optimizing_compiler.cc revision 175dc732c80e6f2afd83209348124df349290ba8 
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#include "optimizing_compiler.h"
18
19#include <fstream>
20#include <stdint.h>
21
22#ifdef ART_ENABLE_CODEGEN_arm64
23#include "instruction_simplifier_arm64.h"
24#endif
25
26#ifdef ART_ENABLE_CODEGEN_x86
27#include "constant_area_fixups_x86.h"
28#endif
29
30#include "art_method-inl.h"
31#include "base/arena_allocator.h"
32#include "base/arena_containers.h"
33#include "base/dumpable.h"
34#include "base/timing_logger.h"
35#include "boolean_simplifier.h"
36#include "bounds_check_elimination.h"
37#include "builder.h"
38#include "code_generator.h"
39#include "compiled_method.h"
40#include "compiler.h"
41#include "constant_folding.h"
42#include "dead_code_elimination.h"
43#include "dex/quick/dex_file_to_method_inliner_map.h"
44#include "dex/verified_method.h"
45#include "dex/verification_results.h"
46#include "driver/compiler_driver.h"
47#include "driver/compiler_driver-inl.h"
48#include "driver/compiler_options.h"
49#include "driver/dex_compilation_unit.h"
50#include "elf_writer_quick.h"
51#include "graph_checker.h"
52#include "graph_visualizer.h"
53#include "gvn.h"
54#include "induction_var_analysis.h"
55#include "inliner.h"
56#include "instruction_simplifier.h"
57#include "intrinsics.h"
58#include "licm.h"
59#include "jni/quick/jni_compiler.h"
60#include "nodes.h"
61#include "prepare_for_register_allocation.h"
62#include "reference_type_propagation.h"
63#include "register_allocator.h"
64#include "side_effects_analysis.h"
65#include "ssa_builder.h"
66#include "ssa_phi_elimination.h"
67#include "ssa_liveness_analysis.h"
68#include "utils/assembler.h"
69#include "verifier/method_verifier.h"
70
71namespace art {
72
73/**
74 * Used by the code generator, to allocate the code in a vector.
75 */
76class CodeVectorAllocator FINAL : public CodeAllocator {
77 public:
78  explicit CodeVectorAllocator(ArenaAllocator* arena)
79      : memory_(arena->Adapter(kArenaAllocCodeBuffer)),
80        size_(0) {}
81
82  virtual uint8_t* Allocate(size_t size) {
83    size_ = size;
84    memory_.resize(size);
85    return &memory_[0];
86  }
87
88  size_t GetSize() const { return size_; }
89  const ArenaVector<uint8_t>& GetMemory() const { return memory_; }
90
91 private:
92  ArenaVector<uint8_t> memory_;
93  size_t size_;
94
95  DISALLOW_COPY_AND_ASSIGN(CodeVectorAllocator);
96};
97
98/**
99 * Filter to apply to the visualizer. Methods whose name contain that filter will
100 * be dumped.
101 */
102static constexpr const char kStringFilter[] = "";
103
104class PassScope;
105
106class PassObserver : public ValueObject {
107 public:
108  PassObserver(HGraph* graph,
109               const char* method_name,
110               CodeGenerator* codegen,
111               std::ostream* visualizer_output,
112               CompilerDriver* compiler_driver)
113      : graph_(graph),
114        method_name_(method_name),
115        timing_logger_enabled_(compiler_driver->GetDumpPasses()),
116        timing_logger_(method_name, true, true),
117        disasm_info_(graph->GetArena()),
118        visualizer_enabled_(!compiler_driver->GetDumpCfgFileName().empty()),
119        visualizer_(visualizer_output, graph, *codegen),
120        graph_in_bad_state_(false) {
121    if (timing_logger_enabled_ || visualizer_enabled_) {
122      if (!IsVerboseMethod(compiler_driver, method_name)) {
123        timing_logger_enabled_ = visualizer_enabled_ = false;
124      }
125      if (visualizer_enabled_) {
126        visualizer_.PrintHeader(method_name_);
127        codegen->SetDisassemblyInformation(&disasm_info_);
128      }
129    }
130  }
131
132  ~PassObserver() {
133    if (timing_logger_enabled_) {
134      LOG(INFO) << "TIMINGS " << method_name_;
135      LOG(INFO) << Dumpable<TimingLogger>(timing_logger_);
136    }
137  }
138
139  void DumpDisassembly() const {
140    if (visualizer_enabled_) {
141      visualizer_.DumpGraphWithDisassembly();
142    }
143  }
144
145  void SetGraphInBadState() { graph_in_bad_state_ = true; }
146
147 private:
148  void StartPass(const char* pass_name) {
149    // Dump graph first, then start timer.
