00:02RSpliet: that's not necessarily important though... on NVA0 there's +e24, +e28 and +e34. None of them seem to have any direct correlation with VBIOS bits, none of them improve stability on reclocking
00:03RSpliet: imirkin: ^ they're likely just... "parameters". idk, tweaks on mem request priority, queue sizes... small things
02:46pmoreau: RSpliet: Did you made progresses on reclocking for G96? O:-)
02:55RSpliet: pmoreau: I've managed to not have any time in the past week
02:55RSpliet: like, literally 0
02:55RSpliet: maybe coming weekend
02:59pmoreau: Eh eh!
02:59pmoreau: Well, same for me unfortunately...
02:59pmoreau: Reading logs was about the only find I could do
07:28hakzsam: joi, Hi, what is the minimum required version of glibc for valgrind-mmt-git?
07:30joi: hakzsam: I have no idea
07:34hakzsam: okay, you have recently added support for glibc 2.21, I'm going to update the AUR package ;)
07:35hakzsam: oh! this package is not mine actually
08:55s0be: I'm running into a glitch triggered by google-chrome. At some point during a session, chrome triggers an interesting state that, if I switch desktops (or workspaces as xfce calls it), it shows the screen state that chrome was on when the bug triggered.
08:57s0be: (Running Linux Kernel 3.19, latest xf86-video-nouveau, nva5/GT216[Quattro 440m])
08:58s0be: What can I do to help track down where the bug resides? I can take a video showing it, if that helps.
09:06Karlton: s0be: Does it still happen if use xf86-video-nouveau-1.0.11 instead of the latest?
09:07s0be: I'll downgrade and test. I've been hitting this for a 'while(months)' but been overwhelmed at work and unable to make time to debug.
09:08imirkin_: s0be: can you take another stab at describing the issue? i didn't quite understand...
09:08s0be: Basically, chrome (flash plugin specifically) at some point seems to take a snapshot of the current workspace it's running in
09:09s0be: then, later, if I switch workspaces, I see that snapshot instead of what should be on another workspace. I'll take a video real fast to show you.
09:09imirkin_: oh, i see
09:09s0be: if I minimize chrome, it stops happening
09:09imirkin_: so basically screen doesn't update
09:09imirkin_: do you see any errors in dmesg when this happens?
09:09imirkin_: sounds like a copy engine issue
09:10imirkin_: how easy is this to reproduce?
09:10imirkin_: does it only happen after you've been running the desktop for a while?
09:10s0be: yeah, it can take a few hours to happen
09:11s0be: I actually went a few weeks without it happening, figuring chrome had fixed the glitch
09:11imirkin_: do you ever suspend/resume?
09:11imirkin_: next time you reboot, try booting with nouveau.config=PCE0=0
09:11imirkin_: this will disable the copy engine
09:11imirkin_: which is the thing that may be breaking here
09:12s0be: if it helps to make sense of it, the 'glitched' screen data is always the same. It's a shot from about 36 hours ago.
09:12imirkin_: although.... someone reported a composite issue too
09:12imirkin_: and there's no corruption on it at all?
09:12imirkin_: maybe i would like a video ;)
09:12s0be: Grabbing one, give me 2 minutes
09:16s0be: (got a phone call as soon as I started video, promise I'm working on getting it
09:16imirkin_: take your time
09:16imirkin_: no rush. i'm not the one with the problem ;)
09:19s0be: And now to figure out uploading a video to youtube from my phone, haha
09:21BadHorsie: GK106M [GTX 765M] Kepler NVE6, have mmio traces on nvidia 346 been taken for this one? Is there anything I could run/test?
09:22imirkin_: BadHorsie: should work fine... except a bunch of them fail to start up. i believe skeggsb is supposed to receive an affected laptop
09:25buhman:would love working gk106
09:25buhman: imirkin_: false
09:25buhman: imirkin_: devinit works fine
09:25imirkin_: buhman: i believe i addressed your case with "a bunch of them fail to start up"
09:26buhman: well, yeah
09:26imirkin_: or do you think that perhaps the people for whom it works don't come in here to complain that it works great for them?
