Radeon IRC Logs For 2008-6-03

dmb: argh, it really sucks when x crashes
MostAwesomeDude: dmb: Yes.
moondrake: someone should decouple the actuall state of the applications from the underlying rendering. So that you can just reset (crash/restart) the driver part without affecting apps
moondrake: only 10 years work or so 😛
dmb: moondrake, agreed though
MostAwesomeDude: moondrake: The problem with that is that 3D rendering is a state machine.
dmb: i went on a minimizing and maximizing spree for no apparent reason, and it just decided to crash
MostAwesomeDude: You would have to reset everything.
dmb: i have an idea, lets make a quantum video card!
moondrake: MostAwesomeDude “state” seems to indicate it is easy to resume, as oposed to “process”. But I admit that i have little knowledge of the 3D rendering. For 2D however I pretty sure it is theorethically possible
z3ro: MostAwesomeDude: I’ve asked bridgman about the reset procedure for the gpu’s (r300-r500, since r600 docs aren’t out yet) but I guess he hasn’t had time yet.
airlied: z3ro: its mostly mentioned in the docs.
airlied: the GA_RESET stuff.. glisse looked at it already I think.
MostAwesomeDude: There’s also the problem of exactly which layers you want to decouple.
dmb: also, i wish there was a better way to reset thing when a graphics hang occurs
z3ro: airlied: hmm iirc I tried that but it wasn’t successful 100% of the time.
dmb: like when keyboard and mouse stop functioning, and you can’t kill X
z3ro: airlied: maybe it needs other stuff like CP teardown first.
airlied: z3ro: when did you execute it? after the whole wait until hang stuff?
z3ro: and I’m not sure if it’s enough to reset it to the “just after box has booted” state (eg before any kernel module has been loaded, X started, etc)
eboettcher: err most of the time I have a unrecoverable crash, xorg is in some sort of race
airlied: z3ro: you most likely would need to reset the CP completely..
z3ro: airlied: I tried just during normal rendering of glxgears, and after a soft lock.
MostAwesomeDude: dmb: If the graphics card is hung, then it doesn’t seem like there is any sane way to reset it without a hard reset.
airlied: eboettcher: its just waiting for the kernel waiting for the gpu
eboettcher: airlied: 100% cpu usage waiting for the kernel?
z3ro: MostAwesomeDude: I’m sure there must be a way to reset the hw to it’s initial state.
dmb: MostAwesomeDude, there should be…
MostAwesomeDude: I mean, you can’t reset it automatically because there’s no way to detect a hardlock, and you can’t reset it by hand except through a hardware switch.
eboettcher: that’s user time as well
airlied: eboettcher: yes…
dmb: why can’t they add that to a video card?
MostAwesomeDude: I mean, it’s semantics, that’s all.
airlied: eboettcher: its stuck in an ioctl which returns and re-enters.
dmb: a way for the driver to reset the card
z3ro: MostAwesomeDude: yeah it wouldn’t work for a hard lock up, but most of my lockups are soft lockups.
z3ro: MostAwesomeDude: eg sshing in would still work.
MostAwesomeDude: z3ro: Ah. That is different.
dmb: yeh, same, i can usually ssh in when my display is locked
z3ro: MostAwesomeDude: yeah. as I understand it a hard lockup is when the gpu brings the pci bus down with it. in which case you’re screwed unless you’ve got jtag or something.
eboettcher: airlied: grr.. I’d look at where that happens, but GDB has trouble attaching to xorg in that state
dmb: never tried killing x to see what would happen though
eboettcher: apparently the latest changes to gdb fix that on linux
moondrake: MostAwesomeDude: “hard” depends on definition. even a dma hang or something is fixable. Just flush the buffer, and lose some pixmaps.
eboettcher: and I have yet to reproduce while X is running under gdb
MostAwesomeDude: dmb: If you kill X, you usually get a hardlock.
z3ro: eboettcher: hmm interesting. I’ve had the same problem. gdb wont’ attach.
airlied: eboettcher: its just a loop around the buffer ioctl..
z3ro: MostAwesomeDude: right. I’ve seen that too.
airlied: turn on drm debugging you’ll see it..
dmb: i can reproduce a lock right now, use the switch users function which brings up another x server, then have that server exit
dmb: it soft/hardlocks my box
eboettcher: airlied: with no timeout counter, timer, or something else?
airlied: eboettcher: well the problem is what do you do when it times out?
airlied: eboettcher: hang the machine 🙂
MostAwesomeDude: …airlied beat me to it.
z3ro: this is where it would be nice to have good hw reset code.
MostAwesomeDude: Well, what could you do, really?
dmb: exactly
moondrake: airlied: call software suspend
eboettcher: well there has to be some other behavior that can be done because it even impedes telling the machine to reboot!
moondrake: it will save my work and reinit the card on boot 😛
airlied: moondrake: if only..
MostAwesomeDude: Dump an error to syslog, return an ioctl error EHWHNG (hardware hang)?
airlied: moondrake: the card is too far gone then..
dmb: what does windows do in that kind of situation?
z3ro: MostAwesomeDude: even if it just saves rebooting, and resets the card to it’s initial state, then kill X, reload drm, and go from there.
eboettcher: tell the clients that the X server is dead, let them deal with that how they will
MostAwesomeDude: And then Mesa has to find a sane state to start over from again. Do we just buffer up commands until the next glFinish?
dmb: tbh, i have never seen a display lock in windows
eboettcher: dmb: I have 🙂
eboettcher: and my experience with ‘vpu recover’ has been “if it gets called, reboot”
eboettcher: else you’re in for another crash 😛
MostAwesomeDude: It seems to me like we COULD just have Mesa deal with that by dying with a nice, verbose “DRM timed out waiting for card response; here’s some info to send to the devs.”
