kristian_: A recent commit made my r300 go all blank with a few 8-10 white stripes across the screen.
kristian_: I did a git bisect and found the commit that broke it.
kristian_: [965a5dbcd9dc4bf1cdd7f2bbdec15e9733b2e090] RADEON: improve support for secondary cards
agd5f: kristian_: is it a secondary card? are you using bsd or linux?
agd5f: do you have your xorg log?
kristian_: agd5f: I only have one integrated card in this laptop, I'm running ubuntu gnu/linux
agd5f: kristian_: can you pastebin your xorg log?
agd5f: also does this patch fix it? http://www.botchco.com/alex/xorg/stefanb.diff
kristian_: agd5f: http://picard.hgo.se/~kristian/Xorg.0.log
agd5f: kristian_: ah ok
agd5f: I see the problem
kristian_: agd5f: Do you still want me to try the patch?
agd5f: no, not relevent for your card
kristian_: agd5f: Do you wan't me to stick around and maybe try a patch later on?
agd5f: yeah, one sec
agd5f: kristian_: try this patch: http://www.botchco.com/alex/xorg/no_legacy_init.diff
kristian_: agd5f: yes this seems to work fine
agd5f: kristian_: good. thanks!
kristian_: agd5f: thanks for the quick help!
agd5f: np :)
MostAwesomeDude: rx__: I thought /quit was left.
Zajec: who can I torture for rebasing valgrind-mmt to glibc 2.8? :)
rx__: fails at logic
Zajec: rx__: logic of driver or logic of my sentence? ;)
rx__: uh neither
rx__: was referring to something else :)
stefan1: he he,airlied pushed r5xx mesa stuff already to Fedora koji as RPM :-)
rx__: that's good news
z3ro_: doh! stupid bug with strtol. should have used strtoul. =/
stefan1: agd5f: Still BIOS read problems with latest git
stefan1: agd5f: Looks like the latest change is only for secondary cards
onestone: MrCooper: ping
MrCooper: onestone: yes, I'm going to push your Mesa fixes :)
onestone: ok thx
MrCooper: or do you have more than the two I've seen?
onestone: I have some other bugs but I don't think that I will be able to fix them
MrCooper: what are they?
onestone: 1. performance drop after suspend 2. vblank hang 3. mplayer hangs X in fullscreen xv without a comp mgr running <- everything r500 related
adamk: agd5f, FYI, after applying that patch and trying the agp x1300 again, X never actually comes up on the display.
adamk: It doesn't crash, but it sits there consuming 100% in top till I kill -9 the process.
adamk: agd5f, So I went back to the agp x700 I have in that box... The latest git code (xserver/drm/mesa/ati), and nearly everything works fine, but compiz looks like this: http://188.8.131.52/Screenshot.png
adamk: I'm going to try using git versions of drm/mesa/ati and switch back to the xserver that ships with F9 (1.5.0 RC 1) to see if I still have the same problem.
peterz: adamk: wouldn't be because you use gnome, right? :-)
peterz: darn, my bad ;-)
adamk: And I doubt that my DE has to do with all that distortion :-)
peterz: shuts up again
agd5f: adamk: are you on BSD by any chance?
adamk: agd5f, Not at the moment, though I do have it installed. With FreeBSD, I stick to the ports tree :-) This is Fedora 9.
agd5f: adamk: ok
adamk: So latest git drm/mesa + ati 6.8.0-14 + xserver 1.5.0 rc1 work fine with compiz.
adamk: So then the problem with the distortion is from either the X server or the 2D driver (or some combination).
adamk: Alright, so even after updating to the ati driver in git, it still works. So it's something to do with the x server.
AMF|DJ-N: how can i compile mesa from latest git? everytime i restart the xserver i see that i only have mesa 7.0.3-rc2
AMF|DJ-N: OpenGL vendor string: Mesa project: www.mesa3d.org
AMF|DJ-N: OpenGL renderer string: Mesa GLX Indirect
AMF|DJ-N: OpenGL version string: 1.4 (2.1 Mesa 7.0.3-rc2)
AMF|DJ-N: i think i do something wrong everytime
adamk: AMF|DJ-N, Use git got grab the latest code and recompile... :-)
AMF|DJ-N: thats exactly what i did
adamk: And did you install it after compiling? Are you sure that glxinfo is linking to the new library?
