/* * Copyright (C) 2014 The Android Open Source Project * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ #include "optimizing_compiler.h" #include #include #include #include #include "art_method-inl.h" #include "base/arena_allocator.h" #include "base/arena_containers.h" #include "base/dumpable.h" #include "base/logging.h" #include "base/macros.h" #include "base/mutex.h" #include "base/scoped_arena_allocator.h" #include "base/timing_logger.h" #include "builder.h" #include "code_generator.h" #include "compiled_method.h" #include "compiler.h" #include "debug/elf_debug_writer.h" #include "debug/method_debug_info.h" #include "dex/dex_file_types.h" #include "dex/verification_results.h" #include "dex/verified_method.h" #include "driver/compiled_method_storage.h" #include "driver/compiler_options.h" #include "driver/dex_compilation_unit.h" #include "graph_checker.h" #include "graph_visualizer.h" #include "inliner.h" #include "jit/debugger_interface.h" #include "jit/jit.h" #include "jit/jit_code_cache.h" #include "jit/jit_logger.h" #include "jni/quick/jni_compiler.h" #include "linker/linker_patch.h" #include "nodes.h" #include "oat_quick_method_header.h" #include "prepare_for_register_allocation.h" #include "reference_type_propagation.h" #include "register_allocator_linear_scan.h" #include "select_generator.h" #include "ssa_builder.h" #include "ssa_liveness_analysis.h" #include "ssa_phi_elimination.h" #include "stack_map_stream.h" #include "utils/assembler.h" #include "verifier/verifier_compiler_binding.h" namespace art { static constexpr size_t kArenaAllocatorMemoryReportThreshold = 8 * MB; static constexpr const char* kPassNameSeparator = "$"; /** * Used by the code generator, to allocate the code in a vector. */ class CodeVectorAllocator final : public CodeAllocator { public: explicit CodeVectorAllocator(ArenaAllocator* allocator) : memory_(allocator->Adapter(kArenaAllocCodeBuffer)) {} uint8_t* Allocate(size_t size) override { memory_.resize(size); return &memory_[0]; } ArrayRef GetMemory() const override { return ArrayRef(memory_); } uint8_t* GetData() { return memory_.data(); } private: ArenaVector memory_; DISALLOW_COPY_AND_ASSIGN(CodeVectorAllocator); }; /** * Filter to apply to the visualizer. Methods whose name contain that filter will * be dumped. */ static constexpr const char kStringFilter[] = ""; class PassScope; class PassObserver : public ValueObject { public: PassObserver(HGraph* graph, CodeGenerator* codegen, std::ostream* visualizer_output, const CompilerOptions& compiler_options, Mutex& dump_mutex) : graph_(graph), last_seen_graph_size_(0), cached_method_name_(), timing_logger_enabled_(compiler_options.GetDumpPassTimings()), timing_logger_(timing_logger_enabled_ ? GetMethodName() : "", true, true), disasm_info_(graph->GetAllocator()), visualizer_oss_(), visualizer_output_(visualizer_output), visualizer_enabled_(!compiler_options.GetDumpCfgFileName().empty()), visualizer_(&visualizer_oss_, graph, *codegen), codegen_(codegen), visualizer_dump_mutex_(dump_mutex), graph_in_bad_state_(false) { if (timing_logger_enabled_ || visualizer_enabled_) { if (!IsVerboseMethod(compiler_options, GetMethodName())) { timing_logger_enabled_ = visualizer_enabled_ = false; } if (visualizer_enabled_) { visualizer_.PrintHeader(GetMethodName()); codegen->SetDisassemblyInformation(&disasm_info_); } } } ~PassObserver() { if (timing_logger_enabled_) { LOG(INFO) << "TIMINGS " << GetMethodName(); LOG(INFO) << Dumpable(timing_logger_); } if (visualizer_enabled_) { FlushVisualizer(); } DCHECK(visualizer_oss_.str().empty()); } void DumpDisassembly() { if (visualizer_enabled_) { visualizer_.DumpGraphWithDisassembly(); } } void SetGraphInBadState() { graph_in_bad_state_ = true; } const char* GetMethodName() { // PrettyMethod() is expensive, so we delay calling it until we actually have to. if (cached_method_name_.empty()) { cached_method_name_ = graph_->GetDexFile().PrettyMethod(graph_->GetMethodIdx()); } return cached_method_name_.c_str(); } private: void StartPass(const char* pass_name) { VLOG(compiler) << "Starting pass: " << pass_name; // Dump graph first, then start timer. if (visualizer_enabled_) { visualizer_.DumpGraph(pass_name, /* is_after_pass= */ false, graph_in_bad_state_); } if (timing_logger_enabled_) { timing_logger_.StartTiming(pass_name); } } void FlushVisualizer() REQUIRES(!visualizer_dump_mutex_) { MutexLock mu(Thread::Current(), visualizer_dump_mutex_); *visualizer_output_ << visualizer_oss_.str(); visualizer_output_->flush(); visualizer_oss_.str(""); visualizer_oss_.clear(); } void EndPass(const char* pass_name, bool pass_change) { // Pause timer first, then dump graph. if (timing_logger_enabled_) { timing_logger_.EndTiming(); } if (visualizer_enabled_) { visualizer_.DumpGraph(pass_name, /* is_after_pass= */ true, graph_in_bad_state_); } // Validate the HGraph if running in debug mode. if (kIsDebugBuild) { if (!graph_in_bad_state_) { GraphChecker checker(graph_, codegen_); last_seen_graph_size_ = checker.Run(pass_change, last_seen_graph_size_); if (!checker.IsValid()) { LOG(FATAL) << "Error after " << pass_name << ": " << Dumpable(checker); } } } } static bool IsVerboseMethod(const CompilerOptions& compiler_options, const char* method_name) { // Test an exact match to --verbose-methods. If verbose-methods is set, this overrides an // empty kStringFilter matching all methods. if (compiler_options.HasVerboseMethods()) { return compiler_options.IsVerboseMethod(method_name); } // Test the kStringFilter sub-string. constexpr helper variable to silence unreachable-code // warning when the string is empty. constexpr bool kStringFilterEmpty = arraysize(kStringFilter) <= 1; if (kStringFilterEmpty || strstr(method_name, kStringFilter) != nullptr) { return true; } return false; } HGraph* const graph_; size_t last_seen_graph_size_; std::string cached_method_name_; bool timing_logger_enabled_; TimingLogger timing_logger_; DisassemblyInformation disasm_info_; std::ostringstream visualizer_oss_; std::ostream* visualizer_output_; bool visualizer_enabled_; HGraphVisualizer visualizer_; CodeGenerator* codegen_; Mutex& visualizer_dump_mutex_; // Flag to be set by the compiler if the pass failed and the graph is not // expected to validate. bool graph_in_bad_state_; friend PassScope; DISALLOW_COPY_AND_ASSIGN(PassObserver); }; class PassScope : public ValueObject { public: PassScope(const char *pass_name, PassObserver* pass_observer) : pass_name_(pass_name), pass_change_(true), // assume change pass_observer_(pass_observer) { pass_observer_->StartPass(pass_name_); } void SetPassNotChanged() { pass_change_ = false; } ~PassScope() { pass_observer_->EndPass(pass_name_, pass_change_); } private: const char* const pass_name_; bool pass_change_; PassObserver* const pass_observer_; }; class OptimizingCompiler final : public Compiler { public: explicit OptimizingCompiler(const CompilerOptions& compiler_options, CompiledMethodStorage* storage); ~OptimizingCompiler() override; bool CanCompileMethod(uint32_t method_idx, const DexFile& dex_file) const override; CompiledMethod* Compile(const dex::CodeItem* code_item, uint32_t access_flags, InvokeType invoke_type, uint16_t class_def_idx, uint32_t method_idx, Handle class_loader, const DexFile& dex_file, Handle dex_cache) const override; CompiledMethod* JniCompile(uint32_t access_flags, uint32_t method_idx, const DexFile& dex_file, Handle dex_cache) const override; uintptr_t GetEntryPointOf(ArtMethod* method) const override REQUIRES_SHARED(Locks::mutator_lock_) { return reinterpret_cast(method->GetEntryPointFromQuickCompiledCodePtrSize( InstructionSetPointerSize(GetCompilerOptions().GetInstructionSet()))); } bool JitCompile(Thread* self, jit::JitCodeCache* code_cache, jit::JitMemoryRegion* region, ArtMethod* method, CompilationKind compilation_kind, jit::JitLogger* jit_logger) override REQUIRES_SHARED(Locks::mutator_lock_); private: bool RunOptimizations(HGraph* graph, CodeGenerator* codegen, const DexCompilationUnit& dex_compilation_unit, PassObserver* pass_observer, const OptimizationDef definitions[], size_t length) const { // Convert definitions to optimization passes. ArenaVector optimizations = ConstructOptimizations( definitions, length, graph->GetAllocator(), graph, compilation_stats_.get(), codegen, dex_compilation_unit); DCHECK_EQ(length, optimizations.size()); // Run the optimization passes one by one. Any "depends_on" pass refers back to // the most recent occurrence of that pass, skipped or executed. std::bitset(OptimizationPass::kLast) + 1u> pass_changes; pass_changes[static_cast(OptimizationPass::kNone)] = true; bool change = false; for (size_t i = 0; i < length; ++i) { if (pass_changes[static_cast(definitions[i].depends_on)]) { // Execute the pass and record whether it changed anything. PassScope scope(optimizations[i]->GetPassName(), pass_observer); bool pass_change = optimizations[i]->Run(); pass_changes[static_cast(definitions[i].pass)] = pass_change; if (pass_change) { change = true; } else { scope.SetPassNotChanged(); } } else { // Skip the pass and record that nothing changed. pass_changes[static_cast(definitions[i].pass)] = false; } } return change; } template bool RunOptimizations( HGraph* graph, CodeGenerator* codegen, const DexCompilationUnit& dex_compilation_unit, PassObserver* pass_observer, const OptimizationDef (&definitions)[length]) const { return RunOptimizations( graph, codegen, dex_compilation_unit, pass_observer, definitions, length); } void RunOptimizations(HGraph* graph, CodeGenerator* codegen, const DexCompilationUnit& dex_compilation_unit, PassObserver* pass_observer) const; private: // Create a 'CompiledMethod' for an optimized graph. CompiledMethod* Emit(ArenaAllocator* allocator, CodeVectorAllocator* code_allocator, CodeGenerator* codegen, const dex::CodeItem* item) const; // Try compiling a method and return the code generator used for // compiling it. // This method: // 1) Builds the graph. Returns null if it failed to build it. // 2) Transforms the graph to SSA. Returns null if it failed. // 3) Runs optimizations on the graph, including register allocator. // 4) Generates code with the `code_allocator` provided. CodeGenerator* TryCompile(ArenaAllocator* allocator, ArenaStack* arena_stack, CodeVectorAllocator* code_allocator, const DexCompilationUnit& dex_compilation_unit, ArtMethod* method, CompilationKind compilation_kind, VariableSizedHandleScope* handles) const; CodeGenerator* TryCompileIntrinsic(ArenaAllocator* allocator, ArenaStack* arena_stack, CodeVectorAllocator* code_allocator, const DexCompilationUnit& dex_compilation_unit, ArtMethod* method, VariableSizedHandleScope* handles) const; bool RunArchOptimizations(HGraph* graph, CodeGenerator* codegen, const DexCompilationUnit& dex_compilation_unit, PassObserver* pass_observer) const; bool RunBaselineOptimizations(HGraph* graph, CodeGenerator* codegen, const DexCompilationUnit& dex_compilation_unit, PassObserver* pass_observer) const; std::vector GenerateJitDebugInfo(const debug::MethodDebugInfo& method_debug_info); // This must be called before any other function that dumps data to the cfg void DumpInstructionSetFeaturesToCfg() const; std::unique_ptr compilation_stats_; std::unique_ptr visualizer_output_; mutable Mutex dump_mutex_; // To synchronize visualizer writing. DISALLOW_COPY_AND_ASSIGN(OptimizingCompiler); }; static const int kMaximumCompilationTimeBeforeWarning = 100; /* ms */ OptimizingCompiler::OptimizingCompiler(const CompilerOptions& compiler_options, CompiledMethodStorage* storage) : Compiler(compiler_options, storage, kMaximumCompilationTimeBeforeWarning), dump_mutex_("Visualizer dump lock") { // Enable C1visualizer output. const std::string& cfg_file_name = compiler_options.GetDumpCfgFileName(); if (!cfg_file_name.empty()) { std::ios_base::openmode cfg_file_mode = compiler_options.GetDumpCfgAppend() ? std::ofstream::app : std::ofstream::out; visualizer_output_.reset(new std::ofstream(cfg_file_name, cfg_file_mode)); DumpInstructionSetFeaturesToCfg(); } if (compiler_options.GetDumpStats()) { compilation_stats_.reset(new OptimizingCompilerStats()); } } OptimizingCompiler::~OptimizingCompiler() { if (compilation_stats_.get() != nullptr) { compilation_stats_->Log(); } } void OptimizingCompiler::DumpInstructionSetFeaturesToCfg() const { const CompilerOptions& compiler_options = GetCompilerOptions(); const InstructionSetFeatures* features = compiler_options.GetInstructionSetFeatures(); std::string isa_string = std::string("isa:") + GetInstructionSetString(features->GetInstructionSet()); std::string features_string = "isa_features:" + features->GetFeatureString(); // It is assumed that visualizer_output_ is empty when calling this function, hence the fake // compilation block containing the ISA features will be printed at the beginning of the .cfg // file. *visualizer_output_ << HGraphVisualizer::InsertMetaDataAsCompilationBlock(isa_string + ' ' + features_string); } bool OptimizingCompiler::CanCompileMethod(uint32_t method_idx ATTRIBUTE_UNUSED, const DexFile& dex_file ATTRIBUTE_UNUSED) const { return true; } static bool IsInstructionSetSupported(InstructionSet instruction_set) { return instruction_set == InstructionSet::kArm || instruction_set == InstructionSet::kArm64 || instruction_set == InstructionSet::kThumb2 || instruction_set == InstructionSet::kX86 || instruction_set == InstructionSet::kX86_64; } bool OptimizingCompiler::RunBaselineOptimizations(HGraph* graph, CodeGenerator* codegen, const DexCompilationUnit& dex_compilation_unit, PassObserver* pass_observer) const { switch (codegen->GetCompilerOptions().GetInstructionSet()) { #ifdef ART_ENABLE_CODEGEN_x86 case InstructionSet::kX86: { OptimizationDef x86_optimizations[] = { OptDef(OptimizationPass::kPcRelativeFixupsX86), }; return RunOptimizations(graph, codegen, dex_compilation_unit, pass_observer, x86_optimizations); } #endif default: UNUSED(graph); UNUSED(codegen); UNUSED(dex_compilation_unit); UNUSED(pass_observer); return false; } } bool OptimizingCompiler::RunArchOptimizations(HGraph* graph, CodeGenerator* codegen, const DexCompilationUnit& dex_compilation_unit, PassObserver* pass_observer) const { switch (codegen->GetCompilerOptions().GetInstructionSet()) { #if defined(ART_ENABLE_CODEGEN_arm) case InstructionSet::kThumb2: case InstructionSet::kArm: { OptimizationDef arm_optimizations[] = { OptDef(OptimizationPass::kInstructionSimplifierArm), OptDef(OptimizationPass::kSideEffectsAnalysis), OptDef(OptimizationPass::kGlobalValueNumbering, "GVN$after_arch"), OptDef(OptimizationPass::kScheduling) }; return RunOptimizations(graph, codegen, dex_compilation_unit, pass_observer, arm_optimizations); } #endif #ifdef ART_ENABLE_CODEGEN_arm64 case InstructionSet::kArm64: { OptimizationDef arm64_optimizations[] = { OptDef(OptimizationPass::kInstructionSimplifierArm64), OptDef(OptimizationPass::kSideEffectsAnalysis), OptDef(OptimizationPass::kGlobalValueNumbering, "GVN$after_arch"), OptDef(OptimizationPass::kScheduling) }; return RunOptimizations(graph, codegen, dex_compilation_unit, pass_observer, arm64_optimizations); } #endif #ifdef ART_ENABLE_CODEGEN_x86 case InstructionSet::kX86: { OptimizationDef x86_optimizations[] = { OptDef(OptimizationPass::kInstructionSimplifierX86), OptDef(OptimizationPass::kSideEffectsAnalysis), OptDef(OptimizationPass::kGlobalValueNumbering, "GVN$after_arch"), OptDef(OptimizationPass::kPcRelativeFixupsX86), OptDef(OptimizationPass::kX86MemoryOperandGeneration) }; return RunOptimizations(graph, codegen, dex_compilation_unit, pass_observer, x86_optimizations); } #endif #ifdef ART_ENABLE_CODEGEN_x86_64 case InstructionSet::kX86_64: { OptimizationDef x86_64_optimizations[] = { OptDef(OptimizationPass::kInstructionSimplifierX86_64), OptDef(OptimizationPass::kSideEffectsAnalysis), OptDef(OptimizationPass::kGlobalValueNumbering, "GVN$after_arch"), OptDef(OptimizationPass::kX86MemoryOperandGeneration) }; return RunOptimizations(graph, codegen, dex_compilation_unit, pass_observer, x86_64_optimizations); } #endif default: return false; } } NO_INLINE // Avoid increasing caller's frame size by large stack-allocated objects. static void AllocateRegisters(HGraph* graph, CodeGenerator* codegen, PassObserver* pass_observer, RegisterAllocator::Strategy strategy, OptimizingCompilerStats* stats) { { PassScope scope(PrepareForRegisterAllocation::kPrepareForRegisterAllocationPassName, pass_observer); PrepareForRegisterAllocation(graph, codegen->GetCompilerOptions(), stats).Run(); } // Use local allocator shared by SSA liveness analysis and register allocator. // (Register allocator creates new objects in the liveness data.) ScopedArenaAllocator local_allocator(graph->GetArenaStack()); SsaLivenessAnalysis liveness(graph, codegen, &local_allocator); { PassScope scope(SsaLivenessAnalysis::kLivenessPassName, pass_observer); liveness.Analyze(); } { PassScope scope(RegisterAllocator::kRegisterAllocatorPassName, pass_observer); std::unique_ptr register_allocator = RegisterAllocator::Create(&local_allocator, codegen, liveness, strategy); register_allocator->AllocateRegisters(); } } // Strip pass name suffix to get optimization name. static std::string ConvertPassNameToOptimizationName(const std::string& pass_name) { size_t pos = pass_name.find(kPassNameSeparator); return pos == std::string::npos ? pass_name : pass_name.substr(0, pos); } void OptimizingCompiler::RunOptimizations(HGraph* graph, CodeGenerator* codegen, const DexCompilationUnit& dex_compilation_unit, PassObserver* pass_observer) const { const std::vector* pass_names = GetCompilerOptions().GetPassesToRun(); if (pass_names != nullptr) { // If passes were defined on command-line, build the optimization // passes and run these instead of the built-in optimizations. // TODO: a way to define depends_on via command-line? const size_t length = pass_names->size(); std::vector optimizations; for (const std::string& pass_name : *pass_names) { std::string opt_name = ConvertPassNameToOptimizationName(pass_name); optimizations.push_back(OptDef(OptimizationPassByName(opt_name), pass_name.c_str())); } RunOptimizations(graph, codegen, dex_compilation_unit, pass_observer, optimizations.data(), length); return; } OptimizationDef optimizations[] = { // Initial optimizations. OptDef(OptimizationPass::kConstantFolding), OptDef(OptimizationPass::kInstructionSimplifier), OptDef(OptimizationPass::kDeadCodeElimination, "dead_code_elimination$initial"), // Inlining. OptDef(OptimizationPass::kInliner), // Simplification (only if inlining occurred). OptDef(OptimizationPass::kConstantFolding, "constant_folding$after_inlining", OptimizationPass::kInliner), OptDef(OptimizationPass::kInstructionSimplifier, "instruction_simplifier$after_inlining", OptimizationPass::kInliner), OptDef(OptimizationPass::kDeadCodeElimination, "dead_code_elimination$after_inlining", OptimizationPass::kInliner), // GVN. OptDef(OptimizationPass::kSideEffectsAnalysis, "side_effects$before_gvn"), OptDef(OptimizationPass::kGlobalValueNumbering), // Simplification (TODO: only if GVN occurred). OptDef(OptimizationPass::kSelectGenerator), OptDef(OptimizationPass::kConstantFolding, "constant_folding$after_gvn"), OptDef(OptimizationPass::kInstructionSimplifier, "instruction_simplifier$after_gvn"), OptDef(OptimizationPass::kDeadCodeElimination, "dead_code_elimination$after_gvn"), // High-level optimizations. OptDef(OptimizationPass::kSideEffectsAnalysis, "side_effects$before_licm"), OptDef(OptimizationPass::kInvariantCodeMotion), OptDef(OptimizationPass::kInductionVarAnalysis), OptDef(OptimizationPass::kBoundsCheckElimination), OptDef(OptimizationPass::kLoopOptimization), // Simplification. OptDef(OptimizationPass::kConstantFolding, "constant_folding$after_bce"), OptDef(OptimizationPass::kAggressiveInstructionSimplifier, "instruction_simplifier$after_bce"), // Other high-level optimizations. OptDef(OptimizationPass::kSideEffectsAnalysis, "side_effects$before_lse"), OptDef(OptimizationPass::kLoadStoreElimination), OptDef(OptimizationPass::kCHAGuardOptimization), OptDef(OptimizationPass::kDeadCodeElimination, "dead_code_elimination$final"), OptDef(OptimizationPass::kCodeSinking), // The codegen has a few assumptions that only the instruction simplifier // can satisfy. For example, the code generator does not expect to see a // HTypeConversion from a type to the same type. OptDef(OptimizationPass::kAggressiveInstructionSimplifier, "instruction_simplifier$before_codegen"), // Eliminate constructor fences after code sinking to avoid // complicated sinking logic to split a fence with many inputs. OptDef(OptimizationPass::kConstructorFenceRedundancyElimination) }; RunOptimizations(graph, codegen, dex_compilation_unit, pass_observer, optimizations); RunArchOptimizations(graph, codegen, dex_compilation_unit, pass_observer); } static ArenaVector EmitAndSortLinkerPatches(CodeGenerator* codegen) { ArenaVector linker_patches(codegen->GetGraph()->GetAllocator()->Adapter()); codegen->EmitLinkerPatches(&linker_patches); // Sort patches by literal offset. Required for .oat_patches encoding. std::sort(linker_patches.begin(), linker_patches.end(), [](const linker::LinkerPatch& lhs, const linker::LinkerPatch& rhs) { return lhs.LiteralOffset() < rhs.LiteralOffset(); }); return linker_patches; } CompiledMethod* OptimizingCompiler::Emit(ArenaAllocator* allocator, CodeVectorAllocator* code_allocator, CodeGenerator* codegen, const dex::CodeItem* code_item_for_osr_check) const { ArenaVector linker_patches = EmitAndSortLinkerPatches(codegen); ScopedArenaVector stack_map = codegen->BuildStackMaps(code_item_for_osr_check); CompiledMethodStorage* storage = GetCompiledMethodStorage(); CompiledMethod* compiled_method = CompiledMethod::SwapAllocCompiledMethod( storage, codegen->GetInstructionSet(), code_allocator->GetMemory(), ArrayRef(stack_map), ArrayRef(*codegen->GetAssembler()->cfi().data()), ArrayRef(linker_patches)); for (const linker::LinkerPatch& patch : linker_patches) { if (codegen->NeedsThunkCode(patch) && storage->GetThunkCode(patch).empty()) { ArenaVector code(allocator->Adapter()); std::string debug_name; codegen->EmitThunkCode(patch, &code, &debug_name); storage->SetThunkCode(patch, ArrayRef(code), debug_name); } } return compiled_method; } CodeGenerator* OptimizingCompiler::TryCompile(ArenaAllocator* allocator, ArenaStack* arena_stack, CodeVectorAllocator* code_allocator, const DexCompilationUnit& dex_compilation_unit, ArtMethod* method, CompilationKind compilation_kind, VariableSizedHandleScope* handles) const { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kAttemptBytecodeCompilation); const CompilerOptions& compiler_options = GetCompilerOptions(); InstructionSet instruction_set = compiler_options.GetInstructionSet(); const DexFile& dex_file = *dex_compilation_unit.GetDexFile(); uint32_t