Revert of [heap] Invoke incremental marking step before allocation. (patchset...

Revert of [heap] Invoke incremental marking step before allocation. (patchset #1 id:1 of https://codereview.chromium.org/2464393002/ )

Reason for revert:
Performance regression on Octane and V8 runtime stats.

Original issue's description:
> [heap] Invoke incremental marking step before allocation.
>
> This ensures that the newly allocated object immediatly precedes the
> linear allocation area, which is needed for allocation folding.
>
> For more info see:
> https://bugs.chromium.org/p/chromium/issues/detail?id=659165#c13
>
> BUG=chromium:659165

TBR=hpayer@chromium.org,mlippautz@chromium.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=chromium:659165

Review-Url: https://codereview.chromium.org/2472043002
Cr-Commit-Position: refs/heads/master@{#40713}

src/heap/incremental-marking.cc

@@ -32,7 +32,9 @@ IncrementalMarking::IncrementalMarking(Heap* heap)
       was_activated_(false),
       black_allocation_(false),
       finalize_marking_completed_(false),
-      request_type_(NONE) {}
+      request_type_(NONE),
+      new_generation_observer_(*this, kAllocatedThreshold),
+      old_generation_observer_(*this, kAllocatedThreshold) {}
 
 bool IncrementalMarking::BaseRecordWrite(HeapObject* obj, Object* value) {
   HeapObject* value_heap_obj = HeapObject::cast(value);
@@ -487,6 +489,16 @@ void IncrementalMarking::Start(GarbageCollectionReason gc_reason) {
     state_ = SWEEPING;
   }
 
+  SpaceIterator it(heap_);
+  while (it.has_next()) {
+    Space* space = it.next();
+    if (space == heap_->new_space()) {
+      space->AddAllocationObserver(&new_generation_observer_);
+    } else {
+      space->AddAllocationObserver(&old_generation_observer_);
+    }
+  }
+
   incremental_marking_job()->Start(heap_);
 }
 
@@ -945,6 +957,16 @@ void IncrementalMarking::Stop() {
         Max(0, old_generation_size_mb - old_generation_limit_mb));
   }
 
+  SpaceIterator it(heap_);
+  while (it.has_next()) {
+    Space* space = it.next();
+    if (space == heap_->new_space()) {
+      space->RemoveAllocationObserver(&new_generation_observer_);
+    } else {
+      space->RemoveAllocationObserver(&old_generation_observer_);
+    }
+  }
+
   IncrementalMarking::set_should_hurry(false);
   if (IsMarking()) {
     PatchIncrementalMarkingRecordWriteStubs(heap_,
@@ -1063,33 +1085,30 @@ void IncrementalMarking::AdvanceIncrementalMarkingOnAllocation() {
     return;
   }
 
-  size_t bytes_to_process = StepSizeToKeepUpWithAllocations();
-
-  if (bytes_to_process < IncrementalMarking::kAllocatedThreshold) {
-    return;
-  }
-
-  bytes_to_process += StepSizeToMakeProgress();
-
-  // The first step after Scavenge will see many allocated bytes.
-  // Cap the step size to distribute the marking work more uniformly.
-  size_t max_step_size = GCIdleTimeHandler::EstimateMarkingStepSize(
-      kMaxStepSizeInMs,
-      heap()->tracer()->IncrementalMarkingSpeedInBytesPerMillisecond());
-  bytes_to_process = Min(bytes_to_process, max_step_size);
-
-  size_t bytes_processed = 0;
-  if (bytes_marked_ahead_of_schedule_ >= bytes_to_process) {
-    // Steps performed in tasks have put us ahead of schedule.
-    // We skip processing of marking dequeue here and thus
-    // shift marking time from inside V8 to standalone tasks.
-    bytes_marked_ahead_of_schedule_ -= bytes_to_process;
-    bytes_processed = bytes_to_process;
-  } else {
-    bytes_processed = Step(bytes_to_process, GC_VIA_STACK_GUARD,
-                           FORCE_COMPLETION, StepOrigin::kV8);
+  size_t bytes_to_process =
+      StepSizeToKeepUpWithAllocations() + StepSizeToMakeProgress();
+
+  if (bytes_to_process >= IncrementalMarking::kAllocatedThreshold) {
+    // The first step after Scavenge will see many allocated bytes.
+    // Cap the step size to distribute the marking work more uniformly.
+    size_t max_step_size = GCIdleTimeHandler::EstimateMarkingStepSize(
+        kMaxStepSizeInMs,
+        heap()->tracer()->IncrementalMarkingSpeedInBytesPerMillisecond());
+    bytes_to_process = Min(bytes_to_process, max_step_size);
+
+    size_t bytes_processed = 0;
+    if (bytes_marked_ahead_of_schedule_ >= bytes_to_process) {
+      // Steps performed in tasks have put us ahead of schedule.
+      // We skip processing of marking dequeue here and thus
+      // shift marking time from inside V8 to standalone tasks.
+      bytes_marked_ahead_of_schedule_ -= bytes_to_process;
+      bytes_processed = bytes_to_process;
+    } else {
+      bytes_processed = Step(bytes_to_process, GC_VIA_STACK_GUARD,
+                             FORCE_COMPLETION, StepOrigin::kV8);
+    }
+    bytes_allocated_ -= Min(bytes_allocated_, bytes_processed);
   }
-  bytes_allocated_ -= Min(bytes_allocated_, bytes_processed);
 }
 
