Home Phoronix Phoronix Forums X.Org Videos From FOSDEM 2008

Radeon IRC Logs For 2008-11-18

Search This Log:


hifi: http://scene.fi/~hifi/r500-openitg.jpg animated arrows, what might cause this or is this already fixed?
hifi: though I was impressed how smoothly the game works
hifi: it looks like the animated image frames are just not aligned properly
MostAwesomeDude: Looks like bad texcoords.
MostAwesomeDude: Do you get any errors on console?
hifi: drmRadeonCmdBuffer: -22
hifi: but that doesn't happen every time so it's not related I suppose
hifi: I can test with R300 at home later
MostAwesomeDude: Damn, thought we had gotten all of those taken care of. :C
hifi: in your face!
MostAwesomeDude: No, that's a *bad* thing.
hifi: you sure the debien experimental branch has the git revision with all those fixed?
MostAwesomeDude: Well, that'd be more of a DRM thing, prolly?
MostAwesomeDude: I dunno. We don't have anybody testing stepmania. :3
hifi: I can test with R300 if the issue is there too
hifi: my boss wouldn't like me booting my work R500 machine all day and hacking with the drm drivers ;)
MostAwesomeDude: But he doesn't mind you playing games?
hifi: just tested out of curiosity how this R500 handles openitg/stepmania :o
dmb: i'll test it
dmb: i have a r515 i think
dmb: oh wait, nm, can't do that right now
dmb: ignore me
dmb: tomorrow i'll test
hifi: RV515 here
dmb: hifi, ati x1400?
hifi: X1300
dmb: oh
hifi: oh no wait
hifi: X1550
dmb: oh :/
hifi: the crippled down 64-bit edition
EruditeHermit: is there any inherent problem with specifying a GARTSize greater than vram?
glisse: MostAwesomeDude: i have started gallium
glisse: doing winsys dri2
MostAwesomeDude: glisse: Woot.
MostAwesomeDude: 'k.
MostAwesomeDude: And I'll craft the pipe.
glisse: then i copy my hw pipe to adapt it
MostAwesomeDude: Hm. You've already got a pipe started?
glisse: one year ago
MostAwesomeDude: Rock.
glisse: there is no compiler in it
MostAwesomeDude: I'm totally in, just gimme a sec. Gotta do some of my paid work.
glisse: so it usefull only for glxgears :)
MostAwesomeDude: I was going to port nha's stuff, but...
MostAwesomeDude: Would it be useful to start working on the LLVM bits right now?
glisse: MostAwesomeDude: would be awesome
glisse: MostAwesomeDude: marcheu done a lot of work on llvm
marcheu: yeah that would be most awesome, dude
MostAwesomeDude: points to nick <---
glisse: i think he is not that far away from a working things for nv40
MostAwesomeDude: marcheu: Could you walk me through the basics, or is your code self-documenting? :3
marcheu: MostAwesomeDude: well I'd say grab the llvm docs really... since I'm using the llvm IR format the only thing you care is the llvm backend interface
marcheu: the theory is that you should not care about the frontend (at least not in the first approach)
glisse: yup llvm.org as great doc
MostAwesomeDude: 'k. Almost done with work. (MediaWiki is a pain to upgrade.) What should I check out?
glisse: allmost a full walkthrough to write your own backend
marcheu: MostAwesomeDude: get the llvm source tree, look at lib/Target/*
marcheu: MostAwesomeDude: that's the different backends. then start doing one for r500
marcheu: TBH I don't really like the (plentiful) llvm doc, I prefer looking at existing code. but that's just me
marcheu: if you like docs then go http://llvm.org/docs/WritingAnLLVMBackend.html
glisse: marcheu: you are not exactly human after all ... :)
MostAwesomeDude: I'm not human either, but in the other direction. :3
marcheu: MostAwesomeDude: also the powerpc and sparc backends are the best existing ones to look at IMO
marcheu: the X86 backend is messy
MostAwesomeDude: So we're looking at turning LLVM into machine code, right?
MostAwesomeDude: GLSL -> TGSI -> LLVM -> machine code?
marcheu: yeah
marcheu: you need to do LLVM -> r500
marcheu: and I have TGSI -> LLVM
marcheu: and I'm also doing LLVM -> nv40
MostAwesomeDude: 'k.