09:26buhman: I don't hear anyone complaining about not-gk106 nearly as much ;p
09:26imirkin_: yeah, most of the people with problems and recent hw do have gk106m's
09:27s0be: I'm complaining about GT216 right now! ;-)
09:27imirkin_: however do most people with gk106m's have issues? that seems like a leap
09:27imirkin_: s0be: are you on mesa-git btw? i seem to have introduced a bug in the nv50 driver
09:28s0be: how recently, I know I've hit this since at least 3.16
09:28imirkin_: i'll probably just roll the change back... but you might consider moving back to mesa 10.5.x
09:28s0be: (linux kernel, so that timetable)
09:28s0be: this isn't a new bug
09:28s0be: it's just a "I've had no time to report it" issue
09:28imirkin_: commit went in on feb 19 or so
09:29imirkin_: commit 5000a5f67b55
09:29imirkin_: whompy: btw, am i waiting on you to test something re that commit?
09:30s0be: yeah, probably not related to my issue, but I'll drop back to stable mesa if disabling copyengine doesn't solve it (or if it does, so can help localize the bug)
09:30imirkin_: s0be: from the sounds of it, you're seeing a stale pixmap or texture or whatever
09:30s0be: here's a dmesg dump if it's helpful: http://pastebin.com/iY7CD7Zx GPU init starts around line 889
09:31BadHorsie: Can you direct me to a good info link on where to start poking at code and compiling/testing best practices for this ? I still see lots of error entries on dmesg and many things freeze my laptop even on 1 monitor...
09:31s0be: video upload at 71%
09:31buhman: BadHorsie: hah
09:31buhman: BadHorsie: even skeggsb couldn't figure it out
09:31imirkin_: BadHorsie: pastebin your dmesg... which errors are you talkinga bout?
09:31buhman:probably the NMI stuff
09:31imirkin_: buhman: i still have faith in him... with the hw in front of him, he'll defeat it
09:32imirkin_: s0be: btw, unrelated, but with 3.19 you can now reclock your gpu if you boot with nouveau.pstate=1
09:32buhman: BadHorsie: I bet your dmesg looks like this: https://bugs.freedesktop.org/attachment.cgi?id=101916
09:33s0be: nice, will play with that. This laptop has a horrible thermal design, single heatpipe for gpu and cpu, so I have had to force my CPU to the lowest C-State while watching videos or it'd thermal shutdown
09:34imirkin_: s0be: well you can also use vdpau to decode the videos... might be easier on your cpu
09:34s0be: I do
09:34imirkin_: s0be: also you can downclock the gpu, which will hopefully help
09:34imirkin_: looks like it comes up in the middle pstate by default
09:35s0be: the problem is, I'm doing android builds and whatnot in the background, and vdpau playing a video with 4 cores pegged at full speed is too much
09:35imirkin_: hehe. perhaps invest in an actual computer :p
09:35imirkin_: they have like 36-core chips now
09:35s0be: BAH! Heresy. Also, yeah, it's planned.
09:35imirkin_: or something ridiculous like that
09:35s0be: Hey, this laptop was great like 4 years ago when I got it.
09:37imirkin_: hm, only 18 cores? maybe i was thinking with HT...
09:38buhman:lugs a quad-socket E7 board at imirkin_
09:38s0be: non power of 2 core counts seems weird to me
09:39imirkin_: s0be: how do you feel about non-power-of-2 words?
09:39imirkin_: like 18-bit or 36-bit words
09:39s0be: like... 24 bit words or something?
09:39imirkin_: as the pdp's of yore had
09:39buhman: how about prime numbers
09:39s0be: that's just.... dirty
09:39BadHorsie: http://pastebin.com/zYhhD0xS imirkin_ very similar, I do not see the first "MMIO read of" tho...
09:40imirkin_: there's no *real* reason for words to be powers of 2
09:40BadHorsie: (In your link)
09:40s0be: next you'll say non zero null pointers and no ring 0
09:40imirkin_: s0be: no, but i will say *valid* null pointers :)
09:40imirkin_: i.e. ptr == 0 is valid. heh.
