LittleBigPlanet 2 [BCES00850/BCES01693] General multiplayer issues

[TorutheRedFox]

Quick summary

Online multiplayer manages to connect only about 10% of the time, and the chances of it not instantly diverging are quite low too.

There's also times where the emulator hard crashes (closes) when attempting to connect sometimes.

Details

All of the connection attempts were done through matchmaking provided by LBP Union's Beacon server, and when trying to connect, most of the time it fails to connect, saying "Failed to connect to [username]" (quite often followed by a crash a second or so later), and other times it seems like it tries to join before immediately failing with an ingame error message saying "failed to connect to players" and booting you back to your pod.

Sometimes it manages to actually connect, ask the game leader to let you join, and if they say yes, it'll connect successfully, but there's a high chance then that less than a second after you join and everything loads, the game will diverge and kick everyone out of the room.

Attach a log file

RPCS3.log (crashed immediately after attempting to dive in)

Attach capture files for visual issues

No response

System configuration

No response

Other details

No response

[TorutheRedFox]

same seems to apply to LittleBigPlanet [BCES00141] too, and probably to LittleBigPlanet 3 [BCES01663] too

[scarletcafe]

After testing, it seems people have mostly realized that a strong cause of divergence is attempting to matchmake between AMD and Intel machines. Operating system and precise processor does not seem to have much of an effect (I have had the Steam Deck (AMD APU) running the Linux version of RPCS3 connect to my Windows 11 host machine (Ryzen 5950x, RTX 3090) running the Windows version of RPCS3 without strong divergence).

In particular, Intel<->Intel sessions and AMD<->AMD sessions seem to be the most stable, they are not free of divergence but I have had them run for >30 minutes at a time online before encountering an issue. In contrast, Intel<->AMD sessions tend to either not connect at all, or diverge effectively immediately (the longest I've had a session hold was about 10 seconds).

Matching settings between versions seems to extend the amount of time marginally, but there is so much variance it is hard to pinpoint which ones are actually having an impact.

Things I've tested that do not seem to fix the issue:

• Using static interpreter SPU & PPU
• Changing XFloat accuracy
• Disabling TSX instructions on the Intel machine
• Changing thread scheduling
• Using the single-threaded shader mode
• Using strict rendering mode
• Ordered & atomic RSX FIFO
• Accurate sleep timers
• Accurate PPU NaN handling/float condition

I will admit I don't know a lot about RPCS3 internals, but I tried various permutations of the settings above to no effect. I do not know exactly what accuracy issue LBP is detecting, but if I get time in the future I may try hosting a custom Lighthouse & RPCN server to see if I can reverse engineer what the issue is at the network level for any hints.

[scarletcafe]

Reposting a theory from the Discord so that it is visible for documentation. In my tests in practice, ASMJIT didn't seem to help the problem, possibly for other reasons, and I already tried XFloat to no effect.

[TorutheRedFox]

doesn't explain most connection attempts ending with a refused to connect error

(by the way this has nothing to do with configuration as attempting multiple times over with only one other person on a closed lighthouse instance resulted in a success after 30 minutes worth of attempts)

[kubacAkaGoomba]

Can confirm with @Gorialis that the divergence issue is persistent despite the changes of CPU SPU/PPU and other potential RPCS3 settings mentioned above that could affect LBP's online multiplayer based on my test between 2 PCs (One that has AMD Ryzen 7 3700X CPU and RTX 3080 and the other - Intel i7-7700k CPU and GTX 970).

Also to shed some more light on issues mentioned by @TorutheRedFox:

• some connectivity issues are most likely unsolvable due to the very nature of peer-to-peer used for LBP's multiplayer which depends on the network conditions of all connected players (the original hardware - PS3 also has issues with connection leading to errors such as "Failed to connect to [player]")
• the aforementioned lobby divergence happening seconds after joining someone's lobby is due to previously mentioned mismatch between AMD and Intel CPUs, if you manage to successfully connect to the player which is running RPCS3 on the same CPU brand then the online session is much more stable, though not without occasional 'The game diverged' error, as it can happen pretty much at random but it's also largely dependent on what's happening in the level and it's also dependent on some player actions. Again, the original hardware also had "The game diverged" error but it was extremely rare compared to RPCS3.

