[Oilpan] Unify memory usage reporters of Oilpan

third_party/WebKit/Source/platform/heap/ThreadState.h

Index: third_party/WebKit/Source/platform/heap/ThreadState.h
diff --git a/third_party/WebKit/Source/platform/heap/ThreadState.h b/third_party/WebKit/Source/platform/heap/ThreadState.h
index d4ad238d2bfdb2a8a0e6bfa325e4ca729b3652a7..e3086377ff457d43f51bbd0de9222ae3fb3a25c7 100644
--- a/third_party/WebKit/Source/platform/heap/ThreadState.h
+++ b/third_party/WebKit/Source/platform/heap/ThreadState.h
@@ -520,6 +520,17 @@ public:
     void leaveStaticReferenceRegistrationDisabledScope();
 #endif
 
+    void reportMemoryToV8();
+
+    size_t heapSize();

[sof, 2016/02/15 20:47:36]

This declaration appears to be without a definition.

[peria, 2016/02/16 03:20:20]

removed.

+    size_t allocatedObjectSize() { return m_allocatedObjectSize; }

[sof, 2016/02/15 20:47:36]

These two can be const methods.

And, if concurrent R/W is an issue for these, then you need to symmetrically use
acquireLoad() if you're using atomicAdd() for the other half.

However, I don't understand how such a race can happen -- the thread associated
with the ThreadState is either parked at a safepoint or running the GC. And we
don't support cross-thread heap updates (e.g., in-place expand/shrink of objects
by another thread), so thread contention on these new fields doesn't appear
possible.

[peria, 2016/02/16 03:20:20]

Yes and they should be, but I removed these accessors.
Agree. We don't need to care for thread race.

[sof, 2016/02/16 06:32:56]

If so, why is the latest patchset still using atomicAdd() et al?

+    size_t markedObjectSize() { return m_markedObjectSize; }
+
+    void resetHeapCounters();
+    void increaseAllocatedObjectSize(size_t);
+    void decreaseAllocatedObjectSize(size_t);
+    void increaseMarkedObjectSize(size_t);
+
 private:
     enum SnapshotType {
         HeapSnapshot,
@@ -678,6 +689,11 @@ private:
     static const int likelyToBePromptlyFreedArraySize = (1 << 8);
     static const int likelyToBePromptlyFreedArrayMask = likelyToBePromptlyFreedArraySize - 1;
     OwnPtr<int[]> m_likelyToBePromptlyFreed;
+
+    // Stats for heap memory of this thread.
+    size_t m_allocatedObjectSize;
+    size_t m_markedObjectSize;
+    size_t m_reportedMemoryToV8;
 };
 
 template<ThreadAffinity affinity> class ThreadStateFor;