xref: /third_party/node/deps/v8/src/compiler/backend/ia32/instruction-selector-ia32.cc

// Copyright 2014 the V8 project authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.

#include <stddef.h>
#include <stdint.h>

#include <limits>
#include <type_traits>
#include <vector>

#include "src/base/bits.h"
#include "src/base/flags.h"
#include "src/base/iterator.h"
#include "src/base/logging.h"
#include "src/base/macros.h"
#include "src/base/platform/wrappers.h"
#include "src/codegen/cpu-features.h"
#include "src/codegen/ia32/assembler-ia32.h"
#include "src/codegen/ia32/register-ia32.h"
#include "src/codegen/machine-type.h"
#include "src/codegen/turbo-assembler.h"
#include "src/common/globals.h"
#include "src/compiler/backend/instruction-codes.h"
#include "src/compiler/backend/instruction-selector-impl.h"
#include "src/compiler/backend/instruction-selector.h"
#include "src/compiler/backend/instruction.h"
#include "src/compiler/common-operator.h"
#include "src/compiler/frame.h"
#include "src/compiler/globals.h"
#include "src/compiler/linkage.h"
#include "src/compiler/machine-operator.h"
#include "src/compiler/node-matchers.h"
#include "src/compiler/node-properties.h"
#include "src/compiler/node.h"
#include "src/compiler/opcodes.h"
#include "src/compiler/operator.h"
#include "src/compiler/write-barrier-kind.h"
#include "src/flags/flags.h"
#include "src/utils/utils.h"
#include "src/zone/zone-containers.h"

#if V8_ENABLE_WEBASSEMBLY
#include "src/wasm/simd-shuffle.h"
#endif  // V8_ENABLE_WEBASSEMBLY

namespace v8 {
namespace internal {
namespace compiler {

// Adds IA32-specific methods for generating operands.
class IA32OperandGenerator final : public OperandGenerator {
 public:
  explicit IA32OperandGenerator(InstructionSelector* selector)
      : OperandGenerator(selector) {}

  InstructionOperand UseByteRegister(Node* node) {
    // TODO(titzer): encode byte register use constraints.
    return UseFixed(node, edx);
  }

  InstructionOperand DefineAsByteRegister(Node* node) {
    // TODO(titzer): encode byte register def constraints.
    return DefineAsRegister(node);
  }

  bool CanBeMemoryOperand(InstructionCode opcode, Node* node, Node* input,
                          int effect_level) {
    if ((input->opcode() != IrOpcode::kLoad &&
         input->opcode() != IrOpcode::kLoadImmutable) ||
        !selector()->CanCover(node, input)) {
      return false;
    }
    if (effect_level != selector()->GetEffectLevel(input)) {
      return false;
    }
    MachineRepresentation rep =
        LoadRepresentationOf(input->op()).representation();
    switch (opcode) {
      case kIA32And:
      case kIA32Or:
      case kIA32Xor:
      case kIA32Add:
      case kIA32Sub:
      case kIA32Cmp:
      case kIA32Test:
        return rep == MachineRepresentation::kWord32 || IsAnyTagged(rep);
      case kIA32Cmp16:
      case kIA32Test16:
        return rep == MachineRepresentation::kWord16;
      case kIA32Cmp8:
      case kIA32Test8:
        return rep == MachineRepresentation::kWord8;
      default:
        break;
    }
    return false;
  }

  bool CanBeImmediate(Node* node) {
    switch (node->opcode()) {
      case IrOpcode::kInt32Constant:
      case IrOpcode::kExternalConstant:
      case IrOpcode::kRelocatableInt32Constant:
      case IrOpcode::kRelocatableInt64Constant:
        return true;
      case IrOpcode::kNumberConstant: {
        const double value = OpParameter<double>(node->op());
        return bit_cast<int64_t>(value) == 0;
      }
      case IrOpcode::kHeapConstant: {
// TODO(bmeurer): We must not dereference handles concurrently. If we
// really have to this here, then we need to find a way to put this
// information on the HeapConstant node already.
#if 0
        // Constants in young generation cannot be used as immediates in V8
        // because the GC does not scan code objects when collecting the young
        // generation.
        Handle<HeapObject> value = HeapConstantOf(node->op());
        return !Heap::InYoungGeneration(*value);
#else
        return false;
#endif
      }
      default:
        return false;
    }
  }

  AddressingMode GenerateMemoryOperandInputs(
      Node* index, int scale, Node* base, int32_t displacement,
      DisplacementMode displacement_mode, InstructionOperand inputs[],
      size_t* input_count, RegisterMode register_mode = kRegister) {
    AddressingMode mode = kMode_MRI;
    if (displacement_mode == kNegativeDisplacement) {
      displacement = base::bits::WraparoundNeg32(displacement);
    }
    if (base != nullptr) {
      if (base->opcode() == IrOpcode::kInt32Constant) {
        displacement = base::bits::WraparoundAdd32(
            displacement, OpParameter<int32_t>(base->op()));
        base = nullptr;
      }
    }
    if (base != nullptr) {
      inputs[(*input_count)++] = UseRegisterWithMode(base, register_mode);
      if (index != nullptr) {
        DCHECK(scale >= 0 && scale <= 3);
        inputs[(*input_count)++] = UseRegisterWithMode(index, register_mode);
        if (displacement != 0) {
          inputs[(*input_count)++] = TempImmediate(displacement);
          static const AddressingMode kMRnI_modes[] = {kMode_MR1I, kMode_MR2I,
                                                       kMode_MR4I, kMode_MR8I};
          mode = kMRnI_modes[scale];
        } else {
          static const AddressingMode kMRn_modes[] = {kMode_MR1, kMode_MR2,
                                                      kMode_MR4, kMode_MR8};
          mode = kMRn_modes[scale];
        }
      } else {
        if (displacement == 0) {
          mode = kMode_MR;
        } else {
          inputs[(*input_count)++] = TempImmediate(displacement);
          mode = kMode_MRI;
        }
      }
    } else {
      DCHECK(scale >= 0 && scale <= 3);
      if (index != nullptr) {
        inputs[(*input_count)++] = UseRegisterWithMode(index, register_mode);
        if (displacement != 0) {
          inputs[(*input_count)++] = TempImmediate(displacement);
          static const AddressingMode kMnI_modes[] = {kMode_MRI, kMode_M2I,
                                                      kMode_M4I, kMode_M8I};
          mode = kMnI_modes[scale];
        } else {
          static const AddressingMode kMn_modes[] = {kMode_MR, kMode_M2,
                                                     kMode_M4, kMode_M8};
          mode = kMn_modes[scale];
        }
      } else {
        inputs[(*input_count)++] = TempImmediate(displacement);
        return kMode_MI;
      }
    }
    return mode;
  }

  AddressingMode GenerateMemoryOperandInputs(
      Node* index, int scale, Node* base, Node* displacement_node,
      DisplacementMode displacement_mode, InstructionOperand inputs[],
      size_t* input_count, RegisterMode register_mode = kRegister) {
    int32_t displacement = (displacement_node == nullptr)
                               ? 0
                               : OpParameter<int32_t>(displacement_node->op());
    return GenerateMemoryOperandInputs(index, scale, base, displacement,
                                       displacement_mode, inputs, input_count,
                                       register_mode);
  }

  AddressingMode GetEffectiveAddressMemoryOperand(
      Node* node, InstructionOperand inputs[], size_t* input_count,
      RegisterMode register_mode = kRegister) {
    {
      LoadMatcher<ExternalReferenceMatcher> m(node);
      if (m.index().HasResolvedValue() && m.object().HasResolvedValue() &&
          selector()->CanAddressRelativeToRootsRegister(
              m.object().ResolvedValue())) {
        ptrdiff_t const delta =
            m.index().ResolvedValue() +
            TurboAssemblerBase::RootRegisterOffsetForExternalReference(
                selector()->isolate(), m.object().ResolvedValue());
        if (is_int32(delta)) {
          inputs[(*input_count)++] = TempImmediate(static_cast<int32_t>(delta));
          return kMode_Root;
        }
      }
    }

    BaseWithIndexAndDisplacement32Matcher m(node, AddressOption::kAllowAll);
    DCHECK(m.matches());
    if ((m.displacement() == nullptr || CanBeImmediate(m.displacement()))) {
      return GenerateMemoryOperandInputs(
          m.index(), m.scale(), m.base(), m.displacement(),
          m.displacement_mode(), inputs, input_count, register_mode);
    } else {
      inputs[(*input_count)++] =
          UseRegisterWithMode(node->InputAt(0), register_mode);
      inputs[(*input_count)++] =
          UseRegisterWithMode(node->InputAt(1), register_mode);
      return kMode_MR1;
    }
  }

  InstructionOperand GetEffectiveIndexOperand(Node* index,
                                              AddressingMode* mode) {
    if (CanBeImmediate(index)) {
      *mode = kMode_MRI;
      return UseImmediate(index);
    } else {
      *mode = kMode_MR1;
      return UseUniqueRegister(index);
    }
  }

  bool CanBeBetterLeftOperand(Node* node) const {
    return !selector()->IsLive(node);
  }
};

namespace {

ArchOpcode GetLoadOpcode(LoadRepresentation load_rep) {
  ArchOpcode opcode;
  switch (load_rep.representation()) {
    case MachineRepresentation::kFloat32:
      opcode = kIA32Movss;
      break;
    case MachineRepresentation::kFloat64:
      opcode = kIA32Movsd;
      break;
    case MachineRepresentation::kBit:  // Fall through.
    case MachineRepresentation::kWord8:
      opcode = load_rep.IsSigned() ? kIA32Movsxbl : kIA32Movzxbl;
      break;
    case MachineRepresentation::kWord16:
      opcode = load_rep.IsSigned() ? kIA32Movsxwl : kIA32Movzxwl;
      break;
    case MachineRepresentation::kTaggedSigned:   // Fall through.
    case MachineRepresentation::kTaggedPointer:  // Fall through.
    case MachineRepresentation::kTagged:         // Fall through.
    case MachineRepresentation::kWord32:
      opcode = kIA32Movl;
      break;
    case MachineRepresentation::kSimd128:
      opcode = kIA32Movdqu;
      break;
    case MachineRepresentation::kCompressedPointer:  // Fall through.
    case MachineRepresentation::kCompressed:         // Fall through.
    case MachineRepresentation::kSandboxedPointer:   // Fall through.
    case MachineRepresentation::kWord64:             // Fall through.
    case MachineRepresentation::kMapWord:            // Fall through.
    case MachineRepresentation::kNone:
      UNREACHABLE();
  }
  return opcode;
}

void VisitRO(InstructionSelector* selector, Node* node, ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  Node* input = node->InputAt(0);
  // We have to use a byte register as input to movsxb.
  InstructionOperand input_op =
      opcode == kIA32Movsxbl ? g.UseFixed(input, eax) : g.Use(input);
  selector->Emit(opcode, g.DefineAsRegister(node), input_op);
}

void VisitROWithTemp(InstructionSelector* selector, Node* node,
                     ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand temps[] = {g.TempRegister()};
  selector->Emit(opcode, g.DefineAsRegister(node), g.Use(node->InputAt(0)),
                 arraysize(temps), temps);
}

void VisitROWithTempSimd(InstructionSelector* selector, Node* node,
                         ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand temps[] = {g.TempSimd128Register()};
  selector->Emit(opcode, g.DefineAsRegister(node),
                 g.UseUniqueRegister(node->InputAt(0)), arraysize(temps),
                 temps);
}

void VisitRR(InstructionSelector* selector, Node* node,
             InstructionCode opcode) {
  IA32OperandGenerator g(selector);
  selector->Emit(opcode, g.DefineAsRegister(node),
                 g.UseRegister(node->InputAt(0)));
}

void VisitRROFloat(InstructionSelector* selector, Node* node,
                   ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
  InstructionOperand operand1 = g.Use(node->InputAt(1));
  if (selector->IsSupported(AVX)) {
    selector->Emit(opcode, g.DefineAsRegister(node), operand0, operand1);
  } else {
    selector->Emit(opcode, g.DefineSameAsFirst(node), operand0, operand1);
  }
}

// For float unary operations. Also allocates a temporary general register for
// used in external operands. If a temp is not required, use VisitRRSimd (since
// float and SIMD registers are the same on IA32.
void VisitFloatUnop(InstructionSelector* selector, Node* node, Node* input,
                    ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand temps[] = {g.TempRegister()};
  // No need for unique because inputs are float but temp is general.
  if (selector->IsSupported(AVX)) {
    selector->Emit(opcode, g.DefineAsRegister(node), g.UseRegister(input),
                   arraysize(temps), temps);
  } else {
    selector->Emit(opcode, g.DefineSameAsFirst(node), g.UseRegister(input),
                   arraysize(temps), temps);
  }
}

void VisitRRSimd(InstructionSelector* selector, Node* node,
                 ArchOpcode avx_opcode, ArchOpcode sse_opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
  if (selector->IsSupported(AVX)) {
    selector->Emit(avx_opcode, g.DefineAsRegister(node), operand0);
  } else {
    selector->Emit(sse_opcode, g.DefineSameAsFirst(node), operand0);
  }
}

void VisitRRSimd(InstructionSelector* selector, Node* node, ArchOpcode opcode) {
  VisitRRSimd(selector, node, opcode, opcode);
}

// TODO(v8:9198): Like VisitRROFloat, but for SIMD. SSE requires operand1 to be
// a register as we don't have memory alignment yet. For AVX, memory operands
// are fine, but can have performance issues if not aligned to 16/32 bytes
// (based on load size), see SDM Vol 1, chapter 14.9
void VisitRROSimd(InstructionSelector* selector, Node* node,
                  ArchOpcode avx_opcode, ArchOpcode sse_opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
  if (selector->IsSupported(AVX)) {
    selector->Emit(avx_opcode, g.DefineAsRegister(node), operand0,
                   g.UseRegister(node->InputAt(1)));
  } else {
    selector->Emit(sse_opcode, g.DefineSameAsFirst(node), operand0,
                   g.UseRegister(node->InputAt(1)));
  }
}
void VisitRRRSimd(InstructionSelector* selector, Node* node,
                  ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand dst = selector->IsSupported(AVX)
                               ? g.DefineAsRegister(node)
                               : g.DefineSameAsFirst(node);
  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
  InstructionOperand operand1 = g.UseRegister(node->InputAt(1));
  selector->Emit(opcode, dst, operand0, operand1);
}

void VisitRRISimd(InstructionSelector* selector, Node* node,
                  ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
  InstructionOperand operand1 =
      g.UseImmediate(OpParameter<int32_t>(node->op()));
  // 8x16 uses movsx_b on dest to extract a byte, which only works
  // if dest is a byte register.
  InstructionOperand dest = opcode == kIA32I8x16ExtractLaneS
                                ? g.DefineAsFixed(node, eax)
                                : g.DefineAsRegister(node);
  selector->Emit(opcode, dest, operand0, operand1);
}

void VisitRRISimd(InstructionSelector* selector, Node* node,
                  ArchOpcode avx_opcode, ArchOpcode sse_opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
  InstructionOperand operand1 =
      g.UseImmediate(OpParameter<int32_t>(node->op()));
  if (selector->IsSupported(AVX)) {
    selector->Emit(avx_opcode, g.DefineAsRegister(node), operand0, operand1);
  } else {
    selector->Emit(sse_opcode, g.DefineSameAsFirst(node), operand0, operand1);
  }
}

void VisitRROSimdShift(InstructionSelector* selector, Node* node,
                       ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  if (g.CanBeImmediate(node->InputAt(1))) {
    selector->Emit(opcode, g.DefineSameAsFirst(node),
                   g.UseRegister(node->InputAt(0)),
                   g.UseImmediate(node->InputAt(1)));
  } else {
    InstructionOperand operand0 = g.UseUniqueRegister(node->InputAt(0));
    InstructionOperand operand1 = g.UseUniqueRegister(node->InputAt(1));
    InstructionOperand temps[] = {g.TempSimd128Register(), g.TempRegister()};
    selector->Emit(opcode, g.DefineSameAsFirst(node), operand0, operand1,
                   arraysize(temps), temps);
  }
}

void VisitRRRR(InstructionSelector* selector, Node* node,
               InstructionCode opcode) {
  IA32OperandGenerator g(selector);
  selector->Emit(
      opcode, g.DefineAsRegister(node), g.UseRegister(node->InputAt(0)),
      g.UseRegister(node->InputAt(1)), g.UseRegister(node->InputAt(2)));
}

void VisitI8x16Shift(InstructionSelector* selector, Node* node,
                     ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand output = CpuFeatures::IsSupported(AVX)
                                  ? g.UseRegister(node)
                                  : g.DefineSameAsFirst(node);

  if (g.CanBeImmediate(node->InputAt(1))) {
    if (opcode == kIA32I8x16ShrS) {
      selector->Emit(opcode, output, g.UseRegister(node->InputAt(0)),
                     g.UseImmediate(node->InputAt(1)));
    } else {
      InstructionOperand temps[] = {g.TempRegister()};
      selector->Emit(opcode, output, g.UseRegister(node->InputAt(0)),
                     g.UseImmediate(node->InputAt(1)), arraysize(temps), temps);
    }
  } else {
    InstructionOperand operand0 = g.UseUniqueRegister(node->InputAt(0));
    InstructionOperand operand1 = g.UseUniqueRegister(node->InputAt(1));
    InstructionOperand temps[] = {g.TempRegister(), g.TempSimd128Register()};
    selector->Emit(opcode, output, operand0, operand1, arraysize(temps), temps);
  }
}
}  // namespace

void InstructionSelector::VisitStackSlot(Node* node) {
  StackSlotRepresentation rep = StackSlotRepresentationOf(node->op());
  int slot = frame_->AllocateSpillSlot(rep.size(), rep.alignment());
  OperandGenerator g(this);