MostAwesomeDude: Just kill the entire app.
airlied: MostAwesomeDude: the X erver goes with it.
z3ro: I might play around with the GA_SOFT_RESET stuff again, and try bringing down the CP properly first.
airlied: MostAwesomeDude: it kinda needs the gpu to render.
MostAwesomeDude: airlied: Well, if we’re using AIGLX, yeah.
z3ro: my fear is that doing that would make the CP flush commands and just make it worse.
airlied: MostAwesomeDude: not just AIGLX.
MrCooper: MostAwesomeDude: doesn’t it do that already?
airlied: MostAwesomeDude: we use the GPU to do 2D and 3D renderingh.
dmb: hardware wise, is there a way to reset the video card without resetting the CPU?
airlied: MostAwesomeDude: same engine..
MostAwesomeDude: airlied: Can’t we just suspend the xserver while we’re in DRM?
airlied: MostAwesomeDude: like we could make it not retry as often, but really you end up in a place its hard to get out off.
MostAwesomeDude: Have DRM handle the softlock by resetting the card on the spot, and then returning the error?
MostAwesomeDude: Or how much reinit do we have to do after the card hangs?
z3ro: brb food
MostAwesomeDude: Hmm.
MrCooper: MostAwesomeDude: basically from the level you reset it, I guess
MostAwesomeDude: I can sorta see how just stopping everything would be preferable to a full reset, though.
airlied: the main problem is if submitting the same info again crashes the card.
airlied: you can end up in a bad loop..
MostAwesomeDude: airlied: Well, then I guess we should take down X too.
MostAwesomeDude: Big error should necessitate a big response.
MrCooper: airlied: right, though it’s not like that isn’t happening now 🙂
airlied: MrCooper: true…
dmb: well, there is a reason people don’t use SIGSEGV to continue there program after a segfault…
dmb: SIGSEGV handler
MostAwesomeDude: dmb: If only we had a way to magically know which pointer fizzled!
osiris_: bridgman: hmm. so we upload texture to gpu address space, then we fill up vertex buffer with x,y,z,w,s,r,t,q and then what? we write TX_OFFSET with uploaded texture addresses?
osiris_: oh, he is gone
osiris_: agd5f?
airlied: osiris_: pretyty much , you set the texture unit up with width/height/offset etc.
arekm: whops, fast tcpdump output kills ability to switch between tabs in kde konsole even after (1-2 months old) fixes 8)
osiris_: airlied: ok. one more thing about stuffed textures. if the normal way of sending tex coords is by putting them in vertex buffer, how are they uploaded for stuffed textures?
airlied: osiris_: I think you setup the GA/SU to put the corrds into the rasteriser instead of using the supplied ones
osiris_: airlied: but where will GA/SU get those coords from, if you don’t pass them using vertex buffer?
airlied: osiris_: some GA regs maybe.. GA_POINT_T0 etc.
airlied: but I’m not 100% sure on how the stuffing works.
EruditeHermit: hey, is it possible to set Adaptor1 as the default Video Adaptor?
glisse: z3ro: the reset procedure for r300-r400 is different from r500 in some way
glisse: i already tell you about it, basicly cut the clock, try to do like r500 reset and if its doesn’t work, cut the clock again and post the card
glisse: there is one last option which should always work but i don’t think we want to use this one
glisse: osiris_: stuffed texture are generated by gpu
glisse: osiris_: ie it’s not you who send them
glisse: you just route them inside the gpu
EruditeHermit: I want to make the TexturedVideo adaptor the default adaptor
elbeardmorez: join /xorg
elbeardmorez: doh!
z3ro: glisse: whats the always work option?
z3ro: I would like to get comments from amd too on how they do it… but I’ve asked bridgman twice now, so…
z3ro: brb
MostAwesomeDude: z3ro: ‘k.
glisse: z3ro: the always work option is to heavy and too dangerous
glisse: z3ro: try the clock trick
glisse: it should work
glisse: you have to wait a bit and bang several times register
z3ro: glisse: well I’m just interested in case it doesn’t. last time I tried it didn’t work for me.
glisse: amd is doing 16x times the procedure
z3ro: so I’ll try the heavy reset too.
glisse: i can’t give you heavy one but the idea is to reset the pcibus
glisse: so it’s like rebooting
glisse: z3ro: i am pretty sure the clock tricj should work
glisse: did you waited enough time ?
glisse: stop cp, stop clock, wait stop clock, … do reset procedure
z3ro: iirc I waited 15ms
glisse: well once you done you should see rbbm_status changing
z3ro: btw did someone at amd tell you this procedure, or it’s guessed from the docs? I’m confused as to where it came from.
airlied: I think someone said 50ms..
z3ro: brb, restart X
glisse: yup better wait 50ms
glisse: but it seems you to repeat several time the procedure
glisse: to make sure the hw get confuse and stop clock
Zajec2: z3ro: i’ve request for you, could you make some work on valgrind-mmt, please? i mean rebasing this to the newest version of valgrind
z3ro: Zajec2: yeah I can do that, it might have to wait until tomorrow after work, though.
Zajec2: oh, would be great 🙂
Zajec2: finally we could kill ddc bug 🙂
z3ro: ok, time to try this reset code… this will most likely end badly. 😉
z3ro: or maybe not if I’m really lucky. 😛
z3ro: hmm didn’t really seem to do anything.
z3ro: which is really weird since it should bring down an important clock. =/
glisse: it worked 😉
Zajec2: glisse: 😀
MostAwesomeDude: Wait, I missed this adventure. What’s going on?
glisse: MostAwesomeDude: z3ro was trying some reset code
glisse: obviously it did reset somethings 🙂
MrCooper: EruditeHermit: you may want to take a look at the intel driver, I think it has an option for this
MostAwesomeDude: Yep. :3
airlied: Zajec2: http://people.freedesktop.org/~airlied/mmt_diff.patch
airlied: Zajec2: that might apply on a valgrind checkout.
