[cairo] Mozilla on Cairo

[Robert O'Callahan]

On Wed, 2005-05-11 at 18:12 -0400, mental at rydia.net wrote:
> Quoting Robert O'Callahan <rocallahan at novell.com>:
> Assuming the client application builds up the image in a naive
> painters' algorithm fashion, which approach is more efficient will
> depend on the complexity of the image (in terms of overlapping) and
> on the overall pixel area that needs to be painted.
> 
> Straight supersampling potentially wastes a lot of time pushing
> pixels by painting opaquely over the same area several times due to
> overlapping, and by supersampling areas that don't really require it
> (e.g. large areas of flat color).  Adaptive supersampling would
> help, but you'd likely need to set up a BSP or quadtree or
> something to do that...
> 
> So I don't think there's a clear-cut performance win one way or the
> other.

It seems to me that in a hardware acceleration situation, almost all the
extra work required for supersampling can be performed in hardware. Much
of it is wasted, but it wastes a plentiful resource (GPU time and VRAM
bandwidth). On the other hand, analyzing the trapezoid scene will
require CPU time which is in short supply.

Rob