code_generator_x86_64.h revision 225b6464a58ebe11c156144653f11a1c6607f4eb 
1/*
2 * Copyright (C) 2014 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 *      http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17#ifndef ART_COMPILER_OPTIMIZING_CODE_GENERATOR_X86_64_H_
18#define ART_COMPILER_OPTIMIZING_CODE_GENERATOR_X86_64_H_
19
20#include "code_generator.h"
21#include "dex/compiler_enums.h"
22#include "driver/compiler_options.h"
23#include "nodes.h"
24#include "parallel_move_resolver.h"
25#include "utils/x86_64/assembler_x86_64.h"
26
27namespace art {
28namespace x86_64 {
29
30// Use a local definition to prevent copying mistakes.
31static constexpr size_t kX86_64WordSize = kX86_64PointerSize;
32
33static constexpr Register kParameterCoreRegisters[] = { RSI, RDX, RCX, R8, R9 };
34static constexpr FloatRegister kParameterFloatRegisters[] =
35    { XMM0, XMM1, XMM2, XMM3, XMM4, XMM5, XMM6, XMM7 };
36
37static constexpr size_t kParameterCoreRegistersLength = arraysize(kParameterCoreRegisters);
38static constexpr size_t kParameterFloatRegistersLength = arraysize(kParameterFloatRegisters);
39
40static constexpr Register kRuntimeParameterCoreRegisters[] = { RDI, RSI, RDX, RCX };
41static constexpr size_t kRuntimeParameterCoreRegistersLength =
42    arraysize(kRuntimeParameterCoreRegisters);
43static constexpr FloatRegister kRuntimeParameterFpuRegisters[] = { XMM0, XMM1 };
44static constexpr size_t kRuntimeParameterFpuRegistersLength =
45    arraysize(kRuntimeParameterFpuRegisters);
46
47class InvokeRuntimeCallingConvention : public CallingConvention<Register, FloatRegister> {
48 public:
49  InvokeRuntimeCallingConvention()
50      : CallingConvention(kRuntimeParameterCoreRegisters,
51                          kRuntimeParameterCoreRegistersLength,
52                          kRuntimeParameterFpuRegisters,
53                          kRuntimeParameterFpuRegistersLength,
54                          kX86_64PointerSize) {}
55
56 private:
57  DISALLOW_COPY_AND_ASSIGN(InvokeRuntimeCallingConvention);
58};
59
60class InvokeDexCallingConvention : public CallingConvention<Register, FloatRegister> {
61 public:
62  InvokeDexCallingConvention() : CallingConvention(
63      kParameterCoreRegisters,
64      kParameterCoreRegistersLength,
65      kParameterFloatRegisters,
66      kParameterFloatRegistersLength,
67      kX86_64PointerSize) {}
68
69 private:
70  DISALLOW_COPY_AND_ASSIGN(InvokeDexCallingConvention);
71};
72
73class InvokeDexCallingConventionVisitorX86_64 : public InvokeDexCallingConventionVisitor {
74 public:
75  InvokeDexCallingConventionVisitorX86_64() {}
76  virtual ~InvokeDexCallingConventionVisitorX86_64() {}
77
78  Location GetNextLocation(Primitive::Type type) OVERRIDE;
79  Location GetReturnLocation(Primitive::Type type) const OVERRIDE;
80  Location GetMethodLocation() const OVERRIDE;
81
82 private:
83  InvokeDexCallingConvention calling_convention;
84
85  DISALLOW_COPY_AND_ASSIGN(InvokeDexCallingConventionVisitorX86_64);
86};
87
88class CodeGeneratorX86_64;
