Ghost In The Shell - Stand Alone Complex (ULUS10020) - Black Textures and missing screens.

[piau9000]

What happens?

Some textures are black and some screens are completely missing (also black screen). The missing screens includes the green text intro at the beginning and chapters names. It's happening since 1.6.3-454. It seems #11380 broke it.

What should happen?

The proper image should be as follows:

What hardware, operating system, and PPSSPP version? On desktop, GPU matters for graphical issues.

It happens on any version after 1.6.3-454.
Windows 7 64-bit, GTX 750 ti, 8gb RAM.

[Panderner]

Have you tried every backends?

[piau9000]

Yes. There's no difference between opengl, vulkan or direct3d. Tried several other graphical settings, but none of them makes a difference.

[hrydgard]

Thanks, good report! Especially tracking down the change responsible.

This should just be a matter of handling bad fog parameters a bit differently. Could you get a GE capture of the situation for easy investigation? https://github.com/hrydgard/ppsspp/wiki/How-to-create-a-frame-dump

[piau9000]

Here is the dump file:

recording.zip

[iota97]

The game render with:

• u_fogcoef = {65535.0f, -65535.0f} // Note the second one is negative
• u_fogcolor = {0.0f, 0.0f, 0.0f}

Vertex shader does this and get a negative number:

v_fogdepth = (viewPos.z + u_fogcoef.x) * u_fogcoef.y;

Fragment shader then clamp to 0 and just render the fog color (black):

float fogCoef = clamp(v_fogdepth, 0.0, 1.0);
v = mix(vec4(u_fogcolor, v.a), v, fogCoef);
fragColor0 = v;

Frame dump (I hope is dumped correctly):
GhostDump.zip

Set both fog coef to same sign make it render fine.

[hrydgard]

Ok, nice debugging! Not quite sure if the game is intending to do something special with those settings or if it's just a game bug that happens to work on the hardware.. so exactly what the correct fix here is I'm not sure..

[iota97]

Here there is a gameplay on read HW (https://youtu.be/TDlABQogoqw?t=240):

It seem pretty the same as with no fog.
I wonder the real HW handle it as some kind of overflow skipping it given the max Esp on both value (I have no idea how psp hadle it tho', so just wondering).

[unknownbrackets]

Hm. So the PSP game itself is using:

Fog End = INF
Fog Density = -NAN

When I use those same parameters in a simple test, I also get black. What's more, when I run that dump on a PSP, I get...

Compare to our rendering (softgpu):

I wonder if the actual problem is somewhere else, where it's calculating these values to even send to the display list.

-[Unknown]

[unknownbrackets]

If you run this game using software rendering (try to use an ingame save if possible), does it render properly?

If yes, it'd really help to have a GE dump while running software rendering too.

-[Unknown]

[iota97]

Here a dump from 1.9.4 Software rendering, same problem (uploading the dump anyway recording.ppdmp.zip)

This game seem to segfault (on linux at least) at start on master, will bisect later for the right commit.

EDIT: 54e1afd is the first bad commit, guess was expected to be this one.

[hrydgard]

@iota97 do you see anything interesting in the log console?

[Panderner]

This how it looks on a android phone running on OpenGL:



Anybody have you tested any backends on latest build version?