nanonyme: DanaG: You'd think Ubuntu devs would at the very least disable broken staging drivers... :(
soreau: reboots airlied
airlied: soreau: kernel didn't boot last time, need to upgrade more bits
kcodyjr: love to know wtf a distro thinks it's doing enabling staging drivers in its common build
airlied: kcodyjr: its one of the reasons staaging exists
airlied: distros were including !upstream drivers on their own
airlied: it was decided that at least havign everyone shipping crap drivers from one place would be better
kcodyjr: ok, i'll buy that. known crap beats strange crap. but crap is still crap; if it was ready, it wouldn't be !upstream, and it wouldn't be staging either
kcodyjr: cluebats the distro maintainers, to no avail.
airlied: kcodyjr: lots of out-of-tree drivers worked but weren't clean enough to ship upstream for various reasons
airlied: ah nuts I'm crap at doing maths
kcodyjr: works != ready for the unwashed masses. what kind of maths?
kcodyjr: last week or so of working with that allocator has been like a flashback to my first c.s. teacher. i'd forgotten all about binary heaps and heapsorts.
airlied: nah I'm failing addition subtraction and divisioin at t he moment
kcodyjr: i think that symptom would cause bender to burn some booze
airlied: soreau: load detect worked on my laptop
airlied: didn't switch on though
soreau: airlied: elaborate please? What do you mean load detect worked? assuming by didn't switch on you mean no output at all?
DanaG: I draw the line at "fails to compile".
airlied: soreau: it cames up as connected
DanaG: Bonus points for the failure being caused by a missing file.
Nightwulf|work: hi all
dileX: hi Nightwulf|work
Nightwulf|work: hi dileX
airlied: soreau: more bad news, kms booted on it now
airlied: soreau: what type of cable you using? s-video or composite?
soreau: airlied: S-video
soreau: airlied: This the the kernel log output of dmesg | egrep "radeon|drm" for both http://pastebin.com/m1d3ed00f
soreau: not sure if that means anything
soreau: airlied: Which chip model are you testing there?
airlied: this is an rv370
airlied: wierd crtc 1 is connected to a TV
airlied: it knows something is there
airlied: hehe laptop finally crashed
airlied: probably need to clean the fan out
soreau: Where could it be failing after that here though?
airlied: ntsc or pal?
soreau: ntsc afiak
soreau: I'm almost absolutely positive it is ntsc
soreau: This is an rv350
soreau: agp card
soreau: radeon 9600 256mb
soreau: has served me well for many years now, about 6 or 7
airlied: I think my 9600 has tv-out also, its just not close to the tv monitor
soreau: airlied: I have a 25ft+ cable I've hacked together ;)
airlied: well my tv-out monitor is at my desk, my 9600 is 3 floors above me in my lab
airlied: I try to keep AGP hw away from me
soreau: Seems you've done a pretty good job in succeeding there
EruditeHermit: you have a lab?
airlied: EruditeHermit: yeah its cold up there though
EruditeHermit: I wish I had a lab at home
airlied: this is in mny office ;)
soreau: I would expect it to be colder in the basement
EruditeHermit: thats different
EruditeHermit: not as cool as a lab at home
airlied: we call those a shed usually
EruditeHermit: I call it the garage
airlied: speaking of which, time to hop on the bike
soreau: airlied: So it at least shows connected via xrandr but still shows nothing on the tele or ?
airlied: soreau: no it all worked when I rebooted
airlied: soreau: did you boot with no VGA plugged in at all for a test?
soreau: airlied: Always with VGA plugged. Never tested without
airlied: I've tested on rv200 and rv370 I'd expect the in between ones
airlied: to work
soreau: airlied: Only tested with VGA+S-video plugged, no secondary monitor plugged in
soreau: So I must have some oddity here
soreau: airlied: Testing on fedora?
airlied: it might be the 25ft cable confusing load detection
airlied: soreau: yup
soreau: airlied: Is there a fedora image I should d/l and test with?
airlied: soreau: well its just upstreak kms code
airlied: the userspace makes no difference
soreau: airlied: You think even though the same cable works with bios splash image and ums, it still could be an issue?
airlied: no thats wierd alright
airlied: so it must be something stupid in the register programming
soreau: I think I want to test a fedora dvd live cd
airlied: try with no VGa plugged in for a go, report more stuff int he bug
airlied: I'm not sure when I merged tv-out, the last fedora test day CD should work
airlied: outta here&
soreau: thanx & thinks
ball: Are there good open source X.org drivers for Radeon HD4200?
dileX: ball: the development is still work-in-progress. you might have a look into
ball: dileX: I may just go with an older board then
ball: ...to be sure it will work.
dileX: ball: which distro do you use?
hifi: ball: also depends what you want to do with the GPU
ball: dileX: NetBSD mostly, perhaps with some Linux also
ball: ...may have to run Vmware ESXi to get both going at the same time
hifi: if you only need accelerated 2D, any older board would do
ball: hifi: accellerated 2D and video playback
ball: 3D would be nice, but is a luxury.
dileX: ball: if you want to get an impression or for testing try latest ubuntu alpha release
ball: dileX: I don't have the hardware yet
ball: ...but thanks
rainbyte: ball, 2D and video are working well for r600/r700
hifi: also if you use a free linux distribution (like Debian or gNewSense) or NetBSD does not approve the firmware blob you might want to consider *really* old boards
hifi: if you care
ball: rainbyte: Is HD4200 an r700 ?
ball: rainbyte: excellent, thanks.
rainbyte: ball, the 3D accel is work-in-progress, but it is improving very fast
ball: rainbyte: okay. It would be nice to try that one day.
rainbyte: that's in Linux, in bsd i think there is only 2D and video
ball: rainbyte: If I can't get virtualisation running, perhaps I can dual-boot
rainbyte: if you want more features now (KMS, 3d, etc) you can buy a r500 (HD1xxx)
ball: What is kms?
ball: None of the boards I'm looking at have HD1xxx
nanonyme: ball: Kernel modesetting; moves graphics-related direct hardware tampering to kernel.
ball: nanonyme: Oh dear.
nanonyme: Well, it allows for eg higher OpenGL versions and better power management so I wouldn't complain. :)
airlied: MrCooper: vertices don't contain erlocs
airlied: so we won't be doing wierd seeks on vertex data packets
airlied: except maybe on r100
MrCooper: airlied: k, I wasn't sure if it was relevant but thought I'd mention it anyway
airlied: btw if get some time testing that path for speed regression might be good.
airlied: on my test box I wasn't seeing a major loss in x11perf with it
MrCooper: has anyone played with putting vertices in a separate BO? It looks like we can only fit 600ish characters in a single CS...
airlied: MrCooper: ould steal that ode from r600
airlied: hmm dodgy c key
airlied: it was a bit of overhead on fire and forget vs re-using the vbos
airlied: roll on gallium
airlied: ddx an use the 3d driver optimised bits
MrCooper: yeah, my long term hopes certainly are on the Xorg state tracker :)
twoerner: agd5f: jup, it was the dram_prim commit
twoerner: airlied, agd5f: ok.. latest mesa checking fixes my drm problem, but there are new render problems
twoerner: just give quake3-demo a run
adamk_: ut2004 is crashing here with Mesa from git as of yesterday... "Kernel failed to parse or rejected command stream. See dmesg for more info" dmesg: http://pastebin.com/m28a0ea2d
suokko: adamk_: Which card is that?
suokko: And did you enable compressed textures?
suokko: pitch is wrong in any case
adamk_: Yes, compress textures was enabled because I had been playing NWN earlier. I disabled it and now ut2004 starts up.
adamk_: Hmmm.. But now my mouse doesn't work in the game.
adamk_: Actually, it doesn't even work in the menu. I noticed the same problem with return to castle wolfenstein, which gives me an X error of failed request: "XF86DGANoDirectVIdeoMode"
suokko: There is really something badly wrong in legacy reference counting :/
suokko: in mesa
suokko: adamk_: I think that DGA video mode was deprecated and you need to do something special to make it work
adamk_: It's just odd.. It was working just a few days ago, and I haven't done anything other than enable KMS.
adamk_: I'm about to reboot with KMS disabled to see if it works.
suokko: adamk_: It is only disabled in KMS I think
adamk_: Well that sucks.
adamk_: It seems I'm always have to decide what I want to trade-off. I can use KMS + DRI2 at home, but only on one monitor. I can use KMS + DRI2 at work, but my mouse is useless in various commercial games :-)
adamk_: I guess I can't really complain. It is still all under development :-) I just hope these quirks are fixed before too long.
spreeuw: just use dri1
adamk_: Well that's another trade off :-) And, besides, if I have KMS enabled, but add 'Disable "dri2"' to the Module section, X hangs :-)
suokko: It doesn't disable dri2 from mesa
adamk_: If I try to start X a second time, it gives me: symbol lookup error: /usr/lib64/xorg/modules/drivers//radeon_drv.so: undefined symbol: DRI2ScreenInit
adamk_: Oh, so the suggestion is to recompile Mesa with dri2 disabled :-)
suokko: and not from ddx even ;)
suokko: adamk_: suggestion is to ghave 2 grub options where one has radeon.modeset=0
adamk_: Which is what I've been doing.
suokko: Dual-boot ;)
adamk_: I'm just hoping these problems get worked out, that's all.
suokko: Another option is to unload radeon and unpin fbcon and then reload radeon module
adamk_: Hmmm... A search for 'kms dga' in the freedesktop bugzilla doesn't bring anything up. Perhaps I should open up a bug.
adamk_: Or, rather, it doesn't bring up this particular bug.
