Op.getOperand(0);
  uint64_t IntrinsicEnum = Op.getConstantOperandVal(1);
  SDValue Op2 = Op.getOperand(2);
  const StringRef ErrorMsgOOR = "argument out of range";
  const StringRef ErrorMsgReqLA64 = "requires loongarch64";
  const StringRef ErrorMsgReqLA32 = "requires loongarch32";
  const StringRef ErrorMsgReqF = "requires basic 'f' target feature";
 
  switch (IntrinsicEnum) {
  default:
    // TODO: Add more Intrinsics.
    return SDValue();
  case Intrinsic::loongarch_cacop_d:
  case Intrinsic::loongarch_cacop_w: {
    if (IntrinsicEnum == Intrinsic::loongarch_cacop_d && !Subtarget.is64Bit())
      return emitIntrinsicErrorMessage(Op, ErrorMsgReqLA64, DAG);
    if (IntrinsicEnum == Intrinsic::loongarch_cacop_w && Subtarget.is64Bit())
      return emitIntrinsicErrorMessage(Op, ErrorMsgReqLA32, DAG);
    // call void @llvm.loongarch.cacop.[d/w](uimm5, rj, simm12)
    unsigned Imm1 = Op2->getAsZExtVal();
    int Imm2 = cast<ConstantSDNode>(Op.getOperand(4))->getSExtValue();
    if (!isUInt<5>(Imm1) || !isInt<12>(Imm2))
      return emitIntrinsicErrorMessage(Op, ErrorMsgOOR, DAG);
    return Op;
  }
  case Intrinsic::loongarch_dbar: {
    unsigned Imm = Op2->getAsZExtVal();
    return !isUInt<15>(Imm)
               ? emitIntrinsicErrorMessage(Op, ErrorMsgOOR, DAG)
               : DAG.getNode(LoongArchISD::DBAR, DL, MVT::Other, Chain,
                             DAG.getConstant(Imm, DL, GRLenVT));
  }
  case Intrinsic::loongarch_ibar: {
    unsigned Imm = Op2->getAsZExtVal();
    return !isUInt<15>(Imm)
               ? emitIntrinsicErrorMessage(Op, ErrorMsgOOR, DAG)
               : DAG.getNode(LoongArchISD::IBAR, DL, MVT::Other, Chain,
                             DAG.getConstant(Imm, DL, GRLenVT));
  }
  case Intrinsic::loongarch_break: {
    unsigned Imm = Op2->getAsZExtVal();
    return !isUInt<15>(Imm)
               ? emitIntrinsicErrorMessage(Op, ErrorMsgOOR, DAG)
               : DAG.getNode(LoongArchISD::BREAK, DL, MVT::Other, Chain,
                             DAG.getConstant(Imm, DL, GRLenVT));
  }
  case Intrinsic::loongarch_movgr2fcsr: {
    if (!Subtarget.hasBasicF())
      return emitIntrinsicErrorMessage(Op, ErrorMsgReqF, DAG);
    unsigned Imm = Op2->getAsZExtVal();
    return !isUInt<2>(Imm)
               ? emitIntrinsicErrorMessage(Op, ErrorMsgOOR, DAG)
               : DAG.getNode(LoongArchISD::MOVGR2FCSR, DL, MVT::Other, Chain,
                             DAG.getConstant(Imm, DL, GRLenVT),
                             DAG.getNode(ISD::ANY_EXTEND, DL, GRLenVT,
                                         Op.getOperand(3)));
  }
  case Intrinsic::loongarch_syscall: {
    unsigned Imm = Op2->getAsZExtVal();
    return !isUInt<15>(Imm)
               ? emitIntrinsicErrorMessage(Op, ErrorMsgOOR, DAG)
               : DAG.getNode(LoongArchISD::SYSCALL, DL, MVT::Other, Chain,
                             DAG.getConstant(Imm, DL, GRLenVT));
  }
#define IOCSRWR_CASE(NAME, NODE)                                               \
  case Intrinsic::loongarch_##NAME: {                                          \
    SDValue Op3 = Op.getOperand(3);                                            \
    return Subtarget.is64Bit()                                                 \
               ? DAG.getNode(LoongArchISD::NODE, DL, MVT::Other, Chain,        \
                             DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op2),  \
                             DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op3))  \
               : DAG.getNode(LoongArchISD::NODE, DL, MVT::Other, Chain, Op2,   \
                             Op3);                                             \
  }
    IOCSRWR_CASE(iocsrwr_b, IOCSRWR_B);
    IOCSRWR_CASE(iocsrwr_h, IOCSRWR_H);
    IOCSRWR_CASE(iocsrwr_w, IOCSRWR_W);
#undef IOCSRWR_CASE
  case Intrinsic::loongarch_iocsrwr_d: {
    return !Subtarget.is64Bit()
               ? emitIntrinsicErrorMessage(Op, ErrorMsgReqLA64, DAG)
               : DAG.getNode(LoongArchISD::IOCSRWR_D, DL, MVT::Other, Chain,
                             Op2,
                             DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64,
                                         Op.getOperand(3)));
  }
#define ASRT_LE_GT_CASE(NAME)                                                  \
  case Intrinsic::loongarch_##NAME: {                                          \
    return !Subtarget.is64Bit()                                                \
               ? emitIntrinsicErrorMessage(Op, ErrorMsgReqLA64, DAG)           \
               : Op;                                                           \
  }
    ASRT_LE_GT_CASE(asrtle_d)
    ASRT_LE_GT_CASE(asrtgt_d)
#undef ASRT_LE_GT_CASE
  case Intrinsic::loongarch_ldpte_d: {
    unsigned Imm = Op.getConstantOperandVal(3);
    return !Subtarget.is64Bit()
               ? emitIntrinsicErrorMessage(Op, ErrorMsgReqLA64, DAG)
           : !isUInt<8>(Imm) ? emitIntrinsicErrorMessage(Op, ErrorMsgOOR, DAG)
                             : Op;
  }
  case Intrinsic::loongarch_lsx_vst:
  case Intrinsic::loongarch_lasx_xvst:
    return !isInt<12>(cast<ConstantSDNode>(Op.getOperand(4))->getSExtValue())
               ? emitIntrinsicErrorMessage(Op, ErrorMsgOOR, DAG)
               : SDValue();
  case Intrinsic::loongarch_lasx_xvstelm_b:
    return (!isInt<8>(cast<ConstantSDNode>(Op.getOperand(4))->getSExtValue()) ||
            !isUInt<5>(Op.getConstantOperandVal(5)))
               ? emitIntrinsicErrorMessage(Op, ErrorMsgOOR, DAG)
               : SDValue();
  case Intrinsic::loongarch_lsx_vstelm_b:
    return (!isInt<8>(cast<ConstantSDNode>(Op.getOperand(4))->getSExtValue()) ||
            !isUInt<4>(Op.getConstantOperandVal(5)))
               ? emitIntrinsicErrorMessage(Op, ErrorMsgOOR, DAG)
               : SDValue();
  case Intrinsic::loongarch_lasx_xvstelm_h:
    return (!isShiftedInt<8, 1>(
                cast<ConstantSDNode>(Op.getOperand(4))->getSExtValue()) ||
            !isUInt<4>(Op.getConstantOperandVal(5)))
               ? emitIntrinsicErrorMessage(
                     Op, "argument out of range or not a multiple of 2", DAG)
               : SDValue();
  case Intrinsic::loongarch_lsx_vstelm_h:
    return (!isShiftedInt<8, 1>(
                cast<ConstantSDNode>(Op.getOperand(4))->getSExtValue()) ||
            !isUInt<3>(Op.getConstantOperandVal(5)))
               ? emitIntrinsicErrorMessage(
                     Op, "argument out of range or not a multiple of 2", DAG)
               : SDValue();
  case Intrinsic::loongarch_lasx_xvstelm_w:
    return (!isShiftedInt<8, 2>(
                cast<ConstantSDNode>(Op.getOperand(4))->getSExtValue()) ||
            !isUInt<3>(Op.getConstantOperandVal(5)))
               ? emitIntrinsicErrorMessage(
                     Op, "argument out of range or not a multiple of 4", DAG)
               : SDValue();
  case Intrinsic::loongarch_lsx_vstelm_w:
    return (!isShiftedInt<8, 2>(
                cast<ConstantSDNode>(Op.getOperand(4))->getSExtValue()) ||
            !isUInt<2>(Op.getConstantOperandVal(5)))
               ? emitIntrinsicErrorMessage(
                     Op, "argument out of range or not a multiple of 4", DAG)
               : SDValue();
  case Intrinsic::loongarch_lasx_xvstelm_d:
    return (!isShiftedInt<8, 3>(
                cast<ConstantSDNode>(Op.getOperand(4))->getSExtValue()) ||
            !isUInt<2>(Op.getConstantOperandVal(5)))
               ? emitIntrinsicErrorMessage(
                     Op, "argument out of range or not a multiple of 8", DAG)
               : SDValue();
  case Intrinsic::loongarch_lsx_vstelm_d:
    return (!isShiftedInt<8, 3>(
                cast<ConstantSDNode>(Op.getOperand(4))->getSExtValue()) ||
            !isUInt<1>(Op.getConstantOperandVal(5)))
               ? emitIntrinsicErrorMessage(
                     Op, "argument out of range or not a multiple of 8", DAG)
               : SDValue();
  }
}
 
SDValue LoongArchTargetLowering::lowerShiftLeftParts(SDValue Op,
                                                     SelectionDAG &DAG) const {
  SDLoc DL(Op);
  SDValue Lo = Op.getOperand(0);
  SDValue Hi = Op.getOperand(1);
  SDValue Shamt = Op.getOperand(2);
  EVT VT = Lo.getValueType();
 
  // if Shamt-GRLen < 0: // Shamt < GRLen
  //   Lo = Lo << Shamt
  //   Hi = (Hi << Shamt) | ((Lo >>u 1) >>u (GRLen-1 ^ Shamt))
  // else:
  //   Lo = 0
  //   Hi = Lo << (Shamt-GRLen)
 
  SDValue Zero = DAG.getConstant(0, DL, VT);
  SDValue One = DAG.getConstant(1, DL, VT);
  SDValue MinusGRLen = DAG.getConstant(-(int)Subtarget.getGRLen(), DL, VT);
  SDValue GRLenMinus1 = DAG.getConstant(Subtarget.getGRLen() - 1, DL, VT);
  SDValue ShamtMinusGRLen = DAG.getNode(ISD::ADD, DL, VT, Shamt, MinusGRLen);
  SDValue GRLenMinus1Shamt = DAG.getNode(ISD::XOR, DL, VT, Shamt, GRLenMinus1);
 
  SDValue LoTrue = DAG.getNode(ISD::SHL, DL, VT, Lo, Shamt);
  SDValue ShiftRight1Lo = DAG.getNode(ISD::SRL, DL, VT, Lo, One);
  SDValue ShiftRightLo =
      DAG.getNode(ISD::SRL, DL, VT, ShiftRight1Lo, GRLenMinus1Shamt);
  SDValue ShiftLeftHi = DAG.getNode(ISD::SHL, DL, VT, Hi, Shamt);
  SDValue HiTrue = DAG.getNode(ISD::OR, DL, VT, ShiftLeftHi, ShiftRightLo);
  SDValue HiFalse = DAG.getNode(ISD::SHL, DL, VT, Lo, ShamtMinusGRLen);
 
  SDValue CC = DAG.getSetCC(DL, VT, ShamtMinusGRLen, Zero, ISD::SETLT);
 
  Lo = DAG.getNode(ISD::SELECT, DL, VT, CC, LoTrue, Zero);
  Hi = DAG.getNode(ISD::SELECT, DL, VT, CC, HiTrue, HiFalse);
 
  SDValue Parts[2] = {Lo, Hi};
  return DAG.getMergeValues(Parts, DL);
}
 
SDValue LoongArchTargetLowering::lowerShiftRightParts(SDValue Op,
                                                      SelectionDAG &DAG,
                                                      bool IsSRA) const {
  SDLoc DL(Op);
  SDValue Lo = Op.getOperand(0);
  SDValue Hi = Op.getOperand(1);
  SDValue Shamt = Op.getOperand(2);
  EVT VT = Lo.getValueType();
 
