[profiler] Implement POC Sampling Heap Profiler

This implements a proof-of-concept sampling based heap profiler inspired by
tcmalloc's heap profiler [1] and Go's mprof/memprofile [2].

The basic idea is the sample allocations using a randomized Poisson process. At
any point in time we can cheaply request the set of live sample objects that
should be a representative sample of heap. Samples include stack-traces from the
allocation sites, making this an effective tool for memory leak debugging.

Unlike AllocationTracking, this is intended to be cheap and usable online in
production.

The proof-of-concept is only sampling new-space allocations at this point.
Support for sampling paged space and native allocations is anticipated in the
future.

[1] http://goog-perftools.sourceforge.net/doc/heap_profiler.html
[2] http://blog.golang.org/profiling-go-programs

Review URL: https://codereview.chromium.org/1555553002

Cr-Commit-Position: refs/heads/master@{#33448}

BUILD.gn

@@ -1185,6 +1185,8 @@ source_set("v8_base") {
     "src/profiler/profile-generator.h",
     "src/profiler/sampler.cc",
     "src/profiler/sampler.h",
+    "src/profiler/sampling-heap-profiler.cc",
+    "src/profiler/sampling-heap-profiler.h",
     "src/profiler/strings-storage.cc",
     "src/profiler/strings-storage.h",
     "src/profiler/unbound-queue-inl.h",

include/v8-profiler.h

@@ -418,6 +418,90 @@ class V8_EXPORT ActivityControl {  // NOLINT
 };
 
 
+/**
+ * AllocationProfile is a sampled profile of allocations done by the program.
+ * This is structured as a call-graph.
+ */
+class V8_EXPORT AllocationProfile {
+ public:
+  struct Allocation {
+    /**
+     * Size of the sampled allocation object.
+     */
+    size_t size;
+
+    /**
+     * The number of objects of such size that were sampled.
+     */
+    unsigned int count;
+  };
+
+  /**
+   * Represents a node in the call-graph.
+   */
+  struct Node {
+    /**
+     * Name of the function. May be empty for anonymous functions or if the
+     * script corresponding to this function has been unloaded.
+     */
+    Local<String> name;
+
+    /**
+     * Name of the script containing the function. May be empty if the script
+     * name is not available, or if the script has been unloaded.
+     */
+    Local<String> script_name;
+
+    /**
+     * id of the script where the function is located. May be equal to
+     * v8::UnboundScript::kNoScriptId in cases where the script doesn't exist.
+     */
+    int script_id;
+
+    /**
+     * Start position of the function in the script.
+     */
+    int start_position;
+
+    /**
+     * 1-indexed line number where the function starts. May be
+     * kNoLineNumberInfo if no line number information is available.
+     */
+    int line_number;
+
+    /**
+     * 1-indexed column number where the function starts. May be
+     * kNoColumnNumberInfo if no line number information is available.
+     */
+    int column_number;
+
+    /**
+     * List of callees called from this node for which we have sampled
+     * allocations. The lifetime of the children is scoped to the containing
+     * AllocationProfile.
+     */
+    std::vector<Node*> children;
+
+    /**
+     * List of self allocations done by this node in the call-graph.
+     */
+    std::vector<Allocation> allocations;
+  };
+
+  /**
+   * Returns the root node of the call-graph. The root node corresponds to an
+   * empty JS call-stack. The lifetime of the returned Node* is scoped to the
+   * containing AllocationProfile.
+   */
+  virtual Node* GetRootNode() = 0;
+
+  virtual ~AllocationProfile() {}
+
+  static const int kNoLineNumberInfo = Message::kNoLineNumberInfo;
+  static const int kNoColumnNumberInfo = Message::kNoColumnInfo;
+};
+
+
 /**
  * Interface for controlling heap profiling. Instance of the
  * profiler can be retrieved using v8::Isolate::GetHeapProfiler.
@@ -521,6 +605,49 @@ class V8_EXPORT HeapProfiler {
    */
   void StopTrackingHeapObjects();
 
