Home Phoronix Phoronix Forums X.Org Videos From FOSDEM 2008

Radeon IRC Logs For 2008-7-09

Search This Log:


MostAwesomeDude: Gah, gotta go get food.
MostAwesomeDude: As soon as I sign off, bridgman will sign on.
MostAwesomeDude: Just watch.
MostAwesomeDude: ...Did I miss him?
dmb: yup
MostAwesomeDude: See, what'd I tell ya.
airlied: MostAwesomeDude: nope.. I think you need new light bulbs :)
dmb: o/j
MostAwesomeDude: airlied: XD
MostAwesomeDude: Wait.
MostAwesomeDude: Light bulbs?
MostAwesomeDude: Is that Aussie slang?
airlied: MostAwesomeDude: nope.. for the bridgman signal.
MostAwesomeDude: Oh.
airlied: I didn't see no ATI logo in the clouds.
MostAwesomeDude: Hmm.
MostAwesomeDude: Oh. I put in the AMD template instead.
MostAwesomeDude: Maybe I need theatrical smoke and lights, too?
MostAwesomeDude: Maybe he's looking for his cape?
dmb: maybe he's asleep :P
MrCooper: merely mentioning 'bridgman' used to work within minutes :)
MostAwesomeDude: MrCooper: It depends. This late, he's prolly asleep.
MostAwesomeDude: I just wanted to get it somewhere into the backlog.
MrCooper: he definitely should be, but I'm pretty sure I've seen him up way beyond this :)
MostAwesomeDude: Yeah, the funny thing about insomnia is that it never kicks in when you need it. :3
spiritssight: Is there any one here that could tell me if I can use the open source drivers for ATI x1300 mobile to use mythtv
legume: got a HD3850 AGP for his birthday :-)
legume: If I want to add it's PCIID - what do the last fields in radeon_chipinfo_gen.h mean?
spiritssight: legume do you know how to setup ati driver open source so that I could watch TV
agd5f: spiritssight: yes, the x1300 is supported
spiritssight: my gf geting mad at me not able to watch TV for a week now
spiritssight: could you PLEASE PLEASE help
agd5f: spiritssight: what's the problem?
spiritssight: when I go into mythtv its blank for watch TV and I was told its a video setup issue
spiritssight: I have the log if you want to see it
agd5f: spiritssight: sounds like a mythtv issue
agd5f: if mythtv requires Xv, make sure you ahve the latest drm installed as that is requierd for Xv on r5xx cards
spiritssight: how do you install XV
spiritssight: Xv
agd5f: spiritssight: the driver and xserver provide it
TobiasTheCommie: woohoo, just got new telly, so no longer need tv-out, now i just need to get opengl to work on my rv530.. guess i'll start seeing if xbmc can work correctly with it next week(since compiz should work with the git version, well, it might work with xbmc as well)
MrCooper: what's xbmc?
TobiasTheCommie: not really relevant here
TobiasTheCommie: just wanted to rejoice
TobiasTheCommie: xbmc = xbox media center
TobiasTheCommie: being ported to linux, still alpha
bridgman: MostAwesomeDude: reading the TGdaily article now; interesting
bridgman: not sure if it's a Good Thing or an IP trap
bridgman: btw there really aren't alarms when my name appears in IRC
bridgman: unless you type it three times of course ;)
glisse: bridgman: while you here do you know anythings on cmdfifo format ? :)
bridgman: specifically...?
glisse: a sec
agd5f: glisse: reg specs don't say anything about the format
glisse: bridgman: http://pastebin.com/m619ee46b dump from cmdfifo at lockup time :)
glisse: i don't understand command fifo content
glisse: agd5f: airlied was saying that r200 doc speak of register/data pair
glisse: but on r200 there is 2 reg to dump cmdfifo
glisse: while on r300 according to doc only 1
agd5f: bridgman: regs 0xe70, 0xe74, 0xe7c
glisse: i am good at asking hard question :)
glisse: i should start a show on tv with such question ;p
bridgman: there are 64 entries so that argues for a single register I guess
bridgman: or are you concerned that's a circular argument; there are 64 cause doc told you 1 reg ;)
jjl: hi guys, i bought yesterday a HD3650 (http://www.kmelektronik.de/main_site/main.php?action=Product_Detail&ArtNr=18127&Shop=0). Im using 2 analog monitors with dvi-vga adapters, but mirror-mode does not work.
