Add heap allocator usage to task profiler.

base/tracked_objects.cc

 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "base/tracked_objects.h"
 
 #include <limits.h>
 #include <stdlib.h>
 
 #include "base/atomicops.h"
 #include "base/base_switches.h"
 #include "base/command_line.h"
 #include "base/compiler_specific.h"
 #include "base/debug/leak_annotations.h"
+#include "base/debug/scoped_thread_heap_usage.h"
 #include "base/logging.h"
 #include "base/process/process_handle.h"
 #include "base/strings/stringprintf.h"
 #include "base/third_party/valgrind/memcheck.h"
 #include "base/threading/worker_pool.h"
 #include "base/tracking_info.h"
 #include "build/build_config.h"
 
 using base::TimeDelta;
 
@@ skipping 42 matching lines @@
              switches::kProfilerTiming) ==
          switches::kProfilerTimingDisabledValue)
             ? DISABLED_TIMING
             : ENABLED_TIMING;
     base::subtle::NoBarrier_Store(&g_profiler_timing_enabled,
                                   current_timing_enabled);
   }
   return current_timing_enabled == ENABLED_TIMING;
 }
 
+void SaturatingAdd(const uint32_t addend, base::subtle::Atomic32* sum) {
+  // Bail quick if no work or alreadu saturated.

[Primiano Tucci (use gerrit), 2016/10/13 20:47:44]

typo: alreadu

[Sigurður Ásgeirsson, 2016/10/14 20:11:35]

Done.

+  if (addend == 0U || *sum == INT_MAX)

[Primiano Tucci (use gerrit), 2016/10/13 20:47:44]

I think you are supposed to use at least NoBarrier_Load or this will stop
building the day we fix base::subtle to use proper atomics

[Sigurður Ásgeirsson, 2016/10/14 20:11:36]

I'm modeling on what's done elsewhere in this file. The threading model here is
quite interesting, as there's only one thread that'll write to each datum. So,
from the perspective of data integrity, it's fine for this thread to read the
data without atomic primitives, as no other thread will ever modify it.

+    return;
+
+  // Check for ovewflow.
+  int32_t new_sum = *sum + addend;
+  if (new_sum < *sum)
+    new_sum = INT_MAX;
+
+  base::subtle::NoBarrier_Store(sum, new_sum);

[Primiano Tucci (use gerrit), 2016/10/13 20:47:44]

this is not atomic anymore now? Don't you want a NoBarrier_AtomicExchange here?

[Sigurður Ásgeirsson, 2016/10/14 20:11:35]

This thread is the only writer, and in the interest of efficiency, we accept low
atomicity guarantees here, as elsewhere in this file. Notably we only get
per-datum atomicity at best.

+}
+
 }  // namespace
 
 //------------------------------------------------------------------------------
 // DeathData tallies durations when a death takes place.
 
 DeathData::DeathData()
     : count_(0),
       sample_probability_count_(0),
       run_duration_sum_(0),
       queue_duration_sum_(0),
       run_duration_max_(0),
       queue_duration_max_(0),
       run_duration_sample_(0),
       queue_duration_sample_(0),
-      last_phase_snapshot_(nullptr) {
-}
+      alloc_ops_(0),
+      free_ops_(0),
+      allocated_bytes_(0),
+      freed_bytes_(0),
+      alloc_overhead_bytes_(0),
+      max_allocated_bytes_(0),
+      last_phase_snapshot_(nullptr) {}
 
 DeathData::DeathData(const DeathData& other)
     : count_(other.count_),
       sample_probability_count_(other.sample_probability_count_),
       run_duration_sum_(other.run_duration_sum_),
       queue_duration_sum_(other.queue_duration_sum_),
       run_duration_max_(other.run_duration_max_),
       queue_duration_max_(other.queue_duration_max_),
       run_duration_sample_(other.run_duration_sample_),
       queue_duration_sample_(other.queue_duration_sample_),
+      alloc_ops_(other.alloc_ops_),
+      free_ops_(other.free_ops_),
+      allocated_bytes_(other.allocated_bytes_),
+      freed_bytes_(other.freed_bytes_),
+      alloc_overhead_bytes_(other.alloc_overhead_bytes_),
+      max_allocated_bytes_(other.max_allocated_bytes_),
       last_phase_snapshot_(nullptr) {
   // This constructor will be used by std::map when adding new DeathData values
   // to the map.  At that point, last_phase_snapshot_ is still NULL, so we don't
   // need to worry about ownership transfer.
   DCHECK(other.last_phase_snapshot_ == nullptr);
 }
 
