How does WinUAE detect joystick inputs on pause?

[BeamCoder]

How does WinUAE detect joystick inputs on pause?

I was testing the input lag of my Pong game and I was only using a single buffer yet the inputs always come out at the maximum delay for a single buffer; 2 frames.

Here is the process of how I test input lag:

1. Pause emulation
2. Hold a joystick direction to move paddle
3. While holding, do a single step.
4. Do single steps up until paddle moves. Always results to 2 frames before paddle moves.

My only guess is WinUAE or maybe all UAE(I also tested PUAE on retroarch) detects joystick inputs at the frame after doing a single step and not at the pause frame.

To add information about my Pong game, it reads inputs just after the end visible line on an NTSC(line $f4).

[Toni Wilen]

Pause resume behavior is "undefined" (and it won't change), sorry. It is not good for latency measurements.

2 frames probably is normal if not in lagless vsync mode but using pause as a starting point can cause side-effects or extra delays, at least in some situations.

Input is read multiple times/frame.

MPO (multiplane overlay) might reduce latency by 1 frame but I still haven't really found any useful documentation how it actually works.. I only know NVidia recently introduced MPO driver support.

[BeamCoder]

Originally Posted by Toni Wilen

Pause resume behavior is "undefined" (and it won't change), sorry. It is not good for latency measurements.

2 frames probably is normal if not in lagless vsync mode but using pause as a starting point can cause side-effects or extra delays, at least in some situations.

Input is read multiple times/frame.

MPO (multiplane overlay) might reduce latency by 1 frame but I still haven't really found any useful documentation how it actually works.. I only know NVidia recently introduced MPO driver support.

Thanks for the reply.

Regarding doing pause-frame advance step in testing input lag, I have learned about this method by Retroarch users doing it for using the Runahead feature of Retroarch. I thought that this is a good way to test input lag on just the game decoupling it from the hardware its being run on(vsync on or lagless vsync, what GPU is used, etc.).

I tried moving the reading of input on raster line 1 and now movement of the paddle appears after the step. My guess now is maybe when doing pause, it stops at vblank. Also, since after pausing a frame is already fully drawn, then it probably did pause at vblank.

I also tested other games:
Solid Gold - consistent 3f of delay on any actions(jumping, moving left and right, moving left right on air). I read somewhere it uses double-buffering

Gianna Sisters - consistent 2f delay, read somewhere it uses only a single buffer.

LionHeart - different moves have different lag however most moves have consistent delay(2f-jumping, 3f when crouching, readying attack, and attacking. Moving left and right is finnicky though; 3f-5f).

All of these where tested on 50Hz lagless vsync 1 slice and standard vsync. Solid gold was one timed tested on 29 slices. All where consistent regardless of vsync setting.

Since delay are consistent on all the games I've tested, maybe pause-resume is not that undefined(stops randomly somewhere at vblank)?