method_idx = dex_compilation_unit.GetDexMethodIndex(); const dex::CodeItem* code_item = dex_compilation_unit.GetCodeItem(); // Always use the Thumb-2 assembler: some runtime functionality // (like implicit stack overflow checks) assume Thumb-2. DCHECK_NE(instruction_set, InstructionSet::kArm); // Do not attempt to compile on architectures we do not support. if (!IsInstructionSetSupported(instruction_set)) { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kNotCompiledUnsupportedIsa); return nullptr; } if (Compiler::IsPathologicalCase(*code_item, method_idx, dex_file)) { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kNotCompiledPathological); return nullptr; } // Implementation of the space filter: do not compile a code item whose size in // code units is bigger than 128. static constexpr size_t kSpaceFilterOptimizingThreshold = 128; if ((compiler_options.GetCompilerFilter() == CompilerFilter::kSpace) && (CodeItemInstructionAccessor(dex_file, code_item).InsnsSizeInCodeUnits() > kSpaceFilterOptimizingThreshold)) { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kNotCompiledSpaceFilter); return nullptr; } CodeItemDebugInfoAccessor code_item_accessor(dex_file, code_item, method_idx); bool dead_reference_safe; ArrayRef interpreter_metadata; // For AOT compilation, we may not get a method, for example if its class is erroneous, // possibly due to an unavailable superclass. JIT should always have a method. DCHECK(Runtime::Current()->IsAotCompiler() || method != nullptr); if (method != nullptr) { const dex::ClassDef* containing_class; { ScopedObjectAccess soa(Thread::Current()); containing_class = &method->GetClassDef(); interpreter_metadata = method->GetQuickenedInfo(); } // MethodContainsRSensitiveAccess is currently slow, but HasDeadReferenceSafeAnnotation() // is currently rarely true. dead_reference_safe = annotations::HasDeadReferenceSafeAnnotation(dex_file, *containing_class) && !annotations::MethodContainsRSensitiveAccess(dex_file, *containing_class, method_idx); } else { // If we could not resolve the class, conservatively assume it's dead-reference unsafe. dead_reference_safe = false; } HGraph* graph = new (allocator) HGraph( allocator, arena_stack, handles, dex_file, method_idx, compiler_options.GetInstructionSet(), kInvalidInvokeType, dead_reference_safe, compiler_options.GetDebuggable(), compilation_kind); if (method != nullptr) { graph->SetArtMethod(method); } std::unique_ptr codegen( CodeGenerator::Create(graph, compiler_options, compilation_stats_.get())); if (codegen.get() == nullptr) { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kNotCompiledNoCodegen); return nullptr; } codegen->GetAssembler()->cfi().SetEnabled(compiler_options.GenerateAnyDebugInfo()); PassObserver pass_observer(graph, codegen.get(), visualizer_output_.get(), compiler_options, dump_mutex_); { VLOG(compiler) << "Building " << pass_observer.GetMethodName(); PassScope scope(HGraphBuilder::kBuilderPassName, &pass_observer); HGraphBuilder builder(graph, code_item_accessor, &dex_compilation_unit, &dex_compilation_unit, codegen.get(), compilation_stats_.get(), interpreter_metadata); GraphAnalysisResult result = builder.BuildGraph(); if (result != kAnalysisSuccess) { switch (result) { case kAnalysisSkipped: { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kNotCompiledSkipped); break; } case kAnalysisInvalidBytecode: { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kNotCompiledInvalidBytecode); break; } case kAnalysisFailThrowCatchLoop: { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kNotCompiledThrowCatchLoop); break; } case kAnalysisFailAmbiguousArrayOp: { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kNotCompiledAmbiguousArrayOp); break; } case kAnalysisFailIrreducibleLoopAndStringInit: { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kNotCompiledIrreducibleLoopAndStringInit); break; } case kAnalysisFailPhiEquivalentInOsr: { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kNotCompiledPhiEquivalentInOsr); break; } case kAnalysisSuccess: UNREACHABLE(); } pass_observer.SetGraphInBadState(); return nullptr; } } if (compilation_kind == CompilationKind::kBaseline) { RunBaselineOptimizations(graph, codegen.get(), dex_compilation_unit, &pass_observer); } else { RunOptimizations(graph, codegen.get(), dex_compilation_unit, &pass_observer); } RegisterAllocator::Strategy regalloc_strategy = compiler_options.GetRegisterAllocationStrategy(); AllocateRegisters(graph, codegen.get(), &pass_observer, regalloc_strategy, compilation_stats_.get()); codegen->Compile(code_allocator); pass_observer.DumpDisassembly(); MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kCompiledBytecode); return codegen.release(); } CodeGenerator* OptimizingCompiler::TryCompileIntrinsic( ArenaAllocator* allocator, ArenaStack* arena_stack, CodeVectorAllocator* code_allocator, const DexCompilationUnit& dex_compilation_unit, ArtMethod* method, VariableSizedHandleScope* handles) const { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kAttemptIntrinsicCompilation); const CompilerOptions& compiler_options = GetCompilerOptions(); InstructionSet instruction_set = compiler_options.GetInstructionSet(); const DexFile& dex_file = *dex_compilation_unit.GetDexFile(); uint32_t method_idx = dex_compilation_unit.GetDexMethodIndex(); // Always use the Thumb-2 assembler: some runtime functionality // (like implicit stack overflow checks) assume Thumb-2. DCHECK_NE(instruction_set, InstructionSet::kArm); // Do not attempt to compile on architectures we do not support. if (!IsInstructionSetSupported(instruction_set)) { return nullptr; } HGraph* graph = new (allocator) HGraph( allocator, arena_stack, handles, dex_file, method_idx, compiler_options.GetInstructionSet(), kInvalidInvokeType, /* dead_reference_safe= */ true, // Intrinsics don't affect dead reference safety. compiler_options.GetDebuggable(), CompilationKind::kOptimized); DCHECK(Runtime::Current()->IsAotCompiler()); DCHECK(method != nullptr); graph->SetArtMethod(method); std::unique_ptr codegen( CodeGenerator::Create(graph, compiler_options, compilation_stats_.get())); if (codegen.get() == nullptr) { return nullptr; } codegen->GetAssembler()->cfi().SetEnabled(compiler_options.GenerateAnyDebugInfo()); PassObserver pass_observer(graph, codegen.get(), visualizer_output_.get(), compiler_options, dump_mutex_); { VLOG(compiler) << "Building intrinsic graph " << pass_observer.GetMethodName(); PassScope scope(HGraphBuilder::kBuilderPassName, &pass_observer); HGraphBuilder builder(graph, CodeItemDebugInfoAccessor(), // Null code item. &dex_compilation_unit, &dex_compilation_unit, codegen.get(), compilation_stats_.get(), /* interpreter_metadata= */ ArrayRef()); builder.BuildIntrinsicGraph(method); } OptimizationDef optimizations[] = { // The codegen has a few assumptions that only the instruction simplifier // can satisfy. OptDef(OptimizationPass::kInstructionSimplifier), }; RunOptimizations(graph, codegen.get(), dex_compilation_unit, &pass_observer, optimizations); RunArchOptimizations(graph, codegen.get(), dex_compilation_unit, &pass_observer); AllocateRegisters(graph, codegen.get(), &pass_observer, compiler_options.GetRegisterAllocationStrategy(), compilation_stats_.get()); if (!codegen->IsLeafMethod()) { VLOG(compiler) << "Intrinsic method is not leaf: " << method->GetIntrinsic() << " " << graph->PrettyMethod(); return nullptr; } codegen->Compile(code_allocator); pass_observer.DumpDisassembly(); VLOG(compiler) << "Compiled intrinsic: " << method->GetIntrinsic() << " " << graph->PrettyMethod(); MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kCompiledIntrinsic); return codegen.release(); } CompiledMethod* OptimizingCompiler::Compile(const dex::CodeItem* code_item, uint32_t access_flags, InvokeType invoke_type, uint16_t class_def_idx, uint32_t method_idx, Handle jclass_loader, const DexFile& dex_file, Handle dex_cache) const { const CompilerOptions& compiler_options = GetCompilerOptions(); DCHECK(compiler_options.IsAotCompiler()); CompiledMethod* compiled_method = nullptr; Runtime* runtime = Runtime::Current(); DCHECK(runtime->IsAotCompiler()); const VerifiedMethod* verified_method = compiler_options.GetVerifiedMethod(&dex_file, method_idx); DCHECK(!verified_method->HasRuntimeThrow()); if (compiler_options.IsMethodVerifiedWithoutFailures(method_idx, class_def_idx, dex_file) || verifier::CanCompilerHandleVerificationFailure( verified_method->GetEncounteredVerificationFailures())) { ArenaAllocator allocator(runtime->GetArenaPool()); ArenaStack arena_stack(runtime->GetArenaPool()); CodeVectorAllocator code_allocator(&allocator); std::unique_ptr codegen; bool compiled_intrinsic = false; { ScopedObjectAccess soa(Thread::Current()); ArtMethod* method = runtime->GetClassLinker()->ResolveMethod( method_idx, dex_cache, jclass_loader, /*referrer=*/ nullptr, invoke_type); DCHECK_EQ(method == nullptr, soa.Self()->IsExceptionPending()); soa.Self()->ClearException(); // Suppress exception if any. VariableSizedHandleScope handles(soa.Self()); Handle compiling_class = handles.NewHandle(method != nullptr ? method->GetDeclaringClass() : nullptr); DexCompilationUnit dex_compilation_unit( jclass_loader, runtime->GetClassLinker(), dex_file, code_item, class_def_idx, method_idx, access_flags, /*verified_method=*/ nullptr, // Not needed by the Optimizing compiler. dex_cache, compiling_class); // Go to native so that we don't block GC during compilation. ScopedThreadSuspension sts(soa.Self(), kNative); if (method != nullptr && UNLIKELY(method->IsIntrinsic())) { DCHECK(compiler_options.IsBootImage()); codegen.reset( TryCompileIntrinsic(&allocator, &arena_stack, &code_allocator, dex_compilation_unit, method, &handles)); if (codegen != nullptr) { compiled_intrinsic = true; } } if (codegen == nullptr) { codegen.reset( TryCompile(&allocator, &arena_stack, &code_allocator, dex_compilation_unit, method, compiler_options.IsBaseline() ? CompilationKind::kBaseline : CompilationKind::kOptimized, &handles)); } } if (codegen.get() != nullptr) { compiled_method = Emit(&allocator, &code_allocator, codegen.get(), compiled_intrinsic ? nullptr : code_item); if (compiled_intrinsic) { compiled_method->MarkAsIntrinsic(); } if (kArenaAllocatorCountAllocations) { codegen.reset(); // Release codegen's ScopedArenaAllocator for memory accounting. size_t total_allocated = allocator.BytesAllocated() + arena_stack.PeakBytesAllocated(); if (total_allocated > kArenaAllocatorMemoryReportThreshold) { MemStats mem_stats(allocator.GetMemStats()); MemStats peak_stats(arena_stack.GetPeakStats()); LOG(INFO) << "Used " << total_allocated << " bytes of arena memory for compiling " << dex_file.PrettyMethod(method_idx) << "\n" << Dumpable(mem_stats) << "\n" << Dumpable(peak_stats); } } } } else { MethodCompilationStat method_stat; if (compiler_options.VerifyAtRuntime()) { method_stat = MethodCompilationStat::kNotCompiledVerifyAtRuntime; } else { method_stat = MethodCompilationStat::kNotCompiledVerificationError; } MaybeRecordStat(compilation_stats_.get(), method_stat); } if (kIsDebugBuild && compiler_options.CompileArtTest() && IsInstructionSetSupported(compiler_options.GetInstructionSet())) { // For testing purposes, we put a special marker on method names // that should be compiled with this compiler (when the // instruction set is supported). This makes sure we're not // regressing. std::string method_name = dex_file.PrettyMethod(method_idx); bool shouldCompile = method_name.find("$opt$") != std::string::npos; DCHECK((compiled_method != nullptr) || !shouldCompile) << "Didn't compile " << method_name; } return compiled_method; } static ScopedArenaVector CreateJniStackMap(ScopedArenaAllocator* allocator, const JniCompiledMethod& jni_compiled_method) { // StackMapStream is quite large, so allocate it using the ScopedArenaAllocator // to stay clear of the frame size limit. std::unique_ptr stack_map_stream( new (allocator) StackMapStream(allocator, jni_compiled_method.GetInstructionSet())); stack_map_stream->BeginMethod( jni_compiled_method.GetFrameSize(), jni_compiled_method.GetCoreSpillMask(), jni_compiled_method.GetFpSpillMask(), /* num_dex_registers= */ 0, /* baseline= */ false); stack_map_stream->EndMethod(); return stack_map_stream->Encode(); } CompiledMethod* OptimizingCompiler::JniCompile(uint32_t access_flags, uint32_t method_idx, const DexFile& dex_file, Handle dex_cache) const { Runtime* runtime = Runtime::Current(); ArenaAllocator allocator(runtime->GetArenaPool()); ArenaStack arena_stack(runtime->GetArenaPool()); const CompilerOptions& compiler_options = GetCompilerOptions(); if (compiler_options.IsBootImage()) { ScopedObjectAccess soa(Thread::Current()); ArtMethod* method = runtime->GetClassLinker()->LookupResolvedMethod( method_idx, dex_cache.Get(), /*class_loader=*/ nullptr); if (method != nullptr && UNLIKELY(method->IsIntrinsic())) { VariableSizedHandleScope handles(soa.Self()); ScopedNullHandle class_loader; // null means boot class path loader. Handle compiling_class = handles.NewHandle(method->GetDeclaringClass()); DexCompilationUnit dex_compilation_unit( class_loader, runtime->GetClassLinker(), dex_file, /*code_item=*/ nullptr, /*class_def_idx=*/ DexFile::kDexNoIndex16, method_idx, access_flags, /*verified_method=*/ nullptr, dex_cache, compiling_class); CodeVectorAllocator code_allocator(&allocator); // Go to native so that we don't block GC during compilation. ScopedThreadSuspension sts(soa.Self(), kNative); std::unique_ptr codegen( TryCompileIntrinsic(&allocator, &arena_stack, &code_allocator, dex_compilation_unit, method, &handles)); if (codegen != nullptr) { CompiledMethod* compiled_method = Emit(&allocator, &code_allocator, codegen.get(), /* item= */ nullptr); compiled_method->MarkAsIntrinsic(); return compiled_method; } } } JniCompiledMethod jni_compiled_method = ArtQuickJniCompileMethod( compiler_options, access_flags, method_idx, dex_file); MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kCompiledNativeStub); ScopedArenaAllocator stack_map_allocator(&arena_stack); // Will hold the stack map. ScopedArenaVector stack_map = CreateJniStackMap(&stack_map_allocator, jni_compiled_method); return CompiledMethod::SwapAllocCompiledMethod( GetCompiledMethodStorage(), jni_compiled_method.GetInstructionSet(), jni_compiled_method.GetCode(), ArrayRef(stack_map), jni_compiled_method.GetCfi(), /* patches= */ ArrayRef()); } Compiler* CreateOptimizingCompiler(const CompilerOptions& compiler_options, CompiledMethodStorage* storage) { return new OptimizingCompiler(compiler_options, storage); } bool EncodeArtMethodInInlineInfo(ArtMethod* method ATTRIBUTE_UNUSED) { // Note: the runtime is null only for unit testing. return Runtime::Current() == nullptr || !Runtime::Current()->IsAotCompiler(); } bool OptimizingCompiler::JitCompile(Thread* self, jit::JitCodeCache* code_cache, jit::JitMemoryRegion* region, ArtMethod* method, CompilationKind compilation_kind, jit::JitLogger* jit_logger) { const CompilerOptions& compiler_options = GetCompilerOptions(); // If the baseline flag was explicitly passed, change the compilation kind // from optimized to baseline. if (compiler_options.IsBaseline() && compilation_kind == CompilationKind::kOptimized) { compilation_kind = CompilationKind::kBaseline; } DCHECK(compiler_options.IsJitCompiler()); DCHECK_EQ(compiler_options.IsJitCompilerForSharedCode(), code_cache->IsSharedRegion(*region)); StackHandleScope<3> hs(self); Handle class_loader(hs.NewHandle( method->GetDeclaringClass()->GetClassLoader())); Handle dex_cache(hs.NewHandle(method->GetDexCache())); DCHECK(method->IsCompilable()); const DexFile* dex_file = method->GetDexFile(); const uint16_t class_def_idx = method->GetClassDefIndex(); const dex::CodeItem* code_item = dex_file->GetCodeItem(method->GetCodeItemOffset()); const uint32_t method_idx = method->GetDexMethodIndex(); const uint32_t access_flags = method->GetAccessFlags(); Runtime* runtime = Runtime::Current(); ArenaAllocator allocator(runtime->GetJitArenaPool()); if (UNLIKELY(method->IsNative())) { JniCompiledMethod jni_compiled_method = ArtQuickJniCompileMethod( compiler_options, access_flags, method_idx, *dex_file); std::vector> roots; ArenaSet> cha_single_implementation_list( allocator.Adapter(kArenaAllocCHA)); ArenaStack arena_stack(runtime->GetJitArenaPool()); // StackMapStream is large and it does not fit into this frame, so we need helper method. ScopedArenaAllocator stack_map_allocator(&arena_stack); // Will hold the stack map. ScopedArenaVector stack_map = CreateJniStackMap(&stack_map_allocator, jni_compiled_method); ArrayRef reserved_code; ArrayRef reserved_data; if (!code_cache->Reserve(self, region, jni_compiled_method.GetCode().size(), stack_map.size(), /* number_of_roots= */ 0, method, /*out*/ &reserved_code, /*out*/ &reserved_data)) { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kJitOutOfMemoryForCommit); return false; } const uint8_t* code = reserved_code.data() + OatQuickMethodHeader::InstructionAlignedSize(); // Add debug info after we know the code location but before we update entry-point. std::vector debug_info; if (compiler_options.GenerateAnyDebugInfo()) { debug::MethodDebugInfo info = {}; info.custom_name = "art_jni_trampoline"; info.dex_file = dex_file; info.class_def_index = class_def_idx; info.dex_method_index = method_idx; info.access_flags = access_flags; info.code_item = code_item; info.isa = jni_compiled_method.GetInstructionSet(); info.deduped = false; info.is_native_debuggable = compiler_options.GetNativeDebuggable(); info.is_optimized = true; info.is_code_address_text_relative = false; info.code_address = reinterpret_cast(code); info.code_size = jni_compiled_method.GetCode().size(); info.frame_size_in_bytes = jni_compiled_method.GetFrameSize(); info.code_info = nullptr; info.cfi = jni_compiled_method.GetCfi(); debug_info = GenerateJitDebugInfo(info); } if (!code_cache->Commit(self, region, method, reserved_code, jni_compiled_method.GetCode(), reserved_data, roots, ArrayRef(stack_map), debug_info, /* is_full_debug_info= */ compiler_options.GetGenerateDebugInfo(), compilation_kind, /* has_should_deoptimize_flag= */ false, cha_single_implementation_list)) { code_cache->Free(self, region, reserved_code.data(), reserved_data.data()); return false; } Runtime::Current()->GetJit()->AddMemoryUsage(method, allocator.BytesUsed()); if (jit_logger != nullptr) { jit_logger->WriteLog(code, jni_compiled_method.GetCode().size(), method); } return true; } ArenaStack arena_stack(runtime->GetJitArenaPool()); CodeVectorAllocator code_allocator(&allocator); VariableSizedHandleScope handles(self); std::unique_ptr codegen; { Handle compiling_class = handles.NewHandle(method->GetDeclaringClass()); DexCompilationUnit dex_compilation_unit( class_loader, runtime->GetClassLinker(), *dex_file, code_item, class_def_idx, method_idx, access_flags, /*verified_method=*/ nullptr, dex_cache, compiling_class); // Go to native so that we don't block GC during compilation. ScopedThreadSuspension sts(self, kNative); codegen.reset( TryCompile(&allocator, &arena_stack, &code_allocator, dex_compilation_unit, method, compilation_kind, &handles)); if (codegen.get() == nullptr) { return false; } } ScopedArenaVector stack_map = codegen->BuildStackMaps(code_item); ArrayRef reserved_code; ArrayRef reserved_data; if (!code_cache->Reserve(self, region, code_allocator.GetMemory().size(), stack_map.size(), /*number_of_roots=*/codegen->GetNumberOfJitRoots(), method, /*out*/ &reserved_code, /*out*/ &reserved_data)) { MaybeRecordStat(compilation_stats_.get(), MethodCompilationStat::kJitOutOfMemoryForCommit); return false; } const uint8_t* code = reserved_code.data() + OatQuickMethodHeader::InstructionAlignedSize(); const uint8_t* roots_data = reserved_data.data(); std::vector> roots; codegen->EmitJitRoots(code_allocator.GetData(), roots_data, &roots); // The root Handle<>s filled by the codegen reference entries in the VariableSizedHandleScope. DCHECK(std::all_of(roots.begin(), roots.end(), [&handles](Handle root){ return handles.Contains(root.GetReference()); })); // Add debug info after we know the code location but before we update entry-point. std::vector debug_info; if (compiler_options.GenerateAnyDebugInfo()) { debug::MethodDebugInfo info = {}; DCHECK(info.custom_name.empty()); info.dex_file = dex_file; info.class_def_index = class_def_idx; info.dex_method_index = method_idx; info.access_flags = access_flags; info.code_item = code_item; info.isa = codegen->GetInstructionSet(); info.deduped = false; info.is_native_debuggable = compiler_options.GetNativeDebuggable(); info.is_optimized = true; info.is_code_address_text_relative = false; info.code_address = reinterpret_cast(code); info.code_size = code_allocator.GetMemory().size(); info.frame_size_in_bytes = codegen->GetFrameSize(); info.code_info = stack_map.size() == 0 ? nullptr : stack_map.data(); info.cfi = ArrayRef(*codegen->GetAssembler()->cfi().data()); debug_info = GenerateJitDebugInfo(info); } if (!code_cache->Commit(self, region, method, reserved_code, code_allocator.GetMemory(), reserved_data, roots, ArrayRef(stack_map), debug_info, /* is_full_debug_info= */ compiler_options.GetGenerateDebugInfo(), compilation_kind, codegen->GetGraph()->HasShouldDeoptimizeFlag(), codegen->GetGraph()->GetCHASingleImplementationList())) { code_cache->Free(self, region, reserved_code.data(), reserved_data.data()); return false; } Runtime::Current()->GetJit()->AddMemoryUsage(method, allocator.BytesUsed()); if (jit_logger != nullptr) { jit_logger->WriteLog(code, code_allocator.GetMemory().size(), method); } if (kArenaAllocatorCountAllocations) { codegen.reset(); // Release codegen's ScopedArenaAllocator for memory accounting. size_t total_allocated = allocator.BytesAllocated() + arena_stack.PeakBytesAllocated(); if (total_allocated > kArenaAllocatorMemoryReportThreshold) { MemStats mem_stats(allocator.GetMemStats()); MemStats peak_stats(arena_stack.GetPeakStats()); LOG(INFO) << "Used " << total_allocated << " bytes of arena memory for compiling " << dex_file->PrettyMethod(method_idx) << "\n" << Dumpable(mem_stats) << "\n" << Dumpable(peak_stats); } } return true; } std::vector OptimizingCompiler::GenerateJitDebugInfo(const debug::MethodDebugInfo& info) { const CompilerOptions& compiler_options = GetCompilerOptions(); if (compiler_options.GenerateAnyDebugInfo()) { // If both flags are passed, generate full debug info. const bool mini_debug_info = !compiler_options.GetGenerateDebugInfo(); // Create entry for the single method that we just compiled. InstructionSet isa = compiler_options.GetInstructionSet(); const InstructionSetFeatures* features = compiler_options.GetInstructionSetFeatures(); return debug::MakeElfFileForJIT(isa, features, mini_debug_info, info); } return std::vector(); } } // namespace art