optimizing_compiler.cc revision a4f81546373f4cb5fa6dfc135307ee0a1d930872 
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 <memory>
21#include <stdint.h>
22
23#ifdef ART_ENABLE_CODEGEN_arm
24#include "dex_cache_array_fixups_arm.h"
25#endif
26
27#ifdef ART_ENABLE_CODEGEN_arm64
28#include "instruction_simplifier_arm64.h"
29#endif
30
31#ifdef ART_ENABLE_CODEGEN_x86
32#include "pc_relative_fixups_x86.h"
33#endif
34
35#include "art_method-inl.h"
36#include "base/arena_allocator.h"
37#include "base/arena_containers.h"
38#include "base/dumpable.h"
39#include "base/macros.h"
40#include "base/timing_logger.h"
41#include "bounds_check_elimination.h"
42#include "builder.h"
43#include "code_generator.h"
44#include "compiled_method.h"
45#include "compiler.h"
46#include "constant_folding.h"
47#include "dead_code_elimination.h"
48#include "debug/elf_debug_writer.h"
49#include "debug/method_debug_info.h"
50#include "dex/quick/dex_file_to_method_inliner_map.h"
51#include "dex/verification_results.h"
52#include "dex/verified_method.h"
53#include "driver/compiler_driver-inl.h"
54#include "driver/compiler_options.h"
55#include "driver/dex_compilation_unit.h"
56#include "elf_writer_quick.h"
57#include "graph_checker.h"
58#include "graph_visualizer.h"
59#include "gvn.h"
60#include "induction_var_analysis.h"
61#include "inliner.h"
62#include "instruction_simplifier.h"
63#include "instruction_simplifier_arm.h"
64#include "intrinsics.h"
65#include "jit/debugger_interface.h"
66#include "jit/jit.h"
67#include "jit/jit_code_cache.h"
68#include "jni/quick/jni_compiler.h"
69#include "licm.h"
70#include "load_store_elimination.h"
71#include "nodes.h"
72#include "oat_quick_method_header.h"
73#include "prepare_for_register_allocation.h"
74#include "reference_type_propagation.h"
75#include "register_allocator.h"
76#include "select_generator.h"
77#include "sharpening.h"
78#include "side_effects_analysis.h"
79#include "ssa_builder.h"
80#include "ssa_liveness_analysis.h"
81#include "ssa_phi_elimination.h"
82#include "utils/assembler.h"
83#include "verifier/method_verifier.h"
84
85namespace art {
86
87/**
88 * Used by the code generator, to allocate the code in a vector.
89 */
90class CodeVectorAllocator FINAL : public CodeAllocator {
91 public:
92  explicit CodeVectorAllocator(ArenaAllocator* arena)
93      : memory_(arena->Adapter(kArenaAllocCodeBuffer)),
94        size_(0) {}
95
96  virtual uint8_t* Allocate(size_t size) {
97    size_ = size;
98    memory_.resize(size);
99    return &memory_[0];
100  }
101
102  size_t GetSize() const { return size_; }
103  const ArenaVector<uint8_t>& GetMemory() const { return memory_; }
104
105 private:
106  ArenaVector<uint8_t> memory_;
107  size_t size_;
108
109  DISALLOW_COPY_AND_ASSIGN(CodeVectorAllocator);
110};
111
112/**
113 * Filter to apply to the visualizer. Methods whose name contain that filter will
114 * be dumped.
115 */
116static constexpr const char kStringFilter[] = "";
117
118class PassScope;
119
120class PassObserver : public ValueObject {
121 public:
122  PassObserver(HGraph* graph,
123               CodeGenerator* codegen,
124               std::ostream* visualizer_output,
125               CompilerDriver* compiler_driver)
126      : graph_(graph),
127        cached_method_name_(),
128        timing_logger_enabled_(compiler_driver->GetDumpPasses()),
129        timing_logger_(timing_logger_enabled_ ? GetMethodName() : "", true, true),
130        disasm_info_(graph->GetArena()),
131        visualizer_enabled_(!compiler_driver->GetCompilerOptions().GetDumpCfgFileName().empty()),
132        visualizer_(visualizer_output, graph, *codegen),
133        graph_in_bad_state_(false) {
134    if (timing_logger_enabled_ || visualizer_enabled_) {
135      if (!IsVerboseMethod(compiler_driver, GetMethodName())) {
136        timing_logger_enabled_ = visualizer_enabled_ = false;
137      }
138      if (visualizer_enabled_) {
139        visualizer_.PrintHeader(GetMethodName());
140        codegen->SetDisassemblyInformation(&disasm_info_);
141      }
142    }
143  }
144
145  ~PassObserver() {
146    if (timing_logger_enabled_) {
147      LOG(INFO) << "TIMINGS " << GetMethodName();
148      LOG(INFO) << Dumpable<TimingLogger>(timing_logger_);
149    }
150  }
151
152  void DumpDisassembly() const {
153    if (visualizer_enabled_) {
154      visualizer_.DumpGraphWithDisassembly();
155    }
156  }
157
158  void SetGraphInBadState() { graph_in_bad_state_ = true; }
159
160  const char* GetMethodName() {
161    // PrettyMethod() is expensive, so we delay calling it until we actually have to.