 size_t IncrementalMarking::Step(size_t bytes_to_process,

src/heap/incremental-marking.h

@@ -99,7 +99,6 @@ class IncrementalMarking {
                                    CompletionAction completion_action,
                                    ForceCompletionAction force_completion,
                                    StepOrigin step_origin);
-  void AdvanceIncrementalMarkingOnAllocation();
 
   // It's hard to know how much work the incremental marker should do to make
   // progress in the face of the mutator creating new work for it.  We start
@@ -219,6 +218,20 @@ class IncrementalMarking {
   void AbortBlackAllocation();
 
  private:
+  class Observer : public AllocationObserver {
+   public:
+    Observer(IncrementalMarking& incremental_marking, intptr_t step_size)
+        : AllocationObserver(step_size),
+          incremental_marking_(incremental_marking) {}
+
+    void Step(int bytes_allocated, Address, size_t) override {
+      incremental_marking_.AdvanceIncrementalMarkingOnAllocation();
+    }
+
+   private:
+    IncrementalMarking& incremental_marking_;
+  };
+
   int64_t SpaceLeftInOldSpace();
 
   void StartMarking();
@@ -256,6 +269,8 @@ class IncrementalMarking {
 
   void IncrementIdleMarkingDelayCounter();
 
+  void AdvanceIncrementalMarkingOnAllocation();
+
   size_t StepSizeToKeepUpWithAllocations();
   size_t StepSizeToMakeProgress();
 
@@ -282,6 +297,8 @@ class IncrementalMarking {
   GCRequestType request_type_;
 
   IncrementalMarkingJob incremental_marking_job_;
+  Observer new_generation_observer_;
+  Observer old_generation_observer_;
 
   DISALLOW_IMPLICIT_CONSTRUCTORS(IncrementalMarking);
 };

src/heap/spaces.cc

@@ -2582,15 +2582,6 @@ HeapObject* FreeList::Allocate(size_t size_in_bytes) {
 
   owner_->heap()->StartIncrementalMarkingIfAllocationLimitIsReached(
       Heap::kNoGCFlags, kNoGCCallbackFlags);
-  // We cannot place incremental marking step in an AllocationObserver because
-  // 1) incremental marking step can change linear allocation area.
-  // 2) allocation observers are called after allocation.
-  // 3) allocation folding assumes that the newly allocated object immediately
-  //    precedes the linear allocation area.
-  // See crbug.com/659165.
-  owner_->heap()
-      ->incremental_marking()
-      ->AdvanceIncrementalMarkingOnAllocation();
 
   size_t new_node_size = 0;
   FreeSpace* new_node = FindNodeFor(size_in_bytes, &new_node_size);
@@ -3020,8 +3011,6 @@ AllocationResult LargeObjectSpace::AllocateRaw(int object_size,
 
   heap()->StartIncrementalMarkingIfAllocationLimitIsReached(Heap::kNoGCFlags,
                                                             kNoGCCallbackFlags);
-  heap()->incremental_marking()->AdvanceIncrementalMarkingOnAllocation();
-
   AllocationStep(object->address(), object_size);
 
   if (heap()->incremental_marking()->black_allocation()) {