150    if (visualizer_enabled_) {
151      visualizer_.DumpGraph(pass_name, /* is_after_pass */ false, graph_in_bad_state_);
152    }
153    if (timing_logger_enabled_) {
154      timing_logger_.StartTiming(pass_name);
155    }
156  }
157
158  void EndPass(const char* pass_name) {
159    // Pause timer first, then dump graph.
160    if (timing_logger_enabled_) {
161      timing_logger_.EndTiming();
162    }
163    if (visualizer_enabled_) {
164      visualizer_.DumpGraph(pass_name, /* is_after_pass */ true, graph_in_bad_state_);
165    }
166
167    // Validate the HGraph if running in debug mode.
168    if (kIsDebugBuild) {
169      if (!graph_in_bad_state_) {
170        if (graph_->IsInSsaForm()) {
171          SSAChecker checker(graph_->GetArena(), graph_);
172          checker.Run();
173          if (!checker.IsValid()) {
174            LOG(FATAL) << "Error after " << pass_name << ": " << Dumpable<SSAChecker>(checker);
175          }
176        } else {
177          GraphChecker checker(graph_->GetArena(), graph_);
178          checker.Run();
179          if (!checker.IsValid()) {
180            LOG(FATAL) << "Error after " << pass_name << ": " << Dumpable<GraphChecker>(checker);
181          }
182        }
183      }
184    }
185  }
186
187  static bool IsVerboseMethod(CompilerDriver* compiler_driver, const char* method_name) {
188    // Test an exact match to --verbose-methods. If verbose-methods is set, this overrides an
189    // empty kStringFilter matching all methods.
190    if (compiler_driver->GetCompilerOptions().HasVerboseMethods()) {
191      return compiler_driver->GetCompilerOptions().IsVerboseMethod(method_name);
192    }
193
194    // Test the kStringFilter sub-string. constexpr helper variable to silence unreachable-code
195    // warning when the string is empty.
196    constexpr bool kStringFilterEmpty = arraysize(kStringFilter) <= 1;
197    if (kStringFilterEmpty || strstr(method_name, kStringFilter) != nullptr) {
198      return true;
199    }
200
201    return false;
202  }
203
204  HGraph* const graph_;
205  const char* method_name_;
206
207  bool timing_logger_enabled_;
208  TimingLogger timing_logger_;
209
210  DisassemblyInformation disasm_info_;
211
212  bool visualizer_enabled_;
213  HGraphVisualizer visualizer_;
214
215  // Flag to be set by the compiler if the pass failed and the graph is not
216  // expected to validate.
217  bool graph_in_bad_state_;
218
219  friend PassScope;
220
221  DISALLOW_COPY_AND_ASSIGN(PassObserver);
222};
223
224class PassScope : public ValueObject {
225 public:
226  PassScope(const char *pass_name, PassObserver* pass_observer)
227      : pass_name_(pass_name),
228        pass_observer_(pass_observer) {
229    pass_observer_->StartPass(pass_name_);
230  }
231
232  ~PassScope() {
233    pass_observer_->EndPass(pass_name_);
234  }
235
236 private:
237  const char* const pass_name_;
238  PassObserver* const pass_observer_;
239};
240
241class OptimizingCompiler FINAL : public Compiler {
242 public:
243  explicit OptimizingCompiler(CompilerDriver* driver);
244  ~OptimizingCompiler();
245
246  bool CanCompileMethod(uint32_t method_idx, const DexFile& dex_file, CompilationUnit* cu) const
247      OVERRIDE;
248
249  CompiledMethod* Compile(const DexFile::CodeItem* code_item,
250                          uint32_t access_flags,
251                          InvokeType invoke_type,
252                          uint16_t class_def_idx,
253                          uint32_t method_idx,
254                          jobject class_loader,
255                          const DexFile& dex_file,
256                          Handle<mirror::DexCache> dex_cache) const OVERRIDE;
257
258  CompiledMethod* TryCompile(const DexFile::CodeItem* code_item,
259                             uint32_t access_flags,
260                             InvokeType invoke_type,
261                             uint16_t class_def_idx,
262                             uint32_t method_idx,
263                             jobject class_loader,
264                             const DexFile& dex_file,
265                             Handle<mirror::DexCache> dex_cache) const;
266
267  CompiledMethod* JniCompile(uint32_t access_flags,
268                             uint32_t method_idx,
269                             const DexFile& dex_file) const OVERRIDE {
270    return ArtQuickJniCompileMethod(GetCompilerDriver(), access_flags, method_idx, dex_file);
271  }
272
273  uintptr_t GetEntryPointOf(ArtMethod* method) const OVERRIDE
274      SHARED_REQUIRES(Locks::mutator_lock_) {
275    return reinterpret_cast<uintptr_t>(method->GetEntryPointFromQuickCompiledCodePtrSize(
276        InstructionSetPointerSize(GetCompilerDriver()->GetInstructionSet())));
277  }
278
279  void InitCompilationUnit(CompilationUnit& cu) const OVERRIDE;
280
281  void Init() OVERRIDE;
282
283  void UnInit() const OVERRIDE;
284
285  void MaybeRecordStat(MethodCompilationStat compilation_stat) const {
286    if (compilation_stats_.get() != nullptr) {
287      compilation_stats_->RecordStat(compilation_stat);
288    }
289  }
290
291 private:
292  // Whether we should run any optimization or register allocation. If false, will
293  // just run the code generation after the graph was built.
