C & Free Software & Hacking & Maemo & ProFUSION Gustavo Sverzut Barbieri on 13 Oct 2008 07:30 pm
x11 benchmark on embedded systems
Unlike most would say, x11 does quite well. See our benchmarks comparing X11, FB, DirectFB on a Fresscale imx31. There you can see how x11-16 bitch-slaps everything… that’s even better on n8×0 and that’s why Canola runs fast 
2 Responses to “x11 benchmark on embedded systems”
Leave a Reply
You must be logged in to post a comment.
on 14 Oct 2008 at 3:42 pm 1.lpotter said …
Probably you should have made the FB and DirectFB also 16 bit by default to avoid the conversion in those, otherwise your numbers and conclusion are meaningless.
Take the X11-16 numbers away, and DirectFB and FB are still faster than X.
on 14 Oct 2008 at 3:48 pm 2.Gustavo Sverzut Barbieri said …
See that the first benchmark is using X11 against FB, both 32bpp, thus the comparison has a meaning, and yes FB is bit faster, but not as fast as people would say.
X11-16 is there as it’s the only 16bpp engine, the other is SDL. I’d need to write DirectFB-16 and FB-16 (this later it’s a good thing to do) in order to do the comparison. Also, x11-16 is not always faster, there are lots of other cases tested that are not in the charts that explore 16bpp packing-unpacking problems and make it bit slower, not much.
As I noticed in the post, Evas X11 and FB does all operations in the client side and push image updates using XShmPutImage() (with clipper rectangles) and equivalent, so it’s all about measuring image exchange and the overhead the communication imposes. It is not high.
But thanks for your points!