  // SRA expansion:
  //   if Shamt-GRLen < 0: // Shamt < GRLen
  //     Lo = (Lo >>u Shamt) | ((Hi << 1) << (ShAmt ^ GRLen-1))
  //     Hi = Hi >>s Shamt
  //   else:
  //     Lo = Hi >>s (Shamt-GRLen);
  //     Hi = Hi >>s (GRLen-1)
  //
  // SRL expansion:
  //   if Shamt-GRLen < 0: // Shamt < GRLen
  //     Lo = (Lo >>u Shamt) | ((Hi << 1) << (ShAmt ^ GRLen-1))
  //     Hi = Hi >>u Shamt
  //   else:
  //     Lo = Hi >>u (Shamt-GRLen);
  //     Hi = 0;
 
  unsigned ShiftRightOp = IsSRA ? ISD::SRA : ISD::SRL;
 
  SDValue Zero = DAG.getConstant(0, DL, VT);
  SDValue One = DAG.getConstant(1, DL, VT);
  SDValue MinusGRLen = DAG.getConstant(-(int)Subtarget.getGRLen(), DL, VT);
  SDValue GRLenMinus1 = DAG.getConstant(Subtarget.getGRLen() - 1, DL, VT);
  SDValue ShamtMinusGRLen = DAG.getNode(ISD::ADD, DL, VT, Shamt, MinusGRLen);
  SDValue GRLenMinus1Shamt = DAG.getNode(ISD::XOR, DL, VT, Shamt, GRLenMinus1);
 
  SDValue ShiftRightLo = DAG.getNode(ISD::SRL, DL, VT, Lo, Shamt);
  SDValue ShiftLeftHi1 = DAG.getNode(ISD::SHL, DL, VT, Hi, One);
  SDValue ShiftLeftHi =
      DAG.getNode(ISD::SHL, DL, VT, ShiftLeftHi1, GRLenMinus1Shamt);
  SDValue LoTrue = DAG.getNode(ISD::OR, DL, VT, ShiftRightLo, ShiftLeftHi);
  SDValue HiTrue = DAG.getNode(ShiftRightOp, DL, VT, Hi, Shamt);
  SDValue LoFalse = DAG.getNode(ShiftRightOp, DL, VT, Hi, ShamtMinusGRLen);
  SDValue HiFalse =
      IsSRA ? DAG.getNode(ISD::SRA, DL, VT, Hi, GRLenMinus1) : Zero;
 
  SDValue CC = DAG.getSetCC(DL, VT, ShamtMinusGRLen, Zero, ISD::SETLT);
 
  Lo = DAG.getNode(ISD::SELECT, DL, VT, CC, LoTrue, LoFalse);
  Hi = DAG.getNode(ISD::SELECT, DL, VT, CC, HiTrue, HiFalse);
 
  SDValue Parts[2] = {Lo, Hi};
  return DAG.getMergeValues(Parts, DL);
}
 
// Returns the opcode of the target-specific SDNode that implements the 32-bit
// form of the given Opcode.
static LoongArchISD::NodeType getLoongArchWOpcode(unsigned Opcode) {
  switch (Opcode) {
  default:
    llvm_unreachable("Unexpected opcode");
  case ISD::UDIV:
    return LoongArchISD::DIV_WU;
  case ISD::UREM:
    return LoongArchISD::MOD_WU;
  case ISD::SHL:
    return LoongArchISD::SLL_W;
  case ISD::SRA:
    return LoongArchISD::SRA_W;
  case ISD::SRL:
    return LoongArchISD::SRL_W;
  case ISD::ROTL:
  case ISD::ROTR:
    return LoongArchISD::ROTR_W;
  case ISD::CTTZ:
    return LoongArchISD::CTZ_W;
  case ISD::CTLZ:
    return LoongArchISD::CLZ_W;
  }
}
 
// Converts the given i8/i16/i32 operation to a target-specific SelectionDAG
// node. Because i8/i16/i32 isn't a legal type for LA64, these operations would
// otherwise be promoted to i64, making it difficult to select the
// SLL_W/.../*W later one because the fact the operation was originally of
// type i8/i16/i32 is lost.
static SDValue customLegalizeToWOp(SDNode *N, SelectionDAG &DAG, int NumOp,
                                   unsigned ExtOpc = ISD::ANY_EXTEND) {
  SDLoc DL(N);
  LoongArchISD::NodeType WOpcode = getLoongArchWOpcode(N->getOpcode());
  SDValue NewOp0, NewRes;
 
  switch (NumOp) {
  default:
    llvm_unreachable("Unexpected NumOp");
  case 1: {
    NewOp0 = DAG.getNode(ExtOpc, DL, MVT::i64, N->getOperand(0));
    NewRes = DAG.getNode(WOpcode, DL, MVT::i64, NewOp0);
    break;
  }
  case 2: {
    NewOp0 = DAG.getNode(ExtOpc, DL, MVT::i64, N->getOperand(0));
    SDValue NewOp1 = DAG.getNode(ExtOpc, DL, MVT::i64, N->getOperand(1));
    if (N->getOpcode() == ISD::ROTL) {
      SDValue TmpOp = DAG.getConstant(32, DL, MVT::i64);
      NewOp1 = DAG.getNode(ISD::SUB, DL, MVT::i64, TmpOp, NewOp1);
    }
    NewRes = DAG.getNode(WOpcode, DL, MVT::i64, NewOp0, NewOp1);
    break;
  }
    // TODO:Handle more NumOp.
  }
 
  // ReplaceNodeResults requires we maintain the same type for the return
  // value.
  return DAG.getNode(ISD::TRUNCATE, DL, N->getValueType(0), NewRes);
}
 
// Converts the given 32-bit operation to a i64 operation with signed extension
// semantic to reduce the signed extension instructions.
static SDValue customLegalizeToWOpWithSExt(SDNode *N, SelectionDAG &DAG) {
  SDLoc DL(N);
  SDValue NewOp0 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(0));
  SDValue NewOp1 = DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(1));
  SDValue NewWOp = DAG.getNode(N->getOpcode(), DL, MVT::i64, NewOp0, NewOp1);
  SDValue NewRes = DAG.getNode(ISD::SIGN_EXTEND_INREG, DL, MVT::i64, NewWOp,
                               DAG.getValueType(MVT::i32));
  return DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, NewRes);
}
 
// Helper function that emits error message for intrinsics with/without chain
// and return a UNDEF or and the chain as the results.
static void emitErrorAndReplaceIntrinsicResults(
    SDNode *N, SmallVectorImpl<SDValue> &Results, SelectionDAG &DAG,
    StringRef ErrorMsg, bool WithChain = true) {
  DAG.getContext()->emitError(N->getOperationName(0) + ": " + ErrorMsg + ".");
  Results.push_back(DAG.getUNDEF(N->getValueType(0)));
  if (!WithChain)
    return;
  Results.push_back(N->getOperand(0));
}
 
template <unsigned N>
static void
replaceVPICKVE2GRResults(SDNode *Node, SmallVectorImpl<SDValue> &Results,
                         SelectionDAG &DAG, const LoongArchSubtarget &Subtarget,
                         unsigned ResOp) {
  const StringRef ErrorMsgOOR = "argument out of range";
  unsigned Imm = Node->getConstantOperandVal(2);
  if (!isUInt<N>(Imm)) {
    emitErrorAndReplaceIntrinsicResults(Node, Results, DAG, ErrorMsgOOR,
                                        /*WithChain=*/false);
    return;
  }
  SDLoc DL(Node);
  SDValue Vec = Node->getOperand(1);
 
  SDValue PickElt =
      DAG.getNode(ResOp, DL, Subtarget.getGRLenVT(), Vec,
                  DAG.getConstant(Imm, DL, Subtarget.getGRLenVT()),
                  DAG.getValueType(Vec.getValueType().getVectorElementType()));
  Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, Node->getValueType(0),
                                PickElt.getValue(0)));
}
 
static void replaceVecCondBranchResults(SDNode *N,
                                        SmallVectorImpl<SDValue> &Results,
                                        SelectionDAG &DAG,
                                        const LoongArchSubtarget &Subtarget,
                                        unsigned ResOp) {
  SDLoc DL(N);
  SDValue Vec = N->getOperand(1);
 
  SDValue CB = DAG.getNode(ResOp, DL, Subtarget.getGRLenVT(), Vec);
  Results.push_back(
      DAG.getNode(ISD::TRUNCATE, DL, N->getValueType(0), CB.getValue(0)));
}
 
static void
replaceINTRINSIC_WO_CHAINResults(SDNode *N, SmallVectorImpl<SDValue> &Results,
                                 SelectionDAG &DAG,
                                 const LoongArchSubtarget &Subtarget) {
  switch (N->getConstantOperandVal(0)) {
  default:
    llvm_unreachable("Unexpected Intrinsic.");
  case Intrinsic::loongarch_lsx_vpickve2gr_b:
    replaceVPICKVE2GRResults<4>(N, Results, DAG, Subtarget,
                                LoongArchISD::VPICK_SEXT_ELT);
    break;
  case Intrinsic::loongarch_lsx_vpickve2gr_h:
  case Intrinsic::loongarch_lasx_xvpickve2gr_w:
    replaceVPICKVE2GRResults<3>(N, Results, DAG, Subtarget,
                                LoongArchISD::VPICK_SEXT_ELT);
    break;
  case Intrinsic::loongarch_lsx_vpickve2gr_w:
    replaceVPICKVE2GRResults<2>(N, Results, DAG, Subtarget,
                                LoongArchISD::VPICK_SEXT_ELT);
    break;
  case Intrinsic::loongarch_lsx_vpickve2gr_bu:
    replaceVPICKVE2GRResults<4>(N, Results, DAG, Subtarget,
                                LoongArchISD::VPICK_ZEXT_ELT);
    break;
  case Intrinsic::loongarch_lsx_vpickve2gr_hu:
  case Intrinsic::loongarch_lasx_xvpickve2gr_wu:
    replaceVPICKVE2GRResults<3>(N, Results, DAG, Subtarget,
                                LoongArchISD::VPICK_ZEXT_ELT);
    break;
  case Intrinsic::loongarch_lsx_vpickve2gr_wu:
    replaceVPICKVE2GRResults<2>(N, Results, DAG, Subtarget,
                                LoongArchISD::VPICK_ZEXT_ELT);
    break;
  case Intrinsic::loongarch_lsx_bz_b:
  case Intrinsic::loongarch_lsx_bz_h:
  case Intrinsic::loongarch_lsx_bz_w:
  case Intrinsic::loongarch_lsx_bz_d:
  case Intrinsic::loongarch_lasx_xbz_b:
  case Intrinsic::loongarch_lasx_xbz_h:
  case Intrinsic::loongarch_lasx_xbz_w:
  case Intrinsic::loongarch_lasx_xbz_d:
    replaceVecCondBranchResults(N, Results, DAG, Subtarget,
                                LoongArchISD::VALL_ZERO);
    break;
  case Intrinsic::loongarch_lsx_bz_v:
  case Intrinsic::loongarch_lasx_xbz_v:
    replaceVecCondBranchResults(N, Results, DAG, Subtarget,
                                LoongArchISD::VANY_ZERO);
    break;
  case Intrinsic::loongarch_lsx_bnz_b:
  case Intrinsic::loongarch_lsx_bnz_h:
  case Intrinsic::loongarch_lsx_bnz_w:
  case Intrinsic::loongarch_lsx_bnz_d:
  case Intrinsic::loongarch_lasx_xbnz_b:
  case Intrinsic::loongarch_lasx_xbnz_h:
  case Intrinsic::loongarch_lasx_xbnz_w:
  case Intrinsic::loongarch_lasx_xbnz_d:
    replaceVecCondBranchResults(N, Results, DAG, Subtarget,
                                LoongArchISD::VALL_NONZERO);
    break;
  case Intrinsic::loongarch_lsx_bnz_v:
  case Intrinsic::loongarch_lasx_xbnz_v:
    replaceVecCondBranchResults(N, Results, DAG, Subtarget,
                                LoongArchISD::VANY_NONZERO);
    break;
  }
}
 