162    if (cached_method_name_.empty()) {
163      cached_method_name_ = PrettyMethod(graph_->GetMethodIdx(), graph_->GetDexFile());
164    }
165    return cached_method_name_.c_str();
166  }
167
168 private:
169  void StartPass(const char* pass_name) {
170    // Dump graph first, then start timer.
171    if (visualizer_enabled_) {
172      visualizer_.DumpGraph(pass_name, /* is_after_pass */ false, graph_in_bad_state_);
173    }
174    if (timing_logger_enabled_) {
175      timing_logger_.StartTiming(pass_name);
176    }
177  }
178
179  void EndPass(const char* pass_name) {
180    // Pause timer first, then dump graph.
181    if (timing_logger_enabled_) {
182      timing_logger_.EndTiming();
183    }
184    if (visualizer_enabled_) {
185      visualizer_.DumpGraph(pass_name, /* is_after_pass */ true, graph_in_bad_state_);
186    }
187
188    // Validate the HGraph if running in debug mode.
189    if (kIsDebugBuild) {
190      if (!graph_in_bad_state_) {
191        GraphChecker checker(graph_);
192        checker.Run();
193        if (!checker.IsValid()) {
194          LOG(FATAL) << "Error after " << pass_name << ": " << Dumpable<GraphChecker>(checker);
195        }
196      }
197    }
198  }
199
200  static bool IsVerboseMethod(CompilerDriver* compiler_driver, const char* method_name) {
201    // Test an exact match to --verbose-methods. If verbose-methods is set, this overrides an
202    // empty kStringFilter matching all methods.
203    if (compiler_driver->GetCompilerOptions().HasVerboseMethods()) {
204      return compiler_driver->GetCompilerOptions().IsVerboseMethod(method_name);
205    }
206
207    // Test the kStringFilter sub-string. constexpr helper variable to silence unreachable-code
208    // warning when the string is empty.
209    constexpr bool kStringFilterEmpty = arraysize(kStringFilter) <= 1;
210    if (kStringFilterEmpty || strstr(method_name, kStringFilter) != nullptr) {
211      return true;
212    }
213
214    return false;
215  }
216
217  HGraph* const graph_;
218
219  std::string cached_method_name_;
220
221  bool timing_logger_enabled_;
222  TimingLogger timing_logger_;
223
224  DisassemblyInformation disasm_info_;
225
226  bool visualizer_enabled_;
227  HGraphVisualizer visualizer_;
228
229  // Flag to be set by the compiler if the pass failed and the graph is not
230  // expected to validate.
231  bool graph_in_bad_state_;
232
233  friend PassScope;
234
235  DISALLOW_COPY_AND_ASSIGN(PassObserver);
236};
237
238class PassScope : public ValueObject {
239 public:
240  PassScope(const char *pass_name, PassObserver* pass_observer)
241      : pass_name_(pass_name),
242        pass_observer_(pass_observer) {
243    pass_observer_->StartPass(pass_name_);
244  }
245
246  ~PassScope() {
247    pass_observer_->EndPass(pass_name_);
248  }
249
250 private:
251  const char* const pass_name_;
252  PassObserver* const pass_observer_;
253};
254
255class OptimizingCompiler FINAL : public Compiler {
256 public:
257  explicit OptimizingCompiler(CompilerDriver* driver);
258  ~OptimizingCompiler();
259
260  bool CanCompileMethod(uint32_t method_idx, const DexFile& dex_file, CompilationUnit* cu) const
261      OVERRIDE;
262
263  CompiledMethod* Compile(const DexFile::CodeItem* code_item,
264                          uint32_t access_flags,
265                          InvokeType invoke_type,
266                          uint16_t class_def_idx,
267                          uint32_t method_idx,
268                          jobject class_loader,
269                          const DexFile& dex_file,
270                          Handle<mirror::DexCache> dex_cache) const OVERRIDE;
271
272  CompiledMethod* JniCompile(uint32_t access_flags,
273                             uint32_t method_idx,
274                             const DexFile& dex_file) const OVERRIDE {
275    return ArtQuickJniCompileMethod(GetCompilerDriver(), access_flags, method_idx, dex_file);
276  }
277
278  uintptr_t GetEntryPointOf(ArtMethod* method) const OVERRIDE
279      SHARED_REQUIRES(Locks::mutator_lock_) {
280    return reinterpret_cast<uintptr_t>(method->GetEntryPointFromQuickCompiledCodePtrSize(
281        InstructionSetPointerSize(GetCompilerDriver()->GetInstructionSet())));
282  }
283
284  void InitCompilationUnit(CompilationUnit& cu) const OVERRIDE;
285
286  void Init() OVERRIDE;
287
288  void UnInit() const OVERRIDE;
289
290  void MaybeRecordStat(MethodCompilationStat compilation_stat) const {
291    if (compilation_stats_.get() != nullptr) {
292      compilation_stats_->RecordStat(compilation_stat);
293    }
294  }
295
296  bool JitCompile(Thread* self, jit::JitCodeCache* code_cache, ArtMethod* method, bool osr)
297      OVERRIDE
298      SHARED_REQUIRES(Locks::mutator_lock_);
299
300 private:
301  // Create a 'CompiledMethod' for an optimized graph.
