WIP r830 Discuss (do not post builds)

[Joaquim Gonçalves]

What's wrong with wide screen mode?

What's wrong with wide screen mode?

Hy Bart, long time since I posted here

Widescreen works only with geometry, in the case of the background (clouds, sky etc ...) that seens to be a kind of image bmp or something like that, we have this problem with the bg´s that will not be drawn on the left and right sides.

This issue happens only with games that dont have a skybox done with polygons meches. In the VF3, the bg´s seens to use this kind of aproach using images, or a texture with a fixed aspect ratio.

On the model 2 emu, to get around with this same problem with bmp bg´s tilemaps, elsemi put an option that stretches the bg TileMap using independent settings for each game. In this case, each game has a .lua script file that can be edited.

Daytona USA example:

Code: Select all

Model2_SetStretchBLow(1)   -- Stretch the bg tilemap (sky & clouds) when widescreen

Yes LsrgcBath, thanks.

Watch here.
https://www.youtube.com/watch?v=N4Xa6drrQD8



In case if we put 4:3 the screen don´t show that mess.

[Ian]

The texture mapping problem plaguing three of the levels (Lau's, Pai's, and Akira's) hasn't yet been solved. If I can find some time, I'd like to look into that further.

Bart if you check the poly number in the poly headers, you'll see virtua fighter is still parsing a large amount of junk data. The original bug, is very much still there, although drastically reduced with your fix. I think this is where the extra polys are coming from in some of the stages, like the Jacky stage which is almost unplayable because of the massive poly traversing the scene. (If I was to guess the texture bug might be related).

I could probably improve the situation by, throwing out any meshes that doesn't start with 0 for the poly number. But that might actually lose parts of the scene too.

[Ian]

Heres a few debug screenshots with the current build

http://imgur.com/a/YemVK

Looks like its either parsing completely junk data, or missing the start of models still. The 2nd screenshot, 94 polys seem to be missing.

[Bart]

Can't seem to zoom the screenshots on my phone here but the polygon on Jacky's stage was a VROM model, I thought? VROM models are not transferred by DMA. Also, a week or two back I spent a few hours investigating textures. It would require a solid few days of work -- the problem appears not to be DMA related. The game code doesn't even appear to try to load the level textures. It loads up 256-byte level descriptors in RAM and one of the bytes that seems to control which set of stage textures to load is inexplicably set to 0. I tried manually changing it but couldn't identify any value that would find the correct textures. It's a big piece of the game engine so slogging through the disassembled code will take a huge amount of time.

[Ian]

I guess I could confirm it is a VROM model by simply disabling all VROM models
I

[Ian]

I just tested, and you are right Bart, it's a VROM model



Didn't actually realise so many of the models were VROM models

[Bart]

I was able to find the offending polygon using Nik's 3D debugger. I forgot to save the address but at the time, nothing about it looked conspicuous to me. It's possible, I suppose, that the transform matrices are somehow wrong or aren't being uploaded due to a DMA or other bug. It would be hard to investigate this. I can't think of any reason why the game would create a scene graph node with this model if it is intended to be invisible...

VROM models can readily be identified by their address, by the way.

[Ian]

Might it be easier ignoring it and checking why its passing bad poly data?
At least that's fairly easy to catch in the debugger
It does it on the 2nd scene in attract mode

[Bart]

Might it be easier ignoring it and checking why its passing bad poly data?
At least that's fairly easy to catch in the debugger
It does it on the 2nd scene in attract mode

I think the data must be valid if it appears in VROM. Not sure that I follow.

The polygon obscuring Jacky's stage (and probably the similar one on Akira's stage) is deliberately placed there in the scene graph. Either we are rendering it wrong or the scene graph is corrupted. The latter is always harder to figure out. Could be similar in nature to the Daytona bug you observed, with the scene periodically dropping out due to a corrupt matrix.

[Ian]

I briefly tested attract mode again by commenting out all ROM models
the bug is definitely happening in poly ram (as well ? )

Maybe the jacky stage is a separate bug

http://i.imgur.com/agogSaB.jpg