sedzimir: btw the scaling code is strictly radeon-specific or is it something more common?
Imperion: when will MostAwesomeDude come to life?
telmich: good morning
telmich: any dev alive? I'm wondering how to further debug https://bugs.freedesktop.org/show_bug.cgi?id=28393
glisse: telmich: what is the bug ?
glisse: also there is no xv accel with kms on r1xx iirc
adamk: No overlay?
adamk: That seems like a significant regression :-)
jcristau: r1xx is probably not very significant anymore
telmich: glisse: the problem is, xv worked beforet he upgrade
Ke: did you use KMS before the upgrade?
Ke: you can use modeset=0 parameter for radeon kernel module to disable KMS
adamk: Personally, I think it would make more sense to keep KMS disabled by default on r1xx in light of that missing feature.
spstarr: vgaswitcheroo seems broken
spstarr: or it doesnt work with KMS(?)
Sarvatt: or just revert http://cgit.freedesktop.org/xorg/driver/xf86-video-ati/commit/?id=eb5665688ef9b52f03f61546351d0848cab54740 :)
Sarvatt: to get xv back with KMS on r1xx
spstarr: [ 392.177215] radeon 0000:01:00.0: r600_cs_track_validate_cb offset 320 not aligned
spstarr: [ 392.177224] radeon 0000:01:00.0: r600_packet3_check:1256 invalid cmd stream 922
spstarr: [ 392.177230] [drm:radeon_cs_ioctl] *ERROR* Invalid command stream !
spstarr: looks for new drm/mesa commits
spstarr: nothing recent
Suprano: radeon-git doesn't compile :(
spstarr: i am getting text glyph corruption with amd gpu also
schlurchz: Hi. I've been using the radeonfb kernel driver on my thinkpad for quite some time, because it suspended the radeon gpu nicely, putting it into D2 sleep. Now, please correct me if I'm wrong: AFAIUI, radeonfb is now deprecated, but with the radeon driver in linux 2.6.32, my gpu again draws lots of power in acpi sleep. Is there a fix for this?
schlurchz: Or did that workaround got lost in the transition?
schlurchz: To be honest, I find the driver situation fairly confusing.
mattst88: schlurchz, lots of new power-management code has gone into the drivers very recently
mattst88: so you might try updating to the latest code from git.
schlurchz: mattst88: thanks, maybe I'll try that. Can't find any reference to something like ACPI D2 suspend in the git kernel log, though, which was the workaround in radeonfb.
mattst88: agd5f, any idea^ ?
telmich: anyone a hint on how to debug this https://bugs.launchpad.net/ubuntu/+source/xserver-xorg-video-ati/+bug/589850?
ossman: upgraded to fc13 and opengl vsync is broken again :/
ossman: agd5f, any known new issues?
zhasha: ossman: Works For Me™
ossman: it is also flickering like crazy, like there are no buffer swaps at all
ossman: zhasha, fc13 with what hw?
ossman: zhasha, I'm on r600 stuff
ossman: which tends to be more "interesting"... :/
rhodan: Hi, using Mozilla Firefox on this page (http://en.wikipedia.org/wiki/Comparison_of_Internet_Relay_Chat_clients), X uses 100% of one core's CPU time.
rhodan: When using vesa, it's much less.
rhodan: Is there a way to make the radeon driver use software rasterizing?
BioTube: if you mean for OpenGL, pass LIBGL_ALWAYS_SOFTWARE=1 to firefox
adamk: I'm pretty sure he's talking about 2D.
BioTube: if you want to turn all acceleration off, set option "DRI" "off" in Section "Device" of your /etc/X11/xorg.conf
BioTube: (at least, I think that's it)
rhodan: I don't *want* to turn off all acceleration.
rhodan: I just want 2D to be snappy.
adamk: Don't use firefox then :-)
rhodan: That probably means turning off acceleration :(
rhodan: adamk: Chromium resizes images in software, FF uses the GPU.
BioTube: try turning off smoothscrolling in firefox
rhodan: That's why my whole X stalls when opening a big JPG in a new tab.
rhodan: At least that's what they told me in #gentoo
rhodan: Driver "ati" or Driver "radeon"?
rhodan: I forgot.
BioTube: either one
rhodan: Didn't have an xorg.conf for a long time now.
adamk: I'm curious where you heard about firefox using the GPU to resize images?
adamk: Also, make sure you are using EXA.
adamk: I was under the impression FF only did that on Windows with Direct2D.
leio: adamk: he did say from #gentoo from a completely unknown source out of that channel, wouldn't trust the information too much.
adamk: Yeah, that's kind of what I thought.
rhodan: Who are you talking about? ;)
Droste: rhodan: xorg.conf add Option "NoAccel" "true" to section device
rhodan: "DRI" "off" is bogus, right?
Droste: oh sorry
adamk: That will disable direct rendering.