89
90class ParallelMoveResolverX86_64 : public ParallelMoveResolverWithSwap {
91 public:
92  ParallelMoveResolverX86_64(ArenaAllocator* allocator, CodeGeneratorX86_64* codegen)
93      : ParallelMoveResolverWithSwap(allocator), codegen_(codegen) {}
94
95  void EmitMove(size_t index) OVERRIDE;
96  void EmitSwap(size_t index) OVERRIDE;
97  void SpillScratch(int reg) OVERRIDE;
98  void RestoreScratch(int reg) OVERRIDE;
99
100  X86_64Assembler* GetAssembler() const;
101
102 private:
103  void Exchange32(CpuRegister reg, int mem);
104  void Exchange32(XmmRegister reg, int mem);
105  void Exchange32(int mem1, int mem2);
106  void Exchange64(CpuRegister reg, int mem);
107  void Exchange64(XmmRegister reg, int mem);
108  void Exchange64(int mem1, int mem2);
109
110  CodeGeneratorX86_64* const codegen_;
111
112  DISALLOW_COPY_AND_ASSIGN(ParallelMoveResolverX86_64);
113};
114
115class LocationsBuilderX86_64 : public HGraphVisitor {
116 public:
117  LocationsBuilderX86_64(HGraph* graph, CodeGeneratorX86_64* codegen)
118      : HGraphVisitor(graph), codegen_(codegen) {}
119
120#define DECLARE_VISIT_INSTRUCTION(name, super)     \
121  void Visit##name(H##name* instr) OVERRIDE;
122
123  FOR_EACH_CONCRETE_INSTRUCTION_COMMON(DECLARE_VISIT_INSTRUCTION)
124  FOR_EACH_CONCRETE_INSTRUCTION_X86_64(DECLARE_VISIT_INSTRUCTION)
125
126#undef DECLARE_VISIT_INSTRUCTION
127
128  void VisitInstruction(HInstruction* instruction) OVERRIDE {
129    LOG(FATAL) << "Unreachable instruction " << instruction->DebugName()
130               << " (id " << instruction->GetId() << ")";
131  }
132
133 private:
134  void HandleInvoke(HInvoke* invoke);
135  void HandleBitwiseOperation(HBinaryOperation* operation);
136  void HandleShift(HBinaryOperation* operation);
137  void HandleFieldSet(HInstruction* instruction, const FieldInfo& field_info);
138  void HandleFieldGet(HInstruction* instruction);
139
140  CodeGeneratorX86_64* const codegen_;
141  InvokeDexCallingConventionVisitorX86_64 parameter_visitor_;
142
143  DISALLOW_COPY_AND_ASSIGN(LocationsBuilderX86_64);
144};
145
146class InstructionCodeGeneratorX86_64 : public HGraphVisitor {
147 public:
148  InstructionCodeGeneratorX86_64(HGraph* graph, CodeGeneratorX86_64* codegen);
149
150#define DECLARE_VISIT_INSTRUCTION(name, super)     \
151  void Visit##name(H##name* instr) OVERRIDE;
152
153  FOR_EACH_CONCRETE_INSTRUCTION_COMMON(DECLARE_VISIT_INSTRUCTION)
154  FOR_EACH_CONCRETE_INSTRUCTION_X86_64(DECLARE_VISIT_INSTRUCTION)
155
156#undef DECLARE_VISIT_INSTRUCTION
157
158  void VisitInstruction(HInstruction* instruction) OVERRIDE {
159    LOG(FATAL) << "Unreachable instruction " << instruction->DebugName()
160               << " (id " << instruction->GetId() << ")";
161  }
162
163  X86_64Assembler* GetAssembler() const { return assembler_; }
164
165 private:
166  // Generate code for the given suspend check. If not null, `successor`
167  // is the block to branch to if the suspend check is not needed, and after
168  // the suspend call.
