Home Phoronix Phoronix Forums X.Org Videos From FOSDEM 2008

Radeon IRC Logs For 2009-11-07

Search This Log:

cxo: whats 256mb in hex?
cxo: 0 x 1000 0000
spstarr: ?
soreau: dileX_: seems enemy territory does not crash X when in kde, only while in my standalone X session
MrSunshine: ello :)
dmb: ): olle
MrSunshine: got a slight problem ... upgraded my linux distro and it upgraded my ati video driver ... and now my screen is all messed up ...
MrSunshine: tried downgrading but then X wont even start :(
MrSunshine: tried latest GIT same problem there
MrSunshine: its on an ATI RS482 [Radeon Xpress 200M]
MrSunshine: i get 4 screens looking the same on the screen ... and colors etc is all messed up
MrSunshine: im atm at git20091014
soreau: MrSunshine: Can you pastebin the X log from the failed session?
MrSunshine: from the ones that doesnt start at all ?
MrSunshine: dont know how to get it to a paste service from non grapical env :/
MrSunshine: module: ABI major version (5) doesn't match the server's version (6)
soreau: MrSunshine: cat /var/log/Xorg.0.log | curl -F 'sprunge=<-' http://sprunge.us
MostAwesomeDude: The ABI error means that you need to rebuild your driver against the new server. Which distro is this?
MrSunshine: got X working by downgrading the x components
MrSunshine: MostAwesomeDude: archlinux
MrSunshine: did a full system upgrade today .. and blam ... graphics got destroyed
MrSunshine: downgraded all X related stuff to the stuff i had befor enow .. and it works
MrSunshine: well all except mouse and stuff :P
soreau: So rebuild your input drivers too (?)
MostAwesomeDude: Uf. I have no idea how to get Arch back on track when it breaks.
soreau: Shouldn't be much different from any other distro
soreau: Except they might not have tools to cater latest ati drivers :)
MrSunshine: forgot to downgrade synaptics driver ))
MrSunshine: =)
MrSunshine: thanks for trying to help tho :)
MrSunshine: gonna quit here and fire up xchat insted :)
MrSunshine: =)
spreeuw: is crossfire supported?
MostAwesomeDude: No.
glisse: agd5f: issue is on r700 hw
glisse: maybe somethings changed their
MostAwesomeDude: Alright, I've deduced that something's wrong with rawhide's kernel and it doesn't appear to matter whether I'm on r400 or r600.
MostAwesomeDude: Now I have to bisect the damn thing. :C
soreau: MostAwesomeDude: What seems to be the trouble?
glisse: MostAwesomeDude: what's your issue ?
MostAwesomeDude: glisse: Random hardlocks after 5-30 minutes of usage.
MostAwesomeDude: HW cursor dies too.
glisse: MostAwesomeDude: with gallium ?
glisse: or normal usage ?
MostAwesomeDude: glisse: Normal usage, no apps running.
MostAwesomeDude: It'll freeze while I'm in vim, or in Fx, or just letting it sit there.
glisse: i saw freeze on r600
MostAwesomeDude: And it happens with both r400 and IGP r600, so I think it's a bug somewhere else.
soreau: MostAwesomeDude: What do you mean by hw cursor dying?
MostAwesomeDude: soreau: If a lock isn't completely and totally killing the kernel, then HW cursors usually still work.
glisse: sadly the lockup thing is widely spread accross our supported hw :(
glisse: got to go
MostAwesomeDude: glisse: This happens when I disable the IGP; it's probably not r600 bug.
glisse: MostAwesomeDude: maybe open a bug
MostAwesomeDude: glisse: Yeah, gonna gather more stuff first, but I need this fixed.
glisse: yeah lockup are annoying to say the least
glisse: bbl
MostAwesomeDude: Later.
airlied: MostAwesomeDude: what kernel?
soreau: MostAwesomeDude: FWIW if I had the hw to reproduce the issue, I would bisect
airlied: no if its Fedora kernel I'd blame me.
MostAwesomeDude: airlied: Not (f11) but (f12) and newer.
MostAwesomeDude: All f12 kernels I have do it, but none of the f11.
airlied: MostAwesomeDude: just get the latest
airlied: -56 is like ancient
airlied: i.e what is that last one?
MostAwesomeDude: airlied: (f12) also does it.
yangman: MostAwesomeDude: I see it on r500 too. 2.6.31 + drm-next
MostAwesomeDude: That's the newest one I've got.
airlied: MostAwesomeDude: thats like ages old
MostAwesomeDude: I'm not totally sure it's video-card-related.
airlied: yum upgade your ass already
MostAwesomeDude: chroots and updates
airlied: I'd blame me.
airlied: 00:19 < MostAwesomeDude> airlied: Not (f11) but (f12)
airlied: and newer.
airlied: oops broken cut-n-paste
airlied: http://kojipkgs.fedoraproject.org/packages/kernel/
MostAwesomeDude: Oh God that's a lot of updates.
MostAwesomeDude: ...and a new glibc. Goodie. Well, I'll report back in once I get this going.
airlied: just get kernel/-ati to fix r600
MostAwesomeDude: r400 is also hardlocking when IGP is disabled; maybe related?
airlied: doesn't ring any bells
airlied: but lets pretend it fixes it
MostAwesomeDude: Whoo, pretend.
MostAwesomeDude: If it doesn't, then I can pop out the r400 for now.