Droste: I didn't read "I don't *want* to turn off all acceleration." ;-)
leio: radeon and intel are drivers whose acceleration probably means really acceleration, so I wouldn't actually think about using that for the long term :P
rhodan: (**) RADEON(0): Option "DRI" "off"
rhodan: (II) RADEON(0): [DRI2] Setup complete
leio: I'd happily use it for many other drivers, who suffer from EXA ping-pong due to not being able to accelerate all common ops :P
rhodan: Is there a way I can debug FF being slow on render operations?
MostAwesomeDude: Fx uses Cairo. Whether Cairo uses the GPU for ops depends on the backend; it's usually Xrender, which is fairly accelerated.
adamk: And, frankly, if Firefox is having problems with that page, or problems in general, does it really matter if it's using the GPU for resizing images vs the CPU. Use whichever browser gives you the best overall experience. For me, that's chrome. FF just sucks at rendering pages with large tables.
rhodan: MostAwesomeDude: If Cairo was compiled with OpenGL, would it default to that?
leio: frankly, I don't see ANY images on that wikipedia page :P
rhodan: leio: OK, I have my problems mixed up ;)
rhodan: When opening a big image, FF stalls my whole X server when the picture isn't fully loaded yet.
leio: Cairo doesn't really decide for itself what backend it uses. You (as a library user) create a specific surface
MostAwesomeDude: rhodan: Kind of. Fx requests Xrender as the backend IIRC.
rhodan: does USE="-opengl" emerge -1 cairo
MostAwesomeDude: The problem's not with Cairo, I think.
leio: rhodan: don't
leio: rhodan: you will end up having to rebuild many packages. So, don't for that reason.
rhodan: leio: What, why?
rhodan: If it doesn't speed up FF I can just emerge -1 cairo.
leio: rhodan: well, I lied anyway
rhodan: I thought this was a serious channel
leio: rhodan: glitz is enabled only with USE=opengl && USE=glitz. If you don't have glitz, it won't make a difference. And either way, it will not make a difference to firefox.
leio: I lied, as in, I just double-checked the ebuild for the exact logic.
spstarr: MostAwesomeDude: i have text glyph corruption and stuff with latest code :(
rhodan: Oh, wait.
rhodan: xcb is active, too.
rhodan: xcb sucks, right?
leio: No, xcb rocks
leio: however, nothing really uses xcb surfaces of cairo at this point yet.
spstarr: MostAwesomeDude: maybe I should just give up and wait for gallium 3d and the new ddx
leio: So, again, it ends up using the backend that firefox creates a surface for...
leio: and that's probably xlib and image surfaces. xlib surfaces end up using XRender
leio: and XRender is accelerated by EXA Composite hooks.
leio: It is possible to disable only these hooks
leio: but that really covers most of modern desktops acceleration :P
rhodan: If Chromium manages to be fast by using software rasterizing, couldn't I tell FF to do the same?
leio: instead of making random guesses on why it is slow for you, I'd instead actually profile it.
leio: I have never noticed firefox being particularly slow on my radeons, compared to other graphics
rhodan: Profile FF or X?
leio: then again, I ditched firefox quite a while ago
leio: profile the whole system while doing the slow operations?
rhodan: That is possible? D:
leio: yes. Sample based profiling
rhodan: I never profiled anything. Do I use OProfile for that? Or is kernel profiling something completely different
leio: e.g, sysprof or oprofile
leio: oprofile can profile userspace as well
leio: do you run 32 or 64bit?
rhodan: Please ;)
leio: ok, there are some issues with getting proper stacktraces of the profiles on x86-64, but it should still point out the process and function that is taking most of the CPU
leio: emerge sysprof and give it a go :P
leio: and let me know of any package issues with sysprof, I maintain it in gentoo :)
rhodan: It's testing ;)
rhodan: Oh wow, that one compiled quickly.
rhodan: Is that conclusive?
leio: I suppose you don't have debug symbols with splitdebug :(
rhodan: What do I have to re-emerge with -fno-omit-frame-pointer?
rhodan: xorg-server? Drivers, too?
leio: that instruction is for trying to get better stacktraces, but you don't even have function names?
leio: libfb.so is from xorg-server
rhodan: Oh, I see.
rhodan: sysprof doesn't seem to allow direct copy-paste
rhodan: It displayed the full path.
leio: you are missing -g or so, see http://www.gentoo.org/proj/en/qa/backtraces.xml
leio: (in short, put -g to CFLAGS, splitdebug to FEATURES and re-emerge xorg-server)
rhodan: And then restart xdm ;)
rhodan: I'm on stable, btw.
rhodan: But I was on testing for years. Did a new install last week. Doesn't feel any different, really.
rhodan: Except less warnings on bootup ;)
leio: yeah, that we can talk on #gentoo-desktop
leio: can you bzip2 and upload the profile?
rhodan: OK, I'm trying to find an image that makes X11 stall.
rhodan: Not all do, strangely.
leio: yeah, you should be profiling the slowness, not end up cumulatively with other non-problematic stuff percentually at top
rhodan: leio: Took me a while ;)
rhodan: Ignore that picture.