AMF|DJ-N: i compiled the drm with ./autogen.sh; ./configure --prefix=/usr; make; sudo make install
AMF|DJ-N: then i compiled mesa with make linux-dri-x86
AMF|DJ-N: but how can i install the new mesa?
adamk: make install
adamk: Or 'sudo make install' if you're not root.
agd5f: AMF|DJ-N: you can use autogen with mesa from git as well
agd5f: otherwise you have to ajust the MESA_INSTALL_DIR var (I foget the exact name)
AMF|DJ-N: if i make ./autogen.sh --prefix=/usr then i can't compile with make linux-dri-x86
AMF|DJ-N: it says that i have to clean it with make realclean
agd5f: AMF|DJ-N: if you use autogen, then just make and make install
AMF|DJ-N: ah, i give it a try :)
agd5f: AMF|DJ-N: e.g.: ./autgen.sh --prefix=/usr --with-driver=dri --with-dri-driverdir=/usr/lib/dri --with-demos=no
AMF|DJ-N: thanks :) i try it
stefanb: agd5f: I tried the latest -ati git. Still can't read the BIOS. But I guess the last git commit was for secondary cards only?
agd5f: stefanb: are you on linux or bsd?
stefanb: agd5f: Linux, Fedora 9, 184.108.40.206
agd5f: stefanb: I haven't fulled addressed the issue yet. I'm not sure why it's happening to you and adamk. I can't reporduce it here
stefanb: agd5f: If I use your hack patch, it loads int10 and then it can read it
agd5f: the problem is, if I enable int10 secondary atom cards don't work
stefanb: agd5f: Ahh.
agd5f: which I guess is better than breaking primaries
mattst88: adamk is the master of finding bizarre hardware problems
agd5f: in your case
stefanb: mattst88: I'm also known to have those "magic hands" sometimes :-)
adamk: Alright, I'm not sure what I did this time around, but all that distortion on the x700 is gone.
mattst88: adamk, is this radeon one of the ones used when finding weird things with skyos?
adamk: mattst88, Different radeon, same maachine.
mattst88: I wonder if there's something bizarre about that motherboard.
adamk: I never had any issues on skyos that were resolved by not using the radeon driver on this machine though, iirc.
stefanb: is airlied still awake?
agd5f: stefanb: no, he's in au
stefanb: Well if I wait until midnight again :-)
stefanb: Ok latest mesa compiled, ut2004demo downloading....
AMF|DJ-N: agd5f: thanks very much for your help :) now it works
AMF|DJ-N: warcraft 3: tft works too with this driver version
spstarr_work: adamk: which card do you have?
spstarr_work: i have dri locking issues too
adamk: I have an agp x1300 and a PCIe x1650. X maxes out at 100% with the x1300, and compiz locks up the x1650 (though ut204 worked fine).
adamk: I think there might have been issues with my build that I just got through, though, as my AGP x700 didn't work properly till just an hour ago or so.
adamk: Though I'm not necessarily expecting it to suddenly work now, I'm going to try the x1300 again :-)
onestone: adamk: do you have vblank_mode=0
onestone: adamk: this should fix the compiz locks
adamk: No, it's set to 1...
adamk: Oh, cool.
adamk: I'll give that a shot on the x1650 when I get home.
adamk: Yeah, the AGP x1300 still has the same problem. X consumes every available resource till I kill -9 it and my screen just stays blank.
adamk: I'll assume this is an AGP related issue :-)
MrCooper: adamk: got a backtrace where it's stuck?
adamk: No, but I can try.
adamk: All that shows up in dmesg that jumps out at me is "[drm] writeback test failed"
adamk: It loads the r500 microcode and detects 1 pipe.
adamk: I'll try it in gdb now.
adamk: I've gotten backtraces from gdb when an app segfaults, but I'm not much of an expert.. Is there some way to grab a backtrace with the app still running?