 DeathData::~DeathData() {
   while (last_phase_snapshot_) {
     const DeathDataPhaseSnapshot* snapshot = last_phase_snapshot_;
     last_phase_snapshot_ = snapshot->prev;
     delete snapshot;
   }
 }
 
 // TODO(jar): I need to see if this macro to optimize branching is worth using.
 //
 // This macro has no branching, so it is surely fast, and is equivalent to:
 //             if (assign_it)
 //               target = source;
 // We use a macro rather than a template to force this to inline.
 // Related code for calculating max is discussed on the web.
 #define CONDITIONAL_ASSIGN(assign_it, target, source) \
   ((target) ^= ((target) ^ (source)) & -static_cast<int32_t>(assign_it))
 
-void DeathData::RecordDeath(const int32_t queue_duration,
-                            const int32_t run_duration,
-                            const uint32_t random_number) {
+void DeathData::RecordDurations(const int32_t queue_duration,
+                                const int32_t run_duration,
+                                const uint32_t random_number) {
   // We'll just clamp at INT_MAX, but we should note this in the UI as such.
   if (count_ < INT_MAX)
     base::subtle::NoBarrier_Store(&count_, count_ + 1);
 
   int sample_probability_count =
       base::subtle::NoBarrier_Load(&sample_probability_count_);
   if (sample_probability_count < INT_MAX)
     ++sample_probability_count;
   base::subtle::NoBarrier_Store(&sample_probability_count_,
                                 sample_probability_count);
@@ skipping 16 matching lines @@
   // but that should be inconsequentially likely).  We ignore the fact that we
   // correlated our selection of a sample to the run and queue times (i.e., we
   // used them to generate random_number).
   CHECK_GT(sample_probability_count, 0);
   if (0 == (random_number % sample_probability_count)) {
     base::subtle::NoBarrier_Store(&queue_duration_sample_, queue_duration);
     base::subtle::NoBarrier_Store(&run_duration_sample_, run_duration);
   }
 }
 
+void DeathData::RecordAllocations(const uint32_t alloc_ops,
+                                  const uint32_t free_ops,
+                                  const uint32_t allocated_bytes,
+                                  const uint32_t freed_bytes,
+                                  const uint32_t alloc_overhead_bytes,
+                                  const uint32_t max_allocated_bytes) {
+  // Use saturating arithmetic.
+  SaturatingAdd(alloc_ops, &alloc_ops_);
+  SaturatingAdd(free_ops, &free_ops_);
+  SaturatingAdd(allocated_bytes, &allocated_bytes_);
+  SaturatingAdd(freed_bytes, &freed_bytes_);
+  SaturatingAdd(alloc_overhead_bytes, &alloc_overhead_bytes_);
+
+  if (max_allocated_bytes > INT_MAX)

[Primiano Tucci (use gerrit), 2016/10/13 20:47:44]

I think base::saturated_cast might help here?

[Sigurður Ásgeirsson, 2016/10/14 20:11:35]

Thanks, neat! I didn't know of that.

+    base::subtle::NoBarrier_Store(&max_allocated_bytes_, INT_MAX);
+  else if (static_cast<int32_t>(max_allocated_bytes) > max_allocated_bytes_)
+    base::subtle::NoBarrier_Store(&max_allocated_bytes_, max_allocated_bytes);
+}
+
 void DeathData::OnProfilingPhaseCompleted(int profiling_phase) {
   // Snapshotting and storing current state.
-  last_phase_snapshot_ = new DeathDataPhaseSnapshot(
-      profiling_phase, count(), run_duration_sum(), run_duration_max(),
-      run_duration_sample(), queue_duration_sum(), queue_duration_max(),
-      queue_duration_sample(), last_phase_snapshot_);
+  last_phase_snapshot_ =
+      new DeathDataPhaseSnapshot(profiling_phase, *this, last_phase_snapshot_);
 