jjl: somebody an idea?
agd5f: jjl: what's the problem?
glisse: bridgman: well what i don't understand is how is stored the register where data should be writen
glisse: bridgman: from my understanding the fifo hold register offset/data to be written
jjl: agd5f: sorry, i mean the mirror-mode is only working
glisse: but i can see anythings that looks like that
jjl: both monitors show the same
glisse: r200 had 2 register to read one command fifo entry
glisse: one register was to get register offset
agd5f: jjl: use xrandr to setup dualhead
glisse: one to get register data to be written
agd5f: jjl: http://wiki.debian.org/XStrikeForce/HowToRandR12
bridgman: yeah, all the different interpretations make sense; I'm trying to get a definitive "what is connected to what" diagram of the chips so maybe that will help
glisse: bridgman: someone still remember how the things is wired ? :)
jjl: agd5f: this does not work, if i use "xrandr --output HDMI-0 --right-of DVI-0" nothing happens
agd5f: jjl: you'll probably need to assign one to another crtc
agd5f: jjl: xrandr --output HDMI-0 --crtc 1 --mode
agd5f: then xrandr --output HDMI-0 --right-of DVI-0
jjl: agd5f: my xorg.conf -> http://rafb.net/p/sTPhsJ62.html
agd5f: xrandr should be fixed to choose independent crtc when non-clone setups are asked for
jjl: some frind has an HD3650 too but only other card (clubD3) and ist works
agd5f: jjl: it should work. got a log?
jjl: agd5f: http://rafb.net/p/1I3sfI38.html
jjl: agd5f: xrandr out -> http://rafb.net/p/sIpgpi14.html
agd5f: jjl: can you change modes ok? xrandr --output DVI-0 --mode 1024x768
jjl: i figured out, that if i first start x with the fglrx-driver (dualhead works) and then swith to free raeon-driver (only X restart) so all works correctly
agd5f: jjl: so it only works if you start fglrx first?
jjl: [17:40] jjl: so it only works if you start fglrx first? => yes
jjl: ok, with the fglrx-driver (under winxp too!) I'm not able to change the refreshrate of the 2. crt
jjl: agd5f: changing modes works
agd5f: jjl: how about turning off one head then turning it back on and then switching to dualhead
jjl: agd5f: xrandr --output HDMI-0 --off => dualhead goes off, but both crts stay on
jjl: agd5f: xrandr --output DVI-0 --off => both crts goes off
agd5f: jjl: weird
jjl: agd5f: dualhead works (my desktop has 2560x1024), the problem is, that both crts show the same
agd5f: jjl: yeah, got it
jjl: i don't know if this is important, but my video card has an white dvi-connector and the other is yellow
agd5f: jjl: shouldn't matter
agd5f: jjl: what version of the driver are you using?
jjl: i also tried the radeonhd-driver, but this was weird too
jjl: [17:59] jjl: what version of the driver are you using? => xserver-xorg-video-radeon 1:6.9.0-1 (Debian - Sid)
agd5f: jjl: does it work if you start X with only one monitor attached and then add the second one after X has started?
jjl: i will try it
MostAwesomeDude: bridgman: It occurs to me that if nVidia is really willing to just let anybody use CUDA, we could get it in writing that they won't sue us, and then add CUDA to Gallium.
Ori_B: MostAwesomeDude: so a CUDA based OpenGL driver?
Ori_B: that would be.. interestnig =P
jjl: agd5f: no this does not work, if i use "xrandr --output HDMI-0 --auto" the 2. crt stys off, but xrandr tell that its on (!)
MostAwesomeDude: Ori_B: No, a CUDA interface to Gallium.
MostAwesomeDude: Just like how the XvMC interface to Gallium doesn't use OpenGL either; it's making commands for the cards without needing to go through Mesa.
bridgman: MostAwesomeDude: yeah, I don't know the legal status of CUDA today. Don't think it is an open standard, it's more like DirectX, but will find out
BigBrain: MostAwesomeDude: However, a CUDA interface wouldn't be much of use in _Gallium_ as it's Nvidia only, isn't it?
agd5f: BigBrain: well, it's juat an API
agd5f: it's just that it's nvidia's and no one else uses it
bridgman: BigBrain: that's the whole question; an outside developer seems to be implementing Cuda on Radeon and getting support from NVidia to do so
BigBrain: ah, okay. Makes more sense then
MostAwesomeDude: BigBrain: It's no different from BTM or Brook+, it's just an API that somebody could write a Gallium driver for.