  Emit(kArchStackSlot, g.DefineAsRegister(node),
       sequence()->AddImmediate(Constant(slot)), 0, nullptr);
}

void InstructionSelector::VisitAbortCSADcheck(Node* node) {
  IA32OperandGenerator g(this);
  Emit(kArchAbortCSADcheck, g.NoOutput(), g.UseFixed(node->InputAt(0), edx));
}

void InstructionSelector::VisitLoadLane(Node* node) {
  LoadLaneParameters params = LoadLaneParametersOf(node->op());
  InstructionCode opcode = kArchNop;
  if (params.rep == MachineType::Int8()) {
    opcode = kIA32Pinsrb;
  } else if (params.rep == MachineType::Int16()) {
    opcode = kIA32Pinsrw;
  } else if (params.rep == MachineType::Int32()) {
    opcode = kIA32Pinsrd;
  } else if (params.rep == MachineType::Int64()) {
    // pinsrq not available on IA32.
    if (params.laneidx == 0) {
      opcode = kIA32Movlps;
    } else {
      DCHECK_EQ(1, params.laneidx);
      opcode = kIA32Movhps;
    }
  } else {
    UNREACHABLE();
  }

  IA32OperandGenerator g(this);
  InstructionOperand outputs[] = {IsSupported(AVX) ? g.DefineAsRegister(node)
                                                   : g.DefineSameAsFirst(node)};
  // Input 0 is value node, 1 is lane idx, and GetEffectiveAddressMemoryOperand
  // uses up to 3 inputs. This ordering is consistent with other operations that
  // use the same opcode.
  InstructionOperand inputs[5];
  size_t input_count = 0;

  inputs[input_count++] = g.UseRegister(node->InputAt(2));
  inputs[input_count++] = g.UseImmediate(params.laneidx);

  AddressingMode mode =
      g.GetEffectiveAddressMemoryOperand(node, inputs, &input_count);
  opcode |= AddressingModeField::encode(mode);

  DCHECK_GE(5, input_count);

  // IA32 supports unaligned loads.
  DCHECK_NE(params.kind, MemoryAccessKind::kUnaligned);
  // Trap handler is not supported on IA32.
  DCHECK_NE(params.kind, MemoryAccessKind::kProtected);

  Emit(opcode, 1, outputs, input_count, inputs);
}

void InstructionSelector::VisitLoadTransform(Node* node) {
  LoadTransformParameters params = LoadTransformParametersOf(node->op());
  InstructionCode opcode;
  switch (params.transformation) {
    case LoadTransformation::kS128Load8Splat:
      opcode = kIA32S128Load8Splat;
      break;
    case LoadTransformation::kS128Load16Splat:
      opcode = kIA32S128Load16Splat;
      break;
    case LoadTransformation::kS128Load32Splat:
      opcode = kIA32S128Load32Splat;
      break;
    case LoadTransformation::kS128Load64Splat:
      opcode = kIA32S128Load64Splat;
      break;
    case LoadTransformation::kS128Load8x8S:
      opcode = kIA32S128Load8x8S;
      break;
    case LoadTransformation::kS128Load8x8U:
      opcode = kIA32S128Load8x8U;
      break;
    case LoadTransformation::kS128Load16x4S:
      opcode = kIA32S128Load16x4S;
      break;
    case LoadTransformation::kS128Load16x4U:
      opcode = kIA32S128Load16x4U;
      break;
    case LoadTransformation::kS128Load32x2S:
      opcode = kIA32S128Load32x2S;
      break;
    case LoadTransformation::kS128Load32x2U:
      opcode = kIA32S128Load32x2U;
      break;
    case LoadTransformation::kS128Load32Zero:
      opcode = kIA32Movss;
      break;
    case LoadTransformation::kS128Load64Zero:
      opcode = kIA32Movsd;
      break;
    default:
      UNREACHABLE();
  }

  // IA32 supports unaligned loads.
  DCHECK_NE(params.kind, MemoryAccessKind::kUnaligned);
  // Trap handler is not supported on IA32.
  DCHECK_NE(params.kind, MemoryAccessKind::kProtected);

  IA32OperandGenerator g(this);
  InstructionOperand outputs[1];
  outputs[0] = g.DefineAsRegister(node);
  InstructionOperand inputs[3];
  size_t input_count = 0;
  AddressingMode mode =
      g.GetEffectiveAddressMemoryOperand(node, inputs, &input_count);
  InstructionCode code = opcode | AddressingModeField::encode(mode);
  Emit(code, 1, outputs, input_count, inputs);
}

void InstructionSelector::VisitLoad(Node* node, Node* value,
                                    InstructionCode opcode) {
  IA32OperandGenerator g(this);
  InstructionOperand outputs[1];
  outputs[0] = g.DefineAsRegister(node);
  InstructionOperand inputs[3];
  size_t input_count = 0;
  AddressingMode mode =
      g.GetEffectiveAddressMemoryOperand(value, inputs, &input_count);
  InstructionCode code = opcode | AddressingModeField::encode(mode);
  Emit(code, 1, outputs, input_count, inputs);
}

void InstructionSelector::VisitLoad(Node* node) {
  LoadRepresentation load_rep = LoadRepresentationOf(node->op());
  DCHECK(!load_rep.IsMapWord());
  VisitLoad(node, node, GetLoadOpcode(load_rep));
}

void InstructionSelector::VisitProtectedLoad(Node* node) {
  // TODO(eholk)
  UNIMPLEMENTED();
}

namespace {

ArchOpcode GetStoreOpcode(MachineRepresentation rep) {
  switch (rep) {
    case MachineRepresentation::kFloat32:
      return kIA32Movss;
    case MachineRepresentation::kFloat64:
      return kIA32Movsd;
    case MachineRepresentation::kBit:  // Fall through.
    case MachineRepresentation::kWord8:
      return kIA32Movb;
    case MachineRepresentation::kWord16:
      return kIA32Movw;
    case MachineRepresentation::kTaggedSigned:   // Fall through.
    case MachineRepresentation::kTaggedPointer:  // Fall through.
    case MachineRepresentation::kTagged:         // Fall through.
    case MachineRepresentation::kWord32:
      return kIA32Movl;
    case MachineRepresentation::kSimd128:
      return kIA32Movdqu;
    case MachineRepresentation::kCompressedPointer:  // Fall through.
    case MachineRepresentation::kCompressed:         // Fall through.
    case MachineRepresentation::kSandboxedPointer:   // Fall through.
    case MachineRepresentation::kWord64:             // Fall through.
    case MachineRepresentation::kMapWord:            // Fall through.
    case MachineRepresentation::kNone:
      UNREACHABLE();
  }
}

ArchOpcode GetSeqCstStoreOpcode(MachineRepresentation rep) {
  switch (rep) {
    case MachineRepresentation::kWord8:
      return kAtomicExchangeInt8;
    case MachineRepresentation::kWord16:
      return kAtomicExchangeInt16;
    case MachineRepresentation::kTaggedSigned:   // Fall through.
    case MachineRepresentation::kTaggedPointer:  // Fall through.
    case MachineRepresentation::kTagged:         // Fall through.
    case MachineRepresentation::kWord32:
      return kAtomicExchangeWord32;
    default:
      UNREACHABLE();
  }
}

void VisitAtomicExchange(InstructionSelector* selector, Node* node,
                         ArchOpcode opcode, MachineRepresentation rep) {
  IA32OperandGenerator g(selector);
  Node* base = node->InputAt(0);
  Node* index = node->InputAt(1);
  Node* value = node->InputAt(2);

  AddressingMode addressing_mode;
  InstructionOperand value_operand = (rep == MachineRepresentation::kWord8)
                                         ? g.UseFixed(value, edx)
                                         : g.UseUniqueRegister(value);
  InstructionOperand inputs[] = {
      value_operand, g.UseUniqueRegister(base),
      g.GetEffectiveIndexOperand(index, &addressing_mode)};
  InstructionOperand outputs[] = {
      (rep == MachineRepresentation::kWord8)
          // Using DefineSameAsFirst requires the register to be unallocated.
          ? g.DefineAsFixed(node, edx)
          : g.DefineSameAsFirst(node)};
  InstructionCode code = opcode | AddressingModeField::encode(addressing_mode);
  selector->Emit(code, 1, outputs, arraysize(inputs), inputs);
}

void VisitStoreCommon(InstructionSelector* selector, Node* node,
                      StoreRepresentation store_rep,
                      base::Optional<AtomicMemoryOrder> atomic_order) {
  IA32OperandGenerator g(selector);
  Node* base = node->InputAt(0);
  Node* index = node->InputAt(1);
  Node* value = node->InputAt(2);

  WriteBarrierKind write_barrier_kind = store_rep.write_barrier_kind();
  MachineRepresentation rep = store_rep.representation();
  const bool is_seqcst =
      atomic_order && *atomic_order == AtomicMemoryOrder::kSeqCst;

  if (FLAG_enable_unconditional_write_barriers && CanBeTaggedPointer(rep)) {
    write_barrier_kind = kFullWriteBarrier;
  }

  if (write_barrier_kind != kNoWriteBarrier && !FLAG_disable_write_barriers) {
    DCHECK(CanBeTaggedPointer(rep));
    AddressingMode addressing_mode;
    InstructionOperand inputs[] = {
        g.UseUniqueRegister(base),
        g.GetEffectiveIndexOperand(index, &addressing_mode),
        g.UseUniqueRegister(value)};
    RecordWriteMode record_write_mode =
        WriteBarrierKindToRecordWriteMode(write_barrier_kind);
    InstructionOperand temps[] = {g.TempRegister(), g.TempRegister()};
    size_t const temp_count = arraysize(temps);
    InstructionCode code = is_seqcst ? kArchAtomicStoreWithWriteBarrier
                                     : kArchStoreWithWriteBarrier;
    code |= AddressingModeField::encode(addressing_mode);
    code |= MiscField::encode(static_cast<int>(record_write_mode));
    selector->Emit(code, 0, nullptr, arraysize(inputs), inputs, temp_count,
                   temps);
  } else if (is_seqcst) {
    VisitAtomicExchange(selector, node, GetSeqCstStoreOpcode(rep), rep);
  } else {
    // Release and non-atomic stores emit MOV.
    // https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html

    InstructionOperand val;
    if (g.CanBeImmediate(value)) {
      val = g.UseImmediate(value);
    } else if (!atomic_order && (rep == MachineRepresentation::kWord8 ||
                                 rep == MachineRepresentation::kBit)) {
      val = g.UseByteRegister(value);
    } else {
      val = g.UseRegister(value);
    }

    InstructionOperand inputs[4];
    size_t input_count = 0;
    AddressingMode addressing_mode =
        g.GetEffectiveAddressMemoryOperand(node, inputs, &input_count);
    InstructionCode code =
        GetStoreOpcode(rep) | AddressingModeField::encode(addressing_mode);
    inputs[input_count++] = val;
    selector->Emit(code, 0, static_cast<InstructionOperand*>(nullptr),
                   input_count, inputs);
  }
}

}  // namespace

void InstructionSelector::VisitStore(Node* node) {
  VisitStoreCommon(this, node, StoreRepresentationOf(node->op()),
                   base::nullopt);
}

void InstructionSelector::VisitProtectedStore(Node* node) {
  // TODO(eholk)
  UNIMPLEMENTED();
}

void InstructionSelector::VisitStoreLane(Node* node) {
  IA32OperandGenerator g(this);

  StoreLaneParameters params = StoreLaneParametersOf(node->op());
  InstructionCode opcode = kArchNop;
  if (params.rep == MachineRepresentation::kWord8) {
    opcode = kIA32Pextrb;
  } else if (params.rep == MachineRepresentation::kWord16) {
    opcode = kIA32Pextrw;
  } else if (params.rep == MachineRepresentation::kWord32) {
    opcode = kIA32S128Store32Lane;
  } else if (params.rep == MachineRepresentation::kWord64) {
    if (params.laneidx == 0) {
      opcode = kIA32Movlps;
    } else {
      DCHECK_EQ(1, params.laneidx);
      opcode = kIA32Movhps;
    }
  } else {
    UNREACHABLE();
  }

  InstructionOperand inputs[4];
  size_t input_count = 0;
  AddressingMode addressing_mode =
      g.GetEffectiveAddressMemoryOperand(node, inputs, &input_count);
  opcode |= AddressingModeField::encode(addressing_mode);

  InstructionOperand value_operand = g.UseRegister(node->InputAt(2));
  inputs[input_count++] = value_operand;
  inputs[input_count++] = g.UseImmediate(params.laneidx);
  DCHECK_GE(4, input_count);
  Emit(opcode, 0, nullptr, input_count, inputs);
}

// Architecture supports unaligned access, therefore VisitLoad is used instead
void InstructionSelector::VisitUnalignedLoad(Node* node) { UNREACHABLE(); }

// Architecture supports unaligned access, therefore VisitStore is used instead
void InstructionSelector::VisitUnalignedStore(Node* node) { UNREACHABLE(); }

namespace {

// Shared routine for multiple binary operations.
void VisitBinop(InstructionSelector* selector, Node* node,
                InstructionCode opcode, FlagsContinuation* cont) {
  IA32OperandGenerator g(selector);
  Int32BinopMatcher m(node);
  Node* left = m.left().node();
  Node* right = m.right().node();
  InstructionOperand inputs[6];
  size_t input_count = 0;
  InstructionOperand outputs[1];
  size_t output_count = 0;

  // TODO(turbofan): match complex addressing modes.
  if (left == right) {
    // If both inputs refer to the same operand, enforce allocating a register
    // for both of them to ensure that we don't end up generating code like
    // this:
    //
    //   mov eax, [ebp-0x10]
    //   add eax, [ebp-0x10]
    //   jo label
    InstructionOperand const input = g.UseRegister(left);
    inputs[input_count++] = input;
    inputs[input_count++] = input;
  } else if (g.CanBeImmediate(right)) {
    inputs[input_count++] = g.UseRegister(left);
    inputs[input_count++] = g.UseImmediate(right);
  } else {
    int effect_level = selector->GetEffectLevel(node, cont);
    if (node->op()->HasProperty(Operator::kCommutative) &&
        g.CanBeBetterLeftOperand(right) &&
        (!g.CanBeBetterLeftOperand(left) ||
         !g.CanBeMemoryOperand(opcode, node, right, effect_level))) {
      std::swap(left, right);
    }
    if (g.CanBeMemoryOperand(opcode, node, right, effect_level)) {
      inputs[input_count++] = g.UseRegister(left);
      AddressingMode addressing_mode =
          g.GetEffectiveAddressMemoryOperand(right, inputs, &input_count);
      opcode |= AddressingModeField::encode(addressing_mode);
    } else {
      inputs[input_count++] = g.UseRegister(left);
      inputs[input_count++] = g.Use(right);
    }
  }

  outputs[output_count++] = g.DefineSameAsFirst(node);

  DCHECK_NE(0u, input_count);
  DCHECK_EQ(1u, output_count);
  DCHECK_GE(arraysize(inputs), input_count);
  DCHECK_GE(arraysize(outputs), output_count);

  selector->EmitWithContinuation(opcode, output_count, outputs, input_count,
                                 inputs, cont);
}

// Shared routine for multiple binary operations.
void VisitBinop(InstructionSelector* selector, Node* node,
                InstructionCode opcode) {
  FlagsContinuation cont;
  VisitBinop(selector, node, opcode, &cont);
}

}  // namespace

void InstructionSelector::VisitWord32And(Node* node) {
  VisitBinop(this, node, kIA32And);
}

void InstructionSelector::VisitWord32Or(Node* node) {
  VisitBinop(this, node, kIA32Or);
}

void InstructionSelector::VisitWord32Xor(Node* node) {
  IA32OperandGenerator g(this);
  Int32BinopMatcher m(node);
  if (m.right().Is(-1)) {
    Emit(kIA32Not, g.DefineSameAsFirst(node), g.UseRegister(m.left().node()));
  } else {
    VisitBinop(this, node, kIA32Xor);
  }
}

void InstructionSelector::VisitStackPointerGreaterThan(
    Node* node, FlagsContinuation* cont) {
  StackCheckKind kind = StackCheckKindOf(node->op());
  InstructionCode opcode =
      kArchStackPointerGreaterThan | MiscField::encode(static_cast<int>(kind));

  int effect_level = GetEffectLevel(node, cont);

  IA32OperandGenerator g(this);

  // No outputs.
  InstructionOperand* const outputs = nullptr;
  const int output_count = 0;

  // Applying an offset to this stack check requires a temp register. Offsets
  // are only applied to the first stack check. If applying an offset, we must
  // ensure the input and temp registers do not alias, thus kUniqueRegister.
  InstructionOperand temps[] = {g.TempRegister()};
  const int temp_count = (kind == StackCheckKind::kJSFunctionEntry) ? 1 : 0;
  const auto register_mode = (kind == StackCheckKind::kJSFunctionEntry)
                                 ? OperandGenerator::kUniqueRegister
                                 : OperandGenerator::kRegister;

