The 3rd Birthday (fog in some GLES gpus)

[brujo5]

in beffered rendering (block transfer on/off is the same,also dissable alpha mask hack do nothing)

in read buffer to memory (gpu) looks a bit better

test build 0.9.8-1574

[aidan0203]

I assume you're not in Windows, are you? Looks fine in here.

My specs:

[brujo5]

My device,iPhone 5S los 7.1.2

[unknownbrackets]

I wonder if we need highp (like for fog) on PowerVR for depal?

-[Unknown]

[unknownbrackets]

Has this improved at all in the latest version (v1.0-101-g1170331 or newer)?

-[Unknown]

[brujo5]

Will test tomorrow early.

[brujo5]

@unknownbrackets
Think fixed, what you think?

[hrydgard]

That looks right. Although the colors are slightly different from the Windows shot above, that could come down to the gamma difference between mac and windows when taking the screenshots.. maybe.

[brujo5]

@hrydgard

Here in non-befered rendering

Here in befered rendering + slow effects ON+ GPU block transfer ON

[unknownbrackets]

That doesn't look any different from before. I think it's just that you had slow effects disabled (which also disables the bloom entirely.)

-[Unknown]

[brujo5]

Yeap that whit befered rendering+ slow effects offs
Note: Powervr alphatest off/on is the same

[unknownbrackets]

The above screenshot doesn't really look wrong to me. It's a bit bluer and darker as hrydgard mentioned. Closing.

-[Unknown]

[daniel229]

The fog is not rendering in above screenshot.

[unknownbrackets]

FWIW, checked quickly - this is what would happen if depal resulted in a completely white texture. It doesn't seem related to dual source blending or framebuffer fetch or anything.

The stages of depal are (may be noted elsewhere, but just to recap):

• Screen -> temporary, extracting top 4 bits of G from 8888 to 4444, but only the R channel has values in CLUT.
• Temporary -> temporary, to blur using blending. In this case all components are set to the same value as R, creating a grayscale blurry version of the top 4 bits of G.
• Temporary -> screen, again mapping R to all components (this is why dualsrc doesn't matter.)

White could occur if the depal index became 15 for all pixels, since that color is white.

Also worth noting is that a single temporary is used, so the temporary -> temporary step is render-to-self (but since it's depal, we should be using a temporary buffer...)

-[Unknown]

[unknownbrackets]

How does this look now?

-[Unknown]

[psennermann]

On my Samsung Galaxy Tab 2 official PPSSPP 1.5.4 still has this problem...is there a particular reason (presumed fix) to test a new development version?

[unknownbrackets]

Latest git builds have lots of changes to OpenGL, some state handling fixes, and recent fixes to depal handling. So yes - worth testing, I think. Will soon be 1.6.0.

When it still has this problem - does it still have wrong fog, or missing fog?

-[Unknown]

[psennermann]

Ok, I've tried 1.5.4-971 and from what I can see the problem is fixed only using Vulkan, while OpenGL still has the "overbrightness" (fixed only by disabling slow effects)

[unknownbrackets]

Okay - it could be the depal stuff then, we have some disabled code to do the same thing for OpenGL we do for Vulkan (on GPUs that support it), but we don't plan to enable until after 1.6.0.

Noting #10908 here so we remember to take another look. Hopefully it's related to that.

-[Unknown]

[psennermann]

I've tried official 1.6.3 and I think that now OpenGL is fixed...

[unknownbrackets]

Great. Going to close this then.

-[Unknown]