code_generator_x86_64.h revision 23a8e35481face09183a24b9d11e505597c75ebb 
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 FieldAccessCallingConvetionX86_64 : public FieldAccessCallingConvetion {
74 public:
75  FieldAccessCallingConvetionX86_64() {}
76
77  Location GetObjectLocation() const OVERRIDE {
78    return Location::RegisterLocation(RSI);
79  }
80  Location GetFieldIndexLocation() const OVERRIDE {
81    return Location::RegisterLocation(RDI);
82  }
83  Location GetReturnLocation(Primitive::Type type ATTRIBUTE_UNUSED) const OVERRIDE {
84    return Location::RegisterLocation(RAX);
85  }
86  Location GetSetValueLocation(
87      Primitive::Type type ATTRIBUTE_UNUSED, bool is_instance) const OVERRIDE {
88    return Location::RegisterLocation(is_instance ? RDX : RSI);
89  }
90  Location GetFpuLocation(Primitive::Type type ATTRIBUTE_UNUSED) const OVERRIDE {
91    return Location::FpuRegisterLocation(XMM0);
92  }
93
94 private:
95  DISALLOW_COPY_AND_ASSIGN(FieldAccessCallingConvetionX86_64);
96};
97
98
99class InvokeDexCallingConventionVisitorX86_64 : public InvokeDexCallingConventionVisitor {
100 public:
101  InvokeDexCallingConventionVisitorX86_64() {}
102  virtual ~InvokeDexCallingConventionVisitorX86_64() {}
103
104  Location GetNextLocation(Primitive::Type type) OVERRIDE;
105  Location GetReturnLocation(Primitive::Type type) const OVERRIDE;
106  Location GetMethodLocation() const OVERRIDE;
107
108 private:
109  InvokeDexCallingConvention calling_convention;
110
111  DISALLOW_COPY_AND_ASSIGN(InvokeDexCallingConventionVisitorX86_64);
112};
113
114class CodeGeneratorX86_64;
115
116class SlowPathCodeX86_64 : public SlowPathCode {
117 public:
118  SlowPathCodeX86_64() : entry_label_(), exit_label_() {}
119
120  Label* GetEntryLabel() { return &entry_label_; }
121  Label* GetExitLabel() { return &exit_label_; }
122
123 private:
124  Label entry_label_;
125  Label exit_label_;
126
127  DISALLOW_COPY_AND_ASSIGN(SlowPathCodeX86_64);
128};
129
130class ParallelMoveResolverX86_64 : public ParallelMoveResolverWithSwap {
131 public:
132  ParallelMoveResolverX86_64(ArenaAllocator* allocator, CodeGeneratorX86_64* codegen)
133      : ParallelMoveResolverWithSwap(allocator), codegen_(codegen) {}
134
135  void EmitMove(size_t index) OVERRIDE;
136  void EmitSwap(size_t index) OVERRIDE;
137  void SpillScratch(int reg) OVERRIDE;
138  void RestoreScratch(int reg) OVERRIDE;
139
140  X86_64Assembler* GetAssembler() const;
141
142 private:
143  void Exchange32(CpuRegister reg, int mem);
144  void Exchange32(XmmRegister reg, int mem);
145  void Exchange32(int mem1, int mem2);
146  void Exchange64(CpuRegister reg, int mem);
147  void Exchange64(XmmRegister reg, int mem);
148  void Exchange64(int mem1, int mem2);
149
150  CodeGeneratorX86_64* const codegen_;
151
152  DISALLOW_COPY_AND_ASSIGN(ParallelMoveResolverX86_64);
153};
154
155class LocationsBuilderX86_64 : public HGraphVisitor {
156 public:
157  LocationsBuilderX86_64(HGraph* graph, CodeGeneratorX86_64* codegen)
158      : HGraphVisitor(graph), codegen_(codegen) {}
159
160#define DECLARE_VISIT_INSTRUCTION(name, super)     \
161  void Visit##name(H##name* instr) OVERRIDE;
162
163  FOR_EACH_CONCRETE_INSTRUCTION_COMMON(DECLARE_VISIT_INSTRUCTION)
164  FOR_EACH_CONCRETE_INSTRUCTION_X86_64(DECLARE_VISIT_INSTRUCTION)
165
166#undef DECLARE_VISIT_INSTRUCTION
167
168  void VisitInstruction(HInstruction* instruction) OVERRIDE {
169    LOG(FATAL) << "Unreachable instruction " << instruction->DebugName()
170               << " (id " << instruction->GetId() << ")";
171  }
172
173 private:
174  void HandleInvoke(HInvoke* invoke);
175  void HandleBitwiseOperation(HBinaryOperation* operation);
176  void HandleShift(HBinaryOperation* operation);
177  void HandleFieldSet(HInstruction* instruction, const FieldInfo& field_info);
178  void HandleFieldGet(HInstruction* instruction);
179
180  CodeGeneratorX86_64* const codegen_;
181  InvokeDexCallingConventionVisitorX86_64 parameter_visitor_;
182
183  DISALLOW_COPY_AND_ASSIGN(LocationsBuilderX86_64);
184};
185
186class InstructionCodeGeneratorX86_64 : public HGraphVisitor {
187 public:
188  InstructionCodeGeneratorX86_64(HGraph* graph, CodeGeneratorX86_64* codegen);
189
190#define DECLARE_VISIT_INSTRUCTION(name, super)     \
191  void Visit##name(H##name* instr) OVERRIDE;
192
193  FOR_EACH_CONCRETE_INSTRUCTION_COMMON(DECLARE_VISIT_INSTRUCTION)
194  FOR_EACH_CONCRETE_INSTRUCTION_X86_64(DECLARE_VISIT_INSTRUCTION)
195
196#undef DECLARE_VISIT_INSTRUCTION
197
198  void VisitInstruction(HInstruction* instruction) OVERRIDE {
199    LOG(FATAL) << "Unreachable instruction " << instruction->DebugName()
200               << " (id " << instruction->GetId() << ")";
201  }
202
203  X86_64Assembler* GetAssembler() const { return assembler_; }
204
205 private:
206  // Generate code for the given suspend check. If not null, `successor`
207  // is the block to branch to if the suspend check is not needed, and after
208  // the suspend call.
