Changes in Dolphin since Revision 6000~

[hungry man]

I haven't used Dolphin since around 6000~ what's changed? I realize there are changelogs, but I'm talking about big picture. I've heard performance is actually worse for the sake of accuracy?

[neobrain]

Some list I created some time ago:

Quote:6000-7000:
ClearScreen fixes (my favorite one).
EFB format changes emulated (my 2nd most favorite one).
Implemented EFB color pokes (my 3rd most favorite one).
MSAA+texture dumping+hires textures+debugger in D3D11.
FIFO fixes.
Texture decoding vastly sped up.
Audio+Video plugins merged.
CMake introduced (think that was after r6000) for Linux.
LLE vastly sped up.
XAudio2 sound backend.
Implemented DL (display list) caching.

Also: EFB to RAM and Real XFB in DX11.

[Starscream]

The biggest problem with the new builds is that AA is broken in some cases. That is a deal breaker for me and renders all new builds not worth using until that gets fixed.

[hungry man]

Pretty cool. AA isn't a big deal for me. I could only ever use it on GC games anyways.

[ladsk10]

and no fog in metroid prime.. :/

[SantaC]

AA is a big deal. Is it broken in both dx9 and dx11?

[neobrain]

SSAA is only broken in D3D9.
Also, it has always been broken, it's not even a regression actually. (Erm, and don't try to argue over this, it's a fact which I explained thousands of times to NaturalViolence ;P)

[NaturalViolence]

Quote:Also, it has always been broken, it's not even a regression actually. (Erm, and don't try to argue over this, it's a fact which I explained thousands of times to NaturalViolence ;P)

Yup. Broken but working perfectly....*rolls eyes*.

Broken: http://i112.photobucket.com/albums/n189/NaturalViolence/fract-4xSSAA.png

I mean just look at all that aliasing! /end sarcasm

Things we still have to explain:
1. You said that the only thing 4xSSAA was doing before 6549 was applying a 2x scale. If that's true then why does a 2x fractional scale after 6549 look totally different and not eliminate aliasing?

2. If it really is the shader that is broken like you said then why was it working perfectly before 6549 (eliminates aliasing without any artifacting/blurring, keep in mind your +/- 1 offset theory has now been dis-proven). If the shader was broken then why did it work?

3. Why does implementing the 4 sample shader not work (causes blurring, which you would expect since it's just blending 4 texture samples).

Until you can answer all three of those questions it logically has to be a regression.

4. SSAA is supposed to be a scale + post-processing box filter. Implementing a post-processing box filter through nvidia inspector after 6549 produces the exact same image quality as native SSAA before 6549. Why would you want to *fix* SSAA by changing the vertex/pixel shader then?

And so once you admit it was a regression you just need to answer:

1. What was changed in 6549?
2. How could any of those changes have affected SSAA?

Only then can we determine a fix.

You managed to convince me that it was not your fault it was the crappy shaders fault. But once I started testing I quickly realized that doesn't make any sense since things weren't "adding up".

Sorry to abuse the thread this way by the way. Sometimes I can't help it.

[neobrain]

It's awesome how you're asking the same questions over and over again. I don't feel like answering for the thousand-and-first time now ;P

---

Fwiw, note that it wasn't applying a 2x scale before r6549, but rather a 1.997 scale or sth like that. That's relevant, since it's the only reason why you perceive SSAA as "broken by r6549".

[NaturalViolence]

Then why didn't the +/- 1 offset fix it? And why are you mucking with the shaders to try and fix it?