BigBrain: Gallium really offers __many__ possibilities, I see :D
MostAwesomeDude: bridgman: I believe that it is completely legal to implement DirectX, as demonstrated by Wine. I'm REALLY hoping that somebody makes a D3D frontend for Gallium. :3
MrCooper: s/driver/state tracker/ :)
bridgman: But the obvious trap there is that if Cuda exists on Radeon but we don't have rights to make a good implementation ourselves then that devaulues our API but leaves the only real good implementation on nvidia products
MostAwesomeDude: bridgman: Kinda, but Gallium has value that nVidia's driver doesn't have.
MrCooper: MostAwesomeDude: already done
MostAwesomeDude: MrCooper: ?
MrCooper: MostAwesomeDude: we did it in-house for one of our customers
MostAwesomeDude: MrCooper: That's awesome. Will you guys be sharing with the rest of us?
bridgman: makes sense
MrCooper: MostAwesomeDude: I think that's up to the customer; note that it's Windows specific
MostAwesomeDude: MrCooper: That's awesome. Will you guys be sharing with the rest of us?
MostAwesomeDude: People historically have used nvidia's driver over nv because it has so many more features, but now with Gallium+nouveau, the community finally is starting to have features that nvidia can't/won't support.
MostAwesomeDude: ...Stupid laptop keyboard.
MostAwesomeDude: CUDA is one of those things that nvidia might really try to hold on to.
agd5f: MostAwesomeDude: I don't rally see much value other than supporting nvidia's API, which is probably tuned to their hw
MostAwesomeDude: agd5f: Yeah. We don't have nv_f_p for that reason.
bridgman: I don't see CUDA being a good fit with Larabee, for example
bridgman: IMO it's the upper level APIs that are more important (GPGPU tools like Rapidmind, Brook), game APIs like Havok & PhysX etc...
MostAwesomeDude: Yeah.
MostAwesomeDude: Definitely.
bridgman: since they are high enough to be largely hardware independent
MostAwesomeDude: Actually, we shouldn't even be thinking of adding more state trackers until we get a Radeon pipe going.
bridgman: I don't think we're pushing CAL as a cross-vendor standard because neither it nor CUDA were designed for that; they're the gateway into specific hardware and drivers
MostAwesomeDude: kicks himself to get going on radeon pipes
bridgman: Gotta run. MostAwesomeDude, remember to pick up light bulbs ;)
MostAwesomeDude: bridgman: Will do.
MostAwesomeDude: See ya.
Ori_B: MostAwesomeDude: ah, I see what you mean.
dmb: MostAwesomeDude, if someone made a d3d frontend to gallium, would that mean easier porting of games?
MostAwesomeDude: dmb: One better. It means that you could bring D3D games directly over to Linux.
MostAwesomeDude: UT2004, for example, would need to change their input code a bit. That's it. They already use OpenAL.
dmb: MostAwesomeDude, well, that would require the use of wine anyway
dmb: oh
nha: basically, it means the Wine people wouldn't have to translate things into OpenGL calls
MostAwesomeDude: Okay, UT isn't a good example, but yeah.
MostAwesomeDude: nha: Yeah, that's the real benefit.
dmb: i thought ut used gl in the first place?
dmb: i guess not
MostAwesomeDude: dmb: On Windows it uses D3D.
dmb: oh
dmb: which reminds me, we will soon being seeing the steam platform on linux soon
MostAwesomeDude: Really?
dmb: in less then 2 months (According to close sources)
nha: huh
dmb: never mind
nha: other sources close to Valve say that's just a big hoax
dmb: nha, nope
dmb: nha, i'm friends with one of the people porting it
dmb: anyway, i shouldn't be talking about that in the first place, work time
beata: Hoy. I'm having framerate trouble with a Radeon 9550 and xorg driver.