302  CompiledMethod* Emit(ArenaAllocator* arena,
303                       CodeVectorAllocator* code_allocator,
304                       CodeGenerator* codegen,
305                       CompilerDriver* driver,
306                       const DexFile::CodeItem* item) const;
307
308  // Try compiling a method and return the code generator used for
309  // compiling it.
310  // This method:
311  // 1) Builds the graph. Returns null if it failed to build it.
312  // 2) Transforms the graph to SSA. Returns null if it failed.
313  // 3) Runs optimizations on the graph, including register allocator.
314  // 4) Generates code with the `code_allocator` provided.
315  CodeGenerator* TryCompile(ArenaAllocator* arena,
316                            CodeVectorAllocator* code_allocator,
317                            const DexFile::CodeItem* code_item,
318                            uint32_t access_flags,
319                            InvokeType invoke_type,
320                            uint16_t class_def_idx,
321                            uint32_t method_idx,
322                            jobject class_loader,
323                            const DexFile& dex_file,
324                            Handle<mirror::DexCache> dex_cache,
325                            bool osr) const;
326
327  std::unique_ptr<OptimizingCompilerStats> compilation_stats_;
328
329  std::unique_ptr<std::ostream> visualizer_output_;
330
331  DISALLOW_COPY_AND_ASSIGN(OptimizingCompiler);
332};
333
334static const int kMaximumCompilationTimeBeforeWarning = 100; /* ms */
335
336OptimizingCompiler::OptimizingCompiler(CompilerDriver* driver)
337    : Compiler(driver, kMaximumCompilationTimeBeforeWarning) {}
338
339void OptimizingCompiler::Init() {
340  // Enable C1visualizer output. Must be done in Init() because the compiler
341  // driver is not fully initialized when passed to the compiler's constructor.
342  CompilerDriver* driver = GetCompilerDriver();
343  const std::string cfg_file_name = driver->GetCompilerOptions().GetDumpCfgFileName();
344  if (!cfg_file_name.empty()) {
345    CHECK_EQ(driver->GetThreadCount(), 1U)
346      << "Graph visualizer requires the compiler to run single-threaded. "
347      << "Invoke the compiler with '-j1'.";
348    std::ios_base::openmode cfg_file_mode =
349        driver->GetCompilerOptions().GetDumpCfgAppend() ? std::ofstream::app : std::ofstream::out;
350    visualizer_output_.reset(new std::ofstream(cfg_file_name, cfg_file_mode));
351  }
352  if (driver->GetDumpStats()) {
353    compilation_stats_.reset(new OptimizingCompilerStats());
354  }
355}
356
357void OptimizingCompiler::UnInit() const {
358}
359
360OptimizingCompiler::~OptimizingCompiler() {
361  if (compilation_stats_.get() != nullptr) {
362    compilation_stats_->Log();
363  }
364}
365
366void OptimizingCompiler::InitCompilationUnit(CompilationUnit& cu ATTRIBUTE_UNUSED) const {
367}
368
369bool OptimizingCompiler::CanCompileMethod(uint32_t method_idx ATTRIBUTE_UNUSED,
370                                          const DexFile& dex_file ATTRIBUTE_UNUSED,
371                                          CompilationUnit* cu ATTRIBUTE_UNUSED) const {
372  return true;
373}
374
375static bool IsInstructionSetSupported(InstructionSet instruction_set) {
376  return (instruction_set == kArm && !kArm32QuickCodeUseSoftFloat)
377      || instruction_set == kArm64
378      || (instruction_set == kThumb2 && !kArm32QuickCodeUseSoftFloat)
379      || instruction_set == kMips
380      || instruction_set == kMips64
381      || instruction_set == kX86
382      || instruction_set == kX86_64;
383}
384
385// Read barrier are supported on ARM, ARM64, x86 and x86-64 at the moment.
