Home Phoronix Phoronix Forums X.Org Videos From FOSDEM 2008

Radeon IRC Logs For 2009-6-18

Search This Log:

airlied: glisse: ouch, r100.c: Why does some hw doesn't have CONFIG_MEMSIZE properly setup? (wtf?)
airlied: glisse: that is totally bogus I guess I get to rewrite the MC setup :)
airlied: this is my biggest issues with clean code bases :)
ssieb: please, please fix RS690... :-(
ssieb: has another hang under a configuration he thought worked
ssieb: may have to go back to the slow but non-hanging radeonhd driver
ssieb: rebooting after stuck X gives a kernel oops in Xorg, top of stack is:
ssieb: kernel: [] ? radeon_get_ring_head+0x28/0x50 [radeon]
glisse: airlied: CONFIG_MEMSIZE is setup in the mc callback
glisse: oh the thing with aper size
glisse: well i can't remember but i am pretty sure i stol it from rawhide
Zajec: glisse: you answered me partly yesterday (thanks!), can you say somethiing more about KMS for r6xx/r7xx? is there anything to do expect filling r600.c?
Zajec: glisse: some other files to touch, some MM related code maybe? DDX side? libdrm?
stuckey_: Anyone here know if the x1300 fanless cards run hot?
Zajec: stuckey_: fanless by design, or fanless by you? :P anyway, maybe just try?
stuckey_: There are some x1300 that are fanless by design
Zajec: stuckey_: you can switch to low power mode with xonf.conf
stuckey_: I can do that with all the cards?
Zajec: but no aby experience to answer you
Zajec: stuckey_: afaik, yes
airlied: glisse: I seriously doubt it
Zajec: stuckey_: r100-r500 are supported with registers i think, and r6xx-r7xx with AtomBIOS
airlied: glisse: I think the test was the other way around
airlied: if the aper size is less than VRAM
Zajec: or maybe r5xx is AtomBIOS also? anyway, shoyuld work
stuckey_: I want to get a r1000 series but I'm not sure which one
airlied: not >
stuckey_: Zajec: what are registers and what is AtomBIOS?
airlied: its very common to have VRAM < aper
airlied: also explains a lot of crap ppl are seeing
Zajec: stuckey_: it's just way of programming GPU
Zajec: stuckey_: as end user, you don't have to worry about it :)
stuckey_: I enjoy this
Zajec: stuckey_: it's important for developers only
stuckey_: Zajec: Which card do you use?
stuckey_: Are the r1000 cards all pretty much the same?
stuckey_: I mean, they're the same chipset right? Just with slightly different speeds?
Zajec: stuckey_: rv620
Zajec: r1000? what do you mean?
stuckey_: oops... I mean R500 (x1000) series
Zajec: stuckey_: just open http://en.wikipedia.org/wiki/Comparison_of_ATI_Graphics_Processing_Units
Zajec: ):)
Zajec: :)
stuckey_: Zajec: I'm looking at that, but it's just a bunch of technical numbers I don't understand.
stuckey_: Zajec: is the rv620 supported by the open-source driver?
Zajec: stuckey_: we have 2D, Xv, 3D for r3xx-r5xx
Zajec: stuckey_: 2D and Xv only for r6xx-r7xx
Zajec: stuckey_: but slowly we are getting 3D as well for r6xx/r7xx
Zajec: don't know about 3D for ancient r1xx-r2xx
stuckey_: Okay
glisse: airlied: yeah i likely get that one wrong
stuckey_: So, are there any differences between the cards in the r500 line, or are they all basically the same card?
Zajec: glisse: over my current conversation asked you some questions, could you scroll up? :)
glisse: stuckey_: there is allmost no differences from the driver pov
glisse: Zajec: you need to hack the ddx to not try to do accel
chithead: 3d for r1xx/r2xx works, including compiz
stuckey_: glisse: What about from the end user who plays the occasional game and watches x264 movies's point of view?
chithead: x264 decoding is done on the cpu atm
glisse: stuckey_: well bigger card are more powerfull
glisse: they have more pipe, higher frequency, ...
glisse: .win 1
airlied: glisse: I've been looking at the DDX, I'm going to try and converge master + kms-support cleanly
airlied: I've got way too many conditionals in the kms-support tree
airlied: drm_mode_setting, drm_mm, new_cs, dri2.enabled
airlied: I'm going to consolidate on kms I think
airlied: MrCooper: if you get a chance please give master a spin for compiz
alpha_one_x86: Hello, I have this crash: http://pastebin.com/mf229eec can you help my to solve it? I'm under R300
MrCooper: airlied: yeah I saw your fix, will do - FWIW torcs still fails that assertion though
airlied: ah well one down hopefully, I saw a few things possibly wrong
glisse: alpha_one_x86: i think both extension your program try to use nolonger exist
alpha_one_x86: What do then?
airlied: the vidmode should still work I think
xmik: Hello, please is someone working on proper texture tiling support?
zhasha: xmik: on which driver? :)
xmik: R500 and integrated mobility GPUs like RS690
zhasha: to simplify: radeon-rewrite+kernel modesetting _or_ pre-kms/ttm stuff
xmik: Ah, I mean on non-kms system
zhasha: should already have wonderful tiling
zhasha: So I take it you have a problem :P
xmik: Really? I have snapshot of xf86-video-ati about 3 months old and texture tiling doesn't work OK. And after reading this: http://airlied.livejournal.com/66163.html it seems like texture tiling should have been worked on after merge to mainline.
xmik: Or is it texture tiling for KMS-enabled systems?
airlied: the old mesa driver should do tiling fine
airlied: it not brilliant at it.
airlied: need to redesign for new mesa diver
zhasha: that article is about FBOs
zhasha: which in all fairness, has pretty much nothing to do with tiling in itself
xmik: Ah, ok, I don't know these thigs well... thanks for explanation
xmik: airlied: Well, it seems to not work for me on my notebook (RS690)... at least shadows in KWin4 (when compositing is on) doesn't work and I have found somewhere on net that it is related tu texture tiling
zhasha: xmik: also, there are 2 drivers. The one you're interested in is the 3d driver (from mesa) called r300 (for r300-r500 cards). What you downloaded is the 2d driver (or DDX) which is just for accelerating X, not 3D
zhasha: -- i.e. OpenGL
xmik: zhasha: I know that 3D part is managed by Mesa, I just thought that X.org driver has also something to do with it...