beata: Has conf and log up at http://dalhagen.net/public/xorg.aiglx.conf and http://dalhagen.net/public/Xorg.aiglx.log, glxinfo says I have direct rendering enabled and that my vendor is SGI.
agd5f: beata: sounds like you have fglrx leftovers still around
beata: I shouldn't..but what would I check for?
agd5f: OpenGL vendor string should say DRI R300 Project
agd5f: beata: fglrx installs it's own libGL and a few other things
jaj: hmm, should hardware acceleration work on a Radeon Mobility X1400 (R500 based I think) ?
redeeman: jaj: in git i believe it should
agd5f: jaj: yes
agd5f: jaj: you need bleeding edge bits however
MostAwesomeDude: jaj: Yeah, use Mesa 7.1rc3.
jaj: I'm trying to make it work under OpenBSD but DRM support is quite experimental so I don't know if the problem comes from the radeon driver or from DRM
MostAwesomeDude: jaj: You'll need DRM 1.29.
beata: I loaded the output of glxinfo into the same place as the other files.
agd5f: jaj: BSD's are a crap shoot, the drm isn't as up to date
jaj: hmm I have Mesa 7.0.3 so that means no ATM I guess :)
agd5f: jaj: you need mesa from git for r5xx
agd5f: 7.0.3 is too old
jaj: agd5f: yeah, I'll just wait for time to pass until it gets imported
MostAwesomeDude: jaj: You might be waiting until Mesa 7.1 is released...
agd5f: beata: looks fine. what's the problem?
beata: I get under 1fps with anything that actually uses it, besides 'glxgears'.
agd5f: beata: use ldd to see if your apps are using some other libGL
agd5f: beata: you might also try LIBGL_DEBUG=verbose
agd5f: to see if there are any error messages
jaj: oh and I have another question, sorry if this is a FAQ but I couldn't find an answer. Why is there radeon and radeonhd? It seems to me like both drivers share the same goals.
tormod: jaj, see http://phoronix.com/forums/showthread.php?t=11244
beata: Uploaded to libgl_debug.log
jaj: tormod: thanks!
beata: It's saying something but I'm not sure what it means.
agd5f: beata: you need to disable low impact fallbacks
agd5f: in your .drirc
beata: I don't know what that is.
beata: Is attempting to look it up though.
arekm: it's a one of these options that allow people to write long posts on web pages explaining "how to get radeon working with x"
beata: I just found out about driconf. It says low impact fallbacks are disabled already.
agd5f: beata: your libgl debug output would seem to inidicate otherwise
agd5f: although maybe that line gets printed anyay
agd5f: yeah, looks like they are not enabled
agd5f: er disabled
beata: Hmm. Not sure. But anyway, having it create a drirc made the app speed up somewhat.
agd5f: beata: keep trying until that message disappears
beata: I think it's working. Getting around 18fps window'd now and no warning.
agd5f: there you go
beata: Now to work on another X server issue: disabling modes my monitor doesn't support.
beata: Which are all the VESA modes. :/
beata: Which isn't (I don't believe) radeon-specific.
agd5f: beata: the xserver adds a variety of modes which fit into the monitor's sync ranges as reported by ddc
beata: Yeah. I'm trying to figure out how to disable them though, since my monitor screwes up when displaying them.
agd5f: beata: you'll either have to edit the xserver code or use xrandr to remove them
agd5f: xrandr --delmode
xnguard: What's the current story on getting acceleration on R5xx? I just upgraded to an RV535 from an R423, lightly modified xorg.conf to reflect the new board, and DRI is no longer working. Do I need to go do something to mesa as well?
MostAwesomeDude: xnguard: You need Mesa 7.1rc3 or newer, and DRM 1.29 or newer.
xnguard: MostAwesomeDude: Aha.
xnguard: MostAwesomeDude: Do you happen to know off-hand if I need to add more overlays, or if the x11 overlay is sufficient?
MostAwesomeDude: xnguard: x11 has everything you need. You may need the mesa-9999 if there's no mesa-7.1rc3 yet.
xnguard: MostAwesomeDude: Cool. Thanks.
