Oilpan: Reduce sizeof(Persistent) to 16 byte

Source/platform/heap/HeapTest.cpp

Index: Source/platform/heap/HeapTest.cpp
diff --git a/Source/platform/heap/HeapTest.cpp b/Source/platform/heap/HeapTest.cpp
index f5e2d275da575e08cfd551cc608fb5691fc76996..2cf6309ed614ca0c27645efba68545bc69854f56 100644
--- a/Source/platform/heap/HeapTest.cpp
+++ b/Source/platform/heap/HeapTest.cpp
@@ -6269,4 +6269,30 @@ TEST(HeapTest, MixinConstructionNoGC)
     EXPECT_GT(mixin->traceCount(), 0);
 }
 
+TEST(HeapTest, PersistentPerfTest)

[haraken, 2015/06/29 11:48:52]

This is a micro-benchmark to measure the performance of this CL (I'm not
planning to commit the benchmark). It seems that this CL regresses the
performance by 4%. (I'd say that a 4% regression in this kind of micro benchmark
is OK, but I was expecting that this CL would improve performance.)

+{
+    const int count = 10000;
+    int sum = 0;
+    Persistent<IntWrapper>** handle = static_cast<Persistent<IntWrapper>**>(malloc(sizeof(Persistent<IntWrapper>*) * count));
+    double begin = WTF::currentTimeMS();
+    for (int i = 0; i < count; i++) {
+        handle[i] = new Persistent<IntWrapper>(IntWrapper::create(i));
+    }
+    for (int delta = 2; delta < 100; delta++) {
+        for (int i = 0; i < count; i+=delta) {
+            sum += handle[i]->get()->value();
+            delete handle[i];
+        }
+        for (int i = 0; i < count; i+=delta) {
+            handle[i] = new Persistent<IntWrapper>(IntWrapper::create(i));
+        }
+    }
+    for (int i = 0; i < count; i++) {
+        sum += handle[i]->get()->value();
+        delete handle[i];
+    }
+    double end =WTF::currentTimeMS();
+    fprintf(stderr, "%.2lf %d\n", end - begin, sum);
+}
+
 } // namespace blink