code_generator_x86_64.h revision ecc4366670e12b4812ef1653f7c8d52234ca1b1f 
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 SlowPathCodeX86_64 : public SlowPathCode {
91 public:
92  SlowPathCodeX86_64() : entry_label_(), exit_label_() {}
93
94  Label* GetEntryLabel() { return &entry_label_; }
95  Label* GetExitLabel() { return &exit_label_; }
96
97 private:
98  Label entry_label_;
99  Label exit_label_;
100
101  DISALLOW_COPY_AND_ASSIGN(SlowPathCodeX86_64);
102};
103
104class ParallelMoveResolverX86_64 : public ParallelMoveResolverWithSwap {
105 public:
106  ParallelMoveResolverX86_64(ArenaAllocator* allocator, CodeGeneratorX86_64* codegen)
107      : ParallelMoveResolverWithSwap(allocator), codegen_(codegen) {}
108
109  void EmitMove(size_t index) OVERRIDE;
110  void EmitSwap(size_t index) OVERRIDE;
111  void SpillScratch(int reg) OVERRIDE;
112  void RestoreScratch(int reg) OVERRIDE;
113
114  X86_64Assembler* GetAssembler() const;
115
116 private:
117  void Exchange32(CpuRegister reg, int mem);
118  void Exchange32(XmmRegister reg, int mem);
119  void Exchange32(int mem1, int mem2);
120  void Exchange64(CpuRegister reg, int mem);
121  void Exchange64(XmmRegister reg, int mem);
122  void Exchange64(int mem1, int mem2);
123
124  CodeGeneratorX86_64* const codegen_;
125
126  DISALLOW_COPY_AND_ASSIGN(ParallelMoveResolverX86_64);
127};
128
129class LocationsBuilderX86_64 : public HGraphVisitor {
130 public:
131  LocationsBuilderX86_64(HGraph* graph, CodeGeneratorX86_64* codegen)
132      : HGraphVisitor(graph), codegen_(codegen) {}
133
134#define DECLARE_VISIT_INSTRUCTION(name, super)     \
135  void Visit##name(H##name* instr) OVERRIDE;
136
137  FOR_EACH_CONCRETE_INSTRUCTION_COMMON(DECLARE_VISIT_INSTRUCTION)
138  FOR_EACH_CONCRETE_INSTRUCTION_X86_64(DECLARE_VISIT_INSTRUCTION)
139
140#undef DECLARE_VISIT_INSTRUCTION
141
142  void VisitInstruction(HInstruction* instruction) OVERRIDE {
143    LOG(FATAL) << "Unreachable instruction " << instruction->DebugName()
144               << " (id " << instruction->GetId() << ")";
145  }
146
147 private:
148  void HandleInvoke(HInvoke* invoke);
149  void HandleBitwiseOperation(HBinaryOperation* operation);
150  void HandleShift(HBinaryOperation* operation);
151  void HandleFieldSet(HInstruction* instruction, const FieldInfo& field_info);
152  void HandleFieldGet(HInstruction* instruction);
153
154  CodeGeneratorX86_64* const codegen_;
155  InvokeDexCallingConventionVisitorX86_64 parameter_visitor_;
156
157  DISALLOW_COPY_AND_ASSIGN(LocationsBuilderX86_64);
158};
159
160class InstructionCodeGeneratorX86_64 : public HGraphVisitor {
161 public:
162  InstructionCodeGeneratorX86_64(HGraph* graph, CodeGeneratorX86_64* codegen);
163
164#define DECLARE_VISIT_INSTRUCTION(name, super)     \
165  void Visit##name(H##name* instr) OVERRIDE;
166
167  FOR_EACH_CONCRETE_INSTRUCTION_COMMON(DECLARE_VISIT_INSTRUCTION)
168  FOR_EACH_CONCRETE_INSTRUCTION_X86_64(DECLARE_VISIT_INSTRUCTION)
169
170#undef DECLARE_VISIT_INSTRUCTION
171
172  void VisitInstruction(HInstruction* instruction) OVERRIDE {
173    LOG(FATAL) << "Unreachable instruction " << instruction->DebugName()
174               << " (id " << instruction->GetId() << ")";
175  }
176
177  X86_64Assembler* GetAssembler() const { return assembler_; }
178
179 private:
180  // Generate code for the given suspend check. If not null, `successor`
181  // is the block to branch to if the suspend check is not needed, and after
182  // the suspend call.