EruditeHermit: MrCooper: man intel? All I can see in this is that I can turn off XVideo. If I do so will it default to Textured?
MrCooper: EruditeHermit: the code; it’s not implemented yet in the radeon driver
EruditeHermit: MrCooper: TexturedVideo is in the latest git snapshots
z3ro: it really sucks trying to debug this without a second box. =/
MrCooper: EruditeHermit: sure, I mean the option to choose the default adaptor
EruditeHermit: oh
EruditeHermit: ah well
EruditeHermit: thanks anyway then
MrCooper: np
osiris_: glisse: I think I know enough now, to implement stipple lines. one more question, what if all 16 texture units are already used for line textures?
MostAwesomeDude: osiris_: Put out an error and fallback, I guess.
MostAwesomeDude: But I don’t think you’re going to find that to be a problem very often.
osiris_: MostAwesomeDude: it would be good to implement it that it never fallback
z3ro: osiris_: afaik mesa still doesn’t support more than 8 texture units, so just use the last unit, and that would be fine (for now, anyway)
MrCooper: has anyone seen a real app that even uses 8? 🙂
z3ro: MrCooper: newer games do, but they would probably run at all of 1fps anyway. 😉
MostAwesomeDude: MrCooper: Not in my experience, no.
MostAwesomeDude: Don’t you only consume one tex unit per texture per quad?
MostAwesomeDude: So you have to be multitexturing to consume multiple tex units?
osiris_: MrCooper: what about texture mipmaps?
MrCooper: MostAwesomeDude: yes
MrCooper: osiris_: those are handled by the texture sampling hardware
MostAwesomeDude: MrCooper: I know that WoW can use up to three texs at max quality.
MostAwesomeDude: But beyond that, I dunno.
MostAwesomeDude: Out of curiosity, how would one get render-to-texture going, and what apps actually use it?
osiris_: ok, will use last unit
MrCooper: anyway, using 8 textures for stippled lines seems extremely useless 🙂
MostAwesomeDude: osiris_: Try to use the last unit present.
MostAwesomeDude: Although FP doesn’t really care; it’ll just peel off tex units in the order it finds them. :#
MrCooper: MostAwesomeDude: using GL_EXT_framebuffer_object, I think most modern games use it for effects like dynamic shadows or motion blur
MostAwesomeDude: MrCooper: Are there that many apps out there using stippled lines, and if so, how many line types do they need?
osiris_: MrCopper: I will use only one unit
MostAwesomeDude: MrCooper: What happened to stencil buf and accum buf?
MrCooper: MostAwesomeDude: different kind of shadows, and we’ve never accelerated the accumulation buffer
Zajec: airlied: it was near… 1 of 6 hunks failed for coregrind/m_syswrap/syswrap-generic.c
Zajec: the rest was fine
MostAwesomeDude: MrCooper: This is true. I don’t think we can accel the accum buf, either.
MrCooper: MostAwesomeDude: e.g. ut2004 uses FBOs for soft character and vehicle shadows (and putting the player name on the car licence plate 🙂
MostAwesomeDude: Ah.
MostAwesomeDude: We need better mem management before we can do FBOs, though.
MrCooper: right
glisse: osiris_: there will never be more than 8texture unit in use
glisse: osiris_: hw can only manage 8 texture coordinate
glisse: additional texture unit is for clever texture scheduling
glisse: osiris_: also when you draw a line stipple you are the only one drawing ie think as if you were the only one ussing the card
glisse: gpu are stream processor, they are not multithreaded the way a cpu is
bridgman: glisse, z3ro: I found some more reset code in tcore; hope to get that our RSN
bridgman: think we’re finished the last full review, just chasing details now
glisse: bridgman: great
bridgman: trying to get a handle on how often we use the hard reset option
bridgman: as glisse says, don’t think it’s a real good idea 😉
glisse: wish to be just chasing details in his thesis too 🙂
bridgman: glisse: I cheated; Alex did most of the work
glisse: 🙂
bridgman: z3ro: glisse and I were talking about reset at XDC
MostAwesomeDude: bridgman: I wish I had that option.
MostAwesomeDude: *) Write code.
MostAwesomeDude: *) Look at glisse and airlied’s ramblings in IRC and run it through lint.
MostAwesomeDude: *) Make agd5f write it.
MostAwesomeDude: *) Play moar UT2k4.
MostAwesomeDude: It’s almost like our very own /. poll.
bridgman: 😉
bridgman: osiris: as I understand it stuffing textures basically means that you don’t
bridgman: pass coordinates down from the app, you tell the GPU to force specific
bridgman: start & end coordinates for the texture
bridgman: mostly used for points & lines, ie things which don’t use textures in a
bridgman: normal way
bridgman: The important thing when you’re visualizing how a GPU works is to remember that
bridgman: about half the hardware you see is “scanning across a triangle” and breaking
bridgman: it down into a bunch of fragments to be drawn, and the other half of the hardware
bridgman: only knows about a single fragment / pixel and works with just the data handed
bridgman: to it by the first half of the hw
bridgman: – application passes texture coordinates and colours for each vertex
bridgman: – interpolators in the GPU create per-pixel coordinates and colours
bridgman: – (think about a triangle with different colors per vertex then imagine
bridgman: – same thing being done with texture coordinates)
glisse: this is related to the famous half theory of is glass half full or half empty
bridgman: – each pixel has the shader run and is handed the precalculated texture
bridgman: coordinates for each tex unit, and the precalculated colour
bridgman: If this sounds like a heck of a lot of precalculation that’s one of the
bridgman: reasons multitexturing was so expensive (and would still be expensive today
bridgman: if we weren’t using tiny fab processes)
bridgman: glisse: even worse, 16 glasses all being drunk at the same time, each
bridgman: with different amounts of water; how full is the glass ?