MostAwesomeDude: LLVM -> r300 and LLVM -> r500 should be easy?
MostAwesomeDude: Well, if one's done, the other will be easy. :3
marcheu: yeah I wouldn't say easy
MostAwesomeDude: Okay, compared to fog, it's easy.
glisse: fog will be easy in gallium, just don't do anythings it's handled for you :)
marcheu: well you'll see :) I'd suggest looking at the .td (tablegen) format first
MostAwesomeDude: 'k. I'll start reading once I finish MW work. :c
marcheu: the .td format is what describes the hw (registers, instructions...)
MostAwesomeDude: So we're using tablegen?
MostAwesomeDude: Also we're writing this in C++?
marcheu: yeah llvm's interface is c++
MostAwesomeDude: Also whose Mesa tree should I be checking out?
marcheu: mesa/mesa:gallium-0.2
marcheu: being able to work on llvm+nv40 was the point of me merging nouveau into gallium-0.2
MostAwesomeDude: Wow, stuff's changed. Cool.
MostAwesomeDude: So where should I put my stuff? Where can I see your guys' stuff?
marcheu: my... "guy's stuff" ? :)
marcheu: the nv40 llvm code is not imported yet
MostAwesomeDude: Ah.
marcheu: the tgsi->llvm is in gallivum/
marcheu: the tgsi->llvm is in gallivm/
MostAwesomeDude: I'm just kinda bad at working in the dark.
marcheu: as I said, get the llvm code, look at lib/Target, and write a backend for your hw
MostAwesomeDude: 'k.
soreau: MostAwesomeDude: hey now, watch the guys stuff ova here ;)
MostAwesomeDude: soreau: :3
MostAwesomeDude: It's always weird diving into a new codebase.
soreau: Like diving into a different colored pool of jello?
soreau: Or some other guys pool?
MostAwesomeDude: Yeah, kinda.
soreau: (ok, I'm done now :)
MostAwesomeDude: Are we building our backends directly against LLVM?
marcheu: MostAwesomeDude: that's what I'm doing ATM
marcheu: it's not difficult to move it later
MostAwesomeDude: marcheu: 'k.
MostAwesomeDude: Duly noted.
marcheu: if by directly, you mean "put it in lib/Target"
MostAwesomeDude: So I'm gonna start writing up tablegen stuff, and then I'm gonna pastebin it so everybody can laugh at me. :3
MostAwesomeDude: Yeah, that's what I'm doin'.
soreau: MostAwesomeDude: You feel the same way I do about my code xD
soreau: So I have a dumb question
soreau: When you guys are working on the same project like this, what prevents one from replicating the others work?
marcheu: newsflash: we talk to each other
soreau: Or even write different implementations of the same thing?
soreau: Is that that one comm -.. uni cat ion thingy?
soreau: marcheu: The reason I ask is because here it is pretty simple with just two for instance, but what about a project with even more teams on board?
marcheu: luckily, we're severly undermanned
soreau: lol
MostAwesomeDude: soreau: Each of us has stuff that we've picked out in advance.
soreau: And who gets left with the stuff no one wants to do?
soreau: still is wondering what it would take to figure out the tv-out timings for 1024x768
MostAwesomeDude: soreau: Grab the GATOS code from gatos.sf.net and start ripping it apart. I'm sure there's somethin' in there.
soreau: That code is already working in the radeon driver
soreau: But only for 800x600
glisse: soreau: it useless to do 1024x768, pal is somethings like 500x300
soreau: gah
glisse: in 800x600 you should already see that you loose content
soreau: glisse: Ok, well tell me how the desktop can expand across both screens with 1024x768 and 800x600 and take a screenshot correctly for instance
soreau: If you use compiz and 3d windows with that setup, it really shows why you'd want to do this
glisse: because screenshot capture what is in the card not what ends up on the display
soreau: Also, the card(s) are capable of up to 1024x768 out, and I think a driver should utilize all features of the underlying hw as options to the user space
glisse: looks at at PAL on wikipedia i am sure they are timing and res
soreau: hmm
glisse: it just seems useless to do 1024x768 as it not a mpeg2 size
glisse: neither a hd size
glisse: and for hd you really want new screen with dvi or hdmi
soreau: I just think a driver should exploit all potential feature of the hw it supports
soreau: Also, while I am complaining, what will it take to get blur working with the open drivers?
glisse: blur ?
soreau: Compiz blur plugin
soreau: Works with fglrx but not with the open driver as it doesn't implement whatever is needed for it
glisse: don't know look at plugin code to see what extension it use
MostAwesomeDude: Accumulation buffer or FBOs.