MostAwesomeDude: xnguard: Sure.
beata: I'm also kind of curious if I'd need to do anything special in xorg.conf because my video card has two subdevices.
agd5f: beata: ignore the .1 one, it's just a place-holder for windows 2000 driver model
beata: 01:00.0 VGA compatible controller: ATI Technologies Inc RV350 AS [Radeon 9550]01:00.1 Display controller: ATI Technologies Inc RV350 AS [Radeon 9550] (Secondary)
beata: Oh, huh.
xnguard: How much of the x11 overlay am I going to have to unmask? I'm up to x11-proto/dri2proto.
MostAwesomeDude: xnguard: I just used autounmask.
MostAwesomeDude: It's a LOT of stuff.
xnguard: MostAwesomeDude: Ah. How's that work?
MostAwesomeDude: xnguard: sudo emerge -v autounmask; sudo autounmask -n media-libs/mesa-9999
MostAwesomeDude: Should do it.
xnguard: nods.
xnguard: This is going to require an svn/git X server, isn't it? :)
MostAwesomeDude: :3
airlied: glisse: looked at moving DDX to command buf type submission but HostDataBlit bites it in the ass.. as we put the image data into the indirect buffer
glisse: airlied: well i think the proper scheme would be to do the new ioctl
glisse: which parse indirect buffer
agd5f: airlied: render uses immediate mode as well, so coords end up in the stream
airlied: agd5f: good point.
agd5f: we could also use the DMA engine for UTS/DFS, but it doesn't do tiling
sroland: hmmm. Isn't it possible to force-enable a dvi output in xorg.conf?
xnguard: MostAwesomeDude: Does 'autounmask x11-base/xorg-server-9999' cover it? That seems like a relatively small number of packages unmasked.
MostAwesomeDude: xnguard: Try and see. Should, but I don't know for sure.
xnguard: Okey-dokey.
glisse: agd5f: ok i think i partialy solved the problem now i only get cmdfifo busy and request on host interface set in rbbm status do you know anythings on this host interface could it be due to us reading to frequently the rbbm status ?
agd5f: glisse: could be. not sure. I'll see what I can find. what did you change?
glisse: i emit the packet3 indx_buffer with the packet3 draw_indx2 in an indirect buffer ib1
spstarr: agd5f: you didnt move to the Thornhill HQ for ATI/AMD?
spstarr: :)
spstarr: no
spstarr: still in VA you are
glisse: this way i don't the same lockup
glisse: ie vap doesn't go crazy
agd5f: spstarr: I work in the VA office :)
spstarr: ah there's an office there :)
agd5f: spstarr: there is an office, a small one ;)
spstarr: agd5f: what sort of people work in the old ATI HQ office?
spstarr: the board designers?
agd5f: spstarr: all kinds
spstarr: i see some of them when I go for lunch at the local foodcourt
sroland: agd5f: does the radeon driver somehow support force-enabling a dvi output through xorg.conf? Using the enable option doesn't seem to work since it'll complain there are no modes.
agd5f: sroland: I don't think those options work
sroland: agd5f: too bad...
agd5f: sroland: in theory maybe if you add modelines to you rmonitor sections
sroland: agd5f: I did...
sroland: agd5f: (since edid from this device is useless anyway...)
agd5f: the xserver should add them I guess, and maybe it's not?
sroland: agd5f: not sure it seems to work for VGA
agd5f: those options are handled in the core randr code
sroland: agd5f: true. Not sure though why DVI would be different from VGA in the randr code.
agd5f: sroland: yeah, me neither
agd5f: although your vga port might actually be detected as attached since it can do load detection
agd5f: DVI relies on edid
sroland: agd5f: though even with VGA it actually doesn't really enable the output but adding the mode at least worked
sroland: agd5f: hmm too bad guess I need some scripts to enable outputs then
sroland: agd5f: difficult to do with xrandr if all you don't get a signal...
sroland: agd5f: why does it rely on edid?
agd5f: sroland: how else can you tell if it is attached?
sroland: ah you mean for load detection
libv: "for"?
libv: sroland: issues with the tmds or the vga part of the dvi connector?
sroland: libv: this is strictly a dvi-d connector
libv: sroland: r5xx or earlier?
sroland: rs690
libv: ah, so it has a DDIA or LVTMA, and no working load detection we know of
libv: sroland: hpd should work rather beautifully
sroland: maybe I misunderstood what the enable option really does in the xorg.conf file, I _thought_ you could use it so a device is just enabled even though all detection says it's not...