glisse: bridgman: theory i am thinking of don’t apply for water 😉
glisse: at least where never experimented with water 😉
bridgman: it’s 6:44 am; I’m thinking water or coffee only 😉
bridgman: yikes it’s 6:44 am, gotta go bye 😉
glisse: oh, it 12h over here, apperitif time 🙂
MostAwesomeDude: 3:45 AM here; I should go to bed.
z3ro_: just missed all that…
z3ro_: reading now
z3ro_: cool, sounds like amd will release that code. 🙂
algol: Is XAA still developed on 9200 cards (that is, if I find a bug is it worth to report it)?
agd5f: algol: it’s probably a bug in your xserver
agd5f: there were some breakages found in the XAA core
algol: agd5f: ah, ok, I’ll wait for next release
adamk: I’m not sure if this is a bug or not… But I’m getting lots of screen corruption with compiz if I set MigrationHeuristic to greedy. http://68.32.29.130/Screenshot.png
adamk: If I comment out the line where I set that, it works fine.
adamk: It did previously work, however.
onestone: agd5f: any idea what could cause the performance drop for glx clients after suspend on r500. EXA performance seams to be stable.
adamk: And, for some reason, xrandr shows 1024×768 as the maximum resolution allowed on VGA-0 with two different monitors (one of which can go as high as 1600×1200 and the other can at least do 1280×1024). This is with the latest xserver and ati driver. Here’s the Xorg log file: http://pastebin.ca/1037729
agd5f: onestone: no idea. suspend/resume is blackmagic. I’m guessing some bad state, but I’m not sure what.
agd5f: adamk: looks like your monitor’s edid is corrupt: (EE) RADEON(0): Unknown EDID version 255
adamk: agd5f, Two separate monitors, though?
adamk: I can try a third 🙂
agd5f: adamk: the other one on DVI-0 works fine
adamk: Yes, I mean two separate monitors on VGA-0… This is the second one to cap at 1024×768.
agd5f: adamk: maybe the port is bad
onestone: agd5f: is there any way to find out what is going wrong there?
adamk: Heck, I’ve even swapped the monitor on DVI-0 and VGA-0, and xrandr won’t go beyond 1024×768 on the monitor that was at 1280×1024.
adamk: And this happens with both an x700 and an x1300.
agd5f: onestone: dump registers before and after resume and compare 🙁
adamk: So I doubt it’s the port, either.
adamk: I can grab logs from every possible combination, if it helps resolve this issue.
onestone: agd5f: a lot of registers changed and changing in each dump
agd5f: onestone: yeah. maybe GB_TILE_CONFIG? make sure all the raster pipes are enabled?
onestone: agd5f: let me look at my dumps
onestone: agd5f: 00004018 00010011 (65553) stays the same
adamk: I’ll throw in my x700, swap the monitors so that the flat panel is on the VGA port and the CRT is on the DVI port and upload a new log file.
agd5f: onestone: does the bios post your card when it resumes?
onestone: agd5f: no idea?
onestone: how can I check it?
adamk: agd5f, http://pastebin.ca/1037737 . Now, with the x700 and the monitors reversed, both monitors are limited to 1024×768.
agd5f: adamk: in this case the driver isn’t able to get an edid from either monitor. could be lots of things unforunately. Is ddcprobe able to get any edid info?
adamk: Excellent question. I don’t have it installed. I’ll see if I can find it somewhere.
adamk: Heh…
adamk: I wonder if I can find the commit that broke it.. It definitely worked before, at least with the flat panel on DVI-0 and the CRT on VGA-0.
adamk: onestone, Here’s the ddcprobe output. I’m not sure I’m reading it properly, but it only seems to detect the flat panel and not the CRT: http://pastebin.ca/1037771
agd5f: adamk: ddcprobe may only query one monitor. try running it with only one monitor attached, switch monitors, repeat
adamk: Yeah, I pulled the other monitor and even rebooted. It’s not showing any timings now.
adamk: I’m going to confirm this with a completely different monitor, on both the x700 and the x1300.
agd5f: yeah, those x700’s seem to have flakey ddc
agd5f: adamk: see if this patch helps: http://www.botchco.com/alex/xorg/adamk.diff
adamk: Hmm… I’ll give it a try.
adamk: A third monitor, an old HP, does show up in ddcprobe.
algol: Is there a test suite for the radeon driver?
algol: I mean, something that can be run on one’s hardware to check some basic functions in a controlled environment.
GerbilSoft: algol: i think that’s what tcore is
GerbilSoft: (hasn’t been released yet, but it’s coming eventually)
glisse: algol: you want to test speed ?
glisse: or test if there is a hw defect ?
algol: glisse: well, basically if everything gets rendered as intended and there aren’t corruptions.
glisse: algol: for opengl or for x ?
glisse: for gl there is piglit iirc
algol: glisse: 2D, as a first step
glisse: glean for gl
algol: glisse: nice, I didn’t know piglit 🙂
glisse: for 2d i think X test suite or gtk test suite like one used in benchmark on phoronix should test for correctness of rendering but i am not sure
algol: glisse: ok, I’ll have a look at them, thanks
mcgreg: gtkperf – it is
adamk: agd5f, Sorry, the patch didn’t help.
adamk: Still getting “unknown EDID version 255”
adamk: So I have one viewsonic CRT and one sony CRT which give the EDID error on the VGA port of both an x700 and x1300. And a viewsonic LCD that gives that error on the VGA port of at least one of the cards, but works fine on the DVI port of both cards. And
adamk: And an old HP CRT that works fine on the VGA port of both cards (but is limited to 1024×768, so is pointless is using as a replacement for the other two).
adamk: Very odd.