MostAwesomeDude: Alright, switched and so far no lockups.
MostAwesomeDude: I'll give it a bit; it usually happens about 10 min in.
soreau: MostAwesomeDude: I always assumed you kept up with all latest updates
EruditeHermit: is the gallium build broken?
dileX: w/ latest mesa GIT?
EruditeHermit: yes
EruditeHermit: No rule to make target `/usr/lib/gcc/i486-linux-gnu/4.3.3/include/stddef.h', needed by `glew.o'. Stop.
dileX: here it is /usr/lib/gcc/i486-linux-gnu/4.3/include/stddef.h
EruditeHermit: does it fail for you too?
dileX: building...
dileX: EruditeHermit: build, install and setup OK
EruditeHermit: hmm
EruditeHermit: weird
dileX: does stddef.h file exist?
EruditeHermit: where?
EruditeHermit: 4.3.3 doesn't exist
EruditeHermit: its 4.3.4
dileX: symlink as temp. fix might help
EruditeHermit: symlink as temp?
dileX: just an idea for a dirty fix: let missing stddef.h (4.3.3) point to the one from 4.3.4
EruditeHermit: ah
EruditeHermit: well
dileX: EruditeHermit: did you build from a fresh git checkout or in an already-built-from?
EruditeHermit: shouldn't ./configure look for it and find it in the right place?
dileX: yupp
EruditeHermit: it was an already built from
dileX: make distclean?
EruditeHermit: but I did do a ./configure again
EruditeHermit: didn't distclean
EruditeHermit: will try that
dileX: no, thats no good idea
dileX: (mean "./configure again")
dileX: is that new in glxgears? shadow behind the paddle-board
orion: i have been trying for about a week to get my 4850 running with mandriva 2010 and getting nowhere it just hangs at a white screen. today though it allowed me at the white screen to ctr+alt+f2 and i got a monitor message saying the horizontal was out of range. Is this problem because i am using an old crt monitor with a DVI output through a converter dongle for the older monitor?
dileX: orion: hi. the output of dmesg, Xorg.log and the paste of an existing xorg.conf could be helpful.
orion: dileX: do i take them to pastebin?
zhasha: orion: sounds likely
zhasha: and yes, just pastebin it
orion: ok the two xorg.conf files are on pastebin under the name colin. one is the working version with no 3d and the other is the fglrx which white screens
orion: zhasha: the monitor said that horizontal was 74.8 but supposedly this monitors max is 70
dileX: orion: paste the URL here, pls
orion: http://pastebin.org/51596
dileX: there are dozens of pastebin services on the World-Wild-Web
orion: :(
zhasha: fglrx, should go in #ati
orion: there are two different xorg.confs pasted
dileX: orion: fglrx and OSS radeon shall not co-exist
dileX: orion: more infos? see build-wiki in topic
zhasha: your screen section doesn't seem to list the resolution your monitor allegedly uses?
orion: they dont they are two different xorg files that i have tried tonight one has the radeon and the other the fglrx
dileX: orion: you can not switch via xorg.conf or sth else. fglrx replaces e.g. libGL.
orion: yeha Ok thanks you obviously arent fololowing me
dileX: to test OSS radeon you have to get rid of fglrx - completely
dileX: I think its time to publish this pricelist here :-)
dileX: any objections :-)?
eosie: the good thing about such people is they leave early and hopefully never return until having a brain transplant
dileX: eosie: pricelist is better option :-)
dileX: (but not realisable)
twnqx: interesting, this java app has screen corruptions i never observed before updating to the latest driver
rozzo: hello
rozzo: i have a big problem
rozzo: biggest
rozzo: when
rozzo: i use radeon opensource have a big rpoblem with it
rozzo: random get 100% of cpu
rozzo: and all xorg is locked
rozzo: ;_;
twnqx: upgrade to latest everythiing
rozzo: twnqx, talk with me?
twnqx: yes
twnqx: i had that until two days ago, too
twnqx: with the very latest kernel and other parts it works now
rozzo: mm i have fedora 11
rozzo: with xorg 1.7
rozzo: and kernel
rozzo: if i have lucky the problem not show
rozzo: for 1 week
rozzo: but after rea appear
rozzo: ;_;
rozzo: are you shure?
twnqx: what chip do you have?
rozzo: ati radeon x1270
rozzo: of family x1200
rozzo: R690
rozzo: chipset
twnqx: yeah, that was some r600 fixes
twnqx: two days old.
rozzo: ?
rozzo: have fixed in the chipset r600?
twnqx: the fixes for my freezes. were done two days ago.
rozzo: of radeon driver cvs?
twnqx: git.
rozzo: ok
twnqx: not driver, kernel
rozzo: io try the git version
rozzo: of radeon
twnqx: not. the. driver.
twnqx: the kernel.
rozzo: the latest kernel?
twnqx: latest drm-next kernel
rozzo: is a car in the ass for fedora
rozzo: ;_;
twnqx: move to gentoo :P
rozzo: what distribution you use?
twnqx: gentoo
rozzo: not have time
rozzo: ok
rozzo: thanks for support i suspect
rozzo: the drm driver
rozzo: now i go out
rozzo: sorry
rozzo: bye
rozzo: and thanks for all fish!
rozzo: :D
MegaMan: anyone know how to get 4 monitors working on an ati Radeon X1650 XT rev 9a in ubuntu 9.10?
chithead: MegaMan: you have only 2 crtcs per card and crossposting questions is usually being frowned upon
MegaMan: 2 crtcs per card?
