C++程序  |  582行  |  22.42 KB

/*
 * Copyright (C) 2014 The Android Open Source Project
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *      http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#ifndef ART_COMPILER_OPTIMIZING_CODE_GENERATOR_X86_64_H_
#define ART_COMPILER_OPTIMIZING_CODE_GENERATOR_X86_64_H_

#include "arch/x86_64/instruction_set_features_x86_64.h"
#include "code_generator.h"
#include "dex/compiler_enums.h"
#include "driver/compiler_options.h"
#include "nodes.h"
#include "parallel_move_resolver.h"
#include "utils/x86_64/assembler_x86_64.h"

namespace art {
namespace x86_64 {

// Use a local definition to prevent copying mistakes.
static constexpr size_t kX86_64WordSize = kX86_64PointerSize;

// Some x86_64 instructions require a register to be available as temp.
static constexpr Register TMP = R11;

static constexpr Register kParameterCoreRegisters[] = { RSI, RDX, RCX, R8, R9 };
static constexpr FloatRegister kParameterFloatRegisters[] =
    { XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7 };

static constexpr size_t kParameterCoreRegistersLength = arraysize(kParameterCoreRegisters);
static constexpr size_t kParameterFloatRegistersLength = arraysize(kParameterFloatRegisters);

static constexpr Register kRuntimeParameterCoreRegisters[] = { RDI, RSI, RDX, RCX };
static constexpr size_t kRuntimeParameterCoreRegistersLength =
    arraysize(kRuntimeParameterCoreRegisters);
static constexpr FloatRegister kRuntimeParameterFpuRegisters[] = { XMM0, XMM1 };
static constexpr size_t kRuntimeParameterFpuRegistersLength =
    arraysize(kRuntimeParameterFpuRegisters);

// These XMM registers are non-volatile in ART ABI, but volatile in native ABI.
// If the ART ABI changes, this list must be updated.  It is used to ensure that
// these are not clobbered by any direct call to native code (such as math intrinsics).
static constexpr FloatRegister non_volatile_xmm_regs[] = { XMM12, XMM13, XMM14, XMM15 };


class InvokeRuntimeCallingConvention : public CallingConvention<Register, FloatRegister> {
 public:
  InvokeRuntimeCallingConvention()
      : CallingConvention(kRuntimeParameterCoreRegisters,
                          kRuntimeParameterCoreRegistersLength,
                          kRuntimeParameterFpuRegisters,
                          kRuntimeParameterFpuRegistersLength,
                          kX86_64PointerSize) {}

 private:
  DISALLOW_COPY_AND_ASSIGN(InvokeRuntimeCallingConvention);
};

class InvokeDexCallingConvention : public CallingConvention<Register, FloatRegister> {
 public:
  InvokeDexCallingConvention() : CallingConvention(
      kParameterCoreRegisters,
      kParameterCoreRegistersLength,
      kParameterFloatRegisters,
      kParameterFloatRegistersLength,
      kX86_64PointerSize) {}

 private:
  DISALLOW_COPY_AND_ASSIGN(InvokeDexCallingConvention);
};

class FieldAccessCallingConventionX86_64 : public FieldAccessCallingConvention {
 public:
  FieldAccessCallingConventionX86_64() {}

  Location GetObjectLocation() const OVERRIDE {
    return Location::RegisterLocation(RSI);
  }
  Location GetFieldIndexLocation() const OVERRIDE {
    return Location::RegisterLocation(RDI);
  }
  Location GetReturnLocation(Primitive::Type type ATTRIBUTE_UNUSED) const OVERRIDE {
    return Location::RegisterLocation(RAX);
  }
  Location GetSetValueLocation(Primitive::Type type, bool is_instance) const OVERRIDE {
    return Primitive::Is64BitType(type)
        ? Location::RegisterLocation(RDX)
        : (is_instance
            ? Location::RegisterLocation(RDX)
            : Location::RegisterLocation(RSI));
  }
  Location GetFpuLocation(Primitive::Type type ATTRIBUTE_UNUSED) const OVERRIDE {
    return Location::FpuRegisterLocation(XMM0);
  }

 private:
  DISALLOW_COPY_AND_ASSIGN(FieldAccessCallingConventionX86_64);
};


class InvokeDexCallingConventionVisitorX86_64 : public InvokeDexCallingConventionVisitor {
 public:
  InvokeDexCallingConventionVisitorX86_64() {}
  virtual ~InvokeDexCallingConventionVisitorX86_64() {}