  Node* const value = node->InputAt(0);
  if (g.CanBeMemoryOperand(kIA32Cmp, node, value, effect_level)) {
    DCHECK(value->opcode() == IrOpcode::kLoad ||
           value->opcode() == IrOpcode::kLoadImmutable);

    // GetEffectiveAddressMemoryOperand can create at most 3 inputs.
    static constexpr int kMaxInputCount = 3;

    size_t input_count = 0;
    InstructionOperand inputs[kMaxInputCount];
    AddressingMode addressing_mode = g.GetEffectiveAddressMemoryOperand(
        value, inputs, &input_count, register_mode);
    opcode |= AddressingModeField::encode(addressing_mode);
    DCHECK_LE(input_count, kMaxInputCount);

    EmitWithContinuation(opcode, output_count, outputs, input_count, inputs,
                         temp_count, temps, cont);
  } else {
    InstructionOperand inputs[] = {g.UseRegisterWithMode(value, register_mode)};
    static constexpr int input_count = arraysize(inputs);
    EmitWithContinuation(opcode, output_count, outputs, input_count, inputs,
                         temp_count, temps, cont);
  }
}

// Shared routine for multiple shift operations.
static inline void VisitShift(InstructionSelector* selector, Node* node,
                              ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  Node* left = node->InputAt(0);
  Node* right = node->InputAt(1);

  if (g.CanBeImmediate(right)) {
    selector->Emit(opcode, g.DefineSameAsFirst(node), g.UseRegister(left),
                   g.UseImmediate(right));
  } else {
    selector->Emit(opcode, g.DefineSameAsFirst(node), g.UseRegister(left),
                   g.UseFixed(right, ecx));
  }
}

namespace {

void VisitMulHigh(InstructionSelector* selector, Node* node,
                  ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand temps[] = {g.TempRegister(eax)};
  selector->Emit(
      opcode, g.DefineAsFixed(node, edx), g.UseFixed(node->InputAt(0), eax),
      g.UseUniqueRegister(node->InputAt(1)), arraysize(temps), temps);
}

void VisitDiv(InstructionSelector* selector, Node* node, ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand temps[] = {g.TempRegister(edx)};
  selector->Emit(opcode, g.DefineAsFixed(node, eax),
                 g.UseFixed(node->InputAt(0), eax),
                 g.UseUnique(node->InputAt(1)), arraysize(temps), temps);
}

void VisitMod(InstructionSelector* selector, Node* node, ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  InstructionOperand temps[] = {g.TempRegister(eax)};
  selector->Emit(opcode, g.DefineAsFixed(node, edx),
                 g.UseFixed(node->InputAt(0), eax),
                 g.UseUnique(node->InputAt(1)), arraysize(temps), temps);
}

void EmitLea(InstructionSelector* selector, Node* result, Node* index,
             int scale, Node* base, Node* displacement,
             DisplacementMode displacement_mode) {
  IA32OperandGenerator g(selector);
  InstructionOperand inputs[4];
  size_t input_count = 0;
  AddressingMode mode =
      g.GenerateMemoryOperandInputs(index, scale, base, displacement,
                                    displacement_mode, inputs, &input_count);

  DCHECK_NE(0u, input_count);
  DCHECK_GE(arraysize(inputs), input_count);

  InstructionOperand outputs[1];
  outputs[0] = g.DefineAsRegister(result);

  InstructionCode opcode = AddressingModeField::encode(mode) | kIA32Lea;

  selector->Emit(opcode, 1, outputs, input_count, inputs);
}

}  // namespace

void InstructionSelector::VisitWord32Shl(Node* node) {
  Int32ScaleMatcher m(node, true);
  if (m.matches()) {
    Node* index = node->InputAt(0);
    Node* base = m.power_of_two_plus_one() ? index : nullptr;
    EmitLea(this, node, index, m.scale(), base, nullptr, kPositiveDisplacement);
    return;
  }
  VisitShift(this, node, kIA32Shl);
}

void InstructionSelector::VisitWord32Shr(Node* node) {
  VisitShift(this, node, kIA32Shr);
}

void InstructionSelector::VisitWord32Sar(Node* node) {
  VisitShift(this, node, kIA32Sar);
}

void InstructionSelector::VisitInt32PairAdd(Node* node) {
  IA32OperandGenerator g(this);

  Node* projection1 = NodeProperties::FindProjection(node, 1);
  if (projection1) {
    // We use UseUniqueRegister here to avoid register sharing with the temp
    // register.
    InstructionOperand inputs[] = {
        g.UseRegister(node->InputAt(0)),
        g.UseUniqueRegisterOrSlotOrConstant(node->InputAt(1)),
        g.UseRegister(node->InputAt(2)), g.UseUniqueRegister(node->InputAt(3))};

    InstructionOperand outputs[] = {g.DefineSameAsFirst(node),
                                    g.DefineAsRegister(projection1)};

    InstructionOperand temps[] = {g.TempRegister()};

    Emit(kIA32AddPair, 2, outputs, 4, inputs, 1, temps);
  } else {
    // The high word of the result is not used, so we emit the standard 32 bit
    // instruction.
    Emit(kIA32Add, g.DefineSameAsFirst(node), g.UseRegister(node->InputAt(0)),
         g.Use(node->InputAt(2)));
  }
}

void InstructionSelector::VisitInt32PairSub(Node* node) {
  IA32OperandGenerator g(this);

  Node* projection1 = NodeProperties::FindProjection(node, 1);
  if (projection1) {
    // We use UseUniqueRegister here to avoid register sharing with the temp
    // register.
    InstructionOperand inputs[] = {
        g.UseRegister(node->InputAt(0)),
        g.UseUniqueRegisterOrSlotOrConstant(node->InputAt(1)),
        g.UseRegister(node->InputAt(2)), g.UseUniqueRegister(node->InputAt(3))};

    InstructionOperand outputs[] = {g.DefineSameAsFirst(node),
                                    g.DefineAsRegister(projection1)};

    InstructionOperand temps[] = {g.TempRegister()};

    Emit(kIA32SubPair, 2, outputs, 4, inputs, 1, temps);
  } else {
    // The high word of the result is not used, so we emit the standard 32 bit
    // instruction.
    Emit(kIA32Sub, g.DefineSameAsFirst(node), g.UseRegister(node->InputAt(0)),
         g.Use(node->InputAt(2)));
  }
}

void InstructionSelector::VisitInt32PairMul(Node* node) {
  IA32OperandGenerator g(this);

  Node* projection1 = NodeProperties::FindProjection(node, 1);
  if (projection1) {
    // InputAt(3) explicitly shares ecx with OutputRegister(1) to save one
    // register and one mov instruction.
    InstructionOperand inputs[] = {
        g.UseUnique(node->InputAt(0)),
        g.UseUniqueRegisterOrSlotOrConstant(node->InputAt(1)),
        g.UseUniqueRegister(node->InputAt(2)),
        g.UseFixed(node->InputAt(3), ecx)};

    InstructionOperand outputs[] = {
        g.DefineAsFixed(node, eax),
        g.DefineAsFixed(NodeProperties::FindProjection(node, 1), ecx)};

    InstructionOperand temps[] = {g.TempRegister(edx)};

    Emit(kIA32MulPair, 2, outputs, 4, inputs, 1, temps);
  } else {
    // The high word of the result is not used, so we emit the standard 32 bit
    // instruction.
    Emit(kIA32Imul, g.DefineSameAsFirst(node), g.UseRegister(node->InputAt(0)),
         g.Use(node->InputAt(2)));
  }
}

void VisitWord32PairShift(InstructionSelector* selector, InstructionCode opcode,
                          Node* node) {
  IA32OperandGenerator g(selector);

  Node* shift = node->InputAt(2);
  InstructionOperand shift_operand;
  if (g.CanBeImmediate(shift)) {
    shift_operand = g.UseImmediate(shift);
  } else {
    shift_operand = g.UseFixed(shift, ecx);
  }
  InstructionOperand inputs[] = {g.UseFixed(node->InputAt(0), eax),
                                 g.UseFixed(node->InputAt(1), edx),
                                 shift_operand};

  InstructionOperand outputs[2];
  InstructionOperand temps[1];
  int32_t output_count = 0;
  int32_t temp_count = 0;
  outputs[output_count++] = g.DefineAsFixed(node, eax);
  Node* projection1 = NodeProperties::FindProjection(node, 1);
  if (projection1) {
    outputs[output_count++] = g.DefineAsFixed(projection1, edx);
  } else {
    temps[temp_count++] = g.TempRegister(edx);
  }

  selector->Emit(opcode, output_count, outputs, 3, inputs, temp_count, temps);
}

void InstructionSelector::VisitWord32PairShl(Node* node) {
  VisitWord32PairShift(this, kIA32ShlPair, node);
}

void InstructionSelector::VisitWord32PairShr(Node* node) {
  VisitWord32PairShift(this, kIA32ShrPair, node);
}

void InstructionSelector::VisitWord32PairSar(Node* node) {
  VisitWord32PairShift(this, kIA32SarPair, node);
}

void InstructionSelector::VisitWord32Rol(Node* node) {
  VisitShift(this, node, kIA32Rol);
}

void InstructionSelector::VisitWord32Ror(Node* node) {
  VisitShift(this, node, kIA32Ror);
}

#define RO_OP_LIST(V)                                        \
  V(Word32Clz, kIA32Lzcnt)                                   \
  V(Word32Ctz, kIA32Tzcnt)                                   \
  V(Word32Popcnt, kIA32Popcnt)                               \
  V(ChangeFloat32ToFloat64, kIA32Float32ToFloat64)           \
  V(RoundInt32ToFloat32, kSSEInt32ToFloat32)                 \
  V(ChangeInt32ToFloat64, kSSEInt32ToFloat64)                \
  V(TruncateFloat32ToInt32, kIA32Float32ToInt32)             \
  V(ChangeFloat64ToInt32, kIA32Float64ToInt32)               \
  V(TruncateFloat64ToFloat32, kIA32Float64ToFloat32)         \
  V(RoundFloat64ToInt32, kIA32Float64ToInt32)                \
  V(BitcastFloat32ToInt32, kIA32BitcastFI)                   \
  V(BitcastInt32ToFloat32, kIA32BitcastIF)                   \
  V(Float32Sqrt, kIA32Float32Sqrt)                           \
  V(Float64Sqrt, kIA32Float64Sqrt)                           \
  V(Float64ExtractLowWord32, kIA32Float64ExtractLowWord32)   \
  V(Float64ExtractHighWord32, kIA32Float64ExtractHighWord32) \
  V(SignExtendWord8ToInt32, kIA32Movsxbl)                    \
  V(SignExtendWord16ToInt32, kIA32Movsxwl)                   \
  V(F64x2Sqrt, kIA32F64x2Sqrt)

#define RO_WITH_TEMP_OP_LIST(V) V(ChangeUint32ToFloat64, kIA32Uint32ToFloat64)

#define RO_WITH_TEMP_SIMD_OP_LIST(V)               \
  V(TruncateFloat32ToUint32, kIA32Float32ToUint32) \
  V(ChangeFloat64ToUint32, kIA32Float64ToUint32)   \
  V(TruncateFloat64ToUint32, kIA32Float64ToUint32)

#define RR_OP_LIST(V)                                                          \
  V(TruncateFloat64ToWord32, kArchTruncateDoubleToI)                           \
  V(Float32RoundDown, kIA32Float32Round | MiscField::encode(kRoundDown))       \
  V(Float64RoundDown, kIA32Float64Round | MiscField::encode(kRoundDown))       \
  V(Float32RoundUp, kIA32Float32Round | MiscField::encode(kRoundUp))           \
  V(Float64RoundUp, kIA32Float64Round | MiscField::encode(kRoundUp))           \
  V(Float32RoundTruncate, kIA32Float32Round | MiscField::encode(kRoundToZero)) \
  V(Float64RoundTruncate, kIA32Float64Round | MiscField::encode(kRoundToZero)) \
  V(Float32RoundTiesEven,                                                      \
    kIA32Float32Round | MiscField::encode(kRoundToNearest))                    \
  V(Float64RoundTiesEven,                                                      \
    kIA32Float64Round | MiscField::encode(kRoundToNearest))                    \
  V(F32x4Ceil, kIA32F32x4Round | MiscField::encode(kRoundUp))                  \
  V(F32x4Floor, kIA32F32x4Round | MiscField::encode(kRoundDown))               \
  V(F32x4Trunc, kIA32F32x4Round | MiscField::encode(kRoundToZero))             \
  V(F32x4NearestInt, kIA32F32x4Round | MiscField::encode(kRoundToNearest))     \
  V(F64x2Ceil, kIA32F64x2Round | MiscField::encode(kRoundUp))                  \
  V(F64x2Floor, kIA32F64x2Round | MiscField::encode(kRoundDown))               \
  V(F64x2Trunc, kIA32F64x2Round | MiscField::encode(kRoundToZero))             \
  V(F64x2NearestInt, kIA32F64x2Round | MiscField::encode(kRoundToNearest))

#define RRO_FLOAT_OP_LIST(V) \
  V(Float32Add, kFloat32Add) \
  V(Float64Add, kFloat64Add) \
  V(Float32Sub, kFloat32Sub) \
  V(Float64Sub, kFloat64Sub) \
  V(Float32Mul, kFloat32Mul) \
  V(Float64Mul, kFloat64Mul) \
  V(Float32Div, kFloat32Div) \
  V(Float64Div, kFloat64Div) \
  V(F64x2Add, kIA32F64x2Add) \
  V(F64x2Sub, kIA32F64x2Sub) \
  V(F64x2Mul, kIA32F64x2Mul) \
  V(F64x2Div, kIA32F64x2Div) \
  V(F64x2Eq, kIA32F64x2Eq)   \
  V(F64x2Ne, kIA32F64x2Ne)   \
  V(F64x2Lt, kIA32F64x2Lt)   \
  V(F64x2Le, kIA32F64x2Le)

#define FLOAT_UNOP_LIST(V)   \
  V(Float32Abs, kFloat32Abs) \
  V(Float64Abs, kFloat64Abs) \
  V(Float32Neg, kFloat32Neg) \
  V(Float64Neg, kFloat64Neg) \
  V(F32x4Abs, kFloat32Abs)   \
  V(F32x4Neg, kFloat32Neg)   \
  V(F64x2Abs, kFloat64Abs)   \
  V(F64x2Neg, kFloat64Neg)

#define RO_VISITOR(Name, opcode)                      \
  void InstructionSelector::Visit##Name(Node* node) { \
    VisitRO(this, node, opcode);                      \
  }
RO_OP_LIST(RO_VISITOR)
#undef RO_VISITOR
#undef RO_OP_LIST

#define RO_WITH_TEMP_VISITOR(Name, opcode)            \
  void InstructionSelector::Visit##Name(Node* node) { \
    VisitROWithTemp(this, node, opcode);              \
  }
RO_WITH_TEMP_OP_LIST(RO_WITH_TEMP_VISITOR)
#undef RO_WITH_TEMP_VISITOR
#undef RO_WITH_TEMP_OP_LIST

#define RO_WITH_TEMP_SIMD_VISITOR(Name, opcode)       \
  void InstructionSelector::Visit##Name(Node* node) { \
    VisitROWithTempSimd(this, node, opcode);          \
  }
RO_WITH_TEMP_SIMD_OP_LIST(RO_WITH_TEMP_SIMD_VISITOR)
#undef RO_WITH_TEMP_SIMD_VISITOR
#undef RO_WITH_TEMP_SIMD_OP_LIST

#define RR_VISITOR(Name, opcode)                      \
  void InstructionSelector::Visit##Name(Node* node) { \
    VisitRR(this, node, opcode);                      \
  }
RR_OP_LIST(RR_VISITOR)
#undef RR_VISITOR
#undef RR_OP_LIST

#define RRO_FLOAT_VISITOR(Name, opcode)               \
  void InstructionSelector::Visit##Name(Node* node) { \
    VisitRROFloat(this, node, opcode);                \
  }
RRO_FLOAT_OP_LIST(RRO_FLOAT_VISITOR)
#undef RRO_FLOAT_VISITOR
#undef RRO_FLOAT_OP_LIST

#define FLOAT_UNOP_VISITOR(Name, opcode)                  \
  void InstructionSelector::Visit##Name(Node* node) {     \
    VisitFloatUnop(this, node, node->InputAt(0), opcode); \
  }
FLOAT_UNOP_LIST(FLOAT_UNOP_VISITOR)
#undef FLOAT_UNOP_VISITOR
#undef FLOAT_UNOP_LIST

void InstructionSelector::VisitWord32ReverseBits(Node* node) { UNREACHABLE(); }

void InstructionSelector::VisitWord64ReverseBytes(Node* node) { UNREACHABLE(); }

void InstructionSelector::VisitWord32ReverseBytes(Node* node) {
  IA32OperandGenerator g(this);
  Emit(kIA32Bswap, g.DefineSameAsFirst(node), g.UseRegister(node->InputAt(0)));
}

void InstructionSelector::VisitSimd128ReverseBytes(Node* node) {
  UNREACHABLE();
}

void InstructionSelector::VisitInt32Add(Node* node) {
  IA32OperandGenerator g(this);

  // Try to match the Add to a lea pattern
  BaseWithIndexAndDisplacement32Matcher m(node);
  if (m.matches() &&
      (m.displacement() == nullptr || g.CanBeImmediate(m.displacement()))) {
    InstructionOperand inputs[4];
    size_t input_count = 0;
    AddressingMode mode = g.GenerateMemoryOperandInputs(
        m.index(), m.scale(), m.base(), m.displacement(), m.displacement_mode(),
        inputs, &input_count);