MegaMan: I was in ati and they told me to go to radeon
adamk_: He's referring to asking the same question in #radeonhd
twnqx: so what about that card with 6 mini-displayport connectors
MegaMan: just trying to ask as many peeps as possible
adamk_: chithead: On #ati I basically told him the same thing about supporting two output devices at once, but he said it works under Windows, so I'm not sure what's happening.
MegaMan: http://paste.ubuntu.com/312484
MegaMan: right now we can get one dvi output working at a time
MegaMan: but not both
MegaMan: so one dvi output has a splitter cable that connects to two monitors
twnqx: MegaMan: abusing the dual link?
MegaMan: http://www.amazon.com/Radeon-X1650XT-512MB-Monitor-DMS-59/dp/B000Q8X8NC
MegaMan: twnqx, abusing? not up on the lingo
twnqx: well, dual link is supposed to connect to one monitor only with two full links - if you connect to two screens instead it's not "originally intended use"
MegaMan: I've never seen one mintor with two of the same input before
twnqx: mh?
twnqx: a "dual link" dvi connector has two full dvi links in a single connector
MegaMan: so what does "link" mean?
twnqx: http://en.wikipedia.org/wiki/Digital_Visual_Interface
MegaMan: so I'm down to two choices? switch to nvidea or switch back to windows
twnqx: just get two cards?
MegaMan: whatever works
twnqx: i don't think nvidia will give you 4 display on one card, even with the cables
twnqx: chithead: though it would be possible to implement four displays this way :P
MegaMan: the catalyst software in windows shows all four displays and allowed you to movie them around to match how they are physically mounted. It even allowed different resolutions for all 4
honk: nvidia cant drive 4 monitors with 1 card either o.O
honk: then again.. I havent seen ati do it yet either ;)
MegaMan: i dont know about what they are capable yet, I just know my experience with getting their cards working on linux have been way better heh
honk: oh I see
honk: that card is actually a dual gpu card ;p
honk: makes a lot more sense =)
twnqx: i'd bet with the breakout cable to could set it up in a way to use the two links to run two panels :P
honk: well, it's not a real dvi to 2dvi adapter.. ;P
MegaMan: it's just weird how we can only get two monitors working at one time
honk: they are running 2 signals to the port at once in the first place
honk: MegaMan: it's not weird at all =)
honk: there is no support for crossfire in the opensource drivers
MegaMan: well it's two out of four at one time. so one half or the other not both
honk: and I dont know if you can actually use the two gpus seperately w/o CF
MegaMan: so we are just chaning the pci bus in the xorg file and that flips it
honk: there is no support for crossfire in the opensource drivers
honk: you'd have to have at least 2 seperate entries in there for each "card"
honk: (if it works at all)
MegaMan: so crossfire is using both?
honk: http://en.wikipedia.org/wiki/ATI_CrossFire :}
MegaMan: I'm wondering if we can fool xog into thinking they are two separate cards
honk: you'd have to have at least 2 seperate entries in there for each "card"
honk: (if it works at all)
honk: what does lspci show?
MegaMan: http://paste.ubuntu.com/312484
honk: that looks promising ;)
honk: you'd have to have at least 2 seperate entries in there for each "card"
honk: are there or are there not 4 entires in your xorg.conf? ;P
honk: *entries
honk: and are your drivers the latest git version? =)
MegaMan: http://paste.ubuntu.com/312512 is the xorg
MegaMan: http://paste.ubuntu.com/312513 is the lsmod
honk: noone asked for lsmod ;P
honk: but yeah, that's similar to what you'd need =)
MegaMan: lol
honk: anyway.. 1) make sure all your packages (mesa, drm, kernel, driver) are up to date (i.e. latest git) 2) hope that it can work at all ;)
honk: if everything else fails - downgrade your kernel and xorg and use fglrx *lol*
honk: or just spend 30 bucks and buy 2 low power gfx cards ;)
MegaMan: on a differnt box we are trying the downgrade and fglrx heh not working either
spstarr: notes VBOs is still broken in r6xx
hnsr: hmm I thought it was working before
MostAwesomeDude: airlied: Latest Rawhide hasn't frozen in 8h; good work. Thanks. :3
spreeuw: hnsr: what can I do to test them?
spreeuw: vbos
MegaMan: honk, we were able to get all four monitors working but now the other half is a mirror of the first
MegaMan: this is with the fglrx
unimatrix: hi i've installed radeon 1.3.0 and got 3D support on my r700 card, but opengl stuff keeps flickering, is this a known problem?
MostAwesomeDude: I'd normally say "turn off compiz," but I didn't think the r600 driver did DRI1.
MostAwesomeDude: ...Are you using compiz? Does turning it off help?
unimatrix: well you're right
unimatrix: it was compiz
adamk_: MostAwesomeDude, r600 definitely does DRI1
MostAwesomeDude: adamk_: Ah, well, there we go.
unimatrix: i can't get 3D on radeonhd though, only on radeon... thought i'd mention that
MostAwesomeDude: Not our problem. :3
spreeuw: MegaMan: other channel
MegaMan: spreeuw, ?
MostAwesomeDude: MegaMan: #ati for fglrx problems.
unimatrix: call me ignorant but i thought radeon and radeonhd were more or less the same driver?
MostAwesomeDude: These days, yes.
biotube: but they're separate codebases
MostAwesomeDude: And separate sets of bugs.
