CPUProfiler: Improve line numbers support in profiler.

src/cpu-profiler.cc

 // Copyright 2012 the V8 project authors. All rights reserved.
 // Redistribution and use in source and binary forms, with or without
 // modification, are permitted provided that the following conditions are
 // met:
 //
 //     * Redistributions of source code must retain the above copyright
 //       notice, this list of conditions and the following disclaimer.
 //     * Redistributions in binary form must reproduce the above
 //       copyright notice, this list of conditions and the following
 //       disclaimer in the documentation and/or other materials provided
@@ skipping 34 matching lines @@
 static const int kProfilerStackSize = 64 * KB;
 
 
 ProfilerEventsProcessor::ProfilerEventsProcessor(ProfileGenerator* generator)
     : Thread(Thread::Options("v8:ProfEvntProc", kProfilerStackSize)),
       generator_(generator),
       running_(true),
       ticks_buffer_(sizeof(TickSampleEventRecord),
                     kTickSamplesBufferChunkSize,
                     kTickSamplesBufferChunksCount),
-      enqueue_order_(0) {
+      last_recorded_code_event_id_(0), last_processed_code_event_id_(0) {
 }
 
 
 void ProfilerEventsProcessor::Enqueue(const CodeEventsContainer& event) {
-  event.generic.order = ++enqueue_order_;
+  event.generic.order = ++last_recorded_code_event_id_;
   events_buffer_.Enqueue(event);
 }
 
 
 void ProfilerEventsProcessor::AddCurrentStack(Isolate* isolate) {
-  TickSampleEventRecord record(enqueue_order_);
+  TickSampleEventRecord record(last_recorded_code_event_id_);
   TickSample* sample = &record.sample;
   sample->state = isolate->current_vm_state();
   sample->pc = reinterpret_cast<Address>(sample);  // Not NULL.
   for (StackTraceFrameIterator it(isolate);
        !it.done() && sample->frames_count < TickSample::kMaxFramesCount;
        it.Advance()) {
     sample->stack[sample->frames_count++] = it.frame()->pc();
   }
   ticks_from_vm_buffer_.Enqueue(record);
 }
 
 
-bool ProfilerEventsProcessor::ProcessCodeEvent(unsigned* dequeue_order) {
+void ProfilerEventsProcessor::Start() {
+  last_processed_code_event_id_ = 0;
+  while (ProcessCodeEvent()) { }

[yurys, 2013/07/05 13:10:00]

Why do we have to do this processing on the main thread? It slows down the start
and adds access to the code map from one more thread.

[loislo, 2013/07/05 13:15:37]

reverted

+  StartSynchronously();
+}
+
+
+void ProfilerEventsProcessor::Stop() {

[yurys, 2013/07/05 13:10:00]

StopAndJoin or StopSynchronously.

[loislo, 2013/07/05 13:15:37]

Done.

+  if (!running_) return;
+  running_ = false;
+  Join();
+}
+
+
+bool ProfilerEventsProcessor::ProcessCodeEvent() {
   CodeEventsContainer record;
   if (events_buffer_.Dequeue(&record)) {
     switch (record.generic.type) {
 #define PROFILER_TYPE_CASE(type, clss)                          \
       case CodeEventRecord::type:                               \
         record.clss##_.UpdateCodeMap(generator_->code_map());   \
         break;
 
       CODE_EVENTS_TYPE_LIST(PROFILER_TYPE_CASE)
 
 #undef PROFILER_TYPE_CASE
       default: return true;  // Skip record.
     }
-    *dequeue_order = record.generic.order;
+    last_processed_code_event_id_ = record.generic.order;
     return true;
   }
   return false;
 }
 
 
-bool ProfilerEventsProcessor::ProcessTicks(unsigned dequeue_order) {
+bool ProfilerEventsProcessor::ProcessTicks() {
   while (true) {
     if (!ticks_from_vm_buffer_.IsEmpty()
-        && ticks_from_vm_buffer_.Peek()->order == dequeue_order) {
+        && ticks_from_vm_buffer_.Peek()->order ==
+           last_processed_code_event_id_) {
       TickSampleEventRecord record;
       ticks_from_vm_buffer_.Dequeue(&record);
       generator_->RecordTickSample(record.sample);
     }
 
