dmb: does anyone have any links to the tcore stuff?
rx__: not out yet
rx__: see http://www.botchco.com/agd5f/?p=20
udovdh: any info about R6xx?
dmb: rx__: but the 3d guide is out for the 5xx right now, just not code examples?
rx__: that's correct
dmb: argh, that 3d guide seems so complicated :D
dmb: is glad he isn't a video card driver develop :P
rx__: well it *is* meant for developers ;)
dmb: like, i'm a programmer, you have to be very special to understand this stuff
dmb: its so long also
dmb: rx__: are you one of those lucky people :D?
dmb: i wonder if nvidia is going to follow and release its specs also
rx__: i'm an interested observer.. i wouldn't consider myself a developer :)
dmb: rx__: ah, join the club :D
dmb: rx__: i'm happy to see a real video card with a real open source driver
rx__: i do send patches whenever i can though :)
dmb: i remember when everyone thought ati was the worst brand of video cards for linux :D:D
dmb: now its looking like nvidia
rx__: i can't speak for nvidia since i've never used nvidia + linux
dmb: I have, wasn't a very good experience
dmb: ironically, the card also seemed to have burned up on linux
dmb: don't know if it was the result of overheating, because the temps were always normal
rx__: i'm just happy that there are many choices for drivers
z3ro: cool, just saw some 3d docs are out. :)
z3ro: I should probably go read the full article though. I just skipped to the download link. ;)
z3ro: heh at least I have some reading after I return home in a drunken state. :)
airlied: z3ro: you can start fixing r300_reg.h ;-)
z3ro: airlied: yeah. I'll probably have r300_reg.h open when I do the first thorougher read through and check for any errors.
z3ro: just from a quick read now, the 3d doc looks quite nice. :)
airlied: yes I've done a couple of quick passes over it..
z3ro: have you already started fixing up r300_reg.h?
airlied: I looked at it :) but it'll be tomorrow before I can do anything if I can get away with it tomorrow :)
z3ro: ok. just wanted to check so I didn't duplicate work.
z3ro: I can probably do some checking tonight (early morning) when I get back... but I won't commit until tomorrow once I can look over it again. :P
z3ro: btw I saw some interesting stuff in there about CP soft reset... could be useful for "soft" lockups.
z3ro: I'll have to try it out sometime.
z3ro: hmm and maybe the frag shader docs have some information so I could get fragment.position working correctly.
soc: coincidence that docs for < R600 will be released before R600? :-)
soc: mhhh "The bottom layer of AMD's new OpenGL driver from last Fall is also being evaluated as possibly being open-sourced in the near future." that sounds promising ...
z3ro: cool docs show hyper z.
airlied: soc: before r600 what?
airlied: your comment about coincidence.. it doesn't parse.
soc: coincidence that docs for < R600 will be released before _R600's docs_?
soc: nevermind ..
z3ro: likely amd wrote the r5xx docs first.
soc: i wonder if their announcement about releasing some parts of the opengl code is just like microsoft's pledge "we now like open-source .... really!" or if they really intend to do so ...
airlied: r500 is a lot like r3/400 so the docs were more useful to get first.
z3ro: hmm I wonder how far glisse is on his radeon gallium work. it would be nice to start directly on gallium, but I'm not sure how stable it is right now.
airlied: soc: they ask what docs we want first and we tell them what we are most interested in..
soc: ah ok
airlied: soc: they might not release any code because it might not be useful to us
airlied: just getting a big block of disconnected code that can't be used is rarely greatly useful..
soc: it's probably easier for amd's reviewers, too, to do r500 after r300/400 ...
MaximLevitsky: Thats true
z3ro: airlied: well I hope tcore is still released... from the chat I had with bridgman it looks like it has a lot of cool debugging stuff.
soc: mh ok, i understand ...
airlied: z3ro: oh yeah tcore shuold be useful for debugging stuff..
z3ro: oh, you were talking about the backend of their 3d stack.
z3ro: yeah, I don't think that would be very useful. tcore, on the other hand...
MaximLevitsky: A bit strange that their doc is just ~250 pages
MaximLevitsky: I bet plenty od information is missing
z3ro: looks reasonably complete from my quick read through.
airlied: definitely enough to bring up a 3D driver.