MrCooper: adamk: gdb -p
MrCooper: but the writeback test failing could indeed indicate an AGP issue
MrCooper: does it come up with the DRI disabled?
adamk: Well, with dri and glx disabled. With dri disabled I get some other error regarding swrast, iirc.
adamk: I have Xorg running in gdb, but it's not crashing, so I'm not sure how to get a backtrace.
MrCooper: adamk: you got a prompt when you attached it, didn't you? :)
MrCooper: adamk: Ctrl-C should get you a prompt again
adamk: Oh, I actally ran 'gdb /usr/bin/Xorg' and then told it to run.
adamk: Let me try this again.
MrCooper: Ctrl-C didn't work?
adamk: Nothing... But maybe it's just because the CPUs are maxed out and I didn't wait long enough.
adamk: I'm rebooting again now and will give it another shot, letting it sit for a few minutes after hitting control-c.
MrCooper: adamk: there should be no delay; try attaching gdb once the X server is running, that should give you either a prompt or a gdb error right away
adamk: Yeah, I'm tying gdb -p now... It just takes about 15-20 seconds after each keystroke on the serial console before it shows up. :-)
MrCooper: adamk: why not ssh instead of serial console? :)
adamk: Good question. serial console actually feels a *little* faster than ssh for some reason.
adamk: And gdb is now "Attaching to process 3060"
adamk: It has been for maybe 30 seconds.
adamk: Alright, time to go stretch my legs for a few minutes. Maybe it'll be attached when I get back :-)
MrCooper: adamk: hmm, sounds like the process could be in some weird state where gdb can't interrupt it
adamk: Yeah, it's still trying to attach.
adamk: I'll give it a few more minutes and then reboot
adamk: If you guys figure out what's wrong with AGP support, or simply want me to test something, you know where to find me :-)
onestone: I now have the backtrace of my hanging mplayer. RADEONDownloadFromScreenCP
adamk: onestone, Does it work if you disable DFS?
adamk: In the driver section:
adamk: Option "AccelDFS" "off"
adamk: They sound similar, so I assume it's related :-) It has to do with EXA, though I couldn't say more than that.
adamk: /var/log/Xorg.0.log should tell you if it's using the EXA DownloadFromScreen hook currently.
adamk: As I understand it, it's disabled by default with AGP cards, but enabled by default with PCIe cards... At least that's how it is for r400 cards.
agd5f: onestone: yeah try disabling DFS as adamk said. it's used for accelerated readback of data from the FB
adamk: Oh, and if MrCooper or agd5f is listening :-) My tests with the x1300 are with DFS disabled. DFS works fine with my x700, so I had it enabled in my xorg.conf file, but disabled it for testing the r500 driver.
onestone: agd5f: mplayer fullscreen still hangs the driver
MrCooper: DFS is probably just where it waits for the hardware, which never catches up because it's locked up
adamk: BTW, guys, great job with the r500 support. It was pretty cool playing ut2004 with my x1650 with open source drivers last night.
onestone: but why should mplayer in fullscreen mode try to get the video back to the server?
MrCooper: onestone: that's probably not what's going on; again, the DFS is probably a red herring
onestone: here is the bt http://www.pastebin.ca/1033230
MrCooper: there's only a couple of points where the driver waits for the hardware to catch up, so that's where it'll end up on a lockup
MrCooper: but that doesn't say what caused the lockup
stefan1: spstarr_work: Seems like KDE4 startup doesn't like compiz-manager as window manager. It doesn't get the necessary feedback "hey window manager is up" and therefore waits 30 seconds with the bouncing launch feedback before the user can do something
spstarr_work: i didnt get compiz working right either, but im forgoing all composite until r3xx dri / radeon drm is stabilized
stefan1: spstarr_work: I'm now using a small compiz-wrapper replacement script that first starts kwin and then the real compiz-manager. Now KDE4 startup recognizes that the window manager has started and doesn't wait anymore
MostAwesomeDude: stefan1: XD, nice hax.
stefan1: spstarr_work: I guess the proper correction would be to use some program to give feedback to klauncher
stefan1: cat .kde/env/kdewm.sh
stefan1: $ cat /home/stefanb2/bin/compiz-manager
stefan1: /usr/bin/kwin $* &
stefan1: sleep 0.2
stefan1: exec /usr/bin/compiz-manager $*
stefan1: Crude, but effective
MostAwesomeDude: Well, doesn't the kde-window-decorator talk to klauncher?