MostAwesomeDude: has heard this one before
soreau: has asked this before
MostAwesomeDude: Accumulation buffer is there, but slooooow. FBOs are WIP. :3
soreau: I see
soreau: glisse: Thanks for the tip on the wiki (duh ;)
quicksilver: MostAwesomeDude: does the hardware support an accelerated accumulation buffer in principle?
quicksilver: MostAwesomeDude: I had heard that modern hardware didn't do accum anyhow
MostAwesomeDude: quicksilver: Most modern HW does not, correct. AFAIK Radeons do not.
quicksilver: nods
quicksilver: so the compiz blur plugin (if it works with fglrx) must be accumulating in software.
MostAwesomeDude: Well, fgrlx has FBOs, which can also be used.
quicksilver: I suppose FBO + MRT can do pretty much everything people used to use an accumulation buffer for?
MostAwesomeDude: Ack, I just can't get started. TableGen is kinda tough.
MostAwesomeDude: quicksilver: Yes, pretty much.
MostAwesomeDude: Well, render-to-tex is needed too, but FBO support will also bring RTT.
quicksilver: I thought FBO gave you RTT? I thought one of the parts of the FBO extension is the ability to bind an FBO as a texture.
quicksilver: or a texture as an FBO, depending how you want to look at it.
MostAwesomeDude: Yep.
quicksilver: wishes he could help, actually. But his free hackign time these days is on the train, without his ati card.
MostAwesomeDude: :C
MostAwesomeDude: marcheu, glisse: Does it get easier after getting the tblgen stuff ready?
MostAwesomeDude: Or am I just being a whiner? :3
glisse: MostAwesomeDude: lot of code to do before being able to compile first simple program :)
MostAwesomeDude: Well, I'm okay with lots of code.
MostAwesomeDude: I'm just very overwhelmed by LLVM.
MostAwesomeDude: But I feel better when I know that you guys are working on this too.
z3ro: damn, asap means asap with these guys...
z3ro: so I leave on Friday. :)
MostAwesomeDude: Yay?
z3ro: yeah, looking forward to it. :) lots to organise though.
z3ro: brb
MostAwesomeDude: Am I doin' this even remotely right? http://pastebin.ca/1260285
marcheu: MostAwesomeDude: yeah at first it's difficult
MostAwesomeDude: Yay.]
marcheu: MostAwesomeDude: you should define your registers first
MostAwesomeDude: I seriously feel like I'm just kinda drifting...
marcheu: z3ro: leave for the job ?
marcheu: MostAwesomeDude: that's the case, at first it quite complex
MostAwesomeDude: Hm. Aren't registers just floating-point groups of four?
z3ro: marcheu: yeah
MostAwesomeDude: Oh, I see.
marcheu: yeah v4fp
marcheu: but you have to tell llvm
marcheu: once you have the regs defined, things start falling into place
marcheu: look at how SSE or Altivec are impelmented in the PowerPC or X86 backends for examples of vector instructions/registers
MostAwesomeDude: 'k.
marcheu: I agree it's pretty complex at first, but you have to describe the full arch in order to get the best performance...
marcheu: really llvm is making no compromise there
MostAwesomeDude: No, I understand. I'm just really workin' to cement myself.
MostAwesomeDude: Once I've got a reference, I do well.
MostAwesomeDude: There's a reason the r5xx FP compiler looked identical to the r3xx FP compiler. :3
marcheu: maybe we should share some llvm tree with experimental stuff at some point
MostAwesomeDude: I'm in.
