This could alleviate the suffering of dual core synchronization a bit.

But I doubt you would notice it in most cases.

git-svn-id: https://dolphin-emu.googlecode.com/svn/trunk@4830 8ced0084-cf51-0410-be5f-012b33b47a6e

Source/Core/Common/Src/Thread.cpp

@@ -117,6 +117,60 @@ void Thread::SetCurrentThreadAffinity(int mask)
 	SetThreadAffinityMask(GetCurrentThread(), mask);
 }
 
+#ifdef _WIN32
+EventEx::EventEx()
+{
+	InterlockedExchange(&m_Lock, 1);
+}
+
+void EventEx::Init()
+{
+	InterlockedExchange(&m_Lock, 1);
+}
+
+void EventEx::Shutdown()
+{
+	InterlockedExchange(&m_Lock, 0);
+}
+
+void EventEx::Set()
+{
+	InterlockedExchange(&m_Lock, 0);
+}
+
+void EventEx::Spin()
+{
+	while (InterlockedCompareExchange(&m_Lock, 1, 0))
+		// This only yields when there is a runnable thread on this core
+		// If not, spin
+		SwitchToThread();
+}
+
+void EventEx::Wait()
+{
+	while (InterlockedCompareExchange(&m_Lock, 1, 0))
+		// This directly enters Ring0 and enforces a sleep about 15ms
+		SleepCurrentThread(1);
+}
+
+bool EventEx::MsgWait()
+{
+	while (InterlockedCompareExchange(&m_Lock, 1, 0))
+	{
+		MSG msg;
+		while (PeekMessage(&msg, 0, 0, 0, PM_REMOVE))
+		{
+			if (msg.message == WM_QUIT) return false;
+			TranslateMessage(&msg);
+			DispatchMessage(&msg);
+		}
+		// This directly enters Ring0 and enforces a sleep about 15ms
+		SleepCurrentThread(1);
+	}
+	return true;
+}
+#endif
+
 // Regular same thread loop based waiting
 Event::Event()
 {
@@ -164,6 +218,7 @@ void Event::MsgWait()
 			if (msg.message == WM_QUIT)  
 				return; 
 			// Otherwise, dispatch the message.
+			TranslateMessage(&msg);
 			DispatchMessage(&msg); 
 		}
 

Source/Core/Common/Src/Thread.h

@@ -137,7 +137,31 @@ private:
 #endif
 };
 
-
+#ifdef _WIN32
+// Event(WaitForSingleObject) is too expensive
+// as it always enters Ring0 regardless of the state of lock
+// This EventEx will try to stay in Ring3 as much as possible
+// If the lock can be obtained in the first time, Ring0 won't be entered at all
+class EventEx
+{
+public:
+	EventEx();
+	void Init();
+	void Shutdown();
+	void Set();
+	// Infinite wait
+	void Spin();
+	// Infinite wait with sleep
+	void Wait();
+	// Wait with message processing and sleep
+	bool MsgWait();
+private:
+	volatile long m_Lock;
+};
+#else
+// TODO: implement for Linux
+#define EventEx	Event
+#endif
 
 class Event
 {
@@ -182,9 +206,10 @@ private:
 void InitThreading();
 void SleepCurrentThread(int ms);
 
-// YieldCPU: Use this function during a spin-wait to make the current thread
-// relax while another thread is working. This may be more efficient than using
-// events because event functions use kernel calls.
+// YieldCPU: This function is only effective on HyperThreading CPU
+// Use this function during a spin-wait to make the current thread
+// relax while another thread is working. This may be more efficient
+// than using events because event functions use kernel calls.
 inline void YieldCPU()
 {
 #ifdef _WIN32