Here are the slides at least. They are of course missing the fullscreen demos.
Cairo: 2D in a 3D World
Did you get Booed or something?
The reception was ok and the short discussion we had on how cairo could be better integrated with clutter was worthwhile. I just had much better talk written than the shambles I presented.
Hi Chris – thanks for looking at Bugzilla bugs – your work is very much appreciated!
I feel bad about asking you this but you do talk about Cairo and stuff so…
What does Chris Blizzard mean when he says Cairo + Pixman’s underlying architecture never became fast? http://derstandard.at/1280984648833/Firefox-4-One-generation-ahead-of-everyone-else-speedwise Is this just one of those things?
I believe he is referring to the lack of good hardware acceleration from the drivers for RENDER, a decade after its first introduction. In terms of composite performance pixman is comparable to all the other blitters, some areas it is weaker (notably downsampling and gradients could do with a lot of performance tuning) but the majority of its operations are fast. Cairo itself is also not that slow – at least in comparing the cairo-traces between pixman, xlib, Qt and skia backends. Critics point out that Qt and skia are designed for fundamentally different modes of operation, but that does not diminish that as immediate mode rasterisers they are below par (nor that cairo is lacking some desirable features).
And I have a good laugh every time people suggest that we should simply move to GL and reap the performance benefits. (See earlier statement about poor drivers.)
Certainly the graphs on http://ickle.wordpress.com/2009/08/28/i915-absolute-times/ show that GL poor for cairo on Intel hardware (which is what I use). I guess this could be different for other drivers/cards. Is it likely that RENDER acceleration will improve at this point on the major cards though? It does sound like this is Linux problem that other platforms won’t suffer…
cairo-gl is universally slow, really slow, at this moment in time. Any driver, any platform. With one exception (that I know about), mesa/i965. A recent comparison is http://blogs.gnome.org/otte/2010/06/26/fun-with-benchmarks. It is not that hard to actually accelerate RENDER, much easier than doing a full GL driver for a shaderful architecture (and the core is the same), just at the moment we all seem to be lost fighting fires – making sure everybody’s computer just works takes precedence over making them go fast and efficiently.
I have to say your stabalisation work is much appreciated. My work machine has a G45 in that was unstable in Slackware 13.0 (I turned off GL screensavers because the machine seemed to be more likely to lock up while running them). In Slackware 13.1 things are so much better and suspend resume is faster and more reliable too. Your priorities are definitely in the right place…
thanks a lot for info
Fill in your details below or click an icon to log in:
You are commenting using your WordPress.com account. ( Log Out / Change )
You are commenting using your Twitter account. ( Log Out / Change )
You are commenting using your Facebook account. ( Log Out / Change )
Connecting to %s
Notify me of follow-up comments via email.
Get every new post delivered to your Inbox.