     const TickSampleEventRecord* rec =
         TickSampleEventRecord::cast(ticks_buffer_.StartDequeue());
     if (rec == NULL) return !ticks_from_vm_buffer_.IsEmpty();
     // Make a local copy of tick sample record to ensure that it won't
     // be modified as we are processing it. This is possible as the
     // sampler writes w/o any sync to the queue, so if the processor
     // will get far behind, a record may be modified right under its
     // feet.
     TickSampleEventRecord record = *rec;
-    if (record.order == dequeue_order) {
-      // A paranoid check to make sure that we don't get a memory overrun
-      // in case of frames_count having a wild value.
-      if (record.sample.frames_count < 0
-          || record.sample.frames_count > TickSample::kMaxFramesCount)
-        record.sample.frames_count = 0;
-      generator_->RecordTickSample(record.sample);
-      ticks_buffer_.FinishDequeue();
-    } else {
-      return true;
-    }
+    if (record.order != last_processed_code_event_id_) return true;
+
+    // A paranoid check to make sure that we don't get a memory overrun
+    // in case of frames_count having a wild value.
+    if (record.sample.frames_count < 0
+        || record.sample.frames_count > TickSample::kMaxFramesCount)
+      record.sample.frames_count = 0;
+    generator_->RecordTickSample(record.sample);
+    ticks_buffer_.FinishDequeue();
   }
 }
 
 
 void ProfilerEventsProcessor::Run() {
-  unsigned dequeue_order = 0;
-
   while (running_) {
     // Process ticks until we have any.
-    if (ProcessTicks(dequeue_order)) {
-      // All ticks of the current dequeue_order are processed,
+    if (ProcessTicks()) {
+      // All ticks of the current last_processed_code_event_id_ are processed,
       // proceed to the next code event.
-      ProcessCodeEvent(&dequeue_order);
+      ProcessCodeEvent();
     }
     YieldCPU();
   }
 
   // Process remaining tick events.
   ticks_buffer_.FlushResidualRecords();
-  // Perform processing until we have tick events, skip remaining code events.
-  while (ProcessTicks(dequeue_order) && ProcessCodeEvent(&dequeue_order)) { }
+  do {
+    ProcessTicks();
+  } while (ProcessCodeEvent());
 }
 
 
 int CpuProfiler::GetProfilesCount() {
   // The count of profiles doesn't depend on a security token.
   return profiles_->Profiles(TokenEnumerator::kNoSecurityToken)->length();
 }
 
 
 CpuProfile* CpuProfiler::GetProfile(Object* security_token, int index) {
@@ skipping 278 matching lines @@
 
 void CpuProfiler::StartProcessorIfNotStarted() {
   if (processor_ == NULL) {
     Logger* logger = isolate_->logger();
     // Disable logging when using the new implementation.
     saved_logging_nesting_ = logger->logging_nesting_;
     logger->logging_nesting_ = 0;
     generator_ = new ProfileGenerator(profiles_);
     processor_ = new ProfilerEventsProcessor(generator_);
     is_profiling_ = true;
-    processor_->StartSynchronously();
     // Enumerate stuff we already have in the heap.
     ASSERT(isolate_->heap()->HasBeenSetUp());
     if (!FLAG_prof_browser_mode) {
       logger->LogCodeObjects();
     }
     logger->LogCompiledFunctions();
     logger->LogAccessorCallbacks();
     LogBuiltins();
+    processor_->Start();

[loislo, 2013/07/05 13:15:37]

reverted

     // Enable stack sampling.
     Sampler* sampler = logger->sampler();
     sampler->IncreaseProfilingDepth();
     if (!sampler->IsActive()) {
       sampler->Start();
       need_to_stop_sampler_ = true;
     }
   }
 }
 
@@ skipping 31 matching lines @@
 void CpuProfiler::StopProcessor() {
   Logger* logger = isolate_->logger();
   Sampler* sampler = reinterpret_cast<Sampler*>(logger->ticker_);
   sampler->DecreaseProfilingDepth();
   if (need_to_stop_sampler_) {
     sampler->Stop();
     need_to_stop_sampler_ = false;
   }
   is_profiling_ = false;
   processor_->Stop();
-  processor_->Join();
   delete processor_;
   delete generator_;
   processor_ = NULL;
   generator_ = NULL;
   logger->logging_nesting_ = saved_logging_nesting_;
 }
 
 
 void CpuProfiler::LogBuiltins() {
   Builtins* builtins = isolate_->builtins();
   ASSERT(builtins->is_initialized());
   for (int i = 0; i < Builtins::builtin_count; i++) {
     CodeEventsContainer evt_rec(CodeEventRecord::REPORT_BUILTIN);
     ReportBuiltinEventRecord* rec = &evt_rec.ReportBuiltinEventRecord_;
     Builtins::Name id = static_cast<Builtins::Name>(i);
     rec->start = builtins->builtin(id)->address();
     rec->builtin_id = id;
     processor_->Enqueue(evt_rec);
   }
 }
 
 
 } }  // namespace v8::internal