MostAwesomeDude: Should we move this chat to #dri-devel or something since it's not Radeon-only?
marcheu: sure
boutz: cheers. I am trying the radeon driver with screen rotation and it's very slow.. is there a way to get it faster?
boutz: some more details: ubuntu hardy, 2 20" TFT, R350 Radeon 9800 Pro
boutz: i tried both EXA and XAA but it doesn't make a big difference in speed
zhasha_netbook: using randr 1.2 to rotate?
boutz: yes
zhasha_netbook: Then I have no idea what's wrong
boutz: should it be hardware accelrated?
zhasha_netbook: as far as I know, yes
boutz: i use "xrandr --output DVI-0 --rotate left;xrandr --output VGA-0 --rotate left" to rotate them currently
boutz: so.. is there a way to get some debugging information
zhasha_netbook: I'm really terrible with these things so I couldn't tell you
boutz: i just noticed another problem.. there is a gap of some pixels between those monitors
zhasha_netbook: what version of radeon and Xserver are you using?
boutz: it's xserver-xorg-video-ati 6.8.0-1
zhasha_netbook: and xserver 1.4 I take it?
boutz: xorg 7.3
zhasha_netbook: that's 1.4
boutz: xserver-xorg-core is 1.4 aye
boutz: 1.4.1 even
zhasha_netbook: I don't think it had accelerated rotation in 1.4
boutz: what's the current version?
zhasha_netbook: can you update to Xorg 7.4 or at least update Xserver to 1.5?
boutz: I'll try
zhasha_netbook: for that you'll need to use radeon 0.6.9
zhasha_netbook: or whatever it's called
zhasha_netbook: 6.9.0
boutz: ok
boutz: looks like i have to upgrade to ubuntu 8.10 to get those versions.. eww
zhasha_netbook: why is that eww?
boutz: i prefer to stay on hardy because it's a LTS version
boutz: i don't want to be forced to upgrade every year or so to continue receiving updated packages
zhasha_netbook: I'm not a big fan of ubuntu to begin with
boutz: what do you use?
zhasha_netbook: archlinux
agd5f: boutz: what card? if it's r4xx or below the max texture size is 2kx2k so if you dektop is bigger than that, no accelerated rotation
zhasha: agd5f, does X.org 7.3 even support accelerated rotation?
agd5f: zhasha: yeah, it should
agd5f: soreau: the tv-out controller scales the image from the crtc down to the native mode of your tv (pal, ntsc, etc.). you'll need to dump the crtc/pll/tv regs when the bios has set a 1024x768 "mode" on the tv
soreau: Can I still use radeondump or tool for this while running fglrx?
agd5f: soreau: radeontool is probably your best bet
soreau: How should I use it to get all those registers data?
agd5f: soreau: (as root) ./radeontool regmatch '*'
soreau: Last time I did that it didn't seem to dump all the data I needed :P
agd5f: radeontool didn't used to dump pll regs, but it does now
soreau: I might need to update it
soreau: 12 files changed ;)
soreau: agd5f: Ok, I'll try again
TobiasTheCommie: woohoo, got me rv770
zhasha_netbook: how is that woohoo material? :P
zhasha_netbook: TobiasTheCommie, er du en dev eller kommer du her bare ofte?
TobiasTheCommie: er her bare ofte
TobiasTheCommie: eller retere, altid
TobiasTheCommie: ved ikke hvordan den er endnu
zhasha_netbook: damn. haabede lidt paa at du kunne hjaelpe mig med at saette mig ind i koden :P
zhasha_netbook: hvorfor har du koebt et radeon kort?
TobiasTheCommie: fordi mit gamle radeon var noget lort
zhasha_netbook: men radeon er noget lort, lige meget hvilken indfaldsvinkel du tager
zhasha_netbook: den aabne driver er ikke kommet videre end gl 1.3 og den lukkede virker ikke det meste af tiden
TobiasTheCommie: de har åbne specs
TobiasTheCommie: den lukkede driver har fungeret fint for mig siden jeg købte mit gamle radeon for et år siden
zhasha_netbook: det er foerst i dag at corbin simpson og glisse er begyndt at arbejde paa gallium
TobiasTheCommie: false
zhasha_netbook: min anden laptop kan ikke engang koere compiz fordi video og gl ikke kan vises ordentligt med fglrx. et problem jeg ikke har med den her intel dims
TobiasTheCommie: works fine here
zhasha_netbook: det er fordi ati+vsync = disaster
TobiasTheCommie: btw, nvidia, compiz, video er også crap
zhasha_netbook: seen it working, ret saa godt endda
TobiasTheCommie: det virker fint for mig, jeg vil godt have åbne specs, amd er bedre end intel
zhasha_netbook: geforce 5100 outperformede mit radeon x1400
TobiasTheCommie: ah
TobiasTheCommie: that is the same as my old one
TobiasTheCommie: den gamle er 1600, men det er stort set samme chip
TobiasTheCommie: den er virkelig langsom
TobiasTheCommie: og dårlig.