MostAwesomeDude: Mwahahaha. Dug my old CRT out of my garage. Let's see if dual-head works.
adamk: MostAwesomeDude, It works here.
adamk: So I imagine it will work there :-)
MostAwesomeDude: adamk: You may be surprised. Can't find power cable...
MostAwesomeDude: Seriously! It's one of those weird ones.
stefan1: MostAwesomeDude: Good question, kde4-window-decorator doesn't seem to...
rx__: oh.. MostAwesomeDude i've got a question
rx__: how do you test these 3d instructions.. is there some program and reference picture?
MostAwesomeDude: rx__: In the mesa tree, progs/fp provides per-inst reference tests.
rx__: ah.. i see
rx__: so this is how you know LIT and such work correctly or not
MostAwesomeDude: rx__: Right.
MostAwesomeDude: ...Dammit, still have to fix LIT.
MostAwesomeDude: I'll do that now.
rx__: well that makes things nice and convenient
rx__: so you just have to program the hw correctly :)
surfer: i like how my x1800 is like 50% faster than michael at phoronix
surfer: must be the faster memory or something
rx__: 50% in what?
surfer: according to his screenshot
eboettcher: DUDE I get 3300fps in gears!!!11!!11 :P
surfer: what card?
eboettcher: surfer: that's irrelevant
surfer: are u scared of society or something
rx__: he gets 3300fps in software ;)
eboettcher: rx__: that would be nice ;P
surfer: ur not normal are you eb
surfer: i get about 5300 fps
eboettcher: if normal is comparing my R423 to your R520 using gears, then no, I'm not normal
eboettcher: I get about 9900fps... with the window occcluded :P
surfer: with the binary drievr i get over 10000 fps
surfer: i have a feeling there's a lot of room for improvement in the open source one
surfer: the funny thing is it still feels faster than that peice of shit
eboettcher: surfer: what do you get say with the demo 'four' in quake3?
surfer: i don't have it installed
Ori_B: surfer: sure, glxgears doesn't test anything especially useful
adamk: Dear God... I never realized there were still people who compared glxgears framerates as if they actually mean something.
mogurakun: i get 160fps on R600. with binary :P
surfer: i guess u probably don't believe in any kind of benchmark what so ever
mogurakun: adamk: +100
surfer: u just go by ur feelings adam?
surfer: i feel like this is fast
surfer: so i'm happy
eboettcher: indeed not
mogurakun: i do believe in benchmarks, but glxgears is no benchmark :P
surfer: of course it is
adamk: surfer, No, if I care about my performance, I use a relevent benchmark.
surfer: if u test 2 systems running the same driver
surfer: and u think the numbers don't mean anything
surfer: that's hilarious
z3ro_: surfer: yes, you could use glxgears as a kind of benchmark, but it's a bad one.
adamk: What's hilarious is that you actually think they mean something :-)
z3ro_: it doesn't stress the gpu very much.
z3ro_: and it's pretty much just how fast you can glClear
eboettcher: oh that reminds me to look into a small test program I had...
rx__: z3ro_; maybe someone should patch out the FPS output ;)
eboettcher: it would glClear(red); draw a flat shaded quad in white, and call glxswapbuffers
eboettcher: this was called in the event loop
eboettcher: the first frame was red with white box
eboettcher: all after that: red
z3ro_: rx__: I guess people would complain then that it's broken or missing a "feature"...
rx__: a misleading feature at that
Ori_B: rx__: hm. I think for a while it had an option you had to add
Ori_B: 'glxgears -iacknolwedgethatthisisnotabenchmark' or something
Ori_B: maybe that needs to come back?
z3ro_: I imagine if someone were to remove it, someone else would just add it back.
z3ro_: or change it to the option.
eboettcher: z3ro_: it would still remind the community of an important fact :P
z3ro_: I really should install a newer X server, that doesn't have an input bug that brings down the whole system...