xmik: Btw. I have mesa 7.4.2
zhasha: the applications aren't aware of whether tiling is on on the card
airlied: xmik: texure tilign can't be seen by apps
airlied: its purely an optimisation in the driver
zhasha: everything is linear in the eyes of mesa as well
zhasha: btw airlied, this:
zhasha: CS section size missmatch start at (r300_cmdbuf.c,emit_cb_offset,254) 16 vs 7
zhasha: CS section end at (r300_cmdbuf.c,emit_cb_offset,264)
xmik: airlied: So maybe it is just wrong information that it is related to tiling... but when I turn on shadows in KWin4, they are there, but looks like solid black bars
zhasha: is extremely annoying
glisse: zhasha: should be fixed in master
zhasha: glisse: using fedora 11 rpms
zhasha: and I don't feel like compiling my own since these do work fairly well
xmik: airlied: Do you have some idea what can be cause of this?
zhasha: just wondering when an update will be available
zhasha: xmik: could you give me the output of glxinfo?
xmik: zhasha: I am now at work, so I can only when I get home (about 18:00, now is 11:13 here)
zhasha: you're not making this easy :P
zhasha: it could just be a KWin bug. It's pretty likely actually
alpha_one_x86: Same crash with 6.12.1 drivers
zhasha: alpha_one_x86: what program is this?
xmik: zhasha: But proprietary Catalyst drivers did work OK with KWin4 (but that was old kernel and old xorg-server of course)
airlied: zhasha: thre is new mesa rpms in updates-tseting
airlied: for F11
xmik: zhasha: So I don't think it is KWin bug...
Zarin: xmik, what KDE version?
Zarin: And just FWI KWin != KWin4. The latter is a name for an old KDE game :)
alpha_one_x86: Allways for my crash I don't see solution, I use gentoo portage tree with xinerama
xmik: Zarin: 4.2.3 (but shadows didn't work with opensource drivers in 4.1.x too)
zhasha: xmik: just because it works with fglrx doesn't mean it's not a bug
zhasha: alpha_one_x86: what program is crashing?
Zarin: xmik, is it just menu shadows or all window shadows? What window decoration are you using if it's all shadows. Also what other effects do you have enabled?
xmik: I have to go now... I will be back in about a hour.
alpha_one_x86: Some 3d programme (KDE4 with effect, warsow) but not glxgears
xmik: It's windeco shadows (all other effects works)
zhasha: oh, but that whole thing is a warsow bug AFAIK
zhasha: it doesn't work right with Xserver 1.6
alpha_one_x86: I had same bug with Xserver 1.5
Zarin: xmik, Do you have blur enabled?
slytherin: airlied: The load_detection trick worked. I could get my TV working. Not all resolutions were detected but that may be related to some changes I did in xorg.conf.
airlied: slytherin: you can add new modes with xrandr if you know the modelines
slytherin: airlied: unfortunately the TV manual is very limited.
airlied: you can generate modes with cvt or gtf commands
scarabeus: hm, what can cause that 640x480 mode is not on the randr list, i know i can add it by hand
scarabeus: but i have it specified in modes section in xorg.conf
slytherin: airlied: I will try. The TV supports full HD (at least in specifications) I will see if it actually works after using modelines.
suokko: glisse: Update about r200 screen corruption. I did compile git version of mesa and lindrm with linus' git tree. Now I haven't yet seen any corruption. Problem as caused probably because some of ppa packages were quite old.
suokko: Too bad video play back is still broken
glisse: suokko: i am booting my self one of my agpbox
glisse: took me time to build lastest everythings one it
suokko: I'm still foricng pci mode
suokko: It takes some time for me too.
glisse: suokko: if after sometimes pci mode works flawlessly try back agp
suokko: sure. Do you have any idea what could cause xv turn everything to greenscale image?
suokko: btw, When I tested the agp mode with ppa packages it was very slow. gtkperf took 47 second while in pci mode it took only 30 seconds
suokko: But PCI mode in with KMS takes large latency if CPU is boot to deep sleep state. I tried disabling deep sleep state with agp mode too but no help to performance
suokko: Specialy text drwaing was very slow in agp mode. It was even visible slow down for VI in terminal
suokko: I hate it when some program crashes after setting 640x480 screen resolution :(
Zajec: can I set CONFIG_DRM_RADEON_KMS as module?
suokko: Zajec: no. That code is part of radeon module
Zajec: ah, so where should I actually set it?
Zajec: in .h of radeon drm, of .config?
suokko: .config
suokko: make oldconfig should ask for it if your current kernel doesn't have KMS
suokko: If it has (like fedora) then it is already set there
Zajec: uh, it's weird... every "make menuconfig" deletes CONFIG_DRM_RADEON_KMS from .config
merkurio: Hi everybody, is it a known issue that KMS doesn't work with a second monitor connected?
merkurio: My laptop's panel is currently borken, I;m awaiting for the pieces to fix it to arrive, so I'm forced to use the DVI output, without KMS both outputs (LVDS and DVI-0) work all right, wheres with KMS I get no vts
merkurio: in fact, no output at all
merkurio: This is with a r500 (X1600 Mobility)
adamk_: Works for me.
adamk_: Of course this isn't a laptop, but is a workstation with two DVI monitors.
adamk_: So it sounds like you've hit a bug.
merkurio: yeah
merkurio: On another note, with all the drivers (libdrm, mesa and the ddx) needed for KMS, when I boot without modeset, Xv is still broken
merkurio: I;ve reported it some time ago, I only see the top left quarter of the video
suokko: torcs-bin: radeon_common.c:1016: radeon_validate_bo: Assertion `radeon->state.validated_bo_count < 32' failed.