294  const bool run_optimizations_;
295
296  // Optimize and compile `graph`.
297  CompiledMethod* CompileOptimized(HGraph* graph,
298                                   CodeGenerator* codegen,
299                                   CompilerDriver* driver,
300                                   const DexCompilationUnit& dex_compilation_unit,
301                                   PassObserver* pass_observer) const;
302
303  // Just compile without doing optimizations.
304  CompiledMethod* CompileBaseline(CodeGenerator* codegen,
305                                  CompilerDriver* driver,
306                                  const DexCompilationUnit& dex_compilation_unit,
307                                  PassObserver* pass_observer) const;
308
309  std::unique_ptr<OptimizingCompilerStats> compilation_stats_;
310
311  std::unique_ptr<std::ostream> visualizer_output_;
312
313  // Delegate to Quick in case the optimizing compiler cannot compile a method.
314  std::unique_ptr<Compiler> delegate_;
315
316  DISALLOW_COPY_AND_ASSIGN(OptimizingCompiler);
317};
318
319static const int kMaximumCompilationTimeBeforeWarning = 100; /* ms */
320
321OptimizingCompiler::OptimizingCompiler(CompilerDriver* driver)
322    : Compiler(driver, kMaximumCompilationTimeBeforeWarning),
323      run_optimizations_(
324          (driver->GetCompilerOptions().GetCompilerFilter() != CompilerOptions::kTime)
325          && !driver->GetCompilerOptions().GetDebuggable()),
326      delegate_(Create(driver, Compiler::Kind::kQuick)) {}
327
328void OptimizingCompiler::Init() {
329  delegate_->Init();
330  // Enable C1visualizer output. Must be done in Init() because the compiler
331  // driver is not fully initialized when passed to the compiler's constructor.
332  CompilerDriver* driver = GetCompilerDriver();
333  const std::string cfg_file_name = driver->GetDumpCfgFileName();
334  if (!cfg_file_name.empty()) {
335    CHECK_EQ(driver->GetThreadCount(), 1U)
336      << "Graph visualizer requires the compiler to run single-threaded. "
337      << "Invoke the compiler with '-j1'.";
338    std::ios_base::openmode cfg_file_mode =
339        driver->GetDumpCfgAppend() ? std::ofstream::app : std::ofstream::out;
340    visualizer_output_.reset(new std::ofstream(cfg_file_name, cfg_file_mode));
341  }
342  if (driver->GetDumpStats()) {
343    compilation_stats_.reset(new OptimizingCompilerStats());
344  }
345}
346
347void OptimizingCompiler::UnInit() const {
348  delegate_->UnInit();
349}
350
351OptimizingCompiler::~OptimizingCompiler() {
352  if (compilation_stats_.get() != nullptr) {
353    compilation_stats_->Log();
354  }
355}
356
357void OptimizingCompiler::InitCompilationUnit(CompilationUnit& cu) const {
358  delegate_->InitCompilationUnit(cu);
359}
360
361bool OptimizingCompiler::CanCompileMethod(uint32_t method_idx ATTRIBUTE_UNUSED,
362                                          const DexFile& dex_file ATTRIBUTE_UNUSED,
363                                          CompilationUnit* cu ATTRIBUTE_UNUSED) const {
364  return true;
365}
366
367static bool IsInstructionSetSupported(InstructionSet instruction_set) {
368  return instruction_set == kArm64
369      || (instruction_set == kThumb2 && !kArm32QuickCodeUseSoftFloat)
370      || instruction_set == kMips64
371      || instruction_set == kX86
372      || instruction_set == kX86_64;
373}
374
375static void RunOptimizations(HOptimization* optimizations[],
376                             size_t length,
377                             PassObserver* pass_observer) {
378  for (size_t i = 0; i < length; ++i) {
379    PassScope scope(optimizations[i]->GetPassName(), pass_observer);
380    optimizations[i]->Run();
381  }
382}
383
384static void MaybeRunInliner(HGraph* graph,
385                            CompilerDriver* driver,
386                            OptimizingCompilerStats* stats,
387                            const DexCompilationUnit& dex_compilation_unit,
388                            PassObserver* pass_observer,
389                            StackHandleScopeCollection* handles) {
390  const CompilerOptions& compiler_options = driver->GetCompilerOptions();
391  bool should_inline = (compiler_options.GetInlineDepthLimit() > 0)
392      && (compiler_options.GetInlineMaxCodeUnits() > 0);
393  if (!should_inline) {
394    return;
395  }
396
397  ArenaAllocator* arena = graph->GetArena();
398  HInliner* inliner = new (arena) HInliner(
399    graph, dex_compilation_unit, dex_compilation_unit, driver, handles, stats);
400  ReferenceTypePropagation* type_propagation =
401    new (arena) ReferenceTypePropagation(graph, handles,
402        "reference_type_propagation_after_inlining");
403
404  HOptimization* optimizations[] = {
405    inliner,
406    // Run another type propagation phase: inlining will open up more opportunities
407    // to remove checkcast/instanceof and null checks.
