02:14 mwk: imirkin_: I have no idea what they are
04:05 karolherbst: is there some kind of visualisation software which can like give you a tree representation of the generated assembly code?
04:06 RSpliet: I wouldn't assume there is
04:06 RSpliet: (but if you want to look if you can modify existing software, look for "control flow graph" :-P)
04:07 karolherbst: ohh right, that was the term
07:22 imirkin_: RSpliet: did you ever look at using addRegPreference for the nv50 mad thing?
07:23 RSpliet: I... can't recall
07:24 imirkin_: ok
07:24 imirkin_: i'm trying to figure out how to properly deal with the "regid must be less than 64" issue
07:24 imirkin_: and happened upon that bit of (unused) functionality
07:26 RSpliet: ah, no I didn't look at addRegPreference
07:26 imirkin_: it seems expressly designed for the situation at hand
07:27 RSpliet: my attempt was a semi-blind "coalesce" of ssrc2 and sdst, and found out that the compiler did it as blindly as I tried (rather than preserve correctness)
07:27 imirkin_: (the mad situation)
07:27 imirkin_: right...
07:27 RSpliet: ( http://lists.freedesktop.org/archives/nouveau/2015-January/019725.html - for reference)
07:28 imirkin_: anyways, i'll give it a shot
07:28 imirkin_: ... at some point
07:32 RSpliet: Soon™
07:32 imirkin_: Not Soon (tm) (r) (sm)
07:38 karolherbst: wuut, there are some games which doesn't do vsync :/ that's really annyoing with prime
07:39 Tom^: O_o which game
07:39 karolherbst: risen
07:40 imirkin_: karolherbst: you can force vsync
07:40 karolherbst: imirkin_: how?
07:40 imirkin_: karolherbst: vblank_mode=3 iirc
07:40 imirkin_: or 1 or 2
07:40 imirkin_: i forget what they all do
07:42 karolherbst: yeah I never found any documentation about that option :D
07:43 karolherbst: imirkin_: vblank_mode=3 works, nice
07:44 imirkin_: it's in driconf
08:04 fugdnscerd: imirkin_: reinstalled with debian+gdm now i get the black screen after it switches to gdm and wont let me switch back to tty1 i assume due to the -novtswitch option being passed to Xorg from a seemingly nonexistent config file. I disabled TV-1 and VGA-1 in kernel args so all thats left is DVI-D-1 which seems to be the port connected to the internal LCD. Any ideas here? maybe a gdm-nouveau issue as you mentioned before?
08:04 fugdnscerd: imirkin: ^^
08:29 imirkin_: fugdnscerd: logs :)
08:32 Tom^: is this a laptop and relatively new?
08:33 Tom^: if so doesnt that mean he has to run X and stuff on the intel gpu and use prime or what its called to offload on the nvidia gpu?
08:33 Tom^: oh. "GeForce FX 5200" its not new. ;)
08:35 imirkin_: unlikely. not a lot of ppc laptops with nv34's being released.
08:35 imirkin_: although i'm guessing this is actually an imac, not a ibook/powerbook
08:39 fugdnscerd: Its a G5 ppc power8
08:39 fugdnscerd: imac
08:39 fugdnscerd: i will upload logs in a minute
08:40 imirkin_: it's most definitely not a power8 :)
08:40 imirkin_: more like power4 iirc
08:42 fugdnscerd: i ran some command i found somewhere and it said it was a power8,2 i believe
08:42 fugdnscerd: cant remember the command or what log file i greped
08:42 imirkin_: powermac8,2 yes
08:42 imirkin_: power is the cpu architecture, which is also versioned
08:43 fugdnscerd: oh my bad, that was probably it lol
08:44 fugdnscerd: here is my Xorg.log file after booting, sshing into the box, and usint /etc/init.d/gdm3 stop. on the command line i got an error saying the CPU #0 froze for 20 seconds or sohttp://paste2.org/Ye3cdYeh
08:45 fugdnscerd: http://paste2.org/36EVE3vL
08:45 fugdnscerd: here is my dmesg log from the same point in time
08:45 imirkin_: awesome
08:45 fugdnscerd: anything else that would be helpful?
