From http://www.opengl.org/discussion_boards/ubbthreads.php?ubb=showflat&Number=196709&page=3

Quote:Just for the record - for my personal, immediate purposes, the following setup seems to work fine :)

if( dt>24) {
  if(vsync) {
    wglSwapIntervalEXT(0);
    vsync=false;
  }
}
else if( dt<12){
  if(!vsync) {
    wglSwapIntervalEXT(1);
    vsync=true;
  }
}
SwapBuffers();

dt being the time between the start of the last render, and the start of the current - ie. one frame lag.

Is this still helpful if triple-buffering is enabled?

anyone?

Quote:Just for the record - for my personal, immediate purposes, the following setup seems to work fine :)

if( dt>24) {  if(vsync) {    wglSwapIntervalEXT(0);    vsync=false;  }}else if( dt<12){  if(!vsync) {    wglSwapIntervalEXT(1);    vsync=true;  }}SwapBuffers();

dt being the time between the start of the last render, and the start of the current - ie. one frame lag.

That is very interesting. I had the same question with no answers a few months ago, and I came out with almost the same idea, but for some reason, I could never get it to work. V-sync never got switched on, or never got switched off, no matter what values I used it the else-if conditions (I also had the two different values to introduce some kind of hysteresis), just didn't work.
So I decided to go the user selection way, and the user can switch to lower resolution, decide how many particles are drawn etc.

Quote:Original post by samoth
This code snippet tries to deal with the problem by collecting some more or less reliable data (last frame time), compare it to some hardcoded value and turn VSYNC on/off on the fly.

Of course, it is trivial to extend to some statistic measure relying on a few past frames, rather than a single frame.

Quote:Here's is the first flaw already. Not only is 24 ms a really weird frame time, this corresponds to something like 42 fps -- few if any devices will have a refresh rate like this -- but also you don't really know the real refresh rate, so hardcoding it is a kind of stupid idea.

I am myself very curious about this choice. As I wrote in the first post, I copied this snipped from elsewhere. The following comment is one of the replies on that forum:

Quote:Mikkel Gjoel: you are right, you snippet works very well !
This is really nice to the eye, even with black+white vertical stripes in variable high-speed horizontal scroll, my worst case as far as display refresh is concerned.
I don't get why your first threshold value works so well, but I could not get better results with a smaller value. However 85 Hz monitor would mean around 11-12 milliseconds, not 24 right ?

There was no explanation in reply though; I myself posted in the thread but got no answer of any sort.
I cannot begin to imagine the reason for the choice, whether latencies for wglSwapIntervalEXT's effect come into play, or what... I'd love to hear some suggestions. Perhaps the choice is simply empirical, but given that a different random user also found this the smallest value that made a difference, it's unlikely that the writer of the snipped had some unusual refresh rate.

Quote:And, whatever you measure is (obviously) after the fact.

That's hardly specific to this issue. Consider a physics simulation where you're determining the delta-t. There is at least a one-frame lag, some smoothness of the dt variation over most frames is always assumed since you cannot predict discontinuous events that would cause a sudden change (since they usually result from user action or arise from system complexity). This is a general aspect of any feedback-based controller system, where the variable is not necessarily just time but could be temperature or whatever.

Quote:I think the right solution here is to let the user choose. Usually I hate it when people just play the "let the user choose" card, but in this case, I think it is really justified.
Not only is this much less work and trouble, but it also is a lot more reliable. And, it accounts for the fact that different people are... well... different. Some people maybe don't mind a bit of tearing but really hate low frame rates. And then, tearing might annoy the hell out of others. You don't know.
Lastly, you don't even know if you can change VSYNC at all. Most drivers allow the user to turn VSYNC on/off globally, so whatever you do might not be good for anything after all.

I'm lucky to be writing for a hardware and software platform that I specify (touchscreen kiosks), which removes these concerns. But even someone without that privilege can provide the user with such an option in addition to having vsync always on or always off, maybe with even tunable threshold dt. (By the way, the NVIDIA driver lets one to force vsync on and off, but also has a setting allowing the application to select it.) So I would really like to get a bit deeper into this issue in order to get best results.

This algorithm is, as you explained, geared to prevents halving of framerate for cases where some portion of frames might exceed one refresh time. The problem of course is that those frames would appear with tearing, so it's a specific compromise between lag to next frame display and visual artifacts.
This got me into thinking about the other solution, triple buffering, which presents a different compromise--adding a lag time of an extra frame to act as a safety margin for the cases where some frames take longer than a vertical retrace. But wouldn't it make sense, in cases where the draw takes less than half of a vertical refresh, to dynamically switch to double-buffering and thus improve latency? This would be somewhat analogous to the dynamically using vsync in the snippet above, and could be combined. But how can one turn triple buffering on/off from the application, given there seems to be no WGL/GLX extension as there is for vsync? Does NVIDIA have any way to ask the driver for it programmatically? Worst case, I'd imagine one could manually handle two back buffers in the application and turn triple-buffering off in the driver...

Then there is the question of how NVIDIA's driver setting of "Maximum pre-rendered frames" affects the above considerations...

I'd love some comments on the triple-buffering stuff :)