All text controls in Firefox and OpenOffice misrender with EXA on R200 (specifically, an ATI FireGL 8800). In addition, Firefox misrenders the backgrounds of elements using specific variations of the -moz-border-radius CSS property. None of these rendering errors occur when EXA is disabled.
Created attachment 19110 [details] mozilla & openoffice text fields
Created attachment 19111 [details] mozilla border radius
Created attachment 19112 [details] test HTML for the second image The relevant bits are the -moz-border-radius and the fact that the border is missing one side.
Some version information: cairo-1.6.4-1.fc9.x86_64 gtk2-2.12.11-1.fc9.x86_64 gtk-nodoka-engine-0.7.1-1.fc9.x86_64 nodoka-theme-gnome-0.3.90-1.fc9.noarch pixman-0.10.0-1.fc9.x86_64 xorg-x11-drv-ati-6.8.0-14.fc9.x86_64 xorg-x11-server-Xorg-1.5.0-1.fc9.x86_64
Please attach the full xorg.conf and Xorg.0.log files. Note that there's already bug 14645, which appears to be about R200 RENDER acceleration problems as well.
Created attachment 19116 [details] X log
Created attachment 19117 [details] X config
Does it also happen without Option "AccelDFS"? Do the other workarounds mentioned in bug 14645 work for you as well? I find it interesting that all three log files there and here show (a derivate of) the original R200 GPU, as opposed to an RV2[58]0 variant. Maybe the radeon driver EXA RENDER acceleration code doesn't correctly handle some texturing related quirk of the original R200.
Option "RenderAccel" "Off" good Option "RenderAccel" "On" Option "AccelDFS" "Off" bad Option "RenderAccel" "On" Option "AccelDFS" "On" Option "MigrationHeuristic" "greedy" good Option "RenderAccel" "On" Option "AccelDFS" "On" Option "MigrationHeuristic" "smart" bad
(In reply to comment #8) > I find it interesting that all three log files there and here show (a derivate > of) the original R200 GPU, as opposed to an RV2[58]0 variant. Maybe the radeon > driver EXA RENDER acceleration code doesn't correctly handle some texturing > related quirk of the original R200. Interesting indeed. The original bug reports though about these quirks in the 3d driver were about lockups, not misrenderings (I'd think though it could be possible depending on other state). Also, one of the workarounds in the 3d driver (the "T0 hang") can't apply here, since it doesn't kick in with plain nearest or linear filtering. The other one ("Texture cache LRU hang") though potentially could as it's applied unconditionally (apparently noone could figure out in which cases it's needed). There could, of course, be more bugs which don't show up in the 3d driver for some reason too.
does ati git master help? Specifically 783cdb7374941bb1d2b63eea375fbf2f1b808cc3?
closing due to lack of feedback.
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.