MaximLevitsky: airlied: I agree, the doc is quite good
mikkoc: so now.. how long until we see basic 3d, in radeonhd? :D
GagaPeter: mikkoc: well, didn't you started already writing? ;D
MaximLevitsky: In other news several Microsoft employers were hospitalized with cold burns, in interview they said : "The hell is frozen!"
mikkoc: muahaha, yea i wish i could understand programming and stuff :P
MaximLevitsky: Finally linux will gain proper 3D acceleration
MaximLevitsky: I will try to help too
GagaPeter: yes, and that out-of-the-box
Sir_Lewk_: without proprietary blobs!
GagaPeter: open source is better for linux, at the moment i have a radeon hd 2600xt and no 3d support because radeonhd doesnt support it and fglrx wont run
mikkoc: i wonder how long it will take until we can say: ati > nvidia :D
MaximLevitsky: Exactly, say the nvidia has whole grapical subsystem, including the opengl library, and kernel module that does all the linking between X and direct rendering
MaximLevitsky: But this doesn't give anything to linux
Tigerchen: and thinkpads switched to nvidia :(
Tigerchen: exactly the wrong moment for that
Sir_Lewk: will the 2d acceleration for r600 cards be enough to use opengl drivers with mplayer?
Sir_Lewk: or does that sort of thing require 3d for some reason?
MaximLevitsky: OprnGL = 3D
Sir_Lewk: yeah, that's what I was afraid of
mikkoc: but radeonhd doesn't even have 2d, or am i wrong?
Sir_Lewk: oh well, sdl still works fine (even without 2d) :D
GagaPeter: Well, i got a quad core here and the mesa libgl got 4 frames per second in cube. how can i change it (temponary)
GagaPeter: i know its OT :D
aneas: still no r500 3d commit in git... stop dreaming and start working, guys! ^^
Ryszu: Yeah, because they sit around doing nothing all day
Sir_Lewk: I have a moderate amount of C experience, is there any way that I could help?
Tigerchen: i think they are all at fosdem listening to whatever talks are there
dli: aneas, airlied promised N+1 day for 3d support, because the docs already compiled in codes
GagaPeter: wow, you guys are quick :D
dli: Sir_Lewk, I guess everyone can help, just get the source, test it, try to fix some bug, send in the patch
aneas: they have a docs->C compiler? now thats a great thing! :D
dli: aneas, it means the codes for 3d support have been written, they just need the permission to release them
aneas: dli, i just tried to be funny ^^
mikkoc: dli: you mean we could have 3d support anytime soon?
slackern: hmm there isn't a logo for radeonhd is there?
dli: mikkoc, I believe so
dli: mikkoc, I guess the 3d support goes to the kernel module: http://gitweb.freedesktop.org/?p=mesa/drm.git;a=summary
mikkoc: ah, so it's not the radeonhd driver itself that provides 3d?
dli: mikkoc, I'm not sure, but we will see soon
GagaPeter: i think it would be good i its in kernel module. 3d in framebuffer
aneas: 3d stays in userspace
aneas: radeonhd sets up dri, mesa provides kernel modules to enable dri
mikkoc: feels ignorant
aneas: the "3d support" itself is in mesa/drivers/dri/radeon
aneas: (i hope thats correct)
libvbe: FOSDEM ATI Talks in 55 minutes :)
libvbe: Room is all wired up ;)
Ryszu: Live commentary please :P
michaellarabel: Ryszu: There will be videos too shortly after each talk.
Ryszu: Ah, cool
michaellarabel: I'll be uploading them to RadeonHD.org.
aneas: here is a livestream of a different room, next talk (perl 6) starts in 10 minutes: http://streaming.linux-magazin.de/en/program_fosdem.htm
libvbe: Ryszu: no such thing; sorry, too much running around without it already...
aneas: michaellarabel, would you please announce it in here after youve uploaded the ati talk?
michaellarabel: sure thing, half way through Bridgman talk right now
michaellarabel: Hopefully the audio turns out well, as the door is open and it's noisy at times.
aneas: as long as your not recording it with a webcam ^^
michaellarabel: Recording it at 640 x 480 to compact flash
aneas: thank you for doing that :)
michaellarabel: Also recording notes that will be posted at Phoronix.com shortly.