   // Not touching fields for which a delta can be computed by comparing with a
   // snapshot from the previous phase. Resetting other fields.  Sample values
   // will be reset upon next death recording because sample_probability_count_
   // is set to 0.
   // We avoid resetting to 0 in favor of deltas whenever possible.  The reason
   // is that for incrementable fields, resetting to 0 from the snapshot thread
   // potentially in parallel with incrementing in the death thread may result in
   // significant data corruption that has a potential to grow with time.  Not
   // resetting incrementable fields and using deltas will cause any
@@ skipping 19 matching lines @@
 }
 
 //------------------------------------------------------------------------------
 DeathDataSnapshot::DeathDataSnapshot()
     : count(-1),
       run_duration_sum(-1),
       run_duration_max(-1),
       run_duration_sample(-1),
       queue_duration_sum(-1),
       queue_duration_max(-1),
-      queue_duration_sample(-1) {
-}
+      queue_duration_sample(-1),
+      alloc_ops(-1),
+      free_ops(-1),
+      allocated_bytes(-1),
+      freed_bytes(-1),
+      alloc_overhead_bytes(-1),
+      max_allocated_bytes(-1) {}
 
 DeathDataSnapshot::DeathDataSnapshot(int count,
                                      int32_t run_duration_sum,
                                      int32_t run_duration_max,
                                      int32_t run_duration_sample,
                                      int32_t queue_duration_sum,
                                      int32_t queue_duration_max,
-                                     int32_t queue_duration_sample)
+                                     int32_t queue_duration_sample,
+                                     int32_t alloc_ops,
+                                     int32_t free_ops,
+                                     int32_t allocated_bytes,
+                                     int32_t freed_bytes,
+                                     int32_t alloc_overhead_bytes,
+                                     int32_t max_allocated_bytes)
     : count(count),
       run_duration_sum(run_duration_sum),
       run_duration_max(run_duration_max),
       run_duration_sample(run_duration_sample),
       queue_duration_sum(queue_duration_sum),
       queue_duration_max(queue_duration_max),
-      queue_duration_sample(queue_duration_sample) {}
+      queue_duration_sample(queue_duration_sample),
+      alloc_ops(alloc_ops),
+      free_ops(free_ops),
+      allocated_bytes(allocated_bytes),
+      freed_bytes(freed_bytes),
+      alloc_overhead_bytes(alloc_overhead_bytes),
+      max_allocated_bytes(max_allocated_bytes) {}
+
+DeathDataSnapshot::DeathDataSnapshot(const DeathData& death_data)
+    : count(death_data.count()),
+      run_duration_sum(death_data.run_duration_sum()),
+      run_duration_max(death_data.run_duration_max()),
+      run_duration_sample(death_data.run_duration_sample()),
+      queue_duration_sum(death_data.queue_duration_sum()),
+      queue_duration_max(death_data.queue_duration_max()),
+      queue_duration_sample(death_data.queue_duration_sample()),
+      alloc_ops(death_data.alloc_ops()),
+      free_ops(death_data.free_ops()),
+      allocated_bytes(death_data.allocated_bytes()),
+      freed_bytes(death_data.freed_bytes()),
+      alloc_overhead_bytes(death_data.alloc_overhead_bytes()),
+      max_allocated_bytes(death_data.max_allocated_bytes()) {}
 
 DeathDataSnapshot::~DeathDataSnapshot() {
 }
 
 DeathDataSnapshot DeathDataSnapshot::Delta(
     const DeathDataSnapshot& older) const {
-  return DeathDataSnapshot(count - older.count,
-                           run_duration_sum - older.run_duration_sum,
-                           run_duration_max, run_duration_sample,
-                           queue_duration_sum - older.queue_duration_sum,
-                           queue_duration_max, queue_duration_sample);
+  return DeathDataSnapshot(
+      count - older.count, run_duration_sum - older.run_duration_sum,
+      run_duration_max, run_duration_sample,
+      queue_duration_sum - older.queue_duration_sum, queue_duration_max,
+      queue_duration_sample, alloc_ops - older.alloc_ops,
+      free_ops - older.free_ops, allocated_bytes - older.allocated_bytes,
+      freed_bytes - older.freed_bytes,
+      alloc_overhead_bytes - older.alloc_overhead_bytes, max_allocated_bytes);
 }
 