    DCHECK_NE(0u, input_count);
    DCHECK_GE(arraysize(inputs), input_count);

    InstructionOperand outputs[1];
    outputs[0] = g.DefineAsRegister(node);

    InstructionCode opcode = AddressingModeField::encode(mode) | kIA32Lea;
    Emit(opcode, 1, outputs, input_count, inputs);
    return;
  }

  // No lea pattern match, use add
  VisitBinop(this, node, kIA32Add);
}

void InstructionSelector::VisitInt32Sub(Node* node) {
  IA32OperandGenerator g(this);
  Int32BinopMatcher m(node);
  if (m.left().Is(0)) {
    Emit(kIA32Neg, g.DefineSameAsFirst(node), g.Use(m.right().node()));
  } else {
    VisitBinop(this, node, kIA32Sub);
  }
}

void InstructionSelector::VisitInt32Mul(Node* node) {
  Int32ScaleMatcher m(node, true);
  if (m.matches()) {
    Node* index = node->InputAt(0);
    Node* base = m.power_of_two_plus_one() ? index : nullptr;
    EmitLea(this, node, index, m.scale(), base, nullptr, kPositiveDisplacement);
    return;
  }
  IA32OperandGenerator g(this);
  Node* left = node->InputAt(0);
  Node* right = node->InputAt(1);
  if (g.CanBeImmediate(right)) {
    Emit(kIA32Imul, g.DefineAsRegister(node), g.Use(left),
         g.UseImmediate(right));
  } else {
    if (g.CanBeBetterLeftOperand(right)) {
      std::swap(left, right);
    }
    Emit(kIA32Imul, g.DefineSameAsFirst(node), g.UseRegister(left),
         g.Use(right));
  }
}

void InstructionSelector::VisitInt32MulHigh(Node* node) {
  VisitMulHigh(this, node, kIA32ImulHigh);
}

void InstructionSelector::VisitUint32MulHigh(Node* node) {
  VisitMulHigh(this, node, kIA32UmulHigh);
}

void InstructionSelector::VisitInt32Div(Node* node) {
  VisitDiv(this, node, kIA32Idiv);
}

void InstructionSelector::VisitUint32Div(Node* node) {
  VisitDiv(this, node, kIA32Udiv);
}

void InstructionSelector::VisitInt32Mod(Node* node) {
  VisitMod(this, node, kIA32Idiv);
}

void InstructionSelector::VisitUint32Mod(Node* node) {
  VisitMod(this, node, kIA32Udiv);
}

void InstructionSelector::VisitRoundUint32ToFloat32(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  Emit(kIA32Uint32ToFloat32, g.DefineAsRegister(node), g.Use(node->InputAt(0)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitFloat64Mod(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister(eax), g.TempRegister()};
  Emit(kIA32Float64Mod, g.DefineSameAsFirst(node),
       g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitFloat32Max(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  Emit(kIA32Float32Max, g.DefineSameAsFirst(node),
       g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitFloat64Max(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  Emit(kIA32Float64Max, g.DefineSameAsFirst(node),
       g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitFloat32Min(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  Emit(kIA32Float32Min, g.DefineSameAsFirst(node),
       g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitFloat64Min(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  Emit(kIA32Float64Min, g.DefineSameAsFirst(node),
       g.UseRegister(node->InputAt(0)), g.Use(node->InputAt(1)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitFloat64RoundTiesAway(Node* node) {
  UNREACHABLE();
}

void InstructionSelector::VisitFloat64Ieee754Binop(Node* node,
                                                   InstructionCode opcode) {
  IA32OperandGenerator g(this);
  Emit(opcode, g.DefineSameAsFirst(node), g.UseRegister(node->InputAt(0)),
       g.UseRegister(node->InputAt(1)))
      ->MarkAsCall();
}

void InstructionSelector::VisitFloat64Ieee754Unop(Node* node,
                                                  InstructionCode opcode) {
  IA32OperandGenerator g(this);
  Emit(opcode, g.DefineSameAsFirst(node), g.UseRegister(node->InputAt(0)))
      ->MarkAsCall();
}

void InstructionSelector::EmitPrepareArguments(
    ZoneVector<PushParameter>* arguments, const CallDescriptor* call_descriptor,
    Node* node) {
  IA32OperandGenerator g(this);

  // Prepare for C function call.
  if (call_descriptor->IsCFunctionCall()) {
    InstructionOperand temps[] = {g.TempRegister()};
    size_t const temp_count = arraysize(temps);
    Emit(kArchPrepareCallCFunction | MiscField::encode(static_cast<int>(
                                         call_descriptor->ParameterCount())),
         0, nullptr, 0, nullptr, temp_count, temps);

    // Poke any stack arguments.
    for (size_t n = 0; n < arguments->size(); ++n) {
      PushParameter input = (*arguments)[n];
      if (input.node) {
        int const slot = static_cast<int>(n);
        InstructionOperand value = g.CanBeImmediate(node)
                                       ? g.UseImmediate(input.node)
                                       : g.UseRegister(input.node);
        Emit(kIA32Poke | MiscField::encode(slot), g.NoOutput(), value);
      }
    }
  } else {
    // Push any stack arguments.
    int effect_level = GetEffectLevel(node);
    int stack_decrement = 0;
    for (PushParameter input : base::Reversed(*arguments)) {
      stack_decrement += kSystemPointerSize;
      // Skip holes in the param array. These represent both extra slots for
      // multi-slot values and padding slots for alignment.
      if (input.node == nullptr) continue;
      InstructionOperand decrement = g.UseImmediate(stack_decrement);
      stack_decrement = 0;
      if (g.CanBeImmediate(input.node)) {
        Emit(kIA32Push, g.NoOutput(), decrement, g.UseImmediate(input.node));
      } else if (IsSupported(INTEL_ATOM) ||
                 sequence()->IsFP(GetVirtualRegister(input.node))) {
        // TODO(bbudge): IA32Push cannot handle stack->stack double moves
        // because there is no way to encode fixed double slots.
        Emit(kIA32Push, g.NoOutput(), decrement, g.UseRegister(input.node));
      } else if (g.CanBeMemoryOperand(kIA32Push, node, input.node,
                                      effect_level)) {
        InstructionOperand outputs[1];
        InstructionOperand inputs[5];
        size_t input_count = 0;
        inputs[input_count++] = decrement;
        AddressingMode mode = g.GetEffectiveAddressMemoryOperand(
            input.node, inputs, &input_count);
        InstructionCode opcode = kIA32Push | AddressingModeField::encode(mode);
        Emit(opcode, 0, outputs, input_count, inputs);
      } else {
        Emit(kIA32Push, g.NoOutput(), decrement, g.UseAny(input.node));
      }
    }
  }
}

void InstructionSelector::EmitPrepareResults(
    ZoneVector<PushParameter>* results, const CallDescriptor* call_descriptor,
    Node* node) {
  IA32OperandGenerator g(this);

  for (PushParameter output : *results) {
    if (!output.location.IsCallerFrameSlot()) continue;
    // Skip any alignment holes in nodes.
    if (output.node != nullptr) {
      DCHECK(!call_descriptor->IsCFunctionCall());
      if (output.location.GetType() == MachineType::Float32()) {
        MarkAsFloat32(output.node);
      } else if (output.location.GetType() == MachineType::Float64()) {
        MarkAsFloat64(output.node);
      } else if (output.location.GetType() == MachineType::Simd128()) {
        MarkAsSimd128(output.node);
      }
      int offset = call_descriptor->GetOffsetToReturns();
      int reverse_slot = -output.location.GetLocation() - offset;
      Emit(kIA32Peek, g.DefineAsRegister(output.node),
           g.UseImmediate(reverse_slot));
    }
  }
}

bool InstructionSelector::IsTailCallAddressImmediate() { return true; }

namespace {

void VisitCompareWithMemoryOperand(InstructionSelector* selector,
                                   InstructionCode opcode, Node* left,
                                   InstructionOperand right,
                                   FlagsContinuation* cont) {
  DCHECK(left->opcode() == IrOpcode::kLoad ||
         left->opcode() == IrOpcode::kLoadImmutable);
  IA32OperandGenerator g(selector);
  size_t input_count = 0;
  InstructionOperand inputs[4];
  AddressingMode addressing_mode =
      g.GetEffectiveAddressMemoryOperand(left, inputs, &input_count);
  opcode |= AddressingModeField::encode(addressing_mode);
  inputs[input_count++] = right;

  selector->EmitWithContinuation(opcode, 0, nullptr, input_count, inputs, cont);
}

// Shared routine for multiple compare operations.
void VisitCompare(InstructionSelector* selector, InstructionCode opcode,
                  InstructionOperand left, InstructionOperand right,
                  FlagsContinuation* cont) {
  selector->EmitWithContinuation(opcode, left, right, cont);
}

// Shared routine for multiple compare operations.
void VisitCompare(InstructionSelector* selector, InstructionCode opcode,
                  Node* left, Node* right, FlagsContinuation* cont,
                  bool commutative) {
  IA32OperandGenerator g(selector);
  if (commutative && g.CanBeBetterLeftOperand(right)) {
    std::swap(left, right);
  }
  VisitCompare(selector, opcode, g.UseRegister(left), g.Use(right), cont);
}

MachineType MachineTypeForNarrow(Node* node, Node* hint_node) {
  if (hint_node->opcode() == IrOpcode::kLoad ||
      hint_node->opcode() == IrOpcode::kLoadImmutable) {
    MachineType hint = LoadRepresentationOf(hint_node->op());
    if (node->opcode() == IrOpcode::kInt32Constant ||
        node->opcode() == IrOpcode::kInt64Constant) {
      int64_t constant = node->opcode() == IrOpcode::kInt32Constant
                             ? OpParameter<int32_t>(node->op())
                             : OpParameter<int64_t>(node->op());
      if (hint == MachineType::Int8()) {
        if (constant >= std::numeric_limits<int8_t>::min() &&
            constant <= std::numeric_limits<int8_t>::max()) {
          return hint;
        }
      } else if (hint == MachineType::Uint8()) {
        if (constant >= std::numeric_limits<uint8_t>::min() &&
            constant <= std::numeric_limits<uint8_t>::max()) {
          return hint;
        }
      } else if (hint == MachineType::Int16()) {
        if (constant >= std::numeric_limits<int16_t>::min() &&
            constant <= std::numeric_limits<int16_t>::max()) {
          return hint;
        }
      } else if (hint == MachineType::Uint16()) {
        if (constant >= std::numeric_limits<uint16_t>::min() &&
            constant <= std::numeric_limits<uint16_t>::max()) {
          return hint;
        }
      } else if (hint == MachineType::Int32()) {
        return hint;
      } else if (hint == MachineType::Uint32()) {
        if (constant >= 0) return hint;
      }
    }
  }
  return node->opcode() == IrOpcode::kLoad ||
                 node->opcode() == IrOpcode::kLoadImmutable
             ? LoadRepresentationOf(node->op())
             : MachineType::None();
}

// Tries to match the size of the given opcode to that of the operands, if
// possible.
InstructionCode TryNarrowOpcodeSize(InstructionCode opcode, Node* left,
                                    Node* right, FlagsContinuation* cont) {
  // TODO(epertoso): we can probably get some size information out of phi nodes.
  // If the load representations don't match, both operands will be
  // zero/sign-extended to 32bit.
  MachineType left_type = MachineTypeForNarrow(left, right);
  MachineType right_type = MachineTypeForNarrow(right, left);
  if (left_type == right_type) {
    switch (left_type.representation()) {
      case MachineRepresentation::kBit:
      case MachineRepresentation::kWord8: {
        if (opcode == kIA32Test) return kIA32Test8;
        if (opcode == kIA32Cmp) {
          if (left_type.semantic() == MachineSemantic::kUint32) {
            cont->OverwriteUnsignedIfSigned();
          } else {
            CHECK_EQ(MachineSemantic::kInt32, left_type.semantic());
          }
          return kIA32Cmp8;
        }
        break;
      }
      case MachineRepresentation::kWord16:
        if (opcode == kIA32Test) return kIA32Test16;
        if (opcode == kIA32Cmp) {
          if (left_type.semantic() == MachineSemantic::kUint32) {
            cont->OverwriteUnsignedIfSigned();
          } else {
            CHECK_EQ(MachineSemantic::kInt32, left_type.semantic());
          }
          return kIA32Cmp16;
        }
        break;
      default:
        break;
    }
  }
  return opcode;
}

// Shared routine for multiple float32 compare operations (inputs commuted).
void VisitFloat32Compare(InstructionSelector* selector, Node* node,
                         FlagsContinuation* cont) {
  Node* const left = node->InputAt(0);
  Node* const right = node->InputAt(1);
  VisitCompare(selector, kIA32Float32Cmp, right, left, cont, false);
}

// Shared routine for multiple float64 compare operations (inputs commuted).
void VisitFloat64Compare(InstructionSelector* selector, Node* node,
                         FlagsContinuation* cont) {
  Node* const left = node->InputAt(0);
  Node* const right = node->InputAt(1);
  VisitCompare(selector, kIA32Float64Cmp, right, left, cont, false);
}

// Shared routine for multiple word compare operations.
void VisitWordCompare(InstructionSelector* selector, Node* node,
                      InstructionCode opcode, FlagsContinuation* cont) {
  IA32OperandGenerator g(selector);
  Node* left = node->InputAt(0);
  Node* right = node->InputAt(1);

  InstructionCode narrowed_opcode =
      TryNarrowOpcodeSize(opcode, left, right, cont);

  int effect_level = selector->GetEffectLevel(node, cont);

  // If one of the two inputs is an immediate, make sure it's on the right, or
  // if one of the two inputs is a memory operand, make sure it's on the left.
  if ((!g.CanBeImmediate(right) && g.CanBeImmediate(left)) ||
      (g.CanBeMemoryOperand(narrowed_opcode, node, right, effect_level) &&
       !g.CanBeMemoryOperand(narrowed_opcode, node, left, effect_level))) {
    if (!node->op()->HasProperty(Operator::kCommutative)) cont->Commute();
    std::swap(left, right);
  }

  // Match immediates on right side of comparison.
  if (g.CanBeImmediate(right)) {
    if (g.CanBeMemoryOperand(narrowed_opcode, node, left, effect_level)) {
      return VisitCompareWithMemoryOperand(selector, narrowed_opcode, left,
                                           g.UseImmediate(right), cont);
    }
    return VisitCompare(selector, opcode, g.Use(left), g.UseImmediate(right),
                        cont);
  }

  // Match memory operands on left side of comparison.
  if (g.CanBeMemoryOperand(narrowed_opcode, node, left, effect_level)) {
    bool needs_byte_register =
        narrowed_opcode == kIA32Test8 || narrowed_opcode == kIA32Cmp8;
    return VisitCompareWithMemoryOperand(
        selector, narrowed_opcode, left,
        needs_byte_register ? g.UseByteRegister(right) : g.UseRegister(right),
        cont);
  }

  return VisitCompare(selector, opcode, left, right, cont,
                      node->op()->HasProperty(Operator::kCommutative));
}

void VisitWordCompare(InstructionSelector* selector, Node* node,
                      FlagsContinuation* cont) {
  VisitWordCompare(selector, node, kIA32Cmp, cont);
}

void VisitAtomicBinOp(InstructionSelector* selector, Node* node,
                      ArchOpcode opcode, MachineRepresentation rep) {
  AddressingMode addressing_mode;
  IA32OperandGenerator g(selector);
  Node* base = node->InputAt(0);
  Node* index = node->InputAt(1);
  Node* value = node->InputAt(2);
  InstructionOperand inputs[] = {
      g.UseUniqueRegister(value), g.UseUniqueRegister(base),
      g.GetEffectiveIndexOperand(index, &addressing_mode)};
  InstructionOperand outputs[] = {g.DefineAsFixed(node, eax)};
  InstructionOperand temp[] = {(rep == MachineRepresentation::kWord8)
                                   ? g.UseByteRegister(node)
                                   : g.TempRegister()};
  InstructionCode code = opcode | AddressingModeField::encode(addressing_mode);
  selector->Emit(code, arraysize(outputs), outputs, arraysize(inputs), inputs,
                 arraysize(temp), temp);
}

void VisitPairAtomicBinOp(InstructionSelector* selector, Node* node,
                          ArchOpcode opcode) {
  IA32OperandGenerator g(selector);
  Node* base = node->InputAt(0);
  Node* index = node->InputAt(1);
  Node* value = node->InputAt(2);
  // For Word64 operations, the value input is split into the a high node,
  // and a low node in the int64-lowering phase.
  Node* value_high = node->InputAt(3);

  // Wasm lives in 32-bit address space, so we do not need to worry about
  // base/index lowering. This will need to be fixed for Wasm64.
  AddressingMode addressing_mode;
  InstructionOperand inputs[] = {
      g.UseUniqueRegisterOrSlotOrConstant(value), g.UseFixed(value_high, ecx),
      g.UseUniqueRegister(base),
      g.GetEffectiveIndexOperand(index, &addressing_mode)};
  InstructionCode code = opcode | AddressingModeField::encode(addressing_mode);
  Node* projection0 = NodeProperties::FindProjection(node, 0);
  Node* projection1 = NodeProperties::FindProjection(node, 1);
  InstructionOperand outputs[2];
  size_t output_count = 0;
  InstructionOperand temps[2];
  size_t temp_count = 0;
  if (projection0) {
    outputs[output_count++] = g.DefineAsFixed(projection0, eax);
  } else {
    temps[temp_count++] = g.TempRegister(eax);
  }
  if (projection1) {
    outputs[output_count++] = g.DefineAsFixed(projection1, edx);
  } else {
    temps[temp_count++] = g.TempRegister(edx);
  }
  selector->Emit(code, output_count, outputs, arraysize(inputs), inputs,
                 temp_count, temps);
}