MostAwesomeDude: And only one is actually worked on.
MostAwesomeDude: osiris_: Working on getting your code to be more modern. Already found a bug in SW TCL though.
MostAwesomeDude: gears is being funky because of the mixed elements/arrays render setup. Detangling them makes the corruption go away.
osiris_: MostAwesomeDude: good, I got sidetracked by blit support for r300c
MostAwesomeDude: osiris_: No worries.
MostAwesomeDude: Are you doing 2D engine or 3D engine?
twnqx: i don't have compiz, but buttons in my java app have "noise" since i moved to KMS :X
osiris_: 3d
MostAwesomeDude: Ooh, unfun. Well, good luck.
hagabaka: I'm also getting "noise", in certain characters, and in various areas of the screen
osiris_: glisse or nha: why there are two paths for buffer clears (one for kms and one for ums)?
MostAwesomeDude: osiris_: Comments on http://cgit.freedesktop.org/~csimpson/mesa/log/?h=r300g-vbo ? I'm going to push this to master since it works well enough and appears to not break things.
MostAwesomeDude: (Well, aside from SW TCL, but that's been broken for a bit.)
MostAwesomeDude: I still need to push down some of the code, but that can be done on master.
osiris_: is taking a look at the code
osiris_: MostAwesomeDude: "r300g: Don't assert on oversized VBOs, just return FALSE." commit is wrong
osiris_: MostAwesomeDude: the condition should be if (count > 65535) not 65536
osiris_: MostAwesomeDude: the first assertion from setup_index_buffer can be removed now
osiris_: MostAwesomeDude: format is supported check should be moved to setup_vertex_attributes
osiris_: MostAwesomeDude: the setup_vertex_array and finish_vertex_arrays_setup can be removed
leokolln: g
osiris_: MostAwesomeDude: in r300_setup_vertex_buffers remove the vbuf_count alias for r300->aos_count or rename it to aos_count to avoid confusion. also I think this function should add a buffer only once (currently if two attributes will be in the same buf, we will call add_buffer twice for it)
MostAwesomeDude: osiris_: Agreed on most of those.
MegaMan: for question on the ati 9.3 drives is that here or ati channel?
MostAwesomeDude: #ati
MostAwesomeDude: This is only for the open-source drivers.
elenril: hi
elenril: should i use radeon or radeonhd driver for Radeon HD 4650?
elenril: and can i get 3d with any of the open drivers?
EruditeHermit: elenril, you will need latest kernel
EruditeHermit: you can use either, but radeon is probably better maintained right now
happycube: you'll get faster 2d but unfinished 3d - it's a fair tradeoff in many cases
elenril: latest means 2.6.31 or rc?
EruditeHermit: for 3D I think you need rc
elenril: ok, thanks for information
elenril: goes to compile himself a new kernel
LordVan: does radeon work with HD4750 yet? ;)
chainsawbike: uses drm-next with compiz - no major problems
chainsawbike: LordVan, well my hd4870 works so id assume so
happycube: 4750? it got renamed 4770 with it beat it's performance goals... and the 5750 isn't
happycube: ^ supported
happycube: but yeah, if you have a 4750 it'd work ;)
LordVan: chainsawbike, i menat 5750
LordVan: ;)
LordVan: typos suck
stikonas: no, it is still WIP
stikonas: somebody has to put http://www.x.org/wiki/RadeonFeature to #radeon topic
agd5f: osiris_: ums did clears in the drm
agd5f: Megaman: the hw only has two display controllers. it can only dirve two independant displays no matter what hw or driver
agd5f: only the new evergreen chips can drive more than two
osiris_: agd5f: I've replaced the clear code from r300 with meta_mesa_Clear and it seems working fine (no perf regression at least on kms - checked with demos/clearspd and glxgears)
Zajec: agd5f: evergreen? does this seriee has 3crtcs?
agd5f: Zajec: 6
Zajec: *have
Zajec: wooow
Zajec: agd5f: of course also 6pll, right?
agd5f: no 2 plls.
agd5f: and 6 dig phys
Zajec: agd5f: er, is that enought for 6 indepentend displays?
agd5f: yeah
Zajec: ok, i'll read about "dig phy"
agd5f: 6 DP
agd5f: 6 DP displays or 2 legacy and 4 DP
Zajec: ah, ok
Zajec: nice, really :)
MostAwesomeDude: Or 1 legacy and 5 DP. :3
agd5f: right :)
Zajec: MostAwesomeDude: ;)
MostAwesomeDude: Too bad that it doesn't have enough CRTCs for 8 displays though.
MostAwesomeDude: (As if any of us could afford that!)