void LoongArchTargetLowering::ReplaceNodeResults(
    SDNode *N, SmallVectorImpl<SDValue> &Results, SelectionDAG &DAG) const {
  SDLoc DL(N);
  EVT VT = N->getValueType(0);
  switch (N->getOpcode()) {
  default:
    llvm_unreachable("Don't know how to legalize this operation");
  case ISD::ADD:
  case ISD::SUB:
    assert(N->getValueType(0) == MVT::i32 && Subtarget.is64Bit() &&
           "Unexpected custom legalisation");
    Results.push_back(customLegalizeToWOpWithSExt(N, DAG));
    break;
  case ISD::UDIV:
  case ISD::UREM:
    assert(VT == MVT::i32 && Subtarget.is64Bit() &&
           "Unexpected custom legalisation");
    Results.push_back(customLegalizeToWOp(N, DAG, 2, ISD::SIGN_EXTEND));
    break;
  case ISD::SHL:
  case ISD::SRA:
  case ISD::SRL:
    assert(VT == MVT::i32 && Subtarget.is64Bit() &&
           "Unexpected custom legalisation");
    if (N->getOperand(1).getOpcode() != ISD::Constant) {
      Results.push_back(customLegalizeToWOp(N, DAG, 2));
      break;
    }
    break;
  case ISD::ROTL:
  case ISD::ROTR:
    assert(VT == MVT::i32 && Subtarget.is64Bit() &&
           "Unexpected custom legalisation");
    Results.push_back(customLegalizeToWOp(N, DAG, 2));
    break;
  case ISD::FP_TO_SINT: {
    assert(VT == MVT::i32 && Subtarget.is64Bit() &&
           "Unexpected custom legalisation");
    SDValue Src = N->getOperand(0);
    EVT FVT = EVT::getFloatingPointVT(N->getValueSizeInBits(0));
    if (getTypeAction(*DAG.getContext(), Src.getValueType()) !=
        TargetLowering::TypeSoftenFloat) {
      SDValue Dst = DAG.getNode(LoongArchISD::FTINT, DL, FVT, Src);
      Results.push_back(DAG.getNode(ISD::BITCAST, DL, VT, Dst));
      return;
    }
    // If the FP type needs to be softened, emit a library call using the 'si'
    // version. If we left it to default legalization we'd end up with 'di'.
    RTLIB::Libcall LC;
    LC = RTLIB::getFPTOSINT(Src.getValueType(), VT);
    MakeLibCallOptions CallOptions;
    EVT OpVT = Src.getValueType();
    CallOptions.setTypeListBeforeSoften(OpVT, VT, true);
    SDValue Chain = SDValue();
    SDValue Result;
    std::tie(Result, Chain) =
        makeLibCall(DAG, LC, VT, Src, CallOptions, DL, Chain);
    Results.push_back(Result);
    break;
  }
  case ISD::BITCAST: {
    SDValue Src = N->getOperand(0);
    EVT SrcVT = Src.getValueType();
    if (VT == MVT::i32 && SrcVT == MVT::f32 && Subtarget.is64Bit() &&
        Subtarget.hasBasicF()) {
      SDValue Dst =
          DAG.getNode(LoongArchISD::MOVFR2GR_S_LA64, DL, MVT::i64, Src);
      Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Dst));
    }
    break;
  }
  case ISD::FP_TO_UINT: {
    assert(VT == MVT::i32 && Subtarget.is64Bit() &&
           "Unexpected custom legalisation");
    auto &TLI = DAG.getTargetLoweringInfo();
    SDValue Tmp1, Tmp2;
    TLI.expandFP_TO_UINT(N, Tmp1, Tmp2, DAG);
    Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, MVT::i32, Tmp1));
    break;
  }
  case ISD::BSWAP: {
    SDValue Src = N->getOperand(0);
    assert((VT == MVT::i16 || VT == MVT::i32) &&
           "Unexpected custom legalization");
    MVT GRLenVT = Subtarget.getGRLenVT();
    SDValue NewSrc = DAG.getNode(ISD::ANY_EXTEND, DL, GRLenVT, Src);
    SDValue Tmp;
    switch (VT.getSizeInBits()) {
    default:
      llvm_unreachable("Unexpected operand width");
    case 16:
      Tmp = DAG.getNode(LoongArchISD::REVB_2H, DL, GRLenVT, NewSrc);
      break;
    case 32:
      // Only LA64 will get to here due to the size mismatch between VT and
      // GRLenVT, LA32 lowering is directly defined in LoongArchInstrInfo.
      Tmp = DAG.getNode(LoongArchISD::REVB_2W, DL, GRLenVT, NewSrc);
      break;
    }
    Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, VT, Tmp));
    break;
  }
  case ISD::BITREVERSE: {
    SDValue Src = N->getOperand(0);
    assert((VT == MVT::i8 || (VT == MVT::i32 && Subtarget.is64Bit())) &&
           "Unexpected custom legalization");
    MVT GRLenVT = Subtarget.getGRLenVT();
    SDValue NewSrc = DAG.getNode(ISD::ANY_EXTEND, DL, GRLenVT, Src);
    SDValue Tmp;
    switch (VT.getSizeInBits()) {
    default:
      llvm_unreachable("Unexpected operand width");
    case 8:
      Tmp = DAG.getNode(LoongArchISD::BITREV_4B, DL, GRLenVT, NewSrc);
      break;
    case 32:
      Tmp = DAG.getNode(LoongArchISD::BITREV_W, DL, GRLenVT, NewSrc);
      break;
    }
    Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, VT, Tmp));
    break;
  }
  case ISD::CTLZ:
  case ISD::CTTZ: {
    assert(VT == MVT::i32 && Subtarget.is64Bit() &&
           "Unexpected custom legalisation");
    Results.push_back(customLegalizeToWOp(N, DAG, 1));
    break;
  }
  case ISD::INTRINSIC_W_CHAIN: {
    SDValue Chain = N->getOperand(0);
    SDValue Op2 = N->getOperand(2);
    MVT GRLenVT = Subtarget.getGRLenVT();
    const StringRef ErrorMsgOOR = "argument out of range";
    const StringRef ErrorMsgReqLA64 = "requires loongarch64";
    const StringRef ErrorMsgReqF = "requires basic 'f' target feature";
 
    switch (N->getConstantOperandVal(1)) {
    default:
      llvm_unreachable("Unexpected Intrinsic.");
    case Intrinsic::loongarch_movfcsr2gr: {
      if (!Subtarget.hasBasicF()) {
        emitErrorAndReplaceIntrinsicResults(N, Results, DAG, ErrorMsgReqF);
        return;
      }
      unsigned Imm = Op2->getAsZExtVal();
      if (!isUInt<2>(Imm)) {
        emitErrorAndReplaceIntrinsicResults(N, Results, DAG, ErrorMsgOOR);
        return;
      }
      SDValue MOVFCSR2GRResults = DAG.getNode(
          LoongArchISD::MOVFCSR2GR, SDLoc(N), {MVT::i64, MVT::Other},
          {Chain, DAG.getConstant(Imm, DL, GRLenVT)});
      Results.push_back(
          DAG.getNode(ISD::TRUNCATE, DL, VT, MOVFCSR2GRResults.getValue(0)));
      Results.push_back(MOVFCSR2GRResults.getValue(1));
      break;
    }
#define CRC_CASE_EXT_BINARYOP(NAME, NODE)                                      \
  case Intrinsic::loongarch_##NAME: {                                          \
    SDValue NODE = DAG.getNode(                                                \
        LoongArchISD::NODE, DL, {MVT::i64, MVT::Other},                        \
        {Chain, DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op2),               \
         DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(3))});       \
    Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, VT, NODE.getValue(0)));   \
    Results.push_back(NODE.getValue(1));                                       \
    break;                                                                     \
  }
      CRC_CASE_EXT_BINARYOP(crc_w_b_w, CRC_W_B_W)
      CRC_CASE_EXT_BINARYOP(crc_w_h_w, CRC_W_H_W)
      CRC_CASE_EXT_BINARYOP(crc_w_w_w, CRC_W_W_W)
      CRC_CASE_EXT_BINARYOP(crcc_w_b_w, CRCC_W_B_W)
      CRC_CASE_EXT_BINARYOP(crcc_w_h_w, CRCC_W_H_W)
      CRC_CASE_EXT_BINARYOP(crcc_w_w_w, CRCC_W_W_W)
#undef CRC_CASE_EXT_BINARYOP
 
#define CRC_CASE_EXT_UNARYOP(NAME, NODE)                                       \
  case Intrinsic::loongarch_##NAME: {                                          \
    SDValue NODE = DAG.getNode(                                                \
        LoongArchISD::NODE, DL, {MVT::i64, MVT::Other},                        \
        {Chain, Op2,                                                           \
         DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(3))});       \
    Results.push_back(DAG.getNode(ISD::TRUNCATE, DL, VT, NODE.getValue(0)));   \
    Results.push_back(NODE.getValue(1));                                       \
    break;                                                                     \
  }
      CRC_CASE_EXT_UNARYOP(crc_w_d_w, CRC_W_D_W)
      CRC_CASE_EXT_UNARYOP(crcc_w_d_w, CRCC_W_D_W)
#undef CRC_CASE_EXT_UNARYOP
#define CSR_CASE(ID)                                                           \
  case Intrinsic::loongarch_##ID: {                                            \
    if (!Subtarget.is64Bit())                                                  \
      emitErrorAndReplaceIntrinsicResults(N, Results, DAG, ErrorMsgReqLA64);   \
    break;                                                                     \
  }
      CSR_CASE(csrrd_d);
      CSR_CASE(csrwr_d);
      CSR_CASE(csrxchg_d);
      CSR_CASE(iocsrrd_d);
#undef CSR_CASE
    case Intrinsic::loongarch_csrrd_w: {
      unsigned Imm = Op2->getAsZExtVal();
      if (!isUInt<14>(Imm)) {
        emitErrorAndReplaceIntrinsicResults(N, Results, DAG, ErrorMsgOOR);
        return;
      }
      SDValue CSRRDResults =
          DAG.getNode(LoongArchISD::CSRRD, DL, {GRLenVT, MVT::Other},
                      {Chain, DAG.getConstant(Imm, DL, GRLenVT)});
      Results.push_back(
          DAG.getNode(ISD::TRUNCATE, DL, VT, CSRRDResults.getValue(0)));
      Results.push_back(CSRRDResults.getValue(1));
      break;
    }
    case Intrinsic::loongarch_csrwr_w: {
      unsigned Imm = N->getConstantOperandVal(3);
      if (!isUInt<14>(Imm)) {
        emitErrorAndReplaceIntrinsicResults(N, Results, DAG, ErrorMsgOOR);
        return;
      }
      SDValue CSRWRResults =
          DAG.getNode(LoongArchISD::CSRWR, DL, {GRLenVT, MVT::Other},
                      {Chain, DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op2),
                       DAG.getConstant(Imm, DL, GRLenVT)});
      Results.push_back(
          DAG.getNode(ISD::TRUNCATE, DL, VT, CSRWRResults.getValue(0)));
      Results.push_back(CSRWRResults.getValue(1));
      break;
    }
    case Intrinsic::loongarch_csrxchg_w: {
      unsigned Imm = N->getConstantOperandVal(4);
      if (!isUInt<14>(Imm)) {
        emitErrorAndReplaceIntrinsicResults(N, Results, DAG, ErrorMsgOOR);
        return;
      }
      SDValue CSRXCHGResults = DAG.getNode(
          LoongArchISD::CSRXCHG, DL, {GRLenVT, MVT::Other},
          {Chain, DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op2),
           DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, N->getOperand(3)),
           DAG.getConstant(Imm, DL, GRLenVT)});
      Results.push_back(
          DAG.getNode(ISD::TRUNCATE, DL, VT, CSRXCHGResults.getValue(0)));
      Results.push_back(CSRXCHGResults.getValue(1));
      break;
    }
#define IOCSRRD_CASE(NAME, NODE)                                               \
  case Intrinsic::loongarch_##NAME: {                                          \
    SDValue IOCSRRDResults =                                                   \
        DAG.getNode(LoongArchISD::NODE, DL, {MVT::i64, MVT::Other},            \
                    {Chain, DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op2)}); \
    Results.push_back(                                                         \
        DAG.getNode(ISD::TRUNCATE, DL, VT, IOCSRRDResults.getValue(0)));       \
    Results.push_back(IOCSRRDResults.getValue(1));                             \
    break;                                                                     \
  }
      IOCSRRD_CASE(iocsrrd_b, IOCSRRD_B);
      IOCSRRD_CASE(iocsrrd_h, IOCSRRD_H);
      IOCSRRD_CASE(iocsrrd_w, IOCSRRD_W);
#undef IOCSRRD_CASE
    case Intrinsic::loongarch_cpucfg: {
      SDValue CPUCFGResults =
          DAG.getNode(LoongArchISD::CPUCFG, DL, {GRLenVT, MVT::Other},
                      {Chain, DAG.getNode(ISD::ANY_EXTEND, DL, MVT::i64, Op2)});
      Results.push_back(
          DAG.getNode(ISD::TRUNCATE, DL, VT, CPUCFGResults.getValue(0)));
      Results.push_back(CPUCFGResults.getValue(1));
      break;
    }
    case Intrinsic::loongarch_lddir_d: {
      if (!Subtarget.is64Bit()) {
        emitErrorAndReplaceIntrinsicResults(N, Results, DAG, ErrorMsgReqLA64);
        return;
      }
      break;
    }
    }
    break;
  }
  case ISD::READ_REGISTER: {
    if (Subtarget.is64Bit())
      DAG.getContext()->emitError(
          "On LA64, only 64-bit registers can be read.");
    else
      DAG.getContext()->emitError(
          "On LA32, only 32-bit registers can be read.");
    Results.push_back(DAG.getUNDEF(VT));
    Results.push_back(N->getOperand(0));
    break;
  }
  case ISD::INTRINSIC_WO_CHAIN: {
    replaceINTRINSIC_WO_CHAINResults(N, Results, DAG, Subtarget);
    break;
  }
  }
}
 
static SDValue performANDCombine(SDNode *N, SelectionDAG &DAG,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const LoongArchSubtarget &Subtarget) {
  if (DCI.isBeforeLegalizeOps())
    return SDValue();
 