twoerner: mesa commit f338de4018f2cfcb462f98c7d76bf6cc36ab6f47 was the latest, which was working for me without problems
twoerner: the draw_prim made it unusable, drm errors
twoerner: latest version is working with drm, but textures are not fully visible and i have lines all over the scene
twoerner: and also crashes
taiu: yes, there's some bits wrong in r700SetupVTXConstants2 for new draw path
taiu: also in r700TryDrawPrims r700PredictRenderSize can flush but we'v already set up index buffer and it asserts after that
suokko: taiu: yes. You have to setup index buffers after prediction
taiu: i'm sure when agd5f wakes up well see some fixing soon :)
suokko: And we should even do allocation first and check mememory and only then copy data to allocated memory but none has yet had problems with too little memory
agd5f: twoerner: you can disable the draw_prims code until it's fixed up. just comment out line 1133 of r700_render.c
twoerner: agd5f: does not work
twoerner: --- a/src/mesa/drivers/dri/r600/r700_render.c
twoerner: +++ b/src/mesa/drivers/dri/r600/r700_render.c
twoerner: @@ -1130,6 +1130,7 @@ static void r700DrawPrims(GLcontext *ctx,
twoerner: context_t *context = R700_CONTEXT(ctx);
twoerner: /* For non indexed drawing, using tnl pipe. */
twoerner: +#ifdef BLUBBER
twoerner: context->ind_buf.bo = NULL;
twoerner: @@ -1138,6 +1139,7 @@ static void r700DrawPrims(GLcontext *ctx,
twoerner: index_bounds_valid, min_index, max_index);
twoerner: r700DrawPrimsRe(ctx, arrays, prim, nr_prims, ib, index_bounds_valid, min
twoerner: this sould do it, right?
twoerner: or should it be used everytime=
twoerner: ahh.. only line 1133 :-)
twoerner: then it works
Georhime: this tnl pipe , what aim/purpose it has, haven't figured it out?
Georhime: NOTES inside tnl suggests like it could be a sw implementation of some transformations, however all drivers use it
Georhime: ah that shoudn't matter same looking functions as main , just to wrap around hw ones, never mind
frej: hmm, the 'vbo' test in progs/perf crashes for me, anyone else?
frej: #0 radeon_cs_space_add_persistent_bo (cs=0x83e7080, bo=0x0, read_domains=2, write_domain=0) at ../../../libdrm/radeon/radeon_cs_space.c:168
frej: #1 0x00fc045b in r300AllocDmaRegions (ctx=0x80a5af0, arrays=0x80e3244, prim=0xbfffeff0, nr_prims=1, ib=0x0, index_bounds_valid=1 '\001', min_index=0, max_index=0) at r300_draw.c:550
frej: #2 r300TryDrawPrims (ctx=0x80a5af0, arrays=0x80e3244, prim=0xbfffeff0, nr_prims=1, ib=0x0, index_bounds_valid=1 '\001', min_index=0, max_index=0) at r300_draw.c:653
frej: #3 r300DrawPrims (ctx=0x80a5af0, arrays=0x80e3244, prim=0xbfffeff0, nr_prims=1, ib=0x0, index_bounds_valid=1 '\001', min_index=0, max_index=0) at r300_draw.c:706
frej: #4 0x0107ef91 in vbo_exec_DrawArrays (mode=0, start=0, count=1) at vbo/vbo_exec_array.c:523
frej: #5 0x01075db7 in neutral_DrawArrays (mode=0, start=0, count=1) at main/vtxfmt_tmp.h:328
frej: #6 0x08067ef0 in UploadVBO (count=2) at vbo.c:73
frej: #7 0x080680b0 in PerfMeasureRate (f=0x8067e60
frej: #8 0x08067e19 in PerfDraw () at vbo.c:198
frej: #9 0x080682c7 in Draw () at glmain.c:185
frej: #10 0x006da5bb in ?? () from /usr/lib/libglut.so.3
frej: #11 0x006ddea3 in fgEnumWindows () from /usr/lib/libglut.so.3
frej: #12 0x006daf80 in glutMainLoopEvent () from /usr/lib/libglut.so.3
frej: #13 0x006db3dd in glutMainLoop () from /usr/lib/libglut.so.3
frej: #14 0x08068285 in main (argc=1, argv=0xbffff424) at glmain.c:265
frej: ARhg.. sorry
frej: http://pastebin.com/m4c7ed20e <- intended paste
frej: wrong buffer :(
agd5f: twoerner: just line 1133
twoerner: agd5f: jup, that makes it work..
twoerner: agd5f: even quake3-demos is working again
twoerner: also gooleearth
agd5f: twoerner: right. the new draw prims is still a wip
nanonyme: agd5f: What benefits does the change to draw_prim bring, btw? (if such a listing is trivial)
agd5f: nanonyme: native support for vbos and index buffers in hw
nanonyme: Ooh, nice. :)
nanonyme: That's plenty good enough a reason to have a broken driver for some time. ^^
Georhime: airlied: anholt is surely right about that flow thing , that what you ment? what bothers me are the false statements in court "..." not a paranoia , i am not willing to accept that shit -- utemps and regs, branching is difficult hw looping with hw registers, now my gateway is also blocked
nanonyme: (writing on a blog at the same time as the other person and competitively rewriting your messages)--
boghog: what's this about :p
boghog: <- curious
nanonyme: Just a small two-message conversation on Phoronix.
nanonyme: Who needs more than two messages when you've an edit button, huh? ;)
nanonyme: Hmm, there appears to be an off-by-one error in r600 KMS in some games.
nanonyme: Right border and bottom aren't rendered properly.
nanonyme: Just experienced with ut2004demo
boghog: a 1 pixel border?
nanonyme: I think so.
boghog: I've seen that too, for a while now
boghog: on my rv770 with kms
nanonyme: Approximately one, of course it's only a by-eye estimate.
boghog: (dunno about non-kms)
nanonyme: Might be two too, of course. :)
nanonyme: Maybe I should try if it's related to the cliprect off-by-one hacks?
nanonyme: (since apparently coordinates aren't enabled the same way in r300 non-KMS and KMS, it brings me to wonder if r600 KMS actually *is* wrongly implemented the same way as r600 non-KMS ;)
nanonyme: Worth a try?
boghog: i dunno, I guess that is a good place to start looking, but really i'm clueless :p
nanonyme: Well, there's a trivial hack that should allow me to see it.
nanonyme: Make the non-KMS coordinate hack in Mesa apply also for KMS and see if it fixes borders.
boghog: is going to see if his other radeon 4850 spontaneously came back to life before requesting an RMA for it
boghog: i wonder why my 360watt power supply has power connectors for two pci-e cards
nanonyme: I think I was right.
nanonyme: r600 KMS rects handling probably differs from r300 KMS rects handling in some very crucial way.
nanonyme: Allowing the DRI1 hack for r600 DRI2 fixed the borders, that is.
nanonyme: (wait a minute, wait a minute... there was more stuff there than just the increase by one, have to check if it matters)
boghog: really? neat
nanonyme: boghog: Well, it's not a fix but I'm pretty sure cliprect handling is somehow wrong in r600 KMS.
nanonyme: (that is, while it seems to be rendering right, doing it like that would break rendering pretty much everywhere else...)
boghog: meh, 4850 is still dead :(
boghog: i so hate havign to rma stuff
legume: boghog: Isn't 360w a bit low for a 4850
nanonyme: So with KMS dPriv->x == 0 and dPriv->y == 0 pretty much always. That would hint towards it actually being just a simple off-by-some bug.
boghog: im not sure, it doesnt have much else plugged in right now, and the cpu shouldnt eat much power
boghog: (amd x2 4850e)
boghog: maybe you're right though
legume: boghog: http://ati.amd.com/products/radeonhd4800/requirements.html says 450
nanonyme: hopes he doesn't have to start kernel hacking if he wants to fix this >.<
nanonyme: also thinks he's getting surprisingly far without having a clue what he's doing
legume: nanonyme: FWIW openarena fullscreen 1280x1024 is more like 4px off at bottom less at right side.
nanonyme: legume: Might be. Using the DRI1 hack fixed it still for me though.
legume: nanonyme: I also don't see it on mesa demos anymore, but used to.
nanonyme: Try ut2004demo?
nanonyme: I got it there.
nanonyme: Anyway, I'm atm trying to figure out how r300 and r600 KMS handle scissors so I could find out if there's differences in the two.
nanonyme: (of course I could just be looking at the wrong point of the code path)
kdekorte: nanonyme, have you run the scissor test is piglet? That is how we found and fixed the DRI1 case on r600
suokko: yes. Someone should sstart to run piglit tests with r600 and upload results
suokko: (or send the results for me so I can upload them)
nanonyme: I wonder what this is about: /* Hardware expects inclusive bottom-right corner */
boghog: are piglit tests included with mesa? or what is piglit actually :p
suokko: It means that you give a coordinate that is inside the draw area and not the first coord outside ;)
nanonyme: It's a test suite, I think.
agd5f: nanonyme: I think r6xx is exclusive. need to check the reg spec
suokko: boghog: Regression test suit and it is in freedesktop git but not with mesa
nanonyme: agd5f: No worries, that's from r300.
nanonyme: I'm reading through both drivers.
Georhime: nanonyme: hei, is the i965 the only open source chip that has had working glsl entryoints, or is there a radeon or some other repo as well?
nanonyme: Why should I know? :D
MostAwesomeDude: Georhime: The ARB entrypoints are already there. Dunno if nha or somebody else has added them yet.