 //------------------------------------------------------------------------------
 BirthOnThread::BirthOnThread(const Location& location,
                              const ThreadData& current)
     : location_(location),
       birth_thread_(&current) {
 }
 
 //------------------------------------------------------------------------------
@@ skipping 198 matching lines @@
 
   // Add births that are still active -- i.e. objects that have tallied a birth,
   // but have not yet tallied a matching death, and hence must be either
   // running, queued up, or being held in limbo for future posting.
   auto* current_phase_tasks =
       &process_data_snapshot->phased_snapshots[current_profiling_phase].tasks;
   for (const auto& birth_count : birth_counts) {
     if (birth_count.second > 0) {
       current_phase_tasks->push_back(
           TaskSnapshot(BirthOnThreadSnapshot(*birth_count.first),
-                       DeathDataSnapshot(birth_count.second, 0, 0, 0, 0, 0, 0),
+                       DeathDataSnapshot(birth_count.second, 0, 0, 0, 0, 0, 0,
+                                         0, 0, 0, 0, 0, 0),
                        "Still_Alive"));
     }
   }
 }
 
 // static
 void ThreadData::OnProfilingPhaseCompleted(int profiling_phase) {
   // Get an unchanging copy of a ThreadData list.
   ThreadData* my_list = ThreadData::first();
 
@@ skipping 38 matching lines @@
       static_cast<uint32_t>(&births - reinterpret_cast<Births*>(0));
 
   DeathMap::iterator it = death_map_.find(&births);
   DeathData* death_data;
   if (it != death_map_.end()) {
     death_data = &it->second;
   } else {
     base::AutoLock lock(map_lock_);  // Lock as the map may get relocated now.
     death_data = &death_map_[&births];
   }  // Release lock ASAP.
-  death_data->RecordDeath(queue_duration, run_duration, random_number_);
+  death_data->RecordDurations(queue_duration, run_duration, random_number_);
+
+#if BUILDFLAG(USE_EXPERIMENTAL_ALLOCATOR_SHIM)
+  // TODO(siggi): Make this conditional on whether heap tracking is enabled.
+  // TODO(siggi): Should these be passed as uint64_t perhaps?
+  // DO NOT SUBMIT
+  base::debug::ThreadAllocatorUsage heap_usage = stopwatch.heap_usage().usage();
+  death_data->RecordAllocations(
+      static_cast<int32_t>(heap_usage.alloc_ops),
+      static_cast<int32_t>(heap_usage.free_ops),
+      static_cast<int32_t>(heap_usage.alloc_bytes),
+      static_cast<int32_t>(heap_usage.free_bytes),
+      static_cast<int32_t>(heap_usage.alloc_overhead_bytes),
+      static_cast<int32_t>(heap_usage.max_allocated_bytes));
+#endif
 }
 
 // static
 Births* ThreadData::TallyABirthIfActive(const Location& location) {
   if (!TrackingStatus())
     return NULL;
   ThreadData* current_thread_data = Get();
   if (!current_thread_data)
     return NULL;
   return current_thread_data->TallyABirth(location);
@@ skipping 118 matching lines @@
                               BirthMap* birth_map,
                               DeathsSnapshot* deaths) {
   base::AutoLock lock(map_lock_);
 
   for (const auto& birth : birth_map_)
     (*birth_map)[birth.first] = birth.second;
 
   for (const auto& death : death_map_) {
     deaths->push_back(std::make_pair(
         death.first,
-        DeathDataPhaseSnapshot(profiling_phase, death.second.count(),
-                               death.second.run_duration_sum(),
-                               death.second.run_duration_max(),
-                               death.second.run_duration_sample(),
-                               death.second.queue_duration_sum(),
-                               death.second.queue_duration_max(),
-                               death.second.queue_duration_sample(),
+        DeathDataPhaseSnapshot(profiling_phase, death.second,
                                death.second.last_phase_snapshot())));
   }
 }
 