Zajec: i compiled kernel with KMS... how can I know (without X starting) it's actually working?
Zajec: it should be working from booting moment, right?
merkurio: you should have a framebuffer the size of your monitor
Zajec: so if my whole console is filled with text, it's alright?
Zajec: but than can be done without KMS...
merkurio: if it's in text mode, it is not
merkurio: I'm talking about graphic framebuffer
merkurio: with the exact size of your monitor
merkurio: 1440*900 or whatever it is
Zajec: ok, but that was possible with fbdev, right?
Zajec: full screen console, UTF-8 chars, mplayer video, etc.
merkurio: I could never get non standard vesa modes before
Zajec: so how can I know that now I use KMS and not fbdev? :)
MrCooper: Zajec: cat /proc/fb
merkurio: see the output of dmesg
Zajec: i think it was working for me... but ok, moment, i'll boot to older kernel
Zajec: dmesg | grep what? :)
Zajec: ahh, vesafb :/
Zajec: what should I see in case of working KMS?
merkurio: radeondrmfb
Zajec: ok :)
Zajec: maybe it didn't compile actually
merkurio: I'd suggest you to disable all the framebuffer drivers
Zajec: merkurio: in kernel configuration?
merkurio: just enable the radeon drm
merkurio: yes
Zajec: ok :)
merkurio: and either the CONFIG_DRM_RADEON_KMS=y opton
Zajec: yes, yes, set this one
merkurio: or add radeon.modeset=1 to your bootloader
Zajec: thanks, going experiment
merkurio: ;)
Zajec: ok, i've compiled kernel without CONFIG_DRM_RADEON_KMS=y, my mistake
Zajec: but... damn, it's not my fault
merkurio: you can just add radeon.modeset=1 to your bootloader
merkurio: the result is the same
Zajec: after I edit .config and type make I get:
Zajec: scripts/kconfig/conf -s arch/x86/Kconfig
Zajec: # configuration written to .config
Zajec: Unknown boot option `radeon.modeset=1': ignoring
Zajec: radeon: Unknown parameter `modeset'
merkurio: which kernel are you using?
Zajec: 2.6.30-git11
Zajec: it's patched
Zajec: with glisse patch
merkurio: it was only added two days ago
merkurio: to the main tree
Zajec: patch that means KMS WIP for r6
Zajec: merkurio: yes, i know, git11 was generated after pulling
Zajec: # bzip2 -dc /usr/src/patch-2.6.30-git11.bz2 | grep RADEON_KMS
Zajec: +config DRM_RADEON_KMS
merkurio: you also need CONFIG_DRM_TTM=y
merkurio: in order to see the other option
Zajec: ah
Zajec: that's sth new
merkurio: it's been it glisse's kernel for a while
merkurio: in
merkurio: but it was not in airlied's
merkurio: if you 'git clone git://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux-2.6'
merkurio: you should have all the needed in-kernel bits since last Tuesday
Zajec: merkurio: shouldn't -git11 be enought for TTM and KMS?
Zajec: as I said, it was released after pulling
merkurio: I really don't what git11 contains
merkurio: don't know
merkurio: to keep current I use git directly
Zajec: eh, ok, cloning linux
Zajec: *linus
Zajec: one more compilation, god... ;)
merkurio: xDD
merkurio: at least you don't need any patches or added pulls anymore
merkurio: http://airlied.livejournal.com/66958.html
merkurio: that's the latest info by airlied himself
Zajec: know that :0
merkurio: ok
Zajec: have it open :)
suokko: :( No nequiz corrupted my screen. So same corruption stil lhappens with latest everything :(
suokko: At least smile are rendered corretly here ;)
Zajec: em, so finally, how should I enable KMS building in kernel?
Zajec: can it be make menuconfig; load my default config; exit; vim .config; make?
Zajec: oh, and I can set staging drivers using make menuconfig
Zajec: i thought it isn't here to make staging drivers enabling harder :)
osiris_: xmik: the kde4 shadows should be fixed in current master
osiris_: master mesa of course
xmik: osiris_: Thanks for info. You mean Mesa master (future 7.6)? What was the problem?
osiris_: xmik: yes. rs690 requires textures to have minimal 64-byte rowstride unlike others GPU (only 32 bytes)
xmik: osiris_: Great! Thanks again...
xmik: I have to go... bye
osiris_: np
suokko: airlied, glisse: Seems like textured video is broken in xorg/driver/xf86-video-ati kms-support but works in ~glisse/xf86-video-ati radeon-gem-cs3. I'm using Linus' kernel with KMS. (r280 card)
Zajec: can sb help me with KMS?
Zajec: http://pastebin.com/m410c1977
Zajec: I'm trying to get KMS in console only on R6xx
suokko: Zajec: You hould get rid of vesafb
Zajec: suokko: i tried...
suokko: All you need is radeon.ko, drm.ko, ttm.ko and fbcon.ko
Zajec: CONFIG_FB=m
Zajec: # CONFIG_FB_DDC is not set
Zajec: is this ok?
suokko: You don't hve to compile kernel for it. That why there i modules
suokko: Yo uhave to jut makes sure that you only load radeon, and fbcon modules
suokko: radeon modules should automaticaly load drm and ttm module
Zajec: suokko: but i don't have any vesa loaded :/
Zajec: lsmod | egrep "vesa|fb" gives empty output
Zajec: hm, maybe it's because I didn't call "make install" after recompiling keren with vesa disabled
Zajec: i just called "make modules_install"
Zajec: vesa was probably buildit eariler, not module
suokko: might be. That why I like make deb-pkg ;)
Zajec: ok, called "make install", i'll reboot
Zajec: suokko: :) (un)fortunately i use suse :)
Zajec: generally it's fine
Zajec: maybe not just for kernel experiments :)
Zajec: thanks for help everybody helping me today :)
Zajec: fianlly i get quite clean error in KMS :)
Zajec: hm, wanted to as about source of some error... log has 4 lines... should I put it here or pastebin? ;)
Zajec: [drm] radeon: Initializing kernel modesetting.