Georhime: hmm, i'm quite a new to opengl miself also, never followed
nanonyme: I'm mostly an end-user who's sometimes more, sometimes less out of his league. ;)
kdekorte: suokko, I got a bad scissor on r600 kms, I'll put it on pastebin
Georhime: MostAwesomeDude: it should go through the same client calls , so arb entrypoints suite too, but vars are obviously different, but tracking them is documented inside shader_api.c
kdekorte: suokko, http://pastebin.com/m1ffc0b7b
kdekorte: suokko, http://pastebin.com/m3d54e255 is the glean output (subset of it)
nanonyme: agd5f: Should the exclusiveness/inclusiveness affect cliprect code in Mesa at all, btw?
agd5f: nanonyme: no. driver deals with it when it sets the scissor regs
Georhime: MostAwesomeDude: so radeon hasn't got a glsl yet? i965 pushes like all the backend+frontend to driver, wonder why that one in shader/ woudn't suite compile.c, but that's all fuzz, maybe you know how to deal with registers, seems like hex bitmangling
suokko1: I think scissors looks like inclusive/eclusive bug in r600 code
MostAwesomeDude: Georhime: Honestly, GLSL should be done in core Mesa. I really really hate how i965 has special setup code.
Georhime: it's fairly large too
suokko1: But Intel doesn't like to share the code in core :/
Georhime: looks to me as well, that in core mesa it should me done
nanonyme: suokko1: Would sound like a very easy explanation for the off-by-one issues...
nanonyme: But as to where exactly it can happen...
suokko1: nanonyme: r700SetScissor
nanonyme: Ah, point.
suokko1: nanonyme: http://nopaste.com/p/a9AIar0yR
suokko1: Others paths should be validated too
suokko1: agd5f: Is top_left exclusive too? Does it work with (-1, -1) coordinate correctly if exclusive?
agd5f: suokko1: I don't think so. I think you want 0, 0 there
suokko1: From test it looks like exclusive
agd5f: maybe not. should say in the reg spec
suokko1: Or some other radeon code modifying it wrongly
agd5f: for the scissor regs
suokko1: I guess that is off-by-one error somewhere else then
nanonyme: Anyway, I'll leave you guys at it. I just thought I'd poke at it as far as I can deduce, didn't think I could actually formulate the fix. ;)
nanonyme: It's just a bug that I hit doing expose on Compiz and this time I got annoyed enough after hitting it elsewhere too. :)
soreau: nanonyme: Scale. It's called scale
nanonyme: soreau: The effect?
soreau: yes. There is no expose in compiz
nanonyme: Well, it's a plagiate from the expose effect in OSX anyway so why not call it that? ;)
suokko1: Apple wouldn't like that name ;)
nanonyme: Oh, good point, it might be a trademark.
suokko1: kdekorte: http://nopaste.com/p/aUYgso5sI ?
nanonyme: suokko1: Err, what have you been doing? o.O
suokko1: Just fixing some funny code related to scissors where they sometimes are exlusive and sometimes inclusive
nanonyme: The original reason why I mentioned the inclusive/exclusive thing is that r300 DRM already decreases x2 and y2 by one in DRM.
NeKit: is there anything new with 3D support for r670?
suokko1: That should only happen in old code
nanonyme: Ah, ok.
soreau: NeKit: Since.. when?
NeKit: is this is still valid?
nanonyme: Since today morning Richard disabled draw_prim so games run again? ;) Otherwise probably nothing out of ordinary. (as in, only gradual progress)
spreeuw: NeKit: it runs ok give it a go
nanonyme: NeKit: Not really, it runs a multitude of games nowadays.
nanonyme: That blog post is ancient history. ;)
spreeuw: NeKit: install drm and mesa from git
spreeuw: drm from agd5f's tree
spreeuw: then you can drop in replace it with dri1
NeKit: I mean, these instructions still can be used?
spreeuw: dont know those details
nanonyme: Well, agd5f has been updating his git repo so I suspect them to work.
nanonyme: Especially since we made sure it got compatted to 2.6.31 API before release. ;)
soreau: NeKit: Follow the link in the topic
NeKit: soreau: ok
nanonyme: I'd assume no one really cares that much of whether or not it compiles against 2.6.32. :p (I don't ^^)
NeKit: will it now work with Dolphin?:)
nanonyme: No promises on particular programs, go test.
soreau: NeKit: Test it and let us know
soreau: I dont see why it wouldnt
nanonyme: Assuming it doesn't require OpenGL higher than 1.4, it might well work.
spreeuw: bo(0x9ef4fa0, 1048576) is mapped (-4) can't valide it.
spreeuw: validated 0x9ef4fa0 [0xD3155000, 0xD315D000]
spreeuw: what does that indicate?
spreeuw: File radeon_dma.c function radeonReleaseDmaRegions line 348
spreeuw: Leaking dma buffer object!
suokko1: spreeuw: That means there is bug with reference counting in legacy code
suokko1: You might want to add assert to that can't validate and post backtrace here
spreeuw: I got it with dangerdeep
spreeuw: may be somethign with the water shader
spreeuw: openarena problem of yesterday is gone now
suokko1: kdekorte: MY change at least passes scissor tests with r200. Can you rerun with r600?
suokko1: nanonyme: Maybe you can run the piglit test for scissors?
Zajec: spreeuw: "r600 : disable draw_prim for now."
kdekorte: suokko1, gonna be a few minutes... got a issue I have to deal with first
suokko1: ok :) No hurry
kdekorte: suokko1, waiting on hold so I thought I would try it... it ALMOST works...
kdekorte: passes all but a couple of the tests
kdekorte: what would you like to see?
spreeuw: Commander is being attacked
spreeuw: Team1(UnnamedPlayer) has been shown the door
spreeuw: drmRadeonCmdBuffer: -16. Kernel failed to parse or rejected command stream. See dmesg for more info.
NeKit: when I was trying to update drm:
spreeuw: dgunned the opponent commander
NeKit: nekit@COREQUAD:~/mesa/drm$ git fetch && git rebase origin
NeKit: linux-core/Makefile.kernel: needs merge
NeKit: shared-core/radeon_drv.h: needs merge
NeKit: shared-core/radeon_irq.c: needs merge
suokko1: kdekorte: The erros would be nice to see
spreeuw: the n[drm] Resetting GPU
spreeuw: [drm:radeon_cs_ioctl] *ERROR* ib_get failed
suokko1: spreeuw: GPU hung :/
soreau: NeKit: Try git clean
spreeuw: I happened when the enemy commander blew up
spreeuw: its a little rough on the gpu
spreeuw: NeKit: you need a special branch
kdekorte: suokko1, ha... ran it again and they all passed...
kdekorte: suokko1, so I guess your patch is good
spreeuw: git clone git://anongit.freedesktop.org/~agd5f/drm
spreeuw: cd drm
spreeuw: git checkout -b r6xx-r7xx-3d origin/r6xx-r7xx-3d
NeKit: spreeuw: clone will download all from start, right?
spreeuw: yeh just do it to make sure
spreeuw: only a few mb
suokko1: nha: Can you run scissors tests from piglit with http://nopaste.com/p/aUYgso5sI ?
spreeuw: NeKit: for mesa you can use main git
suokko1: I would like to see if r300 passes too
spreeuw: NeKit: git clone git://anongit.freedesktop.org/git/mesa/mesa
spreeuw: and you also need git video-ati ddx
nha: suokko1: I'm on my way
NeKit: spreeuw: I use GPRS
spreeuw: flat rate?
spreeuw: in that case give git checkout -b r6xx-r7xx-3d origin/r6xx-r7xx-3d a try
spreeuw: not sure if it works
spreeuw: you have to do it right the first time or you end up fighting git
NeKit: no, looks like I have to do clone...
nanonyme: suokko: Fine fine, I'll take a look when I get back from Helsinki.
nha: suokko1: on RV530, glean/scissor passes, but scissor-depth/stencil-clear and scissor-copypixels regresses; maybe you missed some rectangles in those paths?
suokko1: nha: Maybe. Can you post the errors?
nha: eh what?
nha: [ 513.160197] [drm:r300_cs_track_check] *ERROR* [drm] Buffer too small for color buffer 0 (need 45248 have 45056) !
nha: [ 513.160207] [drm:r300_cs_track_check] *ERROR* [drm] color buffer 0 (112 4 0 101)
nha: [ 513.160211] [drm:radeon_cs_ioctl] *ERROR* Invalid command stream !
nha: in dmesg
nha: for scissor-stencil-clear
suokko1: hmm. Doesn't sounds good
nha: same for depth
suokko1: So it is one picel too large and cs checker disallows it maybe
nha: yup, that could be it
nha: if adding height 1 causes it to just exceed the buffer size
kcodyjr: seems like an awfully odd-sized buffer
agd5f: maybe the drm command checker needs to be fixed for some cases
Pallokala: do we discuss here compilation-problems of x11-drm?