+  /**
+   * Starts gathering a sampling heap profile. A sampling heap profile is
+   * similar to tcmalloc's heap profiler and Go's mprof. It samples object
+   * allocations and builds an online 'sampling' heap profile. At any point in
+   * time, this profile is expected to be a representative sample of objects
+   * currently live in the system. Each sampled allocation includes the stack
+   * trace at the time of allocation, which makes this really useful for memory
+   * leak detection.
+   *
+   * This mechanism is intended to be cheap enough that it can be used in
+   * production with minimal performance overhead.
+   *
+   * Allocations are sampled using a randomized Poisson process. On average, one
+   * allocation will be sampled every |sample_interval| bytes allocated. The
+   * |stack_depth| parameter controls the maximum number of stack frames to be
+   * captured on each allocation.
+   *
+   * NOTE: This is a proof-of-concept at this point. Right now we only sample
+   * newspace allocations. Support for paged space allocation (e.g. pre-tenured
+   * objects, large objects, code objects, etc.) and native allocations
+   * doesn't exist yet, but is anticipated in the future.
+   *
+   * Objects allocated before the sampling is started will not be included in
+   * the profile.
+   *
+   * Returns false if a sampling heap profiler is already running.
+   */
+  bool StartSamplingHeapProfiler(uint64_t sample_interval = 512 * 1024,
+                                 int stack_depth = 16);
+
+  /**
+   * Stops the sampling heap profile and discards the current profile.
+   */
+  void StopSamplingHeapProfiler();
+
+  /**
+   * Returns the sampled profile of allocations allocated (and still live) since
+   * StartSamplingHeapProfiler was called. The ownership of the pointer is
+   * transfered to the caller. Returns nullptr if sampling heap profiler is not
+   * active.
+   */
+  AllocationProfile* GetAllocationProfile();
+
   /**
    * Deletes all snapshots taken. All previously returned pointers to
    * snapshots and their contents become invalid after this call.

src/api.cc

@@ -8277,6 +8277,23 @@ SnapshotObjectId HeapProfiler::GetHeapStats(OutputStream* stream,
 }
 
 
+bool HeapProfiler::StartSamplingHeapProfiler(uint64_t sample_interval,
+                                             int stack_depth) {
+  return reinterpret_cast<i::HeapProfiler*>(this)
+      ->StartSamplingHeapProfiler(sample_interval, stack_depth);
+}
+
+
+void HeapProfiler::StopSamplingHeapProfiler() {
+  reinterpret_cast<i::HeapProfiler*>(this)->StopSamplingHeapProfiler();
+}
+
+
+AllocationProfile* HeapProfiler::GetAllocationProfile() {
+  return reinterpret_cast<i::HeapProfiler*>(this)->GetAllocationProfile();
+}
+
+
 void HeapProfiler::DeleteAllHeapSnapshots() {
   reinterpret_cast<i::HeapProfiler*>(this)->DeleteAllSnapshots();
 }

src/heap/heap.h

@@ -476,6 +476,7 @@ class GCTracer;
 class HeapObjectsFilter;
 class HeapStats;
 class HistogramTimer;
+class InlineAllocationObserver;
 class Isolate;
 class MemoryReducer;
 class ObjectStats;
@@ -2790,6 +2791,60 @@ class PathTracer : public ObjectVisitor {
   DISALLOW_IMPLICIT_CONSTRUCTORS(PathTracer);
 };
 #endif  // DEBUG
+
+// -----------------------------------------------------------------------------
+// Allows observation of inline allocation in the new space.
+class InlineAllocationObserver {
+ public:
+  explicit InlineAllocationObserver(intptr_t step_size)
+      : step_size_(step_size), bytes_to_next_step_(step_size) {
+    DCHECK(step_size >= kPointerSize);
+  }
+  virtual ~InlineAllocationObserver() {}
+
+  // Called each time the new space does an inline allocation step. This may be
+  // more frequently than the step_size we are monitoring (e.g. when there are
+  // multiple observers, or when page or space boundary is encountered.)
+  void InlineAllocationStep(int bytes_allocated, Address soon_object,
+                            size_t size) {
+    bytes_to_next_step_ -= bytes_allocated;
+    if (bytes_to_next_step_ <= 0) {
+      Step(static_cast<int>(step_size_ - bytes_to_next_step_), soon_object,
+           size);
+      step_size_ = GetNextStepSize();
+      bytes_to_next_step_ = step_size_;
+    }
+  }
+
+ protected:
+  intptr_t step_size() const { return step_size_; }
+  intptr_t bytes_to_next_step() const { return bytes_to_next_step_; }
+
+  // Pure virtual method provided by the subclasses that gets called when at
+  // least step_size bytes have been allocated. soon_object is the address just
+  // allocated (but not yet initialized.) size is the size of the object as
+  // requested (i.e. w/o the alignment fillers). Some complexities to be aware
+  // of:
+  // 1) soon_object will be nullptr in cases where we end up observing an
+  //    allocation that happens to be a filler space (e.g. page boundaries.)
+  // 2) size is the requested size at the time of allocation. Right-trimming
+  //    may change the object size dynamically.
+  // 3) soon_object may actually be the first object in an allocation-folding
+  //    group. In such a case size is the size of the group rather than the
+  //    first object.
+  virtual void Step(int bytes_allocated, Address soon_object, size_t size) = 0;
+
+  // Subclasses can override this method to make step size dynamic.
+  virtual intptr_t GetNextStepSize() { return step_size_; }
+
+  intptr_t step_size_;
+  intptr_t bytes_to_next_step_;
+
+ private:
+  friend class NewSpace;
+  DISALLOW_COPY_AND_ASSIGN(InlineAllocationObserver);
+};
+
 }  // namespace internal
 }  // namespace v8
 

