Counters: Tighten tolerances on framelimiter #3785
Conversation
Smooths out frame pace and keeps average framerate closer to target.
Fixes graphics issues in some games
|
Example comparison of master (top) vs PR (bottom) - Midway Arcade Treasures title screen Master: PR: |
|
Bonus Master: PR: |
|
|
||
| // If any integer value of milliseconds exists, sleep it off. | ||
| // Prior comments suggested that 1-2 ms sleeps were inaccurate on some OSes; | ||
| // further testing suggests instead that this was utter bullshit. |
There was a problem hiding this comment.
This is true and might compromise accuracy a bit - it's worth running a test without sleep at all.
There was a problem hiding this comment.
I just tested the theory, it does indeed make it smoother, how noticeable it'll be though is any bodies guess
There was a problem hiding this comment.
Other games such as SotC don't seem to care too much
There was a problem hiding this comment.
Might be that those games don't hit the condition for a sleep?
There was a problem hiding this comment.
I run at well above the frame cap in those games. I get about 70fps in SotC, so there should be about 3.334ms (maybe 2.5ms, my calculation may have been off) my CPU is waiting every frame.
There was a problem hiding this comment.
Another look at the Xenosaga Ep. 1 Elsa save point. Master (top), msec >=1 then sleep(msec) (middle), msec > 1 then sleep(msec - 1) (bottom). I think we have our winner.
There was a problem hiding this comment.
That's a good approach, but I would go even further and make it something like msec - 3 or even 5. The threshold has to be pretty big or else you'll always risk people losing time on those e.g. due to running on a laptop battery.
There was a problem hiding this comment.
Well I don't recommend people run the emulator on a laptop and if they do they should be using high performance mode and not running off battery (that's a really dumb idea :P) I'm honestly not concerned about people crazy enough to do that, it will still be head over shoulders above master (which was 1ms buffer)
There was a problem hiding this comment.
Sure, that should be good enough as a starting point and could be revisited in the future if it's ever needed.
There was a problem hiding this comment.
I'm using a gaming laptop plugged in with high performance mode and it does improve significantly on specific games, u could stare blindly at squeezing an extra 5%-10%.
|
NASCAR Heat 2002 had good frametimes during the race, but entering a pit lane resulted in frametimes being all over the place: With this PR, it doesn't seem to occur anymore: |
|
A different test (Drakan: The Ancients' Gates with FrameratePAL=119.98) Master: 👍 |
|
All tested on default settings (commit-93823a3) Master: PR: Game 2 True Crime NY (Game needs a patch/fix); Master: PR: Game 3 Prince Caspian Chapter 2; Master: PR: Game 4 Rogue Galaxy; Master: PR: |
|
Another change has just been pushed that should (sometimes) tighten this up even further. |
|
Some more testing with new changes (I'll edit if I find anything substantial) CoD Finest Hour (PR on top, master on bottom) |
|
Latest PR (8aaae36)
|
|
Ratchet & Clank 2 Champions - Return to Arms Metal Arms - Glitch in the System 💯 |
|
Added more examples along with my system info Laracroft Legend: Alter Echo: FinalFantasyX: FinalFantasyX-2: |
|
Has anyone checked Linux? RTSS is not available but at least to make sure stability is still good? |
|
Examples when i change the msec if (msec > 1): (original PR) @RedPanda4552
|
|
Just tested on Linux x86 with EGL and GTK2, with a few Kingdom Hearts games (KH1, KH2, Re:CoM). Seems stable. |
|
Discovered this article thanks to a forum user: https://randomascii.wordpress.com/2020/10/04/windows-timer-resolution-the-great-rule-change/ We may need to play with Win10 2004 release. Sounds like they are changing how thread sleeping will behave and sleep(1) is going to be a LOT more latent than ever before, if this is right. |
|
Update on the above: multiple testers have disclosed they did their tests on 2004 and had positive results. I am comfortable with where this is but more testing is always welcome. Up to everyone else. |
|
Frame pacing did seem quite a bit better in a few games I tried, although Xenosaga is one that still kind of gives weird readings. |
Smooths out frame pace and keeps average framerate closer to target.
Comparing PCSX2's old framelimiter with RTSS showed there was room for improvement. Frametimes regularly would jump up and down in various situations. This PR smooths out frame pace substantially (easily visible in RTSS) and, drum roll please, may appease the input lag folks for a time. We'll find out.
Extensive user tests required; some vsync functions were shuffled around, and this was tested on a high performance system. We should test low end systems, and a wide variety of games to make sure there are no outlier games that needed the old behavior for some reason.
Shoutout to Refraction for greasing the wheels and helping with the various vsync aspects.
Before (above) and after (below) of Xenosaga Ep. 1 save point inside the Elsa:
