smoother timing in frame limiter

[water111]

No description provided.

[water111]

This should fix some of the issues we were having with the framelimiter and sets it as the default.

[water111]

this has a math bug in the windows sleep

[VirtualTim]

Why have two different implementations of sleep_us?

[ManDude]

One is for linux and the other is for Windows.

[VirtualTim]

Uh, yeah. I meant why not use

void sleep_us(u64 us) {
  std::this_thread::sleep_for(std::chrono::microseconds(us))
}

for both platforms?

[water111]

We need sleeps that are accurate to a few ms for this to work, but the default windows timer resolution is (sometimes) ~15 ms. that's almost a whole frame long, so it becomes impossible to sleep for the right amount of time. The solution is to use the windows API to request a higher timer resolution and do the sleeps manually.

The C++ standard library sleeps are kinda weird too, they have very poorly specified behavior:
"The standard recommends that a steady clock is used to measure the duration. If an implementation uses a system clock instead, the wait time may also be sensitive to clock adjustments."

and we definitely wouldn't want the frame limiter to be affected by somebody changing their system time...

[VirtualTim]

Makes sense. I did not realise that the std sleeps were so poorly defined. I always assumed they were a steady clock, because as you said, if they're not that's hugely problematic.