169  void GenerateSuspendCheck(HSuspendCheck* instruction, HBasicBlock* successor);
170  void GenerateClassInitializationCheck(SlowPathCode* slow_path, CpuRegister class_reg);
171  void HandleBitwiseOperation(HBinaryOperation* operation);
172  void GenerateRemFP(HRem* rem);
173  void DivRemOneOrMinusOne(HBinaryOperation* instruction);
174  void DivByPowerOfTwo(HDiv* instruction);
175  void GenerateDivRemWithAnyConstant(HBinaryOperation* instruction);
176  void GenerateDivRemIntegral(HBinaryOperation* instruction);
177  void HandleShift(HBinaryOperation* operation);
178  void GenerateMemoryBarrier(MemBarrierKind kind);
179  void HandleFieldSet(HInstruction* instruction,
180                      const FieldInfo& field_info,
181                      bool value_can_be_null);
182  void HandleFieldGet(HInstruction* instruction, const FieldInfo& field_info);
183  void GenerateImplicitNullCheck(HNullCheck* instruction);
184  void GenerateExplicitNullCheck(HNullCheck* instruction);
185  void PushOntoFPStack(Location source, uint32_t temp_offset,
186                       uint32_t stack_adjustment, bool is_float);
187  void GenerateTestAndBranch(HInstruction* instruction,
188                             Label* true_target,
189                             Label* false_target,
190                             Label* always_true_target);
191  void GenerateCompareTestAndBranch(HIf* if_inst,
192                                    HCondition* condition,
193                                    Label* true_target,
194                                    Label* false_target,
195                                    Label* always_true_target);
196  void GenerateFPJumps(HCondition* cond, Label* true_label, Label* false_label);
197  void HandleGoto(HInstruction* got, HBasicBlock* successor);
198
199  X86_64Assembler* const assembler_;
200  CodeGeneratorX86_64* const codegen_;
201
202  DISALLOW_COPY_AND_ASSIGN(InstructionCodeGeneratorX86_64);
203};
204
205class CodeGeneratorX86_64 : public CodeGenerator {
206 public:
207  CodeGeneratorX86_64(HGraph* graph,
208                  const X86_64InstructionSetFeatures& isa_features,
209                  const CompilerOptions& compiler_options,
210                  OptimizingCompilerStats* stats = nullptr);
211  virtual ~CodeGeneratorX86_64() {}
212
213  void GenerateFrameEntry() OVERRIDE;
214  void GenerateFrameExit() OVERRIDE;
215  void Bind(HBasicBlock* block) OVERRIDE;
216  void Move(HInstruction* instruction, Location location, HInstruction* move_for) OVERRIDE;
217  void MoveConstant(Location destination, int32_t value) OVERRIDE;
218  size_t SaveCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
219  size_t RestoreCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
220  size_t SaveFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
221  size_t RestoreFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
222
223  // Generate code to invoke a runtime entry point.
224  void InvokeRuntime(QuickEntrypointEnum entrypoint,
225                     HInstruction* instruction,
226                     uint32_t dex_pc,
227                     SlowPathCode* slow_path) OVERRIDE;
228
229  void InvokeRuntime(int32_t entry_point_offset,
230                     HInstruction* instruction,
231                     uint32_t dex_pc,
232                     SlowPathCode* slow_path);
233
234  size_t GetWordSize() const OVERRIDE {
235    return kX86_64WordSize;
236  }
237
238  size_t GetFloatingPointSpillSlotSize() const OVERRIDE {
239    return kX86_64WordSize;
240  }
241
242  HGraphVisitor* GetLocationBuilder() OVERRIDE {
243    return &location_builder_;
244  }
245
246  HGraphVisitor* GetInstructionVisitor() OVERRIDE {
247    return &instruction_visitor_;
248  }
249
250  X86_64Assembler* GetAssembler() OVERRIDE {
251    return &assembler_;
252  }
253
254  const X86_64Assembler& GetAssembler() const OVERRIDE {
255    return assembler_;
256  }
257
258  ParallelMoveResolverX86_64* GetMoveResolver() OVERRIDE {
259    return &move_resolver_;
260  }
261
262  uintptr_t GetAddressOf(HBasicBlock* block) const OVERRIDE {
263    return GetLabelOf(block)->Position();
264  }
265
266  Location GetStackLocation(HLoadLocal* load) const OVERRIDE;
267
268  void SetupBlockedRegisters(bool is_baseline) const OVERRIDE;
269  Location AllocateFreeRegister(Primitive::Type type) const OVERRIDE;
270  void DumpCoreRegister(std::ostream& stream, int reg) const OVERRIDE;
271  void DumpFloatingPointRegister(std::ostream& stream, int reg) const OVERRIDE;
272  void Finalize(CodeAllocator* allocator) OVERRIDE;
273
274  InstructionSet GetInstructionSet() const OVERRIDE {
275    return InstructionSet::kX86_64;
276  }
277
278  // Emit a write barrier.
