hormelc: vnj on radeon! it is working well so far on r630 :D
hormelc: thank you
hormelc: ArchLinux testing local/xf86-video-ati-git 20100322-1
hormelc: 02:00.0 VGA compatible controller: ATI Technologies Inc RV 630 XT AGP [Radeon HD 2600 XT AGP]
hormelc: wish there were more framebuffer modes built into card but good nonetheless
hormelc: 'working well' is a major understatement btw... dri2 is badass
hormelc: 2d and 3d is working fantastic
thansen: agd5f: will I need to install the ucode files for this firepro 7740?
DanaG: hmm, anyone here have that Onkyo / Kohjinsha dual-screen thingy?
Nightwulf|work: hi all
dileX: hi Nightwulf|work
Nightwulf|work: hi dileX
tstellar: How can I enable glut extensions like GL_ARB_vertex_shader? Is there an environment variable or a compile flag that will do it?
Hiryu: have you looked at libglew?
tstellar: What's in libglew?
Hiryu: library that simplifies dealing with opengl extensions
Hiryu: From the title of the libglew page: "The OpenGL Extension Wrangler Library"
Hiryu: works in windows and OS X just as well as it does linux
tstellar: I was wondering if I could enable those extensions when I built the mesa driver.
Hiryu: wolfenstein enemy territory works great, yay!
Hiryu: I don't know the details of mesa, but I'd guess it doesn't work that way
Hiryu: quake4 fails... and crashes plasma in kde at the same time hahaha
Vash63: Hey, just got some font corruption with new mesa/xf86-video-ati I compiled from git when loading up glxgears and making the window bigger than it probably should be.
Vash63: Messed up chars in Firefox and Konsole.
suokko: Vash63: What card and kernel?
Vash63: R700 and 2.6.33
Vash63: Actually having lots of redraw issues.
Vash63: Scrolling in Firefox is causing bad ones.
suokko: Vash63: what version of video-ati?
Vash63: Latest git. I'm not complaining, just reporting. It wasn't happening last week.
suokko: And did you self compile the driver?
suokko: Vash63: I know I just try to find out what could go wrong. Can yo upaste bin xorg.log?
Vash63: Just checking to see if it's a known issue more or less.
Vash63: I have none.
Vash63: I read that as conf for some reason.
Vash63: Blame 4am
suokko: Not the best time to troubleshoot problems ;)
Vash63: space check failed in flush
Vash63: I think I found an issue... like 500 lines of that.
Vash63: Make that 3000...
twnqx: agd5f: at least one of your patches is bad :P
twnqx: after i applied upto 11 and booted the kernel i got a hard lockup after a few seconds
suokko: twnqx: Can you find the problematic patch?
twnqx: maybe when i'm too bored this evening
Vash63: Yeah, I'm gonna mess with this tomorrow probably. Just seeing if anyone else had the issue for now... too tired to do any real hunting.
twnqx: but there was something with powersaving as well (i guess...), i noticed some "flickering"
suokko: Vash63: That space check looks like cause for trouble
suokko: So reproting it tomorrow if not fixe yet is good idea
Vash63: Yeah, definitely figured as much on the space check.
Vash63: 3000+ line xorg logs are usually a bad sign >_>
suokko: agd5f: I think pm.mutex may cause deadlocks
pht: i have a little problem with suspend & resume (S3)
pht: symptoms are black screen on resume, and complete death if i switch to the xorg VT
pht: hibernate works as expected
pht: Radeon HD 3200, kernel 2.6.33
pht: do you have any tips? maybe the video card bios needs to be proded the right way?
roysjosh: pht, are you using KMS? did it work with 2.6.32?
pht: roysjosh: i'm not
pht: it's my first time with this driver, before that i used proprietary (where it worked)
roysjosh: pht, please pastebin the Xorg log?
pht: roysjosh: http://pastebin.com/qn090iVS
pht: roysjosh: http://pastebin.com/0HedJuBD
pht: line 19 marks the end of suspend
pht: or maybe line 12 ;)
roysjosh: pht, have you tried it with kms at all? sometimes that can work better. also, what version of the xf86-video-ati driver are you using? is it a git version?
pht: KMS mode for ATI is not officially supported at this time. DRI2-based ATI drivers remain experimental and the upstream developers are discouraging widespread distribution until stability is increased. After discussions with upstream developers the Arch Linux team has decided to disable KMS mode by default.
pht: i followed this
kjeldahl: Every second I get the "Failed to parse relocation" message. I'm on xorg-edgers, testing the latest 2.6.34 build from kernel ppa dated 2010-03-25. Everything works,but it's slow, with frequent lockups which unfreeze after a few seconds (up to 30-60 worst case).
roysjosh: pht, I have a M3A78-EM with a radeon 3200 igp that works fine & suspends/resumes under latest f12 (2.6.32, kms)
roysjosh: pht, it might be worth a try- it's been enabled in the past few fedora releases by default, I think
Tommeh: kjeldahl: hi :)
suokko: pht: I would think that is pre 2.6.33 statement
Tommeh: Is that with KMS or UMS?
kjeldahl: If I can't tell, I guess it's with KMS.. ;-) At least the console is a lot wider than 80 chars..
