| Summary: | Slow performance on blogger.com "newsticker" | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Product: | xorg | Reporter: | Albert Astals Cid <aacid> | ||||||||
| Component: | Driver/Radeon | Assignee: | xf86-video-ati maintainers <xorg-driver-ati> | ||||||||
| Status: | RESOLVED WORKSFORME | QA Contact: | Xorg Project Team <xorg-team> | ||||||||
| Severity: | normal | ||||||||||
| Priority: | medium | ||||||||||
| Version: | 7.2 (2007.02) | ||||||||||
| Hardware: | Other | ||||||||||
| OS: | All | ||||||||||
| Whiteboard: | |||||||||||
| i915 platform: | i915 features: | ||||||||||
| Attachments: |
|
||||||||||
|
Description
Albert Astals Cid
2007-10-03 14:56:18 UTC
It's pixmap migration overhead due to software fallbacks - missing RENDER acceleration with konqueror, stippled and tiled fills with firefox (not sure what the stippled fills are for...). If Option "AccelDFS" isn't enabled yet, it might help somewhat. Or you could try Option "MigrationHeuristic" "greedy" or "smart". (In reply to comment #2) > see also bug 13207 This is just about slow performance, not a lockup. Created attachment 15277 [details]
html file for example
Created attachment 15278 [details]
transparent backgroun
Created attachment 15279 [details]
stylesheet to set transparent background on page
I'm able to reproduce this on blogger.com, but this is a more general bug affecting pages Mozilla renders that use nested layers of transparency. oprofile confirms that my test case and the blogger.com ticker hit the same functions. The call stack ends up calling fbPolyFillRect into exaPolyFillRect. I'll attach three files that together make up a pretty clean test case - a div that sets a transparent image for a page of text. The problem is compounded easily by nesting the divs with transparent background. The system begins to become unusable with five or so divs nested. (In reply to comment #7) Does it help if the width and height of the background image are powers of two? Some hardware can't directly accelerate repeating blits otherwise. With xserver master (to become 1.5) exaFillRegionTiled() could be used instead, but that may not be very efficient yet either with small patterns. Also, as mentioned in comment #1, when I profiled blogger.com earlier, at least with Firefox the problem seemed mostly related to stippled fills, not compositing, so I'm not sure this is really (still) the same problem. Can you provide your profiling data? is this still an issue with a more recent xserver/driver? What you would consider a more recent xserver/driver combo? xf86-video-ati 6.9.0 or git master, xserver 1.5 or newer. Seems fine here, on ATI Mobility Radeon X1400. The CPU requirements are at about 10% on my Core 2 Duo T7200. On the blogger.com page. pk@tull:~::pacman -Qi xorg-server Name : xorg-server Version : 1.5.3-2 pk@tull:~::pacman -Qi xf86-video-ati Name : xf86-video-ati Version : 6.9.0-5 Works for me, great work :-) |
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.