code_generator_x86_64.h revision 92e83bf8c0b2df8c977ffbc527989631d94b1819 
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
55 private:
56  DISALLOW_COPY_AND_ASSIGN(InvokeRuntimeCallingConvention);
57};
58
59class InvokeDexCallingConvention : public CallingConvention<Register, FloatRegister> {
60 public:
61  InvokeDexCallingConvention() : CallingConvention(
62      kParameterCoreRegisters,
63      kParameterCoreRegistersLength,
64      kParameterFloatRegisters,
65      kParameterFloatRegistersLength) {}
66
67 private:
68  DISALLOW_COPY_AND_ASSIGN(InvokeDexCallingConvention);
69};
70
71class InvokeDexCallingConventionVisitorX86_64 : public InvokeDexCallingConventionVisitor {
72 public:
73  InvokeDexCallingConventionVisitorX86_64() {}
74  virtual ~InvokeDexCallingConventionVisitorX86_64() {}
75
76  Location GetNextLocation(Primitive::Type type) OVERRIDE;
77
78 private:
79  InvokeDexCallingConvention calling_convention;
80
81  DISALLOW_COPY_AND_ASSIGN(InvokeDexCallingConventionVisitorX86_64);
82};
83
84class CodeGeneratorX86_64;
85
86class SlowPathCodeX86_64 : public SlowPathCode {
87 public:
88  SlowPathCodeX86_64() : entry_label_(), exit_label_() {}
89
90  Label* GetEntryLabel() { return &entry_label_; }
91  Label* GetExitLabel() { return &exit_label_; }
92
93 private:
94  Label entry_label_;
95  Label exit_label_;
96
97  DISALLOW_COPY_AND_ASSIGN(SlowPathCodeX86_64);
98};
99
100class ParallelMoveResolverX86_64 : public ParallelMoveResolverWithSwap {
101 public:
102  ParallelMoveResolverX86_64(ArenaAllocator* allocator, CodeGeneratorX86_64* codegen)
103      : ParallelMoveResolverWithSwap(allocator), codegen_(codegen) {}
104
105  void EmitMove(size_t index) OVERRIDE;
106  void EmitSwap(size_t index) OVERRIDE;
107  void SpillScratch(int reg) OVERRIDE;
108  void RestoreScratch(int reg) OVERRIDE;
109
110  X86_64Assembler* GetAssembler() const;
111
112 private:
113  void Exchange32(CpuRegister reg, int mem);
114  void Exchange32(XmmRegister reg, int mem);
115  void Exchange32(int mem1, int mem2);
116  void Exchange64(CpuRegister reg, int mem);
117  void Exchange64(XmmRegister reg, int mem);
118  void Exchange64(int mem1, int mem2);
119
120  CodeGeneratorX86_64* const codegen_;
121
122  DISALLOW_COPY_AND_ASSIGN(ParallelMoveResolverX86_64);
123};
124
125class LocationsBuilderX86_64 : public HGraphVisitor {
126 public:
127  LocationsBuilderX86_64(HGraph* graph, CodeGeneratorX86_64* codegen)
128      : HGraphVisitor(graph), codegen_(codegen) {}
129
130#define DECLARE_VISIT_INSTRUCTION(name, super)     \
131  void Visit##name(H##name* instr) OVERRIDE;
132
133  FOR_EACH_CONCRETE_INSTRUCTION(DECLARE_VISIT_INSTRUCTION)
134
135#undef DECLARE_VISIT_INSTRUCTION
136
137 private:
138  void HandleInvoke(HInvoke* invoke);
139  void HandleBitwiseOperation(HBinaryOperation* operation);
140  void HandleShift(HBinaryOperation* operation);
141  void HandleFieldSet(HInstruction* instruction, const FieldInfo& field_info);
142  void HandleFieldGet(HInstruction* instruction);
143
144  CodeGeneratorX86_64* const codegen_;
145  InvokeDexCallingConventionVisitorX86_64 parameter_visitor_;
146
147  DISALLOW_COPY_AND_ASSIGN(LocationsBuilderX86_64);
148};
149
150class InstructionCodeGeneratorX86_64 : public HGraphVisitor {
151 public:
152  InstructionCodeGeneratorX86_64(HGraph* graph, CodeGeneratorX86_64* codegen);
153
154#define DECLARE_VISIT_INSTRUCTION(name, super)     \
155  void Visit##name(H##name* instr) OVERRIDE;
156
157  FOR_EACH_CONCRETE_INSTRUCTION(DECLARE_VISIT_INSTRUCTION)
158
159#undef DECLARE_VISIT_INSTRUCTION
160
161  X86_64Assembler* GetAssembler() const { return assembler_; }
162
163 private:
164  // Generate code for the given suspend check. If not null, `successor`
165  // is the block to branch to if the suspend check is not needed, and after
166  // the suspend call.