Zajec: [drm:radeon_driver_load_kms] *ERROR* Failed to initialize radeon, disabling IOCTL
Zajec: radeon 0000:01:00.0: PCI INT A disabled
Zajec: radeon: probe of 0000:01:00.0 failed with error -22
stikonas: Zajec: what distribution are you using?
Zajec: it used to be openSUSE :)
Zajec: openSUSE with own kernel, own Mesa, own libdrm :)
Zajec: own DDX driver of course
stikonas: I also recompiled X-server
Zajec: i just upgraded it to 1.6.1 from RPM... but that shouldn't impact KMS
suokko: Zajec: You can also try drm.debug=1 if you want more verbose output from drm problems
Zajec: oh, great
Zajec: thanks
Zajec: suokko: do you know where I can find value of ENOMEM?
Zajec: nvm errno.h
suokko: btw, "make binrpm-pkg" is also possible with linux :)
Zajec: suokko: i know, but this recompiles whole kernel
Zajec: suokko: i calls "make clean" first
suokko: That's not happening with deb-pkg.
Zajec: errno.h -> errno-base.h
Zajec: suokko: you said I don't have to compile whole kernel
Zajec: how can i compile only stuff in drivers/gpu/drm/radeon/?
Zajec: # make
Zajec: make: *** No targets. Stop.
suokko: make modules should be quite fast at checking if something should be recompiled
suokko: I don't remember how you can build specific module
Zajec: it's slow :)
Zajec: or just bulding radeon module is slow...
Zajec: real 4m31.544s
suokko: make SUBDIRS=drivers/gpu modules
Zajec: wow, thanks!
Zajec: does MC stand for memory controller?
Zajec: ok, hacking officialy started :) first radoen crash ;)
Zajec: suokko: thanks for help
cbmuser: glisse: any news on the firefox-triggered near system freeze ?
cbmuser: it crashes for me even without flash
adamk: rnoland: If X crashes, I regularly get this in my next Xorg log file:
adamk: (EE) RADEON(0): [pci] Out of memory (-12)
adamk: (EE) RADEON(0): [pci] PCI failed to initialize. Disabling the DRI.
adamk: If I unload and reload the radeon kernel module, X will actually refuse to start.
adamk: Have you seen this at all?
moeSizlak: id like to know if this problem is specific to me or not...using f11 and newest rpmfusion mplayer (mplayer-1.0-0.109.20090329svn.fc11.i586), when i watch a vid in full-screen, and press 'e' a few times to pan/scan zoom, the left side of the screen becomes and stays glitchy
moeSizlak: spstarr u around?
spstarr: it's a possibility ;)
moeSizlak: could u possibly check to see it you have this mplayer glitch as well?
spstarr: i dont have an r3xx right now to use, the machine is in retirement
moeSizlak: try it on yours
moeSizlak: might be a mplayer issue
spstarr: im right now using the Intel GMA GPU not the r6xx GPU
moeSizlak: do you have f11 + mplayer?
spstarr: yes
spstarr: well f11/rawhide
moeSizlak: well give it a try if u can
spstarr: try what option
moeSizlak: load a video of any type, make it fullscreen, and hit 'e' a few times
moeSizlak: let me know if the left side of your screen becomes glitchy
spstarr: nope
moeSizlak: lame
spstarr: im on Intel GPU i said
spstarr: its nice that I have a fallback GPU so I don't need to bug the radeon folk
fcami: moeSizlak: KMS or nomodeset ?
moeSizlak: kms
moeSizlak: kernel -191 and new mesa
fcami: could you try with nomodeset in the kernel command line ?
moeSizlak: sure ill try that
moeSizlak: fcami, indeed it works correctly with nomodeset
moeSizlak: actually first time i tried it my machine hard-locked immediately after starting mplayer
moeSizlak: 2nd time around it worked correctly
fcami: moeSizlak: please file a bug at bugzilla.redhat.com and include dmesg and Xorg.0.log log
moeSizlak: sure
fcami: moeSizlak: please try to reproduce without nomodeset but with drm.debug=1 in the command line and include /var/log/messages from that run as well...
moeSizlak: k
moeSizlak: no preoblem reproducing as it happenes every time
moeSizlak: it looks like when i try to zoom it doesnt correctly center the video and whatever was on left side of screen just stays there
moeSizlak: motionless
moeSizlak: rebooting
Zajec: is there some common problem with:
Zajec: <3>[drm:radeon_gart_bind] *ERROR* trying to bind memory to unitialized GART !
moeSizlak: fcami: Unknown boot option `drm.debug=1': ignoring
fcami: that's not a problem
moeSizlak: ok
fcami: the kernel itself does not know what to do but the message gets given to the drm module anyway
moeSizlak: heres a nice screenshot showing the glitch: http://i39.tinypic.com/243kn5w.jpg
moeSizlak: everey time i hit 'e' to zoom in it adds another bar of glitch to the left
liquidxd: helllloooo
liquidxd: i need your hlp
liquidxd: i have a problem in configuration on xorg.conf (ATI Technologies Inc Radeon Mobility M7 LW [Radeon Mobility 7500] (rv200)) if anyone can help me please do!!!!!!!!!ibm thinkpad t30 with radeon driver
suokko: liquidxd: You should have to need any configuration in xorg.conf if you are using resent xorg
liquidxd: i have low fps 300-500
liquidxd: and no 3d accell
suokko: http://pastebin.com/m41ed1014
suokko: That should be enough
suokko: and removing it should work too
suokko: Maybe you could tell some info about your system. Which istro and paste your Xorg.0.log and dmesg
aliverius: after a long journay trying to get many more glx visuals i return to my old stable setup
aliverius: even at 64 visuals i didnt het what i need and the fps was so slow
liquidxd: http://pastebin.com/m43a59f99
aliverius: fglrx dpesnt want to work with my current xorg and x1600
aliverius: =/
aliverius: so now the question is, where are those visuals defined?
airlied: liquidxd: 300-500 fps sound fine for that card... does glxinfo report software
airlied: ?
liquidxd: http://pastebin.com/m3fb000c3
suokko: seems like drm modules aren't loaded
moeSizlak: ok ive filed the bug: https://bugzilla.redhat.com/show_bug.cgi?id=506852
liquidxd: http://pastebin.com/m4d960070
suokko: liquidxd: you could try modprobe -v radeon
MostAwesomeDude: liquidxd: You have
MostAwesomeDude: Option "DRI" "false"
MostAwesomeDude: No idea why.