zhasha_netbook: men jeg er ret sikker paa at radeon driveren naar gallium engang bliver stable, kommer til at blive ekstremt god
TobiasTheCommie: jeg kommer desvære tilbage på fglrx driver nu, fordi rv770 ikke er understøttet godt nok endnu
TobiasTheCommie: alle specs er ikke ude endnu
TobiasTheCommie: og amd har virkelig været langsome med det
TobiasTheCommie: men, jeg købte det alligevel
zhasha_netbook: jeg er paa fglrx fordi radeon ikke understoetter fbos og shaders
TobiasTheCommie: og venter bare spændt på at jeg kan gå væk fra fglrx igen
zhasha_netbook: jeg har taenkt mig at se om jeg kan hjaelpe med gallium driveren
TobiasTheCommie: fbo=
TobiasTheCommie: ?
zhasha_netbook: framebuffer object
TobiasTheCommie: ah
TobiasTheCommie: hm, jeg vidste ikke at opensource driveren ikke havde shaders
TobiasTheCommie: jeg ved dog at dit x1400 har meget gamle og dårlige shaders
zhasha_netbook: den har fragment/vertex_program, men ikke shaders
TobiasTheCommie: hm..
zhasha_netbook: shader extensions kom foerst i gl 2.0
zhasha_netbook: x1400 har shader model 3.0 :P
zhasha_netbook: saa de er OK
zhasha_netbook: de mangler bare geometry shaders
TobiasTheCommie: hm, ret sikker på at min x1650 har lorte shaders
zhasha_netbook: det har ogsaa sm3.0, med de kort har bare ikke saa mange af dem
TobiasTheCommie: nope. du har ret
TobiasTheCommie: mit nye har så shader 4.1
zhasha_netbook: det er ogsaa nice
zhasha_netbook: har du geometry shader extension?
zhasha_netbook: tror nok den hedder GL_NV_geometry_shader4
TobiasTheCommie: no idea
spstarr_work: looks at all the new DDX builds on koji
zhasha_netbook: spstarr_work, do you all use fedora?
spstarr_work: ya
spstarr_work: i do
spstarr_work: airlied does
zhasha_netbook: is it good for bleeding edge stuff like gallium development?
zhasha_netbook: last I tried fedora was FC3 :P
spstarr_work: glisse i working on gallium (soon) but I donno his distro :)
spstarr_work: MostAwesomeDude also wants to help
Trou: btw i have a 3650, if some help is needed
spstarr_work: also on another box
zhasha_netbook: spstarr_work, i want to help as well
spstarr_work: thats an r6xx
spstarr_work: zhasha_netbook: speak to glisse and others on here
spstarr_work: to get a coordinated gallium driver :)
Trou: RV365
zhasha_netbook: and I think they started today
spstarr_work: the HD 3650 is not an r3xx
zhasha_netbook: yes well, I've never done any driver development before, so to begin with it's mostly learning ;)
Trou: i meant globally, not on specific parts right now
Trou: s/globally/generally/
spstarr_work: zhasha: the closest ive done for graphics dev is add EXA support for an old S3 ViRGE card one of the exa accel routines
spstarr_work: with help
zhasha_netbook: This isn't a competition of knowing the least :P
spstarr_work: lol
zhasha_netbook: I'd like to think I know C in and out. I participated in IOI (international olympiad in informatics), I'm fairly good with algorithmics; fat lot of good that does when I don't actually know how to write a driver :P
spstarr_work: there's a LOT of algorithmic in graphics drivers...
zhasha_netbook: That I know, but I don't know where to actually shove it :P
spstarr_work: :)
zhasha_netbook: oh dear god. denmark is being overrun by islamic people
TobiasTheCommie: oh
TobiasTheCommie: ok
TobiasTheCommie: :(
zhasha_netbook: saa lige nyhederne. der er aabenbart ret mange der oensker at vi faar sharia
adamk_: Afternoon all.