08:48 imirkin_: ok so... (a) the softlockup is probably fixed by booting with nouveau.runpm=0
08:48 imirkin_: (b) what version of mesa do you have installed?
08:49 fugdnscerd: not sure how to check
08:49 fugdnscerd: nouveau.runpm=0 is a kernel boot option?
08:50 fugdnscerd: the lockup doesnt occur until I try to stop gdm3
08:50 fugdnscerd: if that makes a difference
08:51 imirkin_: yeah that makes sense
08:51 imirkin_: it's a pretty old kernel, i suspect this has been worked out
08:53 fugdnscerd: hmm im running dist-upgrade now, I thought it was up to date since I just downloaded the install media a couple days ago and ran apt-get upgrade
08:53 imirkin_: ... it's debian
08:54 imirkin_: they tend to ship older versions of things
08:54 fugdnscerd: i wonder if there is a repository for the newer kernels
08:54 fugdnscerd: would backports have them maybe?
08:54 imirkin_: sorry, i don't do debian
08:55 Tom^: debian is where software has to be moldy before its considered to be updated.
08:55 fugdnscerd: lol
08:56 imirkin_: i dunno, i think their policies make sense for a lot of use-cases
08:56 imirkin_: they make stable distros... stable means "if it worked before, it'll keep working"
08:56 Tom^: it sure does, but it sucks for the home tinkerer. running on outdated software with bugs that is fixed years ago :p
08:56 imirkin_: which is not to be confused with "it works"
08:58 fugdnscerd: Ive had a lot of luck with debian on servers.
08:58 fugdnscerd: i almost refuse to use anything else... almost
08:58 imirkin_: and yet you ask me how to find out the version of a package? a bad sign...
09:00 fugdnscerd: installing kernel 4.2.0-0 now
09:00 fugdnscerd: lol, im a little out of practice
09:00 imirkin_: bad idea
09:00 imirkin_: 4.2.0 won't work for you i'm afraid
09:00 fugdnscerd: yea?
09:01 glennk: imirkin_, didn't you actually have a similar machine?
09:01 Tom^: indeed im not saying debian is bad or not having a place, but i still find it a bit uh lacking when it comes to use it as a desktop. some software on linux tends to be lacking features or have a bit of show stopping bugs to be used properly which gets fixed or implented quite fast once there is an interest in it. so when debian sticks to its "stable" old things you get a bit out of the loop
09:01 imirkin_: fugdnscerd: kernels 3.19..4.2 had an issue in loading the OpenFirmware-based vbios
09:01 imirkin_: fugdnscerd: 4.3.0 should work iirc
09:01 imirkin_: glennk: yes indeed
09:02 fugdnscerd: hmm i may be shifting distros then because besides my current kernel 4.2.0 seems to be my only other 'distro-supported' option
09:02 glennk: last ever agp model if i remember correctly imirkin_ ?
09:02 Tom^: fugdnscerd: you could just build the kernel yourself tho :P
09:02 imirkin_: glennk: nv4a was native agp i think
09:03 glennk: imirkin_, i meant the ppc
09:03 imirkin_: glennk: ah yeah
09:03 imirkin_: mine is a powermac7,3 iirc
09:04 fugdnscerd: Tom^: indeed the reason i say "only other 'distro-supported' option"
09:04 fugdnscerd: lol
09:05 karolherbst: mupuf: so the FSRM should be just configured at nouveau load time with resonable thresholds I guess, any idea where to get the right temperatures out of the vbios? Or should I just use fixed values for that?
09:05 glennk: imirkin_, i think those used the pcie<->agp bridge chip
09:05 glennk: same chip as nv36 etc but mounted backwards
09:06 imirkin_: glennk: i think nv4a was native agp. there were other nv4x's with agp, which had bridge chips.
09:06 imirkin_: aka nv44a
09:07 imirkin_: but perhaps you're right.