leio: I can't
boris64: i can't
boris64: i must err...
rhodan: Download it, tail -c 3437 erds.jpg | bzip2 -d > profile
rhodan: Wohoo steganographic pic hoster abuse!
leio: uh, try filebin.ca or something
leio: anyway, got that for now...
rhodan: I didn't know that those existed.
rhodan: I only knew those file hosters that require you to wait a minute and solve a CAPTCHA.
leio: that already shows lots of software work in libpixman, for which you don't have debug symbols :P
rhodan: Do I have to restart X after emerging pixman?
leio: probably would be better, yes, quite likely you won't get the same runtime binary :P
leio: so it'd show quite some wrong things otherwise
rhodan: That one is with kwin compositing on, btw.
rhodan: Do you want one without?
rhodan: Just did a profile, looks very similar.
rhodan: Performs similar, too.
leio: well, I'd personally prefer to not want any at this time of day
boris64: Hey folks, did anyone of you _ever_ experience the cursor disappearing / becoming invisible with the radeon driver +kms? i don't know how to reproduce it (yet), but i happens from time to time mostly when watching movies/dvb-s.
leio: I don't know offhand why it would be calling that bits_image_fetch_bilinear_no_repeat_8888 function so much
rhodan: It's scaling an image.
leio: it seems to be fetching something rather :P
leio: Maybe try Option "AccelDFS" "on" for the heck of it
leio: if it is excessively getting some image from GPU memory for some software operation for some reason, it might help
rhodan: Still the same.
leio: ah, my pixman git clone was quite outdated, found that function after a git pull now
leio: lets see what calls it...
leio: I really should figure out how to get that stacktracing to work on 64bit
leio: would just show in the profile then
rhodan: Scrolling seems faster with AccelDFS.
rhodan: Or is it just my imagination?
leio: may be
leio: If you aren't getting any visible rendering corruptions with it, or other such problems, then its a good thing to have. I think it's not the default due to it having problems on some chips and it being hard to detect which, or something along those lines...
leio: ah, man radeon is helpful.
leio: It's off by default only on AGP, due to issues with some AGP bridges
leio: so I presume you have an AGP card
leio: otherwise you are just perceiving it, as it was enabled already by default :)
rhodan: OK then ;)
rhodan: AccelDFS gave me some glitches in KDE.
rhodan: But DRI2 fixed those.
leio: ok, so fetch_scanline_a8 seems to trigger it probably
leio: can't find what would call that
rhodan: atombios_crtc.c:670: warning: dereferencing pointer ‘spc5_ptr’ does break strict-aliasing rules
rhodan: Please do not file a Gentoo bug and instead report the above QA issues directly to the upstream developers of this software. ;)
rhodan: Whoa. NoAccel *is* fast.
rhodan: leio: http://www.pubbs.net/201001/xorg/61077-bug-26225-new-firefox-slow-with-accel-fast-with-noaccel.html
leio: so follow http://bugs.freedesktop.org/show_bug.cgi?id=26225
leio: rhodan: I guess you can try installing to testing xf86-video-ati and remove that NoAccel again.
rhodan: I forgot mentioning I'm already running 6.13.0 :(
rhodan: Does NoAccel does everything on the CPU?
leio: more or less
leio: 6.13.0 seems to have been released after that commit went it\
rhodan: If so, couldn't I shut off most part of the GPU to save power?
rhodan: Or solder it out? :3
rhodan: While at power saving,
rhodan: Option "ClockGating" "true"
rhodan: seems to not get enabled according to Xorg.0.log
BioTube: rhodan: I didn
BioTube: 't see what card you have, but for r600 and later(HD 3000+), clockgating is built into the hardware and can't be disabled AFAIK
BioTube: that still needs the driver to do it
BioTube: but if you're using KMS(and, scanning the log, it looks like you are), then the ddx wouldn't be able to do it
BioTube: it'd be handled by the kernel
rhodan: Do I need /dev/agpgart (AGP Support) for a PCI card?
BioTube: I think so
BioTube: Not sure *why*, but it probably does something else
edman007: hi, xorg won't give me direct rendering, using slackware64 13.1, radeonhd 3870, the kernel takes a minute to load the microde on boot and ends with a GPU corruption screensaver look if i try to do radeon.modeset=1 on boot to turn on KMS so i have that turned off for now... http://pastebin.ca/1877721 http://pastebin.ca/1877720 http://pastebin.ca/1877719
edman007: oh, and right now i'm using my own kernel, 2.6.35-rc1, the stock slack one does the same
edman007: well i have not tried doing the KMS thing on the slack kernel...but i don't get direct rendering out of it
DanaG: oh yeah, something that'd be cool to add to dynpm: manual selection of max and min clock set.
DanaG: so you have profile=low,high and method auto...
DanaG: or profile high/high to match default, or profile low,high and method "profile" for ac/battery.