Tommeh: grep -i KMS /var/log/dmesg
pht: the module version should be in the paste?
pht: it's definitely not git
Tommeh: grep -i modeset /var/log/dmesg, even
Tommeh: It'll be in there somewhere :D
roysjosh: pht, if it's just 6.12.4 it's quite old :)
[Enrico]: pht: OT: that's the reason why i switched from archlinux to fedora couse my r200 can do 3d only in kms from a lot of months until now (dunno what happened to the UMS code)
pht: i'm not switching the distro
[Enrico]: i said it was OT (off topic :) )
kjeldahl: Tommeh: Nothing, I swear...
pht: not because of whatever philosophy, but because of the work associated
pht: well, it seems to be december 2009 - calling that 'quite old' goes beyond my understanding
pht: ok, let's try this KMS then
kjeldahl: The grub bootline says "radeon.modeset=1", but no mention of KMS in any of the logs.
twnqx: radeon module?
twnqx: or static?
kjeldahl: radeon appears in lsmod...
Tommeh: sudo su -
Tommeh: And then, dmesg | grep -i radeon
kjeldahl: kdm.log says: (II) [KMS] Kernel modesetting enabled.
Tommeh: Weee :)
Tommeh: Next thing to do is try to replicate the problem with UMS.
Tommeh: IIRC you had the same problem though?
kjeldahl: Only thing mentioned in dmesg is: [drm:radeon_cs_ioctl] *ERROR* Failed to parse relocation !
Tommeh: Hmm.. What does grep -i firmware return?
kjeldahl: dmesg? blank. Hmf, I tried upgrading to 2.6.34 because it supposedly had important fixes. Now nothing related is showing up in the logs...
kjeldahl: Wait. Fingertrouble.
kjeldahl: Here's what I get for firmware: [ 1.749671] Firmware did not grant requested _OSC control
kjeldahl: [ 10.427802] platform radeon_cp.0: firmware: requesting radeon/RV730_pfp.bin
kjeldahl: [ 10.668536] platform radeon_cp.0: firmware: requesting radeon/RV730_me.bin
kjeldahl: [ 10.674325] platform radeon_cp.0: firmware: requesting radeon/R700_rlc.bin
kjeldahl: Mixed up dmesg (the command) with dmesg (the logfile). Because of all the errors in dmesg, the command only has lines with the error...
pht: roysjosh: configure: WARNING: unrecognized options: --enable-maintainer-mode, --enable-radeon-experimental-api
roysjosh: pht, libdrm?
roysjosh: pht, what version are you compiling?
kjeldahl: Now, what's UMS?
roysjosh: kjeldahl, User Mode-Setting
roysjosh: pht, --enable-radeon-experimental-api might not be required any more
kjeldahl: How to "test"? radeon.modeset=0?
roysjosh: kjeldahl, yep
roysjosh: kjeldahl, or nomodeset
pht: thought so
kjeldahl: Ok, brb.
kjeldahl: Now running with UMS. Still the same message every other second: [drm:radeon_cs_ioctl] *ERROR* Failed to parse relocation !
Tommeh: Something similar.
kjeldahl: Finding lots of similar messages, but nothing "fixing" it without effectively disabling all acceleration.
Tommeh: Shame :(
Tommeh: So it does fix it if compiz is off, yeah?
kjeldahl: E.g. https://bugs.freedesktop.org/show_bug.cgi?id=26302
kjeldahl: No, turning off compositing (I'm using KDE/kwin) doesn't solve it.
kjeldahl: My quad-CPU i7 feels a lot slower than my dual CPU laptop. I believe it
kjeldahl: 's more than a bit premature to state that ATI is supported under linux. YMMV.
kjeldahl: supported = as in working great.
edwin: well last comment on the bug says that latest PPA packages fix it
kjeldahl: Which PPA is that? I'm already on Linux i7 2.6.34-999-generic #201003251003 SMP Thu Mar 25 10:10:25 UTC 2010 x86_64 GNU/Linux
agd5f: suokko: how so? (pm.mutex causing deadlocks)
suokko: agd5f: Just because it is locked from work_queue
agd5f: suokko: ah
suokko: But I guess it is not possible to deadlock there if lock order is always same everywhere
twnqx: agd5f: my last ping tiemout was a hardlock and half past midnight, so i just went to bed :P
agd5f: twnqx: can you try disabling the memory reclock?
twnqx: gonna try your patches one my one after work
twnqx: one by one*
twnqx: but not right now
suokko: agd5f: Also I think some of reclocking problems in 2.6.34 might be caused because there is no quarentee that cp is idle
agd5f: suokko: yeah. that's changed in my patches
suokko: agd5f: I jsut downlaoded them
suokko: But I still need to pull drt too
pht: roysjosh: man, this stuff takes ages to compile, but i'm nearing
roysjosh: pht, yep. cool- good luck on the first reboot. make sure you have the needed firmware installed, and if you use an initrd image, update that too.