09:41imirkin_: BadHorsie: oh neat. and you get the unknown dp condition 7 thing
09:41imirkin_: iirc i saw that before somewhere
09:41imirkin_: BadHorsie: try booting with nouveau.noaccel=1
09:41imirkin_: that should calm things down i think -- now it's trying to spin that gpu up and almost succeeding
09:42s0be: https://www.youtube.com/watch?v=Nk5qGnIYgGk HA! video. take that internet
09:42s0be: (that shows my glitch imirkin)
09:43BadHorsie: imirkin_: lemme try that
09:43imirkin_: s0be: seems like you just have random windows all over the place
09:43imirkin_: s0be: i blame your compositor
09:43s0be: nope, that's exactly what my screen looked like a few days ago
09:43imirkin_: (please tell me you have a compositor)
09:43imirkin_: (which i can blame)
09:43s0be: yes, I have a compositor I believe (xfce4 has one right?)
09:45s0be: let me dig for a few minutes, figure out if xfce has one built in, and if not, if I may have set one up at some point in the past few years
09:46Karlton: s0be: it does by default, not sure if it is turned on though
09:46s0be: yes, found that it has one built in, checking if it's enabled
09:49s0be: it is, indeed, enabled. Hooray for finding what to blame.
09:49imirkin_: not necessarily its fault, but if i blame it, it can't fight back
09:50s0be: disabling composite fixed the glitch on the fly
09:50imirkin_: ok, so here's the sticking point -- i have another report from someone where the nouveau impl of the composite exa operation is messed up
09:50s0be: re-enabled composite, and the glitch happened again, but the 'corrupt' data was updated to what's currently on my screen.
09:51imirkin_: i don't think this is related to the composite extension though
09:51s0be: Anything else I can do to help track it down?
09:51imirkin_: invite the xfce compositor people to look at it -- i'm currently laying the blame on them :)
09:52imirkin_: of course they'll probably say like "oh yeah, nouveau sucks, it's clearly nouveau's fault" :)
09:52s0be: well duh, noone ever takes blame. It's not my fault my golf ball went into the woods, this club sucks.
09:52imirkin_: but maybe not
09:52imirkin_: and if they can provide some indication that it's nouveau's fault, i'll gladly look at it
09:53imirkin_: unfortunately i don't know that much about the whole composite flow
09:53s0be: In the interim, I can just disable composite
09:53imirkin_: could be that nouveau is updating pixmaps when it's not supposed to
09:53imirkin_: or not updating them when it is
09:53imirkin_: or could be a bug in their thing
09:54buhman: sounds like imirkin_ needs to raid nvidia for datasheets
09:54imirkin_: but i would hope that the people who wrote the compositor know *something* about the composite extension
09:54imirkin_: unlike me, who knows nothing
09:54Karlton: would testing a different compositor like compton help?
09:54s0be: yeah, was just thinking that
09:55s0be: I do mostly coding on this laptop, so I haven't really played with eye candy on it. Will see what I can do to swap compositors.
09:55buhman: imirkin_: http://cgit.freedesktop.org/mesa/mesa/tree/docs/GL3.txt#n115 when are you going to finish this?
09:56BadHorsie: imirkin_: http://pastebin.com/BuD5AbGU
09:57BadHorsie: Just one E, nice, still was able to run glxgears...
09:57buhman: Kernel driver in use: nvidia
09:59imirkin_: BadHorsie: right, you're not going to get hw accel when you say noaccel=1 :)
10:04s0be: I can confirm that switching to compton does not trigger the issue, so it looks like an interaction between the xfce4 composition manager and nouveau specifically.
10:05imirkin_: cool. well you'll definitely have to involve them, i'm not about to figure out how their compositor works :)
10:05s0be: yep. But, you were able to help me find where the issue resides at least
10:06s0be: (I'm still going to blame google-chrome/flash, as it never happens without those open)
10:08BadHorsie: And X dies when I tried to use arandr and extend to use another monitor ouch...
10:22s0be: imirkin, gonna test pstate control, is it 100% manual, or is there a daemon I can run that heuristically changes states?
10:22imirkin_: fully manual
10:22imirkin_: not reliable enough right now
10:22imirkin_: works well for most people though i think
10:23s0be: can I force the boot pstate with a kernel commandline arg?
10:23imirkin_: or whatever
10:23s0be: and I suppose if I had asked modinfo nouveau it would have told me....