Zajec: adamk: your x700 is normal or mobility?
adamk: Normal add-in card.
Zajec: adamk: did you try if fglrx can read edid?
Zajec: i have problem with reading edid on x700 mobility
adamk: I’ve used fglrx on this machine under opensuse and didn’t have any issues, but I’m not sure if that means that fglrx read edid…
Zajec: did amdccc was able to read supported resolutions?
Zajec: (and refresh freq)
agd5f: adamk: that patch only makes a difference on the x700
adamk: I never used amdcccle, to my knowledge.
adamk: agd5f, Ahhh, I’ll check now then..
adamk: I did just throw the x700 back in.
Zajec: agd5f: should i try your pathch in my case? do you rembmber my one?
agd5f: Zajec: you already tried it. didn’t make a difference in your case
Zajec: adamk: i ask as i was told than it can be checked how fglrx makes reading different from radeon driver
Zajec: ah, this one patch. ok
adamk: agd5f, Well on the x700 now I’m not getting the unknown EDID version error, but it’s still limiting VGA-0 to a maximum of 1024×768.
adamk: A search for “Monitor name” in /var/log/Xorg.0.log only shows the flat panel.
agd5f: adamk: ok
adamk: Is it possible to get it to a higher resolution anyway?
agd5f: adamk: yeah.
agd5f: it’s just limiting you because it doesn’t know what the monitor’s capabilities are since it can’t get the edid
adamk: I have to create a –newmode, correct?
agd5f: yup
agd5f: use cvt to generate modes
agd5f: e.g: cvt 1280 1024 60
agd5f: xrandr –newmode “1280x1024_60.00” 109.00 1280 1368 1496 1712 1024 1027 1034 1063 -hsync +vsync
agd5f: xrandr –addmode VGA-0 “1280x1024_60.00”
agd5f: xrandr –output VGA-0 –mode “1280x1024_60.00”
agd5f: for example
Zajec: wow, i didn’t know “cvt”
adamk: agd5f, That got it. Thanks 🙂
adamk: I can continue to test anything you might want… And, of course, I will try newer versions of the ati driver to see if it seems to get resolved 🙂
osiris_: glisse: how to allocate some memory in gart?
osiris_: glisse: nevermind, found it
stefanb: MrCooper: are the latest drm changes the vblank change you have been discussing?
MrCooper: stefanb: yes, plus the corresponding xf86-video-ati change
stefanb: MrCooper: OK, then I’ll update and try to run without vblank_mode=0
osiris_: agd5f: should textures be aligned to micro/macro block size in gart memory?
TobiasTheCommie: ehm, just wondering, the disable tv-out patch, does it even work? as far as i can see it shouldn’t work at all…
TobiasTheCommie: http://gitweb.freedesktop.org/?p=xorg/driver/xf86-video-ati.git;a=blobdiff;h=e71b02e875c2e499a4d6f7b6e1905c3049ffece4;hp=5cc21d5262172b6088f4e7f8bebbf365adec104d;hb=effa245914823371e052cd9aa1143a02350891e7;f=src/radeon_atombios.c
TobiasTheCommie: info->BiosConnector[i].valid = FALSE;
TobiasTheCommie: followed by
TobiasTheCommie: info->BiosConnector[i].valid = TRUE;
agd5f: osiris_: IIRC 4k, but it might be less need to check the tx_offset reg and see what teh align requirements are
TobiasTheCommie: a few lines afterwards
agd5f: TobiasTheCommie: you’ve missed the continue
TobiasTheCommie: oh, heh, yup, sorry…
quicksilver: my mesa only ‘claims’ to support openGL 1.3
quicksilver: is that significant?
quicksilver: in particular, does it mean it won’t do GL_GENERATE_MIPMAP which came in in 1.4?
GerbilSoft: GL_SGIS_generate_mipmap is supported according to glxinfo (RV530)
GerbilSoft: though mipmaps appear to be broken
stefanb: MrCooper: Cool, compiz running without vblank_mode=0!
MrCooper: cool indeed 🙂
stefanb: and googleearth running without lowimpact_… 0
MrCooper: yeah, airlied made that the default
MrCooper: off for tonight, bbl
gustaf1: googleearth is _extremely_ slow
stefanb: gustaf1: Not on my box. And that’s just a laptop
gustaf1: could be my drm or mesa versions..
GerbilSoft: what compiz works without vblank_mode=0 now?
stefanb: Yep. Just installed latest drm & -ati git master
GerbilSoft: oh so that’s what those vblank changes are
GerbilSoft: nice
stefanb: gustaf1: drm git master, drm & radeon kernel modules (same source), mesa git master, xf86-video-ati git master, and just to be sure recompiled xserver SRPM against libdrm-devel & mesa*-devel.
gustaf1: oh, well… i’ll wait for debian packages.. I use rather new mesa (a few days old), and just a few days old -ati and drm, but I guess there have been changes since which makes it a gazillion times faster..
stefanb: gustaf1: Have you tried “disable_lowimpact_fallback=true ./googleearth” ?
gustaf1: no difference….
gustaf1: but I do I know it’s not just software rendering?
stefanb: gustaf1: xdriinfo says “Screen 0: r300”?
stefanb: gustaf1: BTW, did you check in googleearth settings that it uses OpenGL?
gustaf1: yep
gustaf1: can’t find any opengl setting..
gustaf1: Renderer: OpenGL according to its About box
stefanb: gustaf1: BTW you do have a r5xx chip, not a r6xx?
gustaf1: r500.. well, I should update my packages (which means, I’ll wait)
gustaf1: modern mesa and xserver packages haven’t come to neither sid or experimental yet…
gustaf1: nor*
stefanb: gustaf1: Yeah, xserver 1.4.99 or newer might be required…
gustaf1: i’m desperately waiting 😉
stefan1: Boy this hotel WLAN is painstakingly slow…
GerbilSoft: hm
GerbilSoft: how do i add another resolution via xrandr?