408    type_propagation,
409  };
410
411  RunOptimizations(optimizations, arraysize(optimizations), pass_observer);
412}
413
414static void RunArchOptimizations(InstructionSet instruction_set,
415                                 HGraph* graph,
416                                 OptimizingCompilerStats* stats,
417                                 PassObserver* pass_observer) {
418  ArenaAllocator* arena = graph->GetArena();
419  switch (instruction_set) {
420#ifdef ART_ENABLE_CODEGEN_arm64
421    case kArm64: {
422      arm64::InstructionSimplifierArm64* simplifier =
423          new (arena) arm64::InstructionSimplifierArm64(graph, stats);
424      SideEffectsAnalysis* side_effects = new (arena) SideEffectsAnalysis(graph);
425      GVNOptimization* gvn = new (arena) GVNOptimization(graph, *side_effects, "GVN_after_arch");
426      HOptimization* arm64_optimizations[] = {
427        simplifier,
428        side_effects,
429        gvn
430      };
431      RunOptimizations(arm64_optimizations, arraysize(arm64_optimizations), pass_observer);
432      break;
433    }
434#endif
435#ifdef ART_ENABLE_CODEGEN_x86
436    case kX86: {
437      x86::ConstantAreaFixups* constant_area_fixups =
438          new (arena) x86::ConstantAreaFixups(graph, stats);
439      HOptimization* x86_optimizations[] = {
440        constant_area_fixups
441      };
442      RunOptimizations(x86_optimizations, arraysize(x86_optimizations), pass_observer);
443      break;
444    }
445#endif
446    default:
447      break;
448  }
449}
450
451static void RunOptimizations(HGraph* graph,
452                             CompilerDriver* driver,
453                             OptimizingCompilerStats* stats,
454                             const DexCompilationUnit& dex_compilation_unit,
455                             PassObserver* pass_observer,
456                             StackHandleScopeCollection* handles) {
457  ArenaAllocator* arena = graph->GetArena();
458  HDeadCodeElimination* dce1 = new (arena) HDeadCodeElimination(
459      graph, stats, HDeadCodeElimination::kInitialDeadCodeEliminationPassName);
460  HDeadCodeElimination* dce2 = new (arena) HDeadCodeElimination(
461      graph, stats, HDeadCodeElimination::kFinalDeadCodeEliminationPassName);
462  HConstantFolding* fold1 = new (arena) HConstantFolding(graph);
463  InstructionSimplifier* simplify1 = new (arena) InstructionSimplifier(graph, stats);
464  HBooleanSimplifier* boolean_simplify = new (arena) HBooleanSimplifier(graph);
465  HConstantFolding* fold2 = new (arena) HConstantFolding(graph, "constant_folding_after_inlining");
466  SideEffectsAnalysis* side_effects = new (arena) SideEffectsAnalysis(graph);
467  GVNOptimization* gvn = new (arena) GVNOptimization(graph, *side_effects);
468  LICM* licm = new (arena) LICM(graph, *side_effects);
469  HInductionVarAnalysis* induction = new (arena) HInductionVarAnalysis(graph);
470  BoundsCheckElimination* bce = new (arena) BoundsCheckElimination(graph, induction);
471  ReferenceTypePropagation* type_propagation =
472      new (arena) ReferenceTypePropagation(graph, handles);
473  InstructionSimplifier* simplify2 = new (arena) InstructionSimplifier(
474      graph, stats, "instruction_simplifier_after_types");
475  InstructionSimplifier* simplify3 = new (arena) InstructionSimplifier(
476      graph, stats, "instruction_simplifier_after_bce");
477  InstructionSimplifier* simplify4 = new (arena) InstructionSimplifier(
478      graph, stats, "instruction_simplifier_before_codegen");
479
480  IntrinsicsRecognizer* intrinsics = new (arena) IntrinsicsRecognizer(graph, driver);
481
482  HOptimization* optimizations1[] = {
483    intrinsics,
484    fold1,
485    simplify1,
486    type_propagation,
487    dce1,
488    simplify2
489  };
490
491  RunOptimizations(optimizations1, arraysize(optimizations1), pass_observer);
492
493  // TODO: Update passes incompatible with try/catch so we have the same
494  //       pipeline for all methods.