  Location GetNextLocation(Primitive::Type type) OVERRIDE;
  Location GetReturnLocation(Primitive::Type type) const OVERRIDE;
  Location GetMethodLocation() const OVERRIDE;

 private:
  InvokeDexCallingConvention calling_convention;

  DISALLOW_COPY_AND_ASSIGN(InvokeDexCallingConventionVisitorX86_64);
};

class CodeGeneratorX86_64;

class ParallelMoveResolverX86_64 : public ParallelMoveResolverWithSwap {
 public:
  ParallelMoveResolverX86_64(ArenaAllocator* allocator, CodeGeneratorX86_64* codegen)
      : ParallelMoveResolverWithSwap(allocator), codegen_(codegen) {}

  void EmitMove(size_t index) OVERRIDE;
  void EmitSwap(size_t index) OVERRIDE;
  void SpillScratch(int reg) OVERRIDE;
  void RestoreScratch(int reg) OVERRIDE;

  X86_64Assembler* GetAssembler() const;

 private:
  void Exchange32(CpuRegister reg, int mem);
  void Exchange32(XmmRegister reg, int mem);
  void Exchange32(int mem1, int mem2);
  void Exchange64(CpuRegister reg1, CpuRegister reg2);
  void Exchange64(CpuRegister reg, int mem);
  void Exchange64(XmmRegister reg, int mem);
  void Exchange64(int mem1, int mem2);

  CodeGeneratorX86_64* const codegen_;

  DISALLOW_COPY_AND_ASSIGN(ParallelMoveResolverX86_64);
};

class LocationsBuilderX86_64 : public HGraphVisitor {
 public:
  LocationsBuilderX86_64(HGraph* graph, CodeGeneratorX86_64* codegen)
      : HGraphVisitor(graph), codegen_(codegen) {}

#define DECLARE_VISIT_INSTRUCTION(name, super)     \
  void Visit##name(H##name* instr) OVERRIDE;

  FOR_EACH_CONCRETE_INSTRUCTION_COMMON(DECLARE_VISIT_INSTRUCTION)
  FOR_EACH_CONCRETE_INSTRUCTION_X86_64(DECLARE_VISIT_INSTRUCTION)

#undef DECLARE_VISIT_INSTRUCTION

  void VisitInstruction(HInstruction* instruction) OVERRIDE {
    LOG(FATAL) << "Unreachable instruction " << instruction->DebugName()
               << " (id " << instruction->GetId() << ")";
  }

 private:
  void HandleInvoke(HInvoke* invoke);
  void HandleBitwiseOperation(HBinaryOperation* operation);
  void HandleCondition(HCondition* condition);
  void HandleShift(HBinaryOperation* operation);
  void HandleFieldSet(HInstruction* instruction, const FieldInfo& field_info);
  void HandleFieldGet(HInstruction* instruction);

  CodeGeneratorX86_64* const codegen_;
  InvokeDexCallingConventionVisitorX86_64 parameter_visitor_;

  DISALLOW_COPY_AND_ASSIGN(LocationsBuilderX86_64);
};

class InstructionCodeGeneratorX86_64 : public InstructionCodeGenerator {
 public:
  InstructionCodeGeneratorX86_64(HGraph* graph, CodeGeneratorX86_64* codegen);