279  void MarkGCCard(CpuRegister temp,
280                  CpuRegister card,
281                  CpuRegister object,
282                  CpuRegister value,
283                  bool value_can_be_null);
284
285  // Helper method to move a value between two locations.
286  void Move(Location destination, Location source);
287
288  Label* GetLabelOf(HBasicBlock* block) const {
289    return CommonGetLabelOf<Label>(block_labels_, block);
290  }
291
292  void Initialize() OVERRIDE {
293    block_labels_ = CommonInitializeLabels<Label>();
294  }
295
296  bool NeedsTwoRegisters(Primitive::Type type ATTRIBUTE_UNUSED) const OVERRIDE {
297    return false;
298  }
299
300  void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp) OVERRIDE;
301  void GenerateVirtualCall(HInvokeVirtual* invoke, Location temp) OVERRIDE;
302
303  void MoveFromReturnRegister(Location trg, Primitive::Type type) OVERRIDE;
304
305  void EmitLinkerPatches(ArenaVector<LinkerPatch>* linker_patches) OVERRIDE;
306
307  const X86_64InstructionSetFeatures& GetInstructionSetFeatures() const {
308    return isa_features_;
309  }
310
311  int ConstantAreaStart() const {
312    return constant_area_start_;
313  }
314
315  Address LiteralDoubleAddress(double v);
316  Address LiteralFloatAddress(float v);
317  Address LiteralInt32Address(int32_t v);
318  Address LiteralInt64Address(int64_t v);
319
320  // Load a 64 bit value into a register in the most efficient manner.
321  void Load64BitValue(CpuRegister dest, int64_t value);
322
323  // Store a 64 bit value into a DoubleStackSlot in the most efficient manner.
324  void Store64BitValueToStack(Location dest, int64_t value);
325
326 private:
327  struct PcRelativeDexCacheAccessInfo {
328    PcRelativeDexCacheAccessInfo(const DexFile& dex_file, uint32_t element_off)
329        : target_dex_file(dex_file), element_offset(element_off), label() { }
330
331    const DexFile& target_dex_file;
332    uint32_t element_offset;
333    Label label;
334  };
335
336  // Labels for each block that will be compiled.
337  Label* block_labels_;  // Indexed by block id.
338  Label frame_entry_label_;
339  LocationsBuilderX86_64 location_builder_;
340  InstructionCodeGeneratorX86_64 instruction_visitor_;
341  ParallelMoveResolverX86_64 move_resolver_;
342  X86_64Assembler assembler_;
343  const X86_64InstructionSetFeatures& isa_features_;
344
345  // Offset to the start of the constant area in the assembled code.
346  // Used for fixups to the constant area.
347  int constant_area_start_;
348
349  // Method patch info. Using ArenaDeque<> which retains element addresses on push/emplace_back().
350  ArenaDeque<MethodPatchInfo<Label>> method_patches_;
351  ArenaDeque<MethodPatchInfo<Label>> relative_call_patches_;
352  // PC-relative DexCache access info.
353  ArenaDeque<PcRelativeDexCacheAccessInfo> pc_rel_dex_cache_patches_;
354
355  // When we don't know the proper offset for the value, we use kDummy32BitOffset.
356  // We will fix this up in the linker later to have the right value.
357  static constexpr int32_t kDummy32BitOffset = 256;
358
359  DISALLOW_COPY_AND_ASSIGN(CodeGeneratorX86_64);
360};
361
362}  // namespace x86_64
363}  // namespace art
364
365#endif  // ART_COMPILER_OPTIMIZING_CODE_GENERATOR_X86_64_H_
366