10:23imirkin_: the config thing isn't well documented
10:24s0be: k, thanks. Want to see if forcing the low clock can still play videos smoothly, and will just leave it there
10:24imirkin_: even http://nouveau.freedesktop.org/wiki/KernelModuleParameters/ doesn't have NvClkMode
10:24imirkin_: i guess i should add it in
10:25s0be: just to make sure I set it right, NvClkMode=3 (when dmesg lists 3,7,f) would be how I choose that param?
10:25imirkin_: 3, 7, 15 (sadly it doesn't parse hex!)
10:26s0be: k, no issue there, not going for more power
10:26s0be: I don't even touch what my gpu can do other than with video playback
10:29s0be: rebooting, lets see this baby in action!
10:36s0be: Well, the world didn't end form booting with pstate enabled and set
10:36imirkin_: does the pstate file say that you're in the right mode?
10:36imirkin_: (look for -- )
10:36imirkin_: cat /sys/class/drm/card0/device/pstate)
10:37s0be: yep, it's in 03
10:37s0be: want me to paste the 4 lines of output here?
10:37imirkin_: note that if you plug in an external screen, you might have to clock up
10:37s0be: it doesn't have a -- in sysfs, it says 'AC'
10:37imirkin_: blob driver handles all this automatically
10:38imirkin_: oh right. you can set diff preferred levels for AC/DC
10:38imirkin_: imo it's not really the right place to be doing this
10:38imirkin_: but... wtvr
10:38s0be: yeah, that should be a userspace thing, policy shouldn't live in the kernel or whatever the mantra is
10:39imirkin_: i suspect it'll disappear before this becomes more widely available
10:39s0be: I fall into the camp of "it works, don't care about how"
10:39s0be: vdpau test time
10:40s0be: yep, 720p playback is still smooth
10:40s0be: with it in the lowest state, so I should never have to muck with it
10:42s0be: 1080p is fine too
10:42s0be: thanks for pointing that out!
10:42imirkin_: np. hopefully it has a material effect on actual power consumption
10:43s0be: I'm more concerned with temperature (yeah, yeah, power consumed = heat, so two sides of the same coin)
10:43imirkin_: also note that it's not the 720p or 1080p-ness of the video that dictates ability to decode
10:43imirkin_: but rather the bitrate of the encoding
10:44imirkin_: you can have a 100kbit/s 1080p video, or you can have a 100mbit/s 480p video :)
10:45imirkin_: (actually probably 480p can't go that high, even if you effectively make it a mjpeg)
10:47s0be: nouveau has certainly come a long way since I got this laptop
10:47imirkin_: yeah, pretty much all features of that gpu should be reasonably well supported
10:48imirkin_: although the blob has a more sophisticated shader compiler
10:48imirkin_: probably coz they've invested more than 1-2 man-years into it
10:49imirkin_: (if i had to guess, they've put in probably 500+ into it)
10:49s0be: yeah, considering it IS their business
10:50s0be: all the fancy hardware in the world can't make up for junk drivers pushing it
10:50imirkin_: but i think nouveau is at like 80% of the perf in a lot of the cases when reclocking works on the card
10:51imirkin_: i bet we could make up half of it by having instruction scheduling
10:51imirkin_: (half the difference, that is)
10:51s0be: I understood the result you talked about, just not the method of achieving it
10:52imirkin_: although iirc RSpliet had some dissapointing results on his nva0 even clocked all the way up it was like 60% of the perf
10:54s0be: I'll just keep using the work done. I don't know that I ever had to use the blob on this system, so I have no baseline for what it 'could be'. I also don't do anything that pushes the gpu
10:54imirkin_: yeah, these things would only matter if you played games
10:55s0be: I'm just giddy that my laptop is running colder while watching videos, lol
10:56imirkin_: did you actually check that it cooled down?
10:56imirkin_: or just hoping for it?
10:56s0be: I keep a systray with cpu/gpu temps
10:57s0be: I used to hover around 140f while watching a video, now I'm hovering around 120f
10:57imirkin_: nice :)
11:00s0be: So getting composite fixed gets bumped down my 'queue' and you've helped my system stay cool. All in all, you need to be hourly imirkin_
11:03s0be: I'd buy you a drink, but the reason I have time for this is I'm between jobs (Interview tuesday), so I'll put a note 'Owe imirkin beer' on my queue for after employment is found.
11:03imirkin_: good luck
12:55nfk: what's the state of power management for NVC3? does runpm support it?