Ori_B: maybe calling the option something shorter would help
Ori_B: "glxgears -unbenchmark", and have the help say "print out some pseudorandom FPS numbers"
adamk: So this same build of Xorg works with an x850, but fails with my x1650 (which I had working, briefly, last night before updating today). This is the Xorg log file: http://pastebin.ca/1033409
adamk: It's unable to load AIGLX... Calling driver entry point failed?
adamk: Ahhh, on the console I see "unknown chip id 0x71c1, can't guess."
adamk: I'm going to guess that's somehow related to AIGLX not getting enabled?
xnguard: What's the fastest, easily-obtained R5xx card available right now? The X1650?
adamk: Probably depends on the definition of "easily-obtained" :-)
xnguard: adamk: "I can buy it from NewEgg." :)
adamk: Search NewEgg for radeons :-) The x1950 is probably the fastest r500.
xnguard: I see a lot of HD cards for cheap, but I'm unclear on what core GPU those use, and what status the radeonhd project is in.
adamk: The HD cards are all r600, I believe.
adamk: And there is no 3D for them yet.
adamk: Or, rather, no open source 3D for them.
stefanb: Bahh, ut2004demo and UniEngine Sanctuary don't work
adamk: ut2004 ran fine here last night for about 5 minutes.
stefanb: adamk: Some xcb error, not even OpenGL/driver problem :-)
adamk: But I can't get X working again today after updating... So I'm updating again, and just assuming I screwed something up earlier :-)
agd5f: adamk: unknown chip id 0x71c1, can't guess. means your mesa is too old
stefanb: adamk: mesa git master?
adamk: agd5f, Odd, 'ldd /usr/bin/glxinfo' is pointing to the newly compiled version of libGL.. I think I have an old one floating around. I'll move it out of the way.
stefanb: adamk: That's why I recompiled SRPMs with updated git archives: stuff ends up where the system expects it
stefanb: airlied: I guess you are not yet awake?
adamk: agd5f, Updating mesa did it... Which is odd, because all I did was rsync my Xorg installation from one machine where it worked to this one, where it didn't.. Oh well, just something to bear in mind.
adamk: Xorg starts, and compiz works seemingly without problem with my x1650 now that I set vblank_mode to 0 as onestone suggested to me earlier.
stefanb: adamk: Yeah, vblank_mode = 0 is also required on my RV515 to get compiz working
stefanb: I think MrCooper is working on that issue...
stefanb: OK, nightynight....
GerbilSoft: with compiz running, X freezes if i switch virtual desktops
GerbilSoft: (haven't tried with vblank_mode=0 yet)
xnguard: Hm. I'm not seeing X1950[PRO]s available for purchase anywhere. Discontinued?
peterz: its starting hard to get x2xxx series
rx__: xnguard; for US?
xnguard: rx__: Yeah. I'm in .va.us.
rx__: try amazon or compusa
adamk: GerbilSoft, You need to set vblank_mode=0 in your /etc/drirc file (or perhaps just your users ~/.drirc file would work).
xnguard: CompUSA's bankrupt. I wouldn't trust them to ship anything to anywhere, and definitely not at a decent price. :)
adamk: GerbilSoft, This is my /etc/drirc file: http://pastebin.ca/1033464
GerbilSoft: adamk: that fixes the virtual desktop problem?
GerbilSoft: ok, i'll set it
rx__: compusa was bought out.. but i understand where you're coming from
adamk: GerbilSoft, It fixed a few lockups I experienced with compiz... And lockups for a few others.
adamk: It may not be a guaranteed fix, but it's worth testing :-)
xnguard: ...there's a very few X1950PROs available on Amazon for a very large sum of money. The X1650s on NewEgg are going for USD$45 (and USD$15 MiR scam). At least, the Sapphire boards are.