Pallokala: x11-drm from git
suokko1: agd5f: That was real error. But I think there was post at morning that cs checker thinks pitch for compressed texture is 1. I guess that is mesa bug too
suokko1: Pallokala: I think Gentoo has masked it
agd5f: suokko1: compressed textures pitches are weird like half pixels. I suspect the CS checker needs fixes
c_: radeon kms ,, will it generate some sort of vga hotplug events ?
agd5f: c_: eventually
c_: agd5f: someone is working on it ?
agd5f: c_: not at the moment
Pallokala: well x11 overlay has it (x11-drm -ebuild)
Pallokala: and so I thought that I should use it instead of drm in 2.6.31
chithead: Pallokala: it is masked for a reason
agd5f: c_: shouldn't be too hard to add
c_: agd5f:what about kms support for radeon 4*** cards ,, will that be in 2.6.32 ?
agd5f: c_: yes
agd5f: but no interrupts yet on r6xx/r7xx cards
marvin24_: something causes x-plane segfault on rs780
marvin24_: backtrace: http://pastebin.com/m54ec9c68
marvin24_: worked a week ago
c_: agd5f: meaning ,?
agd5f: c_: can't implement hotplug yet on r6xx/r7xx
c_: but it will be possible later ? 2.6.33 or ?
agd5f: c_: yes. it will be possible
agd5f: might even get into 2.6.32 if I get it working in time
c_: but fast mode switching will make it to 2.6.32 ?
agd5f: c_: all the other kms stuff will be there including 3d support
c_: agd5f: your a hero :D
c_: ill start to tell all friends to support ATI
c_: agd5f: you wont happen to know if intel kms with enable vga hotplug as of 2.6.32 ?
c_: i have a labtop with a pesky intel card
agd5f: c_: I think it works
c_: as in 2.6.31 or upcomming ?
agd5f: c_: 31 I think, but not sure. try in #intel-gfx
c_: agd5f: thanks
kdekorte: ajax: question about the -40 rawhide kernel. Did you just take the edid patches or other patches, specifically 5b31aee9d72f529ee6b60e8d66967f817a0e39fc which fixed r600 pitch issues?
ajax: just the edid patches.
kdekorte: ajax, bummer, the pitch issues help my r600 lockup issues, doesn't fix them, but the machine runs better with them
biotube: should KMS on an rv620 result in a mosaic effect?
agd5f: biotube: no
biotube: anything I should look for in the logs? nothing I can think of brings up anything out of the ordindary
agd5f: biotube: make sure you are using the latest code from drm-next
EruditeHermit: airlied, got time to look at my suspend/resume issues today?
soreau: airlied: News@10: S-video works OOTB with the daily F12 snapshot
biotube: agd5f: I pulled the code right before I built the kernel, according to the instructions in the link
soreau: So it must be something wrong with my setup
biotube: it left my ttys unusable, but X was fine until I restarted it with DRI enabled
agd5f: biotube: if you are using kms make sure you have an updated ddx and mesa as well
EruditeHermit: soreau, you have an rv350 right?
soreau: EruditeHermit: Yes
EruditeHermit: soreau, do you have time to test something?
agd5f: biotube: you need to load fbcon to enable console
biotube: agd5f: I built a new ddx and libdrm yesterday(both with --enable-radeon-experimental-api), mesa today(with --with-dri-drivers=radeon,r200,r300,r600,swrast and --enable-gallium-radeon)
soreau: EruditeHermit: Sure, whats up?
agd5f: biotube: ok
EruditeHermit: soreau, in firefox, can you install glazoom or zoomit. Make sure all your files are synced etc to prepare for a crash. Set the default zoom to something non multiple of 100 (i.e. 130%) and then visit engadget.com. Does it crash
biotube: just found this: (II) [KMS] drm report modesetting isn't supported.
agd5f: biotube: check dmesg
kdekorte: biotube: dmesg | grep drm is useful
biotube: both 'radeon' and 'drm' bring up nothing; 'kms' only brings up references to the kernel version(220.127.116.11-kms)
agd5f: biotube: 2.6.31 doesn't have kms support for r6xx/r7xx
agd5f: you need dave's drm-next branch
biotube: agd5f: I merged it in, per the instructions on the page
agd5f: biotube: shoudl should see a bunch of drm modesetting stuff in dmesg
agd5f: did you enable radeon kms in your kernel config?
biotube: could the fact that I started it with kexec change anything?
spreeuw: http://img223.imageshack.us/img223/9118/screen000t.png any idea what causes that weird dot in the fonts?
agd5f: biotube: try loading radeon with modeset=1
biotube: let me reboot
soreau: EruditeHermit: I installed glazoom and it works on engadget.com home page at 133% and when I try typing 130% it goes to 129% and still works
EruditeHermit: soreau, no crash?
soreau: EruditeHermit: I couldn't find an addon named zoomit
soreau: EruditeHermit: No crash of any kind. Works here
EruditeHermit: soreau, oh sorry the 2nd one is called no squint
EruditeHermit: but it shouldn't matter
EruditeHermit: either one of them should test it
EruditeHermit: soreau, are you using firefox 3.5 btw
EruditeHermit: it has to be 3.5 or newer
soreau: EruditeHermit: Mozilla Firefox 3.5.3, Copyright (c) 1998 - 2009 mozilla.org
EruditeHermit: AGP card right?
EruditeHermit: can you try with a zoom level of 160%
soreau: Works with 160% too
soreau: EruditeHermit: I notice there are videos there.. should any be playing at the same time?
EruditeHermit: oh no
EruditeHermit: just on page load it crashes for me
EruditeHermit: most of the time
soreau: Well I loaded the page then set zoom level
EruditeHermit: try setting the global zoom level
EruditeHermit: in glazoom
EruditeHermit: go to options
EruditeHermit: apply the following zoom to all pages
EruditeHermit: then set it to 160
EruditeHermit: then navigate away to another page
EruditeHermit: make sure its zooming the page to 160
EruditeHermit: then visit engadget.com
soreau: Done. Still works
soreau: I can't manage to crash it at all
soreau: EruditeHermit: What crashes, FF or X?
soreau: EruditeHermit: Can you try at 1280x1024 res?
EruditeHermit: let me try at that resolution
EruditeHermit: X crashes
EruditeHermit: actually the whole machine hangs
EruditeHermit: kernel and all
soreau: My guess is you're using a higher resolution since using such zoom levels is ridiculous at 1280x1024 which is my max here ;)
EruditeHermit: I have 1920x1200
EruditeHermit: but firefox doesn't follow DPI internally
EruditeHermit: i.e. the web page window doesn't follow DPI
EruditeHermit: may crash here
biotube: even starting with radeon.modeset=1 left me with a useless screen
EruditeHermit: soreau, hey, its not crashing!
biotube: the ttys were fine, but after X started they just stuck to being black screens
EruditeHermit: soreau, do you happen to have a 1920x1200 screen anywhere?
soreau: EruditeHermit: Nope. Even my LCD that went out is 1280x1024 max
EruditeHermit: soreau, or even can you make X run at that resolution even on your small screen
soreau: EruditeHermit: I doubt it but I can try
soreau: doesn't think this is a good idea
nanonyme: Anyone got r6xx/r7xx who could test something for me?
EruditeHermit: soreau, I think I am hitting some limits somewhere if I am not crashing at low resolution
soreau: EruditeHermit: I think glazoom is buggy because I was experiencing some strangeness when a page was loading and I was trying to click on a FF menu, it wasn't opening then when it did, there were two menus shown, only one of which was active.. it was weird. Wish I would have got a ss
idletask: Good evening
NeKit: Can't create opengl renderer. You might be missing some required opengl extensions, check the logs for more info"
NeKit: reports Dolphin
soreau: EruditeHermit: And I can't add a mode 1920x1200 xrandr: cannot find mode "1920x1200"
kdekorte: ajax, running rawhide latest kernel and glxgears results in X going to powered off screens, but the machine is up however no errors that I can find
soreau: NeKit: What does 'glxinfo|grep renderer' say?
NeKit: IRQ's not enabled, falling back to busy waits: 2 0
NeKit: OpenGL renderer string: Mesa DRI R600 (RV670 950F) 20090101 TCL
soreau: I guess he managed to crash it again ;)
EruditeHermit: soreau, sorry, I crashed
EruditeHermit: were you able to set it to 1920x1200?
NeKit: Dolphin requires OpenGL 2.0, as I know
soreau: NeKit: Well I guess it probably requires OpenGL 2.0/2.1. You can see the version you're currently using with 'glxinfo|grep nGL'
soreau: mesa only supports 1.5 at the latest, GLSL and gl 2.0 should be coming soon to a driver near you though
soreau: EruditeHermit: And I can't add a mode 1920x1200 xrandr: cannot find mode "1920x1200"
EruditeHermit: got an HD TV? =p
NeKit: will 2.0/2.1 be supported for r670?
NeKit: I can't get fglrx to work...
soreau: <- poor
EruditeHermit: no worries
EruditeHermit: me too
EruditeHermit: NeKit, eventually
EruditeHermit: probably will take a year though
soreau: EruditeHermit: I think glazoom is buggy because I did notice some oddity with it. That doesn't mean that apps should be able to crash the system though
nanonyme: NeKit: If you ask for whether it'll be supported soon, then no. I think the hardware itself should be able to do 3.0 or something.
NeKit: DesMuMe under wine works, but 3D is very slow
nanonyme: Maybe 3.1.
nanonyme: So those will eventually work...
EruditeHermit: soreau, I have a feeling I am hitting some limit
EruditeHermit: and it doesn't know what to do
NeKit: NO$GBA under OpenGL mode works, but I can't see any 3D
soreau: EruditeHermit: Anything in syslog?
EruditeHermit: soreau, nope
EruditeHermit: soreau, atleast I don't know what i should see there
nanonyme: Hmm, maybe a time to check again whether or not Linus' tree compiles. :)
Zajec: I can see Mesa 7.6's tree contains r600 driver
Zajec: and I can see it receives some work still
Zajec: but relnotes doesn't contain info about it: http://cgit.freedesktop.org/mesa/mesa/tree/docs/relnotes-7.6.html?h=mesa_7_6_branch
nanonyme: Maybe not considered ready enough with 7.6 yet?
Zajec: and Phoronix seems to don't see r600 in 7.6 as well: http://www.phoronix.com/scan.php?page=news_item&px=NzU1NQ
Zajec: could someone clarify this, please? :)
Zajec: nanonyme: so it will be somehow hardly disabled?
nanonyme: Probably not.
EruditeHermit: 7.6 hasn't been released yet
nanonyme: Zajec: I'd assume if it's not considered ready, just not compiled by default.
Zajec: Erektium: will be relese soon probably...
Zajec: nanonyme: oki :) understood,, thx
EruditeHermit: release info is usually last thing done
nanonyme: But seriously, you don't probably want to be using 7.6 if you want r600. :)
soreau: Don't read the release notes before the release ;)
Zajec: Erektium: incorrect auto-correction
soreau: Erektium: I think that was meant to be for EruditeHermit
Erektium: it appears so
nanonyme: Most people are running 7.7-devel for r600 anyway, me thinks.