MostAwesomeDude: Remove that from your xorg.conf and things should work.
zhasha: so has anyone ever had the black pixels on the screen flicker seemingly randomly to red?
zhasha: and just as I was writing that, it stopped
MostAwesomeDude: Actually, there's a lot of weird options in there. Did you just copy-and-paste from somebody else?
liquidxd: yes from www.thinkwiki.org
aliverius: j #nvidia
suokko: I think everything else should be removed except AccelMethod "EXA" and I'm not even sure if that is what you want
suokko: XAA might give better performance for your card
liquidxd: AccelMethod "XXA"
suokko: aliverius: If you want 3D performance you could try DRI2. But there is still bugs and not yet performance tuned
suokko: aliverius: Or just downgrade to older X&kernel that is supported by fglrx
liquidxd: ok i'll test it and i'll get in again to tell the result
zhasha: and it's back
liquidxd: fglrx did not work
zhasha: I can't shoot it
Zajec: what is "cp" in KMS?
MostAwesomeDude: Zajec: Command Processor.
suokko: It is way to transfer commands from driver to GPU
MostAwesomeDude: suokko: No, EXA is correct.
aliverius: suokko: i just wanted a large pool of visuals
Zajec: is this the sam as command submission?
MostAwesomeDude: Dang it, he left.
MostAwesomeDude: Zajec: CP is a chip on the card that handles reading the ring buffer.
suokko: MostAwesomeDude: I have experience that XAA give a lot more balanced 2D performance if no composite wm ;)
MostAwesomeDude: Command Submission is the architecture that we're using to fill the ring buffer in-kernel.
Zajec: MostAwesomeDude: ah, ok
Zajec: great, understand, thanks
moeSizlak: hmm i get a whole lot of these
moeSizlak: (EE) RADEON(0): ADVANCE_RING count != expected (14 vs 16) at radeon_textured_videofuncs.c:1623
MostAwesomeDude: suokko: For recent drivers, it should be getting a lot better.
MostAwesomeDude: liquidxd: Welcome back. Yeah, a lot of those options are bogus or bad.
suokko: MostAwesomeDude: Sure but I have to stil lset migrationheuristic to smart to get balanced performance from EXA
suokko: 3D performance is clearly better if using EXA :)
liquidxd: http://pastebin.com/m441687be
MostAwesomeDude: suokko: With Xserver 1.6, it shouldn't be a problem.
liquidxd: success
MostAwesomeDude: liquidxd: Your AccelMethod should be either XAA (not XXA) or EXA. Try both, see which one feels better for you.
MostAwesomeDude: Does 3D work?
liquidxd: http://pastebin.com/m77465771
suokko: MostAwesomeDude: I have been using it for quite long time and there is firefox&wine that don't work well with r200 if not setting smart
suokko: liquidxd: Option "AccelMethod" "XAA"
MostAwesomeDude: suokko: It's only gotten better very recently.
suokko: But you should test if EXA or XAA is better for you
MostAwesomeDude: Anyway, at some point, XAA will go away, so don't rely on it. :3
liquidxd: yes i see it
MostAwesomeDude: Ideally, if we can, we'd like for EXA to be outperforming XAA *now*. :3
liquidxd: i think xaa is better
suokko: MostAwesomeDude: How recetly? I'm on Karmic ;)
MostAwesomeDude: suokko: Hm.
MostAwesomeDude: Karmic should have good EXA.
otaylor: liquidxd: xaa is never better
zhasha: alrighty, I can make it appear, and disappear at will... by playing a video using Xv
suokko: ye sit is good but there is still some corner cases that hit performance problems and are taking 100% CPU jut for memcpy
otaylor: liquidxd: it may, under pathological cases, be *faster*, but that doesn't make it better
otaylor: suokko: Basically, even that is just because XAA happens to have the XaaNoOffscreenPixmaps option
suokko: otaylor: He is using debian stable so EXA in't that good there yet
MostAwesomeDude: XAA with no offscreen pixmaps really shouldn't be outperforming EXA on Radeons.
otaylor: suokko: ExaNoOffscreenPixmaps could be added as well, but generally, effort has been put into fixing EXA instead
suokko: And I have been using EXA for years jsut because 3D performance has been always better with it
otaylor: MostAwesomeDude: you could write an app that would force it to be out-performed
otaylor: MostAwesomeDude: if the app constantly drew that called GetImage
otaylor: then called
MostAwesomeDude: otaylor: This is true.
Zajec: MostAwesomeDude: does CP starting/stopping differs between r600 and older?
suokko: otaylor: That is just what wine is doing with some applications ;)
MostAwesomeDude: But I was talking about general performance, not special weirdness.
MostAwesomeDude: suokko: Oh, right, the whole DIB stupidity. I remember now. :C
otaylor: MostAwesomeDude: currently, also, if an app is using old X cruft (say, arcs, or XOR'ing), then NoOffscreenPixmaps is a win
MostAwesomeDude: Zajec: It's different for each generation.
otaylor: MostAwesomeDude: the migration heurisitc stuff is supposed to try and catch that and act that way, but it's a hard battle
suokko: But I agree that EXA is getting a lot better all the time
otaylor: suokko: don't think it's really gotten much better in the last year or so.
Zajec: MostAwesomeDude: according to radeon_asic.h we use r100_cp_init for r1xx-r5xx :)
otaylor: suokko: mostly because most of the stuff that really mattered was already done a year ago
MostAwesomeDude: Zajec: And if you look at it, you'll note that it still uses a different microcode blob for each family..
MostAwesomeDude: ...and that there are family-specific tweaks in there.
