VU1 sample syncronization problem

[Fewnity]

I'm currently trying to draw a more detailed 3D model in the VU1 sample. And for this we have to make more packets every frame to render the scene. But this will cause flickering/missing models on real hardware. (we can get the same problem just by adding more cubes to render in the scene).

According to several people, it's a syncronization problem with the VU1. Like I'm sending data while the VU1 is still rendering.
I tried to add a flush to wait for the VU1 but without success. I'm desperate not to find a solution.

Fixing this problem will help people to create a more complexe project by using the sample as a base so I think it's interesting to fix this.
I planned to add PS2 support in a game engine.

PS2 vs PCSX2

[F0bes]

Can you provide minimum reproducible code for this issue? If the issue is what you say it is, you might just need a FLUSHE in your VIFCode (or a bc2t instruction loop).

[Fewnity]

vu1.zip
@F0bes Here is the vu1 sample (the normal one) but with just more cubes. On emulator you can see few cubes blicking and on real hardware a big amount of cubes are not rendered correctly. I tried many fix, I hope someone can figure out how to fix it :/

[F0bes]

You'll need to optimize, a vblank interval is happening during your drawing because it's too slow.

Here is how I can prove this:

You're doing 1224 cubes.
I used cop0 count and determined that draw_cube call is taking around 5186 cycles on pcsx2. ~9000 on my PS2!

I'll use the PCSX2 number for the math below.

So that's 6,347,664 cycles / frame.
The EE is ~300 Mhz optimistically so that'll take around 0.021 seconds, or 20 miliseconds per frame. (Divide the total cycle count by 300,000,000)
The time per frame (1/60) for 60fps is 16 miliseconds per frame. Faster that what you can render currently.

You can test this yourself, make z only have 12 iterations. This will give you around 14 miliseconds / frame. If you try it on PCSX2, there will be no flickering :)