pht: you mean firmware that comes with kernel or something extra
pht: and no initrd
pht: it's just that the git version of the driver requires git version of everything else, apparently down to the nuclear reactor i get power from
thansen: ok, so I got my new laptop yesterday and was lucky enough to figure out how to get mesa work now I'd just like to get my displayport monitor working. Anyone with some time to help?
agd5f: thansen: pastebin your xorg log and dmesg
thansen: agd5f: does it matter if I have dp connected at the time?
agd5f: thansen: yes, plase have it connected
pht: roysjosh: [ 7.115254] [drm] radeon kernel modesetting enabled.
roysjosh: pht, awesome. does it work any better?
pht: roysjosh: for some reason, kde refuses to enable 'effects'
pht: and, uh, it has problems with even suspending, let alone resume
roysjosh: pht, check `glxinfo | grep render` and make sure mesa/gl is installed properly? it should say: direct rendering: Yes \n OpenGL renderer string: Mesa DRI... hmm, what does it do/not do when you try to suspend?
pht: another problem is i can't swith to a text VT from X
suokko: pht: Xorg.log?
pht: (EE) RADEON(0): Timeout trying to update memory controller settings !
pht: (EE) RADEON(0): You will probably crash now ...
suokko: pht: Can you paste the full log?
pht: ok, the VT issue seems to be solved by the fbcon module
pht: i get a 160 column VT
pht: how do i get a 80x25 one?
thansen: agd5f: ok dmesg -> http://paste.pocoo.org/show/193775/
thansen: xorg log -> http://paste.pocoo.org/show/193776/
thansen: error that should up in /var/log/messages -> http://paste.pocoo.org/show/193777/
pht: roysjosh: i got direct rendering yes
agd5f: thansen: [ 1.952623] [drm:dp_link_train] *ERROR* channel eq failed
agd5f: link trainign is failing
suokko: pht: direct rendering has nothing to do with acceleration
adamk: pht: You can have direct rendering with the software rasterizer though.
adamk: You need to check the renderer string, too.
thansen: agd5f: ok, anything I can do about it?
agd5f: thansen: does your system have two displayports? maybe one's on a docking station
pht: suokko: i just did what he told me
thansen: agd5f: I don't have docking station connected, but I'm pretty sure if I had bought the thing it includes a dp out
thansen: I'v heard others talk about it anyway
adamk: pht: He mentioned checking for the renderer string in addition to checking for direct rendering.
pht: oh that
pht: shit, it says Software
adamk: pht: Did you ever pastebin your full /var/log/Xorg.0.log file?
agd5f: thansen: can you enable drm debugging and then plug in the monitor and try and enable it. echo 1 > /sys/module/drm/parameters/debug
pht: adamk: in my life you mean?
agd5f: and then pastebin the dmesg
thansen: agd5f: will I need to restart X for it to take effect?
suokko: pht: glxinfo |grep renderer
pht: adamk: actually i read it before pasting, so i can see that i forgot to install the -git driver ;)
agd5f: just turn off the monitor wit xrandr
thansen: agd5f: I don't see how to 'turn off' monitor with xrandr (sorry for my ignorance)
agd5f: xrandr --output DP --off
agd5f: actually see if turnign it back on again will work. xrandr --output DP --auto
pht: adamk: in the meantime be happy to answer my 80x25 question ;)
seblabel: i would like help for a radeon 9600 use
seblabel: i mean a 3 D use
agd5f: if it doesn't work, then turn it back off, then (as root) echo 1 > /sys/module/drm/parameters/debug and try and enable it again: xrandr --output DP --auto
agd5f: then pastebin the dmesg
adamk: Sorry, I didn't really see it above. If you are using KMS, you can specify a resolution on the kernel boot line, along the lines of video=640x480
pht: fuck! now i did a killall X and the computer froze.
pht: KMS not so stable after all then?
adamk: seblabel: LEt's start with the basics. Can you pastebin /var/log/Xorg.0.log?
seblabel: wait a moment please
thansen: agd5f: http://paste.pocoo.org/show/193779/
thansen: agd5f: I don't know that everything is in there, it appears it might be getting truncated, but still no worky
adamk: So far so good.
adamk: seblabel: How about the full output of 'LIBGL_DEBUG=verbose glxinfo' ?
adamk: And, btw, what exactly doesn't work for you?
suokko: pht: I suspect your configuration is somehow broken which causes instability
agd5f: thansen: too much ioctl spam in there. do you have remote access to the box?
seblabel: i cannot play a 3D game or using special desktop
pht: ok guys ... i think all problems solved - suspend works, kde works ...
pht: thanks a lot
suokko: pht: It is quite complex to get first time compiling the driver
suokko: good :)
suokko: I hope it stays working now
seblabel: should i enter 'LIBGL_DEBUG=verbose glxinfo' in a terminal?
pht: suokko: what breakage you have in mind?