Zajec: soreau: :P
Zajec: yeah, but for example Ubuntu 10.04... probably will get 7.6
Zajec: so would be nice to have r600 in 7.6... 7.6 maybe even
nanonyme: That would be sad...
Zajec: nanonyme: sad? which part?
nanonyme: You'd hope they would rather have a git snapshot of 7.7, at least.
Zajec: nanonyme: hm, do you think 7.7 will come out so soon?
nanonyme: Zajec: OpenGL version string: 1.4 Mesa 7.7-devel
Zajec: well...soon.. half of year actually
EruditeHermit: soreau, what does this say? xrandr --newmode "1920x1200_60.00" 193.16 1920 2048 2256 2592 1200 1201 1204 1242 -HSync +Vsync
soreau: EruditeHermit: That's weird. It worked but it added the mode to S-video ;)
soreau: EruditeHermit: Hang on, lemme try again
EruditeHermit: do xrandr --addmode VGA 1920x1200_60.00
EruditeHermit: or something
EruditeHermit: VGA or DVI
soreau: yea yea, I got it
nanonyme: Zajec: That's what I meant. They can get 7.7 git snapshots *now*. I'd assume any sensible person would for next April's release.
EruditeHermit: hopefully it gives you a desktop in panning mode
EruditeHermit: where you pan following mouse
agd5f: the r600 driver in 7.6 is just about the same as the version in master
nanonyme: agd5f: Yeah but for April's release? It'll probably move on by then.
Zajec: agd5f: first i was thinking like that :)
Zajec: nanonyme: if 7.6.1 is going to be bug fix release... probably some work will be still backported nicely
soreau: EruditeHermit: I used cvt 1920x1200 to get this modeline, but 'xrandr --addmode VGA-0 "1920x1200_60.00" 193.25 1920 2056 2256 2592 1200 1203 1209 1245 -hsync +vsync' just shows --help
agd5f: nanonyme: 7.7 will have more features, but right now, they aren't much different
nanonyme: Right now, yeah.
nanonyme: But assuming 7.6 will be out soonish, we're talking about a commit difference of many months.
EruditeHermit: soreau, xrandr --output VGA --mode 1920x1200_60.00
agd5f: soreau, EruditeHermit: if you monitor doesn't support 1920x1200, it won't work
EruditeHermit: agd5f, oh
soreau: agd5f: That's what I was thinking
EruditeHermit: agd5f, way back when it used to be that it would create a panning desktop
soreau: EruditeHermit: And I've already uninstalled glazoom so I'm pretty much done with it :)
agd5f: EruditeHermit: that's different
soreau: EruditeHermit: Just makes me depressed I don't have a real monitor lol
EruditeHermit: agd5f, if I get a crash at a resolution of 1920x1200 but not at 1280x800, does that mean I am hitting a limit?
agd5f: EruditeHermit: probably not. depends what the crash is
EruditeHermit: agd5f, you remember this one? http://bugs.freedesktop.org/show_bug.cgi?id=19129
EruditeHermit: I don't know if its useful information that it doesn't crash at low resolutions
agd5f: EruditeHermit: might as well mention it
nanonyme: Yay, my first succesful Linus' tree compile after 2.6.31 release. :)
soreau: nanonyme: What was the problem prior?
nanonyme: I don't remember, some core component failed to compile.
soreau: huh. I've yet to have a kernel fail to compile for me
nanonyme: Probably bad timing on my part.
soreau: Now I'm wondering if fedora uses a different kernel in the daily snapshot live cd than what is in airlied's drm-next tree
kcodyjr: i would kill for a memory map at the moment EruditeHermit's problem occurs
nanonyme: soreau: Probably does.
EruditeHermit: kcodyjr, how do I get that?
legume: What happened to the mesa tree - I am seeing history rearanged :-(
airlied: legume: its not rearranged its just part of the normal insane mesa dev proess
kcodyjr: i'm not even sure there is a way, lockups notwithstanding
airlied: soreau: not really, its drm-next just a few days old
soreau: airlied: That's what I assumed
dileX: nanonyme: guess some staging crap
soreau: airlied: What would be your best guess as to why on my system S-video fails but on the fedora live daily it works right away?
legume: Hmm, it certainly buggers up some going back in time I was about to try.
soreau: airlied: Do they patch any component from git?
airlied: soreau: I'm the fedora maintainer
airlied: the only think that might have happened is a regression in reent drm-next
nanonyme: Yay, this works. Happy dance.
airlied: but I haven't really changed anything in that area sine I wrote it
soreau: airlied: This is pretty suspicious then.. the only difference I could see when on the live cd, it had X1.6.99 and here I have X -version X.Org X Server 18.104.22.168
legume: On a seperate note, is there an obvious reason why mplayer -vo gl* should be OK an kms but borked CPU usage wise non kms (both same kernel and old non kms kernel)?
soreau: airlied: Might this cause a problem?
mzz: stupid question: is there anything interesting crash-prevention-wise for r300 in drm-next relative to vanilla 2.6.31?
soreau: mzz: How are you able to crash on r300?
mzz: soreau: inconsistently.
soreau: That doesn't really answer the 'how' part ;(
mzz: soreau: sometimes I get no video output on resume, very rarely it locks up trying to reset the gpu while I'm doing something in firefox.
soreau: mzz: What resolution?
mzz: soreau: (very rarely used to be about twice a day, but I *think* it got less than that when I disabled agp)
mzz: soreau: 1280x1024
mzz: soreau: only one boring old crt attached
soreau: mzz: cloned?
mzz: soreau: no, only one boring old crt attached.
soreau: oh er..
soreau: I see
EruditeHermit: mzz, is your situation similar to this? http://bugs.freedesktop.org/show_bug.cgi?id=19129
mzz: EruditeHermit: it's not reproducible.
mzz: EruditeHermit: sometimes it'll just lock up while I'm about to navigate to a different page, in the middle of me moving the mouse cursor.
mzz: EruditeHermit: it does seem to frequently happen while I'm doing *something* in firefox, but since the other apps I actually use are emacs and urxvt that doesn't really mean much.
EruditeHermit: I wonder if its related though
mzz: EruditeHermit: haven't read the whole bug report, but what I did is add netconsole, which told me it hangs trying to reset the card
bridgman: I can see other players in Nexuiz now; is that a "known good thing" ?
adamk_: Woohoo... 51.3 fps in openarena (anholt's timedemo) with my HD3450.
adamk_: I wonder how that compares to an x850.
soreau: sees bridgman
EruditeHermit: adamk_, my rv350 gets 124.2 FPS
legume: bridgman: Always handy :-) Last time I tried they were shadows, but were lit by dynamic lights.
EruditeHermit: adamk_, its 3 generations behind your card =p
adamk_: Yeah, I'm expected the x850 to do much better :-)
adamk_: I have an x1900 that I can test with, too :-)
kcodyjr: EruditeHermit, what total feature set are you using, and how much ram is on that 9600?
airlied: soreau: not at fbon
soreau: airlied: Sorry, say what?
EruditeHermit: kcodyjr, 128mb vram, 2GB system ram, 2GHz Pentium M single core
soreau: oh.. I meant just after X starts
airlied: soreau: if tv-out works or doesn't work before X starts won't be affected by the X server
legume: adamk_: better than me - I guess I an CPU limited getting 44 on 3850 (everything set high 1280x1024)
bridgman: x850 is a three generation old 16 pipe chip, 3450 is a current 4 pipe chip, my money is on the x850 ;)
soreau: airlied: I'd be more apt to test again if it didn't take forever to load that live cd
EruditeHermit: bridgman, what pipe are you referring to?
bridgman: do we think it was the addition of draw_prims that made players visible in Nexuiz ?
bridgman: I hate the idea of having to regress something that works now ;)
adamk_: Yeah, I'm sure even with drivers as developed as the r300 driver, the HD3450 can't even compare to the x850 or x1900.
airlied: soreau: put it on a a usb stik ;-)
bridgman: pixels per clock
soreau: airlied: I wish my bios supported usb boot
legume: airlied: Are there any other tv-out settings for kms I could try other than tv=0 or 1?
airlied: legume: its =1 by default
bridgman: what the reviewers like to call rops I guess (cb + db)
bridgman: 3450 is 4, 36xx/43xx/45xx/46xx are 8, 38xx/47xx/48xx are 16
bridgman: 58xx is 32 !
legume: airlied: Yea, just wondering about anything else like pal/ntsc, it doesn't work for me currently.
adamk_: So my CRT blanking with KMS happens with the x850 and x1900, but *not* the HD3450.
EruditeHermit: bridgman, congrats on 58xx
soreau: legume: What distro and what doesnt work exactly?
EruditeHermit: bridgman, looks amazing
nanonyme: And yay, moving to newest from Linus' tree also solved all of my problems with sound card. :)
legume: soreau: LFS and only kms + tv doesn't work.
nanonyme: Looks like 2.6.32 will be an overall win.
bridgman: yeah, it seems to work pretty well
EruditeHermit: bridgman, have you played with one yet?
bridgman: best part is >2 screens without needing shatter ;)
soreau: legume: Elaborate on doesnt work. What have you tried that is not working?
bridgman: I was playing with the 24 screen demo rig
bridgman: before we shipped it down to LA
nanonyme: bridgman: Doesn't it take a bit of of an unoptimal clunk of memory? ;)
airlied: adamk_: hmm its iwerd, I'm leaning towards memory bw calcs doing it wrong
tsamolotoff: Excuse me guys, but are xf86-videos prone to the latest glibc bug?
bridgman: nanonyme; only if you crank up the AA
legume: soreau: s-video tv when kms is started - it works with modeset=0
EruditeHermit: bridgman, is monday release to public?
bridgman: it's already released and in stores AFAIK
EruditeHermit: oh really?
soreau: legume: Have you tried with xrandr after X starts?
bridgman: yeah; next week is 5850 I think
adamk_: airlied, You know... With the x1900 I have had problems in X in the past where I had to do some sort of coherent thing with xrandr for this LCD... Related?