386// TODO: Add support for other architectures and remove this function
387static bool InstructionSetSupportsReadBarrier(InstructionSet instruction_set) {
388  return instruction_set == kArm64
389      || instruction_set == kThumb2
390      || instruction_set == kX86
391      || instruction_set == kX86_64;
392}
393
394static void RunOptimizations(HOptimization* optimizations[],
395                             size_t length,
396                             PassObserver* pass_observer) {
397  for (size_t i = 0; i < length; ++i) {
398    PassScope scope(optimizations[i]->GetPassName(), pass_observer);
399    optimizations[i]->Run();
400  }
401}
402
403static void MaybeRunInliner(HGraph* graph,
404                            CodeGenerator* codegen,
405                            CompilerDriver* driver,
406                            OptimizingCompilerStats* stats,
407                            const DexCompilationUnit& dex_compilation_unit,
408                            PassObserver* pass_observer,
409                            StackHandleScopeCollection* handles) {
410  const CompilerOptions& compiler_options = driver->GetCompilerOptions();
411  bool should_inline = (compiler_options.GetInlineDepthLimit() > 0)
412      && (compiler_options.GetInlineMaxCodeUnits() > 0);
413  if (!should_inline) {
414    return;
415  }
416  size_t number_of_dex_registers = dex_compilation_unit.GetCodeItem()->registers_size_;
417  HInliner* inliner = new (graph->GetArena()) HInliner(
418      graph,
419      graph,
420      codegen,
421      dex_compilation_unit,
422      dex_compilation_unit,
423      driver,
424      handles,
425      stats,
426      number_of_dex_registers,
427      /* depth */ 0);
428  HOptimization* optimizations[] = { inliner };
429
430  RunOptimizations(optimizations, arraysize(optimizations), pass_observer);
431}
432
433static void RunArchOptimizations(InstructionSet instruction_set,
434                                 HGraph* graph,
435                                 CodeGenerator* codegen,
436                                 OptimizingCompilerStats* stats,
437                                 PassObserver* pass_observer) {
438  ArenaAllocator* arena = graph->GetArena();
439  switch (instruction_set) {
440#ifdef ART_ENABLE_CODEGEN_arm
441    case kThumb2:
442    case kArm: {
443      arm::DexCacheArrayFixups* fixups = new (arena) arm::DexCacheArrayFixups(graph, stats);
444      arm::InstructionSimplifierArm* simplifier =
445          new (arena) arm::InstructionSimplifierArm(graph, stats);
446      HOptimization* arm_optimizations[] = {
447        simplifier,
448        fixups
449      };
450      RunOptimizations(arm_optimizations, arraysize(arm_optimizations), pass_observer);
451      break;
452    }
453#endif
454#ifdef ART_ENABLE_CODEGEN_arm64
455    case kArm64: {
456      arm64::InstructionSimplifierArm64* simplifier =
457          new (arena) arm64::InstructionSimplifierArm64(graph, stats);
458      SideEffectsAnalysis* side_effects = new (arena) SideEffectsAnalysis(graph);
459      GVNOptimization* gvn = new (arena) GVNOptimization(graph, *side_effects, "GVN_after_arch");
460      HOptimization* arm64_optimizations[] = {
461        simplifier,
462        side_effects,
463        gvn
464      };
465      RunOptimizations(arm64_optimizations, arraysize(arm64_optimizations), pass_observer);
466      break;
467    }
468#endif
469#ifdef ART_ENABLE_CODEGEN_x86
470    case kX86: {
471      x86::PcRelativeFixups* pc_relative_fixups =
472          new (arena) x86::PcRelativeFixups(graph, codegen, stats);
473      HOptimization* x86_optimizations[] = {
474          pc_relative_fixups
475      };
476      RunOptimizations(x86_optimizations, arraysize(x86_optimizations), pass_observer);
477      break;
478    }
479#endif
480    default:
481      break;
482  }
483}
484
485NO_INLINE  // Avoid increasing caller's frame size by large stack-allocated objects.
486static void AllocateRegisters(HGraph* graph,
487                              CodeGenerator* codegen,
488                              PassObserver* pass_observer) {
489  {
490    PassScope scope(PrepareForRegisterAllocation::kPrepareForRegisterAllocationPassName,
491                    pass_observer);
492    PrepareForRegisterAllocation(graph).Run();
493  }
494  SsaLivenessAnalysis liveness(graph, codegen);
495  {
496    PassScope scope(SsaLivenessAnalysis::kLivenessPassName, pass_observer);
497    liveness.Analyze();
498  }
499  {
500    PassScope scope(RegisterAllocator::kRegisterAllocatorPassName, pass_observer);
501    RegisterAllocator(graph->GetArena(), codegen, liveness).AllocateRegisters();
502  }
503}
504
505static void RunOptimizations(HGraph* graph,
506                             CodeGenerator* codegen,
507                             CompilerDriver* driver,
508                             OptimizingCompilerStats* stats,