}  // namespace

// Shared routine for word comparison with zero.
void InstructionSelector::VisitWordCompareZero(Node* user, Node* value,
                                               FlagsContinuation* cont) {
  // Try to combine with comparisons against 0 by simply inverting the branch.
  while (value->opcode() == IrOpcode::kWord32Equal && CanCover(user, value)) {
    Int32BinopMatcher m(value);
    if (!m.right().Is(0)) break;

    user = value;
    value = m.left().node();
    cont->Negate();
  }

  if (CanCover(user, value)) {
    switch (value->opcode()) {
      case IrOpcode::kWord32Equal:
        cont->OverwriteAndNegateIfEqual(kEqual);
        return VisitWordCompare(this, value, cont);
      case IrOpcode::kInt32LessThan:
        cont->OverwriteAndNegateIfEqual(kSignedLessThan);
        return VisitWordCompare(this, value, cont);
      case IrOpcode::kInt32LessThanOrEqual:
        cont->OverwriteAndNegateIfEqual(kSignedLessThanOrEqual);
        return VisitWordCompare(this, value, cont);
      case IrOpcode::kUint32LessThan:
        cont->OverwriteAndNegateIfEqual(kUnsignedLessThan);
        return VisitWordCompare(this, value, cont);
      case IrOpcode::kUint32LessThanOrEqual:
        cont->OverwriteAndNegateIfEqual(kUnsignedLessThanOrEqual);
        return VisitWordCompare(this, value, cont);
      case IrOpcode::kFloat32Equal:
        cont->OverwriteAndNegateIfEqual(kUnorderedEqual);
        return VisitFloat32Compare(this, value, cont);
      case IrOpcode::kFloat32LessThan:
        cont->OverwriteAndNegateIfEqual(kUnsignedGreaterThan);
        return VisitFloat32Compare(this, value, cont);
      case IrOpcode::kFloat32LessThanOrEqual:
        cont->OverwriteAndNegateIfEqual(kUnsignedGreaterThanOrEqual);
        return VisitFloat32Compare(this, value, cont);
      case IrOpcode::kFloat64Equal:
        cont->OverwriteAndNegateIfEqual(kUnorderedEqual);
        return VisitFloat64Compare(this, value, cont);
      case IrOpcode::kFloat64LessThan:
        cont->OverwriteAndNegateIfEqual(kUnsignedGreaterThan);
        return VisitFloat64Compare(this, value, cont);
      case IrOpcode::kFloat64LessThanOrEqual:
        cont->OverwriteAndNegateIfEqual(kUnsignedGreaterThanOrEqual);
        return VisitFloat64Compare(this, value, cont);
      case IrOpcode::kProjection:
        // Check if this is the overflow output projection of an
        // <Operation>WithOverflow node.
        if (ProjectionIndexOf(value->op()) == 1u) {
          // We cannot combine the <Operation>WithOverflow with this branch
          // unless the 0th projection (the use of the actual value of the
          // <Operation> is either nullptr, which means there's no use of the
          // actual value, or was already defined, which means it is scheduled
          // *AFTER* this branch).
          Node* const node = value->InputAt(0);
          Node* const result = NodeProperties::FindProjection(node, 0);
          if (result == nullptr || IsDefined(result)) {
            switch (node->opcode()) {
              case IrOpcode::kInt32AddWithOverflow:
                cont->OverwriteAndNegateIfEqual(kOverflow);
                return VisitBinop(this, node, kIA32Add, cont);
              case IrOpcode::kInt32SubWithOverflow:
                cont->OverwriteAndNegateIfEqual(kOverflow);
                return VisitBinop(this, node, kIA32Sub, cont);
              case IrOpcode::kInt32MulWithOverflow:
                cont->OverwriteAndNegateIfEqual(kOverflow);
                return VisitBinop(this, node, kIA32Imul, cont);
              default:
                break;
            }
          }
        }
        break;
      case IrOpcode::kInt32Sub:
        return VisitWordCompare(this, value, cont);
      case IrOpcode::kWord32And:
        return VisitWordCompare(this, value, kIA32Test, cont);
      case IrOpcode::kStackPointerGreaterThan:
        cont->OverwriteAndNegateIfEqual(kStackPointerGreaterThanCondition);
        return VisitStackPointerGreaterThan(value, cont);
      default:
        break;
    }
  }

  // Continuation could not be combined with a compare, emit compare against 0.
  IA32OperandGenerator g(this);
  VisitCompare(this, kIA32Cmp, g.Use(value), g.TempImmediate(0), cont);
}

void InstructionSelector::VisitSwitch(Node* node, const SwitchInfo& sw) {
  IA32OperandGenerator g(this);
  InstructionOperand value_operand = g.UseRegister(node->InputAt(0));

  // Emit either ArchTableSwitch or ArchBinarySearchSwitch.
  if (enable_switch_jump_table_ == kEnableSwitchJumpTable) {
    static const size_t kMaxTableSwitchValueRange = 2 << 16;
    size_t table_space_cost = 4 + sw.value_range();
    size_t table_time_cost = 3;
    size_t lookup_space_cost = 3 + 2 * sw.case_count();
    size_t lookup_time_cost = sw.case_count();
    if (sw.case_count() > 4 &&
        table_space_cost + 3 * table_time_cost <=
            lookup_space_cost + 3 * lookup_time_cost &&
        sw.min_value() > std::numeric_limits<int32_t>::min() &&
        sw.value_range() <= kMaxTableSwitchValueRange) {
      InstructionOperand index_operand = value_operand;
      if (sw.min_value()) {
        index_operand = g.TempRegister();
        Emit(kIA32Lea | AddressingModeField::encode(kMode_MRI), index_operand,
             value_operand, g.TempImmediate(-sw.min_value()));
      }
      // Generate a table lookup.
      return EmitTableSwitch(sw, index_operand);
    }
  }

  // Generate a tree of conditional jumps.
  return EmitBinarySearchSwitch(sw, value_operand);
}

void InstructionSelector::VisitWord32Equal(Node* const node) {
  FlagsContinuation cont = FlagsContinuation::ForSet(kEqual, node);
  Int32BinopMatcher m(node);
  if (m.right().Is(0)) {
    return VisitWordCompareZero(m.node(), m.left().node(), &cont);
  }
  VisitWordCompare(this, node, &cont);
}

void InstructionSelector::VisitInt32LessThan(Node* node) {
  FlagsContinuation cont = FlagsContinuation::ForSet(kSignedLessThan, node);
  VisitWordCompare(this, node, &cont);
}

void InstructionSelector::VisitInt32LessThanOrEqual(Node* node) {
  FlagsContinuation cont =
      FlagsContinuation::ForSet(kSignedLessThanOrEqual, node);
  VisitWordCompare(this, node, &cont);
}

void InstructionSelector::VisitUint32LessThan(Node* node) {
  FlagsContinuation cont = FlagsContinuation::ForSet(kUnsignedLessThan, node);
  VisitWordCompare(this, node, &cont);
}

void InstructionSelector::VisitUint32LessThanOrEqual(Node* node) {
  FlagsContinuation cont =
      FlagsContinuation::ForSet(kUnsignedLessThanOrEqual, node);
  VisitWordCompare(this, node, &cont);
}

void InstructionSelector::VisitInt32AddWithOverflow(Node* node) {
  if (Node* ovf = NodeProperties::FindProjection(node, 1)) {
    FlagsContinuation cont = FlagsContinuation::ForSet(kOverflow, ovf);
    return VisitBinop(this, node, kIA32Add, &cont);
  }
  FlagsContinuation cont;
  VisitBinop(this, node, kIA32Add, &cont);
}

void InstructionSelector::VisitInt32SubWithOverflow(Node* node) {
  if (Node* ovf = NodeProperties::FindProjection(node, 1)) {
    FlagsContinuation cont = FlagsContinuation::ForSet(kOverflow, ovf);
    return VisitBinop(this, node, kIA32Sub, &cont);
  }
  FlagsContinuation cont;
  VisitBinop(this, node, kIA32Sub, &cont);
}

void InstructionSelector::VisitInt32MulWithOverflow(Node* node) {
  if (Node* ovf = NodeProperties::FindProjection(node, 1)) {
    FlagsContinuation cont = FlagsContinuation::ForSet(kOverflow, ovf);
    return VisitBinop(this, node, kIA32Imul, &cont);
  }
  FlagsContinuation cont;
  VisitBinop(this, node, kIA32Imul, &cont);
}

void InstructionSelector::VisitFloat32Equal(Node* node) {
  FlagsContinuation cont = FlagsContinuation::ForSet(kUnorderedEqual, node);
  VisitFloat32Compare(this, node, &cont);
}

void InstructionSelector::VisitFloat32LessThan(Node* node) {
  FlagsContinuation cont =
      FlagsContinuation::ForSet(kUnsignedGreaterThan, node);
  VisitFloat32Compare(this, node, &cont);
}

void InstructionSelector::VisitFloat32LessThanOrEqual(Node* node) {
  FlagsContinuation cont =
      FlagsContinuation::ForSet(kUnsignedGreaterThanOrEqual, node);
  VisitFloat32Compare(this, node, &cont);
}

void InstructionSelector::VisitFloat64Equal(Node* node) {
  FlagsContinuation cont = FlagsContinuation::ForSet(kUnorderedEqual, node);
  VisitFloat64Compare(this, node, &cont);
}

void InstructionSelector::VisitFloat64LessThan(Node* node) {
  FlagsContinuation cont =
      FlagsContinuation::ForSet(kUnsignedGreaterThan, node);
  VisitFloat64Compare(this, node, &cont);
}

void InstructionSelector::VisitFloat64LessThanOrEqual(Node* node) {
  FlagsContinuation cont =
      FlagsContinuation::ForSet(kUnsignedGreaterThanOrEqual, node);
  VisitFloat64Compare(this, node, &cont);
}

void InstructionSelector::VisitFloat64InsertLowWord32(Node* node) {
  IA32OperandGenerator g(this);
  Node* left = node->InputAt(0);
  Node* right = node->InputAt(1);
  Float64Matcher mleft(left);
  if (mleft.HasResolvedValue() &&
      (bit_cast<uint64_t>(mleft.ResolvedValue()) >> 32) == 0u) {
    Emit(kIA32Float64LoadLowWord32, g.DefineAsRegister(node), g.Use(right));
    return;
  }
  Emit(kIA32Float64InsertLowWord32, g.DefineSameAsFirst(node),
       g.UseRegister(left), g.Use(right));
}

void InstructionSelector::VisitFloat64InsertHighWord32(Node* node) {
  IA32OperandGenerator g(this);
  Node* left = node->InputAt(0);
  Node* right = node->InputAt(1);
  Emit(kIA32Float64InsertHighWord32, g.DefineSameAsFirst(node),
       g.UseRegister(left), g.Use(right));
}

void InstructionSelector::VisitFloat64SilenceNaN(Node* node) {
  IA32OperandGenerator g(this);
  Emit(kIA32Float64SilenceNaN, g.DefineSameAsFirst(node),
       g.UseRegister(node->InputAt(0)));
}

void InstructionSelector::VisitMemoryBarrier(Node* node) {
  IA32OperandGenerator g(this);
  Emit(kIA32MFence, g.NoOutput());
}

void InstructionSelector::VisitWord32AtomicLoad(Node* node) {
  AtomicLoadParameters atomic_load_params = AtomicLoadParametersOf(node->op());
  LoadRepresentation load_rep = atomic_load_params.representation();
  DCHECK(load_rep.representation() == MachineRepresentation::kWord8 ||
         load_rep.representation() == MachineRepresentation::kWord16 ||
         load_rep.representation() == MachineRepresentation::kWord32 ||
         load_rep.representation() == MachineRepresentation::kTaggedSigned ||
         load_rep.representation() == MachineRepresentation::kTaggedPointer ||
         load_rep.representation() == MachineRepresentation::kTagged);
  USE(load_rep);
  // The memory order is ignored as both acquire and sequentially consistent
  // loads can emit MOV.
  // https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
  VisitLoad(node, node, GetLoadOpcode(load_rep));
}

void InstructionSelector::VisitWord32AtomicStore(Node* node) {
  AtomicStoreParameters store_params = AtomicStoreParametersOf(node->op());
  VisitStoreCommon(this, node, store_params.store_representation(),
                   store_params.order());
}

void InstructionSelector::VisitWord32AtomicExchange(Node* node) {
  IA32OperandGenerator g(this);
  MachineType type = AtomicOpType(node->op());
  ArchOpcode opcode;
  if (type == MachineType::Int8()) {
    opcode = kAtomicExchangeInt8;
  } else if (type == MachineType::Uint8()) {
    opcode = kAtomicExchangeUint8;
  } else if (type == MachineType::Int16()) {
    opcode = kAtomicExchangeInt16;
  } else if (type == MachineType::Uint16()) {
    opcode = kAtomicExchangeUint16;
  } else if (type == MachineType::Int32() || type == MachineType::Uint32()) {
    opcode = kAtomicExchangeWord32;
  } else {
    UNREACHABLE();
  }
  VisitAtomicExchange(this, node, opcode, type.representation());
}

void InstructionSelector::VisitWord32AtomicCompareExchange(Node* node) {
  IA32OperandGenerator g(this);
  Node* base = node->InputAt(0);
  Node* index = node->InputAt(1);
  Node* old_value = node->InputAt(2);
  Node* new_value = node->InputAt(3);

  MachineType type = AtomicOpType(node->op());
  ArchOpcode opcode;
  if (type == MachineType::Int8()) {
    opcode = kAtomicCompareExchangeInt8;
  } else if (type == MachineType::Uint8()) {
    opcode = kAtomicCompareExchangeUint8;
  } else if (type == MachineType::Int16()) {
    opcode = kAtomicCompareExchangeInt16;
  } else if (type == MachineType::Uint16()) {
    opcode = kAtomicCompareExchangeUint16;
  } else if (type == MachineType::Int32() || type == MachineType::Uint32()) {
    opcode = kAtomicCompareExchangeWord32;
  } else {
    UNREACHABLE();
  }
  AddressingMode addressing_mode;
  InstructionOperand new_val_operand =
      (type.representation() == MachineRepresentation::kWord8)
          ? g.UseByteRegister(new_value)
          : g.UseUniqueRegister(new_value);
  InstructionOperand inputs[] = {
      g.UseFixed(old_value, eax), new_val_operand, g.UseUniqueRegister(base),
      g.GetEffectiveIndexOperand(index, &addressing_mode)};
  InstructionOperand outputs[] = {g.DefineAsFixed(node, eax)};
  InstructionCode code = opcode | AddressingModeField::encode(addressing_mode);
  Emit(code, 1, outputs, arraysize(inputs), inputs);
}

void InstructionSelector::VisitWord32AtomicBinaryOperation(
    Node* node, ArchOpcode int8_op, ArchOpcode uint8_op, ArchOpcode int16_op,
    ArchOpcode uint16_op, ArchOpcode word32_op) {
  MachineType type = AtomicOpType(node->op());
  ArchOpcode opcode;
  if (type == MachineType::Int8()) {
    opcode = int8_op;
  } else if (type == MachineType::Uint8()) {
    opcode = uint8_op;
  } else if (type == MachineType::Int16()) {
    opcode = int16_op;
  } else if (type == MachineType::Uint16()) {
    opcode = uint16_op;
  } else if (type == MachineType::Int32() || type == MachineType::Uint32()) {
    opcode = word32_op;
  } else {
    UNREACHABLE();
  }
  VisitAtomicBinOp(this, node, opcode, type.representation());
}

#define VISIT_ATOMIC_BINOP(op)                                           \
  void InstructionSelector::VisitWord32Atomic##op(Node* node) {          \
    VisitWord32AtomicBinaryOperation(                                    \
        node, kAtomic##op##Int8, kAtomic##op##Uint8, kAtomic##op##Int16, \
        kAtomic##op##Uint16, kAtomic##op##Word32);                       \
  }
VISIT_ATOMIC_BINOP(Add)
VISIT_ATOMIC_BINOP(Sub)
VISIT_ATOMIC_BINOP(And)
VISIT_ATOMIC_BINOP(Or)
VISIT_ATOMIC_BINOP(Xor)
#undef VISIT_ATOMIC_BINOP

void InstructionSelector::VisitWord32AtomicPairLoad(Node* node) {
  // Both acquire and sequentially consistent loads can emit MOV.
  // https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html
  IA32OperandGenerator g(this);
  AddressingMode mode;
  Node* base = node->InputAt(0);
  Node* index = node->InputAt(1);
  Node* projection0 = NodeProperties::FindProjection(node, 0);
  Node* projection1 = NodeProperties::FindProjection(node, 1);
  if (projection0 && projection1) {
    InstructionOperand inputs[] = {g.UseUniqueRegister(base),
                                   g.GetEffectiveIndexOperand(index, &mode)};
    InstructionCode code =
        kIA32Word32AtomicPairLoad | AddressingModeField::encode(mode);
    InstructionOperand outputs[] = {g.DefineAsRegister(projection0),
                                    g.DefineAsRegister(projection1)};
    Emit(code, 2, outputs, 2, inputs);
  } else if (projection0 || projection1) {
    // Only one word is needed, so it's enough to load just that.
    ArchOpcode opcode = kIA32Movl;