src/heap/incremental-marking.h

@@ -7,6 +7,7 @@
 
 #include "src/cancelable-task.h"
 #include "src/execution.h"
+#include "src/heap/heap.h"
 #include "src/heap/incremental-marking-job.h"
 #include "src/heap/spaces.h"
 #include "src/objects.h"

src/heap/spaces.h

@@ -20,6 +20,7 @@ namespace v8 {
 namespace internal {
 
 class CompactionSpaceCollection;
+class InlineAllocationObserver;
 class Isolate;
 
 // -----------------------------------------------------------------------------
@@ -2558,54 +2559,6 @@ class NewSpacePageIterator BASE_EMBEDDED {
   NewSpacePage* last_page_;
 };
 
-// -----------------------------------------------------------------------------
-// Allows observation of inline allocation in the new space.
-class InlineAllocationObserver {
- public:
-  explicit InlineAllocationObserver(intptr_t step_size)
-      : step_size_(step_size), bytes_to_next_step_(step_size) {
-    DCHECK(step_size >= kPointerSize);
-  }
-  virtual ~InlineAllocationObserver() {}
-
- private:
-  intptr_t step_size() const { return step_size_; }
-  intptr_t bytes_to_next_step() const { return bytes_to_next_step_; }
-
-  // Pure virtual method provided by the subclasses that gets called when at
-  // least step_size bytes have been allocated. soon_object is the address just
-  // allocated (but not yet initialized.) size is the size of the object as
-  // requested (i.e. w/o the alignment fillers). Some complexities to be aware
-  // of:
-  // 1) soon_object will be nullptr in cases where we end up observing an
-  //    allocation that happens to be a filler space (e.g. page boundaries.)
-  // 2) size is the requested size at the time of allocation. Right-trimming
-  //    may change the object size dynamically.
-  // 3) soon_object may actually be the first object in an allocation-folding
-  //    group. In such a case size is the size of the group rather than the
-  //    first object.
-  virtual void Step(int bytes_allocated, Address soon_object, size_t size) = 0;
-
-  // Called each time the new space does an inline allocation step. This may be
-  // more frequently than the step_size we are monitoring (e.g. when there are
-  // multiple observers, or when page or space boundary is encountered.)
-  void InlineAllocationStep(int bytes_allocated, Address soon_object,
-                            size_t size) {
-    bytes_to_next_step_ -= bytes_allocated;
-    if (bytes_to_next_step_ <= 0) {
-      Step(static_cast<int>(step_size_ - bytes_to_next_step_), soon_object,
-           size);
-      bytes_to_next_step_ = step_size_;
-    }
-  }
-
-  intptr_t step_size_;
-  intptr_t bytes_to_next_step_;
-
-  friend class NewSpace;
-
-  DISALLOW_COPY_AND_ASSIGN(InlineAllocationObserver);
-};
 
 // -----------------------------------------------------------------------------
 // The young generation space.

src/isolate.cc

@@ -1939,13 +1939,14 @@ void Isolate::Deinit() {
   delete basic_block_profiler_;
   basic_block_profiler_ = NULL;
 
+  delete heap_profiler_;
+  heap_profiler_ = NULL;
+
   heap_.TearDown();
   logger_->TearDown();
 
   cancelable_task_manager()->CancelAndWait();
 
-  delete heap_profiler_;
-  heap_profiler_ = NULL;
   delete cpu_profiler_;
   cpu_profiler_ = NULL;
 

src/profiler/heap-profiler.cc

@@ -8,6 +8,7 @@
 #include "src/debug/debug.h"
 #include "src/profiler/allocation-tracker.h"
 #include "src/profiler/heap-snapshot-generator-inl.h"
+#include "src/profiler/sampling-heap-profiler.h"
 
 namespace v8 {
 namespace internal {
@@ -84,6 +85,31 @@ HeapSnapshot* HeapProfiler::TakeSnapshot(
 }
 