Zajec: yes, i noticed microcode difference
Zajec: a few, ok :)
suokko: btw, Has anyone found out why firefox doesn't play nice with exa?
moeSizlak: mine does
suokko: I think there is sometime some corner cases that makes it perform badly
MostAwesomeDude: It massively (ab)uses Xrender.
MostAwesomeDude: And not all of Xr is accelerated (yet) for us.
otaylor: suokko: it doesnt'?
MostAwesomeDude: It's not so bad these days though.
otaylor: suokko: http://cworth.org/intel/performance_measurement/ is a good read
suokko: Firefox 2 scrolling used to be very laggy but it has been getting better
MostAwesomeDude: suokko: Fx2 is slow for completely non-X-related reasons.
MostAwesomeDude: You can make Fx2 and Fx3 perform better by disabling smooth scrolling.
otaylor: suokko: that type of statement is basically useless without quantification
suokko: Firefox 3 stil lhad same problems but was better
liquidxd: thanks a lot guys
MostAwesomeDude: liquidxd: No problem.
MostAwesomeDude: liquidxd: If you *do* have corruption or lockups, and setting AGPMode to 1, 2, 4, or 8, makes it go away, come back and let us know. :3
liquidxd: the only option is the "AccelMethod" "EXA" now
AbortRetryFail: i had some issues with Firefox 3 and EXA yesterday
AbortRetryFail: specifically scrolling up, pieces of text would get 'stuck'
liquidxd: "GARTSize" "256" this option is to share ram?
MostAwesomeDude: No.
otaylor: liquidxd: I'm pretty sure that's the default anyways, just delete it
suokko: I have tesdted CPU usage and how fast the reponse time felt with EXA with different migration heuristics. I did choose smart even tough greedy was pointed by cpu usage but it had some extra latency in my system.
MostAwesomeDude: GART size shouldn't need to be set anymore.
liquidxd: ok
liquidxd: Option "AGPMode" "4" this make any diference?
suokko: So I did choose to use smart because I couldn't see anywhere large latencies like always had often or greedy sometimes
MostAwesomeDude: liquidxd: See above.
otaylor: we need to just kill these options
suokko: in my option the least latency in UI is the most important factor.
otaylor: suokko: I don't think latency has any meaning here
otaylor: suokko: that is, the term just isn't applicable
suokko: There is latency between me clicking scroll bar in firefox and image updating in screen
liquidxd: ok thanks again and see you soon guys
liquidxd: bb
otaylor: suokko: OK, I guess that is an appropriate use of the word
otaylor: suokko: But there's no difference between "latency" and "rendering speed" here
suokko: otaylor: There is difference between CPU usage & latency. I did have higher latency even when there was less CPU usage
suokko: So maybe my GPU is too slow for EXA
MostAwesomeDude: suokko: No, that's not it.
otaylor: suokko: No
suokko: or PCI is too slow :) or what ever.
otaylor: suokko: I think the root problem may just be that firefox (and most X programs) don't do a very good job of scheduling their drawing.
otaylor: suokko: But beyond that, I think you are mostly just tilting at windmills.
suokko: But I don't the code behind so I jsut make pure guess and try to guess correct settings after that ;)
otaylor: suokko: I'm pretty skeptical that migration heuristics do anything for firefox.
otaylor: suokko: the correct settings are simple: delete your Xorg.conf. That's not so fun, though.
suokko: otaylor: Do you have old radeon anywhere around? jsut try the difference with always and smart and yo ushould see it
airlied: the best option to help on AGP is probably AccelDFS
otaylor: airlied: I've never seen that show up in firefox profiles
airlied: otaylor: not unless you are running other things on the desktop
MostAwesomeDude: airlied: But AccelDFS is incredibly unfun on AGP. :3
airlied: MostAwesomeDude: it works on most AGP boxes
otaylor: suokko: well, always is a pretty bad thing to do.
airlied: we probably should do a quirk list for it
MostAwesomeDude: airlied: IIRC it's on by default for all non-AGP, right?
suokko: for wine also AGPFastWrite might help if someone figured out how to make it stable
airlied: MostAwesomeDude: yes.
airlied: heh.. agp fast write never worked.
otaylor: suokko: have you profiled wine?
suokko: sure. 70% of time is used in memcpy :)
suokko: Inside EXA
otaylor: suokko: then AGPFastWrite almost certainly has nothing to do with it
suokko: if using default migration heuristics
otaylor: suokko: because it's pretty much impossible to spend that much time writing *to* memory
otaylor: suokko: AccelDFS is probably the best choice there, though it does tend to break spectacularly
suokko: If I remember correctly agp fast write should accerlate trasnfer from GPU to memory
airlied: suokko: it doesn't
otaylor: suokko: the real thing to do would be to analyze why wine is triggering video memory reads and fix it
suokko: otaylor: DIB engine
otaylor: A DIB is by definition not in video memory
MostAwesomeDude: otaylor: Wine has no DIB handling. It makes X calls for all win32 DIB calls.
suokko: But what X does when wine ask it to do the rendering?
otaylor: MostAwesomeDude: OK, then the "find out why" part isn't an issue, and the "fix it" part spectacularly obvious
otaylor: (if a reasonably amount of wokr)
MostAwesomeDude: otaylor: Yeah. This is why I can't understand why the Wine people have turned down *two* attempts to fix it, both of which included working code.