  SDValue FirstOperand = N->getOperand(0);
  SDValue SecondOperand = N->getOperand(1);
  unsigned FirstOperandOpc = FirstOperand.getOpcode();
  EVT ValTy = N->getValueType(0);
  SDLoc DL(N);
  uint64_t lsb, msb;
  unsigned SMIdx, SMLen;
  ConstantSDNode *CN;
  SDValue NewOperand;
  MVT GRLenVT = Subtarget.getGRLenVT();
 
  // Op's second operand must be a shifted mask.
  if (!(CN = dyn_cast<ConstantSDNode>(SecondOperand)) ||
      !isShiftedMask_64(CN->getZExtValue(), SMIdx, SMLen))
    return SDValue();
 
  if (FirstOperandOpc == ISD::SRA || FirstOperandOpc == ISD::SRL) {
    // Pattern match BSTRPICK.
    //  $dst = and ((sra or srl) $src , lsb), (2**len - 1)
    //  => BSTRPICK $dst, $src, msb, lsb
    //  where msb = lsb + len - 1
 
    // The second operand of the shift must be an immediate.
    if (!(CN = dyn_cast<ConstantSDNode>(FirstOperand.getOperand(1))))
      return SDValue();
 
    lsb = CN->getZExtValue();
 
    // Return if the shifted mask does not start at bit 0 or the sum of its
    // length and lsb exceeds the word's size.
    if (SMIdx != 0 || lsb + SMLen > ValTy.getSizeInBits())
      return SDValue();
 
    NewOperand = FirstOperand.getOperand(0);
  } else {
    // Pattern match BSTRPICK.
    //  $dst = and $src, (2**len- 1) , if len > 12
    //  => BSTRPICK $dst, $src, msb, lsb
    //  where lsb = 0 and msb = len - 1
 
    // If the mask is <= 0xfff, andi can be used instead.
    if (CN->getZExtValue() <= 0xfff)
      return SDValue();
 
    // Return if the MSB exceeds.
    if (SMIdx + SMLen > ValTy.getSizeInBits())
      return SDValue();
 
    if (SMIdx > 0) {
      // Omit if the constant has more than 2 uses. This a conservative
      // decision. Whether it is a win depends on the HW microarchitecture.
      // However it should always be better for 1 and 2 uses.
      if (CN->use_size() > 2)
        return SDValue();
      // Return if the constant can be composed by a single LU12I.W.
      if ((CN->getZExtValue() & 0xfff) == 0)
        return SDValue();
      // Return if the constand can be composed by a single ADDI with
      // the zero register.
      if (CN->getSExtValue() >= -2048 && CN->getSExtValue() < 0)
        return SDValue();
    }
 
    lsb = SMIdx;
    NewOperand = FirstOperand;
  }
 
  msb = lsb + SMLen - 1;
  SDValue NR0 = DAG.getNode(LoongArchISD::BSTRPICK, DL, ValTy, NewOperand,
                            DAG.getConstant(msb, DL, GRLenVT),
                            DAG.getConstant(lsb, DL, GRLenVT));
  if (FirstOperandOpc == ISD::SRA || FirstOperandOpc == ISD::SRL || lsb == 0)
    return NR0;
  // Try to optimize to
  //   bstrpick $Rd, $Rs, msb, lsb
  //   slli     $Rd, $Rd, lsb
  return DAG.getNode(ISD::SHL, DL, ValTy, NR0,
                     DAG.getConstant(lsb, DL, GRLenVT));
}
 
static SDValue performSRLCombine(SDNode *N, SelectionDAG &DAG,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const LoongArchSubtarget &Subtarget) {
  if (DCI.isBeforeLegalizeOps())
    return SDValue();
 
  // $dst = srl (and $src, Mask), Shamt
  // =>
  // BSTRPICK $dst, $src, MaskIdx+MaskLen-1, Shamt
  // when Mask is a shifted mask, and MaskIdx <= Shamt <= MaskIdx+MaskLen-1
  //
 
  SDValue FirstOperand = N->getOperand(0);
  ConstantSDNode *CN;
  EVT ValTy = N->getValueType(0);
  SDLoc DL(N);
  MVT GRLenVT = Subtarget.getGRLenVT();
  unsigned MaskIdx, MaskLen;
  uint64_t Shamt;
 
  // The first operand must be an AND and the second operand of the AND must be
  // a shifted mask.
  if (FirstOperand.getOpcode() != ISD::AND ||
      !(CN = dyn_cast<ConstantSDNode>(FirstOperand.getOperand(1))) ||
      !isShiftedMask_64(CN->getZExtValue(), MaskIdx, MaskLen))
    return SDValue();
 
  // The second operand (shift amount) must be an immediate.
  if (!(CN = dyn_cast<ConstantSDNode>(N->getOperand(1))))
    return SDValue();
 
  Shamt = CN->getZExtValue();
  if (MaskIdx <= Shamt && Shamt <= MaskIdx + MaskLen - 1)
    return DAG.getNode(LoongArchISD::BSTRPICK, DL, ValTy,
                       FirstOperand->getOperand(0),
                       DAG.getConstant(MaskIdx + MaskLen - 1, DL, GRLenVT),
                       DAG.getConstant(Shamt, DL, GRLenVT));
 
  return SDValue();
}
 
static SDValue performORCombine(SDNode *N, SelectionDAG &DAG,
                                TargetLowering::DAGCombinerInfo &DCI,
                                const LoongArchSubtarget &Subtarget) {
  MVT GRLenVT = Subtarget.getGRLenVT();
  EVT ValTy = N->getValueType(0);
  SDValue N0 = N->getOperand(0), N1 = N->getOperand(1);
  ConstantSDNode *CN0, *CN1;
  SDLoc DL(N);
  unsigned ValBits = ValTy.getSizeInBits();
  unsigned MaskIdx0, MaskLen0, MaskIdx1, MaskLen1;
  unsigned Shamt;
  bool SwapAndRetried = false;
 
  if (DCI.isBeforeLegalizeOps())
    return SDValue();
 
  if (ValBits != 32 && ValBits != 64)
    return SDValue();
 
Retry:
  // 1st pattern to match BSTRINS:
  //  R = or (and X, mask0), (and (shl Y, lsb), mask1)
  //  where mask1 = (2**size - 1) << lsb, mask0 = ~mask1
  //  =>
  //  R = BSTRINS X, Y, msb, lsb (where msb = lsb + size - 1)
  if (N0.getOpcode() == ISD::AND &&
      (CN0 = dyn_cast<ConstantSDNode>(N0.getOperand(1))) &&
      isShiftedMask_64(~CN0->getSExtValue(), MaskIdx0, MaskLen0) &&
      N1.getOpcode() == ISD::AND && N1.getOperand(0).getOpcode() == ISD::SHL &&
      (CN1 = dyn_cast<ConstantSDNode>(N1.getOperand(1))) &&
      isShiftedMask_64(CN1->getZExtValue(), MaskIdx1, MaskLen1) &&
      MaskIdx0 == MaskIdx1 && MaskLen0 == MaskLen1 &&
      (CN1 = dyn_cast<ConstantSDNode>(N1.getOperand(0).getOperand(1))) &&
      (Shamt = CN1->getZExtValue()) == MaskIdx0 &&
      (MaskIdx0 + MaskLen0 <= ValBits)) {
    LLVM_DEBUG(dbgs() << "Perform OR combine: match pattern 1\n");
    return DAG.getNode(LoongArchISD::BSTRINS, DL, ValTy, N0.getOperand(0),
                       N1.getOperand(0).getOperand(0),
                       DAG.getConstant((MaskIdx0 + MaskLen0 - 1), DL, GRLenVT),
                       DAG.getConstant(MaskIdx0, DL, GRLenVT));
  }
 
  // 2nd pattern to match BSTRINS:
  //  R = or (and X, mask0), (shl (and Y, mask1), lsb)
  //  where mask1 = (2**size - 1), mask0 = ~(mask1 << lsb)
  //  =>
  //  R = BSTRINS X, Y, msb, lsb (where msb = lsb + size - 1)
  if (N0.getOpcode() == ISD::AND &&
      (CN0 = dyn_cast<ConstantSDNode>(N0.getOperand(1))) &&
      isShiftedMask_64(~CN0->getSExtValue(), MaskIdx0, MaskLen0) &&
      N1.getOpcode() == ISD::SHL && N1.getOperand(0).getOpcode() == ISD::AND &&
      (CN1 = dyn_cast<ConstantSDNode>(N1.getOperand(1))) &&
      (Shamt = CN1->getZExtValue()) == MaskIdx0 &&
      (CN1 = dyn_cast<ConstantSDNode>(N1.getOperand(0).getOperand(1))) &&
      isShiftedMask_64(CN1->getZExtValue(), MaskIdx1, MaskLen1) &&
      MaskLen0 == MaskLen1 && MaskIdx1 == 0 &&
      (MaskIdx0 + MaskLen0 <= ValBits)) {
    LLVM_DEBUG(dbgs() << "Perform OR combine: match pattern 2\n");
    return DAG.getNode(LoongArchISD::BSTRINS, DL, ValTy, N0.getOperand(0),
                       N1.getOperand(0).getOperand(0),
                       DAG.getConstant((MaskIdx0 + MaskLen0 - 1), DL, GRLenVT),
                       DAG.getConstant(MaskIdx0, DL, GRLenVT));
  }
 
  // 3rd pattern to match BSTRINS:
  //  R = or (and X, mask0), (and Y, mask1)
  //  where ~mask0 = (2**size - 1) << lsb, mask0 & mask1 = 0
  //  =>
  //  R = BSTRINS X, (shr (and Y, mask1), lsb), msb, lsb
  //  where msb = lsb + size - 1
  if (N0.getOpcode() == ISD::AND && N1.getOpcode() == ISD::AND &&
      (CN0 = dyn_cast<ConstantSDNode>(N0.getOperand(1))) &&
      isShiftedMask_64(~CN0->getSExtValue(), MaskIdx0, MaskLen0) &&
      (MaskIdx0 + MaskLen0 <= 64) &&
      (CN1 = dyn_cast<ConstantSDNode>(N1->getOperand(1))) &&
      (CN1->getSExtValue() & CN0->getSExtValue()) == 0) {
    LLVM_DEBUG(dbgs() << "Perform OR combine: match pattern 3\n");
    return DAG.getNode(LoongArchISD::BSTRINS, DL, ValTy, N0.getOperand(0),
                       DAG.getNode(ISD::SRL, DL, N1->getValueType(0), N1,
                                   DAG.getConstant(MaskIdx0, DL, GRLenVT)),
                       DAG.getConstant(ValBits == 32
                                           ? (MaskIdx0 + (MaskLen0 & 31) - 1)
                                           : (MaskIdx0 + MaskLen0 - 1),
                                       DL, GRLenVT),
                       DAG.getConstant(MaskIdx0, DL, GRLenVT));
  }
 
