code_generator_x86_64.h revision 38207af82afb6f99c687f64b15601ed20d82220a 
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(DECLARE_VISIT_INSTRUCTION)
138
139#undef DECLARE_VISIT_INSTRUCTION
140
141 private:
142  void HandleInvoke(HInvoke* invoke);
143  void HandleBitwiseOperation(HBinaryOperation* operation);
144  void HandleShift(HBinaryOperation* operation);
145  void HandleFieldSet(HInstruction* instruction, const FieldInfo& field_info);
146  void HandleFieldGet(HInstruction* instruction);
147
148  CodeGeneratorX86_64* const codegen_;
149  InvokeDexCallingConventionVisitorX86_64 parameter_visitor_;
150
151  DISALLOW_COPY_AND_ASSIGN(LocationsBuilderX86_64);
152};
153
154class InstructionCodeGeneratorX86_64 : public HGraphVisitor {
155 public:
156  InstructionCodeGeneratorX86_64(HGraph* graph, CodeGeneratorX86_64* codegen);
157
158#define DECLARE_VISIT_INSTRUCTION(name, super)     \
159  void Visit##name(H##name* instr) OVERRIDE;
160
161  FOR_EACH_CONCRETE_INSTRUCTION(DECLARE_VISIT_INSTRUCTION)
162
163#undef DECLARE_VISIT_INSTRUCTION
164
165  X86_64Assembler* GetAssembler() const { return assembler_; }
166
167 private:
168  // Generate code for the given suspend check. If not null, `successor`
169  // is the block to branch to if the suspend check is not needed, and after
170  // the suspend call.
171  void GenerateSuspendCheck(HSuspendCheck* instruction, HBasicBlock* successor);
172  void GenerateClassInitializationCheck(SlowPathCodeX86_64* slow_path, CpuRegister class_reg);
173  void HandleBitwiseOperation(HBinaryOperation* operation);
174  void GenerateRemFP(HRem *rem);
175  void DivRemOneOrMinusOne(HBinaryOperation* instruction);
176  void DivByPowerOfTwo(HDiv* instruction);
177  void GenerateDivRemWithAnyConstant(HBinaryOperation* instruction);
178  void GenerateDivRemIntegral(HBinaryOperation* instruction);
179  void HandleShift(HBinaryOperation* operation);
180  void GenerateMemoryBarrier(MemBarrierKind kind);
181  void HandleFieldSet(HInstruction* instruction,
182                      const FieldInfo& field_info,
183                      bool value_can_be_null);
184  void HandleFieldGet(HInstruction* instruction, const FieldInfo& field_info);
185  void GenerateImplicitNullCheck(HNullCheck* instruction);
186  void GenerateExplicitNullCheck(HNullCheck* instruction);
187  void PushOntoFPStack(Location source, uint32_t temp_offset,
188                       uint32_t stack_adjustment, bool is_float);
189  void GenerateTestAndBranch(HInstruction* instruction,
190                             Label* true_target,
191                             Label* false_target,
192                             Label* always_true_target);
193
194  X86_64Assembler* const assembler_;
195  CodeGeneratorX86_64* const codegen_;
196
197  DISALLOW_COPY_AND_ASSIGN(InstructionCodeGeneratorX86_64);
198};
199
200class CodeGeneratorX86_64 : public CodeGenerator {
201 public:
202  CodeGeneratorX86_64(HGraph* graph,
203                  const X86_64InstructionSetFeatures& isa_features,
204                  const CompilerOptions& compiler_options);
205  virtual ~CodeGeneratorX86_64() {}
206
207  void GenerateFrameEntry() OVERRIDE;
208  void GenerateFrameExit() OVERRIDE;
209  void Bind(HBasicBlock* block) OVERRIDE;
210  void Move(HInstruction* instruction, Location location, HInstruction* move_for) OVERRIDE;
211  size_t SaveCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
212  size_t RestoreCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
213  size_t SaveFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
214  size_t RestoreFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
215
216  size_t GetWordSize() const OVERRIDE {
217    return kX86_64WordSize;
218  }
219
220  size_t GetFloatingPointSpillSlotSize() const OVERRIDE {
221    return kX86_64WordSize;
222  }
223
224  HGraphVisitor* GetLocationBuilder() OVERRIDE {
225    return &location_builder_;
226  }
227
228  HGraphVisitor* GetInstructionVisitor() OVERRIDE {
229    return &instruction_visitor_;
230  }
231
232  X86_64Assembler* GetAssembler() OVERRIDE {
233    return &assembler_;
234  }
235
236  ParallelMoveResolverX86_64* GetMoveResolver() OVERRIDE {
237    return &move_resolver_;
238  }
239
240  uintptr_t GetAddressOf(HBasicBlock* block) const OVERRIDE {
241    return GetLabelOf(block)->Position();
242  }
243
244  Location GetStackLocation(HLoadLocal* load) const OVERRIDE;
245
246  void SetupBlockedRegisters(bool is_baseline) const OVERRIDE;
247  Location AllocateFreeRegister(Primitive::Type type) const OVERRIDE;
248  void DumpCoreRegister(std::ostream& stream, int reg) const OVERRIDE;
249  void DumpFloatingPointRegister(std::ostream& stream, int reg) const OVERRIDE;
250  void Finalize(CodeAllocator* allocator) OVERRIDE;
251
252  InstructionSet GetInstructionSet() const OVERRIDE {
253    return InstructionSet::kX86_64;
254  }
255
256  // Emit a write barrier.
257  void MarkGCCard(CpuRegister temp,
258                  CpuRegister card,
259                  CpuRegister object,
260                  CpuRegister value,
261                  bool value_can_be_null);
262
263  // Helper method to move a value between two locations.
264  void Move(Location destination, Location source);
265
266  void LoadCurrentMethod(CpuRegister reg);
267
268  Label* GetLabelOf(HBasicBlock* block) const {
269    return CommonGetLabelOf<Label>(block_labels_.GetRawStorage(), block);
270  }
271
272  void Initialize() OVERRIDE {
273    block_labels_.SetSize(GetGraph()->GetBlocks().Size());
274  }
275
276  bool NeedsTwoRegisters(Primitive::Type type ATTRIBUTE_UNUSED) const OVERRIDE {
277    return false;
278  }
279
280  void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp);
281
282  const X86_64InstructionSetFeatures& GetInstructionSetFeatures() const {
283    return isa_features_;
284  }
285
286  int ConstantAreaStart() const {
287    return constant_area_start_;
288  }
289
290  Address LiteralDoubleAddress(double v);
291  Address LiteralFloatAddress(float v);
292  Address LiteralInt32Address(int32_t v);
293  Address LiteralInt64Address(int64_t v);
294
295  // Load a 64 bit value into a register in the most efficient manner.
296  void Load64BitValue(CpuRegister dest, int64_t value);
297
298 private:
299  // Labels for each block that will be compiled.
300  GrowableArray<Label> block_labels_;
301  Label frame_entry_label_;
302  LocationsBuilderX86_64 location_builder_;
303  InstructionCodeGeneratorX86_64 instruction_visitor_;
304  ParallelMoveResolverX86_64 move_resolver_;
305  X86_64Assembler assembler_;
306  const X86_64InstructionSetFeatures& isa_features_;
307
308  // Offset to the start of the constant area in the assembled code.
309  // Used for fixups to the constant area.
310  int constant_area_start_;
311
312  DISALLOW_COPY_AND_ASSIGN(CodeGeneratorX86_64);
313};
314
315}  // namespace x86_64
316}  // namespace art
317
318#endif  // ART_COMPILER_OPTIMIZING_CODE_GENERATOR_X86_64_H_
319