MostAwesomeDude: The third attempt is underway IIUC.
suokko: DIB engine is jsut too big so there is no guarantee that implementations were working all around. (I can at least understand why they say no)
otaylor: MostAwesomeDude: I guess back before ShmPixmaps were turned off in Xorg, using a ShmPixmap for a DIB would have almost been reasonable
MostAwesomeDude: otaylor: Almost, yeah.
otaylor: suokko: mm, raster drawing just isn't that hard. It's certainly not as hard as trying to emulate GDI in terms of X.
otaylor: suokko: though that code is needed in any case.
suokko: It is just massive api.
otaylor: suokko: and when were the wine developers scared of massive apis? ;-)
suokko: But anyway, Is it possible that soe code in r200 DRI driver or DDX could access directly graphics memory?
suokko: I still got some corruption when running KMS&git master mesa
MostAwesomeDude: suokko: What do you mean, "directly?"
suokko: warzone2100 wa able to cause corruption with all the latest bits
MostAwesomeDude: I mean, there's levels of directness. It's trivial to get a pointer to a mapped buffer of VRAM...
suokko: That corruption looks like rendering in screen is happening to same vram location where some textures or pixmaps are located
suokko: So I ay get texture corruption in game or corruption outside the window
suokko: Alo it is possible that my font is corrupted (sometimes only few of characters are corrupted but sometimes all of them are)
suokko: Another possible explanation could be race condition
suokko: again guessing without knowing the code or hardware details
suokko: or maybe there is some driver/lib that I forgot to compile my self that is interfering
suokko: btw, I once saw corruption caue by firefox when I pressed the zoom in large image
spstarr: It's going to take me some time to realize a post-AGP world for his AMD GPUs.. (even though this Intel GMA Is using an AGP bus as a fallback)
spstarr: I still have an r1xx but its just for 2D and TV In
spstarr: welcomes himself to the 'modern' computer bus architecture world
airlied: spstarr: its not using an AGP bus as anything
airlied: you really are clueless :)
spstarr: the kernel says there's AGP though
spstarr: [ 0.000999] No AGP bridge found
spstarr: [ 0.600797] agpgart-intel 0000:00:00.0: AGP aperture is 256M @ 0xd0000000
spstarr: im confused
spstarr: you can have a AGP GART w/o an AGP bridge?
airlied: its just a GART
spstarr: [ 0.594574] agpgart-intel 0000:00:00.0: Intel Mobile IntelĀ® GM45 Express Chipset
spstarr: hmm
airlied: you don't have any AGP anything, its just the name
spstarr: oh
spstarr: that is confusing
airlied: not really.
airlied: first person I've ever seen get confused by it :)
[Enrico]: hi guys :D
spstarr: a GART sure that's not an AGP specific thing
spstarr: but an AGP aperture?
airlied: do you know what an aperture is?
spstarr: its memory association
airlied: you see you need to learn to stop saying things that don't make any sense, like why would your Intel hw use an AGP bus as a fallback?
spstarr: that wouldn't make sense, no, but there is nothing to say its a PCIe
spstarr: there's no PCIe bus even listed in lspci
airlied: its not either, AGP or PCIe are buses.
airlied: Intel GPUs are integrated
[Enrico]: i have some problem with fedora 11 and i think it is radeon related (r2xx card) when i try to login in kdm sometime i get a black screen (but i can switch to tty) and when i try to enable composite i get strage half rendered windows
airlied: [Enrico]: for composite try yum --enablerepo=updates-testing update mesa-libGL
airlied: not sure if it'll help the login issue
spstarr: airlied: I thought even integrated GPUs would be attached to some bus
spstarr: didn't know it was not
airlied: spstarr: they are on the same chip
airlied: so they have internally wired stuff, buts it not like AGP or PCIE, its something internal
[Enrico]: airlied: mhm ok let me try it, thanks :D
spstarr: airlied: so something like AMD's IGP?
airlied: spstarr: yes it has its own thing
spstarr: ah ok
[Enrico]: oh mesa 7.6 cool :D
[Enrico]: very bleeding egde stuff
[Enrico]: mhm now windows are not rendered at all lol
[Enrico]: even with coposite the main menu is not rendered anymore
[Enrico]: without*
[Enrico]: mhm mesa-libGLU is still 7.5.... is this correct airlied ?
airlied: that shouldnt matter
[Enrico]: kk
MostAwesomeDude: GLU and GLEW are largely unchanging.
[Enrico]: reverting to mesa 7.5
pl4nkton: hi
pl4nkton: kms does not work for me on r500 gpu
pl4nkton: it can't read edid data on the lvds display
pl4nkton: http://www.das-labor.org/kms_0.log
[Enrico]: airlied: how to check if kms is running?
airlied: [Enrico]: you have a high res console
[Enrico]: damn so nomodset is useless lol
airlied: nomodeset means turn off kms
[Enrico]: but it remains active :D
airlied: you got a vga= line?
[Enrico]: at least it seems
[Enrico]: airlied: no
[Enrico]: i haven't changed the kernel line except for the nomodset
[Enrico]: but since my monitor is 1024x768 only may be there is no difference
[Enrico]: mhm i've restarted kde after enabling composite, and now it work..... a bit slow may be but it works
[Enrico]: may be i should try again with 7.6
[Enrico]: mhm no mesa 7.6 broken here :'(
[Enrico]: but no crash, just white windows :D
[Enrico]: better than a crash i suppose
[Enrico]: :D
[Enrico]: mhm it was my fault, it is nomodeset i've written nomodset, but now it is worse than before :'(
[Enrico]: no windows at all
[Enrico]: system freezed
DanaG: yay, got my good laptop back.
[Enrico]: mhm i have really no idea what to try to get composite working on F11
_Groo_: hi/2 all
_Groo_: airlied: ping
Jake1: can someone help me w/ my mobility 1300 my xorg.conf has it as vesa but if i put in radeon x won't start
[Enrico]: Jake1: can you provide some log of the non working try? for example the xorg.0.log
Jake1: yeah hold on
Jake1: http://fpaste.org/paste/15733
Jake1: here is the xorg.conf
Jake1: http://www.pastebin.ca/1465680
[Enrico]: Jake1: that was the log using vesa, i want the one using radeon if possible :D
Jake1: the only way to do that would be single user mode after i break it again ahha
[Enrico]: Jake1: or the xorg.0.log.old :D
Jake1: http://www.pastebin.ca/1465682
[Enrico]: O_O radeon is not present?????
[Enrico]: omg!