13:08nfk: ah, so it's only for suspension, fairly useless for me
13:08nfk: unless it can get suspended while X11 is running but display is off
13:18imirkin_: nfk: runpm should work fine with it
13:19imirkin_: nfk: it only gets turned off when the card's not being used at all. and only in optimus setups, since you need to use acpi to actually shut the device off
13:19imirkin_: there's no reclocking for fermi family though
13:21nfk: so pstate does nothing?
13:24imirkin_: it shows you the current pstate :)
13:24nfk: i know that already - it's the middle setting
13:24nfk: decent enough for compositing but not really fast enough for what i'd like it to do while watching videos
13:25imirkin_: ah =/
13:25imirkin_: it should be plenty fast to run vdpau...
13:26nfk: i want to run mpv's vo_openg with some pretty crazy settings
13:26imirkin_: iirc the NV_vdpau_interop thing isn't so great
13:27nfk: it's mostly doing it already except i'd really like to upgrade luma scaler from spline64 to ewa_lanczossharp (basically Jinc if that rings any bells)
13:27nfk: why i'd use vdpau interop?
13:27nfk: my videos are virtually always Hi10p anyway so it's not like Fermi or even Kepler could decode them
13:27nfk: not sure about Maxwell
13:28imirkin_: nfk: that's how most of the opengl outputs work... use vdpau to decode, and then use opengl to display/further process it
13:28imirkin_: maxwell's vdpau doesn't work yet
13:28nfk: sure, mpv can do that but it's as you pointed out buggy and i just want to run some slightly crazy filtering on video using GPU
13:29nfk: it's already almost powerful enough to if i got it to run at full power it should handle it fine
13:29imirkin_: just reduce the crazy ;)
13:29nfk: but i ran some tests and it looks so sexy
13:30nfk: apart from dropping frames from scene to scene
13:30nfk: but the drop amount is not that high so just a bit of an oomph in performance should make it usable
13:30imirkin_: you could look at the shaders being generated and improve the compiler so that it generates better shaders
13:31nfk: it's already impressively improved
13:31imirkin_: iirc nvc0 can dual-issue if you order the instructions properly
13:31imirkin_: but we don't do any instruction scheduling
13:31nfk: half a year ago spline16 for crhoma scaler would make it drop frames
13:31imirkin_: hmmmm... no shader improvements in the last 6 months that i'm aware of...
13:31nfk: nowadays it's spline64 for both and with new frame generation to smooth out motion
13:31imirkin_: perhaps the mpv shaders improved :)
13:32nfk: and all that while hopefullly outputting proper sRGB
13:32nfk: or mesa?
13:32nfk: or both
13:32imirkin_: or perhaps i added support for some extension that mpv optionally uses
13:33nfk: thank you for your work on nouveau, a virtual pat on the back from me
13:33imirkin_: it's fun :)
13:34imirkin_: been spending more time on freedreno lately though -- it has a lot more easy-ish things to do
13:35imirkin_: but when chrisf finishes his core tess patches, i'll switch back and get tess going again... it's already pretty close
13:37nfk: does that mean OpenGL 4 is imminent?
13:37imirkin_: gl4 is always imminent
13:37imirkin_: it's been imminent for a year now
13:37nfk: true, just like wayland and ubuntu on mobile
13:37imirkin_: i had tess working in like june of last year
13:37TheSeven:thinks it's intermittent :P
13:37imirkin_: or maybe it was july
13:38imirkin_: but then there's also ARB_shader_subroutine which is like the black sheep of extensions
13:38imirkin_: it's required for gl4, but no one uses it, and no one wants to implement it
13:39nfk: so do the mesa thing and call it implemented without implementing it? and then get yelled at by angry devs when they do us e it and it's not actually implemented
13:40imirkin_: "do the mesa thing"?
13:40nfk: from what i know, that's the modus operandi for mesa during OpenGL 2 days
13:40imirkin_: mesa tends to have fairly high quality implementations for all the gl it supports...