pht: the video= parameter doesnt work;P
suokko: pht: Possible some race condition in startup or something not compiled against correct version of libdrm
suokko: They are most common problems
thansen: agd5f: if you have a box somewhere I can ssh -R
adamk: seblabel: yes, that's what you should enter.
adamk: pht: Are you using KMS?
pht: adamk: yes
pht: suokko: how could that affect recovery from killing Xorg ?
pht: Xorg is just userspace process and if i understand correctly, with KMS it doesnt touch hardware that much either
adamk: seblabel: Your 3D drivers are setup properly.
suokko: pht: If your driver functions incorrectly it may mess up with the GPU quite badly
adamk: seblabel: How are you trying to start this 'special desktop'? And what game are you trying to play?
pht: so killing it shouldn't kill machine
adamk: X definitely touches the hardware, with or without KMS.
seblabel: i tried tux3d or tuxracer
adamk: pht: But, no, killing X still shouldn't kill the machine.
suokko: pht: Problem is that driver still hs the UMS code in there wihch might kickin because of some problem.
seblabel: but it's very very slow and to playable at all
agd5f: thansen: you'd need a second box
adamk: I have no real experience with those games.
adamk: However, compiz should definitely work.
seblabel: for 2d, it works fine
adamk: Is that what you are talking about when you say 'special desktop'?
thansen: agd5f: for what?
agd5f: thansen: file a bug https://bugs.freedesktop.org
agd5f: thansen: for remote access to your box
pht: ok, one more question
seblabel: i mean desktop with effect
pht: what options do i need from the -*- Support for frame buffer devices --->
agd5f: so that you can try some things that may result in losing access to your screen
pht: shall i use < > ATI Radeon display support ?
thansen: agd5f: I have lots of machines here...just nat'ed
agd5f: thansen: ok
suokko: pht: only frame buffer console
agd5f: can you ssh into your laptop?
thansen: agd5f: yeah, I have the daemon started
pht: and [ ] Enable Video Mode Handling Helpers ?
adamk: seblabel: How are you trying to enable desktop effects?
agd5f: thansen: ok. once you are in, stop X, and unload the radeon drm module (modprobe -r radeon)
agd5f: you may have to unbind the console first
pht: suokko: ok
seblabel: if i try to put it on, it freeze the screen for a sec or two, then everything disappears and appears again
thansen: agd5f: I have it built-in
seblabel: and finally, the effect are turn off like before
adamk: seblabel: No errors, though? It sounds like the effects are starting up fine then.
adamk: seblabel: Bring up a terminal and run 'compiz &' and pastebin the output.
Wizzup: is vsync supported at this time?
agd5f: thansen: ok. can you try any other modes with the DP monitor? like 1024x768? xrandr --output DP --mode 1024x768
agd5f: see if any of those other modes work
Wizzup: I can't seem to get it to work with driconf + gears
seblabel: thank you adamk !
thansen: agd5f: sure (I *really* appreciate the help)
pht: suokko: do you think that the issue will vanish on its own in time? (for example when this stuff is not git only)
adamk: seblabel: Looks like the desktop effects are starting but that you may not have a decorator installed.
adamk: If you are using gnome be sure to install compiz-gnome (or compiz-kde on KDE).
seblabel: what does it mean ?
suokko: pht: Only if the problem has bug report and it is fixed in code
seblabel: sorry, i'm very new at this
seblabel: O.o : it wasn't installed
thansen: agd5f: nothing...if you can still help I think I'll setup port forwarding on my router to hit this machine. can you try?
seblabel: now, it's done
suokko: pht: But to avoid all the most common problems: Make sure that radeon kernel module is fully loaded before X starts, always recompile (make clean && make) mesa and video-ati if you update libdrm
adamk: seblabel: Try and run compiz again like I said.
agd5f: thansen: can't really do much if the module is built in.
agd5f: thansen: actually
thansen: agd5f: I'll rebuild as module if you'd like
agd5f: can you boot with drm.debug=1 on your kernel command line
seblabel: i changed the effects to extra and i got the effect (waved windows) but, i lose the menus and icons in top and bottom bars
agd5f: and then start up your box, but don't start X
agd5f: then dump your dmesg
thansen: agd5f: sure, do you want me to build as module while I'm at it?
agd5f: and pastebin it
agd5f: thansen: that would be easier in the long run
twnqx`: agd5f: generally, setting the memory frequency works with stock kernel (with the >= 4 instead of == 4)
agd5f: twnqx: stock kernel doesn't change the memory clock
agd5f: it's only enabled in my patch series
twnqx: agd5f: http://nopaste.info/2ba9f2ef69.html
seblabel: very strange
twnqx: well, it's drm-radeon-testing
agd5f: twnqx: it's prints out all aspects of the power mode, but doesn't actually touch the mclk or pcie lanes
twnqx: so only core clock?
adamk: seblabel: Now, according to that, desktop effects are running just fine.
twnqx: [ 4092.617644] [drm:r600_cs_packet_next_reloc_mm] *ERROR* No packet3 for relocation for packet at 5072.
twnqx: [ 4092.617650] [drm:radeon_cs_ioctl] *ERROR* Invalid command stream !
seblabel: hum, but is it normal that i lose the different menu bars ?
seblabel: i mean a " extra effect " effect ?
adamk: seblabel: Let's take this conversation to #compiz
Shuren: read pastebin from seblabel, i found: Application calling GLX 1.3 function "glXCreatePixmap" when GLX 1.3 is not supported! and from glxinfo: #
Shuren: server glx version string: 1.2... just curious, it's normal? what is the glx server?
seblabel: what does it mean?