EruditeHermit: I can't afford 5870
bridgman: neither can I ;)
adamk_: airlied, And, in fact, I just booted up with KMS disabled and the LCD goes to sleep when X starts.
EruditeHermit: but 5850 if its near $200
EruditeHermit: is a possibility
legume: soreau: with kms xrandr doesn't list DIN output
bridgman: nanonyme; 8MB per screen for cb, another 8 for db, so 6 screens is <100MB
airlied: legume: you using drm-next?
soreau: legume: DIN?
legume: airlied: yes
airlied: wierd does it show up in dmesg?
adamk_: airlied, I have to go pick up dinner now, but I'll try and track down that xrandr command from before to see if it improves the situation with the x850 once X starts up.
legume: soreau: HD3850 cards have 7 pin DIN connectors.
nanonyme: bridgman: Yeah but my point was that without shatter it ought to be one *ginormous* block of memory, right?
legume: airlied: yes - [drm] TV-12: set mode 1e with console and [drm] TV-12: set mode 1024x768 23 after startx
EruditeHermit: bridgman, also will fglrx support it already?
nanonyme: Instead of one per display.
legume: airlied: It didn't work with non kms until your recentish ddx commits.
soreau: legume: Well I have the same problem but slightly different on my rv350. In short, tv-out doesnt work in kms but works with nomodeset
airlied: legume: I thin I ported all those fixes to kms
legume: airlied: with non kms I can and do specify pal, maybe it's trying ntsc or something - but maybe not as it just cuts the output when I modprobe.
soreau: hmm.. I wonder how to check if its trying to do pal when it should be doing ntsc here..
soreau: of course I get no output at all, not even a flinch on the tele
legume: soreau: but TV works before x / kms - you can see the boot?
nanonyme: yays at suokko for fixing r600 KMS scissors :3
soreau: legume: Well I have radeon compiled into the kernel, so it works until kms happens which is the bios and grub screen, then about half a second when the kernel boots
legume: seconds that yay.
legume: soreau: sounds the same as me then.
soreau: legume: Now you have me thinking it may be trying to set pal when it should be ntsc and vice versa
soreau: legume: Does xrandr list s-video as being disconnected?
agd5f: bridgman: draw_prims is disabled for now. the hidden objects think in nexuiz was a nexuiz bug IIRC
legume: soreau: I don't think you can set it with kms - with non kms it's set in xorg.conf... & xrandr + kms doesn't even list my tv.
suokko: nha, What about silencing warnings for piglit with RADEON_SILENT environment variable or something mesa wide?
legume: soreau: misread what you said there I thought you wanted to set pal.
soreau: legume: Here I use xrandr for nonkms and xrandr still works for the main monitor on kms. It lists svideo and I can assign modes to it but never says connected when it should
soreau: (also it doesnt work obviously)
agd5f: soreau: you can still force it on even if it doesn't say connected
DanaG: oh yeah, how do you switch outputs under KMS?
soreau: agd5f: You never told me how to do that
agd5f: tv dac load detection has always been flakey which is why it's not on by default in the ddx
agd5f: soreau: xrandr --newmode 800x600 38.25 800 832 912 1024 600 603 607 624 -hsync +vsync
Zajec: nanonyme: what are scissors in Mesa meaning?
Zajec: was wondering for some time...
suokko: Zajec: cliping rectngles
agd5f: xrandr --addmode S-video 800x600
suokko: clipping rectangles even :)
Zajec: suokko: and... purpose of clipping? :)
agd5f: xrandr --output S-video --mode 800x600
agd5f: Zajec: clip rendering to a certain rectangle
suokko: Zajec: OpenGL feature for user to render only to limited area of scene
Zajec: ah, ok
airlied: ah I think I need to do something for r600 tv
airlied: looks like we skip it still
Zajec: smarter every day... drop in the sea but who cares ;)
suokko: But I guess FBOs have reduced need for scissors a lot
airlied: legume: able to try a patch?
legume: airlied: sure
soreau: agd5f: That does not work. It adds the mode and puts an '*' next to it, but nothing is ever displayed on the tv
airlied: legume: http://people.freedesktop.org/~airlied/tmp/r600_tv_enable.diff
agd5f: soreau: then I guess there are other problems
soreau: agd5f: Obviously
soreau: agd5f: The weird thing is .. hm
soreau: legume: Which version of X are you using?
airlied: soreau: legume has a very different tv-out than you
nanonyme: suokko: Scissors go clip-clap and cut rectangular shapes out of virtual graphics paper? :3
soreau: I guess that wouldn't even matter since it's supposed to work as soon as kms driver is loaded
soreau: airlied: ok
agd5f: soreau: you could try hacking the drm to always assume the tv is connected
agd5f: soreau: what other monitor is connected to your card?
soreau: agd5f: VGA
airlied: soreau: did you try with the vga out?
soreau: agd5f: I did that setting found = true in the first declaration
airlied: but since it works in ums shouldn't be an isue
agd5f: soreau: if you are using the same DAC for vga as for tv, it probably won't work
soreau: agd5f: All's it did was cause compiz zoom to not zzom an 800x600 square. Was weird
soreau: airlied: I don't follow you exactly, sorry..
soreau: agd5f: It works with nonkms as well as kms on the F12 daily live cd
airlied: soreau: I'll see if I can work out the commint in F12
agd5f: soreau: ok, then it's using the other dac
airlied: so you can see if we regressed
soreau: agd5f: So now I am trying to narrow down what's the difference between that F12 setup and the one I have installed here
soreau: So far the only thing I have found is that X versions are different and the kernel has a different name :p
adamk_: OK, with KMS disabled on the x1900, this same LCD blanks with the x1900 at 1280x1024 but is fine at 1024x768. I'm fairly certain that "--set coherent_mode 1" got this working before with this monitor and this GPU, but it's not doing it now.
adamk_: Wow... This LCD has all sorts of problems.
agd5f: soreau: compare reg dumps between livecd and drm-next with tv forced on as I described
soreau: agd5f: I have done that. They are different. Lemme try and find the pastebin
agd5f: soreau: with tv forced on?
airlied: soreau: 5d93b135fc051be3740cbe55729fc7b86a7069d7
airlied: is the commitid where F12 is using at the moment
airlied: it might be worth checking that out and testing it
soreau: agd5f: Yes, with ums and kms forced on (the * next to the mode for svideo)
soreau: airlied: Kernel.. ok
airlied: adamk_: under kms try "coherent mode"
adamk_: I'll give it a shot.
soreau: agd5f: http://pastebin.com/m7fa4e224
agd5f: soreau: does setting them as they are set under ums help? e.g., radeontool regset RADEON_TV_DAC_CNTL 0x00770103
legume: airlied: Still doesn't work.
adamk_: airlied, 'xrandr --output DVI-0 --mode 1280x1024 --set coherent_mode 1' under KMS results in 'X Error of failed request: BadName (named color or font does not exist)
airlied: legume: paste a dmesg
airlied: adamk_: hence "coherent mode" note no _
adamk_: Wait... so just '--set coherent mode' ?
legume: airlied: It does show on xrandr now - as disconnected.
soreau: agd5f: Nope.. trying to read the regs now to see if they really set
legume: airlied: You need all 3k line dmesg or just the last bit?
soreau: agd5f: Seems it only set one of them out of the three I tried to set
airlied: legume: radeon/drm bits
airlied: adamk_: you made need quotes around the coherent mode
agd5f: soreau: what do you mean?
soreau: agd5f: And something in syslog http://pastebin.com/m7a02f625
adamk_: airlied, Sorry, but no go... 'xrandr --output DVI-0 --mode 1280x1024 --set "coherent mode"' just gives me the xranrd help.
airlied: still need the true
airlied: or 1
adamk_: airlied, Both true and 1 give me the same BadName error.
agd5f: soreau: enable the tv under kms using xrandr as I said. then use radeontool to write the ums values to the regs in question
soreau: agd5f: What I did was radeontool RADEON_TV_DAC_CNTL 0x00770103 for the three regs that were different, then I read them right back with radeontool regmatch 0xffff or whatever
agd5f: soreau: radeontool regset
soreau: agd5f: Bah, that's what I meant
soreau: agd5f: And s-video is forced on http://pastebin.com/m7ff6db01
soreau: agd5f: So anyway, I use regset to set the regs but when I read them back with regmatch, they're not set to the values I gave
soreau: waits for the kernel to compile
agd5f: soreau: what are they set to? BTW, I don;t think radeontool uses the RADEON_ prefix for the regnames
agd5f: radeontool regset TV_DAC_CNTL 0x00770103
soreau: Hmm.. ok
soreau: Oh boy, wecanhas output
legume: airlied: http://www.pastebin.ca/1578353
soreau: agd5f: Well that at least did _something_
soreau: agd5f: Now I get some output from the tv, some blocks though it's a stable picture
agd5f: soreau: try the rest of the potentially interesting regs
soreau: Will it hurt or is it not a good idea to use xrandr as root?
agd5f: then work backwards to figure out way those regs aren't set properly in the drm
agd5f: soreau: it's fine
chithead: you may have to set the DISPLAY variable properly
soreau: agd5f: I'm not sure right off hand where I should find the other interesting regs
soreau: chithead: Well xrandr works as root from a terminal within X, but I just wanted to make sure I'm not screwing myself
soreau: chithead: Either way, I'm sure you'd have to set DISPLAY from tty or ssh
agd5f: soreau: that list you had
agd5f: in the pastebin
soreau: agd5f: But the rest are.. ok
agd5f: soreau: ok
soreau: set the three that were different already
agd5f: soreau: so which reg(s) are the problematic one(s)?