aneas: hehe, guess someone stepped on the wlan cable in your devroom
michaellarabel: Nope, don't believe so, I am sitting right in front of that.
michaellarabel: Recording done... now to Oggify and reduce its 1.5GB size.
michaellarabel: aneas: If you want the highlights, the text is at http://www.phoronix.com/vr.php?view=11887 and the video is being worked on
slackern: you should have my athlon xp processor it would encode it in a flash :p
aneas: great, thanks :)
michaellarabel: slackern: It's on a C2D, but it's in problems because I had recorded the entire speech as one file and it being over 1.5GB and then the CF being FAT.
slackern: ahh, i was being ironic with my processor too :)
michaellarabel: argh, the upload keeps stalling on this connection
aneas: wanna torrent it?
michaellarabel: I might depending upon if these issues continue
aneas: do you have an ETA?
aneas: when its uploaded
michaellarabel: 1.3 hours for the raw video
aneas: textured video: 'real soon' as in days, weeks or months?
michaellarabel: Internet still causing issues... Though if interested, text from Egbert's talk is @ http://www.phoronix.com/scan.php?page=news_item&px=NjM1MA
agd5f: aneas: I already have textured video working on r100-r400, just haven't had a chance to finish cleaning it up and pushing it
aneas: wow, nice! is r500 very different from it?
agd5f: aneas: no. In fact I had it working on r500 without lockups, but, texuted aren't working yet due to the fragment program setup
agd5f: textures even
aneas: that sounds very promising :)
udovdh: back from FOSDEM...
udovdh: nice, infirmative talks
udovdh: the phoronix url above gives some good summary
Tigerchen: i hope with the new stuff also comes the newly introduced bug
Tigerchen: *fix to the
udovdh: the 640x480 issue?
udovdh: did the devs have a log to look over?
udovdh: `X -logverbose 7` is preferred
Tigerchen: i think i can create one with -7
udovdh: that might help them
udovdh: just do init 3
udovdh: then clean out /var/log/Xorg*
udovdh: and do
udovdh: X -logverbose 7
udovdh: switch back to textconsole
Alpo: What is the status of 3D acceleration?
Tigerchen: know know
udovdh: and ctrl-C X
udovdh: will come later Alpo
udovdh: see the phoronix url
udovdh: I left after Egberts talk
udovdh: both gave a big amount of info on the technical side but also the project, the how/when/why etc
udovdh: room could have had some more ventilation...
udovdh: but no other complaints!
Tigerchen: udovdh: http://rafb.net/p/dRX1tU49.html this should be a correct run
udovdh: Tigerchen, I am not a dev
udovdh: but I can have a look
Tigerchen: yeah, thanks
udovdh: (EE) RADEONHD(0): RandR: No valid modes. Disabled.
Tigerchen: i saw that
udovdh: the EE makes me wonder
Tigerchen: me too
udovdh: the xrandr disabled is 'normal'
udovdh: but looks like
Tigerchen: i think that has something to do with the patch
udovdh: (--) RandR disabled
Tigerchen: PATCH: Report PANEL as disconnected when no modes are available
Tigerchen: i think this patch broke it but am not sure
udovdh: doubt it, since stuff is modularized
udovdh: but I am not an expert
Tigerchen: do you have experience with git? perhaps i could find out which commit exactly broke it
Tigerchen: how can i switch to a previos version of the tree?
yangman: git revert
yangman: where commit# is the commit you want to undo, not the commit you want to go back to
yangman: problems with laptop panel?
Tigerchen: will try out later
Tigerchen: ok it was definiively commit 3d29d4ce1183a53810ff8b5c2970b03f9cc3b097 which broke my sys
dmb: is there a place to download any of the fosdom videos?
Tigerchen: radeonhd.org could have them
Arrow: Hi all
Arrow: anyone home?
Arrow: could someone direct me to the GPU/crossfire detect code?
Arrow: trying to start at the begiunning here
Arrow: acually crossfire is 'not' optimal
Arrow: does it pay to use the DVIs to pass data one GPU to the other?