209  void GenerateSuspendCheck(HSuspendCheck* instruction, HBasicBlock* successor);
210  void GenerateClassInitializationCheck(SlowPathCodeX86_64* slow_path, CpuRegister class_reg);
211  void HandleBitwiseOperation(HBinaryOperation* operation);
212  void GenerateRemFP(HRem* rem);
213  void DivRemOneOrMinusOne(HBinaryOperation* instruction);
214  void DivByPowerOfTwo(HDiv* instruction);
215  void GenerateDivRemWithAnyConstant(HBinaryOperation* instruction);
216  void GenerateDivRemIntegral(HBinaryOperation* instruction);
217  void HandleShift(HBinaryOperation* operation);
218  void GenerateMemoryBarrier(MemBarrierKind kind);
219  void HandleFieldSet(HInstruction* instruction,
220                      const FieldInfo& field_info,
221                      bool value_can_be_null);
222  void HandleFieldGet(HInstruction* instruction, const FieldInfo& field_info);
223  void GenerateImplicitNullCheck(HNullCheck* instruction);
224  void GenerateExplicitNullCheck(HNullCheck* instruction);
225  void PushOntoFPStack(Location source, uint32_t temp_offset,
226                       uint32_t stack_adjustment, bool is_float);
227  void GenerateTestAndBranch(HInstruction* instruction,
228                             Label* true_target,
229                             Label* false_target,
230                             Label* always_true_target);
231  void GenerateCompareTestAndBranch(HIf* if_inst,
232                                    HCondition* condition,
233                                    Label* true_target,
234                                    Label* false_target,
235                                    Label* always_true_target);
236  void GenerateFPJumps(HCondition* cond, Label* true_label, Label* false_label);
237  void HandleGoto(HInstruction* got, HBasicBlock* successor);
238
239  X86_64Assembler* const assembler_;
240  CodeGeneratorX86_64* const codegen_;
241
242  DISALLOW_COPY_AND_ASSIGN(InstructionCodeGeneratorX86_64);
243};
244
245class CodeGeneratorX86_64 : public CodeGenerator {
246 public:
247  CodeGeneratorX86_64(HGraph* graph,
248                  const X86_64InstructionSetFeatures& isa_features,
249                  const CompilerOptions& compiler_options,
250                  OptimizingCompilerStats* stats = nullptr);
251  virtual ~CodeGeneratorX86_64() {}
252
253  void GenerateFrameEntry() OVERRIDE;
254  void GenerateFrameExit() OVERRIDE;
255  void Bind(HBasicBlock* block) OVERRIDE;
256  void Move(HInstruction* instruction, Location location, HInstruction* move_for) OVERRIDE;
257  void MoveConstant(Location destination, int32_t value) OVERRIDE;
258  void AddLocationAsTemp(Location location, LocationSummary* locations) OVERRIDE;
259  void MoveLocationToTemp(Location source,
260                          const LocationSummary& locations,
261                          int temp_index,
262                          Primitive::Type type) OVERRIDE;
263  void MoveTempToLocation(const LocationSummary& locations,
264                          int temp_index,
265                          Location destination,
266                          Primitive::Type type) OVERRIDE;
267  size_t SaveCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
268  size_t RestoreCoreRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
269  size_t SaveFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
270  size_t RestoreFloatingPointRegister(size_t stack_index, uint32_t reg_id) OVERRIDE;
271
272  // Generate code to invoke a runtime entry point.
