NFS: Most Wanted: Graphics Dust, Smoke, Lights Glitches on last update.

[patocius]

Game or games this happens in

Tested in Most Wanted and Pulse

What area of the game

Usually happend on lights and smoke.

What happens

You can see a video of the issue in this link.

https://cdn.discordapp.com/attachments/869601237522337862/907711236660150272/VID_20211109_161923.mp4

What should happen

Lights and dust must showed properly like in other versions of ppsspp

GE frame capture

Platform

SDL2

Mobile phone model or graphics card

Anbernic 351P (351Elec)

PPSSPP version affected

v1.12.3-182+g4d3890547

Last working version

v1.11

Graphics backend (3D API)

OpenGL / GLES

Checklist

• [ ] Test in the latest git build in case it's already fixed.
• [X] Search for other reports of the same issue.
• [X] Try resetting settings or older versions and include if the issue is related.
• [X] Try changing graphics settings to determine if one causes the glitch (especially speed hacks or enhancements/replacements.)
• [X] Include logs or screenshots of issue.

[patocius]

Another related issue is when you skip buffer effects the game got unplayable, gui and graphics glitch harder than in other versions of ppssspp.

[ghost]

Related https://github.com/hrydgard/ppsspp/issues/15056? This issue cannot reproduce in other graphic backend right?

[dhwz]

That's not the libretro version, that's the SDL2 version used in the screenshot.

[patocius]

Related #15056? This issue cannot reproduce in other graphic backend right?

I didnt test it in other graphic backend.

[patocius]

That's not the libretro version, that's the SDL2 version used in the screenshot.

Fixed the report and put it as SDL2 thx dhwz.

[unknownbrackets]

Please include a GE frame dump, which is like a 3D screenshot: https://github.com/hrydgard/ppsspp/wiki/How-to-create-a-frame-dump

You've instead attached a 2D screenshot. That's not very helpful.

Also, there are about 1800 different changes between the versions you posted. Without a GE frame dump, we're left guessing. If you could try a git build in between or try the stable version (v1.12.3), that would help at least narrow it down further.

-[Unknown]

[patocius]

Please include a GE frame dump, which is like a 3D screenshot: https://github.com/hrydgard/ppsspp/wiki/How-to-create-a-frame-dump

You've instead attached a 2D screenshot. That's not very helpful.

Also, there are about 1800 different changes between the versions you posted. Without a GE frame dump, we're left guessing. If you could try a git build in between or try the stable version (v1.12.3), that would help at least narrow it down further.

-[Unknown]

ok will try tomorrow thx.

[Panderner]

Here's a GE dump: DUMP.ppdmp.zip

[Panderner]

I can reproduce this issue for me:

[hrydgard]

@Panderner what's your device again? I'm having problems getting this to repro.

Just to confirm, you can repro directly by running the GE dump?

[Panderner]

@Panderner what's your device again? I'm having problems getting this to repro.

@hrydgard i'm using POCO M3

[unknownbrackets]

Based on the positions of the artifacts (which I also don't experience), it looks like maybe draw 398/462 (or maybe 397.)

This should be a texture from RAM that shows as a light spot. Maybe there's some interaction with creating framebuffers in RAM here or something? Not sure if that flag is on for these games.

Another interesting thing: a VRAM texture is set at level 1, but mipmapping is not enabled. I don't think we have a bug there in GL, though...

Anyway, the texture in question is only 32x32, so that's why I kinda suspect RAM framebuffers, the artifacting in the screenshot looks higher res than that, though it is set to wrap textures...

-[Unknown]

[ghost]

I cannot reproduce this on my phone 🤔

[hrydgard]

Bisected a bit. Weirdly, the HUD text changes color from white to black, as reported in #15212, in commit 5315c404c55778b48ab7dbf41295ed3904d0ce54 . Also, a lot of HUD stuff indeed breaks in that commit. Not sure if the smoke is related or not, having trouble getting a solid repro of that.

[hrydgard]

Unknown's related comment in Discord:

I added software transform culling but I didn't actually suppress the incorrect triangle culling that happens on the split triangles because it'd mean passing in prim as a uniform value and I hoped it wouldn't matter...

Wait, which split triangles? you mean splitting rectangles into triangles? And the depth part of it won't be that incorrect, right? These particular rectangles don't seem to be anywhere close getting culled at the sides....

Alternatively, instead of a uniform for prim, maybe we can squeeze a bit into the vertex attributes somewhere... Although a uniform is probably fine too.

[hrydgard]

To get the current master to behave exactly like the build prior to the commit that "broke" this, comment out the following two lines in SoftwareTransform.cpp, in the "else" case of if (!gstate.isDepthClampEnabled()) { in ExpandRectangles:

                if (tlz <= minZValue && brz <= minZValue)
                    continue;

So things get wrongly culled when depth clamp is enabled, and the two depths are below the minvalue. Here we haven't even reached the shader culling yet.

It's very weird to me how this change causes the effects we see, but there might be some offscreen processing step of font textures that break, or something.

[hrydgard]

Closing this as fixed after #15235, since correctly culling rects/triangles is not really something that should affect this once it's working.

[Maarcis12]

thank you very much, that fixed it indeed.