167  void GenerateSuspendCheck(HSuspendCheck* instruction, HBasicBlock* successor);
168  void GenerateClassInitializationCheck(SlowPathCodeX86_64* slow_path, CpuRegister class_reg);
169  void HandleBitwiseOperation(HBinaryOperation* operation);
170  void GenerateRemFP(HRem *rem);
171  void DivRemOneOrMinusOne(HBinaryOperation* instruction);
172  void DivByPowerOfTwo(HDiv* instruction);
173  void GenerateDivRemWithAnyConstant(HBinaryOperation* instruction);
174  void GenerateDivRemIntegral(HBinaryOperation* instruction);
175  void HandleShift(HBinaryOperation* operation);
176  void GenerateMemoryBarrier(MemBarrierKind kind);
177  void HandleFieldSet(HInstruction* instruction, const FieldInfo& field_info);
178  void HandleFieldGet(HInstruction* instruction, const FieldInfo& field_info);
179  void GenerateImplicitNullCheck(HNullCheck* instruction);
180  void GenerateExplicitNullCheck(HNullCheck* instruction);
181  void PushOntoFPStack(Location source, uint32_t temp_offset,
182                       uint32_t stack_adjustment, bool is_float);
183  void GenerateTestAndBranch(HInstruction* instruction,
184                             Label* true_target,
185                             Label* false_target,
186                             Label* always_true_target);
187
188  X86_64Assembler* const assembler_;
189  CodeGeneratorX86_64* const codegen_;
190
191  DISALLOW_COPY_AND_ASSIGN(InstructionCodeGeneratorX86_64);
192};
193
194class CodeGeneratorX86_64 : public CodeGenerator {
195 public:
196  CodeGeneratorX86_64(HGraph* graph,
197                  const X86_64InstructionSetFeatures& isa_features,
198                  const CompilerOptions& compiler_options);
199  virtual ~CodeGeneratorX86_64() {}
200
201  void GenerateFrameEntry() OVERRIDE;
202  void GenerateFrameExit() OVERRIDE;
203  void Bind(HBasicBlock* block) OVERRIDE;
204  void Move(HInstruction* instruction, Location location, HInstruction* move_for) OVERRIDE;
205  size_t SaveCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
206  size_t RestoreCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
207  size_t SaveFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
208  size_t RestoreFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
209
210  size_t GetWordSize() const OVERRIDE {
211    return kX86_64WordSize;
212  }
213
214  size_t GetFloatingPointSpillSlotSize() const OVERRIDE {
215    return kX86_64WordSize;
216  }
217
218  HGraphVisitor* GetLocationBuilder() OVERRIDE {
219    return &location_builder_;
220  }
221
222  HGraphVisitor* GetInstructionVisitor() OVERRIDE {
223    return &instruction_visitor_;
224  }
225
226  X86_64Assembler* GetAssembler() OVERRIDE {
227    return &assembler_;
228  }
229
230  ParallelMoveResolverX86_64* GetMoveResolver() OVERRIDE {
231    return &move_resolver_;
232  }
233
234  uintptr_t GetAddressOf(HBasicBlock* block) const OVERRIDE {
235    return GetLabelOf(block)->Position();
236  }
237
238  Location GetStackLocation(HLoadLocal* load) const OVERRIDE;
239
240  void SetupBlockedRegisters(bool is_baseline) const OVERRIDE;
241  Location AllocateFreeRegister(Primitive::Type type) const OVERRIDE;
242  void DumpCoreRegister(std::ostream& stream, int reg) const OVERRIDE;
243  void DumpFloatingPointRegister(std::ostream& stream, int reg) const OVERRIDE;
244  void Finalize(CodeAllocator* allocator) OVERRIDE;
245
246  InstructionSet GetInstructionSet() const OVERRIDE {
247    return InstructionSet::kX86_64;
248  }
249
250  // Emit a write barrier.
251  void MarkGCCard(CpuRegister temp, CpuRegister card, CpuRegister object, CpuRegister value);
252
253  // Helper method to move a value between two locations.
254  void Move(Location destination, Location source);
255
256  void LoadCurrentMethod(CpuRegister reg);
257
258  Label* GetLabelOf(HBasicBlock* block) const {
259    return CommonGetLabelOf<Label>(block_labels_.GetRawStorage(), block);
260  }
261
262  void Initialize() OVERRIDE {
263    block_labels_.SetSize(GetGraph()->GetBlocks().Size());
264  }
265
266  bool NeedsTwoRegisters(Primitive::Type type ATTRIBUTE_UNUSED) const OVERRIDE {
267    return false;
268  }
269
270  void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, CpuRegister temp);
271
272  const X86_64InstructionSetFeatures& GetInstructionSetFeatures() const {
273    return isa_features_;
274  }
275
276  int ConstantAreaStart() const {
277    return constant_area_start_;
278  }
279
280  Address LiteralDoubleAddress(double v);
281  Address LiteralFloatAddress(float v);
282  Address LiteralInt32Address(int32_t v);
283  Address LiteralInt64Address(int64_t v);
284
285  // Load a 64 bit value into a register in the most efficient manner.
286  void Load64BitValue(CpuRegister dest, int64_t value);
287
288 private:
289  // Labels for each block that will be compiled.
290  GrowableArray<Label> block_labels_;
291  Label frame_entry_label_;
292  LocationsBuilderX86_64 location_builder_;
293  InstructionCodeGeneratorX86_64 instruction_visitor_;
294  ParallelMoveResolverX86_64 move_resolver_;
295  X86_64Assembler assembler_;
296  const X86_64InstructionSetFeatures& isa_features_;
297
298  // Offset to the start of the constant area in the assembled code.
299  // Used for fixups to the constant area.
300  int constant_area_start_;
301
302  DISALLOW_COPY_AND_ASSIGN(CodeGeneratorX86_64);
303};
304
305}  // namespace x86_64
306}  // namespace art
307
308#endif  // ART_COMPILER_OPTIMIZING_CODE_GENERATOR_X86_64_H_
309