GerbilSoft: i want to set my laptop to 1680×1050 using scaling (the native resolution is 1600×1200)
GerbilSoft: but 1680×1050 isn’t in the resolution list
stefan1: xrandr –newmode ….
stefan1: xrandr –addmode OUTPUT MODE
GerbilSoft: i added the 1680×1050 mode, but attempting to add it results in a BadMatch error
stefan1: GerbilSoft: You might need the newest xrandr though
GerbilSoft: david@gs_laptop.gerbilsoft.gsft ~ $ xrandr –addmode LVDS 1680×1050
GerbilSoft: X Error of failed request: BadMatch (invalid parameter attributes)
stefan1: GerbilSoft: Maybe 1680×1050 exceeds the monitor specs connected to the output?
GerbilSoft: the LCD panel is 1600×1200
GerbilSoft: i guess the scaler can only upscale
stefan1: GerbilSoft: Or it can’t upscale 1050 -> 1200
stefan1: GerbilSoft: Or it can only upscale in both directions at the same time with the same scaling facter
GerbilSoft: ok it seems that it can’t downscale
GerbilSoft: 1440×900 works
GerbilSoft: hm
GerbilSoft: possibly
GerbilSoft: i’ll try 1920×1200
stefan1: nope 1600 / 1440 = 1.1111, 900 x 1.111 = 1000 != 1200
GerbilSoft: oh hm
stefan1: GerbilSoft: So the scaler must support different scaling factors for X & Y
GerbilSoft: 1920×1200 doesn’t work
GerbilSoft: so i guess it doesn’t support resolutions higher than the LVDS panel’s native res
stefan1: GerbilSoft: Try VirtualSize 1920×1200 and then the new mode
stefan1: 1920×1200 is probably rejected now, because your screens virtual size is 1600×1200
GerbilSoft: i don’t think i have a monitor section (using autodetection)
GerbilSoft: i’ll test that out later, have work to do
stefan1: Hmm it looks like that without vblank_mode=0 compiz runs more smoothly now…
stefan1: Rotating the cube is really smooth now
mattmatteh: could anyone suggest where i can find out what the max resolution of radeon mobility m9 and m6
osiris_: agd5f: could you take a look at http://pastebin.ca/1037991? it’s stipple lines WIP. no frag prog yet, just generating and setting texture.
osiris_: shouldn’t this give me some visible results?
agd5f: mattmatteh: max in what regard? max texture size? max surface size? max mode?
agd5f: osiris_: you probably need to setup the RS and FP otherwise, it won’t be used
glisse: osiris_: also try with the default shader
glisse: ie the one which directly draw using supplied color
mattmatteh: agd5f, i was trying to find out what graphics the sam440ep had, i googled and found a link that suggest it has an ati radeon r250 Lf mobility 9000 m9 or m6. specifically what display resolution, my lcd is 1920×1200. any information would be helpfull. i went to ati’s web site and that just boast about its technology, nothing useful
glisse: mattmatteh: lspci should tell you what is your card
mattmatteh: glisse, i dont have the card. i am going by some other review on it that showed lspci
mattmatteh: glisse, i am researching before i buy
mattmatteh: glisse, it would be useless to me if it does not support 1920×1200
glisse: radeon r250 is old
glisse: very old
agd5f: mattmatteh: generally 2048×2048 for 3D/tiling, crtc is 4k IIRC
glisse: sam440p is a laptop ?
glisse: must be an old one
mattmatteh: a low power itx board
mattmatteh: agd5f, what do you mean by 4k
glisse: 4096
agd5f: mattmatteh: yeah 4096
mattmatteh: oh, total pixels ?
glisse: 4096×4096
mattmatteh: i was reading http://en.wikipedia.org/wiki/Comparison_of_ATI_Graphics_Processing_Units, but that didnt show resolution, unless i missed it
mattmatteh: oh
mattmatteh: is there any card that can not do 1920×1200 ? or gpu series
glisse: mattmatteh: well this depends also on the ramdac stuff i can’t remember how its on r2xx
agd5f: mattmatteh: shoudl be ok on all radeons at least at 60 hz
mattmatteh: yeah, my lcd is 60 Hz, as most
gustaf1: you’re talking about resolutions. great. When I try to run for instance compiz, it says maximum 3D texture size is 2048. What I’ve heard here is that I need xserver 1.5.. why is that?
otaylor: gustaf1: what card?
gustaf1: r500
otaylor: gustaf1: so you are testing out git mesa/drm?
gustaf1: yeah
otaylor: gustaf1: Weird. I can’t think of any reason why a) youd’ have that limit, unless someone just forgot to update something in the dri driver b) why xserver 1.5 would make a difference for that, though I’d definitely recommend it in general
gustaf1: I don’t want to use compiz, but I really don’t want to use metacity, and so I guess I don’t want to use compiz less than I don’t want to use metacity… So I just wanted to try it out.. Did that make sense? to me it did
otaylor: gustaf1: texture size limits should be 4096 for r500
gustaf1: yeah I hope so..
