Fascinating new anti-aliasing scheme

[Ed Oscuro]

In looking at updates for the Dark Souls injector, I found this page linked in the comments.

And in that page, you find a link to InjectSMAA, which allows you to run SMAA in many programs.

What is SMAA, you ask? It is a new form of anti-aliasing which looks MUCH better than previous styles, but doesn't cost much more in performance. Take a look at the screenshots. In the Unigine Heaven DX11 benchmark, SMAA loses only 4 fps (from 123) but looks markedly better. And the improvement on Halo (and I assume Quake would benefit as well) is remarkable. I hope ports of older games start implementing this soon. It would also be nice if this could be forced from a graphics driver's controller, too.

[Ex-Cyber]

I hope this doesn't turn out to be patented, like Carmack's Reverse did.

[Cuilan]

I hope this doesn't turn out to be patented, like Carmack's Reverse did.

Seriously. At last, a non-resource-intensive post-processing AA method that doesn't have the same result as turning the sharpness down on your display. This would be especially useful on consoles, where system resources are very limited.

[Ed Oscuro]

Well, that first page I link shows that the code and techniques are being freely distributed, and it's a Brazilian (apparently) academic project, so I don't think there's really any chance of it being patented. This is a different situation than Carmack's Reverse, where some guys working for Creative - remember when they were a competitor to 3Dfx? - developed the technique while working there. Here's a bio of half of that team, Mike Songy.

[Raytrace]

while in general graphically im a PC snob, I have found it intriguing the various methods of AA which have come to the fore with consoles this generation

[Obiwanshinobi]

To be fair, some fullscreen antialiasing routine in certain PS2 games impressed the hell out of me (Baldur's Gate: Dark Alliance, The Bouncer, Contra: Shattered Soldier...). Depends on the TV and cable to an extent of course, but such clarity in 480i was remarkable. Super Monkey Ball is another last gen game looking clean above average on my telly.
Technically it might have been fullscreen blur rather than "true" antialiasing, but in those few games worked a treat.

[Ex-Cyber]

Well, that first page I link shows that the code and techniques are being freely distributed, and it's a Brazilian (apparently) academic project, so I don't think there's really any chance of it being patented. This is a different situation than Carmack's Reverse, where some guys working for Creative - remember when they were a competitor to 3Dfx? - developed the technique while working there. Here's a bio of half of that team, Mike Songy.

I'm saying that you don't actually know that it's different. Carmack discovered the technique independently and disclosed it while Creative already had a patent pending. He and the public had no idea there was a patent until Creative came forward to negotiate a license. The only way you can know for sure that something isn't patented is to have prior art older than the length of a patent term.

[Ed Oscuro]

I'm saying that you don't actually know that it's different.

And I haven't said anything that suggested I had proof - I can't prove a negative here! I am more interested in what's likely.

I did a bit more research on this subject, so I can expand further on my reasoning why the "Carmack's Reverse" stencil shadowing situation will not repeat itself.

The Creative Labs situation is quite different from today. In 1998 plenty of cards were still fill rate limited, so funding research on something like stencil shadows was clearly meant to strengthen their future position in graphics - which did not materialize, because shortly after Creative had left the market.

Two of the biggest players in desktop games graphics, AMD and nVidia, have continued researching AA algorithms, and given the nature of their business, they cannot push competitive advantage too far without burning developers (see also: PhysX market takeup) but at the same time they absolutely must do in-house research on these processes, to forestall that process if nothing else. In a short reply to you, I would say this: Both parties have released things like SLAA, and nobody has breathed a word of patenting any of these approaches.

AMD's MLAA has been out there for a while (Intel's presumably is similar), and is not comparable to SLAA (the focus of my first post) in quality or speed (although we should note that it is, like the processes that followed it, a sub-pixel post-process effect; I believe all three and their variants are pixel shaders). However, nVidia's SRAA claims about a tenfold increase in speed (to a 1ms delay for processing) over MLAA. I don't know nVidia's stance in the area of software patents, but in recent times a former nVidia employee who held a compression patent granted immunity to open source projects that utilized his technique. On top of that, the SMAA guys are claiming it is a different technique from nVidia's SRAA - a good example is here on their MLAA page, where they write "In order to avoid further confusion between the different MLAA implementations, we named ours Jimenez's MLAA." Aside from the timeframe being too tight for a patent-based challenge, their approaches have been used in shipping games, which again doesn't bode well for any future patent-based challenge.

These guys are well aware of what SRAA is - they have a hosted copy of the nVidia team's SRAA presentation for SIGGRAPH '11. I find it hard to believe they would do little work to differentiate their approach and would then court disaster by hosting a copy of a damning document! Rather, I believe (or at least they do, whether courts would agree is of course a different matter) that their approach is different enough that it would not be covered by any (so far mythical) patent from nVidia, Intel, or a third-party dark horse.

Here is a snippet, with some CUDA code, of nVidia's white paper describing their SRAA implementation.
For comparison, MLAA is also sub-pixel post-processing but with comparatively terrible performance and not-so-great visual quality.

The bottom line is that while anything's possible, it does not look like any techniques are being duplicated as closely as Carmack's Reverse happened to be.