495  if (graph->HasTryCatch()) {
496    HOptimization* optimizations2[] = {
497      side_effects,
498      gvn,
499      dce2,
500      // The codegen has a few assumptions that only the instruction simplifier
501      // can satisfy. For example, the code generator does not expect to see a
502      // HTypeConversion from a type to the same type.
503      simplify4,
504    };
505
506    RunOptimizations(optimizations2, arraysize(optimizations2), pass_observer);
507  } else {
508    MaybeRunInliner(graph, driver, stats, dex_compilation_unit, pass_observer, handles);
509
510    HOptimization* optimizations2[] = {
511      // BooleanSimplifier depends on the InstructionSimplifier removing
512      // redundant suspend checks to recognize empty blocks.
513      boolean_simplify,
514      fold2,  // TODO: if we don't inline we can also skip fold2.
515      side_effects,
516      gvn,
517      licm,
518      induction,
519      bce,
520      simplify3,
521      dce2,
522      // The codegen has a few assumptions that only the instruction simplifier
523      // can satisfy. For example, the code generator does not expect to see a
524      // HTypeConversion from a type to the same type.
525      simplify4,
526    };
527
528    RunOptimizations(optimizations2, arraysize(optimizations2), pass_observer);
529  }
530
531  RunArchOptimizations(driver->GetInstructionSet(), graph, stats, pass_observer);
532}
533
534// The stack map we generate must be 4-byte aligned on ARM. Since existing
535// maps are generated alongside these stack maps, we must also align them.
536static ArrayRef<const uint8_t> AlignVectorSize(ArenaVector<uint8_t>& vector) {
537  size_t size = vector.size();
538  size_t aligned_size = RoundUp(size, 4);
539  for (; size < aligned_size; ++size) {
540    vector.push_back(0);
541  }
542  return ArrayRef<const uint8_t>(vector);
543}
544
545static void AllocateRegisters(HGraph* graph,
546                              CodeGenerator* codegen,
547                              PassObserver* pass_observer) {
548  PrepareForRegisterAllocation(graph).Run();
549  SsaLivenessAnalysis liveness(graph, codegen);
550  {
551    PassScope scope(SsaLivenessAnalysis::kLivenessPassName, pass_observer);
552    liveness.Analyze();
553  }
554  {
555    PassScope scope(RegisterAllocator::kRegisterAllocatorPassName, pass_observer);
556    RegisterAllocator(graph->GetArena(), codegen, liveness).AllocateRegisters();
557  }
558}
559
560static ArenaVector<LinkerPatch> EmitAndSortLinkerPatches(CodeGenerator* codegen) {
561  ArenaVector<LinkerPatch> linker_patches(codegen->GetGraph()->GetArena()->Adapter());
562  codegen->EmitLinkerPatches(&linker_patches);
563
564  // Sort patches by literal offset. Required for .oat_patches encoding.
565  std::sort(linker_patches.begin(), linker_patches.end(),
566            [](const LinkerPatch& lhs, const LinkerPatch& rhs) {
567    return lhs.LiteralOffset() < rhs.LiteralOffset();
568  });
569
570  return linker_patches;
571}
572
573CompiledMethod* OptimizingCompiler::CompileOptimized(HGraph* graph,
574                                                     CodeGenerator* codegen,
575                                                     CompilerDriver* compiler_driver,
576                                                     const DexCompilationUnit& dex_compilation_unit,
577                                                     PassObserver* pass_observer) const {
578  if (graph->HasTryCatch() && graph->IsDebuggable()) {
579    // TODO: b/24054676, stop creating catch phis eagerly to avoid special cases like phis without
580    // inputs.
581    return nullptr;
582  }
583
584  ScopedObjectAccess soa(Thread::Current());
585  StackHandleScopeCollection handles(soa.Self());
586  soa.Self()->TransitionFromRunnableToSuspended(kNative);
587  RunOptimizations(graph, compiler_driver, compilation_stats_.get(),
588                   dex_compilation_unit, pass_observer, &handles);
589
590  AllocateRegisters(graph, codegen, pass_observer);
591
592  ArenaAllocator* arena = graph->GetArena();
593  CodeVectorAllocator allocator(arena);
594  DefaultSrcMap src_mapping_table;
595  codegen->SetSrcMap(compiler_driver->GetCompilerOptions().GetGenerateDebugInfo()
596                         ? &src_mapping_table
597                         : nullptr);
598  codegen->CompileOptimized(&allocator);
599
600  ArenaVector<LinkerPatch> linker_patches = EmitAndSortLinkerPatches(codegen);
601
602  ArenaVector<uint8_t> stack_map(arena->Adapter(kArenaAllocStackMaps));
603  codegen->BuildStackMaps(&stack_map);
604
605  MaybeRecordStat(MethodCompilationStat::kCompiledOptimized);
606
607  CompiledMethod* compiled_method = CompiledMethod::SwapAllocCompiledMethod(
608      compiler_driver,
609      codegen->GetInstructionSet(),
610      ArrayRef<const uint8_t>(allocator.GetMemory()),
611      // Follow Quick's behavior and set the frame size to zero if it is
612      // considered "empty" (see the definition of
613      // art::CodeGenerator::HasEmptyFrame).