#define DECLARE_VISIT_INSTRUCTION(name, super)     \
  void Visit##name(H##name* instr) OVERRIDE;

  FOR_EACH_CONCRETE_INSTRUCTION_COMMON(DECLARE_VISIT_INSTRUCTION)
  FOR_EACH_CONCRETE_INSTRUCTION_X86_64(DECLARE_VISIT_INSTRUCTION)

#undef DECLARE_VISIT_INSTRUCTION

  void VisitInstruction(HInstruction* instruction) OVERRIDE {
    LOG(FATAL) << "Unreachable instruction " << instruction->DebugName()
               << " (id " << instruction->GetId() << ")";
  }

  X86_64Assembler* GetAssembler() const { return assembler_; }

 private:
  // Generate code for the given suspend check. If not null, `successor`
  // is the block to branch to if the suspend check is not needed, and after
  // the suspend call.
  void GenerateSuspendCheck(HSuspendCheck* instruction, HBasicBlock* successor);
  void GenerateClassInitializationCheck(SlowPathCode* slow_path, CpuRegister class_reg);
  void HandleBitwiseOperation(HBinaryOperation* operation);
  void GenerateRemFP(HRem* rem);
  void DivRemOneOrMinusOne(HBinaryOperation* instruction);
  void DivByPowerOfTwo(HDiv* instruction);
  void GenerateDivRemWithAnyConstant(HBinaryOperation* instruction);
  void GenerateDivRemIntegral(HBinaryOperation* instruction);
  void HandleCondition(HCondition* condition);
  void HandleShift(HBinaryOperation* operation);

  void HandleFieldSet(HInstruction* instruction,
                      const FieldInfo& field_info,
                      bool value_can_be_null);
  void HandleFieldGet(HInstruction* instruction, const FieldInfo& field_info);

  // Generate a heap reference load using one register `out`:
  //
  //   out <- *(out + offset)
  //
  // while honoring heap poisoning and/or read barriers (if any).
  //
  // Location `maybe_temp` is used when generating a read barrier and
  // shall be a register in that case; it may be an invalid location
  // otherwise.
  void GenerateReferenceLoadOneRegister(HInstruction* instruction,
                                        Location out,
                                        uint32_t offset,
                                        Location maybe_temp);
  // Generate a heap reference load using two different registers
  // `out` and `obj`:
  //
  //   out <- *(obj + offset)
  //
  // while honoring heap poisoning and/or read barriers (if any).
  //
  // Location `maybe_temp` is used when generating a Baker's (fast
  // path) read barrier and shall be a register in that case; it may
  // be an invalid location otherwise.
  void GenerateReferenceLoadTwoRegisters(HInstruction* instruction,
                                         Location out,
                                         Location obj,
                                         uint32_t offset,
                                         Location maybe_temp);
  // Generate a GC root reference load:
  //
  //   root <- *address
  //
  // while honoring read barriers (if any).
  void GenerateGcRootFieldLoad(HInstruction* instruction,
                               Location root,
                               const Address& address,
                               Label* fixup_label = nullptr);

  void PushOntoFPStack(Location source, uint32_t temp_offset,
                       uint32_t stack_adjustment, bool is_float);
  void GenerateCompareTest(HCondition* condition);
  template<class LabelType>
  void GenerateTestAndBranch(HInstruction* instruction,
                             size_t condition_input_index,
                             LabelType* true_target,
                             LabelType* false_target);
  template<class LabelType>
  void GenerateCompareTestAndBranch(HCondition* condition,
                                    LabelType* true_target,
                                    LabelType* false_target);
  template<class LabelType>
  void GenerateFPJumps(HCondition* cond, LabelType* true_label, LabelType* false_label);

  void HandleGoto(HInstruction* got, HBasicBlock* successor);

  X86_64Assembler* const assembler_;
  CodeGeneratorX86_64* const codegen_;

  DISALLOW_COPY_AND_ASSIGN(InstructionCodeGeneratorX86_64);
};