soreau: Oh wait.. it is working because I see my mouse.. on the tv
soreau: agd5f: As soon as I set TV_DAC_CNTL, it started working
octoploid: I just hit: "check_context_limits: Assertion `ctx->Const.MaxTextureCoordUnits <= ctx->Const.MaxTextureImageUnits' failed"
octoploid: This error occured everytime I ran glxinfo or compiz. With compiz it would hang my machine.
octoploid: After googling for a while I found out that rm ~/.drirc solved the problem.
octoploid: Should a simple user config file be able to hang a machine?
soreau: agd5f: And when I use xrandr to set my main monitor resolution, it resets that reg to the faulty value
soreau: agd5f: But it's like it clones the upper left 800x600 square of my primary 1280x1024 display but only shows gray with two thick black lines though I can see the mouse cursor on it clearly
soreau: testing the kernel from the commit airlied gave, brb
agd5f: soreau: looks like the crtc routing is reversed between UMS and KMS
spstarr: validated 0x6456930 [0xF0366000, 0xF0367000]
spstarr: bo(0x770a010, 1048576) is mapped (-8) can't valide it.
spstarr: validated 0x770a010 [0xF06E2000, 0xF0707000]
spstarr: bo(0x6456930, 98304) is mapped (1) can't valide it.
spstarr: validated 0x6456930 [0xD4EA2000, 0xD4EA6000]
spstarr: snowglobe-do-not-run-directly: radeon_texture.c:87: radeonFreeTexImageData: Assertion `!image->base.Data' failed.
spstarr: it is failing badly
soreau: airlied: Aaaaannnndd... we have a good commit
spstarr: do not give 512MB texture mem?
agd5f: soreau: now you can bisect :)
soreau: agd5f: Indeed, but..
spstarr: i cannot run snowglobe viewer from second life
spstarr: that assertion
spstarr: is bad
spreeuw: spstarr: got those too in another game\
spstarr: tries 128MB textures only
soreau: airlied: But imagine this: When it begins loading, 800x600 of the main screen is showing the boot up while the tv is at 'normal' 800x600 resolution. Then when X starts, the main monitor is 1280x1024, and the tv only shows 800x600 part of the cloned screen
soreau: tries --right-of VGA-0
spstarr: trying 128mB texture size
spstarr: o(0x77ab030, 262144) is mapped (1) can't valide it.
spstarr: validated 0x77ab030 [0xF0472000, 0xF0473000]
spstarr: File radeon_dma.c function radeonReleaseDmaRegions line 348
spstarr: Leaking dma buffer object!
spstarr: man badness all over :)
soreau: That's about as good as it gets
spstarr: hides and returns to his intel GMA for now :(
soreau: smacks spstarr with a pastebin service
spstarr: valide? :)
spstarr: validate maybe
soreau: Now I don't want to screw this up so..
soreau: agd5f: airlied: So I am on the good commit. How should I start the bisect? git bisect master 5d93b135fc051be3740cbe55729fc7b86a7069d7 ??
soreau: I don't want to waste time by screwing up :)
airlied: soreau: git bisect start
airlied: git bisect good
airlied: git bisect bad
soreau: airlied: Can bad commit be master or do I have to specify your branch?
spstarr: re booting...
airlied: soreau: master should work or drm-next
rnoland: does master resolve to master/HEAD?
soreau: hopes so
soreau: Hi rnoland :)
agd5f: soreau: f066a17d9f8d0a20d01d1aa9badce7f43c7bd6ad is the last commit in drm-next
rnoland: i would just checkout the right branch and then use HEAD for the bad commit...
nanonyme: suokko: And yeah, after that fix of yours, the scissor update function started looking sensible to me for the first time. ;)
soreau: agd5f: I already did bad master and it said: Bisecting: a merge base must be tested [3ff323f89075624b6891e7c428edb8e8a35be13c] Merge branch 'drm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/airlied/drm-2.6
soreau: rnoland: I did git bisect bad master
soreau: rnoland: And that was the output^^
legume: Should an AGP card use 32 bit dma if in PCIE mode or is 40 OK then?
rnoland: soreau: yeah, not sure... but master is a different branch from drm-next
soreau: rnoland: (I am currently on the good branch but went ahead and specified git bisect good
soreau: rnoland: Hmm
soreau: decides to reset
rnoland: if your talking about airlied's tree, then master should be linus tree
agd5f: legume: onboard gart suppors 40 bits
spstarr: back into Intel GMA :(
soreau: rnoland: Yes, drm-next
soreau: I must have started using drm-next *right* after this broke so I never knew it worked
legume: agd5f: OK, thanks.
rnoland: soreau: so checout drm-next then git bisect start HEAD
soreau: rnoland: But I thought I am already on drm-next...
soreau: rnoland: Followed whatever's in the topic link
rnoland: soreau: make sure with git status
rnoland: or git branch actually
agd5f: soreau: reset to the current head rather if you checked out a particular commit
soreau: agd5f: ok
soreau: rnoland: Yea, I checked out a commit so I am not currently on any branch
spstarr: at least, i can test radeon now
rnoland: soreau: so checkout drm-next then start the bisect with HEAD
spstarr: I will try again periodically to see
soreau: Alright. First I need to jot down some notes about radeontool since this is the first time I've ever got it to do anything useful for me ;)
soreau: agd5f: What is the reg header again please?
rnoland: needs to go cut grass....
agd5f: soreau: reg header?
soreau: agd5f: Sorry, the radeon register header file.h
agd5f: soreau: radeon_reg.h
soreau: Thank you
adamk_: So I get the same BadName errors from xrandr when using KMS or not. And the LCD doesn't work in Xorg with the x1900 with KMS disabled, nor does it work in console or Xorg with KMS enabled.
spstarr: oh, with KMS on screen goes blank with r6xx
adamk_: Which, as I recall, is the reason I've been using the x850 with this LCD for the past few months instead of the x1900. Unfortunately, with KMS, the LCD is unusable even with the x850.
agd5f: adamk_: does it work if you swap ports? or boot with only the lcd?
adamk_: With the x850, if I boot with only the LCD it works fine with KMS.
adamk_: Let me confirm that with the x1900.
adamk_: Hmmm... With the x1900, with only the LCD + KMS, it still stays off.
adamk_: You know what.. I think I need to stick with the x850 and wait till this LCD works with that card before I test it with the x1900... I think there may be different problems for each card..
adamk_: So let me switch back to the x850, test just the LCD with KMS.
soreau: There is no drm-next branch listed by git branch. So I switched to the radeon_kms branch and did 'git bisect start HEAD
agd5f: soreau: you're in the kernel tree right?
soreau: agd5f: Yes, linux-2.6.31.y
agd5f: soreau: airlied's tree or something else?
adamk_: As I recall, when I swapped ports with the x850, the monitors were stuck in clone mode and no amount of --right-of or --left-of did any good, but let me confirm that as well.
adamk_: For now the x1900 goes back to work with me where I know it works with both monitors.
soreau: agd5f: It is from whatever's in the topic link. Should be airlied's drm-next tree
soreau: Right now I'm sitting at this: Bisecting: 27 revisions left to test after this (roughly 5 steps) [e87b2c42b397ae64ac13a390011931f7ef222321] drm: fix drm_fb_helper handling of kernel crtcs
agd5f: soreau: oh, you merged airlied's tree into 2.6.31
soreau: agd5f: Isn't that what it says to do there in the topic link?
adamk_: agd5f, OK, with the x850, with KMS and *just* the LCD, the LCD stays awake and receives a signal just fine.
agd5f: soreau: apparently.
soreau: agd5f: So maybe that's why there's no drm-next branch.. because it's already merged in?
agd5f: soreau: ok. so for those instructions, radeon_kms is correct
soreau: agd5f: Alright, thanks
adamk_: Just damn...
adamk_: agd5f, Sorry, but this time, when I switched ports, both monitors stayed active when KMS kicked in.
adamk_: So, in other words, it worked.
soreau: adamk_: intermittent.. that's horrible
adamk_: But I know that before I opened up that report at fdo, I tested this and the monitors stayed in clone mode.
soreau: Especially since you're juggling cards there ;)
adamk_: You know...
adamk_: I did update from 2.6.31 to drm-next since opening that bug report.
agd5f: adamk_: might have been fixed in drm-next
adamk_: Any thoughts about why switching ports made the difference? Not that I'm not happy that it's working :-)
adamk_: And are there any plans on making it possible to configure the console resolution with KMS? Because this bug is really annoying: https://bugs.freedesktop.org/show_bug.cgi?id=24124 (And, frankly, I'd rather run that monitor at 1280x1024 anyway).
soreau: Would git bisect work if you trying to figure out what fixed something? (ie. somewhat in reverse) Like git bisect
soreau: put a start in there somewhere
mzz: soreau: I'd expect it to, although you might have to switch the meaning of "bad" and "good"
soreau: mzz: I just found out.. you are correct sir ;)
mzz: (the concepts behind git things like this one I tend to grasp, I just don't know the commandline ui specifics)
mzz: there might be a way to start a bisect like that where it does that switch for you, for example
agd5f: adamk_: are you sure it was the switching of ports that fixed it and not just the fact that you are using newer code?
adamk_: agd5f, I updated the code a few days ago. Didn't switch the ports then, and the LCD still stayed off with KMS enabled. Switched ports tonight, (but haven't updated code since a few days ago) and the LCD stayed on.
adamk_: I can power off in a bit and switch ports to double check...
soreau: Now I have kernel compile fail: http://pastebin.com/m1a235bec Should I just skip it?
adamk_: agd5f, Alright, I just confirmed it... Switched the ports back and the LCD turns off... So the code update a few days ago may have partially fixed it, but I wouldn't have known that if I hadn't switched ports :-)
nanonyme: Sounds odd...
agd5f: adamk_: can you provide your dmesg and xorg log for both cases?
adamk_: Sure, give me a few minutes.