adamk_: I've installed the Fedora 10 preview on another machine of mine, but I've run into a problem. It has an x1300, two monitor. KMS kicks in fine. Xorg, however, looks like this: http://68.32.29.130/DSC00002.JPG
adamk_: This happens both without an xorg.conf file and with the default one generated by 'Xorg -configure'. And here's the corresponding log file: http://pastebin.ca/1260566
zhasha_netbook: I have a feeling this is where the problem starts: (EE) RADEON(0): [drm] Failed to initialize GART heap manager
kasukka: i'm attempting a setup to have one display adapter to drive TV and pc desktop independently: some multimedia app on TV and regular desktop on pc monitor simultaneously.
kasukka: it's not obvious to me, could anyone confirm it's even possible?
adamk_: zhasha_netbook, Hmmm.
adamk_: I just noticed that there are no AGP modules loaded.
zhasha_netbook: adamk_, it's fairly likely it's a driver problem, not setup
adamk_: Wait...
adamk_: There is no intel_agp module.
zhasha_netbook: try loading it then :P
adamk_: No, I mean it doesn't exist at all.
adamk_: It seems that they compiled it into the system.
adamk_: And /dev/agpgart definitely exists.
kasukka: so should it be something like for x screens: :0.0 as desktop and :0.1 or it there any other options? there's this "RightOf" in xrandr and xorg.conf various sources keep suggesting. it doesn't suggest the tv and desktop need to go together do it?
adamk_: kasukka, If you use 'xrandr' you will get one desktop spanned across both monitors.
adamk_: Which, I assume, you do not want?
zhasha_netbook: kasukka, however if you do, maximizing a window on one monitor will make it span only that monitor, at least with metacity it will
kasukka: adamk_: yea, it's kinda problematic if my desktop stuff keeps appearing on the tv side
adamk_: I'm going to disable kernel modesetting and see if it makes a difference.
zhasha_netbook: it's probably possible what you ask, but I wouldn't know how to do it
kasukka: zhasha_netbook, adamk_: do you think xrandr is out knowing this? sounds a bit kludgy, but it crossed my mind if there was something to keep the spanned screen spaces separate
kasukka: though i would prefer some straight-forward xorg tweak..
adamk_: kasukka, Theoretically, I think you can get away with separate screens in your xorg.conf file, one for each monitor. However, this will disable direct rendering, if that's a problem for you.
adamk_: For anyone interested in my problem, disabling kernel modesetting got X working :-)
adamk_: Now to see if I can get the two of them working together.
kasukka: adamk_: is dri off for both of the displays? for tv side, i'd be using video playback... i wonder how much a perfomance hit would it be then
adamk_: kasukka, Yes, for both displays.
adamk_: Last I heard, anyway.
kasukka: adamk_: ok
adamk_: Alright, so setting 'radeon.agpmode=4' didn't help...
adamk_: Setting it to 2 worked.
adamk_: But I never had to do that in my xorg.conf file with previous versions of Xorg, nor do I have to do it with this version of Xorg if I disable kernel modesetting entirely.
zhasha_netbook: adamk_, it's still beta
adamk_: zhasha_netbook, Very true.
zhasha_netbook: KMS is probably not going to enter the kernel till 2.6.29
adamk_: Just reporting my successes and failures :-)
xnguard: zhasha_netbook: I thought it went into 2.6.28rc.
chithead: that was the intel stuff
zhasha: got disconnected there
adamk_: So it looks like radeon.agpmode=2 results in some graphical glitches, too...
airlied: adamk_: what Fedora bits it running?
airlied: adamk_: I've fixed a few things in the last few days.
jcarlos: I don't know if this is the correct channel, but I don't find one for kstars. Does someone use kstars with the readeon driver ?
airlied: but there is still some glitches on firefox and otherplaces
airlied: just little green spots sometimes.
adamk_: airlied, I run an update today, so it should be the latest. kernel 2.6.27.5-113, xorg-x11-server-common-1.5.3-5, xorg-x11-drv-ati-6.9.0-38
adamk_: I'm not in front of the computer, but I can grab some screenshots either later this evening or tomorrow.
adamk_: fvwm was fine :-) xfce4 was pretty bad. I'm assuming it's render related, as a result. nomodeset resolves the issue entirely.
spstarr_work: airlied: oh interesting bug fixes ;-)
spstarr_work: oooops=)
xglurb: hi all
xglurb: I am using radeon driver that comes with ubuntu 8.10 (6.9.0+git20081003) with Xorg 7.4. I have a _big_ slow down from last version, and an immense slow down compared to what I had when I installed on 8.04 radeon and X packages for DRM and EXA support
xglurb: I fully reinstalled (no update), no errors shown from Xorg.log, and EXO acceleration is enabled (when disabled, no visible changes)
xglurb: I observe very high CPU usage when 2d is loaded (change workspace for example) and very slow 2d drawing (i can see the refresh when changing tab on a simple page of code !)