GerbilSoft: ok added
GerbilSoft: i'll test compiz again once i finish recompiling due to an xcb change
GerbilSoft: (libxcb-9999 in portage removes libxcb-xlib.so)
xnguard: I'm kind of curious to know why core clock and RAM MHz aren't listed for HD cards on NewEgg. Just RAM bit width and num stream processors.
rx__: they are
rx__: see specifications tab
xnguard: I did.
xnguard: Check the specs tab.
rx__: depends on the listing
xnguard: I've looked at about six Sapphire HD products. The only differentiator is SPUs. They all have 0.5GB VRAM @ 128-bit, GDDR3.
xnguard: Is this an ATi marketing ploy, or something? Like the Athlon "MHz don't matter" campaign?
rx__: i think it's just sapphire
xnguard: Well, ATi lists MHz, but their cards are a screaming rip-off. ;)
xnguard: Er, s/ATi/ASUS
rx__: i tend to think of sapphire as bottom barrel :/
xnguard: rx__: Whyso? I've had good luck with Sapphire, and IIRC they were ATi's actual board manufacturer until ATi stopped selling boards.
rx__: i've had good experience w/em too
rx__: but read horror stories once you need tech support
rx__: and for the most part they flood the market with same card different spec
xnguard: ...Sapphire, HIS, and Diamond are the last vendors with X1650PROs on NewEgg, in price order, with Diamond charging screamingly high prices.
rx__: i don't know anyone who buys diamond...
xnguard: Someone must, or NE wouldn't stock 'em. :)
xnguard: Okay, never mind. NewEgg's keyword search is still broken. PowerColor still offers a range of X1650PROs.
xnguard: Oi. And their power search is even more screwed up.
xnguard: Any opinions on Sapphire vs. PowerColor vs. HIS?
mattmatteh: likes fanless
xnguard: Sapphire and HIS seem to get significantly better reviews.
xnguard: ...I'm not seeing any fanless X1650PROs.
otaylor: xnguard: I dont' think they make fanless 1650's, just 1550's
rx__: i think HIS/powercolor have 2yr warranty vs. sapphire's 1 yr
xnguard: With any luck, in a year, we'll have nice, fast, stable support for R600. ;)
otaylor: rx__: though for a 30-40 $ card, the chances you wanting warranty service a year later is low
otaylor: (i've also never had a graphics card die, other than the fan getting noisy)
adamk: I had an x800 that burned up...
adamk: But that was because the fan died.
airlied: has a fair few dead ones :-(
xnguard: I'd just like to throw out this X800GTO for something more stable, and at least as fast, preferably in an R5xx.
airlied: mostly RAM.
airlied: hey look the BIOS screen has funky stripes..
airlied: I've two of those cards.
xnguard: airlied: Make/model?
airlied: xnguard: ah one FireGL of some sort and one r300 all in wonder card
airlied: but I handle cards a lot more and I'm not the most static safe person.
xnguard: *grins* I haven't had an AiW card since my 7500.
xnguard: Pity PVR software was so terrible back then. I was working on building a DIY PVR.
xnguard: Hm. I think I'll put the HIS card on my wishlist for now. It's got external exhaust without messing with aftermarket coolers.
GerbilSoft: i've got an old 8500DV AIW somewhere
GerbilSoft: don't know where the breakout box is though :(
xnguard: Yeah. I gave my 7500 AIW to a friend who was building a cheapo computer to someone, but couldn't find the Super-Sekrit breakout cord thingy.
xnguard: Then I found it three months later while I was getting rid of SCSI cables. :)
GhotiPhud: I'm getting an error while trying to compile the xserver from git
GhotiPhud: could anyone help me?
eboettcher: Error: R300 timed out... exiting
eboettcher: and X is eating up all cpu
eboettcher: linux-nat.c:979: internal-error: linux_nat_attach: Assertion `pid == GET_PID (inferior_ptid) && WIFSTOPPED (status) && WSTOPSIG (status) == SIGSTOP' failed.
eboettcher: yet gdb won't attach :/
eboettcher: err hardlock
mattmatteh: does anyone have idle time working with pidgin ? i have been online all day, but its reporting that i have been idle for over 18 hours
mattmatteh: wrong channel
mattmatteh: that was meant for #gentoo