13:40imirkin_: on average a lot more conformant than the average vendor driver
13:41nfk: i'm pretty sure i remember mpv devs derping about mesa or intel driver claiming to support something while not actually supporting it
13:41imirkin_: some really old functionality does get occasionally thrown into the "wtvr" bucket
13:41nfk: like 2 or 3 years ago
13:41imirkin_: i mean... bugs do happen
13:42imirkin_: but that's hardly the mo
13:42imirkin_: and every so often individual backends lie about things
13:42imirkin_: because the dev is lazy
13:42imirkin_: like me enabling instancing on freedreno
13:42imirkin_: even though it only works up to 256 instances
13:43imirkin_: i really should fix that up
13:43nfk: in fact, i'm pretty sure mpv had to change fbo format because what worked on nvidia either did not work at all or had terrible performance on intel and then it turned out that the new formwat was even worse on mesa
13:44imirkin_: formats are a disaster in GL
13:44nfk: now it's rgba16 which seems to work with all of them
13:44imirkin_: no driver is required to support any particular thing
13:44imirkin_: and it can mark fbo's as incomplete for any reason it feels like
13:44imirkin_: like time of day or phase of moon
13:45imirkin_: applications end up having to try a sequence of formats
13:46imirkin_: and in practice, there are certain things you kinda have to support, otherwise no application will ever run
13:46imirkin_: despite your driver being fully 100% conformant to the spec
13:47imirkin_: but anyways, i'm not aware of any precedent for claiming to support something without actually supporting it at all
13:48imirkin_: definitely not at the mesa core level
18:42whompy: imirkin: Not that I am aware. I tested master with the QUERY_ALLOC_SPACE set to 128 and had no issues, so reverting that part of the commit seems to be at least the short-term fix.
18:42imirkin: no, that'll totally break the other thing
18:43imirkin: i should just revert the commit
18:44imirkin: really what i should do is plug in a tesla and see wtf is going on
18:45whompy: Aha. I wondered if that was the case. That is what I did locally. I can test stuff if you want, but I'm definitely still learning and may need guidance.
18:46imirkin: what was the original issue? segfault right?
18:46imirkin: if you could load it up in gdb and see wtf is going on, that'd be *super* useful
19:08whompy: It was a Steam issue... and appeared to come across as a clean close. I'll need to dig in a little further in a few days.
19:09imirkin: right, but what was the issue?
19:09imirkin: wasn't it a segfault?
19:10whompy: Ah, there it is. Yeah, segfault
19:10imirkin: if you can see what's going on in gdb, that'd be great
19:10imirkin: it might be in steam code
19:10imirkin: which would be unfortunate
19:12whompy: I'm slowly repartitioning my nouveau box right now, so I'll try to dig into it more when I finish.
19:13whompy: I am still very green at using gdb, so I may need some pointers along the way. I think I found a post with some of the debugging options in Steam since my last attempt.
19:14imirkin: with steam iirc you have to use like
19:14imirkin: when it loads, just hit 'r' which will run the thing
19:14whompy: Yeah. There was a STEAM_DEBUG option in there too
19:14imirkin: and then when it dies, run
19:14imirkin: bt full
19:14imirkin: and pastebin the output of that
19:15whompy: Ok, will do
19:44TrumpChild: Does anyone know where I can find a kernel level source for a Radeon HD 7870?
19:45TrumpChild: One that can support Eyefinity (multi-display), vertex and pixel shaders, and GPGPU; basically a "complete" source.
19:45TrumpChild: *A device driver.
19:47TrumpChild: I figured it was accessible from there, but I was looking for someone more direct.
19:47TrumpChild: *something more direct. The Linux kernel source is too massive.
19:48imirkin: btw why are you asking questions about radeon in #nouveau?
19:48TrumpChild: Because it's easier this way.
19:48imirkin: anyways, the driver is part of the linux kernel. look in 'drivers/gpu/drm/radeon'
19:49TrumpChild: You mean the driver is compiled alongside the kernel?
19:49TrumpChild: I thought it was as a "loadable module" like in OS X.
19:49imirkin: i don't understand what question you're asking
19:49TrumpChild: Then read it again.
19:49imirkin: and this channel is about the nvidia open-source hw driver
19:50TrumpChild: Wouldn't you think I understand that by now? I came here manually.
19:50imirkin: as opposed to... automatically?
19:50TrumpChild: I am not an idiot.
19:50imirkin: thanks for clarifying
19:50TrumpChild: Now I believe you'll need to convince me of the same.
19:50imirkin: sounds good
19:51TrumpChild: Your inability to capitalize a sentence shows that you sure aren't competent in English perfection.