183  void GenerateSuspendCheck(HSuspendCheck* instruction, HBasicBlock* successor);
184  void GenerateClassInitializationCheck(SlowPathCodeX86_64* slow_path, CpuRegister class_reg);
185  void HandleBitwiseOperation(HBinaryOperation* operation);
186  void GenerateRemFP(HRem* rem);
187  void DivRemOneOrMinusOne(HBinaryOperation* instruction);
188  void DivByPowerOfTwo(HDiv* instruction);
189  void GenerateDivRemWithAnyConstant(HBinaryOperation* instruction);
190  void GenerateDivRemIntegral(HBinaryOperation* instruction);
191  void HandleShift(HBinaryOperation* operation);
192  void GenerateMemoryBarrier(MemBarrierKind kind);
193  void HandleFieldSet(HInstruction* instruction,
194                      const FieldInfo& field_info,
195                      bool value_can_be_null);
196  void HandleFieldGet(HInstruction* instruction, const FieldInfo& field_info);
197  void GenerateImplicitNullCheck(HNullCheck* instruction);
198  void GenerateExplicitNullCheck(HNullCheck* instruction);
199  void PushOntoFPStack(Location source, uint32_t temp_offset,
200                       uint32_t stack_adjustment, bool is_float);
201  void GenerateTestAndBranch(HInstruction* instruction,
202                             Label* true_target,
203                             Label* false_target,
204                             Label* always_true_target);
205  void GenerateCompareTestAndBranch(HIf* if_inst,
206                                    HCondition* condition,
207                                    Label* true_target,
208                                    Label* false_target,
209                                    Label* always_true_target);
210  void GenerateFPJumps(HCondition* cond, Label* true_label, Label* false_label);
211  void HandleGoto(HInstruction* got, HBasicBlock* successor);
212
213  X86_64Assembler* const assembler_;
214  CodeGeneratorX86_64* const codegen_;
215
216  DISALLOW_COPY_AND_ASSIGN(InstructionCodeGeneratorX86_64);
217};
218
219class CodeGeneratorX86_64 : public CodeGenerator {
220 public:
221  CodeGeneratorX86_64(HGraph* graph,
222                  const X86_64InstructionSetFeatures& isa_features,
223                  const CompilerOptions& compiler_options,
224                  OptimizingCompilerStats* stats = nullptr);
225  virtual ~CodeGeneratorX86_64() {}
226
227  void GenerateFrameEntry() OVERRIDE;
228  void GenerateFrameExit() OVERRIDE;
229  void Bind(HBasicBlock* block) OVERRIDE;
230  void Move(HInstruction* instruction, Location location, HInstruction* move_for) OVERRIDE;
231  size_t SaveCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
232  size_t RestoreCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
233  size_t SaveFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
234  size_t RestoreFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
235
236  // Generate code to invoke a runtime entry point.
237  void InvokeRuntime(Address entry_point,
238                     HInstruction* instruction,
239                     uint32_t dex_pc,
240                     SlowPathCode* slow_path);
241
242  size_t GetWordSize() const OVERRIDE {
243    return kX86_64WordSize;
244  }
245
246  size_t GetFloatingPointSpillSlotSize() const OVERRIDE {
247    return kX86_64WordSize;
248  }
249
250  HGraphVisitor* GetLocationBuilder() OVERRIDE {
251    return &location_builder_;
252  }
253
254  HGraphVisitor* GetInstructionVisitor() OVERRIDE {
255    return &instruction_visitor_;
256  }
257
258  X86_64Assembler* GetAssembler() OVERRIDE {
259    return &assembler_;
260  }
261
262  const X86_64Assembler& GetAssembler() const OVERRIDE {
263    return assembler_;
264  }
265
266  ParallelMoveResolverX86_64* GetMoveResolver() OVERRIDE {
267    return &move_resolver_;
268  }
269
270  uintptr_t GetAddressOf(HBasicBlock* block) const OVERRIDE {
271    return GetLabelOf(block)->Position();
272  }
273
274  Location GetStackLocation(HLoadLocal* load) const OVERRIDE;
275
276  void SetupBlockedRegisters(bool is_baseline) const OVERRIDE;
277  Location AllocateFreeRegister(Primitive::Type type) const OVERRIDE;
278  void DumpCoreRegister(std::ostream& stream, int reg) const OVERRIDE;
279  void DumpFloatingPointRegister(std::ostream& stream, int reg) const OVERRIDE;
280  void Finalize(CodeAllocator* allocator) OVERRIDE;
281
282  InstructionSet GetInstructionSet() const OVERRIDE {
283    return InstructionSet::kX86_64;
284  }
285
286  // Emit a write barrier.
