Jurassic Park: transparency gitch on the first stage

[vanfanel]

Hello there,

Running latest FBNeo GIT code here.
I've noticed a transparency glitch on the first stage of Jurassic Park, where brown-ish transparent bands move up and down with the cursor.
I recorded a video of it:

jpark_glitch.mp4

Seems like a Sega System 32 emulation bug.

[dinkc64]

Hi vanfanel,
Here's how it's supposed to look:
https://youtu.be/Mx0fHZOThlo?si=pflfqBvwG1VW1Y6R&t=153

I'll try to figure it out, but I can't promise anything, the Sega System 32 is kinda complex!

best regards,

• dink

[mahoneyt944]

I believe it's related to row scroll. I managed to get it working for player 2 but not player 1 in mame2003+. Player 1 might just be a priority issue though as its not shown at all but I never got any further with it. Maybe this will help here though.

https://github.com/libretro/mame2003-plus-libretro/blob/3206819c48567e759f993134ac27491b11b3dd80/src/vidhrdw/system32_vidhrdw.c#L1327-L1331

[dinkc64]

Hi mahoneyt944, thanks for the tip!
I'll see if this can be improved upon,

best regards,

• dink

[mahoneyt944]

@dinkc64 I went down the rabbit hole a little by porting newer segas32 video code over to mame2003+.... I get the same issue noted here. Curiously it's not broken for player 2 in any code base, so I'm hopeful that it's a minor adjustment needed for player 1. This issue has been present since the beginning of this video code (around mame 97u2), so it's unlikely my previous comment will help here. Player 2's render might point us in the right direction though.

[mahoneyt944]

So this effect is created by update_tilemap_rowscroll(&layer_data[MIXER_LAYER_NBG2], cliprect, 2); during update_tilemaps() and seemingly works for player 2 (when player 1 is alive) but doesn't work correctly for player 1.

[mahoneyt944]

I found this video which describes hooking up a generic jamma harness to a jpark board and notes that moving the joystick up and down changed the camera view, and button 2 lowered the health bars. I confirmed this behavior by swapping the inputs on jpark with a generic two player set. This makes me think that the "flash light" effect issue in the raptor tunnel might not be the video code at all but an input issue since the effect follows the positioning of the input.

Panning higher and lower:

Video starting around 1:50
https://youtu.be/l__YufFcaSQ?si=7NPpF26CkZO8tiwz

[dinkc64]

Nice find :)

[dinkc64]

lately I've been up a barrage of sh*t, which explains my inactivity. When things hopefully get better, I'd really like to check into your changes and possibly help out, or at least merge the fixes you've made. I really dig the work and findings you've come up with on s32! :)

best regards,

• dink

[mahoneyt944]

If you want to try my fix up for the backgrounds in harddunk we use this macro in place of the utility function in current mame. You'd just need to port the dozen or so places it's used then add the change for the resolution in update_tilemap_zoom(). Here's the PR in Mame to reference

mamedev/mame#13047

/*************************************
 *
 *  Utility Macro
 *
 *************************************/

#define SEXT(value, width) (((int64_t)((value) & ((1ULL << (width)) - 1)) ^ (int64_t)(1ULL << ((width) - 1))) - (1ULL << ((width) - 1)))

[dinkc64]

Thanks :) (I also have a sign extender macro in drv/snes/cx4.cpp)
I have an idea I'd like to explore with the 0x200 default, maybe there is a problem with the math up above which is requiring the sign extend change to begin with? Could be wrong about this, but, it's worth investigating.

best regards,

• dink

[mahoneyt944]

If there's not a math issue, I also theorize that the sign extend amount might be dynamically set based on the zoom value instead of hard coding 9 and 10. But these values seem to at least generate the correct result.

Let me know if you figure anything out.

[mahoneyt944]

Also a side note, it seems one of the main issues in general is sprite transparency. I noticed in rad mobile the title should be transparent when it flips. I was able to get a real hardware capture of this vs the emulation of. Edit: I think the blendmask needs controlled by a register?

There's also sprite priority issues such as the cars showing in the rear view mirror are displaying over the mirror

[mahoneyt944]

Some more digging with jpark,

The effect width for player 1 is being set here. There seems to be a conflict with this extent and the one used by player2. This effect does not track with player1 X analog. You can also override the min and max here to 100 and 200 respectively to see what I mean. There's seems to be an error with this extents list.

FBNeo/src/burn/drv/sega/d_segas32.cpp

Lines 2753 to 2757 in 95c4eea

	else
	{
	*extent++ = cur.nMinx;
	*extent++ = cur.nMaxx;
	}

[mahoneyt944]

Stumbled across this for the rad mobile title screen transparency issue noted previously, apparently the astro city mini's emulator handles this transparency more correctly. I believe this is using flycast or a branch of.

https://m.youtube.com/watch?v=gRmiNvFIU5U&pp=ygUaQXN0cm8gY2l0eSBtaW5pIHJhZCBtb2JpbGU%3D

[dinkc64]

I've dug into this so many times, it's insane!

"at first you don't succeed, try, try, try, try, try, try again" - jack o'neil from stargate sg-1

trying again tonight :)