Hi guys,
come suse hackweek, I finally managed to land $subject in master
(thanks to Cloph, Caolan and Stephan for cleaning up after me btw).
I would not call it production-ready yet (therefore it's listed far
down in the config xml factory list of canvas implementations), but it
showcases some interesting features.
First of all, it does not use any of the tricks the older
hw-accelerated canvas implementations use [1], to get the pixels to
the screen - every drawing operation that touches pixel is native
opengl, with a fair amount of pixel shaders e.g. for gradient
fills. As such, most of the bottlenecks of cairo or gdi+ should be
gone. Of course, input data that is already a bitmap (e.g. a png)
stays that, but it simply gets loaded into a texture w/o pixel
processing.
It does that by recording display lists of rendering operations [2],
that client code calls at the XCanvas interface (conceptually not too
dissimilar from the venerable metafile approach)
Every re-paint of the screen, e.g. because something moved in
slideshow, is then a plain re-display of a slightly modified list of
already-existing display primitives, pretty much exactly what a game
does for every frame.
[1] both cairocanvas and directx canvas are software renderers at the
bottom, just using the hardware to composite already-rendered
content on the screen. hardly an efficient use of today's gpu
capabilities.
[2] there is an obvious duality here between drawinglayer primitives
aka scene graph, and the procedural XCanvas interface, that makes
converting from primitives to canvas to opengl scene graph a bit
of a waste. I'm aware of that, might be another interesting future
project.
Cheers,
-- Thorsten
Attachment:
signature.asc
Description: Digital signature