    InstructionOperand outputs[] = {
        g.DefineAsRegister(projection0 ? projection0 : projection1)};
    InstructionOperand inputs[3];
    size_t input_count = 0;
    // TODO(ahaas): Introduce an enum for {scale} instead of an integer.
    // {scale = 0} means *1 in the generated code.
    int scale = 0;
    AddressingMode mode = g.GenerateMemoryOperandInputs(
        index, scale, base, projection0 ? 0 : 4, kPositiveDisplacement, inputs,
        &input_count);
    InstructionCode code = opcode | AddressingModeField::encode(mode);
    Emit(code, 1, outputs, input_count, inputs);
  }
}

void InstructionSelector::VisitWord32AtomicPairStore(Node* node) {
  // Release pair stores emit a MOVQ via a double register, and sequentially
  // consistent stores emit CMPXCHG8B.
  // https://www.cl.cam.ac.uk/~pes20/cpp/cpp0xmappings.html

  IA32OperandGenerator g(this);
  Node* base = node->InputAt(0);
  Node* index = node->InputAt(1);
  Node* value = node->InputAt(2);
  Node* value_high = node->InputAt(3);

  AtomicMemoryOrder order = OpParameter<AtomicMemoryOrder>(node->op());
  if (order == AtomicMemoryOrder::kAcqRel) {
    AddressingMode addressing_mode;
    InstructionOperand inputs[] = {
        g.UseUniqueRegisterOrSlotOrConstant(value),
        g.UseUniqueRegisterOrSlotOrConstant(value_high),
        g.UseUniqueRegister(base),
        g.GetEffectiveIndexOperand(index, &addressing_mode),
    };
    InstructionCode code = kIA32Word32ReleasePairStore |
                           AddressingModeField::encode(addressing_mode);
    Emit(code, 0, nullptr, arraysize(inputs), inputs);
  } else {
    DCHECK_EQ(order, AtomicMemoryOrder::kSeqCst);

    AddressingMode addressing_mode;
    InstructionOperand inputs[] = {
        g.UseUniqueRegisterOrSlotOrConstant(value), g.UseFixed(value_high, ecx),
        g.UseUniqueRegister(base),
        g.GetEffectiveIndexOperand(index, &addressing_mode)};
    // Allocating temp registers here as stores are performed using an atomic
    // exchange, the output of which is stored in edx:eax, which should be saved
    // and restored at the end of the instruction.
    InstructionOperand temps[] = {g.TempRegister(eax), g.TempRegister(edx)};
    const int num_temps = arraysize(temps);
    InstructionCode code = kIA32Word32SeqCstPairStore |
                           AddressingModeField::encode(addressing_mode);
    Emit(code, 0, nullptr, arraysize(inputs), inputs, num_temps, temps);
  }
}

void InstructionSelector::VisitWord32AtomicPairAdd(Node* node) {
  VisitPairAtomicBinOp(this, node, kIA32Word32AtomicPairAdd);
}

void InstructionSelector::VisitWord32AtomicPairSub(Node* node) {
  VisitPairAtomicBinOp(this, node, kIA32Word32AtomicPairSub);
}

void InstructionSelector::VisitWord32AtomicPairAnd(Node* node) {
  VisitPairAtomicBinOp(this, node, kIA32Word32AtomicPairAnd);
}

void InstructionSelector::VisitWord32AtomicPairOr(Node* node) {
  VisitPairAtomicBinOp(this, node, kIA32Word32AtomicPairOr);
}

void InstructionSelector::VisitWord32AtomicPairXor(Node* node) {
  VisitPairAtomicBinOp(this, node, kIA32Word32AtomicPairXor);
}

void InstructionSelector::VisitWord32AtomicPairExchange(Node* node) {
  VisitPairAtomicBinOp(this, node, kIA32Word32AtomicPairExchange);
}

void InstructionSelector::VisitWord32AtomicPairCompareExchange(Node* node) {
  IA32OperandGenerator g(this);
  Node* index = node->InputAt(1);
  AddressingMode addressing_mode;

  InstructionOperand inputs[] = {
      // High, Low values of old value
      g.UseFixed(node->InputAt(2), eax), g.UseFixed(node->InputAt(3), edx),
      // High, Low values of new value
      g.UseUniqueRegisterOrSlotOrConstant(node->InputAt(4)),
      g.UseFixed(node->InputAt(5), ecx),
      // InputAt(0) => base
      g.UseUniqueRegister(node->InputAt(0)),
      g.GetEffectiveIndexOperand(index, &addressing_mode)};
  Node* projection0 = NodeProperties::FindProjection(node, 0);
  Node* projection1 = NodeProperties::FindProjection(node, 1);
  InstructionCode code = kIA32Word32AtomicPairCompareExchange |
                         AddressingModeField::encode(addressing_mode);

  InstructionOperand outputs[2];
  size_t output_count = 0;
  InstructionOperand temps[2];
  size_t temp_count = 0;
  if (projection0) {
    outputs[output_count++] = g.DefineAsFixed(projection0, eax);
  } else {
    temps[temp_count++] = g.TempRegister(eax);
  }
  if (projection1) {
    outputs[output_count++] = g.DefineAsFixed(projection1, edx);
  } else {
    temps[temp_count++] = g.TempRegister(edx);
  }
  Emit(code, output_count, outputs, arraysize(inputs), inputs, temp_count,
       temps);
}

#define SIMD_INT_TYPES(V) \
  V(I32x4)                \
  V(I16x8)                \
  V(I8x16)

#define SIMD_BINOP_LIST(V) \
  V(I32x4GtU)              \
  V(I32x4GeU)              \
  V(I16x8Ne)               \
  V(I16x8GeS)              \
  V(I16x8GtU)              \
  V(I16x8GeU)              \
  V(I8x16Ne)               \
  V(I8x16GeS)              \
  V(I8x16GtU)              \
  V(I8x16GeU)

#define SIMD_BINOP_UNIFIED_SSE_AVX_LIST(V) \
  V(F32x4Add)                              \
  V(F32x4Sub)                              \
  V(F32x4Mul)                              \
  V(F32x4Div)                              \
  V(F32x4Eq)                               \
  V(F32x4Ne)                               \
  V(F32x4Lt)                               \
  V(F32x4Le)                               \
  V(F32x4Min)                              \
  V(F32x4Max)                              \
  V(I64x2Add)                              \
  V(I64x2Sub)                              \
  V(I64x2Eq)                               \
  V(I64x2Ne)                               \
  V(I32x4Add)                              \
  V(I32x4Sub)                              \
  V(I32x4Mul)                              \
  V(I32x4MinS)                             \
  V(I32x4MaxS)                             \
  V(I32x4Eq)                               \
  V(I32x4Ne)                               \
  V(I32x4GtS)                              \
  V(I32x4GeS)                              \
  V(I32x4MinU)                             \
  V(I32x4MaxU)                             \
  V(I32x4DotI16x8S)                        \
  V(I16x8Add)                              \
  V(I16x8AddSatS)                          \
  V(I16x8Sub)                              \
  V(I16x8SubSatS)                          \
  V(I16x8Mul)                              \
  V(I16x8Eq)                               \
  V(I16x8GtS)                              \
  V(I16x8MinS)                             \
  V(I16x8MaxS)                             \
  V(I16x8AddSatU)                          \
  V(I16x8SubSatU)                          \
  V(I16x8MinU)                             \
  V(I16x8MaxU)                             \
  V(I16x8SConvertI32x4)                    \
  V(I16x8UConvertI32x4)                    \
  V(I16x8RoundingAverageU)                 \
  V(I8x16Add)                              \
  V(I8x16AddSatS)                          \
  V(I8x16Sub)                              \
  V(I8x16SubSatS)                          \
  V(I8x16MinS)                             \
  V(I8x16MaxS)                             \
  V(I8x16Eq)                               \
  V(I8x16GtS)                              \
  V(I8x16AddSatU)                          \
  V(I8x16SubSatU)                          \
  V(I8x16MinU)                             \
  V(I8x16MaxU)                             \
  V(I8x16SConvertI16x8)                    \
  V(I8x16UConvertI16x8)                    \
  V(I8x16RoundingAverageU)                 \
  V(S128And)                               \
  V(S128Or)                                \
  V(S128Xor)

// These opcodes require all inputs to be registers because the codegen is
// simpler with all registers.
#define SIMD_BINOP_RRR(V)  \
  V(I64x2ExtMulLowI32x4S)  \
  V(I64x2ExtMulHighI32x4S) \
  V(I64x2ExtMulLowI32x4U)  \
  V(I64x2ExtMulHighI32x4U) \
  V(I32x4ExtMulLowI16x8S)  \
  V(I32x4ExtMulHighI16x8S) \
  V(I32x4ExtMulLowI16x8U)  \
  V(I32x4ExtMulHighI16x8U) \
  V(I16x8ExtMulLowI8x16S)  \
  V(I16x8ExtMulHighI8x16S) \
  V(I16x8ExtMulLowI8x16U)  \
  V(I16x8ExtMulHighI8x16U) \
  V(I16x8Q15MulRSatS)

#define SIMD_UNOP_LIST(V)   \
  V(F64x2ConvertLowI32x4S)  \
  V(F32x4DemoteF64x2Zero)   \
  V(F32x4Sqrt)              \
  V(F32x4SConvertI32x4)     \
  V(F32x4RecipApprox)       \
  V(F32x4RecipSqrtApprox)   \
  V(I64x2BitMask)           \
  V(I64x2SConvertI32x4Low)  \
  V(I64x2SConvertI32x4High) \
  V(I64x2UConvertI32x4Low)  \
  V(I64x2UConvertI32x4High) \
  V(I32x4SConvertI16x8Low)  \
  V(I32x4SConvertI16x8High) \
  V(I32x4Neg)               \
  V(I32x4UConvertI16x8Low)  \
  V(I32x4UConvertI16x8High) \
  V(I32x4Abs)               \
  V(I32x4BitMask)           \
  V(I16x8SConvertI8x16Low)  \
  V(I16x8SConvertI8x16High) \
  V(I16x8Neg)               \
  V(I16x8UConvertI8x16Low)  \
  V(I16x8UConvertI8x16High) \
  V(I16x8Abs)               \
  V(I8x16Neg)               \
  V(I8x16Abs)               \
  V(I8x16BitMask)           \
  V(S128Not)

#define SIMD_ALLTRUE_LIST(V) \
  V(I64x2AllTrue)            \
  V(I32x4AllTrue)            \
  V(I16x8AllTrue)            \
  V(I8x16AllTrue)

#define SIMD_SHIFT_OPCODES_UNIFED_SSE_AVX(V) \
  V(I64x2Shl)                                \
  V(I64x2ShrU)                               \
  V(I32x4Shl)                                \
  V(I32x4ShrS)                               \
  V(I32x4ShrU)                               \
  V(I16x8Shl)                                \
  V(I16x8ShrS)                               \
  V(I16x8ShrU)

void InstructionSelector::VisitS128Const(Node* node) {
  IA32OperandGenerator g(this);
  static const int kUint32Immediates = kSimd128Size / sizeof(uint32_t);
  uint32_t val[kUint32Immediates];
  memcpy(val, S128ImmediateParameterOf(node->op()).data(), kSimd128Size);
  // If all bytes are zeros or ones, avoid emitting code for generic constants
  bool all_zeros = !(val[0] || val[1] || val[2] || val[3]);
  bool all_ones = val[0] == UINT32_MAX && val[1] == UINT32_MAX &&
                  val[2] == UINT32_MAX && val[3] == UINT32_MAX;
  InstructionOperand dst = g.DefineAsRegister(node);
  if (all_zeros) {
    Emit(kIA32S128Zero, dst);
  } else if (all_ones) {
    Emit(kIA32S128AllOnes, dst);
  } else {
    InstructionOperand inputs[kUint32Immediates];
    for (int i = 0; i < kUint32Immediates; ++i) {
      inputs[i] = g.UseImmediate(val[i]);
    }
    InstructionOperand temp(g.TempRegister());
    Emit(kIA32S128Const, 1, &dst, kUint32Immediates, inputs, 1, &temp);
  }
}

void InstructionSelector::VisitF64x2Min(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
  InstructionOperand operand1 = g.UseRegister(node->InputAt(1));

  if (IsSupported(AVX)) {
    Emit(kIA32F64x2Min, g.DefineAsRegister(node), operand0, operand1);
  } else {
    Emit(kIA32F64x2Min, g.DefineSameAsFirst(node), operand0, operand1);
  }
}

void InstructionSelector::VisitF64x2Max(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
  InstructionOperand operand1 = g.UseRegister(node->InputAt(1));
  if (IsSupported(AVX)) {
    Emit(kIA32F64x2Max, g.DefineAsRegister(node), operand0, operand1);
  } else {
    Emit(kIA32F64x2Max, g.DefineSameAsFirst(node), operand0, operand1);
  }
}

void InstructionSelector::VisitF64x2Splat(Node* node) {
  VisitRRSimd(this, node, kIA32F64x2Splat);
}

void InstructionSelector::VisitF64x2ExtractLane(Node* node) {
  VisitRRISimd(this, node, kIA32F64x2ExtractLane, kIA32F64x2ExtractLane);
}

void InstructionSelector::VisitI64x2SplatI32Pair(Node* node) {
  IA32OperandGenerator g(this);
  Int32Matcher match_left(node->InputAt(0));
  Int32Matcher match_right(node->InputAt(1));
  if (match_left.Is(0) && match_right.Is(0)) {
    Emit(kIA32S128Zero, g.DefineAsRegister(node));
  } else {
    InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
    InstructionOperand operand1 = g.Use(node->InputAt(1));
    Emit(kIA32I64x2SplatI32Pair, g.DefineAsRegister(node), operand0, operand1);
  }
}

void InstructionSelector::VisitI64x2ReplaceLaneI32Pair(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand operand = g.UseRegister(node->InputAt(0));
  InstructionOperand lane = g.UseImmediate(OpParameter<int32_t>(node->op()));
  InstructionOperand low = g.Use(node->InputAt(1));
  InstructionOperand high = g.Use(node->InputAt(2));
  Emit(kIA32I64x2ReplaceLaneI32Pair, g.DefineSameAsFirst(node), operand, lane,
       low, high);
}

void InstructionSelector::VisitI64x2Neg(Node* node) {
  IA32OperandGenerator g(this);
  // If AVX unsupported, make sure dst != src to avoid a move.
  InstructionOperand operand0 = IsSupported(AVX)
                                    ? g.UseRegister(node->InputAt(0))
                                    : g.UseUnique(node->InputAt(0));
  Emit(kIA32I64x2Neg, g.DefineAsRegister(node), operand0);
}

void InstructionSelector::VisitI64x2ShrS(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand dst =
      IsSupported(AVX) ? g.DefineAsRegister(node) : g.DefineSameAsFirst(node);

  if (g.CanBeImmediate(node->InputAt(1))) {
    Emit(kIA32I64x2ShrS, dst, g.UseRegister(node->InputAt(0)),
         g.UseImmediate(node->InputAt(1)));
  } else {
    InstructionOperand temps[] = {g.TempSimd128Register(), g.TempRegister()};
    Emit(kIA32I64x2ShrS, dst, g.UseUniqueRegister(node->InputAt(0)),
         g.UseRegister(node->InputAt(1)), arraysize(temps), temps);
  }
}

void InstructionSelector::VisitI64x2Mul(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempSimd128Register(),
                                g.TempSimd128Register()};
  Emit(kIA32I64x2Mul, g.DefineAsRegister(node),
       g.UseUniqueRegister(node->InputAt(0)),
       g.UseUniqueRegister(node->InputAt(1)), arraysize(temps), temps);
}

void InstructionSelector::VisitF32x4Splat(Node* node) {
  VisitRRSimd(this, node, kIA32F32x4Splat);
}

void InstructionSelector::VisitF32x4ExtractLane(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
  InstructionOperand operand1 =
      g.UseImmediate(OpParameter<int32_t>(node->op()));
  Emit(kIA32F32x4ExtractLane, g.DefineAsRegister(node), operand0, operand1);
}

void InstructionSelector::VisitF32x4UConvertI32x4(Node* node) {
  VisitRRSimd(this, node, kIA32F32x4UConvertI32x4);
}

void InstructionSelector::VisitI32x4SConvertF32x4(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  InstructionOperand dst =
      IsSupported(AVX) ? g.DefineAsRegister(node) : g.DefineSameAsFirst(node);
  Emit(kIA32I32x4SConvertF32x4, dst, g.UseRegister(node->InputAt(0)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitI32x4UConvertF32x4(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempSimd128Register()};
  InstructionCode opcode =
      IsSupported(AVX) ? kAVXI32x4UConvertF32x4 : kSSEI32x4UConvertF32x4;
  Emit(opcode, g.DefineSameAsFirst(node), g.UseRegister(node->InputAt(0)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitS128Zero(Node* node) {
  IA32OperandGenerator g(this);
  Emit(kIA32S128Zero, g.DefineAsRegister(node));
}

void InstructionSelector::VisitS128Select(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand dst =
      IsSupported(AVX) ? g.DefineAsRegister(node) : g.DefineSameAsFirst(node);
  Emit(kIA32S128Select, dst, g.UseRegister(node->InputAt(0)),
       g.UseRegister(node->InputAt(1)), g.UseRegister(node->InputAt(2)));
}

void InstructionSelector::VisitS128AndNot(Node* node) {
  IA32OperandGenerator g(this);
  // andnps a b does ~a & b, but we want a & !b, so flip the input.
  InstructionOperand dst =
      IsSupported(AVX) ? g.DefineAsRegister(node) : g.DefineSameAsFirst(node);
  Emit(kIA32S128AndNot, dst, g.UseRegister(node->InputAt(1)),
       g.UseRegister(node->InputAt(0)));
}