Zajec: MostAwesomeDude: do you have idea of using 8ctcs? :P
Zajec: MostAwesomeDude: or even 6?
agd5f: the launch demo had 24
Zajec: ;)
MostAwesomeDude: agd5f: That was with DP chains, right? Or was that a workstation card?
agd5f: MostAwesomeDude: 4 cards. 6 each
agd5f: DP doesn't support multi-point yet
agd5f: just repeater
agd5f: s
airlied: osiris_: ums clears use a special kernel clear path, kms clears just use meta ops
airlied: ums clears are probably fast since hyperz
airlied: ah alex answered already ;-)
osiris_: airlied: hyperz? I thought it wasn't implemented yet
airlied: osiris_: kms should have been using metaops before now
airlied: we use it in ums
airlied: for z clears
spstarr: agd5f: ping
spstarr: agd5f: it worked as of
spstarr: the changes made to r6xx KMS it seems
agd5f: spstarr: what changes?
dmb: spstarr, you said radeon driver doesn't work with more then 256mb of ram?
osiris_: airlied: and why aren't we using the hyperz clear for kms?
dmb: spstarr, i thought that was AGP
spstarr: agd5f: looking at git log
agd5f: osiris_: multiple Z buffers
agd5f: dmb: we currently clip usable vram to the max bar size (generally 256 MB)
agd5f: dmb: requires changes to ttm to properly support it since the CPU can't map it
osiris_: agd5f: are you saying that for ums there's only one Z buffer for all apps?
agd5f: osiris_: yes
dmb: oh
dmb: so thats only KMS?
agd5f: osiris_: only one front and back as well
osiris_: ah, dri1
agd5f: dmb: all drivers right now
dmb: oh
airlied: osiris_: for kms we'd have to assign hyperz to one app I think
spstarr: http://git.kernel.org/?p=linux/kernel/git/airlied/drm-2.6.git;a=shortlog;h=refs/heads/drm-next
spstarr: agd5f: I had drm-next prior to 3 days ago
airlied: spstarr: try reverting the couple of r600 commits for w500
agd5f: spstarr: any chance you could bisect it?
spstarr: so the 10 days ago
airlied: the VRAM at 0 and the VGA one
spstarr: airlied: those are the ones..
airlied: osiris_: we could also use the hyper-z engine to clear the colorbuffers
airlied: not sure if we know exactly how that can work
airlied: but it can speed things up
spstarr: airlied: when your w500 failed, did it hang on bootup, ie, you could type into console, but kernel wouldnt continue booting up system?
osiris_: airlied: hmm, isn't hyperz writing the Z data in some compressed format?
airlied: spstarr: no it would hang hard nothing more happened
spstarr: hmm
airlied: spstarr: but later when acpi video loaded
airlied: osiris_: it has a cache
airlied: from memory
airlied: but its clear engine is very optimised
airlied: even without the caching
osiris_: airlied: so only cached values are compressed?
airlied: osiris_: I think so, you have the HiZ RAM
agd5f: osiris_: yeah. it's stored compressed in a cache
MostAwesomeDude: The problem with using the HiZ stuff is the context switch.
spstarr: airlied: reverting patch
MostAwesomeDude: You don't want to have to constantly flush the HiZ cache; it sucks so bad.
MostAwesomeDude: If you can guarantee it's only getting used by one app, then you could avoid flushing it.
spstarr: airlied: the DV1VGA/DV2VGA only appears to impact an rv515 GPU?
agd5f: spstarr: r5xx and newer
spstarr: ooh
airlied: MostAwesomeDude: exactly why you need to pick an app to get it
airlied: which should be gears first then any fullscreen app ;-)
spstarr: ok, reverted those two .. building kernel
MostAwesomeDude: airlied: Yeah, b
MostAwesomeDude: Argfl.
MostAwesomeDude: Yeah, but I'm not sure how to detect that right, and I'm also not sure how much userspace co-op is required to do that.
agd5f: MostAwesomeDude: first one available
agd5f: although that would mean compiz would tend to end up with it
MostAwesomeDude: Or wayland, etc.
happycube: is hiz locked to one memory area?
otaylor: well, you could exclude apps that don't use a visual with a depth buffer - compositors don't usually make much use of it,and could be taught not to ask for one
agd5f: happycube: only one cache
happycube: oooo... makes sense
otaylor: (If people haven't seen developer.amd.com/media/gpu_assets/Depth_in-depth.pdf, it's a pretty good overview of the area)
MostAwesomeDude: otaylor: Agreed. Compositors shouldn't ask for them in general, although e.g. compiz would have to be careful about rendering order in that case.
MostAwesomeDude: For cube, atlantis, etc.
otaylor: MostAwesomeDude: I don't think compiz uses depth buffers, except maybe for the cube
MostAwesomeDude: otaylor: Cube is paramount though. We can't ask them to break cube.
spstarr: airlied: assuming it's those two changes only
otaylor: Well, it's easy to do a cube without depth buffers, since faces are either culled or not
MostAwesomeDude: Additionally it's easy to do the faces in the "right" order.
MostAwesomeDude: If you're doing a translucent cube.
otaylor: The other thing you could try to try to tie it to unredirection of full-screen apps - which Compiz doesn't do by default and Mutter doesn't do at all, currently, but is pretty important for squeezing the ends of frame-rate out of games
MostAwesomeDude: This is true.
MostAwesomeDude: It's just that in the not-so-distant future, the first app will be Xorg, not compiz. :3
happycube: figured r600 would improve that
happycube: flipped through the pdf
airlied: spstarr: no they are just my guesses
airlied: if they don't work you'll have to bisect properly
otaylor: MostAwesomeDude: if you are doing translucent, then depth buffers don't you much good to start with
MostAwesomeDude: otaylor: This is true.