  // 4th pattern to match BSTRINS:
  //  R = or (and X, mask), (shl Y, shamt)
  //  where mask = (2**shamt - 1)
  //  =>
  //  R = BSTRINS X, Y, ValBits - 1, shamt
  //  where ValBits = 32 or 64
  if (N0.getOpcode() == ISD::AND && N1.getOpcode() == ISD::SHL &&
      (CN0 = dyn_cast<ConstantSDNode>(N0.getOperand(1))) &&
      isShiftedMask_64(CN0->getZExtValue(), MaskIdx0, MaskLen0) &&
      MaskIdx0 == 0 && (CN1 = dyn_cast<ConstantSDNode>(N1.getOperand(1))) &&
      (Shamt = CN1->getZExtValue()) == MaskLen0 &&
      (MaskIdx0 + MaskLen0 <= ValBits)) {
    LLVM_DEBUG(dbgs() << "Perform OR combine: match pattern 4\n");
    return DAG.getNode(LoongArchISD::BSTRINS, DL, ValTy, N0.getOperand(0),
                       N1.getOperand(0),
                       DAG.getConstant((ValBits - 1), DL, GRLenVT),
                       DAG.getConstant(Shamt, DL, GRLenVT));
  }
 
  // 5th pattern to match BSTRINS:
  //  R = or (and X, mask), const
  //  where ~mask = (2**size - 1) << lsb, mask & const = 0
  //  =>
  //  R = BSTRINS X, (const >> lsb), msb, lsb
  //  where msb = lsb + size - 1
  if (N0.getOpcode() == ISD::AND &&
      (CN0 = dyn_cast<ConstantSDNode>(N0.getOperand(1))) &&
      isShiftedMask_64(~CN0->getSExtValue(), MaskIdx0, MaskLen0) &&
      (CN1 = dyn_cast<ConstantSDNode>(N1)) &&
      (CN1->getSExtValue() & CN0->getSExtValue()) == 0) {
    LLVM_DEBUG(dbgs() << "Perform OR combine: match pattern 5\n");
    return DAG.getNode(
        LoongArchISD::BSTRINS, DL, ValTy, N0.getOperand(0),
        DAG.getConstant(CN1->getSExtValue() >> MaskIdx0, DL, ValTy),
        DAG.getConstant(ValBits == 32 ? (MaskIdx0 + (MaskLen0 & 31) - 1)
                                      : (MaskIdx0 + MaskLen0 - 1),
                        DL, GRLenVT),
        DAG.getConstant(MaskIdx0, DL, GRLenVT));
  }
 
  // 6th pattern.
  // a = b | ((c & mask) << shamt), where all positions in b to be overwritten
  // by the incoming bits are known to be zero.
  // =>
  // a = BSTRINS b, c, shamt + MaskLen - 1, shamt
  //
  // Note that the 1st pattern is a special situation of the 6th, i.e. the 6th
  // pattern is more common than the 1st. So we put the 1st before the 6th in
  // order to match as many nodes as possible.
  ConstantSDNode *CNMask, *CNShamt;
  unsigned MaskIdx, MaskLen;
  if (N1.getOpcode() == ISD::SHL && N1.getOperand(0).getOpcode() == ISD::AND &&
      (CNMask = dyn_cast<ConstantSDNode>(N1.getOperand(0).getOperand(1))) &&
      isShiftedMask_64(CNMask->getZExtValue(), MaskIdx, MaskLen) &&
      MaskIdx == 0 && (CNShamt = dyn_cast<ConstantSDNode>(N1.getOperand(1))) &&
      CNShamt->getZExtValue() + MaskLen <= ValBits) {
    Shamt = CNShamt->getZExtValue();
    APInt ShMask(ValBits, CNMask->getZExtValue() << Shamt);
    if (ShMask.isSubsetOf(DAG.computeKnownBits(N0).Zero)) {
      LLVM_DEBUG(dbgs() << "Perform OR combine: match pattern 6\n");
      return DAG.getNode(LoongArchISD::BSTRINS, DL, ValTy, N0,
                         N1.getOperand(0).getOperand(0),
                         DAG.getConstant(Shamt + MaskLen - 1, DL, GRLenVT),
                         DAG.getConstant(Shamt, DL, GRLenVT));
    }
  }
 
  // 7th pattern.
  // a = b | ((c << shamt) & shifted_mask), where all positions in b to be
  // overwritten by the incoming bits are known to be zero.
  // =>
  // a = BSTRINS b, c, MaskIdx + MaskLen - 1, MaskIdx
  //
  // Similarly, the 7th pattern is more common than the 2nd. So we put the 2nd
  // before the 7th in order to match as many nodes as possible.
  if (N1.getOpcode() == ISD::AND &&
      (CNMask = dyn_cast<ConstantSDNode>(N1.getOperand(1))) &&
      isShiftedMask_64(CNMask->getZExtValue(), MaskIdx, MaskLen) &&
      N1.getOperand(0).getOpcode() == ISD::SHL &&
      (CNShamt = dyn_cast<ConstantSDNode>(N1.getOperand(0).getOperand(1))) &&
      CNShamt->getZExtValue() == MaskIdx) {
    APInt ShMask(ValBits, CNMask->getZExtValue());
    if (ShMask.isSubsetOf(DAG.computeKnownBits(N0).Zero)) {
      LLVM_DEBUG(dbgs() << "Perform OR combine: match pattern 7\n");
      return DAG.getNode(LoongArchISD::BSTRINS, DL, ValTy, N0,
                         N1.getOperand(0).getOperand(0),
                         DAG.getConstant(MaskIdx + MaskLen - 1, DL, GRLenVT),
                         DAG.getConstant(MaskIdx, DL, GRLenVT));
    }
  }
 
  // (or a, b) and (or b, a) are equivalent, so swap the operands and retry.
  if (!SwapAndRetried) {
    std::swap(N0, N1);
    SwapAndRetried = true;
    goto Retry;
  }
 
  SwapAndRetried = false;
Retry2:
  // 8th pattern.
  // a = b | (c & shifted_mask), where all positions in b to be overwritten by
  // the incoming bits are known to be zero.
  // =>
  // a = BSTRINS b, c >> MaskIdx, MaskIdx + MaskLen - 1, MaskIdx
  //
  // Similarly, the 8th pattern is more common than the 4th and 5th patterns. So
  // we put it here in order to match as many nodes as possible or generate less
  // instructions.
  if (N1.getOpcode() == ISD::AND &&
      (CNMask = dyn_cast<ConstantSDNode>(N1.getOperand(1))) &&
      isShiftedMask_64(CNMask->getZExtValue(), MaskIdx, MaskLen)) {
    APInt ShMask(ValBits, CNMask->getZExtValue());
    if (ShMask.isSubsetOf(DAG.computeKnownBits(N0).Zero)) {
      LLVM_DEBUG(dbgs() << "Perform OR combine: match pattern 8\n");
      return DAG.getNode(LoongArchISD::BSTRINS, DL, ValTy, N0,
                         DAG.getNode(ISD::SRL, DL, N1->getValueType(0),
                                     N1->getOperand(0),
                                     DAG.getConstant(MaskIdx, DL, GRLenVT)),
                         DAG.getConstant(MaskIdx + MaskLen - 1, DL, GRLenVT),
                         DAG.getConstant(MaskIdx, DL, GRLenVT));
    }
  }
  // Swap N0/N1 and retry.
  if (!SwapAndRetried) {
    std::swap(N0, N1);
    SwapAndRetried = true;
    goto Retry2;
  }
 
  return SDValue();
}
 
static bool checkValueWidth(SDValue V, ISD::LoadExtType &ExtType) {
  ExtType = ISD::NON_EXTLOAD;
 
  switch (V.getNode()->getOpcode()) {
  case ISD::LOAD: {
    LoadSDNode *LoadNode = cast<LoadSDNode>(V.getNode());
    if ((LoadNode->getMemoryVT() == MVT::i8) ||
        (LoadNode->getMemoryVT() == MVT::i16)) {
      ExtType = LoadNode->getExtensionType();
      return true;
    }
    return false;
  }
  case ISD::AssertSext: {
    VTSDNode *TypeNode = cast<VTSDNode>(V.getNode()->getOperand(1));
    if ((TypeNode->getVT() == MVT::i8) || (TypeNode->getVT() == MVT::i16)) {
      ExtType = ISD::SEXTLOAD;
      return true;
    }
    return false;
  }
  case ISD::AssertZext: {
    VTSDNode *TypeNode = cast<VTSDNode>(V.getNode()->getOperand(1));
    if ((TypeNode->getVT() == MVT::i8) || (TypeNode->getVT() == MVT::i16)) {
      ExtType = ISD::ZEXTLOAD;
      return true;
    }
    return false;
  }
  default:
    return false;
  }
 
  return false;
}
 
// Eliminate redundant truncation and zero-extension nodes.
// * Case 1:
//  +------------+ +------------+ +------------+
//  |   Input1   | |   Input2   | |     CC     |
//  +------------+ +------------+ +------------+
//         |              |              |
//         V              V              +----+
//  +------------+ +------------+             |
//  |  TRUNCATE  | |  TRUNCATE  |             |
//  +------------+ +------------+             |
//         |              |                   |
//         V              V                   |
//  +------------+ +------------+             |
//  |  ZERO_EXT  | |  ZERO_EXT  |             |
//  +------------+ +------------+             |
//         |              |                   |
//         |              +-------------+     |
//         V              V             |     |
//        +----------------+            |     |
//        |      AND       |            |     |
//        +----------------+            |     |
//                |                     |     |
//                +---------------+     |     |
//                                |     |     |
//                                V     V     V
//                               +-------------+
//                               |     CMP     |
//                               +-------------+
// * Case 2:
//  +------------+ +------------+ +-------------+ +------------+ +------------+
//  |   Input1   | |   Input2   | | Constant -1 | | Constant 0 | |     CC     |
//  +------------+ +------------+ +-------------+ +------------+ +------------+
//         |              |             |               |               |
//         V              |             |               |               |
//  +------------+        |             |               |               |
//  |     XOR    |<---------------------+               |               |
//  +------------+        |                             |               |
//         |              |                             |               |
//         V              V             +---------------+               |
//  +------------+ +------------+       |                               |
//  |  TRUNCATE  | |  TRUNCATE  |       |     +-------------------------+
//  +------------+ +------------+       |     |
//         |              |             |     |
//         V              V             |     |
//  +------------+ +------------+       |     |
//  |  ZERO_EXT  | |  ZERO_EXT  |       |     |
//  +------------+ +------------+       |     |
//         |              |             |     |
//         V              V             |     |
//        +----------------+            |     |
//        |      AND       |            |     |
//        +----------------+            |     |
//                |                     |     |
//                +---------------+     |     |
//                                |     |     |
//                                V     V     V
//                               +-------------+
//                               |     CMP     |
//                               +-------------+
static SDValue performSETCCCombine(SDNode *N, SelectionDAG &DAG,
                                   TargetLowering::DAGCombinerInfo &DCI,
                                   const LoongArchSubtarget &Subtarget) {
  ISD::CondCode CC = cast<CondCodeSDNode>(N->getOperand(2))->get();
 
  SDNode *AndNode = N->getOperand(0).getNode();
  if (AndNode->getOpcode() != ISD::AND)
    return SDValue();
 
  SDValue AndInputValue2 = AndNode->getOperand(1);
  if (AndInputValue2.getOpcode() != ISD::ZERO_EXTEND)
    return SDValue();
 
  SDValue CmpInputValue = N->getOperand(1);
  SDValue AndInputValue1 = AndNode->getOperand(0);
  if (AndInputValue1.getOpcode() == ISD::XOR) {
    if (CC != ISD::SETEQ && CC != ISD::SETNE)
      return SDValue();
    ConstantSDNode *CN = dyn_cast<ConstantSDNode>(AndInputValue1.getOperand(1));
    if (!CN || CN->getSExtValue() != -1)
      return SDValue();
    CN = dyn_cast<ConstantSDNode>(CmpInputValue);
    if (!CN || CN->getSExtValue() != 0)
      return SDValue();
    AndInputValue1 = AndInputValue1.getOperand(0);
    if (AndInputValue1.getOpcode() != ISD::ZERO_EXTEND)
      return SDValue();
  } else if (AndInputValue1.getOpcode() == ISD::ZERO_EXTEND) {
    if (AndInputValue2 != CmpInputValue)
      return SDValue();
  } else {
    return SDValue();
  }
 
