GSdx

[Nobbs66]

I'll do some testing once I really wake up here in a few minutes.

[mirh]

m_userhacks_RemoveHoleBetweenSprite was a previous workaround that target only ace combat.

Just for the records, I compiled the second version of your patch, which tbh seems equal to the one you posted here

[gregory]

Yes I think impact will mostly the same for Ace combat.

@Blyss,

You said that it was possible to change the filtering (disabling?). What do you mean exactly? Which option?

Maybe we could do a hack that uses linear filtering for sprites instead of bilinear filtering. Sprites will have more block but it would avoid scaling issue on some sprites. In others word, it will annihilate the anti-aliasing effect of scaling for sprites.

Edit: maybe it is possible to detect edge with the alpha value of the texture. It might worth to give it a shoot

[Blyss Sarania]

Yes I think impact will mostly the same for Ace combat.

@Blyss,

You said that it was possible to change the filtering (disabling?). What do you mean exactly? Which option?

Maybe we could do a hack that uses linear filtering for sprites instead of bilinear filtering. Sprites will have more block but it would avoid scaling issue on some sprites. In others word, it will annihilate the anti-aliasing effect of scaling for sprites.

Edit: maybe it is possible to detect edge with the alpha value of the texture. It might worth to give it a shoot

I meant the normal texture tick box in GSdx config.



For instance in Persona 3, it uses the same sort of "Base Body Sprite" + "Facial animation Sprite" setup. With texture filtering on(full tick) it has boundary issues similar to but not as bad as Ar tonelico. Setting it to off, or shaded(e.g. filter 3d only) will fix it.

However this does NOT help Ar tonelico. The gaps are much bigger in it. The gaps in Persona 3 are maybe 1-2 pixels on one edge. In Ar tonelico games it's more like 2-3 pixels on ALL edges. They are there with any level of upscaling, but not at native HW or in SW. It's definitely an upscaling precision thing.

[gregory]

Ok could you check the option name in gsdx.ini file?

[Blyss Sarania]

I think it's this one:

Code:

filter=1

[gregory]

Ok. Thanks. In my humble opinion the option is wrongly implemented. Well I don't know the reason of this option.

The PS2 samples at full integer position for linear interpolation. But bilinear sampling have an extra half-pixel offset. I think a hack that disable bilinear but keep offset for sprite might help to improve the accuracy (but quality will be bad).

Could you generate me a dump of PS3 where this option help?

My guess is that AT contains severals upscaling issues (likely primitive position+texture wrong filtering)

[Blyss Sarania]

Ok. Thanks. In my humble opinion the option is wrongly implemented. Well I don't know the reason of this option.

The PS2 samples at full integer position for linear interpolation. But bilinear sampling have an extra half-pixel offset. I think a hack that disable bilinear but keep offset for sprite might help to improve the accuracy (but quality will be bad).

Could you generate me a dump of PS3 where this option help?

My guess is that AT contains severals upscaling issues (likely primitive position+texture wrong filtering)

Alright, I'll make a Persona 3 dump in a bit.

[dabore]

@gregory have you tried to border_clamp the textures. just an idea. that might work with the linear filtering. i'm aware it needs a full seperate texture tho. it won't do that with "sub textures" like "sub tile" stuff. and linear or bilinear doesn't matter imo. it's planar 2d. done linear. bilinear is only of interest for min and magnification. for 2d it only gets more blurry. cause of the double filter. ugly imo. i'd not try to fix that. and to fix the blended sampling into the "nothing at border" you'd actually have to start to go down to point sample to get an accurate texel center. and you have to fiddle that with the vendor specific hacks. pretty screwed. imo. i'm aware of that.

and another idea is to do that sprites with subtextures with pointsampling... so they border clamp automaticly and you implement the linear interpolation with a shader. you'd only need the sprite rect to choose whether to sample and lerp inside or don't cause of "out of rect". might be slow with the shader tho. tmu might be faster.

[f8708136]

Ok. Thanks. In my humble opinion the option is wrongly implemented. Well I don't know the reason of this option.

The PS2 samples at full integer position for linear interpolation. But bilinear sampling have an extra half-pixel offset. I think a hack that disable bilinear but keep offset for sprite might help to improve the accuracy (but quality will be bad).

Could you generate me a dump of PS3 where this option help?

My guess is that AT contains severals upscaling issues (likely primitive position+texture wrong filtering)

Hello, can you fix this game?

  1.jpg (Size: 659,91 KB / Downloads: 433)

and

  2.jpg (Size: 214,05 KB / Downloads: 391)


I use your hack but still have some vertical line.