xglurb: Finally, I have a radeon 9600 (mobility)
dmb: xglurb, you probably mean exa
dmb: what does glxinfo say?
xglurb: http://nopaste.info/e8598dfbe8.html
xglurb: the output
dmb: dunno
xglurb: ? sorry, i am not a native english speaker
dmb: looks fine to me
xglurb: ok
xglurb: the slow done is very noticeable
xglurb: hmm this is strange
xglurb: I tried to run gtkperf (which i did on last version, after enabling exa acceleration)
xglurb: and everything is ok
xglurb: but refresh of simple text files (code) is very slow
xglurb: I am thinking that this may appear much more when using QT-based softwares (i am using gnome)
xglurb: anyway, this is not related to radeon driver
dmb: oh
xglurb: maybe something related to AA in qt app ? I don't know
xglurb: I am developping a QT-based app, which has the same problem
xglurb: since ubuntu 8.10, rendering is _much_ slower
xglurb: and kdevelop has too
spstarr: airlied: im in kms + EXA with exa comp on
spstarr: [ 157.093106] [drm:drm_buffer_object_validate] *ERROR* Failed moving buffer. e6000100 1444 4000027 2000020
spstarr: [ 157.093106] [drm:drm_buffer_object_validate] *ERROR* Out of aperture space or DRM memory quota.
spstarr: but it's continuing
spstarr: RADEON DRM CS failure - corruptions/glitches may occur -12
spstarr: bufmgr: last submission : r:0 vs g:22016000 w:0 vs v:19375713
spstarr: your patch to try in vain to not die...
spstarr: continues me
spstarr: interesting
spstarr: Interrupt enable: 02000001
spstarr: Interrupts received: 32908
spstarr: Current sequence: 19666 19666
spstarr: Counter sequence: 19666
spstarr: CS: 9091
spstarr: RADEON_CP_RB_WPTR 00038ef0
spstarr: RADEON_CP_RB_RPTR 00038ef0
airlied: spstarr: -53 on its way
spstarr: :)
airlied: spstarr: try -53 please :)
spstarr: getting
spstarr: testing now..
spstarr: no errors..
spstarr: AGP is off right now?
spstarr: i should try it *on* now
spstarr: brb
spstarr: [ 1.868184] [drm] Forcing AGP to PCI mode
spstarr: brb
spstarr: airlied: oh interesting
spstarr: im in AGP mode
spstarr: MASIVE CORRUPTION
spstarr: let me video.. i cant even see what im typing properly
spstarr: but.. it hasn't locked up yet
spstarr: gnome germinal text is ok
spstarr: no errors
spstarr: Interrupt enable: 02000001
spstarr: Interrupts received: 23952
spstarr: Current sequence: 16988 16988
spstarr: Counter sequence: 16988
spstarr: CS: 8657
spstarr: RADEON_CP_RB_WPTR 000285e0
spstarr: RADEON_CP_RB_RPTR 000285e0
spstarr: [ 0.914011] agpgart-intel 0000:00:00.0: AGP aperture is 256M @ 0xd0000000
spstarr: [ 2.308013] [drm] AGP mode requested: 4
spstarr: [ 2.308013] agpgart-intel 0000:00:00.0: AGP 2.0 bridge
spstarr: [ 2.308013] agpgart-intel 0000:00:00.0: putting AGP V2 device into 4x mode
spstarr: [ 2.308013] pci 0000:01:00.0: putting AGP V2 device into 4x mode
spstarr: This is the corruption i'm seeing ;-)
spstarr: pretty funky
spstarr: no mouse cursor -->
spstarr: airlied: i have video
spstarr: airlied: http://www.sh0n.net/spstarr/agp-kms-corruption.ogg
spstarr: it has not locked up though
spstarr: goes back to non AGP .. airlied this is encouraging however..