gustaf1: I’m running 3200×1200
agd5f: gustaf1: yeah text size for r5xx is 4k
agd5f: text = texture
gustaf1: I can disable the checks (in compiz) and then all I get is a complete white screen (can’t see nothing of _anything_)… but yes, the value is weird
gustaf1: I’ll try the abesolutely latest gits.. I have a few days old stuff
agd5f: gustaf1: you will need a newer xserver unless you built the r500 3D driver in an older mesa since mesa from git needs xserver from git
agd5f: but that has nothign to do with texture size
gustaf1: need? I run xserver 1.4 and mesa/drm/ati from git =)
gustaf1: well, mesa and drm are unofficial ubuntu packages, and I’m on debian..
gustaf1: but it “works”
gustaf1: well, glxinfo says GL_MAX_TEXTURE_SIZE = 4096, but perhaps I should just wait for xserver 1.5 before I initiate you into an investigation..
agd5f: gustaf1: well in that case you’ll probably have issues with AIGLX
agd5f: which is what compiz uses
gustaf1: ah..
gustaf1: I need to write something explicitly in xorg.conf for that?
agd5f: gustaf1: no. as I said, mesa from git requires xserver from git for that to work
gustaf1: agd5f: ah ok..
legume: agd5f: I found a workaround for my monitor going into standby when switching vts or quiting X – if I start and then quit X with vesa driver it works OK after that for radeon.
mattmatteh: gustaf1, what all does max 3d texture size effect ? i dont play games. cant really, i only have a p3 799 Mhz now. i was looking to replace it with that ppc board
Ori_B: mattmatteh: it’ll also affect the maximum window size in compiz, possibly the maximum screen size that can be accelerated.
mattmatteh: Ori_B, i have never tried compiz
mattmatteh: Ori_B, wasnt sure how stable it was yet.
mattmatteh: Ori_B, i was looking for something that can be a simple desktop. just better than what i have now
funda3: i’m getting “(EE) RADEON(0): Failed to determine num pipes from DRM, falling back to manual look-up!” with mesa/drm/xserver/ati all from git on 2.6.25
funda3: i have an agp 9500
Ori_B: mattmatteh: yeah. I was just answering your questions on what max texture size would affect
mattmatteh: ohh
mattmatteh: Ori_B, i dont understand all the details of graphics cards. only a rough idea. thanks 🙂
Ori_B: same here 😉
gustaf1: mattmatteh: I don’t really care about compiz neither, but metacity has started to act like a jack ass with my dual head setup…
gustaf1: and I read that latest compiz has a rewritten window placement ruitine, so I figured I could give it a try
mattmatteh: dual head seems to be popular lately ?
osiris_: glisse: how to create default program and bind it to primitive?
agd5f: funda3: your drm is old. you don’t have to worry about it unless you want to use GL
gustaf1: mattmatteh: I’ve used dual head for years, but it wasn’t until two weeks ago I bought two new kick-ass 1600×1200 S-IPS monitors
rx__: in what kernel is kmmio support coming back?
mattmatteh: gustaf1, i got my 1920×1200 about 9 months ago
gustaf1: currently, I can take advantage of some of their features, and marginal parts of the GPU, but things are moving in the right direction.
mattmatteh: i like one widescreen
rx__: kmmio trace
mattmatteh: i use it for desktop and movies
gustaf1: I like 3200×1200 😉
mattmatteh: gustaf1, i think the line down the middle would bother me
gustaf1: I don’t play and don’t watch movies. I write code, and like to have a browser/terminal/pdf on one monitor, and the code on another.
gustaf1: but if I would only use one monitor, then widescreen it would be yes. a _hell_ lot cheaper too
Ori_B: gustaf1: it’s quite likely you’re running into the same “over max texture size” limits that I was running into with my old desktop
Ori_B: gustaf1: 2048×2048 or less, it was just fine.
Ori_B: bigger, it fell back to software.
Ori_B: (remember, the root window is still a window)
gustaf1: Ori_B: yes the 2048-figure is bogus, but I’ll still need xserver 1.5 and I won’t really compile that myself, so I wait..
rx__: uh.. anyone know the build order for r5xx drm?
agd5f: rx__: libdrm then the kernel modules
rx__: where does X server go in that?
rx__: 🙂
rx__: X is last?
agd5f: everything else, drm, mesa, xserver, ddx
rx__: thanks
mcgreg: agd5f: dunno if care about that at all, but with the current git state it is possible to play WoW with wine 🙂 the gfx are broken (actually only icons) but in general the 3d gfx are all okay 🙂
agd5f: mcgreg: cool!
mcgreg: I was very surprised too , since it didnt work at the beginning but some commits later have changed something and it worked. though the speed is still very very slow 🙂 but in short – I love that
rx__: hum.. i wish cgit posted the git URLs
mcgreg: use freedesktop
rx__: uh..
rx__: what was the fix for
rx__: radeon_screen.c:848: error: .RADEON_PARAM_NUM_GB_PIPES. undeclared (first use in this function)
osiris_: rx__: you didn’t install libdrm headers
rx__: ah
rx__: hm.. the headers are there
osiris_: rx__: the newest one?
rx__: from mesa/drm
rx__: git
rx__: xf86drm.h and xf86mm.h are in /usr/include
osiris_: rx__: check for radeon_drm.h in /usr/include/drm
agd5f: rx__: make sure they are the newest ones and that the mesa build is actually using them
rx__: osiris_; oh.. it’s not defined there.. i guess it isn’t the newest
rx__: oh.. heh
rx__: i ran into this before..
rx__: can’t install linux-core because i unchecked AGP in my custom kernel
legend2440: hi i have a radeon 9600 on ubuntu hardy using the fglrx driver. i am playing with dual head and was wondering how to use radeon drivers instad of fglrx. is it as simple as changing xorg.conf to say radeon instead of fglrx?
tormod: legend2440: System->Administration->HardwareDrivers
mcgreg: legend2440: if you have radeon driver install, it is indeed as easy as that
mcgreg: legend2440: but you should uninstaleld fglrex anyway, since it installed his own libgl , as far as I know
legend2440: mcgreg: thank you.