 
+bool HeapProfiler::StartSamplingHeapProfiler(uint64_t sample_interval,
+                                             int stack_depth) {
+  if (sampling_heap_profiler_.get()) {
+    return false;
+  }
+  sampling_heap_profiler_.Reset(new SamplingHeapProfiler(
+      heap(), names_.get(), sample_interval, stack_depth));
+  return true;
+}
+
+
+void HeapProfiler::StopSamplingHeapProfiler() {
+  sampling_heap_profiler_.Reset(nullptr);
+}
+
+
+v8::AllocationProfile* HeapProfiler::GetAllocationProfile() {
+  if (sampling_heap_profiler_.get()) {
+    return sampling_heap_profiler_->GetAllocationProfile();
+  } else {
+    return nullptr;
+  }
+}
+
+
 void HeapProfiler::StartHeapObjectsTracking(bool track_allocations) {
   ids_->UpdateHeapObjectsMap();
   is_tracking_object_moves_ = true;

src/profiler/heap-profiler.h

@@ -16,6 +16,7 @@ namespace internal {
 class AllocationTracker;
 class HeapObjectsMap;
 class HeapSnapshot;
+class SamplingHeapProfiler;
 class StringsStorage;
 
 class HeapProfiler {
@@ -29,6 +30,10 @@ class HeapProfiler {
       v8::ActivityControl* control,
       v8::HeapProfiler::ObjectNameResolver* resolver);
 
+  bool StartSamplingHeapProfiler(uint64_t sample_interval, int stack_depth);
+  void StopSamplingHeapProfiler();
+  AllocationProfile* GetAllocationProfile();
+
   void StartHeapObjectsTracking(bool track_allocations);
   void StopHeapObjectsTracking();
   AllocationTracker* allocation_tracker() const {
@@ -79,6 +84,7 @@ class HeapProfiler {
   base::SmartPointer<AllocationTracker> allocation_tracker_;
   bool is_tracking_object_moves_;
   base::Mutex profiler_mutex_;
+  base::SmartPointer<SamplingHeapProfiler> sampling_heap_profiler_;
 };
 
