Emulation Timing

[WedNESday]

I am confused by how QueryPerformanceCounter works. I have used QueryPerformanceFrequency to get my CPU ticks, and I know that works because it returns the speed of my CPU in Hz correctly. Now I have searched Google and have come up with this kind of thing;

Code: Select all

QueryPerformanceCounter(&lpPerformanceCount);
(Run a frame)
QueryPerformanceCounter(&lpPerformanceCount);

Now apparently you're supposed to save the before value and compare it to the after one. However when I do the following;

Code: Select all

QueryPerformanceCounter(&lpPerformanceCount);
(Message Box Displaying lpPerformanceCount)
QueryPerformanceCounter(&lpPerformanceCount);
(Message Box Displaying lpPerformanceCount)

It's like 10000000 different when it should only be a couple of thousand. So what does QueryPerformanceCounter actually return? It always seems to wrap at around the 32-bit 4 billion mark.

Any help/examples?

[mic_]

What does counter2.QuadPart - counter1.QuadPart give you?

[mic_]

And if you've got a MessageBox call in between the two calls you could get a high value because MessageBox doesn't return until you press OK/Cancel/whatever.

[WedNESday]

What does counter2.QuadPart - counter1.QuadPart give you?

That appears to give me the amount of time a frame takes, in Hz. Perfect! Holy crap, 21Mhz for just one frame...

Now that I've got the length of one frame, what kind of calculations must I do?

[mic_]

what kind of calculations must I do?

To get what..?

[WedNESday]

To limit the emulator to just 60FPS using Sleep() or whatever, that kind of thing.

[mic_]

Google says:

Code: Select all

//limit FPS to 60
void Sleep()
{
	static LARGE_INTEGER s_lastTime = m_starttime;
	LARGE_INTEGER currentTime;
	float  fps;

	do {
		Sleep(1);
		QueryPerformanceCounter(&currentTime);
		fps = (float)m_ticksPerSecond.QuadPart/(currentTime.QuadPart - s_lastTime.QuadPart);
	} while (fps > 60);

	s_lastTime = currentTime;

}

Though using Sleep(1) will hog the CPU quite a bit, so you should probably do it in two stages. First with longer sleeps, and then with Sleep(1) as you get closer to the limit.

[WedNESday]

And that's the problem for me. I know that I can use methods that will work, but will use the CPU 100% of the time, so I need the kind of method that other emulator authors use.

[dXtr]

Here's another tip for you if you're using QueryPerformacene*

I discovered this one after trying our code at the office on another computer.. it seems like computers with more then one core can have a problem with those functions making it very unstable. To solve this you call SetThreadAffinityMask() before and after the QueryPerformance call.

Example

Code: Select all

DWORD oldMask = SetThreadAffinityMask(GetCurrentThread(), 1);

bool bUsingQPF = QueryPerformanceFrequency(&m_TicksPerSecond);

SetThreadAffinityMask(GetCurrentThread(), oldMask);

edit:
Worth to note though, is that I've heard mixed response to this solution with some saying that this still won't guarantee anything. But it seems to work good so far ;)