Thunderbird: it is correct glXCreatePixmap is a GLX 1.3 call (glXCreateGLXPixmap is the old call) not sure where you are seeing this
agd5f: twnqx: if you are using it master, likely bug 27284
zhasha: Shuren: the glx server is the glx extension to Xserver. the glx client is libGL
Thunderbird: GLX is the glue between OpenGL and X11; lets just say that there are multiple parts which have to implement parts of GLX for most functionality; server is in the Xserver and client in your libGL
Shuren: so, Xserver support only 1.2, when compiz want 1.3?
zhasha: Shuren: you must be using an older Xserver
twnqx: agd5f: well, i noticed that tabs *close* slowly on firefox :P
Shuren: err... xserver support 1.2, but libgl is 1.5 capable... need a newer xserver than or recompile it over new libgl?
adamk: Shuren: It's a harmless warning.
zhasha: libGL and the Xserver are separate projects. As adamk says, it's a harmless warning. You need Xserver 1.7 to get GLX 1.4 I think :/
Shuren: i'm only just curious about the chemical between the 3d parts
seblabel: i 'm too !
suokko: Shuren: It is compiz bug that it calls 1.3 function and not extension function
seblabel: i don't understand all that you said cause of my poor english, but it is interesting
Shuren: well, if i have a opengl 3 capable video card, i need first mesa support for gl3, and also a capable Xserver?
seblabel: should i upgrade xserver so ?
Shuren: seblabel, no, you ask in #compiz about your problem
Shuren: mine only is curiosity, not about your issue
seblabel: thank you anyway!
zhasha: Shuren: GL3 is underway
Thunderbird: I don't care that much about GL3 (it is just a number) but I care more about solid GLSL and FBO support
thansen: agd5f: gonna reboot with that...brb
Shuren: yes, ok, but if i need: You need Xserver 1.7 to get GLX 1.4 I think :/ <--- than i need xserver, mmm 2.8, to get glx 3? or glx versioning is not realeated with opengl one?
Thunderbird: glx versioning is unrelated
Thunderbird: these days it is usually 1.4
Shuren: ok, thank's
zhasha: GLX is just the Xserver extension. The OpenGL driver is clientside and you can technically speaking run GL4 on GLX 1.2
zhasha: GLX is really just a way of preparing a window for GL rendering
Shuren: ok, now it's clear for me, thank you
zhasha: to find out what version of OpenGL you have available to you, use glxinfo | grep "GL version"
thansen: agd5f: fresh boot, drm as module and drm.debug=1 http://paste.pocoo.org/show/193812/
thansen: agd5f: and I punched a hole in the firewall to ssh into this box
agd5f: thansen: much better
thansen: agd5f: pm?
thansen: agd5f: can I PM a few things?
agd5f: I thought you were asking about power management :)
okias: MostAwesomeDude: Hi, what missing from r300g for RS690 cards? I already heard it didn't work, so it's possible help (except coding)? debug?
mjg59: agd5f: Ok, I've made some improvements, I think
mjg59: agd5f: I've gone back to running in process context
mjg59: But from an rt workqueue rather than the generic radeon workqueue
agd5f: mjg59: cool
mjg59: Then waiting 5usec to see if we land in a vblank, and if not delaying the reclock for another frame
mjg59: The 100x100 rectangle test in x11perf seems to be enough to trigger maximum clocking
mjg59: Boom again. Looks like we went several frames without managing to hit a vblank, then things exploded.
agd5f: memory reclocking is particularly touchy
agd5f: especially with gddr memory due to the link training
mjg59: Yeah, I'm leaning towards blaming the memory
mjg59: But that's where the biggest win comes from
mjg59: So it'd be nice to get this stable
mjg59: agd5f: Oh, wait - I just noticed something
mjg59: If we don't get an idle within the timeout, we go ahead and reclock anyway? That doesn't sound ideal
spirov92: hi, I have a RV250 and it's being horribly slow and is eating 70% CPU when I move a window. can anyone help?
suokko: spirov92: use compositor
mjg59: agd5f: Huh.
spirov92: suokko: I tried compiz some time ago, but it was just as slow
mjg59: agd5f: I'm, uh, not a fan of looping to check whether to change the clocks.
mjg59: That's pretty much doubling my idle wakeups
mjg59: (Also, why is r600_audio_update_hdmi firing 10 times a second?)
spirov92: btw I remember compiz working better on a really ancient nVidia...weird
Tommeh: Yeah but you had to mess about with binary blobs then (and still do!)