509                             const DexCompilationUnit& dex_compilation_unit,
510                             PassObserver* pass_observer,
511                             StackHandleScopeCollection* handles) {
512  ArenaAllocator* arena = graph->GetArena();
513  HDeadCodeElimination* dce1 = new (arena) HDeadCodeElimination(
514      graph, stats, HDeadCodeElimination::kInitialDeadCodeEliminationPassName);
515  HDeadCodeElimination* dce2 = new (arena) HDeadCodeElimination(
516      graph, stats, HDeadCodeElimination::kFinalDeadCodeEliminationPassName);
517  HConstantFolding* fold1 = new (arena) HConstantFolding(graph);
518  InstructionSimplifier* simplify1 = new (arena) InstructionSimplifier(graph, stats);
519  HSelectGenerator* select_generator = new (arena) HSelectGenerator(graph, stats);
520  HConstantFolding* fold2 = new (arena) HConstantFolding(graph, "constant_folding_after_inlining");
521  HConstantFolding* fold3 = new (arena) HConstantFolding(graph, "constant_folding_after_bce");
522  SideEffectsAnalysis* side_effects = new (arena) SideEffectsAnalysis(graph);
523  GVNOptimization* gvn = new (arena) GVNOptimization(graph, *side_effects);
524  LICM* licm = new (arena) LICM(graph, *side_effects, stats);
525  LoadStoreElimination* lse = new (arena) LoadStoreElimination(graph, *side_effects);
526  HInductionVarAnalysis* induction = new (arena) HInductionVarAnalysis(graph);
527  BoundsCheckElimination* bce = new (arena) BoundsCheckElimination(graph, *side_effects, induction);
528  HSharpening* sharpening = new (arena) HSharpening(graph, codegen, dex_compilation_unit, driver);
529  InstructionSimplifier* simplify2 = new (arena) InstructionSimplifier(
530      graph, stats, "instruction_simplifier_after_bce");
531  InstructionSimplifier* simplify3 = new (arena) InstructionSimplifier(
532      graph, stats, "instruction_simplifier_before_codegen");
533  IntrinsicsRecognizer* intrinsics = new (arena) IntrinsicsRecognizer(graph, driver, stats);
534
535  HOptimization* optimizations1[] = {
536    intrinsics,
537    sharpening,
538    fold1,
539    simplify1,
540    dce1,
541  };
542  RunOptimizations(optimizations1, arraysize(optimizations1), pass_observer);
543
544  MaybeRunInliner(graph, codegen, driver, stats, dex_compilation_unit, pass_observer, handles);
545
546  HOptimization* optimizations2[] = {
547    // SelectGenerator depends on the InstructionSimplifier removing
548    // redundant suspend checks to recognize empty blocks.
549    select_generator,
550    fold2,  // TODO: if we don't inline we can also skip fold2.
551    side_effects,
552    gvn,
553    licm,
554    induction,
555    bce,
556    fold3,  // evaluates code generated by dynamic bce
557    simplify2,
558    lse,
559    dce2,
560    // The codegen has a few assumptions that only the instruction simplifier
561    // can satisfy. For example, the code generator does not expect to see a
562    // HTypeConversion from a type to the same type.
563    simplify3,
564  };
565  RunOptimizations(optimizations2, arraysize(optimizations2), pass_observer);
566
567  RunArchOptimizations(driver->GetInstructionSet(), graph, codegen, stats, pass_observer);
568  AllocateRegisters(graph, codegen, pass_observer);
569}
570
571static ArenaVector<LinkerPatch> EmitAndSortLinkerPatches(CodeGenerator* codegen) {
572  ArenaVector<LinkerPatch> linker_patches(codegen->GetGraph()->GetArena()->Adapter());
573  codegen->EmitLinkerPatches(&linker_patches);
574
575  // Sort patches by literal offset. Required for .oat_patches encoding.
576  std::sort(linker_patches.begin(), linker_patches.end(),
577            [](const LinkerPatch& lhs, const LinkerPatch& rhs) {
578    return lhs.LiteralOffset() < rhs.LiteralOffset();
579  });
580
581  return linker_patches;
582}
583
584CompiledMethod* OptimizingCompiler::Emit(ArenaAllocator* arena,
585                                         CodeVectorAllocator* code_allocator,
586                                         CodeGenerator* codegen,
587                                         CompilerDriver* compiler_driver,
588                                         const DexFile::CodeItem* code_item) const {
589  ArenaVector<LinkerPatch> linker_patches = EmitAndSortLinkerPatches(codegen);
590  ArenaVector<uint8_t> stack_map(arena->Adapter(kArenaAllocStackMaps));
591  stack_map.resize(codegen->ComputeStackMapsSize());
592  codegen->BuildStackMaps(MemoryRegion(stack_map.data(), stack_map.size()), *code_item);
593
594  CompiledMethod* compiled_method = CompiledMethod::SwapAllocCompiledMethod(
595      compiler_driver,
596      codegen->GetInstructionSet(),
597      ArrayRef<const uint8_t>(code_allocator->GetMemory()),
598      // Follow Quick's behavior and set the frame size to zero if it is
599      // considered "empty" (see the definition of
600      // art::CodeGenerator::HasEmptyFrame).