tormod: legend2440: the GUI does the right thing
legend2440: mcgreg: would you happen to know which driver is easiest to to use for dual head? fglrx,ati or radeon? also is xrandr or xinerama needed?
mcgreg: well, radeon does support xrandr , si it should be rather easy with that but honestly I never did any dual head stuff so I am probbaly the wrong one to ask about
legend2440: ok
agd5f: legend2440: ati and radeon are the same driver
agd5f: dualhead works fine with radoen and xrandr
legend2440: agd5f: ok ty so all i would need to do is change fglrx to radeon in xorg.conf?
agd5f: legend2440: do what tormod said or 3D won’t work
rx__: huh… who maintains the XGI drm module?
agd5f: if you just want to try radeon, you can just change the config file assuming you installed the radeon driver
agd5f: rx__: idr
legend2440: yes radeon is installed. and i do have ati acceelerated graphics enabled in System->Administration->HardwareDrivers
rx__: thx agd5f
agd5f: rx__: np
tormod: legend2440: “ati acceelerated graphics” that sounds like fglrx – disable it -> switch to free driver
legend2440: tormod: yes it is fglrx. so disable it and change fglrx to radeon in xorg.conf is that correct?
tormod: yes, “ATI accelerated graphics driver” = fglrx
tormod: if you use the GUI you don’t need to manually modify xorg.conf !
legend2440: tormod ok that makes sense thanks
rx__: hey agd5f
agd5f: rx__: hey
^^MAg^^: airlied: any chances for new xorg server packages for f9?
airlied: ^^MAg^^: why?
airlied: I already have R500 support mostly line up for F9 updates-testing
airlied: just waintg on the kernel
^^MAg^^: I already installed your packages from koji
^^MAg^^: but composite misses something
spstarr: rawhide 🙂
^^MAg^^: compiz (core) – Fatal: GLX_EXT_texture_from_pixmap is missing
spstarr: I find it odd how fedora works rawhide is supposed to be where you do testing, not stable F9, seems strange to me coming from a Debian world
^^MAg^^: spstarr: I will switch when it stabilizes, I use this laptop for daily work 🙂
spstarr: it cant stabilize if F9-testing is getting NEWER code then rawhide
airlied: spstarr: go away.
^^MAg^^: 😀
airlied: ^^MAg^^: does using the desktop effect applet work?
airlied: ^^MAg^^: you might need the kernel from koji as well for r500..
^^MAg^^: yup I’ve got kernel from kjoji
^^MAg^^: Linux viper.entropy.be 2.6.25.4-38.fc9.x86_64 #1 SMP Wed May 28 01:43:41 EDT 2008 x86_64 x86_64 x86_64 GNU/Linux
^^MAg^^: Xv works flawlessly
airlied: cool, so libdrm .12 and Mesa .33 should be enough for it to work.
^^MAg^^: DRI/3d also
airlied: so try using the applet to enabled desktop effects rather than running compiz.
^^MAg^^: ok but where the hell is this applet?
^^MAg^^: aa got it
^^MAg^^: everything works
^^MAg^^: great!
^^MAg^^: one issue – Option “AccelMethod” “EXA” – this should be by default for r500
airlied: ^^MAg^^: yeah I want to do that.. next -ati might force it.
^^MAg^^: also
^^MAg^^: Option “XAANoOffscreenPixmaps” “true”
^^MAg^^: is this needed now?
rx__: if you’re using EXA my guess is no
rx__: 😉
GerbilSoft: gah this xvideo crash is getting irritating
GerbilSoft: so i had VLC running a DVB stream, and maximizing the window caused a GPU lock
GerbilSoft: at the end of the xorg.log was this line
GerbilSoft: “exaCopyDirty: Pending damage region empty!”
dmb: airlied, if there is going to be another mesa RC (rc2) will this mesa changes for the r5xx be in it?
airlied: dmb: most likely..
dmb: airlied, you think there is going to be an rc2?
airlied: dmb: no idea… I would guess so, he’s not going to directly release 7.1 from rc1.
dmb: i’m going to be package some stuff for a distro, would be a lot better if i packaged an rc as opposed to a git snapshot 😀
airlied: dmb: why?
airlied: dmb: for Fedora we just package git snapshots.
airlied: you just have to add all the fixes to the version you pick in any case.
airlied: if you are going to have a worthwhile distro
dmb: yes
GerbilSoft: airlied: any idea about the xvideo crash?
airlied: GerbilSoft: none, I’m not willing to test it now, can’t lose this desktop 🙂
GerbilSoft: haha
GerbilSoft: ok
airlied: if its reproducible it sounds debuggable..
GerbilSoft: it’s definitely reproducible
GerbilSoft: also it logs this to xorg.log
GerbilSoft: “exaCopyDirty: Pending damage region empty!”
airlied: did you open a bug? 🙂
GerbilSoft: https://bugs.freedesktop.org/show_bug.cgi?id=15824
GerbilSoft: here
GerbilSoft: i’ll add that line as a new comment
airlied: cool I have the same hw at home 🙂
GerbilSoft: hm
GerbilSoft: if it is an exa problem, then turning exa off should make it work
GerbilSoft: i’ll test that later
GerbilSoft: with xaa, merely starting x-video causes a crash
GerbilSoft: i suspect that’s becuase xaa isn’t really supported for r500

Radeon IRC Logs For 2008-6-03

Rig Lab Hardware

You Might Also Like

RadeonHD.org | Radeon IRC Logs For 2008-10-23

RadeonHD.org | Radeon IRC Logs For 2008-3-30

RadeonHD.org | Radeon IRC Logs For 2008-6-06