Tommeh: As well supported as they are, there's no replacement for a decent FOSS driver.
agd5f: mjg59: the audio stuff needs to be switched to use an interrupt, but that info is not approved for release yet
mjg59: agd5f: Ugh.
mjg59: agd5f: It seems to be set up even if there's no HDMI
agd5f: mjg59: yeah. it should only be enabled if hdmi is detected at modeset time
mjg59: Yeah. I'll look at that later.
agd5f: but it's not at the moment
mjg59: I think the reclocking needs to be event driven
spirov92: so, with both compiz and non-compositing kwin I get horrible performance, can anyone help me?
spirov92: (compiz is set to very minimal setup with only a few plugins enabled btw)
suokko: spirov92: update to the latest driver or use mesa 7.5
suokko: They are the best bets to get maximum performance
spirov92: suokko: I'll check for updates
suokko: spirov92: mesa 7.9 (with 2.6.33 kernel and KMS) that I have works well with my rv280
spirov92: suokko: thanks, I'll try it
suokko: mjg59, agd5f: Maybe we should use synchronization primitives to stop rendering on demand for reclocking
mjg59: suokko: Well, right now it simply takes the cp mutex
suokko: mjg59: yes. But I was thinking like stoping CP in middle of IB if there is some nice synchronization primitives in between
suokko: Or is that r600+ feature only?
mjg59: suokko: Right now I just want to try to get rid of the loop that runs repeatedly even if the gpu is entirely idle :)
mjg59: Also, make this stable
suokko: sure :)
suokko: I was jsut wondering if that would be easier solution to make sure that engine is idle when we need it idle
suokko: r500 guide has required packet4 too. (WAIT_SEMAPHORE)
thansen: agd5f: it appears I'm not even getting vga output to work either :(
agd5f: suokko: you can't really stop the CP in the middle of an IB
suokko: agd5f: If command checker or mesa inserts WAIT_SEMAPHOREs between rendering commands what would go wrong?
suokko: So that default for that semaphore is to pass but in rare cases reclockin can write nonzsero value to the register and rendering would stop on demand
agd5f: suokko: well, you can stall the CP, but it (the CP) will still be busy I think as it does the polling
agd5f: and the CP is tied to the sclk IIRC
hnsr: so annoying, i cant use any kernel newer than 2.6.31 because it causes random bus errors on my machine :(
suokko: agd5f: ok. Then sounds like not going to work
agd5f: suokko: you still need to wait for the pipeline to drain
suokko: agd5f: yes. But might be a lot less time than waiting full IB to be processed
Curan: hnsr: that sounds like a regression in the kernel... maybe a bisect and a report to the kernel devs?
hnsr: Curan, yeah I guess I should do that, not really sure how to trigger it though, seems to happen on heavy disk activity, but not all the time
hnsr: when it happens, every process I try to start terminates and just prints 'Bus error' to stdout
suokko: hnsr: Does dmes have any errors when that happens?
hnsr: nah, but i should mention that it's probably nothing related to radeon since I get it with fglrx as well
suokko: hnsr: Also if you can bisect the bad commit it might help fixing the problem
hnsr: I think I'll go do that bisect this weekend, I figured the problem would be fixed eventually, but found out the other day it happens in 2.6.33 as well, so I guess it really is up to me :P
Curan: hnsr: you might try 184.108.40.206 ;) (but I'd say a bisect is your best bet)
pht: suokko: ok, well, obviously it's a minor probability issue, so i won't touch it until the things get out as regular stable packages (i hate to compile all day only to test the latest commit), if the problem prevails then, i'll first report to the distro and then escalate
mjg59: agd5f: Ok, interesting. I'm seeing that we often fail to idle the GPU in that timeout
mjg59: So I suspect that's part of the problem
agd5f: in which timeout?
mjg59: The wait_event_interruptible_timeout(rdev->irq.idle_queue, rdev->pm.gui_idle, blah)
mjg59: Ok, still blowing up
mjg59: I can read the current scanout line, right?
mjg59: I should check that we're at the start of vblank entry and not somewhere near the end of it
Obscene_CNN: now has glxgears running only 10fps slow at a 1GHz CPU speed than at a 2GHz CPU speed :)
agd5f: mjg59: yeah, there's a crtc status reg for that
mjg59: Yeah got it
mjg59: agd5f: So, yeah, sometimes the vblank interrupt fires early:
mjg59: [drm] not in vbl for pm change 00020002 00000000 at entry
mjg59: Vertical position is 1085
mjg59: 1086 seems to be the start of vblank
Wizzup: http://www.phoronix.com/scan.php?page=news_item&px=ODEwMQ Woah...
mjg59: agd5f: Hm. I suspect the pauses that people are seeing with dynpm are basically entirely the blocking on gui idle
agd5f: mjg59: yeah, likely
agd5f: we shouldn't be reclocking while a game is running
GNU\colossus: which software components/versions do I need to have my "ATI Technologies Inc Mobility Radeon HD 3600 Series" board accelerate OpenGL in hardware?