601      codegen->HasEmptyFrame() ? 0 : codegen->GetFrameSize(),
602      codegen->GetCoreSpillMask(),
603      codegen->GetFpuSpillMask(),
604      ArrayRef<const SrcMapElem>(),
605      ArrayRef<const uint8_t>(),  // mapping_table.
606      ArrayRef<const uint8_t>(stack_map),
607      ArrayRef<const uint8_t>(),  // native_gc_map.
608      ArrayRef<const uint8_t>(*codegen->GetAssembler()->cfi().data()),
609      ArrayRef<const LinkerPatch>(linker_patches));
610
611  return compiled_method;
612}
613
614CodeGenerator* OptimizingCompiler::TryCompile(ArenaAllocator* arena,
615                                              CodeVectorAllocator* code_allocator,
616                                              const DexFile::CodeItem* code_item,
617                                              uint32_t access_flags,
618                                              InvokeType invoke_type,
619                                              uint16_t class_def_idx,
620                                              uint32_t method_idx,
621                                              jobject class_loader,
622                                              const DexFile& dex_file,
623                                              Handle<mirror::DexCache> dex_cache,
624                                              bool osr) const {
625  MaybeRecordStat(MethodCompilationStat::kAttemptCompilation);
626  CompilerDriver* compiler_driver = GetCompilerDriver();
627  InstructionSet instruction_set = compiler_driver->GetInstructionSet();
628
629  // Always use the Thumb-2 assembler: some runtime functionality
630  // (like implicit stack overflow checks) assume Thumb-2.
631  if (instruction_set == kArm) {
632    instruction_set = kThumb2;
633  }
634
635  // Do not attempt to compile on architectures we do not support.
636  if (!IsInstructionSetSupported(instruction_set)) {
637    MaybeRecordStat(MethodCompilationStat::kNotCompiledUnsupportedIsa);
638    return nullptr;
639  }
640
641  // When read barriers are enabled, do not attempt to compile for
642  // instruction sets that have no read barrier support.
643  if (kEmitCompilerReadBarrier && !InstructionSetSupportsReadBarrier(instruction_set)) {
644    return nullptr;
645  }
646
647  if (Compiler::IsPathologicalCase(*code_item, method_idx, dex_file)) {
648    MaybeRecordStat(MethodCompilationStat::kNotCompiledPathological);
649    return nullptr;
650  }
651
652  // Implementation of the space filter: do not compile a code item whose size in
653  // code units is bigger than 128.
654  static constexpr size_t kSpaceFilterOptimizingThreshold = 128;
655  const CompilerOptions& compiler_options = compiler_driver->GetCompilerOptions();
656  if ((compiler_options.GetCompilerFilter() == CompilerOptions::kSpace)
657      && (code_item->insns_size_in_code_units_ > kSpaceFilterOptimizingThreshold)) {
658    MaybeRecordStat(MethodCompilationStat::kNotCompiledSpaceFilter);
659    return nullptr;
660  }
661
662  DexCompilationUnit dex_compilation_unit(
663    nullptr, class_loader, Runtime::Current()->GetClassLinker(), dex_file, code_item,
664    class_def_idx, method_idx, access_flags,
665    nullptr, dex_cache);
666
667  bool requires_barrier = dex_compilation_unit.IsConstructor()
668      && compiler_driver->RequiresConstructorBarrier(Thread::Current(),
669                                                     dex_compilation_unit.GetDexFile(),
670                                                     dex_compilation_unit.GetClassDefIndex());
671
672  HGraph* graph = new (arena) HGraph(
673      arena,
674      dex_file,
675      method_idx,
676      requires_barrier,
677      compiler_driver->GetInstructionSet(),
678      kInvalidInvokeType,
679      compiler_driver->GetCompilerOptions().GetDebuggable(),
680      osr);
681
682  const uint8_t* interpreter_metadata = nullptr;
683  {
684    ScopedObjectAccess soa(Thread::Current());
685    StackHandleScope<1> hs(soa.Self());
686    Handle<mirror::ClassLoader> loader(hs.NewHandle(
687        soa.Decode<mirror::ClassLoader*>(class_loader)));
688    ArtMethod* art_method = compiler_driver->ResolveMethod(
689        soa, dex_cache, loader, &dex_compilation_unit, method_idx, invoke_type);
690    // We may not get a method, for example if its class is erroneous.
691    if (art_method != nullptr) {
692      graph->SetArtMethod(art_method);
693      interpreter_metadata = art_method->GetQuickenedInfo();
694    }
695  }
696
697  std::unique_ptr<CodeGenerator> codegen(
698      CodeGenerator::Create(graph,
699                            instruction_set,
700                            *compiler_driver->GetInstructionSetFeatures(),
701                            compiler_driver->GetCompilerOptions()));
702  if (codegen.get() == nullptr) {
703    MaybeRecordStat(MethodCompilationStat::kNotCompiledNoCodegen);
704    return nullptr;
705  }
706  codegen->GetAssembler()->cfi().SetEnabled(
707      compiler_driver->GetCompilerOptions().GenerateAnyDebugInfo());
708
709  PassObserver pass_observer(graph,
710                             codegen.get(),
711                             visualizer_output_.get(),
712                             compiler_driver);
713
714  VLOG(compiler) << "Building " << pass_observer.GetMethodName();
715
716  {
717    ScopedObjectAccess soa(Thread::Current());
718    StackHandleScopeCollection handles(soa.Self());
719    // Do not hold `mutator_lock_` between optimizations.