19:51TrumpChild: Something good to remember when you indirectly call someone an idiot over relay chat.
19:52imirkin: thanks, i'll be sure to keep that in mind
20:17whompy: imirkin: managed to get *something* out of it. Turns out I really suck at gdb and just needed to iterate. Didn't realize it was sitting at a breakpoint.
20:17imirkin: there should be any breakpoints
20:17whompy: Shall pastebin in a moment.
20:17imirkin: when it starts it doesn't auto-run though... so just do 'r' and then it'll run
20:18imirkin: i guess it might sit at a signal... that can happen
20:18imirkin: very confusing at first
20:18whompy: That's where this was odd: It did auto-run. and sit at a signal.
20:18imirkin: i had the same problem on arm... stupid crypto library ends up doing something that causes a SIGILL
20:19imirkin: which it catches and handles, but i didn't know that
20:21imirkin: please confirm that your line 163 == q->data = q->sequence + 1; /* for comparison COND_MODE */
20:21imirkin: can you also tell me what "p q->data" outputs
20:22imirkin: actually, just do "p *q"
20:23imirkin: might be easier if you do a build with -O0... also wouldn't hurt to throw --enable-debug in there
20:24whompy: Ah shoot. I'm using the package I made from my other machine, not my local git tree. I'll run again and confirm.
20:24whompy: What are your recommended flags for a debug build?
20:24imirkin: i see why this crashes
20:25imirkin: the real question is, why didn't this crash earlier
20:25imirkin: i need to think carefully about this
20:26imirkin: if you change
20:26imirkin: if (q->offset - q->base == NV50_QUERY_ALLOC_SPACE)
20:27imirkin: if (q->offset - q->base + 32 >= NV50_QUERY_ALLOC_SPACE)
20:27imirkin: i suspect it'll all work much better
20:27imirkin: or perhaps the data increment should happen later? gr
20:28imirkin: yeah. the nvc0 stuff sets rotate to 32
20:28imirkin: for occlusion
20:29imirkin: so yeah, kill that suggestion... but instead do
20:29imirkin: q->offset += 16
20:29imirkin: make that be += 32
20:29imirkin: and change the line below to also use 32 instead of 16
20:30imirkin: that should work.
20:31imirkin: whompy: let me know if that makes sense
20:31whompy: Ah, looking at lines 153-154 in master, right?
20:32imirkin: whompy: http://hastebin.com/raw/icivigavah
20:32imirkin: apply this patch
20:32whompy: Ah, two spots. will do.
20:33imirkin: this is wholly unrelated to the space increase thing... it was just getting lucky before
20:36whompy: My ability to follow this stuff is elementary and rusty at best, but fun.
20:36whompy: ...that sentence made no sense. I should probably resume testing in the morning at this point.
20:37imirkin: whenever you get a chance, it'll be super-useful
20:37imirkin: due to my extreme laziness around swapping hw
20:38whompy: Not a problem. I just need to build and run it, but I apparently missed something with my build setup when moving stuff around.
20:38whompy: Getting a bit too groggy to go searching for what I mistyped.
20:44imirkin: hehe np
20:55whompy: imirkin: I lied. Turns out I just halfway removed llvm from my config. Tested the patch, and found no issues.
20:55imirkin: wait, so it all works?
20:55imirkin: no more crashes?
20:56imirkin: oh, this probably started happening with tobijk's render condition inverted stuff
20:56whompy: Correct. Applied the patch above to a fresh, clean master and it loaded up no problem. Hopped into a map for good measure to confirm.
20:56imirkin: i think he added those data/5 writes
20:56imirkin: but didn't increase the 16 to 32
20:57whompy: Ahhh, that'd do it.
20:57imirkin: commit 1a170980a09d82201b88e3fe20684a0df2cfea80
20:59imirkin: whompy: can i put on your tested-by?
20:59whompy: Interesting! Well, feel free to add my Tb
21:00whompy: would you like me to throw anything in the bug?
21:01imirkin: there's a bug open?
21:02whompy: Oh yeah, fdo #89679
21:04imirkin: ok, i'll add a ref
21:06whompy: Thanks! I appreciate the debugging help & support. Have a good night (or whatever may be appropriate for your location)