273  void InvokeRuntime(QuickEntrypointEnum entrypoint,
274                     HInstruction* instruction,
275                     uint32_t dex_pc,
276                     SlowPathCode* slow_path) OVERRIDE;
277
278  void InvokeRuntime(int32_t entry_point_offset,
279                     HInstruction* instruction,
280                     uint32_t dex_pc,
281                     SlowPathCode* slow_path);
282
283  size_t GetWordSize() const OVERRIDE {
284    return kX86_64WordSize;
285  }
286
287  size_t GetFloatingPointSpillSlotSize() const OVERRIDE {
288    return kX86_64WordSize;
289  }
290
291  HGraphVisitor* GetLocationBuilder() OVERRIDE {
292    return &location_builder_;
293  }
294
295  HGraphVisitor* GetInstructionVisitor() OVERRIDE {
296    return &instruction_visitor_;
297  }
298
299  X86_64Assembler* GetAssembler() OVERRIDE {
300    return &assembler_;
301  }
302
303  const X86_64Assembler& GetAssembler() const OVERRIDE {
304    return assembler_;
305  }
306
307  ParallelMoveResolverX86_64* GetMoveResolver() OVERRIDE {
308    return &move_resolver_;
309  }
310
311  uintptr_t GetAddressOf(HBasicBlock* block) const OVERRIDE {
312    return GetLabelOf(block)->Position();
313  }
314
315  Location GetStackLocation(HLoadLocal* load) const OVERRIDE;
316
317  void SetupBlockedRegisters(bool is_baseline) const OVERRIDE;
318  Location AllocateFreeRegister(Primitive::Type type) const OVERRIDE;
319  void DumpCoreRegister(std::ostream& stream, int reg) const OVERRIDE;
320  void DumpFloatingPointRegister(std::ostream& stream, int reg) const OVERRIDE;
321  void Finalize(CodeAllocator* allocator) OVERRIDE;
322
323  InstructionSet GetInstructionSet() const OVERRIDE {
324    return InstructionSet::kX86_64;
325  }
326
327  // Emit a write barrier.
328  void MarkGCCard(CpuRegister temp,
329                  CpuRegister card,
330                  CpuRegister object,
331                  CpuRegister value,
332                  bool value_can_be_null);
333
334  // Helper method to move a value between two locations.
335  void Move(Location destination, Location source);
336
337  Label* GetLabelOf(HBasicBlock* block) const {
338    return CommonGetLabelOf<Label>(block_labels_.GetRawStorage(), block);
339  }
340
341  void Initialize() OVERRIDE {
342    block_labels_.SetSize(GetGraph()->GetBlocks().size());
343  }
344
345  bool NeedsTwoRegisters(Primitive::Type type ATTRIBUTE_UNUSED) const OVERRIDE {
346    return false;
347  }
348
349  void GenerateStaticOrDirectCall(HInvokeStaticOrDirect* invoke, Location temp);
350  void GenerateVirtualCall(HInvokeVirtual* invoke, Location temp);
351
352  void EmitLinkerPatches(ArenaVector<LinkerPatch>* linker_patches) OVERRIDE;
353
354  const X86_64InstructionSetFeatures& GetInstructionSetFeatures() const {
355    return isa_features_;
356  }
357
358  int ConstantAreaStart() const {
359    return constant_area_start_;
360  }
361
362  Address LiteralDoubleAddress(double v);
363  Address LiteralFloatAddress(float v);
364  Address LiteralInt32Address(int32_t v);
365  Address LiteralInt64Address(int64_t v);
366
367  // Load a 64 bit value into a register in the most efficient manner.
368  void Load64BitValue(CpuRegister dest, int64_t value);
369
370  // Store a 64 bit value into a DoubleStackSlot in the most efficient manner.
371  void Store64BitValueToStack(Location dest, int64_t value);
372
373 private:
374  struct PcRelativeDexCacheAccessInfo {
375    PcRelativeDexCacheAccessInfo(const DexFile& dex_file, uint32_t element_off)
376        : target_dex_file(dex_file), element_offset(element_off), label() { }
377
378    const DexFile& target_dex_file;
379    uint32_t element_offset;
380    Label label;
381  };
382
383  // Labels for each block that will be compiled.
384  GrowableArray<Label> block_labels_;
385  Label frame_entry_label_;
386  LocationsBuilderX86_64 location_builder_;
387  InstructionCodeGeneratorX86_64 instruction_visitor_;
388  ParallelMoveResolverX86_64 move_resolver_;
389  X86_64Assembler assembler_;
390  const X86_64InstructionSetFeatures& isa_features_;
391
392  // Offset to the start of the constant area in the assembled code.
393  // Used for fixups to the constant area.
394  int constant_area_start_;
395
396  // Method patch info. Using ArenaDeque<> which retains element addresses on push/emplace_back().
397  ArenaDeque<MethodPatchInfo<Label>> method_patches_;
398  ArenaDeque<MethodPatchInfo<Label>> relative_call_patches_;
399  // PC-relative DexCache access info.
400  ArenaDeque<PcRelativeDexCacheAccessInfo> pc_rel_dex_cache_patches_;
401
402  // When we don't know the proper offset for the value, we use kDummy32BitOffset.
403  // We will fix this up in the linker later to have the right value.
404  static constexpr int32_t kDummy32BitOffset = 256;
405
406  DISALLOW_COPY_AND_ASSIGN(CodeGeneratorX86_64);
407};
408
409}  // namespace x86_64
410}  // namespace art
411
412#endif  // ART_COMPILER_OPTIMIZING_CODE_GENERATOR_X86_64_H_
413