720    ScopedThreadSuspension sts(soa.Self(), kNative);
721
722    {
723      PassScope scope(HGraphBuilder::kBuilderPassName, &pass_observer);
724      HGraphBuilder builder(graph,
725                            &dex_compilation_unit,
726                            &dex_compilation_unit,
727                            &dex_file,
728                            compiler_driver,
729                            compilation_stats_.get(),
730                            interpreter_metadata,
731                            dex_cache);
732      GraphAnalysisResult result = builder.BuildGraph(*code_item, &handles);
733      if (result != kAnalysisSuccess) {
734        switch (result) {
735          case kAnalysisInvalidBytecode:
736            break;
737          case kAnalysisFailThrowCatchLoop:
738            MaybeRecordStat(MethodCompilationStat::kNotCompiledThrowCatchLoop);
739            break;
740          case kAnalysisFailAmbiguousArrayOp:
741            MaybeRecordStat(MethodCompilationStat::kNotCompiledAmbiguousArrayOp);
742            break;
743          case kAnalysisSuccess:
744            UNREACHABLE();
745        }
746        pass_observer.SetGraphInBadState();
747        return nullptr;
748      }
749    }
750
751    RunOptimizations(graph,
752                     codegen.get(),
753                     compiler_driver,
754                     compilation_stats_.get(),
755                     dex_compilation_unit,
756                     &pass_observer,
757                     &handles);
758
759    codegen->Compile(code_allocator);
760    pass_observer.DumpDisassembly();
761  }
762
763  if (kArenaAllocatorCountAllocations) {
764    if (arena->BytesAllocated() > 4 * MB) {
765      MemStats mem_stats(arena->GetMemStats());
766      LOG(INFO) << PrettyMethod(method_idx, dex_file) << " " << Dumpable<MemStats>(mem_stats);
767    }
768  }
769
770  return codegen.release();
771}
772
773static bool CanHandleVerificationFailure(const VerifiedMethod* verified_method) {
774  // For access errors the compiler will use the unresolved helpers (e.g. HInvokeUnresolved).
775  uint32_t unresolved_mask = verifier::VerifyError::VERIFY_ERROR_NO_CLASS
776      | verifier::VerifyError::VERIFY_ERROR_ACCESS_CLASS
777      | verifier::VerifyError::VERIFY_ERROR_ACCESS_FIELD
778      | verifier::VerifyError::VERIFY_ERROR_ACCESS_METHOD;
779  return (verified_method->GetEncounteredVerificationFailures() & (~unresolved_mask)) == 0;
780}
781
782CompiledMethod* OptimizingCompiler::Compile(const DexFile::CodeItem* code_item,
783                                            uint32_t access_flags,
784                                            InvokeType invoke_type,
785                                            uint16_t class_def_idx,
786                                            uint32_t method_idx,
787                                            jobject jclass_loader,
788                                            const DexFile& dex_file,
789                                            Handle<mirror::DexCache> dex_cache) const {
790  CompilerDriver* compiler_driver = GetCompilerDriver();
791  CompiledMethod* method = nullptr;
792  DCHECK(Runtime::Current()->IsAotCompiler());
793  const VerifiedMethod* verified_method = compiler_driver->GetVerifiedMethod(&dex_file, method_idx);
794  DCHECK(!verified_method->HasRuntimeThrow());
795  if (compiler_driver->IsMethodVerifiedWithoutFailures(method_idx, class_def_idx, dex_file)
796      || CanHandleVerificationFailure(verified_method)) {
797    ArenaAllocator arena(Runtime::Current()->GetArenaPool());
798    CodeVectorAllocator code_allocator(&arena);
799    std::unique_ptr<CodeGenerator> codegen(
800        TryCompile(&arena,
801                   &code_allocator,
802                   code_item,
803                   access_flags,
804                   invoke_type,
805                   class_def_idx,
806                   method_idx,
807                   jclass_loader,
808                   dex_file,
809                   dex_cache,
810                   /* osr */ false));
811    if (codegen.get() != nullptr) {
812      MaybeRecordStat(MethodCompilationStat::kCompiled);
813      method = Emit(&arena, &code_allocator, codegen.get(), compiler_driver, code_item);
814    }
815  } else {
816    if (compiler_driver->GetCompilerOptions().VerifyAtRuntime()) {
817      MaybeRecordStat(MethodCompilationStat::kNotCompiledVerifyAtRuntime);
818    } else {
819      MaybeRecordStat(MethodCompilationStat::kNotCompiledVerificationError);
820    }
821  }
822
823  if (kIsDebugBuild &&
824      IsCompilingWithCoreImage() &&
825      IsInstructionSetSupported(compiler_driver->GetInstructionSet()) &&
826      (!kEmitCompilerReadBarrier ||
827       InstructionSetSupportsReadBarrier(compiler_driver->GetInstructionSet()))) {
828    // For testing purposes, we put a special marker on method names
829    // that should be compiled with this compiler (when the the
830    // instruction set is supported -- and has support for read
831    // barriers, if they are enabled). This makes sure we're not
832    // regressing.