614      codegen->HasEmptyFrame() ? 0 : codegen->GetFrameSize(),
615      codegen->GetCoreSpillMask(),
616      codegen->GetFpuSpillMask(),
617      &src_mapping_table,
618      ArrayRef<const uint8_t>(),  // mapping_table.
619      ArrayRef<const uint8_t>(stack_map),
620      ArrayRef<const uint8_t>(),  // native_gc_map.
621      ArrayRef<const uint8_t>(*codegen->GetAssembler()->cfi().data()),
622      ArrayRef<const LinkerPatch>(linker_patches));
623  pass_observer->DumpDisassembly();
624
625  soa.Self()->TransitionFromSuspendedToRunnable();
626  return compiled_method;
627}
628
629CompiledMethod* OptimizingCompiler::CompileBaseline(
630    CodeGenerator* codegen,
631    CompilerDriver* compiler_driver,
632    const DexCompilationUnit& dex_compilation_unit,
633    PassObserver* pass_observer) const {
634  ArenaAllocator* arena = codegen->GetGraph()->GetArena();
635  CodeVectorAllocator allocator(arena);
636  DefaultSrcMap src_mapping_table;
637  codegen->SetSrcMap(compiler_driver->GetCompilerOptions().GetGenerateDebugInfo()
638                         ? &src_mapping_table
639                         : nullptr);
640  codegen->CompileBaseline(&allocator);
641
642  ArenaVector<LinkerPatch> linker_patches = EmitAndSortLinkerPatches(codegen);
643
644  ArenaVector<uint8_t> mapping_table(arena->Adapter(kArenaAllocBaselineMaps));
645  codegen->BuildMappingTable(&mapping_table);
646  ArenaVector<uint8_t> vmap_table(arena->Adapter(kArenaAllocBaselineMaps));
647  codegen->BuildVMapTable(&vmap_table);
648  ArenaVector<uint8_t> gc_map(arena->Adapter(kArenaAllocBaselineMaps));
649  codegen->BuildNativeGCMap(&gc_map, dex_compilation_unit);
650
651  MaybeRecordStat(MethodCompilationStat::kCompiledBaseline);
652  CompiledMethod* compiled_method = CompiledMethod::SwapAllocCompiledMethod(
653      compiler_driver,
654      codegen->GetInstructionSet(),
655      ArrayRef<const uint8_t>(allocator.GetMemory()),
656      // Follow Quick's behavior and set the frame size to zero if it is
657      // considered "empty" (see the definition of
658      // art::CodeGenerator::HasEmptyFrame).
659      codegen->HasEmptyFrame() ? 0 : codegen->GetFrameSize(),
660      codegen->GetCoreSpillMask(),
661      codegen->GetFpuSpillMask(),
662      &src_mapping_table,
663      AlignVectorSize(mapping_table),
664      AlignVectorSize(vmap_table),
665      AlignVectorSize(gc_map),
666      ArrayRef<const uint8_t>(*codegen->GetAssembler()->cfi().data()),
667      ArrayRef<const LinkerPatch>(linker_patches));
668  pass_observer->DumpDisassembly();
669  return compiled_method;
670}
671
672CompiledMethod* OptimizingCompiler::TryCompile(const DexFile::CodeItem* code_item,
673                                               uint32_t access_flags,
674                                               InvokeType invoke_type,
675                                               uint16_t class_def_idx,
676                                               uint32_t method_idx,
677                                               jobject class_loader,
678                                               const DexFile& dex_file,
679                                               Handle<mirror::DexCache> dex_cache) const {
680  UNUSED(invoke_type);
681  std::string method_name = PrettyMethod(method_idx, dex_file);
682  MaybeRecordStat(MethodCompilationStat::kAttemptCompilation);
683  CompilerDriver* compiler_driver = GetCompilerDriver();
684  InstructionSet instruction_set = compiler_driver->GetInstructionSet();
685  // Always use the thumb2 assembler: some runtime functionality (like implicit stack
686  // overflow checks) assume thumb2.