#define VISIT_SIMD_SPLAT(Type)                               \
  void InstructionSelector::Visit##Type##Splat(Node* node) { \
    Int32Matcher int32_matcher(node->InputAt(0));            \
    if (int32_matcher.Is(0)) {                               \
      IA32OperandGenerator g(this);                          \
      Emit(kIA32S128Zero, g.DefineAsRegister(node));         \
    } else {                                                 \
      VisitRO(this, node, kIA32##Type##Splat);               \
    }                                                        \
  }
SIMD_INT_TYPES(VISIT_SIMD_SPLAT)
#undef SIMD_INT_TYPES
#undef VISIT_SIMD_SPLAT

void InstructionSelector::VisitI8x16ExtractLaneU(Node* node) {
  VisitRRISimd(this, node, kIA32Pextrb);
}

void InstructionSelector::VisitI8x16ExtractLaneS(Node* node) {
  VisitRRISimd(this, node, kIA32I8x16ExtractLaneS);
}

void InstructionSelector::VisitI16x8ExtractLaneU(Node* node) {
  VisitRRISimd(this, node, kIA32Pextrw);
}

void InstructionSelector::VisitI16x8ExtractLaneS(Node* node) {
  VisitRRISimd(this, node, kIA32I16x8ExtractLaneS);
}

void InstructionSelector::VisitI32x4ExtractLane(Node* node) {
  VisitRRISimd(this, node, kIA32I32x4ExtractLane);
}

#define SIMD_REPLACE_LANE_TYPE_OP(V) \
  V(I32x4, kIA32Pinsrd)              \
  V(I16x8, kIA32Pinsrw)              \
  V(I8x16, kIA32Pinsrb)              \
  V(F32x4, kIA32Insertps)

#define VISIT_SIMD_REPLACE_LANE(TYPE, OPCODE)                              \
  void InstructionSelector::Visit##TYPE##ReplaceLane(Node* node) {         \
    IA32OperandGenerator g(this);                                          \
    InstructionOperand operand0 = g.UseRegister(node->InputAt(0));         \
    InstructionOperand operand1 =                                          \
        g.UseImmediate(OpParameter<int32_t>(node->op()));                  \
    InstructionOperand operand2 = g.Use(node->InputAt(1));                 \
    InstructionOperand dst = IsSupported(AVX) ? g.DefineAsRegister(node)   \
                                              : g.DefineSameAsFirst(node); \
    Emit(OPCODE, dst, operand0, operand1, operand2);                       \
  }
SIMD_REPLACE_LANE_TYPE_OP(VISIT_SIMD_REPLACE_LANE)
#undef VISIT_SIMD_REPLACE_LANE
#undef SIMD_REPLACE_LANE_TYPE_OP

void InstructionSelector::VisitF64x2ReplaceLane(Node* node) {
  IA32OperandGenerator g(this);
  int32_t lane = OpParameter<int32_t>(node->op());
  // When no-AVX, define dst == src to save a move.
  InstructionOperand dst =
      IsSupported(AVX) ? g.DefineAsRegister(node) : g.DefineSameAsFirst(node);
  Emit(kIA32F64x2ReplaceLane, dst, g.UseRegister(node->InputAt(0)),
       g.UseImmediate(lane), g.UseRegister(node->InputAt(1)));
}

#define VISIT_SIMD_SHIFT_UNIFIED_SSE_AVX(Opcode)        \
  void InstructionSelector::Visit##Opcode(Node* node) { \
    VisitRROSimdShift(this, node, kIA32##Opcode);       \
  }
SIMD_SHIFT_OPCODES_UNIFED_SSE_AVX(VISIT_SIMD_SHIFT_UNIFIED_SSE_AVX)
#undef VISIT_SIMD_SHIFT_UNIFIED_SSE_AVX
#undef SIMD_SHIFT_OPCODES_UNIFED_SSE_AVX

// TODO(v8:9198): SSE requires operand0 to be a register as we don't have memory
// alignment yet. For AVX, memory operands are fine, but can have performance
// issues if not aligned to 16/32 bytes (based on load size), see SDM Vol 1,
// chapter 14.9
#define VISIT_SIMD_UNOP(Opcode)                         \
  void InstructionSelector::Visit##Opcode(Node* node) { \
    IA32OperandGenerator g(this);                       \
    Emit(kIA32##Opcode, g.DefineAsRegister(node),       \
         g.UseRegister(node->InputAt(0)));              \
  }
SIMD_UNOP_LIST(VISIT_SIMD_UNOP)
#undef VISIT_SIMD_UNOP
#undef SIMD_UNOP_LIST

void InstructionSelector::VisitV128AnyTrue(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  Emit(kIA32S128AnyTrue, g.DefineAsRegister(node),
       g.UseRegister(node->InputAt(0)), arraysize(temps), temps);
}

#define VISIT_SIMD_ALLTRUE(Opcode)                                            \
  void InstructionSelector::Visit##Opcode(Node* node) {                       \
    IA32OperandGenerator g(this);                                             \
    InstructionOperand temps[] = {g.TempRegister(), g.TempSimd128Register()}; \
    Emit(kIA32##Opcode, g.DefineAsRegister(node),                             \
         g.UseUniqueRegister(node->InputAt(0)), arraysize(temps), temps);     \
  }
SIMD_ALLTRUE_LIST(VISIT_SIMD_ALLTRUE)
#undef VISIT_SIMD_ALLTRUE
#undef SIMD_ALLTRUE_LIST

#define VISIT_SIMD_BINOP(Opcode)                          \
  void InstructionSelector::Visit##Opcode(Node* node) {   \
    VisitRROSimd(this, node, kAVX##Opcode, kSSE##Opcode); \
  }
SIMD_BINOP_LIST(VISIT_SIMD_BINOP)
#undef VISIT_SIMD_BINOP
#undef SIMD_BINOP_LIST

#define VISIT_SIMD_BINOP_UNIFIED_SSE_AVX(Opcode)            \
  void InstructionSelector::Visit##Opcode(Node* node) {     \
    VisitRROSimd(this, node, kIA32##Opcode, kIA32##Opcode); \
  }
SIMD_BINOP_UNIFIED_SSE_AVX_LIST(VISIT_SIMD_BINOP_UNIFIED_SSE_AVX)
#undef VISIT_SIMD_BINOP_UNIFIED_SSE_AVX
#undef SIMD_BINOP_UNIFIED_SSE_AVX_LIST

#define VISIT_SIMD_BINOP_RRR(OPCODE)                    \
  void InstructionSelector::Visit##OPCODE(Node* node) { \
    VisitRRRSimd(this, node, kIA32##OPCODE);            \
  }
SIMD_BINOP_RRR(VISIT_SIMD_BINOP_RRR)
#undef VISIT_SIMD_BINOP_RRR
#undef SIMD_BINOP_RRR

void InstructionSelector::VisitI16x8BitMask(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempSimd128Register()};
  Emit(kIA32I16x8BitMask, g.DefineAsRegister(node),
       g.UseUniqueRegister(node->InputAt(0)), arraysize(temps), temps);
}

void InstructionSelector::VisitI8x16Shl(Node* node) {
  VisitI8x16Shift(this, node, kIA32I8x16Shl);
}

void InstructionSelector::VisitI8x16ShrS(Node* node) {
  VisitI8x16Shift(this, node, kIA32I8x16ShrS);
}

void InstructionSelector::VisitI8x16ShrU(Node* node) {
  VisitI8x16Shift(this, node, kIA32I8x16ShrU);
}

void InstructionSelector::VisitInt32AbsWithOverflow(Node* node) {
  UNREACHABLE();
}

void InstructionSelector::VisitInt64AbsWithOverflow(Node* node) {
  UNREACHABLE();
}

#if V8_ENABLE_WEBASSEMBLY
namespace {

// Returns true if shuffle can be decomposed into two 16x4 half shuffles
// followed by a 16x8 blend.
// E.g. [3 2 1 0 15 14 13 12].
bool TryMatch16x8HalfShuffle(uint8_t* shuffle16x8, uint8_t* blend_mask) {
  *blend_mask = 0;
  for (int i = 0; i < 8; i++) {
    if ((shuffle16x8[i] & 0x4) != (i & 0x4)) return false;
    *blend_mask |= (shuffle16x8[i] > 7 ? 1 : 0) << i;
  }
  return true;
}

struct ShuffleEntry {
  uint8_t shuffle[kSimd128Size];
  ArchOpcode opcode;
  ArchOpcode avx_opcode;
  bool src0_needs_reg;
  bool src1_needs_reg;
};

// Shuffles that map to architecture-specific instruction sequences. These are
// matched very early, so we shouldn't include shuffles that match better in
// later tests, like 32x4 and 16x8 shuffles. In general, these patterns should
// map to either a single instruction, or be finer grained, such as zip/unzip or
// transpose patterns.
static const ShuffleEntry arch_shuffles[] = {
    {{0, 1, 2, 3, 4, 5, 6, 7, 16, 17, 18, 19, 20, 21, 22, 23},
     kIA32S64x2UnpackLow,
     kIA32S64x2UnpackLow,
     true,
     false},
    {{8, 9, 10, 11, 12, 13, 14, 15, 24, 25, 26, 27, 28, 29, 30, 31},
     kIA32S64x2UnpackHigh,
     kIA32S64x2UnpackHigh,
     true,
     false},
    {{0, 1, 2, 3, 16, 17, 18, 19, 4, 5, 6, 7, 20, 21, 22, 23},
     kIA32S32x4UnpackLow,
     kIA32S32x4UnpackLow,
     true,
     false},
    {{8, 9, 10, 11, 24, 25, 26, 27, 12, 13, 14, 15, 28, 29, 30, 31},
     kIA32S32x4UnpackHigh,
     kIA32S32x4UnpackHigh,
     true,
     false},
    {{0, 1, 16, 17, 2, 3, 18, 19, 4, 5, 20, 21, 6, 7, 22, 23},
     kIA32S16x8UnpackLow,
     kIA32S16x8UnpackLow,
     true,
     false},
    {{8, 9, 24, 25, 10, 11, 26, 27, 12, 13, 28, 29, 14, 15, 30, 31},
     kIA32S16x8UnpackHigh,
     kIA32S16x8UnpackHigh,
     true,
     false},
    {{0, 16, 1, 17, 2, 18, 3, 19, 4, 20, 5, 21, 6, 22, 7, 23},
     kIA32S8x16UnpackLow,
     kIA32S8x16UnpackLow,
     true,
     false},
    {{8, 24, 9, 25, 10, 26, 11, 27, 12, 28, 13, 29, 14, 30, 15, 31},
     kIA32S8x16UnpackHigh,
     kIA32S8x16UnpackHigh,
     true,
     false},

    {{0, 1, 4, 5, 8, 9, 12, 13, 16, 17, 20, 21, 24, 25, 28, 29},
     kSSES16x8UnzipLow,
     kAVXS16x8UnzipLow,
     true,
     false},
    {{2, 3, 6, 7, 10, 11, 14, 15, 18, 19, 22, 23, 26, 27, 30, 31},
     kSSES16x8UnzipHigh,
     kAVXS16x8UnzipHigh,
     true,
     true},
    {{0, 2, 4, 6, 8, 10, 12, 14, 16, 18, 20, 22, 24, 26, 28, 30},
     kSSES8x16UnzipLow,
     kAVXS8x16UnzipLow,
     true,
     true},
    {{1, 3, 5, 7, 9, 11, 13, 15, 17, 19, 21, 23, 25, 27, 29, 31},
     kSSES8x16UnzipHigh,
     kAVXS8x16UnzipHigh,
     true,
     true},

    {{0, 16, 2, 18, 4, 20, 6, 22, 8, 24, 10, 26, 12, 28, 14, 30},
     kSSES8x16TransposeLow,
     kAVXS8x16TransposeLow,
     true,
     true},
    {{1, 17, 3, 19, 5, 21, 7, 23, 9, 25, 11, 27, 13, 29, 15, 31},
     kSSES8x16TransposeHigh,
     kAVXS8x16TransposeHigh,
     true,
     true},
    {{7, 6, 5, 4, 3, 2, 1, 0, 15, 14, 13, 12, 11, 10, 9, 8},
     kSSES8x8Reverse,
     kAVXS8x8Reverse,
     true,
     true},
    {{3, 2, 1, 0, 7, 6, 5, 4, 11, 10, 9, 8, 15, 14, 13, 12},
     kSSES8x4Reverse,
     kAVXS8x4Reverse,
     true,
     true},
    {{1, 0, 3, 2, 5, 4, 7, 6, 9, 8, 11, 10, 13, 12, 15, 14},
     kSSES8x2Reverse,
     kAVXS8x2Reverse,
     true,
     true}};

bool TryMatchArchShuffle(const uint8_t* shuffle, const ShuffleEntry* table,
                         size_t num_entries, bool is_swizzle,
                         const ShuffleEntry** arch_shuffle) {
  uint8_t mask = is_swizzle ? kSimd128Size - 1 : 2 * kSimd128Size - 1;
  for (size_t i = 0; i < num_entries; ++i) {
    const ShuffleEntry& entry = table[i];
    int j = 0;
    for (; j < kSimd128Size; ++j) {
      if ((entry.shuffle[j] & mask) != (shuffle[j] & mask)) {
        break;
      }
    }
    if (j == kSimd128Size) {
      *arch_shuffle = &entry;
      return true;
    }
  }
  return false;
}

}  // namespace

void InstructionSelector::VisitI8x16Shuffle(Node* node) {
  uint8_t shuffle[kSimd128Size];
  bool is_swizzle;
  CanonicalizeShuffle(node, shuffle, &is_swizzle);

  int imm_count = 0;
  static const int kMaxImms = 6;
  uint32_t imms[kMaxImms];
  int temp_count = 0;
  static const int kMaxTemps = 2;
  InstructionOperand temps[kMaxTemps];

  IA32OperandGenerator g(this);
  bool use_avx = CpuFeatures::IsSupported(AVX);
  // AVX and swizzles don't generally need DefineSameAsFirst to avoid a move.
  bool no_same_as_first = use_avx || is_swizzle;
  // We generally need UseRegister for input0, Use for input1.
  // TODO(v8:9198): We don't have 16-byte alignment for SIMD operands yet, but
  // we retain this logic (continue setting these in the various shuffle match
  // clauses), but ignore it when selecting registers or slots.
  bool src0_needs_reg = true;
  bool src1_needs_reg = false;
  ArchOpcode opcode = kIA32I8x16Shuffle;  // general shuffle is the default