MostAwesomeDude: Hm. Lemme take this to the compiz guys.
spstarr: prays
spstarr: unless its the initial KMS mode validation stuff
spstarr: ok brb testing kernel
osiris_: I've pushed my accelerated blit branch to private repo. anyone care to take a look? (it's still WIP, but should show the direction I'm heading)
spstarr: airlied / agd5f: nope
spstarr: it hangs with
spstarr: firmware: requesting radeon/RV635_pfp.bin
spstarr: it then sits there. i can ctrl+alt+del though and machine will reboot ok
spstarr: i have radeon w/o KMS on right now
spstarr: my ramdisk has all the firmware bits checking it..
spstarr: ./lib/firmware/2.6.32-rc6-custom/radeon/RV635_me.bin
spstarr: ./lib/firmware/2.6.32-rc6-custom/radeon/RV730_pfp.bin
spstarr: it does not load RV 635_me.bin
spstarr: what is pfp ? and me?
spstarr: OpenGL renderer string: Mesa DRI R600 (RV635 9591) 20090101 TCL
spstarr: this is a r6xx (RV635)
MostAwesomeDude: osiris_: Looks pretty reasonable. You might want to consider using texrect and non-normalized coords. It'll be a bit more math but you'll be able to handle NPOT.
airlied: spstarr: what userspace?
airlied: F12?
spstarr: im on debian at the moment, sid
spstarr: kernel.org git master + drm-next
spstarr: so im looking at the git commits
osiris_: MostAwesomeDude: hw needs normalized texcoords (even for texrect)
spstarr: w/o KMS the firmware loads properly
airlied: sounds like some thing gone wrong in your initrd or initramfs
spstarr: im looking into that
MostAwesomeDude: osiris_: There's a flag to use non-normalized for r500 FP IIRC.
spstarr: but drm.ko, ttm.ko, radeon.ko, firmware bits..
airlied: try bootign with nomodeset, then rmmod radeon modprobe radeon modeset=1
airlied: boot without X starting though
spstarr: drm_kms_helper.ko
osiris_: MostAwesomeDude: oh, didn't know about it. but r300 certainly doesn't have it
spstarr: airlied: sec doing so now..
spstarr: i'll boot into single user mode
MostAwesomeDude: osiris_: Oh, good point.
osiris_: MostAwesomeDude: anyway, what should be the pos coords and tex coords to make it work properly?
osiris_: 0.0, 0.0, 0.0, 0.0 is left upper corner when VTE is disabled, right?
MostAwesomeDude: I think so? It's been a while.
MostAwesomeDude: Check out the Xv code in the DDX; I *know* it works.
soreau: Does anyone know how this can happen? I am using compiz in kde, need to restart it after some glitch so I go to tty and do DISPLAY=:0 compiz-manager & then to my surprise, I get back to X and low and behold, white screen due to swrast
soreau: In other words, it's working then somehow falls back to swrast in 'real time'. An X restart fixes it
osiris_: MostAwesomeDude: ok, thanks
MostAwesomeDude: osiris_: Sure. IIRC it's topleft, topright, bottomright, bottomleft.
spstarr: hmm
spstarr: looking into ramdisk problem
spstarr: I thought i had all the modules
osiris_: agd5f: how are rectangle textures handled by r300 hw? is it enough that tex pitch/bpp is POT?
spstarr: its ok i guess to have both as modules
osiris_: agd5f: i.e. width = 96, cpp = 4 and pitch = 512
airlied: osiris_: PITCH_EN is the main diff
airlied: osiris_: pitch dosen't have to be POT
airlied: its rectangular
spstarr: Starting with kernel 2.6.32, the radeon in-kernel driver will stall the boot process while waiting for the microcode to be supplied. The 2.6.31 kernel did not do this (was it in-kernel before?). I'm using kernel modesetting, and have not tried the non-kms case but I suspect it will stall as well (but maybe later in the boot process). Entries in dmesg look like:
spstarr: ew
spstarr: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=550392
spstarr: i dont get that far though
soreau: spstarr: Where does it stall, starting hal?
spstarr: soreau: during firmware load
soreau: doesn't know when that happens
spstarr: it is an option to build firmware into kernel
osiris_: airlied: hmm, so I just have to set width to next POT, and pitch to width * cpp?
spstarr: tries
airlied: osiris_: no POT at all for rectangular
spstarr: if this is the case, then debian broke something
airlied: width is actual width, pitch is aligned to 32
airlied: or 256
airlied: forgets if 256 is necessary probably only for tiling fbos
airlied: spstarr: we used to have a bug in Fedora where we'd load radeon befre hotplug was running
airlied: but I doubt they have that
MostAwesomeDude: Wasn't RS690 also funky somehow?
airlied: MostAwesomeDude: I think it needs 64 byte pitch
airlied: cacheline alignment stuff
spstarr: airlied: asking...
airlied: its all in classic driver or exa composite hooks
airlied: osiris_: width is ignored vwhen pitch is enabled i think
Antiemes: Hi!
Antiemes: I have a problem with the Radeon driver.
spstarr: airlied: it does work for intel KMS though (they dont have firmware)
biotube: Antiemes: details
Antiemes: With some apps (i think, that with Qt3 apps, but I have tested with just 2 of them), it displays random things in the program windows.
Antiemes: http://dcs.vein.hu/~vakulya/qsstv.png
Antiemes: Like this.
biotube: what version of the driver are you using?
Antiemes: I've also tried it with VESA driver, it is OK.
Antiemes: compiled for, module version = 6.12.99
Antiemes: And Mobility 7500.
biotube: which distro?