833    std::string method_name = PrettyMethod(method_idx, dex_file);
834    bool shouldCompile = method_name.find("$opt$") != std::string::npos;
835    DCHECK((method != nullptr) || !shouldCompile) << "Didn't compile " << method_name;
836  }
837
838  return method;
839}
840
841Compiler* CreateOptimizingCompiler(CompilerDriver* driver) {
842  return new OptimizingCompiler(driver);
843}
844
845bool IsCompilingWithCoreImage() {
846  const std::string& image = Runtime::Current()->GetImageLocation();
847  // TODO: This is under-approximating...
848  if (EndsWith(image, "core.art") || EndsWith(image, "core-optimizing.art")) {
849    return true;
850  }
851  return false;
852}
853
854bool OptimizingCompiler::JitCompile(Thread* self,
855                                    jit::JitCodeCache* code_cache,
856                                    ArtMethod* method,
857                                    bool osr) {
858  StackHandleScope<2> hs(self);
859  Handle<mirror::ClassLoader> class_loader(hs.NewHandle(
860      method->GetDeclaringClass()->GetClassLoader()));
861  Handle<mirror::DexCache> dex_cache(hs.NewHandle(method->GetDexCache()));
862
863  jobject jclass_loader = class_loader.ToJObject();
864  const DexFile* dex_file = method->GetDexFile();
865  const uint16_t class_def_idx = method->GetClassDefIndex();
866  const DexFile::CodeItem* code_item = dex_file->GetCodeItem(method->GetCodeItemOffset());
867  const uint32_t method_idx = method->GetDexMethodIndex();
868  const uint32_t access_flags = method->GetAccessFlags();
869  const InvokeType invoke_type = method->GetInvokeType();
870
871  ArenaAllocator arena(Runtime::Current()->GetJitArenaPool());
872  CodeVectorAllocator code_allocator(&arena);
873  std::unique_ptr<CodeGenerator> codegen;
874  {
875    // Go to native so that we don't block GC during compilation.
876    ScopedThreadSuspension sts(self, kNative);
877    codegen.reset(
878        TryCompile(&arena,
879                   &code_allocator,
880                   code_item,
881                   access_flags,
882                   invoke_type,
883                   class_def_idx,
884                   method_idx,
885                   jclass_loader,
886                   *dex_file,
887                   dex_cache,
888                   osr));
889    if (codegen.get() == nullptr) {
890      return false;
891    }
892  }
893
894  size_t stack_map_size = codegen->ComputeStackMapsSize();
895  uint8_t* stack_map_data = code_cache->ReserveData(self, stack_map_size);
896  if (stack_map_data == nullptr) {
897    return false;
898  }
899  MaybeRecordStat(MethodCompilationStat::kCompiled);
900  codegen->BuildStackMaps(MemoryRegion(stack_map_data, stack_map_size), *code_item);
901  const void* code = code_cache->CommitCode(
902      self,
903      method,
904      nullptr,
905      stack_map_data,
906      nullptr,
907      codegen->HasEmptyFrame() ? 0 : codegen->GetFrameSize(),
908      codegen->GetCoreSpillMask(),
909      codegen->GetFpuSpillMask(),
910      code_allocator.GetMemory().data(),
911      code_allocator.GetSize(),
912      osr);
913
914  if (code == nullptr) {
915    code_cache->ClearData(self, stack_map_data);
916    return false;
917  }
918
919  const CompilerOptions& compiler_options = GetCompilerDriver()->GetCompilerOptions();
920  if (compiler_options.GetGenerateDebugInfo()) {
921    const auto* method_header = reinterpret_cast<const OatQuickMethodHeader*>(code);
922    const uintptr_t code_address = reinterpret_cast<uintptr_t>(method_header->GetCode());
923    debug::MethodDebugInfo info = debug::MethodDebugInfo();
924    info.trampoline_name = nullptr;
925    info.dex_file = dex_file;
926    info.class_def_index = class_def_idx;
927    info.dex_method_index = method_idx;
928    info.access_flags = access_flags;
929    info.code_item = code_item;
930    info.isa = codegen->GetInstructionSet();
931    info.deduped = false;
932    info.is_native_debuggable = compiler_options.GetNativeDebuggable();
933    info.is_optimized = true;
934    info.is_code_address_text_relative = false;
935    info.code_address = code_address;
936    info.code_size = code_allocator.GetSize();
937    info.frame_size_in_bytes = method_header->GetFrameSizeInBytes();
938    info.code_info = stack_map_size == 0 ? nullptr : stack_map_data;
939    info.cfi = ArrayRef<const uint8_t>(*codegen->GetAssembler()->cfi().data());
940    ArrayRef<const uint8_t> elf_file = debug::WriteDebugElfFileForMethods(
941        GetCompilerDriver()->GetInstructionSet(),
942        GetCompilerDriver()->GetInstructionSetFeatures(),
943        ArrayRef<const debug::MethodDebugInfo>(&info, 1));
944    CreateJITCodeEntryForAddress(code_address,
945                                 std::unique_ptr<const uint8_t[]>(elf_file.data()),
946                                 elf_file.size());
947  }
948
949  Runtime::Current()->GetJit()->AddMemoryUsage(method, arena.BytesUsed());
950
951  return true;
952}
953
954}  // namespace art
955