Arrow: might be kinda cludgy
Arrow: I'm a GPGPU guy btw...so straight framerate to a display is not my issue
TGEN: Arrow: 'modern' CrossFire uses internal bridges like nVidia's SLI, but they can also communicate purely via PCIe ports
TGEN: for GPGPU computing, CrossFire or SLI aren't necessary or even particularly useful\
Arrow: I gathered that :)
TGEN: afaik the internal/external bridges are only there for the compositing engine(s) to communicate, not for high-bandwidth data exchange
Arrow: yes...I see that now
TGEN: the 'GPU detect code' lives in src/rhd_driver.c btw
TGEN: but that's just enumerating and matching PCI IDs :)
Arrow: needs to detect multi GPUs and assign PCIe lanes etc. the 790FX board I bought can go 16-0-16-0 or 8-8-8-8 as far as I can tell
TGEN: 16-1-16-1 I reckon
Arrow: builds a table and feeds it back to his loadbalancer
TGEN: Arrow: just query PCI config space, and compare vendor/device/subvendor/subdevice IDs
Arrow: nods right
Arrow: I have 4 PCIe slots so we'll see
Arrow: hopes to have the 'words fastest PC' when he's done ;>
TGEN: the world's fastest PC is probably one thrown down a gravity well
Arrow: there should be a contest.."most performance from a single wall socket (15 amp service)"
TGEN: you forgot to specify voltage, AC or DC, and in case of the former, phases+frequency
Arrow: right 110-115 60hz standard us power ;>
Arrow: is buying the 1600 w PSU
TGEN: if it were 230V European standard, you could get more out of it
Arrow: I know :)
Arrow: you work with what you got
TGEN: -48V DC
Arrow: I own one 3870x2 it will use about 320w
Arrow: three more puts the total system close to that 1600
Arrow: not when your doing Wavlet-Galerkin physics problems
TGEN: wouldn't you be better off having a farm of FPGAs?
Arrow: that's 'meshless' multi-resolution analysis
Arrow: not at 460 us for the 3870x2- 2 GPU card ...it damn cheep for that much power
TGEN: sure, but you're limited to single-precision fp arithmetic afaik
Arrow: no a 3870 GPU can be tweeked to do about 200Gflops DP
Arrow: with 8 in the system that's 1.6Tflops for under 3 grand
TGEN: what do you mean 'no'
Arrow: no you are not limited...the internal fetching is 128 bits wide...a "shader" can be made to do DP at about 40%
TGEN: the FPUs are still single-precision
TGEN: so it does about 1 TFLOP/s max single precision per card
Arrow: times 4
Arrow: 4 Tflop in one PC...and I'll digup the paper on how to do longer precision on a 32bit FP...:)
TGEN: oh, emulating double-precision on single-precision hardware isn't that hard
TGEN: but it does incur a significant performance penalty
Arrow: acually the guys at Univ. of Kentucky like to 'optimise' FP precision to the problem at hand
Arrow: yes it runs at 40%
TGEN: for physical simulations, double precision is the norm rather than the exception
TGEN: but it's not like FPUs these days are one-clock wonders anyways
Arrow: true but for really 'smooth' regions lesser works too
Arrow: the codes all do alot of mult-adds the GPU does one a cycle
Arrow: this is a good overview of the issue
Arrow: well I need to run AMD is supposed to put out GPGPU stuff for Linux "soon " :)
Arrow: I'm sure I'll be helping with the 'graphics' driver in any event :)
Arrow: TGEN see you soon :0
TGEN: ... I'm now afraid to open the door when someone rings ;)
dvandyk: TGEN: was arrow refering to mixed precission iterative refinement, or is there something else that allows to enhance precision on 'single' operations?
TGEN: dvandyk: I assume the first
adam2321: My samsung 223BW (22") monitor comes up blank with radeonhd driver. Using Option "HPD" "swap" gets gdm up but the screen shows banding (horizontal lines across the screen).
adam2321: if I login, the background image is rendered properly, but the taskbar is corrupted (banding). When I hit the logout button, the screen is again corrupted, but if I mouse over "logout" or "shutdown", the icon appears clear.
adam2321: Anyone know what might be happening?
nooop: people should be sleeping now.
nooop: and it's sunday~
adam2321: I'm on the other side of the world :)
adam2321: I'll check back later