[cairo] anti-aliasing quality in glitz/OpenGL vs. Cairo native rasterization?

[David Reveman]

On Fri, 2004-06-11 at 00:43 +0200, Leon Woestenberg wrote: 
> Hello,
> 
> David wrote,
> On Fri, 2004-04-30 at 21:04 +0200, claude M wrote:
>  >> i've tested Cairo+Glitz. Glitz selects the Multisample extension to
>  >> perform antialiasing if it finds it. What i can say is that it worked
> 
>  > Yes thats right, glitz uses the multi-sample extension. Glitz's format
>  > system allows you control anti-aliasing. You can select if you like to
>  > use AA or not and number of samples used for multi-sampling.
> 
> I am just dropping in this mailing list (and sorry for not being able to
> reply properly, I copied the above from the archive).
> 
> How does the OpenGL output compare against the Cairo software rasterizer?

Screenshots here:
http://www.cs.umu.se/~c99drn/pics/

*glitz.png files are OpenGL output and *pixman.png are output from
libpixman (the library cairo uses for software rendering).

>  From what I understand Cairo estimates pixel coverage very accurately, 
> allowing to exploit the full range of (pixel color) quantization levels.

Yes, the software implementation of XRender does. However, I'm guessing
that the XRender specification defines the current output requirements
for cairo's backends, and I don't think that it defines a minimum set of
quantization levels for Imprecise mode rendering of anti-aliased
polygons. I'm not sure, though.

> Contrary to this, a hardware video card will do super- then subsampling
> to have 4 quantization levels for anti-aliasing in 2x2 supersampling,
> or 16 in 4x4 mode. (I know of no higher level supersampling in currently
> available cards).
> 
> Is my assumption true that the typical OpenGL backend output is worse
> than native rasterization in the Cairo software quality-wise?

yes
 
-David