No more Trilinear in the future... just anisotropic

[arjan de lumens]

Bicubic interpolation is about 4-5x as computationally expensive as bilinear (requires 16 texture samples per mipmap per pixel, as opposed to 4 for bilinear), doesn't make that big a difference other than for magnified textures with lots of detail, and is somewhat susceptible to color value over/underflow. Also, if you use bicubic for magnified textures and bilinear for minified textures, you will get visible banding on the boundary between the magnified and minified areas.

So basically it cuts the fillrate by a factor of ~4, and given that the quality increase is far more subtle than for AA/AF, it's just not that attractive for the HW vendors to add support for. Now.

 

[Xmas]

Oh, while we're talking about AF...
Demirug did a port of my texture filtering test to D3D, adding a very interesting texel analyze function.

http://www.forum-3dcenter.net/vbulletin/showthread.php?s=&threadid=58867

 

[Heathen]

So where does Lanczos3 filtering fit in?

 

[RussSchultz]

Oh, while we're talking about AF...
Demirug did a port of my texture filtering test to D3D, adding a very interesting texel analyze function.

http://www.forum-3dcenter.net/vbulletin/showthread.php?s=&threadid=58867

Umm, any translation for the German impaired?

 

[mr]

Umm, any translation for the German impaired?

Sure:

I'm sure you are familiar with SamX's original tool.
This is a conversion to D3D and it needs DX9 installed.

On the settings panel there is a new section called "Texel Analyze".
It writes every texel information of every pixel to a file called TA.DAT in the working directory of the D3DAFTester.exe . This takes ~10min.
After it finished, the screen shows a picture like this where:
Red = 0 texel for a pixel
Green = maximum number of texels for a pixel (This number is found in the TexelAnalyse section left of the Stop button)

In my above example (GF3Ti200, 41.09) I have 1xAF (trilinear) with 8 texels for every pixel that is green.

When I activate 8xAF, I get this

Demirug (the author) is still working on an analysation tool.

 

[Tagrineth]

Anisotropic point sampling should work fine; however it can't be used on R300 - it forces bilinear or trilinear with aniso.

I always meant to check it out on my Kyro II, actually... you can force Aniso in the CP without any other settings, and MDK2 lets you choose Point, Bilinear, and Trilinear from the launcher. Hmm...

 

[EricVonZipper]

bump

 

[akira888]

OpenGL is slightly different:
1.) Min & Mip are set as a single state

IIRC min and mag were set to a single state and mip was set to another, but it's been a couple of years so I might be off.

 

[bloodbob]

Why doesn't anyone have any support for Bicubic filtering? Based on what I can see from enlarging pictures in Photoshop using Bilinear versus Bicubic, the Bicubic setting would dramatically improve skyboxes and other very large textures.

3dlabs would most likely support it I know my r300 doesn't in DX9. Yeah its nice and all but I think you can tell by the whole ATI & NV filtering debates going on that IHV are trying to do less and less sample rather then more.

 

[Xmas]

What exactly was the reason to bump this thread?

OpenGL is slightly different:
1.) Min & Mip are set as a single state

IIRC min and mag were set to a single state and mip was set to another, but it's been a couple of years so I might be off.

Hyp-X is right, in OpenGL min&mip are one setting (e.g. LINEAR_MIPMAP_LINEAR), and mag is another (NEAREST or LINEAR)

 

[Neeyik]

bump

And the point of bumping a one year old thread was...?