// Class for fixups to jump tables.
class JumpTableRIPFixup;

class CodeGeneratorX86_64 : public CodeGenerator {
 public:
  CodeGeneratorX86_64(HGraph* graph,
                  const X86_64InstructionSetFeatures& isa_features,
                  const CompilerOptions& compiler_options,
                  OptimizingCompilerStats* stats = nullptr);
  virtual ~CodeGeneratorX86_64() {}

  void GenerateFrameEntry() OVERRIDE;
  void GenerateFrameExit() OVERRIDE;
  void Bind(HBasicBlock* block) OVERRIDE;
  void MoveConstant(Location destination, int32_t value) OVERRIDE;
  void MoveLocation(Location dst, Location src, Primitive::Type dst_type) OVERRIDE;
  void AddLocationAsTemp(Location location, LocationSummary* locations) OVERRIDE;

  size_t SaveCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
  size_t RestoreCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
  size_t SaveFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
  size_t RestoreFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;

  // Generate code to invoke a runtime entry point.
  void InvokeRuntime(QuickEntrypointEnum entrypoint,
                     HInstruction* instruction,
                     uint32_t dex_pc,
                     SlowPathCode* slow_path) OVERRIDE;

  void InvokeRuntime(int32_t entry_point_offset,
                     HInstruction* instruction,
                     uint32_t dex_pc,
                     SlowPathCode* slow_path);

  size_t GetWordSize() const OVERRIDE {
    return kX86_64WordSize;
  }

  size_t GetFloatingPointSpillSlotSize() const OVERRIDE {
    return kX86_64WordSize;
  }

  HGraphVisitor* GetLocationBuilder() OVERRIDE {
    return &location_builder_;
  }

  HGraphVisitor* GetInstructionVisitor() OVERRIDE {
    return &instruction_visitor_;
  }

  X86_64Assembler* GetAssembler() OVERRIDE {
    return &assembler_;
  }

  const X86_64Assembler& GetAssembler() const OVERRIDE {
    return assembler_;
  }

  ParallelMoveResolverX86_64* GetMoveResolver() OVERRIDE {
    return &move_resolver_;
  }

  uintptr_t GetAddressOf(HBasicBlock* block) OVERRIDE {
    return GetLabelOf(block)->Position();
  }

  void SetupBlockedRegisters() const OVERRIDE;
  void DumpCoreRegister(std::ostream& stream, int reg) const OVERRIDE;
  void DumpFloatingPointRegister(std::ostream& stream, int reg) const OVERRIDE;
  void Finalize(CodeAllocator* allocator) OVERRIDE;

  InstructionSet GetInstructionSet() const OVERRIDE {
    return InstructionSet::kX86_64;
  }

  // Emit a write barrier.
  void MarkGCCard(CpuRegister temp,
                  CpuRegister card,
                  CpuRegister object,
                  CpuRegister value,
                  bool value_can_be_null);

  void GenerateMemoryBarrier(MemBarrierKind kind);

  // Helper method to move a value between two locations.
  void Move(Location destination, Location source);

  Label* GetLabelOf(HBasicBlock* block) const {
    return CommonGetLabelOf<Label>(block_labels_, block);
  }

  void Initialize() OVERRIDE {
    block_labels_ = CommonInitializeLabels<Label>();
  }

  bool NeedsTwoRegisters(Primitive::Type type ATTRIBUTE_UNUSED) const OVERRIDE {
    return false;
  }

  // Check if the desired_string_load_kind is supported. If it is, return it,
  // otherwise return a fall-back kind that should be used instead.
  HLoadString::LoadKind GetSupportedLoadStringKind(
      HLoadString::LoadKind desired_string_load_kind) OVERRIDE;

