we're working on an official mechanism in #1599 in response to your feedback (thank you!), but you c

we're working on an official mechanism in #1599 in response to your feedback (thank you!), but you could try:

const double loopsPerSecond = 10000;

Stopwatch loopStopwatch = Stopwatch.StartNew();
window.Initialize();
while (!window.IsClosing)
{
    do
    {
        if (X86Base.IsSupported)
        {
            X86Base.Pause();
        }
    }
    while (loopStopwatch.Elapsed.TotalSeconds < (1 / loopsPerSecond));
    window.DoEvents();
    if (!window.IsClosing)
    {
        window.DoUpdate();
    }
    if (!window.IsClosing)
    {
        window.DoRender();
    }
}
window.DoEvents();
window.Reset();

const double loopsPerSecond = 10000;

Stopwatch loopStopwatch = Stopwatch.StartNew();
window.Initialize();
while (!window.IsClosing)
{
    do
    {
        if (X86Base.IsSupported)
        {
            X86Base.Pause();
        }
    }
    while (loopStopwatch.Elapsed.TotalSeconds < (1 / loopsPerSecond));
    window.DoEvents();
    if (!window.IsClosing)
    {
        window.DoUpdate();
    }
    if (!window.IsClosing)
    {
        window.DoRender();
    }
}
window.DoEvents();
window.Reset();

const double loopsPerSecond = 10000;

Stopwatch loopStopwatch = Stopwatch.StartNew();
window.Initialize();
while (!window.IsClosing)
{
    do
    {
        if (X86Base.IsSupported)
        {
            X86Base.Pause();
        }
    }
    while (loopStopwatch.Elapsed.TotalSeconds < (1 / loopsPerSecond));
    window.DoEvents();
    if (!window.IsClosing)
    {
        window.DoUpdate();
    }
    if (!window.IsClosing)
    {
        window.DoRender();
    }
}
window.DoEvents();
window.Reset();

const double loopsPerSecond = 10000;

Stopwatch loopStopwatch = Stopwatch.StartNew();
window.Initialize();
while (!window.IsClosing)
{
    do
    {
        if (X86Base.IsSupported)
        {
            X86Base.Pause();
        }
    }
    while (loopStopwatch.Elapsed.TotalSeconds < (1 / loopsPerSecond));
    window.DoEvents();
    if (!window.IsClosing)
    {
        window.DoUpdate();
    }
    if (!window.IsClosing)
    {
        window.DoRender();
    }
}
window.DoEvents();
window.Reset();

King•8/6/23, 6:42 PM

Oh... wait that turned into an actual ticket? :o

khitiara•8/6/23, 6:43 PM

and now a pull request yep!

PerkseyOP•8/6/23, 6:43 PM

https://github.com/dotnet/Silk.NET/pull/1599

King•8/6/23, 6:43 PM

Oh wow... nice

PerkseyOP•8/6/23, 6:44 PM

only hold-up rn is that it doesn't implement something like loopsPerSecond in the above sample, it just loops until one of the events is due, which isn't good for rhythm games that need regular input polling

King•8/6/23, 6:45 PM

I am genuinely impressed you guys managed to figure out something at all, I been googling this whole thing like a madman and everything I find is either "Use vsync" or "Use Sleep(1)" but the latter is just aaaaa because of how the OS scheduler works or something along those lines

PerkseyOP•8/6/23, 6:47 PM

yeah so this actually came from me looking at the macOS kernel source code and discovering that their implementation of nanosleepnanosleep is just a busy loop with a pausepause instruction. we had tried to make our own cross-platform high-resolution sleep in the past, but it only worked on windows, so the pausepause method seemed good enough for me seeing as macOS uses it and their IO seems to be pretty CPU friendly

khitiara•8/6/23, 6:48 PM

wonder if the net7 requirement could be dropped using a small bit of native code (could be written in zig or whatever else) to just export one function for that

khitiara•8/6/23, 6:48 PM

since we already do a lot of native stuff anyway here

King•8/6/23, 6:48 PM

If I may suggest another thing to look at: RayLib. I did some testing with that and apparently their implementation of SetFPS() manages to lock FPS to a value without eating CPU.

khitiara•8/6/23, 6:48 PM

might be too much of a perf hit to do the pinvoke though im not sure

King•8/6/23, 6:49 PM

I don't know nearly enough low-level stuff to even begin to figure out what they do to manage this miracle but it might offer some ideas besides what apple does

