DmitryKo
Veteran
Last edited by a moderator:
Some D3D10.1 antialiasing details and pixel-coverage-masks
- Thread starter DmitryKo
- Start date
OpenGL guy
Veteran
I think that just means that you have to support at least 4 samples, it doesn't preclude 2 samples.
As 10.1 is a strict superset of 10 I read it the same way.
DmitryKo
Veteran
Now what does 'pixel coverage mask' actually mean? Googling for it either links to the same MS presentation, or a research paper on a tile-based edge antialiasing implementation...
Sounds to me like regular multi-sample anti-aliasing, just standardized as part of the Direct3D 10.1 specification. So implementations are no longer free to use other approaches. It could make it easier to work with anti-aliased surfaces by having a standard format.
Perhaps looking for "screen door translucency" might get you somewhere?
I seem to recall a statement somewhere (I'm thinking it was at R3D) from Sireric a couple years back re AA in D3D; something reasonably close to "there is a general consensus that the current api is somewhat broken". Does D3D10 and/or 10.1 address this? And if so, how?
Edit: Well, reasonably close, Dec 2004: http://www.rage3d.com/board/showpost.php?p=1333401610&postcount=5
Edit: Well, reasonably close, Dec 2004: http://www.rage3d.com/board/showpost.php?p=1333401610&postcount=5
So, it basically can be used for the antialiasing of transparent edges similar to NVidia's Transparency AA, but employing dedicated coverage mask instead of alpha values?
There are separate features:
1. sample coverage (mask) - use to globally turn off part of the coverage. use to do transparency of a whole primitive (for example)
2. alpha to coverage - use to convert fragment alpha values to coverage values. For primitives with feathered edges (foliage), it can be used to approximate coverage of the edges though there is no correlation at the sample level with actual edges.
3. shader output of coverage (aka omask) - a generalization of 1) & 2) above since the shader can perform the conversion from alpha itself, incorporate a per-primitive mask, etc.
4. selective-super sampling - perform shading calculations at sample frequency rather than pixel frequency for a primtive. This means that texture lookups and shading computations are performed separately for each sample, perhaps by iterating the shader over each sample. This can improve the alpha texture antialiasing by taking a larger number of texture samples (one per fragment sample) and then alpha testing on each sample.
1 & 2 were available in the multisample extension for OpenGL since SGI first implemented it circa 1993. 3 & 4 are new, though 4 is a generalization of "transparency AA".
here's a picture of that slide 31:
[source: http://www.behardware.com/articles/631-2/directx-10-and-gpus.html ]