There's also few things I want to point out myself:

• Currently friends list do not work in-game and it seems like LBP is unable to fetch RPCN friends overall, meaning you cannot join others directly (though I do recall seeing the screenshots and videos of RPCS3's test build which does show friends list and joining friends feature working, so I'm hoping that build is going to be pushed into PR relatively soon).
• Because friends list do not work, the only two options to join other players remain: Dive-in functionality from the Community's main page in the Pod and the ability to join others playing any story/community level upon loading it. The latter is more reliable due to being able to control who you want to connect with more rather than using Dive-in which will attempt to join any RPCS3 player no matter what CPU brand they have.
• After failing to connect to a certain player (or if the player rejects your request to join) via Dive-in/Join others into the level, the next time you try to join the same player, the game will either ignore the player you are trying to join and look for someone else or try to join that person but with the target host not receiving the join request on the other end eventually resulting in an "We did not receive an answer from the people you were trying to join" error. This, in my opinion, is not an RPCS3 bug, but rather how LBP series handle joining other players via Dive-in/joining levels and it might be due to how LBP 'blacklists' the stranger from trying to rejoin until the game is restarted.

[kubacAkaGoomba]

Apparently in another game (#13599 (comment)) with similar issue to this one you can get AMD/Intel connection to work by just standing still and not changing player coordinates.

The problem with LBP games though is that when spawning in from the entrances/checkpoint after joining another player the Sackboy spawns slightly above the ground making it somewhat impossible to NOT move after joining someone's lobby.

However there are technically two ways to reproduce this behaviour, similarly to the previously mentioned game: one is to create an empty level with an entrance that has Anti-gravity tweaker that will ensure the Entrance stands completely still and the Level Gate which, upon activating, will make joining players unable to spawn in until the level is restarted. The other is to create an Entrance and then using a Merge Glitch, slightly merge the Entrance down to the ground making it so Sackboy will immediately stand on the floor once he spawns into the level. The latter is probably going to be incredibly more difficult to achieve without diverging so I will go with the first.

When I get back home, I'll try to reproduce it.

[kubacAkaGoomba]

I was planning on releasing the footage of my AMD/Intel tests until I saw the PR 😄.
But the tests confirmed more or less what I suspected: I used the Level Gate method as mentioned. The game still instantly diverges with Sackboy standing still in the level, however once I added Controllinator and when Sackboy was sitting on it, the game did not diverge and I was able to switch costumes, place stickers etc.
As soon as I got out of the Controllinator the game instantly diverged.

Surprisingly enough I was albo able to get some basic physics working on AMD+Intel session, emitting objects worked, both static and dynamic ones, Sackbots did not diverge the session and also moving around the object with the Controllinator also worked until certain physics caused it to diverge.

So it turns out the Sackboy physics are the main culprit, considering that even standing still causes the game to diverge and it would also explain why an AMD+Intel session would diverge in matter of seconds.

Anyway, I'm eagerly waiting for PR to be merged to see if that fixes it 🤞

[kubacAkaGoomba]

Alriight, I've yeeted the link for the compiled PR to Beacon Custom Server users to try it out and these guys seem to be going crazy and it seems like the connection between AMD and Intel is working properly now 😅

I'm on my way home right now so I can test the PR between two PCs myself.

[TorutheRedFox]

I tested the PR, it worked first try

[TorutheRedFox]

I tested the PR, it worked first try

worked second try after having to restart after someone else got accidentally rejected too

[kubacAkaGoomba]

Can confirm, AMD - Intel connectivity works!

[kubacAkaGoomba]

My tonight's stream on YouTube speaks for itself 😀
https://www.youtube.com/watch?v=-9bc9xNjxJE

Most of the people I played with are on AMD CPU, however they were a few like SammyGoesHowdy (in the stream mentioned above) who has an Intel CPU and we managed to connect and have prolonging sessions much like how we would with AMD players on a master build.