687  if (instruction_set == kArm) {
688    instruction_set = kThumb2;
689  }
690
691  // Do not attempt to compile on architectures we do not support.
692  if (!IsInstructionSetSupported(instruction_set)) {
693    MaybeRecordStat(MethodCompilationStat::kNotCompiledUnsupportedIsa);
694    return nullptr;
695  }
696
697  if (Compiler::IsPathologicalCase(*code_item, method_idx, dex_file)) {
698    MaybeRecordStat(MethodCompilationStat::kNotCompiledPathological);
699    return nullptr;
700  }
701
702  // Implementation of the space filter: do not compile a code item whose size in
703  // code units is bigger than 128.
704  static constexpr size_t kSpaceFilterOptimizingThreshold = 128;
705  const CompilerOptions& compiler_options = compiler_driver->GetCompilerOptions();
706  if ((compiler_options.GetCompilerFilter() == CompilerOptions::kSpace)
707      && (code_item->insns_size_in_code_units_ > kSpaceFilterOptimizingThreshold)) {
708    MaybeRecordStat(MethodCompilationStat::kNotCompiledSpaceFilter);
709    return nullptr;
710  }
711
712  DexCompilationUnit dex_compilation_unit(
713    nullptr, class_loader, Runtime::Current()->GetClassLinker(), dex_file, code_item,
714    class_def_idx, method_idx, access_flags,
715    compiler_driver->GetVerifiedMethod(&dex_file, method_idx), dex_cache);
716
717  bool requires_barrier = dex_compilation_unit.IsConstructor()
718      && compiler_driver->RequiresConstructorBarrier(Thread::Current(),
719                                                     dex_compilation_unit.GetDexFile(),
720                                                     dex_compilation_unit.GetClassDefIndex());
721  ArenaAllocator arena(Runtime::Current()->GetArenaPool());
722  HGraph* graph = new (&arena) HGraph(
723      &arena, dex_file, method_idx, requires_barrier, compiler_driver->GetInstructionSet(),
724      kInvalidInvokeType, compiler_driver->GetCompilerOptions().GetDebuggable());
725
726  // For testing purposes, we put a special marker on method names that should be compiled
727  // with this compiler. This makes sure we're not regressing.
728  bool shouldCompile = method_name.find("$opt$") != std::string::npos;
729  bool shouldOptimize = method_name.find("$opt$reg$") != std::string::npos && run_optimizations_;
730
731  std::unique_ptr<CodeGenerator> codegen(
732      CodeGenerator::Create(graph,
733                            instruction_set,
734                            *compiler_driver->GetInstructionSetFeatures(),
735                            compiler_driver->GetCompilerOptions()));
736  if (codegen.get() == nullptr) {
737    CHECK(!shouldCompile) << "Could not find code generator for optimizing compiler";
738    MaybeRecordStat(MethodCompilationStat::kNotCompiledNoCodegen);
739    return nullptr;
740  }
741  codegen->GetAssembler()->cfi().SetEnabled(
742      compiler_driver->GetCompilerOptions().GetGenerateDebugInfo());
743
744  PassObserver pass_observer(graph,
745                             method_name.c_str(),
746                             codegen.get(),
747                             visualizer_output_.get(),
748                             compiler_driver);
749
750  const uint8_t* interpreter_metadata = nullptr;
751  {
752    ScopedObjectAccess soa(Thread::Current());
753    StackHandleScope<1> hs(soa.Self());
754    Handle<mirror::ClassLoader> loader(hs.NewHandle(
755        soa.Decode<mirror::ClassLoader*>(class_loader)));
756    ArtMethod* art_method = compiler_driver->ResolveMethod(
757        soa, dex_cache, loader, &dex_compilation_unit, method_idx, invoke_type);
758    // We may not get a method, for example if its class is erroneous.
759    // TODO: Clean this up, the compiler driver should just pass the ArtMethod to compile.
760    if (art_method != nullptr) {
761      interpreter_metadata = art_method->GetQuickenedInfo();
762    }
763  }
764  HGraphBuilder builder(graph,
765                        &dex_compilation_unit,
766                        &dex_compilation_unit,
767                        &dex_file,
768                        compiler_driver,
769                        compilation_stats_.get(),
770                        interpreter_metadata,
771                        dex_cache);
772
773  VLOG(compiler) << "Building " << method_name;
774
775  {
776    PassScope scope(HGraphBuilder::kBuilderPassName, &pass_observer);
777    if (!builder.BuildGraph(*code_item)) {
778      DCHECK(!(IsCompilingWithCoreImage() && shouldCompile))
779          << "Could not build graph in optimizing compiler";
780      pass_observer.SetGraphInBadState();
781      return nullptr;
782    }
783  }
784
785  bool can_allocate_registers = RegisterAllocator::CanAllocateRegistersFor(*graph, instruction_set);
786
787  // `run_optimizations_` is set explicitly (either through a compiler filter
788  // or the debuggable flag). If it is set, we can run baseline. Otherwise, we fall back
789  // to Quick.
