Emulator

This sounds like CPU emulation inaccuracies. Something you could try to do is enabling the debug GUI in Dolphin in interface and try running the game, pausing, and disabling parts of the JIT to see if any of that affects it. I'd suggest using interpreter, but in a case like this I think it'd require you to be running full speed, so this is the best we can do.

Other things to try:

FPRF in the game properties page.
Changing the Emulated CPU clock up/down to see if maybe that affects it.

Having a recording of both sides of the same match would also be nice to see what goes wrong.

Alright, sorry for the late response.
Unfortunately, changing the Emulated CPU Clock did not fix the problem (and under 33%, the game's framerate decreases, which is normal).
I tried pausing and disabling part of the JIT with the debug GUI, but it caused a loss of connection everytime, so I couldn't do that.
The FPRF didn't affect the problem either.

What you're going to have to do is try disabling parts of the JIT in the debug before the game connects and see if maybe one of those fix it while maintaining full speed. If not, does the game support recording replays or something like that? If the physics are desyncing, it may manifest in replays.

Okay, i'm going to try that as soon as possible.
The game does not have a replay feature, unfortunately.

Disabling "JIT Floating Point" fixes the rollbacks, the only problem is that the game crashes during shoot special moves and goalkeeping special moves, but by enabling and disabling it at the right time, the game does not rollback, and the goal is scored.

That's interesting. We need to find out which floating point instruction is broken then. If you can compile Dolphin and check out the JIT tables, you can disable things on an instruction by instruction basis.

If there any differences in gameplay outside of online play where we could study this more closely? Even if it's subtle (like a shot not moving right, a player sliding wrong, etc.)

I actually never considered that, I'll try to test and compare side by side to see if there's this kind of difference.

Hi,
I'm extremely sorry for the late response, but I managed to compile Dolphin, but I don't really understand what I'm supposed to do, is it possible for anyone to guide me through the steps please ?

It looks like it doesn't crash anymore (on the latest versions of Dolphin) with the option off (i just can't reach maximum framerate), it's just that I don't know what to do next.

After some tests, the root of the problem comes from fmadd, there are some inaccuracies with it but I don't have a way to debug on the wii to get some specific results at the moment

Did you test on a recent dev build? https://github.com/dolphin-emu/dolphin/pull/9748 emulates fmadd more accurately for jit64 and was merged 8 days ago.

That change does not affect the behavior unless you're using netplay/TAS or the interpreter.

It has been established that the issue can be avoided by doing FALLBACK_IF(inst.OPCD == 59) or FALLBACK_IF(inst.SUBOP5 == 30) in Jit64::fmaddXX. (This corresponds to fnmsubsx.)

Software FMA fixes the issue for some reason. Not sure why.