[Enrico]: start a fedora OS in vm
[Enrico]: Jake1: hold on
Jake1: ok
[Enrico]: i don't remeber the package name containing the radeon driver
Jake1: ok
[Enrico]: Jake1: btw it is strange, may be you have removed it accidently
Jake1: i did and F11 upgrade it it started without any video
Jake1: it was broken to begin w/
[Enrico]: Jake1: ehehehe be aware of what you remove :D
Jake1: hahahha
Jake1: i noticed
[Enrico]: Jake1: xorg-x11-drv-ati install this
[Enrico]: Jake1: and check also that mesa-libGL and mesa-dri-driver are installed too
[Enrico]: mesa-dri-drivers*
Jake1: i installed those yesterday
Jake1: i though ti did the ati but it jsut said its updating it so thats good news i guess
Jake1: so should i now
Jake1: put radeon in
[Enrico]: Jake1: after the update check if
Jake1: ok
[Enrico]: Jake1: /usr/lib/xorg/modules/drivers/radeon_drv.so is present
Jake1: it seems to be ther enow
[Enrico]: Jake1: ok try again to start with the radeon driver (even without the xorg.conf if you want to try :D )
Jake1: how do i jsut start it w/ the radeon driver?
Jake1: just mv xorg.conf somewhere?
Jake1: and have it create a new one?
[Enrico]: Jake1: without an xorg.conf it should autodetect. or you can just change the driver line from vesa to radeon
Jake1: ok
[Enrico]: Jake1: if you want to try without just mv xorg.conf xorg.conf.bak
[Enrico]: else edit it
Jake1: ok wish me luck :)
[Enrico]: Jake1: good luck again :D
[Enrico]: doh
Jake1: IT Worked thanks :)
[Enrico]: Jake1: eheheheh glade to be helpfull :D
[Enrico]: Jake1: now you should have 3d too
[Enrico]: and composite :D
Jake1: haha whats composite?
[Enrico]: Jake1: do you know compiz? or kde4 kwin desktop effects?
Jake1: yeah
[Enrico]: unluckly they don't work for me with F11
[Enrico]: :'(
[Enrico]: mesa 7.5 buggy stuff i think
[Enrico]: with 7.4 it works like a charm
[Enrico]: i'm sure they will solve it :D
[Enrico]: Jake1: but may be you are more lucky than me, you have a newer card
[Enrico]: i have a r200 card
[Enrico]: you have an r500 card
Jake1: how do i install composite?
MostAwesomeDude: Jake1: It's already there.
[Enrico]: Jake1: ehehehe
MostAwesomeDude: Composite's not a big deal; it magically works on everything that's not fglrx.
[Enrico]: Jake1: what DE are you using? kde? gnome? something else?
Jake1: gnome
[Enrico]: MostAwesomeDude: here works pretty well with fglrx :D
[Enrico]: Jake1: then install compiz
[Enrico]: MostAwesomeDude: and untill you guys make r600 3d working i'm stick with fglrx :'(
[Enrico]: bu i want radeon
[Enrico]: simply loves radeon
[Enrico]: i'm stick with .28 kernel thanks to fglrx, i want some new stuff (yeah i'm a ricer :D )
chithead: metacity/kwin compositing works on r600 even
[Enrico]: i'm corious of trying fscache
[Enrico]: chithead: true but xrender performance are horrible
[Enrico]: i've already tried it
chithead: just be sure to leave the opengl checkbox off ;)
[Enrico]: the opengl backend is really faster
[Enrico]: chithead: it is a combobox not a checkbox, or i'ven't understood what you mean
chithead: on my rv530 both opengl and xrender compositing work at acceptable speed
chithead: yes note the ;)
[Enrico]: chithead: here it is not. even fglrx has poor performance with xrender (radeon makes better than fglrx anyway :D )
chithead: there is no reason to use fglrx if xrender is enough
[Enrico]: btw i think that with opengl backend selected kwin just doesn't start since there is no 3d
[Enrico]: chithead: eheheheheh well not really, anyway i've said it, my xrender performance are really horrible
[Enrico]: huge lag for simplest operations
[Enrico]: even the jumping icon when you start a program is really slow
[Enrico]: i've asked some hint...... no luck
[Enrico]: more crrectly, no answer at all :D
airlied: hehe.. r200.
[Enrico]: airlied: eheheh a bit to old ya? i guess r300-500 come first, no problem anyway :D
airlied: [Enrico]: well its not exactly first on the testing shelf
[Enrico]: airlied: and even better but r600 on the top of the list :D
airlied: you have fglrx on it?
[Enrico]: airlied: no i have fglrx on my r600
airlied: ah explains that then
[Enrico]: eheheheh r200 is not supported from 8.24 by fglrx!
airlied: some people still run that.
[Enrico]: bleah
airlied: if you are running F11, get a kernel from updatest-testing as well
airlied: it can make AGP cards go a bit faster
[Enrico]: radeon really well with that card (excpet some bugs in mesa 7.5 :D ) why use fglrx?
[Enrico]: airlied: you mean .30? sure!
[Enrico]: love really new stuff :D
airlied: [Enrico]: no there should be a new 29.5
[Enrico]: airlied: thanks for the hint
airlied: in updatest-testing
[Enrico]: ok i will update :D
[Enrico]: now i'm recovering my arch system
airlied: http://kojipkgs.fedoraproject.org/packages/kernel/
[Enrico]: fedora 11 has detected the corruption :D
airlied: that one has the AGP go faster bits in it
[Enrico]: airlied: really a good news :D
airlied: glisse: please re-write the gart set page functions for IGP and PCIE to be correct :)
[Enrico]: airlied: thank you very much
airlied: they both look nothing like the version in drm-rawhide
[Enrico]: is exited, it is the first time he talks with a red hat guy
[Enrico]: [Enrico] love really new stuff :D <---- btw that's why i use gentoo too :D
[Enrico]: copying 28/33 GB this is endless lol
[Enrico]: airlied: there is no kernel update in updates-testing. i think i should download and install it manually :D
airlied: [Enrico]: ah it must not be tagged yet
[Enrico]: airlied: eheheh don't worry i've downloaded it manually, and also the firmware (that is needed) installing now
[Enrico]: airlied: thanks again
[Enrico]: finished, time to reboot and see the benefints :D
[Enrico]: airlied: you are right with the new kernel things seems more smooth, thank you really :D
[Enrico]: wish to be able to try composite with that soon :;D