09:07 imirkin_: and i guess i now have an agp mobo in that powermac :)
09:11 glennk: imirkin_, hmm, looks like nv40 was native agp, nv41 pcie
09:12 imirkin_: oh right, well there's that too
09:12 imirkin_: but they did make an agp-only nv4a chip too
09:13 glennk: looks like it by the board pictures
09:38 fugdnscerd: imirkin_: what distro are you using on yours
09:39 imirkin_: fugdnscerd: gentoo
09:44 fugdnscerd: i might look into that.. as you expected the 4.2.0 kernel didnt work :-) sorry but I often have to learn the hard way. With the old kernel and the nouveau kernel option you suggested above it seems to boot, still does softlockup occasionally, but now allows me to switch vts. Still black screen but new info in dmesg http://paste2.org/0tDV71JU
09:45 imirkin_: well, all i know is that i got mine to work with 4.3.0 after submitting a few patches after the rewrite that happened there
09:45 imirkin_: as well as a few mesa patches to make it work better but still not great
09:46 imirkin_: i don't have a clear picture in my head of when the BE <-> LE conversions need to be done/not done.
09:47 fugdnscerd: ok well thanks for your help
09:47 imirkin_: you'll want mesa 11.0.x as well
09:54 Guest65072: Hi, I just installed xubuntu 15.10 and am using a gtx750 ti nvidia card. The text was garbled with some parts of the font missing. I've had the problem before with this card on the open source driver and couldn't solve the problem then. This time, I managed to solve it by unchecking 'antialias' in the fonts settings. However, now my fonts look cr*p. Is there anyway to overcome this?
09:57 Guest65072: actually, I've just changed the fonts to bold and increased the size slightly, and they are now fine. Thanks.
09:58 imirkin_: yeah, that seems like the optimal solution. heh.
10:16 joi: karolherbst, prg_: fyi, if you want to rebase mmt to valgrind 3.11, there are up-to-date valgrind & VEX repositories on my github account
10:21 prg_: joi, that would be a lot of work for someone with absolutely no clue of mmt and valgrind internals, right?
10:22 joi: nope
10:22 prg_: huh, should just apply cleanly?
10:22 joi: they don't change API between core and plugins that much
10:23 joi: I think it's possible
10:23 prg_: what's your github account then?
10:23 joi: last time there were some build system changes needed
10:24 joi: https://github.com/mslusarz/valgrind
10:24 prg_: will have a look, thanks
10:48 karolherbst: joi: nice
10:48 karolherbst: joi: I wanted to debug wasteland 2, because it just randomly crashes/hangs for me, but it uses unity
10:49 karolherbst: so it is mono baes
10:49 karolherbst: joi: what's your account name by the way?
10:49 joi: mslusarz, it's 5 lines above :)
10:50 karolherbst: ahh
10:50 karolherbst: yeah found it in the valgrind history :D
10:50 karolherbst: ohhh
10:50 karolherbst: the link :D
10:52 karolherbst: ...
10:52 karolherbst: priv/ir_opt.c:6328:13: error: ‘for’ loop initial declarations are only allowed in C99 or C11 mode
10:53 karolherbst: really :D
10:54 karolherbst: joi: VEX_3_11_BRANCH and VALGRIND_3_11_BRANCH ?
10:54 karolherbst: or which branches do I have to use?
10:54 joi: yeah, those branches
10:55 prg_: only conflicts in NEWS, configure.ac and some xml file in docs/
10:55 prg_: let's see if this actually works...
10:55 karolherbst: prg_: what do you want to do?
10:56 prg_: um, use mmt with valgrind 3.11
10:57 prg_: (hoping it won't crash anymore when tracing wine-preloader)
10:57 karolherbst: prg_: why don't you use joi repositories?
10:57 prg_: mmt needs rebasing?
10:57 karolherbst: ohhh
10:57 karolherbst: I thought the mmt stuff is inside there already
10:57 prg_: rebasing worked fine, but... ../include/pub_tool_basics.h:70:30: error: too many arguments to function ‘vgPlain_get_filename’
10:58 karolherbst: joi: your repostiory contains a 3.11 based mmt already, right?
10:58 karolherbst: or not?
10:59 joi: nope, just bare valgrind
10:59 prg_: now i'd need to understand the code i guess...
