EXA performance ( XCOMPOSITE and XRENDER ) performance regression from xorg 7.0 to 7.1 ( r200 )

Michel Dänzer michel at tungstengraphics.com
Mon Jul 3 02:57:18 PDT 2006


On Mon, 2006-07-03 at 02:19 -0700, James Blanford wrote:
> 
> I ran the evas rendering performance test.  Here are the results on my
> system with git xserver and ati-driver:

Interesting, did you have the exa-damagetrack bits in the xserver tree?

> 		Fluxbox - No compositing manager
> 
> XAA			0.09
> EXA - No AccelDFS	2.4
> EXA - DFS - SMART	3.7
> EXA - DFS - ALWAYS	4.0
> EXA - DFS - GREEDY	0.18
> 
> 		Compiz
> 
> EXA - No AccelDFS	2.1
> EXA - DFS - SMART	2.0
> EXA - DFS - ALWAYS	2.1
> EXA - DFS - GREEDY	0.61

[...]

> So the "greedy" heuristic didn't seem to work well in these tests 

Yeah, I merely suggested trying this in the absence of the improved EXA
and driver bits because it used to be the default in 7.0, so some of the
performance difference in 7.1 could be due to smart vs. greedy. In my
experience, greedy barely uses offscreen memory (and thus acceleration)
at all. Oddly enough, that seems to actually help in some
circumstances...

> and the download from screen hooks didn't seem to help in compiz.

That's not too surprising, the bottleneck is likely
GLX_EXT_texture_from_pixmap or glTexSubImage there.


-- 
Earthling Michel Dänzer           |          http://tungstengraphics.com
Libre software enthusiast         |          Debian, X and DRI developer





More information about the xorg mailing list