AppSpeed

[gamecreator]

I was testing my program to see how it would behave on slower computers, by adding various Sleep amounts in my code. To my surprise, the program didn't act as I expected. So, I decided to write a simple test program to see why this was the case. Here is the relevant part:

 

float x=0.0;
double t1, t2;

t1=t2=timeGetTime();

while(!KeyHit(KEY_ESCAPE))
{
Sleep(1);

x+=0.01*AppSpeed();

if(x>5)
{
	t2=timeGetTime();
	x=-10000.0;
}

UpdateFramework();
RenderFramework();

DrawText(0,20,"x: %f",x);
DrawText(0,40,"AppSpeed: %f",AppSpeed());
DrawText(0,60,"time: %f",t2-t1);

Flip(0);
}

 

I noticed that as I increased the Sleep amount, the time it took x to go from 0 to 5 took longer and longer.

 

Sleep(amount): time

1: 8185ms

10: 8295ms

50: 8757ms

100: 9322ms

 

Is this difference expected? I get that we're talking milliseconds here but this could be the difference between making a jump and missing a cliff, especially if the fps drops in certain scenes.

[Pixel Perfect]

I don't know how accurate the AppSpeed() function is but certainly the sleep function is not that accurate and is dependent on the tick resolution of the system its being run on. It tend to be less accurate on low values (< 10ms) and even when it completes, the thread is not guaranteed to restart immediately.

 

You might want to do a comparison using a high resolution timer that simply loops rather than sleeps the thread.

[gamecreator]

Fair enough. Would this be better tested by decreasing the frame rate by putting a lot of triangles on the screen?

[Pixel Perfect]

That might be more representative of the type of situation you would really face but there would normally be a non rendering cost associated with running game logic/AI too (unless it was a very simple game) which is better simulated by loops. A combination of both might be better. But you could try increasing polygon rendering initially, that should start to bring the frame rate down fairly quickly from an initial high value.

[gamecreator]

I wrote the code to only include loaded models (not Sleep). This is what I got:

 

models - fps - time

0 - 1080 - 8169

25 - 189 - 8216

50 - 100 - 8265

75 - 68 - 8326

100 - 51 - 8376

150 - 34 - 8476

200 - 26 - 8560

300 - 17 - 8756

500 - 10 - 9182

 

If you do the conversion between these results and the results in the first post, you'll find they're actually pretty close and linear (though I did my bit of rounding so this is hardly exact). For example, the 10 frames per second with a heavy model load brought the result time to 9,182ms whereas above, with a 100ms Sleep (also 10fps), the time was 9,322.

 

Does this mean that AppSpeed can't be relied upon to give you the same results on two different computers?

[Roland]

This might be a bit off-topic, but what does AppTime actually return. Its said the current application time and nothing more.

Is that in milliseconds and whats the resolution (is it msec with resoltuon 15 msec)? It does not seem to be the application time either as it the would start will 0.

[DaDonik]

If you really want to know how your game will behave on a slower PC, it's not enough to slow down you CPU. You should also remove your GPU and put in a really bad one.

 

Does this mean that AppSpeed can't be relied upon to give you the same results on two different computers?

AppSpeed returns 1.0 if your game runs at 60fps, 2.0 at 30fps and 0.5 at 120 fps. I tested this on 3 different PC's by just increasing/decreasing the amount of cubes i render. It was not a precise measurement, but at around 30fps AppSpeed returns ~2.0.

 

It does not seem to be the application time either as it the would start will 0.

I could be wrong, but i think that AppTime() just returns the system time (time in milliseconds since startup).

[Road Kill Kenny]

I could be wrong, but i think that AppTime() just returns the system time (time in milliseconds since startup).

 

Yes it does. And it loops at some number but I can't remember what that number is.

[Roland]

Yes it does. And it loops at some number but I can't remember what that number is.

Now how strange isn't that Made a quick test and it seems to increment in 64 msec steps.

[Josh]

AppTime returns the time at the last UpdateTime() call. This way the time value stays the same for your entire loop, until the next time update.

[Roland]

Thanks for that clarification Josh.

[gamecreator]

AppSpeed returns 1.0 if your game runs at 60fps, 2.0 at 30fps and 0.5 at 120 fps. I tested this on 3 different PC's by just increasing/decreasing the amount of cubes i render. It was not a precise measurement, but at around 30fps AppSpeed returns ~2.0.

Then any idea what's wrong with my code? Shouldn't this make x increase at a constant speed on all computers?

 

x+=0.01*AppSpeed();

[DaDonik]

When using the curve function i do:

fMove = Curve ((float)(KeyDown (KEY_W) - KeyDown (KEY_S)), fMove, 5.0f / AppSpeed());

For movements i do:

float fMove = fMove * ((1.0f / 1000.0f) * AppSpeed());

After that fMove is measured in meter/second. To make it easier for me i made myself a macro:

#define M_S     (0.016666f * AppSpeed())