 void ThreadData::OnProfilingPhaseCompletedOnThread(int profiling_phase) {
   base::AutoLock lock(map_lock_);
 
   for (auto& death : death_map_) {
     death.second.OnProfilingPhaseCompleted(profiling_phase);
   }
@@ skipping 159 matching lines @@
 #endif
 }
 
 void TaskStopwatch::Start() {
 #if DCHECK_IS_ON()
   DCHECK(state_ == CREATED);
   state_ = RUNNING;
 #endif
 
   start_time_ = ThreadData::Now();
+#if BUILDFLAG(USE_EXPERIMENTAL_ALLOCATOR_SHIM)
+  heap_usage_.Start();
+#endif
 
   current_thread_data_ = ThreadData::Get();
   if (!current_thread_data_)
     return;
 
   parent_ = current_thread_data_->current_stopwatch_;
 #if DCHECK_IS_ON()
   if (parent_) {
     DCHECK(parent_->state_ == RUNNING);
     DCHECK(parent_->child_ == NULL);
     parent_->child_ = this;
   }
 #endif
   current_thread_data_->current_stopwatch_ = this;
 }
 
 void TaskStopwatch::Stop() {
   const TrackedTime end_time = ThreadData::Now();
 #if DCHECK_IS_ON()
   DCHECK(state_ == RUNNING);
   state_ = STOPPED;
   DCHECK(child_ == NULL);
+
+#endif
+#if BUILDFLAG(USE_EXPERIMENTAL_ALLOCATOR_SHIM)
+  heap_usage_.Stop(true);
 #endif
 
   if (!start_time_.is_null() && !end_time.is_null()) {
     wallclock_duration_ms_ = (end_time - start_time_).InMilliseconds();
   }
 
   if (!current_thread_data_)
     return;
 
   DCHECK(current_thread_data_->current_stopwatch_ == this);
@@ skipping 19 matching lines @@
 }
 
 int32_t TaskStopwatch::RunDurationMs() const {
 #if DCHECK_IS_ON()
   DCHECK(state_ == STOPPED);
 #endif
 
   return wallclock_duration_ms_ - excluded_duration_ms_;
 }
 
+#if BUILDFLAG(USE_EXPERIMENTAL_ALLOCATOR_SHIM)
+const base::debug::HeapUsageTracker& TaskStopwatch::heap_usage() const {
+  return heap_usage_;
+}
+#endif
+
 ThreadData* TaskStopwatch::GetThreadData() const {
 #if DCHECK_IS_ON()
   DCHECK(state_ != CREATED);
 #endif
 
   return current_thread_data_;
 }
 
 //------------------------------------------------------------------------------
 // DeathDataPhaseSnapshot
 
 DeathDataPhaseSnapshot::DeathDataPhaseSnapshot(
     int profiling_phase,
-    int count,
-    int32_t run_duration_sum,
-    int32_t run_duration_max,
-    int32_t run_duration_sample,
-    int32_t queue_duration_sum,
-    int32_t queue_duration_max,
-    int32_t queue_duration_sample,
+    const DeathData& death,
     const DeathDataPhaseSnapshot* prev)
-    : profiling_phase(profiling_phase),
-      death_data(count,
-                 run_duration_sum,
-                 run_duration_max,
-                 run_duration_sample,
-                 queue_duration_sum,
-                 queue_duration_max,
-                 queue_duration_sample),
-      prev(prev) {}
+    : profiling_phase(profiling_phase), death_data(death), prev(prev) {}
 
 //------------------------------------------------------------------------------
 // TaskSnapshot
 
 TaskSnapshot::TaskSnapshot() {
 }
 
 TaskSnapshot::TaskSnapshot(const BirthOnThreadSnapshot& birth,
                            const DeathDataSnapshot& death_data,
                            const std::string& death_thread_name)
@@ skipping 28 matching lines @@
 #endif
 }
 
 ProcessDataSnapshot::ProcessDataSnapshot(const ProcessDataSnapshot& other) =
     default;
 
 ProcessDataSnapshot::~ProcessDataSnapshot() {
 }
 
 }  // namespace tracked_objects