Oilpan: mark ThreadAffinity as MainThreadOnly for Node or CSSValue inheriting classes.

Source/heap/ThreadState.h

 /*
  * Copyright (C) 2013 Google Inc. All rights reserved.
  *
  * Redistribution and use in source and binary forms, with or without
  * modification, are permitted provided that the following conditions are
  * met:
  *
  *     * Redistributions of source code must retain the above copyright
  * notice, this list of conditions and the following disclaimer.
  *     * Redistributions in binary form must reproduce the above
@@ skipping 53 matching lines @@
 //
 // FIXME: We should evaluate the performance gain. Having
 // ThreadAffinity is complicating the implementation and we should get
 // rid of it if it is fast enough to go through thread-local storage
 // always.
 enum ThreadAffinity {
     AnyThread,
     MainThreadOnly,
 };
 
-// By default all types are considered to be used on the main thread only.
+class Node;
+class CSSValue;
+
+template<typename T>
+class DerivesNodeOrCSSValue {
+private:
+    typedef char YesType;
+    struct NoType {
+        char padding[8];
+    };
+
+    static YesType check(Node*);
+    static YesType check(CSSValue*);
+    static NoType check(void*);
+
+public:
+    static const bool value = sizeof(check(static_cast<T*>(0))) == sizeof(YesType);
+};
+
+template<typename T, bool derivesNodeOrCSSValue = DerivesNodeOrCSSValue<T>::value > class DefaultThreadingTrait;

[zerny-chromium, 2014/02/21 07:04:54]

I think this can use:

  derivesNodeOrCSSValue = IsSubclass<T, Node>::value || IsSubClass<T,
CSSValue>::value

and then eliminate the DerivesNodeOrCSSValue template.

+
+template<typename T>
+struct DefaultThreadingTrait<T, false> {

[haraken, 2014/02/21 06:51:08]

Would you add a comment to mention that only Node and CSSValue use
MainThreadOnly?

+    static const ThreadAffinity Affinity = MainThreadOnly;
+};
+
+template<typename T>
+struct DefaultThreadingTrait<T, true> {
+    static const ThreadAffinity Affinity = AnyThread;
+};
+
 template<typename T>
 struct ThreadingTrait {
-    static const ThreadAffinity Affinity = MainThreadOnly;
+    static const ThreadAffinity Affinity = DefaultThreadingTrait<T>::Affinity;
 };
 
 // Marks the specified class as being used from multiple threads. When
 // a class is used from multiple threads we go through thread local
 // storage to get the heap in which to allocate an object of that type
 // and when allocating a Persistent handle for an object with that
 // type. Notice that marking the base class does not automatically
 // mark its descendants and they have to be explicitly marked.
 #define USED_FROM_MULTIPLE_THREADS(Class)                 \
     class Class;                                          \
@@ skipping 450 matching lines @@
         ASSERT(ThreadState::isMainThread());
         return ThreadState::mainThreadState();
     }
 };
 
 template<> class ThreadStateFor<AnyThread> {
 public:
     static ThreadState* state() { return ThreadState::current(); }
 };
 
+/*
 // FIXME: Experiment if the threading affinity really matters for performance.
 // FIXME: Move these macros and other related structures to a separate file.
 USED_FROM_MULTIPLE_THREADS(Algorithm);
 USED_FROM_MULTIPLE_THREADS(Crypto);
 USED_FROM_MULTIPLE_THREADS(DOMParser);
 USED_FROM_MULTIPLE_THREADS(DeprecatedStorageQuota);
 USED_FROM_MULTIPLE_THREADS(Key);
 USED_FROM_MULTIPLE_THREADS(KeyPair);
 USED_FROM_MULTIPLE_THREADS(MemoryInfo);
 USED_FROM_MULTIPLE_THREADS(Notification);
@@ skipping 11 matching lines @@
 USED_FROM_MULTIPLE_THREADS(WebKitNotification);
 USED_FROM_MULTIPLE_THREADS(WorkerCrypto);
 USED_FROM_MULTIPLE_THREADS(WorkerPerformance);
 USED_FROM_MULTIPLE_THREADS(XMLHttpRequest);
 USED_FROM_MULTIPLE_THREADS(XMLSerializer);
 USED_FROM_MULTIPLE_THREADS(XPathEvaluator);
 USED_FROM_MULTIPLE_THREADS(XPathExpression);
 USED_FROM_MULTIPLE_THREADS(XPathNSResolver);
 USED_FROM_MULTIPLE_THREADS(XPathResult);
 USED_FROM_MULTIPLE_THREADS(XSLTProcessor);
+*/
 
 }
 
 #endif // ThreadState_h