PerkseyOP•8/6/23, 6:50 PM

interesting, they use nanosleepnanosleep

khitiara•8/6/23, 6:50 PM

yup heres the excerpt from raylib

PerkseyOP•8/6/23, 6:51 PM

https://github.com/raysan5/raylib/blob/master/src/rcore.c#L5088

PerkseyOP•8/6/23, 6:51 PM

cc @baebey

ritchie ❤•8/6/23, 6:51 PM

I have two questions

Kkhitiara yup heres the excerpt from raylib

PerkseyOP•8/6/23, 6:52 PM

this is actually really intelligent, it gets around nanosleepnanosleep being like 1ms out by just busy waiting for a certain percentage of the loop and sleeping for the rest

PerkseyOP•8/6/23, 6:53 PM

so it gets "almost there" on nanosleep but preserves the accuracy of busy waiting

khitiara•8/6/23, 6:53 PM

yeah it looks real smart

khitiara•8/6/23, 6:53 PM

i dont understand it

khitiara•8/6/23, 6:53 PM

but it looks good

King•8/6/23, 6:57 PM

This is why I ended up asking about how to dodge the silk event handler. What I could gather from the magic raylib does is that it checks frametime and kinda sleeps until it's time to do a frame, and that ends up limiting FPS while also keeping cpu low.

King•8/6/23, 6:57 PM

So I figured I'd do the same thing with Thread.Sleep(1). Just fire off that whenever frametime goes too low and you kinda sorta get a settable FPS with limited cpu usage

King•8/6/23, 6:58 PM

It kinda works... but is really really messy and bouncing all over the place due to sleep not being 1ms but several of them

King•8/6/23, 6:58 PM

But at least I have now learned why minecraft's fps slider is so unreliable and microstutter-y

Aqua•8/6/23, 6:58 PM

Thread.SleepThread.Sleep is simply not accurate enough to be used as a timer like this

King•8/6/23, 6:59 PM

Yeeeeeah... wtb better OS thread scheduler ;-;

Aqua•8/6/23, 6:59 PM

the solution is to either use a SpinWaitSpinWait, which i've found works well, although not at FPS values above 60 the last time I tested

Aqua•8/6/23, 6:59 PM

or to just waste cpu cycles, which is the most accurate

PerkseyOP•8/6/23, 6:59 PM

so windows does actually have high-resolution sleeping, but every other platform tends to be like 1ms out

PPerksey we're working on an official mechanism in #1599 in response to your feedback (th...

PerkseyOP•8/6/23, 6:59 PM

before jumping into anything, can you try this out and see what improvement (if any) it makes?

PPerksey before jumping into anything, can you try this out and see what improvement (if ...

King•8/6/23, 7:01 PM

Uhm... give me a moment... my code right now is so messy due to the attempt at my own FPS limiter... need to make sure I comment all that out proper xD

King•8/6/23, 7:04 PM

Okay so the first problem I'm having is

King•8/6/23, 7:04 PM

King•8/6/23, 7:05 PM

Am I missing some using?

khitiara•8/6/23, 7:05 PM

what dotnet version you using?

King•8/6/23, 7:05 PM

khitiara•8/6/23, 7:05 PM

hmm

King•8/6/23, 7:05 PM

I think...

King•8/6/23, 7:05 PM

lemmie check

King•8/6/23, 7:05 PM

Oh, right... old project

King•8/6/23, 7:05 PM

was 6

King•8/6/23, 7:06 PM

Set it to 7 and that warning went away, good good

khitiara•8/6/23, 7:06 PM

yeah that method was introduced in 7

khitiara•8/6/23, 7:06 PM

thats why theres the #if NET7_OR_GREATER in the pr

King•8/6/23, 7:08 PM

Hmm... sadly the code doesn't seem to do anything, unless I"m being a dummy

King•8/6/23, 7:08 PM

FPS and UPS is still going mad at 5000 according to deltatime

King•8/6/23, 7:09 PM

and I did set const double loopsPerSecond = 500;const double loopsPerSecond = 500;

King•8/6/23, 7:09 PM

And CPU and GPU is going mad

we're working on an official mechanism in #1599 in response to your feedback (thank you!), but you c

Similar Threads

Similar Threads

Similar Threads