11:00 karolherbst: mhhh
11:00 karolherbst: 3.11 also segfaults with wasteland 2 :/
11:00 karolherbst: meh
11:00 karolherbst: ohh okay, then I missunderstood that :D
11:01 karolherbst: ohhhh
11:01 karolherbst: valgrind 3.11 now executes the 3.10 mmt
11:01 karolherbst: :D
12:01 prg_: okay, minor api change. still segfaults though
12:02 prg_: host stacktrace: ==27072== at 0x380080F0: post_mmap.isra.6 (mmt_trace.c:739)
12:05 prg_: if (!FD_ISSET(fd, &trace_fds))
12:05 prg_: how does that segfault?
12:08 prg_: p fd $1 = 4278171452, p trace_fds $2 = {__fds_bits = {0 <repeats 32 times>}}
12:08 prg_: hm.
12:09 prg_: p args $3 = <optimized out>
12:09 prg_: sigh.
12:18 Tom^: karolherbst: whats wrong with mono?
12:18 Tom^: karolherbst: :p
12:38 prg_: well... crash can be avoided with
12:38 prg_: - if (res._isError || (int)fd == -1)
12:38 prg_: + if (res._isError || (int)fd <= -1)
12:38 prg_: a bit above that line, but nfi if that's a reasonable thing to do
12:39 prg_: (still doesn't work, dies with Illegal instruction now)
13:13 pmoreau: Hum… my use of enormous amounts of temporaries as disappeared, as mysteriously as it had appeard :-D
13:13 pmoreau: I won't complain!
14:17 pmoreau: imirkin_: Can store take an immediate as value to store?
14:17 imirkin_: pmoreau: nfc... check envydis
14:19 imirkin_: i don't see any
14:26 mupuf: karolherbst: since you have different thresholds, I would say let's have a look at mmiotraces
14:26 mupuf: and check for the same values in the bios
14:28 karolherbst: mhh makes sense
14:34 imirkin_: pmoreau: always use loadImm and let the propagation passes take care of it fo ryou
15:42 karolherbst: mupuf: so I guess (because the values are pretty high) the FSRM is intented to hit when nouveau already blasts the fan to max, because the thresholds seems to be always above 90°C
15:42 mupuf: yes
15:42 karolherbst: seems like something which isn't meant to be used, but just as a fail safe before something really bad happens
15:43 karolherbst: mwk: your traces is too big :p
15:44 karolherbst: I always remove it before running my loop over the traces, because it just spends minutes with yours :D
15:45 karolherbst: mupuf: what bothers me is, that the blob usually not change those thresholds, only for a handful of cards
15:48 mupuf: karolherbst: well, does the blob at least read them?
15:48 karolherbst: yeah
15:49 mupuf: then it is checking
15:49 mupuf: you may want to set the value to something else before loading the blob
15:49 karolherbst: well on one card it writes the same values back
15:49 mupuf: and check that it actually changes them
15:49 mupuf: brb
15:50 karolherbst: mupuf: and it isn't read out on amxwell
15:54 mupuf: karolherbst: sure, this is the job of the pmu to set that
15:54 mupuf: you can't set it from the host anymore :s
15:54 karolherbst: ahhhh
15:54 karolherbst: :D
15:55 mupuf: do you feel the love? :D
15:55 mupuf: if there is one thing that makes sense, it is this
15:55 mupuf: making sure that the gpu never dies due to overheating
15:55 mupuf: fan management is kind of OK to put in the pmu
15:56 mupuf: but forcing signing the pgraph firmware? Why dude?
16:00 karolherbst: well, maybe it would be possible to just upload a new pgraph firmware on the fly and then inject some code, which writes stuff into other regs, and maleware will have a working screen screener to read out passwords you know then ;)
16:02 karolherbst: anyway, I will head to bed, I hope I will be able to have a working prototype ready this week, any stuff you want to have there?
16:03 imirkin_: RSpliet: if you're still awake, check out the top 2 commits at https://github.com/imirkin/mesa/commits/atomic3 [don't worry, nothing atomic-related], and let me know if they make sense
16:03 imirkin_: i've yet to test them out on hw
16:54 imirkin: neat! i think my short reg thing worked (after a minor bugfix due to weirdo lvalues)
17:15 imirkin: and it looks like the mad arg pref thing works too