Antiemes: Arch.
biotube: well, try the latest git and see if that helps
Antiemes: OK.
osiris_: how do I create a macro that would expand SWZ(A, R, G, B) into something like R500_C0_SEL_A | R500_C1_SEL_R | R500_C2_SEL_G | R500_C3_SEL_B
biotube: R500_C0_SEL_##A IIRC
MostAwesomeDude: That.
osiris_: biotube: thanks
spstarr: ok two things to try
spstarr: the hotplug script someone provided into ramdisk and built in firmware...brb
spstarr: closes the other bug
spstarr: airlied: its distro problem, adding hotplug script fixes
spstarr: KMS is also *NOT* hanging with r6xx now
spstarr: :))
spstarr: so no regressions at all
spstarr: logs bug with debian for them to fix that
spstarr: my issue is solved.
spstarr: agd5f: the only issue i have now is FBOs seem broken
spstarr: wow
spstarr: with KMS on 3D is quite sluggish at the moment, which is known
spstarr: thats ok, because KMS + DRI2 isn't ready yetish
spstarr: boots into non-kMS
MostAwesomeDude: agd5f: Ping.
spstarr: I also notice the laptop is less hot
soreau: spstarr: It's all in your mind
soreau: It's actually raging inferno hell in there
spstarr: lol
spstarr: its hot, but touching the bottom isn't enough to cause me to remove my hand from a spot
soreau: You're hands have callused from the heat tests over time
soreau: Your* ;)
spstarr: hahahaha
agd5f: glisse, airlied: I wonder if the problems are stemming from VGA_HDP_BASE
MostAwesomeDude: agd5f: Not to be a bother, but... there's a few vert attr types at the end of the PSC table that are missing.
MostAwesomeDude: 13, 14, and 15.
MostAwesomeDude: Are they reserved, or just not there?
MostAwesomeDude: I was really hoping to discover something like long int attrs.
MostAwesomeDude: But if the support's just not there, then I'll ignore it.
EruditeHermit: tormod, hey
tormod: hi
EruditeHermit: tormod, did you ever get 2.6.32-rcX to load radeon kms properly?
EruditeHermit: remember I had that issue where i had to logout and log back in to get it to work
EruditeHermit: adding it to /etc/initramfs-tools/modules and rebuilding initramfs didn't solve the problem
tormod: EruditeHermit, did you try the fbcon loading like on the wiki.ubuntu.com/RadeonKMS
EruditeHermit: tormod, page does not exist
EruditeHermit: I tried adding fbcon to /etc/initramfs-tools/modules yes
EruditeHermit: and rebuilding initramfs
EruditeHermit: also to /etc/modules.conf
MostAwesomeDude: Oh God, r600 errata's almost as good as r500.
tormod: /etc/modules (without rebuilding) should do
EruditeHermit: yeah tried that too
tormod: https://wiki.ubuntu.com/X/RadeonKMS
MostAwesomeDude: "If 8 MRTs are in use, there are 8 samples per pixel, and MRT7 had a fast-clear performed on it..."
MostAwesomeDude: Boy, is that ever a likely combination. :3
EruditeHermit: tormod, yeah that didn't work
tormod: EruditeHermit, so if you do it manually: boot without KMS, stop gdm, unload radeon, load fbcon, modprobe radeon modeset=1, start gdm, it still does not work? logs?
EruditeHermit: tormod, it works if I do it manually
EruditeHermit: tormod, what I normally do is the following
EruditeHermit: boot with grub radeon.modeset=1
EruditeHermit: login
EruditeHermit: logout
EruditeHermit: and then it works
EruditeHermit: login again
EruditeHermit: X starts before fbcon
m03sizlak: hey radeon team, particularly airlied, great fscking job!
m03sizlak: kernel -122 is by far the fastest and most stable my radeon has been ever
m03sizlak: and my system even wakes up from suspend now...for the first time since i put f12 alpha on
happycube: nice
MostAwesomeDude: Yeah, 122 appears to be pretty fucking sweet.
m03sizlak: so, THANK YOU
m03sizlak: of course its not perfect yet, but its come a LONG way
osiris_: zhasha: what's the status of your CS decoder?
happycube: mostawesomedude - where's the errata posted?
MostAwesomeDude: happycube: In the r600 docs.
MostAwesomeDude: x.org/docs/AMD
MostAwesomeDude: I enjoy them because of how silly they are.
happycube: cool
Kaapa: wow, I take that back
Kaapa: after reinstalling original mesa, both suspend and hibernate work
Kaapa: .oO( and just crashed after I did a logout on X ? wtf?! )
Kaapa: question about a behavior of my M72: after resuming from suspension, the system crashes if I try to switch to the console (or do logout on X)
Kaapa: is this familiar to anyone?
Kaapa: this does *not* happen on a suspend to disk, only to ram
Kaapa: and it's a M76, not M72
orion: (EE) RADEON(0): [dri] RADEONDRIGetVersion failed to open the DRM
orion: [dri] Disabling DRI. <-- is this my problem
chithead: the actual error is probably before that
orion: that is the frist error in log there is a warning
EruditeHermit: orion, post the log
orion: ok hold 2
orion: http://pastebin.com/m522ae941
orion: the card is a 4850 with dual dvi outputs run through a dongle to an oldish 17" svga style monitor
orion: i have tried both outputs
orion: it allowed me once to ctl+alt+f2 to another window as it usually white screens and locks hard and I was getting an info message from the montior saying that the horizontal was 74.8 when as far as I am aware this monitors max is 70
orion: and thats what it says its max is in xorg.conf also
soreau: orion: Did you compile you kernel with kms support as well as libdrm, mesa and ddx with modesetting? (as the link in the topic describes)
orion: i have tried evrything written absolutley everywhere on the inetrnet over the last 8 days so yes
soreau: Do you have fbcon module loading before radeon ?