  SDValue TruncValue1 = AndInputValue1.getNode()->getOperand(0);
  if (TruncValue1.getOpcode() != ISD::TRUNCATE)
    return SDValue();
 
  SDValue TruncValue2 = AndInputValue2.getNode()->getOperand(0);
  if (TruncValue2.getOpcode() != ISD::TRUNCATE)
    return SDValue();
 
  SDValue TruncInputValue1 = TruncValue1.getNode()->getOperand(0);
  SDValue TruncInputValue2 = TruncValue2.getNode()->getOperand(0);
  ISD::LoadExtType ExtType1;
  ISD::LoadExtType ExtType2;
 
  if (!checkValueWidth(TruncInputValue1, ExtType1) ||
      !checkValueWidth(TruncInputValue2, ExtType2))
    return SDValue();
 
  if (TruncInputValue1->getValueType(0) != TruncInputValue2->getValueType(0) ||
      AndNode->getValueType(0) != TruncInputValue1->getValueType(0))
    return SDValue();
 
  if ((ExtType2 != ISD::ZEXTLOAD) &&
      ((ExtType2 != ISD::SEXTLOAD) && (ExtType1 != ISD::SEXTLOAD)))
    return SDValue();
 
  // These truncation and zero-extension nodes are not necessary, remove them.
  SDValue NewAnd = DAG.getNode(ISD::AND, SDLoc(N), AndNode->getValueType(0),
                               TruncInputValue1, TruncInputValue2);
  SDValue NewSetCC =
      DAG.getSetCC(SDLoc(N), N->getValueType(0), NewAnd, TruncInputValue2, CC);
  DAG.ReplaceAllUsesWith(N, NewSetCC.getNode());
  return SDValue(N, 0);
}
 
// Combine (loongarch_bitrev_w (loongarch_revb_2w X)) to loongarch_bitrev_4b.
static SDValue performBITREV_WCombine(SDNode *N, SelectionDAG &DAG,
                                      TargetLowering::DAGCombinerInfo &DCI,
                                      const LoongArchSubtarget &Subtarget) {
  if (DCI.isBeforeLegalizeOps())
    return SDValue();
 
  SDValue Src = N->getOperand(0);
  if (Src.getOpcode() != LoongArchISD::REVB_2W)
    return SDValue();
 
  return DAG.getNode(LoongArchISD::BITREV_4B, SDLoc(N), N->getValueType(0),
                     Src.getOperand(0));
}
 
template <unsigned N>
static SDValue legalizeIntrinsicImmArg(SDNode *Node, unsigned ImmOp,
                                       SelectionDAG &DAG,
                                       const LoongArchSubtarget &Subtarget,
                                       bool IsSigned = false) {
  SDLoc DL(Node);
  auto *CImm = cast<ConstantSDNode>(Node->getOperand(ImmOp));
  // Check the ImmArg.
  if ((IsSigned && !isInt<N>(CImm->getSExtValue())) ||
      (!IsSigned && !isUInt<N>(CImm->getZExtValue()))) {
    DAG.getContext()->emitError(Node->getOperationName(0) +
                                ": argument out of range.");
    return DAG.getNode(ISD::UNDEF, DL, Subtarget.getGRLenVT());
  }
  return DAG.getConstant(CImm->getZExtValue(), DL, Subtarget.getGRLenVT());
}
 
template <unsigned N>
static SDValue lowerVectorSplatImm(SDNode *Node, unsigned ImmOp,
                                   SelectionDAG &DAG, bool IsSigned = false) {
  SDLoc DL(Node);
  EVT ResTy = Node->getValueType(0);
  auto *CImm = cast<ConstantSDNode>(Node->getOperand(ImmOp));
 
  // Check the ImmArg.
  if ((IsSigned && !isInt<N>(CImm->getSExtValue())) ||
      (!IsSigned && !isUInt<N>(CImm->getZExtValue()))) {
    DAG.getContext()->emitError(Node->getOperationName(0) +
                                ": argument out of range.");
    return DAG.getNode(ISD::UNDEF, DL, ResTy);
  }
  return DAG.getConstant(
      APInt(ResTy.getScalarType().getSizeInBits(),
            IsSigned ? CImm->getSExtValue() : CImm->getZExtValue(), IsSigned),
      DL, ResTy);
}
 
static SDValue truncateVecElts(SDNode *Node, SelectionDAG &DAG) {
  SDLoc DL(Node);
  EVT ResTy = Node->getValueType(0);
  SDValue Vec = Node->getOperand(2);
  SDValue Mask = DAG.getConstant(Vec.getScalarValueSizeInBits() - 1, DL, ResTy);
  return DAG.getNode(ISD::AND, DL, ResTy, Vec, Mask);
}
 
static SDValue lowerVectorBitClear(SDNode *Node, SelectionDAG &DAG) {
  SDLoc DL(Node);
  EVT ResTy = Node->getValueType(0);
  SDValue One = DAG.getConstant(1, DL, ResTy);
  SDValue Bit =
      DAG.getNode(ISD::SHL, DL, ResTy, One, truncateVecElts(Node, DAG));
 
  return DAG.getNode(ISD::AND, DL, ResTy, Node->getOperand(1),
                     DAG.getNOT(DL, Bit, ResTy));
}
 
template <unsigned N>
static SDValue lowerVectorBitClearImm(SDNode *Node, SelectionDAG &DAG) {
  SDLoc DL(Node);
  EVT ResTy = Node->getValueType(0);
  auto *CImm = cast<ConstantSDNode>(Node->getOperand(2));
  // Check the unsigned ImmArg.
  if (!isUInt<N>(CImm->getZExtValue())) {
    DAG.getContext()->emitError(Node->getOperationName(0) +
                                ": argument out of range.");
    return DAG.getNode(ISD::UNDEF, DL, ResTy);
  }
 
  APInt BitImm = APInt(ResTy.getScalarSizeInBits(), 1) << CImm->getAPIntValue();
  SDValue Mask = DAG.getConstant(~BitImm, DL, ResTy);
 
  return DAG.getNode(ISD::AND, DL, ResTy, Node->getOperand(1), Mask);
}
 
template <unsigned N>
static SDValue lowerVectorBitSetImm(SDNode *Node, SelectionDAG &DAG) {
  SDLoc DL(Node);
  EVT ResTy = Node->getValueType(0);
  auto *CImm = cast<ConstantSDNode>(Node->getOperand(2));
  // Check the unsigned ImmArg.
  if (!isUInt<N>(CImm->getZExtValue())) {
    DAG.getContext()->emitError(Node->getOperationName(0) +
                                ": argument out of range.");
    return DAG.getNode(ISD::UNDEF, DL, ResTy);
  }
 
  APInt Imm = APInt(ResTy.getScalarSizeInBits(), 1) << CImm->getAPIntValue();
  SDValue BitImm = DAG.getConstant(Imm, DL, ResTy);
  return DAG.getNode(ISD::OR, DL, ResTy, Node->getOperand(1), BitImm);
}
 
template <unsigned N>
static SDValue lowerVectorBitRevImm(SDNode *Node, SelectionDAG &DAG) {
  SDLoc DL(Node);
  EVT ResTy = Node->getValueType(0);
  auto *CImm = cast<ConstantSDNode>(Node->getOperand(2));
  // Check the unsigned ImmArg.
  if (!isUInt<N>(CImm->getZExtValue())) {
    DAG.getContext()->emitError(Node->getOperationName(0) +
                                ": argument out of range.");
    return DAG.getNode(ISD::UNDEF, DL, ResTy);
  }
 
  APInt Imm = APInt(ResTy.getScalarSizeInBits(), 1) << CImm->getAPIntValue();
  SDValue BitImm = DAG.getConstant(Imm, DL, ResTy);
  return DAG.getNode(ISD::XOR, DL, ResTy, Node->getOperand(1), BitImm);
}
 