 }  // namespace internal

src/profiler/sampling-heap-profiler.cc

@@ -0,0 +1,250 @@
+// Copyright 2015 the V8 project 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 "src/profiler/sampling-heap-profiler.h"
+
+#include <stdint.h>
+#include <memory>
+#include "src/api.h"
+#include "src/base/utils/random-number-generator.h"
+#include "src/frames-inl.h"
+#include "src/heap/heap.h"
+#include "src/isolate.h"
+#include "src/profiler/strings-storage.h"
+
+namespace v8 {
+namespace internal {
+
+SamplingHeapProfiler::SamplingHeapProfiler(Heap* heap, StringsStorage* names,
+                                           uint64_t rate, int stack_depth)
+    : InlineAllocationObserver(GetNextSampleInterval(
+          heap->isolate()->random_number_generator(), rate)),
+      isolate_(heap->isolate()),
+      heap_(heap),
+      random_(isolate_->random_number_generator()),
+      names_(names),
+      samples_(),
+      rate_(rate),
+      stack_depth_(stack_depth) {
+  heap->new_space()->AddInlineAllocationObserver(this);
+}
+
+
+SamplingHeapProfiler::~SamplingHeapProfiler() {
+  heap_->new_space()->RemoveInlineAllocationObserver(this);
+
+  // Clear samples and drop all the weak references we are keeping.
+  std::set<SampledAllocation*>::iterator it;
+  for (it = samples_.begin(); it != samples_.end(); ++it) {
+    delete *it;
+  }
+  std::set<SampledAllocation*> empty;
+  samples_.swap(empty);
+}
+
+void SamplingHeapProfiler::Step(int bytes_allocated, Address soon_object,
+                                size_t size) {
+  DCHECK(heap_->gc_state() == Heap::NOT_IN_GC);
+  DCHECK(soon_object);
+  SampleObject(soon_object, size);
+}
+
+
+void SamplingHeapProfiler::SampleObject(Address soon_object, size_t size) {
+  DisallowHeapAllocation no_allocation;
+
+  HandleScope scope(isolate_);
+  HeapObject* heap_object = HeapObject::FromAddress(soon_object);
+  Handle<Object> obj(heap_object, isolate_);
+
+  // Mark the new block as FreeSpace to make sure the heap is iterable while we
+  // are taking the sample.
+  heap()->CreateFillerObjectAt(soon_object, static_cast<int>(size));
+
+  Local<v8::Value> loc = v8::Utils::ToLocal(obj);
+
+  SampledAllocation* sample =
+      new SampledAllocation(this, isolate_, loc, size, stack_depth_);
+  samples_.insert(sample);
+}
+
+
+// We sample with a Poisson process, with constant average sampling interval.
+// This follows the exponential probability distribution with parameter
+// λ = 1/rate where rate is the average number of bytes between samples.
+//
+// Let u be a uniformly distributed random number between 0 and 1, then
+// next_sample = (- ln u) / λ
+intptr_t SamplingHeapProfiler::GetNextSampleInterval(
+    base::RandomNumberGenerator* random, uint64_t rate) {
+  double u = random->NextDouble();
+  double next = (-std::log(u)) * rate;
+  return next < kPointerSize
+             ? kPointerSize
+             : (next > INT_MAX ? INT_MAX : static_cast<intptr_t>(next));
+}
+
+
+void SamplingHeapProfiler::SampledAllocation::OnWeakCallback(
+    const WeakCallbackInfo<SampledAllocation>& data) {
+  SampledAllocation* sample = data.GetParameter();
+  sample->sampling_heap_profiler_->samples_.erase(sample);
+  delete sample;
+}
+
+
+SamplingHeapProfiler::FunctionInfo::FunctionInfo(SharedFunctionInfo* shared,
+                                                 StringsStorage* names)
+    : name_(names->GetFunctionName(shared->DebugName())),
+      script_name_(""),
+      script_id_(v8::UnboundScript::kNoScriptId),
+      start_position_(shared->start_position()) {
+  if (shared->script()->IsScript()) {
+    Script* script = Script::cast(shared->script());
+    script_id_ = script->id();
+    if (script->name()->IsName()) {
+      Name* name = Name::cast(script->name());
+      script_name_ = names->GetName(name);
+    }
+  }
+}
+
+
+SamplingHeapProfiler::SampledAllocation::SampledAllocation(
+    SamplingHeapProfiler* sampling_heap_profiler, Isolate* isolate,
+    Local<Value> local, size_t size, int max_frames)
+    : sampling_heap_profiler_(sampling_heap_profiler),
+      global_(reinterpret_cast<v8::Isolate*>(isolate), local),
+      size_(size) {
+  global_.SetWeak(this, OnWeakCallback, WeakCallbackType::kParameter);
+
+  StackTraceFrameIterator it(isolate);
