User-Agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.8.0.1) Gecko/20060402 Firefox/1.5.0.1 Build Identifier: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.8.0.1) Gecko/20060402 Firefox/1.5.0.1 I'm using r200 radeon, with (**) RADEON(0): Using EXA acceleration architecture ("fastest" available xrender implentation) with X.org 7.0, xserver-xorg version 1.0.99.2 and xfwm4 says my xrender implentation is too slow to enable compositor.. why? xcompmgr runs well, and is amazingly fast. Reproducible: Always Steps to Reproduce:
(In reply to comment #0) > I'm using r200 radeon, with (**) RADEON(0): Using EXA acceleration architecture > ("fastest" available xrender implentation) with X.org 7.0, xserver-xorg version > 1.0.99.2 and xfwm4 says my xrender implentation is too slow to enable > compositor.. why? xcompmgr runs well, and is amazingly fast. xfwm4 integrates an autoimatix xrender benchmark and determines that your fastest implementation in the world is not fast enough (xfwm4 compositor can be quite demanding on the XRender if you set transparent windows on move/resize for example, so I do think the threshold value is acceptable). The same way, unless specified to do otherwise, the xfwm4 compositor will disable itself when running in XOrg 6.8.x as the XRender/XFixes was buggy. You may disagree with the result of the internal benchmark and may decide to bypass that test by specifying "--compositor=on" on the command line and force the compositor to be active. From a terminal, killall xfwm4 && xfwm4 --compositor=on & then save your session on logout. I'm closing this bug as invalid because it works just fine with my old NVidia GeForce 4 that doesn't claim to be the "fastest" available xrender implentation...
Humm, being curious though, if you force the compositor=on in xfwm4, does it give acceptable results when moving transparent windows? In revision 20890 (just commited), it gives the target value and measured value of the XRender pass: XRender test passed (target 0.0100 sec., measured 0.0027 sec.). I'd be curious to learn about the values on your system, and compared about how it "feels" when used.
XRender test failed (target 0.0100 sec., measured 0.0822 sec.). The thing is, it doesn't disable itself with X.org 7.0 (xorg-server-1.0.2), only with X.org 7.1RC1 (xorg-server-1.0.99.x). And after manually enabling it, it's fast.. well.. fast as any 2D operation would be, moving transparent window around the screen is realtime, and the things behind is.. is realtime.. no slowness.. I wonder how that XRender test you got there works, umm no.. more like what broke it, with 7.1
The XRender test is in the display.c source file. If you want to take a look at the code, look for myDisplayTestXrender() in http://svn.xfce.org/svn/xfce/xfwm4/trunk/src/display.c