static SDValue
performINTRINSIC_WO_CHAINCombine(SDNode *N, SelectionDAG &DAG,
                                 TargetLowering::DAGCombinerInfo &DCI,
                                 const LoongArchSubtarget &Subtarget) {
  SDLoc DL(N);
  switch (N->getConstantOperandVal(0)) {
  default:
    break;
  case Intrinsic::loongarch_lsx_vadd_b:
  case Intrinsic::loongarch_lsx_vadd_h:
  case Intrinsic::loongarch_lsx_vadd_w:
  case Intrinsic::loongarch_lsx_vadd_d:
  case Intrinsic::loongarch_lasx_xvadd_b:
  case Intrinsic::loongarch_lasx_xvadd_h:
  case Intrinsic::loongarch_lasx_xvadd_w:
  case Intrinsic::loongarch_lasx_xvadd_d:
    return DAG.getNode(ISD::ADD, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vaddi_bu:
  case Intrinsic::loongarch_lsx_vaddi_hu:
  case Intrinsic::loongarch_lsx_vaddi_wu:
  case Intrinsic::loongarch_lsx_vaddi_du:
  case Intrinsic::loongarch_lasx_xvaddi_bu:
  case Intrinsic::loongarch_lasx_xvaddi_hu:
  case Intrinsic::loongarch_lasx_xvaddi_wu:
  case Intrinsic::loongarch_lasx_xvaddi_du:
    return DAG.getNode(ISD::ADD, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<5>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vsub_b:
  case Intrinsic::loongarch_lsx_vsub_h:
  case Intrinsic::loongarch_lsx_vsub_w:
  case Intrinsic::loongarch_lsx_vsub_d:
  case Intrinsic::loongarch_lasx_xvsub_b:
  case Intrinsic::loongarch_lasx_xvsub_h:
  case Intrinsic::loongarch_lasx_xvsub_w:
  case Intrinsic::loongarch_lasx_xvsub_d:
    return DAG.getNode(ISD::SUB, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vsubi_bu:
  case Intrinsic::loongarch_lsx_vsubi_hu:
  case Intrinsic::loongarch_lsx_vsubi_wu:
  case Intrinsic::loongarch_lsx_vsubi_du:
  case Intrinsic::loongarch_lasx_xvsubi_bu:
  case Intrinsic::loongarch_lasx_xvsubi_hu:
  case Intrinsic::loongarch_lasx_xvsubi_wu:
  case Intrinsic::loongarch_lasx_xvsubi_du:
    return DAG.getNode(ISD::SUB, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<5>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vneg_b:
  case Intrinsic::loongarch_lsx_vneg_h:
  case Intrinsic::loongarch_lsx_vneg_w:
  case Intrinsic::loongarch_lsx_vneg_d:
  case Intrinsic::loongarch_lasx_xvneg_b:
  case Intrinsic::loongarch_lasx_xvneg_h:
  case Intrinsic::loongarch_lasx_xvneg_w:
  case Intrinsic::loongarch_lasx_xvneg_d:
    return DAG.getNode(
        ISD::SUB, DL, N->getValueType(0),
        DAG.getConstant(
            APInt(N->getValueType(0).getScalarType().getSizeInBits(), 0,
                  /*isSigned=*/true),
            SDLoc(N), N->getValueType(0)),
        N->getOperand(1));
  case Intrinsic::loongarch_lsx_vmax_b:
  case Intrinsic::loongarch_lsx_vmax_h:
  case Intrinsic::loongarch_lsx_vmax_w:
  case Intrinsic::loongarch_lsx_vmax_d:
  case Intrinsic::loongarch_lasx_xvmax_b:
  case Intrinsic::loongarch_lasx_xvmax_h:
  case Intrinsic::loongarch_lasx_xvmax_w:
  case Intrinsic::loongarch_lasx_xvmax_d:
    return DAG.getNode(ISD::SMAX, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vmax_bu:
  case Intrinsic::loongarch_lsx_vmax_hu:
  case Intrinsic::loongarch_lsx_vmax_wu:
  case Intrinsic::loongarch_lsx_vmax_du:
  case Intrinsic::loongarch_lasx_xvmax_bu:
  case Intrinsic::loongarch_lasx_xvmax_hu:
  case Intrinsic::loongarch_lasx_xvmax_wu:
  case Intrinsic::loongarch_lasx_xvmax_du:
    return DAG.getNode(ISD::UMAX, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vmaxi_b:
  case Intrinsic::loongarch_lsx_vmaxi_h:
  case Intrinsic::loongarch_lsx_vmaxi_w:
  case Intrinsic::loongarch_lsx_vmaxi_d:
  case Intrinsic::loongarch_lasx_xvmaxi_b:
  case Intrinsic::loongarch_lasx_xvmaxi_h:
  case Intrinsic::loongarch_lasx_xvmaxi_w:
  case Intrinsic::loongarch_lasx_xvmaxi_d:
    return DAG.getNode(ISD::SMAX, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<5>(N, 2, DAG, /*IsSigned=*/true));
  case Intrinsic::loongarch_lsx_vmaxi_bu:
  case Intrinsic::loongarch_lsx_vmaxi_hu:
  case Intrinsic::loongarch_lsx_vmaxi_wu:
  case Intrinsic::loongarch_lsx_vmaxi_du:
  case Intrinsic::loongarch_lasx_xvmaxi_bu:
  case Intrinsic::loongarch_lasx_xvmaxi_hu:
  case Intrinsic::loongarch_lasx_xvmaxi_wu:
  case Intrinsic::loongarch_lasx_xvmaxi_du:
    return DAG.getNode(ISD::UMAX, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<5>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vmin_b:
  case Intrinsic::loongarch_lsx_vmin_h:
  case Intrinsic::loongarch_lsx_vmin_w:
  case Intrinsic::loongarch_lsx_vmin_d:
  case Intrinsic::loongarch_lasx_xvmin_b:
  case Intrinsic::loongarch_lasx_xvmin_h:
  case Intrinsic::loongarch_lasx_xvmin_w:
  case Intrinsic::loongarch_lasx_xvmin_d:
    return DAG.getNode(ISD::SMIN, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vmin_bu:
  case Intrinsic::loongarch_lsx_vmin_hu:
  case Intrinsic::loongarch_lsx_vmin_wu:
  case Intrinsic::loongarch_lsx_vmin_du:
  case Intrinsic::loongarch_lasx_xvmin_bu:
  case Intrinsic::loongarch_lasx_xvmin_hu:
  case Intrinsic::loongarch_lasx_xvmin_wu:
  case Intrinsic::loongarch_lasx_xvmin_du:
    return DAG.getNode(ISD::UMIN, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vmini_b:
  case Intrinsic::loongarch_lsx_vmini_h:
  case Intrinsic::loongarch_lsx_vmini_w:
  case Intrinsic::loongarch_lsx_vmini_d:
  case Intrinsic::loongarch_lasx_xvmini_b:
  case Intrinsic::loongarch_lasx_xvmini_h:
  case Intrinsic::loongarch_lasx_xvmini_w:
  case Intrinsic::loongarch_lasx_xvmini_d:
    return DAG.getNode(ISD::SMIN, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<5>(N, 2, DAG, /*IsSigned=*/true));
  case Intrinsic::loongarch_lsx_vmini_bu:
  case Intrinsic::loongarch_lsx_vmini_hu:
  case Intrinsic::loongarch_lsx_vmini_wu:
  case Intrinsic::loongarch_lsx_vmini_du:
  case Intrinsic::loongarch_lasx_xvmini_bu:
  case Intrinsic::loongarch_lasx_xvmini_hu:
  case Intrinsic::loongarch_lasx_xvmini_wu:
  case Intrinsic::loongarch_lasx_xvmini_du:
    return DAG.getNode(ISD::UMIN, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<5>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vmul_b:
  case Intrinsic::loongarch_lsx_vmul_h:
  case Intrinsic::loongarch_lsx_vmul_w:
  case Intrinsic::loongarch_lsx_vmul_d:
  case Intrinsic::loongarch_lasx_xvmul_b:
  case Intrinsic::loongarch_lasx_xvmul_h:
  case Intrinsic::loongarch_lasx_xvmul_w:
  case Intrinsic::loongarch_lasx_xvmul_d:
    return DAG.getNode(ISD::MUL, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vmadd_b:
  case Intrinsic::loongarch_lsx_vmadd_h:
  case Intrinsic::loongarch_lsx_vmadd_w:
  case Intrinsic::loongarch_lsx_vmadd_d:
  case Intrinsic::loongarch_lasx_xvmadd_b:
  case Intrinsic::loongarch_lasx_xvmadd_h:
  case Intrinsic::loongarch_lasx_xvmadd_w:
  case Intrinsic::loongarch_lasx_xvmadd_d: {
    EVT ResTy = N->getValueType(0);
    return DAG.getNode(ISD::ADD, SDLoc(N), ResTy, N->getOperand(1),
                       DAG.getNode(ISD::MUL, SDLoc(N), ResTy, N->getOperand(2),
                                   N->getOperand(3)));
  }
  case Intrinsic::loongarch_lsx_vmsub_b:
  case Intrinsic::loongarch_lsx_vmsub_h:
  case Intrinsic::loongarch_lsx_vmsub_w:
  case Intrinsic::loongarch_lsx_vmsub_d:
  case Intrinsic::loongarch_lasx_xvmsub_b:
  case Intrinsic::loongarch_lasx_xvmsub_h:
  case Intrinsic::loongarch_lasx_xvmsub_w:
  case Intrinsic::loongarch_lasx_xvmsub_d: {
    EVT ResTy = N->getValueType(0);
    return DAG.getNode(ISD::SUB, SDLoc(N), ResTy, N->getOperand(1),
                       DAG.getNode(ISD::MUL, SDLoc(N), ResTy, N->getOperand(2),
                                   N->getOperand(3)));
  }
  case Intrinsic::loongarch_lsx_vdiv_b:
  case Intrinsic::loongarch_lsx_vdiv_h:
  case Intrinsic::loongarch_lsx_vdiv_w:
  case Intrinsic::loongarch_lsx_vdiv_d:
  case Intrinsic::loongarch_lasx_xvdiv_b:
  case Intrinsic::loongarch_lasx_xvdiv_h:
  case Intrinsic::loongarch_lasx_xvdiv_w:
  case Intrinsic::loongarch_lasx_xvdiv_d:
    return DAG.getNode(ISD::SDIV, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vdiv_bu:
  case Intrinsic::loongarch_lsx_vdiv_hu:
  case Intrinsic::loongarch_lsx_vdiv_wu:
  case Intrinsic::loongarch_lsx_vdiv_du:
  case Intrinsic::loongarch_lasx_xvdiv_bu:
  case Intrinsic::loongarch_lasx_xvdiv_hu:
  case Intrinsic::loongarch_lasx_xvdiv_wu:
  case Intrinsic::loongarch_lasx_xvdiv_du:
    return DAG.getNode(ISD::UDIV, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vmod_b:
  case Intrinsic::loongarch_lsx_vmod_h:
  case Intrinsic::loongarch_lsx_vmod_w:
  case Intrinsic::loongarch_lsx_vmod_d:
  case Intrinsic::loongarch_lasx_xvmod_b:
  case Intrinsic::loongarch_lasx_xvmod_h:
  case Intrinsic::loongarch_lasx_xvmod_w:
  case Intrinsic::loongarch_lasx_xvmod_d:
    return DAG.getNode(ISD::SREM, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vmod_bu:
  case Intrinsic::loongarch_lsx_vmod_hu:
  case Intrinsic::loongarch_lsx_vmod_wu:
  case Intrinsic::loongarch_lsx_vmod_du:
  case Intrinsic::loongarch_lasx_xvmod_bu:
  case Intrinsic::loongarch_lasx_xvmod_hu:
  case Intrinsic::loongarch_lasx_xvmod_wu:
  case Intrinsic::loongarch_lasx_xvmod_du:
    return DAG.getNode(ISD::UREM, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vand_v:
  case Intrinsic::loongarch_lasx_xvand_v:
    return DAG.getNode(ISD::AND, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vor_v:
  case Intrinsic::loongarch_lasx_xvor_v:
    return DAG.getNode(ISD::OR, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vxor_v:
  case Intrinsic::loongarch_lasx_xvxor_v:
    return DAG.getNode(ISD::XOR, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vnor_v:
  case Intrinsic::loongarch_lasx_xvnor_v: {
    SDValue Res = DAG.getNode(ISD::OR, DL, N->getValueType(0), N->getOperand(1),
                              N->getOperand(2));
    return DAG.getNOT(DL, Res, Res->getValueType(0));
  }
  case Intrinsic::loongarch_lsx_vandi_b:
  case Intrinsic::loongarch_lasx_xvandi_b:
    return DAG.getNode(ISD::AND, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<8>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vori_b:
  case Intrinsic::loongarch_lasx_xvori_b:
    return DAG.getNode(ISD::OR, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<8>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vxori_b:
  case Intrinsic::loongarch_lasx_xvxori_b:
    return DAG.getNode(ISD::XOR, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<8>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vsll_b:
  case Intrinsic::loongarch_lsx_vsll_h:
  case Intrinsic::loongarch_lsx_vsll_w:
  case Intrinsic::loongarch_lsx_vsll_d:
  case Intrinsic::loongarch_lasx_xvsll_b:
  case Intrinsic::loongarch_lasx_xvsll_h:
  case Intrinsic::loongarch_lasx_xvsll_w:
  case Intrinsic::loongarch_lasx_xvsll_d:
    return DAG.getNode(ISD::SHL, DL, N->getValueType(0), N->getOperand(1),
                       truncateVecElts(N, DAG));
  case Intrinsic::loongarch_lsx_vslli_b:
  case Intrinsic::loongarch_lasx_xvslli_b:
    return DAG.getNode(ISD::SHL, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<3>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vslli_h:
  case Intrinsic::loongarch_lasx_xvslli_h:
    return DAG.getNode(ISD::SHL, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<4>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vslli_w:
  case Intrinsic::loongarch_lasx_xvslli_w:
    return DAG.getNode(ISD::SHL, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<5>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vslli_d:
  case Intrinsic::loongarch_lasx_xvslli_d:
    return DAG.getNode(ISD::SHL, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<6>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vsrl_b:
  case Intrinsic::loongarch_lsx_vsrl_h:
  case Intrinsic::loongarch_lsx_vsrl_w:
  case Intrinsic::loongarch_lsx_vsrl_d:
  case Intrinsic::loongarch_lasx_xvsrl_b:
  case Intrinsic::loongarch_lasx_xvsrl_h:
  case Intrinsic::loongarch_lasx_xvsrl_w:
  case Intrinsic::loongarch_lasx_xvsrl_d:
    return DAG.getNode(ISD::SRL, DL, N->getValueType(0), N->getOperand(1),
                       truncateVecElts(N, DAG));
  case Intrinsic::loongarch_lsx_vsrli_b:
  case Intrinsic::loongarch_lasx_xvsrli_b:
    return DAG.getNode(ISD::SRL, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<3>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vsrli_h:
  case Intrinsic::loongarch_lasx_xvsrli_h:
    return DAG.getNode(ISD::SRL, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<4>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vsrli_w:
  case Intrinsic::loongarch_lasx_xvsrli_w:
    return DAG.getNode(ISD::SRL, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<5>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vsrli_d:
  case Intrinsic::loongarch_lasx_xvsrli_d:
    return DAG.getNode(ISD::SRL, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<6>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vsra_b:
  case Intrinsic::loongarch_lsx_vsra_h:
  case Intrinsic::loongarch_lsx_vsra_w:
  case Intrinsic::loongarch_lsx_vsra_d:
  case Intrinsic::loongarch_lasx_xvsra_b:
  case Intrinsic::loongarch_lasx_xvsra_h:
  case Intrinsic::loongarch_lasx_xvsra_w:
  case Intrinsic::loongarch_lasx_xvsra_d:
    return DAG.getNode(ISD::SRA, DL, N->getValueType(0), N->getOperand(1),
                       truncateVecElts(N, DAG));
  case Intrinsic::loongarch_lsx_vsrai_b:
  case Intrinsic::loongarch_lasx_xvsrai_b:
    return DAG.getNode(ISD::SRA, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<3>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vsrai_h:
  case Intrinsic::loongarch_lasx_xvsrai_h:
    return DAG.getNode(ISD::SRA, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<4>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vsrai_w:
  case Intrinsic::loongarch_lasx_xvsrai_w:
    return DAG.getNode(ISD::SRA, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<5>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vsrai_d:
  case Intrinsic::loongarch_lasx_xvsrai_d:
    return DAG.getNode(ISD::SRA, DL, N->getValueType(0), N->getOperand(1),
                       lowerVectorSplatImm<6>(N, 2, DAG));
  case Intrinsic::loongarch_lsx_vclz_b:
  case Intrinsic::loongarch_lsx_vclz_h:
  case Intrinsic::loongarch_lsx_vclz_w:
  case Intrinsic::loongarch_lsx_vclz_d:
  case Intrinsic::loongarch_lasx_xvclz_b:
  case Intrinsic::loongarch_lasx_xvclz_h:
  case Intrinsic::loongarch_lasx_xvclz_w:
  case Intrinsic::loongarch_lasx_xvclz_d:
    return DAG.getNode(ISD::CTLZ, DL, N->getValueType(0), N->getOperand(1));
  case Intrinsic::loongarch_lsx_vpcnt_b:
  case Intrinsic::loongarch_lsx_vpcnt_h:
  case Intrinsic::loongarch_lsx_vpcnt_w:
  case Intrinsic::loongarch_lsx_vpcnt_d:
  case Intrinsic::loongarch_lasx_xvpcnt_b:
  case Intrinsic::loongarch_lasx_xvpcnt_h:
  case Intrinsic::loongarch_lasx_xvpcnt_w:
  case Intrinsic::loongarch_lasx_xvpcnt_d:
    return DAG.getNode(ISD::CTPOP, DL, N->getValueType(0), N->getOperand(1));
  case Intrinsic::loongarch_lsx_vbitclr_b:
  case Intrinsic::loongarch_lsx_vbitclr_h:
  case Intrinsic::loongarch_lsx_vbitclr_w:
  case Intrinsic::loongarch_lsx_vbitclr_d:
  case Intrinsic::loongarch_lasx_xvbitclr_b:
  case Intrinsic::loongarch_lasx_xvbitclr_h:
  case Intrinsic::loongarch_lasx_xvbitclr_w:
  case Intrinsic::loongarch_lasx_xvbitclr_d:
    return lowerVectorBitClear(N, DAG);
  case Intrinsic::loongarch_lsx_vbitclri_b:
  case Intrinsic::loongarch_lasx_xvbitclri_b:
    return lowerVectorBitClearImm<3>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitclri_h:
  case Intrinsic::loongarch_lasx_xvbitclri_h:
    return lowerVectorBitClearImm<4>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitclri_w:
  case Intrinsic::loongarch_lasx_xvbitclri_w:
    return lowerVectorBitClearImm<5>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitclri_d:
  case Intrinsic::loongarch_lasx_xvbitclri_d:
    return lowerVectorBitClearImm<6>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitset_b:
  case Intrinsic::loongarch_lsx_vbitset_h:
  case Intrinsic::loongarch_lsx_vbitset_w:
  case Intrinsic::loongarch_lsx_vbitset_d:
  case Intrinsic::loongarch_lasx_xvbitset_b:
  case Intrinsic::loongarch_lasx_xvbitset_h:
  case Intrinsic::loongarch_lasx_xvbitset_w:
  case Intrinsic::loongarch_lasx_xvbitset_d: {
    EVT VecTy = N->getValueType(0);
    SDValue One = DAG.getConstant(1, DL, VecTy);
    return DAG.getNode(
        ISD::OR, DL, VecTy, N->getOperand(1),
        DAG.getNode(ISD::SHL, DL, VecTy, One, truncateVecElts(N, DAG)));
  }
  case Intrinsic::loongarch_lsx_vbitseti_b:
  case Intrinsic::loongarch_lasx_xvbitseti_b:
    return lowerVectorBitSetImm<3>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitseti_h:
  case Intrinsic::loongarch_lasx_xvbitseti_h:
    return lowerVectorBitSetImm<4>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitseti_w:
  case Intrinsic::loongarch_lasx_xvbitseti_w:
    return lowerVectorBitSetImm<5>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitseti_d:
  case Intrinsic::loongarch_lasx_xvbitseti_d:
    return lowerVectorBitSetImm<6>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitrev_b:
  case Intrinsic::loongarch_lsx_vbitrev_h:
  case Intrinsic::loongarch_lsx_vbitrev_w:
  case Intrinsic::loongarch_lsx_vbitrev_d:
  case Intrinsic::loongarch_lasx_xvbitrev_b:
  case Intrinsic::loongarch_lasx_xvbitrev_h:
  case Intrinsic::loongarch_lasx_xvbitrev_w:
  case Intrinsic::loongarch_lasx_xvbitrev_d: {
    EVT VecTy = N->getValueType(0);
    SDValue One = DAG.getConstant(1, DL, VecTy);
    return DAG.getNode(
        ISD::XOR, DL, VecTy, N->getOperand(1),
        DAG.getNode(ISD::SHL, DL, VecTy, One, truncateVecElts(N, DAG)));
  }
  case Intrinsic::loongarch_lsx_vbitrevi_b:
  case Intrinsic::loongarch_lasx_xvbitrevi_b:
    return lowerVectorBitRevImm<3>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitrevi_h:
  case Intrinsic::loongarch_lasx_xvbitrevi_h:
    return lowerVectorBitRevImm<4>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitrevi_w:
  case Intrinsic::loongarch_lasx_xvbitrevi_w:
    return lowerVectorBitRevImm<5>(N, DAG);
  case Intrinsic::loongarch_lsx_vbitrevi_d:
  case Intrinsic::loongarch_lasx_xvbitrevi_d:
    return lowerVectorBitRevImm<6>(N, DAG);
  case Intrinsic::loongarch_lsx_vfadd_s:
  case Intrinsic::loongarch_lsx_vfadd_d:
  case Intrinsic::loongarch_lasx_xvfadd_s:
  case Intrinsic::loongarch_lasx_xvfadd_d:
    return DAG.getNode(ISD::FADD, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vfsub_s:
  case Intrinsic::loongarch_lsx_vfsub_d:
  case Intrinsic::loongarch_lasx_xvfsub_s:
  case Intrinsic::loongarch_lasx_xvfsub_d:
    return DAG.getNode(ISD::FSUB, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vfmul_s:
  case Intrinsic::loongarch_lsx_vfmul_d:
  case Intrinsic::loongarch_lasx_xvfmul_s:
  case Intrinsic::loongarch_lasx_xvfmul_d:
    return DAG.getNode(ISD::FMUL, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vfdiv_s:
  case Intrinsic::loongarch_lsx_vfdiv_d:
  case Intrinsic::loongarch_lasx_xvfdiv_s:
  case Intrinsic::loongarch_lasx_xvfdiv_d:
    return DAG.getNode(ISD::FDIV, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2));
  case Intrinsic::loongarch_lsx_vfmadd_s:
  case Intrinsic::loongarch_lsx_vfmadd_d:
  case Intrinsic::loongarch_lasx_xvfmadd_s:
  case Intrinsic::loongarch_lasx_xvfmadd_d:
    return DAG.getNode(ISD::FMA, DL, N->getValueType(0), N->getOperand(1),
                       N->getOperand(2), N->getOperand(3));
  case Intrinsic::loongarch_lsx_vinsgr2vr_b:
    return DAG.getNode(ISD::INSERT_VECTOR_ELT, SDLoc(N), N->getValueType(0),
                       N->getOperand(1), N->getOperand(2),
                       legalizeIntrinsicImmArg<4>(N, 3, DAG, Subtarget));
  case Intrinsic::loongarch_lsx_vinsgr2vr_h:
  case Intrinsic::loongarch_lasx_xvinsgr2vr_w:
    return DAG.getNode(ISD::INSERT_VECTOR_ELT, SDLoc(N), N->getValueType(0),
                       N->getOperand(1), N->getOperand(2),
                       legalizeIntrinsicImmArg<3>(N, 3, DAG, Subtarget));
  case Intrinsic::loongarch_lsx_vinsgr2vr_w:
  case Intrinsic::loongarch_lasx_xvinsgr2vr_d:
    return DAG.getNode(ISD::INSERT_VECTOR_ELT, SDLoc(N), N->getValueType(0),
                       N->getOperand(1), N->getOperand(2),
                       legalizeIntrinsicImmArg<2>(N, 3, DAG, Subtarget));
  case Intrinsic::loongarch_lsx_vinsgr2vr_d:
    return DAG.getNode(ISD::INSERT_VECTOR_ELT, SDLoc(N), N->getValueType(0),
                       N->getOperand(1), N->getOperand(2),
                       legalizeIntrinsicImmArg<1>(N, 3, DAG, Subtarget));
  case Intrinsic::loongarch_lsx_vreplgr2vr_b:
  case Intrinsic::loongarch_lsx_vreplgr2vr_h:
  case Intrinsic::loongarch_lsx_vreplgr2vr_w:
  case Intrinsic::loongarch_lsx_vreplgr2vr_d:
  case Intrinsic::loongarch_lasx_xvreplgr2vr_b:
  case Intrinsic::loongarch_lasx_xvreplgr2vr_h:
  case Intrinsic::loongarch_lasx_xvreplgr2vr_w:
  case Intrinsic::loongarch_lasx_xvreplgr2vr_d: {
    EVT ResTy = N->getValueType(0);
    SmallVector<SDValue> Ops(ResTy.getVectorNumElements(), N->getOperand(1));
    return DAG.getBuildVector(ResTy, DL, Ops);
  }
  case Intrinsic::loongarch_lsx_vreplve_b:
  case Intrinsic::loongarch_lsx_vreplve_h:
  case Intrinsic::loongarch_lsx_vreplve_w:
  case Intrinsic::loongarch_lsx_vreplve_d:
  case Intrinsic::loongarch_lasx_xvreplve_b:
  case Intrinsic::loongarch_lasx_xvreplve_h:
  case Intrinsic::loongarch_lasx_xvreplve_w:
  case Intrinsic::loongarch_lasx_xvreplve_d:
    return DAG.getNode(LoongArchISD::VREPLVE, DL, N->getValueType(0),
                       N->getOperand(1),
                       DAG.getNode(ISD::ANY_EXTEND, DL, Subtarget.getGRLenVT(),
                                   N->getOperand(2)));
  }
  return SDValue();
}
 