agd5f: adamk_: is either monitor DVI or are both VGA?
adamk_: There was some flag for radeon.modeset that enabled more verbose debugging information.. Does anyone remember what it is? :-)
adamk_: agd5f, The LCD is DVI, the CRT is VGA (with a VGA->DVI dongle).
agd5f: adamk_: I suspect there's an issue bringing up the external tmds encoder
agd5f: on your board
agd5f: since pre-r5xx boards only had one tmds encoder
adamk_: agd5f, Xorg without it working: http://pastebin.com/m9d9bb12 dmesg without it working: http://pastebin.com/m301f66f6
agd5f: adamk_ : non-kms works fine on either port?
adamk_: Yes, both monitors work fine in Xorg no matter the port they are plugged into.
adamk_: I'm rebooting with the ports switched and KMS enabled.
agd5f: adamk_: I've got a similar card. I'll pop it in tomorrow and see if I can repoduce it
agd5f: used to work, but I haven't tried it in a while
adamk_: Xorg.0.log http://pastebin.com/m70bb1d54 dmesg: http://pastebin.com/m47b23807
adamk_: Those are with it working after swapping ports.
adamk_: Honestly, as long as it works one way, I'm pretty happy :-)
adamk_: I'd say the flickering at the default resolution is a bigger issue for me now :-)
adamk_: Alright, I need to run out for a bit.
agd5f: adamk_: try loading radeon with r4xx_atom=1
agd5f: when you get a chance
adamk_: If I have radeon built into the kernel, can I pass that as radeon.r4xx_atom=1 on the kernel line?
MostAwesomeDude: airlied: Are we going to support the pageflip DRI2 ioctl?
airlied: MostAwesomeDude: whenever it gets written
MostAwesomeDude: airlied: Ah. Do we know enough to be able to do that?
airlied: well awaiting krh/jbarnes final version for intel before we tyouch it
airlied: otherwise we'd have to keep up with them
airlied: but once they are done I'll definitely look at it
MostAwesomeDude: airlied: They're talkin' about it right now. :3
airlied: this rv370 laptop is fubar, booting from HDD it'll stay one a while, connect USB HDD, falls over, bad earthing somewhere
adamk_: agd5f, With that option, the boot hangs.
adamk_: agd5f, From the serial console: http://pastebin.com/m221c818c
soreau: airlied: Bisect complete http://pastebin.com/m50ed2b8b
airlied: soreau: oh nice one
airlied: thanks for that I wonder what I screwed up there
airlied: oops forgot to set dac load detect
soreau: agd5f: Remember I was complaining about that 'static/noise' on the right part of my screen and you said something needed to be ported to kms but instead I ended up just removing rdev->disp_priority == 2 from r100.c line 2338? What was it again that needed to be ported to kms?
airlied: I suck
soreau: airlied: No you don't ;)
soreau: airlied: But did you read my message earlier about the way it behaves? When tv-out is not working, the boot is at 1280x1024. But with tv-out working, tv is at 800x600 and the main monitor is 1280x1024 but displays the output in the upper left 800x600 portion of the screen, cramming it all up there
airlied: soreau: thats how it works in fbcon
soreau: airlied: After X starts, it behaves more sanely (I guess) by having the main monitor at 1280x1024 (and filling it fully) while the tv only shows the upper 800x600 portion of the main screen
airlied: fbcon tries to show on as many screens as it can
airlied: so people can see something
soreau: So is there a way to disable S-video at first while being able to use xrandr to enable it later?
airlied: not just yet, we'll have a command line option in drm-next soon
airlied: soreau: oops try http://people.freedesktop.org/~airlied/radeon_load_detect.patch
soreau: airlied: From this commit? (I'm thinking this is latest) drm/edid: Fix standard timing parse for EDID <= 1.2
airlied: soreau: on top of drm-next
airlied: yeah that sounds like it
soreau: airlied: This gave no output.. was it supposed to? linux-2.6.31.y $ git apply ../radeon_load_detect.patch (sorry for noob questions)
airlied: generally I'd just patch -p1 from the top of the tree
soreau: just checks the files
soreau: I guess git apply worked
soreau: airlied: Can you answer the last question I directed to agd5f perhaps? I wanted to take a look if I can port it
edt: soreau no output if git apply works is normal
soreau: edt: Ok, thanks
edt: do not forget to go commit -a
soreau: I usually use something like patch -i
DanaG: oh yeah, if I boot with one display, and then hotplug another.... how does KMS deal with that?
soreau: edt: eh?
edt: git commit -a
soreau: edt: What does that do?
airlied: DanaG: same as non-kms, does nothing
soreau: I'm just testing the code really..
edt: makes sure git know about the change and creates a log entry
edt: so git log will show the change
DanaG: hmm, I should try my output-switch hotkey under uvesafb.... I seem to recall that it did work.
soreau: edt: Maybe that's why it appends -dirty (because I don't do commit -a)
edt: no thats a kernel config
edt: got me too till I found it
edt: no commit -a will mean git complains when you switch branches
edt: commit -a will show you what files you have changed in your version of the tree
edt: If you just quite the vi session thats all it does, If you insert a comment and save it commits the change to your tree (in what ever branch you have checkedout)
soreau: airlied: Yup, that worked
airlied: soreau: \o/
soreau: airlied: Though it might be a good idea to make s-video off by default since it does several odd things until X completely starts
airlied: soreau: its not odd
airlied: kms doesn't care about X
edt: soreau see: Automatically append version information to the version string
edt: for the dirty stuff
airlied: so has no way to know if X starts etc
soreau: airlied: For instance, when X first displays anything, it shows 1280x1024 on the main screen and a bunch of crazy scan lines in the upper 800x600 portion as well as for all of the 800x600 of the tv
soreau: airlied: After it finally starts 'fully' things are 'normal'
airlied: does it happen with X -retro with no desktop env?
airlied: the resizing of things is whats messing it up
soreau: airlied: Yea, that's xfce messing with it. X -retro starts very cleanly
virtuald: resizes airlied
soreau: But even with --output S-video --off, the tty is still 800x600 in the corner of a black 1280x1024 screen (sorry to be redundant)
soreau: I guess it will be like that no matter what if svideo is plugged in
airlied: the console ignores what X does
soreau: Yea, I keep having trouble distinguishing/wrapping my head around it :)
airlied: you can't display the whole console on the main monitor since then the tvone is truncated
soreau: And I can't unload the radeon module since I compiled it in..
soreau: oh well. At least it's working
soreau: airlied: Thanks :)
airlied: we need a redtect for fbcon alright
soreau: airlied: Now can you say what needs to be ported to KMS to fix the noise on the right side of the screen when 3D happens? fixed by removing rdev->disp_priority == 2 from r100.c line 2338
[Enrico]: do i need git drm to be able to try experimental 3d for hd cards?
airlied: soreau: I tihnk thats the disp priroity option
airlied: in ums we have an xorg.conf option
airlied: we'd need a kernel module option in kms
soreau: oh :p
[Enrico]: oh nvm ;D
legume: What are read domain reloc failure?
soreau: [Enrico]: Short answer, yes
[Enrico]: soreau: yeah i found it in the buildhowto (i didn't found it before, i'm a bit blind lol )
airlied: legume: bad things I just saw some myself
[Enrico]: soreau: btw thanks :D
[Enrico]: btw (if i can disturbe you a bit) what's the current status of 3d for r600 cards? usable? unstable? even more unstable? :D
legume: airlied: Oh, I've got lots in vt1 doesn't look like they are getting logged though.
airlied: legume: yeah I just noticed them, I'll fix that it sounds like a bad thing
edt: [Enrico] unstable but not unusable (nonkms) with kms I last from minutes to hours before I have to reboot
[Enrico]: edt: lol ok thanks :D
edt: hey I am sticking with the new stuff
[Enrico]: edt: well i can wait for kms, but may be a i can take a look to the new shiny 3d stuff :F
edt: I can live it unstable for a few weeks if the end result is 3d
edt: actually, until it crashes, some apps, like googleearth, work better with kms...
edt: muses that working 3d is more important than fast unstable 3d
mjr: would also opine that fuller features outweigh raw speed, but ymmv
biotube: I just want extra screens to stop running slower with DRI eneabled on the first than with it completely disabled
_moep_: its funny
airlied: legume: looks harmless, though might cause some over zealous flushing
legume: airlied: Ahh good
airlied: now I fix that I get a write domain one which isn't so harmless ;-
_moep_: i'm testing with a ubuntulive cd cause my silly radeon bug
_moep_: and the normal boot I here only the sound and get a black screen, like my desktop
legume: Oh, haven't had one of those yet
_moep_: but on the 2th boot option i can login, but I'm not sure, which driver I'm using
_moep_: lsmod | grep radeon show me nothing
legume: Hmm scrollback with fbcon doesn't seem to work very well.
_moep_: airlied: I updated bug #22562
airlied: just pushing some EXA r600 space checking fixes, let me know if it breaks anyone
soreau: Is there a page where I can see all the drm-next commits? Like cgit or so?
airlied: git.kernel.org dig around
soreau: The closest I can find is http://git.kernel.org/?p=linux/kernel/git/airlied/drm-2.6.git;a=log;h=drm-next and it doesn't show latest commits afaict
soreau: airlied: There is a bug now with compiz ezoom. When zooming in, the top 800x600 portion of the screen doesn't zoom while the rest of the screen does. Almost looks as if the rest of the screen zooms 'under' the top 800x600 corner
soreau: I know it happens when bool found = true regardless if tv-out is working because I tried setting found to true instead of false trying to get svideo working out of desperation
airlied: soreau: it should have them
airlied: I don't push anywhere else
soreau: Well maybe I'm looking in the wrong place or it's really slow to update
airlied: well I haven't pushed anything since yesterday
soreau: That would make sense then I guess