sroland: libv: the problem is the device attached (a projector) apparently can't be detected if it's in standby so I thought I'd just bring up the server like it's on anyway.
agd5f: sroland: I think that was the theory, but in practice, I don't think it works
libv: sroland: yeah, not all monitors use the +5V to power the edid slave
libv: sroland: thing is, unless some assumptions are made about the monitor that is attached (or the user is able to provide valid information, or a previous result is cached), it cannot be guaranteed that the mode that is brought up eventually will actually be acceptable by the given monitor
libv: but i guess you already covered that part
sroland: libv: yes of course but that's what modelines are for
sroland: libv: I need a modeline anyway since in the edid data there's only one mode, which doesn't work at all....
libv: sroland: urgh
sroland: libv: fun isn't it... not sure if edid is broken or if it's due to interlace handling
sroland: libv: prefered mode would be a 1920x1080 interlace, however read-edid and the driver apparently read that as 1920x540
libv: i think that in this case, it could be the latter, apparently, for interlaced, you need to do more work
libv: unrelated: it is kind of funny though.
sroland: libv: but anyway this mode is not a good idea anyway since 1280x720 would be a better choice for this device...
libv: via dropped interlaced support to their hardware in 2003, thinking no-one would ever need it anymore
libv: they dropped it when moving from the savages to the castlerock/unichrome
sroland: libv: maybe there would be more modes in edid, get-edid claims there's one more block to get but it fails, so not sure.
sroland: interlace is sooooooo yesterday from analog 1930-days....
libv: sroland: get-edid is vbe based
libv: sroland: and my guess is that the vbe bios on the radeonhd hw is using the atombios calls
libv: sroland: most of which are barely able to do anything
libv: for DDC
sroland: is there some event somewhere generated when a monitor is plugged in?
libv: you usually can get the 128byte EDID block
sroland: yup that's what I got
libv: sroland: yeah, but then you need to be able to match the hpd pin to the right connector
libv: sroland: because you don't want to wildly enable _everything_ until you can find an edid slave
sroland: hmm I just thought running something like xrandr --auto on monitor detect
libv: sroland: but what if there's no working load detection for an output, no hpd handling in the driver and the monitors edid doesn't work, like here?
libv: i know this is asking your own question right back at you :(
sroland: libv: well I guess this still could sort of work, since when I switch on the projector xrandr shows it with the right mode (thanks to the PreferedMode option in xorg.conf), it's just not enabled.
agd5f: sroland: eventually we can get an interrupt in the drm for monitor plug events
libv: still doesn't solve the lack of hpd->connector handling
libv: sure, ddc is a good fallback, but it has a rate of failure equal or higher than hpd
MostAwesomeDude: glisse: Ping. What's up with radeon winsys?
Nyle: enouf, yo
enouf: yo
enouf: Nyle: some turkey in ##linux was asking, so i told him come here, if not using Debian
enouf: Nyle: [22:00:04] does anyone know how to get a radeon x1650 working on ANY DISTRO with 2d and 3d support?
Nyle: yeh
Nyle: use radeonhd from git
Nyle: even sid doesn't have shit atm
Nyle: is eagerly waiting for binary builds
enouf: Nyle: thanks, the weenie didn't even answer me - so F 'em .. :-P
enouf: Nyle: i just took him out of the b0z0 box earlier today - was in for about a week :-)
Nyle: heh
spstarr: airlied: interesting this plymouth stuff
spiritssight1: Hello would someone be willing to help me get my video setup correctly for mythTV
spiritssight1: I have ATI radeon mobilie x1300
spiritssight1: on a Dell E1505
spstarr: spiritssight1: TV IN?
spstarr: MythTV does not work for Radeon All In Wonder cards right now (if you have a AIW)
spstarr: it may work with TV Wonder cards i read
spiritssight1: don't have either
spiritssight1: I have a ATI Mobile radeon x1300
spstarr: whats that have to do with the radeon driver channel?
spstarr: MythTV aspect
spiritssight1: because mine is not setup correcly and also some thing to do with Vx
spiritssight1: this is my video card
spiritssight1: 01:00.0 VGA compatible controller: ATI Technologies Inc M52 [Mobility Radeon X1300]
spiritssight1: its was myth's logs that showed that something with the video was not working right