SDValue LoongArchTargetLowering::PerformDAGCombine(SDNode *N,
                                                   DAGCombinerInfo &DCI) const {
  SelectionDAG &DAG = DCI.DAG;
  switch (N->getOpcode()) {
  default:
    break;
  case ISD::AND:
    return performANDCombine(N, DAG, DCI, Subtarget);
  case ISD::OR:
    return performORCombine(N, DAG, DCI, Subtarget);
  case ISD::SETCC:
    return performSETCCCombine(N, DAG, DCI, Subtarget);
  case ISD::SRL:
    return performSRLCombine(N, DAG, DCI, Subtarget);
  case LoongArchISD::BITREV_W:
    return performBITREV_WCombine(N, DAG, DCI, Subtarget);
  case ISD::INTRINSIC_WO_CHAIN:
    return performINTRINSIC_WO_CHAINCombine(N, DAG, DCI, Subtarget);
  }
  return SDValue();
}
 
static MachineBasicBlock *insertDivByZeroTrap(MachineInstr &MI,
                                              MachineBasicBlock *MBB) {
  if (!ZeroDivCheck)
    return MBB;
 
  // Build instructions:
  // MBB:
  //   div(or mod)   $dst, $dividend, $divisor
  //   bnez          $divisor, SinkMBB
  // BreakMBB:
  //   break         7 // BRK_DIVZERO
  // SinkMBB:
  //   fallthrough
  const BasicBlock *LLVM_BB = MBB->getBasicBlock();
  MachineFunction::iterator It = ++MBB->getIterator();
  MachineFunction *MF = MBB->getParent();
  auto BreakMBB = MF->CreateMachineBasicBlock(LLVM_BB);
  auto SinkMBB = MF->CreateMachineBasicBlock(LLVM_BB);
  MF->insert(It, BreakMBB);
  MF->insert(It, SinkMBB);
 
  // Transfer the remainder of MBB and its successor edges to SinkMBB.
  SinkMBB->splice(SinkMBB->end(), MBB, std::next(MI.getIterator()), MBB->end());
  SinkMBB->transferSuccessorsAndUpdatePHIs(