spstarr: before this would have locked up long ago WITH AGP on
spstarr: corruption without lockups is good
robokopp: spstarr the video is 403
spstarr: sec
spstarr: try now
spstarr: http://www.sh0n.net/spstarr/agp-kms-corruption.ogg
robokopp: works
spstarr: back to kms PCI mode
spstarr: is it possible EXA was causing corruption to the command submission? but the fixes + removal of flushing has fixed that?
spstarr: now we're just getting garden variety corruption
airlied: spstarr: not EXA was just making the driver send dodgy crap to the card.
spstarr: oh
spstarr: airlied: im ready to test AGP again when you resume that side
spstarr: the stability of it not locking was there
Remosi: Has anyone looked at Matthew Garrots stuff?
spstarr: i will keep EXA + exacomp mode, and not enable COMPOSITE for a while tonight
spstarr: if it passes, then enable eyecandy
airlied: Remosi: bits of it.
airlied: spstarr: its wierd that video :)
spstarr: yes
spstarr: note the text in the gnome-terminal wasn't corrupt like that the video was saved 1/2 sized
spstarr: there was little corruption in gnome-terminal
RyanPrior: Which cards should I look into buying if I want 3D and xrandr support that works right now?
MostAwesomeDude: X1950XT is the top of the market that we support, but X1400 or X1650 might be more in your budget.
MostAwesomeDude: :3
RyanPrior: =D
RyanPrior: an X1950 on Newegg is only 70 dollars - is that considered out-of-budget for many?
MostAwesomeDude: is a radeon dev
MostAwesomeDude: No, but OTOH keep in mind that our support doesn't reach the top of the market.
MostAwesomeDude: So it's to be expected that our "best" cards are cheap.
RyanPrior: naturally
MostAwesomeDude: But yeah, for me, I'd rather grab a $20 X1400, but I have a very small budget.
MostAwesomeDude: The irony, of course, being that I've got an HD3650 sitting next to me.
RyanPrior: =D
RyanPrior: If I got an X1400, would that be enough juice to power my 22-inch screen?
MostAwesomeDude: Probably, yes.
RyanPrior: I'm seeing cards listed by categories like "HD 2000 series" and "HD 3000 series" rather than X1400, X1950, etc. How do they translate?
RyanPrior: Also, curse ATI's marketing department for making everything so complicated.
MostAwesomeDude: HD 2000 is the card right after the X1950.
MostAwesomeDude: Not so complex; the numbers just keep going up.
MostAwesomeDude: There's usually a X1k category or such.
RyanPrior: So, X1650 looks like a compromise.
RyanPrior: I can get one for pretty cheap, but it's got 512MB of RAM and an 128-bit interface
MostAwesomeDude: Yep.
RyanPrior: Well, y'all have made a sale for ATI. If nvidia starts being more nice to the free software crowd, they might be back in the running next time I want to buy a card.
robokopp: or buy a used card, i always found that used cards are much cheaper
MostAwesomeDude: robokopp: When they work, yes. :3
robokopp: i never had problems :)
MostAwesomeDude: They're still selling new-in-box X1650s, no reason to go find one on eBay for only $10 less.
robokopp: ok dont know about ebay, but there are many hardware forums and its usually cheaper there
spstarr: airlied: im still in kms + exa/exacomp and its stable.
airlied: yeah breaking radeon to fix intel isn't on my plans, I can pull the latest keithp patch in, but it needs more testing
airlied: oops wrong channel
spstarr: smirks
spstarr: that is such a slip :)
Rabenklaue: hi, are there any known issues with the free radeon driver and the "hibernate-ram" script?
Rabenklaue: When being in console without having loaded the radeon driver hibernate2ram and the wakeup works quite fine, but while started a X-server instance or having loaded the radeon driver at all, my system won't wake up and the info/error led is blinking. :(
Rabenklaue: cat /var/log/Xorg.0.log | grep "(EE)" doesn't show anything
Rabenklaue: and cat /var/log/Xorg.0.log | grep "(WW)" you can see here:
Rabenklaue: http://rafb.net/p/VQIoHq63.html
Rabenklaue: my xorg.conf looks like this:
Rabenklaue: http://rafb.net/p/XOR4Js65.html
Rabenklaue: kernel /boot/linux-2.6.27-gentoo-r1/vmlinuz root=/dev/sda2 usbcore autosuspend=1 resume2=swap:/dev/sda5 acpi_sleep=s3_bios,s3_mode
Rabenklaue: that's how my grub kernel line looks like