790  bool can_use_baseline = !run_optimizations_ && builder.CanUseBaselineForStringInit();
791  CompiledMethod* compiled_method = nullptr;
792  if (run_optimizations_ && can_allocate_registers) {
793    VLOG(compiler) << "Optimizing " << method_name;
794
795    {
796      PassScope scope(SsaBuilder::kSsaBuilderPassName, &pass_observer);
797      if (!graph->TryBuildingSsa()) {
798        // We could not transform the graph to SSA, bailout.
799        LOG(INFO) << "Skipping compilation of " << method_name << ": it contains a non natural loop";
800        MaybeRecordStat(MethodCompilationStat::kNotCompiledCannotBuildSSA);
801        pass_observer.SetGraphInBadState();
802        return nullptr;
803      }
804    }
805
806    compiled_method = CompileOptimized(graph,
807                                       codegen.get(),
808                                       compiler_driver,
809                                       dex_compilation_unit,
810                                       &pass_observer);
811  } else if (shouldOptimize && can_allocate_registers) {
812    LOG(FATAL) << "Could not allocate registers in optimizing compiler";
813    UNREACHABLE();
814  } else if (can_use_baseline) {
815    VLOG(compiler) << "Compile baseline " << method_name;
816
817    if (!run_optimizations_) {
818      MaybeRecordStat(MethodCompilationStat::kNotOptimizedDisabled);
819    } else if (!can_allocate_registers) {
820      MaybeRecordStat(MethodCompilationStat::kNotOptimizedRegisterAllocator);
821    }
822
823    compiled_method = CompileBaseline(codegen.get(),
824                                      compiler_driver,
825                                      dex_compilation_unit,
826                                      &pass_observer);
827  }
828
829  if (kArenaAllocatorCountAllocations) {
830    if (arena.BytesAllocated() > 4 * MB) {
831      MemStats mem_stats(arena.GetMemStats());
832      LOG(INFO) << PrettyMethod(method_idx, dex_file) << " " << Dumpable<MemStats>(mem_stats);
833    }
834  }
835
836  return compiled_method;
837}
838
839static bool HasOnlyUnresolvedFailures(const VerifiedMethod* verified_method) {
840  uint32_t unresolved_mask = verifier::VerifyError::VERIFY_ERROR_NO_CLASS;
841  return (verified_method->GetEncounteredVerificationFailures() & (~unresolved_mask)) == 0;
842}
843
844CompiledMethod* OptimizingCompiler::Compile(const DexFile::CodeItem* code_item,
845                                            uint32_t access_flags,
846                                            InvokeType invoke_type,
847                                            uint16_t class_def_idx,
848                                            uint32_t method_idx,
849                                            jobject jclass_loader,
850                                            const DexFile& dex_file,
851                                            Handle<mirror::DexCache> dex_cache) const {
852  CompilerDriver* compiler_driver = GetCompilerDriver();
853  CompiledMethod* method = nullptr;
854  const VerifiedMethod* verified_method = compiler_driver->GetVerifiedMethod(&dex_file, method_idx);
855  DCHECK(!verified_method->HasRuntimeThrow());
856  if (compiler_driver->IsMethodVerifiedWithoutFailures(method_idx, class_def_idx, dex_file)
857      || HasOnlyUnresolvedFailures(verified_method)) {
858     method = TryCompile(code_item, access_flags, invoke_type, class_def_idx,
859                         method_idx, jclass_loader, dex_file, dex_cache);
860  } else {
861    if (compiler_driver->GetCompilerOptions().VerifyAtRuntime()) {
862      MaybeRecordStat(MethodCompilationStat::kNotCompiledVerifyAtRuntime);
863    } else {
864      MaybeRecordStat(MethodCompilationStat::kNotCompiledClassNotVerified);
865    }
866  }
867
868  if (method != nullptr) {
869    return method;
870  }
871  method = delegate_->Compile(code_item, access_flags, invoke_type, class_def_idx, method_idx,
872                              jclass_loader, dex_file, dex_cache);
873
874  if (method != nullptr) {
875    MaybeRecordStat(MethodCompilationStat::kCompiledQuick);
876  }
877  return method;
878}
879
880Compiler* CreateOptimizingCompiler(CompilerDriver* driver) {
881  return new OptimizingCompiler(driver);
882}
883
884bool IsCompilingWithCoreImage() {
885  const std::string& image = Runtime::Current()->GetImageLocation();
886  return EndsWith(image, "core.art") || EndsWith(image, "core-optimizing.art");
887}
888
889}  // namespace art
890