Emulator

The problem could be that the EFB is missing a blur effect.

I could merge this to issue 300, but this somehow needs to be verified.

The LoZ series seems to use the EFB efficiently (As does Luigi's Mansion and SMS as
we can see)

I looked at some documentation and found that between the EFB and the XFB, with the
exception of gamma correction and color value clamping, there is no filtering.

omegadox: You don't have to wait for donko to notice this, just Cc ;)

Maybe this is not this issue but...

I have problem with bubbles in Mario Party 4 in r2046 adn r2037:

http://heroesworld.narod.ru/mario_part_4_2.JPG

I will search when this issue happen...

This isuue inter r1842 and r1911.

Thanks XK, I forgot about Cc :P

@ ekshenman: This has nothing to do with the bubbles in MP4, those are just shaded
textures. The Bubbles in SMS are drawn from the copied EFB and they use some sort of
post-effect blur to mix the 4 colors in the bubble together.

omegadox: Wow you really love this Cc thing :)

Solves all our problems ;D. Maybe it can somehow fix Dolphin for us ;P.

omegadox: yep. But in MP4 this is problem too:)

Play with safe cache option, but I got them to work ok for me, but issue is not for MP4.

Would it be fine if i opened an issue to fix the fact that you can't clean the goo?
Anyone know what i mean? it makes the game hard to play.

There is a change that the game uses mipmapping tricks to do the blurring effect of
the bubbles.

Issue 3430 has been merged into this issue.

Please respond if this issue is still valid, or it will be closed.

afaik its still valid... was there a commit lately, that should have fixed it?

These bubbles have IA8 format, and to me it looks like mirroring issue, before the mirroring was fixed i have the same looking icons in Wii Menu

For reference, the bubbles don't have an IA8 format...

The bubbles are created using RGB5A3 EFB copies which can be proven by disabling those (the bubbles won't appear then). The EFB copies get reused as RGB5A3 textures lateron.

Uhm, sorry I need to correct myself - they're actually RGB565 EFB copies which are used as RGB565 textures.
Fwiw, the source EFB rectangle is 640x448 which is being scaled down to 320x224.

So, is there a way fix the bubbles now, Neo?

.. if I knew where the problem is, I would've said so.. ;P

Here's my guess: The pixel shader used to render those bubbles is simply using the wrong texture coordinates for sampling from the copied EFB texture. The reason it looks like a checkerboard is that it's incorrectly sampling around the corner of the texture, so you're seeing colors from all 4 corners of the screen in the bubble. If the texture coordinates were correct, you would instead see colors coming from near where the bubble is onscreen.

That's always possible but not entirely likely, I don't think. A texture coordinate of -1 might wrap around to the other side of the texture, but more often than not I find that kind of crap throws an error.

I don't think texture wrapping can throw an error... the exact coordinates aren't known until the pixel shader runs, and pixel shaders don't have time to be checking and throwing errors like that. And it's quite possible that the sample mode is set to wrap instead of clamp in this case. You can easily verify that the 4 colors in the bubble are coming from the corners of the screen by moving the camera around while bubbles are visible.

i've not tried to move the camera around, is it coming from viewable area or not? perhaps it is meant to sample from an area of the framebuffer which is meant to be offscreen or not displayed at all. (many effects work like this already, so it's doubtful...)

It's sampling the whole visible scene before the goop and HUD is getting drawn.

Tried the bubbles on software
so here it is
http://i5.minus.com/iZ4ivIaDE1qXL.jpg
Pretty invisible, but this is the way they're supposed to be

For reference, the shader that is being used in OGL is http://pastie.org/private/jjt5qzj5odlcwdytanbva . Most importantly, "tevcoord.xy = wrappedcoord + indtevtrans0;" is causing problems. If I force indtevtrans0 to zero, the bubbles show up fine. Given that the software renderer uses 24 bit values for the final texture coordinate, I guess this is once again a failure to properly emulate the bitness of intermediate values.

Fixed in tev_fixes_new.