GNU\colossus: I get "direct rendering: Yes" in glxinfo, but KDE4 doesn't let me enable compositing. glxgear is also rather slow at 680fps - my old intel IGP could do about as good as that :)
chithead: glxgears is not a benchmark
GNU\colossus: I'm aware of that.
chithead: also check the opengl renderer string
GNU\colossus: it's "OpenGL renderer string: Software Rasterizer" - I guess that's my problem.
chithead: whether you get 60 or 600 or 6000 fps in glxgears is not indicative of actual 3d performance
mjg59: agd5f: The problem is that we need to clock up
mjg59: agd5f: And that's tricky if somehow we're having trouble clearing the cp in 200msec
maligor: GNU\colossus, try LIBGL_DEBUG=verbose glxgears
maligor: GNU\colossus, or the xorg logs
agd5f: mjg59: well, one stall at the beginning is better then repeated stalls
GNU\colossus: maligor: what am I looking for, specifically?
mjg59: agd5f: Only if there's enough load at the beginning to trigger an upclock
mjg59: agd5f: What about compiz effects, say?
maligor: GNU\colossus, well, most of the stuff should be pretty verbose, so if you know your computer lingo, it's readable ;P
GNU\colossus: maligor: well, it tells me that the software rasterizer is used to render. how would I go about changing that?
maligor: GNU\colossus, check /var/log/Xorg.0.log
GNU\colossus: I get
GNU\colossus: 622 (WW) RADEON(0): Direct rendering disabled
GNU\colossus: 623 (EE) RADEON(0): Acceleration initialization failed
GNU\colossus: no reason given though
maligor: probably missing kernel driver
chithead: the reason is usually somewhere above that
chithead: or in dmesg
maligor: that too, yes
maligor: GNU\colossus, it might be things like mismatching versions for the drm and dri parts ect
maligor: or unable to open the drm device/missing kernel module
GNU\colossus: I'll try configuring the radeon driver as to be compiled into the kernel blob
GNU\colossus: thanks for your advice so far! :)
mjg59: agd5f: So ideally we'd have some other way of stopping things...
mjg59: agd5f: However, even making sure that everything's taking place during the vblank, I'm still seeing hangs
mjg59: Screen just suddenly goes entirely white and the machine is dead
mjg59: I'm going to try disabling memory reclocking and see if it still happens
agd5f: yeah, try that
agd5f: we may also have to disable active memory clients (crtcs) when we reclock memory which could get tricky
suokko: GNU\colossus: pastebining logs is also good idea so people who have seen similar errors before can check them
maligor: suokko, it however decreases the possibility of learning how to deal with similiar issues in the future :P
mjg59: agd5f: Ugh. Just spend about 5 seconds failing to idle the GPU with nothing but 2D running
mjg59: It surprises me that there can be enough commands in flight that 200ms later the engine is still busy?
agd5f: mjg59: unless it's hung
mjg59: agd5f: Well, it eventually clears
mjg59: There's no GPU reset
GNU\colossus: ok, the module built correctly now, and after installing the non-free firmware blobs, everything works fine :)
mjg59: agd5f: Looks stable without the memory reclocking
mjg59: agd5f: I guess you don't get to look at how the Windows code handles this...
mjg59: agd5f: Ooh. There's a bit to stop the CRTC reading from memory, without disabling it.
suokko: I guess that might explain why no memory reclocking with multiscreen setup if crtc disabling is required
mjg59: suokko: Well, if it has to be done during vblank then you're basically screwed with multihead
mjg59: ANd having to do it in vblank is pretty much a given. There's no way you can reprogram the memory while you're reading it out to dump it to screen
mjg59: If we had a small section of RAM we could dump the framebuffer contents to for one frame while we reclock the rest, that'd be awesome
Obscene_CNN: don't some of the IGP's have that
ajax: yeah, it's called sideport
mjg59: agd5f: This might even have worked. I'm going to leave it for a while and see if it's still up...
mjg59: Well, shit. I say that and then it falls over
mjg59: Seemed better, though
Obscene_CNN: can anybody get this java app to show them imagery? http://worldwind.arc.nasa.gov/java/demos/ApplicationTemplate.jnlp
Curan: Obscene_CNN: no, gives me http://pastebin.com/Hgg84P7Z
Obscene_CNN: Thanks Curan
Obscene_CNN: I get a globe view but no imagery in the globe
EdwardIII: when i run lsmod i've got 'raedon' listed so i guess this is the right channel!
Curan: (though that might be due to the fact, that I use OpenJDK... on the other hand I didn't come across anything else so far, that didn't work)
maligor: EdwardIII, yeah, fglrx dropped r200 support quite a while back
Obscene_CNN: I'm using sun's stuff
EdwardIII: i've got an RV280. i've rammed a dvi->hdmi converter from my rv280 to my t.v.s hdmi input but i'm really struggling to get it working
maligor: EdwardIII, do you get any output?
EdwardIII: the t.v. is receiving input over VGA ok, except for the interference, thus my desire to switch to digital
EdwardIII: none whatsoever maligor
EdwardIII: and when i switch the t.v. back to vga and run xrandr it shows DVI-0 as 'disconnected'
maligor: and the device has a dvi output?
