| === ripps_ is now known as ripps | ||
| === ripps_ is now known as ripps | ||
| tjaalton | sweet, only eight packages non-synced/merged with debian | 07:54 |
|---|---|---|
| tjaalton | I'll merge a couple later today | 07:54 |
| Sarvatt | tjaalton: dont suppose xorg-server is in there huh? :D | 08:52 |
| Sarvatt | wow nice, I just applied this to xorg-server 2:1.6.0-0ubuntu15 and its amazing how fast VT>X switch is in KMS now. http://git.debian.org/?p=pkg-xorg/xserver/xorg-server.git;a=commit;h=fdbb6fd3d3c0ce7078f8faaf089af51cc36cbcb3 | 10:02 |
| tjaalton | Sarvatt: sure. I'll add that commit, since there is a noticeable lag when switching back to X | 11:03 |
| tjaalton | oh right, it's included in 1.6.1.901 | 11:07 |
| tjaalton | bryce: there are a bunch of patches against xorg-server by you. plans to send them upstream? | 12:23 |
| Sarvatt | xvmc-vld branch of xf86-video-intel is looking good, xvmc's working in UXA now | 13:42 |
| Sarvatt | your xorg-server 1.6.2rc1 works great here tjaalton - http://sarvatt.com/downloads/Xorg.0.log.txt | 17:13 |
| Sarvatt | nice, UXA hangs in firefox are fixed now, just finished compiling a new kernel/drm/intel driver to test the patches. bad news is it needs updates to all of those to be fixed.. https://bugs.freedesktop.org/show_bug.cgi?id=20152 | 20:24 |
| ubottu | Freedesktop bug 20152 in Driver/intel "[G45/GM965 UXA] cannot view JPG in firefox when running UXA (lots of errors in dmesg)" [Major,Resolved: fixed] | 20:24 |
| albert23 | Last exa migration before the 965 freeze occurs: -> 0x5750ec0 (0xffe0a00) (1280x25) (c) | 20:36 |
| albert23 | assuming size of the pixmap is 4 bpp x 1280 x 25, the pixmap ends at FFFFE00 | 20:36 |
| albert23 | Yesterday I found pushing back exa offscreen from the end of aperture prevented the freeze | 20:37 |
| albert23 | So that seems to match quite nicely | 20:37 |
| albert23 | Could there be any reason the gpu reads more then the pixmap size, for example due to tiling? | 20:38 |
| albert23 | jbarnes maybe^? | 20:39 |
| tjaalton | better ask on #intel-gfx | 20:42 |
| tjaalton | Sarvatt: ok, good to know | 20:42 |
| jbarnes | albert23: yes, that's one reason... it also prefetches data past the end of objects | 20:46 |
| albert23 | jbarnes: how large would the prefetch be? Pushing back exa offscreen 4 KB didn't help. 1MB seems to be enough | 20:47 |
| jbarnes | generally shouldn't got past a page of prefetch | 20:47 |
| jbarnes | but tiling can increase the object's size to the next power of two | 20:48 |
| jbarnes | or at the very least its width will be increased to the next multiple of the tile size | 20:48 |
| jbarnes | which can add a lot | 20:48 |
| albert23 | jbarnes: note we had the 32MB classic texture buffer until beginning of April | 20:49 |
| jbarnes | yeah I saw that add... interesting | 20:49 |
| albert23 | so this matches also with the beginning of the freeze bug | 20:49 |
| albert23 | Now we only need a safe value for the push back... | 20:49 |
| albert23 | The 32 MB classic was causing problems on some older 8xx cards | 20:50 |
| albert23 | Assuming maximum display height of 4096 lines, 8MB should be safe | 20:52 |
| albert23 | Well, we can always calculate the required unused memory dynamically, and possibly only if IS_I965G(pI830) | 21:06 |
| === ripps_ is now known as ripps | ||
| albert23 | This is what I am testing now. Maybe someone else who has the freeze can test this as well? http://paste.ubuntu.com/175953/ | 21:58 |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!