+  int frames_captured = 0;
+  while (!it.done() && frames_captured < max_frames) {
+    JavaScriptFrame* frame = it.frame();
+    SharedFunctionInfo* shared = frame->function()->shared();
+    stack_.push_back(new FunctionInfo(shared, sampling_heap_profiler->names()));
+
+    frames_captured++;
+    it.Advance();
+  }
+
+  if (frames_captured == 0) {
+    const char* name = nullptr;
+    switch (isolate->current_vm_state()) {
+      case GC:
+        name = "(GC)";
+        break;
+      case COMPILER:
+        name = "(COMPILER)";
+        break;
+      case OTHER:
+        name = "(V8 API)";
+        break;
+      case EXTERNAL:
+        name = "(EXTERNAL)";
+        break;
+      case IDLE:
+        name = "(IDLE)";
+        break;
+      case JS:
+        name = "(JS)";
+        break;
+    }
+    stack_.push_back(new FunctionInfo(name));
+  }
+}
+
+
+SamplingHeapProfiler::Node* SamplingHeapProfiler::AllocateNode(
+    AllocationProfile* profile, const std::map<int, Script*>& scripts,
+    FunctionInfo* function_info) {
+  DCHECK(function_info->get_name());
+  DCHECK(function_info->get_script_name());
+
+  int line = v8::AllocationProfile::kNoLineNumberInfo;
+  int column = v8::AllocationProfile::kNoColumnNumberInfo;
+
+  if (function_info->get_script_id() != v8::UnboundScript::kNoScriptId) {
+    // Cannot use std::map<T>::at because it is not available on android.
+    auto non_const_scripts = const_cast<std::map<int, Script*>&>(scripts);
+    Handle<Script> script(non_const_scripts[function_info->get_script_id()]);
+
+    line =
+        1 + Script::GetLineNumber(script, function_info->get_start_position());
+    column = 1 + Script::GetColumnNumber(script,
+                                         function_info->get_start_position());
+  }
+
+  profile->nodes().push_back(
+      Node({ToApiHandle<v8::String>(isolate_->factory()->InternalizeUtf8String(
+                function_info->get_name())),
+            ToApiHandle<v8::String>(isolate_->factory()->InternalizeUtf8String(
+                function_info->get_script_name())),
+            function_info->get_script_id(), function_info->get_start_position(),
+            line, column, std::vector<Node*>(),
+            std::vector<v8::AllocationProfile::Allocation>()}));
+
+  return &profile->nodes().back();
+}
+
+
+SamplingHeapProfiler::Node* SamplingHeapProfiler::FindOrAddChildNode(
+    AllocationProfile* profile, const std::map<int, Script*>& scripts,
+    Node* parent, FunctionInfo* function_info) {
+  for (Node* child : parent->children) {
+    if (child->script_id == function_info->get_script_id() &&
+        child->start_position == function_info->get_start_position())
+      return child;
+  }
+  Node* child = AllocateNode(profile, scripts, function_info);
+  parent->children.push_back(child);
+  return child;
+}
+
+
+SamplingHeapProfiler::Node* SamplingHeapProfiler::AddStack(
+    AllocationProfile* profile, const std::map<int, Script*>& scripts,
+    const std::vector<FunctionInfo*>& stack) {
+  Node* node = profile->GetRootNode();
+
+  // We need to process the stack in reverse order as the top of the stack is
+  // the first element in the list.
+  for (auto it = stack.rbegin(); it != stack.rend(); ++it) {
+    FunctionInfo* function_info = *it;
+    node = FindOrAddChildNode(profile, scripts, node, function_info);
+  }
+  return node;
+}
+
+
+v8::AllocationProfile* SamplingHeapProfiler::GetAllocationProfile() {
+  // To resolve positions to line/column numbers, we will need to look up
+  // scripts. Build a map to allow fast mapping from script id to script.
+  std::map<int, Script*> scripts;
+  {
+    Script::Iterator iterator(isolate_);
+    Script* script;
+    while ((script = iterator.Next())) {
+      scripts[script->id()] = script;
+    }
+  }
+
+  auto profile = new v8::internal::AllocationProfile();
+
+  // Create the root node.
+  FunctionInfo function_info("(root)");
+  AllocateNode(profile, scripts, &function_info);
+
+  for (SampledAllocation* allocation : samples_) {
+    Node* node = AddStack(profile, scripts, allocation->get_stack());
+    node->allocations.push_back({allocation->get_size(), 1});
+  }
+
+  return profile;
+}
+
+
+}  // namespace internal
+}  // namespace v8