EdwardIII: i'm back to a default xorg.conf (pretty much empty) so i am leaving it to auto-detection (which is OK for the VGA)
EdwardIII: maligor: yep - i've plugged a dvi->hdmi convertor into it
EdwardIII: just a converter cable that is
EdwardIII: maligor: it looks a bit like this: http://a248.e.akamai.net/pix.crutchfield.com/graphics/infolib/homelib/dvi_output_jack600.jpg
maligor: EdwardIII, yeah, that's a normal dvi jack
EdwardIII: i haven't just rammed a dvi plug into a vga socket until it fits, although that would be quite amusing, and i have felt a bit like doing it at times
maligor: r200 is very early for dvi output tho, I don't really know much about what issues it would have
EdwardIII: any ideas where to go next maligor? i'm at a dead end. i've tried loads of different xorg.conf combos
EdwardIII: does it generally 'just work' with newer, fglrx compatible models?
maligor: I'm sure there's people better at helping you with it here than me :P
maligor: EdwardIII, well, the basic test would be to make sure the cables and dvi->hdmi part works, but as I don't really know about r200 status, I can't say
EdwardIII: what card have you got maligor?
maligor: this system is on a HD 3870, I also have a r100 and a r200, but neither of those have DVI
maligor: (they're laptops)
suokko: EdwardIII: So what appens with dvi?
EdwardIII: i'm a bit of a massive jew but i'd consider getting another card. most of the hd cards i've seen on ebay cost more than the p.c. though
suokko: Does xrandr report it disconnected?
EdwardIII: suokko: it does
suokko: So detection doesn't for hdmi converter?
EdwardIII: well, it doesn't seem to suokko but it might be a misconfiguration on my part
EdwardIII: at this stage i've just tried leaving it up to xorg to auto-detect
agd5f: EdwardIII: most likely your tv isn't giving an edid over hdmi/dvi
suokko: EdwardIII: You could try to force the output on to DVI
EdwardIII: suokko: through xorg.conf tweaking?
agd5f: EdwardIII: you can force it via xrandr
EdwardIII: agd5f: OK
EdwardIII: i *think* i tried that, but i've tried many, many things
EdwardIII: lemme take another look
EdwardIII: hrm like i said i'm running a clean xorg config so most stuff is just happening automatically - to force xrandr to output via DVI-0 am i going to need to add it to xorg.conf first?
suokko: EdwardIII: You can do all with xrandr in runtime
EdwardIII: hmm ok
suokko: xorg.conf is then good for static configuration when you find working configuration with xrandr
EdwardIII: this looks promising: http://wiki.debian.org/XStrikeForce/HowToRandR12
maligor: I would just imagine DVI would have some mechanism on the physical layer for detecting connection
EdwardIII: me too
EdwardIII: ok the t.v. over VGA is saying it supports 1360x760, 1152x864 and some other stuff
suokko: EdwardIII: try xrandr --verbose to get full mode and add that mode to dvi
EdwardIII: suokko: when you say full mode do you mean just res+frequency?
EdwardIII: i am looking at the verbose output now
EdwardIII: hrm this is odd, under the verbose output for VGA it says Clones: DVI-0? is this normal?
suokko: EdwardIII: "II.6. Adding new modes"
EdwardIII: suokko: yeaaas
EdwardIII: i did bang on one of the modes that was available in VGA
EdwardIII: i ran xrandr --output DVI-0 --mode 1360x768 --rate 60
suokko: You need to use --newmode
EdwardIII: sorry i ran that first (i was getting mode not found warning)
EdwardIII: still no hdmi output, and it still says DVI-0 is disco
EdwardIII: just went a re-seated all the cables to be sure
EdwardIII: oh sorry i ran 'addmode' rather than 'newmode' - but it seemed to have the desired effect
EdwardIII: hmm that's odd, i said rate 60 but on xrandr it shows the rate as 59.8*?
suokko: closes to 60 that is possible
EdwardIII: seems weird though
maligor: not really
EdwardIII: any clues where to go now?
EdwardIII: i'm not ruling out ordering a new card if i can get one on the cheap off ebay. anyone had success with dvi->hdmi on an ati card?
EdwardIII: gotta be no greater than agp4x though
maligor: dvi to hdmi shouldn't really be an issue
maligor: hdmi mostly just adds stuff ontop of the dvi protocol
EdwardIII: i am inclined to think it's more like the DDID stupidness on the t.v.
EdwardIII: but would that explain the 'disconnected' report from randr?
maligor: so the converters are just solid blocks for connecting wires
maligor: no, it doesn't explain it
soreau: EdwardIII: You can force it to be 'connected'
maligor: EdwardIII, if I had to guess, the card must be one of the earliest mass market cards with dvi
EdwardIII: soreau: i can't figure out howto do that?
maligor: EdwardIII, https://help.ubuntu.com/community/Radeon_9200/9250_(RV280)_and_DVI
soreau: Try xrandr --output