High Quality Anisotropic Filtering

[EmbraceUnity]

I just saw this commit.

http://cgit.freedesktop.org/mesa/mesa/commit/?id=3996e493693d5aa048fed1c2fd6db4027cb47df4

It claims that it is an undocumented feature only useful for benchmarking. Yet in the Catalyst Control Center on Windows I have always been able to turn this on no problem.

Why not turn it on by default?

[bridgman]

I think the addition here was an extremely high level of aniso, presumably higher (and slower) than you were able to set in CCC. Just a guess though...

[marek]

Yes, the level of aniso is extremely and needlessly high. I won't enable it by default because I can't see a difference (unless consecutive mipmaps are completely different images). The standard 16x aniso gives outstanding quality already.

-Marek

[curaga]

How high are we talking? 64x?

[Alex_V]

I think the addition here was an extremely high level of aniso, presumably higher (and slower) than you were able to set in CCC. Just a guess though...

Hmm, based on my recollection of the hardware, and what went on back in the day, I'd guess this is more likely to be the exposing of less angle variant sampling rather than a higher sample count. It was available in the Windows drivers as a separate option for R5xx hardware, and was one of the perks back in the time. It became the standard/preferred sampling pattern for R6xx and upwards. This should represent it accurately, methinks: http://www.beyond3d.com/content/reviews/27/13

It was expensive back in the day, mainly due to low TU counts and small tex caches.

[marek]

32x, maybe.