orion: how to check
orion: ?
soreau: lsmod?
orion: there is no radeon and no fbcon in lsmod
orion: mind you this boot is with the ati named driver as it wont work otherwise
soreau: Did you build your kernel with FRAMEBUFFER_CONSOLE support?
orion: sevral installs ago yes
soreau: what about your current install?
osiris_: airlied: is there a reason why the bo and mt pointers are stored in radeon_tex_obj and all it's images?
orion: mandriva 2010.0 stock standard. fglrx module loaded into kernel. very barry basic as I dont want to screw anything up this time
soreau: Seems your kernel does not have kms enabled either
soreau: orion: fglrx? did you *read* the topic
orion: as i said i have tried everything i have found online but have had to reinstall the system several times. so this time around no
soreau: orion: You can't follow every bit of information you find on the internet. Try actually *reading* the wiki link in the topic.
orion: i did and follwed it faithfully and still got the white screen and crash
soreau: read the wiki more carefully
MostAwesomeDude: orion: Okay, you can't have fglrx and radeon installed at the same time.
orion: i have done it twice on two clean installs prior to this one
soreau: Reinstalling != reading
orion: i am prepared to try again. yes i did read and on two occassions the crashes corrupted me completel;y
orion: no response?
soreau: Did you ever actually compile drm-next as described in the wiki?
orion: i fololowd the ionstrutions to the letter on two seperate occassions
soreau: and completely remove fglrx, again, as described in the wiki?
orion: yes yes and yes
soreau: Does 'modinfo fbcon' show anything?
orion: copuld not find
orion: i am goign to try the wiki once again. I am sorry if i seem short with you but this is driving me nuts
soreau: Then you either compiled it into your kernel or did not compile it ata ll
orion: as i said this is a new clean install
soreau: You should compile drm>radeon and fbcon as modules at least for debugging
soreau: and use the drm-next kernel sources as described in the wiki
orion: i will go through it again and come back and let you know the results
orion: just so as we are on the same page
orion: what aprt of the instructions do you want me to do?
soreau: You need to 1) Download the kernel and checkout the drm-next branch 2) Compile libdrm from git with --enable-radeon-experimental-api 3) Compile mesa and ddx against that version of libdrm
soreau: mesa and ddx from git as well
orion: and remove as per instructions all fglrx stuff
soreau: Yes, definitely get rid of everything fglrx
soreau: and in your kernel config, you want device drivers>graphics>drm>radeon as module
soreau: at the very least
orion: ok shall do
orion: god i hope it works this time
cxo: I just noticed that both mplayer and xine crash when I use opengl video output with my r700
cxo: is that to be expected?
MostAwesomeDude: Are there any important texture format differences between r500 and r600?
MostAwesomeDude: The only one I can find is that NPOT is now totally legal.
soreau: Will npot be aavailable on r3xx when gl2.x is supported?
MostAwesomeDude: Other way around.
soreau: I'll take that as a yes
MostAwesomeDude: I'm going to make mesa-st do hax to pretend NPOT works. :3
MostAwesomeDude: But yeah, it's too important to leave out.
MostAwesomeDude: I've just been working on more important things. :3
EruditeHermit: MAD: what state is the gallium driver in now?
MostAwesomeDude: EruditeHermit: Works for HW TCL. glxgears is great; haven't really tested much else but things should be starting to work.
MostAwesomeDude: compiz is glitchy as hell but doesn't crash, for example.
EruditeHermit: ah
EruditeHermit: how about GLSL?
MostAwesomeDude: r600g is in its infancy; it can't even do glxinfo yet.
MostAwesomeDude: r300g lies and claims to do GLSL.
EruditeHermit: I am compiling it now
EruditeHermit: =p
MostAwesomeDude: YMMV; lemme know things that it doesn't work on.
EruditeHermit: MAD: can you help me with the compile for a second?
EruditeHermit: it used to build but now it errors
EruditeHermit: http://pastebin.com/m687934b6
MostAwesomeDude: You need to make clean, reconfigure, make again.
soreau: MostAwesomeDude: Do you need to use state tracker driver aka modesetting to do this?
EruditeHermit: ./configure --prefix=/home/user/xorgit/mesa --with-dri-drivers=r300 --enable-gallium-radeon --with-state-trackers=dri,g3dvl,xorg
EruditeHermit: does that work?
soreau: EruditeHermit: I am using ./autogen.sh --prefix=/opt/xorg --with-dri-drivers="" --enable-gallium-radeon --with-state-trackers=dri,glx --disable-egl
EruditeHermit: ah ok
soreau: hmm
soreau: Seems compiz works fine here except messages like /usr/bin/compiz (core) - Error: Couldn't load plugin '/usr/lib/compiz/libneg.so' : /usr/lib/compiz/libneg.so: undefined symbol: _ZN17GLWindowInterface13glAddGeometryERKSt6vectorIN9GLTexture6MatrixESaIS2_EERK10CompRegionS9_
soreau: hm
cxo: what an ugly symbol name
Luzipher: wow, there is r600g code already ?
Luzipher: cheers
MostAwesomeDude: Cool your horses; it's gonna take me a while because I don't know the HW at all.