BBA/HLE: Increase polling efficiency

[sepalani]

This PR is based on https://github.com/dolphin-emu/dolphin/pull/12559 and updates the way BBA/HLE polls data. It does so by sleeping less often and pulling data as soon as possible and as much as possible.

I suspect this PR has improved the matchmaking stability. My Windows build seems to match consistently my Android phone and the other way around. It might be a coincidence since I don't have a lot of device to test it.

This PR is ready to be tested.

[JMC47]

Sorry for the late response here. Is the optimal way to test this with a Windows computer? Is there any need to test it on Linux/Android connectivity, which already worked fine for me.

[sepalani]

Sorry for the late response here. Is the optimal way to test this with a Windows computer?

Not necessarily, Dolphin on Linux/Android were working fine for me as well. Whereas I had matchmaking/stability issues on Windows which seems less present on this build. Though, I wasn't able to confirm that's related.

Is there any need to test it on Linux/Android connectivity, which already worked fine for me.

It might be worth checking if there is a speed improvement compared to the current build. I was only able to confirm that by sleeping more often, it degrades the performance (FPS) significantly on Mario Kart: Double Dash and Kirby Air Ride. That's why I suppose that sleeping less often might improve the performance.

Sadly my actual setup (Windows PC + Linux VM + Android phone) don't allow me to explore this further in a reliable way. I might be able to push the testing further by adding two of my Wii into the mix (using Nintendont) but it will have to wait 2 weeks or so as I don't have the time to do so currently.

[ShiftaDeband]

I was able to test the Universal MacOS binary on an M-series chip just now and had no issues with PSO. I also tested the ARM64 build as well without issue.

It did feel like the network speed was faster, but I can also attribute that to a placebo effect, and I'm not sure one would really see much improvement for a game that was made for the days of dial up internet, though I swear it felt better. Who knows.

Thank you for this.

E: Tested on Windows X64 and also had no issues with online functionality.

[AdmiralCurtiss]

You can rebase this now.

[sepalani]

I did some extensive testing with this PR. When I increase the sleep period it does decrease performance, sleeping less often seem to improve performance to some extent. However, when comparing with latest master and this PR, I didn't notice a difference when playing.

My testing environment

I tested this PR using Mario Kart: Double Dash! (60Hz mode) and in my testing environment (where I have a TV and my Nintendo Wiis). This room is different from where I usually use my Windows 10 laptop. All devices were connected to the same Wi-Fi network and my Windows machine has UPnP service disabled and Windows' network protection turned off.

Devices

• Nintendo Wii (using Nintendont)
• Windows 10
• Linux (using a VM)
• Android (ASUS ROG Phone 2)

Matchmaking tests

Here is the result of different matchmaking orders when using this PR :

• [KO] Nintendont, Android
• [KO] Nintendont, Linux
• [KO] Nintendont, Windows
• [~OK] Windows, Android
  • Failed on my test environment
• [KO] Windows, Linux
• [OK] Windows, Nintendont
• [~OK] Linux, Android
  • Failed on my test environment
• [OK] Linux, Nintendont
• [KO] Linux, Nintendont, Android
• [KO] Linux, Nintendont, Windows
• [OK] Linux, Windows
• [~OK] Linux, Windows, Android
  • Failed on my test environment
• [OK] Linux, Windows, Nintendont
• [OK] Android, Linux
• [OK] Android, Linux, Windows
• [OK] Android, Linux, Windows, Nintendont
• [OK] Android, Nintendont
• [OK] Android, Windows
• [KO] Android, Windows, Linux (stuck)
• [KO] Android, Windows, Linux (stuck), Nintendont (stuck)
• [OK] Android, Windows, Nintendont

Some takeaways

• My Windows host doesn't like its Linux guest (VM) nor my Android phone.
  • So I suppose that Windows hosts should join last or before Nintendont.
• The opposite is not true, both Android/Linux seem to accept the others fine.
• I can confirm that Nintendont has to join last for some reasons.
• The Wi-Fi signal in the room where I have my testing environment isn't as strong as where I use to have my laptop/phone.
  • That might be the reason why some testing involving Linux/Android failed.
  • This suggests that the network performance is also a key factor.
  • It seems MKDD countdown is stuck when one of the pairs can't detect a pair advertised by the other pairs.

These tests aren't as rigorous/stable as proper hardware testing (especially since I don't have a BBA) so they should be taken as what they are, some (bruteforce) testing within an environment where I try to keep the same settings as a base for each test case.

[JMC47]

In my experience (prior to this PR) real BBA worked fine as long as it wasn't the first thing to connect. Kirby Air Ride was way more stringent than Mario Kart though.

[JMC47]

I tested it on Linux a while back, but I can't test it on Windows because I don't run that OS anymore, and that's where the actual benefits supposedly show.

[AdmiralCurtiss]

Let's just get this in, people will yell if it breaks stuff anyway...