src/profiler/sampling-heap-profiler.h

@@ -0,0 +1,135 @@
+// Copyright 2015 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#ifndef V8_PROFILER_SAMPLING_HEAP_PROFILER_H_
+#define V8_PROFILER_SAMPLING_HEAP_PROFILER_H_
+
+#include <deque>
+#include <map>
+#include <set>
+#include "include/v8-profiler.h"
+#include "src/heap/heap.h"
+#include "src/profiler/strings-storage.h"
+
+namespace v8 {
+
+namespace base {
+class RandomNumberGenerator;
+}
+
+namespace internal {
+
+
+class AllocationProfile : public v8::AllocationProfile {
+ public:
+  AllocationProfile() : nodes_() {}
+
+  Node* GetRootNode() override {
+    return nodes_.size() == 0 ? nullptr : &nodes_.front();
+  }
+
+  std::deque<Node>& nodes() { return nodes_; }
+
+ private:
+  std::deque<Node> nodes_;
+
+  DISALLOW_COPY_AND_ASSIGN(AllocationProfile);
+};
+
+class SamplingHeapProfiler : public InlineAllocationObserver {
+ public:
+  SamplingHeapProfiler(Heap* heap, StringsStorage* names, uint64_t rate,
+                       int stack_depth);
+  ~SamplingHeapProfiler();
+
+  v8::AllocationProfile* GetAllocationProfile();
+
+  void Step(int bytes_allocated, Address soon_object, size_t size) override;
+  intptr_t GetNextStepSize() override {
+    return GetNextSampleInterval(random_, rate_);
+  }
+
+  StringsStorage* names() const { return names_; }
+
+  class FunctionInfo {
+   public:
+    FunctionInfo(SharedFunctionInfo* shared, StringsStorage* names);
+    explicit FunctionInfo(const char* name)
+        : name_(name),
+          script_name_(""),
+          script_id_(v8::UnboundScript::kNoScriptId),
+          start_position_(0) {}
+
+    const char* get_name() const { return name_; }
+    const char* get_script_name() const { return script_name_; }
+    int get_script_id() const { return script_id_; }
+    int get_start_position() const { return start_position_; }
+
+   private:
+    const char* const name_;
+    const char* script_name_;
+    int script_id_;
+    const int start_position_;
+  };
+
+  class SampledAllocation {
+   public:
+    SampledAllocation(SamplingHeapProfiler* sampling_heap_profiler,
+                      Isolate* isolate, Local<Value> local, size_t size,
+                      int max_frames);
+    ~SampledAllocation() {
+      for (auto info : stack_) {
+        delete info;
+      }
+      global_.Reset();  // drop the reference.
+    }
+    size_t get_size() const { return size_; }
+    const std::vector<FunctionInfo*>& get_stack() const { return stack_; }
+
+   private:
+    static void OnWeakCallback(const WeakCallbackInfo<SampledAllocation>& data);
+
+    SamplingHeapProfiler* const sampling_heap_profiler_;
+    Global<Value> global_;
+    std::vector<FunctionInfo*> stack_;
+    const size_t size_;
+
+    DISALLOW_COPY_AND_ASSIGN(SampledAllocation);
+  };
+
+ private:
+  using Node = v8::AllocationProfile::Node;
+
+  Heap* heap() const { return heap_; }
+
+  void SampleObject(Address soon_object, size_t size);
+
+  static intptr_t GetNextSampleInterval(base::RandomNumberGenerator* random,
+                                        uint64_t rate);
+
+  // Methods that construct v8::AllocationProfile.
+  Node* AddStack(AllocationProfile* profile,
+                 const std::map<int, Script*>& scripts,
+                 const std::vector<FunctionInfo*>& stack);
+  Node* FindOrAddChildNode(AllocationProfile* profile,
+                           const std::map<int, Script*>& scripts, Node* parent,
+                           FunctionInfo* function_info);
+  Node* AllocateNode(AllocationProfile* profile,
+                     const std::map<int, Script*>& scripts,
+                     FunctionInfo* function_info);
+
+  Isolate* const isolate_;
+  Heap* const heap_;
+  base::RandomNumberGenerator* const random_;
+  StringsStorage* const names_;
+  std::set<SampledAllocation*> samples_;
+  const uint64_t rate_;
+  const int stack_depth_;
+};
+
+
+}  // namespace internal
+}  // namespace v8
+
+#endif  // V8_PROFILER_SAMPLING_HEAP_PROFILER_H_

test/cctest/test-heap-profiler.cc

@@ -2852,3 +2852,148 @@ TEST(AddressToTraceMap) {
   CHECK_EQ(0u, map.size());
   CHECK_EQ(0u, map.GetTraceNodeId(ToAddress(0x400)));
 }
+
+
+static const v8::AllocationProfile::Node* FindAllocationProfileNode(
+    v8::AllocationProfile& profile, const Vector<const char*>& names) {
+  v8::AllocationProfile::Node* node = profile.GetRootNode();
+  for (int i = 0; node != nullptr && i < names.length(); ++i) {
+    const char* name = names[i];
+    auto children = node->children;
+    node = nullptr;
+    for (v8::AllocationProfile::Node* child : children) {
+      v8::String::Utf8Value child_name(child->name);
+      if (strcmp(*child_name, name) == 0) {
+        node = child;
+        break;
+      }
+    }
+  }
+  return node;
+}
+
+
+TEST(SamplingHeapProfiler) {
+  v8::HandleScope scope(v8::Isolate::GetCurrent());
+  LocalContext env;
+  v8::HeapProfiler* heap_profiler = env->GetIsolate()->GetHeapProfiler();
+
+  // Turn off always_opt. Inlining can cause stack traces to be shorter than
+  // what we expect in this test.
+  v8::internal::FLAG_always_opt = false;
+
+  const char* script_source =
+      "var A = [];\n"
+      "function bar(size) { return new Array(size); }\n"
+      "var foo = function() {\n"
+      "  for (var i = 0; i < 1024; ++i) {\n"
+      "    A[i] = bar(1024);\n"
+      "  }\n"
+      "}\n"
+      "foo();";
+
+  // Sample should be empty if requested before sampling has started.
+  {
+    v8::AllocationProfile* profile = heap_profiler->GetAllocationProfile();
+    CHECK(profile == nullptr);
+  }
+
+  int count_512kb = 0;
+  {
+    heap_profiler->StartSamplingHeapProfiler(512 * 1024);
+    CompileRun(script_source);
+
+    v8::base::SmartPointer<v8::AllocationProfile> profile(
+        heap_profiler->GetAllocationProfile());
+    CHECK(!profile.is_empty());
+
+    const char* names[] = {"", "foo", "bar"};
+    auto node_bar = FindAllocationProfileNode(
+        *profile, Vector<const char*>(names, arraysize(names)));
+    CHECK(node_bar);
+
+    // Count the number of allocations we sampled from bar.
+    for (auto allocation : node_bar->allocations) {
+      count_512kb += allocation.count;
+    }
+
+    heap_profiler->StopSamplingHeapProfiler();
+  }
+
+  // Samples should get cleared once sampling is stopped.
+  {
+    v8::AllocationProfile* profile = heap_profiler->GetAllocationProfile();
+    CHECK(profile == nullptr);
+  }
+
+  // Sampling at a higher rate should give us more sampled objects.
+  {
+    heap_profiler->StartSamplingHeapProfiler(32 * 1024);
+    CompileRun(script_source);
+
+    v8::base::SmartPointer<v8::AllocationProfile> profile(
+        heap_profiler->GetAllocationProfile());
+    CHECK(!profile.is_empty());
+
+    const char* names[] = {"", "foo", "bar"};
+    auto node_bar = FindAllocationProfileNode(
+        *profile, Vector<const char*>(names, arraysize(names)));
+    CHECK(node_bar);
+
+    // Count the number of allocations we sampled from bar.
+    int count_32kb = 0;
+    for (auto allocation : node_bar->allocations) {
+      count_32kb += allocation.count;
+    }
+
+    // We should have roughly 16x as many sampled allocations. However, since
+    // sampling is a randomized process, we use a weaker test.
+    CHECK_GT(count_32kb, count_512kb);
+
+    heap_profiler->StopSamplingHeapProfiler();
+  }
+
+  // A more complicated test cases with deeper call graph and dynamically
+  // generated function names.
+  {
+    heap_profiler->StartSamplingHeapProfiler(128);
+    CompileRun(record_trace_tree_source);
+
+    v8::base::SmartPointer<v8::AllocationProfile> profile(
+        heap_profiler->GetAllocationProfile());
+    CHECK(!profile.is_empty());
+
+    const char* names1[] = {"", "start", "f_0_0", "f_0_1", "f_0_2"};
+    auto node1 = FindAllocationProfileNode(
+        *profile, Vector<const char*>(names1, arraysize(names1)));
+    CHECK(node1);
+
+    const char* names2[] = {"", "generateFunctions"};
+    auto node2 = FindAllocationProfileNode(
+        *profile, Vector<const char*>(names2, arraysize(names2)));
+    CHECK(node2);
+
+    heap_profiler->StopSamplingHeapProfiler();
+  }
+}
+
+
+TEST(SamplingHeapProfilerApiAllocation) {
+  v8::HandleScope scope(v8::Isolate::GetCurrent());
+  LocalContext env;
+  v8::HeapProfiler* heap_profiler = env->GetIsolate()->GetHeapProfiler();
+
+  heap_profiler->StartSamplingHeapProfiler(256);
+
+  for (int i = 0; i < 8 * 1024; ++i) v8::Object::New(env->GetIsolate());
+
+  v8::base::SmartPointer<v8::AllocationProfile> profile(
+      heap_profiler->GetAllocationProfile());
+  CHECK(!profile.is_empty());
+  const char* names[] = {"(V8 API)"};
+  auto node = FindAllocationProfileNode(
+      *profile, Vector<const char*>(names, arraysize(names)));
+  CHECK(node);
+
+  heap_profiler->StopSamplingHeapProfiler();
+}

tools/gyp/v8.gyp

@@ -955,6 +955,8 @@
         '../../src/profiler/profile-generator.h',
         '../../src/profiler/sampler.cc',
         '../../src/profiler/sampler.h',
+        '../../src/profiler/sampling-heap-profiler.cc',
+        '../../src/profiler/sampling-heap-profiler.h',
         '../../src/profiler/strings-storage.cc',
         '../../src/profiler/strings-storage.h',
         '../../src/profiler/unbound-queue-inl.h',