  uint8_t offset;
  uint8_t shuffle32x4[4];
  uint8_t shuffle16x8[8];
  int index;
  const ShuffleEntry* arch_shuffle;
  if (wasm::SimdShuffle::TryMatchConcat(shuffle, &offset)) {
    if (wasm::SimdShuffle::TryMatch32x4Rotate(shuffle, shuffle32x4,
                                              is_swizzle)) {
      uint8_t shuffle_mask = wasm::SimdShuffle::PackShuffle4(shuffle32x4);
      opcode = kIA32S32x4Rotate;
      imms[imm_count++] = shuffle_mask;
    } else {
      // Swap inputs from the normal order for (v)palignr.
      SwapShuffleInputs(node);
      is_swizzle = false;  // It's simpler to just handle the general case.
      no_same_as_first = use_avx;  // SSE requires same-as-first.
      opcode = kIA32S8x16Alignr;
      // palignr takes a single imm8 offset.
      imms[imm_count++] = offset;
    }
  } else if (TryMatchArchShuffle(shuffle, arch_shuffles,
                                 arraysize(arch_shuffles), is_swizzle,
                                 &arch_shuffle)) {
    opcode = use_avx ? arch_shuffle->avx_opcode : arch_shuffle->opcode;
    src0_needs_reg = !use_avx || arch_shuffle->src0_needs_reg;
    // SSE can't take advantage of both operands in registers and needs
    // same-as-first.
    src1_needs_reg = use_avx && arch_shuffle->src1_needs_reg;
    no_same_as_first = use_avx;
  } else if (wasm::SimdShuffle::TryMatch32x4Shuffle(shuffle, shuffle32x4)) {
    uint8_t shuffle_mask = wasm::SimdShuffle::PackShuffle4(shuffle32x4);
    if (is_swizzle) {
      if (wasm::SimdShuffle::TryMatchIdentity(shuffle)) {
        // Bypass normal shuffle code generation in this case.
        EmitIdentity(node);
        return;
      } else {
        // pshufd takes a single imm8 shuffle mask.
        opcode = kIA32S32x4Swizzle;
        no_same_as_first = true;
        // TODO(v8:9198): This doesn't strictly require a register, forcing the
        // swizzles to always use registers until generation of incorrect memory
        // operands can be fixed.
        src0_needs_reg = true;
        imms[imm_count++] = shuffle_mask;
      }
    } else {
      // 2 operand shuffle
      // A blend is more efficient than a general 32x4 shuffle; try it first.
      if (wasm::SimdShuffle::TryMatchBlend(shuffle)) {
        opcode = kIA32S16x8Blend;
        uint8_t blend_mask = wasm::SimdShuffle::PackBlend4(shuffle32x4);
        imms[imm_count++] = blend_mask;
      } else {
        opcode = kIA32S32x4Shuffle;
        no_same_as_first = true;
        // TODO(v8:9198): src0 and src1 is used by pshufd in codegen, which
        // requires memory to be 16-byte aligned, since we cannot guarantee that
        // yet, force using a register here.
        src0_needs_reg = true;
        src1_needs_reg = true;
        imms[imm_count++] = shuffle_mask;
        int8_t blend_mask = wasm::SimdShuffle::PackBlend4(shuffle32x4);
        imms[imm_count++] = blend_mask;
      }
    }
  } else if (wasm::SimdShuffle::TryMatch16x8Shuffle(shuffle, shuffle16x8)) {
    uint8_t blend_mask;
    if (wasm::SimdShuffle::TryMatchBlend(shuffle)) {
      opcode = kIA32S16x8Blend;
      blend_mask = wasm::SimdShuffle::PackBlend8(shuffle16x8);
      imms[imm_count++] = blend_mask;
    } else if (wasm::SimdShuffle::TryMatchSplat<8>(shuffle, &index)) {
      opcode = kIA32S16x8Dup;
      src0_needs_reg = false;
      imms[imm_count++] = index;
    } else if (TryMatch16x8HalfShuffle(shuffle16x8, &blend_mask)) {
      opcode = is_swizzle ? kIA32S16x8HalfShuffle1 : kIA32S16x8HalfShuffle2;
      // Half-shuffles don't need DefineSameAsFirst or UseRegister(src0).
      no_same_as_first = true;
      src0_needs_reg = false;
      uint8_t mask_lo = wasm::SimdShuffle::PackShuffle4(shuffle16x8);
      uint8_t mask_hi = wasm::SimdShuffle::PackShuffle4(shuffle16x8 + 4);
      imms[imm_count++] = mask_lo;
      imms[imm_count++] = mask_hi;
      if (!is_swizzle) imms[imm_count++] = blend_mask;
    }
  } else if (wasm::SimdShuffle::TryMatchSplat<16>(shuffle, &index)) {
    opcode = kIA32S8x16Dup;
    no_same_as_first = use_avx;
    src0_needs_reg = true;
    imms[imm_count++] = index;
  }
  if (opcode == kIA32I8x16Shuffle) {
    // Use same-as-first for general swizzle, but not shuffle.
    no_same_as_first = !is_swizzle;
    src0_needs_reg = !no_same_as_first;
    imms[imm_count++] = wasm::SimdShuffle::Pack4Lanes(shuffle);
    imms[imm_count++] = wasm::SimdShuffle::Pack4Lanes(shuffle + 4);
    imms[imm_count++] = wasm::SimdShuffle::Pack4Lanes(shuffle + 8);
    imms[imm_count++] = wasm::SimdShuffle::Pack4Lanes(shuffle + 12);
    temps[temp_count++] = g.TempRegister();
  }

  // Use DefineAsRegister(node) and Use(src0) if we can without forcing an extra
  // move instruction in the CodeGenerator.
  Node* input0 = node->InputAt(0);
  InstructionOperand dst =
      no_same_as_first ? g.DefineAsRegister(node) : g.DefineSameAsFirst(node);
  // TODO(v8:9198): Use src0_needs_reg when we have memory alignment for SIMD.
  InstructionOperand src0 = g.UseRegister(input0);
  USE(src0_needs_reg);

  int input_count = 0;
  InstructionOperand inputs[2 + kMaxImms + kMaxTemps];
  inputs[input_count++] = src0;
  if (!is_swizzle) {
    Node* input1 = node->InputAt(1);
    // TODO(v8:9198): Use src1_needs_reg when we have memory alignment for SIMD.
    inputs[input_count++] = g.UseRegister(input1);
    USE(src1_needs_reg);
  }
  for (int i = 0; i < imm_count; ++i) {
    inputs[input_count++] = g.UseImmediate(imms[i]);
  }
  Emit(opcode, 1, &dst, input_count, inputs, temp_count, temps);
}

void InstructionSelector::VisitI8x16Swizzle(Node* node) {
  InstructionCode op = kIA32I8x16Swizzle;

  bool relaxed = OpParameter<bool>(node->op());
  if (relaxed) {
    op |= MiscField::encode(true);
  } else {
    auto m = V128ConstMatcher(node->InputAt(1));
    if (m.HasResolvedValue()) {
      // If the indices vector is a const, check if they are in range, or if the
      // top bit is set, then we can avoid the paddusb in the codegen and simply
      // emit a pshufb.
      auto imms = m.ResolvedValue().immediate();
      op |= MiscField::encode(wasm::SimdSwizzle::AllInRangeOrTopBitSet(imms));
    }
  }

  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  Emit(op,
       IsSupported(AVX) ? g.DefineAsRegister(node) : g.DefineSameAsFirst(node),
       g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)),
       arraysize(temps), temps);
}
#else
void InstructionSelector::VisitI8x16Shuffle(Node* node) { UNREACHABLE(); }
void InstructionSelector::VisitI8x16Swizzle(Node* node) { UNREACHABLE(); }
#endif  // V8_ENABLE_WEBASSEMBLY

namespace {
void VisitMinOrMax(InstructionSelector* selector, Node* node, ArchOpcode opcode,
                   bool flip_inputs) {
  // Due to the way minps/minpd work, we want the dst to be same as the second
  // input: b = pmin(a, b) directly maps to minps b a.
  IA32OperandGenerator g(selector);
  InstructionOperand dst = selector->IsSupported(AVX)
                               ? g.DefineAsRegister(node)
                               : g.DefineSameAsFirst(node);
  if (flip_inputs) {
    // Due to the way minps/minpd work, we want the dst to be same as the second
    // input: b = pmin(a, b) directly maps to minps b a.
    selector->Emit(opcode, dst, g.UseRegister(node->InputAt(1)),
                   g.UseRegister(node->InputAt(0)));
  } else {
    selector->Emit(opcode, dst, g.UseRegister(node->InputAt(0)),
                   g.UseRegister(node->InputAt(1)));
  }
}
}  // namespace

void InstructionSelector::VisitF32x4Pmin(Node* node) {
  VisitMinOrMax(this, node, kIA32Minps, true);
}

void InstructionSelector::VisitF32x4Pmax(Node* node) {
  VisitMinOrMax(this, node, kIA32Maxps, true);
}

void InstructionSelector::VisitF64x2Pmin(Node* node) {
  VisitMinOrMax(this, node, kIA32Minpd, true);
}

void InstructionSelector::VisitF64x2Pmax(Node* node) {
  VisitMinOrMax(this, node, kIA32Maxpd, true);
}

void InstructionSelector::VisitF32x4RelaxedMin(Node* node) {
  VisitMinOrMax(this, node, kIA32Minps, false);
}

void InstructionSelector::VisitF32x4RelaxedMax(Node* node) {
  VisitMinOrMax(this, node, kIA32Maxps, false);
}

void InstructionSelector::VisitF64x2RelaxedMin(Node* node) {
  VisitMinOrMax(this, node, kIA32Minpd, false);
}

void InstructionSelector::VisitF64x2RelaxedMax(Node* node) {
  VisitMinOrMax(this, node, kIA32Maxpd, false);
}

namespace {
void VisitExtAddPairwise(InstructionSelector* selector, Node* node,
                         ArchOpcode opcode, bool need_temp) {
  IA32OperandGenerator g(selector);
  InstructionOperand operand0 = g.UseRegister(node->InputAt(0));
  InstructionOperand dst = (selector->IsSupported(AVX))
                               ? g.DefineAsRegister(node)
                               : g.DefineSameAsFirst(node);
  if (need_temp) {
    InstructionOperand temps[] = {g.TempRegister()};
    selector->Emit(opcode, dst, operand0, arraysize(temps), temps);
  } else {
    selector->Emit(opcode, dst, operand0);
  }
}
}  // namespace

void InstructionSelector::VisitI32x4ExtAddPairwiseI16x8S(Node* node) {
  VisitExtAddPairwise(this, node, kIA32I32x4ExtAddPairwiseI16x8S, true);
}

void InstructionSelector::VisitI32x4ExtAddPairwiseI16x8U(Node* node) {
  VisitExtAddPairwise(this, node, kIA32I32x4ExtAddPairwiseI16x8U, false);
}

void InstructionSelector::VisitI16x8ExtAddPairwiseI8x16S(Node* node) {
  VisitExtAddPairwise(this, node, kIA32I16x8ExtAddPairwiseI8x16S, true);
}

void InstructionSelector::VisitI16x8ExtAddPairwiseI8x16U(Node* node) {
  VisitExtAddPairwise(this, node, kIA32I16x8ExtAddPairwiseI8x16U, true);
}

void InstructionSelector::VisitI8x16Popcnt(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand dst = CpuFeatures::IsSupported(AVX)
                               ? g.DefineAsRegister(node)
                               : g.DefineAsRegister(node);
  InstructionOperand temps[] = {g.TempSimd128Register(), g.TempRegister()};
  Emit(kIA32I8x16Popcnt, dst, g.UseUniqueRegister(node->InputAt(0)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitF64x2ConvertLowI32x4U(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  InstructionOperand dst =
      IsSupported(AVX) ? g.DefineAsRegister(node) : g.DefineSameAsFirst(node);
  Emit(kIA32F64x2ConvertLowI32x4U, dst, g.UseRegister(node->InputAt(0)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitI32x4TruncSatF64x2SZero(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  if (IsSupported(AVX)) {
    // Requires dst != src.
    Emit(kIA32I32x4TruncSatF64x2SZero, g.DefineAsRegister(node),
         g.UseUniqueRegister(node->InputAt(0)), arraysize(temps), temps);
  } else {
    Emit(kIA32I32x4TruncSatF64x2SZero, g.DefineSameAsFirst(node),
         g.UseRegister(node->InputAt(0)), arraysize(temps), temps);
  }
}

void InstructionSelector::VisitI32x4TruncSatF64x2UZero(Node* node) {
  IA32OperandGenerator g(this);
  InstructionOperand temps[] = {g.TempRegister()};
  InstructionOperand dst =
      IsSupported(AVX) ? g.DefineAsRegister(node) : g.DefineSameAsFirst(node);
  Emit(kIA32I32x4TruncSatF64x2UZero, dst, g.UseRegister(node->InputAt(0)),
       arraysize(temps), temps);
}

void InstructionSelector::VisitI32x4RelaxedTruncF64x2SZero(Node* node) {
  VisitRRSimd(this, node, kIA32Cvttpd2dq);
}

void InstructionSelector::VisitI32x4RelaxedTruncF64x2UZero(Node* node) {
  VisitFloatUnop(this, node, node->InputAt(0), kIA32I32x4TruncF64x2UZero);
}

void InstructionSelector::VisitI32x4RelaxedTruncF32x4S(Node* node) {
  VisitRRSimd(this, node, kIA32Cvttps2dq);
}

void InstructionSelector::VisitI32x4RelaxedTruncF32x4U(Node* node) {
  VisitFloatUnop(this, node, node->InputAt(0), kIA32I32x4TruncF32x4U);
}

void InstructionSelector::VisitI64x2GtS(Node* node) {
  IA32OperandGenerator g(this);
  if (CpuFeatures::IsSupported(AVX)) {
    Emit(kIA32I64x2GtS, g.DefineAsRegister(node),
         g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)));
  } else if (CpuFeatures::IsSupported(SSE4_2)) {
    Emit(kIA32I64x2GtS, g.DefineSameAsFirst(node),
         g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)));
  } else {
    Emit(kIA32I64x2GtS, g.DefineAsRegister(node),
         g.UseUniqueRegister(node->InputAt(0)),
         g.UseUniqueRegister(node->InputAt(1)));
  }
}

void InstructionSelector::VisitI64x2GeS(Node* node) {
  IA32OperandGenerator g(this);
  if (CpuFeatures::IsSupported(AVX)) {
    Emit(kIA32I64x2GeS, g.DefineAsRegister(node),
         g.UseRegister(node->InputAt(0)), g.UseRegister(node->InputAt(1)));
  } else if (CpuFeatures::IsSupported(SSE4_2)) {
    Emit(kIA32I64x2GeS, g.DefineAsRegister(node),
         g.UseUniqueRegister(node->InputAt(0)),
         g.UseRegister(node->InputAt(1)));
  } else {
    Emit(kIA32I64x2GeS, g.DefineAsRegister(node),
         g.UseUniqueRegister(node->InputAt(0)),
         g.UseUniqueRegister(node->InputAt(1)));
  }
}

void InstructionSelector::VisitI64x2Abs(Node* node) {
  VisitRRSimd(this, node, kIA32I64x2Abs, kIA32I64x2Abs);
}

void InstructionSelector::VisitF64x2PromoteLowF32x4(Node* node) {
  IA32OperandGenerator g(this);
  InstructionCode code = kIA32F64x2PromoteLowF32x4;
  Node* input = node->InputAt(0);
  LoadTransformMatcher m(input);

  if (m.Is(LoadTransformation::kS128Load64Zero) && CanCover(node, input)) {
    // Trap handler is not supported on IA32.
    DCHECK_NE(m.ResolvedValue().kind, MemoryAccessKind::kProtected);
    // LoadTransforms cannot be eliminated, so they are visited even if
    // unused. Mark it as defined so that we don't visit it.
    MarkAsDefined(input);
    VisitLoad(node, input, code);
    return;
  }

  VisitRR(this, node, code);
}

namespace {
// pblendvb is a correct implementation for all the various relaxed lane select,
// see https://github.com/WebAssembly/relaxed-simd/issues/17.
void VisitRelaxedLaneSelect(InstructionSelector* selector, Node* node) {
  IA32OperandGenerator g(selector);
  // pblendvb copies src2 when mask is set, opposite from Wasm semantics.
  // node's inputs are: mask, lhs, rhs (determined in wasm-compiler.cc).
  if (selector->IsSupported(AVX)) {
    selector->Emit(kIA32Pblendvb, g.DefineAsRegister(node),
                   g.UseRegister(node->InputAt(2)),
                   g.UseRegister(node->InputAt(1)),
                   g.UseRegister(node->InputAt(0)));
  } else {
    // SSE4.1 pblendvb requires xmm0 to hold the mask as an implicit operand.
    selector->Emit(kIA32Pblendvb, g.DefineSameAsFirst(node),
                   g.UseRegister(node->InputAt(2)),
                   g.UseRegister(node->InputAt(1)),
                   g.UseFixed(node->InputAt(0), xmm0));
  }
}
}  // namespace

void InstructionSelector::VisitI8x16RelaxedLaneSelect(Node* node) {
  VisitRelaxedLaneSelect(this, node);
}
void InstructionSelector::VisitI16x8RelaxedLaneSelect(Node* node) {
  VisitRelaxedLaneSelect(this, node);
}
void InstructionSelector::VisitI32x4RelaxedLaneSelect(Node* node) {
  VisitRelaxedLaneSelect(this, node);
}
void InstructionSelector::VisitI64x2RelaxedLaneSelect(Node* node) {
  VisitRelaxedLaneSelect(this, node);
}

void InstructionSelector::VisitF64x2Qfma(Node* node) {
  VisitRRRR(this, node, kIA32F64x2Qfma);
}

void InstructionSelector::VisitF64x2Qfms(Node* node) {
  VisitRRRR(this, node, kIA32F64x2Qfms);
}

void InstructionSelector::VisitF32x4Qfma(Node* node) {
  VisitRRRR(this, node, kIA32F32x4Qfma);
}

void InstructionSelector::VisitF32x4Qfms(Node* node) {
  VisitRRRR(this, node, kIA32F32x4Qfms);
}

void InstructionSelector::AddOutputToSelectContinuation(OperandGenerator* g,
                                                        int first_input_index,
                                                        Node* node) {
  UNREACHABLE();
}

// static
MachineOperatorBuilder::Flags
InstructionSelector::SupportedMachineOperatorFlags() {
  MachineOperatorBuilder::Flags flags =
      MachineOperatorBuilder::kWord32ShiftIsSafe |
      MachineOperatorBuilder::kWord32Ctz | MachineOperatorBuilder::kWord32Rol;
  if (CpuFeatures::IsSupported(POPCNT)) {
    flags |= MachineOperatorBuilder::kWord32Popcnt;
  }
  if (CpuFeatures::IsSupported(SSE4_1)) {
    flags |= MachineOperatorBuilder::kFloat32RoundDown |
             MachineOperatorBuilder::kFloat64RoundDown |
             MachineOperatorBuilder::kFloat32RoundUp |
             MachineOperatorBuilder::kFloat64RoundUp |
             MachineOperatorBuilder::kFloat32RoundTruncate |
             MachineOperatorBuilder::kFloat64RoundTruncate |
             MachineOperatorBuilder::kFloat32RoundTiesEven |
             MachineOperatorBuilder::kFloat64RoundTiesEven;
  }
  return flags;
}

// static
MachineOperatorBuilder::AlignmentRequirements
InstructionSelector::AlignmentRequirements() {
  return MachineOperatorBuilder::AlignmentRequirements::
      FullUnalignedAccessSupport();
}

}  // namespace compiler
}  // namespace internal
}  // namespace v8