  // Check if the desired_dispatch_info is supported. If it is, return it,
  // otherwise return a fall-back info that should be used instead.
  HInvokeStaticOrDirect::DispatchInfo GetSupportedInvokeStaticOrDirectDispatch(
      const HInvokeStaticOrDirect::DispatchInfo& desired_dispatch_info,
      MethodReference target_method) OVERRIDE;

  void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) OVERRIDE;
  void GenerateVirtualCall(HInvokeVirtual* invoke, Location temp) OVERRIDE;

  void RecordSimplePatch();
  void RecordStringPatch(HLoadString* load_string);
  Label* NewPcRelativeDexCacheArrayPatch(const DexFile& dex_file, uint32_t element_offset);

  void MoveFromReturnRegister(Location trg, Primitive::Type type) OVERRIDE;

  void EmitLinkerPatches(ArenaVector<LinkerPatch>* linker_patches) OVERRIDE;

  const X86_64InstructionSetFeatures& GetInstructionSetFeatures() const {
    return isa_features_;
  }

  // Fast path implementation of ReadBarrier::Barrier for a heap
  // reference field load when Baker's read barriers are used.
  void GenerateFieldLoadWithBakerReadBarrier(HInstruction* instruction,
                                             Location ref,
                                             CpuRegister obj,
                                             uint32_t offset,
                                             Location temp,
                                             bool needs_null_check);
  // Fast path implementation of ReadBarrier::Barrier for a heap
  // reference array load when Baker's read barriers are used.
  void GenerateArrayLoadWithBakerReadBarrier(HInstruction* instruction,
                                             Location ref,
                                             CpuRegister obj,
                                             uint32_t data_offset,
                                             Location index,
                                             Location temp,
                                             bool needs_null_check);

  // Generate a read barrier for a heap reference within `instruction`
  // using a slow path.
  //
  // A read barrier for an object reference read from the heap is
  // implemented as a call to the artReadBarrierSlow runtime entry
  // point, which is passed the values in locations `ref`, `obj`, and
  // `offset`:
  //
  //   mirror::Object* artReadBarrierSlow(mirror::Object* ref,
  //                                      mirror::Object* obj,
  //                                      uint32_t offset);
  //
  // The `out` location contains the value returned by
  // artReadBarrierSlow.
  //
  // When `index` provided (i.e., when it is different from
  // Location::NoLocation()), the offset value passed to
  // artReadBarrierSlow is adjusted to take `index` into account.
  void GenerateReadBarrierSlow(HInstruction* instruction,
                               Location out,
                               Location ref,
                               Location obj,
                               uint32_t offset,
                               Location index = Location::NoLocation());

  // If read barriers are enabled, generate a read barrier for a heap
  // reference using a slow path. If heap poisoning is enabled, also
  // unpoison the reference in `out`.
  void MaybeGenerateReadBarrierSlow(HInstruction* instruction,
                                    Location out,
                                    Location ref,
                                    Location obj,
                                    uint32_t offset,
                                    Location index = Location::NoLocation());

  // Generate a read barrier for a GC root within `instruction` using
  // a slow path.
  //
  // A read barrier for an object reference GC root is implemented as
  // a call to the artReadBarrierForRootSlow runtime entry point,
  // which is passed the value in location `root`:
  //
  //   mirror::Object* artReadBarrierForRootSlow(GcRoot<mirror::Object>* root);
  //
  // The `out` location contains the value returned by
  // artReadBarrierForRootSlow.
  void GenerateReadBarrierForRootSlow(HInstruction* instruction, Location out, Location root);

  int ConstantAreaStart() const {
    return constant_area_start_;
  }

  Address LiteralDoubleAddress(double v);
  Address LiteralFloatAddress(float v);
  Address LiteralInt32Address(int32_t v);
  Address LiteralInt64Address(int64_t v);

  // Load a 32/64-bit value into a register in the most efficient manner.
  void Load32BitValue(CpuRegister dest, int32_t value);
  void Load64BitValue(CpuRegister dest, int64_t value);
  void Load32BitValue(XmmRegister dest, int32_t value);
  void Load64BitValue(XmmRegister dest, int64_t value);
  void Load32BitValue(XmmRegister dest, float value);
  void Load64BitValue(XmmRegister dest, double value);