287  void MarkGCCard(CpuRegister temp,
288                  CpuRegister card,
289                  CpuRegister object,
290                  CpuRegister value,
291                  bool value_can_be_null);
292
293  // Helper method to move a value between two locations.
294  void Move(Location destination, Location source);
295
296  Label* GetLabelOf(HBasicBlock* block) const {
297    return CommonGetLabelOf<Label>(block_labels_.GetRawStorage(), block);
298  }
299
300  void Initialize() OVERRIDE {
301    block_labels_.SetSize(GetGraph()->GetBlocks().Size());
302  }
303
304  bool NeedsTwoRegisters(Primitive::Type type ATTRIBUTE_UNUSED) const OVERRIDE {
305    return false;
306  }
307
308  void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp);
309
310  void EmitLinkerPatches(ArenaVector<LinkerPatch>* linker_patches) OVERRIDE;
311
312  const X86_64InstructionSetFeatures& GetInstructionSetFeatures() const {
313    return isa_features_;
314  }
315
316  int ConstantAreaStart() const {
317    return constant_area_start_;
318  }
319
320  Address LiteralDoubleAddress(double v);
321  Address LiteralFloatAddress(float v);
322  Address LiteralInt32Address(int32_t v);
323  Address LiteralInt64Address(int64_t v);
324
325  // Load a 64 bit value into a register in the most efficient manner.
326  void Load64BitValue(CpuRegister dest, int64_t value);
327
328  // Store a 64 bit value into a DoubleStackSlot in the most efficient manner.
329  void Store64BitValueToStack(Location dest, int64_t value);
330
331 private:
332  struct PcRelativeDexCacheAccessInfo {
333    PcRelativeDexCacheAccessInfo(const DexFile& dex_file, uint32_t element_off)
334        : target_dex_file(dex_file), element_offset(element_off), label() { }
335
336    const DexFile& target_dex_file;
337    uint32_t element_offset;
338    Label label;
339  };
340
341  // Labels for each block that will be compiled.
342  GrowableArray<Label> block_labels_;
343  Label frame_entry_label_;
344  LocationsBuilderX86_64 location_builder_;
345  InstructionCodeGeneratorX86_64 instruction_visitor_;
346  ParallelMoveResolverX86_64 move_resolver_;
347  X86_64Assembler assembler_;
348  const X86_64InstructionSetFeatures& isa_features_;
349
350  // Offset to the start of the constant area in the assembled code.
351  // Used for fixups to the constant area.
352  int constant_area_start_;
353
354  // Method patch info. Using ArenaDeque<> which retains element addresses on push/emplace_back().
355  ArenaDeque<MethodPatchInfo<Label>> method_patches_;
356  ArenaDeque<MethodPatchInfo<Label>> relative_call_patches_;
357  // PC-relative DexCache access info.
358  ArenaDeque<PcRelativeDexCacheAccessInfo> pc_rel_dex_cache_patches_;
359
360  // When we don't know the proper offset for the value, we use kDummy32BitOffset.
361  // We will fix this up in the linker later to have the right value.
362  static constexpr int32_t kDummy32BitOffset = 256;
363
364  DISALLOW_COPY_AND_ASSIGN(CodeGeneratorX86_64);
365};
366
367}  // namespace x86_64
368}  // namespace art
369
370#endif  // ART_COMPILER_OPTIMIZING_CODE_GENERATOR_X86_64_H_
371