  // Compare a register with a 32/64-bit value in the most efficient manner.
  void Compare32BitValue(CpuRegister dest, int32_t value);
  void Compare64BitValue(CpuRegister dest, int64_t value);

  Address LiteralCaseTable(HPackedSwitch* switch_instr);

  // Store a 64 bit value into a DoubleStackSlot in the most efficient manner.
  void Store64BitValueToStack(Location dest, int64_t value);

  // Assign a 64 bit constant to an address.
  void MoveInt64ToAddress(const Address& addr_low,
                          const Address& addr_high,
                          int64_t v,
                          HInstruction* instruction);

  // Ensure that prior stores complete to memory before subsequent loads.
  // The locked add implementation will avoid serializing device memory, but will
  // touch (but not change) the top of the stack. The locked add should not be used for
  // ordering non-temporal stores.
  void MemoryFence(bool force_mfence = false) {
    if (!force_mfence && isa_features_.PrefersLockedAddSynchronization()) {
      assembler_.lock()->addl(Address(CpuRegister(RSP), 0), Immediate(0));
    } else {
      assembler_.mfence();
    }
  }

  void GenerateNop();
  void GenerateImplicitNullCheck(HNullCheck* instruction);
  void GenerateExplicitNullCheck(HNullCheck* instruction);

  // When we don't know the proper offset for the value, we use kDummy32BitOffset.
  // We will fix this up in the linker later to have the right value.
  static constexpr int32_t kDummy32BitOffset = 256;

 private:
  // Factored implementation of GenerateFieldLoadWithBakerReadBarrier
  // and GenerateArrayLoadWithBakerReadBarrier.
  void GenerateReferenceLoadWithBakerReadBarrier(HInstruction* instruction,
                                                 Location ref,
                                                 CpuRegister obj,
                                                 const Address& src,
                                                 Location temp,
                                                 bool needs_null_check);

  struct PcRelativeDexCacheAccessInfo {
    PcRelativeDexCacheAccessInfo(const DexFile& dex_file, uint32_t element_off)
        : target_dex_file(dex_file), element_offset(element_off), label() { }

    const DexFile& target_dex_file;
    uint32_t element_offset;
    Label label;
  };

  // Labels for each block that will be compiled.
  Label* block_labels_;  // Indexed by block id.
  Label frame_entry_label_;
  LocationsBuilderX86_64 location_builder_;
  InstructionCodeGeneratorX86_64 instruction_visitor_;
  ParallelMoveResolverX86_64 move_resolver_;
  X86_64Assembler assembler_;
  const X86_64InstructionSetFeatures& isa_features_;

  // Offset to the start of the constant area in the assembled code.
  // Used for fixups to the constant area.
  int constant_area_start_;

  // Method patch info. Using ArenaDeque<> which retains element addresses on push/emplace_back().
  ArenaDeque<MethodPatchInfo<Label>> method_patches_;
  ArenaDeque<MethodPatchInfo<Label>> relative_call_patches_;
  // PC-relative DexCache access info.
  ArenaDeque<PcRelativeDexCacheAccessInfo> pc_relative_dex_cache_patches_;
  // Patch locations for patchoat where the linker doesn't do any other work.
  ArenaDeque<Label> simple_patches_;
  // String patch locations.
  ArenaDeque<StringPatchInfo<Label>> string_patches_;

  // Fixups for jump tables need to be handled specially.
  ArenaVector<JumpTableRIPFixup*> fixups_to_jump_tables_;

  DISALLOW_COPY_AND_ASSIGN(CodeGeneratorX86_64);
};

}  // namespace x86_64
}  // namespace art

#endif  // ART_COMPILER_OPTIMIZING_CODE_GENERATOR_X86_64_H_